libceph: use cursor for bio reads
[linux-3.10.git] / net / ceph / messenger.c
1 #include <linux/ceph/ceph_debug.h>
2
3 #include <linux/crc32c.h>
4 #include <linux/ctype.h>
5 #include <linux/highmem.h>
6 #include <linux/inet.h>
7 #include <linux/kthread.h>
8 #include <linux/net.h>
9 #include <linux/slab.h>
10 #include <linux/socket.h>
11 #include <linux/string.h>
12 #ifdef  CONFIG_BLOCK
13 #include <linux/bio.h>
14 #endif  /* CONFIG_BLOCK */
15 #include <linux/dns_resolver.h>
16 #include <net/tcp.h>
17
18 #include <linux/ceph/libceph.h>
19 #include <linux/ceph/messenger.h>
20 #include <linux/ceph/decode.h>
21 #include <linux/ceph/pagelist.h>
22 #include <linux/export.h>
23
24 #define list_entry_next(pos, member)                                    \
25         list_entry(pos->member.next, typeof(*pos), member)
26
27 /*
28  * Ceph uses the messenger to exchange ceph_msg messages with other
29  * hosts in the system.  The messenger provides ordered and reliable
30  * delivery.  We tolerate TCP disconnects by reconnecting (with
31  * exponential backoff) in the case of a fault (disconnection, bad
32  * crc, protocol error).  Acks allow sent messages to be discarded by
33  * the sender.
34  */
35
36 /*
37  * We track the state of the socket on a given connection using
38  * values defined below.  The transition to a new socket state is
39  * handled by a function which verifies we aren't coming from an
40  * unexpected state.
41  *
42  *      --------
43  *      | NEW* |  transient initial state
44  *      --------
45  *          | con_sock_state_init()
46  *          v
47  *      ----------
48  *      | CLOSED |  initialized, but no socket (and no
49  *      ----------  TCP connection)
50  *       ^      \
51  *       |       \ con_sock_state_connecting()
52  *       |        ----------------------
53  *       |                              \
54  *       + con_sock_state_closed()       \
55  *       |+---------------------------    \
56  *       | \                          \    \
57  *       |  -----------                \    \
58  *       |  | CLOSING |  socket event;  \    \
59  *       |  -----------  await close     \    \
60  *       |       ^                        \   |
61  *       |       |                         \  |
62  *       |       + con_sock_state_closing() \ |
63  *       |      / \                         | |
64  *       |     /   ---------------          | |
65  *       |    /                   \         v v
66  *       |   /                    --------------
67  *       |  /    -----------------| CONNECTING |  socket created, TCP
68  *       |  |   /                 --------------  connect initiated
69  *       |  |   | con_sock_state_connected()
70  *       |  |   v
71  *      -------------
72  *      | CONNECTED |  TCP connection established
73  *      -------------
74  *
75  * State values for ceph_connection->sock_state; NEW is assumed to be 0.
76  */
77
78 #define CON_SOCK_STATE_NEW              0       /* -> CLOSED */
79 #define CON_SOCK_STATE_CLOSED           1       /* -> CONNECTING */
80 #define CON_SOCK_STATE_CONNECTING       2       /* -> CONNECTED or -> CLOSING */
81 #define CON_SOCK_STATE_CONNECTED        3       /* -> CLOSING or -> CLOSED */
82 #define CON_SOCK_STATE_CLOSING          4       /* -> CLOSED */
83
84 /*
85  * connection states
86  */
87 #define CON_STATE_CLOSED        1  /* -> PREOPEN */
88 #define CON_STATE_PREOPEN       2  /* -> CONNECTING, CLOSED */
89 #define CON_STATE_CONNECTING    3  /* -> NEGOTIATING, CLOSED */
90 #define CON_STATE_NEGOTIATING   4  /* -> OPEN, CLOSED */
91 #define CON_STATE_OPEN          5  /* -> STANDBY, CLOSED */
92 #define CON_STATE_STANDBY       6  /* -> PREOPEN, CLOSED */
93
94 /*
95  * ceph_connection flag bits
96  */
97 #define CON_FLAG_LOSSYTX           0  /* we can close channel or drop
98                                        * messages on errors */
99 #define CON_FLAG_KEEPALIVE_PENDING 1  /* we need to send a keepalive */
100 #define CON_FLAG_WRITE_PENDING     2  /* we have data ready to send */
101 #define CON_FLAG_SOCK_CLOSED       3  /* socket state changed to closed */
102 #define CON_FLAG_BACKOFF           4  /* need to retry queuing delayed work */
103
104 static bool con_flag_valid(unsigned long con_flag)
105 {
106         switch (con_flag) {
107         case CON_FLAG_LOSSYTX:
108         case CON_FLAG_KEEPALIVE_PENDING:
109         case CON_FLAG_WRITE_PENDING:
110         case CON_FLAG_SOCK_CLOSED:
111         case CON_FLAG_BACKOFF:
112                 return true;
113         default:
114                 return false;
115         }
116 }
117
118 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
119 {
120         BUG_ON(!con_flag_valid(con_flag));
121
122         clear_bit(con_flag, &con->flags);
123 }
124
125 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
126 {
127         BUG_ON(!con_flag_valid(con_flag));
128
129         set_bit(con_flag, &con->flags);
130 }
131
132 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
133 {
134         BUG_ON(!con_flag_valid(con_flag));
135
136         return test_bit(con_flag, &con->flags);
137 }
138
139 static bool con_flag_test_and_clear(struct ceph_connection *con,
140                                         unsigned long con_flag)
141 {
142         BUG_ON(!con_flag_valid(con_flag));
143
144         return test_and_clear_bit(con_flag, &con->flags);
145 }
146
147 static bool con_flag_test_and_set(struct ceph_connection *con,
148                                         unsigned long con_flag)
149 {
150         BUG_ON(!con_flag_valid(con_flag));
151
152         return test_and_set_bit(con_flag, &con->flags);
153 }
154
155 /* static tag bytes (protocol control messages) */
156 static char tag_msg = CEPH_MSGR_TAG_MSG;
157 static char tag_ack = CEPH_MSGR_TAG_ACK;
158 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
159
160 #ifdef CONFIG_LOCKDEP
161 static struct lock_class_key socket_class;
162 #endif
163
164 /*
165  * When skipping (ignoring) a block of input we read it into a "skip
166  * buffer," which is this many bytes in size.
167  */
168 #define SKIP_BUF_SIZE   1024
169
170 static void queue_con(struct ceph_connection *con);
171 static void con_work(struct work_struct *);
172 static void con_fault(struct ceph_connection *con);
173
174 /*
175  * Nicely render a sockaddr as a string.  An array of formatted
176  * strings is used, to approximate reentrancy.
177  */
178 #define ADDR_STR_COUNT_LOG      5       /* log2(# address strings in array) */
179 #define ADDR_STR_COUNT          (1 << ADDR_STR_COUNT_LOG)
180 #define ADDR_STR_COUNT_MASK     (ADDR_STR_COUNT - 1)
181 #define MAX_ADDR_STR_LEN        64      /* 54 is enough */
182
183 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
184 static atomic_t addr_str_seq = ATOMIC_INIT(0);
185
186 static struct page *zero_page;          /* used in certain error cases */
187
188 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
189 {
190         int i;
191         char *s;
192         struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
193         struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
194
195         i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
196         s = addr_str[i];
197
198         switch (ss->ss_family) {
199         case AF_INET:
200                 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
201                          ntohs(in4->sin_port));
202                 break;
203
204         case AF_INET6:
205                 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
206                          ntohs(in6->sin6_port));
207                 break;
208
209         default:
210                 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
211                          ss->ss_family);
212         }
213
214         return s;
215 }
216 EXPORT_SYMBOL(ceph_pr_addr);
217
218 static void encode_my_addr(struct ceph_messenger *msgr)
219 {
220         memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
221         ceph_encode_addr(&msgr->my_enc_addr);
222 }
223
224 /*
225  * work queue for all reading and writing to/from the socket.
226  */
227 static struct workqueue_struct *ceph_msgr_wq;
228
229 static void _ceph_msgr_exit(void)
230 {
231         if (ceph_msgr_wq) {
232                 destroy_workqueue(ceph_msgr_wq);
233                 ceph_msgr_wq = NULL;
234         }
235
236         BUG_ON(zero_page == NULL);
237         kunmap(zero_page);
238         page_cache_release(zero_page);
239         zero_page = NULL;
240 }
241
242 int ceph_msgr_init(void)
243 {
244         BUG_ON(zero_page != NULL);
245         zero_page = ZERO_PAGE(0);
246         page_cache_get(zero_page);
247
248         ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
249         if (ceph_msgr_wq)
250                 return 0;
251
252         pr_err("msgr_init failed to create workqueue\n");
253         _ceph_msgr_exit();
254
255         return -ENOMEM;
256 }
257 EXPORT_SYMBOL(ceph_msgr_init);
258
259 void ceph_msgr_exit(void)
260 {
261         BUG_ON(ceph_msgr_wq == NULL);
262
263         _ceph_msgr_exit();
264 }
265 EXPORT_SYMBOL(ceph_msgr_exit);
266
267 void ceph_msgr_flush(void)
268 {
269         flush_workqueue(ceph_msgr_wq);
270 }
271 EXPORT_SYMBOL(ceph_msgr_flush);
272
273 /* Connection socket state transition functions */
274
275 static void con_sock_state_init(struct ceph_connection *con)
276 {
277         int old_state;
278
279         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
280         if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
281                 printk("%s: unexpected old state %d\n", __func__, old_state);
282         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
283              CON_SOCK_STATE_CLOSED);
284 }
285
286 static void con_sock_state_connecting(struct ceph_connection *con)
287 {
288         int old_state;
289
290         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
291         if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
292                 printk("%s: unexpected old state %d\n", __func__, old_state);
293         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
294              CON_SOCK_STATE_CONNECTING);
295 }
296
297 static void con_sock_state_connected(struct ceph_connection *con)
298 {
299         int old_state;
300
301         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
302         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
303                 printk("%s: unexpected old state %d\n", __func__, old_state);
304         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
305              CON_SOCK_STATE_CONNECTED);
306 }
307
308 static void con_sock_state_closing(struct ceph_connection *con)
309 {
310         int old_state;
311
312         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
313         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
314                         old_state != CON_SOCK_STATE_CONNECTED &&
315                         old_state != CON_SOCK_STATE_CLOSING))
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_CLOSING);
319 }
320
321 static void con_sock_state_closed(struct ceph_connection *con)
322 {
323         int old_state;
324
325         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
327                     old_state != CON_SOCK_STATE_CLOSING &&
328                     old_state != CON_SOCK_STATE_CONNECTING &&
329                     old_state != CON_SOCK_STATE_CLOSED))
330                 printk("%s: unexpected old state %d\n", __func__, old_state);
331         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
332              CON_SOCK_STATE_CLOSED);
333 }
334
335 /*
336  * socket callback functions
337  */
338
339 /* data available on socket, or listen socket received a connect */
340 static void ceph_sock_data_ready(struct sock *sk, int count_unused)
341 {
342         struct ceph_connection *con = sk->sk_user_data;
343         if (atomic_read(&con->msgr->stopping)) {
344                 return;
345         }
346
347         if (sk->sk_state != TCP_CLOSE_WAIT) {
348                 dout("%s on %p state = %lu, queueing work\n", __func__,
349                      con, con->state);
350                 queue_con(con);
351         }
352 }
353
354 /* socket has buffer space for writing */
355 static void ceph_sock_write_space(struct sock *sk)
356 {
357         struct ceph_connection *con = sk->sk_user_data;
358
359         /* only queue to workqueue if there is data we want to write,
360          * and there is sufficient space in the socket buffer to accept
361          * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
362          * doesn't get called again until try_write() fills the socket
363          * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
364          * and net/core/stream.c:sk_stream_write_space().
365          */
366         if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
367                 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
368                         dout("%s %p queueing write work\n", __func__, con);
369                         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
370                         queue_con(con);
371                 }
372         } else {
373                 dout("%s %p nothing to write\n", __func__, con);
374         }
375 }
376
377 /* socket's state has changed */
378 static void ceph_sock_state_change(struct sock *sk)
379 {
380         struct ceph_connection *con = sk->sk_user_data;
381
382         dout("%s %p state = %lu sk_state = %u\n", __func__,
383              con, con->state, sk->sk_state);
384
385         switch (sk->sk_state) {
386         case TCP_CLOSE:
387                 dout("%s TCP_CLOSE\n", __func__);
388         case TCP_CLOSE_WAIT:
389                 dout("%s TCP_CLOSE_WAIT\n", __func__);
390                 con_sock_state_closing(con);
391                 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
392                 queue_con(con);
393                 break;
394         case TCP_ESTABLISHED:
395                 dout("%s TCP_ESTABLISHED\n", __func__);
396                 con_sock_state_connected(con);
397                 queue_con(con);
398                 break;
399         default:        /* Everything else is uninteresting */
400                 break;
401         }
402 }
403
404 /*
405  * set up socket callbacks
406  */
407 static void set_sock_callbacks(struct socket *sock,
408                                struct ceph_connection *con)
409 {
410         struct sock *sk = sock->sk;
411         sk->sk_user_data = con;
412         sk->sk_data_ready = ceph_sock_data_ready;
413         sk->sk_write_space = ceph_sock_write_space;
414         sk->sk_state_change = ceph_sock_state_change;
415 }
416
417
418 /*
419  * socket helpers
420  */
421
422 /*
423  * initiate connection to a remote socket.
424  */
425 static int ceph_tcp_connect(struct ceph_connection *con)
426 {
427         struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
428         struct socket *sock;
429         int ret;
430
431         BUG_ON(con->sock);
432         ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
433                                IPPROTO_TCP, &sock);
434         if (ret)
435                 return ret;
436         sock->sk->sk_allocation = GFP_NOFS;
437
438 #ifdef CONFIG_LOCKDEP
439         lockdep_set_class(&sock->sk->sk_lock, &socket_class);
440 #endif
441
442         set_sock_callbacks(sock, con);
443
444         dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
445
446         con_sock_state_connecting(con);
447         ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
448                                  O_NONBLOCK);
449         if (ret == -EINPROGRESS) {
450                 dout("connect %s EINPROGRESS sk_state = %u\n",
451                      ceph_pr_addr(&con->peer_addr.in_addr),
452                      sock->sk->sk_state);
453         } else if (ret < 0) {
454                 pr_err("connect %s error %d\n",
455                        ceph_pr_addr(&con->peer_addr.in_addr), ret);
456                 sock_release(sock);
457                 con->error_msg = "connect error";
458
459                 return ret;
460         }
461         con->sock = sock;
462         return 0;
463 }
464
465 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
466 {
467         struct kvec iov = {buf, len};
468         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
469         int r;
470
471         r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
472         if (r == -EAGAIN)
473                 r = 0;
474         return r;
475 }
476
477 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
478                      int page_offset, size_t length)
479 {
480         void *kaddr;
481         int ret;
482
483         BUG_ON(page_offset + length > PAGE_SIZE);
484
485         kaddr = kmap(page);
486         BUG_ON(!kaddr);
487         ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
488         kunmap(page);
489
490         return ret;
491 }
492
493 /*
494  * write something.  @more is true if caller will be sending more data
495  * shortly.
496  */
497 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
498                      size_t kvlen, size_t len, int more)
499 {
500         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
501         int r;
502
503         if (more)
504                 msg.msg_flags |= MSG_MORE;
505         else
506                 msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
507
508         r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
509         if (r == -EAGAIN)
510                 r = 0;
511         return r;
512 }
513
514 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
515                      int offset, size_t size, bool more)
516 {
517         int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
518         int ret;
519
520         ret = kernel_sendpage(sock, page, offset, size, flags);
521         if (ret == -EAGAIN)
522                 ret = 0;
523
524         return ret;
525 }
526
527
528 /*
529  * Shutdown/close the socket for the given connection.
530  */
531 static int con_close_socket(struct ceph_connection *con)
532 {
533         int rc = 0;
534
535         dout("con_close_socket on %p sock %p\n", con, con->sock);
536         if (con->sock) {
537                 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
538                 sock_release(con->sock);
539                 con->sock = NULL;
540         }
541
542         /*
543          * Forcibly clear the SOCK_CLOSED flag.  It gets set
544          * independent of the connection mutex, and we could have
545          * received a socket close event before we had the chance to
546          * shut the socket down.
547          */
548         con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
549
550         con_sock_state_closed(con);
551         return rc;
552 }
553
554 /*
555  * Reset a connection.  Discard all incoming and outgoing messages
556  * and clear *_seq state.
557  */
558 static void ceph_msg_remove(struct ceph_msg *msg)
559 {
560         list_del_init(&msg->list_head);
561         BUG_ON(msg->con == NULL);
562         msg->con->ops->put(msg->con);
563         msg->con = NULL;
564
565         ceph_msg_put(msg);
566 }
567 static void ceph_msg_remove_list(struct list_head *head)
568 {
569         while (!list_empty(head)) {
570                 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
571                                                         list_head);
572                 ceph_msg_remove(msg);
573         }
574 }
575
576 static void reset_connection(struct ceph_connection *con)
577 {
578         /* reset connection, out_queue, msg_ and connect_seq */
579         /* discard existing out_queue and msg_seq */
580         dout("reset_connection %p\n", con);
581         ceph_msg_remove_list(&con->out_queue);
582         ceph_msg_remove_list(&con->out_sent);
583
584         if (con->in_msg) {
585                 BUG_ON(con->in_msg->con != con);
586                 con->in_msg->con = NULL;
587                 ceph_msg_put(con->in_msg);
588                 con->in_msg = NULL;
589                 con->ops->put(con);
590         }
591
592         con->connect_seq = 0;
593         con->out_seq = 0;
594         if (con->out_msg) {
595                 ceph_msg_put(con->out_msg);
596                 con->out_msg = NULL;
597         }
598         con->in_seq = 0;
599         con->in_seq_acked = 0;
600 }
601
602 /*
603  * mark a peer down.  drop any open connections.
604  */
605 void ceph_con_close(struct ceph_connection *con)
606 {
607         mutex_lock(&con->mutex);
608         dout("con_close %p peer %s\n", con,
609              ceph_pr_addr(&con->peer_addr.in_addr));
610         con->state = CON_STATE_CLOSED;
611
612         con_flag_clear(con, CON_FLAG_LOSSYTX);  /* so we retry next connect */
613         con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
614         con_flag_clear(con, CON_FLAG_WRITE_PENDING);
615         con_flag_clear(con, CON_FLAG_BACKOFF);
616
617         reset_connection(con);
618         con->peer_global_seq = 0;
619         cancel_delayed_work(&con->work);
620         con_close_socket(con);
621         mutex_unlock(&con->mutex);
622 }
623 EXPORT_SYMBOL(ceph_con_close);
624
625 /*
626  * Reopen a closed connection, with a new peer address.
627  */
628 void ceph_con_open(struct ceph_connection *con,
629                    __u8 entity_type, __u64 entity_num,
630                    struct ceph_entity_addr *addr)
631 {
632         mutex_lock(&con->mutex);
633         dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
634
635         WARN_ON(con->state != CON_STATE_CLOSED);
636         con->state = CON_STATE_PREOPEN;
637
638         con->peer_name.type = (__u8) entity_type;
639         con->peer_name.num = cpu_to_le64(entity_num);
640
641         memcpy(&con->peer_addr, addr, sizeof(*addr));
642         con->delay = 0;      /* reset backoff memory */
643         mutex_unlock(&con->mutex);
644         queue_con(con);
645 }
646 EXPORT_SYMBOL(ceph_con_open);
647
648 /*
649  * return true if this connection ever successfully opened
650  */
651 bool ceph_con_opened(struct ceph_connection *con)
652 {
653         return con->connect_seq > 0;
654 }
655
656 /*
657  * initialize a new connection.
658  */
659 void ceph_con_init(struct ceph_connection *con, void *private,
660         const struct ceph_connection_operations *ops,
661         struct ceph_messenger *msgr)
662 {
663         dout("con_init %p\n", con);
664         memset(con, 0, sizeof(*con));
665         con->private = private;
666         con->ops = ops;
667         con->msgr = msgr;
668
669         con_sock_state_init(con);
670
671         mutex_init(&con->mutex);
672         INIT_LIST_HEAD(&con->out_queue);
673         INIT_LIST_HEAD(&con->out_sent);
674         INIT_DELAYED_WORK(&con->work, con_work);
675
676         con->state = CON_STATE_CLOSED;
677 }
678 EXPORT_SYMBOL(ceph_con_init);
679
680
681 /*
682  * We maintain a global counter to order connection attempts.  Get
683  * a unique seq greater than @gt.
684  */
685 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
686 {
687         u32 ret;
688
689         spin_lock(&msgr->global_seq_lock);
690         if (msgr->global_seq < gt)
691                 msgr->global_seq = gt;
692         ret = ++msgr->global_seq;
693         spin_unlock(&msgr->global_seq_lock);
694         return ret;
695 }
696
697 static void con_out_kvec_reset(struct ceph_connection *con)
698 {
699         con->out_kvec_left = 0;
700         con->out_kvec_bytes = 0;
701         con->out_kvec_cur = &con->out_kvec[0];
702 }
703
704 static void con_out_kvec_add(struct ceph_connection *con,
705                                 size_t size, void *data)
706 {
707         int index;
708
709         index = con->out_kvec_left;
710         BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
711
712         con->out_kvec[index].iov_len = size;
713         con->out_kvec[index].iov_base = data;
714         con->out_kvec_left++;
715         con->out_kvec_bytes += size;
716 }
717
718 #ifdef CONFIG_BLOCK
719 static void init_bio_iter(struct bio *bio, struct bio **bio_iter,
720                         unsigned int *bio_seg)
721 {
722         if (!bio) {
723                 *bio_iter = NULL;
724                 *bio_seg = 0;
725                 return;
726         }
727         *bio_iter = bio;
728         *bio_seg = (unsigned int) bio->bi_idx;
729 }
730
731 static void iter_bio_next(struct bio **bio_iter, unsigned int *seg)
732 {
733         if (*bio_iter == NULL)
734                 return;
735
736         BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
737
738         (*seg)++;
739         if (*seg == (*bio_iter)->bi_vcnt)
740                 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
741 }
742
743 /*
744  * For a bio data item, a piece is whatever remains of the next
745  * entry in the current bio iovec, or the first entry in the next
746  * bio in the list.
747  */
748 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data *data,
749                                         size_t length)
750 {
751         struct ceph_msg_data_cursor *cursor = &data->cursor;
752         struct bio *bio;
753
754         BUG_ON(data->type != CEPH_MSG_DATA_BIO);
755
756         bio = data->bio;
757         BUG_ON(!bio);
758         BUG_ON(!bio->bi_vcnt);
759
760         cursor->resid = length;
761         cursor->bio = bio;
762         cursor->vector_index = 0;
763         cursor->vector_offset = 0;
764         cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
765 }
766
767 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data *data,
768                                                 size_t *page_offset,
769                                                 size_t *length)
770 {
771         struct ceph_msg_data_cursor *cursor = &data->cursor;
772         struct bio *bio;
773         struct bio_vec *bio_vec;
774         unsigned int index;
775
776         BUG_ON(data->type != CEPH_MSG_DATA_BIO);
777
778         bio = cursor->bio;
779         BUG_ON(!bio);
780
781         index = cursor->vector_index;
782         BUG_ON(index >= (unsigned int) bio->bi_vcnt);
783
784         bio_vec = &bio->bi_io_vec[index];
785         BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
786         *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
787         BUG_ON(*page_offset >= PAGE_SIZE);
788         if (cursor->last_piece) /* pagelist offset is always 0 */
789                 *length = cursor->resid;
790         else
791                 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
792         BUG_ON(*length > PAGE_SIZE);
793         BUG_ON(*length > cursor->resid);
794
795         return bio_vec->bv_page;
796 }
797
798 static bool ceph_msg_data_bio_advance(struct ceph_msg_data *data, size_t bytes)
799 {
800         struct ceph_msg_data_cursor *cursor = &data->cursor;
801         struct bio *bio;
802         struct bio_vec *bio_vec;
803         unsigned int index;
804
805         BUG_ON(data->type != CEPH_MSG_DATA_BIO);
806
807         bio = cursor->bio;
808         BUG_ON(!bio);
809
810         index = cursor->vector_index;
811         BUG_ON(index >= (unsigned int) bio->bi_vcnt);
812         bio_vec = &bio->bi_io_vec[index];
813
814         /* Advance the cursor offset */
815
816         BUG_ON(cursor->resid < bytes);
817         cursor->resid -= bytes;
818         cursor->vector_offset += bytes;
819         if (cursor->vector_offset < bio_vec->bv_len)
820                 return false;   /* more bytes to process in this segment */
821         BUG_ON(cursor->vector_offset != bio_vec->bv_len);
822
823         /* Move on to the next segment, and possibly the next bio */
824
825         if (++index == (unsigned int) bio->bi_vcnt) {
826                 bio = bio->bi_next;
827                 index = 0;
828         }
829         cursor->bio = bio;
830         cursor->vector_index = index;
831         cursor->vector_offset = 0;
832
833         if (!cursor->last_piece) {
834                 BUG_ON(!cursor->resid);
835                 BUG_ON(!bio);
836                 /* A short read is OK, so use <= rather than == */
837                 if (cursor->resid <= bio->bi_io_vec[index].bv_len)
838                         cursor->last_piece = true;
839         }
840
841         return true;
842 }
843 #endif
844
845 /*
846  * For a page array, a piece comes from the first page in the array
847  * that has not already been fully consumed.
848  */
849 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data *data,
850                                         size_t length)
851 {
852         struct ceph_msg_data_cursor *cursor = &data->cursor;
853         int page_count;
854
855         BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
856
857         BUG_ON(!data->pages);
858         BUG_ON(!data->length);
859         BUG_ON(length != data->length);
860
861         cursor->resid = length;
862         page_count = calc_pages_for(data->alignment, (u64)data->length);
863         cursor->page_offset = data->alignment & ~PAGE_MASK;
864         cursor->page_index = 0;
865         BUG_ON(page_count > (int) USHRT_MAX);
866         cursor->page_count = (unsigned short) page_count;
867         cursor->last_piece = length <= PAGE_SIZE;
868 }
869
870 static struct page *ceph_msg_data_pages_next(struct ceph_msg_data *data,
871                                         size_t *page_offset,
872                                         size_t *length)
873 {
874         struct ceph_msg_data_cursor *cursor = &data->cursor;
875
876         BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
877
878         BUG_ON(cursor->page_index >= cursor->page_count);
879         BUG_ON(cursor->page_offset >= PAGE_SIZE);
880
881         *page_offset = cursor->page_offset;
882         if (cursor->last_piece)
883                 *length = cursor->resid;
884         else
885                 *length = PAGE_SIZE - *page_offset;
886
887         return data->pages[cursor->page_index];
888 }
889
890 static bool ceph_msg_data_pages_advance(struct ceph_msg_data *data,
891                                                 size_t bytes)
892 {
893         struct ceph_msg_data_cursor *cursor = &data->cursor;
894
895         BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
896
897         BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
898
899         /* Advance the cursor page offset */
900
901         cursor->resid -= bytes;
902         cursor->page_offset += bytes;
903         if (!bytes || cursor->page_offset & ~PAGE_MASK)
904                 return false;   /* more bytes to process in the current page */
905
906         /* Move on to the next page */
907
908         BUG_ON(cursor->page_index >= cursor->page_count);
909         cursor->page_offset = 0;
910         cursor->page_index++;
911         cursor->last_piece = cursor->resid <= PAGE_SIZE;
912
913         return true;
914 }
915
916 /*
917  * For a pagelist, a piece is whatever remains to be consumed in the
918  * first page in the list, or the front of the next page.
919  */
920 static void ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data *data,
921                                         size_t length)
922 {
923         struct ceph_msg_data_cursor *cursor = &data->cursor;
924         struct ceph_pagelist *pagelist;
925         struct page *page;
926
927         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
928
929         pagelist = data->pagelist;
930         BUG_ON(!pagelist);
931         BUG_ON(length != pagelist->length);
932
933         if (!length)
934                 return;         /* pagelist can be assigned but empty */
935
936         BUG_ON(list_empty(&pagelist->head));
937         page = list_first_entry(&pagelist->head, struct page, lru);
938
939         cursor->resid = length;
940         cursor->page = page;
941         cursor->offset = 0;
942         cursor->last_piece = length <= PAGE_SIZE;
943 }
944
945 static struct page *ceph_msg_data_pagelist_next(struct ceph_msg_data *data,
946                                                 size_t *page_offset,
947                                                 size_t *length)
948 {
949         struct ceph_msg_data_cursor *cursor = &data->cursor;
950         struct ceph_pagelist *pagelist;
951
952         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
953
954         pagelist = data->pagelist;
955         BUG_ON(!pagelist);
956
957         BUG_ON(!cursor->page);
958         BUG_ON(cursor->offset + cursor->resid != pagelist->length);
959
960         *page_offset = cursor->offset & ~PAGE_MASK;
961         if (cursor->last_piece) /* pagelist offset is always 0 */
962                 *length = cursor->resid;
963         else
964                 *length = PAGE_SIZE - *page_offset;
965
966         return data->cursor.page;
967 }
968
969 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data *data,
970                                                 size_t bytes)
971 {
972         struct ceph_msg_data_cursor *cursor = &data->cursor;
973         struct ceph_pagelist *pagelist;
974
975         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
976
977         pagelist = data->pagelist;
978         BUG_ON(!pagelist);
979
980         BUG_ON(cursor->offset + cursor->resid != pagelist->length);
981         BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
982
983         /* Advance the cursor offset */
984
985         cursor->resid -= bytes;
986         cursor->offset += bytes;
987         /* pagelist offset is always 0 */
988         if (!bytes || cursor->offset & ~PAGE_MASK)
989                 return false;   /* more bytes to process in the current page */
990
991         /* Move on to the next page */
992
993         BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
994         cursor->page = list_entry_next(cursor->page, lru);
995         cursor->last_piece = cursor->resid <= PAGE_SIZE;
996
997         return true;
998 }
999
1000 /*
1001  * Message data is handled (sent or received) in pieces, where each
1002  * piece resides on a single page.  The network layer might not
1003  * consume an entire piece at once.  A data item's cursor keeps
1004  * track of which piece is next to process and how much remains to
1005  * be processed in that piece.  It also tracks whether the current
1006  * piece is the last one in the data item.
1007  */
1008 static void ceph_msg_data_cursor_init(struct ceph_msg_data *data,
1009                                         size_t length)
1010 {
1011         switch (data->type) {
1012         case CEPH_MSG_DATA_PAGELIST:
1013                 ceph_msg_data_pagelist_cursor_init(data, length);
1014                 break;
1015         case CEPH_MSG_DATA_PAGES:
1016                 ceph_msg_data_pages_cursor_init(data, length);
1017                 break;
1018 #ifdef CONFIG_BLOCK
1019         case CEPH_MSG_DATA_BIO:
1020                 ceph_msg_data_bio_cursor_init(data, length);
1021                 break;
1022 #endif /* CONFIG_BLOCK */
1023         case CEPH_MSG_DATA_NONE:
1024         default:
1025                 /* BUG(); */
1026                 break;
1027         }
1028 }
1029
1030 /*
1031  * Return the page containing the next piece to process for a given
1032  * data item, and supply the page offset and length of that piece.
1033  * Indicate whether this is the last piece in this data item.
1034  */
1035 static struct page *ceph_msg_data_next(struct ceph_msg_data *data,
1036                                         size_t *page_offset,
1037                                         size_t *length,
1038                                         bool *last_piece)
1039 {
1040         struct page *page;
1041
1042         switch (data->type) {
1043         case CEPH_MSG_DATA_PAGELIST:
1044                 page = ceph_msg_data_pagelist_next(data, page_offset, length);
1045                 break;
1046         case CEPH_MSG_DATA_PAGES:
1047                 page = ceph_msg_data_pages_next(data, page_offset, length);
1048                 break;
1049 #ifdef CONFIG_BLOCK
1050         case CEPH_MSG_DATA_BIO:
1051                 page = ceph_msg_data_bio_next(data, page_offset, length);
1052                 break;
1053 #endif /* CONFIG_BLOCK */
1054         case CEPH_MSG_DATA_NONE:
1055         default:
1056                 page = NULL;
1057                 break;
1058         }
1059         BUG_ON(!page);
1060         BUG_ON(*page_offset + *length > PAGE_SIZE);
1061         BUG_ON(!*length);
1062         if (last_piece)
1063                 *last_piece = data->cursor.last_piece;
1064
1065         return page;
1066 }
1067
1068 /*
1069  * Returns true if the result moves the cursor on to the next piece
1070  * of the data item.
1071  */
1072 static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes)
1073 {
1074         struct ceph_msg_data_cursor *cursor = &data->cursor;
1075         bool new_piece;
1076
1077         BUG_ON(bytes > cursor->resid);
1078         switch (data->type) {
1079         case CEPH_MSG_DATA_PAGELIST:
1080                 new_piece = ceph_msg_data_pagelist_advance(data, bytes);
1081                 break;
1082         case CEPH_MSG_DATA_PAGES:
1083                 new_piece = ceph_msg_data_pages_advance(data, bytes);
1084                 break;
1085 #ifdef CONFIG_BLOCK
1086         case CEPH_MSG_DATA_BIO:
1087                 new_piece = ceph_msg_data_bio_advance(data, bytes);
1088                 break;
1089 #endif /* CONFIG_BLOCK */
1090         case CEPH_MSG_DATA_NONE:
1091         default:
1092                 BUG();
1093                 break;
1094         }
1095
1096         return new_piece;
1097 }
1098
1099 static void prepare_message_data(struct ceph_msg *msg,
1100                                 struct ceph_msg_pos *msg_pos)
1101 {
1102         size_t data_len;
1103
1104         BUG_ON(!msg);
1105
1106         data_len = le32_to_cpu(msg->hdr.data_len);
1107         BUG_ON(!data_len);
1108
1109         /* initialize page iterator */
1110         msg_pos->page = 0;
1111         if (ceph_msg_has_pages(msg))
1112                 msg_pos->page_pos = msg->p.alignment;
1113         else
1114                 msg_pos->page_pos = 0;
1115 #ifdef CONFIG_BLOCK
1116         if (ceph_msg_has_bio(msg))
1117                 init_bio_iter(msg->b.bio, &msg->b.bio_iter, &msg->b.bio_seg);
1118 #endif
1119         msg_pos->data_pos = 0;
1120
1121         /* Initialize data cursors */
1122
1123 #ifdef CONFIG_BLOCK
1124         if (ceph_msg_has_bio(msg))
1125                 ceph_msg_data_cursor_init(&msg->b, data_len);
1126 #endif /* CONFIG_BLOCK */
1127         if (ceph_msg_has_pages(msg))
1128                 ceph_msg_data_cursor_init(&msg->p, data_len);
1129         if (ceph_msg_has_pagelist(msg))
1130                 ceph_msg_data_cursor_init(&msg->l, data_len);
1131
1132         msg_pos->did_page_crc = false;
1133 }
1134
1135 /*
1136  * Prepare footer for currently outgoing message, and finish things
1137  * off.  Assumes out_kvec* are already valid.. we just add on to the end.
1138  */
1139 static void prepare_write_message_footer(struct ceph_connection *con)
1140 {
1141         struct ceph_msg *m = con->out_msg;
1142         int v = con->out_kvec_left;
1143
1144         m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1145
1146         dout("prepare_write_message_footer %p\n", con);
1147         con->out_kvec_is_msg = true;
1148         con->out_kvec[v].iov_base = &m->footer;
1149         con->out_kvec[v].iov_len = sizeof(m->footer);
1150         con->out_kvec_bytes += sizeof(m->footer);
1151         con->out_kvec_left++;
1152         con->out_more = m->more_to_follow;
1153         con->out_msg_done = true;
1154 }
1155
1156 /*
1157  * Prepare headers for the next outgoing message.
1158  */
1159 static void prepare_write_message(struct ceph_connection *con)
1160 {
1161         struct ceph_msg *m;
1162         u32 crc;
1163
1164         con_out_kvec_reset(con);
1165         con->out_kvec_is_msg = true;
1166         con->out_msg_done = false;
1167
1168         /* Sneak an ack in there first?  If we can get it into the same
1169          * TCP packet that's a good thing. */
1170         if (con->in_seq > con->in_seq_acked) {
1171                 con->in_seq_acked = con->in_seq;
1172                 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1173                 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1174                 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1175                         &con->out_temp_ack);
1176         }
1177
1178         BUG_ON(list_empty(&con->out_queue));
1179         m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1180         con->out_msg = m;
1181         BUG_ON(m->con != con);
1182
1183         /* put message on sent list */
1184         ceph_msg_get(m);
1185         list_move_tail(&m->list_head, &con->out_sent);
1186
1187         /*
1188          * only assign outgoing seq # if we haven't sent this message
1189          * yet.  if it is requeued, resend with it's original seq.
1190          */
1191         if (m->needs_out_seq) {
1192                 m->hdr.seq = cpu_to_le64(++con->out_seq);
1193                 m->needs_out_seq = false;
1194         }
1195
1196         dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n",
1197              m, con->out_seq, le16_to_cpu(m->hdr.type),
1198              le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1199              le32_to_cpu(m->hdr.data_len), m->p.length);
1200         BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1201
1202         /* tag + hdr + front + middle */
1203         con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1204         con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1205         con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1206
1207         if (m->middle)
1208                 con_out_kvec_add(con, m->middle->vec.iov_len,
1209                         m->middle->vec.iov_base);
1210
1211         /* fill in crc (except data pages), footer */
1212         crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1213         con->out_msg->hdr.crc = cpu_to_le32(crc);
1214         con->out_msg->footer.flags = 0;
1215
1216         crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1217         con->out_msg->footer.front_crc = cpu_to_le32(crc);
1218         if (m->middle) {
1219                 crc = crc32c(0, m->middle->vec.iov_base,
1220                                 m->middle->vec.iov_len);
1221                 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1222         } else
1223                 con->out_msg->footer.middle_crc = 0;
1224         dout("%s front_crc %u middle_crc %u\n", __func__,
1225              le32_to_cpu(con->out_msg->footer.front_crc),
1226              le32_to_cpu(con->out_msg->footer.middle_crc));
1227
1228         /* is there a data payload? */
1229         con->out_msg->footer.data_crc = 0;
1230         if (m->hdr.data_len) {
1231                 prepare_message_data(con->out_msg, &con->out_msg_pos);
1232                 con->out_more = 1;  /* data + footer will follow */
1233         } else {
1234                 /* no, queue up footer too and be done */
1235                 prepare_write_message_footer(con);
1236         }
1237
1238         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1239 }
1240
1241 /*
1242  * Prepare an ack.
1243  */
1244 static void prepare_write_ack(struct ceph_connection *con)
1245 {
1246         dout("prepare_write_ack %p %llu -> %llu\n", con,
1247              con->in_seq_acked, con->in_seq);
1248         con->in_seq_acked = con->in_seq;
1249
1250         con_out_kvec_reset(con);
1251
1252         con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1253
1254         con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1255         con_out_kvec_add(con, sizeof (con->out_temp_ack),
1256                                 &con->out_temp_ack);
1257
1258         con->out_more = 1;  /* more will follow.. eventually.. */
1259         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1260 }
1261
1262 /*
1263  * Prepare to share the seq during handshake
1264  */
1265 static void prepare_write_seq(struct ceph_connection *con)
1266 {
1267         dout("prepare_write_seq %p %llu -> %llu\n", con,
1268              con->in_seq_acked, con->in_seq);
1269         con->in_seq_acked = con->in_seq;
1270
1271         con_out_kvec_reset(con);
1272
1273         con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1274         con_out_kvec_add(con, sizeof (con->out_temp_ack),
1275                          &con->out_temp_ack);
1276
1277         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1278 }
1279
1280 /*
1281  * Prepare to write keepalive byte.
1282  */
1283 static void prepare_write_keepalive(struct ceph_connection *con)
1284 {
1285         dout("prepare_write_keepalive %p\n", con);
1286         con_out_kvec_reset(con);
1287         con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1288         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1289 }
1290
1291 /*
1292  * Connection negotiation.
1293  */
1294
1295 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1296                                                 int *auth_proto)
1297 {
1298         struct ceph_auth_handshake *auth;
1299
1300         if (!con->ops->get_authorizer) {
1301                 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1302                 con->out_connect.authorizer_len = 0;
1303                 return NULL;
1304         }
1305
1306         /* Can't hold the mutex while getting authorizer */
1307         mutex_unlock(&con->mutex);
1308         auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1309         mutex_lock(&con->mutex);
1310
1311         if (IS_ERR(auth))
1312                 return auth;
1313         if (con->state != CON_STATE_NEGOTIATING)
1314                 return ERR_PTR(-EAGAIN);
1315
1316         con->auth_reply_buf = auth->authorizer_reply_buf;
1317         con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1318         return auth;
1319 }
1320
1321 /*
1322  * We connected to a peer and are saying hello.
1323  */
1324 static void prepare_write_banner(struct ceph_connection *con)
1325 {
1326         con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1327         con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1328                                         &con->msgr->my_enc_addr);
1329
1330         con->out_more = 0;
1331         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1332 }
1333
1334 static int prepare_write_connect(struct ceph_connection *con)
1335 {
1336         unsigned int global_seq = get_global_seq(con->msgr, 0);
1337         int proto;
1338         int auth_proto;
1339         struct ceph_auth_handshake *auth;
1340
1341         switch (con->peer_name.type) {
1342         case CEPH_ENTITY_TYPE_MON:
1343                 proto = CEPH_MONC_PROTOCOL;
1344                 break;
1345         case CEPH_ENTITY_TYPE_OSD:
1346                 proto = CEPH_OSDC_PROTOCOL;
1347                 break;
1348         case CEPH_ENTITY_TYPE_MDS:
1349                 proto = CEPH_MDSC_PROTOCOL;
1350                 break;
1351         default:
1352                 BUG();
1353         }
1354
1355         dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1356              con->connect_seq, global_seq, proto);
1357
1358         con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1359         con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1360         con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1361         con->out_connect.global_seq = cpu_to_le32(global_seq);
1362         con->out_connect.protocol_version = cpu_to_le32(proto);
1363         con->out_connect.flags = 0;
1364
1365         auth_proto = CEPH_AUTH_UNKNOWN;
1366         auth = get_connect_authorizer(con, &auth_proto);
1367         if (IS_ERR(auth))
1368                 return PTR_ERR(auth);
1369
1370         con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1371         con->out_connect.authorizer_len = auth ?
1372                 cpu_to_le32(auth->authorizer_buf_len) : 0;
1373
1374         con_out_kvec_add(con, sizeof (con->out_connect),
1375                                         &con->out_connect);
1376         if (auth && auth->authorizer_buf_len)
1377                 con_out_kvec_add(con, auth->authorizer_buf_len,
1378                                         auth->authorizer_buf);
1379
1380         con->out_more = 0;
1381         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1382
1383         return 0;
1384 }
1385
1386 /*
1387  * write as much of pending kvecs to the socket as we can.
1388  *  1 -> done
1389  *  0 -> socket full, but more to do
1390  * <0 -> error
1391  */
1392 static int write_partial_kvec(struct ceph_connection *con)
1393 {
1394         int ret;
1395
1396         dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1397         while (con->out_kvec_bytes > 0) {
1398                 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1399                                        con->out_kvec_left, con->out_kvec_bytes,
1400                                        con->out_more);
1401                 if (ret <= 0)
1402                         goto out;
1403                 con->out_kvec_bytes -= ret;
1404                 if (con->out_kvec_bytes == 0)
1405                         break;            /* done */
1406
1407                 /* account for full iov entries consumed */
1408                 while (ret >= con->out_kvec_cur->iov_len) {
1409                         BUG_ON(!con->out_kvec_left);
1410                         ret -= con->out_kvec_cur->iov_len;
1411                         con->out_kvec_cur++;
1412                         con->out_kvec_left--;
1413                 }
1414                 /* and for a partially-consumed entry */
1415                 if (ret) {
1416                         con->out_kvec_cur->iov_len -= ret;
1417                         con->out_kvec_cur->iov_base += ret;
1418                 }
1419         }
1420         con->out_kvec_left = 0;
1421         con->out_kvec_is_msg = false;
1422         ret = 1;
1423 out:
1424         dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1425              con->out_kvec_bytes, con->out_kvec_left, ret);
1426         return ret;  /* done! */
1427 }
1428
1429 static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
1430                         size_t len, size_t sent)
1431 {
1432         struct ceph_msg *msg = con->out_msg;
1433         struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
1434         bool need_crc = false;
1435
1436         BUG_ON(!msg);
1437         BUG_ON(!sent);
1438
1439         msg_pos->data_pos += sent;
1440         msg_pos->page_pos += sent;
1441         if (ceph_msg_has_pages(msg))
1442                 need_crc = ceph_msg_data_advance(&msg->p, sent);
1443         else if (ceph_msg_has_pagelist(msg))
1444                 need_crc = ceph_msg_data_advance(&msg->l, sent);
1445 #ifdef CONFIG_BLOCK
1446         else if (ceph_msg_has_bio(msg))
1447                 need_crc = ceph_msg_data_advance(&msg->b, sent);
1448 #endif /* CONFIG_BLOCK */
1449         BUG_ON(need_crc && sent != len);
1450
1451         if (sent < len)
1452                 return;
1453
1454         BUG_ON(sent != len);
1455         msg_pos->page_pos = 0;
1456         msg_pos->page++;
1457         msg_pos->did_page_crc = false;
1458 }
1459
1460 static void in_msg_pos_next(struct ceph_connection *con, size_t len,
1461                                 size_t received)
1462 {
1463         struct ceph_msg *msg = con->in_msg;
1464         struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
1465
1466         BUG_ON(!msg);
1467         BUG_ON(!received);
1468
1469         msg_pos->data_pos += received;
1470         msg_pos->page_pos += received;
1471 #ifdef CONFIG_BLOCK
1472         if (ceph_msg_has_bio(msg))
1473                 (void) ceph_msg_data_advance(&msg->b, received);
1474 #endif /* CONFIG_BLOCK */
1475         if (received < len)
1476                 return;
1477
1478         BUG_ON(received != len);
1479         msg_pos->page_pos = 0;
1480         msg_pos->page++;
1481 #ifdef CONFIG_BLOCK
1482         if (msg->b.bio)
1483                 iter_bio_next(&msg->b.bio_iter, &msg->b.bio_seg);
1484 #endif /* CONFIG_BLOCK */
1485 }
1486
1487 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1488                                 unsigned int page_offset,
1489                                 unsigned int length)
1490 {
1491         char *kaddr;
1492
1493         kaddr = kmap(page);
1494         BUG_ON(kaddr == NULL);
1495         crc = crc32c(crc, kaddr + page_offset, length);
1496         kunmap(page);
1497
1498         return crc;
1499 }
1500 /*
1501  * Write as much message data payload as we can.  If we finish, queue
1502  * up the footer.
1503  *  1 -> done, footer is now queued in out_kvec[].
1504  *  0 -> socket full, but more to do
1505  * <0 -> error
1506  */
1507 static int write_partial_message_data(struct ceph_connection *con)
1508 {
1509         struct ceph_msg *msg = con->out_msg;
1510         struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
1511         unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
1512         bool do_datacrc = !con->msgr->nocrc;
1513         int ret;
1514
1515         dout("%s %p msg %p page %d offset %d\n", __func__,
1516              con, msg, msg_pos->page, msg_pos->page_pos);
1517
1518         /*
1519          * Iterate through each page that contains data to be
1520          * written, and send as much as possible for each.
1521          *
1522          * If we are calculating the data crc (the default), we will
1523          * need to map the page.  If we have no pages, they have
1524          * been revoked, so use the zero page.
1525          */
1526         while (data_len > msg_pos->data_pos) {
1527                 struct page *page;
1528                 size_t page_offset;
1529                 size_t length;
1530                 bool last_piece;
1531
1532                 if (ceph_msg_has_pages(msg)) {
1533                         page = ceph_msg_data_next(&msg->p, &page_offset,
1534                                                         &length, &last_piece);
1535                 } else if (ceph_msg_has_pagelist(msg)) {
1536                         page = ceph_msg_data_next(&msg->l, &page_offset,
1537                                                         &length, &last_piece);
1538 #ifdef CONFIG_BLOCK
1539                 } else if (ceph_msg_has_bio(msg)) {
1540                         page = ceph_msg_data_next(&msg->b, &page_offset,
1541                                                         &length, &last_piece);
1542 #endif
1543                 } else {
1544                         size_t resid = data_len - msg_pos->data_pos;
1545
1546                         page = zero_page;
1547                         page_offset = msg_pos->page_pos;
1548                         length = PAGE_SIZE - page_offset;
1549                         length = min(resid, length);
1550                         last_piece = length == resid;
1551                 }
1552                 if (do_datacrc && !msg_pos->did_page_crc) {
1553                         u32 crc = le32_to_cpu(msg->footer.data_crc);
1554
1555                         crc = ceph_crc32c_page(crc, page, page_offset, length);
1556                         msg->footer.data_crc = cpu_to_le32(crc);
1557                         msg_pos->did_page_crc = true;
1558                 }
1559                 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1560                                       length, last_piece);
1561                 if (ret <= 0)
1562                         goto out;
1563
1564                 out_msg_pos_next(con, page, length, (size_t) ret);
1565         }
1566
1567         dout("%s %p msg %p done\n", __func__, con, msg);
1568
1569         /* prepare and queue up footer, too */
1570         if (!do_datacrc)
1571                 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1572         con_out_kvec_reset(con);
1573         prepare_write_message_footer(con);
1574         ret = 1;
1575 out:
1576         return ret;
1577 }
1578
1579 /*
1580  * write some zeros
1581  */
1582 static int write_partial_skip(struct ceph_connection *con)
1583 {
1584         int ret;
1585
1586         while (con->out_skip > 0) {
1587                 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1588
1589                 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1590                 if (ret <= 0)
1591                         goto out;
1592                 con->out_skip -= ret;
1593         }
1594         ret = 1;
1595 out:
1596         return ret;
1597 }
1598
1599 /*
1600  * Prepare to read connection handshake, or an ack.
1601  */
1602 static void prepare_read_banner(struct ceph_connection *con)
1603 {
1604         dout("prepare_read_banner %p\n", con);
1605         con->in_base_pos = 0;
1606 }
1607
1608 static void prepare_read_connect(struct ceph_connection *con)
1609 {
1610         dout("prepare_read_connect %p\n", con);
1611         con->in_base_pos = 0;
1612 }
1613
1614 static void prepare_read_ack(struct ceph_connection *con)
1615 {
1616         dout("prepare_read_ack %p\n", con);
1617         con->in_base_pos = 0;
1618 }
1619
1620 static void prepare_read_seq(struct ceph_connection *con)
1621 {
1622         dout("prepare_read_seq %p\n", con);
1623         con->in_base_pos = 0;
1624         con->in_tag = CEPH_MSGR_TAG_SEQ;
1625 }
1626
1627 static void prepare_read_tag(struct ceph_connection *con)
1628 {
1629         dout("prepare_read_tag %p\n", con);
1630         con->in_base_pos = 0;
1631         con->in_tag = CEPH_MSGR_TAG_READY;
1632 }
1633
1634 /*
1635  * Prepare to read a message.
1636  */
1637 static int prepare_read_message(struct ceph_connection *con)
1638 {
1639         dout("prepare_read_message %p\n", con);
1640         BUG_ON(con->in_msg != NULL);
1641         con->in_base_pos = 0;
1642         con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1643         return 0;
1644 }
1645
1646
1647 static int read_partial(struct ceph_connection *con,
1648                         int end, int size, void *object)
1649 {
1650         while (con->in_base_pos < end) {
1651                 int left = end - con->in_base_pos;
1652                 int have = size - left;
1653                 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1654                 if (ret <= 0)
1655                         return ret;
1656                 con->in_base_pos += ret;
1657         }
1658         return 1;
1659 }
1660
1661
1662 /*
1663  * Read all or part of the connect-side handshake on a new connection
1664  */
1665 static int read_partial_banner(struct ceph_connection *con)
1666 {
1667         int size;
1668         int end;
1669         int ret;
1670
1671         dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1672
1673         /* peer's banner */
1674         size = strlen(CEPH_BANNER);
1675         end = size;
1676         ret = read_partial(con, end, size, con->in_banner);
1677         if (ret <= 0)
1678                 goto out;
1679
1680         size = sizeof (con->actual_peer_addr);
1681         end += size;
1682         ret = read_partial(con, end, size, &con->actual_peer_addr);
1683         if (ret <= 0)
1684                 goto out;
1685
1686         size = sizeof (con->peer_addr_for_me);
1687         end += size;
1688         ret = read_partial(con, end, size, &con->peer_addr_for_me);
1689         if (ret <= 0)
1690                 goto out;
1691
1692 out:
1693         return ret;
1694 }
1695
1696 static int read_partial_connect(struct ceph_connection *con)
1697 {
1698         int size;
1699         int end;
1700         int ret;
1701
1702         dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1703
1704         size = sizeof (con->in_reply);
1705         end = size;
1706         ret = read_partial(con, end, size, &con->in_reply);
1707         if (ret <= 0)
1708                 goto out;
1709
1710         size = le32_to_cpu(con->in_reply.authorizer_len);
1711         end += size;
1712         ret = read_partial(con, end, size, con->auth_reply_buf);
1713         if (ret <= 0)
1714                 goto out;
1715
1716         dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1717              con, (int)con->in_reply.tag,
1718              le32_to_cpu(con->in_reply.connect_seq),
1719              le32_to_cpu(con->in_reply.global_seq));
1720 out:
1721         return ret;
1722
1723 }
1724
1725 /*
1726  * Verify the hello banner looks okay.
1727  */
1728 static int verify_hello(struct ceph_connection *con)
1729 {
1730         if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1731                 pr_err("connect to %s got bad banner\n",
1732                        ceph_pr_addr(&con->peer_addr.in_addr));
1733                 con->error_msg = "protocol error, bad banner";
1734                 return -1;
1735         }
1736         return 0;
1737 }
1738
1739 static bool addr_is_blank(struct sockaddr_storage *ss)
1740 {
1741         switch (ss->ss_family) {
1742         case AF_INET:
1743                 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1744         case AF_INET6:
1745                 return
1746                      ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1747                      ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1748                      ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1749                      ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1750         }
1751         return false;
1752 }
1753
1754 static int addr_port(struct sockaddr_storage *ss)
1755 {
1756         switch (ss->ss_family) {
1757         case AF_INET:
1758                 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1759         case AF_INET6:
1760                 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1761         }
1762         return 0;
1763 }
1764
1765 static void addr_set_port(struct sockaddr_storage *ss, int p)
1766 {
1767         switch (ss->ss_family) {
1768         case AF_INET:
1769                 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1770                 break;
1771         case AF_INET6:
1772                 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1773                 break;
1774         }
1775 }
1776
1777 /*
1778  * Unlike other *_pton function semantics, zero indicates success.
1779  */
1780 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1781                 char delim, const char **ipend)
1782 {
1783         struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1784         struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1785
1786         memset(ss, 0, sizeof(*ss));
1787
1788         if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1789                 ss->ss_family = AF_INET;
1790                 return 0;
1791         }
1792
1793         if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1794                 ss->ss_family = AF_INET6;
1795                 return 0;
1796         }
1797
1798         return -EINVAL;
1799 }
1800
1801 /*
1802  * Extract hostname string and resolve using kernel DNS facility.
1803  */
1804 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1805 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1806                 struct sockaddr_storage *ss, char delim, const char **ipend)
1807 {
1808         const char *end, *delim_p;
1809         char *colon_p, *ip_addr = NULL;
1810         int ip_len, ret;
1811
1812         /*
1813          * The end of the hostname occurs immediately preceding the delimiter or
1814          * the port marker (':') where the delimiter takes precedence.
1815          */
1816         delim_p = memchr(name, delim, namelen);
1817         colon_p = memchr(name, ':', namelen);
1818
1819         if (delim_p && colon_p)
1820                 end = delim_p < colon_p ? delim_p : colon_p;
1821         else if (!delim_p && colon_p)
1822                 end = colon_p;
1823         else {
1824                 end = delim_p;
1825                 if (!end) /* case: hostname:/ */
1826                         end = name + namelen;
1827         }
1828
1829         if (end <= name)
1830                 return -EINVAL;
1831
1832         /* do dns_resolve upcall */
1833         ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1834         if (ip_len > 0)
1835                 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1836         else
1837                 ret = -ESRCH;
1838
1839         kfree(ip_addr);
1840
1841         *ipend = end;
1842
1843         pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1844                         ret, ret ? "failed" : ceph_pr_addr(ss));
1845
1846         return ret;
1847 }
1848 #else
1849 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1850                 struct sockaddr_storage *ss, char delim, const char **ipend)
1851 {
1852         return -EINVAL;
1853 }
1854 #endif
1855
1856 /*
1857  * Parse a server name (IP or hostname). If a valid IP address is not found
1858  * then try to extract a hostname to resolve using userspace DNS upcall.
1859  */
1860 static int ceph_parse_server_name(const char *name, size_t namelen,
1861                         struct sockaddr_storage *ss, char delim, const char **ipend)
1862 {
1863         int ret;
1864
1865         ret = ceph_pton(name, namelen, ss, delim, ipend);
1866         if (ret)
1867                 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1868
1869         return ret;
1870 }
1871
1872 /*
1873  * Parse an ip[:port] list into an addr array.  Use the default
1874  * monitor port if a port isn't specified.
1875  */
1876 int ceph_parse_ips(const char *c, const char *end,
1877                    struct ceph_entity_addr *addr,
1878                    int max_count, int *count)
1879 {
1880         int i, ret = -EINVAL;
1881         const char *p = c;
1882
1883         dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1884         for (i = 0; i < max_count; i++) {
1885                 const char *ipend;
1886                 struct sockaddr_storage *ss = &addr[i].in_addr;
1887                 int port;
1888                 char delim = ',';
1889
1890                 if (*p == '[') {
1891                         delim = ']';
1892                         p++;
1893                 }
1894
1895                 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1896                 if (ret)
1897                         goto bad;
1898                 ret = -EINVAL;
1899
1900                 p = ipend;
1901
1902                 if (delim == ']') {
1903                         if (*p != ']') {
1904                                 dout("missing matching ']'\n");
1905                                 goto bad;
1906                         }
1907                         p++;
1908                 }
1909
1910                 /* port? */
1911                 if (p < end && *p == ':') {
1912                         port = 0;
1913                         p++;
1914                         while (p < end && *p >= '0' && *p <= '9') {
1915                                 port = (port * 10) + (*p - '0');
1916                                 p++;
1917                         }
1918                         if (port > 65535 || port == 0)
1919                                 goto bad;
1920                 } else {
1921                         port = CEPH_MON_PORT;
1922                 }
1923
1924                 addr_set_port(ss, port);
1925
1926                 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1927
1928                 if (p == end)
1929                         break;
1930                 if (*p != ',')
1931                         goto bad;
1932                 p++;
1933         }
1934
1935         if (p != end)
1936                 goto bad;
1937
1938         if (count)
1939                 *count = i + 1;
1940         return 0;
1941
1942 bad:
1943         pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1944         return ret;
1945 }
1946 EXPORT_SYMBOL(ceph_parse_ips);
1947
1948 static int process_banner(struct ceph_connection *con)
1949 {
1950         dout("process_banner on %p\n", con);
1951
1952         if (verify_hello(con) < 0)
1953                 return -1;
1954
1955         ceph_decode_addr(&con->actual_peer_addr);
1956         ceph_decode_addr(&con->peer_addr_for_me);
1957
1958         /*
1959          * Make sure the other end is who we wanted.  note that the other
1960          * end may not yet know their ip address, so if it's 0.0.0.0, give
1961          * them the benefit of the doubt.
1962          */
1963         if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1964                    sizeof(con->peer_addr)) != 0 &&
1965             !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1966               con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1967                 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1968                            ceph_pr_addr(&con->peer_addr.in_addr),
1969                            (int)le32_to_cpu(con->peer_addr.nonce),
1970                            ceph_pr_addr(&con->actual_peer_addr.in_addr),
1971                            (int)le32_to_cpu(con->actual_peer_addr.nonce));
1972                 con->error_msg = "wrong peer at address";
1973                 return -1;
1974         }
1975
1976         /*
1977          * did we learn our address?
1978          */
1979         if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1980                 int port = addr_port(&con->msgr->inst.addr.in_addr);
1981
1982                 memcpy(&con->msgr->inst.addr.in_addr,
1983                        &con->peer_addr_for_me.in_addr,
1984                        sizeof(con->peer_addr_for_me.in_addr));
1985                 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1986                 encode_my_addr(con->msgr);
1987                 dout("process_banner learned my addr is %s\n",
1988                      ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1989         }
1990
1991         return 0;
1992 }
1993
1994 static int process_connect(struct ceph_connection *con)
1995 {
1996         u64 sup_feat = con->msgr->supported_features;
1997         u64 req_feat = con->msgr->required_features;
1998         u64 server_feat = le64_to_cpu(con->in_reply.features);
1999         int ret;
2000
2001         dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2002
2003         switch (con->in_reply.tag) {
2004         case CEPH_MSGR_TAG_FEATURES:
2005                 pr_err("%s%lld %s feature set mismatch,"
2006                        " my %llx < server's %llx, missing %llx\n",
2007                        ENTITY_NAME(con->peer_name),
2008                        ceph_pr_addr(&con->peer_addr.in_addr),
2009                        sup_feat, server_feat, server_feat & ~sup_feat);
2010                 con->error_msg = "missing required protocol features";
2011                 reset_connection(con);
2012                 return -1;
2013
2014         case CEPH_MSGR_TAG_BADPROTOVER:
2015                 pr_err("%s%lld %s protocol version mismatch,"
2016                        " my %d != server's %d\n",
2017                        ENTITY_NAME(con->peer_name),
2018                        ceph_pr_addr(&con->peer_addr.in_addr),
2019                        le32_to_cpu(con->out_connect.protocol_version),
2020                        le32_to_cpu(con->in_reply.protocol_version));
2021                 con->error_msg = "protocol version mismatch";
2022                 reset_connection(con);
2023                 return -1;
2024
2025         case CEPH_MSGR_TAG_BADAUTHORIZER:
2026                 con->auth_retry++;
2027                 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2028                      con->auth_retry);
2029                 if (con->auth_retry == 2) {
2030                         con->error_msg = "connect authorization failure";
2031                         return -1;
2032                 }
2033                 con_out_kvec_reset(con);
2034                 ret = prepare_write_connect(con);
2035                 if (ret < 0)
2036                         return ret;
2037                 prepare_read_connect(con);
2038                 break;
2039
2040         case CEPH_MSGR_TAG_RESETSESSION:
2041                 /*
2042                  * If we connected with a large connect_seq but the peer
2043                  * has no record of a session with us (no connection, or
2044                  * connect_seq == 0), they will send RESETSESION to indicate
2045                  * that they must have reset their session, and may have
2046                  * dropped messages.
2047                  */
2048                 dout("process_connect got RESET peer seq %u\n",
2049                      le32_to_cpu(con->in_reply.connect_seq));
2050                 pr_err("%s%lld %s connection reset\n",
2051                        ENTITY_NAME(con->peer_name),
2052                        ceph_pr_addr(&con->peer_addr.in_addr));
2053                 reset_connection(con);
2054                 con_out_kvec_reset(con);
2055                 ret = prepare_write_connect(con);
2056                 if (ret < 0)
2057                         return ret;
2058                 prepare_read_connect(con);
2059
2060                 /* Tell ceph about it. */
2061                 mutex_unlock(&con->mutex);
2062                 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2063                 if (con->ops->peer_reset)
2064                         con->ops->peer_reset(con);
2065                 mutex_lock(&con->mutex);
2066                 if (con->state != CON_STATE_NEGOTIATING)
2067                         return -EAGAIN;
2068                 break;
2069
2070         case CEPH_MSGR_TAG_RETRY_SESSION:
2071                 /*
2072                  * If we sent a smaller connect_seq than the peer has, try
2073                  * again with a larger value.
2074                  */
2075                 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2076                      le32_to_cpu(con->out_connect.connect_seq),
2077                      le32_to_cpu(con->in_reply.connect_seq));
2078                 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2079                 con_out_kvec_reset(con);
2080                 ret = prepare_write_connect(con);
2081                 if (ret < 0)
2082                         return ret;
2083                 prepare_read_connect(con);
2084                 break;
2085
2086         case CEPH_MSGR_TAG_RETRY_GLOBAL:
2087                 /*
2088                  * If we sent a smaller global_seq than the peer has, try
2089                  * again with a larger value.
2090                  */
2091                 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2092                      con->peer_global_seq,
2093                      le32_to_cpu(con->in_reply.global_seq));
2094                 get_global_seq(con->msgr,
2095                                le32_to_cpu(con->in_reply.global_seq));
2096                 con_out_kvec_reset(con);
2097                 ret = prepare_write_connect(con);
2098                 if (ret < 0)
2099                         return ret;
2100                 prepare_read_connect(con);
2101                 break;
2102
2103         case CEPH_MSGR_TAG_SEQ:
2104         case CEPH_MSGR_TAG_READY:
2105                 if (req_feat & ~server_feat) {
2106                         pr_err("%s%lld %s protocol feature mismatch,"
2107                                " my required %llx > server's %llx, need %llx\n",
2108                                ENTITY_NAME(con->peer_name),
2109                                ceph_pr_addr(&con->peer_addr.in_addr),
2110                                req_feat, server_feat, req_feat & ~server_feat);
2111                         con->error_msg = "missing required protocol features";
2112                         reset_connection(con);
2113                         return -1;
2114                 }
2115
2116                 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2117                 con->state = CON_STATE_OPEN;
2118                 con->auth_retry = 0;    /* we authenticated; clear flag */
2119                 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2120                 con->connect_seq++;
2121                 con->peer_features = server_feat;
2122                 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2123                      con->peer_global_seq,
2124                      le32_to_cpu(con->in_reply.connect_seq),
2125                      con->connect_seq);
2126                 WARN_ON(con->connect_seq !=
2127                         le32_to_cpu(con->in_reply.connect_seq));
2128
2129                 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2130                         con_flag_set(con, CON_FLAG_LOSSYTX);
2131
2132                 con->delay = 0;      /* reset backoff memory */
2133
2134                 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2135                         prepare_write_seq(con);
2136                         prepare_read_seq(con);
2137                 } else {
2138                         prepare_read_tag(con);
2139                 }
2140                 break;
2141
2142         case CEPH_MSGR_TAG_WAIT:
2143                 /*
2144                  * If there is a connection race (we are opening
2145                  * connections to each other), one of us may just have
2146                  * to WAIT.  This shouldn't happen if we are the
2147                  * client.
2148                  */
2149                 pr_err("process_connect got WAIT as client\n");
2150                 con->error_msg = "protocol error, got WAIT as client";
2151                 return -1;
2152
2153         default:
2154                 pr_err("connect protocol error, will retry\n");
2155                 con->error_msg = "protocol error, garbage tag during connect";
2156                 return -1;
2157         }
2158         return 0;
2159 }
2160
2161
2162 /*
2163  * read (part of) an ack
2164  */
2165 static int read_partial_ack(struct ceph_connection *con)
2166 {
2167         int size = sizeof (con->in_temp_ack);
2168         int end = size;
2169
2170         return read_partial(con, end, size, &con->in_temp_ack);
2171 }
2172
2173 /*
2174  * We can finally discard anything that's been acked.
2175  */
2176 static void process_ack(struct ceph_connection *con)
2177 {
2178         struct ceph_msg *m;
2179         u64 ack = le64_to_cpu(con->in_temp_ack);
2180         u64 seq;
2181
2182         while (!list_empty(&con->out_sent)) {
2183                 m = list_first_entry(&con->out_sent, struct ceph_msg,
2184                                      list_head);
2185                 seq = le64_to_cpu(m->hdr.seq);
2186                 if (seq > ack)
2187                         break;
2188                 dout("got ack for seq %llu type %d at %p\n", seq,
2189                      le16_to_cpu(m->hdr.type), m);
2190                 m->ack_stamp = jiffies;
2191                 ceph_msg_remove(m);
2192         }
2193         prepare_read_tag(con);
2194 }
2195
2196
2197 static int read_partial_message_section(struct ceph_connection *con,
2198                                         struct kvec *section,
2199                                         unsigned int sec_len, u32 *crc)
2200 {
2201         int ret, left;
2202
2203         BUG_ON(!section);
2204
2205         while (section->iov_len < sec_len) {
2206                 BUG_ON(section->iov_base == NULL);
2207                 left = sec_len - section->iov_len;
2208                 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2209                                        section->iov_len, left);
2210                 if (ret <= 0)
2211                         return ret;
2212                 section->iov_len += ret;
2213         }
2214         if (section->iov_len == sec_len)
2215                 *crc = crc32c(0, section->iov_base, section->iov_len);
2216
2217         return 1;
2218 }
2219
2220 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2221
2222 static int read_partial_message_pages(struct ceph_connection *con,
2223                                       unsigned int data_len, bool do_datacrc)
2224 {
2225         struct ceph_msg *msg = con->in_msg;
2226         struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2227         struct page **pages;
2228         struct page *page;
2229         size_t page_offset;
2230         size_t length;
2231         unsigned int left;
2232         int ret;
2233
2234         /* (page) data */
2235         pages = msg->p.pages;
2236         BUG_ON(pages == NULL);
2237         page = pages[msg_pos->page];
2238         page_offset = msg_pos->page_pos;
2239         BUG_ON(msg_pos->data_pos >= data_len);
2240         left = data_len - msg_pos->data_pos;
2241         BUG_ON(page_offset >= PAGE_SIZE);
2242         length = min_t(unsigned int, PAGE_SIZE - page_offset, left);
2243
2244         ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2245         if (ret <= 0)
2246                 return ret;
2247
2248         if (do_datacrc)
2249                 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2250                                                         page_offset, ret);
2251
2252         in_msg_pos_next(con, length, ret);
2253
2254         return ret;
2255 }
2256
2257 #ifdef CONFIG_BLOCK
2258 static int read_partial_message_bio(struct ceph_connection *con,
2259                                     unsigned int data_len, bool do_datacrc)
2260 {
2261         struct ceph_msg *msg = con->in_msg;
2262         struct page *page;
2263         size_t page_offset;
2264         size_t length;
2265         int ret;
2266
2267         BUG_ON(!msg);
2268
2269         page = ceph_msg_data_next(&msg->b, &page_offset, &length, NULL);
2270
2271         ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2272         if (ret <= 0)
2273                 return ret;
2274
2275         if (do_datacrc)
2276                 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2277                                                         page_offset, ret);
2278
2279         in_msg_pos_next(con, length, ret);
2280
2281         return ret;
2282 }
2283 #endif
2284
2285 static int read_partial_msg_data(struct ceph_connection *con)
2286 {
2287         struct ceph_msg *msg = con->in_msg;
2288         struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2289         const bool do_datacrc = !con->msgr->nocrc;
2290         unsigned int data_len;
2291         int ret;
2292
2293         BUG_ON(!msg);
2294
2295         data_len = le32_to_cpu(con->in_hdr.data_len);
2296         while (msg_pos->data_pos < data_len) {
2297                 if (ceph_msg_has_pages(msg)) {
2298                         ret = read_partial_message_pages(con, data_len,
2299                                                         do_datacrc);
2300                         if (ret <= 0)
2301                                 return ret;
2302 #ifdef CONFIG_BLOCK
2303                 } else if (ceph_msg_has_bio(msg)) {
2304                         ret = read_partial_message_bio(con,
2305                                                  data_len, do_datacrc);
2306                         if (ret <= 0)
2307                                 return ret;
2308 #endif
2309                 } else {
2310                         BUG_ON(1);
2311                 }
2312         }
2313
2314         return 1;       /* must return > 0 to indicate success */
2315 }
2316
2317 /*
2318  * read (part of) a message.
2319  */
2320 static int read_partial_message(struct ceph_connection *con)
2321 {
2322         struct ceph_msg *m = con->in_msg;
2323         int size;
2324         int end;
2325         int ret;
2326         unsigned int front_len, middle_len, data_len;
2327         bool do_datacrc = !con->msgr->nocrc;
2328         u64 seq;
2329         u32 crc;
2330
2331         dout("read_partial_message con %p msg %p\n", con, m);
2332
2333         /* header */
2334         size = sizeof (con->in_hdr);
2335         end = size;
2336         ret = read_partial(con, end, size, &con->in_hdr);
2337         if (ret <= 0)
2338                 return ret;
2339
2340         crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2341         if (cpu_to_le32(crc) != con->in_hdr.crc) {
2342                 pr_err("read_partial_message bad hdr "
2343                        " crc %u != expected %u\n",
2344                        crc, con->in_hdr.crc);
2345                 return -EBADMSG;
2346         }
2347
2348         front_len = le32_to_cpu(con->in_hdr.front_len);
2349         if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2350                 return -EIO;
2351         middle_len = le32_to_cpu(con->in_hdr.middle_len);
2352         if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2353                 return -EIO;
2354         data_len = le32_to_cpu(con->in_hdr.data_len);
2355         if (data_len > CEPH_MSG_MAX_DATA_LEN)
2356                 return -EIO;
2357
2358         /* verify seq# */
2359         seq = le64_to_cpu(con->in_hdr.seq);
2360         if ((s64)seq - (s64)con->in_seq < 1) {
2361                 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2362                         ENTITY_NAME(con->peer_name),
2363                         ceph_pr_addr(&con->peer_addr.in_addr),
2364                         seq, con->in_seq + 1);
2365                 con->in_base_pos = -front_len - middle_len - data_len -
2366                         sizeof(m->footer);
2367                 con->in_tag = CEPH_MSGR_TAG_READY;
2368                 return 0;
2369         } else if ((s64)seq - (s64)con->in_seq > 1) {
2370                 pr_err("read_partial_message bad seq %lld expected %lld\n",
2371                        seq, con->in_seq + 1);
2372                 con->error_msg = "bad message sequence # for incoming message";
2373                 return -EBADMSG;
2374         }
2375
2376         /* allocate message? */
2377         if (!con->in_msg) {
2378                 int skip = 0;
2379
2380                 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2381                      front_len, data_len);
2382                 ret = ceph_con_in_msg_alloc(con, &skip);
2383                 if (ret < 0)
2384                         return ret;
2385                 if (skip) {
2386                         /* skip this message */
2387                         dout("alloc_msg said skip message\n");
2388                         BUG_ON(con->in_msg);
2389                         con->in_base_pos = -front_len - middle_len - data_len -
2390                                 sizeof(m->footer);
2391                         con->in_tag = CEPH_MSGR_TAG_READY;
2392                         con->in_seq++;
2393                         return 0;
2394                 }
2395
2396                 BUG_ON(!con->in_msg);
2397                 BUG_ON(con->in_msg->con != con);
2398                 m = con->in_msg;
2399                 m->front.iov_len = 0;    /* haven't read it yet */
2400                 if (m->middle)
2401                         m->middle->vec.iov_len = 0;
2402
2403                 /* prepare for data payload, if any */
2404
2405                 if (data_len)
2406                         prepare_message_data(con->in_msg, &con->in_msg_pos);
2407         }
2408
2409         /* front */
2410         ret = read_partial_message_section(con, &m->front, front_len,
2411                                            &con->in_front_crc);
2412         if (ret <= 0)
2413                 return ret;
2414
2415         /* middle */
2416         if (m->middle) {
2417                 ret = read_partial_message_section(con, &m->middle->vec,
2418                                                    middle_len,
2419                                                    &con->in_middle_crc);
2420                 if (ret <= 0)
2421                         return ret;
2422         }
2423
2424         /* (page) data */
2425         if (data_len) {
2426                 ret = read_partial_msg_data(con);
2427                 if (ret <= 0)
2428                         return ret;
2429         }
2430
2431         /* footer */
2432         size = sizeof (m->footer);
2433         end += size;
2434         ret = read_partial(con, end, size, &m->footer);
2435         if (ret <= 0)
2436                 return ret;
2437
2438         dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2439              m, front_len, m->footer.front_crc, middle_len,
2440              m->footer.middle_crc, data_len, m->footer.data_crc);
2441
2442         /* crc ok? */
2443         if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2444                 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2445                        m, con->in_front_crc, m->footer.front_crc);
2446                 return -EBADMSG;
2447         }
2448         if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2449                 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2450                        m, con->in_middle_crc, m->footer.middle_crc);
2451                 return -EBADMSG;
2452         }
2453         if (do_datacrc &&
2454             (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2455             con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2456                 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2457                        con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2458                 return -EBADMSG;
2459         }
2460
2461         return 1; /* done! */
2462 }
2463
2464 /*
2465  * Process message.  This happens in the worker thread.  The callback should
2466  * be careful not to do anything that waits on other incoming messages or it
2467  * may deadlock.
2468  */
2469 static void process_message(struct ceph_connection *con)
2470 {
2471         struct ceph_msg *msg;
2472
2473         BUG_ON(con->in_msg->con != con);
2474         con->in_msg->con = NULL;
2475         msg = con->in_msg;
2476         con->in_msg = NULL;
2477         con->ops->put(con);
2478
2479         /* if first message, set peer_name */
2480         if (con->peer_name.type == 0)
2481                 con->peer_name = msg->hdr.src;
2482
2483         con->in_seq++;
2484         mutex_unlock(&con->mutex);
2485
2486         dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2487              msg, le64_to_cpu(msg->hdr.seq),
2488              ENTITY_NAME(msg->hdr.src),
2489              le16_to_cpu(msg->hdr.type),
2490              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2491              le32_to_cpu(msg->hdr.front_len),
2492              le32_to_cpu(msg->hdr.data_len),
2493              con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2494         con->ops->dispatch(con, msg);
2495
2496         mutex_lock(&con->mutex);
2497 }
2498
2499
2500 /*
2501  * Write something to the socket.  Called in a worker thread when the
2502  * socket appears to be writeable and we have something ready to send.
2503  */
2504 static int try_write(struct ceph_connection *con)
2505 {
2506         int ret = 1;
2507
2508         dout("try_write start %p state %lu\n", con, con->state);
2509
2510 more:
2511         dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2512
2513         /* open the socket first? */
2514         if (con->state == CON_STATE_PREOPEN) {
2515                 BUG_ON(con->sock);
2516                 con->state = CON_STATE_CONNECTING;
2517
2518                 con_out_kvec_reset(con);
2519                 prepare_write_banner(con);
2520                 prepare_read_banner(con);
2521
2522                 BUG_ON(con->in_msg);
2523                 con->in_tag = CEPH_MSGR_TAG_READY;
2524                 dout("try_write initiating connect on %p new state %lu\n",
2525                      con, con->state);
2526                 ret = ceph_tcp_connect(con);
2527                 if (ret < 0) {
2528                         con->error_msg = "connect error";
2529                         goto out;
2530                 }
2531         }
2532
2533 more_kvec:
2534         /* kvec data queued? */
2535         if (con->out_skip) {
2536                 ret = write_partial_skip(con);
2537                 if (ret <= 0)
2538                         goto out;
2539         }
2540         if (con->out_kvec_left) {
2541                 ret = write_partial_kvec(con);
2542                 if (ret <= 0)
2543                         goto out;
2544         }
2545
2546         /* msg pages? */
2547         if (con->out_msg) {
2548                 if (con->out_msg_done) {
2549                         ceph_msg_put(con->out_msg);
2550                         con->out_msg = NULL;   /* we're done with this one */
2551                         goto do_next;
2552                 }
2553
2554                 ret = write_partial_message_data(con);
2555                 if (ret == 1)
2556                         goto more_kvec;  /* we need to send the footer, too! */
2557                 if (ret == 0)
2558                         goto out;
2559                 if (ret < 0) {
2560                         dout("try_write write_partial_message_data err %d\n",
2561                              ret);
2562                         goto out;
2563                 }
2564         }
2565
2566 do_next:
2567         if (con->state == CON_STATE_OPEN) {
2568                 /* is anything else pending? */
2569                 if (!list_empty(&con->out_queue)) {
2570                         prepare_write_message(con);
2571                         goto more;
2572                 }
2573                 if (con->in_seq > con->in_seq_acked) {
2574                         prepare_write_ack(con);
2575                         goto more;
2576                 }
2577                 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2578                         prepare_write_keepalive(con);
2579                         goto more;
2580                 }
2581         }
2582
2583         /* Nothing to do! */
2584         con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2585         dout("try_write nothing else to write.\n");
2586         ret = 0;
2587 out:
2588         dout("try_write done on %p ret %d\n", con, ret);
2589         return ret;
2590 }
2591
2592
2593
2594 /*
2595  * Read what we can from the socket.
2596  */
2597 static int try_read(struct ceph_connection *con)
2598 {
2599         int ret = -1;
2600
2601 more:
2602         dout("try_read start on %p state %lu\n", con, con->state);
2603         if (con->state != CON_STATE_CONNECTING &&
2604             con->state != CON_STATE_NEGOTIATING &&
2605             con->state != CON_STATE_OPEN)
2606                 return 0;
2607
2608         BUG_ON(!con->sock);
2609
2610         dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2611              con->in_base_pos);
2612
2613         if (con->state == CON_STATE_CONNECTING) {
2614                 dout("try_read connecting\n");
2615                 ret = read_partial_banner(con);
2616                 if (ret <= 0)
2617                         goto out;
2618                 ret = process_banner(con);
2619                 if (ret < 0)
2620                         goto out;
2621
2622                 con->state = CON_STATE_NEGOTIATING;
2623
2624                 /*
2625                  * Received banner is good, exchange connection info.
2626                  * Do not reset out_kvec, as sending our banner raced
2627                  * with receiving peer banner after connect completed.
2628                  */
2629                 ret = prepare_write_connect(con);
2630                 if (ret < 0)
2631                         goto out;
2632                 prepare_read_connect(con);
2633
2634                 /* Send connection info before awaiting response */
2635                 goto out;
2636         }
2637
2638         if (con->state == CON_STATE_NEGOTIATING) {
2639                 dout("try_read negotiating\n");
2640                 ret = read_partial_connect(con);
2641                 if (ret <= 0)
2642                         goto out;
2643                 ret = process_connect(con);
2644                 if (ret < 0)
2645                         goto out;
2646                 goto more;
2647         }
2648
2649         WARN_ON(con->state != CON_STATE_OPEN);
2650
2651         if (con->in_base_pos < 0) {
2652                 /*
2653                  * skipping + discarding content.
2654                  *
2655                  * FIXME: there must be a better way to do this!
2656                  */
2657                 static char buf[SKIP_BUF_SIZE];
2658                 int skip = min((int) sizeof (buf), -con->in_base_pos);
2659
2660                 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2661                 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2662                 if (ret <= 0)
2663                         goto out;
2664                 con->in_base_pos += ret;
2665                 if (con->in_base_pos)
2666                         goto more;
2667         }
2668         if (con->in_tag == CEPH_MSGR_TAG_READY) {
2669                 /*
2670                  * what's next?
2671                  */
2672                 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2673                 if (ret <= 0)
2674                         goto out;
2675                 dout("try_read got tag %d\n", (int)con->in_tag);
2676                 switch (con->in_tag) {
2677                 case CEPH_MSGR_TAG_MSG:
2678                         prepare_read_message(con);
2679                         break;
2680                 case CEPH_MSGR_TAG_ACK:
2681                         prepare_read_ack(con);
2682                         break;
2683                 case CEPH_MSGR_TAG_CLOSE:
2684                         con_close_socket(con);
2685                         con->state = CON_STATE_CLOSED;
2686                         goto out;
2687                 default:
2688                         goto bad_tag;
2689                 }
2690         }
2691         if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2692                 ret = read_partial_message(con);
2693                 if (ret <= 0) {
2694                         switch (ret) {
2695                         case -EBADMSG:
2696                                 con->error_msg = "bad crc";
2697                                 ret = -EIO;
2698                                 break;
2699                         case -EIO:
2700                                 con->error_msg = "io error";
2701                                 break;
2702                         }
2703                         goto out;
2704                 }
2705                 if (con->in_tag == CEPH_MSGR_TAG_READY)
2706                         goto more;
2707                 process_message(con);
2708                 if (con->state == CON_STATE_OPEN)
2709                         prepare_read_tag(con);
2710                 goto more;
2711         }
2712         if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2713             con->in_tag == CEPH_MSGR_TAG_SEQ) {
2714                 /*
2715                  * the final handshake seq exchange is semantically
2716                  * equivalent to an ACK
2717                  */
2718                 ret = read_partial_ack(con);
2719                 if (ret <= 0)
2720                         goto out;
2721                 process_ack(con);
2722                 goto more;
2723         }
2724
2725 out:
2726         dout("try_read done on %p ret %d\n", con, ret);
2727         return ret;
2728
2729 bad_tag:
2730         pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2731         con->error_msg = "protocol error, garbage tag";
2732         ret = -1;
2733         goto out;
2734 }
2735
2736
2737 /*
2738  * Atomically queue work on a connection after the specified delay.
2739  * Bump @con reference to avoid races with connection teardown.
2740  * Returns 0 if work was queued, or an error code otherwise.
2741  */
2742 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2743 {
2744         if (!con->ops->get(con)) {
2745                 dout("%s %p ref count 0\n", __func__, con);
2746
2747                 return -ENOENT;
2748         }
2749
2750         if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2751                 dout("%s %p - already queued\n", __func__, con);
2752                 con->ops->put(con);
2753
2754                 return -EBUSY;
2755         }
2756
2757         dout("%s %p %lu\n", __func__, con, delay);
2758
2759         return 0;
2760 }
2761
2762 static void queue_con(struct ceph_connection *con)
2763 {
2764         (void) queue_con_delay(con, 0);
2765 }
2766
2767 static bool con_sock_closed(struct ceph_connection *con)
2768 {
2769         if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2770                 return false;
2771
2772 #define CASE(x)                                                         \
2773         case CON_STATE_ ## x:                                           \
2774                 con->error_msg = "socket closed (con state " #x ")";    \
2775                 break;
2776
2777         switch (con->state) {
2778         CASE(CLOSED);
2779         CASE(PREOPEN);
2780         CASE(CONNECTING);
2781         CASE(NEGOTIATING);
2782         CASE(OPEN);
2783         CASE(STANDBY);
2784         default:
2785                 pr_warning("%s con %p unrecognized state %lu\n",
2786                         __func__, con, con->state);
2787                 con->error_msg = "unrecognized con state";
2788                 BUG();
2789                 break;
2790         }
2791 #undef CASE
2792
2793         return true;
2794 }
2795
2796 static bool con_backoff(struct ceph_connection *con)
2797 {
2798         int ret;
2799
2800         if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2801                 return false;
2802
2803         ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2804         if (ret) {
2805                 dout("%s: con %p FAILED to back off %lu\n", __func__,
2806                         con, con->delay);
2807                 BUG_ON(ret == -ENOENT);
2808                 con_flag_set(con, CON_FLAG_BACKOFF);
2809         }
2810
2811         return true;
2812 }
2813
2814 /* Finish fault handling; con->mutex must *not* be held here */
2815
2816 static void con_fault_finish(struct ceph_connection *con)
2817 {
2818         /*
2819          * in case we faulted due to authentication, invalidate our
2820          * current tickets so that we can get new ones.
2821          */
2822         if (con->auth_retry && con->ops->invalidate_authorizer) {
2823                 dout("calling invalidate_authorizer()\n");
2824                 con->ops->invalidate_authorizer(con);
2825         }
2826
2827         if (con->ops->fault)
2828                 con->ops->fault(con);
2829 }
2830
2831 /*
2832  * Do some work on a connection.  Drop a connection ref when we're done.
2833  */
2834 static void con_work(struct work_struct *work)
2835 {
2836         struct ceph_connection *con = container_of(work, struct ceph_connection,
2837                                                    work.work);
2838         bool fault;
2839
2840         mutex_lock(&con->mutex);
2841         while (true) {
2842                 int ret;
2843
2844                 if ((fault = con_sock_closed(con))) {
2845                         dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2846                         break;
2847                 }
2848                 if (con_backoff(con)) {
2849                         dout("%s: con %p BACKOFF\n", __func__, con);
2850                         break;
2851                 }
2852                 if (con->state == CON_STATE_STANDBY) {
2853                         dout("%s: con %p STANDBY\n", __func__, con);
2854                         break;
2855                 }
2856                 if (con->state == CON_STATE_CLOSED) {
2857                         dout("%s: con %p CLOSED\n", __func__, con);
2858                         BUG_ON(con->sock);
2859                         break;
2860                 }
2861                 if (con->state == CON_STATE_PREOPEN) {
2862                         dout("%s: con %p PREOPEN\n", __func__, con);
2863                         BUG_ON(con->sock);
2864                 }
2865
2866                 ret = try_read(con);
2867                 if (ret < 0) {
2868                         if (ret == -EAGAIN)
2869                                 continue;
2870                         con->error_msg = "socket error on read";
2871                         fault = true;
2872                         break;
2873                 }
2874
2875                 ret = try_write(con);
2876                 if (ret < 0) {
2877                         if (ret == -EAGAIN)
2878                                 continue;
2879                         con->error_msg = "socket error on write";
2880                         fault = true;
2881                 }
2882
2883                 break;  /* If we make it to here, we're done */
2884         }
2885         if (fault)
2886                 con_fault(con);
2887         mutex_unlock(&con->mutex);
2888
2889         if (fault)
2890                 con_fault_finish(con);
2891
2892         con->ops->put(con);
2893 }
2894
2895 /*
2896  * Generic error/fault handler.  A retry mechanism is used with
2897  * exponential backoff
2898  */
2899 static void con_fault(struct ceph_connection *con)
2900 {
2901         pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2902                ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2903         dout("fault %p state %lu to peer %s\n",
2904              con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2905
2906         WARN_ON(con->state != CON_STATE_CONNECTING &&
2907                con->state != CON_STATE_NEGOTIATING &&
2908                con->state != CON_STATE_OPEN);
2909
2910         con_close_socket(con);
2911
2912         if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2913                 dout("fault on LOSSYTX channel, marking CLOSED\n");
2914                 con->state = CON_STATE_CLOSED;
2915                 return;
2916         }
2917
2918         if (con->in_msg) {
2919                 BUG_ON(con->in_msg->con != con);
2920                 con->in_msg->con = NULL;
2921                 ceph_msg_put(con->in_msg);
2922                 con->in_msg = NULL;
2923                 con->ops->put(con);
2924         }
2925
2926         /* Requeue anything that hasn't been acked */
2927         list_splice_init(&con->out_sent, &con->out_queue);
2928
2929         /* If there are no messages queued or keepalive pending, place
2930          * the connection in a STANDBY state */
2931         if (list_empty(&con->out_queue) &&
2932             !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2933                 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2934                 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2935                 con->state = CON_STATE_STANDBY;
2936         } else {
2937                 /* retry after a delay. */
2938                 con->state = CON_STATE_PREOPEN;
2939                 if (con->delay == 0)
2940                         con->delay = BASE_DELAY_INTERVAL;
2941                 else if (con->delay < MAX_DELAY_INTERVAL)
2942                         con->delay *= 2;
2943                 con_flag_set(con, CON_FLAG_BACKOFF);
2944                 queue_con(con);
2945         }
2946 }
2947
2948
2949
2950 /*
2951  * initialize a new messenger instance
2952  */
2953 void ceph_messenger_init(struct ceph_messenger *msgr,
2954                         struct ceph_entity_addr *myaddr,
2955                         u32 supported_features,
2956                         u32 required_features,
2957                         bool nocrc)
2958 {
2959         msgr->supported_features = supported_features;
2960         msgr->required_features = required_features;
2961
2962         spin_lock_init(&msgr->global_seq_lock);
2963
2964         if (myaddr)
2965                 msgr->inst.addr = *myaddr;
2966
2967         /* select a random nonce */
2968         msgr->inst.addr.type = 0;
2969         get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2970         encode_my_addr(msgr);
2971         msgr->nocrc = nocrc;
2972
2973         atomic_set(&msgr->stopping, 0);
2974
2975         dout("%s %p\n", __func__, msgr);
2976 }
2977 EXPORT_SYMBOL(ceph_messenger_init);
2978
2979 static void clear_standby(struct ceph_connection *con)
2980 {
2981         /* come back from STANDBY? */
2982         if (con->state == CON_STATE_STANDBY) {
2983                 dout("clear_standby %p and ++connect_seq\n", con);
2984                 con->state = CON_STATE_PREOPEN;
2985                 con->connect_seq++;
2986                 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2987                 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2988         }
2989 }
2990
2991 /*
2992  * Queue up an outgoing message on the given connection.
2993  */
2994 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2995 {
2996         /* set src+dst */
2997         msg->hdr.src = con->msgr->inst.name;
2998         BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2999         msg->needs_out_seq = true;
3000
3001         mutex_lock(&con->mutex);
3002
3003         if (con->state == CON_STATE_CLOSED) {
3004                 dout("con_send %p closed, dropping %p\n", con, msg);
3005                 ceph_msg_put(msg);
3006                 mutex_unlock(&con->mutex);
3007                 return;
3008         }
3009
3010         BUG_ON(msg->con != NULL);
3011         msg->con = con->ops->get(con);
3012         BUG_ON(msg->con == NULL);
3013
3014         BUG_ON(!list_empty(&msg->list_head));
3015         list_add_tail(&msg->list_head, &con->out_queue);
3016         dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3017              ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3018              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3019              le32_to_cpu(msg->hdr.front_len),
3020              le32_to_cpu(msg->hdr.middle_len),
3021              le32_to_cpu(msg->hdr.data_len));
3022
3023         clear_standby(con);
3024         mutex_unlock(&con->mutex);
3025
3026         /* if there wasn't anything waiting to send before, queue
3027          * new work */
3028         if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3029                 queue_con(con);
3030 }
3031 EXPORT_SYMBOL(ceph_con_send);
3032
3033 /*
3034  * Revoke a message that was previously queued for send
3035  */
3036 void ceph_msg_revoke(struct ceph_msg *msg)
3037 {
3038         struct ceph_connection *con = msg->con;
3039
3040         if (!con)
3041                 return;         /* Message not in our possession */
3042
3043         mutex_lock(&con->mutex);
3044         if (!list_empty(&msg->list_head)) {
3045                 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3046                 list_del_init(&msg->list_head);
3047                 BUG_ON(msg->con == NULL);
3048                 msg->con->ops->put(msg->con);
3049                 msg->con = NULL;
3050                 msg->hdr.seq = 0;
3051
3052                 ceph_msg_put(msg);
3053         }
3054         if (con->out_msg == msg) {
3055                 dout("%s %p msg %p - was sending\n", __func__, con, msg);
3056                 con->out_msg = NULL;
3057                 if (con->out_kvec_is_msg) {
3058                         con->out_skip = con->out_kvec_bytes;
3059                         con->out_kvec_is_msg = false;
3060                 }
3061                 msg->hdr.seq = 0;
3062
3063                 ceph_msg_put(msg);
3064         }
3065         mutex_unlock(&con->mutex);
3066 }
3067
3068 /*
3069  * Revoke a message that we may be reading data into
3070  */
3071 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3072 {
3073         struct ceph_connection *con;
3074
3075         BUG_ON(msg == NULL);
3076         if (!msg->con) {
3077                 dout("%s msg %p null con\n", __func__, msg);
3078
3079                 return;         /* Message not in our possession */
3080         }
3081
3082         con = msg->con;
3083         mutex_lock(&con->mutex);
3084         if (con->in_msg == msg) {
3085                 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3086                 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3087                 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3088
3089                 /* skip rest of message */
3090                 dout("%s %p msg %p revoked\n", __func__, con, msg);
3091                 con->in_base_pos = con->in_base_pos -
3092                                 sizeof(struct ceph_msg_header) -
3093                                 front_len -
3094                                 middle_len -
3095                                 data_len -
3096                                 sizeof(struct ceph_msg_footer);
3097                 ceph_msg_put(con->in_msg);
3098                 con->in_msg = NULL;
3099                 con->in_tag = CEPH_MSGR_TAG_READY;
3100                 con->in_seq++;
3101         } else {
3102                 dout("%s %p in_msg %p msg %p no-op\n",
3103                      __func__, con, con->in_msg, msg);
3104         }
3105         mutex_unlock(&con->mutex);
3106 }
3107
3108 /*
3109  * Queue a keepalive byte to ensure the tcp connection is alive.
3110  */
3111 void ceph_con_keepalive(struct ceph_connection *con)
3112 {
3113         dout("con_keepalive %p\n", con);
3114         mutex_lock(&con->mutex);
3115         clear_standby(con);
3116         mutex_unlock(&con->mutex);
3117         if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3118             con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3119                 queue_con(con);
3120 }
3121 EXPORT_SYMBOL(ceph_con_keepalive);
3122
3123 static void ceph_msg_data_init(struct ceph_msg_data *data)
3124 {
3125         data->type = CEPH_MSG_DATA_NONE;
3126 }
3127
3128 void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
3129                 size_t length, size_t alignment)
3130 {
3131         BUG_ON(!pages);
3132         BUG_ON(!length);
3133         BUG_ON(msg->p.type != CEPH_MSG_DATA_NONE);
3134
3135         msg->p.type = CEPH_MSG_DATA_PAGES;
3136         msg->p.pages = pages;
3137         msg->p.length = length;
3138         msg->p.alignment = alignment & ~PAGE_MASK;
3139 }
3140 EXPORT_SYMBOL(ceph_msg_data_set_pages);
3141
3142 void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3143                                 struct ceph_pagelist *pagelist)
3144 {
3145         BUG_ON(!pagelist);
3146         BUG_ON(!pagelist->length);
3147         BUG_ON(msg->l.type != CEPH_MSG_DATA_NONE);
3148
3149         msg->l.type = CEPH_MSG_DATA_PAGELIST;
3150         msg->l.pagelist = pagelist;
3151 }
3152 EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3153
3154 void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio)
3155 {
3156         BUG_ON(!bio);
3157         BUG_ON(msg->b.type != CEPH_MSG_DATA_NONE);
3158
3159         msg->b.type = CEPH_MSG_DATA_BIO;
3160         msg->b.bio = bio;
3161 }
3162 EXPORT_SYMBOL(ceph_msg_data_set_bio);
3163
3164 /*
3165  * construct a new message with given type, size
3166  * the new msg has a ref count of 1.
3167  */
3168 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3169                               bool can_fail)
3170 {
3171         struct ceph_msg *m;
3172
3173         m = kzalloc(sizeof(*m), flags);
3174         if (m == NULL)
3175                 goto out;
3176
3177         m->hdr.type = cpu_to_le16(type);
3178         m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3179         m->hdr.front_len = cpu_to_le32(front_len);
3180
3181         INIT_LIST_HEAD(&m->list_head);
3182         kref_init(&m->kref);
3183
3184         ceph_msg_data_init(&m->p);
3185         ceph_msg_data_init(&m->l);
3186         ceph_msg_data_init(&m->b);
3187
3188         /* front */
3189         m->front_max = front_len;
3190         if (front_len) {
3191                 if (front_len > PAGE_CACHE_SIZE) {
3192                         m->front.iov_base = __vmalloc(front_len, flags,
3193                                                       PAGE_KERNEL);
3194                         m->front_is_vmalloc = true;
3195                 } else {
3196                         m->front.iov_base = kmalloc(front_len, flags);
3197                 }
3198                 if (m->front.iov_base == NULL) {
3199                         dout("ceph_msg_new can't allocate %d bytes\n",
3200                              front_len);
3201                         goto out2;
3202                 }
3203         } else {
3204                 m->front.iov_base = NULL;
3205         }
3206         m->front.iov_len = front_len;
3207
3208         dout("ceph_msg_new %p front %d\n", m, front_len);
3209         return m;
3210
3211 out2:
3212         ceph_msg_put(m);
3213 out:
3214         if (!can_fail) {
3215                 pr_err("msg_new can't create type %d front %d\n", type,
3216                        front_len);
3217                 WARN_ON(1);
3218         } else {
3219                 dout("msg_new can't create type %d front %d\n", type,
3220                      front_len);
3221         }
3222         return NULL;
3223 }
3224 EXPORT_SYMBOL(ceph_msg_new);
3225
3226 /*
3227  * Allocate "middle" portion of a message, if it is needed and wasn't
3228  * allocated by alloc_msg.  This allows us to read a small fixed-size
3229  * per-type header in the front and then gracefully fail (i.e.,
3230  * propagate the error to the caller based on info in the front) when
3231  * the middle is too large.
3232  */
3233 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3234 {
3235         int type = le16_to_cpu(msg->hdr.type);
3236         int middle_len = le32_to_cpu(msg->hdr.middle_len);
3237
3238         dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3239              ceph_msg_type_name(type), middle_len);
3240         BUG_ON(!middle_len);
3241         BUG_ON(msg->middle);
3242
3243         msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3244         if (!msg->middle)
3245                 return -ENOMEM;
3246         return 0;
3247 }
3248
3249 /*
3250  * Allocate a message for receiving an incoming message on a
3251  * connection, and save the result in con->in_msg.  Uses the
3252  * connection's private alloc_msg op if available.
3253  *
3254  * Returns 0 on success, or a negative error code.
3255  *
3256  * On success, if we set *skip = 1:
3257  *  - the next message should be skipped and ignored.
3258  *  - con->in_msg == NULL
3259  * or if we set *skip = 0:
3260  *  - con->in_msg is non-null.
3261  * On error (ENOMEM, EAGAIN, ...),
3262  *  - con->in_msg == NULL
3263  */
3264 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3265 {
3266         struct ceph_msg_header *hdr = &con->in_hdr;
3267         int middle_len = le32_to_cpu(hdr->middle_len);
3268         struct ceph_msg *msg;
3269         int ret = 0;
3270
3271         BUG_ON(con->in_msg != NULL);
3272         BUG_ON(!con->ops->alloc_msg);
3273
3274         mutex_unlock(&con->mutex);
3275         msg = con->ops->alloc_msg(con, hdr, skip);
3276         mutex_lock(&con->mutex);
3277         if (con->state != CON_STATE_OPEN) {
3278                 if (msg)
3279                         ceph_msg_put(msg);
3280                 return -EAGAIN;
3281         }
3282         if (msg) {
3283                 BUG_ON(*skip);
3284                 con->in_msg = msg;
3285                 con->in_msg->con = con->ops->get(con);
3286                 BUG_ON(con->in_msg->con == NULL);
3287         } else {
3288                 /*
3289                  * Null message pointer means either we should skip
3290                  * this message or we couldn't allocate memory.  The
3291                  * former is not an error.
3292                  */
3293                 if (*skip)
3294                         return 0;
3295                 con->error_msg = "error allocating memory for incoming message";
3296
3297                 return -ENOMEM;
3298         }
3299         memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3300
3301         if (middle_len && !con->in_msg->middle) {
3302                 ret = ceph_alloc_middle(con, con->in_msg);
3303                 if (ret < 0) {
3304                         ceph_msg_put(con->in_msg);
3305                         con->in_msg = NULL;
3306                 }
3307         }
3308
3309         return ret;
3310 }
3311
3312
3313 /*
3314  * Free a generically kmalloc'd message.
3315  */
3316 void ceph_msg_kfree(struct ceph_msg *m)
3317 {
3318         dout("msg_kfree %p\n", m);
3319         if (m->front_is_vmalloc)
3320                 vfree(m->front.iov_base);
3321         else
3322                 kfree(m->front.iov_base);
3323         kfree(m);
3324 }
3325
3326 /*
3327  * Drop a msg ref.  Destroy as needed.
3328  */
3329 void ceph_msg_last_put(struct kref *kref)
3330 {
3331         struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3332
3333         dout("ceph_msg_put last one on %p\n", m);
3334         WARN_ON(!list_empty(&m->list_head));
3335
3336         /* drop middle, data, if any */
3337         if (m->middle) {
3338                 ceph_buffer_put(m->middle);
3339                 m->middle = NULL;
3340         }
3341         if (ceph_msg_has_pages(m)) {
3342                 m->p.length = 0;
3343                 m->p.pages = NULL;
3344                 m->p.type = CEPH_OSD_DATA_TYPE_NONE;
3345         }
3346         if (ceph_msg_has_pagelist(m)) {
3347                 ceph_pagelist_release(m->l.pagelist);
3348                 kfree(m->l.pagelist);
3349                 m->l.pagelist = NULL;
3350                 m->l.type = CEPH_OSD_DATA_TYPE_NONE;
3351         }
3352         if (ceph_msg_has_bio(m)) {
3353                 m->b.bio = NULL;
3354                 m->b.type = CEPH_OSD_DATA_TYPE_NONE;
3355         }
3356
3357         if (m->pool)
3358                 ceph_msgpool_put(m->pool, m);
3359         else
3360                 ceph_msg_kfree(m);
3361 }
3362 EXPORT_SYMBOL(ceph_msg_last_put);
3363
3364 void ceph_msg_dump(struct ceph_msg *msg)
3365 {
3366         pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
3367                  msg->front_max, msg->p.length);
3368         print_hex_dump(KERN_DEBUG, "header: ",
3369                        DUMP_PREFIX_OFFSET, 16, 1,
3370                        &msg->hdr, sizeof(msg->hdr), true);
3371         print_hex_dump(KERN_DEBUG, " front: ",
3372                        DUMP_PREFIX_OFFSET, 16, 1,
3373                        msg->front.iov_base, msg->front.iov_len, true);
3374         if (msg->middle)
3375                 print_hex_dump(KERN_DEBUG, "middle: ",
3376                                DUMP_PREFIX_OFFSET, 16, 1,
3377                                msg->middle->vec.iov_base,
3378                                msg->middle->vec.iov_len, true);
3379         print_hex_dump(KERN_DEBUG, "footer: ",
3380                        DUMP_PREFIX_OFFSET, 16, 1,
3381                        &msg->footer, sizeof(msg->footer), true);
3382 }
3383 EXPORT_SYMBOL(ceph_msg_dump);