Merge tag 'ceph-for-4.20-rc1' of git://github.com/ceph/ceph-client
[sfrench/cifs-2.6.git] / net / ceph / messenger.c
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/crc32c.h>
5 #include <linux/ctype.h>
6 #include <linux/highmem.h>
7 #include <linux/inet.h>
8 #include <linux/kthread.h>
9 #include <linux/net.h>
10 #include <linux/nsproxy.h>
11 #include <linux/sched/mm.h>
12 #include <linux/slab.h>
13 #include <linux/socket.h>
14 #include <linux/string.h>
15 #ifdef  CONFIG_BLOCK
16 #include <linux/bio.h>
17 #endif  /* CONFIG_BLOCK */
18 #include <linux/dns_resolver.h>
19 #include <net/tcp.h>
20
21 #include <linux/ceph/ceph_features.h>
22 #include <linux/ceph/libceph.h>
23 #include <linux/ceph/messenger.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/pagelist.h>
26 #include <linux/export.h>
27
28 /*
29  * Ceph uses the messenger to exchange ceph_msg messages with other
30  * hosts in the system.  The messenger provides ordered and reliable
31  * delivery.  We tolerate TCP disconnects by reconnecting (with
32  * exponential backoff) in the case of a fault (disconnection, bad
33  * crc, protocol error).  Acks allow sent messages to be discarded by
34  * the sender.
35  */
36
37 /*
38  * We track the state of the socket on a given connection using
39  * values defined below.  The transition to a new socket state is
40  * handled by a function which verifies we aren't coming from an
41  * unexpected state.
42  *
43  *      --------
44  *      | NEW* |  transient initial state
45  *      --------
46  *          | con_sock_state_init()
47  *          v
48  *      ----------
49  *      | CLOSED |  initialized, but no socket (and no
50  *      ----------  TCP connection)
51  *       ^      \
52  *       |       \ con_sock_state_connecting()
53  *       |        ----------------------
54  *       |                              \
55  *       + con_sock_state_closed()       \
56  *       |+---------------------------    \
57  *       | \                          \    \
58  *       |  -----------                \    \
59  *       |  | CLOSING |  socket event;  \    \
60  *       |  -----------  await close     \    \
61  *       |       ^                        \   |
62  *       |       |                         \  |
63  *       |       + con_sock_state_closing() \ |
64  *       |      / \                         | |
65  *       |     /   ---------------          | |
66  *       |    /                   \         v v
67  *       |   /                    --------------
68  *       |  /    -----------------| CONNECTING |  socket created, TCP
69  *       |  |   /                 --------------  connect initiated
70  *       |  |   | con_sock_state_connected()
71  *       |  |   v
72  *      -------------
73  *      | CONNECTED |  TCP connection established
74  *      -------------
75  *
76  * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77  */
78
79 #define CON_SOCK_STATE_NEW              0       /* -> CLOSED */
80 #define CON_SOCK_STATE_CLOSED           1       /* -> CONNECTING */
81 #define CON_SOCK_STATE_CONNECTING       2       /* -> CONNECTED or -> CLOSING */
82 #define CON_SOCK_STATE_CONNECTED        3       /* -> CLOSING or -> CLOSED */
83 #define CON_SOCK_STATE_CLOSING          4       /* -> CLOSED */
84
85 /*
86  * connection states
87  */
88 #define CON_STATE_CLOSED        1  /* -> PREOPEN */
89 #define CON_STATE_PREOPEN       2  /* -> CONNECTING, CLOSED */
90 #define CON_STATE_CONNECTING    3  /* -> NEGOTIATING, CLOSED */
91 #define CON_STATE_NEGOTIATING   4  /* -> OPEN, CLOSED */
92 #define CON_STATE_OPEN          5  /* -> STANDBY, CLOSED */
93 #define CON_STATE_STANDBY       6  /* -> PREOPEN, CLOSED */
94
95 /*
96  * ceph_connection flag bits
97  */
98 #define CON_FLAG_LOSSYTX           0  /* we can close channel or drop
99                                        * messages on errors */
100 #define CON_FLAG_KEEPALIVE_PENDING 1  /* we need to send a keepalive */
101 #define CON_FLAG_WRITE_PENDING     2  /* we have data ready to send */
102 #define CON_FLAG_SOCK_CLOSED       3  /* socket state changed to closed */
103 #define CON_FLAG_BACKOFF           4  /* need to retry queuing delayed work */
104
105 static bool con_flag_valid(unsigned long con_flag)
106 {
107         switch (con_flag) {
108         case CON_FLAG_LOSSYTX:
109         case CON_FLAG_KEEPALIVE_PENDING:
110         case CON_FLAG_WRITE_PENDING:
111         case CON_FLAG_SOCK_CLOSED:
112         case CON_FLAG_BACKOFF:
113                 return true;
114         default:
115                 return false;
116         }
117 }
118
119 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120 {
121         BUG_ON(!con_flag_valid(con_flag));
122
123         clear_bit(con_flag, &con->flags);
124 }
125
126 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127 {
128         BUG_ON(!con_flag_valid(con_flag));
129
130         set_bit(con_flag, &con->flags);
131 }
132
133 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134 {
135         BUG_ON(!con_flag_valid(con_flag));
136
137         return test_bit(con_flag, &con->flags);
138 }
139
140 static bool con_flag_test_and_clear(struct ceph_connection *con,
141                                         unsigned long con_flag)
142 {
143         BUG_ON(!con_flag_valid(con_flag));
144
145         return test_and_clear_bit(con_flag, &con->flags);
146 }
147
148 static bool con_flag_test_and_set(struct ceph_connection *con,
149                                         unsigned long con_flag)
150 {
151         BUG_ON(!con_flag_valid(con_flag));
152
153         return test_and_set_bit(con_flag, &con->flags);
154 }
155
156 /* Slab caches for frequently-allocated structures */
157
158 static struct kmem_cache        *ceph_msg_cache;
159
160 /* static tag bytes (protocol control messages) */
161 static char tag_msg = CEPH_MSGR_TAG_MSG;
162 static char tag_ack = CEPH_MSGR_TAG_ACK;
163 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
164 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
165
166 #ifdef CONFIG_LOCKDEP
167 static struct lock_class_key socket_class;
168 #endif
169
170 static void queue_con(struct ceph_connection *con);
171 static void cancel_con(struct ceph_connection *con);
172 static void ceph_con_workfn(struct work_struct *);
173 static void con_fault(struct ceph_connection *con);
174
175 /*
176  * Nicely render a sockaddr as a string.  An array of formatted
177  * strings is used, to approximate reentrancy.
178  */
179 #define ADDR_STR_COUNT_LOG      5       /* log2(# address strings in array) */
180 #define ADDR_STR_COUNT          (1 << ADDR_STR_COUNT_LOG)
181 #define ADDR_STR_COUNT_MASK     (ADDR_STR_COUNT - 1)
182 #define MAX_ADDR_STR_LEN        64      /* 54 is enough */
183
184 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
185 static atomic_t addr_str_seq = ATOMIC_INIT(0);
186
187 static struct page *zero_page;          /* used in certain error cases */
188
189 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
190 {
191         int i;
192         char *s;
193         struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
194         struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
195
196         i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
197         s = addr_str[i];
198
199         switch (ss->ss_family) {
200         case AF_INET:
201                 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
202                          ntohs(in4->sin_port));
203                 break;
204
205         case AF_INET6:
206                 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
207                          ntohs(in6->sin6_port));
208                 break;
209
210         default:
211                 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
212                          ss->ss_family);
213         }
214
215         return s;
216 }
217 EXPORT_SYMBOL(ceph_pr_addr);
218
219 static void encode_my_addr(struct ceph_messenger *msgr)
220 {
221         memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
222         ceph_encode_addr(&msgr->my_enc_addr);
223 }
224
225 /*
226  * work queue for all reading and writing to/from the socket.
227  */
228 static struct workqueue_struct *ceph_msgr_wq;
229
230 static int ceph_msgr_slab_init(void)
231 {
232         BUG_ON(ceph_msg_cache);
233         ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
234         if (!ceph_msg_cache)
235                 return -ENOMEM;
236
237         return 0;
238 }
239
240 static void ceph_msgr_slab_exit(void)
241 {
242         BUG_ON(!ceph_msg_cache);
243         kmem_cache_destroy(ceph_msg_cache);
244         ceph_msg_cache = NULL;
245 }
246
247 static void _ceph_msgr_exit(void)
248 {
249         if (ceph_msgr_wq) {
250                 destroy_workqueue(ceph_msgr_wq);
251                 ceph_msgr_wq = NULL;
252         }
253
254         BUG_ON(zero_page == NULL);
255         put_page(zero_page);
256         zero_page = NULL;
257
258         ceph_msgr_slab_exit();
259 }
260
261 int __init ceph_msgr_init(void)
262 {
263         if (ceph_msgr_slab_init())
264                 return -ENOMEM;
265
266         BUG_ON(zero_page != NULL);
267         zero_page = ZERO_PAGE(0);
268         get_page(zero_page);
269
270         /*
271          * The number of active work items is limited by the number of
272          * connections, so leave @max_active at default.
273          */
274         ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
275         if (ceph_msgr_wq)
276                 return 0;
277
278         pr_err("msgr_init failed to create workqueue\n");
279         _ceph_msgr_exit();
280
281         return -ENOMEM;
282 }
283
284 void ceph_msgr_exit(void)
285 {
286         BUG_ON(ceph_msgr_wq == NULL);
287
288         _ceph_msgr_exit();
289 }
290
291 void ceph_msgr_flush(void)
292 {
293         flush_workqueue(ceph_msgr_wq);
294 }
295 EXPORT_SYMBOL(ceph_msgr_flush);
296
297 /* Connection socket state transition functions */
298
299 static void con_sock_state_init(struct ceph_connection *con)
300 {
301         int old_state;
302
303         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
304         if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
305                 printk("%s: unexpected old state %d\n", __func__, old_state);
306         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
307              CON_SOCK_STATE_CLOSED);
308 }
309
310 static void con_sock_state_connecting(struct ceph_connection *con)
311 {
312         int old_state;
313
314         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
315         if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
316                 printk("%s: unexpected old state %d\n", __func__, old_state);
317         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
318              CON_SOCK_STATE_CONNECTING);
319 }
320
321 static void con_sock_state_connected(struct ceph_connection *con)
322 {
323         int old_state;
324
325         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
326         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
327                 printk("%s: unexpected old state %d\n", __func__, old_state);
328         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
329              CON_SOCK_STATE_CONNECTED);
330 }
331
332 static void con_sock_state_closing(struct ceph_connection *con)
333 {
334         int old_state;
335
336         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
337         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
338                         old_state != CON_SOCK_STATE_CONNECTED &&
339                         old_state != CON_SOCK_STATE_CLOSING))
340                 printk("%s: unexpected old state %d\n", __func__, old_state);
341         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
342              CON_SOCK_STATE_CLOSING);
343 }
344
345 static void con_sock_state_closed(struct ceph_connection *con)
346 {
347         int old_state;
348
349         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
350         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
351                     old_state != CON_SOCK_STATE_CLOSING &&
352                     old_state != CON_SOCK_STATE_CONNECTING &&
353                     old_state != CON_SOCK_STATE_CLOSED))
354                 printk("%s: unexpected old state %d\n", __func__, old_state);
355         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
356              CON_SOCK_STATE_CLOSED);
357 }
358
359 /*
360  * socket callback functions
361  */
362
363 /* data available on socket, or listen socket received a connect */
364 static void ceph_sock_data_ready(struct sock *sk)
365 {
366         struct ceph_connection *con = sk->sk_user_data;
367         if (atomic_read(&con->msgr->stopping)) {
368                 return;
369         }
370
371         if (sk->sk_state != TCP_CLOSE_WAIT) {
372                 dout("%s on %p state = %lu, queueing work\n", __func__,
373                      con, con->state);
374                 queue_con(con);
375         }
376 }
377
378 /* socket has buffer space for writing */
379 static void ceph_sock_write_space(struct sock *sk)
380 {
381         struct ceph_connection *con = sk->sk_user_data;
382
383         /* only queue to workqueue if there is data we want to write,
384          * and there is sufficient space in the socket buffer to accept
385          * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
386          * doesn't get called again until try_write() fills the socket
387          * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
388          * and net/core/stream.c:sk_stream_write_space().
389          */
390         if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
391                 if (sk_stream_is_writeable(sk)) {
392                         dout("%s %p queueing write work\n", __func__, con);
393                         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
394                         queue_con(con);
395                 }
396         } else {
397                 dout("%s %p nothing to write\n", __func__, con);
398         }
399 }
400
401 /* socket's state has changed */
402 static void ceph_sock_state_change(struct sock *sk)
403 {
404         struct ceph_connection *con = sk->sk_user_data;
405
406         dout("%s %p state = %lu sk_state = %u\n", __func__,
407              con, con->state, sk->sk_state);
408
409         switch (sk->sk_state) {
410         case TCP_CLOSE:
411                 dout("%s TCP_CLOSE\n", __func__);
412                 /* fall through */
413         case TCP_CLOSE_WAIT:
414                 dout("%s TCP_CLOSE_WAIT\n", __func__);
415                 con_sock_state_closing(con);
416                 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
417                 queue_con(con);
418                 break;
419         case TCP_ESTABLISHED:
420                 dout("%s TCP_ESTABLISHED\n", __func__);
421                 con_sock_state_connected(con);
422                 queue_con(con);
423                 break;
424         default:        /* Everything else is uninteresting */
425                 break;
426         }
427 }
428
429 /*
430  * set up socket callbacks
431  */
432 static void set_sock_callbacks(struct socket *sock,
433                                struct ceph_connection *con)
434 {
435         struct sock *sk = sock->sk;
436         sk->sk_user_data = con;
437         sk->sk_data_ready = ceph_sock_data_ready;
438         sk->sk_write_space = ceph_sock_write_space;
439         sk->sk_state_change = ceph_sock_state_change;
440 }
441
442
443 /*
444  * socket helpers
445  */
446
447 /*
448  * initiate connection to a remote socket.
449  */
450 static int ceph_tcp_connect(struct ceph_connection *con)
451 {
452         struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
453         struct socket *sock;
454         unsigned int noio_flag;
455         int ret;
456
457         BUG_ON(con->sock);
458
459         /* sock_create_kern() allocates with GFP_KERNEL */
460         noio_flag = memalloc_noio_save();
461         ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
462                                SOCK_STREAM, IPPROTO_TCP, &sock);
463         memalloc_noio_restore(noio_flag);
464         if (ret)
465                 return ret;
466         sock->sk->sk_allocation = GFP_NOFS;
467
468 #ifdef CONFIG_LOCKDEP
469         lockdep_set_class(&sock->sk->sk_lock, &socket_class);
470 #endif
471
472         set_sock_callbacks(sock, con);
473
474         dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
475
476         con_sock_state_connecting(con);
477         ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
478                                  O_NONBLOCK);
479         if (ret == -EINPROGRESS) {
480                 dout("connect %s EINPROGRESS sk_state = %u\n",
481                      ceph_pr_addr(&con->peer_addr.in_addr),
482                      sock->sk->sk_state);
483         } else if (ret < 0) {
484                 pr_err("connect %s error %d\n",
485                        ceph_pr_addr(&con->peer_addr.in_addr), ret);
486                 sock_release(sock);
487                 return ret;
488         }
489
490         if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
491                 int optval = 1;
492
493                 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
494                                         (char *)&optval, sizeof(optval));
495                 if (ret)
496                         pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
497                                ret);
498         }
499
500         con->sock = sock;
501         return 0;
502 }
503
504 /*
505  * If @buf is NULL, discard up to @len bytes.
506  */
507 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
508 {
509         struct kvec iov = {buf, len};
510         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
511         int r;
512
513         if (!buf)
514                 msg.msg_flags |= MSG_TRUNC;
515
516         iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, &iov, 1, len);
517         r = sock_recvmsg(sock, &msg, msg.msg_flags);
518         if (r == -EAGAIN)
519                 r = 0;
520         return r;
521 }
522
523 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
524                      int page_offset, size_t length)
525 {
526         struct bio_vec bvec = {
527                 .bv_page = page,
528                 .bv_offset = page_offset,
529                 .bv_len = length
530         };
531         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
532         int r;
533
534         BUG_ON(page_offset + length > PAGE_SIZE);
535         iov_iter_bvec(&msg.msg_iter, READ | ITER_BVEC, &bvec, 1, length);
536         r = sock_recvmsg(sock, &msg, msg.msg_flags);
537         if (r == -EAGAIN)
538                 r = 0;
539         return r;
540 }
541
542 /*
543  * write something.  @more is true if caller will be sending more data
544  * shortly.
545  */
546 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
547                      size_t kvlen, size_t len, int more)
548 {
549         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
550         int r;
551
552         if (more)
553                 msg.msg_flags |= MSG_MORE;
554         else
555                 msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
556
557         r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
558         if (r == -EAGAIN)
559                 r = 0;
560         return r;
561 }
562
563 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
564                      int offset, size_t size, bool more)
565 {
566         int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
567         int ret;
568
569         ret = kernel_sendpage(sock, page, offset, size, flags);
570         if (ret == -EAGAIN)
571                 ret = 0;
572
573         return ret;
574 }
575
576 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
577                      int offset, size_t size, bool more)
578 {
579         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
580         struct bio_vec bvec;
581         int ret;
582
583         /* sendpage cannot properly handle pages with page_count == 0,
584          * we need to fallback to sendmsg if that's the case */
585         if (page_count(page) >= 1)
586                 return __ceph_tcp_sendpage(sock, page, offset, size, more);
587
588         bvec.bv_page = page;
589         bvec.bv_offset = offset;
590         bvec.bv_len = size;
591
592         if (more)
593                 msg.msg_flags |= MSG_MORE;
594         else
595                 msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
596
597         iov_iter_bvec(&msg.msg_iter, WRITE | ITER_BVEC, &bvec, 1, size);
598         ret = sock_sendmsg(sock, &msg);
599         if (ret == -EAGAIN)
600                 ret = 0;
601
602         return ret;
603 }
604
605 /*
606  * Shutdown/close the socket for the given connection.
607  */
608 static int con_close_socket(struct ceph_connection *con)
609 {
610         int rc = 0;
611
612         dout("con_close_socket on %p sock %p\n", con, con->sock);
613         if (con->sock) {
614                 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
615                 sock_release(con->sock);
616                 con->sock = NULL;
617         }
618
619         /*
620          * Forcibly clear the SOCK_CLOSED flag.  It gets set
621          * independent of the connection mutex, and we could have
622          * received a socket close event before we had the chance to
623          * shut the socket down.
624          */
625         con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
626
627         con_sock_state_closed(con);
628         return rc;
629 }
630
631 /*
632  * Reset a connection.  Discard all incoming and outgoing messages
633  * and clear *_seq state.
634  */
635 static void ceph_msg_remove(struct ceph_msg *msg)
636 {
637         list_del_init(&msg->list_head);
638
639         ceph_msg_put(msg);
640 }
641 static void ceph_msg_remove_list(struct list_head *head)
642 {
643         while (!list_empty(head)) {
644                 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
645                                                         list_head);
646                 ceph_msg_remove(msg);
647         }
648 }
649
650 static void reset_connection(struct ceph_connection *con)
651 {
652         /* reset connection, out_queue, msg_ and connect_seq */
653         /* discard existing out_queue and msg_seq */
654         dout("reset_connection %p\n", con);
655         ceph_msg_remove_list(&con->out_queue);
656         ceph_msg_remove_list(&con->out_sent);
657
658         if (con->in_msg) {
659                 BUG_ON(con->in_msg->con != con);
660                 ceph_msg_put(con->in_msg);
661                 con->in_msg = NULL;
662         }
663
664         con->connect_seq = 0;
665         con->out_seq = 0;
666         if (con->out_msg) {
667                 BUG_ON(con->out_msg->con != con);
668                 ceph_msg_put(con->out_msg);
669                 con->out_msg = NULL;
670         }
671         con->in_seq = 0;
672         con->in_seq_acked = 0;
673
674         con->out_skip = 0;
675 }
676
677 /*
678  * mark a peer down.  drop any open connections.
679  */
680 void ceph_con_close(struct ceph_connection *con)
681 {
682         mutex_lock(&con->mutex);
683         dout("con_close %p peer %s\n", con,
684              ceph_pr_addr(&con->peer_addr.in_addr));
685         con->state = CON_STATE_CLOSED;
686
687         con_flag_clear(con, CON_FLAG_LOSSYTX);  /* so we retry next connect */
688         con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
689         con_flag_clear(con, CON_FLAG_WRITE_PENDING);
690         con_flag_clear(con, CON_FLAG_BACKOFF);
691
692         reset_connection(con);
693         con->peer_global_seq = 0;
694         cancel_con(con);
695         con_close_socket(con);
696         mutex_unlock(&con->mutex);
697 }
698 EXPORT_SYMBOL(ceph_con_close);
699
700 /*
701  * Reopen a closed connection, with a new peer address.
702  */
703 void ceph_con_open(struct ceph_connection *con,
704                    __u8 entity_type, __u64 entity_num,
705                    struct ceph_entity_addr *addr)
706 {
707         mutex_lock(&con->mutex);
708         dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
709
710         WARN_ON(con->state != CON_STATE_CLOSED);
711         con->state = CON_STATE_PREOPEN;
712
713         con->peer_name.type = (__u8) entity_type;
714         con->peer_name.num = cpu_to_le64(entity_num);
715
716         memcpy(&con->peer_addr, addr, sizeof(*addr));
717         con->delay = 0;      /* reset backoff memory */
718         mutex_unlock(&con->mutex);
719         queue_con(con);
720 }
721 EXPORT_SYMBOL(ceph_con_open);
722
723 /*
724  * return true if this connection ever successfully opened
725  */
726 bool ceph_con_opened(struct ceph_connection *con)
727 {
728         return con->connect_seq > 0;
729 }
730
731 /*
732  * initialize a new connection.
733  */
734 void ceph_con_init(struct ceph_connection *con, void *private,
735         const struct ceph_connection_operations *ops,
736         struct ceph_messenger *msgr)
737 {
738         dout("con_init %p\n", con);
739         memset(con, 0, sizeof(*con));
740         con->private = private;
741         con->ops = ops;
742         con->msgr = msgr;
743
744         con_sock_state_init(con);
745
746         mutex_init(&con->mutex);
747         INIT_LIST_HEAD(&con->out_queue);
748         INIT_LIST_HEAD(&con->out_sent);
749         INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
750
751         con->state = CON_STATE_CLOSED;
752 }
753 EXPORT_SYMBOL(ceph_con_init);
754
755
756 /*
757  * We maintain a global counter to order connection attempts.  Get
758  * a unique seq greater than @gt.
759  */
760 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
761 {
762         u32 ret;
763
764         spin_lock(&msgr->global_seq_lock);
765         if (msgr->global_seq < gt)
766                 msgr->global_seq = gt;
767         ret = ++msgr->global_seq;
768         spin_unlock(&msgr->global_seq_lock);
769         return ret;
770 }
771
772 static void con_out_kvec_reset(struct ceph_connection *con)
773 {
774         BUG_ON(con->out_skip);
775
776         con->out_kvec_left = 0;
777         con->out_kvec_bytes = 0;
778         con->out_kvec_cur = &con->out_kvec[0];
779 }
780
781 static void con_out_kvec_add(struct ceph_connection *con,
782                                 size_t size, void *data)
783 {
784         int index = con->out_kvec_left;
785
786         BUG_ON(con->out_skip);
787         BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
788
789         con->out_kvec[index].iov_len = size;
790         con->out_kvec[index].iov_base = data;
791         con->out_kvec_left++;
792         con->out_kvec_bytes += size;
793 }
794
795 /*
796  * Chop off a kvec from the end.  Return residual number of bytes for
797  * that kvec, i.e. how many bytes would have been written if the kvec
798  * hadn't been nuked.
799  */
800 static int con_out_kvec_skip(struct ceph_connection *con)
801 {
802         int off = con->out_kvec_cur - con->out_kvec;
803         int skip = 0;
804
805         if (con->out_kvec_bytes > 0) {
806                 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
807                 BUG_ON(con->out_kvec_bytes < skip);
808                 BUG_ON(!con->out_kvec_left);
809                 con->out_kvec_bytes -= skip;
810                 con->out_kvec_left--;
811         }
812
813         return skip;
814 }
815
816 #ifdef CONFIG_BLOCK
817
818 /*
819  * For a bio data item, a piece is whatever remains of the next
820  * entry in the current bio iovec, or the first entry in the next
821  * bio in the list.
822  */
823 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
824                                         size_t length)
825 {
826         struct ceph_msg_data *data = cursor->data;
827         struct ceph_bio_iter *it = &cursor->bio_iter;
828
829         cursor->resid = min_t(size_t, length, data->bio_length);
830         *it = data->bio_pos;
831         if (cursor->resid < it->iter.bi_size)
832                 it->iter.bi_size = cursor->resid;
833
834         BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
835         cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
836 }
837
838 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
839                                                 size_t *page_offset,
840                                                 size_t *length)
841 {
842         struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
843                                            cursor->bio_iter.iter);
844
845         *page_offset = bv.bv_offset;
846         *length = bv.bv_len;
847         return bv.bv_page;
848 }
849
850 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
851                                         size_t bytes)
852 {
853         struct ceph_bio_iter *it = &cursor->bio_iter;
854
855         BUG_ON(bytes > cursor->resid);
856         BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
857         cursor->resid -= bytes;
858         bio_advance_iter(it->bio, &it->iter, bytes);
859
860         if (!cursor->resid) {
861                 BUG_ON(!cursor->last_piece);
862                 return false;   /* no more data */
863         }
864
865         if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done))
866                 return false;   /* more bytes to process in this segment */
867
868         if (!it->iter.bi_size) {
869                 it->bio = it->bio->bi_next;
870                 it->iter = it->bio->bi_iter;
871                 if (cursor->resid < it->iter.bi_size)
872                         it->iter.bi_size = cursor->resid;
873         }
874
875         BUG_ON(cursor->last_piece);
876         BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
877         cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
878         return true;
879 }
880 #endif /* CONFIG_BLOCK */
881
882 static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
883                                         size_t length)
884 {
885         struct ceph_msg_data *data = cursor->data;
886         struct bio_vec *bvecs = data->bvec_pos.bvecs;
887
888         cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
889         cursor->bvec_iter = data->bvec_pos.iter;
890         cursor->bvec_iter.bi_size = cursor->resid;
891
892         BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
893         cursor->last_piece =
894             cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
895 }
896
897 static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
898                                                 size_t *page_offset,
899                                                 size_t *length)
900 {
901         struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
902                                            cursor->bvec_iter);
903
904         *page_offset = bv.bv_offset;
905         *length = bv.bv_len;
906         return bv.bv_page;
907 }
908
909 static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
910                                         size_t bytes)
911 {
912         struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
913
914         BUG_ON(bytes > cursor->resid);
915         BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
916         cursor->resid -= bytes;
917         bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
918
919         if (!cursor->resid) {
920                 BUG_ON(!cursor->last_piece);
921                 return false;   /* no more data */
922         }
923
924         if (!bytes || cursor->bvec_iter.bi_bvec_done)
925                 return false;   /* more bytes to process in this segment */
926
927         BUG_ON(cursor->last_piece);
928         BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
929         cursor->last_piece =
930             cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
931         return true;
932 }
933
934 /*
935  * For a page array, a piece comes from the first page in the array
936  * that has not already been fully consumed.
937  */
938 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
939                                         size_t length)
940 {
941         struct ceph_msg_data *data = cursor->data;
942         int page_count;
943
944         BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
945
946         BUG_ON(!data->pages);
947         BUG_ON(!data->length);
948
949         cursor->resid = min(length, data->length);
950         page_count = calc_pages_for(data->alignment, (u64)data->length);
951         cursor->page_offset = data->alignment & ~PAGE_MASK;
952         cursor->page_index = 0;
953         BUG_ON(page_count > (int)USHRT_MAX);
954         cursor->page_count = (unsigned short)page_count;
955         BUG_ON(length > SIZE_MAX - cursor->page_offset);
956         cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
957 }
958
959 static struct page *
960 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
961                                         size_t *page_offset, size_t *length)
962 {
963         struct ceph_msg_data *data = cursor->data;
964
965         BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
966
967         BUG_ON(cursor->page_index >= cursor->page_count);
968         BUG_ON(cursor->page_offset >= PAGE_SIZE);
969
970         *page_offset = cursor->page_offset;
971         if (cursor->last_piece)
972                 *length = cursor->resid;
973         else
974                 *length = PAGE_SIZE - *page_offset;
975
976         return data->pages[cursor->page_index];
977 }
978
979 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
980                                                 size_t bytes)
981 {
982         BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
983
984         BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
985
986         /* Advance the cursor page offset */
987
988         cursor->resid -= bytes;
989         cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
990         if (!bytes || cursor->page_offset)
991                 return false;   /* more bytes to process in the current page */
992
993         if (!cursor->resid)
994                 return false;   /* no more data */
995
996         /* Move on to the next page; offset is already at 0 */
997
998         BUG_ON(cursor->page_index >= cursor->page_count);
999         cursor->page_index++;
1000         cursor->last_piece = cursor->resid <= PAGE_SIZE;
1001
1002         return true;
1003 }
1004
1005 /*
1006  * For a pagelist, a piece is whatever remains to be consumed in the
1007  * first page in the list, or the front of the next page.
1008  */
1009 static void
1010 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
1011                                         size_t length)
1012 {
1013         struct ceph_msg_data *data = cursor->data;
1014         struct ceph_pagelist *pagelist;
1015         struct page *page;
1016
1017         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1018
1019         pagelist = data->pagelist;
1020         BUG_ON(!pagelist);
1021
1022         if (!length)
1023                 return;         /* pagelist can be assigned but empty */
1024
1025         BUG_ON(list_empty(&pagelist->head));
1026         page = list_first_entry(&pagelist->head, struct page, lru);
1027
1028         cursor->resid = min(length, pagelist->length);
1029         cursor->page = page;
1030         cursor->offset = 0;
1031         cursor->last_piece = cursor->resid <= PAGE_SIZE;
1032 }
1033
1034 static struct page *
1035 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1036                                 size_t *page_offset, size_t *length)
1037 {
1038         struct ceph_msg_data *data = cursor->data;
1039         struct ceph_pagelist *pagelist;
1040
1041         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1042
1043         pagelist = data->pagelist;
1044         BUG_ON(!pagelist);
1045
1046         BUG_ON(!cursor->page);
1047         BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1048
1049         /* offset of first page in pagelist is always 0 */
1050         *page_offset = cursor->offset & ~PAGE_MASK;
1051         if (cursor->last_piece)
1052                 *length = cursor->resid;
1053         else
1054                 *length = PAGE_SIZE - *page_offset;
1055
1056         return cursor->page;
1057 }
1058
1059 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1060                                                 size_t bytes)
1061 {
1062         struct ceph_msg_data *data = cursor->data;
1063         struct ceph_pagelist *pagelist;
1064
1065         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1066
1067         pagelist = data->pagelist;
1068         BUG_ON(!pagelist);
1069
1070         BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1071         BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1072
1073         /* Advance the cursor offset */
1074
1075         cursor->resid -= bytes;
1076         cursor->offset += bytes;
1077         /* offset of first page in pagelist is always 0 */
1078         if (!bytes || cursor->offset & ~PAGE_MASK)
1079                 return false;   /* more bytes to process in the current page */
1080
1081         if (!cursor->resid)
1082                 return false;   /* no more data */
1083
1084         /* Move on to the next page */
1085
1086         BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1087         cursor->page = list_next_entry(cursor->page, lru);
1088         cursor->last_piece = cursor->resid <= PAGE_SIZE;
1089
1090         return true;
1091 }
1092
1093 /*
1094  * Message data is handled (sent or received) in pieces, where each
1095  * piece resides on a single page.  The network layer might not
1096  * consume an entire piece at once.  A data item's cursor keeps
1097  * track of which piece is next to process and how much remains to
1098  * be processed in that piece.  It also tracks whether the current
1099  * piece is the last one in the data item.
1100  */
1101 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1102 {
1103         size_t length = cursor->total_resid;
1104
1105         switch (cursor->data->type) {
1106         case CEPH_MSG_DATA_PAGELIST:
1107                 ceph_msg_data_pagelist_cursor_init(cursor, length);
1108                 break;
1109         case CEPH_MSG_DATA_PAGES:
1110                 ceph_msg_data_pages_cursor_init(cursor, length);
1111                 break;
1112 #ifdef CONFIG_BLOCK
1113         case CEPH_MSG_DATA_BIO:
1114                 ceph_msg_data_bio_cursor_init(cursor, length);
1115                 break;
1116 #endif /* CONFIG_BLOCK */
1117         case CEPH_MSG_DATA_BVECS:
1118                 ceph_msg_data_bvecs_cursor_init(cursor, length);
1119                 break;
1120         case CEPH_MSG_DATA_NONE:
1121         default:
1122                 /* BUG(); */
1123                 break;
1124         }
1125         cursor->need_crc = true;
1126 }
1127
1128 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1129 {
1130         struct ceph_msg_data_cursor *cursor = &msg->cursor;
1131
1132         BUG_ON(!length);
1133         BUG_ON(length > msg->data_length);
1134         BUG_ON(!msg->num_data_items);
1135
1136         cursor->total_resid = length;
1137         cursor->data = msg->data;
1138
1139         __ceph_msg_data_cursor_init(cursor);
1140 }
1141
1142 /*
1143  * Return the page containing the next piece to process for a given
1144  * data item, and supply the page offset and length of that piece.
1145  * Indicate whether this is the last piece in this data item.
1146  */
1147 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1148                                         size_t *page_offset, size_t *length,
1149                                         bool *last_piece)
1150 {
1151         struct page *page;
1152
1153         switch (cursor->data->type) {
1154         case CEPH_MSG_DATA_PAGELIST:
1155                 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1156                 break;
1157         case CEPH_MSG_DATA_PAGES:
1158                 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1159                 break;
1160 #ifdef CONFIG_BLOCK
1161         case CEPH_MSG_DATA_BIO:
1162                 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1163                 break;
1164 #endif /* CONFIG_BLOCK */
1165         case CEPH_MSG_DATA_BVECS:
1166                 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1167                 break;
1168         case CEPH_MSG_DATA_NONE:
1169         default:
1170                 page = NULL;
1171                 break;
1172         }
1173
1174         BUG_ON(!page);
1175         BUG_ON(*page_offset + *length > PAGE_SIZE);
1176         BUG_ON(!*length);
1177         BUG_ON(*length > cursor->resid);
1178         if (last_piece)
1179                 *last_piece = cursor->last_piece;
1180
1181         return page;
1182 }
1183
1184 /*
1185  * Returns true if the result moves the cursor on to the next piece
1186  * of the data item.
1187  */
1188 static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1189                                   size_t bytes)
1190 {
1191         bool new_piece;
1192
1193         BUG_ON(bytes > cursor->resid);
1194         switch (cursor->data->type) {
1195         case CEPH_MSG_DATA_PAGELIST:
1196                 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1197                 break;
1198         case CEPH_MSG_DATA_PAGES:
1199                 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1200                 break;
1201 #ifdef CONFIG_BLOCK
1202         case CEPH_MSG_DATA_BIO:
1203                 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1204                 break;
1205 #endif /* CONFIG_BLOCK */
1206         case CEPH_MSG_DATA_BVECS:
1207                 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1208                 break;
1209         case CEPH_MSG_DATA_NONE:
1210         default:
1211                 BUG();
1212                 break;
1213         }
1214         cursor->total_resid -= bytes;
1215
1216         if (!cursor->resid && cursor->total_resid) {
1217                 WARN_ON(!cursor->last_piece);
1218                 cursor->data++;
1219                 __ceph_msg_data_cursor_init(cursor);
1220                 new_piece = true;
1221         }
1222         cursor->need_crc = new_piece;
1223 }
1224
1225 static size_t sizeof_footer(struct ceph_connection *con)
1226 {
1227         return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1228             sizeof(struct ceph_msg_footer) :
1229             sizeof(struct ceph_msg_footer_old);
1230 }
1231
1232 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1233 {
1234         /* Initialize data cursor */
1235
1236         ceph_msg_data_cursor_init(msg, (size_t)data_len);
1237 }
1238
1239 /*
1240  * Prepare footer for currently outgoing message, and finish things
1241  * off.  Assumes out_kvec* are already valid.. we just add on to the end.
1242  */
1243 static void prepare_write_message_footer(struct ceph_connection *con)
1244 {
1245         struct ceph_msg *m = con->out_msg;
1246
1247         m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1248
1249         dout("prepare_write_message_footer %p\n", con);
1250         con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1251         if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1252                 if (con->ops->sign_message)
1253                         con->ops->sign_message(m);
1254                 else
1255                         m->footer.sig = 0;
1256         } else {
1257                 m->old_footer.flags = m->footer.flags;
1258         }
1259         con->out_more = m->more_to_follow;
1260         con->out_msg_done = true;
1261 }
1262
1263 /*
1264  * Prepare headers for the next outgoing message.
1265  */
1266 static void prepare_write_message(struct ceph_connection *con)
1267 {
1268         struct ceph_msg *m;
1269         u32 crc;
1270
1271         con_out_kvec_reset(con);
1272         con->out_msg_done = false;
1273
1274         /* Sneak an ack in there first?  If we can get it into the same
1275          * TCP packet that's a good thing. */
1276         if (con->in_seq > con->in_seq_acked) {
1277                 con->in_seq_acked = con->in_seq;
1278                 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1279                 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1280                 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1281                         &con->out_temp_ack);
1282         }
1283
1284         BUG_ON(list_empty(&con->out_queue));
1285         m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1286         con->out_msg = m;
1287         BUG_ON(m->con != con);
1288
1289         /* put message on sent list */
1290         ceph_msg_get(m);
1291         list_move_tail(&m->list_head, &con->out_sent);
1292
1293         /*
1294          * only assign outgoing seq # if we haven't sent this message
1295          * yet.  if it is requeued, resend with it's original seq.
1296          */
1297         if (m->needs_out_seq) {
1298                 m->hdr.seq = cpu_to_le64(++con->out_seq);
1299                 m->needs_out_seq = false;
1300
1301                 if (con->ops->reencode_message)
1302                         con->ops->reencode_message(m);
1303         }
1304
1305         dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1306              m, con->out_seq, le16_to_cpu(m->hdr.type),
1307              le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1308              m->data_length);
1309         WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1310         WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1311
1312         /* tag + hdr + front + middle */
1313         con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1314         con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1315         con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1316
1317         if (m->middle)
1318                 con_out_kvec_add(con, m->middle->vec.iov_len,
1319                         m->middle->vec.iov_base);
1320
1321         /* fill in hdr crc and finalize hdr */
1322         crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1323         con->out_msg->hdr.crc = cpu_to_le32(crc);
1324         memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1325
1326         /* fill in front and middle crc, footer */
1327         crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1328         con->out_msg->footer.front_crc = cpu_to_le32(crc);
1329         if (m->middle) {
1330                 crc = crc32c(0, m->middle->vec.iov_base,
1331                                 m->middle->vec.iov_len);
1332                 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1333         } else
1334                 con->out_msg->footer.middle_crc = 0;
1335         dout("%s front_crc %u middle_crc %u\n", __func__,
1336              le32_to_cpu(con->out_msg->footer.front_crc),
1337              le32_to_cpu(con->out_msg->footer.middle_crc));
1338         con->out_msg->footer.flags = 0;
1339
1340         /* is there a data payload? */
1341         con->out_msg->footer.data_crc = 0;
1342         if (m->data_length) {
1343                 prepare_message_data(con->out_msg, m->data_length);
1344                 con->out_more = 1;  /* data + footer will follow */
1345         } else {
1346                 /* no, queue up footer too and be done */
1347                 prepare_write_message_footer(con);
1348         }
1349
1350         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1351 }
1352
1353 /*
1354  * Prepare an ack.
1355  */
1356 static void prepare_write_ack(struct ceph_connection *con)
1357 {
1358         dout("prepare_write_ack %p %llu -> %llu\n", con,
1359              con->in_seq_acked, con->in_seq);
1360         con->in_seq_acked = con->in_seq;
1361
1362         con_out_kvec_reset(con);
1363
1364         con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1365
1366         con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1367         con_out_kvec_add(con, sizeof (con->out_temp_ack),
1368                                 &con->out_temp_ack);
1369
1370         con->out_more = 1;  /* more will follow.. eventually.. */
1371         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1372 }
1373
1374 /*
1375  * Prepare to share the seq during handshake
1376  */
1377 static void prepare_write_seq(struct ceph_connection *con)
1378 {
1379         dout("prepare_write_seq %p %llu -> %llu\n", con,
1380              con->in_seq_acked, con->in_seq);
1381         con->in_seq_acked = con->in_seq;
1382
1383         con_out_kvec_reset(con);
1384
1385         con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1386         con_out_kvec_add(con, sizeof (con->out_temp_ack),
1387                          &con->out_temp_ack);
1388
1389         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1390 }
1391
1392 /*
1393  * Prepare to write keepalive byte.
1394  */
1395 static void prepare_write_keepalive(struct ceph_connection *con)
1396 {
1397         dout("prepare_write_keepalive %p\n", con);
1398         con_out_kvec_reset(con);
1399         if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1400                 struct timespec64 now;
1401
1402                 ktime_get_real_ts64(&now);
1403                 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1404                 ceph_encode_timespec64(&con->out_temp_keepalive2, &now);
1405                 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1406                                  &con->out_temp_keepalive2);
1407         } else {
1408                 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1409         }
1410         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1411 }
1412
1413 /*
1414  * Connection negotiation.
1415  */
1416
1417 static int get_connect_authorizer(struct ceph_connection *con)
1418 {
1419         struct ceph_auth_handshake *auth;
1420         int auth_proto;
1421
1422         if (!con->ops->get_authorizer) {
1423                 con->auth = NULL;
1424                 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1425                 con->out_connect.authorizer_len = 0;
1426                 return 0;
1427         }
1428
1429         auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
1430         if (IS_ERR(auth))
1431                 return PTR_ERR(auth);
1432
1433         con->auth = auth;
1434         con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1435         con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1436         return 0;
1437 }
1438
1439 /*
1440  * We connected to a peer and are saying hello.
1441  */
1442 static void prepare_write_banner(struct ceph_connection *con)
1443 {
1444         con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1445         con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1446                                         &con->msgr->my_enc_addr);
1447
1448         con->out_more = 0;
1449         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1450 }
1451
1452 static void __prepare_write_connect(struct ceph_connection *con)
1453 {
1454         con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1455         if (con->auth)
1456                 con_out_kvec_add(con, con->auth->authorizer_buf_len,
1457                                  con->auth->authorizer_buf);
1458
1459         con->out_more = 0;
1460         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1461 }
1462
1463 static int prepare_write_connect(struct ceph_connection *con)
1464 {
1465         unsigned int global_seq = get_global_seq(con->msgr, 0);
1466         int proto;
1467         int ret;
1468
1469         switch (con->peer_name.type) {
1470         case CEPH_ENTITY_TYPE_MON:
1471                 proto = CEPH_MONC_PROTOCOL;
1472                 break;
1473         case CEPH_ENTITY_TYPE_OSD:
1474                 proto = CEPH_OSDC_PROTOCOL;
1475                 break;
1476         case CEPH_ENTITY_TYPE_MDS:
1477                 proto = CEPH_MDSC_PROTOCOL;
1478                 break;
1479         default:
1480                 BUG();
1481         }
1482
1483         dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1484              con->connect_seq, global_seq, proto);
1485
1486         con->out_connect.features =
1487             cpu_to_le64(from_msgr(con->msgr)->supported_features);
1488         con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1489         con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1490         con->out_connect.global_seq = cpu_to_le32(global_seq);
1491         con->out_connect.protocol_version = cpu_to_le32(proto);
1492         con->out_connect.flags = 0;
1493
1494         ret = get_connect_authorizer(con);
1495         if (ret)
1496                 return ret;
1497
1498         __prepare_write_connect(con);
1499         return 0;
1500 }
1501
1502 /*
1503  * write as much of pending kvecs to the socket as we can.
1504  *  1 -> done
1505  *  0 -> socket full, but more to do
1506  * <0 -> error
1507  */
1508 static int write_partial_kvec(struct ceph_connection *con)
1509 {
1510         int ret;
1511
1512         dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1513         while (con->out_kvec_bytes > 0) {
1514                 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1515                                        con->out_kvec_left, con->out_kvec_bytes,
1516                                        con->out_more);
1517                 if (ret <= 0)
1518                         goto out;
1519                 con->out_kvec_bytes -= ret;
1520                 if (con->out_kvec_bytes == 0)
1521                         break;            /* done */
1522
1523                 /* account for full iov entries consumed */
1524                 while (ret >= con->out_kvec_cur->iov_len) {
1525                         BUG_ON(!con->out_kvec_left);
1526                         ret -= con->out_kvec_cur->iov_len;
1527                         con->out_kvec_cur++;
1528                         con->out_kvec_left--;
1529                 }
1530                 /* and for a partially-consumed entry */
1531                 if (ret) {
1532                         con->out_kvec_cur->iov_len -= ret;
1533                         con->out_kvec_cur->iov_base += ret;
1534                 }
1535         }
1536         con->out_kvec_left = 0;
1537         ret = 1;
1538 out:
1539         dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1540              con->out_kvec_bytes, con->out_kvec_left, ret);
1541         return ret;  /* done! */
1542 }
1543
1544 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1545                                 unsigned int page_offset,
1546                                 unsigned int length)
1547 {
1548         char *kaddr;
1549
1550         kaddr = kmap(page);
1551         BUG_ON(kaddr == NULL);
1552         crc = crc32c(crc, kaddr + page_offset, length);
1553         kunmap(page);
1554
1555         return crc;
1556 }
1557 /*
1558  * Write as much message data payload as we can.  If we finish, queue
1559  * up the footer.
1560  *  1 -> done, footer is now queued in out_kvec[].
1561  *  0 -> socket full, but more to do
1562  * <0 -> error
1563  */
1564 static int write_partial_message_data(struct ceph_connection *con)
1565 {
1566         struct ceph_msg *msg = con->out_msg;
1567         struct ceph_msg_data_cursor *cursor = &msg->cursor;
1568         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1569         u32 crc;
1570
1571         dout("%s %p msg %p\n", __func__, con, msg);
1572
1573         if (!msg->num_data_items)
1574                 return -EINVAL;
1575
1576         /*
1577          * Iterate through each page that contains data to be
1578          * written, and send as much as possible for each.
1579          *
1580          * If we are calculating the data crc (the default), we will
1581          * need to map the page.  If we have no pages, they have
1582          * been revoked, so use the zero page.
1583          */
1584         crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1585         while (cursor->total_resid) {
1586                 struct page *page;
1587                 size_t page_offset;
1588                 size_t length;
1589                 bool last_piece;
1590                 int ret;
1591
1592                 if (!cursor->resid) {
1593                         ceph_msg_data_advance(cursor, 0);
1594                         continue;
1595                 }
1596
1597                 page = ceph_msg_data_next(cursor, &page_offset, &length,
1598                                           &last_piece);
1599                 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1600                                         length, !last_piece);
1601                 if (ret <= 0) {
1602                         if (do_datacrc)
1603                                 msg->footer.data_crc = cpu_to_le32(crc);
1604
1605                         return ret;
1606                 }
1607                 if (do_datacrc && cursor->need_crc)
1608                         crc = ceph_crc32c_page(crc, page, page_offset, length);
1609                 ceph_msg_data_advance(cursor, (size_t)ret);
1610         }
1611
1612         dout("%s %p msg %p done\n", __func__, con, msg);
1613
1614         /* prepare and queue up footer, too */
1615         if (do_datacrc)
1616                 msg->footer.data_crc = cpu_to_le32(crc);
1617         else
1618                 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1619         con_out_kvec_reset(con);
1620         prepare_write_message_footer(con);
1621
1622         return 1;       /* must return > 0 to indicate success */
1623 }
1624
1625 /*
1626  * write some zeros
1627  */
1628 static int write_partial_skip(struct ceph_connection *con)
1629 {
1630         int ret;
1631
1632         dout("%s %p %d left\n", __func__, con, con->out_skip);
1633         while (con->out_skip > 0) {
1634                 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1635
1636                 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1637                 if (ret <= 0)
1638                         goto out;
1639                 con->out_skip -= ret;
1640         }
1641         ret = 1;
1642 out:
1643         return ret;
1644 }
1645
1646 /*
1647  * Prepare to read connection handshake, or an ack.
1648  */
1649 static void prepare_read_banner(struct ceph_connection *con)
1650 {
1651         dout("prepare_read_banner %p\n", con);
1652         con->in_base_pos = 0;
1653 }
1654
1655 static void prepare_read_connect(struct ceph_connection *con)
1656 {
1657         dout("prepare_read_connect %p\n", con);
1658         con->in_base_pos = 0;
1659 }
1660
1661 static void prepare_read_ack(struct ceph_connection *con)
1662 {
1663         dout("prepare_read_ack %p\n", con);
1664         con->in_base_pos = 0;
1665 }
1666
1667 static void prepare_read_seq(struct ceph_connection *con)
1668 {
1669         dout("prepare_read_seq %p\n", con);
1670         con->in_base_pos = 0;
1671         con->in_tag = CEPH_MSGR_TAG_SEQ;
1672 }
1673
1674 static void prepare_read_tag(struct ceph_connection *con)
1675 {
1676         dout("prepare_read_tag %p\n", con);
1677         con->in_base_pos = 0;
1678         con->in_tag = CEPH_MSGR_TAG_READY;
1679 }
1680
1681 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1682 {
1683         dout("prepare_read_keepalive_ack %p\n", con);
1684         con->in_base_pos = 0;
1685 }
1686
1687 /*
1688  * Prepare to read a message.
1689  */
1690 static int prepare_read_message(struct ceph_connection *con)
1691 {
1692         dout("prepare_read_message %p\n", con);
1693         BUG_ON(con->in_msg != NULL);
1694         con->in_base_pos = 0;
1695         con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1696         return 0;
1697 }
1698
1699
1700 static int read_partial(struct ceph_connection *con,
1701                         int end, int size, void *object)
1702 {
1703         while (con->in_base_pos < end) {
1704                 int left = end - con->in_base_pos;
1705                 int have = size - left;
1706                 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1707                 if (ret <= 0)
1708                         return ret;
1709                 con->in_base_pos += ret;
1710         }
1711         return 1;
1712 }
1713
1714
1715 /*
1716  * Read all or part of the connect-side handshake on a new connection
1717  */
1718 static int read_partial_banner(struct ceph_connection *con)
1719 {
1720         int size;
1721         int end;
1722         int ret;
1723
1724         dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1725
1726         /* peer's banner */
1727         size = strlen(CEPH_BANNER);
1728         end = size;
1729         ret = read_partial(con, end, size, con->in_banner);
1730         if (ret <= 0)
1731                 goto out;
1732
1733         size = sizeof (con->actual_peer_addr);
1734         end += size;
1735         ret = read_partial(con, end, size, &con->actual_peer_addr);
1736         if (ret <= 0)
1737                 goto out;
1738
1739         size = sizeof (con->peer_addr_for_me);
1740         end += size;
1741         ret = read_partial(con, end, size, &con->peer_addr_for_me);
1742         if (ret <= 0)
1743                 goto out;
1744
1745 out:
1746         return ret;
1747 }
1748
1749 static int read_partial_connect(struct ceph_connection *con)
1750 {
1751         int size;
1752         int end;
1753         int ret;
1754
1755         dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1756
1757         size = sizeof (con->in_reply);
1758         end = size;
1759         ret = read_partial(con, end, size, &con->in_reply);
1760         if (ret <= 0)
1761                 goto out;
1762
1763         if (con->auth) {
1764                 size = le32_to_cpu(con->in_reply.authorizer_len);
1765                 if (size > con->auth->authorizer_reply_buf_len) {
1766                         pr_err("authorizer reply too big: %d > %zu\n", size,
1767                                con->auth->authorizer_reply_buf_len);
1768                         ret = -EINVAL;
1769                         goto out;
1770                 }
1771
1772                 end += size;
1773                 ret = read_partial(con, end, size,
1774                                    con->auth->authorizer_reply_buf);
1775                 if (ret <= 0)
1776                         goto out;
1777         }
1778
1779         dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1780              con, (int)con->in_reply.tag,
1781              le32_to_cpu(con->in_reply.connect_seq),
1782              le32_to_cpu(con->in_reply.global_seq));
1783 out:
1784         return ret;
1785 }
1786
1787 /*
1788  * Verify the hello banner looks okay.
1789  */
1790 static int verify_hello(struct ceph_connection *con)
1791 {
1792         if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1793                 pr_err("connect to %s got bad banner\n",
1794                        ceph_pr_addr(&con->peer_addr.in_addr));
1795                 con->error_msg = "protocol error, bad banner";
1796                 return -1;
1797         }
1798         return 0;
1799 }
1800
1801 static bool addr_is_blank(struct sockaddr_storage *ss)
1802 {
1803         struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1804         struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1805
1806         switch (ss->ss_family) {
1807         case AF_INET:
1808                 return addr->s_addr == htonl(INADDR_ANY);
1809         case AF_INET6:
1810                 return ipv6_addr_any(addr6);
1811         default:
1812                 return true;
1813         }
1814 }
1815
1816 static int addr_port(struct sockaddr_storage *ss)
1817 {
1818         switch (ss->ss_family) {
1819         case AF_INET:
1820                 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1821         case AF_INET6:
1822                 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1823         }
1824         return 0;
1825 }
1826
1827 static void addr_set_port(struct sockaddr_storage *ss, int p)
1828 {
1829         switch (ss->ss_family) {
1830         case AF_INET:
1831                 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1832                 break;
1833         case AF_INET6:
1834                 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1835                 break;
1836         }
1837 }
1838
1839 /*
1840  * Unlike other *_pton function semantics, zero indicates success.
1841  */
1842 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1843                 char delim, const char **ipend)
1844 {
1845         struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1846         struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1847
1848         memset(ss, 0, sizeof(*ss));
1849
1850         if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1851                 ss->ss_family = AF_INET;
1852                 return 0;
1853         }
1854
1855         if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1856                 ss->ss_family = AF_INET6;
1857                 return 0;
1858         }
1859
1860         return -EINVAL;
1861 }
1862
1863 /*
1864  * Extract hostname string and resolve using kernel DNS facility.
1865  */
1866 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1867 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1868                 struct sockaddr_storage *ss, char delim, const char **ipend)
1869 {
1870         const char *end, *delim_p;
1871         char *colon_p, *ip_addr = NULL;
1872         int ip_len, ret;
1873
1874         /*
1875          * The end of the hostname occurs immediately preceding the delimiter or
1876          * the port marker (':') where the delimiter takes precedence.
1877          */
1878         delim_p = memchr(name, delim, namelen);
1879         colon_p = memchr(name, ':', namelen);
1880
1881         if (delim_p && colon_p)
1882                 end = delim_p < colon_p ? delim_p : colon_p;
1883         else if (!delim_p && colon_p)
1884                 end = colon_p;
1885         else {
1886                 end = delim_p;
1887                 if (!end) /* case: hostname:/ */
1888                         end = name + namelen;
1889         }
1890
1891         if (end <= name)
1892                 return -EINVAL;
1893
1894         /* do dns_resolve upcall */
1895         ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1896         if (ip_len > 0)
1897                 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1898         else
1899                 ret = -ESRCH;
1900
1901         kfree(ip_addr);
1902
1903         *ipend = end;
1904
1905         pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1906                         ret, ret ? "failed" : ceph_pr_addr(ss));
1907
1908         return ret;
1909 }
1910 #else
1911 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1912                 struct sockaddr_storage *ss, char delim, const char **ipend)
1913 {
1914         return -EINVAL;
1915 }
1916 #endif
1917
1918 /*
1919  * Parse a server name (IP or hostname). If a valid IP address is not found
1920  * then try to extract a hostname to resolve using userspace DNS upcall.
1921  */
1922 static int ceph_parse_server_name(const char *name, size_t namelen,
1923                         struct sockaddr_storage *ss, char delim, const char **ipend)
1924 {
1925         int ret;
1926
1927         ret = ceph_pton(name, namelen, ss, delim, ipend);
1928         if (ret)
1929                 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1930
1931         return ret;
1932 }
1933
1934 /*
1935  * Parse an ip[:port] list into an addr array.  Use the default
1936  * monitor port if a port isn't specified.
1937  */
1938 int ceph_parse_ips(const char *c, const char *end,
1939                    struct ceph_entity_addr *addr,
1940                    int max_count, int *count)
1941 {
1942         int i, ret = -EINVAL;
1943         const char *p = c;
1944
1945         dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1946         for (i = 0; i < max_count; i++) {
1947                 const char *ipend;
1948                 struct sockaddr_storage *ss = &addr[i].in_addr;
1949                 int port;
1950                 char delim = ',';
1951
1952                 if (*p == '[') {
1953                         delim = ']';
1954                         p++;
1955                 }
1956
1957                 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1958                 if (ret)
1959                         goto bad;
1960                 ret = -EINVAL;
1961
1962                 p = ipend;
1963
1964                 if (delim == ']') {
1965                         if (*p != ']') {
1966                                 dout("missing matching ']'\n");
1967                                 goto bad;
1968                         }
1969                         p++;
1970                 }
1971
1972                 /* port? */
1973                 if (p < end && *p == ':') {
1974                         port = 0;
1975                         p++;
1976                         while (p < end && *p >= '0' && *p <= '9') {
1977                                 port = (port * 10) + (*p - '0');
1978                                 p++;
1979                         }
1980                         if (port == 0)
1981                                 port = CEPH_MON_PORT;
1982                         else if (port > 65535)
1983                                 goto bad;
1984                 } else {
1985                         port = CEPH_MON_PORT;
1986                 }
1987
1988                 addr_set_port(ss, port);
1989
1990                 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1991
1992                 if (p == end)
1993                         break;
1994                 if (*p != ',')
1995                         goto bad;
1996                 p++;
1997         }
1998
1999         if (p != end)
2000                 goto bad;
2001
2002         if (count)
2003                 *count = i + 1;
2004         return 0;
2005
2006 bad:
2007         pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
2008         return ret;
2009 }
2010 EXPORT_SYMBOL(ceph_parse_ips);
2011
2012 static int process_banner(struct ceph_connection *con)
2013 {
2014         dout("process_banner on %p\n", con);
2015
2016         if (verify_hello(con) < 0)
2017                 return -1;
2018
2019         ceph_decode_addr(&con->actual_peer_addr);
2020         ceph_decode_addr(&con->peer_addr_for_me);
2021
2022         /*
2023          * Make sure the other end is who we wanted.  note that the other
2024          * end may not yet know their ip address, so if it's 0.0.0.0, give
2025          * them the benefit of the doubt.
2026          */
2027         if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2028                    sizeof(con->peer_addr)) != 0 &&
2029             !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2030               con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2031                 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2032                         ceph_pr_addr(&con->peer_addr.in_addr),
2033                         (int)le32_to_cpu(con->peer_addr.nonce),
2034                         ceph_pr_addr(&con->actual_peer_addr.in_addr),
2035                         (int)le32_to_cpu(con->actual_peer_addr.nonce));
2036                 con->error_msg = "wrong peer at address";
2037                 return -1;
2038         }
2039
2040         /*
2041          * did we learn our address?
2042          */
2043         if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2044                 int port = addr_port(&con->msgr->inst.addr.in_addr);
2045
2046                 memcpy(&con->msgr->inst.addr.in_addr,
2047                        &con->peer_addr_for_me.in_addr,
2048                        sizeof(con->peer_addr_for_me.in_addr));
2049                 addr_set_port(&con->msgr->inst.addr.in_addr, port);
2050                 encode_my_addr(con->msgr);
2051                 dout("process_banner learned my addr is %s\n",
2052                      ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2053         }
2054
2055         return 0;
2056 }
2057
2058 static int process_connect(struct ceph_connection *con)
2059 {
2060         u64 sup_feat = from_msgr(con->msgr)->supported_features;
2061         u64 req_feat = from_msgr(con->msgr)->required_features;
2062         u64 server_feat = le64_to_cpu(con->in_reply.features);
2063         int ret;
2064
2065         dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2066
2067         if (con->auth) {
2068                 /*
2069                  * Any connection that defines ->get_authorizer()
2070                  * should also define ->add_authorizer_challenge() and
2071                  * ->verify_authorizer_reply().
2072                  *
2073                  * See get_connect_authorizer().
2074                  */
2075                 if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
2076                         ret = con->ops->add_authorizer_challenge(
2077                                     con, con->auth->authorizer_reply_buf,
2078                                     le32_to_cpu(con->in_reply.authorizer_len));
2079                         if (ret < 0)
2080                                 return ret;
2081
2082                         con_out_kvec_reset(con);
2083                         __prepare_write_connect(con);
2084                         prepare_read_connect(con);
2085                         return 0;
2086                 }
2087
2088                 ret = con->ops->verify_authorizer_reply(con);
2089                 if (ret < 0) {
2090                         con->error_msg = "bad authorize reply";
2091                         return ret;
2092                 }
2093         }
2094
2095         switch (con->in_reply.tag) {
2096         case CEPH_MSGR_TAG_FEATURES:
2097                 pr_err("%s%lld %s feature set mismatch,"
2098                        " my %llx < server's %llx, missing %llx\n",
2099                        ENTITY_NAME(con->peer_name),
2100                        ceph_pr_addr(&con->peer_addr.in_addr),
2101                        sup_feat, server_feat, server_feat & ~sup_feat);
2102                 con->error_msg = "missing required protocol features";
2103                 reset_connection(con);
2104                 return -1;
2105
2106         case CEPH_MSGR_TAG_BADPROTOVER:
2107                 pr_err("%s%lld %s protocol version mismatch,"
2108                        " my %d != server's %d\n",
2109                        ENTITY_NAME(con->peer_name),
2110                        ceph_pr_addr(&con->peer_addr.in_addr),
2111                        le32_to_cpu(con->out_connect.protocol_version),
2112                        le32_to_cpu(con->in_reply.protocol_version));
2113                 con->error_msg = "protocol version mismatch";
2114                 reset_connection(con);
2115                 return -1;
2116
2117         case CEPH_MSGR_TAG_BADAUTHORIZER:
2118                 con->auth_retry++;
2119                 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2120                      con->auth_retry);
2121                 if (con->auth_retry == 2) {
2122                         con->error_msg = "connect authorization failure";
2123                         return -1;
2124                 }
2125                 con_out_kvec_reset(con);
2126                 ret = prepare_write_connect(con);
2127                 if (ret < 0)
2128                         return ret;
2129                 prepare_read_connect(con);
2130                 break;
2131
2132         case CEPH_MSGR_TAG_RESETSESSION:
2133                 /*
2134                  * If we connected with a large connect_seq but the peer
2135                  * has no record of a session with us (no connection, or
2136                  * connect_seq == 0), they will send RESETSESION to indicate
2137                  * that they must have reset their session, and may have
2138                  * dropped messages.
2139                  */
2140                 dout("process_connect got RESET peer seq %u\n",
2141                      le32_to_cpu(con->in_reply.connect_seq));
2142                 pr_err("%s%lld %s connection reset\n",
2143                        ENTITY_NAME(con->peer_name),
2144                        ceph_pr_addr(&con->peer_addr.in_addr));
2145                 reset_connection(con);
2146                 con_out_kvec_reset(con);
2147                 ret = prepare_write_connect(con);
2148                 if (ret < 0)
2149                         return ret;
2150                 prepare_read_connect(con);
2151
2152                 /* Tell ceph about it. */
2153                 mutex_unlock(&con->mutex);
2154                 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2155                 if (con->ops->peer_reset)
2156                         con->ops->peer_reset(con);
2157                 mutex_lock(&con->mutex);
2158                 if (con->state != CON_STATE_NEGOTIATING)
2159                         return -EAGAIN;
2160                 break;
2161
2162         case CEPH_MSGR_TAG_RETRY_SESSION:
2163                 /*
2164                  * If we sent a smaller connect_seq than the peer has, try
2165                  * again with a larger value.
2166                  */
2167                 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2168                      le32_to_cpu(con->out_connect.connect_seq),
2169                      le32_to_cpu(con->in_reply.connect_seq));
2170                 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2171                 con_out_kvec_reset(con);
2172                 ret = prepare_write_connect(con);
2173                 if (ret < 0)
2174                         return ret;
2175                 prepare_read_connect(con);
2176                 break;
2177
2178         case CEPH_MSGR_TAG_RETRY_GLOBAL:
2179                 /*
2180                  * If we sent a smaller global_seq than the peer has, try
2181                  * again with a larger value.
2182                  */
2183                 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2184                      con->peer_global_seq,
2185                      le32_to_cpu(con->in_reply.global_seq));
2186                 get_global_seq(con->msgr,
2187                                le32_to_cpu(con->in_reply.global_seq));
2188                 con_out_kvec_reset(con);
2189                 ret = prepare_write_connect(con);
2190                 if (ret < 0)
2191                         return ret;
2192                 prepare_read_connect(con);
2193                 break;
2194
2195         case CEPH_MSGR_TAG_SEQ:
2196         case CEPH_MSGR_TAG_READY:
2197                 if (req_feat & ~server_feat) {
2198                         pr_err("%s%lld %s protocol feature mismatch,"
2199                                " my required %llx > server's %llx, need %llx\n",
2200                                ENTITY_NAME(con->peer_name),
2201                                ceph_pr_addr(&con->peer_addr.in_addr),
2202                                req_feat, server_feat, req_feat & ~server_feat);
2203                         con->error_msg = "missing required protocol features";
2204                         reset_connection(con);
2205                         return -1;
2206                 }
2207
2208                 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2209                 con->state = CON_STATE_OPEN;
2210                 con->auth_retry = 0;    /* we authenticated; clear flag */
2211                 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2212                 con->connect_seq++;
2213                 con->peer_features = server_feat;
2214                 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2215                      con->peer_global_seq,
2216                      le32_to_cpu(con->in_reply.connect_seq),
2217                      con->connect_seq);
2218                 WARN_ON(con->connect_seq !=
2219                         le32_to_cpu(con->in_reply.connect_seq));
2220
2221                 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2222                         con_flag_set(con, CON_FLAG_LOSSYTX);
2223
2224                 con->delay = 0;      /* reset backoff memory */
2225
2226                 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2227                         prepare_write_seq(con);
2228                         prepare_read_seq(con);
2229                 } else {
2230                         prepare_read_tag(con);
2231                 }
2232                 break;
2233
2234         case CEPH_MSGR_TAG_WAIT:
2235                 /*
2236                  * If there is a connection race (we are opening
2237                  * connections to each other), one of us may just have
2238                  * to WAIT.  This shouldn't happen if we are the
2239                  * client.
2240                  */
2241                 con->error_msg = "protocol error, got WAIT as client";
2242                 return -1;
2243
2244         default:
2245                 con->error_msg = "protocol error, garbage tag during connect";
2246                 return -1;
2247         }
2248         return 0;
2249 }
2250
2251
2252 /*
2253  * read (part of) an ack
2254  */
2255 static int read_partial_ack(struct ceph_connection *con)
2256 {
2257         int size = sizeof (con->in_temp_ack);
2258         int end = size;
2259
2260         return read_partial(con, end, size, &con->in_temp_ack);
2261 }
2262
2263 /*
2264  * We can finally discard anything that's been acked.
2265  */
2266 static void process_ack(struct ceph_connection *con)
2267 {
2268         struct ceph_msg *m;
2269         u64 ack = le64_to_cpu(con->in_temp_ack);
2270         u64 seq;
2271         bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2272         struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
2273
2274         /*
2275          * In the reconnect case, con_fault() has requeued messages
2276          * in out_sent. We should cleanup old messages according to
2277          * the reconnect seq.
2278          */
2279         while (!list_empty(list)) {
2280                 m = list_first_entry(list, struct ceph_msg, list_head);
2281                 if (reconnect && m->needs_out_seq)
2282                         break;
2283                 seq = le64_to_cpu(m->hdr.seq);
2284                 if (seq > ack)
2285                         break;
2286                 dout("got ack for seq %llu type %d at %p\n", seq,
2287                      le16_to_cpu(m->hdr.type), m);
2288                 m->ack_stamp = jiffies;
2289                 ceph_msg_remove(m);
2290         }
2291
2292         prepare_read_tag(con);
2293 }
2294
2295
2296 static int read_partial_message_section(struct ceph_connection *con,
2297                                         struct kvec *section,
2298                                         unsigned int sec_len, u32 *crc)
2299 {
2300         int ret, left;
2301
2302         BUG_ON(!section);
2303
2304         while (section->iov_len < sec_len) {
2305                 BUG_ON(section->iov_base == NULL);
2306                 left = sec_len - section->iov_len;
2307                 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2308                                        section->iov_len, left);
2309                 if (ret <= 0)
2310                         return ret;
2311                 section->iov_len += ret;
2312         }
2313         if (section->iov_len == sec_len)
2314                 *crc = crc32c(0, section->iov_base, section->iov_len);
2315
2316         return 1;
2317 }
2318
2319 static int read_partial_msg_data(struct ceph_connection *con)
2320 {
2321         struct ceph_msg *msg = con->in_msg;
2322         struct ceph_msg_data_cursor *cursor = &msg->cursor;
2323         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2324         struct page *page;
2325         size_t page_offset;
2326         size_t length;
2327         u32 crc = 0;
2328         int ret;
2329
2330         if (!msg->num_data_items)
2331                 return -EIO;
2332
2333         if (do_datacrc)
2334                 crc = con->in_data_crc;
2335         while (cursor->total_resid) {
2336                 if (!cursor->resid) {
2337                         ceph_msg_data_advance(cursor, 0);
2338                         continue;
2339                 }
2340
2341                 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2342                 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2343                 if (ret <= 0) {
2344                         if (do_datacrc)
2345                                 con->in_data_crc = crc;
2346
2347                         return ret;
2348                 }
2349
2350                 if (do_datacrc)
2351                         crc = ceph_crc32c_page(crc, page, page_offset, ret);
2352                 ceph_msg_data_advance(cursor, (size_t)ret);
2353         }
2354         if (do_datacrc)
2355                 con->in_data_crc = crc;
2356
2357         return 1;       /* must return > 0 to indicate success */
2358 }
2359
2360 /*
2361  * read (part of) a message.
2362  */
2363 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2364
2365 static int read_partial_message(struct ceph_connection *con)
2366 {
2367         struct ceph_msg *m = con->in_msg;
2368         int size;
2369         int end;
2370         int ret;
2371         unsigned int front_len, middle_len, data_len;
2372         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2373         bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2374         u64 seq;
2375         u32 crc;
2376
2377         dout("read_partial_message con %p msg %p\n", con, m);
2378
2379         /* header */
2380         size = sizeof (con->in_hdr);
2381         end = size;
2382         ret = read_partial(con, end, size, &con->in_hdr);
2383         if (ret <= 0)
2384                 return ret;
2385
2386         crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2387         if (cpu_to_le32(crc) != con->in_hdr.crc) {
2388                 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2389                        crc, con->in_hdr.crc);
2390                 return -EBADMSG;
2391         }
2392
2393         front_len = le32_to_cpu(con->in_hdr.front_len);
2394         if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2395                 return -EIO;
2396         middle_len = le32_to_cpu(con->in_hdr.middle_len);
2397         if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2398                 return -EIO;
2399         data_len = le32_to_cpu(con->in_hdr.data_len);
2400         if (data_len > CEPH_MSG_MAX_DATA_LEN)
2401                 return -EIO;
2402
2403         /* verify seq# */
2404         seq = le64_to_cpu(con->in_hdr.seq);
2405         if ((s64)seq - (s64)con->in_seq < 1) {
2406                 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2407                         ENTITY_NAME(con->peer_name),
2408                         ceph_pr_addr(&con->peer_addr.in_addr),
2409                         seq, con->in_seq + 1);
2410                 con->in_base_pos = -front_len - middle_len - data_len -
2411                         sizeof_footer(con);
2412                 con->in_tag = CEPH_MSGR_TAG_READY;
2413                 return 1;
2414         } else if ((s64)seq - (s64)con->in_seq > 1) {
2415                 pr_err("read_partial_message bad seq %lld expected %lld\n",
2416                        seq, con->in_seq + 1);
2417                 con->error_msg = "bad message sequence # for incoming message";
2418                 return -EBADE;
2419         }
2420
2421         /* allocate message? */
2422         if (!con->in_msg) {
2423                 int skip = 0;
2424
2425                 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2426                      front_len, data_len);
2427                 ret = ceph_con_in_msg_alloc(con, &skip);
2428                 if (ret < 0)
2429                         return ret;
2430
2431                 BUG_ON(!con->in_msg ^ skip);
2432                 if (skip) {
2433                         /* skip this message */
2434                         dout("alloc_msg said skip message\n");
2435                         con->in_base_pos = -front_len - middle_len - data_len -
2436                                 sizeof_footer(con);
2437                         con->in_tag = CEPH_MSGR_TAG_READY;
2438                         con->in_seq++;
2439                         return 1;
2440                 }
2441
2442                 BUG_ON(!con->in_msg);
2443                 BUG_ON(con->in_msg->con != con);
2444                 m = con->in_msg;
2445                 m->front.iov_len = 0;    /* haven't read it yet */
2446                 if (m->middle)
2447                         m->middle->vec.iov_len = 0;
2448
2449                 /* prepare for data payload, if any */
2450
2451                 if (data_len)
2452                         prepare_message_data(con->in_msg, data_len);
2453         }
2454
2455         /* front */
2456         ret = read_partial_message_section(con, &m->front, front_len,
2457                                            &con->in_front_crc);
2458         if (ret <= 0)
2459                 return ret;
2460
2461         /* middle */
2462         if (m->middle) {
2463                 ret = read_partial_message_section(con, &m->middle->vec,
2464                                                    middle_len,
2465                                                    &con->in_middle_crc);
2466                 if (ret <= 0)
2467                         return ret;
2468         }
2469
2470         /* (page) data */
2471         if (data_len) {
2472                 ret = read_partial_msg_data(con);
2473                 if (ret <= 0)
2474                         return ret;
2475         }
2476
2477         /* footer */
2478         size = sizeof_footer(con);
2479         end += size;
2480         ret = read_partial(con, end, size, &m->footer);
2481         if (ret <= 0)
2482                 return ret;
2483
2484         if (!need_sign) {
2485                 m->footer.flags = m->old_footer.flags;
2486                 m->footer.sig = 0;
2487         }
2488
2489         dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2490              m, front_len, m->footer.front_crc, middle_len,
2491              m->footer.middle_crc, data_len, m->footer.data_crc);
2492
2493         /* crc ok? */
2494         if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2495                 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2496                        m, con->in_front_crc, m->footer.front_crc);
2497                 return -EBADMSG;
2498         }
2499         if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2500                 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2501                        m, con->in_middle_crc, m->footer.middle_crc);
2502                 return -EBADMSG;
2503         }
2504         if (do_datacrc &&
2505             (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2506             con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2507                 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2508                        con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2509                 return -EBADMSG;
2510         }
2511
2512         if (need_sign && con->ops->check_message_signature &&
2513             con->ops->check_message_signature(m)) {
2514                 pr_err("read_partial_message %p signature check failed\n", m);
2515                 return -EBADMSG;
2516         }
2517
2518         return 1; /* done! */
2519 }
2520
2521 /*
2522  * Process message.  This happens in the worker thread.  The callback should
2523  * be careful not to do anything that waits on other incoming messages or it
2524  * may deadlock.
2525  */
2526 static void process_message(struct ceph_connection *con)
2527 {
2528         struct ceph_msg *msg = con->in_msg;
2529
2530         BUG_ON(con->in_msg->con != con);
2531         con->in_msg = NULL;
2532
2533         /* if first message, set peer_name */
2534         if (con->peer_name.type == 0)
2535                 con->peer_name = msg->hdr.src;
2536
2537         con->in_seq++;
2538         mutex_unlock(&con->mutex);
2539
2540         dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2541              msg, le64_to_cpu(msg->hdr.seq),
2542              ENTITY_NAME(msg->hdr.src),
2543              le16_to_cpu(msg->hdr.type),
2544              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2545              le32_to_cpu(msg->hdr.front_len),
2546              le32_to_cpu(msg->hdr.data_len),
2547              con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2548         con->ops->dispatch(con, msg);
2549
2550         mutex_lock(&con->mutex);
2551 }
2552
2553 static int read_keepalive_ack(struct ceph_connection *con)
2554 {
2555         struct ceph_timespec ceph_ts;
2556         size_t size = sizeof(ceph_ts);
2557         int ret = read_partial(con, size, size, &ceph_ts);
2558         if (ret <= 0)
2559                 return ret;
2560         ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
2561         prepare_read_tag(con);
2562         return 1;
2563 }
2564
2565 /*
2566  * Write something to the socket.  Called in a worker thread when the
2567  * socket appears to be writeable and we have something ready to send.
2568  */
2569 static int try_write(struct ceph_connection *con)
2570 {
2571         int ret = 1;
2572
2573         dout("try_write start %p state %lu\n", con, con->state);
2574         if (con->state != CON_STATE_PREOPEN &&
2575             con->state != CON_STATE_CONNECTING &&
2576             con->state != CON_STATE_NEGOTIATING &&
2577             con->state != CON_STATE_OPEN)
2578                 return 0;
2579
2580         /* open the socket first? */
2581         if (con->state == CON_STATE_PREOPEN) {
2582                 BUG_ON(con->sock);
2583                 con->state = CON_STATE_CONNECTING;
2584
2585                 con_out_kvec_reset(con);
2586                 prepare_write_banner(con);
2587                 prepare_read_banner(con);
2588
2589                 BUG_ON(con->in_msg);
2590                 con->in_tag = CEPH_MSGR_TAG_READY;
2591                 dout("try_write initiating connect on %p new state %lu\n",
2592                      con, con->state);
2593                 ret = ceph_tcp_connect(con);
2594                 if (ret < 0) {
2595                         con->error_msg = "connect error";
2596                         goto out;
2597                 }
2598         }
2599
2600 more:
2601         dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2602         BUG_ON(!con->sock);
2603
2604         /* kvec data queued? */
2605         if (con->out_kvec_left) {
2606                 ret = write_partial_kvec(con);
2607                 if (ret <= 0)
2608                         goto out;
2609         }
2610         if (con->out_skip) {
2611                 ret = write_partial_skip(con);
2612                 if (ret <= 0)
2613                         goto out;
2614         }
2615
2616         /* msg pages? */
2617         if (con->out_msg) {
2618                 if (con->out_msg_done) {
2619                         ceph_msg_put(con->out_msg);
2620                         con->out_msg = NULL;   /* we're done with this one */
2621                         goto do_next;
2622                 }
2623
2624                 ret = write_partial_message_data(con);
2625                 if (ret == 1)
2626                         goto more;  /* we need to send the footer, too! */
2627                 if (ret == 0)
2628                         goto out;
2629                 if (ret < 0) {
2630                         dout("try_write write_partial_message_data err %d\n",
2631                              ret);
2632                         goto out;
2633                 }
2634         }
2635
2636 do_next:
2637         if (con->state == CON_STATE_OPEN) {
2638                 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2639                         prepare_write_keepalive(con);
2640                         goto more;
2641                 }
2642                 /* is anything else pending? */
2643                 if (!list_empty(&con->out_queue)) {
2644                         prepare_write_message(con);
2645                         goto more;
2646                 }
2647                 if (con->in_seq > con->in_seq_acked) {
2648                         prepare_write_ack(con);
2649                         goto more;
2650                 }
2651         }
2652
2653         /* Nothing to do! */
2654         con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2655         dout("try_write nothing else to write.\n");
2656         ret = 0;
2657 out:
2658         dout("try_write done on %p ret %d\n", con, ret);
2659         return ret;
2660 }
2661
2662 /*
2663  * Read what we can from the socket.
2664  */
2665 static int try_read(struct ceph_connection *con)
2666 {
2667         int ret = -1;
2668
2669 more:
2670         dout("try_read start on %p state %lu\n", con, con->state);
2671         if (con->state != CON_STATE_CONNECTING &&
2672             con->state != CON_STATE_NEGOTIATING &&
2673             con->state != CON_STATE_OPEN)
2674                 return 0;
2675
2676         BUG_ON(!con->sock);
2677
2678         dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2679              con->in_base_pos);
2680
2681         if (con->state == CON_STATE_CONNECTING) {
2682                 dout("try_read connecting\n");
2683                 ret = read_partial_banner(con);
2684                 if (ret <= 0)
2685                         goto out;
2686                 ret = process_banner(con);
2687                 if (ret < 0)
2688                         goto out;
2689
2690                 con->state = CON_STATE_NEGOTIATING;
2691
2692                 /*
2693                  * Received banner is good, exchange connection info.
2694                  * Do not reset out_kvec, as sending our banner raced
2695                  * with receiving peer banner after connect completed.
2696                  */
2697                 ret = prepare_write_connect(con);
2698                 if (ret < 0)
2699                         goto out;
2700                 prepare_read_connect(con);
2701
2702                 /* Send connection info before awaiting response */
2703                 goto out;
2704         }
2705
2706         if (con->state == CON_STATE_NEGOTIATING) {
2707                 dout("try_read negotiating\n");
2708                 ret = read_partial_connect(con);
2709                 if (ret <= 0)
2710                         goto out;
2711                 ret = process_connect(con);
2712                 if (ret < 0)
2713                         goto out;
2714                 goto more;
2715         }
2716
2717         WARN_ON(con->state != CON_STATE_OPEN);
2718
2719         if (con->in_base_pos < 0) {
2720                 /*
2721                  * skipping + discarding content.
2722                  */
2723                 ret = ceph_tcp_recvmsg(con->sock, NULL, -con->in_base_pos);
2724                 if (ret <= 0)
2725                         goto out;
2726                 dout("skipped %d / %d bytes\n", ret, -con->in_base_pos);
2727                 con->in_base_pos += ret;
2728                 if (con->in_base_pos)
2729                         goto more;
2730         }
2731         if (con->in_tag == CEPH_MSGR_TAG_READY) {
2732                 /*
2733                  * what's next?
2734                  */
2735                 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2736                 if (ret <= 0)
2737                         goto out;
2738                 dout("try_read got tag %d\n", (int)con->in_tag);
2739                 switch (con->in_tag) {
2740                 case CEPH_MSGR_TAG_MSG:
2741                         prepare_read_message(con);
2742                         break;
2743                 case CEPH_MSGR_TAG_ACK:
2744                         prepare_read_ack(con);
2745                         break;
2746                 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2747                         prepare_read_keepalive_ack(con);
2748                         break;
2749                 case CEPH_MSGR_TAG_CLOSE:
2750                         con_close_socket(con);
2751                         con->state = CON_STATE_CLOSED;
2752                         goto out;
2753                 default:
2754                         goto bad_tag;
2755                 }
2756         }
2757         if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2758                 ret = read_partial_message(con);
2759                 if (ret <= 0) {
2760                         switch (ret) {
2761                         case -EBADMSG:
2762                                 con->error_msg = "bad crc/signature";
2763                                 /* fall through */
2764                         case -EBADE:
2765                                 ret = -EIO;
2766                                 break;
2767                         case -EIO:
2768                                 con->error_msg = "io error";
2769                                 break;
2770                         }
2771                         goto out;
2772                 }
2773                 if (con->in_tag == CEPH_MSGR_TAG_READY)
2774                         goto more;
2775                 process_message(con);
2776                 if (con->state == CON_STATE_OPEN)
2777                         prepare_read_tag(con);
2778                 goto more;
2779         }
2780         if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2781             con->in_tag == CEPH_MSGR_TAG_SEQ) {
2782                 /*
2783                  * the final handshake seq exchange is semantically
2784                  * equivalent to an ACK
2785                  */
2786                 ret = read_partial_ack(con);
2787                 if (ret <= 0)
2788                         goto out;
2789                 process_ack(con);
2790                 goto more;
2791         }
2792         if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2793                 ret = read_keepalive_ack(con);
2794                 if (ret <= 0)
2795                         goto out;
2796                 goto more;
2797         }
2798
2799 out:
2800         dout("try_read done on %p ret %d\n", con, ret);
2801         return ret;
2802
2803 bad_tag:
2804         pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2805         con->error_msg = "protocol error, garbage tag";
2806         ret = -1;
2807         goto out;
2808 }
2809
2810
2811 /*
2812  * Atomically queue work on a connection after the specified delay.
2813  * Bump @con reference to avoid races with connection teardown.
2814  * Returns 0 if work was queued, or an error code otherwise.
2815  */
2816 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2817 {
2818         if (!con->ops->get(con)) {
2819                 dout("%s %p ref count 0\n", __func__, con);
2820                 return -ENOENT;
2821         }
2822
2823         if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2824                 dout("%s %p - already queued\n", __func__, con);
2825                 con->ops->put(con);
2826                 return -EBUSY;
2827         }
2828
2829         dout("%s %p %lu\n", __func__, con, delay);
2830         return 0;
2831 }
2832
2833 static void queue_con(struct ceph_connection *con)
2834 {
2835         (void) queue_con_delay(con, 0);
2836 }
2837
2838 static void cancel_con(struct ceph_connection *con)
2839 {
2840         if (cancel_delayed_work(&con->work)) {
2841                 dout("%s %p\n", __func__, con);
2842                 con->ops->put(con);
2843         }
2844 }
2845
2846 static bool con_sock_closed(struct ceph_connection *con)
2847 {
2848         if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2849                 return false;
2850
2851 #define CASE(x)                                                         \
2852         case CON_STATE_ ## x:                                           \
2853                 con->error_msg = "socket closed (con state " #x ")";    \
2854                 break;
2855
2856         switch (con->state) {
2857         CASE(CLOSED);
2858         CASE(PREOPEN);
2859         CASE(CONNECTING);
2860         CASE(NEGOTIATING);
2861         CASE(OPEN);
2862         CASE(STANDBY);
2863         default:
2864                 pr_warn("%s con %p unrecognized state %lu\n",
2865                         __func__, con, con->state);
2866                 con->error_msg = "unrecognized con state";
2867                 BUG();
2868                 break;
2869         }
2870 #undef CASE
2871
2872         return true;
2873 }
2874
2875 static bool con_backoff(struct ceph_connection *con)
2876 {
2877         int ret;
2878
2879         if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2880                 return false;
2881
2882         ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2883         if (ret) {
2884                 dout("%s: con %p FAILED to back off %lu\n", __func__,
2885                         con, con->delay);
2886                 BUG_ON(ret == -ENOENT);
2887                 con_flag_set(con, CON_FLAG_BACKOFF);
2888         }
2889
2890         return true;
2891 }
2892
2893 /* Finish fault handling; con->mutex must *not* be held here */
2894
2895 static void con_fault_finish(struct ceph_connection *con)
2896 {
2897         dout("%s %p\n", __func__, con);
2898
2899         /*
2900          * in case we faulted due to authentication, invalidate our
2901          * current tickets so that we can get new ones.
2902          */
2903         if (con->auth_retry) {
2904                 dout("auth_retry %d, invalidating\n", con->auth_retry);
2905                 if (con->ops->invalidate_authorizer)
2906                         con->ops->invalidate_authorizer(con);
2907                 con->auth_retry = 0;
2908         }
2909
2910         if (con->ops->fault)
2911                 con->ops->fault(con);
2912 }
2913
2914 /*
2915  * Do some work on a connection.  Drop a connection ref when we're done.
2916  */
2917 static void ceph_con_workfn(struct work_struct *work)
2918 {
2919         struct ceph_connection *con = container_of(work, struct ceph_connection,
2920                                                    work.work);
2921         bool fault;
2922
2923         mutex_lock(&con->mutex);
2924         while (true) {
2925                 int ret;
2926
2927                 if ((fault = con_sock_closed(con))) {
2928                         dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2929                         break;
2930                 }
2931                 if (con_backoff(con)) {
2932                         dout("%s: con %p BACKOFF\n", __func__, con);
2933                         break;
2934                 }
2935                 if (con->state == CON_STATE_STANDBY) {
2936                         dout("%s: con %p STANDBY\n", __func__, con);
2937                         break;
2938                 }
2939                 if (con->state == CON_STATE_CLOSED) {
2940                         dout("%s: con %p CLOSED\n", __func__, con);
2941                         BUG_ON(con->sock);
2942                         break;
2943                 }
2944                 if (con->state == CON_STATE_PREOPEN) {
2945                         dout("%s: con %p PREOPEN\n", __func__, con);
2946                         BUG_ON(con->sock);
2947                 }
2948
2949                 ret = try_read(con);
2950                 if (ret < 0) {
2951                         if (ret == -EAGAIN)
2952                                 continue;
2953                         if (!con->error_msg)
2954                                 con->error_msg = "socket error on read";
2955                         fault = true;
2956                         break;
2957                 }
2958
2959                 ret = try_write(con);
2960                 if (ret < 0) {
2961                         if (ret == -EAGAIN)
2962                                 continue;
2963                         if (!con->error_msg)
2964                                 con->error_msg = "socket error on write";
2965                         fault = true;
2966                 }
2967
2968                 break;  /* If we make it to here, we're done */
2969         }
2970         if (fault)
2971                 con_fault(con);
2972         mutex_unlock(&con->mutex);
2973
2974         if (fault)
2975                 con_fault_finish(con);
2976
2977         con->ops->put(con);
2978 }
2979
2980 /*
2981  * Generic error/fault handler.  A retry mechanism is used with
2982  * exponential backoff
2983  */
2984 static void con_fault(struct ceph_connection *con)
2985 {
2986         dout("fault %p state %lu to peer %s\n",
2987              con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2988
2989         pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2990                 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2991         con->error_msg = NULL;
2992
2993         WARN_ON(con->state != CON_STATE_CONNECTING &&
2994                con->state != CON_STATE_NEGOTIATING &&
2995                con->state != CON_STATE_OPEN);
2996
2997         con_close_socket(con);
2998
2999         if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
3000                 dout("fault on LOSSYTX channel, marking CLOSED\n");
3001                 con->state = CON_STATE_CLOSED;
3002                 return;
3003         }
3004
3005         if (con->in_msg) {
3006                 BUG_ON(con->in_msg->con != con);
3007                 ceph_msg_put(con->in_msg);
3008                 con->in_msg = NULL;
3009         }
3010
3011         /* Requeue anything that hasn't been acked */
3012         list_splice_init(&con->out_sent, &con->out_queue);
3013
3014         /* If there are no messages queued or keepalive pending, place
3015          * the connection in a STANDBY state */
3016         if (list_empty(&con->out_queue) &&
3017             !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
3018                 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
3019                 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
3020                 con->state = CON_STATE_STANDBY;
3021         } else {
3022                 /* retry after a delay. */
3023                 con->state = CON_STATE_PREOPEN;
3024                 if (con->delay == 0)
3025                         con->delay = BASE_DELAY_INTERVAL;
3026                 else if (con->delay < MAX_DELAY_INTERVAL)
3027                         con->delay *= 2;
3028                 con_flag_set(con, CON_FLAG_BACKOFF);
3029                 queue_con(con);
3030         }
3031 }
3032
3033
3034
3035 /*
3036  * initialize a new messenger instance
3037  */
3038 void ceph_messenger_init(struct ceph_messenger *msgr,
3039                          struct ceph_entity_addr *myaddr)
3040 {
3041         spin_lock_init(&msgr->global_seq_lock);
3042
3043         if (myaddr)
3044                 msgr->inst.addr = *myaddr;
3045
3046         /* select a random nonce */
3047         msgr->inst.addr.type = 0;
3048         get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3049         encode_my_addr(msgr);
3050
3051         atomic_set(&msgr->stopping, 0);
3052         write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3053
3054         dout("%s %p\n", __func__, msgr);
3055 }
3056 EXPORT_SYMBOL(ceph_messenger_init);
3057
3058 void ceph_messenger_fini(struct ceph_messenger *msgr)
3059 {
3060         put_net(read_pnet(&msgr->net));
3061 }
3062 EXPORT_SYMBOL(ceph_messenger_fini);
3063
3064 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3065 {
3066         if (msg->con)
3067                 msg->con->ops->put(msg->con);
3068
3069         msg->con = con ? con->ops->get(con) : NULL;
3070         BUG_ON(msg->con != con);
3071 }
3072
3073 static void clear_standby(struct ceph_connection *con)
3074 {
3075         /* come back from STANDBY? */
3076         if (con->state == CON_STATE_STANDBY) {
3077                 dout("clear_standby %p and ++connect_seq\n", con);
3078                 con->state = CON_STATE_PREOPEN;
3079                 con->connect_seq++;
3080                 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3081                 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3082         }
3083 }
3084
3085 /*
3086  * Queue up an outgoing message on the given connection.
3087  */
3088 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3089 {
3090         /* set src+dst */
3091         msg->hdr.src = con->msgr->inst.name;
3092         BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3093         msg->needs_out_seq = true;
3094
3095         mutex_lock(&con->mutex);
3096
3097         if (con->state == CON_STATE_CLOSED) {
3098                 dout("con_send %p closed, dropping %p\n", con, msg);
3099                 ceph_msg_put(msg);
3100                 mutex_unlock(&con->mutex);
3101                 return;
3102         }
3103
3104         msg_con_set(msg, con);
3105
3106         BUG_ON(!list_empty(&msg->list_head));
3107         list_add_tail(&msg->list_head, &con->out_queue);
3108         dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3109              ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3110              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3111              le32_to_cpu(msg->hdr.front_len),
3112              le32_to_cpu(msg->hdr.middle_len),
3113              le32_to_cpu(msg->hdr.data_len));
3114
3115         clear_standby(con);
3116         mutex_unlock(&con->mutex);
3117
3118         /* if there wasn't anything waiting to send before, queue
3119          * new work */
3120         if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3121                 queue_con(con);
3122 }
3123 EXPORT_SYMBOL(ceph_con_send);
3124
3125 /*
3126  * Revoke a message that was previously queued for send
3127  */
3128 void ceph_msg_revoke(struct ceph_msg *msg)
3129 {
3130         struct ceph_connection *con = msg->con;
3131
3132         if (!con) {
3133                 dout("%s msg %p null con\n", __func__, msg);
3134                 return;         /* Message not in our possession */
3135         }
3136
3137         mutex_lock(&con->mutex);
3138         if (!list_empty(&msg->list_head)) {
3139                 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3140                 list_del_init(&msg->list_head);
3141                 msg->hdr.seq = 0;
3142
3143                 ceph_msg_put(msg);
3144         }
3145         if (con->out_msg == msg) {
3146                 BUG_ON(con->out_skip);
3147                 /* footer */
3148                 if (con->out_msg_done) {
3149                         con->out_skip += con_out_kvec_skip(con);
3150                 } else {
3151                         BUG_ON(!msg->data_length);
3152                         con->out_skip += sizeof_footer(con);
3153                 }
3154                 /* data, middle, front */
3155                 if (msg->data_length)
3156                         con->out_skip += msg->cursor.total_resid;
3157                 if (msg->middle)
3158                         con->out_skip += con_out_kvec_skip(con);
3159                 con->out_skip += con_out_kvec_skip(con);
3160
3161                 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3162                      __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3163                 msg->hdr.seq = 0;
3164                 con->out_msg = NULL;
3165                 ceph_msg_put(msg);
3166         }
3167
3168         mutex_unlock(&con->mutex);
3169 }
3170
3171 /*
3172  * Revoke a message that we may be reading data into
3173  */
3174 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3175 {
3176         struct ceph_connection *con = msg->con;
3177
3178         if (!con) {
3179                 dout("%s msg %p null con\n", __func__, msg);
3180                 return;         /* Message not in our possession */
3181         }
3182
3183         mutex_lock(&con->mutex);
3184         if (con->in_msg == msg) {
3185                 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3186                 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3187                 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3188
3189                 /* skip rest of message */
3190                 dout("%s %p msg %p revoked\n", __func__, con, msg);
3191                 con->in_base_pos = con->in_base_pos -
3192                                 sizeof(struct ceph_msg_header) -
3193                                 front_len -
3194                                 middle_len -
3195                                 data_len -
3196                                 sizeof(struct ceph_msg_footer);
3197                 ceph_msg_put(con->in_msg);
3198                 con->in_msg = NULL;
3199                 con->in_tag = CEPH_MSGR_TAG_READY;
3200                 con->in_seq++;
3201         } else {
3202                 dout("%s %p in_msg %p msg %p no-op\n",
3203                      __func__, con, con->in_msg, msg);
3204         }
3205         mutex_unlock(&con->mutex);
3206 }
3207
3208 /*
3209  * Queue a keepalive byte to ensure the tcp connection is alive.
3210  */
3211 void ceph_con_keepalive(struct ceph_connection *con)
3212 {
3213         dout("con_keepalive %p\n", con);
3214         mutex_lock(&con->mutex);
3215         clear_standby(con);
3216         mutex_unlock(&con->mutex);
3217         if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3218             con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3219                 queue_con(con);
3220 }
3221 EXPORT_SYMBOL(ceph_con_keepalive);
3222
3223 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3224                                unsigned long interval)
3225 {
3226         if (interval > 0 &&
3227             (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3228                 struct timespec64 now;
3229                 struct timespec64 ts;
3230                 ktime_get_real_ts64(&now);
3231                 jiffies_to_timespec64(interval, &ts);
3232                 ts = timespec64_add(con->last_keepalive_ack, ts);
3233                 return timespec64_compare(&now, &ts) >= 0;
3234         }
3235         return false;
3236 }
3237
3238 static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
3239 {
3240         BUG_ON(msg->num_data_items >= msg->max_data_items);
3241         return &msg->data[msg->num_data_items++];
3242 }
3243
3244 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3245 {
3246         if (data->type == CEPH_MSG_DATA_PAGELIST)
3247                 ceph_pagelist_release(data->pagelist);
3248 }
3249
3250 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3251                 size_t length, size_t alignment)
3252 {
3253         struct ceph_msg_data *data;
3254
3255         BUG_ON(!pages);
3256         BUG_ON(!length);
3257
3258         data = ceph_msg_data_add(msg);
3259         data->type = CEPH_MSG_DATA_PAGES;
3260         data->pages = pages;
3261         data->length = length;
3262         data->alignment = alignment & ~PAGE_MASK;
3263
3264         msg->data_length += length;
3265 }
3266 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3267
3268 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3269                                 struct ceph_pagelist *pagelist)
3270 {
3271         struct ceph_msg_data *data;
3272
3273         BUG_ON(!pagelist);
3274         BUG_ON(!pagelist->length);
3275
3276         data = ceph_msg_data_add(msg);
3277         data->type = CEPH_MSG_DATA_PAGELIST;
3278         refcount_inc(&pagelist->refcnt);
3279         data->pagelist = pagelist;
3280
3281         msg->data_length += pagelist->length;
3282 }
3283 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3284
3285 #ifdef  CONFIG_BLOCK
3286 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
3287                            u32 length)
3288 {
3289         struct ceph_msg_data *data;
3290
3291         data = ceph_msg_data_add(msg);
3292         data->type = CEPH_MSG_DATA_BIO;
3293         data->bio_pos = *bio_pos;
3294         data->bio_length = length;
3295
3296         msg->data_length += length;
3297 }
3298 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3299 #endif  /* CONFIG_BLOCK */
3300
3301 void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
3302                              struct ceph_bvec_iter *bvec_pos)
3303 {
3304         struct ceph_msg_data *data;
3305
3306         data = ceph_msg_data_add(msg);
3307         data->type = CEPH_MSG_DATA_BVECS;
3308         data->bvec_pos = *bvec_pos;
3309
3310         msg->data_length += bvec_pos->iter.bi_size;
3311 }
3312 EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
3313
3314 /*
3315  * construct a new message with given type, size
3316  * the new msg has a ref count of 1.
3317  */
3318 struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
3319                                gfp_t flags, bool can_fail)
3320 {
3321         struct ceph_msg *m;
3322
3323         m = kmem_cache_zalloc(ceph_msg_cache, flags);
3324         if (m == NULL)
3325                 goto out;
3326
3327         m-&