Merge tag 'fbdev-v4.20' of https://github.com/bzolnier/linux
[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 static struct kmem_cache        *ceph_msg_data_cache;
160
161 /* static tag bytes (protocol control messages) */
162 static char tag_msg = CEPH_MSGR_TAG_MSG;
163 static char tag_ack = CEPH_MSGR_TAG_ACK;
164 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
165 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
166
167 #ifdef CONFIG_LOCKDEP
168 static struct lock_class_key socket_class;
169 #endif
170
171 static void queue_con(struct ceph_connection *con);
172 static void cancel_con(struct ceph_connection *con);
173 static void ceph_con_workfn(struct work_struct *);
174 static void con_fault(struct ceph_connection *con);
175
176 /*
177  * Nicely render a sockaddr as a string.  An array of formatted
178  * strings is used, to approximate reentrancy.
179  */
180 #define ADDR_STR_COUNT_LOG      5       /* log2(# address strings in array) */
181 #define ADDR_STR_COUNT          (1 << ADDR_STR_COUNT_LOG)
182 #define ADDR_STR_COUNT_MASK     (ADDR_STR_COUNT - 1)
183 #define MAX_ADDR_STR_LEN        64      /* 54 is enough */
184
185 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
186 static atomic_t addr_str_seq = ATOMIC_INIT(0);
187
188 static struct page *zero_page;          /* used in certain error cases */
189
190 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
191 {
192         int i;
193         char *s;
194         struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
195         struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
196
197         i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
198         s = addr_str[i];
199
200         switch (ss->ss_family) {
201         case AF_INET:
202                 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
203                          ntohs(in4->sin_port));
204                 break;
205
206         case AF_INET6:
207                 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
208                          ntohs(in6->sin6_port));
209                 break;
210
211         default:
212                 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
213                          ss->ss_family);
214         }
215
216         return s;
217 }
218 EXPORT_SYMBOL(ceph_pr_addr);
219
220 static void encode_my_addr(struct ceph_messenger *msgr)
221 {
222         memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
223         ceph_encode_addr(&msgr->my_enc_addr);
224 }
225
226 /*
227  * work queue for all reading and writing to/from the socket.
228  */
229 static struct workqueue_struct *ceph_msgr_wq;
230
231 static int ceph_msgr_slab_init(void)
232 {
233         BUG_ON(ceph_msg_cache);
234         ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
235         if (!ceph_msg_cache)
236                 return -ENOMEM;
237
238         BUG_ON(ceph_msg_data_cache);
239         ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
240         if (ceph_msg_data_cache)
241                 return 0;
242
243         kmem_cache_destroy(ceph_msg_cache);
244         ceph_msg_cache = NULL;
245
246         return -ENOMEM;
247 }
248
249 static void ceph_msgr_slab_exit(void)
250 {
251         BUG_ON(!ceph_msg_data_cache);
252         kmem_cache_destroy(ceph_msg_data_cache);
253         ceph_msg_data_cache = NULL;
254
255         BUG_ON(!ceph_msg_cache);
256         kmem_cache_destroy(ceph_msg_cache);
257         ceph_msg_cache = NULL;
258 }
259
260 static void _ceph_msgr_exit(void)
261 {
262         if (ceph_msgr_wq) {
263                 destroy_workqueue(ceph_msgr_wq);
264                 ceph_msgr_wq = NULL;
265         }
266
267         BUG_ON(zero_page == NULL);
268         put_page(zero_page);
269         zero_page = NULL;
270
271         ceph_msgr_slab_exit();
272 }
273
274 int __init ceph_msgr_init(void)
275 {
276         if (ceph_msgr_slab_init())
277                 return -ENOMEM;
278
279         BUG_ON(zero_page != NULL);
280         zero_page = ZERO_PAGE(0);
281         get_page(zero_page);
282
283         /*
284          * The number of active work items is limited by the number of
285          * connections, so leave @max_active at default.
286          */
287         ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
288         if (ceph_msgr_wq)
289                 return 0;
290
291         pr_err("msgr_init failed to create workqueue\n");
292         _ceph_msgr_exit();
293
294         return -ENOMEM;
295 }
296
297 void ceph_msgr_exit(void)
298 {
299         BUG_ON(ceph_msgr_wq == NULL);
300
301         _ceph_msgr_exit();
302 }
303
304 void ceph_msgr_flush(void)
305 {
306         flush_workqueue(ceph_msgr_wq);
307 }
308 EXPORT_SYMBOL(ceph_msgr_flush);
309
310 /* Connection socket state transition functions */
311
312 static void con_sock_state_init(struct ceph_connection *con)
313 {
314         int old_state;
315
316         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
317         if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
318                 printk("%s: unexpected old state %d\n", __func__, old_state);
319         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
320              CON_SOCK_STATE_CLOSED);
321 }
322
323 static void con_sock_state_connecting(struct ceph_connection *con)
324 {
325         int old_state;
326
327         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
328         if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
329                 printk("%s: unexpected old state %d\n", __func__, old_state);
330         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
331              CON_SOCK_STATE_CONNECTING);
332 }
333
334 static void con_sock_state_connected(struct ceph_connection *con)
335 {
336         int old_state;
337
338         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
339         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
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_CONNECTED);
343 }
344
345 static void con_sock_state_closing(struct ceph_connection *con)
346 {
347         int old_state;
348
349         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
350         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
351                         old_state != CON_SOCK_STATE_CONNECTED &&
352                         old_state != CON_SOCK_STATE_CLOSING))
353                 printk("%s: unexpected old state %d\n", __func__, old_state);
354         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
355              CON_SOCK_STATE_CLOSING);
356 }
357
358 static void con_sock_state_closed(struct ceph_connection *con)
359 {
360         int old_state;
361
362         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
363         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
364                     old_state != CON_SOCK_STATE_CLOSING &&
365                     old_state != CON_SOCK_STATE_CONNECTING &&
366                     old_state != CON_SOCK_STATE_CLOSED))
367                 printk("%s: unexpected old state %d\n", __func__, old_state);
368         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
369              CON_SOCK_STATE_CLOSED);
370 }
371
372 /*
373  * socket callback functions
374  */
375
376 /* data available on socket, or listen socket received a connect */
377 static void ceph_sock_data_ready(struct sock *sk)
378 {
379         struct ceph_connection *con = sk->sk_user_data;
380         if (atomic_read(&con->msgr->stopping)) {
381                 return;
382         }
383
384         if (sk->sk_state != TCP_CLOSE_WAIT) {
385                 dout("%s on %p state = %lu, queueing work\n", __func__,
386                      con, con->state);
387                 queue_con(con);
388         }
389 }
390
391 /* socket has buffer space for writing */
392 static void ceph_sock_write_space(struct sock *sk)
393 {
394         struct ceph_connection *con = sk->sk_user_data;
395
396         /* only queue to workqueue if there is data we want to write,
397          * and there is sufficient space in the socket buffer to accept
398          * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
399          * doesn't get called again until try_write() fills the socket
400          * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
401          * and net/core/stream.c:sk_stream_write_space().
402          */
403         if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
404                 if (sk_stream_is_writeable(sk)) {
405                         dout("%s %p queueing write work\n", __func__, con);
406                         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
407                         queue_con(con);
408                 }
409         } else {
410                 dout("%s %p nothing to write\n", __func__, con);
411         }
412 }
413
414 /* socket's state has changed */
415 static void ceph_sock_state_change(struct sock *sk)
416 {
417         struct ceph_connection *con = sk->sk_user_data;
418
419         dout("%s %p state = %lu sk_state = %u\n", __func__,
420              con, con->state, sk->sk_state);
421
422         switch (sk->sk_state) {
423         case TCP_CLOSE:
424                 dout("%s TCP_CLOSE\n", __func__);
425                 /* fall through */
426         case TCP_CLOSE_WAIT:
427                 dout("%s TCP_CLOSE_WAIT\n", __func__);
428                 con_sock_state_closing(con);
429                 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
430                 queue_con(con);
431                 break;
432         case TCP_ESTABLISHED:
433                 dout("%s TCP_ESTABLISHED\n", __func__);
434                 con_sock_state_connected(con);
435                 queue_con(con);
436                 break;
437         default:        /* Everything else is uninteresting */
438                 break;
439         }
440 }
441
442 /*
443  * set up socket callbacks
444  */
445 static void set_sock_callbacks(struct socket *sock,
446                                struct ceph_connection *con)
447 {
448         struct sock *sk = sock->sk;
449         sk->sk_user_data = con;
450         sk->sk_data_ready = ceph_sock_data_ready;
451         sk->sk_write_space = ceph_sock_write_space;
452         sk->sk_state_change = ceph_sock_state_change;
453 }
454
455
456 /*
457  * socket helpers
458  */
459
460 /*
461  * initiate connection to a remote socket.
462  */
463 static int ceph_tcp_connect(struct ceph_connection *con)
464 {
465         struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
466         struct socket *sock;
467         unsigned int noio_flag;
468         int ret;
469
470         BUG_ON(con->sock);
471
472         /* sock_create_kern() allocates with GFP_KERNEL */
473         noio_flag = memalloc_noio_save();
474         ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
475                                SOCK_STREAM, IPPROTO_TCP, &sock);
476         memalloc_noio_restore(noio_flag);
477         if (ret)
478                 return ret;
479         sock->sk->sk_allocation = GFP_NOFS;
480
481 #ifdef CONFIG_LOCKDEP
482         lockdep_set_class(&sock->sk->sk_lock, &socket_class);
483 #endif
484
485         set_sock_callbacks(sock, con);
486
487         dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
488
489         con_sock_state_connecting(con);
490         ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
491                                  O_NONBLOCK);
492         if (ret == -EINPROGRESS) {
493                 dout("connect %s EINPROGRESS sk_state = %u\n",
494                      ceph_pr_addr(&con->peer_addr.in_addr),
495                      sock->sk->sk_state);
496         } else if (ret < 0) {
497                 pr_err("connect %s error %d\n",
498                        ceph_pr_addr(&con->peer_addr.in_addr), ret);
499                 sock_release(sock);
500                 return ret;
501         }
502
503         if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
504                 int optval = 1;
505
506                 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
507                                         (char *)&optval, sizeof(optval));
508                 if (ret)
509                         pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
510                                ret);
511         }
512
513         con->sock = sock;
514         return 0;
515 }
516
517 /*
518  * If @buf is NULL, discard up to @len bytes.
519  */
520 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
521 {
522         struct kvec iov = {buf, len};
523         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
524         int r;
525
526         if (!buf)
527                 msg.msg_flags |= MSG_TRUNC;
528
529         iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, &iov, 1, len);
530         r = sock_recvmsg(sock, &msg, msg.msg_flags);
531         if (r == -EAGAIN)
532                 r = 0;
533         return r;
534 }
535
536 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
537                      int page_offset, size_t length)
538 {
539         struct bio_vec bvec = {
540                 .bv_page = page,
541                 .bv_offset = page_offset,
542                 .bv_len = length
543         };
544         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
545         int r;
546
547         BUG_ON(page_offset + length > PAGE_SIZE);
548         iov_iter_bvec(&msg.msg_iter, READ | ITER_BVEC, &bvec, 1, length);
549         r = sock_recvmsg(sock, &msg, msg.msg_flags);
550         if (r == -EAGAIN)
551                 r = 0;
552         return r;
553 }
554
555 /*
556  * write something.  @more is true if caller will be sending more data
557  * shortly.
558  */
559 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
560                      size_t kvlen, size_t len, int more)
561 {
562         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
563         int r;
564
565         if (more)
566                 msg.msg_flags |= MSG_MORE;
567         else
568                 msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
569
570         r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
571         if (r == -EAGAIN)
572                 r = 0;
573         return r;
574 }
575
576 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
577                      int offset, size_t size, bool more)
578 {
579         int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
580         int ret;
581
582         ret = kernel_sendpage(sock, page, offset, size, flags);
583         if (ret == -EAGAIN)
584                 ret = 0;
585
586         return ret;
587 }
588
589 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
590                      int offset, size_t size, bool more)
591 {
592         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
593         struct bio_vec bvec;
594         int ret;
595
596         /* sendpage cannot properly handle pages with page_count == 0,
597          * we need to fallback to sendmsg if that's the case */
598         if (page_count(page) >= 1)
599                 return __ceph_tcp_sendpage(sock, page, offset, size, more);
600
601         bvec.bv_page = page;
602         bvec.bv_offset = offset;
603         bvec.bv_len = size;
604
605         if (more)
606                 msg.msg_flags |= MSG_MORE;
607         else
608                 msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
609
610         iov_iter_bvec(&msg.msg_iter, WRITE | ITER_BVEC, &bvec, 1, size);
611         ret = sock_sendmsg(sock, &msg);
612         if (ret == -EAGAIN)
613                 ret = 0;
614
615         return ret;
616 }
617
618 /*
619  * Shutdown/close the socket for the given connection.
620  */
621 static int con_close_socket(struct ceph_connection *con)
622 {
623         int rc = 0;
624
625         dout("con_close_socket on %p sock %p\n", con, con->sock);
626         if (con->sock) {
627                 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
628                 sock_release(con->sock);
629                 con->sock = NULL;
630         }
631
632         /*
633          * Forcibly clear the SOCK_CLOSED flag.  It gets set
634          * independent of the connection mutex, and we could have
635          * received a socket close event before we had the chance to
636          * shut the socket down.
637          */
638         con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
639
640         con_sock_state_closed(con);
641         return rc;
642 }
643
644 /*
645  * Reset a connection.  Discard all incoming and outgoing messages
646  * and clear *_seq state.
647  */
648 static void ceph_msg_remove(struct ceph_msg *msg)
649 {
650         list_del_init(&msg->list_head);
651
652         ceph_msg_put(msg);
653 }
654 static void ceph_msg_remove_list(struct list_head *head)
655 {
656         while (!list_empty(head)) {
657                 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
658                                                         list_head);
659                 ceph_msg_remove(msg);
660         }
661 }
662
663 static void reset_connection(struct ceph_connection *con)
664 {
665         /* reset connection, out_queue, msg_ and connect_seq */
666         /* discard existing out_queue and msg_seq */
667         dout("reset_connection %p\n", con);
668         ceph_msg_remove_list(&con->out_queue);
669         ceph_msg_remove_list(&con->out_sent);
670
671         if (con->in_msg) {
672                 BUG_ON(con->in_msg->con != con);
673                 ceph_msg_put(con->in_msg);
674                 con->in_msg = NULL;
675         }
676
677         con->connect_seq = 0;
678         con->out_seq = 0;
679         if (con->out_msg) {
680                 BUG_ON(con->out_msg->con != con);
681                 ceph_msg_put(con->out_msg);
682                 con->out_msg = NULL;
683         }
684         con->in_seq = 0;
685         con->in_seq_acked = 0;
686
687         con->out_skip = 0;
688 }
689
690 /*
691  * mark a peer down.  drop any open connections.
692  */
693 void ceph_con_close(struct ceph_connection *con)
694 {
695         mutex_lock(&con->mutex);
696         dout("con_close %p peer %s\n", con,
697              ceph_pr_addr(&con->peer_addr.in_addr));
698         con->state = CON_STATE_CLOSED;
699
700         con_flag_clear(con, CON_FLAG_LOSSYTX);  /* so we retry next connect */
701         con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
702         con_flag_clear(con, CON_FLAG_WRITE_PENDING);
703         con_flag_clear(con, CON_FLAG_BACKOFF);
704
705         reset_connection(con);
706         con->peer_global_seq = 0;
707         cancel_con(con);
708         con_close_socket(con);
709         mutex_unlock(&con->mutex);
710 }
711 EXPORT_SYMBOL(ceph_con_close);
712
713 /*
714  * Reopen a closed connection, with a new peer address.
715  */
716 void ceph_con_open(struct ceph_connection *con,
717                    __u8 entity_type, __u64 entity_num,
718                    struct ceph_entity_addr *addr)
719 {
720         mutex_lock(&con->mutex);
721         dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
722
723         WARN_ON(con->state != CON_STATE_CLOSED);
724         con->state = CON_STATE_PREOPEN;
725
726         con->peer_name.type = (__u8) entity_type;
727         con->peer_name.num = cpu_to_le64(entity_num);
728
729         memcpy(&con->peer_addr, addr, sizeof(*addr));
730         con->delay = 0;      /* reset backoff memory */
731         mutex_unlock(&con->mutex);
732         queue_con(con);
733 }
734 EXPORT_SYMBOL(ceph_con_open);
735
736 /*
737  * return true if this connection ever successfully opened
738  */
739 bool ceph_con_opened(struct ceph_connection *con)
740 {
741         return con->connect_seq > 0;
742 }
743
744 /*
745  * initialize a new connection.
746  */
747 void ceph_con_init(struct ceph_connection *con, void *private,
748         const struct ceph_connection_operations *ops,
749         struct ceph_messenger *msgr)
750 {
751         dout("con_init %p\n", con);
752         memset(con, 0, sizeof(*con));
753         con->private = private;
754         con->ops = ops;
755         con->msgr = msgr;
756
757         con_sock_state_init(con);
758
759         mutex_init(&con->mutex);
760         INIT_LIST_HEAD(&con->out_queue);
761         INIT_LIST_HEAD(&con->out_sent);
762         INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
763
764         con->state = CON_STATE_CLOSED;
765 }
766 EXPORT_SYMBOL(ceph_con_init);
767
768
769 /*
770  * We maintain a global counter to order connection attempts.  Get
771  * a unique seq greater than @gt.
772  */
773 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
774 {
775         u32 ret;
776
777         spin_lock(&msgr->global_seq_lock);
778         if (msgr->global_seq < gt)
779                 msgr->global_seq = gt;
780         ret = ++msgr->global_seq;
781         spin_unlock(&msgr->global_seq_lock);
782         return ret;
783 }
784
785 static void con_out_kvec_reset(struct ceph_connection *con)
786 {
787         BUG_ON(con->out_skip);
788
789         con->out_kvec_left = 0;
790         con->out_kvec_bytes = 0;
791         con->out_kvec_cur = &con->out_kvec[0];
792 }
793
794 static void con_out_kvec_add(struct ceph_connection *con,
795                                 size_t size, void *data)
796 {
797         int index = con->out_kvec_left;
798
799         BUG_ON(con->out_skip);
800         BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
801
802         con->out_kvec[index].iov_len = size;
803         con->out_kvec[index].iov_base = data;
804         con->out_kvec_left++;
805         con->out_kvec_bytes += size;
806 }
807
808 /*
809  * Chop off a kvec from the end.  Return residual number of bytes for
810  * that kvec, i.e. how many bytes would have been written if the kvec
811  * hadn't been nuked.
812  */
813 static int con_out_kvec_skip(struct ceph_connection *con)
814 {
815         int off = con->out_kvec_cur - con->out_kvec;
816         int skip = 0;
817
818         if (con->out_kvec_bytes > 0) {
819                 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
820                 BUG_ON(con->out_kvec_bytes < skip);
821                 BUG_ON(!con->out_kvec_left);
822                 con->out_kvec_bytes -= skip;
823                 con->out_kvec_left--;
824         }
825
826         return skip;
827 }
828
829 #ifdef CONFIG_BLOCK
830
831 /*
832  * For a bio data item, a piece is whatever remains of the next
833  * entry in the current bio iovec, or the first entry in the next
834  * bio in the list.
835  */
836 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
837                                         size_t length)
838 {
839         struct ceph_msg_data *data = cursor->data;
840         struct ceph_bio_iter *it = &cursor->bio_iter;
841
842         cursor->resid = min_t(size_t, length, data->bio_length);
843         *it = data->bio_pos;
844         if (cursor->resid < it->iter.bi_size)
845                 it->iter.bi_size = cursor->resid;
846
847         BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
848         cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
849 }
850
851 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
852                                                 size_t *page_offset,
853                                                 size_t *length)
854 {
855         struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
856                                            cursor->bio_iter.iter);
857
858         *page_offset = bv.bv_offset;
859         *length = bv.bv_len;
860         return bv.bv_page;
861 }
862
863 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
864                                         size_t bytes)
865 {
866         struct ceph_bio_iter *it = &cursor->bio_iter;
867
868         BUG_ON(bytes > cursor->resid);
869         BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
870         cursor->resid -= bytes;
871         bio_advance_iter(it->bio, &it->iter, bytes);
872
873         if (!cursor->resid) {
874                 BUG_ON(!cursor->last_piece);
875                 return false;   /* no more data */
876         }
877
878         if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done))
879                 return false;   /* more bytes to process in this segment */
880
881         if (!it->iter.bi_size) {
882                 it->bio = it->bio->bi_next;
883                 it->iter = it->bio->bi_iter;
884                 if (cursor->resid < it->iter.bi_size)
885                         it->iter.bi_size = cursor->resid;
886         }
887
888         BUG_ON(cursor->last_piece);
889         BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
890         cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
891         return true;
892 }
893 #endif /* CONFIG_BLOCK */
894
895 static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
896                                         size_t length)
897 {
898         struct ceph_msg_data *data = cursor->data;
899         struct bio_vec *bvecs = data->bvec_pos.bvecs;
900
901         cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
902         cursor->bvec_iter = data->bvec_pos.iter;
903         cursor->bvec_iter.bi_size = cursor->resid;
904
905         BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
906         cursor->last_piece =
907             cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
908 }
909
910 static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
911                                                 size_t *page_offset,
912                                                 size_t *length)
913 {
914         struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
915                                            cursor->bvec_iter);
916
917         *page_offset = bv.bv_offset;
918         *length = bv.bv_len;
919         return bv.bv_page;
920 }
921
922 static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
923                                         size_t bytes)
924 {
925         struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
926
927         BUG_ON(bytes > cursor->resid);
928         BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
929         cursor->resid -= bytes;
930         bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
931
932         if (!cursor->resid) {
933                 BUG_ON(!cursor->last_piece);
934                 return false;   /* no more data */
935         }
936
937         if (!bytes || cursor->bvec_iter.bi_bvec_done)
938                 return false;   /* more bytes to process in this segment */
939
940         BUG_ON(cursor->last_piece);
941         BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
942         cursor->last_piece =
943             cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
944         return true;
945 }
946
947 /*
948  * For a page array, a piece comes from the first page in the array
949  * that has not already been fully consumed.
950  */
951 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
952                                         size_t length)
953 {
954         struct ceph_msg_data *data = cursor->data;
955         int page_count;
956
957         BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
958
959         BUG_ON(!data->pages);
960         BUG_ON(!data->length);
961
962         cursor->resid = min(length, data->length);
963         page_count = calc_pages_for(data->alignment, (u64)data->length);
964         cursor->page_offset = data->alignment & ~PAGE_MASK;
965         cursor->page_index = 0;
966         BUG_ON(page_count > (int)USHRT_MAX);
967         cursor->page_count = (unsigned short)page_count;
968         BUG_ON(length > SIZE_MAX - cursor->page_offset);
969         cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
970 }
971
972 static struct page *
973 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
974                                         size_t *page_offset, size_t *length)
975 {
976         struct ceph_msg_data *data = cursor->data;
977
978         BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
979
980         BUG_ON(cursor->page_index >= cursor->page_count);
981         BUG_ON(cursor->page_offset >= PAGE_SIZE);
982
983         *page_offset = cursor->page_offset;
984         if (cursor->last_piece)
985                 *length = cursor->resid;
986         else
987                 *length = PAGE_SIZE - *page_offset;
988
989         return data->pages[cursor->page_index];
990 }
991
992 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
993                                                 size_t bytes)
994 {
995         BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
996
997         BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
998
999         /* Advance the cursor page offset */
1000
1001         cursor->resid -= bytes;
1002         cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
1003         if (!bytes || cursor->page_offset)
1004                 return false;   /* more bytes to process in the current page */
1005
1006         if (!cursor->resid)
1007                 return false;   /* no more data */
1008
1009         /* Move on to the next page; offset is already at 0 */
1010
1011         BUG_ON(cursor->page_index >= cursor->page_count);
1012         cursor->page_index++;
1013         cursor->last_piece = cursor->resid <= PAGE_SIZE;
1014
1015         return true;
1016 }
1017
1018 /*
1019  * For a pagelist, a piece is whatever remains to be consumed in the
1020  * first page in the list, or the front of the next page.
1021  */
1022 static void
1023 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
1024                                         size_t length)
1025 {
1026         struct ceph_msg_data *data = cursor->data;
1027         struct ceph_pagelist *pagelist;
1028         struct page *page;
1029
1030         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1031
1032         pagelist = data->pagelist;
1033         BUG_ON(!pagelist);
1034
1035         if (!length)
1036                 return;         /* pagelist can be assigned but empty */
1037
1038         BUG_ON(list_empty(&pagelist->head));
1039         page = list_first_entry(&pagelist->head, struct page, lru);
1040
1041         cursor->resid = min(length, pagelist->length);
1042         cursor->page = page;
1043         cursor->offset = 0;
1044         cursor->last_piece = cursor->resid <= PAGE_SIZE;
1045 }
1046
1047 static struct page *
1048 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1049                                 size_t *page_offset, size_t *length)
1050 {
1051         struct ceph_msg_data *data = cursor->data;
1052         struct ceph_pagelist *pagelist;
1053
1054         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1055
1056         pagelist = data->pagelist;
1057         BUG_ON(!pagelist);
1058
1059         BUG_ON(!cursor->page);
1060         BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1061
1062         /* offset of first page in pagelist is always 0 */
1063         *page_offset = cursor->offset & ~PAGE_MASK;
1064         if (cursor->last_piece)
1065                 *length = cursor->resid;
1066         else
1067                 *length = PAGE_SIZE - *page_offset;
1068
1069         return cursor->page;
1070 }
1071
1072 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1073                                                 size_t bytes)
1074 {
1075         struct ceph_msg_data *data = cursor->data;
1076         struct ceph_pagelist *pagelist;
1077
1078         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1079
1080         pagelist = data->pagelist;
1081         BUG_ON(!pagelist);
1082
1083         BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1084         BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1085
1086         /* Advance the cursor offset */
1087
1088         cursor->resid -= bytes;
1089         cursor->offset += bytes;
1090         /* offset of first page in pagelist is always 0 */
1091         if (!bytes || cursor->offset & ~PAGE_MASK)
1092                 return false;   /* more bytes to process in the current page */
1093
1094         if (!cursor->resid)
1095                 return false;   /* no more data */
1096
1097         /* Move on to the next page */
1098
1099         BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1100         cursor->page = list_next_entry(cursor->page, lru);
1101         cursor->last_piece = cursor->resid <= PAGE_SIZE;
1102
1103         return true;
1104 }
1105
1106 /*
1107  * Message data is handled (sent or received) in pieces, where each
1108  * piece resides on a single page.  The network layer might not
1109  * consume an entire piece at once.  A data item's cursor keeps
1110  * track of which piece is next to process and how much remains to
1111  * be processed in that piece.  It also tracks whether the current
1112  * piece is the last one in the data item.
1113  */
1114 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1115 {
1116         size_t length = cursor->total_resid;
1117
1118         switch (cursor->data->type) {
1119         case CEPH_MSG_DATA_PAGELIST:
1120                 ceph_msg_data_pagelist_cursor_init(cursor, length);
1121                 break;
1122         case CEPH_MSG_DATA_PAGES:
1123                 ceph_msg_data_pages_cursor_init(cursor, length);
1124                 break;
1125 #ifdef CONFIG_BLOCK
1126         case CEPH_MSG_DATA_BIO:
1127                 ceph_msg_data_bio_cursor_init(cursor, length);
1128                 break;
1129 #endif /* CONFIG_BLOCK */
1130         case CEPH_MSG_DATA_BVECS:
1131                 ceph_msg_data_bvecs_cursor_init(cursor, length);
1132                 break;
1133         case CEPH_MSG_DATA_NONE:
1134         default:
1135                 /* BUG(); */
1136                 break;
1137         }
1138         cursor->need_crc = true;
1139 }
1140
1141 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1142 {
1143         struct ceph_msg_data_cursor *cursor = &msg->cursor;
1144         struct ceph_msg_data *data;
1145
1146         BUG_ON(!length);
1147         BUG_ON(length > msg->data_length);
1148         BUG_ON(list_empty(&msg->data));
1149
1150         cursor->data_head = &msg->data;
1151         cursor->total_resid = length;
1152         data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1153         cursor->data = data;
1154
1155         __ceph_msg_data_cursor_init(cursor);
1156 }
1157
1158 /*
1159  * Return the page containing the next piece to process for a given
1160  * data item, and supply the page offset and length of that piece.
1161  * Indicate whether this is the last piece in this data item.
1162  */
1163 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1164                                         size_t *page_offset, size_t *length,
1165                                         bool *last_piece)
1166 {
1167         struct page *page;
1168
1169         switch (cursor->data->type) {
1170         case CEPH_MSG_DATA_PAGELIST:
1171                 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1172                 break;
1173         case CEPH_MSG_DATA_PAGES:
1174                 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1175                 break;
1176 #ifdef CONFIG_BLOCK
1177         case CEPH_MSG_DATA_BIO:
1178                 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1179                 break;
1180 #endif /* CONFIG_BLOCK */
1181         case CEPH_MSG_DATA_BVECS:
1182                 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1183                 break;
1184         case CEPH_MSG_DATA_NONE:
1185         default:
1186                 page = NULL;
1187                 break;
1188         }
1189
1190         BUG_ON(!page);
1191         BUG_ON(*page_offset + *length > PAGE_SIZE);
1192         BUG_ON(!*length);
1193         BUG_ON(*length > cursor->resid);
1194         if (last_piece)
1195                 *last_piece = cursor->last_piece;
1196
1197         return page;
1198 }
1199
1200 /*
1201  * Returns true if the result moves the cursor on to the next piece
1202  * of the data item.
1203  */
1204 static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1205                                   size_t bytes)
1206 {
1207         bool new_piece;
1208
1209         BUG_ON(bytes > cursor->resid);
1210         switch (cursor->data->type) {
1211         case CEPH_MSG_DATA_PAGELIST:
1212                 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1213                 break;
1214         case CEPH_MSG_DATA_PAGES:
1215                 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1216                 break;
1217 #ifdef CONFIG_BLOCK
1218         case CEPH_MSG_DATA_BIO:
1219                 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1220                 break;
1221 #endif /* CONFIG_BLOCK */
1222         case CEPH_MSG_DATA_BVECS:
1223                 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1224                 break;
1225         case CEPH_MSG_DATA_NONE:
1226         default:
1227                 BUG();
1228                 break;
1229         }
1230         cursor->total_resid -= bytes;
1231
1232         if (!cursor->resid && cursor->total_resid) {
1233                 WARN_ON(!cursor->last_piece);
1234                 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1235                 cursor->data = list_next_entry(cursor->data, links);
1236                 __ceph_msg_data_cursor_init(cursor);
1237                 new_piece = true;
1238         }
1239         cursor->need_crc = new_piece;
1240 }
1241
1242 static size_t sizeof_footer(struct ceph_connection *con)
1243 {
1244         return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1245             sizeof(struct ceph_msg_footer) :
1246             sizeof(struct ceph_msg_footer_old);
1247 }
1248
1249 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1250 {
1251         BUG_ON(!msg);
1252         BUG_ON(!data_len);
1253
1254         /* Initialize data cursor */
1255
1256         ceph_msg_data_cursor_init(msg, (size_t)data_len);
1257 }
1258
1259 /*
1260  * Prepare footer for currently outgoing message, and finish things
1261  * off.  Assumes out_kvec* are already valid.. we just add on to the end.
1262  */
1263 static void prepare_write_message_footer(struct ceph_connection *con)
1264 {
1265         struct ceph_msg *m = con->out_msg;
1266
1267         m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1268
1269         dout("prepare_write_message_footer %p\n", con);
1270         con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1271         if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1272                 if (con->ops->sign_message)
1273                         con->ops->sign_message(m);
1274                 else
1275                         m->footer.sig = 0;
1276         } else {
1277                 m->old_footer.flags = m->footer.flags;
1278         }
1279         con->out_more = m->more_to_follow;
1280         con->out_msg_done = true;
1281 }
1282
1283 /*
1284  * Prepare headers for the next outgoing message.
1285  */
1286 static void prepare_write_message(struct ceph_connection *con)
1287 {
1288         struct ceph_msg *m;
1289         u32 crc;
1290
1291         con_out_kvec_reset(con);
1292         con->out_msg_done = false;
1293
1294         /* Sneak an ack in there first?  If we can get it into the same
1295          * TCP packet that's a good thing. */
1296         if (con->in_seq > con->in_seq_acked) {
1297                 con->in_seq_acked = con->in_seq;
1298                 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1299                 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1300                 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1301                         &con->out_temp_ack);
1302         }
1303
1304         BUG_ON(list_empty(&con->out_queue));
1305         m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1306         con->out_msg = m;
1307         BUG_ON(m->con != con);
1308
1309         /* put message on sent list */
1310         ceph_msg_get(m);
1311         list_move_tail(&m->list_head, &con->out_sent);
1312
1313         /*
1314          * only assign outgoing seq # if we haven't sent this message
1315          * yet.  if it is requeued, resend with it's original seq.
1316          */
1317         if (m->needs_out_seq) {
1318                 m->hdr.seq = cpu_to_le64(++con->out_seq);
1319                 m->needs_out_seq = false;
1320
1321                 if (con->ops->reencode_message)
1322                         con->ops->reencode_message(m);
1323         }
1324
1325         dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1326              m, con->out_seq, le16_to_cpu(m->hdr.type),
1327              le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1328              m->data_length);
1329         WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1330         WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1331
1332         /* tag + hdr + front + middle */
1333         con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1334         con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1335         con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1336
1337         if (m->middle)
1338                 con_out_kvec_add(con, m->middle->vec.iov_len,
1339                         m->middle->vec.iov_base);
1340
1341         /* fill in hdr crc and finalize hdr */
1342         crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1343         con->out_msg->hdr.crc = cpu_to_le32(crc);
1344         memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1345
1346         /* fill in front and middle crc, footer */
1347         crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1348         con->out_msg->footer.front_crc = cpu_to_le32(crc);
1349         if (m->middle) {
1350                 crc = crc32c(0, m->middle->vec.iov_base,
1351                                 m->middle->vec.iov_len);
1352                 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1353         } else
1354                 con->out_msg->footer.middle_crc = 0;
1355         dout("%s front_crc %u middle_crc %u\n", __func__,
1356              le32_to_cpu(con->out_msg->footer.front_crc),
1357              le32_to_cpu(con->out_msg->footer.middle_crc));
1358         con->out_msg->footer.flags = 0;
1359
1360         /* is there a data payload? */
1361         con->out_msg->footer.data_crc = 0;
1362         if (m->data_length) {
1363                 prepare_message_data(con->out_msg, m->data_length);
1364                 con->out_more = 1;  /* data + footer will follow */
1365         } else {
1366                 /* no, queue up footer too and be done */
1367                 prepare_write_message_footer(con);
1368         }
1369
1370         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1371 }
1372
1373 /*
1374  * Prepare an ack.
1375  */
1376 static void prepare_write_ack(struct ceph_connection *con)
1377 {
1378         dout("prepare_write_ack %p %llu -> %llu\n", con,
1379              con->in_seq_acked, con->in_seq);
1380         con->in_seq_acked = con->in_seq;
1381
1382         con_out_kvec_reset(con);
1383
1384         con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1385
1386         con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1387         con_out_kvec_add(con, sizeof (con->out_temp_ack),
1388                                 &con->out_temp_ack);
1389
1390         con->out_more = 1;  /* more will follow.. eventually.. */
1391         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1392 }
1393
1394 /*
1395  * Prepare to share the seq during handshake
1396  */
1397 static void prepare_write_seq(struct ceph_connection *con)
1398 {
1399         dout("prepare_write_seq %p %llu -> %llu\n", con,
1400              con->in_seq_acked, con->in_seq);
1401         con->in_seq_acked = con->in_seq;
1402
1403         con_out_kvec_reset(con);
1404
1405         con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1406         con_out_kvec_add(con, sizeof (con->out_temp_ack),
1407                          &con->out_temp_ack);
1408
1409         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1410 }
1411
1412 /*
1413  * Prepare to write keepalive byte.
1414  */
1415 static void prepare_write_keepalive(struct ceph_connection *con)
1416 {
1417         dout("prepare_write_keepalive %p\n", con);
1418         con_out_kvec_reset(con);
1419         if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1420                 struct timespec64 now;
1421
1422                 ktime_get_real_ts64(&now);
1423                 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1424                 ceph_encode_timespec64(&con->out_temp_keepalive2, &now);
1425                 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1426                                  &con->out_temp_keepalive2);
1427         } else {
1428                 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1429         }
1430         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1431 }
1432
1433 /*
1434  * Connection negotiation.
1435  */
1436
1437 static int get_connect_authorizer(struct ceph_connection *con)
1438 {
1439         struct ceph_auth_handshake *auth;
1440         int auth_proto;
1441
1442         if (!con->ops->get_authorizer) {
1443                 con->auth = NULL;
1444                 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1445                 con->out_connect.authorizer_len = 0;
1446                 return 0;
1447         }
1448
1449         auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
1450         if (IS_ERR(auth))
1451                 return PTR_ERR(auth);
1452
1453         con->auth = auth;
1454         con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1455         con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1456         return 0;
1457 }
1458
1459 /*
1460  * We connected to a peer and are saying hello.
1461  */
1462 static void prepare_write_banner(struct ceph_connection *con)
1463 {
1464         con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1465         con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1466                                         &con->msgr->my_enc_addr);
1467
1468         con->out_more = 0;
1469         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1470 }
1471
1472 static void __prepare_write_connect(struct ceph_connection *con)
1473 {
1474         con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1475         if (con->auth)
1476                 con_out_kvec_add(con, con->auth->authorizer_buf_len,
1477                                  con->auth->authorizer_buf);
1478
1479         con->out_more = 0;
1480         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1481 }
1482
1483 static int prepare_write_connect(struct ceph_connection *con)
1484 {
1485         unsigned int global_seq = get_global_seq(con->msgr, 0);
1486         int proto;
1487         int ret;
1488
1489         switch (con->peer_name.type) {
1490         case CEPH_ENTITY_TYPE_MON:
1491                 proto = CEPH_MONC_PROTOCOL;
1492                 break;
1493         case CEPH_ENTITY_TYPE_OSD:
1494                 proto = CEPH_OSDC_PROTOCOL;
1495                 break;
1496         case CEPH_ENTITY_TYPE_MDS:
1497                 proto = CEPH_MDSC_PROTOCOL;
1498                 break;
1499         default:
1500                 BUG();
1501         }
1502
1503         dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1504              con->connect_seq, global_seq, proto);
1505
1506         con->out_connect.features =
1507             cpu_to_le64(from_msgr(con->msgr)->supported_features);
1508         con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1509         con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1510         con->out_connect.global_seq = cpu_to_le32(global_seq);
1511         con->out_connect.protocol_version = cpu_to_le32(proto);
1512         con->out_connect.flags = 0;
1513
1514         ret = get_connect_authorizer(con);
1515         if (ret)
1516                 return ret;
1517
1518         __prepare_write_connect(con);
1519         return 0;
1520 }
1521
1522 /*
1523  * write as much of pending kvecs to the socket as we can.
1524  *  1 -> done
1525  *  0 -> socket full, but more to do
1526  * <0 -> error
1527  */
1528 static int write_partial_kvec(struct ceph_connection *con)
1529 {
1530         int ret;
1531
1532         dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1533         while (con->out_kvec_bytes > 0) {
1534                 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1535                                        con->out_kvec_left, con->out_kvec_bytes,
1536                                        con->out_more);
1537                 if (ret <= 0)
1538                         goto out;
1539                 con->out_kvec_bytes -= ret;
1540                 if (con->out_kvec_bytes == 0)
1541                         break;            /* done */
1542
1543                 /* account for full iov entries consumed */
1544                 while (ret >= con->out_kvec_cur->iov_len) {
1545                         BUG_ON(!con->out_kvec_left);
1546                         ret -= con->out_kvec_cur->iov_len;
1547                         con->out_kvec_cur++;
1548                         con->out_kvec_left--;
1549                 }
1550                 /* and for a partially-consumed entry */
1551                 if (ret) {
1552                         con->out_kvec_cur->iov_len -= ret;
1553                         con->out_kvec_cur->iov_base += ret;
1554                 }
1555         }
1556         con->out_kvec_left = 0;
1557         ret = 1;
1558 out:
1559         dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1560              con->out_kvec_bytes, con->out_kvec_left, ret);
1561         return ret;  /* done! */
1562 }
1563
1564 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1565                                 unsigned int page_offset,
1566                                 unsigned int length)
1567 {
1568         char *kaddr;
1569
1570         kaddr = kmap(page);
1571         BUG_ON(kaddr == NULL);
1572         crc = crc32c(crc, kaddr + page_offset, length);
1573         kunmap(page);
1574
1575         return crc;
1576 }
1577 /*
1578  * Write as much message data payload as we can.  If we finish, queue
1579  * up the footer.
1580  *  1 -> done, footer is now queued in out_kvec[].
1581  *  0 -> socket full, but more to do
1582  * <0 -> error
1583  */
1584 static int write_partial_message_data(struct ceph_connection *con)
1585 {
1586         struct ceph_msg *msg = con->out_msg;
1587         struct ceph_msg_data_cursor *cursor = &msg->cursor;
1588         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1589         u32 crc;
1590
1591         dout("%s %p msg %p\n", __func__, con, msg);
1592
1593         if (list_empty(&msg->data))
1594                 return -EINVAL;
1595
1596         /*
1597          * Iterate through each page that contains data to be
1598          * written, and send as much as possible for each.
1599          *
1600          * If we are calculating the data crc (the default), we will
1601          * need to map the page.  If we have no pages, they have
1602          * been revoked, so use the zero page.
1603          */
1604         crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1605         while (cursor->total_resid) {
1606                 struct page *page;
1607                 size_t page_offset;
1608                 size_t length;
1609                 bool last_piece;
1610                 int ret;
1611
1612                 if (!cursor->resid) {
1613                         ceph_msg_data_advance(cursor, 0);
1614                         continue;
1615                 }
1616
1617                 page = ceph_msg_data_next(cursor, &page_offset, &length,
1618                                           &last_piece);
1619                 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1620                                         length, !last_piece);
1621                 if (ret <= 0) {
1622                         if (do_datacrc)
1623                                 msg->footer.data_crc = cpu_to_le32(crc);
1624
1625                         return ret;
1626                 }
1627                 if (do_datacrc && cursor->need_crc)
1628                         crc = ceph_crc32c_page(crc, page, page_offset, length);
1629                 ceph_msg_data_advance(cursor, (size_t)ret);
1630         }
1631
1632         dout("%s %p msg %p done\n", __func__, con, msg);
1633
1634         /* prepare and queue up footer, too */
1635         if (do_datacrc)
1636                 msg->footer.data_crc = cpu_to_le32(crc);
1637         else
1638                 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1639         con_out_kvec_reset(con);
1640         prepare_write_message_footer(con);
1641
1642         return 1;       /* must return > 0 to indicate success */
1643 }
1644
1645 /*
1646  * write some zeros
1647  */
1648 static int write_partial_skip(struct ceph_connection *con)
1649 {
1650         int ret;
1651
1652         dout("%s %p %d left\n", __func__, con, con->out_skip);
1653         while (con->out_skip > 0) {
1654                 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1655
1656                 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1657                 if (ret <= 0)
1658                         goto out;
1659                 con->out_skip -= ret;
1660         }
1661         ret = 1;
1662 out:
1663         return ret;
1664 }
1665
1666 /*
1667  * Prepare to read connection handshake, or an ack.
1668  */
1669 static void prepare_read_banner(struct ceph_connection *con)
1670 {
1671         dout("prepare_read_banner %p\n", con);
1672         con->in_base_pos = 0;
1673 }
1674
1675 static void prepare_read_connect(struct ceph_connection *con)
1676 {
1677         dout("prepare_read_connect %p\n", con);
1678         con->in_base_pos = 0;
1679 }
1680
1681 static void prepare_read_ack(struct ceph_connection *con)
1682 {
1683         dout("prepare_read_ack %p\n", con);
1684         con->in_base_pos = 0;
1685 }
1686
1687 static void prepare_read_seq(struct ceph_connection *con)
1688 {
1689         dout("prepare_read_seq %p\n", con);
1690         con->in_base_pos = 0;
1691         con->in_tag = CEPH_MSGR_TAG_SEQ;
1692 }
1693
1694 static void prepare_read_tag(struct ceph_connection *con)
1695 {
1696         dout("prepare_read_tag %p\n", con);
1697         con->in_base_pos = 0;
1698         con->in_tag = CEPH_MSGR_TAG_READY;
1699 }
1700
1701 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1702 {
1703         dout("prepare_read_keepalive_ack %p\n", con);
1704         con->in_base_pos = 0;
1705 }
1706
1707 /*
1708  * Prepare to read a message.
1709  */
1710 static int prepare_read_message(struct ceph_connection *con)
1711 {
1712         dout("prepare_read_message %p\n", con);
1713         BUG_ON(con->in_msg != NULL);
1714         con->in_base_pos = 0;
1715         con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1716         return 0;
1717 }
1718
1719
1720 static int read_partial(struct ceph_connection *con,
1721                         int end, int size, void *object)
1722 {
1723         while (con->in_base_pos < end) {
1724                 int left = end - con->in_base_pos;
1725                 int have = size - left;
1726                 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1727                 if (ret <= 0)
1728                         return ret;
1729                 con->in_base_pos += ret;
1730         }
1731         return 1;
1732 }
1733
1734
1735 /*
1736  * Read all or part of the connect-side handshake on a new connection
1737  */
1738 static int read_partial_banner(struct ceph_connection *con)
1739 {
1740         int size;
1741         int end;
1742         int ret;
1743
1744         dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1745
1746         /* peer's banner */
1747         size = strlen(CEPH_BANNER);
1748         end = size;
1749         ret = read_partial(con, end, size, con->in_banner);
1750         if (ret <= 0)
1751                 goto out;
1752
1753         size = sizeof (con->actual_peer_addr);
1754         end += size;
1755         ret = read_partial(con, end, size, &con->actual_peer_addr);
1756         if (ret <= 0)
1757                 goto out;
1758
1759         size = sizeof (con->peer_addr_for_me);
1760         end += size;
1761         ret = read_partial(con, end, size, &con->peer_addr_for_me);
1762         if (ret <= 0)
1763                 goto out;
1764
1765 out:
1766         return ret;
1767 }
1768
1769 static int read_partial_connect(struct ceph_connection *con)
1770 {
1771         int size;
1772         int end;
1773         int ret;
1774
1775         dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1776
1777         size = sizeof (con->in_reply);
1778         end = size;
1779         ret = read_partial(con, end, size, &con->in_reply);
1780         if (ret <= 0)
1781                 goto out;
1782
1783         if (con->auth) {
1784                 size = le32_to_cpu(con->in_reply.authorizer_len);
1785                 if (size > con->auth->authorizer_reply_buf_len) {
1786                         pr_err("authorizer reply too big: %d > %zu\n", size,
1787                                con->auth->authorizer_reply_buf_len);
1788                         ret = -EINVAL;
1789                         goto out;
1790                 }
1791
1792                 end += size;
1793                 ret = read_partial(con, end, size,
1794                                    con->auth->authorizer_reply_buf);
1795                 if (ret <= 0)
1796                         goto out;
1797         }
1798
1799         dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1800              con, (int)con->in_reply.tag,
1801              le32_to_cpu(con->in_reply.connect_seq),
1802              le32_to_cpu(con->in_reply.global_seq));
1803 out:
1804         return ret;
1805 }
1806
1807 /*
1808  * Verify the hello banner looks okay.
1809  */
1810 static int verify_hello(struct ceph_connection *con)
1811 {
1812         if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1813                 pr_err("connect to %s got bad banner\n",
1814                        ceph_pr_addr(&con->peer_addr.in_addr));
1815                 con->error_msg = "protocol error, bad banner";
1816                 return -1;
1817         }
1818         return 0;
1819 }
1820
1821 static bool addr_is_blank(struct sockaddr_storage *ss)
1822 {
1823         struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1824         struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1825
1826         switch (ss->ss_family) {
1827         case AF_INET:
1828                 return addr->s_addr == htonl(INADDR_ANY);
1829         case AF_INET6:
1830                 return ipv6_addr_any(addr6);
1831         default:
1832                 return true;
1833         }
1834 }
1835
1836 static int addr_port(struct sockaddr_storage *ss)
1837 {
1838         switch (ss->ss_family) {
1839         case AF_INET:
1840                 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1841         case AF_INET6:
1842                 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1843         }
1844         return 0;
1845 }
1846
1847 static void addr_set_port(struct sockaddr_storage *ss, int p)
1848 {
1849         switch (ss->ss_family) {
1850         case AF_INET:
1851                 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1852                 break;
1853         case AF_INET6:
1854                 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1855                 break;
1856         }
1857 }
1858
1859 /*
1860  * Unlike other *_pton function semantics, zero indicates success.
1861  */
1862 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1863                 char delim, const char **ipend)
1864 {
1865         struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1866         struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1867
1868         memset(ss, 0, sizeof(*ss));
1869
1870         if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1871                 ss->ss_family = AF_INET;
1872                 return 0;
1873         }
1874
1875         if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1876                 ss->ss_family = AF_INET6;
1877                 return 0;
1878         }
1879
1880         return -EINVAL;
1881 }
1882
1883 /*
1884  * Extract hostname string and resolve using kernel DNS facility.
1885  */
1886 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1887 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1888                 struct sockaddr_storage *ss, char delim, const char **ipend)
1889 {
1890         const char *end, *delim_p;
1891         char *colon_p, *ip_addr = NULL;
1892         int ip_len, ret;
1893
1894         /*
1895          * The end of the hostname occurs immediately preceding the delimiter or
1896          * the port marker (':') where the delimiter takes precedence.
1897          */
1898         delim_p = memchr(name, delim, namelen);
1899         colon_p = memchr(name, ':', namelen);
1900
1901         if (delim_p && colon_p)
1902                 end = delim_p < colon_p ? delim_p : colon_p;
1903         else if (!delim_p && colon_p)
1904                 end = colon_p;
1905         else {
1906                 end = delim_p;
1907                 if (!end) /* case: hostname:/ */
1908                         end = name + namelen;
1909         }
1910
1911         if (end <= name)
1912                 return -EINVAL;
1913
1914         /* do dns_resolve upcall */
1915         ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1916         if (ip_len > 0)
1917                 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1918         else
1919                 ret = -ESRCH;
1920
1921         kfree(ip_addr);
1922
1923         *ipend = end;
1924
1925         pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1926                         ret, ret ? "failed" : ceph_pr_addr(ss));
1927
1928         return ret;
1929 }
1930 #else
1931 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1932                 struct sockaddr_storage *ss, char delim, const char **ipend)
1933 {
1934         return -EINVAL;
1935 }
1936 #endif
1937
1938 /*
1939  * Parse a server name (IP or hostname). If a valid IP address is not found
1940  * then try to extract a hostname to resolve using userspace DNS upcall.
1941  */
1942 static int ceph_parse_server_name(const char *name, size_t namelen,
1943                         struct sockaddr_storage *ss, char delim, const char **ipend)
1944 {
1945         int ret;
1946
1947         ret = ceph_pton(name, namelen, ss, delim, ipend);
1948         if (ret)
1949                 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1950
1951         return ret;
1952 }
1953
1954 /*
1955  * Parse an ip[:port] list into an addr array.  Use the default
1956  * monitor port if a port isn't specified.
1957  */
1958 int ceph_parse_ips(const char *c, const char *end,
1959                    struct ceph_entity_addr *addr,
1960                    int max_count, int *count)
1961 {
1962         int i, ret = -EINVAL;
1963         const char *p = c;
1964
1965         dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1966         for (i = 0; i < max_count; i++) {
1967                 const char *ipend;
1968                 struct sockaddr_storage *ss = &addr[i].in_addr;
1969                 int port;
1970                 char delim = ',';
1971
1972                 if (*p == '[') {
1973                         delim = ']';
1974                         p++;
1975                 }
1976
1977                 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1978                 if (ret)
1979                         goto bad;
1980                 ret = -EINVAL;
1981
1982                 p = ipend;
1983
1984                 if (delim == ']') {
1985                         if (*p != ']') {
1986                                 dout("missing matching ']'\n");
1987                                 goto bad;
1988                         }
1989                         p++;
1990                 }
1991
1992                 /* port? */
1993                 if (p < end && *p == ':') {
1994                         port = 0;
1995                         p++;
1996                         while (p < end && *p >= '0' && *p <= '9') {
1997                                 port = (port * 10) + (*p - '0');
1998                                 p++;
1999                         }
2000                         if (port == 0)
2001                                 port = CEPH_MON_PORT;
2002                         else if (port > 65535)
2003                                 goto bad;
2004                 } else {
2005                         port = CEPH_MON_PORT;
2006                 }
2007
2008                 addr_set_port(ss, port);
2009
2010                 dout("parse_ips got %s\n", ceph_pr_addr(ss));
2011
2012                 if (p == end)
2013                         break;
2014                 if (*p != ',')
2015                         goto bad;
2016                 p++;
2017         }
2018
2019         if (p != end)
2020                 goto bad;
2021
2022         if (count)
2023                 *count = i + 1;
2024         return 0;
2025
2026 bad:
2027         pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
2028         return ret;
2029 }
2030 EXPORT_SYMBOL(ceph_parse_ips);
2031
2032 static int process_banner(struct ceph_connection *con)
2033 {
2034         dout("process_banner on %p\n", con);
2035
2036         if (verify_hello(con) < 0)
2037                 return -1;
2038
2039         ceph_decode_addr(&con->actual_peer_addr);
2040         ceph_decode_addr(&con->peer_addr_for_me);
2041
2042         /*
2043          * Make sure the other end is who we wanted.  note that the other
2044          * end may not yet know their ip address, so if it's 0.0.0.0, give
2045          * them the benefit of the doubt.
2046          */
2047         if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2048                    sizeof(con->peer_addr)) != 0 &&
2049             !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2050               con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2051                 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2052                         ceph_pr_addr(&con->peer_addr.in_addr),
2053                         (int)le32_to_cpu(con->peer_addr.nonce),
2054                         ceph_pr_addr(&con->actual_peer_addr.in_addr),
2055                         (int)le32_to_cpu(con->actual_peer_addr.nonce));
2056                 con->error_msg = "wrong peer at address";
2057                 return -1;
2058         }
2059
2060         /*
2061          * did we learn our address?
2062          */
2063         if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2064                 int port = addr_port(&con->msgr->inst.addr.in_addr);
2065
2066                 memcpy(&con->msgr->inst.addr.in_addr,
2067                        &con->peer_addr_for_me.in_addr,
2068                        sizeof(con->peer_addr_for_me.in_addr));
2069                 addr_set_port(&con->msgr->inst.addr.in_addr, port);
2070                 encode_my_addr(con->msgr);
2071                 dout("process_banner learned my addr is %s\n",
2072                      ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2073         }
2074
2075         return 0;
2076 }
2077
2078 static int process_connect(struct ceph_connection *con)
2079 {
2080         u64 sup_feat = from_msgr(con->msgr)->supported_features;
2081         u64 req_feat = from_msgr(con->msgr)->required_features;
2082         u64 server_feat = le64_to_cpu(con->in_reply.features);
2083         int ret;
2084
2085         dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2086
2087         if (con->auth) {
2088                 /*
2089                  * Any connection that defines ->get_authorizer()
2090                  * should also define ->add_authorizer_challenge() and
2091                  * ->verify_authorizer_reply().
2092                  *
2093                  * See get_connect_authorizer().
2094                  */
2095                 if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
2096                         ret = con->ops->add_authorizer_challenge(
2097                                     con, con->auth->authorizer_reply_buf,
2098                                     le32_to_cpu(con->in_reply.authorizer_len));
2099                         if (ret < 0)
2100                                 return ret;
2101
2102                         con_out_kvec_reset(con);
2103                         __prepare_write_connect(con);
2104                         prepare_read_connect(con);
2105                         return 0;
2106                 }
2107
2108                 ret = con->ops->verify_authorizer_reply(con);
2109                 if (ret < 0) {
2110                         con->error_msg = "bad authorize reply";
2111                         return ret;
2112                 }
2113         }
2114
2115         switch (con->in_reply.tag) {
2116         case CEPH_MSGR_TAG_FEATURES:
2117                 pr_err("%s%lld %s feature set mismatch,"
2118                        " my %llx < server's %llx, missing %llx\n",
2119                        ENTITY_NAME(con->peer_name),
2120                        ceph_pr_addr(&con->peer_addr.in_addr),
2121                        sup_feat, server_feat, server_feat & ~sup_feat);
2122                 con->error_msg = "missing required protocol features";
2123                 reset_connection(con);
2124                 return -1;
2125
2126         case CEPH_MSGR_TAG_BADPROTOVER:
2127                 pr_err("%s%lld %s protocol version mismatch,"
2128                        " my %d != server's %d\n",
2129                        ENTITY_NAME(con->peer_name),
2130                        ceph_pr_addr(&con->peer_addr.in_addr),
2131                        le32_to_cpu(con->out_connect.protocol_version),
2132                        le32_to_cpu(con->in_reply.protocol_version));
2133                 con->error_msg = "protocol version mismatch";
2134                 reset_connection(con);
2135                 return -1;
2136
2137         case CEPH_MSGR_TAG_BADAUTHORIZER:
2138                 con->auth_retry++;
2139                 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2140                      con->auth_retry);
2141                 if (con->auth_retry == 2) {
2142                         con->error_msg = "connect authorization failure";
2143                         return -1;
2144                 }
2145                 con_out_kvec_reset(con);
2146                 ret = prepare_write_connect(con);
2147                 if (ret < 0)
2148                         return ret;
2149                 prepare_read_connect(con);
2150                 break;
2151
2152         case CEPH_MSGR_TAG_RESETSESSION:
2153                 /*
2154                  * If we connected with a large connect_seq but the peer
2155                  * has no record of a session with us (no connection, or
2156                  * connect_seq == 0), they will send RESETSESION to indicate
2157                  * that they must have reset their session, and may have
2158                  * dropped messages.
2159                  */
2160                 dout("process_connect got RESET peer seq %u\n",
2161                      le32_to_cpu(con->in_reply.connect_seq));
2162                 pr_err("%s%lld %s connection reset\n",
2163                        ENTITY_NAME(con->peer_name),
2164                        ceph_pr_addr(&con->peer_addr.in_addr));
2165                 reset_connection(con);
2166                 con_out_kvec_reset(con);
2167                 ret = prepare_write_connect(con);
2168                 if (ret < 0)
2169                         return ret;
2170                 prepare_read_connect(con);
2171
2172                 /* Tell ceph about it. */
2173                 mutex_unlock(&con->mutex);
2174                 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2175                 if (con->ops->peer_reset)
2176                         con->ops->peer_reset(con);
2177                 mutex_lock(&con->mutex);
2178                 if (con->state != CON_STATE_NEGOTIATING)
2179                         return -EAGAIN;
2180                 break;
2181
2182         case CEPH_MSGR_TAG_RETRY_SESSION:
2183                 /*
2184                  * If we sent a smaller connect_seq than the peer has, try
2185                  * again with a larger value.
2186                  */
2187                 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2188                      le32_to_cpu(con->out_connect.connect_seq),
2189                      le32_to_cpu(con->in_reply.connect_seq));
2190                 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2191                 con_out_kvec_reset(con);
2192                 ret = prepare_write_connect(con);
2193                 if (ret < 0)
2194                         return ret;
2195                 prepare_read_connect(con);
2196                 break;
2197
2198         case CEPH_MSGR_TAG_RETRY_GLOBAL:
2199                 /*
2200                  * If we sent a smaller global_seq than the peer has, try
2201                  * again with a larger value.
2202                  */
2203                 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2204                      con->peer_global_seq,
2205                      le32_to_cpu(con->in_reply.global_seq));
2206                 get_global_seq(con->msgr,
2207                                le32_to_cpu(con->in_reply.global_seq));
2208                 con_out_kvec_reset(con);
2209                 ret = prepare_write_connect(con);
2210                 if (ret < 0)
2211                         return ret;
2212                 prepare_read_connect(con);
2213                 break;
2214
2215         case CEPH_MSGR_TAG_SEQ:
2216         case CEPH_MSGR_TAG_READY:
2217                 if (req_feat & ~server_feat) {
2218                         pr_err("%s%lld %s protocol feature mismatch,"
2219                                " my required %llx > server's %llx, need %llx\n",
2220                                ENTITY_NAME(con->peer_name),
2221                                ceph_pr_addr(&con->peer_addr.in_addr),
2222                                req_feat, server_feat, req_feat & ~server_feat);
2223                         con->error_msg = "missing required protocol features";
2224                         reset_connection(con);
2225                         return -1;
2226                 }
2227
2228                 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2229                 con->state = CON_STATE_OPEN;
2230                 con->auth_retry = 0;    /* we authenticated; clear flag */
2231                 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2232                 con->connect_seq++;
2233                 con->peer_features = server_feat;
2234                 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2235                      con->peer_global_seq,
2236                      le32_to_cpu(con->in_reply.connect_seq),
2237                      con->connect_seq);
2238                 WARN_ON(con->connect_seq !=
2239                         le32_to_cpu(con->in_reply.connect_seq));
2240
2241                 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2242                         con_flag_set(con, CON_FLAG_LOSSYTX);
2243
2244                 con->delay = 0;      /* reset backoff memory */
2245
2246                 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2247                         prepare_write_seq(con);
2248                         prepare_read_seq(con);
2249                 } else {
2250                         prepare_read_tag(con);
2251                 }
2252                 break;
2253
2254         case CEPH_MSGR_TAG_WAIT:
2255                 /*
2256                  * If there is a connection race (we are opening
2257                  * connections to each other), one of us may just have
2258                  * to WAIT.  This shouldn't happen if we are the
2259                  * client.
2260                  */
2261                 con->error_msg = "protocol error, got WAIT as client";
2262                 return -1;
2263
2264         default:
2265                 con->error_msg = "protocol error, garbage tag during connect";
2266                 return -1;
2267         }
2268         return 0;
2269 }
2270
2271
2272 /*
2273  * read (part of) an ack
2274  */
2275 static int read_partial_ack(struct ceph_connection *con)
2276 {
2277         int size = sizeof (con->in_temp_ack);
2278         int end = size;
2279
2280         return read_partial(con, end, size, &con->in_temp_ack);
2281 }
2282
2283 /*
2284  * We can finally discard anything that's been acked.
2285  */
2286 static void process_ack(struct ceph_connection *con)
2287 {
2288         struct ceph_msg *m;
2289         u64 ack = le64_to_cpu(con->in_temp_ack);
2290         u64 seq;
2291         bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2292         struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
2293
2294         /*
2295          * In the reconnect case, con_fault() has requeued messages
2296          * in out_sent. We should cleanup old messages according to
2297          * the reconnect seq.
2298          */
2299         while (!list_empty(list)) {
2300                 m = list_first_entry(list, struct ceph_msg, list_head);
2301                 if (reconnect && m->needs_out_seq)
2302                         break;
2303                 seq = le64_to_cpu(m->hdr.seq);
2304                 if (seq > ack)
2305                         break;
2306                 dout("got ack for seq %llu type %d at %p\n", seq,
2307                      le16_to_cpu(m->hdr.type), m);
2308                 m->ack_stamp = jiffies;
2309                 ceph_msg_remove(m);
2310         }
2311
2312         prepare_read_tag(con);
2313 }
2314
2315
2316 static int read_partial_message_section(struct ceph_connection *con,
2317                                         struct kvec *section,
2318                                         unsigned int sec_len, u32 *crc)
2319 {
2320         int ret, left;
2321
2322         BUG_ON(!section);
2323
2324         while (section->iov_len < sec_len) {
2325                 BUG_ON(section->iov_base == NULL);
2326                 left = sec_len - section->iov_len;
2327                 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2328                                        section->iov_len, left);
2329                 if (ret <= 0)
2330                         return ret;
2331                 section->iov_len += ret;
2332         }
2333         if (section->iov_len == sec_len)
2334                 *crc = crc32c(0, section->iov_base, section->iov_len);
2335
2336         return 1;
2337 }
2338
2339 static int read_partial_msg_data(struct ceph_connection *con)
2340 {
2341         struct ceph_msg *msg = con->in_msg;
2342         struct ceph_msg_data_cursor *cursor = &msg->cursor;
2343         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2344         struct page *page;
2345         size_t page_offset;
2346         size_t length;
2347         u32 crc = 0;
2348         int ret;
2349
2350         BUG_ON(!msg);
2351         if (list_empty(&msg->data))
2352                 return -EIO;
2353
2354         if (do_datacrc)
2355                 crc = con->in_data_crc;
2356         while (cursor->total_resid) {
2357                 if (!cursor->resid) {
2358                         ceph_msg_data_advance(cursor, 0);
2359                         continue;
2360                 }
2361
2362                 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2363                 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2364                 if (ret <= 0) {
2365                         if (do_datacrc)
2366                                 con->in_data_crc = crc;
2367
2368                         return ret;
2369                 }
2370
2371                 if (do_datacrc)
2372                         crc = ceph_crc32c_page(crc, page, page_offset, ret);
2373                 ceph_msg_data_advance(cursor, (size_t)ret);
2374         }
2375         if (do_datacrc)
2376                 con->in_data_crc = crc;
2377
2378         return 1;       /* must return > 0 to indicate success */
2379 }
2380
2381 /*
2382  * read (part of) a message.
2383  */
2384 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2385
2386 static int read_partial_message(struct ceph_connection *con)
2387 {
2388         struct ceph_msg *m = con->in_msg;
2389         int size;
2390         int end;
2391         int ret;
2392         unsigned int front_len, middle_len, data_len;
2393         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2394         bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2395         u64 seq;
2396         u32 crc;
2397
2398         dout("read_partial_message con %p msg %p\n", con, m);
2399
2400         /* header */
2401         size = sizeof (con->in_hdr);
2402         end = size;
2403         ret = read_partial(con, end, size, &con->in_hdr);
2404         if (ret <= 0)
2405                 return ret;
2406
2407         crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2408         if (cpu_to_le32(crc) != con->in_hdr.crc) {
2409                 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2410                        crc, con->in_hdr.crc);
2411                 return -EBADMSG;
2412         }
2413
2414         front_len = le32_to_cpu(con->in_hdr.front_len);
2415         if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2416                 return -EIO;
2417         middle_len = le32_to_cpu(con->in_hdr.middle_len);
2418         if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2419                 return -EIO;
2420         data_len = le32_to_cpu(con->in_hdr.data_len);
2421         if (data_len > CEPH_MSG_MAX_DATA_LEN)
2422                 return -EIO;
2423
2424         /* verify seq# */
2425         seq = le64_to_cpu(con->in_hdr.seq);
2426         if ((s64)seq - (s64)con->in_seq < 1) {
2427                 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2428                         ENTITY_NAME(con->peer_name),
2429                         ceph_pr_addr(&con->peer_addr.in_addr),
2430                         seq, con->in_seq + 1);
2431                 con->in_base_pos = -front_len - middle_len - data_len -
2432                         sizeof_footer(con);
2433                 con->in_tag = CEPH_MSGR_TAG_READY;
2434                 return 1;
2435         } else if ((s64)seq - (s64)con->in_seq > 1) {
2436                 pr_err("read_partial_message bad seq %lld expected %lld\n",
2437                        seq, con->in_seq + 1);
2438                 con->error_msg = "bad message sequence # for incoming message";
2439                 return -EBADE;
2440         }
2441
2442         /* allocate message? */
2443         if (!con->in_msg) {
2444                 int skip = 0;
2445
2446                 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2447                      front_len, data_len);
2448                 ret = ceph_con_in_msg_alloc(con, &skip);
2449                 if (ret < 0)
2450                         return ret;
2451
2452                 BUG_ON(!con->in_msg ^ skip);
2453                 if (skip) {
2454                         /* skip this message */
2455                         dout("alloc_msg said skip message\n");
2456                         con->in_base_pos = -front_len - middle_len - data_len -
2457                                 sizeof_footer(con);
2458                         con->in_tag = CEPH_MSGR_TAG_READY;
2459                         con->in_seq++;
2460                         return 1;
2461                 }
2462
2463                 BUG_ON(!con->in_msg);
2464                 BUG_ON(con->in_msg->con != con);
2465                 m = con->in_msg;
2466                 m->front.iov_len = 0;    /* haven't read it yet */
2467                 if (m->middle)
2468                         m->middle->vec.iov_len = 0;
2469
2470                 /* prepare for data payload, if any */
2471
2472                 if (data_len)
2473                         prepare_message_data(con->in_msg, data_len);
2474         }
2475
2476         /* front */
2477         ret = read_partial_message_section(con, &m->front, front_len,
2478                                            &con->in_front_crc);
2479         if (ret <= 0)
2480                 return ret;
2481
2482         /* middle */
2483         if (m->middle) {
2484                 ret = read_partial_message_section(con, &m->middle->vec,
2485                                                    middle_len,
2486                                                    &con->in_middle_crc);
2487                 if (ret <= 0)
2488                         return ret;
2489         }
2490
2491         /* (page) data */
2492         if (data_len) {
2493                 ret = read_partial_msg_data(con);
2494                 if (ret <= 0)
2495                         return ret;
2496         }
2497
2498         /* footer */
2499         size = sizeof_footer(con);
2500         end += size;
2501         ret = read_partial(con, end, size, &m->footer);
2502         if (ret <= 0)
2503                 return ret;
2504
2505         if (!need_sign) {
2506                 m->footer.flags = m->old_footer.flags;
2507                 m->footer.sig = 0;
2508         }
2509
2510         dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2511              m, front_len, m->footer.front_crc, middle_len,
2512              m->footer.middle_crc, data_len, m->footer.data_crc);
2513
2514         /* crc ok? */
2515         if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2516                 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2517                        m, con->in_front_crc, m->footer.front_crc);
2518                 return -EBADMSG;
2519         }
2520         if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2521                 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2522                        m, con->in_middle_crc, m->footer.middle_crc);
2523                 return -EBADMSG;
2524         }
2525         if (do_datacrc &&
2526             (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2527             con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2528                 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2529                        con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2530                 return -EBADMSG;
2531         }
2532
2533         if (need_sign && con->ops->check_message_signature &&
2534             con->ops->check_message_signature(m)) {
2535                 pr_err("read_partial_message %p signature check failed\n", m);
2536                 return -EBADMSG;
2537         }
2538
2539         return 1; /* done! */
2540 }
2541
2542 /*
2543  * Process message.  This happens in the worker thread.  The callback should
2544  * be careful not to do anything that waits on other incoming messages or it
2545  * may deadlock.
2546  */
2547 static void process_message(struct ceph_connection *con)
2548 {
2549         struct ceph_msg *msg = con->in_msg;
2550
2551         BUG_ON(con->in_msg->con != con);
2552         con->in_msg = NULL;
2553
2554         /* if first message, set peer_name */
2555         if (con->peer_name.type == 0)
2556                 con->peer_name = msg->hdr.src;
2557
2558         con->in_seq++;
2559         mutex_unlock(&con->mutex);
2560
2561         dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2562              msg, le64_to_cpu(msg->hdr.seq),
2563              ENTITY_NAME(msg->hdr.src),
2564              le16_to_cpu(msg->hdr.type),
2565              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2566              le32_to_cpu(msg->hdr.front_len),
2567              le32_to_cpu(msg->hdr.data_len),
2568              con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2569         con->ops->dispatch(con, msg);
2570
2571         mutex_lock(&con->mutex);
2572 }
2573
2574 static int read_keepalive_ack(struct ceph_connection *con)
2575 {
2576         struct ceph_timespec ceph_ts;
2577         size_t size = sizeof(ceph_ts);
2578         int ret = read_partial(con, size, size, &ceph_ts);
2579         if (ret <= 0)
2580                 return ret;
2581         ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
2582         prepare_read_tag(con);
2583         return 1;
2584 }
2585
2586 /*
2587  * Write something to the socket.  Called in a worker thread when the
2588  * socket appears to be writeable and we have something ready to send.
2589  */
2590 static int try_write(struct ceph_connection *con)
2591 {
2592         int ret = 1;
2593
2594         dout("try_write start %p state %lu\n", con, con->state);
2595         if (con->state != CON_STATE_PREOPEN &&
2596             con->state != CON_STATE_CONNECTING &&
2597             con->state != CON_STATE_NEGOTIATING &&
2598             con->state != CON_STATE_OPEN)
2599                 return 0;
2600
2601         /* open the socket first? */
2602         if (con->state == CON_STATE_PREOPEN) {
2603                 BUG_ON(con->sock);
2604                 con->state = CON_STATE_CONNECTING;
2605
2606                 con_out_kvec_reset(con);
2607                 prepare_write_banner(con);
2608                 prepare_read_banner(con);
2609
2610                 BUG_ON(con->in_msg);
2611                 con->in_tag = CEPH_MSGR_TAG_READY;
2612                 dout("try_write initiating connect on %p new state %lu\n",
2613                      con, con->state);
2614                 ret = ceph_tcp_connect(con);
2615                 if (ret < 0) {
2616                         con->error_msg = "connect error";
2617                         goto out;
2618                 }
2619         }
2620
2621 more:
2622         dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2623         BUG_ON(!con->sock);
2624
2625         /* kvec data queued? */
2626         if (con->out_kvec_left) {
2627                 ret = write_partial_kvec(con);
2628                 if (ret <= 0)
2629                         goto out;
2630         }
2631         if (con->out_skip) {
2632                 ret = write_partial_skip(con);
2633                 if (ret <= 0)
2634                         goto out;
2635         }
2636
2637         /* msg pages? */
2638         if (con->out_msg) {
2639                 if (con->out_msg_done) {
2640                         ceph_msg_put(con->out_msg);
2641                         con->out_msg = NULL;   /* we're done with this one */
2642                         goto do_next;
2643                 }
2644
2645                 ret = write_partial_message_data(con);
2646                 if (ret == 1)
2647                         goto more;  /* we need to send the footer, too! */
2648                 if (ret == 0)
2649                         goto out;
2650                 if (ret < 0) {
2651                         dout("try_write write_partial_message_data err %d\n",
2652                              ret);
2653                         goto out;
2654                 }
2655         }
2656
2657 do_next:
2658         if (con->state == CON_STATE_OPEN) {
2659                 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2660                         prepare_write_keepalive(con);
2661                         goto more;
2662                 }
2663                 /* is anything else pending? */
2664                 if (!list_empty(&con->out_queue)) {
2665                         prepare_write_message(con);
2666                         goto more;
2667                 }
2668                 if (con->in_seq > con->in_seq_acked) {
2669                         prepare_write_ack(con);
2670                         goto more;
2671                 }
2672         }
2673
2674         /* Nothing to do! */
2675         con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2676         dout("try_write nothing else to write.\n");
2677         ret = 0;
2678 out:
2679         dout("try_write done on %p ret %d\n", con, ret);
2680         return ret;
2681 }
2682
2683 /*
2684  * Read what we can from the socket.
2685  */
2686 static int try_read(struct ceph_connection *con)
2687 {
2688         int ret = -1;
2689
2690 more:
2691         dout("try_read start on %p state %lu\n", con, con->state);
2692         if (con->state != CON_STATE_CONNECTING &&
2693             con->state != CON_STATE_NEGOTIATING &&
2694             con->state != CON_STATE_OPEN)
2695                 return 0;
2696
2697         BUG_ON(!con->sock);
2698
2699         dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2700              con->in_base_pos);
2701
2702         if (con->state == CON_STATE_CONNECTING) {
2703                 dout("try_read connecting\n");
2704                 ret = read_partial_banner(con);
2705                 if (ret <= 0)
2706                         goto out;
2707                 ret = process_banner(con);
2708                 if (ret < 0)
2709                         goto out;
2710
2711                 con->state = CON_STATE_NEGOTIATING;
2712
2713                 /*
2714                  * Received banner is good, exchange connection info.
2715                  * Do not reset out_kvec, as sending our banner raced
2716                  * with receiving peer banner after connect completed.
2717                  */
2718                 ret = prepare_write_connect(con);
2719                 if (ret < 0)
2720                         goto out;
2721                 prepare_read_connect(con);
2722
2723                 /* Send connection info before awaiting response */
2724                 goto out;
2725         }
2726
2727         if (con->state == CON_STATE_NEGOTIATING) {
2728                 dout("try_read negotiating\n");
2729                 ret = read_partial_connect(con);
2730                 if (ret <= 0)
2731                         goto out;
2732                 ret = process_connect(con);
2733                 if (ret < 0)
2734                         goto out;
2735                 goto more;
2736         }
2737
2738         WARN_ON(con->state != CON_STATE_OPEN);
2739
2740         if (con->in_base_pos < 0) {
2741                 /*
2742                  * skipping + discarding content.
2743                  */
2744                 ret = ceph_tcp_recvmsg(con->sock, NULL, -con->in_base_pos);
2745                 if (ret <= 0)
2746                         goto out;
2747                 dout("skipped %d / %d bytes\n", ret, -con->in_base_pos);
2748                 con->in_base_pos += ret;
2749                 if (con->in_base_pos)
2750                         goto more;
2751         }
2752         if (con->in_tag == CEPH_MSGR_TAG_READY) {
2753                 /*
2754                  * what's next?
2755                  */
2756                 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2757                 if (ret <= 0)
2758                         goto out;
2759                 dout("try_read got tag %d\n", (int)con->in_tag);
2760                 switch (con->in_tag) {
2761                 case CEPH_MSGR_TAG_MSG:
2762                         prepare_read_message(con);
2763                         break;
2764                 case CEPH_MSGR_TAG_ACK:
2765                         prepare_read_ack(con);
2766                         break;
2767                 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2768                         prepare_read_keepalive_ack(con);
2769                         break;
2770                 case CEPH_MSGR_TAG_CLOSE:
2771                         con_close_socket(con);
2772                         con->state = CON_STATE_CLOSED;
2773                         goto out;
2774                 default:
2775                         goto bad_tag;
2776                 }
2777         }
2778         if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2779                 ret = read_partial_message(con);
2780                 if (ret <= 0) {
2781                         switch (ret) {
2782                         case -EBADMSG:
2783                                 con->error_msg = "bad crc/signature";
2784                                 /* fall through */
2785                         case -EBADE:
2786                                 ret = -EIO;
2787                                 break;
2788                         case -EIO:
2789                                 con->error_msg = "io error";
2790                                 break;
2791                         }
2792                         goto out;
2793                 }
2794                 if (con->in_tag == CEPH_MSGR_TAG_READY)
2795                         goto more;
2796                 process_message(con);
2797                 if (con->state == CON_STATE_OPEN)
2798                         prepare_read_tag(con);
2799                 goto more;
2800         }
2801         if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2802             con->in_tag == CEPH_MSGR_TAG_SEQ) {
2803                 /*
2804                  * the final handshake seq exchange is semantically
2805                  * equivalent to an ACK
2806                  */
2807                 ret = read_partial_ack(con);
2808                 if (ret <= 0)
2809                         goto out;
2810                 process_ack(con);
2811                 goto more;
2812         }
2813         if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2814                 ret = read_keepalive_ack(con);
2815                 if (ret <= 0)
2816                         goto out;
2817                 goto more;
2818         }
2819
2820 out:
2821         dout("try_read done on %p ret %d\n", con, ret);
2822         return ret;
2823
2824 bad_tag:
2825         pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2826         con->error_msg = "protocol error, garbage tag";
2827         ret = -1;
2828         goto out;
2829 }
2830
2831
2832 /*
2833  * Atomically queue work on a connection after the specified delay.
2834  * Bump @con reference to avoid races with connection teardown.
2835  * Returns 0 if work was queued, or an error code otherwise.
2836  */
2837 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2838 {
2839         if (!con->ops->get(con)) {
2840                 dout("%s %p ref count 0\n", __func__, con);
2841                 return -ENOENT;
2842         }
2843
2844         if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2845                 dout("%s %p - already queued\n", __func__, con);
2846                 con->ops->put(con);
2847                 return -EBUSY;
2848         }
2849
2850         dout("%s %p %lu\n", __func__, con, delay);
2851         return 0;
2852 }
2853
2854 static void queue_con(struct ceph_connection *con)
2855 {
2856         (void) queue_con_delay(con, 0);
2857 }
2858
2859 static void cancel_con(struct ceph_connection *con)
2860 {
2861         if (cancel_delayed_work(&con->work)) {
2862                 dout("%s %p\n", __func__, con);
2863                 con->ops->put(con);
2864         }
2865 }
2866
2867 static bool con_sock_closed(struct ceph_connection *con)
2868 {
2869         if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2870                 return false;
2871
2872 #define CASE(x)                                                         \
2873         case CON_STATE_ ## x:                                           \
2874                 con->error_msg = "socket closed (con state " #x ")";    \
2875                 break;
2876
2877         switch (con->state) {
2878         CASE(CLOSED);
2879         CASE(PREOPEN);
2880         CASE(CONNECTING);
2881         CASE(NEGOTIATING);
2882         CASE(OPEN);
2883         CASE(STANDBY);
2884         default:
2885                 pr_warn("%s con %p unrecognized state %lu\n",
2886                         __func__, con, con->state);
2887                 con->error_msg = "unrecognized con state";
2888                 BUG();
2889                 break;
2890         }
2891 #undef CASE
2892
2893         return true;
2894 }
2895
2896 static bool con_backoff(struct ceph_connection *con)
2897 {
2898         int ret;
2899
2900         if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2901                 return false;
2902
2903         ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2904         if (ret) {
2905                 dout("%s: con %p FAILED to back off %lu\n", __func__,
2906                         con, con->delay);
2907                 BUG_ON(ret == -ENOENT);
2908                 con_flag_set(con, CON_FLAG_BACKOFF);
2909         }
2910
2911         return true;
2912 }
2913
2914 /* Finish fault handling; con->mutex must *not* be held here */
2915
2916 static void con_fault_finish(struct ceph_connection *con)
2917 {
2918         dout("%s %p\n", __func__, con);
2919
2920         /*
2921          * in case we faulted due to authentication, invalidate our
2922          * current tickets so that we can get new ones.
2923          */
2924         if (con->auth_retry) {
2925                 dout("auth_retry %d, invalidating\n", con->auth_retry);
2926                 if (con->ops->invalidate_authorizer)
2927                         con->ops->invalidate_authorizer(con);
2928                 con->auth_retry = 0;
2929         }
2930
2931         if (con->ops->fault)
2932                 con->ops->fault(con);
2933 }
2934
2935 /*
2936  * Do some work on a connection.  Drop a connection ref when we're done.
2937  */
2938 static void ceph_con_workfn(struct work_struct *work)
2939 {
2940         struct ceph_connection *con = container_of(work, struct ceph_connection,
2941                                                    work.work);
2942         bool fault;
2943
2944         mutex_lock(&con->mutex);
2945         while (true) {
2946                 int ret;
2947
2948                 if ((fault = con_sock_closed(con))) {
2949                         dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2950                         break;
2951                 }
2952                 if (con_backoff(con)) {
2953                         dout("%s: con %p BACKOFF\n", __func__, con);
2954                         break;
2955                 }
2956                 if (con->state == CON_STATE_STANDBY) {
2957                         dout("%s: con %p STANDBY\n", __func__, con);
2958                         break;
2959                 }
2960                 if (con->state == CON_STATE_CLOSED) {
2961                         dout("%s: con %p CLOSED\n", __func__, con);
2962                         BUG_ON(con->sock);
2963                         break;
2964                 }
2965                 if (con->state == CON_STATE_PREOPEN) {
2966                         dout("%s: con %p PREOPEN\n", __func__, con);
2967                         BUG_ON(con->sock);
2968                 }
2969
2970                 ret = try_read(con);
2971                 if (ret < 0) {
2972                         if (ret == -EAGAIN)
2973                                 continue;
2974                         if (!con->error_msg)
2975                                 con->error_msg = "socket error on read";
2976                         fault = true;
2977                         break;
2978                 }
2979
2980                 ret = try_write(con);
2981                 if (ret < 0) {
2982                         if (ret == -EAGAIN)
2983                                 continue;
2984                         if (!con->error_msg)
2985                                 con->error_msg = "socket error on write";
2986                         fault = true;
2987                 }
2988
2989                 break;  /* If we make it to here, we're done */
2990         }
2991         if (fault)
2992                 con_fault(con);
2993         mutex_unlock(&con->mutex);
2994
2995         if (fault)
2996                 con_fault_finish(con);
2997
2998         con->ops->put(con);
2999 }
3000
3001 /*
3002  * Generic error/fault handler.  A retry mechanism is used with
3003  * exponential backoff
3004  */
3005 static void con_fault(struct ceph_connection *con)
3006 {
3007         dout("fault %p state %lu to peer %s\n",
3008              con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
3009
3010         pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
3011                 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
3012         con->error_msg = NULL;
3013
3014         WARN_ON(con->state != CON_STATE_CONNECTING &&
3015                con->state != CON_STATE_NEGOTIATING &&
3016                con->state != CON_STATE_OPEN);
3017
3018         con_close_socket(con);
3019
3020         if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
3021                 dout("fault on LOSSYTX channel, marking CLOSED\n");
3022                 con->state = CON_STATE_CLOSED;
3023                 return;
3024         }
3025
3026         if (con->in_msg) {
3027                 BUG_ON(con->in_msg->con != con);
3028                 ceph_msg_put(con->in_msg);
3029                 con->in_msg = NULL;
3030         }
3031
3032         /* Requeue anything that hasn't been acked */
3033         list_splice_init(&con->out_sent, &con->out_queue);
3034
3035         /* If there are no messages queued or keepalive pending, place
3036          * the connection in a STANDBY state */
3037         if (list_empty(&con->out_queue) &&
3038             !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
3039                 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
3040                 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
3041                 con->state = CON_STATE_STANDBY;
3042         } else {
3043                 /* retry after a delay. */
3044                 con->state = CON_STATE_PREOPEN;
3045                 if (con->delay == 0)
3046                         con->delay = BASE_DELAY_INTERVAL;
3047                 else if (con->delay < MAX_DELAY_INTERVAL)
3048                         con->delay *= 2;
3049                 con_flag_set(con, CON_FLAG_BACKOFF);
3050                 queue_con(con);
3051         }
3052 }
3053
3054
3055
3056 /*
3057  * initialize a new messenger instance
3058  */
3059 void ceph_messenger_init(struct ceph_messenger *msgr,
3060                          struct ceph_entity_addr *myaddr)
3061 {
3062         spin_lock_init(&msgr->global_seq_lock);
3063
3064         if (myaddr)
3065                 msgr->inst.addr = *myaddr;
3066
3067         /* select a random nonce */
3068         msgr->inst.addr.type = 0;
3069         get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3070         encode_my_addr(msgr);
3071
3072         atomic_set(&msgr->stopping, 0);
3073         write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3074
3075         dout("%s %p\n", __func__, msgr);
3076 }
3077 EXPORT_SYMBOL(ceph_messenger_init);
3078
3079 void ceph_messenger_fini(struct ceph_messenger *msgr)
3080 {
3081         put_net(read_pnet(&msgr->net));
3082 }
3083 EXPORT_SYMBOL(ceph_messenger_fini);
3084
3085 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3086 {
3087         if (msg->con)
3088                 msg->con->ops->put(msg->con);
3089
3090         msg->con = con ? con->ops->get(con) : NULL;
3091         BUG_ON(msg->con != con);
3092 }
3093
3094 static void clear_standby(struct ceph_connection *con)
3095 {
3096         /* come back from STANDBY? */
3097         if (con->state == CON_STATE_STANDBY) {
3098                 dout("clear_standby %p and ++connect_seq\n", con);
3099                 con->state = CON_STATE_PREOPEN;
3100                 con->connect_seq++;
3101                 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3102                 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3103         }
3104 }
3105
3106 /*
3107  * Queue up an outgoing message on the given connection.
3108  */
3109 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3110 {
3111         /* set src+dst */
3112         msg->hdr.src = con->msgr->inst.name;
3113         BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3114         msg->needs_out_seq = true;
3115
3116         mutex_lock(&con->mutex);
3117
3118         if (con->state == CON_STATE_CLOSED) {
3119                 dout("con_send %p closed, dropping %p\n", con, msg);
3120                 ceph_msg_put(msg);
3121                 mutex_unlock(&con->mutex);
3122                 return;
3123         }
3124
3125         msg_con_set(msg, con);
3126
3127         BUG_ON(!list_empty(&msg->list_head));
3128         list_add_tail(&msg->list_head, &con->out_queue);
3129         dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3130              ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3131              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3132              le32_to_cpu(msg->hdr.front_len),
3133              le32_to_cpu(msg->hdr.middle_len),
3134              le32_to_cpu(msg->hdr.data_len));
3135
3136         clear_standby(con);
3137         mutex_unlock(&con->mutex);
3138
3139         /* if there wasn't anything waiting to send before, queue
3140          * new work */
3141         if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3142                 queue_con(con);
3143 }
3144 EXPORT_SYMBOL(ceph_con_send);
3145
3146 /*
3147  * Revoke a message that was previously queued for send
3148  */
3149 void ceph_msg_revoke(struct ceph_msg *msg)
3150 {
3151         struct ceph_connection *con = msg->con;
3152
3153         if (!con) {
3154                 dout("%s msg %p null con\n", __func__, msg);
3155                 return;         /* Message not in our possession */
3156         }
3157
3158         mutex_lock(&con->mutex);
3159         if (!list_empty(&msg->list_head)) {
3160                 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3161                 list_del_init(&msg->list_head);
3162                 msg->hdr.seq = 0;
3163
3164                 ceph_msg_put(msg);
3165         }
3166         if (con->out_msg == msg) {
3167                 BUG_ON(con->out_skip);
3168                 /* footer */
3169                 if (con->out_msg_done) {
3170                         con->out_skip += con_out_kvec_skip(con);
3171                 } else {
3172                         BUG_ON(!msg->data_length);
3173                         con->out_skip += sizeof_footer(con);
3174                 }
3175                 /* data, middle, front */
3176                 if (msg->data_length)
3177                         con->out_skip += msg->cursor.total_resid;
3178                 if (msg->middle)
3179                         con->out_skip += con_out_kvec_skip(con);
3180                 con->out_skip += con_out_kvec_skip(con);
3181
3182                 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3183                      __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3184                 msg->hdr.seq = 0;
3185                 con->out_msg = NULL;
3186                 ceph_msg_put(msg);
3187         }
3188
3189         mutex_unlock(&con->mutex);
3190 }
3191
3192 /*
3193  * Revoke a message that we may be reading data into
3194  */
3195 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3196 {
3197         struct ceph_connection *con = msg->con;
3198
3199         if (!con) {
3200                 dout("%s msg %p null con\n", __func__, msg);
3201                 return;         /* Message not in our possession */
3202         }
3203
3204         mutex_lock(&con->mutex);
3205         if (con->in_msg == msg) {
3206                 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3207                 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3208                 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3209
3210                 /* skip rest of message */
3211                 dout("%s %p msg %p revoked\n", __func__, con, msg);
3212                 con->in_base_pos = con->in_base_pos -
3213                                 sizeof(struct ceph_msg_header) -
3214                                 front_len -
3215                                 middle_len -
3216                                 data_len -
3217                                 sizeof(struct ceph_msg_footer);
3218                 ceph_msg_put(con->in_msg);
3219                 con->in_msg = NULL;
3220                 con->in_tag = CEPH_MSGR_TAG_READY;
3221                 con->in_seq++;
3222         } else {
3223                 dout("%s %p in_msg %p msg %p no-op\n",
3224                      __func__, con, con->in_msg, msg);
3225         }
3226         mutex_unlock(&con->mutex);
3227 }
3228
3229 /*
3230  * Queue a keepalive byte to ensure the tcp connection is alive.
3231  */
3232 void ceph_con_keepalive(struct ceph_connection *con)
3233 {
3234         dout("con_keepalive %p\n", con);
3235         mutex_lock(&con->mutex);
3236         clear_standby(con);
3237         mutex_unlock(&con->mutex);
3238         if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3239             con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3240                 queue_con(con);
3241 }
3242 EXPORT_SYMBOL(ceph_con_keepalive);
3243
3244 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3245                                unsigned long interval)
3246 {
3247         if (interval > 0 &&
3248             (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3249                 struct timespec64 now;
3250                 struct timespec64 ts;
3251                 ktime_get_real_ts64(&now);
3252                 jiffies_to_timespec64(interval, &ts);
3253                 ts = timespec64_add(con->last_keepalive_ack, ts);
3254                 return timespec64_compare(&now, &ts) >= 0;
3255         }
3256         return false;
3257 }
3258
3259 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3260 {
3261         struct ceph_msg_data *data;
3262
3263         if (WARN_ON(!ceph_msg_data_type_valid(type)))
3264                 return NULL;
3265
3266         data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3267         if (!data)
3268                 return NULL;
3269
3270         data->type = type;
3271         INIT_LIST_HEAD(&data->links);
3272
3273         return data;
3274 }
3275
3276 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3277 {
3278         if (!data)
3279                 return;
3280
3281         WARN_ON(!list_empty(&data->links));
3282         if (data->type == CEPH_MSG_DATA_PAGELIST)
3283                 ceph_pagelist_release(data->pagelist);
3284         kmem_cache_free(ceph_msg_data_cache, data);
3285 }
3286
3287 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3288                 size_t length, size_t alignment)
3289 {
3290         struct ceph_msg_data *data;
3291
3292         BUG_ON(!pages);
3293         BUG_ON(!length);
3294
3295         data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3296         BUG_ON(!data);
3297         data->pages = pages;
3298         data->length = length;
3299         data->alignment = alignment & ~PAGE_MASK;
3300
3301         list_add_tail(&data->links, &msg->data);
3302         msg->data_length += length;
3303 }
3304 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3305
3306 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3307                                 struct ceph_pagelist *pagelist)
3308 {
3309         struct ceph_msg_data *data;
3310
3311         BUG_ON(!pagelist);
3312         BUG_ON(!pagelist->length);
3313
3314         data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3315         BUG_ON(!data);
3316         data->pagelist = pagelist;
3317
3318         list_add_tail(&data->links, &msg->data);
3319         msg->data_length += pagelist->length;
3320 }
3321 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3322
3323 #ifdef  CONFIG_BLOCK
3324 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
3325                            u32 length)