~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/ceph/messenger.c

Version: ~ [ linux-5.15-rc6 ] ~ [ linux-5.14.14 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.75 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.155 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.213 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.252 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.287 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.289 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 /*
172  * When skipping (ignoring) a block of input we read it into a "skip
173  * buffer," which is this many bytes in size.
174  */
175 #define SKIP_BUF_SIZE   1024
176 
177 static void queue_con(struct ceph_connection *con);
178 static void cancel_con(struct ceph_connection *con);
179 static void ceph_con_workfn(struct work_struct *);
180 static void con_fault(struct ceph_connection *con);
181 
182 /*
183  * Nicely render a sockaddr as a string.  An array of formatted
184  * strings is used, to approximate reentrancy.
185  */
186 #define ADDR_STR_COUNT_LOG      5       /* log2(# address strings in array) */
187 #define ADDR_STR_COUNT          (1 << ADDR_STR_COUNT_LOG)
188 #define ADDR_STR_COUNT_MASK     (ADDR_STR_COUNT - 1)
189 #define MAX_ADDR_STR_LEN        64      /* 54 is enough */
190 
191 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
192 static atomic_t addr_str_seq = ATOMIC_INIT(0);
193 
194 static struct page *zero_page;          /* used in certain error cases */
195 
196 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
197 {
198         int i;
199         char *s;
200         struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
201         struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
202 
203         i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
204         s = addr_str[i];
205 
206         switch (ss->ss_family) {
207         case AF_INET:
208                 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
209                          ntohs(in4->sin_port));
210                 break;
211 
212         case AF_INET6:
213                 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
214                          ntohs(in6->sin6_port));
215                 break;
216 
217         default:
218                 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
219                          ss->ss_family);
220         }
221 
222         return s;
223 }
224 EXPORT_SYMBOL(ceph_pr_addr);
225 
226 static void encode_my_addr(struct ceph_messenger *msgr)
227 {
228         memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
229         ceph_encode_addr(&msgr->my_enc_addr);
230 }
231 
232 /*
233  * work queue for all reading and writing to/from the socket.
234  */
235 static struct workqueue_struct *ceph_msgr_wq;
236 
237 static int ceph_msgr_slab_init(void)
238 {
239         BUG_ON(ceph_msg_cache);
240         ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
241         if (!ceph_msg_cache)
242                 return -ENOMEM;
243 
244         BUG_ON(ceph_msg_data_cache);
245         ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
246         if (ceph_msg_data_cache)
247                 return 0;
248 
249         kmem_cache_destroy(ceph_msg_cache);
250         ceph_msg_cache = NULL;
251 
252         return -ENOMEM;
253 }
254 
255 static void ceph_msgr_slab_exit(void)
256 {
257         BUG_ON(!ceph_msg_data_cache);
258         kmem_cache_destroy(ceph_msg_data_cache);
259         ceph_msg_data_cache = NULL;
260 
261         BUG_ON(!ceph_msg_cache);
262         kmem_cache_destroy(ceph_msg_cache);
263         ceph_msg_cache = NULL;
264 }
265 
266 static void _ceph_msgr_exit(void)
267 {
268         if (ceph_msgr_wq) {
269                 destroy_workqueue(ceph_msgr_wq);
270                 ceph_msgr_wq = NULL;
271         }
272 
273         BUG_ON(zero_page == NULL);
274         put_page(zero_page);
275         zero_page = NULL;
276 
277         ceph_msgr_slab_exit();
278 }
279 
280 int __init ceph_msgr_init(void)
281 {
282         if (ceph_msgr_slab_init())
283                 return -ENOMEM;
284 
285         BUG_ON(zero_page != NULL);
286         zero_page = ZERO_PAGE(0);
287         get_page(zero_page);
288 
289         /*
290          * The number of active work items is limited by the number of
291          * connections, so leave @max_active at default.
292          */
293         ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
294         if (ceph_msgr_wq)
295                 return 0;
296 
297         pr_err("msgr_init failed to create workqueue\n");
298         _ceph_msgr_exit();
299 
300         return -ENOMEM;
301 }
302 
303 void ceph_msgr_exit(void)
304 {
305         BUG_ON(ceph_msgr_wq == NULL);
306 
307         _ceph_msgr_exit();
308 }
309 
310 void ceph_msgr_flush(void)
311 {
312         flush_workqueue(ceph_msgr_wq);
313 }
314 EXPORT_SYMBOL(ceph_msgr_flush);
315 
316 /* Connection socket state transition functions */
317 
318 static void con_sock_state_init(struct ceph_connection *con)
319 {
320         int old_state;
321 
322         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
323         if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
324                 printk("%s: unexpected old state %d\n", __func__, old_state);
325         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
326              CON_SOCK_STATE_CLOSED);
327 }
328 
329 static void con_sock_state_connecting(struct ceph_connection *con)
330 {
331         int old_state;
332 
333         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
334         if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
335                 printk("%s: unexpected old state %d\n", __func__, old_state);
336         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
337              CON_SOCK_STATE_CONNECTING);
338 }
339 
340 static void con_sock_state_connected(struct ceph_connection *con)
341 {
342         int old_state;
343 
344         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
345         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
346                 printk("%s: unexpected old state %d\n", __func__, old_state);
347         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
348              CON_SOCK_STATE_CONNECTED);
349 }
350 
351 static void con_sock_state_closing(struct ceph_connection *con)
352 {
353         int old_state;
354 
355         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
356         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
357                         old_state != CON_SOCK_STATE_CONNECTED &&
358                         old_state != CON_SOCK_STATE_CLOSING))
359                 printk("%s: unexpected old state %d\n", __func__, old_state);
360         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
361              CON_SOCK_STATE_CLOSING);
362 }
363 
364 static void con_sock_state_closed(struct ceph_connection *con)
365 {
366         int old_state;
367 
368         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
369         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
370                     old_state != CON_SOCK_STATE_CLOSING &&
371                     old_state != CON_SOCK_STATE_CONNECTING &&
372                     old_state != CON_SOCK_STATE_CLOSED))
373                 printk("%s: unexpected old state %d\n", __func__, old_state);
374         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
375              CON_SOCK_STATE_CLOSED);
376 }
377 
378 /*
379  * socket callback functions
380  */
381 
382 /* data available on socket, or listen socket received a connect */
383 static void ceph_sock_data_ready(struct sock *sk)
384 {
385         struct ceph_connection *con = sk->sk_user_data;
386         if (atomic_read(&con->msgr->stopping)) {
387                 return;
388         }
389 
390         if (sk->sk_state != TCP_CLOSE_WAIT) {
391                 dout("%s on %p state = %lu, queueing work\n", __func__,
392                      con, con->state);
393                 queue_con(con);
394         }
395 }
396 
397 /* socket has buffer space for writing */
398 static void ceph_sock_write_space(struct sock *sk)
399 {
400         struct ceph_connection *con = sk->sk_user_data;
401 
402         /* only queue to workqueue if there is data we want to write,
403          * and there is sufficient space in the socket buffer to accept
404          * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
405          * doesn't get called again until try_write() fills the socket
406          * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
407          * and net/core/stream.c:sk_stream_write_space().
408          */
409         if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
410                 if (sk_stream_is_writeable(sk)) {
411                         dout("%s %p queueing write work\n", __func__, con);
412                         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
413                         queue_con(con);
414                 }
415         } else {
416                 dout("%s %p nothing to write\n", __func__, con);
417         }
418 }
419 
420 /* socket's state has changed */
421 static void ceph_sock_state_change(struct sock *sk)
422 {
423         struct ceph_connection *con = sk->sk_user_data;
424 
425         dout("%s %p state = %lu sk_state = %u\n", __func__,
426              con, con->state, sk->sk_state);
427 
428         switch (sk->sk_state) {
429         case TCP_CLOSE:
430                 dout("%s TCP_CLOSE\n", __func__);
431                 /* fall through */
432         case TCP_CLOSE_WAIT:
433                 dout("%s TCP_CLOSE_WAIT\n", __func__);
434                 con_sock_state_closing(con);
435                 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
436                 queue_con(con);
437                 break;
438         case TCP_ESTABLISHED:
439                 dout("%s TCP_ESTABLISHED\n", __func__);
440                 con_sock_state_connected(con);
441                 queue_con(con);
442                 break;
443         default:        /* Everything else is uninteresting */
444                 break;
445         }
446 }
447 
448 /*
449  * set up socket callbacks
450  */
451 static void set_sock_callbacks(struct socket *sock,
452                                struct ceph_connection *con)
453 {
454         struct sock *sk = sock->sk;
455         sk->sk_user_data = con;
456         sk->sk_data_ready = ceph_sock_data_ready;
457         sk->sk_write_space = ceph_sock_write_space;
458         sk->sk_state_change = ceph_sock_state_change;
459 }
460 
461 
462 /*
463  * socket helpers
464  */
465 
466 /*
467  * initiate connection to a remote socket.
468  */
469 static int ceph_tcp_connect(struct ceph_connection *con)
470 {
471         struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
472         struct socket *sock;
473         unsigned int noio_flag;
474         int ret;
475 
476         BUG_ON(con->sock);
477 
478         /* sock_create_kern() allocates with GFP_KERNEL */
479         noio_flag = memalloc_noio_save();
480         ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
481                                SOCK_STREAM, IPPROTO_TCP, &sock);
482         memalloc_noio_restore(noio_flag);
483         if (ret)
484                 return ret;
485         sock->sk->sk_allocation = GFP_NOFS;
486 
487 #ifdef CONFIG_LOCKDEP
488         lockdep_set_class(&sock->sk->sk_lock, &socket_class);
489 #endif
490 
491         set_sock_callbacks(sock, con);
492 
493         dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
494 
495         con_sock_state_connecting(con);
496         ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
497                                  O_NONBLOCK);
498         if (ret == -EINPROGRESS) {
499                 dout("connect %s EINPROGRESS sk_state = %u\n",
500                      ceph_pr_addr(&con->peer_addr.in_addr),
501                      sock->sk->sk_state);
502         } else if (ret < 0) {
503                 pr_err("connect %s error %d\n",
504                        ceph_pr_addr(&con->peer_addr.in_addr), ret);
505                 sock_release(sock);
506                 return ret;
507         }
508 
509         if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
510                 int optval = 1;
511 
512                 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
513                                         (char *)&optval, sizeof(optval));
514                 if (ret)
515                         pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
516                                ret);
517         }
518 
519         con->sock = sock;
520         return 0;
521 }
522 
523 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
524 {
525         struct kvec iov = {buf, len};
526         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
527         int r;
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 timespec now;
1421 
1422                 ktime_get_real_ts(&now);
1423                 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1424                 ceph_encode_timespec(&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 struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1438                                                 int *auth_proto)
1439 {
1440         struct ceph_auth_handshake *auth;
1441 
1442         if (!con->ops->get_authorizer) {
1443                 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1444                 con->out_connect.authorizer_len = 0;
1445                 return NULL;
1446         }
1447 
1448         auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1449         if (IS_ERR(auth))
1450                 return auth;
1451 
1452         con->auth_reply_buf = auth->authorizer_reply_buf;
1453         con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1454         return auth;
1455 }
1456 
1457 /*
1458  * We connected to a peer and are saying hello.
1459  */
1460 static void prepare_write_banner(struct ceph_connection *con)
1461 {
1462         con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1463         con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1464                                         &con->msgr->my_enc_addr);
1465 
1466         con->out_more = 0;
1467         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1468 }
1469 
1470 static int prepare_write_connect(struct ceph_connection *con)
1471 {
1472         unsigned int global_seq = get_global_seq(con->msgr, 0);
1473         int proto;
1474         int auth_proto;
1475         struct ceph_auth_handshake *auth;
1476 
1477         switch (con->peer_name.type) {
1478         case CEPH_ENTITY_TYPE_MON:
1479                 proto = CEPH_MONC_PROTOCOL;
1480                 break;
1481         case CEPH_ENTITY_TYPE_OSD:
1482                 proto = CEPH_OSDC_PROTOCOL;
1483                 break;
1484         case CEPH_ENTITY_TYPE_MDS:
1485                 proto = CEPH_MDSC_PROTOCOL;
1486                 break;
1487         default:
1488                 BUG();
1489         }
1490 
1491         dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1492              con->connect_seq, global_seq, proto);
1493 
1494         con->out_connect.features =
1495             cpu_to_le64(from_msgr(con->msgr)->supported_features);
1496         con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1497         con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1498         con->out_connect.global_seq = cpu_to_le32(global_seq);
1499         con->out_connect.protocol_version = cpu_to_le32(proto);
1500         con->out_connect.flags = 0;
1501 
1502         auth_proto = CEPH_AUTH_UNKNOWN;
1503         auth = get_connect_authorizer(con, &auth_proto);
1504         if (IS_ERR(auth))
1505                 return PTR_ERR(auth);
1506 
1507         con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1508         con->out_connect.authorizer_len = auth ?
1509                 cpu_to_le32(auth->authorizer_buf_len) : 0;
1510 
1511         con_out_kvec_add(con, sizeof (con->out_connect),
1512                                         &con->out_connect);
1513         if (auth && auth->authorizer_buf_len)
1514                 con_out_kvec_add(con, auth->authorizer_buf_len,
1515                                         auth->authorizer_buf);
1516 
1517         con->out_more = 0;
1518         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1519 
1520         return 0;
1521 }
1522 
1523 /*
1524  * write as much of pending kvecs to the socket as we can.
1525  *  1 -> done
1526  *  0 -> socket full, but more to do
1527  * <0 -> error
1528  */
1529 static int write_partial_kvec(struct ceph_connection *con)
1530 {
1531         int ret;
1532 
1533         dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1534         while (con->out_kvec_bytes > 0) {
1535                 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1536                                        con->out_kvec_left, con->out_kvec_bytes,
1537                                        con->out_more);
1538                 if (ret <= 0)
1539                         goto out;
1540                 con->out_kvec_bytes -= ret;
1541                 if (con->out_kvec_bytes == 0)
1542                         break;            /* done */
1543 
1544                 /* account for full iov entries consumed */
1545                 while (ret >= con->out_kvec_cur->iov_len) {
1546                         BUG_ON(!con->out_kvec_left);
1547                         ret -= con->out_kvec_cur->iov_len;
1548                         con->out_kvec_cur++;
1549                         con->out_kvec_left--;
1550                 }
1551                 /* and for a partially-consumed entry */
1552                 if (ret) {
1553                         con->out_kvec_cur->iov_len -= ret;
1554                         con->out_kvec_cur->iov_base += ret;
1555                 }
1556         }
1557         con->out_kvec_left = 0;
1558         ret = 1;
1559 out:
1560         dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1561              con->out_kvec_bytes, con->out_kvec_left, ret);
1562         return ret;  /* done! */
1563 }
1564 
1565 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1566                                 unsigned int page_offset,
1567                                 unsigned int length)
1568 {
1569         char *kaddr;
1570 
1571         kaddr = kmap(page);
1572         BUG_ON(kaddr == NULL);
1573         crc = crc32c(crc, kaddr + page_offset, length);
1574         kunmap(page);
1575 
1576         return crc;
1577 }
1578 /*
1579  * Write as much message data payload as we can.  If we finish, queue
1580  * up the footer.
1581  *  1 -> done, footer is now queued in out_kvec[].
1582  *  0 -> socket full, but more to do
1583  * <0 -> error
1584  */
1585 static int write_partial_message_data(struct ceph_connection *con)
1586 {
1587         struct ceph_msg *msg = con->out_msg;
1588         struct ceph_msg_data_cursor *cursor = &msg->cursor;
1589         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1590         u32 crc;
1591 
1592         dout("%s %p msg %p\n", __func__, con, msg);
1593 
1594         if (list_empty(&msg->data))
1595                 return -EINVAL;
1596 
1597         /*
1598          * Iterate through each page that contains data to be
1599          * written, and send as much as possible for each.
1600          *
1601          * If we are calculating the data crc (the default), we will
1602          * need to map the page.  If we have no pages, they have
1603          * been revoked, so use the zero page.
1604          */
1605         crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1606         while (cursor->total_resid) {
1607                 struct page *page;
1608                 size_t page_offset;
1609                 size_t length;
1610                 bool last_piece;
1611                 int ret;
1612 
1613                 if (!cursor->resid) {
1614                         ceph_msg_data_advance(cursor, 0);
1615                         continue;
1616                 }
1617 
1618                 page = ceph_msg_data_next(cursor, &page_offset, &length,
1619                                           &last_piece);
1620                 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1621                                         length, !last_piece);
1622                 if (ret <= 0) {
1623                         if (do_datacrc)
1624                                 msg->footer.data_crc = cpu_to_le32(crc);
1625 
1626                         return ret;
1627                 }
1628                 if (do_datacrc && cursor->need_crc)
1629                         crc = ceph_crc32c_page(crc, page, page_offset, length);
1630                 ceph_msg_data_advance(cursor, (size_t)ret);
1631         }
1632 
1633         dout("%s %p msg %p done\n", __func__, con, msg);
1634 
1635         /* prepare and queue up footer, too */
1636         if (do_datacrc)
1637                 msg->footer.data_crc = cpu_to_le32(crc);
1638         else
1639                 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1640         con_out_kvec_reset(con);
1641         prepare_write_message_footer(con);
1642 
1643         return 1;       /* must return > 0 to indicate success */
1644 }
1645 
1646 /*
1647  * write some zeros
1648  */
1649 static int write_partial_skip(struct ceph_connection *con)
1650 {
1651         int ret;
1652 
1653         dout("%s %p %d left\n", __func__, con, con->out_skip);
1654         while (con->out_skip > 0) {
1655                 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1656 
1657                 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1658                 if (ret <= 0)
1659                         goto out;
1660                 con->out_skip -= ret;
1661         }
1662         ret = 1;
1663 out:
1664         return ret;
1665 }
1666 
1667 /*
1668  * Prepare to read connection handshake, or an ack.
1669  */
1670 static void prepare_read_banner(struct ceph_connection *con)
1671 {
1672         dout("prepare_read_banner %p\n", con);
1673         con->in_base_pos = 0;
1674 }
1675 
1676 static void prepare_read_connect(struct ceph_connection *con)
1677 {
1678         dout("prepare_read_connect %p\n", con);
1679         con->in_base_pos = 0;
1680 }
1681 
1682 static void prepare_read_ack(struct ceph_connection *con)
1683 {
1684         dout("prepare_read_ack %p\n", con);
1685         con->in_base_pos = 0;
1686 }
1687 
1688 static void prepare_read_seq(struct ceph_connection *con)
1689 {
1690         dout("prepare_read_seq %p\n", con);
1691         con->in_base_pos = 0;
1692         con->in_tag = CEPH_MSGR_TAG_SEQ;
1693 }
1694 
1695 static void prepare_read_tag(struct ceph_connection *con)
1696 {
1697         dout("prepare_read_tag %p\n", con);
1698         con->in_base_pos = 0;
1699         con->in_tag = CEPH_MSGR_TAG_READY;
1700 }
1701 
1702 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1703 {
1704         dout("prepare_read_keepalive_ack %p\n", con);
1705         con->in_base_pos = 0;
1706 }
1707 
1708 /*
1709  * Prepare to read a message.
1710  */
1711 static int prepare_read_message(struct ceph_connection *con)
1712 {
1713         dout("prepare_read_message %p\n", con);
1714         BUG_ON(con->in_msg != NULL);
1715         con->in_base_pos = 0;
1716         con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1717         return 0;
1718 }
1719 
1720 
1721 static int read_partial(struct ceph_connection *con,
1722                         int end, int size, void *object)
1723 {
1724         while (con->in_base_pos < end) {
1725                 int left = end - con->in_base_pos;
1726                 int have = size - left;
1727                 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1728                 if (ret <= 0)
1729                         return ret;
1730                 con->in_base_pos += ret;
1731         }
1732         return 1;
1733 }
1734 
1735 
1736 /*
1737  * Read all or part of the connect-side handshake on a new connection
1738  */
1739 static int read_partial_banner(struct ceph_connection *con)
1740 {
1741         int size;
1742         int end;
1743         int ret;
1744 
1745         dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1746 
1747         /* peer's banner */
1748         size = strlen(CEPH_BANNER);
1749         end = size;
1750         ret = read_partial(con, end, size, con->in_banner);
1751         if (ret <= 0)
1752                 goto out;
1753 
1754         size = sizeof (con->actual_peer_addr);
1755         end += size;
1756         ret = read_partial(con, end, size, &con->actual_peer_addr);
1757         if (ret <= 0)
1758                 goto out;
1759 
1760         size = sizeof (con->peer_addr_for_me);
1761         end += size;
1762         ret = read_partial(con, end, size, &con->peer_addr_for_me);
1763         if (ret <= 0)
1764                 goto out;
1765 
1766 out:
1767         return ret;
1768 }
1769 
1770 static int read_partial_connect(struct ceph_connection *con)
1771 {
1772         int size;
1773         int end;
1774         int ret;
1775 
1776         dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1777 
1778         size = sizeof (con->in_reply);
1779         end = size;
1780         ret = read_partial(con, end, size, &con->in_reply);
1781         if (ret <= 0)
1782                 goto out;
1783 
1784         size = le32_to_cpu(con->in_reply.authorizer_len);
1785         end += size;
1786         ret = read_partial(con, end, size, con->auth_reply_buf);
1787         if (ret <= 0)
1788                 goto out;
1789 
1790         dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1791              con, (int)con->in_reply.tag,
1792              le32_to_cpu(con->in_reply.connect_seq),
1793              le32_to_cpu(con->in_reply.global_seq));
1794 out:
1795         return ret;
1796 
1797 }
1798 
1799 /*
1800  * Verify the hello banner looks okay.
1801  */
1802 static int verify_hello(struct ceph_connection *con)
1803 {
1804         if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1805                 pr_err("connect to %s got bad banner\n",
1806                        ceph_pr_addr(&con->peer_addr.in_addr));
1807                 con->error_msg = "protocol error, bad banner";
1808                 return -1;
1809         }
1810         return 0;
1811 }
1812 
1813 static bool addr_is_blank(struct sockaddr_storage *ss)
1814 {
1815         struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1816         struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1817 
1818         switch (ss->ss_family) {
1819         case AF_INET:
1820                 return addr->s_addr == htonl(INADDR_ANY);
1821         case AF_INET6:
1822                 return ipv6_addr_any(addr6);
1823         default:
1824                 return true;
1825         }
1826 }
1827 
1828 static int addr_port(struct sockaddr_storage *ss)
1829 {
1830         switch (ss->ss_family) {
1831         case AF_INET:
1832                 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1833         case AF_INET6:
1834                 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1835         }
1836         return 0;
1837 }
1838 
1839 static void addr_set_port(struct sockaddr_storage *ss, int p)
1840 {
1841         switch (ss->ss_family) {
1842         case AF_INET:
1843                 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1844                 break;
1845         case AF_INET6:
1846                 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1847                 break;
1848         }
1849 }
1850 
1851 /*
1852  * Unlike other *_pton function semantics, zero indicates success.
1853  */
1854 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1855                 char delim, const char **ipend)
1856 {
1857         struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1858         struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1859 
1860         memset(ss, 0, sizeof(*ss));
1861 
1862         if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1863                 ss->ss_family = AF_INET;
1864                 return 0;
1865         }
1866 
1867         if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1868                 ss->ss_family = AF_INET6;
1869                 return 0;
1870         }
1871 
1872         return -EINVAL;
1873 }
1874 
1875 /*
1876  * Extract hostname string and resolve using kernel DNS facility.
1877  */
1878 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1879 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1880                 struct sockaddr_storage *ss, char delim, const char **ipend)
1881 {
1882         const char *end, *delim_p;
1883         char *colon_p, *ip_addr = NULL;
1884         int ip_len, ret;
1885 
1886         /*
1887          * The end of the hostname occurs immediately preceding the delimiter or
1888          * the port marker (':') where the delimiter takes precedence.
1889          */
1890         delim_p = memchr(name, delim, namelen);
1891         colon_p = memchr(name, ':', namelen);
1892 
1893         if (delim_p && colon_p)
1894                 end = delim_p < colon_p ? delim_p : colon_p;
1895         else if (!delim_p && colon_p)
1896                 end = colon_p;
1897         else {
1898                 end = delim_p;
1899                 if (!end) /* case: hostname:/ */
1900                         end = name + namelen;
1901         }
1902 
1903         if (end <= name)
1904                 return -EINVAL;
1905 
1906         /* do dns_resolve upcall */
1907         ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1908         if (ip_len > 0)
1909                 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1910         else
1911                 ret = -ESRCH;
1912 
1913         kfree(ip_addr);
1914 
1915         *ipend = end;
1916 
1917         pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1918                         ret, ret ? "failed" : ceph_pr_addr(ss));
1919 
1920         return ret;
1921 }
1922 #else
1923 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1924                 struct sockaddr_storage *ss, char delim, const char **ipend)
1925 {
1926         return -EINVAL;
1927 }
1928 #endif
1929 
1930 /*
1931  * Parse a server name (IP or hostname). If a valid IP address is not found
1932  * then try to extract a hostname to resolve using userspace DNS upcall.
1933  */
1934 static int ceph_parse_server_name(const char *name, size_t namelen,
1935                         struct sockaddr_storage *ss, char delim, const char **ipend)
1936 {
1937         int ret;
1938 
1939         ret = ceph_pton(name, namelen, ss, delim, ipend);
1940         if (ret)
1941                 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1942 
1943         return ret;
1944 }
1945 
1946 /*
1947  * Parse an ip[:port] list into an addr array.  Use the default
1948  * monitor port if a port isn't specified.
1949  */
1950 int ceph_parse_ips(const char *c, const char *end,
1951                    struct ceph_entity_addr *addr,
1952                    int max_count, int *count)
1953 {
1954         int i, ret = -EINVAL;
1955         const char *p = c;
1956 
1957         dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1958         for (i = 0; i < max_count; i++) {
1959                 const char *ipend;
1960                 struct sockaddr_storage *ss = &addr[i].in_addr;
1961                 int port;
1962                 char delim = ',';
1963 
1964                 if (*p == '[') {
1965                         delim = ']';
1966                         p++;
1967                 }
1968 
1969                 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1970                 if (ret)
1971                         goto bad;
1972                 ret = -EINVAL;
1973 
1974                 p = ipend;
1975 
1976                 if (delim == ']') {
1977                         if (*p != ']') {
1978                                 dout("missing matching ']'\n");
1979                                 goto bad;
1980                         }
1981                         p++;
1982                 }
1983 
1984                 /* port? */
1985                 if (p < end && *p == ':') {
1986                         port = 0;
1987                         p++;
1988                         while (p < end && *p >= '' && *p <= '9') {
1989                                 port = (port * 10) + (*p - '');
1990                                 p++;
1991                         }
1992                         if (port == 0)
1993                                 port = CEPH_MON_PORT;
1994                         else if (port > 65535)
1995                                 goto bad;
1996                 } else {
1997                         port = CEPH_MON_PORT;
1998                 }
1999 
2000                 addr_set_port(ss, port);
2001 
2002                 dout("parse_ips got %s\n", ceph_pr_addr(ss));
2003 
2004                 if (p == end)
2005                         break;
2006                 if (*p != ',')
2007                         goto bad;
2008                 p++;
2009         }
2010 
2011         if (p != end)
2012                 goto bad;
2013 
2014         if (count)
2015                 *count = i + 1;
2016         return 0;
2017 
2018 bad:
2019         pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
2020         return ret;
2021 }
2022 EXPORT_SYMBOL(ceph_parse_ips);
2023 
2024 static int process_banner(struct ceph_connection *con)
2025 {
2026         dout("process_banner on %p\n", con);
2027 
2028         if (verify_hello(con) < 0)
2029                 return -1;
2030 
2031         ceph_decode_addr(&con->actual_peer_addr);
2032         ceph_decode_addr(&con->peer_addr_for_me);
2033 
2034         /*
2035          * Make sure the other end is who we wanted.  note that the other
2036          * end may not yet know their ip address, so if it's 0.0.0.0, give
2037          * them the benefit of the doubt.
2038          */
2039         if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2040                    sizeof(con->peer_addr)) != 0 &&
2041             !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2042               con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2043                 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2044                         ceph_pr_addr(&con->peer_addr.in_addr),
2045                         (int)le32_to_cpu(con->peer_addr.nonce),
2046                         ceph_pr_addr(&con->actual_peer_addr.in_addr),
2047                         (int)le32_to_cpu(con->actual_peer_addr.nonce));
2048                 con->error_msg = "wrong peer at address";
2049                 return -1;
2050         }
2051 
2052         /*
2053          * did we learn our address?
2054          */
2055         if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2056                 int port = addr_port(&con->msgr->inst.addr.in_addr);
2057 
2058                 memcpy(&con->msgr->inst.addr.in_addr,
2059                        &con->peer_addr_for_me.in_addr,
2060                        sizeof(con->peer_addr_for_me.in_addr));
2061                 addr_set_port(&con->msgr->inst.addr.in_addr, port);
2062                 encode_my_addr(con->msgr);
2063                 dout("process_banner learned my addr is %s\n",
2064                      ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2065         }
2066 
2067         return 0;
2068 }
2069 
2070 static int process_connect(struct ceph_connection *con)
2071 {
2072         u64 sup_feat = from_msgr(con->msgr)->supported_features;
2073         u64 req_feat = from_msgr(con->msgr)->required_features;
2074         u64 server_feat = le64_to_cpu(con->in_reply.features);
2075         int ret;
2076 
2077         dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2078 
2079         if (con->auth_reply_buf) {
2080                 /*
2081                  * Any connection that defines ->get_authorizer()
2082                  * should also define ->verify_authorizer_reply().
2083                  * See get_connect_authorizer().
2084                  */
2085                 ret = con->ops->verify_authorizer_reply(con);
2086                 if (ret < 0) {
2087                         con->error_msg = "bad authorize reply";
2088                         return ret;
2089                 }
2090         }
2091 
2092         switch (con->in_reply.tag) {
2093         case CEPH_MSGR_TAG_FEATURES:
2094                 pr_err("%s%lld %s feature set mismatch,"
2095                        " my %llx < server's %llx, missing %llx\n",
2096                        ENTITY_NAME(con->peer_name),
2097                        ceph_pr_addr(&con->peer_addr.in_addr),
2098                        sup_feat, server_feat, server_feat & ~sup_feat);
2099                 con->error_msg = "missing required protocol features";
2100                 reset_connection(con);
2101                 return -1;
2102 
2103         case CEPH_MSGR_TAG_BADPROTOVER:
2104                 pr_err("%s%lld %s protocol version mismatch,"
2105                        " my %d != server's %d\n",
2106                        ENTITY_NAME(con->peer_name),
2107                        ceph_pr_addr(&con->peer_addr.in_addr),
2108                        le32_to_cpu(con->out_connect.protocol_version),
2109                        le32_to_cpu(con->in_reply.protocol_version));
2110                 con->error_msg = "protocol version mismatch";
2111                 reset_connection(con);
2112                 return -1;
2113 
2114         case CEPH_MSGR_TAG_BADAUTHORIZER:
2115                 con->auth_retry++;
2116                 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2117                      con->auth_retry);
2118                 if (con->auth_retry == 2) {
2119                         con->error_msg = "connect authorization failure";
2120                         return -1;
2121                 }
2122                 con_out_kvec_reset(con);
2123                 ret = prepare_write_connect(con);
2124                 if (ret < 0)
2125                         return ret;
2126                 prepare_read_connect(con);
2127                 break;
2128 
2129         case CEPH_MSGR_TAG_RESETSESSION:
2130                 /*
2131                  * If we connected with a large connect_seq but the peer
2132                  * has no record of a session with us (no connection, or
2133                  * connect_seq == 0), they will send RESETSESION to indicate
2134                  * that they must have reset their session, and may have
2135                  * dropped messages.
2136                  */
2137                 dout("process_connect got RESET peer seq %u\n",
2138                      le32_to_cpu(con->in_reply.connect_seq));
2139                 pr_err("%s%lld %s connection reset\n",
2140                        ENTITY_NAME(con->peer_name),
2141                        ceph_pr_addr(&con->peer_addr.in_addr));
2142                 reset_connection(con);
2143                 con_out_kvec_reset(con);
2144                 ret = prepare_write_connect(con);
2145                 if (ret < 0)
2146                         return ret;
2147                 prepare_read_connect(con);
2148 
2149                 /* Tell ceph about it. */
2150                 mutex_unlock(&con->mutex);
2151                 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2152                 if (con->ops->peer_reset)
2153                         con->ops->peer_reset(con);
2154                 mutex_lock(&con->mutex);
2155                 if (con->state != CON_STATE_NEGOTIATING)
2156                         return -EAGAIN;
2157                 break;
2158 
2159         case CEPH_MSGR_TAG_RETRY_SESSION:
2160                 /*
2161                  * If we sent a smaller connect_seq than the peer has, try
2162                  * again with a larger value.
2163                  */
2164                 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2165                      le32_to_cpu(con->out_connect.connect_seq),
2166                      le32_to_cpu(con->in_reply.connect_seq));
2167                 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2168                 con_out_kvec_reset(con);
2169                 ret = prepare_write_connect(con);
2170                 if (ret < 0)
2171                         return ret;
2172                 prepare_read_connect(con);
2173                 break;
2174 
2175         case CEPH_MSGR_TAG_RETRY_GLOBAL:
2176                 /*
2177                  * If we sent a smaller global_seq than the peer has, try
2178                  * again with a larger value.
2179                  */
2180                 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2181                      con->peer_global_seq,
2182                      le32_to_cpu(con->in_reply.global_seq));
2183                 get_global_seq(con->msgr,
2184                                le32_to_cpu(con->in_reply.global_seq));
2185                 con_out_kvec_reset(con);
2186                 ret = prepare_write_connect(con);
2187                 if (ret < 0)
2188                         return ret;
2189                 prepare_read_connect(con);
2190                 break;
2191 
2192         case CEPH_MSGR_TAG_SEQ:
2193         case CEPH_MSGR_TAG_READY:
2194                 if (req_feat & ~server_feat) {
2195                         pr_err("%s%lld %s protocol feature mismatch,"
2196                                " my required %llx > server's %llx, need %llx\n",
2197                                ENTITY_NAME(con->peer_name),
2198                                ceph_pr_addr(&con->peer_addr.in_addr),
2199                                req_feat, server_feat, req_feat & ~server_feat);
2200                         con->error_msg = "missing required protocol features";
2201                         reset_connection(con);
2202                         return -1;
2203                 }
2204 
2205                 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2206                 con->state = CON_STATE_OPEN;
2207                 con->auth_retry = 0;    /* we authenticated; clear flag */
2208                 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2209                 con->connect_seq++;
2210                 con->peer_features = server_feat;
2211                 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2212                      con->peer_global_seq,
2213                      le32_to_cpu(con->in_reply.connect_seq),
2214                      con->connect_seq);
2215                 WARN_ON(con->connect_seq !=
2216                         le32_to_cpu(con->in_reply.connect_seq));
2217 
2218                 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2219                         con_flag_set(con, CON_FLAG_LOSSYTX);
2220 
2221                 con->delay = 0;      /* reset backoff memory */
2222 
2223                 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2224                         prepare_write_seq(con);
2225                         prepare_read_seq(con);
2226                 } else {
2227                         prepare_read_tag(con);
2228                 }
2229                 break;
2230 
2231         case CEPH_MSGR_TAG_WAIT:
2232                 /*
2233                  * If there is a connection race (we are opening
2234                  * connections to each other), one of us may just have
2235                  * to WAIT.  This shouldn't happen if we are the
2236                  * client.
2237                  */
2238                 con->error_msg = "protocol error, got WAIT as client";
2239                 return -1;
2240 
2241         default:
2242                 con->error_msg = "protocol error, garbage tag during connect";
2243                 return -1;
2244         }
2245         return 0;
2246 }
2247 
2248 
2249 /*
2250  * read (part of) an ack
2251  */
2252 static int read_partial_ack(struct ceph_connection *con)
2253 {
2254         int size = sizeof (con->in_temp_ack);
2255         int end = size;
2256 
2257         return read_partial(con, end, size, &con->in_temp_ack);
2258 }
2259 
2260 /*
2261  * We can finally discard anything that's been acked.
2262  */
2263 static void process_ack(struct ceph_connection *con)
2264 {
2265         struct ceph_msg *m;
2266         u64 ack = le64_to_cpu(con->in_temp_ack);
2267         u64 seq;
2268         bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2269         struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
2270 
2271         /*
2272          * In the reconnect case, con_fault() has requeued messages
2273          * in out_sent. We should cleanup old messages according to
2274          * the reconnect seq.
2275          */
2276         while (!list_empty(list)) {
2277                 m = list_first_entry(list, struct ceph_msg, list_head);
2278                 if (reconnect && m->needs_out_seq)
2279                         break;
2280                 seq = le64_to_cpu(m->hdr.seq);
2281                 if (seq > ack)
2282                         break;
2283                 dout("got ack for seq %llu type %d at %p\n", seq,
2284                      le16_to_cpu(m->hdr.type), m);
2285                 m->ack_stamp = jiffies;
2286                 ceph_msg_remove(m);
2287         }
2288 
2289         prepare_read_tag(con);
2290 }
2291 
2292 
2293 static int read_partial_message_section(struct ceph_connection *con,
2294                                         struct kvec *section,
2295                                         unsigned int sec_len, u32 *crc)
2296 {
2297         int ret, left;
2298 
2299         BUG_ON(!section);
2300 
2301         while (section->iov_len < sec_len) {
2302                 BUG_ON(section->iov_base == NULL);
2303                 left = sec_len - section->iov_len;
2304                 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2305                                        section->iov_len, left);
2306                 if (ret <= 0)
2307                         return ret;
2308                 section->iov_len += ret;
2309         }
2310         if (section->iov_len == sec_len)
2311                 *crc = crc32c(0, section->iov_base, section->iov_len);
2312 
2313         return 1;
2314 }
2315 
2316 static int read_partial_msg_data(struct ceph_connection *con)
2317 {
2318         struct ceph_msg *msg = con->in_msg;
2319         struct ceph_msg_data_cursor *cursor = &msg->cursor;
2320         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2321         struct page *page;
2322         size_t page_offset;
2323         size_t length;
2324         u32 crc = 0;
2325         int ret;
2326 
2327         BUG_ON(!msg);
2328         if (list_empty(&msg->data))
2329                 return -EIO;
2330 
2331         if (do_datacrc)
2332                 crc = con->in_data_crc;
2333         while (cursor->total_resid) {
2334                 if (!cursor->resid) {
2335                         ceph_msg_data_advance(cursor, 0);
2336                         continue;
2337                 }
2338 
2339                 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2340                 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2341                 if (ret <= 0) {
2342                         if (do_datacrc)
2343                                 con->in_data_crc = crc;
2344 
2345                         return ret;
2346                 }
2347 
2348                 if (do_datacrc)
2349                         crc = ceph_crc32c_page(crc, page, page_offset, ret);
2350                 ceph_msg_data_advance(cursor, (size_t)ret);
2351         }
2352         if (do_datacrc)
2353                 con->in_data_crc = crc;
2354 
2355         return 1;       /* must return > 0 to indicate success */
2356 }
2357 
2358 /*
2359  * read (part of) a message.
2360  */
2361 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2362 
2363 static int read_partial_message(struct ceph_connection *con)
2364 {
2365         struct ceph_msg *m = con->in_msg;
2366         int size;
2367         int end;
2368         int ret;
2369         unsigned int front_len, middle_len, data_len;
2370         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2371         bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2372         u64 seq;
2373         u32 crc;
2374 
2375         dout("read_partial_message con %p msg %p\n", con, m);
2376 
2377         /* header */
2378         size = sizeof (con->in_hdr);
2379         end = size;
2380         ret = read_partial(con, end, size, &con->in_hdr);
2381         if (ret <= 0)
2382                 return ret;
2383 
2384         crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2385         if (cpu_to_le32(crc) != con->in_hdr.crc) {
2386                 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2387                        crc, con->in_hdr.crc);
2388                 return -EBADMSG;
2389         }
2390 
2391         front_len = le32_to_cpu(con->in_hdr.front_len);
2392         if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2393                 return -EIO;
2394         middle_len = le32_to_cpu(con->in_hdr.middle_len);
2395         if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2396                 return -EIO;
2397         data_len = le32_to_cpu(con->in_hdr.data_len);
2398         if (data_len > CEPH_MSG_MAX_DATA_LEN)
2399                 return -EIO;
2400 
2401         /* verify seq# */
2402         seq = le64_to_cpu(con->in_hdr.seq);
2403         if ((s64)seq - (s64)con->in_seq < 1) {
2404                 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2405                         ENTITY_NAME(con->peer_name),
2406                         ceph_pr_addr(&con->peer_addr.in_addr),
2407                         seq, con->in_seq + 1);
2408                 con->in_base_pos = -front_len - middle_len - data_len -
2409                         sizeof_footer(con);
2410                 con->in_tag = CEPH_MSGR_TAG_READY;
2411                 return 1;
2412         } else if ((s64)seq - (s64)con->in_seq > 1) {
2413                 pr_err("read_partial_message bad seq %lld expected %lld\n",
2414                        seq, con->in_seq + 1);
2415                 con->error_msg = "bad message sequence # for incoming message";
2416                 return -EBADE;
2417         }
2418 
2419         /* allocate message? */
2420         if (!con->in_msg) {
2421                 int skip = 0;
2422 
2423                 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2424                      front_len, data_len);
2425                 ret = ceph_con_in_msg_alloc(con, &skip);
2426                 if (ret < 0)
2427                         return ret;
2428 
2429                 BUG_ON(!con->in_msg ^ skip);
2430                 if (skip) {
2431                         /* skip this message */
2432                         dout("alloc_msg said skip message\n");
2433                         con->in_base_pos = -front_len - middle_len - data_len -
2434                                 sizeof_footer(con);
2435                         con->in_tag = CEPH_MSGR_TAG_READY;
2436                         con->in_seq++;
2437                         return 1;
2438                 }
2439 
2440                 BUG_ON(!con->in_msg);
2441                 BUG_ON(con->in_msg->con != con);
2442                 m = con->in_msg;
2443                 m->front.iov_len = 0;    /* haven't read it yet */
2444                 if (m->middle)
2445                         m->middle->vec.iov_len = 0;
2446 
2447                 /* prepare for data payload, if any */
2448 
2449                 if (data_len)
2450                         prepare_message_data(con->in_msg, data_len);
2451         }
2452 
2453         /* front */
2454         ret = read_partial_message_section(con, &m->front, front_len,
2455                                            &con->in_front_crc);
2456         if (ret <= 0)
2457                 return ret;
2458 
2459         /* middle */
2460         if (m->middle) {
2461                 ret = read_partial_message_section(con, &m->middle->vec,
2462                                                    middle_len,
2463                                                    &con->in_middle_crc);
2464                 if (ret <= 0)
2465                         return ret;
2466         }
2467 
2468         /* (page) data */
2469         if (data_len) {
2470                 ret = read_partial_msg_data(con);
2471                 if (ret <= 0)
2472                         return ret;
2473         }
2474 
2475         /* footer */
2476         size = sizeof_footer(con);
2477         end += size;
2478         ret = read_partial(con, end, size, &m->footer);
2479         if (ret <= 0)
2480                 return ret;
2481 
2482         if (!need_sign) {
2483                 m->footer.flags = m->old_footer.flags;
2484                 m->footer.sig = 0;
2485         }
2486 
2487         dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2488              m, front_len, m->footer.front_crc, middle_len,
2489              m->footer.middle_crc, data_len, m->footer.data_crc);
2490 
2491         /* crc ok? */
2492         if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2493                 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2494                        m, con->in_front_crc, m->footer.front_crc);
2495                 return -EBADMSG;
2496         }
2497         if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2498                 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2499                        m, con->in_middle_crc, m->footer.middle_crc);
2500                 return -EBADMSG;
2501         }
2502         if (do_datacrc &&
2503             (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2504             con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2505                 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2506                        con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2507                 return -EBADMSG;
2508         }
2509 
2510         if (need_sign && con->ops->check_message_signature &&
2511             con->ops->check_message_signature(m)) {
2512                 pr_err("read_partial_message %p signature check failed\n", m);
2513                 return -EBADMSG;
2514         }
2515 
2516         return 1; /* done! */
2517 }
2518 
2519 /*
2520  * Process message.  This happens in the worker thread.  The callback should
2521  * be careful not to do anything that waits on other incoming messages or it
2522  * may deadlock.
2523  */
2524 static void process_message(struct ceph_connection *con)
2525 {
2526         struct ceph_msg *msg = con->in_msg;
2527 
2528         BUG_ON(con->in_msg->con != con);
2529         con->in_msg = NULL;
2530 
2531         /* if first message, set peer_name */
2532         if (con->peer_name.type == 0)
2533                 con->peer_name = msg->hdr.src;
2534 
2535         con->in_seq++;
2536         mutex_unlock(&con->mutex);
2537 
2538         dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2539              msg, le64_to_cpu(msg->hdr.seq),
2540              ENTITY_NAME(msg->hdr.src),
2541              le16_to_cpu(msg->hdr.type),
2542              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2543              le32_to_cpu(msg->hdr.front_len),
2544              le32_to_cpu(msg->hdr.data_len),
2545              con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2546         con->ops->dispatch(con, msg);
2547 
2548         mutex_lock(&con->mutex);
2549 }
2550 
2551 static int read_keepalive_ack(struct ceph_connection *con)
2552 {
2553         struct ceph_timespec ceph_ts;
2554         size_t size = sizeof(ceph_ts);
2555         int ret = read_partial(con, size, size, &ceph_ts);
2556         if (ret <= 0)
2557                 return ret;
2558         ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2559         prepare_read_tag(con);
2560         return 1;
2561 }
2562 
2563 /*
2564  * Write something to the socket.  Called in a worker thread when the
2565  * socket appears to be writeable and we have something ready to send.
2566  */
2567 static int try_write(struct ceph_connection *con)
2568 {
2569         int ret = 1;
2570 
2571         dout("try_write start %p state %lu\n", con, con->state);
2572         if (con->state != CON_STATE_PREOPEN &&
2573             con->state != CON_STATE_CONNECTING &&
2574             con->state != CON_STATE_NEGOTIATING &&
2575             con->state != CON_STATE_OPEN)
2576                 return 0;
2577 
2578 more:
2579         dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2580 
2581         /* open the socket first? */
2582         if (con->state == CON_STATE_PREOPEN) {
2583                 BUG_ON(con->sock);
2584                 con->state = CON_STATE_CONNECTING;
2585 
2586                 con_out_kvec_reset(con);
2587                 prepare_write_banner(con);
2588                 prepare_read_banner(con);
2589 
2590                 BUG_ON(con->in_msg);
2591                 con->in_tag = CEPH_MSGR_TAG_READY;
2592                 dout("try_write initiating connect on %p new state %lu\n",
2593                      con, con->state);
2594                 ret = ceph_tcp_connect(con);
2595                 if (ret < 0) {
2596                         con->error_msg = "connect error";
2597                         goto out;
2598                 }
2599         }
2600 
2601 more_kvec:
2602         BUG_ON(!con->sock);
2603 
2604         /* kvec data queued? */
2605         if (con->out_kvec_left) {
2606                 ret = write_partial_kvec(con);
2607                 if (ret <= 0)
2608                         goto out;
2609         }
2610         if (con->out_skip) {
2611                 ret = write_partial_skip(con);
2612                 if (ret <= 0)
2613                         goto out;
2614         }
2615 
2616         /* msg pages? */
2617         if (con->out_msg) {
2618                 if (con->out_msg_done) {
2619                         ceph_msg_put(con->out_msg);
2620                         con->out_msg = NULL;   /* we're done with this one */
2621                         goto do_next;
2622                 }
2623 
2624                 ret = write_partial_message_data(con);
2625                 if (ret == 1)
2626                         goto more_kvec;  /* we need to send the footer, too! */
2627                 if (ret == 0)
2628                         goto out;
2629                 if (ret < 0) {
2630                         dout("try_write write_partial_message_data err %d\n",
2631                              ret);
2632                         goto out;
2633                 }
2634         }
2635 
2636 do_next:
2637         if (con->state == CON_STATE_OPEN) {
2638                 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2639                         prepare_write_keepalive(con);
2640                         goto more;
2641                 }
2642                 /* is anything else pending? */
2643                 if (!list_empty(&con->out_queue)) {
2644                         prepare_write_message(con);
2645                         goto more;
2646                 }
2647                 if (con->in_seq > con->in_seq_acked) {
2648                         prepare_write_ack(con);
2649                         goto more;
2650                 }
2651         }
2652 
2653         /* Nothing to do! */
2654         con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2655         dout("try_write nothing else to write.\n");
2656         ret = 0;
2657 out:
2658         dout("try_write done on %p ret %d\n", con, ret);
2659         return ret;
2660 }
2661 
2662 
2663 
2664 /*
2665  * Read what we can from the socket.
2666  */
2667 static int try_read(struct ceph_connection *con)
2668 {
2669         int ret = -1;
2670 
2671 more:
2672         dout("try_read start on %p state %lu\n", con, con->state);
2673         if (con->state != CON_STATE_CONNECTING &&
2674             con->state != CON_STATE_NEGOTIATING &&
2675             con->state != CON_STATE_OPEN)
2676                 return 0;
2677 
2678         BUG_ON(!con->sock);
2679 
2680         dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2681              con->in_base_pos);
2682 
2683         if (con->state == CON_STATE_CONNECTING) {
2684                 dout("try_read connecting\n");
2685                 ret = read_partial_banner(con);
2686                 if (ret <= 0)
2687                         goto out;
2688                 ret = process_banner(con);
2689                 if (ret < 0)
2690                         goto out;
2691 
2692                 con->state = CON_STATE_NEGOTIATING;
2693 
2694                 /*
2695                  * Received banner is good, exchange connection info.
2696                  * Do not reset out_kvec, as sending our banner raced
2697                  * with receiving peer banner after connect completed.
2698                  */
2699                 ret = prepare_write_connect(con);
2700                 if (ret < 0)
2701                         goto out;
2702                 prepare_read_connect(con);
2703 
2704                 /* Send connection info before awaiting response */
2705                 goto out;
2706         }
2707 
2708         if (con->state == CON_STATE_NEGOTIATING) {
2709                 dout("try_read negotiating\n");
2710                 ret = read_partial_connect(con);
2711                 if (ret <= 0)
2712                         goto out;
2713                 ret = process_connect(con);
2714                 if (ret < 0)
2715                         goto out;
2716                 goto more;
2717         }
2718 
2719         WARN_ON(con->state != CON_STATE_OPEN);
2720 
2721         if (con->in_base_pos < 0) {
2722                 /*
2723                  * skipping + discarding content.
2724                  *
2725                  * FIXME: there must be a better way to do this!
2726                  */
2727                 static char buf[SKIP_BUF_SIZE];
2728                 int skip = min((int) sizeof (buf), -con->in_base_pos);
2729 
2730                 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2731                 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2732                 if (ret <= 0)
2733                         goto out;
2734                 con->in_base_pos += ret;
2735                 if (con->in_base_pos)
2736                         goto more;
2737         }
2738         if (con->in_tag == CEPH_MSGR_TAG_READY) {
2739                 /*
2740                  * what's next?
2741                  */
2742                 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2743                 if (ret <= 0)
2744                         goto out;
2745                 dout("try_read got tag %d\n", (int)con->in_tag);
2746                 switch (con->in_tag) {
2747                 case CEPH_MSGR_TAG_MSG:
2748                         prepare_read_message(con);
2749                         break;
2750                 case CEPH_MSGR_TAG_ACK:
2751                         prepare_read_ack(con);
2752                         break;
2753                 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2754                         prepare_read_keepalive_ack(con);
2755                         break;
2756                 case CEPH_MSGR_TAG_CLOSE:
2757                         con_close_socket(con);
2758                         con->state = CON_STATE_CLOSED;
2759                         goto out;
2760                 default:
2761                         goto bad_tag;
2762                 }
2763         }
2764         if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2765                 ret = read_partial_message(con);
2766                 if (ret <= 0) {
2767                         switch (ret) {
2768                         case -EBADMSG:
2769                                 con->error_msg = "bad crc/signature";
2770                                 /* fall through */
2771                         case -EBADE:
2772                                 ret = -EIO;
2773                                 break;
2774                         case -EIO:
2775                                 con->error_msg = "io error";
2776                                 break;
2777                         }
2778                         goto out;
2779                 }
2780                 if (con->in_tag == CEPH_MSGR_TAG_READY)
2781                         goto more;
2782                 process_message(con);
2783                 if (con->state == CON_STATE_OPEN)
2784                         prepare_read_tag(con);
2785                 goto more;
2786         }
2787         if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2788             con->in_tag == CEPH_MSGR_TAG_SEQ) {
2789                 /*
2790                  * the final handshake seq exchange is semantically
2791                  * equivalent to an ACK
2792                  */
2793                 ret = read_partial_ack(con);
2794                 if (ret <= 0)
2795                         goto out;
2796                 process_ack(con);
2797                 goto more;
2798         }
2799         if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2800                 ret = read_keepalive_ack(con);
2801                 if (ret <= 0)
2802                         goto out;
2803                 goto more;
2804         }
2805 
2806 out:
2807         dout("try_read done on %p ret %d\n", con, ret);
2808         return ret;
2809 
2810 bad_tag:
2811         pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2812         con->error_msg = "protocol error, garbage tag";
2813         ret = -1;
2814         goto out;
2815 }
2816 
2817 
2818 /*
2819  * Atomically queue work on a connection after the specified delay.
2820  * Bump @con reference to avoid races with connection teardown.
2821  * Returns 0 if work was queued, or an error code otherwise.
2822  */
2823 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2824 {
2825         if (!con->ops->get(con)) {
2826                 dout("%s %p ref count 0\n", __func__, con);
2827                 return -ENOENT;
2828         }
2829 
2830         if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2831                 dout("%s %p - already queued\n", __func__, con);
2832                 con->ops->put(con);
2833                 return -EBUSY;
2834         }
2835 
2836         dout("%s %p %lu\n", __func__, con, delay);
2837         return 0;
2838 }
2839 
2840 static void queue_con(struct ceph_connection *con)
2841 {
2842         (void) queue_con_delay(con, 0);
2843 }
2844 
2845 static void cancel_con(struct ceph_connection *con)
2846 {
2847         if (cancel_delayed_work(&con->work)) {
2848                 dout("%s %p\n", __func__, con);
2849                 con->ops->put(con);
2850         }
2851 }
2852 
2853 static bool con_sock_closed(struct ceph_connection *con)
2854 {
2855         if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2856                 return false;
2857 
2858 #define CASE(x)                                                         \
2859         case CON_STATE_ ## x:                                           \
2860                 con->error_msg = "socket closed (con state " #x ")";    \
2861                 break;
2862 
2863         switch (con->state) {
2864         CASE(CLOSED);
2865         CASE(PREOPEN);
2866         CASE(CONNECTING);
2867         CASE(NEGOTIATING);
2868         CASE(OPEN);
2869         CASE(STANDBY);
2870         default:
2871                 pr_warn("%s con %p unrecognized state %lu\n",
2872                         __func__, con, con->state);
2873                 con->error_msg = "unrecognized con state";
2874                 BUG();
2875                 break;
2876         }
2877 #undef CASE
2878 
2879         return true;
2880 }
2881 
2882 static bool con_backoff(struct ceph_connection *con)
2883 {
2884         int ret;
2885 
2886         if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2887                 return false;
2888 
2889         ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2890         if (ret) {
2891                 dout("%s: con %p FAILED to back off %lu\n", __func__,
2892                         con, con->delay);
2893                 BUG_ON(ret == -ENOENT);
2894                 con_flag_set(con, CON_FLAG_BACKOFF);
2895         }
2896 
2897         return true;
2898 }
2899 
2900 /* Finish fault handling; con->mutex must *not* be held here */
2901 
2902 static void con_fault_finish(struct ceph_connection *con)
2903 {
2904         dout("%s %p\n", __func__, con);
2905 
2906         /*
2907          * in case we faulted due to authentication, invalidate our
2908          * current tickets so that we can get new ones.
2909          */
2910         if (con->auth_retry) {
2911                 dout("auth_retry %d, invalidating\n", con->auth_retry);
2912                 if (con->ops->invalidate_authorizer)
2913                         con->ops->invalidate_authorizer(con);
2914                 con->auth_retry = 0;
2915         }
2916 
2917         if (con->ops->fault)
2918                 con->ops->fault(con);
2919 }
2920 
2921 /*
2922  * Do some work on a connection.  Drop a connection ref when we're done.
2923  */
2924 static void ceph_con_workfn(struct work_struct *work)
2925 {
2926         struct ceph_connection *con = container_of(work, struct ceph_connection,
2927                                                    work.work);
2928         bool fault;
2929 
2930         mutex_lock(&con->mutex);
2931         while (true) {
2932                 int ret;
2933 
2934                 if ((fault = con_sock_closed(con))) {
2935                         dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2936                         break;
2937                 }
2938                 if (con_backoff(con)) {
2939                         dout("%s: con %p BACKOFF\n", __func__, con);
2940                         break;
2941                 }
2942                 if (con->state == CON_STATE_STANDBY) {
2943                         dout("%s: con %p STANDBY\n", __func__, con);
2944                         break;
2945                 }
2946                 if (con->state == CON_STATE_CLOSED) {
2947                         dout("%s: con %p CLOSED\n", __func__, con);
2948                         BUG_ON(con->sock);
2949                         break;
2950                 }
2951                 if (con->state == CON_STATE_PREOPEN) {
2952                         dout("%s: con %p PREOPEN\n", __func__, con);
2953                         BUG_ON(con->sock);
2954                 }
2955 
2956                 ret = try_read(con);
2957                 if (ret < 0) {
2958                         if (ret == -EAGAIN)
2959                                 continue;
2960                         if (!con->error_msg)
2961                                 con->error_msg = "socket error on read";
2962                         fault = true;
2963                         break;
2964                 }
2965 
2966                 ret = try_write(con);
2967                 if (ret < 0) {
2968                         if (ret == -EAGAIN)
2969                                 continue;
2970                         if (!con->error_msg)
2971                                 con->error_msg = "socket error on write";
2972                         fault = true;
2973                 }
2974 
2975                 break;  /* If we make it to here, we're done */
2976         }
2977         if (fault)
2978                 con_fault(con);
2979         mutex_unlock(&con->mutex);
2980 
2981         if (fault)
2982                 con_fault_finish(con);
2983 
2984         con->ops->put(con);
2985 }
2986 
2987 /*
2988  * Generic error/fault handler.  A retry mechanism is used with
2989  * exponential backoff
2990  */
2991 static void con_fault(struct ceph_connection *con)
2992 {
2993         dout("fault %p state %lu to peer %s\n",
2994              con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2995 
2996         pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2997                 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2998         con->error_msg = NULL;
2999 
3000         WARN_ON(con->state != CON_STATE_CONNECTING &&
3001                con->state != CON_STATE_NEGOTIATING &&
3002                con->state != CON_STATE_OPEN);
3003 
3004         con_close_socket(con);
3005 
3006         if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
3007                 dout("fault on LOSSYTX channel, marking CLOSED\n");
3008                 con->state = CON_STATE_CLOSED;
3009                 return;
3010         }
3011 
3012         if (con->in_msg) {
3013                 BUG_ON(con->in_msg->con != con);
3014                 ceph_msg_put(con->in_msg);
3015                 con->in_msg = NULL;
3016         }
3017 
3018         /* Requeue anything that hasn't been acked */
3019         list_splice_init(&con->out_sent, &con->out_queue);
3020 
3021         /* If there are no messages queued or keepalive pending, place
3022          * the connection in a STANDBY state */
3023         if (list_empty(&con->out_queue) &&
3024             !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
3025                 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
3026                 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
3027                 con->state = CON_STATE_STANDBY;
3028         } else {
3029                 /* retry after a delay. */
3030                 con->state = CON_STATE_PREOPEN;
3031                 if (con->delay == 0)
3032                         con->delay = BASE_DELAY_INTERVAL;
3033                 else if (con->delay < MAX_DELAY_INTERVAL)
3034                         con->delay *= 2;
3035                 con_flag_set(con, CON_FLAG_BACKOFF);
3036                 queue_con(con);
3037         }
3038 }
3039 
3040 
3041 
3042 /*
3043  * initialize a new messenger instance
3044  */
3045 void ceph_messenger_init(struct ceph_messenger *msgr,
3046                          struct ceph_entity_addr *myaddr)
3047 {
3048         spin_lock_init(&msgr->global_seq_lock);
3049 
3050         if (myaddr)
3051                 msgr->inst.addr = *myaddr;
3052 
3053         /* select a random nonce */
3054         msgr->inst.addr.type = 0;
3055         get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3056         encode_my_addr(msgr);
3057 
3058         atomic_set(&msgr->stopping, 0);
3059         write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3060 
3061         dout("%s %p\n", __func__, msgr);
3062 }
3063 EXPORT_SYMBOL(ceph_messenger_init);
3064 
3065 void ceph_messenger_fini(struct ceph_messenger *msgr)
3066 {
3067         put_net(read_pnet(&msgr->net));
3068 }
3069 EXPORT_SYMBOL(ceph_messenger_fini);
3070 
3071 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3072 {
3073         if (msg->con)
3074                 msg->con->ops->put(msg->con);
3075 
3076         msg->con = con ? con->ops->get(con) : NULL;
3077         BUG_ON(msg->con != con);
3078 }
3079 
3080 static void clear_standby(struct ceph_connection *con)
3081 {
3082         /* come back from STANDBY? */
3083         if (con->state == CON_STATE_STANDBY) {
3084                 dout("clear_standby %p and ++connect_seq\n", con);
3085                 con->state = CON_STATE_PREOPEN;
3086                 con->connect_seq++;
3087                 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3088                 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3089         }
3090 }
3091 
3092 /*
3093  * Queue up an outgoing message on the given connection.
3094  */
3095 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3096 {
3097         /* set src+dst */
3098         msg->hdr.src = con->msgr->inst.name;
3099         BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3100         msg->needs_out_seq = true;
3101 
3102         mutex_lock(&con->mutex);
3103 
3104         if (con->state == CON_STATE_CLOSED) {
3105                 dout("con_send %p closed, dropping %p\n", con, msg);
3106                 ceph_msg_put(msg);
3107                 mutex_unlock(&con->mutex);
3108                 return;
3109         }
3110 
3111         msg_con_set(msg, con);
3112 
3113         BUG_ON(!list_empty(&msg->list_head));
3114         list_add_tail(&msg->list_head, &con->out_queue);
3115         dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3116              ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3117              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3118              le32_to_cpu(msg->hdr.front_len),
3119              le32_to_cpu(msg->hdr.middle_len),
3120              le32_to_cpu(msg->hdr.data_len));
3121 
3122         clear_standby(con);
3123         mutex_unlock(&con->mutex);
3124 
3125         /* if there wasn't anything waiting to send before, queue
3126          * new work */
3127         if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3128                 queue_con(con);
3129 }
3130 EXPORT_SYMBOL(ceph_con_send);
3131 
3132 /*
3133  * Revoke a message that was previously queued for send
3134  */
3135 void ceph_msg_revoke(struct ceph_msg *msg)
3136 {
3137         struct ceph_connection *con = msg->con;
3138 
3139         if (!con) {
3140                 dout("%s msg %p null con\n", __func__, msg);
3141                 return;         /* Message not in our possession */
3142         }
3143 
3144         mutex_lock(&con->mutex);
3145         if (!list_empty(&msg->list_head)) {
3146                 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3147                 list_del_init(&msg->list_head);
3148                 msg->hdr.seq = 0;
3149 
3150                 ceph_msg_put(msg);
3151         }
3152         if (con->out_msg == msg) {
3153                 BUG_ON(con->out_skip);
3154                 /* footer */
3155                 if (con->out_msg_done) {
3156                         con->out_skip += con_out_kvec_skip(con);
3157                 } else {
3158                         BUG_ON(!msg->data_length);
3159                         con->out_skip += sizeof_footer(con);
3160                 }
3161                 /* data, middle, front */
3162                 if (msg->data_length)
3163                         con->out_skip += msg->cursor.total_resid;
3164                 if (msg->middle)
3165                         con->out_skip += con_out_kvec_skip(con);
3166                 con->out_skip += con_out_kvec_skip(con);
3167 
3168                 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3169                      __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3170                 msg->hdr.seq = 0;
3171                 con->out_msg = NULL;
3172                 ceph_msg_put(msg);
3173         }
3174 
3175         mutex_unlock(&con->mutex);
3176 }
3177 
3178 /*
3179  * Revoke a message that we may be reading data into
3180  */
3181 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3182 {
3183         struct ceph_connection *con = msg->con;
3184 
3185         if (!con) {
3186                 dout("%s msg %p null con\n", __func__, msg);
3187                 return;         /* Message not in our possession */
3188         }
3189 
3190         mutex_lock(&con->mutex);
3191         if (con->in_msg == msg) {
3192                 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3193                 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3194                 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3195 
3196                 /* skip rest of message */
3197                 dout("%s %p msg %p revoked\n", __func__, con, msg);
3198                 con->in_base_pos = con->in_base_pos -
3199                                 sizeof(struct ceph_msg_header) -
3200                                 front_len -
3201                                 middle_len -
3202                                 data_len -
3203                                 sizeof(struct ceph_msg_footer);
3204                 ceph_msg_put(con->in_msg);
3205                 con->in_msg = NULL;
3206                 con->in_tag = CEPH_MSGR_TAG_READY;
3207                 con->in_seq++;
3208         } else {
3209                 dout("%s %p in_msg %p msg %p no-op\n",
3210                      __func__, con, con->in_msg, msg);
3211         }
3212         mutex_unlock(&con->mutex);
3213 }
3214 
3215 /*
3216  * Queue a keepalive byte to ensure the tcp connection is alive.
3217  */
3218 void ceph_con_keepalive(struct ceph_connection *con)
3219 {
3220         dout("con_keepalive %p\n", con);
3221         mutex_lock(&con->mutex);
3222         clear_standby(con);
3223         mutex_unlock(&con->mutex);
3224         if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3225             con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3226                 queue_con(con);
3227 }
3228 EXPORT_SYMBOL(ceph_con_keepalive);
3229 
3230 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3231                                unsigned long interval)
3232 {
3233         if (interval > 0 &&
3234             (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3235                 struct timespec now;
3236                 struct timespec ts;
3237                 ktime_get_real_ts(&now);
3238                 jiffies_to_timespec(interval, &ts);
3239                 ts = timespec_add(con->last_keepalive_ack, ts);
3240                 return timespec_compare(&now, &ts) >= 0;
3241         }
3242         return false;
3243 }
3244 
3245 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3246 {
3247         struct ceph_msg_data *data;
3248 
3249         if (WARN_ON(!ceph_msg_data_type_valid(type)))
3250                 return NULL;
3251 
3252         data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3253         if (!data)
3254                 return NULL;
3255 
3256         data->type = type;
3257         INIT_LIST_HEAD(&data->links);
3258 
3259         return data;
3260 }
3261 
3262 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3263 {
3264         if (!data)
3265                 return;
3266 
3267         WARN_ON(!list_empty(&data->links));
3268         if (data->type == CEPH_MSG_DATA_PAGELIST)
3269                 ceph_pagelist_release(data->pagelist);
3270         kmem_cache_free(ceph_msg_data_cache, data);
3271 }
3272 
3273 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3274                 size_t length, size_t alignment)
3275 {
3276         struct ceph_msg_data *data;
3277 
3278         BUG_ON(!pages);
3279         BUG_ON(!length);
3280 
3281         data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3282         BUG_ON(!data);
3283         data->pages = pages;
3284         data->length = length;
3285         data->alignment = alignment & ~PAGE_MASK;
3286 
3287         list_add_tail(&data->links, &msg->data);
3288         msg->data_length += length;
3289 }
3290 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3291 
3292 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3293                                 struct ceph_pagelist *pagelist)
3294 {
3295         struct ceph_msg_data *data;
3296 
3297         BUG_ON(!pagelist);
3298         BUG_ON(!pagelist->length);
3299 
3300         data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3301         BUG_ON(!data);
3302         data->pagelist = pagelist;
3303 
3304         list_add_tail(&data->links, &msg->data);
3305         msg->data_length += pagelist->length;
3306 }
3307 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3308 
3309 #ifdef  CONFIG_BLOCK
3310 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
3311                            u32 length)
3312 {
3313         struct ceph_msg_data *data;
3314 
3315         data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3316         BUG_ON(!data);
3317         data->bio_pos = *bio_pos;
3318         data->bio_length = length;
3319 
3320         list_add_tail(&data->links, &msg->data);
3321         msg->data_length += length;
3322 }
3323 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3324 #endif  /* CONFIG_BLOCK */
3325 
3326 void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
3327                              struct ceph_bvec_iter *bvec_pos)
3328 {
3329         struct ceph_msg_data *data;
3330 
3331         data = ceph_msg_data_create(CEPH_MSG_DATA_BVECS);
3332         BUG_ON(!data);
3333         data->bvec_pos = *bvec_pos;
3334 
3335         list_add_tail(&data->links, &msg->data);
3336         msg->data_length += bvec_pos->iter.bi_size;
3337 }
3338 EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
3339 
3340 /*
3341  * construct a new message with given type, size
3342  * the new msg has a ref count of 1.
3343  */
3344 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3345                               bool can_fail)
3346 {
3347         struct ceph_msg *m;
3348 
3349         m = kmem_cache_zalloc(ceph_msg_cache, flags);
3350         if (m == NULL)
3351                 goto out;
3352 
3353         m->hdr.type = cpu_to_le16(type);
3354         m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3355         m->hdr.front_len = cpu_to_le32(front_len);
3356 
3357         INIT_LIST_HEAD(&m->list_head);
3358         kref_init(&m->kref);
3359         INIT_LIST_HEAD(&m->data);
3360 
3361         /* front */
3362         if (front_len) {
3363                 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3364                 if (m->front.iov_base == NULL) {
3365                         dout("ceph_msg_new can't allocate %d bytes\n",
3366                              front_len);
3367                         goto out2;
3368                 }
3369         } else {
3370                 m->front.iov_base = NULL;
3371         }
3372         m->front_alloc_len = m->front.iov_len = front_len;
3373 
3374         dout("ceph_msg_new %p front %d\n", m, front_len);
3375         return m;
3376 
3377 out2:
3378         ceph_msg_put(m);
3379 out:
3380         if (!can_fail) {
3381                 pr_err("msg_new can't create type %d front %d\n", type,
3382                        front_len);
3383                 WARN_ON(1);
3384         } else {
3385                 dout("msg_new can't create type %d front %d\n", type,
3386                      front_len);
3387         }
3388         return NULL;
3389 }
3390 EXPORT_SYMBOL(ceph_msg_new);
3391 
3392 /*
3393  * Allocate "middle" portion of a message, if it is needed and wasn't
3394  * allocated by alloc_msg.  This allows us to read a small fixed-size
3395  * per-type header in the front and then gracefully fail (i.e.,
3396  * propagate the error to the caller based on info in the front) when
3397  * the middle is too large.
3398  */
3399 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3400 {
3401         int type = le16_to_cpu(msg->hdr.type);
3402         int middle_len = le32_to_cpu(msg->hdr.middle_len);
3403 
3404         dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3405              ceph_msg_type_name(type), middle_len);
3406         BUG_ON(!middle_len);
3407         BUG_ON(msg->middle);
3408 
3409         msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3410         if (!msg->middle)
3411                 return -ENOMEM;
3412         return 0;
3413 }
3414 
3415 /*
3416  * Allocate a message for receiving an incoming message on a
3417  * connection, and save the result in con->in_msg.  Uses the
3418  * connection's private alloc_msg op if available.
3419  *
3420  * Returns 0 on success, or a negative error code.
3421  *
3422  * On success, if we set *skip = 1:
3423  *  - the next message should be skipped and ignored.
3424  *  - con->in_msg == NULL
3425  * or if we set *skip = 0:
3426  *  - con->in_msg is non-null.
3427  * On error (ENOMEM, EAGAIN, ...),
3428  *  - con->in_msg == NULL
3429  */
3430 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3431 {
3432         struct ceph_msg_header *hdr = &con->in_hdr;
3433         int middle_len = le32_to_cpu(hdr->middle_len);
3434         struct ceph_msg *msg;
3435         int ret = 0;
3436 
3437         BUG_ON(con->in_msg != NULL);
3438         BUG_ON(!con->ops->alloc_msg);
3439 
3440         mutex_unlock(&con->mutex);
3441         msg = con->ops->alloc_msg(con, hdr, skip);
3442         mutex_lock(&con->mutex);
3443         if (con->state != CON_STATE_OPEN) {
3444                 if (msg)
3445                         ceph_msg_put(msg);
3446                 return -EAGAIN;
3447         }
3448         if (msg) {
3449                 BUG_ON(*skip);
3450                 msg_con_set(msg, con);
3451                 con->in_msg = msg;
3452         } else {
3453                 /*
3454                  * Null message pointer means either we should skip
3455                  * this message or we couldn't allocate memory.  The
3456                  * former is not an error.
3457                  */
3458                 if (*skip)
3459                         return 0;
3460 
3461                 con->error_msg = "error allocating memory for incoming message";
3462                 return -ENOMEM;
3463         }
3464         memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3465 
3466         if (middle_len && !con->in_msg->middle) {
3467                 ret = ceph_alloc_middle(con, con->in_msg);
3468                 if (ret < 0) {
3469                         ceph_msg_put(con->in_msg);
3470                         con->in_msg = NULL;
3471                 }
3472         }
3473 
3474         return ret;
3475 }
3476 
3477 
3478 /*
3479  * Free a generically kmalloc'd message.
3480  */
3481 static void ceph_msg_free(struct ceph_msg *m)
3482 {
3483         dout("%s %p\n", __func__, m);
3484         kvfree(m->front.iov_base);
3485         kmem_cache_free(ceph_msg_cache, m);
3486 }
3487 
3488 static void ceph_msg_release(struct kref *kref)
3489 {
3490         struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3491         struct ceph_msg_data *data, *next;
3492 
3493         dout("%s %p\n", __func__, m);
3494         WARN_ON(!list_empty(&m->list_head));
3495 
3496         msg_con_set(m, NULL);
3497 
3498         /* drop middle, data, if any */
3499         if (m->middle) {
3500                 ceph_buffer_put(m->middle);
3501                 m->middle = NULL;
3502         }
3503 
3504         list_for_each_entry_safe(data, next, &m->data, links) {
3505                 list_del_init(&data->links);
3506                 ceph_msg_data_destroy(data);
3507         }
3508         m->data_length = 0;
3509 
3510         if (m->pool)
3511                 ceph_msgpool_put(m->pool, m);
3512         else
3513                 ceph_msg_free(m);
3514 }
3515 
3516 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3517 {
3518         dout("%s %p (was %d)\n", __func__, msg,
3519              kref_read(&msg->kref));
3520         kref_get(&msg->kref);
3521         return msg;
3522 }
3523 EXPORT_SYMBOL(ceph_msg_get);
3524 
3525 void ceph_msg_put(struct ceph_msg *msg)
3526 {
3527         dout("%s %p (was %d)\n", __func__, msg,
3528              kref_read(&msg->kref));
3529         kref_put(&msg->kref, ceph_msg_release);
3530 }
3531 EXPORT_SYMBOL(ceph_msg_put);
3532 
3533 void ceph_msg_dump(struct ceph_msg *msg)
3534 {
3535         pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3536                  msg->front_alloc_len, msg->data_length);
3537         print_hex_dump(KERN_DEBUG, "header: ",
3538                        DUMP_PREFIX_OFFSET, 16, 1,
3539                        &msg->hdr, sizeof(msg->hdr), true);
3540         print_hex_dump(KERN_DEBUG, " front: ",
3541                        DUMP_PREFIX_OFFSET, 16, 1,
3542                        msg->front.iov_base, msg->front.iov_len, true);
3543         if (msg->middle)
3544                 print_hex_dump(KERN_DEBUG, "middle: ",
3545                                DUMP_PREFIX_OFFSET, 16, 1,
3546                                msg->middle->vec.iov_base,
3547                                msg->middle->vec.iov_len, true);
3548         print_hex_dump(KERN_DEBUG, "footer: ",
3549                        DUMP_PREFIX_OFFSET, 16, 1,
3550                        &msg->footer, sizeof(msg->footer), true);
3551 }
3552 EXPORT_SYMBOL(ceph_msg_dump);
3553 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp