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

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

Version: ~ [ linux-5.1-rc1 ] ~ [ linux-5.0.3 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.30 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.107 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.164 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.176 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.136 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.63 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ 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 
160 /* static tag bytes (protocol control messages) */
161 static char tag_msg = CEPH_MSGR_TAG_MSG;
162 static char tag_ack = CEPH_MSGR_TAG_ACK;
163 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
164 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
165 
166 #ifdef CONFIG_LOCKDEP
167 static struct lock_class_key socket_class;
168 #endif
169 
170 static void queue_con(struct ceph_connection *con);
171 static void cancel_con(struct ceph_connection *con);
172 static void ceph_con_workfn(struct work_struct *);
173 static void con_fault(struct ceph_connection *con);
174 
175 /*
176  * Nicely render a sockaddr as a string.  An array of formatted
177  * strings is used, to approximate reentrancy.
178  */
179 #define ADDR_STR_COUNT_LOG      5       /* log2(# address strings in array) */
180 #define ADDR_STR_COUNT          (1 << ADDR_STR_COUNT_LOG)
181 #define ADDR_STR_COUNT_MASK     (ADDR_STR_COUNT - 1)
182 #define MAX_ADDR_STR_LEN        64      /* 54 is enough */
183 
184 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
185 static atomic_t addr_str_seq = ATOMIC_INIT(0);
186 
187 static struct page *zero_page;          /* used in certain error cases */
188 
189 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
190 {
191         int i;
192         char *s;
193         struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
194         struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
195 
196         i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
197         s = addr_str[i];
198 
199         switch (ss->ss_family) {
200         case AF_INET:
201                 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
202                          ntohs(in4->sin_port));
203                 break;
204 
205         case AF_INET6:
206                 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
207                          ntohs(in6->sin6_port));
208                 break;
209 
210         default:
211                 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
212                          ss->ss_family);
213         }
214 
215         return s;
216 }
217 EXPORT_SYMBOL(ceph_pr_addr);
218 
219 static void encode_my_addr(struct ceph_messenger *msgr)
220 {
221         memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
222         ceph_encode_addr(&msgr->my_enc_addr);
223 }
224 
225 /*
226  * work queue for all reading and writing to/from the socket.
227  */
228 static struct workqueue_struct *ceph_msgr_wq;
229 
230 static int ceph_msgr_slab_init(void)
231 {
232         BUG_ON(ceph_msg_cache);
233         ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
234         if (!ceph_msg_cache)
235                 return -ENOMEM;
236 
237         return 0;
238 }
239 
240 static void ceph_msgr_slab_exit(void)
241 {
242         BUG_ON(!ceph_msg_cache);
243         kmem_cache_destroy(ceph_msg_cache);
244         ceph_msg_cache = NULL;
245 }
246 
247 static void _ceph_msgr_exit(void)
248 {
249         if (ceph_msgr_wq) {
250                 destroy_workqueue(ceph_msgr_wq);
251                 ceph_msgr_wq = NULL;
252         }
253 
254         BUG_ON(zero_page == NULL);
255         put_page(zero_page);
256         zero_page = NULL;
257 
258         ceph_msgr_slab_exit();
259 }
260 
261 int __init ceph_msgr_init(void)
262 {
263         if (ceph_msgr_slab_init())
264                 return -ENOMEM;
265 
266         BUG_ON(zero_page != NULL);
267         zero_page = ZERO_PAGE(0);
268         get_page(zero_page);
269 
270         /*
271          * The number of active work items is limited by the number of
272          * connections, so leave @max_active at default.
273          */
274         ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
275         if (ceph_msgr_wq)
276                 return 0;
277 
278         pr_err("msgr_init failed to create workqueue\n");
279         _ceph_msgr_exit();
280 
281         return -ENOMEM;
282 }
283 
284 void ceph_msgr_exit(void)
285 {
286         BUG_ON(ceph_msgr_wq == NULL);
287 
288         _ceph_msgr_exit();
289 }
290 
291 void ceph_msgr_flush(void)
292 {
293         flush_workqueue(ceph_msgr_wq);
294 }
295 EXPORT_SYMBOL(ceph_msgr_flush);
296 
297 /* Connection socket state transition functions */
298 
299 static void con_sock_state_init(struct ceph_connection *con)
300 {
301         int old_state;
302 
303         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
304         if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
305                 printk("%s: unexpected old state %d\n", __func__, old_state);
306         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
307              CON_SOCK_STATE_CLOSED);
308 }
309 
310 static void con_sock_state_connecting(struct ceph_connection *con)
311 {
312         int old_state;
313 
314         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
315         if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
316                 printk("%s: unexpected old state %d\n", __func__, old_state);
317         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
318              CON_SOCK_STATE_CONNECTING);
319 }
320 
321 static void con_sock_state_connected(struct ceph_connection *con)
322 {
323         int old_state;
324 
325         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
326         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
327                 printk("%s: unexpected old state %d\n", __func__, old_state);
328         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
329              CON_SOCK_STATE_CONNECTED);
330 }
331 
332 static void con_sock_state_closing(struct ceph_connection *con)
333 {
334         int old_state;
335 
336         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
337         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
338                         old_state != CON_SOCK_STATE_CONNECTED &&
339                         old_state != CON_SOCK_STATE_CLOSING))
340                 printk("%s: unexpected old state %d\n", __func__, old_state);
341         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
342              CON_SOCK_STATE_CLOSING);
343 }
344 
345 static void con_sock_state_closed(struct ceph_connection *con)
346 {
347         int old_state;
348 
349         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
350         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
351                     old_state != CON_SOCK_STATE_CLOSING &&
352                     old_state != CON_SOCK_STATE_CONNECTING &&
353                     old_state != CON_SOCK_STATE_CLOSED))
354                 printk("%s: unexpected old state %d\n", __func__, old_state);
355         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
356              CON_SOCK_STATE_CLOSED);
357 }
358 
359 /*
360  * socket callback functions
361  */
362 
363 /* data available on socket, or listen socket received a connect */
364 static void ceph_sock_data_ready(struct sock *sk)
365 {
366         struct ceph_connection *con = sk->sk_user_data;
367         if (atomic_read(&con->msgr->stopping)) {
368                 return;
369         }
370 
371         if (sk->sk_state != TCP_CLOSE_WAIT) {
372                 dout("%s on %p state = %lu, queueing work\n", __func__,
373                      con, con->state);
374                 queue_con(con);
375         }
376 }
377 
378 /* socket has buffer space for writing */
379 static void ceph_sock_write_space(struct sock *sk)
380 {
381         struct ceph_connection *con = sk->sk_user_data;
382 
383         /* only queue to workqueue if there is data we want to write,
384          * and there is sufficient space in the socket buffer to accept
385          * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
386          * doesn't get called again until try_write() fills the socket
387          * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
388          * and net/core/stream.c:sk_stream_write_space().
389          */
390         if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
391                 if (sk_stream_is_writeable(sk)) {
392                         dout("%s %p queueing write work\n", __func__, con);
393                         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
394                         queue_con(con);
395                 }
396         } else {
397                 dout("%s %p nothing to write\n", __func__, con);
398         }
399 }
400 
401 /* socket's state has changed */
402 static void ceph_sock_state_change(struct sock *sk)
403 {
404         struct ceph_connection *con = sk->sk_user_data;
405 
406         dout("%s %p state = %lu sk_state = %u\n", __func__,
407              con, con->state, sk->sk_state);
408 
409         switch (sk->sk_state) {
410         case TCP_CLOSE:
411                 dout("%s TCP_CLOSE\n", __func__);
412                 /* fall through */
413         case TCP_CLOSE_WAIT:
414                 dout("%s TCP_CLOSE_WAIT\n", __func__);
415                 con_sock_state_closing(con);
416                 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
417                 queue_con(con);
418                 break;
419         case TCP_ESTABLISHED:
420                 dout("%s TCP_ESTABLISHED\n", __func__);
421                 con_sock_state_connected(con);
422                 queue_con(con);
423                 break;
424         default:        /* Everything else is uninteresting */
425                 break;
426         }
427 }
428 
429 /*
430  * set up socket callbacks
431  */
432 static void set_sock_callbacks(struct socket *sock,
433                                struct ceph_connection *con)
434 {
435         struct sock *sk = sock->sk;
436         sk->sk_user_data = con;
437         sk->sk_data_ready = ceph_sock_data_ready;
438         sk->sk_write_space = ceph_sock_write_space;
439         sk->sk_state_change = ceph_sock_state_change;
440 }
441 
442 
443 /*
444  * socket helpers
445  */
446 
447 /*
448  * initiate connection to a remote socket.
449  */
450 static int ceph_tcp_connect(struct ceph_connection *con)
451 {
452         struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
453         struct socket *sock;
454         unsigned int noio_flag;
455         int ret;
456 
457         BUG_ON(con->sock);
458 
459         /* sock_create_kern() allocates with GFP_KERNEL */
460         noio_flag = memalloc_noio_save();
461         ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
462                                SOCK_STREAM, IPPROTO_TCP, &sock);
463         memalloc_noio_restore(noio_flag);
464         if (ret)
465                 return ret;
466         sock->sk->sk_allocation = GFP_NOFS;
467 
468 #ifdef CONFIG_LOCKDEP
469         lockdep_set_class(&sock->sk->sk_lock, &socket_class);
470 #endif
471 
472         set_sock_callbacks(sock, con);
473 
474         dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
475 
476         con_sock_state_connecting(con);
477         ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
478                                  O_NONBLOCK);
479         if (ret == -EINPROGRESS) {
480                 dout("connect %s EINPROGRESS sk_state = %u\n",
481                      ceph_pr_addr(&con->peer_addr.in_addr),
482                      sock->sk->sk_state);
483         } else if (ret < 0) {
484                 pr_err("connect %s error %d\n",
485                        ceph_pr_addr(&con->peer_addr.in_addr), ret);
486                 sock_release(sock);
487                 return ret;
488         }
489 
490         if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
491                 int optval = 1;
492 
493                 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
494                                         (char *)&optval, sizeof(optval));
495                 if (ret)
496                         pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
497                                ret);
498         }
499 
500         con->sock = sock;
501         return 0;
502 }
503 
504 /*
505  * If @buf is NULL, discard up to @len bytes.
506  */
507 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
508 {
509         struct kvec iov = {buf, len};
510         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
511         int r;
512 
513         if (!buf)
514                 msg.msg_flags |= MSG_TRUNC;
515 
516         iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, len);
517         r = sock_recvmsg(sock, &msg, msg.msg_flags);
518         if (r == -EAGAIN)
519                 r = 0;
520         return r;
521 }
522 
523 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
524                      int page_offset, size_t length)
525 {
526         struct bio_vec bvec = {
527                 .bv_page = page,
528                 .bv_offset = page_offset,
529                 .bv_len = length
530         };
531         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
532         int r;
533 
534         BUG_ON(page_offset + length > PAGE_SIZE);
535         iov_iter_bvec(&msg.msg_iter, READ, &bvec, 1, length);
536         r = sock_recvmsg(sock, &msg, msg.msg_flags);
537         if (r == -EAGAIN)
538                 r = 0;
539         return r;
540 }
541 
542 /*
543  * write something.  @more is true if caller will be sending more data
544  * shortly.
545  */
546 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
547                             size_t kvlen, size_t len, bool more)
548 {
549         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
550         int r;
551 
552         if (more)
553                 msg.msg_flags |= MSG_MORE;
554         else
555                 msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
556 
557         r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
558         if (r == -EAGAIN)
559                 r = 0;
560         return r;
561 }
562 
563 /*
564  * @more: either or both of MSG_MORE and MSG_SENDPAGE_NOTLAST
565  */
566 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
567                              int offset, size_t size, int more)
568 {
569         ssize_t (*sendpage)(struct socket *sock, struct page *page,
570                             int offset, size_t size, int flags);
571         int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more;
572         int ret;
573 
574         /*
575          * sendpage cannot properly handle pages with page_count == 0,
576          * we need to fall back to sendmsg if that's the case.
577          *
578          * Same goes for slab pages: skb_can_coalesce() allows
579          * coalescing neighboring slab objects into a single frag which
580          * triggers one of hardened usercopy checks.
581          */
582         if (page_count(page) >= 1 && !PageSlab(page))
583                 sendpage = sock->ops->sendpage;
584         else
585                 sendpage = sock_no_sendpage;
586 
587         ret = sendpage(sock, page, offset, size, flags);
588         if (ret == -EAGAIN)
589                 ret = 0;
590 
591         return ret;
592 }
593 
594 /*
595  * Shutdown/close the socket for the given connection.
596  */
597 static int con_close_socket(struct ceph_connection *con)
598 {
599         int rc = 0;
600 
601         dout("con_close_socket on %p sock %p\n", con, con->sock);
602         if (con->sock) {
603                 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
604                 sock_release(con->sock);
605                 con->sock = NULL;
606         }
607 
608         /*
609          * Forcibly clear the SOCK_CLOSED flag.  It gets set
610          * independent of the connection mutex, and we could have
611          * received a socket close event before we had the chance to
612          * shut the socket down.
613          */
614         con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
615 
616         con_sock_state_closed(con);
617         return rc;
618 }
619 
620 /*
621  * Reset a connection.  Discard all incoming and outgoing messages
622  * and clear *_seq state.
623  */
624 static void ceph_msg_remove(struct ceph_msg *msg)
625 {
626         list_del_init(&msg->list_head);
627 
628         ceph_msg_put(msg);
629 }
630 static void ceph_msg_remove_list(struct list_head *head)
631 {
632         while (!list_empty(head)) {
633                 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
634                                                         list_head);
635                 ceph_msg_remove(msg);
636         }
637 }
638 
639 static void reset_connection(struct ceph_connection *con)
640 {
641         /* reset connection, out_queue, msg_ and connect_seq */
642         /* discard existing out_queue and msg_seq */
643         dout("reset_connection %p\n", con);
644         ceph_msg_remove_list(&con->out_queue);
645         ceph_msg_remove_list(&con->out_sent);
646 
647         if (con->in_msg) {
648                 BUG_ON(con->in_msg->con != con);
649                 ceph_msg_put(con->in_msg);
650                 con->in_msg = NULL;
651         }
652 
653         con->connect_seq = 0;
654         con->out_seq = 0;
655         if (con->out_msg) {
656                 BUG_ON(con->out_msg->con != con);
657                 ceph_msg_put(con->out_msg);
658                 con->out_msg = NULL;
659         }
660         con->in_seq = 0;
661         con->in_seq_acked = 0;
662 
663         con->out_skip = 0;
664 }
665 
666 /*
667  * mark a peer down.  drop any open connections.
668  */
669 void ceph_con_close(struct ceph_connection *con)
670 {
671         mutex_lock(&con->mutex);
672         dout("con_close %p peer %s\n", con,
673              ceph_pr_addr(&con->peer_addr.in_addr));
674         con->state = CON_STATE_CLOSED;
675 
676         con_flag_clear(con, CON_FLAG_LOSSYTX);  /* so we retry next connect */
677         con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
678         con_flag_clear(con, CON_FLAG_WRITE_PENDING);
679         con_flag_clear(con, CON_FLAG_BACKOFF);
680 
681         reset_connection(con);
682         con->peer_global_seq = 0;
683         cancel_con(con);
684         con_close_socket(con);
685         mutex_unlock(&con->mutex);
686 }
687 EXPORT_SYMBOL(ceph_con_close);
688 
689 /*
690  * Reopen a closed connection, with a new peer address.
691  */
692 void ceph_con_open(struct ceph_connection *con,
693                    __u8 entity_type, __u64 entity_num,
694                    struct ceph_entity_addr *addr)
695 {
696         mutex_lock(&con->mutex);
697         dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
698 
699         WARN_ON(con->state != CON_STATE_CLOSED);
700         con->state = CON_STATE_PREOPEN;
701 
702         con->peer_name.type = (__u8) entity_type;
703         con->peer_name.num = cpu_to_le64(entity_num);
704 
705         memcpy(&con->peer_addr, addr, sizeof(*addr));
706         con->delay = 0;      /* reset backoff memory */
707         mutex_unlock(&con->mutex);
708         queue_con(con);
709 }
710 EXPORT_SYMBOL(ceph_con_open);
711 
712 /*
713  * return true if this connection ever successfully opened
714  */
715 bool ceph_con_opened(struct ceph_connection *con)
716 {
717         return con->connect_seq > 0;
718 }
719 
720 /*
721  * initialize a new connection.
722  */
723 void ceph_con_init(struct ceph_connection *con, void *private,
724         const struct ceph_connection_operations *ops,
725         struct ceph_messenger *msgr)
726 {
727         dout("con_init %p\n", con);
728         memset(con, 0, sizeof(*con));
729         con->private = private;
730         con->ops = ops;
731         con->msgr = msgr;
732 
733         con_sock_state_init(con);
734 
735         mutex_init(&con->mutex);
736         INIT_LIST_HEAD(&con->out_queue);
737         INIT_LIST_HEAD(&con->out_sent);
738         INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
739 
740         con->state = CON_STATE_CLOSED;
741 }
742 EXPORT_SYMBOL(ceph_con_init);
743 
744 
745 /*
746  * We maintain a global counter to order connection attempts.  Get
747  * a unique seq greater than @gt.
748  */
749 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
750 {
751         u32 ret;
752 
753         spin_lock(&msgr->global_seq_lock);
754         if (msgr->global_seq < gt)
755                 msgr->global_seq = gt;
756         ret = ++msgr->global_seq;
757         spin_unlock(&msgr->global_seq_lock);
758         return ret;
759 }
760 
761 static void con_out_kvec_reset(struct ceph_connection *con)
762 {
763         BUG_ON(con->out_skip);
764 
765         con->out_kvec_left = 0;
766         con->out_kvec_bytes = 0;
767         con->out_kvec_cur = &con->out_kvec[0];
768 }
769 
770 static void con_out_kvec_add(struct ceph_connection *con,
771                                 size_t size, void *data)
772 {
773         int index = con->out_kvec_left;
774 
775         BUG_ON(con->out_skip);
776         BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
777 
778         con->out_kvec[index].iov_len = size;
779         con->out_kvec[index].iov_base = data;
780         con->out_kvec_left++;
781         con->out_kvec_bytes += size;
782 }
783 
784 /*
785  * Chop off a kvec from the end.  Return residual number of bytes for
786  * that kvec, i.e. how many bytes would have been written if the kvec
787  * hadn't been nuked.
788  */
789 static int con_out_kvec_skip(struct ceph_connection *con)
790 {
791         int off = con->out_kvec_cur - con->out_kvec;
792         int skip = 0;
793 
794         if (con->out_kvec_bytes > 0) {
795                 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
796                 BUG_ON(con->out_kvec_bytes < skip);
797                 BUG_ON(!con->out_kvec_left);
798                 con->out_kvec_bytes -= skip;
799                 con->out_kvec_left--;
800         }
801 
802         return skip;
803 }
804 
805 #ifdef CONFIG_BLOCK
806 
807 /*
808  * For a bio data item, a piece is whatever remains of the next
809  * entry in the current bio iovec, or the first entry in the next
810  * bio in the list.
811  */
812 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
813                                         size_t length)
814 {
815         struct ceph_msg_data *data = cursor->data;
816         struct ceph_bio_iter *it = &cursor->bio_iter;
817 
818         cursor->resid = min_t(size_t, length, data->bio_length);
819         *it = data->bio_pos;
820         if (cursor->resid < it->iter.bi_size)
821                 it->iter.bi_size = cursor->resid;
822 
823         BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
824         cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
825 }
826 
827 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
828                                                 size_t *page_offset,
829                                                 size_t *length)
830 {
831         struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
832                                            cursor->bio_iter.iter);
833 
834         *page_offset = bv.bv_offset;
835         *length = bv.bv_len;
836         return bv.bv_page;
837 }
838 
839 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
840                                         size_t bytes)
841 {
842         struct ceph_bio_iter *it = &cursor->bio_iter;
843 
844         BUG_ON(bytes > cursor->resid);
845         BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
846         cursor->resid -= bytes;
847         bio_advance_iter(it->bio, &it->iter, bytes);
848 
849         if (!cursor->resid) {
850                 BUG_ON(!cursor->last_piece);
851                 return false;   /* no more data */
852         }
853 
854         if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done))
855                 return false;   /* more bytes to process in this segment */
856 
857         if (!it->iter.bi_size) {
858                 it->bio = it->bio->bi_next;
859                 it->iter = it->bio->bi_iter;
860                 if (cursor->resid < it->iter.bi_size)
861                         it->iter.bi_size = cursor->resid;
862         }
863 
864         BUG_ON(cursor->last_piece);
865         BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
866         cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
867         return true;
868 }
869 #endif /* CONFIG_BLOCK */
870 
871 static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
872                                         size_t length)
873 {
874         struct ceph_msg_data *data = cursor->data;
875         struct bio_vec *bvecs = data->bvec_pos.bvecs;
876 
877         cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
878         cursor->bvec_iter = data->bvec_pos.iter;
879         cursor->bvec_iter.bi_size = cursor->resid;
880 
881         BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
882         cursor->last_piece =
883             cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
884 }
885 
886 static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
887                                                 size_t *page_offset,
888                                                 size_t *length)
889 {
890         struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
891                                            cursor->bvec_iter);
892 
893         *page_offset = bv.bv_offset;
894         *length = bv.bv_len;
895         return bv.bv_page;
896 }
897 
898 static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
899                                         size_t bytes)
900 {
901         struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
902 
903         BUG_ON(bytes > cursor->resid);
904         BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
905         cursor->resid -= bytes;
906         bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
907 
908         if (!cursor->resid) {
909                 BUG_ON(!cursor->last_piece);
910                 return false;   /* no more data */
911         }
912 
913         if (!bytes || cursor->bvec_iter.bi_bvec_done)
914                 return false;   /* more bytes to process in this segment */
915 
916         BUG_ON(cursor->last_piece);
917         BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
918         cursor->last_piece =
919             cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
920         return true;
921 }
922 
923 /*
924  * For a page array, a piece comes from the first page in the array
925  * that has not already been fully consumed.
926  */
927 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
928                                         size_t length)
929 {
930         struct ceph_msg_data *data = cursor->data;
931         int page_count;
932 
933         BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
934 
935         BUG_ON(!data->pages);
936         BUG_ON(!data->length);
937 
938         cursor->resid = min(length, data->length);
939         page_count = calc_pages_for(data->alignment, (u64)data->length);
940         cursor->page_offset = data->alignment & ~PAGE_MASK;
941         cursor->page_index = 0;
942         BUG_ON(page_count > (int)USHRT_MAX);
943         cursor->page_count = (unsigned short)page_count;
944         BUG_ON(length > SIZE_MAX - cursor->page_offset);
945         cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
946 }
947 
948 static struct page *
949 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
950                                         size_t *page_offset, size_t *length)
951 {
952         struct ceph_msg_data *data = cursor->data;
953 
954         BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
955 
956         BUG_ON(cursor->page_index >= cursor->page_count);
957         BUG_ON(cursor->page_offset >= PAGE_SIZE);
958 
959         *page_offset = cursor->page_offset;
960         if (cursor->last_piece)
961                 *length = cursor->resid;
962         else
963                 *length = PAGE_SIZE - *page_offset;
964 
965         return data->pages[cursor->page_index];
966 }
967 
968 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
969                                                 size_t bytes)
970 {
971         BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
972 
973         BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
974 
975         /* Advance the cursor page offset */
976 
977         cursor->resid -= bytes;
978         cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
979         if (!bytes || cursor->page_offset)
980                 return false;   /* more bytes to process in the current page */
981 
982         if (!cursor->resid)
983                 return false;   /* no more data */
984 
985         /* Move on to the next page; offset is already at 0 */
986 
987         BUG_ON(cursor->page_index >= cursor->page_count);
988         cursor->page_index++;
989         cursor->last_piece = cursor->resid <= PAGE_SIZE;
990 
991         return true;
992 }
993 
994 /*
995  * For a pagelist, a piece is whatever remains to be consumed in the
996  * first page in the list, or the front of the next page.
997  */
998 static void
999 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
1000                                         size_t length)
1001 {
1002         struct ceph_msg_data *data = cursor->data;
1003         struct ceph_pagelist *pagelist;
1004         struct page *page;
1005 
1006         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1007 
1008         pagelist = data->pagelist;
1009         BUG_ON(!pagelist);
1010 
1011         if (!length)
1012                 return;         /* pagelist can be assigned but empty */
1013 
1014         BUG_ON(list_empty(&pagelist->head));
1015         page = list_first_entry(&pagelist->head, struct page, lru);
1016 
1017         cursor->resid = min(length, pagelist->length);
1018         cursor->page = page;
1019         cursor->offset = 0;
1020         cursor->last_piece = cursor->resid <= PAGE_SIZE;
1021 }
1022 
1023 static struct page *
1024 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1025                                 size_t *page_offset, size_t *length)
1026 {
1027         struct ceph_msg_data *data = cursor->data;
1028         struct ceph_pagelist *pagelist;
1029 
1030         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1031 
1032         pagelist = data->pagelist;
1033         BUG_ON(!pagelist);
1034 
1035         BUG_ON(!cursor->page);
1036         BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1037 
1038         /* offset of first page in pagelist is always 0 */
1039         *page_offset = cursor->offset & ~PAGE_MASK;
1040         if (cursor->last_piece)
1041                 *length = cursor->resid;
1042         else
1043                 *length = PAGE_SIZE - *page_offset;
1044 
1045         return cursor->page;
1046 }
1047 
1048 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1049                                                 size_t bytes)
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->offset + cursor->resid != pagelist->length);
1060         BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1061 
1062         /* Advance the cursor offset */
1063 
1064         cursor->resid -= bytes;
1065         cursor->offset += bytes;
1066         /* offset of first page in pagelist is always 0 */
1067         if (!bytes || cursor->offset & ~PAGE_MASK)
1068                 return false;   /* more bytes to process in the current page */
1069 
1070         if (!cursor->resid)
1071                 return false;   /* no more data */
1072 
1073         /* Move on to the next page */
1074 
1075         BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1076         cursor->page = list_next_entry(cursor->page, lru);
1077         cursor->last_piece = cursor->resid <= PAGE_SIZE;
1078 
1079         return true;
1080 }
1081 
1082 /*
1083  * Message data is handled (sent or received) in pieces, where each
1084  * piece resides on a single page.  The network layer might not
1085  * consume an entire piece at once.  A data item's cursor keeps
1086  * track of which piece is next to process and how much remains to
1087  * be processed in that piece.  It also tracks whether the current
1088  * piece is the last one in the data item.
1089  */
1090 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1091 {
1092         size_t length = cursor->total_resid;
1093 
1094         switch (cursor->data->type) {
1095         case CEPH_MSG_DATA_PAGELIST:
1096                 ceph_msg_data_pagelist_cursor_init(cursor, length);
1097                 break;
1098         case CEPH_MSG_DATA_PAGES:
1099                 ceph_msg_data_pages_cursor_init(cursor, length);
1100                 break;
1101 #ifdef CONFIG_BLOCK
1102         case CEPH_MSG_DATA_BIO:
1103                 ceph_msg_data_bio_cursor_init(cursor, length);
1104                 break;
1105 #endif /* CONFIG_BLOCK */
1106         case CEPH_MSG_DATA_BVECS:
1107                 ceph_msg_data_bvecs_cursor_init(cursor, length);
1108                 break;
1109         case CEPH_MSG_DATA_NONE:
1110         default:
1111                 /* BUG(); */
1112                 break;
1113         }
1114         cursor->need_crc = true;
1115 }
1116 
1117 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1118 {
1119         struct ceph_msg_data_cursor *cursor = &msg->cursor;
1120 
1121         BUG_ON(!length);
1122         BUG_ON(length > msg->data_length);
1123         BUG_ON(!msg->num_data_items);
1124 
1125         cursor->total_resid = length;
1126         cursor->data = msg->data;
1127 
1128         __ceph_msg_data_cursor_init(cursor);
1129 }
1130 
1131 /*
1132  * Return the page containing the next piece to process for a given
1133  * data item, and supply the page offset and length of that piece.
1134  * Indicate whether this is the last piece in this data item.
1135  */
1136 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1137                                         size_t *page_offset, size_t *length,
1138                                         bool *last_piece)
1139 {
1140         struct page *page;
1141 
1142         switch (cursor->data->type) {
1143         case CEPH_MSG_DATA_PAGELIST:
1144                 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1145                 break;
1146         case CEPH_MSG_DATA_PAGES:
1147                 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1148                 break;
1149 #ifdef CONFIG_BLOCK
1150         case CEPH_MSG_DATA_BIO:
1151                 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1152                 break;
1153 #endif /* CONFIG_BLOCK */
1154         case CEPH_MSG_DATA_BVECS:
1155                 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1156                 break;
1157         case CEPH_MSG_DATA_NONE:
1158         default:
1159                 page = NULL;
1160                 break;
1161         }
1162 
1163         BUG_ON(!page);
1164         BUG_ON(*page_offset + *length > PAGE_SIZE);
1165         BUG_ON(!*length);
1166         BUG_ON(*length > cursor->resid);
1167         if (last_piece)
1168                 *last_piece = cursor->last_piece;
1169 
1170         return page;
1171 }
1172 
1173 /*
1174  * Returns true if the result moves the cursor on to the next piece
1175  * of the data item.
1176  */
1177 static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1178                                   size_t bytes)
1179 {
1180         bool new_piece;
1181 
1182         BUG_ON(bytes > cursor->resid);
1183         switch (cursor->data->type) {
1184         case CEPH_MSG_DATA_PAGELIST:
1185                 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1186                 break;
1187         case CEPH_MSG_DATA_PAGES:
1188                 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1189                 break;
1190 #ifdef CONFIG_BLOCK
1191         case CEPH_MSG_DATA_BIO:
1192                 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1193                 break;
1194 #endif /* CONFIG_BLOCK */
1195         case CEPH_MSG_DATA_BVECS:
1196                 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1197                 break;
1198         case CEPH_MSG_DATA_NONE:
1199         default:
1200                 BUG();
1201                 break;
1202         }
1203         cursor->total_resid -= bytes;
1204 
1205         if (!cursor->resid && cursor->total_resid) {
1206                 WARN_ON(!cursor->last_piece);
1207                 cursor->data++;
1208                 __ceph_msg_data_cursor_init(cursor);
1209                 new_piece = true;
1210         }
1211         cursor->need_crc = new_piece;
1212 }
1213 
1214 static size_t sizeof_footer(struct ceph_connection *con)
1215 {
1216         return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1217             sizeof(struct ceph_msg_footer) :
1218             sizeof(struct ceph_msg_footer_old);
1219 }
1220 
1221 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1222 {
1223         /* Initialize data cursor */
1224 
1225         ceph_msg_data_cursor_init(msg, (size_t)data_len);
1226 }
1227 
1228 /*
1229  * Prepare footer for currently outgoing message, and finish things
1230  * off.  Assumes out_kvec* are already valid.. we just add on to the end.
1231  */
1232 static void prepare_write_message_footer(struct ceph_connection *con)
1233 {
1234         struct ceph_msg *m = con->out_msg;
1235 
1236         m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1237 
1238         dout("prepare_write_message_footer %p\n", con);
1239         con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1240         if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1241                 if (con->ops->sign_message)
1242                         con->ops->sign_message(m);
1243                 else
1244                         m->footer.sig = 0;
1245         } else {
1246                 m->old_footer.flags = m->footer.flags;
1247         }
1248         con->out_more = m->more_to_follow;
1249         con->out_msg_done = true;
1250 }
1251 
1252 /*
1253  * Prepare headers for the next outgoing message.
1254  */
1255 static void prepare_write_message(struct ceph_connection *con)
1256 {
1257         struct ceph_msg *m;
1258         u32 crc;
1259 
1260         con_out_kvec_reset(con);
1261         con->out_msg_done = false;
1262 
1263         /* Sneak an ack in there first?  If we can get it into the same
1264          * TCP packet that's a good thing. */
1265         if (con->in_seq > con->in_seq_acked) {
1266                 con->in_seq_acked = con->in_seq;
1267                 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1268                 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1269                 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1270                         &con->out_temp_ack);
1271         }
1272 
1273         BUG_ON(list_empty(&con->out_queue));
1274         m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1275         con->out_msg = m;
1276         BUG_ON(m->con != con);
1277 
1278         /* put message on sent list */
1279         ceph_msg_get(m);
1280         list_move_tail(&m->list_head, &con->out_sent);
1281 
1282         /*
1283          * only assign outgoing seq # if we haven't sent this message
1284          * yet.  if it is requeued, resend with it's original seq.
1285          */
1286         if (m->needs_out_seq) {
1287                 m->hdr.seq = cpu_to_le64(++con->out_seq);
1288                 m->needs_out_seq = false;
1289 
1290                 if (con->ops->reencode_message)
1291                         con->ops->reencode_message(m);
1292         }
1293 
1294         dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1295              m, con->out_seq, le16_to_cpu(m->hdr.type),
1296              le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1297              m->data_length);
1298         WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1299         WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1300 
1301         /* tag + hdr + front + middle */
1302         con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1303         con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1304         con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1305 
1306         if (m->middle)
1307                 con_out_kvec_add(con, m->middle->vec.iov_len,
1308                         m->middle->vec.iov_base);
1309 
1310         /* fill in hdr crc and finalize hdr */
1311         crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1312         con->out_msg->hdr.crc = cpu_to_le32(crc);
1313         memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1314 
1315         /* fill in front and middle crc, footer */
1316         crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1317         con->out_msg->footer.front_crc = cpu_to_le32(crc);
1318         if (m->middle) {
1319                 crc = crc32c(0, m->middle->vec.iov_base,
1320                                 m->middle->vec.iov_len);
1321                 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1322         } else
1323                 con->out_msg->footer.middle_crc = 0;
1324         dout("%s front_crc %u middle_crc %u\n", __func__,
1325              le32_to_cpu(con->out_msg->footer.front_crc),
1326              le32_to_cpu(con->out_msg->footer.middle_crc));
1327         con->out_msg->footer.flags = 0;
1328 
1329         /* is there a data payload? */
1330         con->out_msg->footer.data_crc = 0;
1331         if (m->data_length) {
1332                 prepare_message_data(con->out_msg, m->data_length);
1333                 con->out_more = 1;  /* data + footer will follow */
1334         } else {
1335                 /* no, queue up footer too and be done */
1336                 prepare_write_message_footer(con);
1337         }
1338 
1339         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1340 }
1341 
1342 /*
1343  * Prepare an ack.
1344  */
1345 static void prepare_write_ack(struct ceph_connection *con)
1346 {
1347         dout("prepare_write_ack %p %llu -> %llu\n", con,
1348              con->in_seq_acked, con->in_seq);
1349         con->in_seq_acked = con->in_seq;
1350 
1351         con_out_kvec_reset(con);
1352 
1353         con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1354 
1355         con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1356         con_out_kvec_add(con, sizeof (con->out_temp_ack),
1357                                 &con->out_temp_ack);
1358 
1359         con->out_more = 1;  /* more will follow.. eventually.. */
1360         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1361 }
1362 
1363 /*
1364  * Prepare to share the seq during handshake
1365  */
1366 static void prepare_write_seq(struct ceph_connection *con)
1367 {
1368         dout("prepare_write_seq %p %llu -> %llu\n", con,
1369              con->in_seq_acked, con->in_seq);
1370         con->in_seq_acked = con->in_seq;
1371 
1372         con_out_kvec_reset(con);
1373 
1374         con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1375         con_out_kvec_add(con, sizeof (con->out_temp_ack),
1376                          &con->out_temp_ack);
1377 
1378         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1379 }
1380 
1381 /*
1382  * Prepare to write keepalive byte.
1383  */
1384 static void prepare_write_keepalive(struct ceph_connection *con)
1385 {
1386         dout("prepare_write_keepalive %p\n", con);
1387         con_out_kvec_reset(con);
1388         if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1389                 struct timespec64 now;
1390 
1391                 ktime_get_real_ts64(&now);
1392                 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1393                 ceph_encode_timespec64(&con->out_temp_keepalive2, &now);
1394                 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1395                                  &con->out_temp_keepalive2);
1396         } else {
1397                 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1398         }
1399         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1400 }
1401 
1402 /*
1403  * Connection negotiation.
1404  */
1405 
1406 static int get_connect_authorizer(struct ceph_connection *con)
1407 {
1408         struct ceph_auth_handshake *auth;
1409         int auth_proto;
1410 
1411         if (!con->ops->get_authorizer) {
1412                 con->auth = NULL;
1413                 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1414                 con->out_connect.authorizer_len = 0;
1415                 return 0;
1416         }
1417 
1418         auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
1419         if (IS_ERR(auth))
1420                 return PTR_ERR(auth);
1421 
1422         con->auth = auth;
1423         con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1424         con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1425         return 0;
1426 }
1427 
1428 /*
1429  * We connected to a peer and are saying hello.
1430  */
1431 static void prepare_write_banner(struct ceph_connection *con)
1432 {
1433         con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1434         con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1435                                         &con->msgr->my_enc_addr);
1436 
1437         con->out_more = 0;
1438         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1439 }
1440 
1441 static void __prepare_write_connect(struct ceph_connection *con)
1442 {
1443         con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1444         if (con->auth)
1445                 con_out_kvec_add(con, con->auth->authorizer_buf_len,
1446                                  con->auth->authorizer_buf);
1447 
1448         con->out_more = 0;
1449         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1450 }
1451 
1452 static int prepare_write_connect(struct ceph_connection *con)
1453 {
1454         unsigned int global_seq = get_global_seq(con->msgr, 0);
1455         int proto;
1456         int ret;
1457 
1458         switch (con->peer_name.type) {
1459         case CEPH_ENTITY_TYPE_MON:
1460                 proto = CEPH_MONC_PROTOCOL;
1461                 break;
1462         case CEPH_ENTITY_TYPE_OSD:
1463                 proto = CEPH_OSDC_PROTOCOL;
1464                 break;
1465         case CEPH_ENTITY_TYPE_MDS:
1466                 proto = CEPH_MDSC_PROTOCOL;
1467                 break;
1468         default:
1469                 BUG();
1470         }
1471 
1472         dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1473              con->connect_seq, global_seq, proto);
1474 
1475         con->out_connect.features =
1476             cpu_to_le64(from_msgr(con->msgr)->supported_features);
1477         con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1478         con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1479         con->out_connect.global_seq = cpu_to_le32(global_seq);
1480         con->out_connect.protocol_version = cpu_to_le32(proto);
1481         con->out_connect.flags = 0;
1482 
1483         ret = get_connect_authorizer(con);
1484         if (ret)
1485                 return ret;
1486 
1487         __prepare_write_connect(con);
1488         return 0;
1489 }
1490 
1491 /*
1492  * write as much of pending kvecs to the socket as we can.
1493  *  1 -> done
1494  *  0 -> socket full, but more to do
1495  * <0 -> error
1496  */
1497 static int write_partial_kvec(struct ceph_connection *con)
1498 {
1499         int ret;
1500 
1501         dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1502         while (con->out_kvec_bytes > 0) {
1503                 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1504                                        con->out_kvec_left, con->out_kvec_bytes,
1505                                        con->out_more);
1506                 if (ret <= 0)
1507                         goto out;
1508                 con->out_kvec_bytes -= ret;
1509                 if (con->out_kvec_bytes == 0)
1510                         break;            /* done */
1511 
1512                 /* account for full iov entries consumed */
1513                 while (ret >= con->out_kvec_cur->iov_len) {
1514                         BUG_ON(!con->out_kvec_left);
1515                         ret -= con->out_kvec_cur->iov_len;
1516                         con->out_kvec_cur++;
1517                         con->out_kvec_left--;
1518                 }
1519                 /* and for a partially-consumed entry */
1520                 if (ret) {
1521                         con->out_kvec_cur->iov_len -= ret;
1522                         con->out_kvec_cur->iov_base += ret;
1523                 }
1524         }
1525         con->out_kvec_left = 0;
1526         ret = 1;
1527 out:
1528         dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1529              con->out_kvec_bytes, con->out_kvec_left, ret);
1530         return ret;  /* done! */
1531 }
1532 
1533 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1534                                 unsigned int page_offset,
1535                                 unsigned int length)
1536 {
1537         char *kaddr;
1538 
1539         kaddr = kmap(page);
1540         BUG_ON(kaddr == NULL);
1541         crc = crc32c(crc, kaddr + page_offset, length);
1542         kunmap(page);
1543 
1544         return crc;
1545 }
1546 /*
1547  * Write as much message data payload as we can.  If we finish, queue
1548  * up the footer.
1549  *  1 -> done, footer is now queued in out_kvec[].
1550  *  0 -> socket full, but more to do
1551  * <0 -> error
1552  */
1553 static int write_partial_message_data(struct ceph_connection *con)
1554 {
1555         struct ceph_msg *msg = con->out_msg;
1556         struct ceph_msg_data_cursor *cursor = &msg->cursor;
1557         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1558         int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
1559         u32 crc;
1560 
1561         dout("%s %p msg %p\n", __func__, con, msg);
1562 
1563         if (!msg->num_data_items)
1564                 return -EINVAL;
1565 
1566         /*
1567          * Iterate through each page that contains data to be
1568          * written, and send as much as possible for each.
1569          *
1570          * If we are calculating the data crc (the default), we will
1571          * need to map the page.  If we have no pages, they have
1572          * been revoked, so use the zero page.
1573          */
1574         crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1575         while (cursor->total_resid) {
1576                 struct page *page;
1577                 size_t page_offset;
1578                 size_t length;
1579                 int ret;
1580 
1581                 if (!cursor->resid) {
1582                         ceph_msg_data_advance(cursor, 0);
1583                         continue;
1584                 }
1585 
1586                 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
1587                 if (length == cursor->total_resid)
1588                         more = MSG_MORE;
1589                 ret = ceph_tcp_sendpage(con->sock, page, page_offset, length,
1590                                         more);
1591                 if (ret <= 0) {
1592                         if (do_datacrc)
1593                                 msg->footer.data_crc = cpu_to_le32(crc);
1594 
1595                         return ret;
1596                 }
1597                 if (do_datacrc && cursor->need_crc)
1598                         crc = ceph_crc32c_page(crc, page, page_offset, length);
1599                 ceph_msg_data_advance(cursor, (size_t)ret);
1600         }
1601 
1602         dout("%s %p msg %p done\n", __func__, con, msg);
1603 
1604         /* prepare and queue up footer, too */
1605         if (do_datacrc)
1606                 msg->footer.data_crc = cpu_to_le32(crc);
1607         else
1608                 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1609         con_out_kvec_reset(con);
1610         prepare_write_message_footer(con);
1611 
1612         return 1;       /* must return > 0 to indicate success */
1613 }
1614 
1615 /*
1616  * write some zeros
1617  */
1618 static int write_partial_skip(struct ceph_connection *con)
1619 {
1620         int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
1621         int ret;
1622 
1623         dout("%s %p %d left\n", __func__, con, con->out_skip);
1624         while (con->out_skip > 0) {
1625                 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1626 
1627                 if (size == con->out_skip)
1628                         more = MSG_MORE;
1629                 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, more);
1630                 if (ret <= 0)
1631                         goto out;
1632                 con->out_skip -= ret;
1633         }
1634         ret = 1;
1635 out:
1636         return ret;
1637 }
1638 
1639 /*
1640  * Prepare to read connection handshake, or an ack.
1641  */
1642 static void prepare_read_banner(struct ceph_connection *con)
1643 {
1644         dout("prepare_read_banner %p\n", con);
1645         con->in_base_pos = 0;
1646 }
1647 
1648 static void prepare_read_connect(struct ceph_connection *con)
1649 {
1650         dout("prepare_read_connect %p\n", con);
1651         con->in_base_pos = 0;
1652 }
1653 
1654 static void prepare_read_ack(struct ceph_connection *con)
1655 {
1656         dout("prepare_read_ack %p\n", con);
1657         con->in_base_pos = 0;
1658 }
1659 
1660 static void prepare_read_seq(struct ceph_connection *con)
1661 {
1662         dout("prepare_read_seq %p\n", con);
1663         con->in_base_pos = 0;
1664         con->in_tag = CEPH_MSGR_TAG_SEQ;
1665 }
1666 
1667 static void prepare_read_tag(struct ceph_connection *con)
1668 {
1669         dout("prepare_read_tag %p\n", con);
1670         con->in_base_pos = 0;
1671         con->in_tag = CEPH_MSGR_TAG_READY;
1672 }
1673 
1674 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1675 {
1676         dout("prepare_read_keepalive_ack %p\n", con);
1677         con->in_base_pos = 0;
1678 }
1679 
1680 /*
1681  * Prepare to read a message.
1682  */
1683 static int prepare_read_message(struct ceph_connection *con)
1684 {
1685         dout("prepare_read_message %p\n", con);
1686         BUG_ON(con->in_msg != NULL);
1687         con->in_base_pos = 0;
1688         con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1689         return 0;
1690 }
1691 
1692 
1693 static int read_partial(struct ceph_connection *con,
1694                         int end, int size, void *object)
1695 {
1696         while (con->in_base_pos < end) {
1697                 int left = end - con->in_base_pos;
1698                 int have = size - left;
1699                 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1700                 if (ret <= 0)
1701                         return ret;
1702                 con->in_base_pos += ret;
1703         }
1704         return 1;
1705 }
1706 
1707 
1708 /*
1709  * Read all or part of the connect-side handshake on a new connection
1710  */
1711 static int read_partial_banner(struct ceph_connection *con)
1712 {
1713         int size;
1714         int end;
1715         int ret;
1716 
1717         dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1718 
1719         /* peer's banner */
1720         size = strlen(CEPH_BANNER);
1721         end = size;
1722         ret = read_partial(con, end, size, con->in_banner);
1723         if (ret <= 0)
1724                 goto out;
1725 
1726         size = sizeof (con->actual_peer_addr);
1727         end += size;
1728         ret = read_partial(con, end, size, &con->actual_peer_addr);
1729         if (ret <= 0)
1730                 goto out;
1731 
1732         size = sizeof (con->peer_addr_for_me);
1733         end += size;
1734         ret = read_partial(con, end, size, &con->peer_addr_for_me);
1735         if (ret <= 0)
1736                 goto out;
1737 
1738 out:
1739         return ret;
1740 }
1741 
1742 static int read_partial_connect(struct ceph_connection *con)
1743 {
1744         int size;
1745         int end;
1746         int ret;
1747 
1748         dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1749 
1750         size = sizeof (con->in_reply);
1751         end = size;
1752         ret = read_partial(con, end, size, &con->in_reply);
1753         if (ret <= 0)
1754                 goto out;
1755 
1756         if (con->auth) {
1757                 size = le32_to_cpu(con->in_reply.authorizer_len);
1758                 if (size > con->auth->authorizer_reply_buf_len) {
1759                         pr_err("authorizer reply too big: %d > %zu\n", size,
1760                                con->auth->authorizer_reply_buf_len);
1761                         ret = -EINVAL;
1762                         goto out;
1763                 }
1764 
1765                 end += size;
1766                 ret = read_partial(con, end, size,
1767                                    con->auth->authorizer_reply_buf);
1768                 if (ret <= 0)
1769                         goto out;
1770         }
1771 
1772         dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1773              con, (int)con->in_reply.tag,
1774              le32_to_cpu(con->in_reply.connect_seq),
1775              le32_to_cpu(con->in_reply.global_seq));
1776 out:
1777         return ret;
1778 }
1779 
1780 /*
1781  * Verify the hello banner looks okay.
1782  */
1783 static int verify_hello(struct ceph_connection *con)
1784 {
1785         if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1786                 pr_err("connect to %s got bad banner\n",
1787                        ceph_pr_addr(&con->peer_addr.in_addr));
1788                 con->error_msg = "protocol error, bad banner";
1789                 return -1;
1790         }
1791         return 0;
1792 }
1793 
1794 static bool addr_is_blank(struct sockaddr_storage *ss)
1795 {
1796         struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1797         struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1798 
1799         switch (ss->ss_family) {
1800         case AF_INET:
1801                 return addr->s_addr == htonl(INADDR_ANY);
1802         case AF_INET6:
1803                 return ipv6_addr_any(addr6);
1804         default:
1805                 return true;
1806         }
1807 }
1808 
1809 static int addr_port(struct sockaddr_storage *ss)
1810 {
1811         switch (ss->ss_family) {
1812         case AF_INET:
1813                 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1814         case AF_INET6:
1815                 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1816         }
1817         return 0;
1818 }
1819 
1820 static void addr_set_port(struct sockaddr_storage *ss, int p)
1821 {
1822         switch (ss->ss_family) {
1823         case AF_INET:
1824                 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1825                 break;
1826         case AF_INET6:
1827                 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1828                 break;
1829         }
1830 }
1831 
1832 /*
1833  * Unlike other *_pton function semantics, zero indicates success.
1834  */
1835 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1836                 char delim, const char **ipend)
1837 {
1838         struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1839         struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1840 
1841         memset(ss, 0, sizeof(*ss));
1842 
1843         if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1844                 ss->ss_family = AF_INET;
1845                 return 0;
1846         }
1847 
1848         if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1849                 ss->ss_family = AF_INET6;
1850                 return 0;
1851         }
1852 
1853         return -EINVAL;
1854 }
1855 
1856 /*
1857  * Extract hostname string and resolve using kernel DNS facility.
1858  */
1859 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1860 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1861                 struct sockaddr_storage *ss, char delim, const char **ipend)
1862 {
1863         const char *end, *delim_p;
1864         char *colon_p, *ip_addr = NULL;
1865         int ip_len, ret;
1866 
1867         /*
1868          * The end of the hostname occurs immediately preceding the delimiter or
1869          * the port marker (':') where the delimiter takes precedence.
1870          */
1871         delim_p = memchr(name, delim, namelen);
1872         colon_p = memchr(name, ':', namelen);
1873 
1874         if (delim_p && colon_p)
1875                 end = delim_p < colon_p ? delim_p : colon_p;
1876         else if (!delim_p && colon_p)
1877                 end = colon_p;
1878         else {
1879                 end = delim_p;
1880                 if (!end) /* case: hostname:/ */
1881                         end = name + namelen;
1882         }
1883 
1884         if (end <= name)
1885                 return -EINVAL;
1886 
1887         /* do dns_resolve upcall */
1888         ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1889         if (ip_len > 0)
1890                 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1891         else
1892                 ret = -ESRCH;
1893 
1894         kfree(ip_addr);
1895 
1896         *ipend = end;
1897 
1898         pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1899                         ret, ret ? "failed" : ceph_pr_addr(ss));
1900 
1901         return ret;
1902 }
1903 #else
1904 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1905                 struct sockaddr_storage *ss, char delim, const char **ipend)
1906 {
1907         return -EINVAL;
1908 }
1909 #endif
1910 
1911 /*
1912  * Parse a server name (IP or hostname). If a valid IP address is not found
1913  * then try to extract a hostname to resolve using userspace DNS upcall.
1914  */
1915 static int ceph_parse_server_name(const char *name, size_t namelen,
1916                         struct sockaddr_storage *ss, char delim, const char **ipend)
1917 {
1918         int ret;
1919 
1920         ret = ceph_pton(name, namelen, ss, delim, ipend);
1921         if (ret)
1922                 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1923 
1924         return ret;
1925 }
1926 
1927 /*
1928  * Parse an ip[:port] list into an addr array.  Use the default
1929  * monitor port if a port isn't specified.
1930  */
1931 int ceph_parse_ips(const char *c, const char *end,
1932                    struct ceph_entity_addr *addr,
1933                    int max_count, int *count)
1934 {
1935         int i, ret = -EINVAL;
1936         const char *p = c;
1937 
1938         dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1939         for (i = 0; i < max_count; i++) {
1940                 const char *ipend;
1941                 struct sockaddr_storage *ss = &addr[i].in_addr;
1942                 int port;
1943                 char delim = ',';
1944 
1945                 if (*p == '[') {
1946                         delim = ']';
1947                         p++;
1948                 }
1949 
1950                 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1951                 if (ret)
1952                         goto bad;
1953                 ret = -EINVAL;
1954 
1955                 p = ipend;
1956 
1957                 if (delim == ']') {
1958                         if (*p != ']') {
1959                                 dout("missing matching ']'\n");
1960                                 goto bad;
1961                         }
1962                         p++;
1963                 }
1964 
1965                 /* port? */
1966                 if (p < end && *p == ':') {
1967                         port = 0;
1968                         p++;
1969                         while (p < end && *p >= '' && *p <= '9') {
1970                                 port = (port * 10) + (*p - '');
1971                                 p++;
1972                         }
1973                         if (port == 0)
1974                                 port = CEPH_MON_PORT;
1975                         else if (port > 65535)
1976                                 goto bad;
1977                 } else {
1978                         port = CEPH_MON_PORT;
1979                 }
1980 
1981                 addr_set_port(ss, port);
1982 
1983                 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1984 
1985                 if (p == end)
1986                         break;
1987                 if (*p != ',')
1988                         goto bad;
1989                 p++;
1990         }
1991 
1992         if (p != end)
1993                 goto bad;
1994 
1995         if (count)
1996                 *count = i + 1;
1997         return 0;
1998 
1999 bad:
2000         pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
2001         return ret;
2002 }
2003 EXPORT_SYMBOL(ceph_parse_ips);
2004 
2005 static int process_banner(struct ceph_connection *con)
2006 {
2007         dout("process_banner on %p\n", con);
2008 
2009         if (verify_hello(con) < 0)
2010                 return -1;
2011 
2012         ceph_decode_addr(&con->actual_peer_addr);
2013         ceph_decode_addr(&con->peer_addr_for_me);
2014 
2015         /*
2016          * Make sure the other end is who we wanted.  note that the other
2017          * end may not yet know their ip address, so if it's 0.0.0.0, give
2018          * them the benefit of the doubt.
2019          */
2020         if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2021                    sizeof(con->peer_addr)) != 0 &&
2022             !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2023               con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2024                 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2025                         ceph_pr_addr(&con->peer_addr.in_addr),
2026                         (int)le32_to_cpu(con->peer_addr.nonce),
2027                         ceph_pr_addr(&con->actual_peer_addr.in_addr),
2028                         (int)le32_to_cpu(con->actual_peer_addr.nonce));
2029                 con->error_msg = "wrong peer at address";
2030                 return -1;
2031         }
2032 
2033         /*
2034          * did we learn our address?
2035          */
2036         if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2037                 int port = addr_port(&con->msgr->inst.addr.in_addr);
2038 
2039                 memcpy(&con->msgr->inst.addr.in_addr,
2040                        &con->peer_addr_for_me.in_addr,
2041                        sizeof(con->peer_addr_for_me.in_addr));
2042                 addr_set_port(&con->msgr->inst.addr.in_addr, port);
2043                 encode_my_addr(con->msgr);
2044                 dout("process_banner learned my addr is %s\n",
2045                      ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2046         }
2047 
2048         return 0;
2049 }
2050 
2051 static int process_connect(struct ceph_connection *con)
2052 {
2053         u64 sup_feat = from_msgr(con->msgr)->supported_features;
2054         u64 req_feat = from_msgr(con->msgr)->required_features;
2055         u64 server_feat = le64_to_cpu(con->in_reply.features);
2056         int ret;
2057 
2058         dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2059 
2060         if (con->auth) {
2061                 int len = le32_to_cpu(con->in_reply.authorizer_len);
2062 
2063                 /*
2064                  * Any connection that defines ->get_authorizer()
2065                  * should also define ->add_authorizer_challenge() and
2066                  * ->verify_authorizer_reply().
2067                  *
2068                  * See get_connect_authorizer().
2069                  */
2070                 if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
2071                         ret = con->ops->add_authorizer_challenge(
2072                                     con, con->auth->authorizer_reply_buf, len);
2073                         if (ret < 0)
2074                                 return ret;
2075 
2076                         con_out_kvec_reset(con);
2077                         __prepare_write_connect(con);
2078                         prepare_read_connect(con);
2079                         return 0;
2080                 }
2081 
2082                 if (len) {
2083                         ret = con->ops->verify_authorizer_reply(con);
2084                         if (ret < 0) {
2085                                 con->error_msg = "bad authorize reply";
2086                                 return ret;
2087                         }
2088                 }
2089         }
2090 
2091         switch (con->in_reply.tag) {
2092         case CEPH_MSGR_TAG_FEATURES:
2093                 pr_err("%s%lld %s feature set mismatch,"
2094                        " my %llx < server's %llx, missing %llx\n",
2095                        ENTITY_NAME(con->peer_name),
2096                        ceph_pr_addr(&con->peer_addr.in_addr),
2097                        sup_feat, server_feat, server_feat & ~sup_feat);
2098                 con->error_msg = "missing required protocol features";
2099                 reset_connection(con);
2100                 return -1;
2101 
2102         case CEPH_MSGR_TAG_BADPROTOVER:
2103                 pr_err("%s%lld %s protocol version mismatch,"
2104                        " my %d != server's %d\n",
2105                        ENTITY_NAME(con->peer_name),
2106                        ceph_pr_addr(&con->peer_addr.in_addr),
2107                        le32_to_cpu(con->out_connect.protocol_version),
2108                        le32_to_cpu(con->in_reply.protocol_version));
2109                 con->error_msg = "protocol version mismatch";
2110                 reset_connection(con);
2111                 return -1;
2112 
2113         case CEPH_MSGR_TAG_BADAUTHORIZER:
2114                 con->auth_retry++;
2115                 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2116                      con->auth_retry);
2117                 if (con->auth_retry == 2) {
2118                         con->error_msg = "connect authorization failure";
2119                         return -1;
2120                 }
2121                 con_out_kvec_reset(con);
2122                 ret = prepare_write_connect(con);
2123                 if (ret < 0)
2124                         return ret;
2125                 prepare_read_connect(con);
2126                 break;
2127 
2128         case CEPH_MSGR_TAG_RESETSESSION:
2129                 /*
2130                  * If we connected with a large connect_seq but the peer
2131                  * has no record of a session with us (no connection, or
2132                  * connect_seq == 0), they will send RESETSESION to indicate
2133                  * that they must have reset their session, and may have
2134                  * dropped messages.
2135                  */
2136                 dout("process_connect got RESET peer seq %u\n",
2137                      le32_to_cpu(con->in_reply.connect_seq));
2138                 pr_err("%s%lld %s connection reset\n",
2139                        ENTITY_NAME(con->peer_name),
2140                        ceph_pr_addr(&con->peer_addr.in_addr));
2141                 reset_connection(con);
2142                 con_out_kvec_reset(con);
2143                 ret = prepare_write_connect(con);
2144                 if (ret < 0)
2145                         return ret;
2146                 prepare_read_connect(con);
2147 
2148                 /* Tell ceph about it. */
2149                 mutex_unlock(&con->mutex);
2150                 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2151                 if (con->ops->peer_reset)
2152                         con->ops->peer_reset(con);
2153                 mutex_lock(&con->mutex);
2154                 if (con->state != CON_STATE_NEGOTIATING)
2155                         return -EAGAIN;
2156                 break;
2157 
2158         case CEPH_MSGR_TAG_RETRY_SESSION:
2159                 /*
2160                  * If we sent a smaller connect_seq than the peer has, try
2161                  * again with a larger value.
2162                  */
2163                 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2164                      le32_to_cpu(con->out_connect.connect_seq),
2165                      le32_to_cpu(con->in_reply.connect_seq));
2166                 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2167                 con_out_kvec_reset(con);
2168                 ret = prepare_write_connect(con);
2169                 if (ret < 0)
2170                         return ret;
2171                 prepare_read_connect(con);
2172                 break;
2173 
2174         case CEPH_MSGR_TAG_RETRY_GLOBAL:
2175                 /*
2176                  * If we sent a smaller global_seq than the peer has, try
2177                  * again with a larger value.
2178                  */
2179                 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2180                      con->peer_global_seq,
2181                      le32_to_cpu(con->in_reply.global_seq));
2182                 get_global_seq(con->msgr,
2183                                le32_to_cpu(con->in_reply.global_seq));
2184                 con_out_kvec_reset(con);
2185                 ret = prepare_write_connect(con);
2186                 if (ret < 0)
2187                         return ret;
2188                 prepare_read_connect(con);
2189                 break;
2190 
2191         case CEPH_MSGR_TAG_SEQ:
2192         case CEPH_MSGR_TAG_READY:
2193                 if (req_feat & ~server_feat) {
2194                         pr_err("%s%lld %s protocol feature mismatch,"
2195                                " my required %llx > server's %llx, need %llx\n",
2196                                ENTITY_NAME(con->peer_name),
2197                                ceph_pr_addr(&con->peer_addr.in_addr),
2198                                req_feat, server_feat, req_feat & ~server_feat);
2199                         con->error_msg = "missing required protocol features";
2200                         reset_connection(con);
2201                         return -1;
2202                 }
2203 
2204                 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2205                 con->state = CON_STATE_OPEN;
2206                 con->auth_retry = 0;    /* we authenticated; clear flag */
2207                 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2208                 con->connect_seq++;
2209                 con->peer_features = server_feat;
2210                 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2211                      con->peer_global_seq,
2212                      le32_to_cpu(con->in_reply.connect_seq),
2213                      con->connect_seq);
2214                 WARN_ON(con->connect_seq !=
2215                         le32_to_cpu(con->in_reply.connect_seq));
2216 
2217                 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2218                         con_flag_set(con, CON_FLAG_LOSSYTX);
2219 
2220                 con->delay = 0;      /* reset backoff memory */
2221 
2222                 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2223                         prepare_write_seq(con);
2224                         prepare_read_seq(con);
2225                 } else {
2226                         prepare_read_tag(con);
2227                 }
2228                 break;
2229 
2230         case CEPH_MSGR_TAG_WAIT:
2231                 /*
2232                  * If there is a connection race (we are opening
2233                  * connections to each other), one of us may just have
2234                  * to WAIT.  This shouldn't happen if we are the
2235                  * client.
2236                  */
2237                 con->error_msg = "protocol error, got WAIT as client";
2238                 return -1;
2239 
2240         default:
2241                 con->error_msg = "protocol error, garbage tag during connect";
2242                 return -1;
2243         }
2244         return 0;
2245 }
2246 
2247 
2248 /*
2249  * read (part of) an ack
2250  */
2251 static int read_partial_ack(struct ceph_connection *con)
2252 {
2253         int size = sizeof (con->in_temp_ack);
2254         int end = size;
2255 
2256         return read_partial(con, end, size, &con->in_temp_ack);
2257 }
2258 
2259 /*
2260  * We can finally discard anything that's been acked.
2261  */
2262 static void process_ack(struct ceph_connection *con)
2263 {
2264         struct ceph_msg *m;
2265         u64 ack = le64_to_cpu(con->in_temp_ack);
2266         u64 seq;
2267         bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2268         struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
2269 
2270         /*
2271          * In the reconnect case, con_fault() has requeued messages
2272          * in out_sent. We should cleanup old messages according to
2273          * the reconnect seq.
2274          */
2275         while (!list_empty(list)) {
2276                 m = list_first_entry(list, struct ceph_msg, list_head);
2277                 if (reconnect && m->needs_out_seq)
2278                         break;
2279                 seq = le64_to_cpu(m->hdr.seq);
2280                 if (seq > ack)
2281                         break;
2282                 dout("got ack for seq %llu type %d at %p\n", seq,
2283                      le16_to_cpu(m->hdr.type), m);
2284                 m->ack_stamp = jiffies;
2285                 ceph_msg_remove(m);
2286         }
2287 
2288         prepare_read_tag(con);
2289 }
2290 
2291 
2292 static int read_partial_message_section(struct ceph_connection *con,
2293                                         struct kvec *section,
2294                                         unsigned int sec_len, u32 *crc)
2295 {
2296         int ret, left;
2297 
2298         BUG_ON(!section);
2299 
2300         while (section->iov_len < sec_len) {
2301                 BUG_ON(section->iov_base == NULL);
2302                 left = sec_len - section->iov_len;
2303                 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2304                                        section->iov_len, left);
2305                 if (ret <= 0)
2306                         return ret;
2307                 section->iov_len += ret;
2308         }
2309         if (section->iov_len == sec_len)
2310                 *crc = crc32c(0, section->iov_base, section->iov_len);
2311 
2312         return 1;
2313 }
2314 
2315 static int read_partial_msg_data(struct ceph_connection *con)
2316 {
2317         struct ceph_msg *msg = con->in_msg;
2318         struct ceph_msg_data_cursor *cursor = &msg->cursor;
2319         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2320         struct page *page;
2321         size_t page_offset;
2322         size_t length;
2323         u32 crc = 0;
2324         int ret;
2325 
2326         if (!msg->num_data_items)
2327                 return -EIO;
2328 
2329         if (do_datacrc)
2330                 crc = con->in_data_crc;
2331         while (cursor->total_resid) {
2332                 if (!cursor->resid) {
2333                         ceph_msg_data_advance(cursor, 0);
2334                         continue;
2335                 }
2336 
2337                 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2338                 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2339                 if (ret <= 0) {
2340                         if (do_datacrc)
2341                                 con->in_data_crc = crc;
2342 
2343                         return ret;
2344                 }
2345 
2346                 if (do_datacrc)
2347                         crc = ceph_crc32c_page(crc, page, page_offset, ret);
2348                 ceph_msg_data_advance(cursor, (size_t)ret);
2349         }
2350         if (do_datacrc)
2351                 con->in_data_crc = crc;
2352 
2353         return 1;       /* must return > 0 to indicate success */
2354 }
2355 
2356 /*
2357  * read (part of) a message.
2358  */
2359 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2360 
2361 static int read_partial_message(struct ceph_connection *con)
2362 {
2363         struct ceph_msg *m = con->in_msg;
2364         int size;
2365         int end;
2366         int ret;
2367         unsigned int front_len, middle_len, data_len;
2368         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2369         bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2370         u64 seq;
2371         u32 crc;
2372 
2373         dout("read_partial_message con %p msg %p\n", con, m);
2374 
2375         /* header */
2376         size = sizeof (con->in_hdr);
2377         end = size;
2378         ret = read_partial(con, end, size, &con->in_hdr);
2379         if (ret <= 0)
2380                 return ret;
2381 
2382         crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2383         if (cpu_to_le32(crc) != con->in_hdr.crc) {
2384                 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2385                        crc, con->in_hdr.crc);
2386                 return -EBADMSG;
2387         }
2388 
2389         front_len = le32_to_cpu(con->in_hdr.front_len);
2390         if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2391                 return -EIO;
2392         middle_len = le32_to_cpu(con->in_hdr.middle_len);
2393         if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2394                 return -EIO;
2395         data_len = le32_to_cpu(con->in_hdr.data_len);
2396         if (data_len > CEPH_MSG_MAX_DATA_LEN)
2397                 return -EIO;
2398 
2399         /* verify seq# */
2400         seq = le64_to_cpu(con->in_hdr.seq);
2401         if ((s64)seq - (s64)con->in_seq < 1) {
2402                 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2403                         ENTITY_NAME(con->peer_name),
2404                         ceph_pr_addr(&con->peer_addr.in_addr),
2405                         seq, con->in_seq + 1);
2406                 con->in_base_pos = -front_len - middle_len - data_len -
2407                         sizeof_footer(con);
2408                 con->in_tag = CEPH_MSGR_TAG_READY;
2409                 return 1;
2410         } else if ((s64)seq - (s64)con->in_seq > 1) {
2411                 pr_err("read_partial_message bad seq %lld expected %lld\n",
2412                        seq, con->in_seq + 1);
2413                 con->error_msg = "bad message sequence # for incoming message";
2414                 return -EBADE;
2415         }
2416 
2417         /* allocate message? */
2418         if (!con->in_msg) {
2419                 int skip = 0;
2420 
2421                 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2422                      front_len, data_len);
2423                 ret = ceph_con_in_msg_alloc(con, &skip);
2424                 if (ret < 0)
2425                         return ret;
2426 
2427                 BUG_ON(!con->in_msg ^ skip);
2428                 if (skip) {
2429                         /* skip this message */
2430                         dout("alloc_msg said skip message\n");
2431                         con->in_base_pos = -front_len - middle_len - data_len -
2432                                 sizeof_footer(con);
2433                         con->in_tag = CEPH_MSGR_TAG_READY;
2434                         con->in_seq++;
2435                         return 1;
2436                 }
2437 
2438                 BUG_ON(!con->in_msg);
2439                 BUG_ON(con->in_msg->con != con);
2440                 m = con->in_msg;
2441                 m->front.iov_len = 0;    /* haven't read it yet */
2442                 if (m->middle)
2443                         m->middle->vec.iov_len = 0;
2444 
2445                 /* prepare for data payload, if any */
2446 
2447                 if (data_len)
2448                         prepare_message_data(con->in_msg, data_len);
2449         }
2450 
2451         /* front */
2452         ret = read_partial_message_section(con, &m->front, front_len,
2453                                            &con->in_front_crc);
2454         if (ret <= 0)
2455                 return ret;
2456 
2457         /* middle */
2458         if (m->middle) {
2459                 ret = read_partial_message_section(con, &m->middle->vec,
2460                                                    middle_len,
2461                                                    &con->in_middle_crc);
2462                 if (ret <= 0)
2463                         return ret;
2464         }
2465 
2466         /* (page) data */
2467         if (data_len) {
2468                 ret = read_partial_msg_data(con);
2469                 if (ret <= 0)
2470                         return ret;
2471         }
2472 
2473         /* footer */
2474         size = sizeof_footer(con);
2475         end += size;
2476         ret = read_partial(con, end, size, &m->footer);
2477         if (ret <= 0)
2478                 return ret;
2479 
2480         if (!need_sign) {
2481                 m->footer.flags = m->old_footer.flags;
2482                 m->footer.sig = 0;
2483         }
2484 
2485         dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2486              m, front_len, m->footer.front_crc, middle_len,
2487              m->footer.middle_crc, data_len, m->footer.data_crc);
2488 
2489         /* crc ok? */
2490         if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2491                 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2492                        m, con->in_front_crc, m->footer.front_crc);
2493                 return -EBADMSG;
2494         }
2495         if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2496                 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2497                        m, con->in_middle_crc, m->footer.middle_crc);
2498                 return -EBADMSG;
2499         }
2500         if (do_datacrc &&
2501             (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2502             con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2503                 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2504                        con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2505                 return -EBADMSG;
2506         }
2507 
2508         if (need_sign && con->ops->check_message_signature &&
2509             con->ops->check_message_signature(m)) {
2510                 pr_err("read_partial_message %p signature check failed\n", m);
2511                 return -EBADMSG;
2512         }
2513 
2514         return 1; /* done! */
2515 }
2516 
2517 /*
2518  * Process message.  This happens in the worker thread.  The callback should
2519  * be careful not to do anything that waits on other incoming messages or it
2520  * may deadlock.
2521  */
2522 static void process_message(struct ceph_connection *con)
2523 {
2524         struct ceph_msg *msg = con->in_msg;
2525 
2526         BUG_ON(con->in_msg->con != con);
2527         con->in_msg = NULL;
2528 
2529         /* if first message, set peer_name */
2530         if (con->peer_name.type == 0)
2531                 con->peer_name = msg->hdr.src;
2532 
2533         con->in_seq++;
2534         mutex_unlock(&con->mutex);
2535 
2536         dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2537              msg, le64_to_cpu(msg->hdr.seq),
2538              ENTITY_NAME(msg->hdr.src),
2539              le16_to_cpu(msg->hdr.type),
2540              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2541              le32_to_cpu(msg->hdr.front_len),
2542              le32_to_cpu(msg->hdr.data_len),
2543              con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2544         con->ops->dispatch(con, msg);
2545 
2546         mutex_lock(&con->mutex);
2547 }
2548 
2549 static int read_keepalive_ack(struct ceph_connection *con)
2550 {
2551         struct ceph_timespec ceph_ts;
2552         size_t size = sizeof(ceph_ts);
2553         int ret = read_partial(con, size, size, &ceph_ts);
2554         if (ret <= 0)
2555                 return ret;
2556         ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
2557         prepare_read_tag(con);
2558         return 1;
2559 }
2560 
2561 /*
2562  * Write something to the socket.  Called in a worker thread when the
2563  * socket appears to be writeable and we have something ready to send.
2564  */
2565 static int try_write(struct ceph_connection *con)
2566 {
2567         int ret = 1;
2568 
2569         dout("try_write start %p state %lu\n", con, con->state);
2570         if (con->state != CON_STATE_PREOPEN &&
2571             con->state != CON_STATE_CONNECTING &&
2572             con->state != CON_STATE_NEGOTIATING &&
2573             con->state != CON_STATE_OPEN)
2574                 return 0;
2575 
2576         /* open the socket first? */
2577         if (con->state == CON_STATE_PREOPEN) {
2578                 BUG_ON(con->sock);
2579                 con->state = CON_STATE_CONNECTING;
2580 
2581                 con_out_kvec_reset(con);
2582                 prepare_write_banner(con);
2583                 prepare_read_banner(con);
2584 
2585                 BUG_ON(con->in_msg);
2586                 con->in_tag = CEPH_MSGR_TAG_READY;
2587                 dout("try_write initiating connect on %p new state %lu\n",
2588                      con, con->state);
2589                 ret = ceph_tcp_connect(con);
2590                 if (ret < 0) {
2591                         con->error_msg = "connect error";
2592                         goto out;
2593                 }
2594         }
2595 
2596 more:
2597         dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2598         BUG_ON(!con->sock);
2599 
2600         /* kvec data queued? */
2601         if (con->out_kvec_left) {
2602                 ret = write_partial_kvec(con);
2603                 if (ret <= 0)
2604                         goto out;
2605         }
2606         if (con->out_skip) {
2607                 ret = write_partial_skip(con);
2608                 if (ret <= 0)
2609                         goto out;
2610         }
2611 
2612         /* msg pages? */
2613         if (con->out_msg) {
2614                 if (con->out_msg_done) {
2615                         ceph_msg_put(con->out_msg);
2616                         con->out_msg = NULL;   /* we're done with this one */
2617                         goto do_next;
2618                 }
2619 
2620                 ret = write_partial_message_data(con);
2621                 if (ret == 1)
2622                         goto more;  /* we need to send the footer, too! */
2623                 if (ret == 0)
2624                         goto out;
2625                 if (ret < 0) {
2626                         dout("try_write write_partial_message_data err %d\n",
2627                              ret);
2628                         goto out;
2629                 }
2630         }
2631 
2632 do_next:
2633         if (con->state == CON_STATE_OPEN) {
2634                 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2635                         prepare_write_keepalive(con);
2636                         goto more;
2637                 }
2638                 /* is anything else pending? */
2639                 if (!list_empty(&con->out_queue)) {
2640                         prepare_write_message(con);
2641                         goto more;
2642                 }
2643                 if (con->in_seq > con->in_seq_acked) {
2644                         prepare_write_ack(con);
2645                         goto more;
2646                 }
2647         }
2648 
2649         /* Nothing to do! */
2650         con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2651         dout("try_write nothing else to write.\n");
2652         ret = 0;
2653 out:
2654         dout("try_write done on %p ret %d\n", con, ret);
2655         return ret;
2656 }
2657 
2658 /*
2659  * Read what we can from the socket.
2660  */
2661 static int try_read(struct ceph_connection *con)
2662 {
2663         int ret = -1;
2664 
2665 more:
2666         dout("try_read start on %p state %lu\n", con, con->state);
2667         if (con->state != CON_STATE_CONNECTING &&
2668             con->state != CON_STATE_NEGOTIATING &&
2669             con->state != CON_STATE_OPEN)
2670                 return 0;
2671 
2672         BUG_ON(!con->sock);
2673 
2674         dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2675              con->in_base_pos);
2676 
2677         if (con->state == CON_STATE_CONNECTING) {
2678                 dout("try_read connecting\n");
2679                 ret = read_partial_banner(con);
2680                 if (ret <= 0)
2681                         goto out;
2682                 ret = process_banner(con);
2683                 if (ret < 0)
2684                         goto out;
2685 
2686                 con->state = CON_STATE_NEGOTIATING;
2687 
2688                 /*
2689                  * Received banner is good, exchange connection info.
2690                  * Do not reset out_kvec, as sending our banner raced
2691                  * with receiving peer banner after connect completed.
2692                  */
2693                 ret = prepare_write_connect(con);
2694                 if (ret < 0)
2695                         goto out;
2696                 prepare_read_connect(con);
2697 
2698                 /* Send connection info before awaiting response */
2699                 goto out;
2700         }
2701 
2702         if (con->state == CON_STATE_NEGOTIATING) {
2703                 dout("try_read negotiating\n");
2704                 ret = read_partial_connect(con);
2705                 if (ret <= 0)
2706                         goto out;
2707                 ret = process_connect(con);
2708                 if (ret < 0)
2709                         goto out;
2710                 goto more;
2711         }
2712 
2713         WARN_ON(con->state != CON_STATE_OPEN);
2714 
2715         if (con->in_base_pos < 0) {
2716                 /*
2717                  * skipping + discarding content.
2718                  */
2719                 ret = ceph_tcp_recvmsg(con->sock, NULL, -con->in_base_pos);
2720                 if (ret <= 0)
2721                         goto out;
2722                 dout("skipped %d / %d bytes\n", ret, -con->in_base_pos);
2723                 con->in_base_pos += ret;
2724                 if (con->in_base_pos)
2725                         goto more;
2726         }
2727         if (con->in_tag == CEPH_MSGR_TAG_READY) {
2728                 /*
2729                  * what's next?
2730                  */
2731                 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2732                 if (ret <= 0)
2733                         goto out;
2734                 dout("try_read got tag %d\n", (int)con->in_tag);
2735                 switch (con->in_tag) {
2736                 case CEPH_MSGR_TAG_MSG:
2737                         prepare_read_message(con);
2738                         break;
2739                 case CEPH_MSGR_TAG_ACK:
2740                         prepare_read_ack(con);
2741                         break;
2742                 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2743                         prepare_read_keepalive_ack(con);
2744                         break;
2745                 case CEPH_MSGR_TAG_CLOSE:
2746                         con_close_socket(con);
2747                         con->state = CON_STATE_CLOSED;
2748                         goto out;
2749                 default:
2750                         goto bad_tag;
2751                 }
2752         }
2753         if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2754                 ret = read_partial_message(con);
2755                 if (ret <= 0) {
2756                         switch (ret) {
2757                         case -EBADMSG:
2758                                 con->error_msg = "bad crc/signature";
2759                                 /* fall through */
2760                         case -EBADE:
2761                                 ret = -EIO;
2762                                 break;
2763                         case -EIO:
2764                                 con->error_msg = "io error";
2765                                 break;
2766                         }
2767                         goto out;
2768                 }
2769                 if (con->in_tag == CEPH_MSGR_TAG_READY)
2770                         goto more;
2771                 process_message(con);
2772                 if (con->state == CON_STATE_OPEN)
2773                         prepare_read_tag(con);
2774                 goto more;
2775         }
2776         if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2777             con->in_tag == CEPH_MSGR_TAG_SEQ) {
2778                 /*
2779                  * the final handshake seq exchange is semantically
2780                  * equivalent to an ACK
2781                  */
2782                 ret = read_partial_ack(con);
2783                 if (ret <= 0)
2784                         goto out;
2785                 process_ack(con);
2786                 goto more;
2787         }
2788         if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2789                 ret = read_keepalive_ack(con);
2790                 if (ret <= 0)
2791                         goto out;
2792                 goto more;
2793         }
2794 
2795 out:
2796         dout("try_read done on %p ret %d\n", con, ret);
2797         return ret;
2798 
2799 bad_tag:
2800         pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2801         con->error_msg = "protocol error, garbage tag";
2802         ret = -1;
2803         goto out;
2804 }
2805 
2806 
2807 /*
2808  * Atomically queue work on a connection after the specified delay.
2809  * Bump @con reference to avoid races with connection teardown.
2810  * Returns 0 if work was queued, or an error code otherwise.
2811  */
2812 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2813 {
2814         if (!con->ops->get(con)) {
2815                 dout("%s %p ref count 0\n", __func__, con);
2816                 return -ENOENT;
2817         }
2818 
2819         if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2820                 dout("%s %p - already queued\n", __func__, con);
2821                 con->ops->put(con);
2822                 return -EBUSY;
2823         }
2824 
2825         dout("%s %p %lu\n", __func__, con, delay);
2826         return 0;
2827 }
2828 
2829 static void queue_con(struct ceph_connection *con)
2830 {
2831         (void) queue_con_delay(con, 0);
2832 }
2833 
2834 static void cancel_con(struct ceph_connection *con)
2835 {
2836         if (cancel_delayed_work(&con->work)) {
2837                 dout("%s %p\n", __func__, con);
2838                 con->ops->put(con);
2839         }
2840 }
2841 
2842 static bool con_sock_closed(struct ceph_connection *con)
2843 {
2844         if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2845                 return false;
2846 
2847 #define CASE(x)                                                         \
2848         case CON_STATE_ ## x:                                           \
2849                 con->error_msg = "socket closed (con state " #x ")";    \
2850                 break;
2851 
2852         switch (con->state) {
2853         CASE(CLOSED);
2854         CASE(PREOPEN);
2855         CASE(CONNECTING);
2856         CASE(NEGOTIATING);
2857         CASE(OPEN);
2858         CASE(STANDBY);
2859         default:
2860                 pr_warn("%s con %p unrecognized state %lu\n",
2861                         __func__, con, con->state);
2862                 con->error_msg = "unrecognized con state";
2863                 BUG();
2864                 break;
2865         }
2866 #undef CASE
2867 
2868         return true;
2869 }
2870 
2871 static bool con_backoff(struct ceph_connection *con)
2872 {
2873         int ret;
2874 
2875         if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2876                 return false;
2877 
2878         ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2879         if (ret) {
2880                 dout("%s: con %p FAILED to back off %lu\n", __func__,
2881                         con, con->delay);
2882                 BUG_ON(ret == -ENOENT);
2883                 con_flag_set(con, CON_FLAG_BACKOFF);
2884         }
2885 
2886         return true;
2887 }
2888 
2889 /* Finish fault handling; con->mutex must *not* be held here */
2890 
2891 static void con_fault_finish(struct ceph_connection *con)
2892 {
2893         dout("%s %p\n", __func__, con);
2894 
2895         /*
2896          * in case we faulted due to authentication, invalidate our
2897          * current tickets so that we can get new ones.
2898          */
2899         if (con->auth_retry) {
2900                 dout("auth_retry %d, invalidating\n", con->auth_retry);
2901                 if (con->ops->invalidate_authorizer)
2902                         con->ops->invalidate_authorizer(con);
2903                 con->auth_retry = 0;
2904         }
2905 
2906         if (con->ops->fault)
2907                 con->ops->fault(con);
2908 }
2909 
2910 /*
2911  * Do some work on a connection.  Drop a connection ref when we're done.
2912  */
2913 static void ceph_con_workfn(struct work_struct *work)
2914 {
2915         struct ceph_connection *con = container_of(work, struct ceph_connection,
2916                                                    work.work);
2917         bool fault;
2918 
2919         mutex_lock(&con->mutex);
2920         while (true) {
2921                 int ret;
2922 
2923                 if ((fault = con_sock_closed(con))) {
2924                         dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2925                         break;
2926                 }
2927                 if (con_backoff(con)) {
2928                         dout("%s: con %p BACKOFF\n", __func__, con);
2929                         break;
2930                 }
2931                 if (con->state == CON_STATE_STANDBY) {
2932                         dout("%s: con %p STANDBY\n", __func__, con);
2933                         break;
2934                 }
2935                 if (con->state == CON_STATE_CLOSED) {
2936                         dout("%s: con %p CLOSED\n", __func__, con);
2937                         BUG_ON(con->sock);
2938                         break;
2939                 }
2940                 if (con->state == CON_STATE_PREOPEN) {
2941                         dout("%s: con %p PREOPEN\n", __func__, con);
2942                         BUG_ON(con->sock);
2943                 }
2944 
2945                 ret = try_read(con);
2946                 if (ret < 0) {
2947                         if (ret == -EAGAIN)
2948                                 continue;
2949                         if (!con->error_msg)
2950                                 con->error_msg = "socket error on read";
2951                         fault = true;
2952                         break;
2953                 }
2954 
2955                 ret = try_write(con);
2956                 if (ret < 0) {
2957                         if (ret == -EAGAIN)
2958                                 continue;
2959                         if (!con->error_msg)
2960                                 con->error_msg = "socket error on write";
2961                         fault = true;
2962                 }
2963 
2964                 break;  /* If we make it to here, we're done */
2965         }
2966         if (fault)
2967                 con_fault(con);
2968         mutex_unlock(&con->mutex);
2969 
2970         if (fault)
2971                 con_fault_finish(con);
2972 
2973         con->ops->put(con);
2974 }
2975 
2976 /*
2977  * Generic error/fault handler.  A retry mechanism is used with
2978  * exponential backoff
2979  */
2980 static void con_fault(struct ceph_connection *con)
2981 {
2982         dout("fault %p state %lu to peer %s\n",
2983              con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2984 
2985         pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2986                 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2987         con->error_msg = NULL;
2988 
2989         WARN_ON(con->state != CON_STATE_CONNECTING &&
2990                con->state != CON_STATE_NEGOTIATING &&
2991                con->state != CON_STATE_OPEN);
2992 
2993         con_close_socket(con);
2994 
2995         if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2996                 dout("fault on LOSSYTX channel, marking CLOSED\n");
2997                 con->state = CON_STATE_CLOSED;
2998                 return;
2999         }
3000 
3001         if (con->in_msg) {
3002                 BUG_ON(con->in_msg->con != con);
3003                 ceph_msg_put(con->in_msg);
3004                 con->in_msg = NULL;
3005         }
3006 
3007         /* Requeue anything that hasn't been acked */
3008         list_splice_init(&con->out_sent, &con->out_queue);
3009 
3010         /* If there are no messages queued or keepalive pending, place
3011          * the connection in a STANDBY state */
3012         if (list_empty(&con->out_queue) &&
3013             !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
3014                 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
3015                 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
3016                 con->state = CON_STATE_STANDBY;
3017         } else {
3018                 /* retry after a delay. */
3019                 con->state = CON_STATE_PREOPEN;
3020                 if (con->delay == 0)
3021                         con->delay = BASE_DELAY_INTERVAL;
3022                 else if (con->delay < MAX_DELAY_INTERVAL)
3023                         con->delay *= 2;
3024                 con_flag_set(con, CON_FLAG_BACKOFF);
3025                 queue_con(con);
3026         }
3027 }
3028 
3029 
3030 
3031 /*
3032  * initialize a new messenger instance
3033  */
3034 void ceph_messenger_init(struct ceph_messenger *msgr,
3035                          struct ceph_entity_addr *myaddr)
3036 {
3037         spin_lock_init(&msgr->global_seq_lock);
3038 
3039         if (myaddr)
3040                 msgr->inst.addr = *myaddr;
3041 
3042         /* select a random nonce */
3043         msgr->inst.addr.type = 0;
3044         get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3045         encode_my_addr(msgr);
3046 
3047         atomic_set(&msgr->stopping, 0);
3048         write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3049 
3050         dout("%s %p\n", __func__, msgr);
3051 }
3052 EXPORT_SYMBOL(ceph_messenger_init);
3053 
3054 void ceph_messenger_fini(struct ceph_messenger *msgr)
3055 {
3056         put_net(read_pnet(&msgr->net));
3057 }
3058 EXPORT_SYMBOL(ceph_messenger_fini);
3059 
3060 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3061 {
3062         if (msg->con)
3063                 msg->con->ops->put(msg->con);
3064 
3065         msg->con = con ? con->ops->get(con) : NULL;
3066         BUG_ON(msg->con != con);
3067 }
3068 
3069 static void clear_standby(struct ceph_connection *con)
3070 {
3071         /* come back from STANDBY? */
3072         if (con->state == CON_STATE_STANDBY) {
3073                 dout("clear_standby %p and ++connect_seq\n", con);
3074                 con->state = CON_STATE_PREOPEN;
3075                 con->connect_seq++;
3076                 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3077                 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3078         }
3079 }
3080 
3081 /*
3082  * Queue up an outgoing message on the given connection.
3083  */
3084 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3085 {
3086         /* set src+dst */
3087         msg->hdr.src = con->msgr->inst.name;
3088         BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3089         msg->needs_out_seq = true;
3090 
3091         mutex_lock(&con->mutex);
3092 
3093         if (con->state == CON_STATE_CLOSED) {
3094                 dout("con_send %p closed, dropping %p\n", con, msg);
3095                 ceph_msg_put(msg);
3096                 mutex_unlock(&con->mutex);
3097                 return;
3098         }
3099 
3100         msg_con_set(msg, con);
3101 
3102         BUG_ON(!list_empty(&msg->list_head));
3103         list_add_tail(&msg->list_head, &con->out_queue);
3104         dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3105              ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3106              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3107              le32_to_cpu(msg->hdr.front_len),
3108              le32_to_cpu(msg->hdr.middle_len),
3109              le32_to_cpu(msg->hdr.data_len));
3110 
3111         clear_standby(con);
3112         mutex_unlock(&con->mutex);
3113 
3114         /* if there wasn't anything waiting to send before, queue
3115          * new work */
3116         if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3117                 queue_con(con);
3118 }
3119 EXPORT_SYMBOL(ceph_con_send);
3120 
3121 /*
3122  * Revoke a message that was previously queued for send
3123  */
3124 void ceph_msg_revoke(struct ceph_msg *msg)
3125 {
3126         struct ceph_connection *con = msg->con;
3127 
3128         if (!con) {
3129                 dout("%s msg %p null con\n", __func__, msg);
3130                 return;         /* Message not in our possession */
3131         }
3132 
3133         mutex_lock(&con->mutex);
3134         if (!list_empty(&msg->list_head)) {
3135                 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3136                 list_del_init(&msg->list_head);
3137                 msg->hdr.seq = 0;
3138 
3139                 ceph_msg_put(msg);
3140         }
3141         if (con->out_msg == msg) {
3142                 BUG_ON(con->out_skip);
3143                 /* footer */
3144                 if (con->out_msg_done) {
3145                         con->out_skip += con_out_kvec_skip(con);
3146                 } else {
3147                         BUG_ON(!msg->data_length);
3148                         con->out_skip += sizeof_footer(con);
3149                 }
3150                 /* data, middle, front */
3151                 if (msg->data_length)
3152                         con->out_skip += msg->cursor.total_resid;
3153                 if (msg->middle)
3154                         con->out_skip += con_out_kvec_skip(con);
3155                 con->out_skip += con_out_kvec_skip(con);
3156 
3157                 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3158                      __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3159                 msg->hdr.seq = 0;
3160                 con->out_msg = NULL;
3161                 ceph_msg_put(msg);
3162         }
3163 
3164         mutex_unlock(&con->mutex);
3165 }
3166 
3167 /*
3168  * Revoke a message that we may be reading data into
3169  */
3170 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3171 {
3172         struct ceph_connection *con = msg->con;
3173 
3174         if (!con) {
3175                 dout("%s msg %p null con\n", __func__, msg);
3176                 return;         /* Message not in our possession */
3177         }
3178 
3179         mutex_lock(&con->mutex);
3180         if (con->in_msg == msg) {
3181                 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3182                 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3183                 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3184 
3185                 /* skip rest of message */
3186                 dout("%s %p msg %p revoked\n", __func__, con, msg);
3187                 con->in_base_pos = con->in_base_pos -
3188                                 sizeof(struct ceph_msg_header) -
3189                                 front_len -
3190                                 middle_len -
3191                                 data_len -
3192                                 sizeof(struct ceph_msg_footer);
3193                 ceph_msg_put(con->in_msg);
3194                 con->in_msg = NULL;
3195                 con->in_tag = CEPH_MSGR_TAG_READY;
3196                 con->in_seq++;
3197         } else {
3198                 dout("%s %p in_msg %p msg %p no-op\n",
3199                      __func__, con, con->in_msg, msg);
3200         }
3201         mutex_unlock(&con->mutex);
3202 }
3203 
3204 /*
3205  * Queue a keepalive byte to ensure the tcp connection is alive.
3206  */
3207 void ceph_con_keepalive(struct ceph_connection *con)
3208 {
3209         dout("con_keepalive %p\n", con);
3210         mutex_lock(&con->mutex);
3211         clear_standby(con);
3212         con_flag_set(con, CON_FLAG_KEEPALIVE_PENDING);
3213         mutex_unlock(&con->mutex);
3214 
3215         if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3216                 queue_con(con);
3217 }
3218 EXPORT_SYMBOL(ceph_con_keepalive);
3219 
3220 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3221                                unsigned long interval)
3222 {
3223         if (interval > 0 &&
3224             (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3225                 struct timespec64 now;
3226                 struct timespec64 ts;
3227                 ktime_get_real_ts64(&now);
3228                 jiffies_to_timespec64(interval, &ts);
3229                 ts = timespec64_add(con->last_keepalive_ack, ts);
3230                 return timespec64_compare(&now, &ts) >= 0;
3231         }
3232         return false;
3233 }
3234 
3235 static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
3236 {
3237         BUG_ON(msg->num_data_items >= msg->max_data_items);
3238         return &msg->data[msg->num_data_items++];
3239 }
3240 
3241 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3242 {
3243         if (data->type == CEPH_MSG_DATA_PAGELIST)
3244                 ceph_pagelist_release(data->pagelist);
3245 }
3246 
3247 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3248                 size_t length, size_t alignment)
3249 {
3250         struct ceph_msg_data *data;
3251 
3252         BUG_ON(!pages);
3253         BUG_ON(!length);
3254 
3255         data = ceph_msg_data_add(msg);
3256         data->type = CEPH_MSG_DATA_PAGES;
3257         data->pages = pages;
3258         data->length = length;
3259         data->alignment = alignment & ~PAGE_MASK;
3260 
3261         msg->data_length += length;
3262 }
3263 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3264 
3265 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3266                                 struct ceph_pagelist *pagelist)
3267 {
3268         struct ceph_msg_data *data;
3269 
3270         BUG_ON(!pagelist);
3271         BUG_ON(!pagelist->length);
3272 
3273         data = ceph_msg_data_add(msg);
3274         data->type = CEPH_MSG_DATA_PAGELIST;
3275         refcount_inc(&pagelist->refcnt);
3276         data->pagelist = pagelist;
3277 
3278         msg->data_length += pagelist->length;
3279 }
3280 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3281 
3282 #ifdef  CONFIG_BLOCK
3283 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
3284                            u32 length)
3285 {
3286         struct ceph_msg_data *data;
3287 
3288         data = ceph_msg_data_add(msg);
3289         data->type = CEPH_MSG_DATA_BIO;
3290         data->bio_pos = *bio_pos;
3291         data->bio_length = length;
3292 
3293         msg->data_length += length;
3294 }
3295 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3296 #endif  /* CONFIG_BLOCK */
3297 
3298 void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
3299                              struct ceph_bvec_iter *bvec_pos)
3300 {
3301         struct ceph_msg_data *data;
3302 
3303         data = ceph_msg_data_add(msg);
3304         data->type = CEPH_MSG_DATA_BVECS;
3305         data->bvec_pos = *bvec_pos;
3306 
3307         msg->data_length += bvec_pos->iter.bi_size;
3308 }
3309 EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
3310 
3311 /*
3312  * construct a new message with given type, size
3313  * the new msg has a ref count of 1.
3314  */
3315 struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
3316                                gfp_t flags, bool can_fail)
3317 {
3318         struct ceph_msg *m;
3319 
3320         m = kmem_cache_zalloc(ceph_msg_cache, flags);
3321         if (m == NULL)
3322                 goto out;
3323 
3324         m->hdr.type = cpu_to_le16(type);
3325         m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3326         m->hdr.front_len = cpu_to_le32(front_len);
3327 
3328         INIT_LIST_HEAD(&m->list_head);
3329         kref_init(&m->kref);
3330 
3331         /* front */
3332         if (front_len) {
3333                 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3334                 if (m->front.iov_base == NULL) {
3335                         dout("ceph_msg_new can't allocate %d bytes\n",
3336                              front_len);
3337                         goto out2;
3338                 }
3339         } else {
3340                 m->front.iov_base = NULL;
3341         }
3342         m->front_alloc_len = m->front.iov_len = front_len;
3343 
3344         if (max_data_items) {
3345                 m->data = kmalloc_array(max_data_items, sizeof(*m->data),
3346                                         flags);
3347                 if (!m->data)
3348                         goto out2;
3349 
3350                 m->max_data_items = max_data_items;
3351         }
3352 
3353         dout("ceph_msg_new %p front %d\n", m, front_len);
3354         return m;
3355 
3356 out2:
3357         ceph_msg_put(m);
3358 out:
3359         if (!can_fail) {
3360                 pr_err("msg_new can't create type %d front %d\n", type,
3361                        front_len);
3362                 WARN_ON(1);
3363         } else {
3364                 dout("msg_new can't create type %d front %d\n", type,
3365                      front_len);
3366         }
3367         return NULL;
3368 }
3369 EXPORT_SYMBOL(ceph_msg_new2);
3370 
3371 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3372                               bool can_fail)
3373 {
3374         return ceph_msg_new2(type, front_len, 0, flags, can_fail);
3375 }
3376 EXPORT_SYMBOL(ceph_msg_new);
3377 
3378 /*
3379  * Allocate "middle" portion of a message, if it is needed and wasn't
3380  * allocated by alloc_msg.  This allows us to read a small fixed-size
3381  * per-type header in the front and then gracefully fail (i.e.,
3382  * propagate the error to the caller based on info in the front) when
3383  * the middle is too large.
3384  */
3385 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3386 {
3387         int type = le16_to_cpu(msg->hdr.type);
3388         int middle_len = le32_to_cpu(msg->hdr.middle_len);
3389 
3390         dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3391              ceph_msg_type_name(type), middle_len);
3392         BUG_ON(!middle_len);
3393         BUG_ON(msg->middle);
3394 
3395         msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3396         if (!msg->middle)
3397                 return -ENOMEM;
3398         return 0;
3399 }
3400 
3401 /*
3402  * Allocate a message for receiving an incoming message on a
3403  * connection, and save the result in con->in_msg.  Uses the
3404  * connection's private alloc_msg op if available.
3405  *
3406  * Returns 0 on success, or a negative error code.
3407  *
3408  * On success, if we set *skip = 1:
3409  *  - the next message should be skipped and ignored.
3410  *  - con->in_msg == NULL
3411  * or if we set *skip = 0:
3412  *  - con->in_msg is non-null.
3413  * On error (ENOMEM, EAGAIN, ...),
3414  *  - con->in_msg == NULL
3415  */
3416 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3417 {
3418         struct ceph_msg_header *hdr = &con->in_hdr;
3419         int middle_len = le32_to_cpu(hdr->middle_len);
3420         struct ceph_msg *msg;
3421         int ret = 0;
3422 
3423         BUG_ON(con->in_msg != NULL);
3424         BUG_ON(!con->ops->alloc_msg);
3425 
3426         mutex_unlock(&con->mutex);
3427         msg = con->ops->alloc_msg(con, hdr, skip);
3428         mutex_lock(&con->mutex);
3429         if (con->state != CON_STATE_OPEN) {
3430                 if (msg)
3431                         ceph_msg_put(msg);
3432                 return -EAGAIN;
3433         }
3434         if (msg) {
3435                 BUG_ON(*skip);
3436                 msg_con_set(msg, con);
3437                 con->in_msg = msg;
3438         } else {
3439                 /*
3440                  * Null message pointer means either we should skip
3441                  * this message or we couldn't allocate memory.  The
3442                  * former is not an error.
3443                  */
3444                 if (*skip)
3445                         return 0;
3446 
3447                 con->error_msg = "error allocating memory for incoming message";
3448                 return -ENOMEM;
3449         }
3450         memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3451 
3452         if (middle_len && !con->in_msg->middle) {
3453                 ret = ceph_alloc_middle(con, con->in_msg);
3454                 if (ret < 0) {
3455                         ceph_msg_put(con->in_msg);
3456                         con->in_msg = NULL;
3457                 }
3458         }
3459 
3460         return ret;
3461 }
3462 
3463 
3464 /*
3465  * Free a generically kmalloc'd message.
3466  */
3467 static void ceph_msg_free(struct ceph_msg *m)
3468 {
3469         dout("%s %p\n", __func__, m);
3470         kvfree(m->front.iov_base);
3471         kfree(m->data);
3472         kmem_cache_free(ceph_msg_cache, m);
3473 }
3474 
3475 static void ceph_msg_release(struct kref *kref)
3476 {
3477         struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3478         int i;
3479 
3480         dout("%s %p\n", __func__, m);
3481         WARN_ON(!list_empty(&m->list_head));
3482 
3483         msg_con_set(m, NULL);
3484 
3485         /* drop middle, data, if any */
3486         if (m->middle) {
3487                 ceph_buffer_put(m->middle);
3488                 m->middle = NULL;
3489         }
3490 
3491         for (i = 0; i < m->num_data_items; i++)
3492                 ceph_msg_data_destroy(&m->data[i]);
3493 
3494         if (m->pool)
3495                 ceph_msgpool_put(m->pool, m);
3496         else
3497                 ceph_msg_free(m);
3498 }
3499 
3500 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3501 {
3502         dout("%s %p (was %d)\n", __func__, msg,
3503              kref_read(&msg->kref));
3504         kref_get(&msg->kref);
3505         return msg;
3506 }
3507 EXPORT_SYMBOL(ceph_msg_get);
3508 
3509 void ceph_msg_put(struct ceph_msg *msg)
3510 {
3511         dout("%s %p (was %d)\n", __func__, msg,
3512              kref_read(&msg->kref));
3513         kref_put(&msg->kref, ceph_msg_release);
3514 }
3515 EXPORT_SYMBOL(ceph_msg_put);
3516 
3517 void ceph_msg_dump(struct ceph_msg *msg)
3518 {
3519         pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3520                  msg->front_alloc_len, msg->data_length);
3521         print_hex_dump(KERN_DEBUG, "header: ",
3522                        DUMP_PREFIX_OFFSET, 16, 1,
3523                        &msg->hdr, sizeof(msg->hdr), true);
3524         print_hex_dump(KERN_DEBUG, " front: ",
3525                        DUMP_PREFIX_OFFSET, 16, 1,
3526                        msg->front.iov_base, msg->front.iov_len, true);
3527         if (msg->middle)
3528                 print_hex_dump(KERN_DEBUG, "middle: ",
3529                                DUMP_PREFIX_OFFSET, 16, 1,
3530                                msg->middle->vec.iov_base,
3531                                msg->middle->vec.iov_len, true);
3532         print_hex_dump(KERN_DEBUG, "footer: ",
3533                        DUMP_PREFIX_OFFSET, 16, 1,
3534                        &msg->footer, sizeof(msg->footer), true);
3535 }
3536 EXPORT_SYMBOL(ceph_msg_dump);
3537 

~ [ 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