projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: kill message trail
[deliverable/linux.git] / net / ceph / messenger.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
5a0e3ad6 9#include <linux/slab.h>
31b8006e
SW		 10#include <linux/socket.h>
11#include <linux/string.h>
3ebc21f7 12#ifdef CONFIG_BLOCK
68b4476b 13#include <linux/bio.h>
3ebc21f7 14#endif /* CONFIG_BLOCK */
ee3b56f2 15#include <linux/dns_resolver.h>
31b8006e
SW		 16#include <net/tcp.h>
17
3d14c5d2
YS		 18#include <linux/ceph/libceph.h>
19#include <linux/ceph/messenger.h>
20#include <linux/ceph/decode.h>
21#include <linux/ceph/pagelist.h>
bc3b2d7f 22#include <linux/export.h>
31b8006e 23
fe38a2b6
AE		 24#define list_entry_next(pos, member) \
25 list_entry(pos->member.next, typeof(*pos), member)
26
31b8006e
SW		 27/*
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
33 * the sender.
34 */
35
bc18f4b1
AE		 36/*
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
40 * unexpected state.
41 *
42 * --------
43 * | NEW* | transient initial state
44 * --------
45 * | con_sock_state_init()
46 * v
47 * ----------
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
50 * ^ \
51 * | \ con_sock_state_connecting()
52 * | ----------------------
53 * | \
54 * + con_sock_state_closed() \
fbb85a47
SW		 55 * |+--------------------------- \
56 * | \ \ \
57 * | ----------- \ \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
60 * | ^ \ |
61 * | | \ |
62 * | + con_sock_state_closing() \ |
63 * | / \ | |
64 * | / --------------- | |
65 * | / \ v v
bc18f4b1
AE		 66 * | / --------------
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
70 * | | v
71 * -------------
72 * | CONNECTED | TCP connection established
73 * -------------
74 *
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
76 */
ce2c8903
AE		 77
78#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
83
8dacc7da
SW		 84/*
85 * connection states
86 */
87#define CON_STATE_CLOSED 1 /* -> PREOPEN */
88#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
93
4a861692
SW		 94/*
95 * ceph_connection flag bits
96 */
97#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 103
c9ffc77a
AE		 104static bool con_flag_valid(unsigned long con_flag)
105{
106 switch (con_flag) {
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
112 return true;
113 default:
114 return false;
115 }
116}
117
118static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
119{
120 BUG_ON(!con_flag_valid(con_flag));
121
122 clear_bit(con_flag, &con->flags);
123}
124
125static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
126{
127 BUG_ON(!con_flag_valid(con_flag));
128
129 set_bit(con_flag, &con->flags);
130}
131
132static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
133{
134 BUG_ON(!con_flag_valid(con_flag));
135
136 return test_bit(con_flag, &con->flags);
137}
138
139static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
141{
142 BUG_ON(!con_flag_valid(con_flag));
143
144 return test_and_clear_bit(con_flag, &con->flags);
145}
146
147static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
149{
150 BUG_ON(!con_flag_valid(con_flag));
151
152 return test_and_set_bit(con_flag, &con->flags);
153}
154
31b8006e
SW		 155/* static tag bytes (protocol control messages) */
156static char tag_msg = CEPH_MSGR_TAG_MSG;
157static char tag_ack = CEPH_MSGR_TAG_ACK;
158static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
159
a6a5349d
SW		 160#ifdef CONFIG_LOCKDEP
161static struct lock_class_key socket_class;
162#endif
163
84495f49
AE		 164/*
165 * When skipping (ignoring) a block of input we read it into a "skip
166 * buffer," which is this many bytes in size.
167 */
168#define SKIP_BUF_SIZE 1024
31b8006e
SW		 169
170static void queue_con(struct ceph_connection *con);
171static void con_work(struct work_struct *);
93209264 172static void con_fault(struct ceph_connection *con);
31b8006e 173
31b8006e 174/*
f64a9317
AE		 175 * Nicely render a sockaddr as a string. An array of formatted
176 * strings is used, to approximate reentrancy.
31b8006e 177 */
f64a9317
AE		 178#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
179#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
180#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
181#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
182
183static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
184static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 185
57666519 186static struct page *zero_page; /* used in certain error cases */
57666519 187
3d14c5d2 188const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 189{
190 int i;
191 char *s;
99f0f3b2
AE		 192 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
193 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 194
f64a9317 195 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 196 s = addr_str[i];
197
198 switch (ss->ss_family) {
199 case AF_INET:
bd406145
AE		 200 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
201 ntohs(in4->sin_port));
31b8006e
SW		 202 break;
203
204 case AF_INET6:
bd406145
AE		 205 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
206 ntohs(in6->sin6_port));
31b8006e
SW		 207 break;
208
209 default:
d3002b97
AE		 210 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
211 ss->ss_family);
31b8006e
SW		 212 }
213
214 return s;
215}
3d14c5d2 216EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 217
63f2d211
SW		 218static void encode_my_addr(struct ceph_messenger *msgr)
219{
220 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
221 ceph_encode_addr(&msgr->my_enc_addr);
222}
223
31b8006e
SW		 224/*
225 * work queue for all reading and writing to/from the socket.
226 */
e0f43c94 227static struct workqueue_struct *ceph_msgr_wq;
31b8006e 228
15417167 229static void _ceph_msgr_exit(void)
6173d1f0 230{
d3002b97 231 if (ceph_msgr_wq) {
6173d1f0 232 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 233 ceph_msgr_wq = NULL;
234 }
6173d1f0 235
6173d1f0
AE		 236 BUG_ON(zero_page == NULL);
237 kunmap(zero_page);
238 page_cache_release(zero_page);
239 zero_page = NULL;
240}
241
3d14c5d2 242int ceph_msgr_init(void)
31b8006e 243{
57666519
AE		 244 BUG_ON(zero_page != NULL);
245 zero_page = ZERO_PAGE(0);
246 page_cache_get(zero_page);
247
f363e45f 248 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
6173d1f0
AE		 249 if (ceph_msgr_wq)
250 return 0;
57666519 251
6173d1f0
AE		 252 pr_err("msgr_init failed to create workqueue\n");
253 _ceph_msgr_exit();
57666519 254
6173d1f0 255 return -ENOMEM;
31b8006e 256}
3d14c5d2 257EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 258
259void ceph_msgr_exit(void)
260{
57666519 261 BUG_ON(ceph_msgr_wq == NULL);
57666519 262
6173d1f0 263 _ceph_msgr_exit();
31b8006e 264}
3d14c5d2 265EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 266
cd84db6e 267void ceph_msgr_flush(void)
a922d38f
SW		 268{
269 flush_workqueue(ceph_msgr_wq);
270}
3d14c5d2 271EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 272
ce2c8903
AE		 273/* Connection socket state transition functions */
274
275static void con_sock_state_init(struct ceph_connection *con)
276{
277 int old_state;
278
279 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
280 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
281 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 282 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
283 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 284}
285
286static void con_sock_state_connecting(struct ceph_connection *con)
287{
288 int old_state;
289
290 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
291 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
292 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 293 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
294 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 295}
296
297static void con_sock_state_connected(struct ceph_connection *con)
298{
299 int old_state;
300
301 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
302 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
303 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 304 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
305 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 306}
307
308static void con_sock_state_closing(struct ceph_connection *con)
309{
310 int old_state;
311
312 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
313 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
314 old_state != CON_SOCK_STATE_CONNECTED &&
315 old_state != CON_SOCK_STATE_CLOSING))
316 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 317 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
318 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 319}
320
321static void con_sock_state_closed(struct ceph_connection *con)
322{
323 int old_state;
324
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 327 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 328 old_state != CON_SOCK_STATE_CONNECTING &&
329 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 330 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 331 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
332 CON_SOCK_STATE_CLOSED);
ce2c8903 333}
a922d38f 334
31b8006e
SW		 335/*
336 * socket callback functions
337 */
338
339/* data available on socket, or listen socket received a connect */
327800bd 340static void ceph_sock_data_ready(struct sock *sk, int count_unused)
31b8006e 341{
bd406145 342 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 343 if (atomic_read(&con->msgr->stopping)) {
344 return;
345 }
bd406145 346
31b8006e 347 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 348 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 349 con, con->state);
350 queue_con(con);
351 }
352}
353
354/* socket has buffer space for writing */
327800bd 355static void ceph_sock_write_space(struct sock *sk)
31b8006e 356{
d3002b97 357 struct ceph_connection *con = sk->sk_user_data;
31b8006e 358
182fac26
JS		 359 /* only queue to workqueue if there is data we want to write,
360 * and there is sufficient space in the socket buffer to accept
327800bd 361 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 362 * doesn't get called again until try_write() fills the socket
363 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
364 * and net/core/stream.c:sk_stream_write_space().
365 */
c9ffc77a 366 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
182fac26 367 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
327800bd 368 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 369 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
370 queue_con(con);
371 }
31b8006e 372 } else {
327800bd 373 dout("%s %p nothing to write\n", __func__, con);
31b8006e 374 }
31b8006e
SW		 375}
376
377/* socket's state has changed */
327800bd 378static void ceph_sock_state_change(struct sock *sk)
31b8006e 379{
bd406145 380 struct ceph_connection *con = sk->sk_user_data;
31b8006e 381
327800bd 382 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 383 con, con->state, sk->sk_state);
384
31b8006e
SW		 385 switch (sk->sk_state) {
386 case TCP_CLOSE:
327800bd 387 dout("%s TCP_CLOSE\n", __func__);
31b8006e 388 case TCP_CLOSE_WAIT:
327800bd 389 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 390 con_sock_state_closing(con);
c9ffc77a 391 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 392 queue_con(con);
31b8006e
SW		 393 break;
394 case TCP_ESTABLISHED:
327800bd 395 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 396 con_sock_state_connected(con);
31b8006e
SW		 397 queue_con(con);
398 break;
d3002b97
AE		 399 default: /* Everything else is uninteresting */
400 break;
31b8006e
SW		 401 }
402}
403
404/*
405 * set up socket callbacks
406 */
407static void set_sock_callbacks(struct socket *sock,
408 struct ceph_connection *con)
409{
410 struct sock *sk = sock->sk;
bd406145 411 sk->sk_user_data = con;
327800bd
AE		 412 sk->sk_data_ready = ceph_sock_data_ready;
413 sk->sk_write_space = ceph_sock_write_space;
414 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 415}
416
417
418/*
419 * socket helpers
420 */
421
422/*
423 * initiate connection to a remote socket.
424 */
41617d0c 425static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 426{
f91d3471 427 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e
SW		 428 struct socket *sock;
429 int ret;
430
431 BUG_ON(con->sock);
f91d3471
SW		 432 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
433 IPPROTO_TCP, &sock);
31b8006e 434 if (ret)
41617d0c 435 return ret;
31b8006e
SW		 436 sock->sk->sk_allocation = GFP_NOFS;
437
a6a5349d
SW		 438#ifdef CONFIG_LOCKDEP
439 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
440#endif
441
31b8006e
SW		 442 set_sock_callbacks(sock, con);
443
3d14c5d2 444 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 445
89a86be0 446 con_sock_state_connecting(con);
f91d3471
SW		 447 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
448 O_NONBLOCK);
31b8006e
SW		 449 if (ret == -EINPROGRESS) {
450 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 451 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 452 sock->sk->sk_state);
a5bc3129 453 } else if (ret < 0) {
31b8006e 454 pr_err("connect %s error %d\n",
3d14c5d2 455 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 456 sock_release(sock);
31b8006e 457 con->error_msg = "connect error";
31b8006e 458
41617d0c 459 return ret;
a5bc3129
AE		 460 }
461 con->sock = sock;
41617d0c 462 return 0;
31b8006e
SW		 463}
464
465static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
466{
467 struct kvec iov = {buf, len};
468 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 469 int r;
31b8006e 470
98bdb0aa
SW		 471 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
472 if (r == -EAGAIN)
473 r = 0;
474 return r;
31b8006e
SW		 475}
476
afb3d90e
AE		 477static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
478 int page_offset, size_t length)
479{
480 void *kaddr;
481 int ret;
482
483 BUG_ON(page_offset + length > PAGE_SIZE);
484
485 kaddr = kmap(page);
486 BUG_ON(!kaddr);
487 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
488 kunmap(page);
489
490 return ret;
491}
492
31b8006e
SW		 493/*
494 * write something. @more is true if caller will be sending more data
495 * shortly.
496 */
497static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
498 size_t kvlen, size_t len, int more)
499{
500 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 501 int r;
31b8006e
SW		 502
503 if (more)
504 msg.msg_flags |= MSG_MORE;
505 else
506 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
507
42961d23
SW		 508 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
509 if (r == -EAGAIN)
510 r = 0;
511 return r;
31b8006e
SW		 512}
513
31739139 514static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 515 int offset, size_t size, bool more)
31739139
AE		 516{
517 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
518 int ret;
519
520 ret = kernel_sendpage(sock, page, offset, size, flags);
521 if (ret == -EAGAIN)
522 ret = 0;
523
524 return ret;
525}
526
31b8006e
SW		 527
528/*
529 * Shutdown/close the socket for the given connection.
530 */
531static int con_close_socket(struct ceph_connection *con)
532{
8007b8d6 533 int rc = 0;
31b8006e
SW		 534
535 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 536 if (con->sock) {
537 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
538 sock_release(con->sock);
539 con->sock = NULL;
540 }
456ea468
AE		 541
542 /*
4a861692 543 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 544 * independent of the connection mutex, and we could have
545 * received a socket close event before we had the chance to
546 * shut the socket down.
547 */
c9ffc77a 548 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 549
ce2c8903 550 con_sock_state_closed(con);
31b8006e
SW		 551 return rc;
552}
553
554/*
555 * Reset a connection. Discard all incoming and outgoing messages
556 * and clear *_seq state.
557 */
558static void ceph_msg_remove(struct ceph_msg *msg)
559{
560 list_del_init(&msg->list_head);
38941f80 561 BUG_ON(msg->con == NULL);
36eb71aa 562 msg->con->ops->put(msg->con);
38941f80
AE		 563 msg->con = NULL;
564
31b8006e
SW		 565 ceph_msg_put(msg);
566}
567static void ceph_msg_remove_list(struct list_head *head)
568{
569 while (!list_empty(head)) {
570 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
571 list_head);
572 ceph_msg_remove(msg);
573 }
574}
575
576static void reset_connection(struct ceph_connection *con)
577{
578 /* reset connection, out_queue, msg_ and connect_seq */
579 /* discard existing out_queue and msg_seq */
0fa6ebc6 580 dout("reset_connection %p\n", con);
31b8006e
SW		 581 ceph_msg_remove_list(&con->out_queue);
582 ceph_msg_remove_list(&con->out_sent);
583
cf3e5c40 584 if (con->in_msg) {
38941f80
AE		 585 BUG_ON(con->in_msg->con != con);
586 con->in_msg->con = NULL;
cf3e5c40
SW		 587 ceph_msg_put(con->in_msg);
588 con->in_msg = NULL;
36eb71aa 589 con->ops->put(con);
cf3e5c40
SW		 590 }
591
31b8006e
SW		 592 con->connect_seq = 0;
593 con->out_seq = 0;
c86a2930
SW		 594 if (con->out_msg) {
595 ceph_msg_put(con->out_msg);
596 con->out_msg = NULL;
597 }
31b8006e 598 con->in_seq = 0;
0e0d5e0c 599 con->in_seq_acked = 0;
31b8006e
SW		 600}
601
602/*
603 * mark a peer down. drop any open connections.
604 */
605void ceph_con_close(struct ceph_connection *con)
606{
8c50c817 607 mutex_lock(&con->mutex);
3d14c5d2
YS		 608 dout("con_close %p peer %s\n", con,
609 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 610 con->state = CON_STATE_CLOSED;
a5988c49 611
c9ffc77a
AE		 612 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
613 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
614 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
615 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 616
31b8006e 617 reset_connection(con);
6f2bc3ff 618 con->peer_global_seq = 0;
91e45ce3 619 cancel_delayed_work(&con->work);
ee76e073 620 con_close_socket(con);
ec302645 621 mutex_unlock(&con->mutex);
31b8006e 622}
3d14c5d2 623EXPORT_SYMBOL(ceph_con_close);
31b8006e 624
31b8006e
SW		 625/*
626 * Reopen a closed connection, with a new peer address.
627 */
b7a9e5dd
SW		 628void ceph_con_open(struct ceph_connection *con,
629 __u8 entity_type, __u64 entity_num,
630 struct ceph_entity_addr *addr)
31b8006e 631{
5469155f 632 mutex_lock(&con->mutex);
3d14c5d2 633 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 634
122070a2 635 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 636 con->state = CON_STATE_PREOPEN;
a5988c49 637
b7a9e5dd
SW		 638 con->peer_name.type = (__u8) entity_type;
639 con->peer_name.num = cpu_to_le64(entity_num);
640
31b8006e 641 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 642 con->delay = 0; /* reset backoff memory */
5469155f 643 mutex_unlock(&con->mutex);
31b8006e
SW		 644 queue_con(con);
645}
3d14c5d2 646EXPORT_SYMBOL(ceph_con_open);
31b8006e 647
87b315a5
SW		 648/*
649 * return true if this connection ever successfully opened
650 */
651bool ceph_con_opened(struct ceph_connection *con)
652{
653 return con->connect_seq > 0;
654}
655
31b8006e
SW		 656/*
657 * initialize a new connection.
658 */
1bfd89f4
AE		 659void ceph_con_init(struct ceph_connection *con, void *private,
660 const struct ceph_connection_operations *ops,
b7a9e5dd 661 struct ceph_messenger *msgr)
31b8006e
SW		 662{
663 dout("con_init %p\n", con);
664 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 665 con->private = private;
666 con->ops = ops;
31b8006e 667 con->msgr = msgr;
ce2c8903
AE		 668
669 con_sock_state_init(con);
670
ec302645 671 mutex_init(&con->mutex);
31b8006e
SW		 672 INIT_LIST_HEAD(&con->out_queue);
673 INIT_LIST_HEAD(&con->out_sent);
674 INIT_DELAYED_WORK(&con->work, con_work);
a5988c49 675
8dacc7da 676 con->state = CON_STATE_CLOSED;
31b8006e 677}
3d14c5d2 678EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 679
680
681/*
682 * We maintain a global counter to order connection attempts. Get
683 * a unique seq greater than @gt.
684 */
685static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
686{
687 u32 ret;
688
689 spin_lock(&msgr->global_seq_lock);
690 if (msgr->global_seq < gt)
691 msgr->global_seq = gt;
692 ret = ++msgr->global_seq;
693 spin_unlock(&msgr->global_seq_lock);
694 return ret;
695}
696
e2200423 697static void con_out_kvec_reset(struct ceph_connection *con)
859eb799
AE		 698{
699 con->out_kvec_left = 0;
700 con->out_kvec_bytes = 0;
701 con->out_kvec_cur = &con->out_kvec[0];
702}
703
e2200423 704static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 705 size_t size, void *data)
706{
707 int index;
708
709 index = con->out_kvec_left;
710 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
711
712 con->out_kvec[index].iov_len = size;
713 con->out_kvec[index].iov_base = data;
714 con->out_kvec_left++;
715 con->out_kvec_bytes += size;
716}
31b8006e 717
df6ad1f9 718#ifdef CONFIG_BLOCK
07c09b72
AE		 719static void init_bio_iter(struct bio *bio, struct bio **bio_iter,
720 unsigned int *bio_seg)
df6ad1f9
AE		 721{
722 if (!bio) {
07c09b72
AE		 723 *bio_iter = NULL;
724 *bio_seg = 0;
df6ad1f9
AE		 725 return;
726 }
07c09b72
AE		 727 *bio_iter = bio;
728 *bio_seg = (unsigned int) bio->bi_idx;
df6ad1f9
AE		 729}
730
07c09b72 731static void iter_bio_next(struct bio **bio_iter, unsigned int *seg)
df6ad1f9
AE		 732{
733 if (*bio_iter == NULL)
734 return;
735
736 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
737
738 (*seg)++;
739 if (*seg == (*bio_iter)->bi_vcnt)
740 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
741}
6aaa4511
AE		 742
743/*
744 * For a bio data item, a piece is whatever remains of the next
745 * entry in the current bio iovec, or the first entry in the next
746 * bio in the list.
747 */
748static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data *data)
749{
750 struct ceph_msg_data_cursor *cursor = &data->cursor;
751 struct bio *bio;
752
753 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
754
755 bio = data->bio;
756 BUG_ON(!bio);
757 BUG_ON(!bio->bi_vcnt);
758 /* resid = bio->bi_size */
759
760 cursor->bio = bio;
761 cursor->vector_index = 0;
762 cursor->vector_offset = 0;
763 cursor->last_piece = !bio->bi_next && bio->bi_vcnt == 1;
764}
765
766static struct page *ceph_msg_data_bio_next(struct ceph_msg_data *data,
767 size_t *page_offset,
768 size_t *length)
769{
770 struct ceph_msg_data_cursor *cursor = &data->cursor;
771 struct bio *bio;
772 struct bio_vec *bio_vec;
773 unsigned int index;
774
775 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
776
777 bio = cursor->bio;
778 BUG_ON(!bio);
779
780 index = cursor->vector_index;
781 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
782
783 bio_vec = &bio->bi_io_vec[index];
784 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
785 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
786 BUG_ON(*page_offset >= PAGE_SIZE);
787 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
788 BUG_ON(*length > PAGE_SIZE);
789
790 return bio_vec->bv_page;
791}
792
793static bool ceph_msg_data_bio_advance(struct ceph_msg_data *data, size_t bytes)
794{
795 struct ceph_msg_data_cursor *cursor = &data->cursor;
796 struct bio *bio;
797 struct bio_vec *bio_vec;
798 unsigned int index;
799
800 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
801
802 bio = cursor->bio;
803 BUG_ON(!bio);
804
805 index = cursor->vector_index;
806 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
807 bio_vec = &bio->bi_io_vec[index];
808 BUG_ON(cursor->vector_offset + bytes > bio_vec->bv_len);
809
810 /* Advance the cursor offset */
811
812 cursor->vector_offset += bytes;
813 if (cursor->vector_offset < bio_vec->bv_len)
814 return false; /* more bytes to process in this segment */
815
816 /* Move on to the next segment, and possibly the next bio */
817
818 if (++cursor->vector_index == (unsigned int) bio->bi_vcnt) {
819 bio = bio->bi_next;
820 cursor->bio = bio;
821 cursor->vector_index = 0;
822 }
823 cursor->vector_offset = 0;
824
825 if (!cursor->last_piece && bio && !bio->bi_next)
826 if (cursor->vector_index == (unsigned int) bio->bi_vcnt - 1)
827 cursor->last_piece = true;
828
829 return true;
830}
df6ad1f9
AE		 831#endif
832
e766d7b5
AE		 833/*
834 * For a page array, a piece comes from the first page in the array
835 * that has not already been fully consumed.
836 */
837static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data *data)
838{
839 struct ceph_msg_data_cursor *cursor = &data->cursor;
840 int page_count;
841
842 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
843
844 BUG_ON(!data->pages);
845 BUG_ON(!data->length);
846
847 page_count = calc_pages_for(data->alignment, (u64)data->length);
848 BUG_ON(page_count > (int) USHRT_MAX);
849 cursor->resid = data->length;
850 cursor->page_offset = data->alignment & ~PAGE_MASK;
851 cursor->page_index = 0;
852 cursor->page_count = (unsigned short) page_count;
853 cursor->last_piece = cursor->page_count == 1;
854}
855
856static struct page *ceph_msg_data_pages_next(struct ceph_msg_data *data,
857 size_t *page_offset,
858 size_t *length)
859{
860 struct ceph_msg_data_cursor *cursor = &data->cursor;
861
862 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
863
864 BUG_ON(cursor->page_index >= cursor->page_count);
865 BUG_ON(cursor->page_offset >= PAGE_SIZE);
866 BUG_ON(!cursor->resid);
867
868 *page_offset = cursor->page_offset;
869 if (cursor->last_piece) {
870 BUG_ON(*page_offset + cursor->resid > PAGE_SIZE);
871 *length = cursor->resid;
872 } else {
873 *length = PAGE_SIZE - *page_offset;
874 }
875
876 return data->pages[cursor->page_index];
877}
878
879static bool ceph_msg_data_pages_advance(struct ceph_msg_data *data,
880 size_t bytes)
881{
882 struct ceph_msg_data_cursor *cursor = &data->cursor;
883
884 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
885
886 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
887 BUG_ON(bytes > cursor->resid);
888
889 /* Advance the cursor page offset */
890
891 cursor->resid -= bytes;
892 cursor->page_offset += bytes;
893 if (!bytes || cursor->page_offset & ~PAGE_MASK)
894 return false; /* more bytes to process in the current page */
895
896 /* Move on to the next page */
897
898 BUG_ON(cursor->page_index >= cursor->page_count);
899 cursor->page_offset = 0;
900 cursor->page_index++;
901 cursor->last_piece = cursor->page_index == cursor->page_count - 1;
902
903 return true;
904}
905
fe38a2b6 906/*
dd236fcb
AE		 907 * For a pagelist, a piece is whatever remains to be consumed in the
908 * first page in the list, or the front of the next page.
fe38a2b6 909 */
dd236fcb 910static void ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data *data)
fe38a2b6
AE		 911{
912 struct ceph_msg_data_cursor *cursor = &data->cursor;
913 struct ceph_pagelist *pagelist;
914 struct page *page;
915
dd236fcb 916 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 917
918 pagelist = data->pagelist;
919 BUG_ON(!pagelist);
920 if (!pagelist->length)
921 return; /* pagelist can be assigned but empty */
922
923 BUG_ON(list_empty(&pagelist->head));
924 page = list_first_entry(&pagelist->head, struct page, lru);
925
926 cursor->page = page;
927 cursor->offset = 0;
928 cursor->last_piece = pagelist->length <= PAGE_SIZE;
929}
930
dd236fcb 931static struct page *ceph_msg_data_pagelist_next(struct ceph_msg_data *data,
fe38a2b6 932 size_t *page_offset,
dd236fcb 933 size_t *length)
fe38a2b6
AE		 934{
935 struct ceph_msg_data_cursor *cursor = &data->cursor;
936 struct ceph_pagelist *pagelist;
937 size_t piece_end;
938
939 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
940
941 pagelist = data->pagelist;
942 BUG_ON(!pagelist);
943
944 BUG_ON(!cursor->page);
945 BUG_ON(cursor->offset >= pagelist->length);
946
dd236fcb 947 if (cursor->last_piece) {
fe38a2b6
AE		 948 /* pagelist offset is always 0 */
949 piece_end = pagelist->length & ~PAGE_MASK;
950 if (!piece_end)
951 piece_end = PAGE_SIZE;
952 } else {
953 piece_end = PAGE_SIZE;
954 }
955 *page_offset = cursor->offset & ~PAGE_MASK;
956 *length = piece_end - *page_offset;
957
958 return data->cursor.page;
959}
960
dd236fcb
AE		 961static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data *data,
962 size_t bytes)
fe38a2b6
AE		 963{
964 struct ceph_msg_data_cursor *cursor = &data->cursor;
965 struct ceph_pagelist *pagelist;
966
967 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
968
969 pagelist = data->pagelist;
970 BUG_ON(!pagelist);
971 BUG_ON(!cursor->page);
972 BUG_ON(cursor->offset + bytes > pagelist->length);
973 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
974
975 /* Advance the cursor offset */
976
977 cursor->offset += bytes;
978 /* pagelist offset is always 0 */
979 if (!bytes || cursor->offset & ~PAGE_MASK)
980 return false; /* more bytes to process in the current page */
981
982 /* Move on to the next page */
983
984 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
985 cursor->page = list_entry_next(cursor->page, lru);
986
987 /* cursor offset is at page boundary; pagelist offset is always 0 */
988 if (pagelist->length - cursor->offset <= PAGE_SIZE)
989 cursor->last_piece = true;
990
991 return true;
992}
993
dd236fcb
AE		 994/*
995 * Message data is handled (sent or received) in pieces, where each
996 * piece resides on a single page. The network layer might not
997 * consume an entire piece at once. A data item's cursor keeps
998 * track of which piece is next to process and how much remains to
999 * be processed in that piece. It also tracks whether the current
1000 * piece is the last one in the data item.
1001 */
1002static void ceph_msg_data_cursor_init(struct ceph_msg_data *data)
1003{
1004 switch (data->type) {
1005 case CEPH_MSG_DATA_PAGELIST:
1006 ceph_msg_data_pagelist_cursor_init(data);
1007 break;
e766d7b5
AE		 1008 case CEPH_MSG_DATA_PAGES:
1009 ceph_msg_data_pages_cursor_init(data);
1010 break;
dd236fcb
AE		 1011#ifdef CONFIG_BLOCK
1012 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1013 ceph_msg_data_bio_cursor_init(data);
1014 break;
dd236fcb 1015#endif /* CONFIG_BLOCK */
6aaa4511 1016 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1017 default:
1018 /* BUG(); */
1019 break;
1020 }
1021}
1022
1023/*
1024 * Return the page containing the next piece to process for a given
1025 * data item, and supply the page offset and length of that piece.
1026 * Indicate whether this is the last piece in this data item.
1027 */
1028static struct page *ceph_msg_data_next(struct ceph_msg_data *data,
1029 size_t *page_offset,
1030 size_t *length,
1031 bool *last_piece)
1032{
1033 struct page *page;
1034
1035 switch (data->type) {
1036 case CEPH_MSG_DATA_PAGELIST:
1037 page = ceph_msg_data_pagelist_next(data, page_offset, length);
1038 break;
e766d7b5
AE		 1039 case CEPH_MSG_DATA_PAGES:
1040 page = ceph_msg_data_pages_next(data, page_offset, length);
1041 break;
dd236fcb
AE		 1042#ifdef CONFIG_BLOCK
1043 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1044 page = ceph_msg_data_bio_next(data, page_offset, length);
1045 break;
dd236fcb 1046#endif /* CONFIG_BLOCK */
6aaa4511 1047 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1048 default:
1049 page = NULL;
1050 break;
1051 }
1052 BUG_ON(!page);
1053 BUG_ON(*page_offset + *length > PAGE_SIZE);
1054 BUG_ON(!*length);
1055 if (last_piece)
1056 *last_piece = data->cursor.last_piece;
1057
1058 return page;
1059}
1060
1061/*
1062 * Returns true if the result moves the cursor on to the next piece
1063 * of the data item.
1064 */
1065static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes)
1066{
1067 bool new_piece;
1068
1069 switch (data->type) {
1070 case CEPH_MSG_DATA_PAGELIST:
1071 new_piece = ceph_msg_data_pagelist_advance(data, bytes);
1072 break;
e766d7b5
AE		 1073 case CEPH_MSG_DATA_PAGES:
1074 new_piece = ceph_msg_data_pages_advance(data, bytes);
1075 break;
dd236fcb
AE		 1076#ifdef CONFIG_BLOCK
1077 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1078 new_piece = ceph_msg_data_bio_advance(data, bytes);
1079 break;
dd236fcb 1080#endif /* CONFIG_BLOCK */
6aaa4511 1081 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1082 default:
1083 BUG();
1084 break;
1085 }
1086
1087 return new_piece;
1088}
1089
78625051
AE		 1090static void prepare_message_data(struct ceph_msg *msg,
1091 struct ceph_msg_pos *msg_pos)
739c905b 1092{
739c905b
AE		 1093 BUG_ON(!msg);
1094 BUG_ON(!msg->hdr.data_len);
1095
1096 /* initialize page iterator */
bae6acd9 1097 msg_pos->page = 0;
97fb1c7f 1098 if (ceph_msg_has_pages(msg))
f9e15777 1099 msg_pos->page_pos = msg->p.alignment;
739c905b 1100 else
bae6acd9 1101 msg_pos->page_pos = 0;
572c588e 1102#ifdef CONFIG_BLOCK
97fb1c7f 1103 if (ceph_msg_has_bio(msg))
f9e15777 1104 init_bio_iter(msg->b.bio, &msg->b.bio_iter, &msg->b.bio_seg);
572c588e 1105#endif
bae6acd9 1106 msg_pos->data_pos = 0;
fe38a2b6 1107
7fe1e5e5 1108 /* Initialize data cursors */
fe38a2b6 1109
6aaa4511
AE		 1110#ifdef CONFIG_BLOCK
1111 if (ceph_msg_has_bio(msg))
1112 ceph_msg_data_cursor_init(&msg->b);
1113#endif /* CONFIG_BLOCK */
e766d7b5
AE		 1114 if (ceph_msg_has_pages(msg))
1115 ceph_msg_data_cursor_init(&msg->p);
7fe1e5e5
AE		 1116 if (ceph_msg_has_pagelist(msg))
1117 ceph_msg_data_cursor_init(&msg->l);
fe38a2b6 1118
bae6acd9 1119 msg_pos->did_page_crc = false;
739c905b
AE		 1120}
1121
31b8006e
SW		 1122/*
1123 * Prepare footer for currently outgoing message, and finish things
1124 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1125 */
859eb799 1126static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1127{
1128 struct ceph_msg *m = con->out_msg;
859eb799 1129 int v = con->out_kvec_left;
31b8006e 1130
fd154f3c
AE		 1131 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1132
31b8006e
SW		 1133 dout("prepare_write_message_footer %p\n", con);
1134 con->out_kvec_is_msg = true;
1135 con->out_kvec[v].iov_base = &m->footer;
1136 con->out_kvec[v].iov_len = sizeof(m->footer);
1137 con->out_kvec_bytes += sizeof(m->footer);
1138 con->out_kvec_left++;
1139 con->out_more = m->more_to_follow;
c86a2930 1140 con->out_msg_done = true;
31b8006e
SW		 1141}
1142
1143/*
1144 * Prepare headers for the next outgoing message.
1145 */
1146static void prepare_write_message(struct ceph_connection *con)
1147{
1148 struct ceph_msg *m;
a9a0c51a 1149 u32 crc;
31b8006e 1150
e2200423 1151 con_out_kvec_reset(con);
31b8006e 1152 con->out_kvec_is_msg = true;
c86a2930 1153 con->out_msg_done = false;
31b8006e
SW		 1154
1155 /* Sneak an ack in there first? If we can get it into the same
1156 * TCP packet that's a good thing. */
1157 if (con->in_seq > con->in_seq_acked) {
1158 con->in_seq_acked = con->in_seq;
e2200423 1159 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1160 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1161 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1162 &con->out_temp_ack);
31b8006e
SW		 1163 }
1164
38941f80 1165 BUG_ON(list_empty(&con->out_queue));
859eb799 1166 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1167 con->out_msg = m;
38941f80 1168 BUG_ON(m->con != con);
4cf9d544
SW		 1169
1170 /* put message on sent list */
1171 ceph_msg_get(m);
1172 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1173
e84346b7
SW		 1174 /*
1175 * only assign outgoing seq # if we haven't sent this message
1176 * yet. if it is requeued, resend with it's original seq.
1177 */
1178 if (m->needs_out_seq) {
1179 m->hdr.seq = cpu_to_le64(++con->out_seq);
1180 m->needs_out_seq = false;
1181 }
31b8006e 1182
4a73ef27 1183 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n",
31b8006e
SW		 1184 m, con->out_seq, le16_to_cpu(m->hdr.type),
1185 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
f9e15777 1186 le32_to_cpu(m->hdr.data_len), m->p.length);
31b8006e
SW		 1187 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1188
1189 /* tag + hdr + front + middle */
e2200423
AE		 1190 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1191 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1192 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1193
31b8006e 1194 if (m->middle)
e2200423 1195 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1196 m->middle->vec.iov_base);
31b8006e
SW		 1197
1198 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1199 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1200 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1201 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1202
1203 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1204 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1205 if (m->middle) {
1206 crc = crc32c(0, m->middle->vec.iov_base,
1207 m->middle->vec.iov_len);
1208 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1209 } else
31b8006e 1210 con->out_msg->footer.middle_crc = 0;
739c905b 1211 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1212 le32_to_cpu(con->out_msg->footer.front_crc),
1213 le32_to_cpu(con->out_msg->footer.middle_crc));
1214
1215 /* is there a data payload? */
739c905b 1216 con->out_msg->footer.data_crc = 0;
78625051
AE		 1217 if (m->hdr.data_len) {
1218 prepare_message_data(con->out_msg, &con->out_msg_pos);
1219 con->out_more = 1; /* data + footer will follow */
1220 } else {
31b8006e 1221 /* no, queue up footer too and be done */
859eb799 1222 prepare_write_message_footer(con);
78625051 1223 }
31b8006e 1224
c9ffc77a 1225 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1226}
1227
1228/*
1229 * Prepare an ack.
1230 */
1231static void prepare_write_ack(struct ceph_connection *con)
1232{
1233 dout("prepare_write_ack %p %llu -> %llu\n", con,
1234 con->in_seq_acked, con->in_seq);
1235 con->in_seq_acked = con->in_seq;
1236
e2200423 1237 con_out_kvec_reset(con);
859eb799 1238
e2200423 1239 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1240
31b8006e 1241 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1242 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1243 &con->out_temp_ack);
1244
31b8006e 1245 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1246 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1247}
1248
1249/*
1250 * Prepare to write keepalive byte.
1251 */
1252static void prepare_write_keepalive(struct ceph_connection *con)
1253{
1254 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1255 con_out_kvec_reset(con);
1256 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1257 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1258}
1259
1260/*
1261 * Connection negotiation.
1262 */
1263
dac1e716
AE		 1264static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1265 int *auth_proto)
4e7a5dcd 1266{
a3530df3 1267 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1268
1269 if (!con->ops->get_authorizer) {
1270 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1271 con->out_connect.authorizer_len = 0;
729796be 1272 return NULL;
b1c6b980
AE		 1273 }
1274
1275 /* Can't hold the mutex while getting authorizer */
ec302645 1276 mutex_unlock(&con->mutex);
dac1e716 1277 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1278 mutex_lock(&con->mutex);
4e7a5dcd 1279
a3530df3 1280 if (IS_ERR(auth))
729796be 1281 return auth;
8dacc7da 1282 if (con->state != CON_STATE_NEGOTIATING)
729796be 1283 return ERR_PTR(-EAGAIN);
0da5d703 1284
8f43fb53
AE		 1285 con->auth_reply_buf = auth->authorizer_reply_buf;
1286 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1287 return auth;
4e7a5dcd
SW		 1288}
1289
31b8006e
SW		 1290/*
1291 * We connected to a peer and are saying hello.
1292 */
e825a66d 1293static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1294{
e2200423
AE		 1295 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1296 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1297 &con->msgr->my_enc_addr);
eed0ef2c 1298
eed0ef2c 1299 con->out_more = 0;
c9ffc77a 1300 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1301}
1302
e825a66d 1303static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1304{
95c96174 1305 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1306 int proto;
dac1e716 1307 int auth_proto;
729796be 1308 struct ceph_auth_handshake *auth;
31b8006e
SW		 1309
1310 switch (con->peer_name.type) {
1311 case CEPH_ENTITY_TYPE_MON:
1312 proto = CEPH_MONC_PROTOCOL;
1313 break;
1314 case CEPH_ENTITY_TYPE_OSD:
1315 proto = CEPH_OSDC_PROTOCOL;
1316 break;
1317 case CEPH_ENTITY_TYPE_MDS:
1318 proto = CEPH_MDSC_PROTOCOL;
1319 break;
1320 default:
1321 BUG();
1322 }
1323
1324 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1325 con->connect_seq, global_seq, proto);
4e7a5dcd 1326
e825a66d 1327 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1328 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1329 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1330 con->out_connect.global_seq = cpu_to_le32(global_seq);
1331 con->out_connect.protocol_version = cpu_to_le32(proto);
1332 con->out_connect.flags = 0;
31b8006e 1333
dac1e716
AE		 1334 auth_proto = CEPH_AUTH_UNKNOWN;
1335 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1336 if (IS_ERR(auth))
1337 return PTR_ERR(auth);
3da54776 1338
dac1e716 1339 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1340 con->out_connect.authorizer_len = auth ?
1341 cpu_to_le32(auth->authorizer_buf_len) : 0;
1342
e2200423 1343 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1344 &con->out_connect);
1345 if (auth && auth->authorizer_buf_len)
e2200423 1346 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1347 auth->authorizer_buf);
859eb799 1348
31b8006e 1349 con->out_more = 0;
c9ffc77a 1350 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1351
e10c758e 1352 return 0;
31b8006e
SW		 1353}
1354
31b8006e
SW		 1355/*
1356 * write as much of pending kvecs to the socket as we can.
1357 * 1 -> done
1358 * 0 -> socket full, but more to do
1359 * <0 -> error
1360 */
1361static int write_partial_kvec(struct ceph_connection *con)
1362{
1363 int ret;
1364
1365 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1366 while (con->out_kvec_bytes > 0) {
1367 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1368 con->out_kvec_left, con->out_kvec_bytes,
1369 con->out_more);
1370 if (ret <= 0)
1371 goto out;
1372 con->out_kvec_bytes -= ret;
1373 if (con->out_kvec_bytes == 0)
1374 break; /* done */
f42299e6
AE		 1375
1376 /* account for full iov entries consumed */
1377 while (ret >= con->out_kvec_cur->iov_len) {
1378 BUG_ON(!con->out_kvec_left);
1379 ret -= con->out_kvec_cur->iov_len;
1380 con->out_kvec_cur++;
1381 con->out_kvec_left--;
1382 }
1383 /* and for a partially-consumed entry */
1384 if (ret) {
1385 con->out_kvec_cur->iov_len -= ret;
1386 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1387 }
1388 }
1389 con->out_kvec_left = 0;
1390 con->out_kvec_is_msg = false;
1391 ret = 1;
1392out:
1393 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1394 con->out_kvec_bytes, con->out_kvec_left, ret);
1395 return ret; /* done! */
1396}
1397
84ca8fc8 1398static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
9d2a06c2 1399 size_t len, size_t sent)
68b4476b 1400{
84ca8fc8 1401 struct ceph_msg *msg = con->out_msg;
bae6acd9 1402 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
7fe1e5e5 1403 bool need_crc = false;
68b4476b 1404
84ca8fc8
AE		 1405 BUG_ON(!msg);
1406 BUG_ON(!sent);
68b4476b 1407
bae6acd9
AE		 1408 msg_pos->data_pos += sent;
1409 msg_pos->page_pos += sent;
9d2a06c2 1410 if (ceph_msg_has_pages(msg))
e766d7b5 1411 need_crc = ceph_msg_data_advance(&msg->p, sent);
7fe1e5e5
AE		 1412 else if (ceph_msg_has_pagelist(msg))
1413 need_crc = ceph_msg_data_advance(&msg->l, sent);
6aaa4511
AE		 1414#ifdef CONFIG_BLOCK
1415 else if (ceph_msg_has_bio(msg))
1416 need_crc = ceph_msg_data_advance(&msg->b, sent);
1417#endif /* CONFIG_BLOCK */
7fe1e5e5
AE		 1418 BUG_ON(need_crc && sent != len);
1419
5821bd8c
AE		 1420 if (sent < len)
1421 return;
68b4476b 1422
5821bd8c 1423 BUG_ON(sent != len);
bae6acd9
AE		 1424 msg_pos->page_pos = 0;
1425 msg_pos->page++;
1426 msg_pos->did_page_crc = false;
84ca8fc8 1427}
68b4476b 1428
e788182f
AE		 1429static void in_msg_pos_next(struct ceph_connection *con, size_t len,
1430 size_t received)
1431{
1432 struct ceph_msg *msg = con->in_msg;
bae6acd9 1433 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
e788182f
AE		 1434
1435 BUG_ON(!msg);
1436 BUG_ON(!received);
1437
bae6acd9
AE		 1438 msg_pos->data_pos += received;
1439 msg_pos->page_pos += received;
e788182f
AE		 1440 if (received < len)
1441 return;
1442
1443 BUG_ON(received != len);
bae6acd9
AE		 1444 msg_pos->page_pos = 0;
1445 msg_pos->page++;
e788182f 1446#ifdef CONFIG_BLOCK
f9e15777
AE		 1447 if (msg->b.bio)
1448 iter_bio_next(&msg->b.bio_iter, &msg->b.bio_seg);
e788182f
AE		 1449#endif /* CONFIG_BLOCK */
1450}
1451
35b62808
AE		 1452static u32 ceph_crc32c_page(u32 crc, struct page *page,
1453 unsigned int page_offset,
1454 unsigned int length)
1455{
1456 char *kaddr;
1457
1458 kaddr = kmap(page);
1459 BUG_ON(kaddr == NULL);
1460 crc = crc32c(crc, kaddr + page_offset, length);
1461 kunmap(page);
1462
1463 return crc;
1464}
31b8006e
SW		 1465/*
1466 * Write as much message data payload as we can. If we finish, queue
1467 * up the footer.
1468 * 1 -> done, footer is now queued in out_kvec[].
1469 * 0 -> socket full, but more to do
1470 * <0 -> error
1471 */
34d2d200 1472static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1473{
1474 struct ceph_msg *msg = con->out_msg;
bae6acd9 1475 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
95c96174 1476 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
37675b0f 1477 bool do_datacrc = !con->msgr->nocrc;
31b8006e 1478 int ret;
68b4476b 1479 int total_max_write;
31b8006e 1480
34d2d200 1481 dout("%s %p msg %p page %d offset %d\n", __func__,
bae6acd9 1482 con, msg, msg_pos->page, msg_pos->page_pos);
31b8006e 1483
5821bd8c
AE		 1484 /*
1485 * Iterate through each page that contains data to be
1486 * written, and send as much as possible for each.
1487 *
1488 * If we are calculating the data crc (the default), we will
1489 * need to map the page. If we have no pages, they have
1490 * been revoked, so use the zero page.
1491 */
bae6acd9 1492 while (data_len > msg_pos->data_pos) {
31b8006e 1493 struct page *page = NULL;
e387d525
AE		 1494 size_t page_offset;
1495 size_t length;
fe38a2b6
AE		 1496 bool use_cursor = false;
1497 bool last_piece = true; /* preserve existing behavior */
68b4476b 1498
9d2a06c2 1499 total_max_write = data_len - msg_pos->data_pos;
68b4476b 1500
9d2a06c2 1501 if (ceph_msg_has_pages(msg)) {
e766d7b5
AE		 1502 use_cursor = true;
1503 page = ceph_msg_data_next(&msg->p, &page_offset,
1504 &length, &last_piece);
97fb1c7f 1505 } else if (ceph_msg_has_pagelist(msg)) {
7fe1e5e5
AE		 1506 use_cursor = true;
1507 page = ceph_msg_data_next(&msg->l, &page_offset,
1508 &length, &last_piece);
68b4476b 1509#ifdef CONFIG_BLOCK
97fb1c7f 1510 } else if (ceph_msg_has_bio(msg)) {
6aaa4511
AE		 1511 use_cursor = true;
1512 page = ceph_msg_data_next(&msg->b, &page_offset,
1513 &length, &last_piece);
68b4476b 1514#endif
31b8006e 1515 } else {
57666519 1516 page = zero_page;
31b8006e 1517 }
6aaa4511
AE		 1518 if (!use_cursor) {
1519 length = min_t(int, PAGE_SIZE - msg_pos->page_pos,
fe38a2b6 1520 total_max_write);
68b4476b 1521
6aaa4511
AE		 1522 page_offset = msg_pos->page_pos;
1523 }
bae6acd9 1524 if (do_datacrc && !msg_pos->did_page_crc) {
5821bd8c 1525 u32 crc = le32_to_cpu(msg->footer.data_crc);
31b8006e 1526
35b62808 1527 crc = ceph_crc32c_page(crc, page, page_offset, length);
84ca8fc8 1528 msg->footer.data_crc = cpu_to_le32(crc);
bae6acd9 1529 msg_pos->did_page_crc = true;
31b8006e 1530 }
e387d525 1531 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1532 length, last_piece);
31b8006e
SW		 1533 if (ret <= 0)
1534 goto out;
1535
9d2a06c2 1536 out_msg_pos_next(con, page, length, (size_t) ret);
31b8006e
SW		 1537 }
1538
34d2d200 1539 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1540
1541 /* prepare and queue up footer, too */
37675b0f 1542 if (!do_datacrc)
84ca8fc8 1543 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1544 con_out_kvec_reset(con);
859eb799 1545 prepare_write_message_footer(con);
31b8006e
SW		 1546 ret = 1;
1547out:
1548 return ret;
1549}
1550
1551/*
1552 * write some zeros
1553 */
1554static int write_partial_skip(struct ceph_connection *con)
1555{
1556 int ret;
1557
1558 while (con->out_skip > 0) {
31739139 1559 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1560
e1dcb128 1561 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1562 if (ret <= 0)
1563 goto out;
1564 con->out_skip -= ret;
1565 }
1566 ret = 1;
1567out:
1568 return ret;
1569}
1570
1571/*
1572 * Prepare to read connection handshake, or an ack.
1573 */
eed0ef2c
SW		 1574static void prepare_read_banner(struct ceph_connection *con)
1575{
1576 dout("prepare_read_banner %p\n", con);
1577 con->in_base_pos = 0;
1578}
1579
31b8006e
SW		 1580static void prepare_read_connect(struct ceph_connection *con)
1581{
1582 dout("prepare_read_connect %p\n", con);
1583 con->in_base_pos = 0;
1584}
1585
1586static void prepare_read_ack(struct ceph_connection *con)
1587{
1588 dout("prepare_read_ack %p\n", con);
1589 con->in_base_pos = 0;
1590}
1591
1592static void prepare_read_tag(struct ceph_connection *con)
1593{
1594 dout("prepare_read_tag %p\n", con);
1595 con->in_base_pos = 0;
1596 con->in_tag = CEPH_MSGR_TAG_READY;
1597}
1598
1599/*
1600 * Prepare to read a message.
1601 */
1602static int prepare_read_message(struct ceph_connection *con)
1603{
1604 dout("prepare_read_message %p\n", con);
1605 BUG_ON(con->in_msg != NULL);
1606 con->in_base_pos = 0;
1607 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1608 return 0;
1609}
1610
1611
1612static int read_partial(struct ceph_connection *con,
fd51653f 1613 int end, int size, void *object)
31b8006e 1614{
e6cee71f
AE		 1615 while (con->in_base_pos < end) {
1616 int left = end - con->in_base_pos;
31b8006e
SW		 1617 int have = size - left;
1618 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1619 if (ret <= 0)
1620 return ret;
1621 con->in_base_pos += ret;
1622 }
1623 return 1;
1624}
1625
1626
1627/*
1628 * Read all or part of the connect-side handshake on a new connection
1629 */
eed0ef2c 1630static int read_partial_banner(struct ceph_connection *con)
31b8006e 1631{
fd51653f
AE		 1632 int size;
1633 int end;
1634 int ret;
31b8006e 1635
eed0ef2c 1636 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1637
1638 /* peer's banner */
fd51653f
AE		 1639 size = strlen(CEPH_BANNER);
1640 end = size;
1641 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1642 if (ret <= 0)
1643 goto out;
fd51653f
AE		 1644
1645 size = sizeof (con->actual_peer_addr);
1646 end += size;
1647 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1648 if (ret <= 0)
1649 goto out;
fd51653f
AE		 1650
1651 size = sizeof (con->peer_addr_for_me);
1652 end += size;
1653 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1654 if (ret <= 0)
1655 goto out;
fd51653f 1656
eed0ef2c
SW		 1657out:
1658 return ret;
1659}
1660
1661static int read_partial_connect(struct ceph_connection *con)
1662{
fd51653f
AE		 1663 int size;
1664 int end;
1665 int ret;
eed0ef2c
SW		 1666
1667 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1668
fd51653f
AE		 1669 size = sizeof (con->in_reply);
1670 end = size;
1671 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1672 if (ret <= 0)
1673 goto out;
fd51653f
AE		 1674
1675 size = le32_to_cpu(con->in_reply.authorizer_len);
1676 end += size;
1677 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1678 if (ret <= 0)
1679 goto out;
31b8006e 1680
4e7a5dcd
SW		 1681 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1682 con, (int)con->in_reply.tag,
1683 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1684 le32_to_cpu(con->in_reply.global_seq));
1685out:
1686 return ret;
eed0ef2c 1687
31b8006e
SW		 1688}
1689
1690/*
1691 * Verify the hello banner looks okay.
1692 */
1693static int verify_hello(struct ceph_connection *con)
1694{
1695 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1696 pr_err("connect to %s got bad banner\n",
3d14c5d2 1697 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1698 con->error_msg = "protocol error, bad banner";
1699 return -1;
1700 }
1701 return 0;
1702}
1703
1704static bool addr_is_blank(struct sockaddr_storage *ss)
1705{
1706 switch (ss->ss_family) {
1707 case AF_INET:
1708 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1709 case AF_INET6:
1710 return
1711 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1712 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1713 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1714 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1715 }
1716 return false;
1717}
1718
1719static int addr_port(struct sockaddr_storage *ss)
1720{
1721 switch (ss->ss_family) {
1722 case AF_INET:
f28bcfbe 1723 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1724 case AF_INET6:
f28bcfbe 1725 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1726 }
1727 return 0;
1728}
1729
1730static void addr_set_port(struct sockaddr_storage *ss, int p)
1731{
1732 switch (ss->ss_family) {
1733 case AF_INET:
1734 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1735 break;
31b8006e
SW		 1736 case AF_INET6:
1737 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1738 break;
31b8006e
SW		 1739 }
1740}
1741
ee3b56f2
NW		 1742/*
1743 * Unlike other *_pton function semantics, zero indicates success.
1744 */
1745static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1746 char delim, const char **ipend)
1747{
99f0f3b2
AE		 1748 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1749 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1750
1751 memset(ss, 0, sizeof(*ss));
1752
1753 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1754 ss->ss_family = AF_INET;
1755 return 0;
1756 }
1757
1758 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1759 ss->ss_family = AF_INET6;
1760 return 0;
1761 }
1762
1763 return -EINVAL;
1764}
1765
1766/*
1767 * Extract hostname string and resolve using kernel DNS facility.
1768 */
1769#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1770static int ceph_dns_resolve_name(const char *name, size_t namelen,
1771 struct sockaddr_storage *ss, char delim, const char **ipend)
1772{
1773 const char *end, *delim_p;
1774 char *colon_p, *ip_addr = NULL;
1775 int ip_len, ret;
1776
1777 /*
1778 * The end of the hostname occurs immediately preceding the delimiter or
1779 * the port marker (':') where the delimiter takes precedence.
1780 */
1781 delim_p = memchr(name, delim, namelen);
1782 colon_p = memchr(name, ':', namelen);
1783
1784 if (delim_p && colon_p)
1785 end = delim_p < colon_p ? delim_p : colon_p;
1786 else if (!delim_p && colon_p)
1787 end = colon_p;
1788 else {
1789 end = delim_p;
1790 if (!end) /* case: hostname:/ */
1791 end = name + namelen;
1792 }
1793
1794 if (end <= name)
1795 return -EINVAL;
1796
1797 /* do dns_resolve upcall */
1798 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1799 if (ip_len > 0)
1800 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1801 else
1802 ret = -ESRCH;
1803
1804 kfree(ip_addr);
1805
1806 *ipend = end;
1807
1808 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1809 ret, ret ? "failed" : ceph_pr_addr(ss));
1810
1811 return ret;
1812}
1813#else
1814static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1815 struct sockaddr_storage *ss, char delim, const char **ipend)
1816{
1817 return -EINVAL;
1818}
1819#endif
1820
1821/*
1822 * Parse a server name (IP or hostname). If a valid IP address is not found
1823 * then try to extract a hostname to resolve using userspace DNS upcall.
1824 */
1825static int ceph_parse_server_name(const char *name, size_t namelen,
1826 struct sockaddr_storage *ss, char delim, const char **ipend)
1827{
1828 int ret;
1829
1830 ret = ceph_pton(name, namelen, ss, delim, ipend);
1831 if (ret)
1832 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1833
1834 return ret;
1835}
1836
31b8006e
SW		 1837/*
1838 * Parse an ip[:port] list into an addr array. Use the default
1839 * monitor port if a port isn't specified.
1840 */
1841int ceph_parse_ips(const char *c, const char *end,
1842 struct ceph_entity_addr *addr,
1843 int max_count, int *count)
1844{
ee3b56f2 1845 int i, ret = -EINVAL;
31b8006e
SW		 1846 const char *p = c;
1847
1848 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1849 for (i = 0; i < max_count; i++) {
1850 const char *ipend;
1851 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1852 int port;
39139f64
SW		 1853 char delim = ',';
1854
1855 if (*p == '[') {
1856 delim = ']';
1857 p++;
1858 }
31b8006e 1859
ee3b56f2
NW		 1860 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1861 if (ret)
31b8006e 1862 goto bad;
ee3b56f2
NW		 1863 ret = -EINVAL;
1864
31b8006e
SW		 1865 p = ipend;
1866
39139f64
SW		 1867 if (delim == ']') {
1868 if (*p != ']') {
1869 dout("missing matching ']'\n");
1870 goto bad;
1871 }
1872 p++;
1873 }
1874
31b8006e
SW		 1875 /* port? */
1876 if (p < end && *p == ':') {
1877 port = 0;
1878 p++;
1879 while (p < end && *p >= '0' && *p <= '9') {
1880 port = (port * 10) + (*p - '0');
1881 p++;
1882 }
1883 if (port > 65535 || port == 0)
1884 goto bad;
1885 } else {
1886 port = CEPH_MON_PORT;
1887 }
1888
1889 addr_set_port(ss, port);
1890
3d14c5d2 1891 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1892
1893 if (p == end)
1894 break;
1895 if (*p != ',')
1896 goto bad;
1897 p++;
1898 }
1899
1900 if (p != end)
1901 goto bad;
1902
1903 if (count)
1904 *count = i + 1;
1905 return 0;
1906
1907bad:
39139f64 1908 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1909 return ret;
31b8006e 1910}
3d14c5d2 1911EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1912
eed0ef2c 1913static int process_banner(struct ceph_connection *con)
31b8006e 1914{
eed0ef2c 1915 dout("process_banner on %p\n", con);
31b8006e
SW		 1916
1917 if (verify_hello(con) < 0)
1918 return -1;
1919
63f2d211
SW		 1920 ceph_decode_addr(&con->actual_peer_addr);
1921 ceph_decode_addr(&con->peer_addr_for_me);
1922
31b8006e
SW		 1923 /*
1924 * Make sure the other end is who we wanted. note that the other
1925 * end may not yet know their ip address, so if it's 0.0.0.0, give
1926 * them the benefit of the doubt.
1927 */
103e2d3a
SW		 1928 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1929 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1930 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1931 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
cd84db6e 1932 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
3d14c5d2 1933 ceph_pr_addr(&con->peer_addr.in_addr),
cd84db6e 1934 (int)le32_to_cpu(con->peer_addr.nonce),
3d14c5d2 1935 ceph_pr_addr(&con->actual_peer_addr.in_addr),
cd84db6e 1936 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1937 con->error_msg = "wrong peer at address";
31b8006e
SW		 1938 return -1;
1939 }
1940
1941 /*
1942 * did we learn our address?
1943 */
1944 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1945 int port = addr_port(&con->msgr->inst.addr.in_addr);
1946
1947 memcpy(&con->msgr->inst.addr.in_addr,
1948 &con->peer_addr_for_me.in_addr,
1949 sizeof(con->peer_addr_for_me.in_addr));
1950 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1951 encode_my_addr(con->msgr);
eed0ef2c 1952 dout("process_banner learned my addr is %s\n",
3d14c5d2 1953 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1954 }
1955
eed0ef2c
SW		 1956 return 0;
1957}
1958
1959static int process_connect(struct ceph_connection *con)
1960{
3d14c5d2
YS		 1961 u64 sup_feat = con->msgr->supported_features;
1962 u64 req_feat = con->msgr->required_features;
04a419f9 1963 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 1964 int ret;
04a419f9 1965
eed0ef2c
SW		 1966 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1967
31b8006e 1968 switch (con->in_reply.tag) {
04a419f9
SW		 1969 case CEPH_MSGR_TAG_FEATURES:
1970 pr_err("%s%lld %s feature set mismatch,"
1971 " my %llx < server's %llx, missing %llx\n",
1972 ENTITY_NAME(con->peer_name),
3d14c5d2 1973 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1974 sup_feat, server_feat, server_feat & ~sup_feat);
1975 con->error_msg = "missing required protocol features";
0fa6ebc6 1976 reset_connection(con);
04a419f9
SW		 1977 return -1;
1978
31b8006e 1979 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 1980 pr_err("%s%lld %s protocol version mismatch,"
1981 " my %d != server's %d\n",
1982 ENTITY_NAME(con->peer_name),
3d14c5d2 1983 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 1984 le32_to_cpu(con->out_connect.protocol_version),
1985 le32_to_cpu(con->in_reply.protocol_version));
1986 con->error_msg = "protocol version mismatch";
0fa6ebc6 1987 reset_connection(con);
31b8006e
SW		 1988 return -1;
1989
4e7a5dcd
SW		 1990 case CEPH_MSGR_TAG_BADAUTHORIZER:
1991 con->auth_retry++;
1992 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1993 con->auth_retry);
1994 if (con->auth_retry == 2) {
1995 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 1996 return -1;
1997 }
1998 con->auth_retry = 1;
6d4221b5 1999 con_out_kvec_reset(con);
e825a66d 2000 ret = prepare_write_connect(con);
0da5d703
SW		 2001 if (ret < 0)
2002 return ret;
63733a0f 2003 prepare_read_connect(con);
4e7a5dcd 2004 break;
31b8006e
SW		 2005
2006 case CEPH_MSGR_TAG_RESETSESSION:
2007 /*
2008 * If we connected with a large connect_seq but the peer
2009 * has no record of a session with us (no connection, or
2010 * connect_seq == 0), they will send RESETSESION to indicate
2011 * that they must have reset their session, and may have
2012 * dropped messages.
2013 */
2014 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2015 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2016 pr_err("%s%lld %s connection reset\n",
2017 ENTITY_NAME(con->peer_name),
3d14c5d2 2018 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2019 reset_connection(con);
6d4221b5 2020 con_out_kvec_reset(con);
5a0f8fdd
AE		 2021 ret = prepare_write_connect(con);
2022 if (ret < 0)
2023 return ret;
31b8006e
SW		 2024 prepare_read_connect(con);
2025
2026 /* Tell ceph about it. */
ec302645 2027 mutex_unlock(&con->mutex);
31b8006e
SW		 2028 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2029 if (con->ops->peer_reset)
2030 con->ops->peer_reset(con);
ec302645 2031 mutex_lock(&con->mutex);
8dacc7da 2032 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2033 return -EAGAIN;
31b8006e
SW		 2034 break;
2035
2036 case CEPH_MSGR_TAG_RETRY_SESSION:
2037 /*
2038 * If we sent a smaller connect_seq than the peer has, try
2039 * again with a larger value.
2040 */
5bdca4e0 2041 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2042 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2043 le32_to_cpu(con->in_reply.connect_seq));
2044 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2045 con_out_kvec_reset(con);
5a0f8fdd
AE		 2046 ret = prepare_write_connect(con);
2047 if (ret < 0)
2048 return ret;
31b8006e
SW		 2049 prepare_read_connect(con);
2050 break;
2051
2052 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2053 /*
2054 * If we sent a smaller global_seq than the peer has, try
2055 * again with a larger value.
2056 */
eed0ef2c 2057 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2058 con->peer_global_seq,
5bdca4e0 2059 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2060 get_global_seq(con->msgr,
5bdca4e0 2061 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2062 con_out_kvec_reset(con);
5a0f8fdd
AE		 2063 ret = prepare_write_connect(con);
2064 if (ret < 0)
2065 return ret;
31b8006e
SW		 2066 prepare_read_connect(con);
2067 break;
2068
2069 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2070 if (req_feat & ~server_feat) {
2071 pr_err("%s%lld %s protocol feature mismatch,"
2072 " my required %llx > server's %llx, need %llx\n",
2073 ENTITY_NAME(con->peer_name),
3d14c5d2 2074 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2075 req_feat, server_feat, req_feat & ~server_feat);
2076 con->error_msg = "missing required protocol features";
0fa6ebc6 2077 reset_connection(con);
04a419f9
SW		 2078 return -1;
2079 }
8dacc7da 2080
122070a2 2081 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da
SW		 2082 con->state = CON_STATE_OPEN;
2083
31b8006e
SW		 2084 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2085 con->connect_seq++;
aba558e2 2086 con->peer_features = server_feat;
31b8006e
SW		 2087 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2088 con->peer_global_seq,
2089 le32_to_cpu(con->in_reply.connect_seq),
2090 con->connect_seq);
2091 WARN_ON(con->connect_seq !=
2092 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2093
2094 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2095 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2096
85effe18 2097 con->delay = 0; /* reset backoff memory */
92ac41d0 2098
31b8006e
SW		 2099 prepare_read_tag(con);
2100 break;
2101
2102 case CEPH_MSGR_TAG_WAIT:
2103 /*
2104 * If there is a connection race (we are opening
2105 * connections to each other), one of us may just have
2106 * to WAIT. This shouldn't happen if we are the
2107 * client.
2108 */
04177882
SW		 2109 pr_err("process_connect got WAIT as client\n");
2110 con->error_msg = "protocol error, got WAIT as client";
2111 return -1;
31b8006e
SW		 2112
2113 default:
2114 pr_err("connect protocol error, will retry\n");
2115 con->error_msg = "protocol error, garbage tag during connect";
2116 return -1;
2117 }
2118 return 0;
2119}
2120
2121
2122/*
2123 * read (part of) an ack
2124 */
2125static int read_partial_ack(struct ceph_connection *con)
2126{
fd51653f
AE		 2127 int size = sizeof (con->in_temp_ack);
2128 int end = size;
31b8006e 2129
fd51653f 2130 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2131}
2132
2133
2134/*
2135 * We can finally discard anything that's been acked.
2136 */
2137static void process_ack(struct ceph_connection *con)
2138{
2139 struct ceph_msg *m;
2140 u64 ack = le64_to_cpu(con->in_temp_ack);
2141 u64 seq;
2142
31b8006e
SW		 2143 while (!list_empty(&con->out_sent)) {
2144 m = list_first_entry(&con->out_sent, struct ceph_msg,
2145 list_head);
2146 seq = le64_to_cpu(m->hdr.seq);
2147 if (seq > ack)
2148 break;
2149 dout("got ack for seq %llu type %d at %p\n", seq,
2150 le16_to_cpu(m->hdr.type), m);
4cf9d544 2151 m->ack_stamp = jiffies;
31b8006e
SW		 2152 ceph_msg_remove(m);
2153 }
31b8006e
SW		 2154 prepare_read_tag(con);
2155}
2156
2157
2158
2159
2450418c 2160static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2161 struct kvec *section,
2162 unsigned int sec_len, u32 *crc)
2450418c 2163{
68b4476b 2164 int ret, left;
2450418c
YS		 2165
2166 BUG_ON(!section);
2167
2168 while (section->iov_len < sec_len) {
2169 BUG_ON(section->iov_base == NULL);
2170 left = sec_len - section->iov_len;
2171 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2172 section->iov_len, left);
2173 if (ret <= 0)
2174 return ret;
2175 section->iov_len += ret;
2450418c 2176 }
fe3ad593
AE		 2177 if (section->iov_len == sec_len)
2178 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2179
2450418c
YS		 2180 return 1;
2181}
31b8006e 2182
4740a623 2183static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
68b4476b
YS		 2184
2185static int read_partial_message_pages(struct ceph_connection *con,
2186 struct page **pages,
95c96174 2187 unsigned int data_len, bool do_datacrc)
68b4476b 2188{
bae6acd9 2189 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
e788182f 2190 struct page *page;
afb3d90e
AE		 2191 size_t page_offset;
2192 size_t length;
2193 unsigned int left;
68b4476b 2194 int ret;
68b4476b 2195
68b4476b
YS		 2196 /* (page) data */
2197 BUG_ON(pages == NULL);
bae6acd9 2198 page = pages[msg_pos->page];
afb3d90e
AE		 2199 page_offset = msg_pos->page_pos;
2200 BUG_ON(msg_pos->data_pos >= data_len);
2201 left = data_len - msg_pos->data_pos;
2202 BUG_ON(page_offset >= PAGE_SIZE);
2203 length = min_t(unsigned int, PAGE_SIZE - page_offset, left);
2204
2205 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
68b4476b
YS		 2206 if (ret <= 0)
2207 return ret;
e788182f 2208
35b62808
AE		 2209 if (do_datacrc)
2210 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2211 page_offset, ret);
afb3d90e
AE		 2212
2213 in_msg_pos_next(con, length, ret);
68b4476b
YS		 2214
2215 return ret;
2216}
2217
2218#ifdef CONFIG_BLOCK
2219static int read_partial_message_bio(struct ceph_connection *con,
95c96174 2220 unsigned int data_len, bool do_datacrc)
68b4476b 2221{
b3d56fab 2222 struct ceph_msg *msg = con->in_msg;
bae6acd9 2223 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
b3d56fab 2224 struct bio_vec *bv;
e788182f 2225 struct page *page;
afb3d90e
AE		 2226 size_t page_offset;
2227 size_t length;
2228 unsigned int left;
2229 int ret;
68b4476b 2230
b3d56fab 2231 BUG_ON(!msg);
f9e15777
AE		 2232 BUG_ON(!msg->b.bio_iter);
2233 bv = bio_iovec_idx(msg->b.bio_iter, msg->b.bio_seg);
e788182f 2234 page = bv->bv_page;
afb3d90e
AE		 2235 page_offset = bv->bv_offset + msg_pos->page_pos;
2236 BUG_ON(msg_pos->data_pos >= data_len);
2237 left = data_len - msg_pos->data_pos;
2238 BUG_ON(msg_pos->page_pos >= bv->bv_len);
2239 length = min_t(unsigned int, bv->bv_len - msg_pos->page_pos, left);
68b4476b 2240
afb3d90e 2241 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
68b4476b
YS		 2242 if (ret <= 0)
2243 return ret;
e788182f 2244
35b62808
AE		 2245 if (do_datacrc)
2246 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2247 page_offset, ret);
afb3d90e
AE		 2248
2249 in_msg_pos_next(con, length, ret);
68b4476b
YS		 2250
2251 return ret;
2252}
2253#endif
2254
34d2d200
AE		 2255static int read_partial_msg_data(struct ceph_connection *con)
2256{
2257 struct ceph_msg *msg = con->in_msg;
2258 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2259 const bool do_datacrc = !con->msgr->nocrc;
2260 unsigned int data_len;
2261 int ret;
2262
2263 BUG_ON(!msg);
2264
2265 data_len = le32_to_cpu(con->in_hdr.data_len);
2266 while (msg_pos->data_pos < data_len) {
97fb1c7f 2267 if (ceph_msg_has_pages(msg)) {
f9e15777 2268 ret = read_partial_message_pages(con, msg->p.pages,
34d2d200
AE		 2269 data_len, do_datacrc);
2270 if (ret <= 0)
2271 return ret;
2272#ifdef CONFIG_BLOCK
97fb1c7f 2273 } else if (ceph_msg_has_bio(msg)) {
34d2d200
AE		 2274 ret = read_partial_message_bio(con,
2275 data_len, do_datacrc);
2276 if (ret <= 0)
2277 return ret;
2278#endif
2279 } else {
2280 BUG_ON(1);
2281 }
2282 }
2283
2284 return 1; /* must return > 0 to indicate success */
2285}
2286
31b8006e
SW		 2287/*
2288 * read (part of) a message.
2289 */
2290static int read_partial_message(struct ceph_connection *con)
2291{
2292 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2293 int size;
2294 int end;
31b8006e 2295 int ret;
95c96174 2296 unsigned int front_len, middle_len, data_len;
37675b0f 2297 bool do_datacrc = !con->msgr->nocrc;
ae18756b 2298 u64 seq;
fe3ad593 2299 u32 crc;
31b8006e
SW		 2300
2301 dout("read_partial_message con %p msg %p\n", con, m);
2302
2303 /* header */
fd51653f
AE		 2304 size = sizeof (con->in_hdr);
2305 end = size;
2306 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2307 if (ret <= 0)
2308 return ret;
fe3ad593
AE		 2309
2310 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2311 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2312 pr_err("read_partial_message bad hdr "
2313 " crc %u != expected %u\n",
2314 crc, con->in_hdr.crc);
2315 return -EBADMSG;
2316 }
2317
31b8006e
SW		 2318 front_len = le32_to_cpu(con->in_hdr.front_len);
2319 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2320 return -EIO;
2321 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2322 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2323 return -EIO;
2324 data_len = le32_to_cpu(con->in_hdr.data_len);
2325 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2326 return -EIO;
2327
ae18756b
SW		 2328 /* verify seq# */
2329 seq = le64_to_cpu(con->in_hdr.seq);
2330 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2331 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2332 ENTITY_NAME(con->peer_name),
3d14c5d2 2333 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2334 seq, con->in_seq + 1);
2335 con->in_base_pos = -front_len - middle_len - data_len -
2336 sizeof(m->footer);
2337 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2338 return 0;
2339 } else if ((s64)seq - (s64)con->in_seq > 1) {
2340 pr_err("read_partial_message bad seq %lld expected %lld\n",
2341 seq, con->in_seq + 1);
2342 con->error_msg = "bad message sequence # for incoming message";
2343 return -EBADMSG;
2344 }
2345
31b8006e
SW		 2346 /* allocate message? */
2347 if (!con->in_msg) {
4740a623
SW		 2348 int skip = 0;
2349
31b8006e 2350 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2351 front_len, data_len);
4740a623
SW		 2352 ret = ceph_con_in_msg_alloc(con, &skip);
2353 if (ret < 0)
2354 return ret;
2450418c 2355 if (skip) {
31b8006e 2356 /* skip this message */
a79832f2 2357 dout("alloc_msg said skip message\n");
ae32be31 2358 BUG_ON(con->in_msg);
31b8006e
SW		 2359 con->in_base_pos = -front_len - middle_len - data_len -
2360 sizeof(m->footer);
2361 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2362 con->in_seq++;
31b8006e
SW		 2363 return 0;
2364 }
38941f80 2365
4740a623 2366 BUG_ON(!con->in_msg);
38941f80 2367 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2368 m = con->in_msg;
2369 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2370 if (m->middle)
2371 m->middle->vec.iov_len = 0;
9d7f0f13 2372
78625051 2373 /* prepare for data payload, if any */
a4107026 2374
78625051
AE		 2375 if (data_len)
2376 prepare_message_data(con->in_msg, &con->in_msg_pos);
31b8006e
SW		 2377 }
2378
2379 /* front */
2450418c
YS		 2380 ret = read_partial_message_section(con, &m->front, front_len,
2381 &con->in_front_crc);
2382 if (ret <= 0)
2383 return ret;
31b8006e
SW		 2384
2385 /* middle */
2450418c 2386 if (m->middle) {
213c99ee
SW		 2387 ret = read_partial_message_section(con, &m->middle->vec,
2388 middle_len,
2450418c 2389 &con->in_middle_crc);
31b8006e
SW		 2390 if (ret <= 0)
2391 return ret;
31b8006e
SW		 2392 }
2393
2394 /* (page) data */
34d2d200
AE		 2395 if (data_len) {
2396 ret = read_partial_msg_data(con);
2397 if (ret <= 0)
2398 return ret;
31b8006e
SW		 2399 }
2400
31b8006e 2401 /* footer */
fd51653f
AE		 2402 size = sizeof (m->footer);
2403 end += size;
2404 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2405 if (ret <= 0)
2406 return ret;
2407
31b8006e
SW		 2408 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2409 m, front_len, m->footer.front_crc, middle_len,
2410 m->footer.middle_crc, data_len, m->footer.data_crc);
2411
2412 /* crc ok? */
2413 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2414 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2415 m, con->in_front_crc, m->footer.front_crc);
2416 return -EBADMSG;
2417 }
2418 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2419 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2420 m, con->in_middle_crc, m->footer.middle_crc);
2421 return -EBADMSG;
2422 }
bca064d2 2423 if (do_datacrc &&
31b8006e
SW		 2424 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2425 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2426 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2427 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2428 return -EBADMSG;
2429 }
2430
2431 return 1; /* done! */
2432}
2433
2434/*
2435 * Process message. This happens in the worker thread. The callback should
2436 * be careful not to do anything that waits on other incoming messages or it
2437 * may deadlock.
2438 */
2439static void process_message(struct ceph_connection *con)
2440{
5e095e8b 2441 struct ceph_msg *msg;
31b8006e 2442
38941f80
AE		 2443 BUG_ON(con->in_msg->con != con);
2444 con->in_msg->con = NULL;
5e095e8b 2445 msg = con->in_msg;
31b8006e 2446 con->in_msg = NULL;
36eb71aa 2447 con->ops->put(con);
31b8006e
SW		 2448
2449 /* if first message, set peer_name */
2450 if (con->peer_name.type == 0)
dbad185d 2451 con->peer_name = msg->hdr.src;
31b8006e 2452
31b8006e 2453 con->in_seq++;
ec302645 2454 mutex_unlock(&con->mutex);
31b8006e
SW		 2455
2456 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2457 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2458 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2459 le16_to_cpu(msg->hdr.type),
2460 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2461 le32_to_cpu(msg->hdr.front_len),
2462 le32_to_cpu(msg->hdr.data_len),
2463 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2464 con->ops->dispatch(con, msg);
ec302645
SW		 2465
2466 mutex_lock(&con->mutex);
31b8006e
SW		 2467}
2468
2469
2470/*
2471 * Write something to the socket. Called in a worker thread when the
2472 * socket appears to be writeable and we have something ready to send.
2473 */
2474static int try_write(struct ceph_connection *con)
2475{
31b8006e
SW		 2476 int ret = 1;
2477
d59315ca 2478 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2479
31b8006e
SW		 2480more:
2481 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2482
2483 /* open the socket first? */
8dacc7da
SW		 2484 if (con->state == CON_STATE_PREOPEN) {
2485 BUG_ON(con->sock);
2486 con->state = CON_STATE_CONNECTING;
a5988c49 2487
e2200423 2488 con_out_kvec_reset(con);
e825a66d 2489 prepare_write_banner(con);
eed0ef2c 2490 prepare_read_banner(con);
31b8006e 2491
cf3e5c40 2492 BUG_ON(con->in_msg);
31b8006e
SW		 2493 con->in_tag = CEPH_MSGR_TAG_READY;
2494 dout("try_write initiating connect on %p new state %lu\n",
2495 con, con->state);
41617d0c
AE		 2496 ret = ceph_tcp_connect(con);
2497 if (ret < 0) {
31b8006e 2498 con->error_msg = "connect error";
31b8006e
SW		 2499 goto out;
2500 }
2501 }
2502
2503more_kvec:
2504 /* kvec data queued? */
2505 if (con->out_skip) {
2506 ret = write_partial_skip(con);
2507 if (ret <= 0)
42961d23 2508 goto out;
31b8006e
SW		 2509 }
2510 if (con->out_kvec_left) {
2511 ret = write_partial_kvec(con);
2512 if (ret <= 0)
42961d23 2513 goto out;
31b8006e
SW		 2514 }
2515
2516 /* msg pages? */
2517 if (con->out_msg) {
c86a2930
SW		 2518 if (con->out_msg_done) {
2519 ceph_msg_put(con->out_msg);
2520 con->out_msg = NULL; /* we're done with this one */
2521 goto do_next;
2522 }
2523
34d2d200 2524 ret = write_partial_message_data(con);
31b8006e
SW		 2525 if (ret == 1)
2526 goto more_kvec; /* we need to send the footer, too! */
2527 if (ret == 0)
42961d23 2528 goto out;
31b8006e 2529 if (ret < 0) {
34d2d200 2530 dout("try_write write_partial_message_data err %d\n",
31b8006e 2531 ret);
42961d23 2532 goto out;
31b8006e
SW		 2533 }
2534 }
2535
c86a2930 2536do_next:
8dacc7da 2537 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2538 /* is anything else pending? */
2539 if (!list_empty(&con->out_queue)) {
2540 prepare_write_message(con);
2541 goto more;
2542 }
2543 if (con->in_seq > con->in_seq_acked) {
2544 prepare_write_ack(con);
2545 goto more;
2546 }
c9ffc77a 2547 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2548 prepare_write_keepalive(con);
2549 goto more;
2550 }
2551 }
2552
2553 /* Nothing to do! */
c9ffc77a 2554 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2555 dout("try_write nothing else to write.\n");
31b8006e
SW		 2556 ret = 0;
2557out:
42961d23 2558 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2559 return ret;
2560}
2561
2562
2563
2564/*
2565 * Read what we can from the socket.
2566 */
2567static int try_read(struct ceph_connection *con)
2568{
31b8006e
SW		 2569 int ret = -1;
2570
8dacc7da
SW		 2571more:
2572 dout("try_read start on %p state %lu\n", con, con->state);
2573 if (con->state != CON_STATE_CONNECTING &&
2574 con->state != CON_STATE_NEGOTIATING &&
2575 con->state != CON_STATE_OPEN)
31b8006e
SW		 2576 return 0;
2577
8dacc7da 2578 BUG_ON(!con->sock);
ec302645 2579
31b8006e
SW		 2580 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2581 con->in_base_pos);
0da5d703 2582
8dacc7da 2583 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2584 dout("try_read connecting\n");
2585 ret = read_partial_banner(con);
2586 if (ret <= 0)
ab166d5a 2587 goto out;
7593af92
AE		 2588 ret = process_banner(con);
2589 if (ret < 0)
2590 goto out;
2591
8dacc7da 2592 con->state = CON_STATE_NEGOTIATING;
7593af92 2593
6d4221b5
JS		 2594 /*
2595 * Received banner is good, exchange connection info.
2596 * Do not reset out_kvec, as sending our banner raced
2597 * with receiving peer banner after connect completed.
2598 */
7593af92
AE		 2599 ret = prepare_write_connect(con);
2600 if (ret < 0)
2601 goto out;
2602 prepare_read_connect(con);
2603
2604 /* Send connection info before awaiting response */
0da5d703
SW		 2605 goto out;
2606 }
2607
8dacc7da 2608 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2609 dout("try_read negotiating\n");
31b8006e
SW		 2610 ret = read_partial_connect(con);
2611 if (ret <= 0)
31b8006e 2612 goto out;
98bdb0aa
SW		 2613 ret = process_connect(con);
2614 if (ret < 0)
2615 goto out;
31b8006e
SW		 2616 goto more;
2617 }
2618
122070a2 2619 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2620
31b8006e
SW		 2621 if (con->in_base_pos < 0) {
2622 /*
2623 * skipping + discarding content.
2624 *
2625 * FIXME: there must be a better way to do this!
2626 */
84495f49
AE		 2627 static char buf[SKIP_BUF_SIZE];
2628 int skip = min((int) sizeof (buf), -con->in_base_pos);
2629
31b8006e
SW		 2630 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2631 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2632 if (ret <= 0)
98bdb0aa 2633 goto out;
31b8006e
SW		 2634 con->in_base_pos += ret;
2635 if (con->in_base_pos)
2636 goto more;
2637 }
2638 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2639 /*
2640 * what's next?
2641 */
2642 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2643 if (ret <= 0)
98bdb0aa 2644 goto out;
31b8006e
SW		 2645 dout("try_read got tag %d\n", (int)con->in_tag);
2646 switch (con->in_tag) {
2647 case CEPH_MSGR_TAG_MSG:
2648 prepare_read_message(con);
2649 break;
2650 case CEPH_MSGR_TAG_ACK:
2651 prepare_read_ack(con);
2652 break;
2653 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2654 con_close_socket(con);
2655 con->state = CON_STATE_CLOSED;
98bdb0aa 2656 goto out;
31b8006e
SW		 2657 default:
2658 goto bad_tag;
2659 }
2660 }
2661 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2662 ret = read_partial_message(con);
2663 if (ret <= 0) {
2664 switch (ret) {
2665 case -EBADMSG:
2666 con->error_msg = "bad crc";
2667 ret = -EIO;
98bdb0aa 2668 break;
31b8006e
SW		 2669 case -EIO:
2670 con->error_msg = "io error";
98bdb0aa 2671 break;
31b8006e 2672 }
98bdb0aa 2673 goto out;
31b8006e
SW		 2674 }
2675 if (con->in_tag == CEPH_MSGR_TAG_READY)
2676 goto more;
2677 process_message(con);
7b862e07
SW		 2678 if (con->state == CON_STATE_OPEN)
2679 prepare_read_tag(con);
31b8006e
SW		 2680 goto more;
2681 }
2682 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2683 ret = read_partial_ack(con);
2684 if (ret <= 0)
98bdb0aa 2685 goto out;
31b8006e
SW		 2686 process_ack(con);
2687 goto more;
2688 }
2689
31b8006e 2690out:
98bdb0aa 2691 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2692 return ret;
2693
2694bad_tag:
2695 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2696 con->error_msg = "protocol error, garbage tag";
2697 ret = -1;
2698 goto out;
2699}
2700
2701
2702/*
802c6d96
AE		 2703 * Atomically queue work on a connection after the specified delay.
2704 * Bump @con reference to avoid races with connection teardown.
2705 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2706 */
802c6d96 2707static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2708{
31b8006e 2709 if (!con->ops->get(con)) {
802c6d96
AE		 2710 dout("%s %p ref count 0\n", __func__, con);
2711
2712 return -ENOENT;
31b8006e
SW		 2713 }
2714
802c6d96
AE		 2715 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2716 dout("%s %p - already queued\n", __func__, con);
31b8006e 2717 con->ops->put(con);
802c6d96
AE		 2718
2719 return -EBUSY;
31b8006e 2720 }
802c6d96
AE		 2721
2722 dout("%s %p %lu\n", __func__, con, delay);
2723
2724 return 0;
2725}
2726
2727static void queue_con(struct ceph_connection *con)
2728{
2729 (void) queue_con_delay(con, 0);
31b8006e
SW		 2730}
2731
7bb21d68
AE		 2732static bool con_sock_closed(struct ceph_connection *con)
2733{
c9ffc77a 2734 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2735 return false;
2736
2737#define CASE(x) \
2738 case CON_STATE_ ## x: \
2739 con->error_msg = "socket closed (con state " #x ")"; \
2740 break;
2741
2742 switch (con->state) {
2743 CASE(CLOSED);
2744 CASE(PREOPEN);
2745 CASE(CONNECTING);
2746 CASE(NEGOTIATING);
2747 CASE(OPEN);
2748 CASE(STANDBY);
2749 default:
2750 pr_warning("%s con %p unrecognized state %lu\n",
2751 __func__, con, con->state);
2752 con->error_msg = "unrecognized con state";
2753 BUG();
2754 break;
2755 }
2756#undef CASE
2757
2758 return true;
2759}
2760
f20a39fd
AE		 2761static bool con_backoff(struct ceph_connection *con)
2762{
2763 int ret;
2764
2765 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2766 return false;
2767
2768 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2769 if (ret) {
2770 dout("%s: con %p FAILED to back off %lu\n", __func__,
2771 con, con->delay);
2772 BUG_ON(ret == -ENOENT);
2773 con_flag_set(con, CON_FLAG_BACKOFF);
2774 }
2775
2776 return true;
2777}
2778
93209264
AE		 2779/* Finish fault handling; con->mutex must *not* be held here */
2780
2781static void con_fault_finish(struct ceph_connection *con)
2782{
2783 /*
2784 * in case we faulted due to authentication, invalidate our
2785 * current tickets so that we can get new ones.
2786 */
2787 if (con->auth_retry && con->ops->invalidate_authorizer) {
2788 dout("calling invalidate_authorizer()\n");
2789 con->ops->invalidate_authorizer(con);
2790 }
2791
2792 if (con->ops->fault)
2793 con->ops->fault(con);
2794}
2795
31b8006e
SW		 2796/*
2797 * Do some work on a connection. Drop a connection ref when we're done.
2798 */
2799static void con_work(struct work_struct *work)
2800{
2801 struct ceph_connection *con = container_of(work, struct ceph_connection,
2802 work.work);
49659416 2803 bool fault;
31b8006e 2804
9dd4658d 2805 mutex_lock(&con->mutex);
49659416
AE		 2806 while (true) {
2807 int ret;
31b8006e 2808
49659416
AE		 2809 if ((fault = con_sock_closed(con))) {
2810 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2811 break;
2812 }
2813 if (con_backoff(con)) {
2814 dout("%s: con %p BACKOFF\n", __func__, con);
2815 break;
2816 }
2817 if (con->state == CON_STATE_STANDBY) {
2818 dout("%s: con %p STANDBY\n", __func__, con);
2819 break;
2820 }
2821 if (con->state == CON_STATE_CLOSED) {
2822 dout("%s: con %p CLOSED\n", __func__, con);
2823 BUG_ON(con->sock);
2824 break;
2825 }
2826 if (con->state == CON_STATE_PREOPEN) {
2827 dout("%s: con %p PREOPEN\n", __func__, con);
2828 BUG_ON(con->sock);
2829 }
0da5d703 2830
49659416
AE		 2831 ret = try_read(con);
2832 if (ret < 0) {
2833 if (ret == -EAGAIN)
2834 continue;
2835 con->error_msg = "socket error on read";
2836 fault = true;
2837 break;
2838 }
2839
2840 ret = try_write(con);
2841 if (ret < 0) {
2842 if (ret == -EAGAIN)
2843 continue;
2844 con->error_msg = "socket error on write";
2845 fault = true;
2846 }
2847
2848 break; /* If we make it to here, we're done */
3a140a0d 2849 }
b6e7b6a1
AE		 2850 if (fault)
2851 con_fault(con);
9dd4658d 2852 mutex_unlock(&con->mutex);
0da5d703 2853
b6e7b6a1
AE		 2854 if (fault)
2855 con_fault_finish(con);
2856
2857 con->ops->put(con);
31b8006e
SW		 2858}
2859
31b8006e
SW		 2860/*
2861 * Generic error/fault handler. A retry mechanism is used with
2862 * exponential backoff
2863 */
93209264 2864static void con_fault(struct ceph_connection *con)
31b8006e 2865{
28362986 2866 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
3d14c5d2 2867 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2868 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2869 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2870
122070a2 2871 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2872 con->state != CON_STATE_NEGOTIATING &&
2873 con->state != CON_STATE_OPEN);
ec302645 2874
31b8006e 2875 con_close_socket(con);
5e095e8b 2876
c9ffc77a 2877 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2878 dout("fault on LOSSYTX channel, marking CLOSED\n");
2879 con->state = CON_STATE_CLOSED;
93209264 2880 return;
3b5ede07
SW		 2881 }
2882
5e095e8b 2883 if (con->in_msg) {
38941f80
AE		 2884 BUG_ON(con->in_msg->con != con);
2885 con->in_msg->con = NULL;
5e095e8b
SW		 2886 ceph_msg_put(con->in_msg);
2887 con->in_msg = NULL;
36eb71aa 2888 con->ops->put(con);
5e095e8b 2889 }
31b8006e 2890
e80a52d1
SW		 2891 /* Requeue anything that hasn't been acked */
2892 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2893
e76661d0
SW		 2894 /* If there are no messages queued or keepalive pending, place
2895 * the connection in a STANDBY state */
2896 if (list_empty(&con->out_queue) &&
c9ffc77a 2897 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2898 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2899 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2900 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2901 } else {
2902 /* retry after a delay. */
8dacc7da 2903 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2904 if (con->delay == 0)
2905 con->delay = BASE_DELAY_INTERVAL;
2906 else if (con->delay < MAX_DELAY_INTERVAL)
2907 con->delay *= 2;
c9ffc77a 2908 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2909 queue_con(con);
31b8006e 2910 }
31b8006e
SW		 2911}
2912
2913
2914
2915/*
15d9882c 2916 * initialize a new messenger instance
31b8006e 2917 */
15d9882c
AE		 2918void ceph_messenger_init(struct ceph_messenger *msgr,
2919 struct ceph_entity_addr *myaddr,
2920 u32 supported_features,
2921 u32 required_features,
2922 bool nocrc)
31b8006e 2923{
3d14c5d2
YS		 2924 msgr->supported_features = supported_features;
2925 msgr->required_features = required_features;
2926
31b8006e
SW		 2927 spin_lock_init(&msgr->global_seq_lock);
2928
31b8006e
SW		 2929 if (myaddr)
2930 msgr->inst.addr = *myaddr;
2931
2932 /* select a random nonce */
ac8839d7 2933 msgr->inst.addr.type = 0;
103e2d3a 2934 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2935 encode_my_addr(msgr);
15d9882c 2936 msgr->nocrc = nocrc;
31b8006e 2937
a2a32584 2938 atomic_set(&msgr->stopping, 0);
31b8006e 2939
15d9882c 2940 dout("%s %p\n", __func__, msgr);
31b8006e 2941}
15d9882c 2942EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2943
e00de341
SW		 2944static void clear_standby(struct ceph_connection *con)
2945{
2946 /* come back from STANDBY? */
8dacc7da 2947 if (con->state == CON_STATE_STANDBY) {
e00de341 2948 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2949 con->state = CON_STATE_PREOPEN;
e00de341 2950 con->connect_seq++;
c9ffc77a
AE		 2951 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2952 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2953 }
2954}
2955
31b8006e
SW		 2956/*
2957 * Queue up an outgoing message on the given connection.
2958 */
2959void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2960{
31b8006e 2961 /* set src+dst */
dbad185d 2962 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2963 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2964 msg->needs_out_seq = true;
2965
ec302645 2966 mutex_lock(&con->mutex);
92ce034b 2967
8dacc7da 2968 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2969 dout("con_send %p closed, dropping %p\n", con, msg);
2970 ceph_msg_put(msg);
2971 mutex_unlock(&con->mutex);
2972 return;
2973 }
2974
38941f80 2975 BUG_ON(msg->con != NULL);
36eb71aa 2976 msg->con = con->ops->get(con);
92ce034b
AE		 2977 BUG_ON(msg->con == NULL);
2978
31b8006e
SW		 2979 BUG_ON(!list_empty(&msg->list_head));
2980 list_add_tail(&msg->list_head, &con->out_queue);
2981 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2982 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2983 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2984 le32_to_cpu(msg->hdr.front_len),
2985 le32_to_cpu(msg->hdr.middle_len),
2986 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 2987
2988 clear_standby(con);
ec302645 2989 mutex_unlock(&con->mutex);
31b8006e
SW		 2990
2991 /* if there wasn't anything waiting to send before, queue
2992 * new work */
c9ffc77a 2993 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2994 queue_con(con);
2995}
3d14c5d2 2996EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 2997
2998/*
2999 * Revoke a message that was previously queued for send
3000 */
6740a845 3001void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3002{
6740a845
AE		 3003 struct ceph_connection *con = msg->con;
3004
3005 if (!con)
3006 return; /* Message not in our possession */
3007
ec302645 3008 mutex_lock(&con->mutex);
31b8006e 3009 if (!list_empty(&msg->list_head)) {
38941f80 3010 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3011 list_del_init(&msg->list_head);
38941f80 3012 BUG_ON(msg->con == NULL);
36eb71aa 3013 msg->con->ops->put(msg->con);
38941f80 3014 msg->con = NULL;
31b8006e 3015 msg->hdr.seq = 0;
38941f80 3016
31b8006e 3017 ceph_msg_put(msg);
ed98adad
SW		 3018 }
3019 if (con->out_msg == msg) {
38941f80 3020 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 3021 con->out_msg = NULL;
31b8006e
SW		 3022 if (con->out_kvec_is_msg) {
3023 con->out_skip = con->out_kvec_bytes;
3024 con->out_kvec_is_msg = false;
3025 }
ed98adad 3026 msg->hdr.seq = 0;
92ce034b
AE		 3027
3028 ceph_msg_put(msg);
31b8006e 3029 }
ec302645 3030 mutex_unlock(&con->mutex);
31b8006e
SW		 3031}
3032
350b1c32 3033/*
0d59ab81 3034 * Revoke a message that we may be reading data into
350b1c32 3035 */
8921d114 3036void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3037{
8921d114
AE		 3038 struct ceph_connection *con;
3039
3040 BUG_ON(msg == NULL);
3041 if (!msg->con) {
3042 dout("%s msg %p null con\n", __func__, msg);
3043
3044 return; /* Message not in our possession */
3045 }
3046
3047 con = msg->con;
350b1c32 3048 mutex_lock(&con->mutex);
8921d114 3049 if (con->in_msg == msg) {
95c96174
ED		 3050 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3051 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3052 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3053
3054 /* skip rest of message */
8921d114
AE		 3055 dout("%s %p msg %p revoked\n", __func__, con, msg);
3056 con->in_base_pos = con->in_base_pos -
350b1c32 3057 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3058 front_len -
3059 middle_len -
3060 data_len -
350b1c32 3061 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3062 ceph_msg_put(con->in_msg);
3063 con->in_msg = NULL;
3064 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3065 con->in_seq++;
350b1c32 3066 } else {
8921d114
AE		 3067 dout("%s %p in_msg %p msg %p no-op\n",
3068 __func__, con, con->in_msg, msg);
350b1c32
SW		 3069 }
3070 mutex_unlock(&con->mutex);
3071}
3072
31b8006e
SW		 3073/*
3074 * Queue a keepalive byte to ensure the tcp connection is alive.
3075 */
3076void ceph_con_keepalive(struct ceph_connection *con)
3077{
e00de341 3078 dout("con_keepalive %p\n", con);
00650931 3079 mutex_lock(&con->mutex);
e00de341 3080 clear_standby(con);
00650931 3081 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3082 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3083 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3084 queue_con(con);
3085}
3d14c5d2 3086EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3087
43794509
AE		 3088static void ceph_msg_data_init(struct ceph_msg_data *data)
3089{
3090 data->type = CEPH_MSG_DATA_NONE;
3091}
3092
02afca6c 3093void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3094 size_t length, size_t alignment)
02afca6c 3095{
07aa1558
AE		 3096 BUG_ON(!pages);
3097 BUG_ON(!length);
43794509 3098 BUG_ON(msg->p.type != CEPH_MSG_DATA_NONE);
02afca6c 3099
43794509 3100 msg->p.type = CEPH_MSG_DATA_PAGES;
f9e15777
AE		 3101 msg->p.pages = pages;
3102 msg->p.length = length;
3103 msg->p.alignment = alignment & ~PAGE_MASK;
02afca6c
AE		 3104}
3105EXPORT_SYMBOL(ceph_msg_data_set_pages);
31b8006e 3106
27fa8385
AE		 3107void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3108 struct ceph_pagelist *pagelist)
3109{
07aa1558
AE		 3110 BUG_ON(!pagelist);
3111 BUG_ON(!pagelist->length);
43794509 3112 BUG_ON(msg->l.type != CEPH_MSG_DATA_NONE);
27fa8385 3113
43794509 3114 msg->l.type = CEPH_MSG_DATA_PAGELIST;
f9e15777 3115 msg->l.pagelist = pagelist;
27fa8385
AE		 3116}
3117EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3118
3119void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio)
3120{
07aa1558 3121 BUG_ON(!bio);
43794509 3122 BUG_ON(msg->b.type != CEPH_MSG_DATA_NONE);
27fa8385 3123
43794509 3124 msg->b.type = CEPH_MSG_DATA_BIO;
f9e15777 3125 msg->b.bio = bio;
27fa8385
AE		 3126}
3127EXPORT_SYMBOL(ceph_msg_data_set_bio);
3128
31b8006e
SW		 3129/*
3130 * construct a new message with given type, size
3131 * the new msg has a ref count of 1.
3132 */
b61c2763
SW		 3133struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3134 bool can_fail)
31b8006e
SW		 3135{
3136 struct ceph_msg *m;
3137
9516e45b 3138 m = kzalloc(sizeof(*m), flags);
31b8006e
SW		 3139 if (m == NULL)
3140 goto out;
31b8006e
SW		 3141
3142 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3143 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3144 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3145
9516e45b
AE		 3146 INIT_LIST_HEAD(&m->list_head);
3147 kref_init(&m->kref);
ca20892d 3148
43794509
AE		 3149 ceph_msg_data_init(&m->p);
3150 ceph_msg_data_init(&m->l);
3151 ceph_msg_data_init(&m->b);
43794509 3152
31b8006e 3153 /* front */
9516e45b 3154 m->front_max = front_len;
31b8006e
SW		 3155 if (front_len) {
3156 if (front_len > PAGE_CACHE_SIZE) {
34d23762 3157 m->front.iov_base = __vmalloc(front_len, flags,
31b8006e
SW		 3158 PAGE_KERNEL);
3159 m->front_is_vmalloc = true;
3160 } else {
34d23762 3161 m->front.iov_base = kmalloc(front_len, flags);
31b8006e
SW		 3162 }
3163 if (m->front.iov_base == NULL) {
b61c2763 3164 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3165 front_len);
3166 goto out2;
3167 }
3168 } else {
3169 m->front.iov_base = NULL;
3170 }
3171 m->front.iov_len = front_len;
3172
bb257664 3173 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3174 return m;
3175
3176out2:
3177 ceph_msg_put(m);
3178out:
b61c2763
SW		 3179 if (!can_fail) {
3180 pr_err("msg_new can't create type %d front %d\n", type,
3181 front_len);
f0ed1b7c 3182 WARN_ON(1);
b61c2763
SW		 3183 } else {
3184 dout("msg_new can't create type %d front %d\n", type,
3185 front_len);
3186 }
a79832f2 3187 return NULL;
31b8006e 3188}
3d14c5d2 3189EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3190
31b8006e
SW		 3191/*
3192 * Allocate "middle" portion of a message, if it is needed and wasn't
3193 * allocated by alloc_msg. This allows us to read a small fixed-size
3194 * per-type header in the front and then gracefully fail (i.e.,
3195 * propagate the error to the caller based on info in the front) when
3196 * the middle is too large.
3197 */
2450418c 3198static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3199{
3200 int type = le16_to_cpu(msg->hdr.type);
3201 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3202
3203 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3204 ceph_msg_type_name(type), middle_len);
3205 BUG_ON(!middle_len);
3206 BUG_ON(msg->middle);
3207
b6c1d5b8 3208 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3209 if (!msg->middle)
3210 return -ENOMEM;
3211 return 0;
3212}
3213
2450418c 3214/*
1c20f2d2
AE		 3215 * Allocate a message for receiving an incoming message on a
3216 * connection, and save the result in con->in_msg. Uses the
3217 * connection's private alloc_msg op if available.
3218 *
4740a623
SW		 3219 * Returns 0 on success, or a negative error code.
3220 *
3221 * On success, if we set *skip = 1:
3222 * - the next message should be skipped and ignored.
3223 * - con->in_msg == NULL
3224 * or if we set *skip = 0:
3225 * - con->in_msg is non-null.
3226 * On error (ENOMEM, EAGAIN, ...),
3227 * - con->in_msg == NULL
2450418c 3228 */
4740a623 3229static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3230{
4740a623 3231 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3232 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3233 struct ceph_msg *msg;
4740a623 3234 int ret = 0;
2450418c 3235
1c20f2d2 3236 BUG_ON(con->in_msg != NULL);
53ded495 3237 BUG_ON(!con->ops->alloc_msg);
2450418c 3238
53ded495
AE		 3239 mutex_unlock(&con->mutex);
3240 msg = con->ops->alloc_msg(con, hdr, skip);
3241 mutex_lock(&con->mutex);
3242 if (con->state != CON_STATE_OPEN) {
3243 if (msg)
1d866d1c 3244 ceph_msg_put(msg);
53ded495
AE		 3245 return -EAGAIN;
3246 }
4137577a
AE		 3247 if (msg) {
3248 BUG_ON(*skip);
3249 con->in_msg = msg;
36eb71aa 3250 con->in_msg->con = con->ops->get(con);
92ce034b 3251 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3252 } else {
3253 /*
3254 * Null message pointer means either we should skip
3255 * this message or we couldn't allocate memory. The
3256 * former is not an error.
3257 */
3258 if (*skip)
3259 return 0;
3260 con->error_msg = "error allocating memory for incoming message";
3261
53ded495 3262 return -ENOMEM;
2450418c 3263 }
1c20f2d2 3264 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3265
1c20f2d2
AE		 3266 if (middle_len && !con->in_msg->middle) {
3267 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3268 if (ret < 0) {
1c20f2d2
AE		 3269 ceph_msg_put(con->in_msg);
3270 con->in_msg = NULL;
2450418c
YS		 3271 }
3272 }
9d7f0f13 3273
4740a623 3274 return ret;
2450418c
YS		 3275}
3276
31b8006e
SW		 3277
3278/*
3279 * Free a generically kmalloc'd message.
3280 */
3281void ceph_msg_kfree(struct ceph_msg *m)
3282{
3283 dout("msg_kfree %p\n", m);
3284 if (m->front_is_vmalloc)
3285 vfree(m->front.iov_base);
3286 else
3287 kfree(m->front.iov_base);
3288 kfree(m);
3289}
3290
3291/*
3292 * Drop a msg ref. Destroy as needed.
3293 */
c2e552e7
SW		 3294void ceph_msg_last_put(struct kref *kref)
3295{
3296 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
31b8006e 3297
c2e552e7
SW		 3298 dout("ceph_msg_put last one on %p\n", m);
3299 WARN_ON(!list_empty(&m->list_head));
3300
3301 /* drop middle, data, if any */
3302 if (m->middle) {
3303 ceph_buffer_put(m->middle);
3304 m->middle = NULL;
31b8006e 3305 }
97fb1c7f 3306 if (ceph_msg_has_pages(m)) {
f9e15777
AE		 3307 m->p.length = 0;
3308 m->p.pages = NULL;
97fb1c7f 3309 }
c2e552e7 3310
97fb1c7f 3311 if (ceph_msg_has_pagelist(m)) {
f9e15777
AE		 3312 ceph_pagelist_release(m->l.pagelist);
3313 kfree(m->l.pagelist);
3314 m->l.pagelist = NULL;
58bb3b37
SW		 3315 }
3316
c2e552e7
SW		 3317 if (m->pool)
3318 ceph_msgpool_put(m->pool, m);
3319 else
3320 ceph_msg_kfree(m);
31b8006e 3321}
3d14c5d2 3322EXPORT_SYMBOL(ceph_msg_last_put);
9ec7cab1
SW		 3323
3324void ceph_msg_dump(struct ceph_msg *msg)
3325{
4a73ef27 3326 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
f9e15777 3327 msg->front_max, msg->p.length);
9ec7cab1
SW		 3328 print_hex_dump(KERN_DEBUG, "header: ",
3329 DUMP_PREFIX_OFFSET, 16, 1,
3330 &msg->hdr, sizeof(msg->hdr), true);
3331 print_hex_dump(KERN_DEBUG, " front: ",
3332 DUMP_PREFIX_OFFSET, 16, 1,
3333 msg->front.iov_base, msg->front.iov_len, true);
3334 if (msg->middle)
3335 print_hex_dump(KERN_DEBUG, "middle: ",
3336 DUMP_PREFIX_OFFSET, 16, 1,
3337 msg->middle->vec.iov_base,
3338 msg->middle->vec.iov_len, true);
3339 print_hex_dump(KERN_DEBUG, "footer: ",
3340 DUMP_PREFIX_OFFSET, 16, 1,
3341 &msg->footer, sizeof(msg->footer), true);
3342}
3d14c5d2 3343EXPORT_SYMBOL(ceph_msg_dump);
Linux kernel - Deliverables
This page took 0.348401 seconds and 5 git commands to generate.