projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: get rid of read helpers
[GitHub/mt8127/android_kernel_alcatel_ttab.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
6aaa4511
AE		 719
720/*
721 * For a bio data item, a piece is whatever remains of the next
722 * entry in the current bio iovec, or the first entry in the next
723 * bio in the list.
724 */
25aff7c5
AE		 725static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data *data,
726 size_t length)
6aaa4511
AE		 727{
728 struct ceph_msg_data_cursor *cursor = &data->cursor;
729 struct bio *bio;
730
731 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
732
733 bio = data->bio;
734 BUG_ON(!bio);
735 BUG_ON(!bio->bi_vcnt);
6aaa4511 736
25aff7c5 737 cursor->resid = length;
6aaa4511
AE		 738 cursor->bio = bio;
739 cursor->vector_index = 0;
740 cursor->vector_offset = 0;
25aff7c5 741 cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
6aaa4511
AE		 742}
743
744static struct page *ceph_msg_data_bio_next(struct ceph_msg_data *data,
745 size_t *page_offset,
746 size_t *length)
747{
748 struct ceph_msg_data_cursor *cursor = &data->cursor;
749 struct bio *bio;
750 struct bio_vec *bio_vec;
751 unsigned int index;
752
753 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
754
755 bio = cursor->bio;
756 BUG_ON(!bio);
757
758 index = cursor->vector_index;
759 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
760
761 bio_vec = &bio->bi_io_vec[index];
762 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
763 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
764 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 765 if (cursor->last_piece) /* pagelist offset is always 0 */
766 *length = cursor->resid;
767 else
768 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
6aaa4511 769 BUG_ON(*length > PAGE_SIZE);
25aff7c5 770 BUG_ON(*length > cursor->resid);
6aaa4511
AE		 771
772 return bio_vec->bv_page;
773}
774
775static bool ceph_msg_data_bio_advance(struct ceph_msg_data *data, size_t bytes)
776{
777 struct ceph_msg_data_cursor *cursor = &data->cursor;
778 struct bio *bio;
779 struct bio_vec *bio_vec;
780 unsigned int index;
781
782 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
783
784 bio = cursor->bio;
785 BUG_ON(!bio);
786
787 index = cursor->vector_index;
788 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
789 bio_vec = &bio->bi_io_vec[index];
6aaa4511
AE		 790
791 /* Advance the cursor offset */
792
25aff7c5
AE		 793 BUG_ON(cursor->resid < bytes);
794 cursor->resid -= bytes;
6aaa4511
AE		 795 cursor->vector_offset += bytes;
796 if (cursor->vector_offset < bio_vec->bv_len)
797 return false; /* more bytes to process in this segment */
25aff7c5 798 BUG_ON(cursor->vector_offset != bio_vec->bv_len);
6aaa4511
AE		 799
800 /* Move on to the next segment, and possibly the next bio */
801
25aff7c5 802 if (++index == (unsigned int) bio->bi_vcnt) {
6aaa4511 803 bio = bio->bi_next;
25aff7c5 804 index = 0;
6aaa4511 805 }
25aff7c5
AE		 806 cursor->bio = bio;
807 cursor->vector_index = index;
6aaa4511
AE		 808 cursor->vector_offset = 0;
809
25aff7c5
AE		 810 if (!cursor->last_piece) {
811 BUG_ON(!cursor->resid);
812 BUG_ON(!bio);
813 /* A short read is OK, so use <= rather than == */
814 if (cursor->resid <= bio->bi_io_vec[index].bv_len)
6aaa4511 815 cursor->last_piece = true;
25aff7c5 816 }
6aaa4511
AE		 817
818 return true;
819}
df6ad1f9
AE		 820#endif
821
e766d7b5
AE		 822/*
823 * For a page array, a piece comes from the first page in the array
824 * that has not already been fully consumed.
825 */
25aff7c5
AE		 826static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data *data,
827 size_t length)
e766d7b5
AE		 828{
829 struct ceph_msg_data_cursor *cursor = &data->cursor;
830 int page_count;
831
832 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
833
834 BUG_ON(!data->pages);
835 BUG_ON(!data->length);
25aff7c5 836 BUG_ON(length != data->length);
e766d7b5 837
25aff7c5 838 cursor->resid = length;
e766d7b5 839 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 840 cursor->page_offset = data->alignment & ~PAGE_MASK;
841 cursor->page_index = 0;
25aff7c5 842 BUG_ON(page_count > (int) USHRT_MAX);
e766d7b5 843 cursor->page_count = (unsigned short) page_count;
25aff7c5 844 cursor->last_piece = length <= PAGE_SIZE;
e766d7b5
AE		 845}
846
847static struct page *ceph_msg_data_pages_next(struct ceph_msg_data *data,
848 size_t *page_offset,
849 size_t *length)
850{
851 struct ceph_msg_data_cursor *cursor = &data->cursor;
852
853 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
854
855 BUG_ON(cursor->page_index >= cursor->page_count);
856 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 857
858 *page_offset = cursor->page_offset;
25aff7c5 859 if (cursor->last_piece)
e766d7b5 860 *length = cursor->resid;
25aff7c5 861 else
e766d7b5 862 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 863
864 return data->pages[cursor->page_index];
865}
866
867static bool ceph_msg_data_pages_advance(struct ceph_msg_data *data,
868 size_t bytes)
869{
870 struct ceph_msg_data_cursor *cursor = &data->cursor;
871
872 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
873
874 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 875
876 /* Advance the cursor page offset */
877
878 cursor->resid -= bytes;
879 cursor->page_offset += bytes;
880 if (!bytes || cursor->page_offset & ~PAGE_MASK)
881 return false; /* more bytes to process in the current page */
882
883 /* Move on to the next page */
884
885 BUG_ON(cursor->page_index >= cursor->page_count);
886 cursor->page_offset = 0;
887 cursor->page_index++;
25aff7c5 888 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 889
890 return true;
891}
892
fe38a2b6 893/*
dd236fcb
AE		 894 * For a pagelist, a piece is whatever remains to be consumed in the
895 * first page in the list, or the front of the next page.
fe38a2b6 896 */
25aff7c5
AE		 897static void ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data *data,
898 size_t length)
fe38a2b6
AE		 899{
900 struct ceph_msg_data_cursor *cursor = &data->cursor;
901 struct ceph_pagelist *pagelist;
902 struct page *page;
903
dd236fcb 904 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 905
906 pagelist = data->pagelist;
907 BUG_ON(!pagelist);
25aff7c5
AE		 908 BUG_ON(length != pagelist->length);
909
910 if (!length)
fe38a2b6
AE		 911 return; /* pagelist can be assigned but empty */
912
913 BUG_ON(list_empty(&pagelist->head));
914 page = list_first_entry(&pagelist->head, struct page, lru);
915
25aff7c5 916 cursor->resid = length;
fe38a2b6
AE		 917 cursor->page = page;
918 cursor->offset = 0;
25aff7c5 919 cursor->last_piece = length <= PAGE_SIZE;
fe38a2b6
AE		 920}
921
dd236fcb 922static struct page *ceph_msg_data_pagelist_next(struct ceph_msg_data *data,
fe38a2b6 923 size_t *page_offset,
dd236fcb 924 size_t *length)
fe38a2b6
AE		 925{
926 struct ceph_msg_data_cursor *cursor = &data->cursor;
927 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 928
929 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
930
931 pagelist = data->pagelist;
932 BUG_ON(!pagelist);
933
934 BUG_ON(!cursor->page);
25aff7c5 935 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 936
fe38a2b6 937 *page_offset = cursor->offset & ~PAGE_MASK;
25aff7c5
AE		 938 if (cursor->last_piece) /* pagelist offset is always 0 */
939 *length = cursor->resid;
940 else
941 *length = PAGE_SIZE - *page_offset;
fe38a2b6
AE		 942
943 return data->cursor.page;
944}
945
dd236fcb
AE		 946static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data *data,
947 size_t bytes)
fe38a2b6
AE		 948{
949 struct ceph_msg_data_cursor *cursor = &data->cursor;
950 struct ceph_pagelist *pagelist;
951
952 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
953
954 pagelist = data->pagelist;
955 BUG_ON(!pagelist);
25aff7c5
AE		 956
957 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 958 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
959
960 /* Advance the cursor offset */
961
25aff7c5 962 cursor->resid -= bytes;
fe38a2b6
AE		 963 cursor->offset += bytes;
964 /* pagelist offset is always 0 */
965 if (!bytes || cursor->offset & ~PAGE_MASK)
966 return false; /* more bytes to process in the current page */
967
968 /* Move on to the next page */
969
970 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
971 cursor->page = list_entry_next(cursor->page, lru);
25aff7c5 972 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 973
974 return true;
975}
976
dd236fcb
AE		 977/*
978 * Message data is handled (sent or received) in pieces, where each
979 * piece resides on a single page. The network layer might not
980 * consume an entire piece at once. A data item's cursor keeps
981 * track of which piece is next to process and how much remains to
982 * be processed in that piece. It also tracks whether the current
983 * piece is the last one in the data item.
984 */
25aff7c5
AE		 985static void ceph_msg_data_cursor_init(struct ceph_msg_data *data,
986 size_t length)
dd236fcb
AE		 987{
988 switch (data->type) {
989 case CEPH_MSG_DATA_PAGELIST:
25aff7c5 990 ceph_msg_data_pagelist_cursor_init(data, length);
dd236fcb 991 break;
e766d7b5 992 case CEPH_MSG_DATA_PAGES:
25aff7c5 993 ceph_msg_data_pages_cursor_init(data, length);
e766d7b5 994 break;
dd236fcb
AE		 995#ifdef CONFIG_BLOCK
996 case CEPH_MSG_DATA_BIO:
25aff7c5 997 ceph_msg_data_bio_cursor_init(data, length);
6aaa4511 998 break;
dd236fcb 999#endif /* CONFIG_BLOCK */
6aaa4511 1000 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1001 default:
1002 /* BUG(); */
1003 break;
1004 }
1005}
1006
1007/*
1008 * Return the page containing the next piece to process for a given
1009 * data item, and supply the page offset and length of that piece.
1010 * Indicate whether this is the last piece in this data item.
1011 */
1012static struct page *ceph_msg_data_next(struct ceph_msg_data *data,
1013 size_t *page_offset,
1014 size_t *length,
1015 bool *last_piece)
1016{
1017 struct page *page;
1018
1019 switch (data->type) {
1020 case CEPH_MSG_DATA_PAGELIST:
1021 page = ceph_msg_data_pagelist_next(data, page_offset, length);
1022 break;
e766d7b5
AE		 1023 case CEPH_MSG_DATA_PAGES:
1024 page = ceph_msg_data_pages_next(data, page_offset, length);
1025 break;
dd236fcb
AE		 1026#ifdef CONFIG_BLOCK
1027 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1028 page = ceph_msg_data_bio_next(data, page_offset, length);
1029 break;
dd236fcb 1030#endif /* CONFIG_BLOCK */
6aaa4511 1031 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1032 default:
1033 page = NULL;
1034 break;
1035 }
1036 BUG_ON(!page);
1037 BUG_ON(*page_offset + *length > PAGE_SIZE);
1038 BUG_ON(!*length);
1039 if (last_piece)
1040 *last_piece = data->cursor.last_piece;
1041
1042 return page;
1043}
1044
1045/*
1046 * Returns true if the result moves the cursor on to the next piece
1047 * of the data item.
1048 */
1049static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes)
1050{
25aff7c5 1051 struct ceph_msg_data_cursor *cursor = &data->cursor;
dd236fcb
AE		 1052 bool new_piece;
1053
25aff7c5 1054 BUG_ON(bytes > cursor->resid);
dd236fcb
AE		 1055 switch (data->type) {
1056 case CEPH_MSG_DATA_PAGELIST:
1057 new_piece = ceph_msg_data_pagelist_advance(data, bytes);
1058 break;
e766d7b5
AE		 1059 case CEPH_MSG_DATA_PAGES:
1060 new_piece = ceph_msg_data_pages_advance(data, bytes);
1061 break;
dd236fcb
AE		 1062#ifdef CONFIG_BLOCK
1063 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1064 new_piece = ceph_msg_data_bio_advance(data, bytes);
1065 break;
dd236fcb 1066#endif /* CONFIG_BLOCK */
6aaa4511 1067 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1068 default:
1069 BUG();
1070 break;
1071 }
1072
1073 return new_piece;
1074}
1075
78625051
AE		 1076static void prepare_message_data(struct ceph_msg *msg,
1077 struct ceph_msg_pos *msg_pos)
739c905b 1078{
25aff7c5
AE		 1079 size_t data_len;
1080
739c905b 1081 BUG_ON(!msg);
25aff7c5
AE		 1082
1083 data_len = le32_to_cpu(msg->hdr.data_len);
1084 BUG_ON(!data_len);
739c905b
AE		 1085
1086 /* initialize page iterator */
bae6acd9 1087 msg_pos->page = 0;
97fb1c7f 1088 if (ceph_msg_has_pages(msg))
f9e15777 1089 msg_pos->page_pos = msg->p.alignment;
739c905b 1090 else
bae6acd9 1091 msg_pos->page_pos = 0;
bae6acd9 1092 msg_pos->data_pos = 0;
fe38a2b6 1093
7fe1e5e5 1094 /* Initialize data cursors */
fe38a2b6 1095
6aaa4511
AE		 1096#ifdef CONFIG_BLOCK
1097 if (ceph_msg_has_bio(msg))
25aff7c5 1098 ceph_msg_data_cursor_init(&msg->b, data_len);
6aaa4511 1099#endif /* CONFIG_BLOCK */
e766d7b5 1100 if (ceph_msg_has_pages(msg))
25aff7c5 1101 ceph_msg_data_cursor_init(&msg->p, data_len);
7fe1e5e5 1102 if (ceph_msg_has_pagelist(msg))
25aff7c5 1103 ceph_msg_data_cursor_init(&msg->l, data_len);
fe38a2b6 1104
bae6acd9 1105 msg_pos->did_page_crc = false;
739c905b
AE		 1106}
1107
31b8006e
SW		 1108/*
1109 * Prepare footer for currently outgoing message, and finish things
1110 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1111 */
859eb799 1112static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1113{
1114 struct ceph_msg *m = con->out_msg;
859eb799 1115 int v = con->out_kvec_left;
31b8006e 1116
fd154f3c
AE		 1117 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1118
31b8006e
SW		 1119 dout("prepare_write_message_footer %p\n", con);
1120 con->out_kvec_is_msg = true;
1121 con->out_kvec[v].iov_base = &m->footer;
1122 con->out_kvec[v].iov_len = sizeof(m->footer);
1123 con->out_kvec_bytes += sizeof(m->footer);
1124 con->out_kvec_left++;
1125 con->out_more = m->more_to_follow;
c86a2930 1126 con->out_msg_done = true;
31b8006e
SW		 1127}
1128
1129/*
1130 * Prepare headers for the next outgoing message.
1131 */
1132static void prepare_write_message(struct ceph_connection *con)
1133{
1134 struct ceph_msg *m;
a9a0c51a 1135 u32 crc;
31b8006e 1136
e2200423 1137 con_out_kvec_reset(con);
31b8006e 1138 con->out_kvec_is_msg = true;
c86a2930 1139 con->out_msg_done = false;
31b8006e
SW		 1140
1141 /* Sneak an ack in there first? If we can get it into the same
1142 * TCP packet that's a good thing. */
1143 if (con->in_seq > con->in_seq_acked) {
1144 con->in_seq_acked = con->in_seq;
e2200423 1145 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1146 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1147 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1148 &con->out_temp_ack);
31b8006e
SW		 1149 }
1150
38941f80 1151 BUG_ON(list_empty(&con->out_queue));
859eb799 1152 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1153 con->out_msg = m;
38941f80 1154 BUG_ON(m->con != con);
4cf9d544
SW		 1155
1156 /* put message on sent list */
1157 ceph_msg_get(m);
1158 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1159
e84346b7
SW		 1160 /*
1161 * only assign outgoing seq # if we haven't sent this message
1162 * yet. if it is requeued, resend with it's original seq.
1163 */
1164 if (m->needs_out_seq) {
1165 m->hdr.seq = cpu_to_le64(++con->out_seq);
1166 m->needs_out_seq = false;
1167 }
31b8006e 1168
4a73ef27 1169 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n",
31b8006e
SW		 1170 m, con->out_seq, le16_to_cpu(m->hdr.type),
1171 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
f9e15777 1172 le32_to_cpu(m->hdr.data_len), m->p.length);
31b8006e
SW		 1173 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1174
1175 /* tag + hdr + front + middle */
e2200423
AE		 1176 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1177 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1178 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1179
31b8006e 1180 if (m->middle)
e2200423 1181 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1182 m->middle->vec.iov_base);
31b8006e
SW		 1183
1184 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1185 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1186 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1187 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1188
1189 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1190 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1191 if (m->middle) {
1192 crc = crc32c(0, m->middle->vec.iov_base,
1193 m->middle->vec.iov_len);
1194 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1195 } else
31b8006e 1196 con->out_msg->footer.middle_crc = 0;
739c905b 1197 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1198 le32_to_cpu(con->out_msg->footer.front_crc),
1199 le32_to_cpu(con->out_msg->footer.middle_crc));
1200
1201 /* is there a data payload? */
739c905b 1202 con->out_msg->footer.data_crc = 0;
78625051
AE		 1203 if (m->hdr.data_len) {
1204 prepare_message_data(con->out_msg, &con->out_msg_pos);
1205 con->out_more = 1; /* data + footer will follow */
1206 } else {
31b8006e 1207 /* no, queue up footer too and be done */
859eb799 1208 prepare_write_message_footer(con);
78625051 1209 }
31b8006e 1210
c9ffc77a 1211 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1212}
1213
1214/*
1215 * Prepare an ack.
1216 */
1217static void prepare_write_ack(struct ceph_connection *con)
1218{
1219 dout("prepare_write_ack %p %llu -> %llu\n", con,
1220 con->in_seq_acked, con->in_seq);
1221 con->in_seq_acked = con->in_seq;
1222
e2200423 1223 con_out_kvec_reset(con);
859eb799 1224
e2200423 1225 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1226
31b8006e 1227 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1228 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1229 &con->out_temp_ack);
1230
31b8006e 1231 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1232 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1233}
1234
3a23083b
SW		 1235/*
1236 * Prepare to share the seq during handshake
1237 */
1238static void prepare_write_seq(struct ceph_connection *con)
1239{
1240 dout("prepare_write_seq %p %llu -> %llu\n", con,
1241 con->in_seq_acked, con->in_seq);
1242 con->in_seq_acked = con->in_seq;
1243
1244 con_out_kvec_reset(con);
1245
1246 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1247 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1248 &con->out_temp_ack);
1249
1250 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1251}
1252
31b8006e
SW		 1253/*
1254 * Prepare to write keepalive byte.
1255 */
1256static void prepare_write_keepalive(struct ceph_connection *con)
1257{
1258 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1259 con_out_kvec_reset(con);
1260 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1261 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1262}
1263
1264/*
1265 * Connection negotiation.
1266 */
1267
dac1e716
AE		 1268static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1269 int *auth_proto)
4e7a5dcd 1270{
a3530df3 1271 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1272
1273 if (!con->ops->get_authorizer) {
1274 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1275 con->out_connect.authorizer_len = 0;
729796be 1276 return NULL;
b1c6b980
AE		 1277 }
1278
1279 /* Can't hold the mutex while getting authorizer */
ec302645 1280 mutex_unlock(&con->mutex);
dac1e716 1281 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1282 mutex_lock(&con->mutex);
4e7a5dcd 1283
a3530df3 1284 if (IS_ERR(auth))
729796be 1285 return auth;
8dacc7da 1286 if (con->state != CON_STATE_NEGOTIATING)
729796be 1287 return ERR_PTR(-EAGAIN);
0da5d703 1288
8f43fb53
AE		 1289 con->auth_reply_buf = auth->authorizer_reply_buf;
1290 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1291 return auth;
4e7a5dcd
SW		 1292}
1293
31b8006e
SW		 1294/*
1295 * We connected to a peer and are saying hello.
1296 */
e825a66d 1297static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1298{
e2200423
AE		 1299 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1300 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1301 &con->msgr->my_enc_addr);
eed0ef2c 1302
eed0ef2c 1303 con->out_more = 0;
c9ffc77a 1304 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1305}
1306
e825a66d 1307static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1308{
95c96174 1309 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1310 int proto;
dac1e716 1311 int auth_proto;
729796be 1312 struct ceph_auth_handshake *auth;
31b8006e
SW		 1313
1314 switch (con->peer_name.type) {
1315 case CEPH_ENTITY_TYPE_MON:
1316 proto = CEPH_MONC_PROTOCOL;
1317 break;
1318 case CEPH_ENTITY_TYPE_OSD:
1319 proto = CEPH_OSDC_PROTOCOL;
1320 break;
1321 case CEPH_ENTITY_TYPE_MDS:
1322 proto = CEPH_MDSC_PROTOCOL;
1323 break;
1324 default:
1325 BUG();
1326 }
1327
1328 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1329 con->connect_seq, global_seq, proto);
4e7a5dcd 1330
e825a66d 1331 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1332 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1333 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1334 con->out_connect.global_seq = cpu_to_le32(global_seq);
1335 con->out_connect.protocol_version = cpu_to_le32(proto);
1336 con->out_connect.flags = 0;
31b8006e 1337
dac1e716
AE		 1338 auth_proto = CEPH_AUTH_UNKNOWN;
1339 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1340 if (IS_ERR(auth))
1341 return PTR_ERR(auth);
3da54776 1342
dac1e716 1343 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1344 con->out_connect.authorizer_len = auth ?
1345 cpu_to_le32(auth->authorizer_buf_len) : 0;
1346
e2200423 1347 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1348 &con->out_connect);
1349 if (auth && auth->authorizer_buf_len)
e2200423 1350 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1351 auth->authorizer_buf);
859eb799 1352
31b8006e 1353 con->out_more = 0;
c9ffc77a 1354 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1355
e10c758e 1356 return 0;
31b8006e
SW		 1357}
1358
31b8006e
SW		 1359/*
1360 * write as much of pending kvecs to the socket as we can.
1361 * 1 -> done
1362 * 0 -> socket full, but more to do
1363 * <0 -> error
1364 */
1365static int write_partial_kvec(struct ceph_connection *con)
1366{
1367 int ret;
1368
1369 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1370 while (con->out_kvec_bytes > 0) {
1371 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1372 con->out_kvec_left, con->out_kvec_bytes,
1373 con->out_more);
1374 if (ret <= 0)
1375 goto out;
1376 con->out_kvec_bytes -= ret;
1377 if (con->out_kvec_bytes == 0)
1378 break; /* done */
f42299e6
AE		 1379
1380 /* account for full iov entries consumed */
1381 while (ret >= con->out_kvec_cur->iov_len) {
1382 BUG_ON(!con->out_kvec_left);
1383 ret -= con->out_kvec_cur->iov_len;
1384 con->out_kvec_cur++;
1385 con->out_kvec_left--;
1386 }
1387 /* and for a partially-consumed entry */
1388 if (ret) {
1389 con->out_kvec_cur->iov_len -= ret;
1390 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1391 }
1392 }
1393 con->out_kvec_left = 0;
1394 con->out_kvec_is_msg = false;
1395 ret = 1;
1396out:
1397 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1398 con->out_kvec_bytes, con->out_kvec_left, ret);
1399 return ret; /* done! */
1400}
1401
84ca8fc8 1402static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
9d2a06c2 1403 size_t len, size_t sent)
68b4476b 1404{
84ca8fc8 1405 struct ceph_msg *msg = con->out_msg;
bae6acd9 1406 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
7fe1e5e5 1407 bool need_crc = false;
68b4476b 1408
84ca8fc8
AE		 1409 BUG_ON(!msg);
1410 BUG_ON(!sent);
68b4476b 1411
bae6acd9
AE		 1412 msg_pos->data_pos += sent;
1413 msg_pos->page_pos += sent;
9d2a06c2 1414 if (ceph_msg_has_pages(msg))
e766d7b5 1415 need_crc = ceph_msg_data_advance(&msg->p, sent);
7fe1e5e5
AE		 1416 else if (ceph_msg_has_pagelist(msg))
1417 need_crc = ceph_msg_data_advance(&msg->l, sent);
6aaa4511
AE		 1418#ifdef CONFIG_BLOCK
1419 else if (ceph_msg_has_bio(msg))
1420 need_crc = ceph_msg_data_advance(&msg->b, sent);
1421#endif /* CONFIG_BLOCK */
7fe1e5e5
AE		 1422 BUG_ON(need_crc && sent != len);
1423
5821bd8c
AE		 1424 if (sent < len)
1425 return;
68b4476b 1426
5821bd8c 1427 BUG_ON(sent != len);
bae6acd9
AE		 1428 msg_pos->page_pos = 0;
1429 msg_pos->page++;
1430 msg_pos->did_page_crc = false;
84ca8fc8 1431}
68b4476b 1432
e788182f
AE		 1433static void in_msg_pos_next(struct ceph_connection *con, size_t len,
1434 size_t received)
1435{
1436 struct ceph_msg *msg = con->in_msg;
bae6acd9 1437 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
e788182f
AE		 1438
1439 BUG_ON(!msg);
1440 BUG_ON(!received);
1441
bae6acd9
AE		 1442 msg_pos->data_pos += received;
1443 msg_pos->page_pos += received;
878efabd
AE		 1444 if (ceph_msg_has_pages(msg))
1445 (void) ceph_msg_data_advance(&msg->p, received);
463207aa 1446#ifdef CONFIG_BLOCK
878efabd 1447 else if (ceph_msg_has_bio(msg))
463207aa
AE		 1448 (void) ceph_msg_data_advance(&msg->b, received);
1449#endif /* CONFIG_BLOCK */
e788182f
AE		 1450 if (received < len)
1451 return;
1452
1453 BUG_ON(received != len);
bae6acd9
AE		 1454 msg_pos->page_pos = 0;
1455 msg_pos->page++;
e788182f
AE		 1456}
1457
35b62808
AE		 1458static u32 ceph_crc32c_page(u32 crc, struct page *page,
1459 unsigned int page_offset,
1460 unsigned int length)
1461{
1462 char *kaddr;
1463
1464 kaddr = kmap(page);
1465 BUG_ON(kaddr == NULL);
1466 crc = crc32c(crc, kaddr + page_offset, length);
1467 kunmap(page);
1468
1469 return crc;
1470}
31b8006e
SW		 1471/*
1472 * Write as much message data payload as we can. If we finish, queue
1473 * up the footer.
1474 * 1 -> done, footer is now queued in out_kvec[].
1475 * 0 -> socket full, but more to do
1476 * <0 -> error
1477 */
34d2d200 1478static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1479{
1480 struct ceph_msg *msg = con->out_msg;
bae6acd9 1481 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
95c96174 1482 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
37675b0f 1483 bool do_datacrc = !con->msgr->nocrc;
31b8006e
SW		 1484 int ret;
1485
34d2d200 1486 dout("%s %p msg %p page %d offset %d\n", __func__,
bae6acd9 1487 con, msg, msg_pos->page, msg_pos->page_pos);
31b8006e 1488
5821bd8c
AE		 1489 /*
1490 * Iterate through each page that contains data to be
1491 * written, and send as much as possible for each.
1492 *
1493 * If we are calculating the data crc (the default), we will
1494 * need to map the page. If we have no pages, they have
1495 * been revoked, so use the zero page.
1496 */
bae6acd9 1497 while (data_len > msg_pos->data_pos) {
8a166d05 1498 struct page *page;
e387d525
AE		 1499 size_t page_offset;
1500 size_t length;
8a166d05 1501 bool last_piece;
68b4476b 1502
9d2a06c2 1503 if (ceph_msg_has_pages(msg)) {
e766d7b5
AE		 1504 page = ceph_msg_data_next(&msg->p, &page_offset,
1505 &length, &last_piece);
97fb1c7f 1506 } else if (ceph_msg_has_pagelist(msg)) {
7fe1e5e5
AE		 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 page = ceph_msg_data_next(&msg->b, &page_offset,
1512 &length, &last_piece);
68b4476b 1513#endif
31b8006e 1514 } else {
61fcdc97
AE		 1515 WARN(1, "con %p data_len %u but no outbound data\n",
1516 con, data_len);
1517 ret = -EINVAL;
1518 goto out;
6aaa4511 1519 }
bae6acd9 1520 if (do_datacrc && !msg_pos->did_page_crc) {
5821bd8c 1521 u32 crc = le32_to_cpu(msg->footer.data_crc);
31b8006e 1522
35b62808 1523 crc = ceph_crc32c_page(crc, page, page_offset, length);
84ca8fc8 1524 msg->footer.data_crc = cpu_to_le32(crc);
bae6acd9 1525 msg_pos->did_page_crc = true;
31b8006e 1526 }
e387d525 1527 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1528 length, last_piece);
31b8006e
SW		 1529 if (ret <= 0)
1530 goto out;
1531
9d2a06c2 1532 out_msg_pos_next(con, page, length, (size_t) ret);
31b8006e
SW		 1533 }
1534
34d2d200 1535 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1536
1537 /* prepare and queue up footer, too */
37675b0f 1538 if (!do_datacrc)
84ca8fc8 1539 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1540 con_out_kvec_reset(con);
859eb799 1541 prepare_write_message_footer(con);
31b8006e
SW		 1542 ret = 1;
1543out:
1544 return ret;
1545}
1546
1547/*
1548 * write some zeros
1549 */
1550static int write_partial_skip(struct ceph_connection *con)
1551{
1552 int ret;
1553
1554 while (con->out_skip > 0) {
31739139 1555 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1556
e1dcb128 1557 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1558 if (ret <= 0)
1559 goto out;
1560 con->out_skip -= ret;
1561 }
1562 ret = 1;
1563out:
1564 return ret;
1565}
1566
1567/*
1568 * Prepare to read connection handshake, or an ack.
1569 */
eed0ef2c
SW		 1570static void prepare_read_banner(struct ceph_connection *con)
1571{
1572 dout("prepare_read_banner %p\n", con);
1573 con->in_base_pos = 0;
1574}
1575
31b8006e
SW		 1576static void prepare_read_connect(struct ceph_connection *con)
1577{
1578 dout("prepare_read_connect %p\n", con);
1579 con->in_base_pos = 0;
1580}
1581
1582static void prepare_read_ack(struct ceph_connection *con)
1583{
1584 dout("prepare_read_ack %p\n", con);
1585 con->in_base_pos = 0;
1586}
1587
3a23083b
SW		 1588static void prepare_read_seq(struct ceph_connection *con)
1589{
1590 dout("prepare_read_seq %p\n", con);
1591 con->in_base_pos = 0;
1592 con->in_tag = CEPH_MSGR_TAG_SEQ;
1593}
1594
31b8006e
SW		 1595static void prepare_read_tag(struct ceph_connection *con)
1596{
1597 dout("prepare_read_tag %p\n", con);
1598 con->in_base_pos = 0;
1599 con->in_tag = CEPH_MSGR_TAG_READY;
1600}
1601
1602/*
1603 * Prepare to read a message.
1604 */
1605static int prepare_read_message(struct ceph_connection *con)
1606{
1607 dout("prepare_read_message %p\n", con);
1608 BUG_ON(con->in_msg != NULL);
1609 con->in_base_pos = 0;
1610 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1611 return 0;
1612}
1613
1614
1615static int read_partial(struct ceph_connection *con,
fd51653f 1616 int end, int size, void *object)
31b8006e 1617{
e6cee71f
AE		 1618 while (con->in_base_pos < end) {
1619 int left = end - con->in_base_pos;
31b8006e
SW		 1620 int have = size - left;
1621 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1622 if (ret <= 0)
1623 return ret;
1624 con->in_base_pos += ret;
1625 }
1626 return 1;
1627}
1628
1629
1630/*
1631 * Read all or part of the connect-side handshake on a new connection
1632 */
eed0ef2c 1633static int read_partial_banner(struct ceph_connection *con)
31b8006e 1634{
fd51653f
AE		 1635 int size;
1636 int end;
1637 int ret;
31b8006e 1638
eed0ef2c 1639 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1640
1641 /* peer's banner */
fd51653f
AE		 1642 size = strlen(CEPH_BANNER);
1643 end = size;
1644 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1645 if (ret <= 0)
1646 goto out;
fd51653f
AE		 1647
1648 size = sizeof (con->actual_peer_addr);
1649 end += size;
1650 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1651 if (ret <= 0)
1652 goto out;
fd51653f
AE		 1653
1654 size = sizeof (con->peer_addr_for_me);
1655 end += size;
1656 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1657 if (ret <= 0)
1658 goto out;
fd51653f 1659
eed0ef2c
SW		 1660out:
1661 return ret;
1662}
1663
1664static int read_partial_connect(struct ceph_connection *con)
1665{
fd51653f
AE		 1666 int size;
1667 int end;
1668 int ret;
eed0ef2c
SW		 1669
1670 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1671
fd51653f
AE		 1672 size = sizeof (con->in_reply);
1673 end = size;
1674 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1675 if (ret <= 0)
1676 goto out;
fd51653f
AE		 1677
1678 size = le32_to_cpu(con->in_reply.authorizer_len);
1679 end += size;
1680 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1681 if (ret <= 0)
1682 goto out;
31b8006e 1683
4e7a5dcd
SW		 1684 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1685 con, (int)con->in_reply.tag,
1686 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1687 le32_to_cpu(con->in_reply.global_seq));
1688out:
1689 return ret;
eed0ef2c 1690
31b8006e
SW		 1691}
1692
1693/*
1694 * Verify the hello banner looks okay.
1695 */
1696static int verify_hello(struct ceph_connection *con)
1697{
1698 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1699 pr_err("connect to %s got bad banner\n",
3d14c5d2 1700 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1701 con->error_msg = "protocol error, bad banner";
1702 return -1;
1703 }
1704 return 0;
1705}
1706
1707static bool addr_is_blank(struct sockaddr_storage *ss)
1708{
1709 switch (ss->ss_family) {
1710 case AF_INET:
1711 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1712 case AF_INET6:
1713 return
1714 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1715 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1716 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1717 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1718 }
1719 return false;
1720}
1721
1722static int addr_port(struct sockaddr_storage *ss)
1723{
1724 switch (ss->ss_family) {
1725 case AF_INET:
f28bcfbe 1726 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1727 case AF_INET6:
f28bcfbe 1728 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1729 }
1730 return 0;
1731}
1732
1733static void addr_set_port(struct sockaddr_storage *ss, int p)
1734{
1735 switch (ss->ss_family) {
1736 case AF_INET:
1737 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1738 break;
31b8006e
SW		 1739 case AF_INET6:
1740 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1741 break;
31b8006e
SW		 1742 }
1743}
1744
ee3b56f2
NW		 1745/*
1746 * Unlike other *_pton function semantics, zero indicates success.
1747 */
1748static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1749 char delim, const char **ipend)
1750{
99f0f3b2
AE		 1751 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1752 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1753
1754 memset(ss, 0, sizeof(*ss));
1755
1756 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1757 ss->ss_family = AF_INET;
1758 return 0;
1759 }
1760
1761 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1762 ss->ss_family = AF_INET6;
1763 return 0;
1764 }
1765
1766 return -EINVAL;
1767}
1768
1769/*
1770 * Extract hostname string and resolve using kernel DNS facility.
1771 */
1772#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1773static int ceph_dns_resolve_name(const char *name, size_t namelen,
1774 struct sockaddr_storage *ss, char delim, const char **ipend)
1775{
1776 const char *end, *delim_p;
1777 char *colon_p, *ip_addr = NULL;
1778 int ip_len, ret;
1779
1780 /*
1781 * The end of the hostname occurs immediately preceding the delimiter or
1782 * the port marker (':') where the delimiter takes precedence.
1783 */
1784 delim_p = memchr(name, delim, namelen);
1785 colon_p = memchr(name, ':', namelen);
1786
1787 if (delim_p && colon_p)
1788 end = delim_p < colon_p ? delim_p : colon_p;
1789 else if (!delim_p && colon_p)
1790 end = colon_p;
1791 else {
1792 end = delim_p;
1793 if (!end) /* case: hostname:/ */
1794 end = name + namelen;
1795 }
1796
1797 if (end <= name)
1798 return -EINVAL;
1799
1800 /* do dns_resolve upcall */
1801 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1802 if (ip_len > 0)
1803 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1804 else
1805 ret = -ESRCH;
1806
1807 kfree(ip_addr);
1808
1809 *ipend = end;
1810
1811 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1812 ret, ret ? "failed" : ceph_pr_addr(ss));
1813
1814 return ret;
1815}
1816#else
1817static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1818 struct sockaddr_storage *ss, char delim, const char **ipend)
1819{
1820 return -EINVAL;
1821}
1822#endif
1823
1824/*
1825 * Parse a server name (IP or hostname). If a valid IP address is not found
1826 * then try to extract a hostname to resolve using userspace DNS upcall.
1827 */
1828static int ceph_parse_server_name(const char *name, size_t namelen,
1829 struct sockaddr_storage *ss, char delim, const char **ipend)
1830{
1831 int ret;
1832
1833 ret = ceph_pton(name, namelen, ss, delim, ipend);
1834 if (ret)
1835 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1836
1837 return ret;
1838}
1839
31b8006e
SW		 1840/*
1841 * Parse an ip[:port] list into an addr array. Use the default
1842 * monitor port if a port isn't specified.
1843 */
1844int ceph_parse_ips(const char *c, const char *end,
1845 struct ceph_entity_addr *addr,
1846 int max_count, int *count)
1847{
ee3b56f2 1848 int i, ret = -EINVAL;
31b8006e
SW		 1849 const char *p = c;
1850
1851 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1852 for (i = 0; i < max_count; i++) {
1853 const char *ipend;
1854 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1855 int port;
39139f64
SW		 1856 char delim = ',';
1857
1858 if (*p == '[') {
1859 delim = ']';
1860 p++;
1861 }
31b8006e 1862
ee3b56f2
NW		 1863 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1864 if (ret)
31b8006e 1865 goto bad;
ee3b56f2
NW		 1866 ret = -EINVAL;
1867
31b8006e
SW		 1868 p = ipend;
1869
39139f64
SW		 1870 if (delim == ']') {
1871 if (*p != ']') {
1872 dout("missing matching ']'\n");
1873 goto bad;
1874 }
1875 p++;
1876 }
1877
31b8006e
SW		 1878 /* port? */
1879 if (p < end && *p == ':') {
1880 port = 0;
1881 p++;
1882 while (p < end && *p >= '0' && *p <= '9') {
1883 port = (port * 10) + (*p - '0');
1884 p++;
1885 }
1886 if (port > 65535 || port == 0)
1887 goto bad;
1888 } else {
1889 port = CEPH_MON_PORT;
1890 }
1891
1892 addr_set_port(ss, port);
1893
3d14c5d2 1894 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1895
1896 if (p == end)
1897 break;
1898 if (*p != ',')
1899 goto bad;
1900 p++;
1901 }
1902
1903 if (p != end)
1904 goto bad;
1905
1906 if (count)
1907 *count = i + 1;
1908 return 0;
1909
1910bad:
39139f64 1911 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1912 return ret;
31b8006e 1913}
3d14c5d2 1914EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1915
eed0ef2c 1916static int process_banner(struct ceph_connection *con)
31b8006e 1917{
eed0ef2c 1918 dout("process_banner on %p\n", con);
31b8006e
SW		 1919
1920 if (verify_hello(con) < 0)
1921 return -1;
1922
63f2d211
SW		 1923 ceph_decode_addr(&con->actual_peer_addr);
1924 ceph_decode_addr(&con->peer_addr_for_me);
1925
31b8006e
SW		 1926 /*
1927 * Make sure the other end is who we wanted. note that the other
1928 * end may not yet know their ip address, so if it's 0.0.0.0, give
1929 * them the benefit of the doubt.
1930 */
103e2d3a
SW		 1931 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1932 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1933 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1934 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
cd84db6e 1935 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
3d14c5d2 1936 ceph_pr_addr(&con->peer_addr.in_addr),
cd84db6e 1937 (int)le32_to_cpu(con->peer_addr.nonce),
3d14c5d2 1938 ceph_pr_addr(&con->actual_peer_addr.in_addr),
cd84db6e 1939 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1940 con->error_msg = "wrong peer at address";
31b8006e
SW		 1941 return -1;
1942 }
1943
1944 /*
1945 * did we learn our address?
1946 */
1947 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1948 int port = addr_port(&con->msgr->inst.addr.in_addr);
1949
1950 memcpy(&con->msgr->inst.addr.in_addr,
1951 &con->peer_addr_for_me.in_addr,
1952 sizeof(con->peer_addr_for_me.in_addr));
1953 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1954 encode_my_addr(con->msgr);
eed0ef2c 1955 dout("process_banner learned my addr is %s\n",
3d14c5d2 1956 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1957 }
1958
eed0ef2c
SW		 1959 return 0;
1960}
1961
1962static int process_connect(struct ceph_connection *con)
1963{
3d14c5d2
YS		 1964 u64 sup_feat = con->msgr->supported_features;
1965 u64 req_feat = con->msgr->required_features;
04a419f9 1966 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 1967 int ret;
04a419f9 1968
eed0ef2c
SW		 1969 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1970
31b8006e 1971 switch (con->in_reply.tag) {
04a419f9
SW		 1972 case CEPH_MSGR_TAG_FEATURES:
1973 pr_err("%s%lld %s feature set mismatch,"
1974 " my %llx < server's %llx, missing %llx\n",
1975 ENTITY_NAME(con->peer_name),
3d14c5d2 1976 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1977 sup_feat, server_feat, server_feat & ~sup_feat);
1978 con->error_msg = "missing required protocol features";
0fa6ebc6 1979 reset_connection(con);
04a419f9
SW		 1980 return -1;
1981
31b8006e 1982 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 1983 pr_err("%s%lld %s protocol version mismatch,"
1984 " my %d != server's %d\n",
1985 ENTITY_NAME(con->peer_name),
3d14c5d2 1986 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 1987 le32_to_cpu(con->out_connect.protocol_version),
1988 le32_to_cpu(con->in_reply.protocol_version));
1989 con->error_msg = "protocol version mismatch";
0fa6ebc6 1990 reset_connection(con);
31b8006e
SW		 1991 return -1;
1992
4e7a5dcd
SW		 1993 case CEPH_MSGR_TAG_BADAUTHORIZER:
1994 con->auth_retry++;
1995 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1996 con->auth_retry);
1997 if (con->auth_retry == 2) {
1998 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 1999 return -1;
2000 }
6d4221b5 2001 con_out_kvec_reset(con);
e825a66d 2002 ret = prepare_write_connect(con);
0da5d703
SW		 2003 if (ret < 0)
2004 return ret;
63733a0f 2005 prepare_read_connect(con);
4e7a5dcd 2006 break;
31b8006e
SW		 2007
2008 case CEPH_MSGR_TAG_RESETSESSION:
2009 /*
2010 * If we connected with a large connect_seq but the peer
2011 * has no record of a session with us (no connection, or
2012 * connect_seq == 0), they will send RESETSESION to indicate
2013 * that they must have reset their session, and may have
2014 * dropped messages.
2015 */
2016 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2017 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2018 pr_err("%s%lld %s connection reset\n",
2019 ENTITY_NAME(con->peer_name),
3d14c5d2 2020 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2021 reset_connection(con);
6d4221b5 2022 con_out_kvec_reset(con);
5a0f8fdd
AE		 2023 ret = prepare_write_connect(con);
2024 if (ret < 0)
2025 return ret;
31b8006e
SW		 2026 prepare_read_connect(con);
2027
2028 /* Tell ceph about it. */
ec302645 2029 mutex_unlock(&con->mutex);
31b8006e
SW		 2030 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2031 if (con->ops->peer_reset)
2032 con->ops->peer_reset(con);
ec302645 2033 mutex_lock(&con->mutex);
8dacc7da 2034 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2035 return -EAGAIN;
31b8006e
SW		 2036 break;
2037
2038 case CEPH_MSGR_TAG_RETRY_SESSION:
2039 /*
2040 * If we sent a smaller connect_seq than the peer has, try
2041 * again with a larger value.
2042 */
5bdca4e0 2043 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2044 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2045 le32_to_cpu(con->in_reply.connect_seq));
2046 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2047 con_out_kvec_reset(con);
5a0f8fdd
AE		 2048 ret = prepare_write_connect(con);
2049 if (ret < 0)
2050 return ret;
31b8006e
SW		 2051 prepare_read_connect(con);
2052 break;
2053
2054 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2055 /*
2056 * If we sent a smaller global_seq than the peer has, try
2057 * again with a larger value.
2058 */
eed0ef2c 2059 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2060 con->peer_global_seq,
5bdca4e0 2061 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2062 get_global_seq(con->msgr,
5bdca4e0 2063 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2064 con_out_kvec_reset(con);
5a0f8fdd
AE		 2065 ret = prepare_write_connect(con);
2066 if (ret < 0)
2067 return ret;
31b8006e
SW		 2068 prepare_read_connect(con);
2069 break;
2070
3a23083b 2071 case CEPH_MSGR_TAG_SEQ:
31b8006e 2072 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2073 if (req_feat & ~server_feat) {
2074 pr_err("%s%lld %s protocol feature mismatch,"
2075 " my required %llx > server's %llx, need %llx\n",
2076 ENTITY_NAME(con->peer_name),
3d14c5d2 2077 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2078 req_feat, server_feat, req_feat & ~server_feat);
2079 con->error_msg = "missing required protocol features";
0fa6ebc6 2080 reset_connection(con);
04a419f9
SW		 2081 return -1;
2082 }
8dacc7da 2083
122070a2 2084 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2085 con->state = CON_STATE_OPEN;
20e55c4c 2086 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2087 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2088 con->connect_seq++;
aba558e2 2089 con->peer_features = server_feat;
31b8006e
SW		 2090 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2091 con->peer_global_seq,
2092 le32_to_cpu(con->in_reply.connect_seq),
2093 con->connect_seq);
2094 WARN_ON(con->connect_seq !=
2095 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2096
2097 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2098 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2099
85effe18 2100 con->delay = 0; /* reset backoff memory */
92ac41d0 2101
3a23083b
SW		 2102 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2103 prepare_write_seq(con);
2104 prepare_read_seq(con);
2105 } else {
2106 prepare_read_tag(con);
2107 }
31b8006e
SW		 2108 break;
2109
2110 case CEPH_MSGR_TAG_WAIT:
2111 /*
2112 * If there is a connection race (we are opening
2113 * connections to each other), one of us may just have
2114 * to WAIT. This shouldn't happen if we are the
2115 * client.
2116 */
04177882
SW		 2117 pr_err("process_connect got WAIT as client\n");
2118 con->error_msg = "protocol error, got WAIT as client";
2119 return -1;
31b8006e
SW		 2120
2121 default:
2122 pr_err("connect protocol error, will retry\n");
2123 con->error_msg = "protocol error, garbage tag during connect";
2124 return -1;
2125 }
2126 return 0;
2127}
2128
2129
2130/*
2131 * read (part of) an ack
2132 */
2133static int read_partial_ack(struct ceph_connection *con)
2134{
fd51653f
AE		 2135 int size = sizeof (con->in_temp_ack);
2136 int end = size;
31b8006e 2137
fd51653f 2138 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2139}
2140
31b8006e
SW		 2141/*
2142 * We can finally discard anything that's been acked.
2143 */
2144static void process_ack(struct ceph_connection *con)
2145{
2146 struct ceph_msg *m;
2147 u64 ack = le64_to_cpu(con->in_temp_ack);
2148 u64 seq;
2149
31b8006e
SW		 2150 while (!list_empty(&con->out_sent)) {
2151 m = list_first_entry(&con->out_sent, struct ceph_msg,
2152 list_head);
2153 seq = le64_to_cpu(m->hdr.seq);
2154 if (seq > ack)
2155 break;
2156 dout("got ack for seq %llu type %d at %p\n", seq,
2157 le16_to_cpu(m->hdr.type), m);
4cf9d544 2158 m->ack_stamp = jiffies;
31b8006e
SW		 2159 ceph_msg_remove(m);
2160 }
31b8006e
SW		 2161 prepare_read_tag(con);
2162}
2163
2164
2450418c 2165static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2166 struct kvec *section,
2167 unsigned int sec_len, u32 *crc)
2450418c 2168{
68b4476b 2169 int ret, left;
2450418c
YS		 2170
2171 BUG_ON(!section);
2172
2173 while (section->iov_len < sec_len) {
2174 BUG_ON(section->iov_base == NULL);
2175 left = sec_len - section->iov_len;
2176 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2177 section->iov_len, left);
2178 if (ret <= 0)
2179 return ret;
2180 section->iov_len += ret;
2450418c 2181 }
fe3ad593
AE		 2182 if (section->iov_len == sec_len)
2183 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2184
2450418c
YS		 2185 return 1;
2186}
31b8006e 2187
34d2d200
AE		 2188static int read_partial_msg_data(struct ceph_connection *con)
2189{
2190 struct ceph_msg *msg = con->in_msg;
2191 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2192 const bool do_datacrc = !con->msgr->nocrc;
2193 unsigned int data_len;
686be208
AE		 2194 struct page *page;
2195 size_t page_offset;
2196 size_t length;
34d2d200
AE		 2197 int ret;
2198
2199 BUG_ON(!msg);
2200
2201 data_len = le32_to_cpu(con->in_hdr.data_len);
2202 while (msg_pos->data_pos < data_len) {
97fb1c7f 2203 if (ceph_msg_has_pages(msg)) {
686be208
AE		 2204 page = ceph_msg_data_next(&msg->p, &page_offset,
2205 &length, NULL);
34d2d200 2206#ifdef CONFIG_BLOCK
97fb1c7f 2207 } else if (ceph_msg_has_bio(msg)) {
686be208
AE		 2208 page = ceph_msg_data_next(&msg->b, &page_offset,
2209 &length, NULL);
34d2d200
AE		 2210#endif
2211 } else {
2212 BUG_ON(1);
2213 }
686be208
AE		 2214 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2215 if (ret <= 0)
2216 return ret;
2217
2218 if (do_datacrc)
2219 con->in_data_crc = ceph_crc32c_page(con->in_data_crc,
2220 page, page_offset, ret);
2221
2222 in_msg_pos_next(con, length, ret);
34d2d200
AE		 2223 }
2224
2225 return 1; /* must return > 0 to indicate success */
2226}
2227
31b8006e
SW		 2228/*
2229 * read (part of) a message.
2230 */
686be208
AE		 2231static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2232
31b8006e
SW		 2233static int read_partial_message(struct ceph_connection *con)
2234{
2235 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2236 int size;
2237 int end;
31b8006e 2238 int ret;
95c96174 2239 unsigned int front_len, middle_len, data_len;
37675b0f 2240 bool do_datacrc = !con->msgr->nocrc;
ae18756b 2241 u64 seq;
fe3ad593 2242 u32 crc;
31b8006e
SW		 2243
2244 dout("read_partial_message con %p msg %p\n", con, m);
2245
2246 /* header */
fd51653f
AE		 2247 size = sizeof (con->in_hdr);
2248 end = size;
2249 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2250 if (ret <= 0)
2251 return ret;
fe3ad593
AE		 2252
2253 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2254 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2255 pr_err("read_partial_message bad hdr "
2256 " crc %u != expected %u\n",
2257 crc, con->in_hdr.crc);
2258 return -EBADMSG;
2259 }
2260
31b8006e
SW		 2261 front_len = le32_to_cpu(con->in_hdr.front_len);
2262 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2263 return -EIO;
2264 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2265 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2266 return -EIO;
2267 data_len = le32_to_cpu(con->in_hdr.data_len);
2268 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2269 return -EIO;
2270
ae18756b
SW		 2271 /* verify seq# */
2272 seq = le64_to_cpu(con->in_hdr.seq);
2273 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2274 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2275 ENTITY_NAME(con->peer_name),
3d14c5d2 2276 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2277 seq, con->in_seq + 1);
2278 con->in_base_pos = -front_len - middle_len - data_len -
2279 sizeof(m->footer);
2280 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2281 return 0;
2282 } else if ((s64)seq - (s64)con->in_seq > 1) {
2283 pr_err("read_partial_message bad seq %lld expected %lld\n",
2284 seq, con->in_seq + 1);
2285 con->error_msg = "bad message sequence # for incoming message";
2286 return -EBADMSG;
2287 }
2288
31b8006e
SW		 2289 /* allocate message? */
2290 if (!con->in_msg) {
4740a623
SW		 2291 int skip = 0;
2292
31b8006e 2293 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2294 front_len, data_len);
4740a623
SW		 2295 ret = ceph_con_in_msg_alloc(con, &skip);
2296 if (ret < 0)
2297 return ret;
2450418c 2298 if (skip) {
31b8006e 2299 /* skip this message */
a79832f2 2300 dout("alloc_msg said skip message\n");
ae32be31 2301 BUG_ON(con->in_msg);
31b8006e
SW		 2302 con->in_base_pos = -front_len - middle_len - data_len -
2303 sizeof(m->footer);
2304 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2305 con->in_seq++;
31b8006e
SW		 2306 return 0;
2307 }
38941f80 2308
4740a623 2309 BUG_ON(!con->in_msg);
38941f80 2310 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2311 m = con->in_msg;
2312 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2313 if (m->middle)
2314 m->middle->vec.iov_len = 0;
9d7f0f13 2315
78625051 2316 /* prepare for data payload, if any */
a4107026 2317
78625051
AE		 2318 if (data_len)
2319 prepare_message_data(con->in_msg, &con->in_msg_pos);
31b8006e
SW		 2320 }
2321
2322 /* front */
2450418c
YS		 2323 ret = read_partial_message_section(con, &m->front, front_len,
2324 &con->in_front_crc);
2325 if (ret <= 0)
2326 return ret;
31b8006e
SW		 2327
2328 /* middle */
2450418c 2329 if (m->middle) {
213c99ee
SW		 2330 ret = read_partial_message_section(con, &m->middle->vec,
2331 middle_len,
2450418c 2332 &con->in_middle_crc);
31b8006e
SW		 2333 if (ret <= 0)
2334 return ret;
31b8006e
SW		 2335 }
2336
2337 /* (page) data */
34d2d200
AE		 2338 if (data_len) {
2339 ret = read_partial_msg_data(con);
2340 if (ret <= 0)
2341 return ret;
31b8006e
SW		 2342 }
2343
31b8006e 2344 /* footer */
fd51653f
AE		 2345 size = sizeof (m->footer);
2346 end += size;
2347 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2348 if (ret <= 0)
2349 return ret;
2350
31b8006e
SW		 2351 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2352 m, front_len, m->footer.front_crc, middle_len,
2353 m->footer.middle_crc, data_len, m->footer.data_crc);
2354
2355 /* crc ok? */
2356 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2357 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2358 m, con->in_front_crc, m->footer.front_crc);
2359 return -EBADMSG;
2360 }
2361 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2362 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2363 m, con->in_middle_crc, m->footer.middle_crc);
2364 return -EBADMSG;
2365 }
bca064d2 2366 if (do_datacrc &&
31b8006e
SW		 2367 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2368 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2369 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2370 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2371 return -EBADMSG;
2372 }
2373
2374 return 1; /* done! */
2375}
2376
2377/*
2378 * Process message. This happens in the worker thread. The callback should
2379 * be careful not to do anything that waits on other incoming messages or it
2380 * may deadlock.
2381 */
2382static void process_message(struct ceph_connection *con)
2383{
5e095e8b 2384 struct ceph_msg *msg;
31b8006e 2385
38941f80
AE		 2386 BUG_ON(con->in_msg->con != con);
2387 con->in_msg->con = NULL;
5e095e8b 2388 msg = con->in_msg;
31b8006e 2389 con->in_msg = NULL;
36eb71aa 2390 con->ops->put(con);
31b8006e
SW		 2391
2392 /* if first message, set peer_name */
2393 if (con->peer_name.type == 0)
dbad185d 2394 con->peer_name = msg->hdr.src;
31b8006e 2395
31b8006e 2396 con->in_seq++;
ec302645 2397 mutex_unlock(&con->mutex);
31b8006e
SW		 2398
2399 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2400 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2401 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2402 le16_to_cpu(msg->hdr.type),
2403 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2404 le32_to_cpu(msg->hdr.front_len),
2405 le32_to_cpu(msg->hdr.data_len),
2406 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2407 con->ops->dispatch(con, msg);
ec302645
SW		 2408
2409 mutex_lock(&con->mutex);
31b8006e
SW		 2410}
2411
2412
2413/*
2414 * Write something to the socket. Called in a worker thread when the
2415 * socket appears to be writeable and we have something ready to send.
2416 */
2417static int try_write(struct ceph_connection *con)
2418{
31b8006e
SW		 2419 int ret = 1;
2420
d59315ca 2421 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2422
31b8006e
SW		 2423more:
2424 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2425
2426 /* open the socket first? */
8dacc7da
SW		 2427 if (con->state == CON_STATE_PREOPEN) {
2428 BUG_ON(con->sock);
2429 con->state = CON_STATE_CONNECTING;
a5988c49 2430
e2200423 2431 con_out_kvec_reset(con);
e825a66d 2432 prepare_write_banner(con);
eed0ef2c 2433 prepare_read_banner(con);
31b8006e 2434
cf3e5c40 2435 BUG_ON(con->in_msg);
31b8006e
SW		 2436 con->in_tag = CEPH_MSGR_TAG_READY;
2437 dout("try_write initiating connect on %p new state %lu\n",
2438 con, con->state);
41617d0c
AE		 2439 ret = ceph_tcp_connect(con);
2440 if (ret < 0) {
31b8006e 2441 con->error_msg = "connect error";
31b8006e
SW		 2442 goto out;
2443 }
2444 }
2445
2446more_kvec:
2447 /* kvec data queued? */
2448 if (con->out_skip) {
2449 ret = write_partial_skip(con);
2450 if (ret <= 0)
42961d23 2451 goto out;
31b8006e
SW		 2452 }
2453 if (con->out_kvec_left) {
2454 ret = write_partial_kvec(con);
2455 if (ret <= 0)
42961d23 2456 goto out;
31b8006e
SW		 2457 }
2458
2459 /* msg pages? */
2460 if (con->out_msg) {
c86a2930
SW		 2461 if (con->out_msg_done) {
2462 ceph_msg_put(con->out_msg);
2463 con->out_msg = NULL; /* we're done with this one */
2464 goto do_next;
2465 }
2466
34d2d200 2467 ret = write_partial_message_data(con);
31b8006e
SW		 2468 if (ret == 1)
2469 goto more_kvec; /* we need to send the footer, too! */
2470 if (ret == 0)
42961d23 2471 goto out;
31b8006e 2472 if (ret < 0) {
34d2d200 2473 dout("try_write write_partial_message_data err %d\n",
31b8006e 2474 ret);
42961d23 2475 goto out;
31b8006e
SW		 2476 }
2477 }
2478
c86a2930 2479do_next:
8dacc7da 2480 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2481 /* is anything else pending? */
2482 if (!list_empty(&con->out_queue)) {
2483 prepare_write_message(con);
2484 goto more;
2485 }
2486 if (con->in_seq > con->in_seq_acked) {
2487 prepare_write_ack(con);
2488 goto more;
2489 }
c9ffc77a 2490 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2491 prepare_write_keepalive(con);
2492 goto more;
2493 }
2494 }
2495
2496 /* Nothing to do! */
c9ffc77a 2497 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2498 dout("try_write nothing else to write.\n");
31b8006e
SW		 2499 ret = 0;
2500out:
42961d23 2501 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2502 return ret;
2503}
2504
2505
2506
2507/*
2508 * Read what we can from the socket.
2509 */
2510static int try_read(struct ceph_connection *con)
2511{
31b8006e
SW		 2512 int ret = -1;
2513
8dacc7da
SW		 2514more:
2515 dout("try_read start on %p state %lu\n", con, con->state);
2516 if (con->state != CON_STATE_CONNECTING &&
2517 con->state != CON_STATE_NEGOTIATING &&
2518 con->state != CON_STATE_OPEN)
31b8006e
SW		 2519 return 0;
2520
8dacc7da 2521 BUG_ON(!con->sock);
ec302645 2522
31b8006e
SW		 2523 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2524 con->in_base_pos);
0da5d703 2525
8dacc7da 2526 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2527 dout("try_read connecting\n");
2528 ret = read_partial_banner(con);
2529 if (ret <= 0)
ab166d5a 2530 goto out;
7593af92
AE		 2531 ret = process_banner(con);
2532 if (ret < 0)
2533 goto out;
2534
8dacc7da 2535 con->state = CON_STATE_NEGOTIATING;
7593af92 2536
6d4221b5
JS		 2537 /*
2538 * Received banner is good, exchange connection info.
2539 * Do not reset out_kvec, as sending our banner raced
2540 * with receiving peer banner after connect completed.
2541 */
7593af92
AE		 2542 ret = prepare_write_connect(con);
2543 if (ret < 0)
2544 goto out;
2545 prepare_read_connect(con);
2546
2547 /* Send connection info before awaiting response */
0da5d703
SW		 2548 goto out;
2549 }
2550
8dacc7da 2551 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2552 dout("try_read negotiating\n");
31b8006e
SW		 2553 ret = read_partial_connect(con);
2554 if (ret <= 0)
31b8006e 2555 goto out;
98bdb0aa
SW		 2556 ret = process_connect(con);
2557 if (ret < 0)
2558 goto out;
31b8006e
SW		 2559 goto more;
2560 }
2561
122070a2 2562 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2563
31b8006e
SW		 2564 if (con->in_base_pos < 0) {
2565 /*
2566 * skipping + discarding content.
2567 *
2568 * FIXME: there must be a better way to do this!
2569 */
84495f49
AE		 2570 static char buf[SKIP_BUF_SIZE];
2571 int skip = min((int) sizeof (buf), -con->in_base_pos);
2572
31b8006e
SW		 2573 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2574 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2575 if (ret <= 0)
98bdb0aa 2576 goto out;
31b8006e
SW		 2577 con->in_base_pos += ret;
2578 if (con->in_base_pos)
2579 goto more;
2580 }
2581 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2582 /*
2583 * what's next?
2584 */
2585 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2586 if (ret <= 0)
98bdb0aa 2587 goto out;
31b8006e
SW		 2588 dout("try_read got tag %d\n", (int)con->in_tag);
2589 switch (con->in_tag) {
2590 case CEPH_MSGR_TAG_MSG:
2591 prepare_read_message(con);
2592 break;
2593 case CEPH_MSGR_TAG_ACK:
2594 prepare_read_ack(con);
2595 break;
2596 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2597 con_close_socket(con);
2598 con->state = CON_STATE_CLOSED;
98bdb0aa 2599 goto out;
31b8006e
SW		 2600 default:
2601 goto bad_tag;
2602 }
2603 }
2604 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2605 ret = read_partial_message(con);
2606 if (ret <= 0) {
2607 switch (ret) {
2608 case -EBADMSG:
2609 con->error_msg = "bad crc";
2610 ret = -EIO;
98bdb0aa 2611 break;
31b8006e
SW		 2612 case -EIO:
2613 con->error_msg = "io error";
98bdb0aa 2614 break;
31b8006e 2615 }
98bdb0aa 2616 goto out;
31b8006e
SW		 2617 }
2618 if (con->in_tag == CEPH_MSGR_TAG_READY)
2619 goto more;
2620 process_message(con);
7b862e07
SW		 2621 if (con->state == CON_STATE_OPEN)
2622 prepare_read_tag(con);
31b8006e
SW		 2623 goto more;
2624 }
3a23083b
SW		 2625 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2626 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2627 /*
2628 * the final handshake seq exchange is semantically
2629 * equivalent to an ACK
2630 */
31b8006e
SW		 2631 ret = read_partial_ack(con);
2632 if (ret <= 0)
98bdb0aa 2633 goto out;
31b8006e
SW		 2634 process_ack(con);
2635 goto more;
2636 }
2637
31b8006e 2638out:
98bdb0aa 2639 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2640 return ret;
2641
2642bad_tag:
2643 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2644 con->error_msg = "protocol error, garbage tag";
2645 ret = -1;
2646 goto out;
2647}
2648
2649
2650/*
802c6d96
AE		 2651 * Atomically queue work on a connection after the specified delay.
2652 * Bump @con reference to avoid races with connection teardown.
2653 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2654 */
802c6d96 2655static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2656{
31b8006e 2657 if (!con->ops->get(con)) {
802c6d96
AE		 2658 dout("%s %p ref count 0\n", __func__, con);
2659
2660 return -ENOENT;
31b8006e
SW		 2661 }
2662
802c6d96
AE		 2663 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2664 dout("%s %p - already queued\n", __func__, con);
31b8006e 2665 con->ops->put(con);
802c6d96
AE		 2666
2667 return -EBUSY;
31b8006e 2668 }
802c6d96
AE		 2669
2670 dout("%s %p %lu\n", __func__, con, delay);
2671
2672 return 0;
2673}
2674
2675static void queue_con(struct ceph_connection *con)
2676{
2677 (void) queue_con_delay(con, 0);
31b8006e
SW		 2678}
2679
7bb21d68
AE		 2680static bool con_sock_closed(struct ceph_connection *con)
2681{
c9ffc77a 2682 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2683 return false;
2684
2685#define CASE(x) \
2686 case CON_STATE_ ## x: \
2687 con->error_msg = "socket closed (con state " #x ")"; \
2688 break;
2689
2690 switch (con->state) {
2691 CASE(CLOSED);
2692 CASE(PREOPEN);
2693 CASE(CONNECTING);
2694 CASE(NEGOTIATING);
2695 CASE(OPEN);
2696 CASE(STANDBY);
2697 default:
2698 pr_warning("%s con %p unrecognized state %lu\n",
2699 __func__, con, con->state);
2700 con->error_msg = "unrecognized con state";
2701 BUG();
2702 break;
2703 }
2704#undef CASE
2705
2706 return true;
2707}
2708
f20a39fd
AE		 2709static bool con_backoff(struct ceph_connection *con)
2710{
2711 int ret;
2712
2713 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2714 return false;
2715
2716 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2717 if (ret) {
2718 dout("%s: con %p FAILED to back off %lu\n", __func__,
2719 con, con->delay);
2720 BUG_ON(ret == -ENOENT);
2721 con_flag_set(con, CON_FLAG_BACKOFF);
2722 }
2723
2724 return true;
2725}
2726
93209264
AE		 2727/* Finish fault handling; con->mutex must *not* be held here */
2728
2729static void con_fault_finish(struct ceph_connection *con)
2730{
2731 /*
2732 * in case we faulted due to authentication, invalidate our
2733 * current tickets so that we can get new ones.
2734 */
2735 if (con->auth_retry && con->ops->invalidate_authorizer) {
2736 dout("calling invalidate_authorizer()\n");
2737 con->ops->invalidate_authorizer(con);
2738 }
2739
2740 if (con->ops->fault)
2741 con->ops->fault(con);
2742}
2743
31b8006e
SW		 2744/*
2745 * Do some work on a connection. Drop a connection ref when we're done.
2746 */
2747static void con_work(struct work_struct *work)
2748{
2749 struct ceph_connection *con = container_of(work, struct ceph_connection,
2750 work.work);
49659416 2751 bool fault;
31b8006e 2752
9dd4658d 2753 mutex_lock(&con->mutex);
49659416
AE		 2754 while (true) {
2755 int ret;
31b8006e 2756
49659416
AE		 2757 if ((fault = con_sock_closed(con))) {
2758 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2759 break;
2760 }
2761 if (con_backoff(con)) {
2762 dout("%s: con %p BACKOFF\n", __func__, con);
2763 break;
2764 }
2765 if (con->state == CON_STATE_STANDBY) {
2766 dout("%s: con %p STANDBY\n", __func__, con);
2767 break;
2768 }
2769 if (con->state == CON_STATE_CLOSED) {
2770 dout("%s: con %p CLOSED\n", __func__, con);
2771 BUG_ON(con->sock);
2772 break;
2773 }
2774 if (con->state == CON_STATE_PREOPEN) {
2775 dout("%s: con %p PREOPEN\n", __func__, con);
2776 BUG_ON(con->sock);
2777 }
0da5d703 2778
49659416
AE		 2779 ret = try_read(con);
2780 if (ret < 0) {
2781 if (ret == -EAGAIN)
2782 continue;
2783 con->error_msg = "socket error on read";
2784 fault = true;
2785 break;
2786 }
2787
2788 ret = try_write(con);
2789 if (ret < 0) {
2790 if (ret == -EAGAIN)
2791 continue;
2792 con->error_msg = "socket error on write";
2793 fault = true;
2794 }
2795
2796 break; /* If we make it to here, we're done */
3a140a0d 2797 }
b6e7b6a1
AE		 2798 if (fault)
2799 con_fault(con);
9dd4658d 2800 mutex_unlock(&con->mutex);
0da5d703 2801
b6e7b6a1
AE		 2802 if (fault)
2803 con_fault_finish(con);
2804
2805 con->ops->put(con);
31b8006e
SW		 2806}
2807
31b8006e
SW		 2808/*
2809 * Generic error/fault handler. A retry mechanism is used with
2810 * exponential backoff
2811 */
93209264 2812static void con_fault(struct ceph_connection *con)
31b8006e 2813{
28362986 2814 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
3d14c5d2 2815 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2816 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2817 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2818
122070a2 2819 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2820 con->state != CON_STATE_NEGOTIATING &&
2821 con->state != CON_STATE_OPEN);
ec302645 2822
31b8006e 2823 con_close_socket(con);
5e095e8b 2824
c9ffc77a 2825 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2826 dout("fault on LOSSYTX channel, marking CLOSED\n");
2827 con->state = CON_STATE_CLOSED;
93209264 2828 return;
3b5ede07
SW		 2829 }
2830
5e095e8b 2831 if (con->in_msg) {
38941f80
AE		 2832 BUG_ON(con->in_msg->con != con);
2833 con->in_msg->con = NULL;
5e095e8b
SW		 2834 ceph_msg_put(con->in_msg);
2835 con->in_msg = NULL;
36eb71aa 2836 con->ops->put(con);
5e095e8b 2837 }
31b8006e 2838
e80a52d1
SW		 2839 /* Requeue anything that hasn't been acked */
2840 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2841
e76661d0
SW		 2842 /* If there are no messages queued or keepalive pending, place
2843 * the connection in a STANDBY state */
2844 if (list_empty(&con->out_queue) &&
c9ffc77a 2845 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2846 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2847 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2848 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2849 } else {
2850 /* retry after a delay. */
8dacc7da 2851 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2852 if (con->delay == 0)
2853 con->delay = BASE_DELAY_INTERVAL;
2854 else if (con->delay < MAX_DELAY_INTERVAL)
2855 con->delay *= 2;
c9ffc77a 2856 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2857 queue_con(con);
31b8006e 2858 }
31b8006e
SW		 2859}
2860
2861
2862
2863/*
15d9882c 2864 * initialize a new messenger instance
31b8006e 2865 */
15d9882c
AE		 2866void ceph_messenger_init(struct ceph_messenger *msgr,
2867 struct ceph_entity_addr *myaddr,
2868 u32 supported_features,
2869 u32 required_features,
2870 bool nocrc)
31b8006e 2871{
3d14c5d2
YS		 2872 msgr->supported_features = supported_features;
2873 msgr->required_features = required_features;
2874
31b8006e
SW		 2875 spin_lock_init(&msgr->global_seq_lock);
2876
31b8006e
SW		 2877 if (myaddr)
2878 msgr->inst.addr = *myaddr;
2879
2880 /* select a random nonce */
ac8839d7 2881 msgr->inst.addr.type = 0;
103e2d3a 2882 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2883 encode_my_addr(msgr);
15d9882c 2884 msgr->nocrc = nocrc;
31b8006e 2885
a2a32584 2886 atomic_set(&msgr->stopping, 0);
31b8006e 2887
15d9882c 2888 dout("%s %p\n", __func__, msgr);
31b8006e 2889}
15d9882c 2890EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2891
e00de341
SW		 2892static void clear_standby(struct ceph_connection *con)
2893{
2894 /* come back from STANDBY? */
8dacc7da 2895 if (con->state == CON_STATE_STANDBY) {
e00de341 2896 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2897 con->state = CON_STATE_PREOPEN;
e00de341 2898 con->connect_seq++;
c9ffc77a
AE		 2899 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2900 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2901 }
2902}
2903
31b8006e
SW		 2904/*
2905 * Queue up an outgoing message on the given connection.
2906 */
2907void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2908{
31b8006e 2909 /* set src+dst */
dbad185d 2910 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2911 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2912 msg->needs_out_seq = true;
2913
ec302645 2914 mutex_lock(&con->mutex);
92ce034b 2915
8dacc7da 2916 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2917 dout("con_send %p closed, dropping %p\n", con, msg);
2918 ceph_msg_put(msg);
2919 mutex_unlock(&con->mutex);
2920 return;
2921 }
2922
38941f80 2923 BUG_ON(msg->con != NULL);
36eb71aa 2924 msg->con = con->ops->get(con);
92ce034b
AE		 2925 BUG_ON(msg->con == NULL);
2926
31b8006e
SW		 2927 BUG_ON(!list_empty(&msg->list_head));
2928 list_add_tail(&msg->list_head, &con->out_queue);
2929 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2930 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2931 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2932 le32_to_cpu(msg->hdr.front_len),
2933 le32_to_cpu(msg->hdr.middle_len),
2934 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 2935
2936 clear_standby(con);
ec302645 2937 mutex_unlock(&con->mutex);
31b8006e
SW		 2938
2939 /* if there wasn't anything waiting to send before, queue
2940 * new work */
c9ffc77a 2941 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2942 queue_con(con);
2943}
3d14c5d2 2944EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 2945
2946/*
2947 * Revoke a message that was previously queued for send
2948 */
6740a845 2949void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 2950{
6740a845
AE		 2951 struct ceph_connection *con = msg->con;
2952
2953 if (!con)
2954 return; /* Message not in our possession */
2955
ec302645 2956 mutex_lock(&con->mutex);
31b8006e 2957 if (!list_empty(&msg->list_head)) {
38941f80 2958 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 2959 list_del_init(&msg->list_head);
38941f80 2960 BUG_ON(msg->con == NULL);
36eb71aa 2961 msg->con->ops->put(msg->con);
38941f80 2962 msg->con = NULL;
31b8006e 2963 msg->hdr.seq = 0;
38941f80 2964
31b8006e 2965 ceph_msg_put(msg);
ed98adad
SW		 2966 }
2967 if (con->out_msg == msg) {
38941f80 2968 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 2969 con->out_msg = NULL;
31b8006e
SW		 2970 if (con->out_kvec_is_msg) {
2971 con->out_skip = con->out_kvec_bytes;
2972 con->out_kvec_is_msg = false;
2973 }
ed98adad 2974 msg->hdr.seq = 0;
92ce034b
AE		 2975
2976 ceph_msg_put(msg);
31b8006e 2977 }
ec302645 2978 mutex_unlock(&con->mutex);
31b8006e
SW		 2979}
2980
350b1c32 2981/*
0d59ab81 2982 * Revoke a message that we may be reading data into
350b1c32 2983 */
8921d114 2984void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 2985{
8921d114
AE		 2986 struct ceph_connection *con;
2987
2988 BUG_ON(msg == NULL);
2989 if (!msg->con) {
2990 dout("%s msg %p null con\n", __func__, msg);
2991
2992 return; /* Message not in our possession */
2993 }
2994
2995 con = msg->con;
350b1c32 2996 mutex_lock(&con->mutex);
8921d114 2997 if (con->in_msg == msg) {
95c96174
ED		 2998 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2999 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3000 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3001
3002 /* skip rest of message */
8921d114
AE		 3003 dout("%s %p msg %p revoked\n", __func__, con, msg);
3004 con->in_base_pos = con->in_base_pos -
350b1c32 3005 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3006 front_len -
3007 middle_len -
3008 data_len -
350b1c32 3009 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3010 ceph_msg_put(con->in_msg);
3011 con->in_msg = NULL;
3012 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3013 con->in_seq++;
350b1c32 3014 } else {
8921d114
AE		 3015 dout("%s %p in_msg %p msg %p no-op\n",
3016 __func__, con, con->in_msg, msg);
350b1c32
SW		 3017 }
3018 mutex_unlock(&con->mutex);
3019}
3020
31b8006e
SW		 3021/*
3022 * Queue a keepalive byte to ensure the tcp connection is alive.
3023 */
3024void ceph_con_keepalive(struct ceph_connection *con)
3025{
e00de341 3026 dout("con_keepalive %p\n", con);
00650931 3027 mutex_lock(&con->mutex);
e00de341 3028 clear_standby(con);
00650931 3029 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3030 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3031 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3032 queue_con(con);
3033}
3d14c5d2 3034EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3035
43794509
AE		 3036static void ceph_msg_data_init(struct ceph_msg_data *data)
3037{
3038 data->type = CEPH_MSG_DATA_NONE;
3039}
3040
02afca6c 3041void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3042 size_t length, size_t alignment)
02afca6c 3043{
07aa1558
AE		 3044 BUG_ON(!pages);
3045 BUG_ON(!length);
43794509 3046 BUG_ON(msg->p.type != CEPH_MSG_DATA_NONE);
02afca6c 3047
43794509 3048 msg->p.type = CEPH_MSG_DATA_PAGES;
f9e15777
AE		 3049 msg->p.pages = pages;
3050 msg->p.length = length;
3051 msg->p.alignment = alignment & ~PAGE_MASK;
02afca6c
AE		 3052}
3053EXPORT_SYMBOL(ceph_msg_data_set_pages);
31b8006e 3054
27fa8385
AE		 3055void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3056 struct ceph_pagelist *pagelist)
3057{
07aa1558
AE		 3058 BUG_ON(!pagelist);
3059 BUG_ON(!pagelist->length);
43794509 3060 BUG_ON(msg->l.type != CEPH_MSG_DATA_NONE);
27fa8385 3061
43794509 3062 msg->l.type = CEPH_MSG_DATA_PAGELIST;
f9e15777 3063 msg->l.pagelist = pagelist;
27fa8385
AE		 3064}
3065EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3066
3067void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio)
3068{
07aa1558 3069 BUG_ON(!bio);
43794509 3070 BUG_ON(msg->b.type != CEPH_MSG_DATA_NONE);
27fa8385 3071
43794509 3072 msg->b.type = CEPH_MSG_DATA_BIO;
f9e15777 3073 msg->b.bio = bio;
27fa8385
AE		 3074}
3075EXPORT_SYMBOL(ceph_msg_data_set_bio);
3076
31b8006e
SW		 3077/*
3078 * construct a new message with given type, size
3079 * the new msg has a ref count of 1.
3080 */
b61c2763
SW		 3081struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3082 bool can_fail)
31b8006e
SW		 3083{
3084 struct ceph_msg *m;
3085
9516e45b 3086 m = kzalloc(sizeof(*m), flags);
31b8006e
SW		 3087 if (m == NULL)
3088 goto out;
31b8006e
SW		 3089
3090 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3091 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3092 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3093
9516e45b
AE		 3094 INIT_LIST_HEAD(&m->list_head);
3095 kref_init(&m->kref);
ca20892d 3096
43794509
AE		 3097 ceph_msg_data_init(&m->p);
3098 ceph_msg_data_init(&m->l);
3099 ceph_msg_data_init(&m->b);
43794509 3100
31b8006e 3101 /* front */
9516e45b 3102 m->front_max = front_len;
31b8006e
SW		 3103 if (front_len) {
3104 if (front_len > PAGE_CACHE_SIZE) {
34d23762 3105 m->front.iov_base = __vmalloc(front_len, flags,
31b8006e
SW		 3106 PAGE_KERNEL);
3107 m->front_is_vmalloc = true;
3108 } else {
34d23762 3109 m->front.iov_base = kmalloc(front_len, flags);
31b8006e
SW		 3110 }
3111 if (m->front.iov_base == NULL) {
b61c2763 3112 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3113 front_len);
3114 goto out2;
3115 }
3116 } else {
3117 m->front.iov_base = NULL;
3118 }
3119 m->front.iov_len = front_len;
3120
bb257664 3121 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3122 return m;
3123
3124out2:
3125 ceph_msg_put(m);
3126out:
b61c2763
SW		 3127 if (!can_fail) {
3128 pr_err("msg_new can't create type %d front %d\n", type,
3129 front_len);
f0ed1b7c 3130 WARN_ON(1);
b61c2763
SW		 3131 } else {
3132 dout("msg_new can't create type %d front %d\n", type,
3133 front_len);
3134 }
a79832f2 3135 return NULL;
31b8006e 3136}
3d14c5d2 3137EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3138
31b8006e
SW		 3139/*
3140 * Allocate "middle" portion of a message, if it is needed and wasn't
3141 * allocated by alloc_msg. This allows us to read a small fixed-size
3142 * per-type header in the front and then gracefully fail (i.e.,
3143 * propagate the error to the caller based on info in the front) when
3144 * the middle is too large.
3145 */
2450418c 3146static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3147{
3148 int type = le16_to_cpu(msg->hdr.type);
3149 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3150
3151 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3152 ceph_msg_type_name(type), middle_len);
3153 BUG_ON(!middle_len);
3154 BUG_ON(msg->middle);
3155
b6c1d5b8 3156 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3157 if (!msg->middle)
3158 return -ENOMEM;
3159 return 0;
3160}
3161
2450418c 3162/*
1c20f2d2
AE		 3163 * Allocate a message for receiving an incoming message on a
3164 * connection, and save the result in con->in_msg. Uses the
3165 * connection's private alloc_msg op if available.
3166 *
4740a623
SW		 3167 * Returns 0 on success, or a negative error code.
3168 *
3169 * On success, if we set *skip = 1:
3170 * - the next message should be skipped and ignored.
3171 * - con->in_msg == NULL
3172 * or if we set *skip = 0:
3173 * - con->in_msg is non-null.
3174 * On error (ENOMEM, EAGAIN, ...),
3175 * - con->in_msg == NULL
2450418c 3176 */
4740a623 3177static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3178{
4740a623 3179 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3180 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3181 struct ceph_msg *msg;
4740a623 3182 int ret = 0;
2450418c 3183
1c20f2d2 3184 BUG_ON(con->in_msg != NULL);
53ded495 3185 BUG_ON(!con->ops->alloc_msg);
2450418c 3186
53ded495
AE		 3187 mutex_unlock(&con->mutex);
3188 msg = con->ops->alloc_msg(con, hdr, skip);
3189 mutex_lock(&con->mutex);
3190 if (con->state != CON_STATE_OPEN) {
3191 if (msg)
1d866d1c 3192 ceph_msg_put(msg);
53ded495
AE		 3193 return -EAGAIN;
3194 }
4137577a
AE		 3195 if (msg) {
3196 BUG_ON(*skip);
3197 con->in_msg = msg;
36eb71aa 3198 con->in_msg->con = con->ops->get(con);
92ce034b 3199 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3200 } else {
3201 /*
3202 * Null message pointer means either we should skip
3203 * this message or we couldn't allocate memory. The
3204 * former is not an error.
3205 */
3206 if (*skip)
3207 return 0;
3208 con->error_msg = "error allocating memory for incoming message";
3209
53ded495 3210 return -ENOMEM;
2450418c 3211 }
1c20f2d2 3212 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3213
1c20f2d2
AE		 3214 if (middle_len && !con->in_msg->middle) {
3215 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3216 if (ret < 0) {
1c20f2d2
AE		 3217 ceph_msg_put(con->in_msg);
3218 con->in_msg = NULL;
2450418c
YS		 3219 }
3220 }
9d7f0f13 3221
4740a623 3222 return ret;
2450418c
YS		 3223}
3224
31b8006e
SW		 3225
3226/*
3227 * Free a generically kmalloc'd message.
3228 */
3229void ceph_msg_kfree(struct ceph_msg *m)
3230{
3231 dout("msg_kfree %p\n", m);
3232 if (m->front_is_vmalloc)
3233 vfree(m->front.iov_base);
3234 else
3235 kfree(m->front.iov_base);
3236 kfree(m);
3237}
3238
3239/*
3240 * Drop a msg ref. Destroy as needed.
3241 */
c2e552e7
SW		 3242void ceph_msg_last_put(struct kref *kref)
3243{
3244 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
31b8006e 3245
c2e552e7
SW		 3246 dout("ceph_msg_put last one on %p\n", m);
3247 WARN_ON(!list_empty(&m->list_head));
3248
3249 /* drop middle, data, if any */
3250 if (m->middle) {
3251 ceph_buffer_put(m->middle);
3252 m->middle = NULL;
31b8006e 3253 }
97fb1c7f 3254 if (ceph_msg_has_pages(m)) {
f9e15777
AE		 3255 m->p.length = 0;
3256 m->p.pages = NULL;
888334f9 3257 m->p.type = CEPH_OSD_DATA_TYPE_NONE;
97fb1c7f 3258 }
97fb1c7f 3259 if (ceph_msg_has_pagelist(m)) {
f9e15777
AE		 3260 ceph_pagelist_release(m->l.pagelist);
3261 kfree(m->l.pagelist);
3262 m->l.pagelist = NULL;
888334f9
AE		 3263 m->l.type = CEPH_OSD_DATA_TYPE_NONE;
3264 }
3265 if (ceph_msg_has_bio(m)) {
3266 m->b.bio = NULL;
3267 m->b.type = CEPH_OSD_DATA_TYPE_NONE;
58bb3b37
SW		 3268 }
3269
c2e552e7
SW		 3270 if (m->pool)
3271 ceph_msgpool_put(m->pool, m);
3272 else
3273 ceph_msg_kfree(m);
31b8006e 3274}
3d14c5d2 3275EXPORT_SYMBOL(ceph_msg_last_put);
9ec7cab1
SW		 3276
3277void ceph_msg_dump(struct ceph_msg *msg)
3278{
4a73ef27 3279 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
f9e15777 3280 msg->front_max, msg->p.length);
9ec7cab1
SW		 3281 print_hex_dump(KERN_DEBUG, "header: ",
3282 DUMP_PREFIX_OFFSET, 16, 1,
3283 &msg->hdr, sizeof(msg->hdr), true);
3284 print_hex_dump(KERN_DEBUG, " front: ",
3285 DUMP_PREFIX_OFFSET, 16, 1,
3286 msg->front.iov_base, msg->front.iov_len, true);
3287 if (msg->middle)
3288 print_hex_dump(KERN_DEBUG, "middle: ",
3289 DUMP_PREFIX_OFFSET, 16, 1,
3290 msg->middle->vec.iov_base,
3291 msg->middle->vec.iov_len, true);
3292 print_hex_dump(KERN_DEBUG, "footer: ",
3293 DUMP_PREFIX_OFFSET, 16, 1,
3294 &msg->footer, sizeof(msg->footer), true);
3295}
3d14c5d2 3296EXPORT_SYMBOL(ceph_msg_dump);
kernel source for telekom puls (alcatel/mediatek)