projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
rbd: allocate image object names with a slab allocator
[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 */
8ae4f4f5 725static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 726 size_t length)
6aaa4511 727{
8ae4f4f5 728 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 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
ca8b3a69 737 cursor->resid = min(length, data->bio_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
8ae4f4f5 744static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 745 size_t *page_offset,
746 size_t *length)
747{
8ae4f4f5 748 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 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);
25aff7c5 769 BUG_ON(*length > cursor->resid);
5df521b1 770 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511
AE		 771
772 return bio_vec->bv_page;
773}
774
8ae4f4f5
AE		 775static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
776 size_t bytes)
6aaa4511 777{
6aaa4511
AE		 778 struct bio *bio;
779 struct bio_vec *bio_vec;
780 unsigned int index;
781
8ae4f4f5 782 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
6aaa4511
AE		 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}
ea96571f 820#endif /* CONFIG_BLOCK */
df6ad1f9 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 */
8ae4f4f5 826static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 827 size_t length)
e766d7b5 828{
8ae4f4f5 829 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 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);
836
ca8b3a69 837 cursor->resid = min(length, data->length);
e766d7b5 838 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 839 cursor->page_offset = data->alignment & ~PAGE_MASK;
840 cursor->page_index = 0;
56fc5659
AE		 841 BUG_ON(page_count > (int)USHRT_MAX);
842 cursor->page_count = (unsigned short)page_count;
843 BUG_ON(length > SIZE_MAX - cursor->page_offset);
844 cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE;
e766d7b5
AE		 845}
846
8ae4f4f5
AE		 847static struct page *
848ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
849 size_t *page_offset, size_t *length)
e766d7b5 850{
8ae4f4f5 851 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 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
8ae4f4f5 867static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 868 size_t bytes)
869{
8ae4f4f5 870 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 871
872 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 873
874 /* Advance the cursor page offset */
875
876 cursor->resid -= bytes;
5df521b1
AE		 877 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
878 if (!bytes || cursor->page_offset)
e766d7b5
AE		 879 return false; /* more bytes to process in the current page */
880
5df521b1 881 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 882
883 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 884 cursor->page_index++;
25aff7c5 885 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 886
887 return true;
888}
889
fe38a2b6 890/*
dd236fcb
AE		 891 * For a pagelist, a piece is whatever remains to be consumed in the
892 * first page in the list, or the front of the next page.
fe38a2b6 893 */
8ae4f4f5
AE		 894static void
895ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 896 size_t length)
fe38a2b6 897{
8ae4f4f5 898 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 899 struct ceph_pagelist *pagelist;
900 struct page *page;
901
dd236fcb 902 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 903
904 pagelist = data->pagelist;
905 BUG_ON(!pagelist);
25aff7c5
AE		 906
907 if (!length)
fe38a2b6
AE		 908 return; /* pagelist can be assigned but empty */
909
910 BUG_ON(list_empty(&pagelist->head));
911 page = list_first_entry(&pagelist->head, struct page, lru);
912
ca8b3a69 913 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 914 cursor->page = page;
915 cursor->offset = 0;
a51b272e 916 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 917}
918
8ae4f4f5
AE		 919static struct page *
920ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
921 size_t *page_offset, size_t *length)
fe38a2b6 922{
8ae4f4f5 923 struct ceph_msg_data *data = cursor->data;
fe38a2b6 924 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 925
926 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
927
928 pagelist = data->pagelist;
929 BUG_ON(!pagelist);
930
931 BUG_ON(!cursor->page);
25aff7c5 932 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 933
5df521b1 934 /* offset of first page in pagelist is always 0 */
fe38a2b6 935 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 936 if (cursor->last_piece)
25aff7c5
AE		 937 *length = cursor->resid;
938 else
939 *length = PAGE_SIZE - *page_offset;
fe38a2b6 940
8ae4f4f5 941 return cursor->page;
fe38a2b6
AE		 942}
943
8ae4f4f5 944static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 945 size_t bytes)
fe38a2b6 946{
8ae4f4f5 947 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 948 struct ceph_pagelist *pagelist;
949
950 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
951
952 pagelist = data->pagelist;
953 BUG_ON(!pagelist);
25aff7c5
AE		 954
955 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 956 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
957
958 /* Advance the cursor offset */
959
25aff7c5 960 cursor->resid -= bytes;
fe38a2b6 961 cursor->offset += bytes;
5df521b1 962 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 963 if (!bytes || cursor->offset & ~PAGE_MASK)
964 return false; /* more bytes to process in the current page */
965
966 /* Move on to the next page */
967
968 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
969 cursor->page = list_entry_next(cursor->page, lru);
25aff7c5 970 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 971
972 return true;
973}
974
dd236fcb
AE		 975/*
976 * Message data is handled (sent or received) in pieces, where each
977 * piece resides on a single page. The network layer might not
978 * consume an entire piece at once. A data item's cursor keeps
979 * track of which piece is next to process and how much remains to
980 * be processed in that piece. It also tracks whether the current
981 * piece is the last one in the data item.
982 */
ca8b3a69 983static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 984{
ca8b3a69 985 size_t length = cursor->total_resid;
8ae4f4f5 986
8ae4f4f5 987 switch (cursor->data->type) {
dd236fcb 988 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 989 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 990 break;
e766d7b5 991 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 992 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 993 break;
dd236fcb
AE		 994#ifdef CONFIG_BLOCK
995 case CEPH_MSG_DATA_BIO:
8ae4f4f5 996 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 997 break;
dd236fcb 998#endif /* CONFIG_BLOCK */
6aaa4511 999 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1000 default:
1001 /* BUG(); */
1002 break;
1003 }
8ae4f4f5 1004 cursor->need_crc = true;
dd236fcb
AE		 1005}
1006
ca8b3a69
AE		 1007static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1008{
1009 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1010 struct ceph_msg_data *data;
1011
1012 BUG_ON(!length);
1013 BUG_ON(length > msg->data_length);
1014 BUG_ON(list_empty(&msg->data));
1015
ca8b3a69
AE		 1016 cursor->data_head = &msg->data;
1017 cursor->total_resid = length;
1018 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1019 cursor->data = data;
1020
1021 __ceph_msg_data_cursor_init(cursor);
1022}
1023
dd236fcb
AE		 1024/*
1025 * Return the page containing the next piece to process for a given
1026 * data item, and supply the page offset and length of that piece.
1027 * Indicate whether this is the last piece in this data item.
1028 */
8ae4f4f5
AE		 1029static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1030 size_t *page_offset, size_t *length,
dd236fcb
AE		 1031 bool *last_piece)
1032{
1033 struct page *page;
1034
8ae4f4f5 1035 switch (cursor->data->type) {
dd236fcb 1036 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1037 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1038 break;
e766d7b5 1039 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1040 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1041 break;
dd236fcb
AE		 1042#ifdef CONFIG_BLOCK
1043 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1044 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1045 break;
dd236fcb 1046#endif /* CONFIG_BLOCK */
6aaa4511 1047 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1048 default:
1049 page = NULL;
1050 break;
1051 }
1052 BUG_ON(!page);
1053 BUG_ON(*page_offset + *length > PAGE_SIZE);
1054 BUG_ON(!*length);
1055 if (last_piece)
8ae4f4f5 1056 *last_piece = cursor->last_piece;
dd236fcb
AE		 1057
1058 return page;
1059}
1060
1061/*
1062 * Returns true if the result moves the cursor on to the next piece
1063 * of the data item.
1064 */
8ae4f4f5
AE		 1065static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1066 size_t bytes)
dd236fcb
AE		 1067{
1068 bool new_piece;
1069
25aff7c5 1070 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1071 switch (cursor->data->type) {
dd236fcb 1072 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1073 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1074 break;
e766d7b5 1075 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1076 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1077 break;
dd236fcb
AE		 1078#ifdef CONFIG_BLOCK
1079 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1080 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1081 break;
dd236fcb 1082#endif /* CONFIG_BLOCK */
6aaa4511 1083 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1084 default:
1085 BUG();
1086 break;
1087 }
ca8b3a69 1088 cursor->total_resid -= bytes;
dd236fcb 1089
ca8b3a69
AE		 1090 if (!cursor->resid && cursor->total_resid) {
1091 WARN_ON(!cursor->last_piece);
1092 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1093 cursor->data = list_entry_next(cursor->data, links);
1094 __ceph_msg_data_cursor_init(cursor);
a51b272e 1095 new_piece = true;
ca8b3a69 1096 }
a51b272e 1097 cursor->need_crc = new_piece;
ca8b3a69 1098
dd236fcb
AE		 1099 return new_piece;
1100}
1101
98fa5dd8 1102static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1103{
739c905b 1104 BUG_ON(!msg);
25aff7c5 1105 BUG_ON(!data_len);
739c905b 1106
4c59b4a2 1107 /* Initialize data cursor */
fe38a2b6 1108
8ae4f4f5 1109 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1110}
1111
31b8006e
SW		 1112/*
1113 * Prepare footer for currently outgoing message, and finish things
1114 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1115 */
859eb799 1116static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1117{
1118 struct ceph_msg *m = con->out_msg;
859eb799 1119 int v = con->out_kvec_left;
31b8006e 1120
fd154f3c
AE		 1121 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1122
31b8006e
SW		 1123 dout("prepare_write_message_footer %p\n", con);
1124 con->out_kvec_is_msg = true;
1125 con->out_kvec[v].iov_base = &m->footer;
1126 con->out_kvec[v].iov_len = sizeof(m->footer);
1127 con->out_kvec_bytes += sizeof(m->footer);
1128 con->out_kvec_left++;
1129 con->out_more = m->more_to_follow;
c86a2930 1130 con->out_msg_done = true;
31b8006e
SW		 1131}
1132
1133/*
1134 * Prepare headers for the next outgoing message.
1135 */
1136static void prepare_write_message(struct ceph_connection *con)
1137{
1138 struct ceph_msg *m;
a9a0c51a 1139 u32 crc;
31b8006e 1140
e2200423 1141 con_out_kvec_reset(con);
31b8006e 1142 con->out_kvec_is_msg = true;
c86a2930 1143 con->out_msg_done = false;
31b8006e
SW		 1144
1145 /* Sneak an ack in there first? If we can get it into the same
1146 * TCP packet that's a good thing. */
1147 if (con->in_seq > con->in_seq_acked) {
1148 con->in_seq_acked = con->in_seq;
e2200423 1149 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1150 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1151 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1152 &con->out_temp_ack);
31b8006e
SW		 1153 }
1154
38941f80 1155 BUG_ON(list_empty(&con->out_queue));
859eb799 1156 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1157 con->out_msg = m;
38941f80 1158 BUG_ON(m->con != con);
4cf9d544
SW		 1159
1160 /* put message on sent list */
1161 ceph_msg_get(m);
1162 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1163
e84346b7
SW		 1164 /*
1165 * only assign outgoing seq # if we haven't sent this message
1166 * yet. if it is requeued, resend with it's original seq.
1167 */
1168 if (m->needs_out_seq) {
1169 m->hdr.seq = cpu_to_le64(++con->out_seq);
1170 m->needs_out_seq = false;
1171 }
98fa5dd8 1172 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1173
98fa5dd8 1174 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1175 m, con->out_seq, le16_to_cpu(m->hdr.type),
1176 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1177 m->data_length);
31b8006e
SW		 1178 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1179
1180 /* tag + hdr + front + middle */
e2200423
AE		 1181 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1182 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1183 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1184
31b8006e 1185 if (m->middle)
e2200423 1186 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1187 m->middle->vec.iov_base);
31b8006e
SW		 1188
1189 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1190 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1191 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1192 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1193
1194 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1195 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1196 if (m->middle) {
1197 crc = crc32c(0, m->middle->vec.iov_base,
1198 m->middle->vec.iov_len);
1199 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1200 } else
31b8006e 1201 con->out_msg->footer.middle_crc = 0;
739c905b 1202 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1203 le32_to_cpu(con->out_msg->footer.front_crc),
1204 le32_to_cpu(con->out_msg->footer.middle_crc));
1205
1206 /* is there a data payload? */
739c905b 1207 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1208 if (m->data_length) {
1209 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1210 con->out_more = 1; /* data + footer will follow */
1211 } else {
31b8006e 1212 /* no, queue up footer too and be done */
859eb799 1213 prepare_write_message_footer(con);
78625051 1214 }
31b8006e 1215
c9ffc77a 1216 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1217}
1218
1219/*
1220 * Prepare an ack.
1221 */
1222static void prepare_write_ack(struct ceph_connection *con)
1223{
1224 dout("prepare_write_ack %p %llu -> %llu\n", con,
1225 con->in_seq_acked, con->in_seq);
1226 con->in_seq_acked = con->in_seq;
1227
e2200423 1228 con_out_kvec_reset(con);
859eb799 1229
e2200423 1230 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1231
31b8006e 1232 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1233 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1234 &con->out_temp_ack);
1235
31b8006e 1236 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1237 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1238}
1239
3a23083b
SW		 1240/*
1241 * Prepare to share the seq during handshake
1242 */
1243static void prepare_write_seq(struct ceph_connection *con)
1244{
1245 dout("prepare_write_seq %p %llu -> %llu\n", con,
1246 con->in_seq_acked, con->in_seq);
1247 con->in_seq_acked = con->in_seq;
1248
1249 con_out_kvec_reset(con);
1250
1251 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1252 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1253 &con->out_temp_ack);
1254
1255 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1256}
1257
31b8006e
SW		 1258/*
1259 * Prepare to write keepalive byte.
1260 */
1261static void prepare_write_keepalive(struct ceph_connection *con)
1262{
1263 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1264 con_out_kvec_reset(con);
1265 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1266 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1267}
1268
1269/*
1270 * Connection negotiation.
1271 */
1272
dac1e716
AE		 1273static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1274 int *auth_proto)
4e7a5dcd 1275{
a3530df3 1276 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1277
1278 if (!con->ops->get_authorizer) {
1279 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1280 con->out_connect.authorizer_len = 0;
729796be 1281 return NULL;
b1c6b980
AE		 1282 }
1283
1284 /* Can't hold the mutex while getting authorizer */
ec302645 1285 mutex_unlock(&con->mutex);
dac1e716 1286 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1287 mutex_lock(&con->mutex);
4e7a5dcd 1288
a3530df3 1289 if (IS_ERR(auth))
729796be 1290 return auth;
8dacc7da 1291 if (con->state != CON_STATE_NEGOTIATING)
729796be 1292 return ERR_PTR(-EAGAIN);
0da5d703 1293
8f43fb53
AE		 1294 con->auth_reply_buf = auth->authorizer_reply_buf;
1295 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1296 return auth;
4e7a5dcd
SW		 1297}
1298
31b8006e
SW		 1299/*
1300 * We connected to a peer and are saying hello.
1301 */
e825a66d 1302static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1303{
e2200423
AE		 1304 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1305 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1306 &con->msgr->my_enc_addr);
eed0ef2c 1307
eed0ef2c 1308 con->out_more = 0;
c9ffc77a 1309 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1310}
1311
e825a66d 1312static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1313{
95c96174 1314 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1315 int proto;
dac1e716 1316 int auth_proto;
729796be 1317 struct ceph_auth_handshake *auth;
31b8006e
SW		 1318
1319 switch (con->peer_name.type) {
1320 case CEPH_ENTITY_TYPE_MON:
1321 proto = CEPH_MONC_PROTOCOL;
1322 break;
1323 case CEPH_ENTITY_TYPE_OSD:
1324 proto = CEPH_OSDC_PROTOCOL;
1325 break;
1326 case CEPH_ENTITY_TYPE_MDS:
1327 proto = CEPH_MDSC_PROTOCOL;
1328 break;
1329 default:
1330 BUG();
1331 }
1332
1333 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1334 con->connect_seq, global_seq, proto);
4e7a5dcd 1335
e825a66d 1336 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1337 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1338 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1339 con->out_connect.global_seq = cpu_to_le32(global_seq);
1340 con->out_connect.protocol_version = cpu_to_le32(proto);
1341 con->out_connect.flags = 0;
31b8006e 1342
dac1e716
AE		 1343 auth_proto = CEPH_AUTH_UNKNOWN;
1344 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1345 if (IS_ERR(auth))
1346 return PTR_ERR(auth);
3da54776 1347
dac1e716 1348 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1349 con->out_connect.authorizer_len = auth ?
1350 cpu_to_le32(auth->authorizer_buf_len) : 0;
1351
e2200423 1352 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1353 &con->out_connect);
1354 if (auth && auth->authorizer_buf_len)
e2200423 1355 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1356 auth->authorizer_buf);
859eb799 1357
31b8006e 1358 con->out_more = 0;
c9ffc77a 1359 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1360
e10c758e 1361 return 0;
31b8006e
SW		 1362}
1363
31b8006e
SW		 1364/*
1365 * write as much of pending kvecs to the socket as we can.
1366 * 1 -> done
1367 * 0 -> socket full, but more to do
1368 * <0 -> error
1369 */
1370static int write_partial_kvec(struct ceph_connection *con)
1371{
1372 int ret;
1373
1374 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1375 while (con->out_kvec_bytes > 0) {
1376 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1377 con->out_kvec_left, con->out_kvec_bytes,
1378 con->out_more);
1379 if (ret <= 0)
1380 goto out;
1381 con->out_kvec_bytes -= ret;
1382 if (con->out_kvec_bytes == 0)
1383 break; /* done */
f42299e6
AE		 1384
1385 /* account for full iov entries consumed */
1386 while (ret >= con->out_kvec_cur->iov_len) {
1387 BUG_ON(!con->out_kvec_left);
1388 ret -= con->out_kvec_cur->iov_len;
1389 con->out_kvec_cur++;
1390 con->out_kvec_left--;
1391 }
1392 /* and for a partially-consumed entry */
1393 if (ret) {
1394 con->out_kvec_cur->iov_len -= ret;
1395 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1396 }
1397 }
1398 con->out_kvec_left = 0;
1399 con->out_kvec_is_msg = false;
1400 ret = 1;
1401out:
1402 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1403 con->out_kvec_bytes, con->out_kvec_left, ret);
1404 return ret; /* done! */
1405}
1406
35b62808
AE		 1407static u32 ceph_crc32c_page(u32 crc, struct page *page,
1408 unsigned int page_offset,
1409 unsigned int length)
1410{
1411 char *kaddr;
1412
1413 kaddr = kmap(page);
1414 BUG_ON(kaddr == NULL);
1415 crc = crc32c(crc, kaddr + page_offset, length);
1416 kunmap(page);
1417
1418 return crc;
1419}
31b8006e
SW		 1420/*
1421 * Write as much message data payload as we can. If we finish, queue
1422 * up the footer.
1423 * 1 -> done, footer is now queued in out_kvec[].
1424 * 0 -> socket full, but more to do
1425 * <0 -> error
1426 */
34d2d200 1427static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1428{
1429 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1430 struct ceph_msg_data_cursor *cursor = &msg->cursor;
37675b0f 1431 bool do_datacrc = !con->msgr->nocrc;
f5db90bc 1432 u32 crc;
31b8006e 1433
859a35d5 1434 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1435
5240d9f9 1436 if (list_empty(&msg->data))
4c59b4a2
AE		 1437 return -EINVAL;
1438
5821bd8c
AE		 1439 /*
1440 * Iterate through each page that contains data to be
1441 * written, and send as much as possible for each.
1442 *
1443 * If we are calculating the data crc (the default), we will
1444 * need to map the page. If we have no pages, they have
1445 * been revoked, so use the zero page.
1446 */
f5db90bc 1447 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1448 while (cursor->resid) {
8a166d05 1449 struct page *page;
e387d525
AE		 1450 size_t page_offset;
1451 size_t length;
8a166d05 1452 bool last_piece;
8ea299bc 1453 bool need_crc;
f5db90bc 1454 int ret;
68b4476b 1455
8ae4f4f5 1456 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 1457 &last_piece);
e387d525 1458 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1459 length, last_piece);
f5db90bc
AE		 1460 if (ret <= 0) {
1461 if (do_datacrc)
1462 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1463
f5db90bc
AE		 1464 return ret;
1465 }
143334ff
AE		 1466 if (do_datacrc && cursor->need_crc)
1467 crc = ceph_crc32c_page(crc, page, page_offset, length);
8ae4f4f5 1468 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
31b8006e
SW		 1469 }
1470
34d2d200 1471 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1472
1473 /* prepare and queue up footer, too */
f5db90bc
AE		 1474 if (do_datacrc)
1475 msg->footer.data_crc = cpu_to_le32(crc);
1476 else
84ca8fc8 1477 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1478 con_out_kvec_reset(con);
859eb799 1479 prepare_write_message_footer(con);
f5db90bc
AE		 1480
1481 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1482}
1483
1484/*
1485 * write some zeros
1486 */
1487static int write_partial_skip(struct ceph_connection *con)
1488{
1489 int ret;
1490
1491 while (con->out_skip > 0) {
31739139 1492 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1493
e1dcb128 1494 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1495 if (ret <= 0)
1496 goto out;
1497 con->out_skip -= ret;
1498 }
1499 ret = 1;
1500out:
1501 return ret;
1502}
1503
1504/*
1505 * Prepare to read connection handshake, or an ack.
1506 */
eed0ef2c
SW		 1507static void prepare_read_banner(struct ceph_connection *con)
1508{
1509 dout("prepare_read_banner %p\n", con);
1510 con->in_base_pos = 0;
1511}
1512
31b8006e
SW		 1513static void prepare_read_connect(struct ceph_connection *con)
1514{
1515 dout("prepare_read_connect %p\n", con);
1516 con->in_base_pos = 0;
1517}
1518
1519static void prepare_read_ack(struct ceph_connection *con)
1520{
1521 dout("prepare_read_ack %p\n", con);
1522 con->in_base_pos = 0;
1523}
1524
3a23083b
SW		 1525static void prepare_read_seq(struct ceph_connection *con)
1526{
1527 dout("prepare_read_seq %p\n", con);
1528 con->in_base_pos = 0;
1529 con->in_tag = CEPH_MSGR_TAG_SEQ;
1530}
1531
31b8006e
SW		 1532static void prepare_read_tag(struct ceph_connection *con)
1533{
1534 dout("prepare_read_tag %p\n", con);
1535 con->in_base_pos = 0;
1536 con->in_tag = CEPH_MSGR_TAG_READY;
1537}
1538
1539/*
1540 * Prepare to read a message.
1541 */
1542static int prepare_read_message(struct ceph_connection *con)
1543{
1544 dout("prepare_read_message %p\n", con);
1545 BUG_ON(con->in_msg != NULL);
1546 con->in_base_pos = 0;
1547 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1548 return 0;
1549}
1550
1551
1552static int read_partial(struct ceph_connection *con,
fd51653f 1553 int end, int size, void *object)
31b8006e 1554{
e6cee71f
AE		 1555 while (con->in_base_pos < end) {
1556 int left = end - con->in_base_pos;
31b8006e
SW		 1557 int have = size - left;
1558 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1559 if (ret <= 0)
1560 return ret;
1561 con->in_base_pos += ret;
1562 }
1563 return 1;
1564}
1565
1566
1567/*
1568 * Read all or part of the connect-side handshake on a new connection
1569 */
eed0ef2c 1570static int read_partial_banner(struct ceph_connection *con)
31b8006e 1571{
fd51653f
AE		 1572 int size;
1573 int end;
1574 int ret;
31b8006e 1575
eed0ef2c 1576 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1577
1578 /* peer's banner */
fd51653f
AE		 1579 size = strlen(CEPH_BANNER);
1580 end = size;
1581 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1582 if (ret <= 0)
1583 goto out;
fd51653f
AE		 1584
1585 size = sizeof (con->actual_peer_addr);
1586 end += size;
1587 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1588 if (ret <= 0)
1589 goto out;
fd51653f
AE		 1590
1591 size = sizeof (con->peer_addr_for_me);
1592 end += size;
1593 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1594 if (ret <= 0)
1595 goto out;
fd51653f 1596
eed0ef2c
SW		 1597out:
1598 return ret;
1599}
1600
1601static int read_partial_connect(struct ceph_connection *con)
1602{
fd51653f
AE		 1603 int size;
1604 int end;
1605 int ret;
eed0ef2c
SW		 1606
1607 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1608
fd51653f
AE		 1609 size = sizeof (con->in_reply);
1610 end = size;
1611 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1612 if (ret <= 0)
1613 goto out;
fd51653f
AE		 1614
1615 size = le32_to_cpu(con->in_reply.authorizer_len);
1616 end += size;
1617 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1618 if (ret <= 0)
1619 goto out;
31b8006e 1620
4e7a5dcd
SW		 1621 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1622 con, (int)con->in_reply.tag,
1623 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1624 le32_to_cpu(con->in_reply.global_seq));
1625out:
1626 return ret;
eed0ef2c 1627
31b8006e
SW		 1628}
1629
1630/*
1631 * Verify the hello banner looks okay.
1632 */
1633static int verify_hello(struct ceph_connection *con)
1634{
1635 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1636 pr_err("connect to %s got bad banner\n",
3d14c5d2 1637 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1638 con->error_msg = "protocol error, bad banner";
1639 return -1;
1640 }
1641 return 0;
1642}
1643
1644static bool addr_is_blank(struct sockaddr_storage *ss)
1645{
1646 switch (ss->ss_family) {
1647 case AF_INET:
1648 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1649 case AF_INET6:
1650 return
1651 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1652 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1653 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1654 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1655 }
1656 return false;
1657}
1658
1659static int addr_port(struct sockaddr_storage *ss)
1660{
1661 switch (ss->ss_family) {
1662 case AF_INET:
f28bcfbe 1663 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1664 case AF_INET6:
f28bcfbe 1665 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1666 }
1667 return 0;
1668}
1669
1670static void addr_set_port(struct sockaddr_storage *ss, int p)
1671{
1672 switch (ss->ss_family) {
1673 case AF_INET:
1674 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1675 break;
31b8006e
SW		 1676 case AF_INET6:
1677 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1678 break;
31b8006e
SW		 1679 }
1680}
1681
ee3b56f2
NW		 1682/*
1683 * Unlike other *_pton function semantics, zero indicates success.
1684 */
1685static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1686 char delim, const char **ipend)
1687{
99f0f3b2
AE		 1688 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1689 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1690
1691 memset(ss, 0, sizeof(*ss));
1692
1693 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1694 ss->ss_family = AF_INET;
1695 return 0;
1696 }
1697
1698 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1699 ss->ss_family = AF_INET6;
1700 return 0;
1701 }
1702
1703 return -EINVAL;
1704}
1705
1706/*
1707 * Extract hostname string and resolve using kernel DNS facility.
1708 */
1709#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1710static int ceph_dns_resolve_name(const char *name, size_t namelen,
1711 struct sockaddr_storage *ss, char delim, const char **ipend)
1712{
1713 const char *end, *delim_p;
1714 char *colon_p, *ip_addr = NULL;
1715 int ip_len, ret;
1716
1717 /*
1718 * The end of the hostname occurs immediately preceding the delimiter or
1719 * the port marker (':') where the delimiter takes precedence.
1720 */
1721 delim_p = memchr(name, delim, namelen);
1722 colon_p = memchr(name, ':', namelen);
1723
1724 if (delim_p && colon_p)
1725 end = delim_p < colon_p ? delim_p : colon_p;
1726 else if (!delim_p && colon_p)
1727 end = colon_p;
1728 else {
1729 end = delim_p;
1730 if (!end) /* case: hostname:/ */
1731 end = name + namelen;
1732 }
1733
1734 if (end <= name)
1735 return -EINVAL;
1736
1737 /* do dns_resolve upcall */
1738 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1739 if (ip_len > 0)
1740 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1741 else
1742 ret = -ESRCH;
1743
1744 kfree(ip_addr);
1745
1746 *ipend = end;
1747
1748 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1749 ret, ret ? "failed" : ceph_pr_addr(ss));
1750
1751 return ret;
1752}
1753#else
1754static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1755 struct sockaddr_storage *ss, char delim, const char **ipend)
1756{
1757 return -EINVAL;
1758}
1759#endif
1760
1761/*
1762 * Parse a server name (IP or hostname). If a valid IP address is not found
1763 * then try to extract a hostname to resolve using userspace DNS upcall.
1764 */
1765static int ceph_parse_server_name(const char *name, size_t namelen,
1766 struct sockaddr_storage *ss, char delim, const char **ipend)
1767{
1768 int ret;
1769
1770 ret = ceph_pton(name, namelen, ss, delim, ipend);
1771 if (ret)
1772 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1773
1774 return ret;
1775}
1776
31b8006e
SW		 1777/*
1778 * Parse an ip[:port] list into an addr array. Use the default
1779 * monitor port if a port isn't specified.
1780 */
1781int ceph_parse_ips(const char *c, const char *end,
1782 struct ceph_entity_addr *addr,
1783 int max_count, int *count)
1784{
ee3b56f2 1785 int i, ret = -EINVAL;
31b8006e
SW		 1786 const char *p = c;
1787
1788 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1789 for (i = 0; i < max_count; i++) {
1790 const char *ipend;
1791 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1792 int port;
39139f64
SW		 1793 char delim = ',';
1794
1795 if (*p == '[') {
1796 delim = ']';
1797 p++;
1798 }
31b8006e 1799
ee3b56f2
NW		 1800 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1801 if (ret)
31b8006e 1802 goto bad;
ee3b56f2
NW		 1803 ret = -EINVAL;
1804
31b8006e
SW		 1805 p = ipend;
1806
39139f64
SW		 1807 if (delim == ']') {
1808 if (*p != ']') {
1809 dout("missing matching ']'\n");
1810 goto bad;
1811 }
1812 p++;
1813 }
1814
31b8006e
SW		 1815 /* port? */
1816 if (p < end && *p == ':') {
1817 port = 0;
1818 p++;
1819 while (p < end && *p >= '0' && *p <= '9') {
1820 port = (port * 10) + (*p - '0');
1821 p++;
1822 }
1823 if (port > 65535 || port == 0)
1824 goto bad;
1825 } else {
1826 port = CEPH_MON_PORT;
1827 }
1828
1829 addr_set_port(ss, port);
1830
3d14c5d2 1831 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1832
1833 if (p == end)
1834 break;
1835 if (*p != ',')
1836 goto bad;
1837 p++;
1838 }
1839
1840 if (p != end)
1841 goto bad;
1842
1843 if (count)
1844 *count = i + 1;
1845 return 0;
1846
1847bad:
39139f64 1848 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1849 return ret;
31b8006e 1850}
3d14c5d2 1851EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1852
eed0ef2c 1853static int process_banner(struct ceph_connection *con)
31b8006e 1854{
eed0ef2c 1855 dout("process_banner on %p\n", con);
31b8006e
SW		 1856
1857 if (verify_hello(con) < 0)
1858 return -1;
1859
63f2d211
SW		 1860 ceph_decode_addr(&con->actual_peer_addr);
1861 ceph_decode_addr(&con->peer_addr_for_me);
1862
31b8006e
SW		 1863 /*
1864 * Make sure the other end is who we wanted. note that the other
1865 * end may not yet know their ip address, so if it's 0.0.0.0, give
1866 * them the benefit of the doubt.
1867 */
103e2d3a
SW		 1868 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1869 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1870 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1871 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
cd84db6e 1872 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
3d14c5d2 1873 ceph_pr_addr(&con->peer_addr.in_addr),
cd84db6e 1874 (int)le32_to_cpu(con->peer_addr.nonce),
3d14c5d2 1875 ceph_pr_addr(&con->actual_peer_addr.in_addr),
cd84db6e 1876 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1877 con->error_msg = "wrong peer at address";
31b8006e
SW		 1878 return -1;
1879 }
1880
1881 /*
1882 * did we learn our address?
1883 */
1884 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1885 int port = addr_port(&con->msgr->inst.addr.in_addr);
1886
1887 memcpy(&con->msgr->inst.addr.in_addr,
1888 &con->peer_addr_for_me.in_addr,
1889 sizeof(con->peer_addr_for_me.in_addr));
1890 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1891 encode_my_addr(con->msgr);
eed0ef2c 1892 dout("process_banner learned my addr is %s\n",
3d14c5d2 1893 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1894 }
1895
eed0ef2c
SW		 1896 return 0;
1897}
1898
1899static int process_connect(struct ceph_connection *con)
1900{
3d14c5d2
YS		 1901 u64 sup_feat = con->msgr->supported_features;
1902 u64 req_feat = con->msgr->required_features;
04a419f9 1903 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 1904 int ret;
04a419f9 1905
eed0ef2c
SW		 1906 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1907
31b8006e 1908 switch (con->in_reply.tag) {
04a419f9
SW		 1909 case CEPH_MSGR_TAG_FEATURES:
1910 pr_err("%s%lld %s feature set mismatch,"
1911 " my %llx < server's %llx, missing %llx\n",
1912 ENTITY_NAME(con->peer_name),
3d14c5d2 1913 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1914 sup_feat, server_feat, server_feat & ~sup_feat);
1915 con->error_msg = "missing required protocol features";
0fa6ebc6 1916 reset_connection(con);
04a419f9
SW		 1917 return -1;
1918
31b8006e 1919 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 1920 pr_err("%s%lld %s protocol version mismatch,"
1921 " my %d != server's %d\n",
1922 ENTITY_NAME(con->peer_name),
3d14c5d2 1923 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 1924 le32_to_cpu(con->out_connect.protocol_version),
1925 le32_to_cpu(con->in_reply.protocol_version));
1926 con->error_msg = "protocol version mismatch";
0fa6ebc6 1927 reset_connection(con);
31b8006e
SW		 1928 return -1;
1929
4e7a5dcd
SW		 1930 case CEPH_MSGR_TAG_BADAUTHORIZER:
1931 con->auth_retry++;
1932 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1933 con->auth_retry);
1934 if (con->auth_retry == 2) {
1935 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 1936 return -1;
1937 }
6d4221b5 1938 con_out_kvec_reset(con);
e825a66d 1939 ret = prepare_write_connect(con);
0da5d703
SW		 1940 if (ret < 0)
1941 return ret;
63733a0f 1942 prepare_read_connect(con);
4e7a5dcd 1943 break;
31b8006e
SW		 1944
1945 case CEPH_MSGR_TAG_RESETSESSION:
1946 /*
1947 * If we connected with a large connect_seq but the peer
1948 * has no record of a session with us (no connection, or
1949 * connect_seq == 0), they will send RESETSESION to indicate
1950 * that they must have reset their session, and may have
1951 * dropped messages.
1952 */
1953 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 1954 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 1955 pr_err("%s%lld %s connection reset\n",
1956 ENTITY_NAME(con->peer_name),
3d14c5d2 1957 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 1958 reset_connection(con);
6d4221b5 1959 con_out_kvec_reset(con);
5a0f8fdd
AE		 1960 ret = prepare_write_connect(con);
1961 if (ret < 0)
1962 return ret;
31b8006e
SW		 1963 prepare_read_connect(con);
1964
1965 /* Tell ceph about it. */
ec302645 1966 mutex_unlock(&con->mutex);
31b8006e
SW		 1967 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1968 if (con->ops->peer_reset)
1969 con->ops->peer_reset(con);
ec302645 1970 mutex_lock(&con->mutex);
8dacc7da 1971 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 1972 return -EAGAIN;
31b8006e
SW		 1973 break;
1974
1975 case CEPH_MSGR_TAG_RETRY_SESSION:
1976 /*
1977 * If we sent a smaller connect_seq than the peer has, try
1978 * again with a larger value.
1979 */
5bdca4e0 1980 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 1981 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 1982 le32_to_cpu(con->in_reply.connect_seq));
1983 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 1984 con_out_kvec_reset(con);
5a0f8fdd
AE		 1985 ret = prepare_write_connect(con);
1986 if (ret < 0)
1987 return ret;
31b8006e
SW		 1988 prepare_read_connect(con);
1989 break;
1990
1991 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1992 /*
1993 * If we sent a smaller global_seq than the peer has, try
1994 * again with a larger value.
1995 */
eed0ef2c 1996 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 1997 con->peer_global_seq,
5bdca4e0 1998 le32_to_cpu(con->in_reply.global_seq));
31b8006e 1999 get_global_seq(con->msgr,
5bdca4e0 2000 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2001 con_out_kvec_reset(con);
5a0f8fdd
AE		 2002 ret = prepare_write_connect(con);
2003 if (ret < 0)
2004 return ret;
31b8006e
SW		 2005 prepare_read_connect(con);
2006 break;
2007
3a23083b 2008 case CEPH_MSGR_TAG_SEQ:
31b8006e 2009 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2010 if (req_feat & ~server_feat) {
2011 pr_err("%s%lld %s protocol feature mismatch,"
2012 " my required %llx > server's %llx, need %llx\n",
2013 ENTITY_NAME(con->peer_name),
3d14c5d2 2014 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2015 req_feat, server_feat, req_feat & ~server_feat);
2016 con->error_msg = "missing required protocol features";
0fa6ebc6 2017 reset_connection(con);
04a419f9
SW		 2018 return -1;
2019 }
8dacc7da 2020
122070a2 2021 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2022 con->state = CON_STATE_OPEN;
20e55c4c 2023 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2024 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2025 con->connect_seq++;
aba558e2 2026 con->peer_features = server_feat;
31b8006e
SW		 2027 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2028 con->peer_global_seq,
2029 le32_to_cpu(con->in_reply.connect_seq),
2030 con->connect_seq);
2031 WARN_ON(con->connect_seq !=
2032 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2033
2034 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2035 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2036
85effe18 2037 con->delay = 0; /* reset backoff memory */
92ac41d0 2038
3a23083b
SW		 2039 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2040 prepare_write_seq(con);
2041 prepare_read_seq(con);
2042 } else {
2043 prepare_read_tag(con);
2044 }
31b8006e
SW		 2045 break;
2046
2047 case CEPH_MSGR_TAG_WAIT:
2048 /*
2049 * If there is a connection race (we are opening
2050 * connections to each other), one of us may just have
2051 * to WAIT. This shouldn't happen if we are the
2052 * client.
2053 */
04177882
SW		 2054 pr_err("process_connect got WAIT as client\n");
2055 con->error_msg = "protocol error, got WAIT as client";
2056 return -1;
31b8006e
SW		 2057
2058 default:
2059 pr_err("connect protocol error, will retry\n");
2060 con->error_msg = "protocol error, garbage tag during connect";
2061 return -1;
2062 }
2063 return 0;
2064}
2065
2066
2067/*
2068 * read (part of) an ack
2069 */
2070static int read_partial_ack(struct ceph_connection *con)
2071{
fd51653f
AE		 2072 int size = sizeof (con->in_temp_ack);
2073 int end = size;
31b8006e 2074
fd51653f 2075 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2076}
2077
31b8006e
SW		 2078/*
2079 * We can finally discard anything that's been acked.
2080 */
2081static void process_ack(struct ceph_connection *con)
2082{
2083 struct ceph_msg *m;
2084 u64 ack = le64_to_cpu(con->in_temp_ack);
2085 u64 seq;
2086
31b8006e
SW		 2087 while (!list_empty(&con->out_sent)) {
2088 m = list_first_entry(&con->out_sent, struct ceph_msg,
2089 list_head);
2090 seq = le64_to_cpu(m->hdr.seq);
2091 if (seq > ack)
2092 break;
2093 dout("got ack for seq %llu type %d at %p\n", seq,
2094 le16_to_cpu(m->hdr.type), m);
4cf9d544 2095 m->ack_stamp = jiffies;
31b8006e
SW		 2096 ceph_msg_remove(m);
2097 }
31b8006e
SW		 2098 prepare_read_tag(con);
2099}
2100
2101
2450418c 2102static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2103 struct kvec *section,
2104 unsigned int sec_len, u32 *crc)
2450418c 2105{
68b4476b 2106 int ret, left;
2450418c
YS		 2107
2108 BUG_ON(!section);
2109
2110 while (section->iov_len < sec_len) {
2111 BUG_ON(section->iov_base == NULL);
2112 left = sec_len - section->iov_len;
2113 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2114 section->iov_len, left);
2115 if (ret <= 0)
2116 return ret;
2117 section->iov_len += ret;
2450418c 2118 }
fe3ad593
AE		 2119 if (section->iov_len == sec_len)
2120 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2121
2450418c
YS		 2122 return 1;
2123}
31b8006e 2124
34d2d200
AE		 2125static int read_partial_msg_data(struct ceph_connection *con)
2126{
2127 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2128 struct ceph_msg_data_cursor *cursor = &msg->cursor;
34d2d200 2129 const bool do_datacrc = !con->msgr->nocrc;
686be208
AE		 2130 struct page *page;
2131 size_t page_offset;
2132 size_t length;
f5db90bc 2133 u32 crc = 0;
34d2d200
AE		 2134 int ret;
2135
2136 BUG_ON(!msg);
5240d9f9 2137 if (list_empty(&msg->data))
4c59b4a2 2138 return -EIO;
34d2d200 2139
f5db90bc
AE		 2140 if (do_datacrc)
2141 crc = con->in_data_crc;
643c68a4 2142 while (cursor->resid) {
8ae4f4f5 2143 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 2144 NULL);
686be208 2145 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2146 if (ret <= 0) {
2147 if (do_datacrc)
2148 con->in_data_crc = crc;
2149
686be208 2150 return ret;
f5db90bc 2151 }
686be208
AE		 2152
2153 if (do_datacrc)
f5db90bc 2154 crc = ceph_crc32c_page(crc, page, page_offset, ret);
8ae4f4f5 2155 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
34d2d200 2156 }
f5db90bc
AE		 2157 if (do_datacrc)
2158 con->in_data_crc = crc;
34d2d200
AE		 2159
2160 return 1; /* must return > 0 to indicate success */
2161}
2162
31b8006e
SW		 2163/*
2164 * read (part of) a message.
2165 */
686be208
AE		 2166static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2167
31b8006e
SW		 2168static int read_partial_message(struct ceph_connection *con)
2169{
2170 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2171 int size;
2172 int end;
31b8006e 2173 int ret;
95c96174 2174 unsigned int front_len, middle_len, data_len;
37675b0f 2175 bool do_datacrc = !con->msgr->nocrc;
ae18756b 2176 u64 seq;
fe3ad593 2177 u32 crc;
31b8006e
SW		 2178
2179 dout("read_partial_message con %p msg %p\n", con, m);
2180
2181 /* header */
fd51653f
AE		 2182 size = sizeof (con->in_hdr);
2183 end = size;
2184 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2185 if (ret <= 0)
2186 return ret;
fe3ad593
AE		 2187
2188 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2189 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2190 pr_err("read_partial_message bad hdr "
2191 " crc %u != expected %u\n",
2192 crc, con->in_hdr.crc);
2193 return -EBADMSG;
2194 }
2195
31b8006e
SW		 2196 front_len = le32_to_cpu(con->in_hdr.front_len);
2197 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2198 return -EIO;
2199 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2200 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2201 return -EIO;
2202 data_len = le32_to_cpu(con->in_hdr.data_len);
2203 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2204 return -EIO;
2205
ae18756b
SW		 2206 /* verify seq# */
2207 seq = le64_to_cpu(con->in_hdr.seq);
2208 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2209 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2210 ENTITY_NAME(con->peer_name),
3d14c5d2 2211 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2212 seq, con->in_seq + 1);
2213 con->in_base_pos = -front_len - middle_len - data_len -
2214 sizeof(m->footer);
2215 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2216 return 0;
2217 } else if ((s64)seq - (s64)con->in_seq > 1) {
2218 pr_err("read_partial_message bad seq %lld expected %lld\n",
2219 seq, con->in_seq + 1);
2220 con->error_msg = "bad message sequence # for incoming message";
2221 return -EBADMSG;
2222 }
2223
31b8006e
SW		 2224 /* allocate message? */
2225 if (!con->in_msg) {
4740a623
SW		 2226 int skip = 0;
2227
31b8006e 2228 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2229 front_len, data_len);
4740a623
SW		 2230 ret = ceph_con_in_msg_alloc(con, &skip);
2231 if (ret < 0)
2232 return ret;
f759ebb9
AE		 2233
2234 BUG_ON(!con->in_msg ^ skip);
2235 if (con->in_msg && data_len > con->in_msg->data_length) {
2236 pr_warning("%s skipping long message (%u > %zd)\n",
2237 __func__, data_len, con->in_msg->data_length);
2238 ceph_msg_put(con->in_msg);
2239 con->in_msg = NULL;
2240 skip = 1;
2241 }
2450418c 2242 if (skip) {
31b8006e 2243 /* skip this message */
a79832f2 2244 dout("alloc_msg said skip message\n");
31b8006e
SW		 2245 con->in_base_pos = -front_len - middle_len - data_len -
2246 sizeof(m->footer);
2247 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2248 con->in_seq++;
31b8006e
SW		 2249 return 0;
2250 }
38941f80 2251
4740a623 2252 BUG_ON(!con->in_msg);
38941f80 2253 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2254 m = con->in_msg;
2255 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2256 if (m->middle)
2257 m->middle->vec.iov_len = 0;
9d7f0f13 2258
78625051 2259 /* prepare for data payload, if any */
a4107026 2260
78625051 2261 if (data_len)
98fa5dd8 2262 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2263 }
2264
2265 /* front */
2450418c
YS		 2266 ret = read_partial_message_section(con, &m->front, front_len,
2267 &con->in_front_crc);
2268 if (ret <= 0)
2269 return ret;
31b8006e
SW		 2270
2271 /* middle */
2450418c 2272 if (m->middle) {
213c99ee
SW		 2273 ret = read_partial_message_section(con, &m->middle->vec,
2274 middle_len,
2450418c 2275 &con->in_middle_crc);
31b8006e
SW		 2276 if (ret <= 0)
2277 return ret;
31b8006e
SW		 2278 }
2279
2280 /* (page) data */
34d2d200
AE		 2281 if (data_len) {
2282 ret = read_partial_msg_data(con);
2283 if (ret <= 0)
2284 return ret;
31b8006e
SW		 2285 }
2286
31b8006e 2287 /* footer */
fd51653f
AE		 2288 size = sizeof (m->footer);
2289 end += size;
2290 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2291 if (ret <= 0)
2292 return ret;
2293
31b8006e
SW		 2294 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2295 m, front_len, m->footer.front_crc, middle_len,
2296 m->footer.middle_crc, data_len, m->footer.data_crc);
2297
2298 /* crc ok? */
2299 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2300 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2301 m, con->in_front_crc, m->footer.front_crc);
2302 return -EBADMSG;
2303 }
2304 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2305 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2306 m, con->in_middle_crc, m->footer.middle_crc);
2307 return -EBADMSG;
2308 }
bca064d2 2309 if (do_datacrc &&
31b8006e
SW		 2310 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2311 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2312 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2313 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2314 return -EBADMSG;
2315 }
2316
2317 return 1; /* done! */
2318}
2319
2320/*
2321 * Process message. This happens in the worker thread. The callback should
2322 * be careful not to do anything that waits on other incoming messages or it
2323 * may deadlock.
2324 */
2325static void process_message(struct ceph_connection *con)
2326{
5e095e8b 2327 struct ceph_msg *msg;
31b8006e 2328
38941f80
AE		 2329 BUG_ON(con->in_msg->con != con);
2330 con->in_msg->con = NULL;
5e095e8b 2331 msg = con->in_msg;
31b8006e 2332 con->in_msg = NULL;
36eb71aa 2333 con->ops->put(con);
31b8006e
SW		 2334
2335 /* if first message, set peer_name */
2336 if (con->peer_name.type == 0)
dbad185d 2337 con->peer_name = msg->hdr.src;
31b8006e 2338
31b8006e 2339 con->in_seq++;
ec302645 2340 mutex_unlock(&con->mutex);
31b8006e
SW		 2341
2342 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2343 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2344 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2345 le16_to_cpu(msg->hdr.type),
2346 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2347 le32_to_cpu(msg->hdr.front_len),
2348 le32_to_cpu(msg->hdr.data_len),
2349 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2350 con->ops->dispatch(con, msg);
ec302645
SW		 2351
2352 mutex_lock(&con->mutex);
31b8006e
SW		 2353}
2354
2355
2356/*
2357 * Write something to the socket. Called in a worker thread when the
2358 * socket appears to be writeable and we have something ready to send.
2359 */
2360static int try_write(struct ceph_connection *con)
2361{
31b8006e
SW		 2362 int ret = 1;
2363
d59315ca 2364 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2365
31b8006e
SW		 2366more:
2367 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2368
2369 /* open the socket first? */
8dacc7da
SW		 2370 if (con->state == CON_STATE_PREOPEN) {
2371 BUG_ON(con->sock);
2372 con->state = CON_STATE_CONNECTING;
a5988c49 2373
e2200423 2374 con_out_kvec_reset(con);
e825a66d 2375 prepare_write_banner(con);
eed0ef2c 2376 prepare_read_banner(con);
31b8006e 2377
cf3e5c40 2378 BUG_ON(con->in_msg);
31b8006e
SW		 2379 con->in_tag = CEPH_MSGR_TAG_READY;
2380 dout("try_write initiating connect on %p new state %lu\n",
2381 con, con->state);
41617d0c
AE		 2382 ret = ceph_tcp_connect(con);
2383 if (ret < 0) {
31b8006e 2384 con->error_msg = "connect error";
31b8006e
SW		 2385 goto out;
2386 }
2387 }
2388
2389more_kvec:
2390 /* kvec data queued? */
2391 if (con->out_skip) {
2392 ret = write_partial_skip(con);
2393 if (ret <= 0)
42961d23 2394 goto out;
31b8006e
SW		 2395 }
2396 if (con->out_kvec_left) {
2397 ret = write_partial_kvec(con);
2398 if (ret <= 0)
42961d23 2399 goto out;
31b8006e
SW		 2400 }
2401
2402 /* msg pages? */
2403 if (con->out_msg) {
c86a2930
SW		 2404 if (con->out_msg_done) {
2405 ceph_msg_put(con->out_msg);
2406 con->out_msg = NULL; /* we're done with this one */
2407 goto do_next;
2408 }
2409
34d2d200 2410 ret = write_partial_message_data(con);
31b8006e
SW		 2411 if (ret == 1)
2412 goto more_kvec; /* we need to send the footer, too! */
2413 if (ret == 0)
42961d23 2414 goto out;
31b8006e 2415 if (ret < 0) {
34d2d200 2416 dout("try_write write_partial_message_data err %d\n",
31b8006e 2417 ret);
42961d23 2418 goto out;
31b8006e
SW		 2419 }
2420 }
2421
c86a2930 2422do_next:
8dacc7da 2423 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2424 /* is anything else pending? */
2425 if (!list_empty(&con->out_queue)) {
2426 prepare_write_message(con);
2427 goto more;
2428 }
2429 if (con->in_seq > con->in_seq_acked) {
2430 prepare_write_ack(con);
2431 goto more;
2432 }
c9ffc77a 2433 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2434 prepare_write_keepalive(con);
2435 goto more;
2436 }
2437 }
2438
2439 /* Nothing to do! */
c9ffc77a 2440 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2441 dout("try_write nothing else to write.\n");
31b8006e
SW		 2442 ret = 0;
2443out:
42961d23 2444 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2445 return ret;
2446}
2447
2448
2449
2450/*
2451 * Read what we can from the socket.
2452 */
2453static int try_read(struct ceph_connection *con)
2454{
31b8006e
SW		 2455 int ret = -1;
2456
8dacc7da
SW		 2457more:
2458 dout("try_read start on %p state %lu\n", con, con->state);
2459 if (con->state != CON_STATE_CONNECTING &&
2460 con->state != CON_STATE_NEGOTIATING &&
2461 con->state != CON_STATE_OPEN)
31b8006e
SW		 2462 return 0;
2463
8dacc7da 2464 BUG_ON(!con->sock);
ec302645 2465
31b8006e
SW		 2466 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2467 con->in_base_pos);
0da5d703 2468
8dacc7da 2469 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2470 dout("try_read connecting\n");
2471 ret = read_partial_banner(con);
2472 if (ret <= 0)
ab166d5a 2473 goto out;
7593af92
AE		 2474 ret = process_banner(con);
2475 if (ret < 0)
2476 goto out;
2477
8dacc7da 2478 con->state = CON_STATE_NEGOTIATING;
7593af92 2479
6d4221b5
JS		 2480 /*
2481 * Received banner is good, exchange connection info.
2482 * Do not reset out_kvec, as sending our banner raced
2483 * with receiving peer banner after connect completed.
2484 */
7593af92
AE		 2485 ret = prepare_write_connect(con);
2486 if (ret < 0)
2487 goto out;
2488 prepare_read_connect(con);
2489
2490 /* Send connection info before awaiting response */
0da5d703
SW		 2491 goto out;
2492 }
2493
8dacc7da 2494 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2495 dout("try_read negotiating\n");
31b8006e
SW		 2496 ret = read_partial_connect(con);
2497 if (ret <= 0)
31b8006e 2498 goto out;
98bdb0aa
SW		 2499 ret = process_connect(con);
2500 if (ret < 0)
2501 goto out;
31b8006e
SW		 2502 goto more;
2503 }
2504
122070a2 2505 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2506
31b8006e
SW		 2507 if (con->in_base_pos < 0) {
2508 /*
2509 * skipping + discarding content.
2510 *
2511 * FIXME: there must be a better way to do this!
2512 */
84495f49
AE		 2513 static char buf[SKIP_BUF_SIZE];
2514 int skip = min((int) sizeof (buf), -con->in_base_pos);
2515
31b8006e
SW		 2516 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2517 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2518 if (ret <= 0)
98bdb0aa 2519 goto out;
31b8006e
SW		 2520 con->in_base_pos += ret;
2521 if (con->in_base_pos)
2522 goto more;
2523 }
2524 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2525 /*
2526 * what's next?
2527 */
2528 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2529 if (ret <= 0)
98bdb0aa 2530 goto out;
31b8006e
SW		 2531 dout("try_read got tag %d\n", (int)con->in_tag);
2532 switch (con->in_tag) {
2533 case CEPH_MSGR_TAG_MSG:
2534 prepare_read_message(con);
2535 break;
2536 case CEPH_MSGR_TAG_ACK:
2537 prepare_read_ack(con);
2538 break;
2539 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2540 con_close_socket(con);
2541 con->state = CON_STATE_CLOSED;
98bdb0aa 2542 goto out;
31b8006e
SW		 2543 default:
2544 goto bad_tag;
2545 }
2546 }
2547 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2548 ret = read_partial_message(con);
2549 if (ret <= 0) {
2550 switch (ret) {
2551 case -EBADMSG:
2552 con->error_msg = "bad crc";
2553 ret = -EIO;
98bdb0aa 2554 break;
31b8006e
SW		 2555 case -EIO:
2556 con->error_msg = "io error";
98bdb0aa 2557 break;
31b8006e 2558 }
98bdb0aa 2559 goto out;
31b8006e
SW		 2560 }
2561 if (con->in_tag == CEPH_MSGR_TAG_READY)
2562 goto more;
2563 process_message(con);
7b862e07
SW		 2564 if (con->state == CON_STATE_OPEN)
2565 prepare_read_tag(con);
31b8006e
SW		 2566 goto more;
2567 }
3a23083b
SW		 2568 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2569 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2570 /*
2571 * the final handshake seq exchange is semantically
2572 * equivalent to an ACK
2573 */
31b8006e
SW		 2574 ret = read_partial_ack(con);
2575 if (ret <= 0)
98bdb0aa 2576 goto out;
31b8006e
SW		 2577 process_ack(con);
2578 goto more;
2579 }
2580
31b8006e 2581out:
98bdb0aa 2582 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2583 return ret;
2584
2585bad_tag:
2586 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2587 con->error_msg = "protocol error, garbage tag";
2588 ret = -1;
2589 goto out;
2590}
2591
2592
2593/*
802c6d96
AE		 2594 * Atomically queue work on a connection after the specified delay.
2595 * Bump @con reference to avoid races with connection teardown.
2596 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2597 */
802c6d96 2598static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2599{
31b8006e 2600 if (!con->ops->get(con)) {
802c6d96
AE		 2601 dout("%s %p ref count 0\n", __func__, con);
2602
2603 return -ENOENT;
31b8006e
SW		 2604 }
2605
802c6d96
AE		 2606 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2607 dout("%s %p - already queued\n", __func__, con);
31b8006e 2608 con->ops->put(con);
802c6d96
AE		 2609
2610 return -EBUSY;
31b8006e 2611 }
802c6d96
AE		 2612
2613 dout("%s %p %lu\n", __func__, con, delay);
2614
2615 return 0;
2616}
2617
2618static void queue_con(struct ceph_connection *con)
2619{
2620 (void) queue_con_delay(con, 0);
31b8006e
SW		 2621}
2622
7bb21d68
AE		 2623static bool con_sock_closed(struct ceph_connection *con)
2624{
c9ffc77a 2625 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2626 return false;
2627
2628#define CASE(x) \
2629 case CON_STATE_ ## x: \
2630 con->error_msg = "socket closed (con state " #x ")"; \
2631 break;
2632
2633 switch (con->state) {
2634 CASE(CLOSED);
2635 CASE(PREOPEN);
2636 CASE(CONNECTING);
2637 CASE(NEGOTIATING);
2638 CASE(OPEN);
2639 CASE(STANDBY);
2640 default:
2641 pr_warning("%s con %p unrecognized state %lu\n",
2642 __func__, con, con->state);
2643 con->error_msg = "unrecognized con state";
2644 BUG();
2645 break;
2646 }
2647#undef CASE
2648
2649 return true;
2650}
2651
f20a39fd
AE		 2652static bool con_backoff(struct ceph_connection *con)
2653{
2654 int ret;
2655
2656 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2657 return false;
2658
2659 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2660 if (ret) {
2661 dout("%s: con %p FAILED to back off %lu\n", __func__,
2662 con, con->delay);
2663 BUG_ON(ret == -ENOENT);
2664 con_flag_set(con, CON_FLAG_BACKOFF);
2665 }
2666
2667 return true;
2668}
2669
93209264
AE		 2670/* Finish fault handling; con->mutex must *not* be held here */
2671
2672static void con_fault_finish(struct ceph_connection *con)
2673{
2674 /*
2675 * in case we faulted due to authentication, invalidate our
2676 * current tickets so that we can get new ones.
2677 */
2678 if (con->auth_retry && con->ops->invalidate_authorizer) {
2679 dout("calling invalidate_authorizer()\n");
2680 con->ops->invalidate_authorizer(con);
2681 }
2682
2683 if (con->ops->fault)
2684 con->ops->fault(con);
2685}
2686
31b8006e
SW		 2687/*
2688 * Do some work on a connection. Drop a connection ref when we're done.
2689 */
2690static void con_work(struct work_struct *work)
2691{
2692 struct ceph_connection *con = container_of(work, struct ceph_connection,
2693 work.work);
49659416 2694 bool fault;
31b8006e 2695
9dd4658d 2696 mutex_lock(&con->mutex);
49659416
AE		 2697 while (true) {
2698 int ret;
31b8006e 2699
49659416
AE		 2700 if ((fault = con_sock_closed(con))) {
2701 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2702 break;
2703 }
2704 if (con_backoff(con)) {
2705 dout("%s: con %p BACKOFF\n", __func__, con);
2706 break;
2707 }
2708 if (con->state == CON_STATE_STANDBY) {
2709 dout("%s: con %p STANDBY\n", __func__, con);
2710 break;
2711 }
2712 if (con->state == CON_STATE_CLOSED) {
2713 dout("%s: con %p CLOSED\n", __func__, con);
2714 BUG_ON(con->sock);
2715 break;
2716 }
2717 if (con->state == CON_STATE_PREOPEN) {
2718 dout("%s: con %p PREOPEN\n", __func__, con);
2719 BUG_ON(con->sock);
2720 }
0da5d703 2721
49659416
AE		 2722 ret = try_read(con);
2723 if (ret < 0) {
2724 if (ret == -EAGAIN)
2725 continue;
2726 con->error_msg = "socket error on read";
2727 fault = true;
2728 break;
2729 }
2730
2731 ret = try_write(con);
2732 if (ret < 0) {
2733 if (ret == -EAGAIN)
2734 continue;
2735 con->error_msg = "socket error on write";
2736 fault = true;
2737 }
2738
2739 break; /* If we make it to here, we're done */
3a140a0d 2740 }
b6e7b6a1
AE		 2741 if (fault)
2742 con_fault(con);
9dd4658d 2743 mutex_unlock(&con->mutex);
0da5d703 2744
b6e7b6a1
AE		 2745 if (fault)
2746 con_fault_finish(con);
2747
2748 con->ops->put(con);
31b8006e
SW		 2749}
2750
31b8006e
SW		 2751/*
2752 * Generic error/fault handler. A retry mechanism is used with
2753 * exponential backoff
2754 */
93209264 2755static void con_fault(struct ceph_connection *con)
31b8006e 2756{
28362986 2757 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
3d14c5d2 2758 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2759 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2760 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2761
122070a2 2762 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2763 con->state != CON_STATE_NEGOTIATING &&
2764 con->state != CON_STATE_OPEN);
ec302645 2765
31b8006e 2766 con_close_socket(con);
5e095e8b 2767
c9ffc77a 2768 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2769 dout("fault on LOSSYTX channel, marking CLOSED\n");
2770 con->state = CON_STATE_CLOSED;
93209264 2771 return;
3b5ede07
SW		 2772 }
2773
5e095e8b 2774 if (con->in_msg) {
38941f80
AE		 2775 BUG_ON(con->in_msg->con != con);
2776 con->in_msg->con = NULL;
5e095e8b
SW		 2777 ceph_msg_put(con->in_msg);
2778 con->in_msg = NULL;
36eb71aa 2779 con->ops->put(con);
5e095e8b 2780 }
31b8006e 2781
e80a52d1
SW		 2782 /* Requeue anything that hasn't been acked */
2783 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2784
e76661d0
SW		 2785 /* If there are no messages queued or keepalive pending, place
2786 * the connection in a STANDBY state */
2787 if (list_empty(&con->out_queue) &&
c9ffc77a 2788 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2789 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2790 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2791 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2792 } else {
2793 /* retry after a delay. */
8dacc7da 2794 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2795 if (con->delay == 0)
2796 con->delay = BASE_DELAY_INTERVAL;
2797 else if (con->delay < MAX_DELAY_INTERVAL)
2798 con->delay *= 2;
c9ffc77a 2799 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2800 queue_con(con);
31b8006e 2801 }
31b8006e
SW		 2802}
2803
2804
2805
2806/*
15d9882c 2807 * initialize a new messenger instance
31b8006e 2808 */
15d9882c
AE		 2809void ceph_messenger_init(struct ceph_messenger *msgr,
2810 struct ceph_entity_addr *myaddr,
2811 u32 supported_features,
2812 u32 required_features,
2813 bool nocrc)
31b8006e 2814{
3d14c5d2
YS		 2815 msgr->supported_features = supported_features;
2816 msgr->required_features = required_features;
2817
31b8006e
SW		 2818 spin_lock_init(&msgr->global_seq_lock);
2819
31b8006e
SW		 2820 if (myaddr)
2821 msgr->inst.addr = *myaddr;
2822
2823 /* select a random nonce */
ac8839d7 2824 msgr->inst.addr.type = 0;
103e2d3a 2825 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2826 encode_my_addr(msgr);
15d9882c 2827 msgr->nocrc = nocrc;
31b8006e 2828
a2a32584 2829 atomic_set(&msgr->stopping, 0);
31b8006e 2830
15d9882c 2831 dout("%s %p\n", __func__, msgr);
31b8006e 2832}
15d9882c 2833EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2834
e00de341
SW		 2835static void clear_standby(struct ceph_connection *con)
2836{
2837 /* come back from STANDBY? */
8dacc7da 2838 if (con->state == CON_STATE_STANDBY) {
e00de341 2839 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2840 con->state = CON_STATE_PREOPEN;
e00de341 2841 con->connect_seq++;
c9ffc77a
AE		 2842 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2843 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2844 }
2845}
2846
31b8006e
SW		 2847/*
2848 * Queue up an outgoing message on the given connection.
2849 */
2850void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2851{
31b8006e 2852 /* set src+dst */
dbad185d 2853 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2854 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2855 msg->needs_out_seq = true;
2856
ec302645 2857 mutex_lock(&con->mutex);
92ce034b 2858
8dacc7da 2859 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2860 dout("con_send %p closed, dropping %p\n", con, msg);
2861 ceph_msg_put(msg);
2862 mutex_unlock(&con->mutex);
2863 return;
2864 }
2865
38941f80 2866 BUG_ON(msg->con != NULL);
36eb71aa 2867 msg->con = con->ops->get(con);
92ce034b
AE		 2868 BUG_ON(msg->con == NULL);
2869
31b8006e
SW		 2870 BUG_ON(!list_empty(&msg->list_head));
2871 list_add_tail(&msg->list_head, &con->out_queue);
2872 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2873 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2874 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2875 le32_to_cpu(msg->hdr.front_len),
2876 le32_to_cpu(msg->hdr.middle_len),
2877 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 2878
2879 clear_standby(con);
ec302645 2880 mutex_unlock(&con->mutex);
31b8006e
SW		 2881
2882 /* if there wasn't anything waiting to send before, queue
2883 * new work */
c9ffc77a 2884 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2885 queue_con(con);
2886}
3d14c5d2 2887EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 2888
2889/*
2890 * Revoke a message that was previously queued for send
2891 */
6740a845 2892void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 2893{
6740a845
AE		 2894 struct ceph_connection *con = msg->con;
2895
2896 if (!con)
2897 return; /* Message not in our possession */
2898
ec302645 2899 mutex_lock(&con->mutex);
31b8006e 2900 if (!list_empty(&msg->list_head)) {
38941f80 2901 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 2902 list_del_init(&msg->list_head);
38941f80 2903 BUG_ON(msg->con == NULL);
36eb71aa 2904 msg->con->ops->put(msg->con);
38941f80 2905 msg->con = NULL;
31b8006e 2906 msg->hdr.seq = 0;
38941f80 2907
31b8006e 2908 ceph_msg_put(msg);
ed98adad
SW		 2909 }
2910 if (con->out_msg == msg) {
38941f80 2911 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 2912 con->out_msg = NULL;
31b8006e
SW		 2913 if (con->out_kvec_is_msg) {
2914 con->out_skip = con->out_kvec_bytes;
2915 con->out_kvec_is_msg = false;
2916 }
ed98adad 2917 msg->hdr.seq = 0;
92ce034b
AE		 2918
2919 ceph_msg_put(msg);
31b8006e 2920 }
ec302645 2921 mutex_unlock(&con->mutex);
31b8006e
SW		 2922}
2923
350b1c32 2924/*
0d59ab81 2925 * Revoke a message that we may be reading data into
350b1c32 2926 */
8921d114 2927void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 2928{
8921d114
AE		 2929 struct ceph_connection *con;
2930
2931 BUG_ON(msg == NULL);
2932 if (!msg->con) {
2933 dout("%s msg %p null con\n", __func__, msg);
2934
2935 return; /* Message not in our possession */
2936 }
2937
2938 con = msg->con;
350b1c32 2939 mutex_lock(&con->mutex);
8921d114 2940 if (con->in_msg == msg) {
95c96174
ED		 2941 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2942 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2943 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 2944
2945 /* skip rest of message */
8921d114
AE		 2946 dout("%s %p msg %p revoked\n", __func__, con, msg);
2947 con->in_base_pos = con->in_base_pos -
350b1c32 2948 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 2949 front_len -
2950 middle_len -
2951 data_len -
350b1c32 2952 sizeof(struct ceph_msg_footer);
350b1c32
SW		 2953 ceph_msg_put(con->in_msg);
2954 con->in_msg = NULL;
2955 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2956 con->in_seq++;
350b1c32 2957 } else {
8921d114
AE		 2958 dout("%s %p in_msg %p msg %p no-op\n",
2959 __func__, con, con->in_msg, msg);
350b1c32
SW		 2960 }
2961 mutex_unlock(&con->mutex);
2962}
2963
31b8006e
SW		 2964/*
2965 * Queue a keepalive byte to ensure the tcp connection is alive.
2966 */
2967void ceph_con_keepalive(struct ceph_connection *con)
2968{
e00de341 2969 dout("con_keepalive %p\n", con);
00650931 2970 mutex_lock(&con->mutex);
e00de341 2971 clear_standby(con);
00650931 2972 mutex_unlock(&con->mutex);
c9ffc77a
AE		 2973 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
2974 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2975 queue_con(con);
2976}
3d14c5d2 2977EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 2978
6644ed7b 2979static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 2980{
6644ed7b
AE		 2981 struct ceph_msg_data *data;
2982
2983 if (WARN_ON(!ceph_msg_data_type_valid(type)))
2984 return NULL;
2985
2986 data = kzalloc(sizeof (*data), GFP_NOFS);
2987 if (data)
2988 data->type = type;
5240d9f9 2989 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 2990
2991 return data;
2992}
2993
2994static void ceph_msg_data_destroy(struct ceph_msg_data *data)
2995{
2996 if (!data)
2997 return;
2998
5240d9f9 2999 WARN_ON(!list_empty(&data->links));
6644ed7b
AE		 3000 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3001 ceph_pagelist_release(data->pagelist);
3002 kfree(data->pagelist);
3003 }
3004 kfree(data);
43794509
AE		 3005}
3006
90af3602 3007void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3008 size_t length, size_t alignment)
02afca6c 3009{
6644ed7b
AE		 3010 struct ceph_msg_data *data;
3011
07aa1558
AE		 3012 BUG_ON(!pages);
3013 BUG_ON(!length);
6644ed7b
AE		 3014
3015 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3016 BUG_ON(!data);
3017 data->pages = pages;
3018 data->length = length;
3019 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3020
5240d9f9
AE		 3021 list_add_tail(&data->links, &msg->data);
3022 msg->data_length += length;
02afca6c 3023}
90af3602 3024EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3025
90af3602 3026void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3027 struct ceph_pagelist *pagelist)
3028{
6644ed7b
AE		 3029 struct ceph_msg_data *data;
3030
07aa1558
AE		 3031 BUG_ON(!pagelist);
3032 BUG_ON(!pagelist->length);
27fa8385 3033
6644ed7b
AE		 3034 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3035 BUG_ON(!data);
3036 data->pagelist = pagelist;
3037
5240d9f9
AE		 3038 list_add_tail(&data->links, &msg->data);
3039 msg->data_length += pagelist->length;
27fa8385 3040}
90af3602 3041EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3042
ea96571f 3043#ifdef CONFIG_BLOCK
90af3602 3044void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3045 size_t length)
27fa8385 3046{
6644ed7b
AE		 3047 struct ceph_msg_data *data;
3048
07aa1558 3049 BUG_ON(!bio);
27fa8385 3050
6644ed7b
AE		 3051 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3052 BUG_ON(!data);
3053 data->bio = bio;
c851c495 3054 data->bio_length = length;
6644ed7b 3055
5240d9f9
AE		 3056 list_add_tail(&data->links, &msg->data);
3057 msg->data_length += length;
27fa8385 3058}
90af3602 3059EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3060#endif /* CONFIG_BLOCK */
27fa8385 3061
31b8006e
SW		 3062/*
3063 * construct a new message with given type, size
3064 * the new msg has a ref count of 1.
3065 */
b61c2763
SW		 3066struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3067 bool can_fail)
31b8006e
SW		 3068{
3069 struct ceph_msg *m;
3070
9516e45b 3071 m = kzalloc(sizeof(*m), flags);
31b8006e
SW		 3072 if (m == NULL)
3073 goto out;
31b8006e
SW		 3074
3075 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3076 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3077 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3078
9516e45b
AE		 3079 INIT_LIST_HEAD(&m->list_head);
3080 kref_init(&m->kref);
5240d9f9 3081 INIT_LIST_HEAD(&m->data);
ca20892d 3082
31b8006e 3083 /* front */
9516e45b 3084 m->front_max = front_len;
31b8006e
SW		 3085 if (front_len) {
3086 if (front_len > PAGE_CACHE_SIZE) {
34d23762 3087 m->front.iov_base = __vmalloc(front_len, flags,
31b8006e
SW		 3088 PAGE_KERNEL);
3089 m->front_is_vmalloc = true;
3090 } else {
34d23762 3091 m->front.iov_base = kmalloc(front_len, flags);
31b8006e
SW		 3092 }
3093 if (m->front.iov_base == NULL) {
b61c2763 3094 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3095 front_len);
3096 goto out2;
3097 }
3098 } else {
3099 m->front.iov_base = NULL;
3100 }
3101 m->front.iov_len = front_len;
3102
bb257664 3103 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3104 return m;
3105
3106out2:
3107 ceph_msg_put(m);
3108out:
b61c2763
SW		 3109 if (!can_fail) {
3110 pr_err("msg_new can't create type %d front %d\n", type,
3111 front_len);
f0ed1b7c 3112 WARN_ON(1);
b61c2763
SW		 3113 } else {
3114 dout("msg_new can't create type %d front %d\n", type,
3115 front_len);
3116 }
a79832f2 3117 return NULL;
31b8006e 3118}
3d14c5d2 3119EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3120
31b8006e
SW		 3121/*
3122 * Allocate "middle" portion of a message, if it is needed and wasn't
3123 * allocated by alloc_msg. This allows us to read a small fixed-size
3124 * per-type header in the front and then gracefully fail (i.e.,
3125 * propagate the error to the caller based on info in the front) when
3126 * the middle is too large.
3127 */
2450418c 3128static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3129{
3130 int type = le16_to_cpu(msg->hdr.type);
3131 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3132
3133 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3134 ceph_msg_type_name(type), middle_len);
3135 BUG_ON(!middle_len);
3136 BUG_ON(msg->middle);
3137
b6c1d5b8 3138 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3139 if (!msg->middle)
3140 return -ENOMEM;
3141 return 0;
3142}
3143
2450418c 3144/*
1c20f2d2
AE		 3145 * Allocate a message for receiving an incoming message on a
3146 * connection, and save the result in con->in_msg. Uses the
3147 * connection's private alloc_msg op if available.
3148 *
4740a623
SW		 3149 * Returns 0 on success, or a negative error code.
3150 *
3151 * On success, if we set *skip = 1:
3152 * - the next message should be skipped and ignored.
3153 * - con->in_msg == NULL
3154 * or if we set *skip = 0:
3155 * - con->in_msg is non-null.
3156 * On error (ENOMEM, EAGAIN, ...),
3157 * - con->in_msg == NULL
2450418c 3158 */
4740a623 3159static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3160{
4740a623 3161 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3162 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3163 struct ceph_msg *msg;
4740a623 3164 int ret = 0;
2450418c 3165
1c20f2d2 3166 BUG_ON(con->in_msg != NULL);
53ded495 3167 BUG_ON(!con->ops->alloc_msg);
2450418c 3168
53ded495
AE		 3169 mutex_unlock(&con->mutex);
3170 msg = con->ops->alloc_msg(con, hdr, skip);
3171 mutex_lock(&con->mutex);
3172 if (con->state != CON_STATE_OPEN) {
3173 if (msg)
1d866d1c 3174 ceph_msg_put(msg);
53ded495
AE		 3175 return -EAGAIN;
3176 }
4137577a
AE		 3177 if (msg) {
3178 BUG_ON(*skip);
3179 con->in_msg = msg;
36eb71aa 3180 con->in_msg->con = con->ops->get(con);
92ce034b 3181 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3182 } else {
3183 /*
3184 * Null message pointer means either we should skip
3185 * this message or we couldn't allocate memory. The
3186 * former is not an error.
3187 */
3188 if (*skip)
3189 return 0;
3190 con->error_msg = "error allocating memory for incoming message";
3191
53ded495 3192 return -ENOMEM;
2450418c 3193 }
1c20f2d2 3194 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3195
1c20f2d2
AE		 3196 if (middle_len && !con->in_msg->middle) {
3197 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3198 if (ret < 0) {
1c20f2d2
AE		 3199 ceph_msg_put(con->in_msg);
3200 con->in_msg = NULL;
2450418c
YS		 3201 }
3202 }
9d7f0f13 3203
4740a623 3204 return ret;
2450418c
YS		 3205}
3206
31b8006e
SW		 3207
3208/*
3209 * Free a generically kmalloc'd message.
3210 */
3211void ceph_msg_kfree(struct ceph_msg *m)
3212{
3213 dout("msg_kfree %p\n", m);
3214 if (m->front_is_vmalloc)
3215 vfree(m->front.iov_base);
3216 else
3217 kfree(m->front.iov_base);
3218 kfree(m);
3219}
3220
3221/*
3222 * Drop a msg ref. Destroy as needed.
3223 */
c2e552e7
SW		 3224void ceph_msg_last_put(struct kref *kref)
3225{
3226 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
5240d9f9
AE		 3227 LIST_HEAD(data);
3228 struct list_head *links;
3229 struct list_head *next;
31b8006e 3230
c2e552e7
SW		 3231 dout("ceph_msg_put last one on %p\n", m);
3232 WARN_ON(!list_empty(&m->list_head));
3233
3234 /* drop middle, data, if any */
3235 if (m->middle) {
3236 ceph_buffer_put(m->middle);
3237 m->middle = NULL;
31b8006e 3238 }
5240d9f9
AE		 3239
3240 list_splice_init(&m->data, &data);
3241 list_for_each_safe(links, next, &data) {
3242 struct ceph_msg_data *data;
3243
3244 data = list_entry(links, struct ceph_msg_data, links);
3245 list_del_init(links);
3246 ceph_msg_data_destroy(data);
3247 }
a1930804 3248 m->data_length = 0;
58bb3b37 3249
c2e552e7
SW		 3250 if (m->pool)
3251 ceph_msgpool_put(m->pool, m);
3252 else
3253 ceph_msg_kfree(m);
31b8006e 3254}
3d14c5d2 3255EXPORT_SYMBOL(ceph_msg_last_put);
9ec7cab1
SW		 3256
3257void ceph_msg_dump(struct ceph_msg *msg)
3258{
4a73ef27 3259 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
5240d9f9 3260 msg->front_max, msg->data_length);
9ec7cab1
SW		 3261 print_hex_dump(KERN_DEBUG, "header: ",
3262 DUMP_PREFIX_OFFSET, 16, 1,
3263 &msg->hdr, sizeof(msg->hdr), true);
3264 print_hex_dump(KERN_DEBUG, " front: ",
3265 DUMP_PREFIX_OFFSET, 16, 1,
3266 msg->front.iov_base, msg->front.iov_len, true);
3267 if (msg->middle)
3268 print_hex_dump(KERN_DEBUG, "middle: ",
3269 DUMP_PREFIX_OFFSET, 16, 1,
3270 msg->middle->vec.iov_base,
3271 msg->middle->vec.iov_len, true);
3272 print_hex_dump(KERN_DEBUG, "footer: ",
3273 DUMP_PREFIX_OFFSET, 16, 1,
3274 &msg->footer, sizeof(msg->footer), true);
3275}
3d14c5d2 3276EXPORT_SYMBOL(ceph_msg_dump);
kernel source for telekom puls (alcatel/mediatek)