projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: replace message data pointer with list
[deliverable/linux.git] / net / ceph / messenger.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
5a0e3ad6 9#include <linux/slab.h>
31b8006e
SW		 10#include <linux/socket.h>
11#include <linux/string.h>
3ebc21f7 12#ifdef CONFIG_BLOCK
68b4476b 13#include <linux/bio.h>
3ebc21f7 14#endif /* CONFIG_BLOCK */
ee3b56f2 15#include <linux/dns_resolver.h>
31b8006e
SW		 16#include <net/tcp.h>
17
3d14c5d2
YS		 18#include <linux/ceph/libceph.h>
19#include <linux/ceph/messenger.h>
20#include <linux/ceph/decode.h>
21#include <linux/ceph/pagelist.h>
bc3b2d7f 22#include <linux/export.h>
31b8006e 23
fe38a2b6
AE		 24#define list_entry_next(pos, member) \
25 list_entry(pos->member.next, typeof(*pos), member)
26
31b8006e
SW		 27/*
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
33 * the sender.
34 */
35
bc18f4b1
AE		 36/*
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
40 * unexpected state.
41 *
42 * --------
43 * | NEW* | transient initial state
44 * --------
45 * | con_sock_state_init()
46 * v
47 * ----------
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
50 * ^ \
51 * | \ con_sock_state_connecting()
52 * | ----------------------
53 * | \
54 * + con_sock_state_closed() \
fbb85a47
SW		 55 * |+--------------------------- \
56 * | \ \ \
57 * | ----------- \ \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
60 * | ^ \ |
61 * | | \ |
62 * | + con_sock_state_closing() \ |
63 * | / \ | |
64 * | / --------------- | |
65 * | / \ v v
bc18f4b1
AE		 66 * | / --------------
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
70 * | | v
71 * -------------
72 * | CONNECTED | TCP connection established
73 * -------------
74 *
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
76 */
ce2c8903
AE		 77
78#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
83
8dacc7da
SW		 84/*
85 * connection states
86 */
87#define CON_STATE_CLOSED 1 /* -> PREOPEN */
88#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
93
4a861692
SW		 94/*
95 * ceph_connection flag bits
96 */
97#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 103
c9ffc77a
AE		 104static bool con_flag_valid(unsigned long con_flag)
105{
106 switch (con_flag) {
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
112 return true;
113 default:
114 return false;
115 }
116}
117
118static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
119{
120 BUG_ON(!con_flag_valid(con_flag));
121
122 clear_bit(con_flag, &con->flags);
123}
124
125static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
126{
127 BUG_ON(!con_flag_valid(con_flag));
128
129 set_bit(con_flag, &con->flags);
130}
131
132static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
133{
134 BUG_ON(!con_flag_valid(con_flag));
135
136 return test_bit(con_flag, &con->flags);
137}
138
139static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
141{
142 BUG_ON(!con_flag_valid(con_flag));
143
144 return test_and_clear_bit(con_flag, &con->flags);
145}
146
147static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
149{
150 BUG_ON(!con_flag_valid(con_flag));
151
152 return test_and_set_bit(con_flag, &con->flags);
153}
154
31b8006e
SW		 155/* static tag bytes (protocol control messages) */
156static char tag_msg = CEPH_MSGR_TAG_MSG;
157static char tag_ack = CEPH_MSGR_TAG_ACK;
158static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
159
a6a5349d
SW		 160#ifdef CONFIG_LOCKDEP
161static struct lock_class_key socket_class;
162#endif
163
84495f49
AE		 164/*
165 * When skipping (ignoring) a block of input we read it into a "skip
166 * buffer," which is this many bytes in size.
167 */
168#define SKIP_BUF_SIZE 1024
31b8006e
SW		 169
170static void queue_con(struct ceph_connection *con);
171static void con_work(struct work_struct *);
93209264 172static void con_fault(struct ceph_connection *con);
31b8006e 173
31b8006e 174/*
f64a9317
AE		 175 * Nicely render a sockaddr as a string. An array of formatted
176 * strings is used, to approximate reentrancy.
31b8006e 177 */
f64a9317
AE		 178#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
179#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
180#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
181#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
182
183static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
184static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 185
57666519 186static struct page *zero_page; /* used in certain error cases */
57666519 187
3d14c5d2 188const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 189{
190 int i;
191 char *s;
99f0f3b2
AE		 192 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
193 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 194
f64a9317 195 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 196 s = addr_str[i];
197
198 switch (ss->ss_family) {
199 case AF_INET:
bd406145
AE		 200 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
201 ntohs(in4->sin_port));
31b8006e
SW		 202 break;
203
204 case AF_INET6:
bd406145
AE		 205 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
206 ntohs(in6->sin6_port));
31b8006e
SW		 207 break;
208
209 default:
d3002b97
AE		 210 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
211 ss->ss_family);
31b8006e
SW		 212 }
213
214 return s;
215}
3d14c5d2 216EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 217
63f2d211
SW		 218static void encode_my_addr(struct ceph_messenger *msgr)
219{
220 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
221 ceph_encode_addr(&msgr->my_enc_addr);
222}
223
31b8006e
SW		 224/*
225 * work queue for all reading and writing to/from the socket.
226 */
e0f43c94 227static struct workqueue_struct *ceph_msgr_wq;
31b8006e 228
15417167 229static void _ceph_msgr_exit(void)
6173d1f0 230{
d3002b97 231 if (ceph_msgr_wq) {
6173d1f0 232 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 233 ceph_msgr_wq = NULL;
234 }
6173d1f0 235
6173d1f0
AE		 236 BUG_ON(zero_page == NULL);
237 kunmap(zero_page);
238 page_cache_release(zero_page);
239 zero_page = NULL;
240}
241
3d14c5d2 242int ceph_msgr_init(void)
31b8006e 243{
57666519
AE		 244 BUG_ON(zero_page != NULL);
245 zero_page = ZERO_PAGE(0);
246 page_cache_get(zero_page);
247
f363e45f 248 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
6173d1f0
AE		 249 if (ceph_msgr_wq)
250 return 0;
57666519 251
6173d1f0
AE		 252 pr_err("msgr_init failed to create workqueue\n");
253 _ceph_msgr_exit();
57666519 254
6173d1f0 255 return -ENOMEM;
31b8006e 256}
3d14c5d2 257EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 258
259void ceph_msgr_exit(void)
260{
57666519 261 BUG_ON(ceph_msgr_wq == NULL);
57666519 262
6173d1f0 263 _ceph_msgr_exit();
31b8006e 264}
3d14c5d2 265EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 266
cd84db6e 267void ceph_msgr_flush(void)
a922d38f
SW		 268{
269 flush_workqueue(ceph_msgr_wq);
270}
3d14c5d2 271EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 272
ce2c8903
AE		 273/* Connection socket state transition functions */
274
275static void con_sock_state_init(struct ceph_connection *con)
276{
277 int old_state;
278
279 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
280 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
281 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 282 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
283 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 284}
285
286static void con_sock_state_connecting(struct ceph_connection *con)
287{
288 int old_state;
289
290 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
291 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
292 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 293 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
294 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 295}
296
297static void con_sock_state_connected(struct ceph_connection *con)
298{
299 int old_state;
300
301 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
302 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
303 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 304 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
305 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 306}
307
308static void con_sock_state_closing(struct ceph_connection *con)
309{
310 int old_state;
311
312 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
313 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
314 old_state != CON_SOCK_STATE_CONNECTED &&
315 old_state != CON_SOCK_STATE_CLOSING))
316 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 317 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
318 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 319}
320
321static void con_sock_state_closed(struct ceph_connection *con)
322{
323 int old_state;
324
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 327 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 328 old_state != CON_SOCK_STATE_CONNECTING &&
329 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 330 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 331 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
332 CON_SOCK_STATE_CLOSED);
ce2c8903 333}
a922d38f 334
31b8006e
SW		 335/*
336 * socket callback functions
337 */
338
339/* data available on socket, or listen socket received a connect */
327800bd 340static void ceph_sock_data_ready(struct sock *sk, int count_unused)
31b8006e 341{
bd406145 342 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 343 if (atomic_read(&con->msgr->stopping)) {
344 return;
345 }
bd406145 346
31b8006e 347 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 348 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 349 con, con->state);
350 queue_con(con);
351 }
352}
353
354/* socket has buffer space for writing */
327800bd 355static void ceph_sock_write_space(struct sock *sk)
31b8006e 356{
d3002b97 357 struct ceph_connection *con = sk->sk_user_data;
31b8006e 358
182fac26
JS		 359 /* only queue to workqueue if there is data we want to write,
360 * and there is sufficient space in the socket buffer to accept
327800bd 361 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 362 * doesn't get called again until try_write() fills the socket
363 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
364 * and net/core/stream.c:sk_stream_write_space().
365 */
c9ffc77a 366 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
182fac26 367 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
327800bd 368 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 369 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
370 queue_con(con);
371 }
31b8006e 372 } else {
327800bd 373 dout("%s %p nothing to write\n", __func__, con);
31b8006e 374 }
31b8006e
SW		 375}
376
377/* socket's state has changed */
327800bd 378static void ceph_sock_state_change(struct sock *sk)
31b8006e 379{
bd406145 380 struct ceph_connection *con = sk->sk_user_data;
31b8006e 381
327800bd 382 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 383 con, con->state, sk->sk_state);
384
31b8006e
SW		 385 switch (sk->sk_state) {
386 case TCP_CLOSE:
327800bd 387 dout("%s TCP_CLOSE\n", __func__);
31b8006e 388 case TCP_CLOSE_WAIT:
327800bd 389 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 390 con_sock_state_closing(con);
c9ffc77a 391 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 392 queue_con(con);
31b8006e
SW		 393 break;
394 case TCP_ESTABLISHED:
327800bd 395 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 396 con_sock_state_connected(con);
31b8006e
SW		 397 queue_con(con);
398 break;
d3002b97
AE		 399 default: /* Everything else is uninteresting */
400 break;
31b8006e
SW		 401 }
402}
403
404/*
405 * set up socket callbacks
406 */
407static void set_sock_callbacks(struct socket *sock,
408 struct ceph_connection *con)
409{
410 struct sock *sk = sock->sk;
bd406145 411 sk->sk_user_data = con;
327800bd
AE		 412 sk->sk_data_ready = ceph_sock_data_ready;
413 sk->sk_write_space = ceph_sock_write_space;
414 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 415}
416
417
418/*
419 * socket helpers
420 */
421
422/*
423 * initiate connection to a remote socket.
424 */
41617d0c 425static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 426{
f91d3471 427 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e
SW		 428 struct socket *sock;
429 int ret;
430
431 BUG_ON(con->sock);
f91d3471
SW		 432 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
433 IPPROTO_TCP, &sock);
31b8006e 434 if (ret)
41617d0c 435 return ret;
31b8006e
SW		 436 sock->sk->sk_allocation = GFP_NOFS;
437
a6a5349d
SW		 438#ifdef CONFIG_LOCKDEP
439 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
440#endif
441
31b8006e
SW		 442 set_sock_callbacks(sock, con);
443
3d14c5d2 444 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 445
89a86be0 446 con_sock_state_connecting(con);
f91d3471
SW		 447 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
448 O_NONBLOCK);
31b8006e
SW		 449 if (ret == -EINPROGRESS) {
450 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 451 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 452 sock->sk->sk_state);
a5bc3129 453 } else if (ret < 0) {
31b8006e 454 pr_err("connect %s error %d\n",
3d14c5d2 455 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 456 sock_release(sock);
31b8006e 457 con->error_msg = "connect error";
31b8006e 458
41617d0c 459 return ret;
a5bc3129
AE		 460 }
461 con->sock = sock;
41617d0c 462 return 0;
31b8006e
SW		 463}
464
465static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
466{
467 struct kvec iov = {buf, len};
468 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 469 int r;
31b8006e 470
98bdb0aa
SW		 471 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
472 if (r == -EAGAIN)
473 r = 0;
474 return r;
31b8006e
SW		 475}
476
afb3d90e
AE		 477static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
478 int page_offset, size_t length)
479{
480 void *kaddr;
481 int ret;
482
483 BUG_ON(page_offset + length > PAGE_SIZE);
484
485 kaddr = kmap(page);
486 BUG_ON(!kaddr);
487 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
488 kunmap(page);
489
490 return ret;
491}
492
31b8006e
SW		 493/*
494 * write something. @more is true if caller will be sending more data
495 * shortly.
496 */
497static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
498 size_t kvlen, size_t len, int more)
499{
500 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 501 int r;
31b8006e
SW		 502
503 if (more)
504 msg.msg_flags |= MSG_MORE;
505 else
506 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
507
42961d23
SW		 508 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
509 if (r == -EAGAIN)
510 r = 0;
511 return r;
31b8006e
SW		 512}
513
31739139 514static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 515 int offset, size_t size, bool more)
31739139
AE		 516{
517 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
518 int ret;
519
520 ret = kernel_sendpage(sock, page, offset, size, flags);
521 if (ret == -EAGAIN)
522 ret = 0;
523
524 return ret;
525}
526
31b8006e
SW		 527
528/*
529 * Shutdown/close the socket for the given connection.
530 */
531static int con_close_socket(struct ceph_connection *con)
532{
8007b8d6 533 int rc = 0;
31b8006e
SW		 534
535 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 536 if (con->sock) {
537 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
538 sock_release(con->sock);
539 con->sock = NULL;
540 }
456ea468
AE		 541
542 /*
4a861692 543 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 544 * independent of the connection mutex, and we could have
545 * received a socket close event before we had the chance to
546 * shut the socket down.
547 */
c9ffc77a 548 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 549
ce2c8903 550 con_sock_state_closed(con);
31b8006e
SW		 551 return rc;
552}
553
554/*
555 * Reset a connection. Discard all incoming and outgoing messages
556 * and clear *_seq state.
557 */
558static void ceph_msg_remove(struct ceph_msg *msg)
559{
560 list_del_init(&msg->list_head);
38941f80 561 BUG_ON(msg->con == NULL);
36eb71aa 562 msg->con->ops->put(msg->con);
38941f80
AE		 563 msg->con = NULL;
564
31b8006e
SW		 565 ceph_msg_put(msg);
566}
567static void ceph_msg_remove_list(struct list_head *head)
568{
569 while (!list_empty(head)) {
570 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
571 list_head);
572 ceph_msg_remove(msg);
573 }
574}
575
576static void reset_connection(struct ceph_connection *con)
577{
578 /* reset connection, out_queue, msg_ and connect_seq */
579 /* discard existing out_queue and msg_seq */
0fa6ebc6 580 dout("reset_connection %p\n", con);
31b8006e
SW		 581 ceph_msg_remove_list(&con->out_queue);
582 ceph_msg_remove_list(&con->out_sent);
583
cf3e5c40 584 if (con->in_msg) {
38941f80
AE		 585 BUG_ON(con->in_msg->con != con);
586 con->in_msg->con = NULL;
cf3e5c40
SW		 587 ceph_msg_put(con->in_msg);
588 con->in_msg = NULL;
36eb71aa 589 con->ops->put(con);
cf3e5c40
SW		 590 }
591
31b8006e
SW		 592 con->connect_seq = 0;
593 con->out_seq = 0;
c86a2930
SW		 594 if (con->out_msg) {
595 ceph_msg_put(con->out_msg);
596 con->out_msg = NULL;
597 }
31b8006e 598 con->in_seq = 0;
0e0d5e0c 599 con->in_seq_acked = 0;
31b8006e
SW		 600}
601
602/*
603 * mark a peer down. drop any open connections.
604 */
605void ceph_con_close(struct ceph_connection *con)
606{
8c50c817 607 mutex_lock(&con->mutex);
3d14c5d2
YS		 608 dout("con_close %p peer %s\n", con,
609 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 610 con->state = CON_STATE_CLOSED;
a5988c49 611
c9ffc77a
AE		 612 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
613 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
614 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
615 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 616
31b8006e 617 reset_connection(con);
6f2bc3ff 618 con->peer_global_seq = 0;
91e45ce3 619 cancel_delayed_work(&con->work);
ee76e073 620 con_close_socket(con);
ec302645 621 mutex_unlock(&con->mutex);
31b8006e 622}
3d14c5d2 623EXPORT_SYMBOL(ceph_con_close);
31b8006e 624
31b8006e
SW		 625/*
626 * Reopen a closed connection, with a new peer address.
627 */
b7a9e5dd
SW		 628void ceph_con_open(struct ceph_connection *con,
629 __u8 entity_type, __u64 entity_num,
630 struct ceph_entity_addr *addr)
31b8006e 631{
5469155f 632 mutex_lock(&con->mutex);
3d14c5d2 633 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 634
122070a2 635 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 636 con->state = CON_STATE_PREOPEN;
a5988c49 637
b7a9e5dd
SW		 638 con->peer_name.type = (__u8) entity_type;
639 con->peer_name.num = cpu_to_le64(entity_num);
640
31b8006e 641 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 642 con->delay = 0; /* reset backoff memory */
5469155f 643 mutex_unlock(&con->mutex);
31b8006e
SW		 644 queue_con(con);
645}
3d14c5d2 646EXPORT_SYMBOL(ceph_con_open);
31b8006e 647
87b315a5
SW		 648/*
649 * return true if this connection ever successfully opened
650 */
651bool ceph_con_opened(struct ceph_connection *con)
652{
653 return con->connect_seq > 0;
654}
655
31b8006e
SW		 656/*
657 * initialize a new connection.
658 */
1bfd89f4
AE		 659void ceph_con_init(struct ceph_connection *con, void *private,
660 const struct ceph_connection_operations *ops,
b7a9e5dd 661 struct ceph_messenger *msgr)
31b8006e
SW		 662{
663 dout("con_init %p\n", con);
664 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 665 con->private = private;
666 con->ops = ops;
31b8006e 667 con->msgr = msgr;
ce2c8903
AE		 668
669 con_sock_state_init(con);
670
ec302645 671 mutex_init(&con->mutex);
31b8006e
SW		 672 INIT_LIST_HEAD(&con->out_queue);
673 INIT_LIST_HEAD(&con->out_sent);
674 INIT_DELAYED_WORK(&con->work, con_work);
a5988c49 675
8dacc7da 676 con->state = CON_STATE_CLOSED;
31b8006e 677}
3d14c5d2 678EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 679
680
681/*
682 * We maintain a global counter to order connection attempts. Get
683 * a unique seq greater than @gt.
684 */
685static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
686{
687 u32 ret;
688
689 spin_lock(&msgr->global_seq_lock);
690 if (msgr->global_seq < gt)
691 msgr->global_seq = gt;
692 ret = ++msgr->global_seq;
693 spin_unlock(&msgr->global_seq_lock);
694 return ret;
695}
696
e2200423 697static void con_out_kvec_reset(struct ceph_connection *con)
859eb799
AE		 698{
699 con->out_kvec_left = 0;
700 con->out_kvec_bytes = 0;
701 con->out_kvec_cur = &con->out_kvec[0];
702}
703
e2200423 704static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 705 size_t size, void *data)
706{
707 int index;
708
709 index = con->out_kvec_left;
710 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
711
712 con->out_kvec[index].iov_len = size;
713 con->out_kvec[index].iov_base = data;
714 con->out_kvec_left++;
715 con->out_kvec_bytes += size;
716}
31b8006e 717
df6ad1f9 718#ifdef CONFIG_BLOCK
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
25aff7c5 737 cursor->resid = length;
6aaa4511
AE		 738 cursor->bio = bio;
739 cursor->vector_index = 0;
740 cursor->vector_offset = 0;
25aff7c5 741 cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
6aaa4511
AE		 742}
743
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);
1190bf06 836 BUG_ON(length > data->length); /* short reads are OK */
e766d7b5 837
25aff7c5 838 cursor->resid = length;
e766d7b5 839 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 840 cursor->page_offset = data->alignment & ~PAGE_MASK;
841 cursor->page_index = 0;
56fc5659
AE		 842 BUG_ON(page_count > (int)USHRT_MAX);
843 cursor->page_count = (unsigned short)page_count;
844 BUG_ON(length > SIZE_MAX - cursor->page_offset);
845 cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE;
e766d7b5
AE		 846}
847
8ae4f4f5
AE		 848static struct page *
849ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
850 size_t *page_offset, size_t *length)
e766d7b5 851{
8ae4f4f5 852 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 853
854 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
855
856 BUG_ON(cursor->page_index >= cursor->page_count);
857 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 858
859 *page_offset = cursor->page_offset;
25aff7c5 860 if (cursor->last_piece)
e766d7b5 861 *length = cursor->resid;
25aff7c5 862 else
e766d7b5 863 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 864
865 return data->pages[cursor->page_index];
866}
867
8ae4f4f5 868static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 869 size_t bytes)
870{
8ae4f4f5 871 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 872
873 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 874
875 /* Advance the cursor page offset */
876
877 cursor->resid -= bytes;
5df521b1
AE		 878 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
879 if (!bytes || cursor->page_offset)
e766d7b5
AE		 880 return false; /* more bytes to process in the current page */
881
5df521b1 882 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 883
884 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 885 cursor->page_index++;
25aff7c5 886 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 887
888 return true;
889}
890
fe38a2b6 891/*
dd236fcb
AE		 892 * For a pagelist, a piece is whatever remains to be consumed in the
893 * first page in the list, or the front of the next page.
fe38a2b6 894 */
8ae4f4f5
AE		 895static void
896ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 897 size_t length)
fe38a2b6 898{
8ae4f4f5 899 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 900 struct ceph_pagelist *pagelist;
901 struct page *page;
902
dd236fcb 903 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 904
905 pagelist = data->pagelist;
906 BUG_ON(!pagelist);
1190bf06 907 BUG_ON(length > pagelist->length); /* short reads are OK */
25aff7c5
AE		 908
909 if (!length)
fe38a2b6
AE		 910 return; /* pagelist can be assigned but empty */
911
912 BUG_ON(list_empty(&pagelist->head));
913 page = list_first_entry(&pagelist->head, struct page, lru);
914
25aff7c5 915 cursor->resid = length;
fe38a2b6
AE		 916 cursor->page = page;
917 cursor->offset = 0;
25aff7c5 918 cursor->last_piece = length <= PAGE_SIZE;
fe38a2b6
AE		 919}
920
8ae4f4f5
AE		 921static struct page *
922ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
923 size_t *page_offset, size_t *length)
fe38a2b6 924{
8ae4f4f5 925 struct ceph_msg_data *data = cursor->data;
fe38a2b6 926 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 927
928 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
929
930 pagelist = data->pagelist;
931 BUG_ON(!pagelist);
932
933 BUG_ON(!cursor->page);
25aff7c5 934 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 935
5df521b1 936 /* offset of first page in pagelist is always 0 */
fe38a2b6 937 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 938 if (cursor->last_piece)
25aff7c5
AE		 939 *length = cursor->resid;
940 else
941 *length = PAGE_SIZE - *page_offset;
fe38a2b6 942
8ae4f4f5 943 return cursor->page;
fe38a2b6
AE		 944}
945
8ae4f4f5 946static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 947 size_t bytes)
fe38a2b6 948{
8ae4f4f5 949 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 950 struct ceph_pagelist *pagelist;
951
952 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
953
954 pagelist = data->pagelist;
955 BUG_ON(!pagelist);
25aff7c5
AE		 956
957 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 958 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
959
960 /* Advance the cursor offset */
961
25aff7c5 962 cursor->resid -= bytes;
fe38a2b6 963 cursor->offset += bytes;
5df521b1 964 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 965 if (!bytes || cursor->offset & ~PAGE_MASK)
966 return false; /* more bytes to process in the current page */
967
968 /* Move on to the next page */
969
970 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
971 cursor->page = list_entry_next(cursor->page, lru);
25aff7c5 972 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 973
974 return true;
975}
976
dd236fcb
AE		 977/*
978 * Message data is handled (sent or received) in pieces, where each
979 * piece resides on a single page. The network layer might not
980 * consume an entire piece at once. A data item's cursor keeps
981 * track of which piece is next to process and how much remains to
982 * be processed in that piece. It also tracks whether the current
983 * piece is the last one in the data item.
984 */
8ae4f4f5 985static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
dd236fcb 986{
8ae4f4f5 987 struct ceph_msg_data_cursor *cursor = &msg->cursor;
5240d9f9 988 struct ceph_msg_data *data;
8ae4f4f5 989
5240d9f9
AE		 990 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
991 cursor->data = data;
8ae4f4f5 992 switch (cursor->data->type) {
dd236fcb 993 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 994 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 995 break;
e766d7b5 996 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 997 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 998 break;
dd236fcb
AE		 999#ifdef CONFIG_BLOCK
1000 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1001 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1002 break;
dd236fcb 1003#endif /* CONFIG_BLOCK */
6aaa4511 1004 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1005 default:
1006 /* BUG(); */
1007 break;
1008 }
8ae4f4f5 1009 cursor->need_crc = true;
dd236fcb
AE		 1010}
1011
1012/*
1013 * Return the page containing the next piece to process for a given
1014 * data item, and supply the page offset and length of that piece.
1015 * Indicate whether this is the last piece in this data item.
1016 */
8ae4f4f5
AE		 1017static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1018 size_t *page_offset, size_t *length,
dd236fcb
AE		 1019 bool *last_piece)
1020{
1021 struct page *page;
1022
8ae4f4f5 1023 switch (cursor->data->type) {
dd236fcb 1024 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1025 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1026 break;
e766d7b5 1027 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1028 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1029 break;
dd236fcb
AE		 1030#ifdef CONFIG_BLOCK
1031 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1032 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1033 break;
dd236fcb 1034#endif /* CONFIG_BLOCK */
6aaa4511 1035 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1036 default:
1037 page = NULL;
1038 break;
1039 }
1040 BUG_ON(!page);
1041 BUG_ON(*page_offset + *length > PAGE_SIZE);
1042 BUG_ON(!*length);
1043 if (last_piece)
8ae4f4f5 1044 *last_piece = cursor->last_piece;
dd236fcb
AE		 1045
1046 return page;
1047}
1048
1049/*
1050 * Returns true if the result moves the cursor on to the next piece
1051 * of the data item.
1052 */
8ae4f4f5
AE		 1053static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1054 size_t bytes)
dd236fcb
AE		 1055{
1056 bool new_piece;
1057
25aff7c5 1058 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1059 switch (cursor->data->type) {
dd236fcb 1060 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1061 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1062 break;
e766d7b5 1063 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1064 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1065 break;
dd236fcb
AE		 1066#ifdef CONFIG_BLOCK
1067 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1068 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1069 break;
dd236fcb 1070#endif /* CONFIG_BLOCK */
6aaa4511 1071 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1072 default:
1073 BUG();
1074 break;
1075 }
8ae4f4f5 1076 cursor->need_crc = new_piece;
dd236fcb
AE		 1077
1078 return new_piece;
1079}
1080
98fa5dd8 1081static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1082{
739c905b 1083 BUG_ON(!msg);
25aff7c5 1084 BUG_ON(!data_len);
739c905b 1085
4c59b4a2 1086 /* Initialize data cursor */
fe38a2b6 1087
8ae4f4f5 1088 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1089}
1090
31b8006e
SW		 1091/*
1092 * Prepare footer for currently outgoing message, and finish things
1093 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1094 */
859eb799 1095static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1096{
1097 struct ceph_msg *m = con->out_msg;
859eb799 1098 int v = con->out_kvec_left;
31b8006e 1099
fd154f3c
AE		 1100 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1101
31b8006e
SW		 1102 dout("prepare_write_message_footer %p\n", con);
1103 con->out_kvec_is_msg = true;
1104 con->out_kvec[v].iov_base = &m->footer;
1105 con->out_kvec[v].iov_len = sizeof(m->footer);
1106 con->out_kvec_bytes += sizeof(m->footer);
1107 con->out_kvec_left++;
1108 con->out_more = m->more_to_follow;
c86a2930 1109 con->out_msg_done = true;
31b8006e
SW		 1110}
1111
1112/*
1113 * Prepare headers for the next outgoing message.
1114 */
1115static void prepare_write_message(struct ceph_connection *con)
1116{
1117 struct ceph_msg *m;
a9a0c51a 1118 u32 crc;
31b8006e 1119
e2200423 1120 con_out_kvec_reset(con);
31b8006e 1121 con->out_kvec_is_msg = true;
c86a2930 1122 con->out_msg_done = false;
31b8006e
SW		 1123
1124 /* Sneak an ack in there first? If we can get it into the same
1125 * TCP packet that's a good thing. */
1126 if (con->in_seq > con->in_seq_acked) {
1127 con->in_seq_acked = con->in_seq;
e2200423 1128 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1129 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1130 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1131 &con->out_temp_ack);
31b8006e
SW		 1132 }
1133
38941f80 1134 BUG_ON(list_empty(&con->out_queue));
859eb799 1135 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1136 con->out_msg = m;
38941f80 1137 BUG_ON(m->con != con);
4cf9d544
SW		 1138
1139 /* put message on sent list */
1140 ceph_msg_get(m);
1141 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1142
e84346b7
SW		 1143 /*
1144 * only assign outgoing seq # if we haven't sent this message
1145 * yet. if it is requeued, resend with it's original seq.
1146 */
1147 if (m->needs_out_seq) {
1148 m->hdr.seq = cpu_to_le64(++con->out_seq);
1149 m->needs_out_seq = false;
1150 }
98fa5dd8 1151 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1152
98fa5dd8 1153 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1154 m, con->out_seq, le16_to_cpu(m->hdr.type),
1155 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1156 m->data_length);
31b8006e
SW		 1157 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1158
1159 /* tag + hdr + front + middle */
e2200423
AE		 1160 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1161 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1162 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1163
31b8006e 1164 if (m->middle)
e2200423 1165 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1166 m->middle->vec.iov_base);
31b8006e
SW		 1167
1168 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1169 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1170 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1171 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1172
1173 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1174 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1175 if (m->middle) {
1176 crc = crc32c(0, m->middle->vec.iov_base,
1177 m->middle->vec.iov_len);
1178 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1179 } else
31b8006e 1180 con->out_msg->footer.middle_crc = 0;
739c905b 1181 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1182 le32_to_cpu(con->out_msg->footer.front_crc),
1183 le32_to_cpu(con->out_msg->footer.middle_crc));
1184
1185 /* is there a data payload? */
739c905b 1186 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1187 if (m->data_length) {
1188 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1189 con->out_more = 1; /* data + footer will follow */
1190 } else {
31b8006e 1191 /* no, queue up footer too and be done */
859eb799 1192 prepare_write_message_footer(con);
78625051 1193 }
31b8006e 1194
c9ffc77a 1195 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1196}
1197
1198/*
1199 * Prepare an ack.
1200 */
1201static void prepare_write_ack(struct ceph_connection *con)
1202{
1203 dout("prepare_write_ack %p %llu -> %llu\n", con,
1204 con->in_seq_acked, con->in_seq);
1205 con->in_seq_acked = con->in_seq;
1206
e2200423 1207 con_out_kvec_reset(con);
859eb799 1208
e2200423 1209 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1210
31b8006e 1211 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1212 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1213 &con->out_temp_ack);
1214
31b8006e 1215 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1216 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1217}
1218
3a23083b
SW		 1219/*
1220 * Prepare to share the seq during handshake
1221 */
1222static void prepare_write_seq(struct ceph_connection *con)
1223{
1224 dout("prepare_write_seq %p %llu -> %llu\n", con,
1225 con->in_seq_acked, con->in_seq);
1226 con->in_seq_acked = con->in_seq;
1227
1228 con_out_kvec_reset(con);
1229
1230 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1231 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1232 &con->out_temp_ack);
1233
1234 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1235}
1236
31b8006e
SW		 1237/*
1238 * Prepare to write keepalive byte.
1239 */
1240static void prepare_write_keepalive(struct ceph_connection *con)
1241{
1242 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1243 con_out_kvec_reset(con);
1244 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1245 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1246}
1247
1248/*
1249 * Connection negotiation.
1250 */
1251
dac1e716
AE		 1252static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1253 int *auth_proto)
4e7a5dcd 1254{
a3530df3 1255 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1256
1257 if (!con->ops->get_authorizer) {
1258 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1259 con->out_connect.authorizer_len = 0;
729796be 1260 return NULL;
b1c6b980
AE		 1261 }
1262
1263 /* Can't hold the mutex while getting authorizer */
ec302645 1264 mutex_unlock(&con->mutex);
dac1e716 1265 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1266 mutex_lock(&con->mutex);
4e7a5dcd 1267
a3530df3 1268 if (IS_ERR(auth))
729796be 1269 return auth;
8dacc7da 1270 if (con->state != CON_STATE_NEGOTIATING)
729796be 1271 return ERR_PTR(-EAGAIN);
0da5d703 1272
8f43fb53
AE		 1273 con->auth_reply_buf = auth->authorizer_reply_buf;
1274 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1275 return auth;
4e7a5dcd
SW		 1276}
1277
31b8006e
SW		 1278/*
1279 * We connected to a peer and are saying hello.
1280 */
e825a66d 1281static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1282{
e2200423
AE		 1283 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1284 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1285 &con->msgr->my_enc_addr);
eed0ef2c 1286
eed0ef2c 1287 con->out_more = 0;
c9ffc77a 1288 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1289}
1290
e825a66d 1291static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1292{
95c96174 1293 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1294 int proto;
dac1e716 1295 int auth_proto;
729796be 1296 struct ceph_auth_handshake *auth;
31b8006e
SW		 1297
1298 switch (con->peer_name.type) {
1299 case CEPH_ENTITY_TYPE_MON:
1300 proto = CEPH_MONC_PROTOCOL;
1301 break;
1302 case CEPH_ENTITY_TYPE_OSD:
1303 proto = CEPH_OSDC_PROTOCOL;
1304 break;
1305 case CEPH_ENTITY_TYPE_MDS:
1306 proto = CEPH_MDSC_PROTOCOL;
1307 break;
1308 default:
1309 BUG();
1310 }
1311
1312 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1313 con->connect_seq, global_seq, proto);
4e7a5dcd 1314
e825a66d 1315 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1316 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1317 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1318 con->out_connect.global_seq = cpu_to_le32(global_seq);
1319 con->out_connect.protocol_version = cpu_to_le32(proto);
1320 con->out_connect.flags = 0;
31b8006e 1321
dac1e716
AE		 1322 auth_proto = CEPH_AUTH_UNKNOWN;
1323 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1324 if (IS_ERR(auth))
1325 return PTR_ERR(auth);
3da54776 1326
dac1e716 1327 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1328 con->out_connect.authorizer_len = auth ?
1329 cpu_to_le32(auth->authorizer_buf_len) : 0;
1330
e2200423 1331 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1332 &con->out_connect);
1333 if (auth && auth->authorizer_buf_len)
e2200423 1334 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1335 auth->authorizer_buf);
859eb799 1336
31b8006e 1337 con->out_more = 0;
c9ffc77a 1338 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1339
e10c758e 1340 return 0;
31b8006e
SW		 1341}
1342
31b8006e
SW		 1343/*
1344 * write as much of pending kvecs to the socket as we can.
1345 * 1 -> done
1346 * 0 -> socket full, but more to do
1347 * <0 -> error
1348 */
1349static int write_partial_kvec(struct ceph_connection *con)
1350{
1351 int ret;
1352
1353 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1354 while (con->out_kvec_bytes > 0) {
1355 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1356 con->out_kvec_left, con->out_kvec_bytes,
1357 con->out_more);
1358 if (ret <= 0)
1359 goto out;
1360 con->out_kvec_bytes -= ret;
1361 if (con->out_kvec_bytes == 0)
1362 break; /* done */
f42299e6
AE		 1363
1364 /* account for full iov entries consumed */
1365 while (ret >= con->out_kvec_cur->iov_len) {
1366 BUG_ON(!con->out_kvec_left);
1367 ret -= con->out_kvec_cur->iov_len;
1368 con->out_kvec_cur++;
1369 con->out_kvec_left--;
1370 }
1371 /* and for a partially-consumed entry */
1372 if (ret) {
1373 con->out_kvec_cur->iov_len -= ret;
1374 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1375 }
1376 }
1377 con->out_kvec_left = 0;
1378 con->out_kvec_is_msg = false;
1379 ret = 1;
1380out:
1381 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1382 con->out_kvec_bytes, con->out_kvec_left, ret);
1383 return ret; /* done! */
1384}
1385
35b62808
AE		 1386static u32 ceph_crc32c_page(u32 crc, struct page *page,
1387 unsigned int page_offset,
1388 unsigned int length)
1389{
1390 char *kaddr;
1391
1392 kaddr = kmap(page);
1393 BUG_ON(kaddr == NULL);
1394 crc = crc32c(crc, kaddr + page_offset, length);
1395 kunmap(page);
1396
1397 return crc;
1398}
31b8006e
SW		 1399/*
1400 * Write as much message data payload as we can. If we finish, queue
1401 * up the footer.
1402 * 1 -> done, footer is now queued in out_kvec[].
1403 * 0 -> socket full, but more to do
1404 * <0 -> error
1405 */
34d2d200 1406static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1407{
1408 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1409 struct ceph_msg_data_cursor *cursor = &msg->cursor;
37675b0f 1410 bool do_datacrc = !con->msgr->nocrc;
f5db90bc 1411 u32 crc;
31b8006e 1412
859a35d5 1413 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1414
5240d9f9 1415 if (list_empty(&msg->data))
4c59b4a2
AE		 1416 return -EINVAL;
1417
5821bd8c
AE		 1418 /*
1419 * Iterate through each page that contains data to be
1420 * written, and send as much as possible for each.
1421 *
1422 * If we are calculating the data crc (the default), we will
1423 * need to map the page. If we have no pages, they have
1424 * been revoked, so use the zero page.
1425 */
f5db90bc 1426 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1427 while (cursor->resid) {
8a166d05 1428 struct page *page;
e387d525
AE		 1429 size_t page_offset;
1430 size_t length;
8a166d05 1431 bool last_piece;
8ea299bc 1432 bool need_crc;
f5db90bc 1433 int ret;
68b4476b 1434
8ae4f4f5 1435 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 1436 &last_piece);
e387d525 1437 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1438 length, last_piece);
f5db90bc
AE		 1439 if (ret <= 0) {
1440 if (do_datacrc)
1441 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1442
f5db90bc
AE		 1443 return ret;
1444 }
143334ff
AE		 1445 if (do_datacrc && cursor->need_crc)
1446 crc = ceph_crc32c_page(crc, page, page_offset, length);
8ae4f4f5 1447 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
31b8006e
SW		 1448 }
1449
34d2d200 1450 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1451
1452 /* prepare and queue up footer, too */
f5db90bc
AE		 1453 if (do_datacrc)
1454 msg->footer.data_crc = cpu_to_le32(crc);
1455 else
84ca8fc8 1456 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1457 con_out_kvec_reset(con);
859eb799 1458 prepare_write_message_footer(con);
f5db90bc
AE		 1459
1460 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1461}
1462
1463/*
1464 * write some zeros
1465 */
1466static int write_partial_skip(struct ceph_connection *con)
1467{
1468 int ret;
1469
1470 while (con->out_skip > 0) {
31739139 1471 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1472
e1dcb128 1473 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1474 if (ret <= 0)
1475 goto out;
1476 con->out_skip -= ret;
1477 }
1478 ret = 1;
1479out:
1480 return ret;
1481}
1482
1483/*
1484 * Prepare to read connection handshake, or an ack.
1485 */
eed0ef2c
SW		 1486static void prepare_read_banner(struct ceph_connection *con)
1487{
1488 dout("prepare_read_banner %p\n", con);
1489 con->in_base_pos = 0;
1490}
1491
31b8006e
SW		 1492static void prepare_read_connect(struct ceph_connection *con)
1493{
1494 dout("prepare_read_connect %p\n", con);
1495 con->in_base_pos = 0;
1496}
1497
1498static void prepare_read_ack(struct ceph_connection *con)
1499{
1500 dout("prepare_read_ack %p\n", con);
1501 con->in_base_pos = 0;
1502}
1503
3a23083b
SW		 1504static void prepare_read_seq(struct ceph_connection *con)
1505{
1506 dout("prepare_read_seq %p\n", con);
1507 con->in_base_pos = 0;
1508 con->in_tag = CEPH_MSGR_TAG_SEQ;
1509}
1510
31b8006e
SW		 1511static void prepare_read_tag(struct ceph_connection *con)
1512{
1513 dout("prepare_read_tag %p\n", con);
1514 con->in_base_pos = 0;
1515 con->in_tag = CEPH_MSGR_TAG_READY;
1516}
1517
1518/*
1519 * Prepare to read a message.
1520 */
1521static int prepare_read_message(struct ceph_connection *con)
1522{
1523 dout("prepare_read_message %p\n", con);
1524 BUG_ON(con->in_msg != NULL);
1525 con->in_base_pos = 0;
1526 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1527 return 0;
1528}
1529
1530
1531static int read_partial(struct ceph_connection *con,
fd51653f 1532 int end, int size, void *object)
31b8006e 1533{
e6cee71f
AE		 1534 while (con->in_base_pos < end) {
1535 int left = end - con->in_base_pos;
31b8006e
SW		 1536 int have = size - left;
1537 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1538 if (ret <= 0)
1539 return ret;
1540 con->in_base_pos += ret;
1541 }
1542 return 1;
1543}
1544
1545
1546/*
1547 * Read all or part of the connect-side handshake on a new connection
1548 */
eed0ef2c 1549static int read_partial_banner(struct ceph_connection *con)
31b8006e 1550{
fd51653f
AE		 1551 int size;
1552 int end;
1553 int ret;
31b8006e 1554
eed0ef2c 1555 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1556
1557 /* peer's banner */
fd51653f
AE		 1558 size = strlen(CEPH_BANNER);
1559 end = size;
1560 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1561 if (ret <= 0)
1562 goto out;
fd51653f
AE		 1563
1564 size = sizeof (con->actual_peer_addr);
1565 end += size;
1566 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1567 if (ret <= 0)
1568 goto out;
fd51653f
AE		 1569
1570 size = sizeof (con->peer_addr_for_me);
1571 end += size;
1572 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1573 if (ret <= 0)
1574 goto out;
fd51653f 1575
eed0ef2c
SW		 1576out:
1577 return ret;
1578}
1579
1580static int read_partial_connect(struct ceph_connection *con)
1581{
fd51653f
AE		 1582 int size;
1583 int end;
1584 int ret;
eed0ef2c
SW		 1585
1586 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1587
fd51653f
AE		 1588 size = sizeof (con->in_reply);
1589 end = size;
1590 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1591 if (ret <= 0)
1592 goto out;
fd51653f
AE		 1593
1594 size = le32_to_cpu(con->in_reply.authorizer_len);
1595 end += size;
1596 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1597 if (ret <= 0)
1598 goto out;
31b8006e 1599
4e7a5dcd
SW		 1600 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1601 con, (int)con->in_reply.tag,
1602 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1603 le32_to_cpu(con->in_reply.global_seq));
1604out:
1605 return ret;
eed0ef2c 1606
31b8006e
SW		 1607}
1608
1609/*
1610 * Verify the hello banner looks okay.
1611 */
1612static int verify_hello(struct ceph_connection *con)
1613{
1614 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1615 pr_err("connect to %s got bad banner\n",
3d14c5d2 1616 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1617 con->error_msg = "protocol error, bad banner";
1618 return -1;
1619 }
1620 return 0;
1621}
1622
1623static bool addr_is_blank(struct sockaddr_storage *ss)
1624{
1625 switch (ss->ss_family) {
1626 case AF_INET:
1627 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1628 case AF_INET6:
1629 return
1630 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1631 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1632 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1633 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1634 }
1635 return false;
1636}
1637
1638static int addr_port(struct sockaddr_storage *ss)
1639{
1640 switch (ss->ss_family) {
1641 case AF_INET:
f28bcfbe 1642 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1643 case AF_INET6:
f28bcfbe 1644 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1645 }
1646 return 0;
1647}
1648
1649static void addr_set_port(struct sockaddr_storage *ss, int p)
1650{
1651 switch (ss->ss_family) {
1652 case AF_INET:
1653 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1654 break;
31b8006e
SW		 1655 case AF_INET6:
1656 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1657 break;
31b8006e
SW		 1658 }
1659}
1660
ee3b56f2
NW		 1661/*
1662 * Unlike other *_pton function semantics, zero indicates success.
1663 */
1664static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1665 char delim, const char **ipend)
1666{
99f0f3b2
AE		 1667 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1668 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1669
1670 memset(ss, 0, sizeof(*ss));
1671
1672 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1673 ss->ss_family = AF_INET;
1674 return 0;
1675 }
1676
1677 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1678 ss->ss_family = AF_INET6;
1679 return 0;
1680 }
1681
1682 return -EINVAL;
1683}
1684
1685/*
1686 * Extract hostname string and resolve using kernel DNS facility.
1687 */
1688#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1689static int ceph_dns_resolve_name(const char *name, size_t namelen,
1690 struct sockaddr_storage *ss, char delim, const char **ipend)
1691{
1692 const char *end, *delim_p;
1693 char *colon_p, *ip_addr = NULL;
1694 int ip_len, ret;
1695
1696 /*
1697 * The end of the hostname occurs immediately preceding the delimiter or
1698 * the port marker (':') where the delimiter takes precedence.
1699 */
1700 delim_p = memchr(name, delim, namelen);
1701 colon_p = memchr(name, ':', namelen);
1702
1703 if (delim_p && colon_p)
1704 end = delim_p < colon_p ? delim_p : colon_p;
1705 else if (!delim_p && colon_p)
1706 end = colon_p;
1707 else {
1708 end = delim_p;
1709 if (!end) /* case: hostname:/ */
1710 end = name + namelen;
1711 }
1712
1713 if (end <= name)
1714 return -EINVAL;
1715
1716 /* do dns_resolve upcall */
1717 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1718 if (ip_len > 0)
1719 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1720 else
1721 ret = -ESRCH;
1722
1723 kfree(ip_addr);
1724
1725 *ipend = end;
1726
1727 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1728 ret, ret ? "failed" : ceph_pr_addr(ss));
1729
1730 return ret;
1731}
1732#else
1733static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1734 struct sockaddr_storage *ss, char delim, const char **ipend)
1735{
1736 return -EINVAL;
1737}
1738#endif
1739
1740/*
1741 * Parse a server name (IP or hostname). If a valid IP address is not found
1742 * then try to extract a hostname to resolve using userspace DNS upcall.
1743 */
1744static int ceph_parse_server_name(const char *name, size_t namelen,
1745 struct sockaddr_storage *ss, char delim, const char **ipend)
1746{
1747 int ret;
1748
1749 ret = ceph_pton(name, namelen, ss, delim, ipend);
1750 if (ret)
1751 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1752
1753 return ret;
1754}
1755
31b8006e
SW		 1756/*
1757 * Parse an ip[:port] list into an addr array. Use the default
1758 * monitor port if a port isn't specified.
1759 */
1760int ceph_parse_ips(const char *c, const char *end,
1761 struct ceph_entity_addr *addr,
1762 int max_count, int *count)
1763{
ee3b56f2 1764 int i, ret = -EINVAL;
31b8006e
SW		 1765 const char *p = c;
1766
1767 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1768 for (i = 0; i < max_count; i++) {
1769 const char *ipend;
1770 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1771 int port;
39139f64
SW		 1772 char delim = ',';
1773
1774 if (*p == '[') {
1775 delim = ']';
1776 p++;
1777 }
31b8006e 1778
ee3b56f2
NW		 1779 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1780 if (ret)
31b8006e 1781 goto bad;
ee3b56f2
NW		 1782 ret = -EINVAL;
1783
31b8006e
SW		 1784 p = ipend;
1785
39139f64
SW		 1786 if (delim == ']') {
1787 if (*p != ']') {
1788 dout("missing matching ']'\n");
1789 goto bad;
1790 }
1791 p++;
1792 }
1793
31b8006e
SW		 1794 /* port? */
1795 if (p < end && *p == ':') {
1796 port = 0;
1797 p++;
1798 while (p < end && *p >= '0' && *p <= '9') {
1799 port = (port * 10) + (*p - '0');
1800 p++;
1801 }
1802 if (port > 65535 || port == 0)
1803 goto bad;
1804 } else {
1805 port = CEPH_MON_PORT;
1806 }
1807
1808 addr_set_port(ss, port);
1809
3d14c5d2 1810 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1811
1812 if (p == end)
1813 break;
1814 if (*p != ',')
1815 goto bad;
1816 p++;
1817 }
1818
1819 if (p != end)
1820 goto bad;
1821
1822 if (count)
1823 *count = i + 1;
1824 return 0;
1825
1826bad:
39139f64 1827 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1828 return ret;
31b8006e 1829}
3d14c5d2 1830EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1831
eed0ef2c 1832static int process_banner(struct ceph_connection *con)
31b8006e 1833{
eed0ef2c 1834 dout("process_banner on %p\n", con);
31b8006e
SW		 1835
1836 if (verify_hello(con) < 0)
1837 return -1;
1838
63f2d211
SW		 1839 ceph_decode_addr(&con->actual_peer_addr);
1840 ceph_decode_addr(&con->peer_addr_for_me);
1841
31b8006e
SW		 1842 /*
1843 * Make sure the other end is who we wanted. note that the other
1844 * end may not yet know their ip address, so if it's 0.0.0.0, give
1845 * them the benefit of the doubt.
1846 */
103e2d3a
SW		 1847 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1848 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1849 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1850 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
cd84db6e 1851 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
3d14c5d2 1852 ceph_pr_addr(&con->peer_addr.in_addr),
cd84db6e 1853 (int)le32_to_cpu(con->peer_addr.nonce),
3d14c5d2 1854 ceph_pr_addr(&con->actual_peer_addr.in_addr),
cd84db6e 1855 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1856 con->error_msg = "wrong peer at address";
31b8006e
SW		 1857 return -1;
1858 }
1859
1860 /*
1861 * did we learn our address?
1862 */
1863 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1864 int port = addr_port(&con->msgr->inst.addr.in_addr);
1865
1866 memcpy(&con->msgr->inst.addr.in_addr,
1867 &con->peer_addr_for_me.in_addr,
1868 sizeof(con->peer_addr_for_me.in_addr));
1869 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1870 encode_my_addr(con->msgr);
eed0ef2c 1871 dout("process_banner learned my addr is %s\n",
3d14c5d2 1872 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1873 }
1874
eed0ef2c
SW		 1875 return 0;
1876}
1877
1878static int process_connect(struct ceph_connection *con)
1879{
3d14c5d2
YS		 1880 u64 sup_feat = con->msgr->supported_features;
1881 u64 req_feat = con->msgr->required_features;
04a419f9 1882 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 1883 int ret;
04a419f9 1884
eed0ef2c
SW		 1885 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1886
31b8006e 1887 switch (con->in_reply.tag) {
04a419f9
SW		 1888 case CEPH_MSGR_TAG_FEATURES:
1889 pr_err("%s%lld %s feature set mismatch,"
1890 " my %llx < server's %llx, missing %llx\n",
1891 ENTITY_NAME(con->peer_name),
3d14c5d2 1892 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1893 sup_feat, server_feat, server_feat & ~sup_feat);
1894 con->error_msg = "missing required protocol features";
0fa6ebc6 1895 reset_connection(con);
04a419f9
SW		 1896 return -1;
1897
31b8006e 1898 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 1899 pr_err("%s%lld %s protocol version mismatch,"
1900 " my %d != server's %d\n",
1901 ENTITY_NAME(con->peer_name),
3d14c5d2 1902 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 1903 le32_to_cpu(con->out_connect.protocol_version),
1904 le32_to_cpu(con->in_reply.protocol_version));
1905 con->error_msg = "protocol version mismatch";
0fa6ebc6 1906 reset_connection(con);
31b8006e
SW		 1907 return -1;
1908
4e7a5dcd
SW		 1909 case CEPH_MSGR_TAG_BADAUTHORIZER:
1910 con->auth_retry++;
1911 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1912 con->auth_retry);
1913 if (con->auth_retry == 2) {
1914 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 1915 return -1;
1916 }
6d4221b5 1917 con_out_kvec_reset(con);
e825a66d 1918 ret = prepare_write_connect(con);
0da5d703
SW		 1919 if (ret < 0)
1920 return ret;
63733a0f 1921 prepare_read_connect(con);
4e7a5dcd 1922 break;
31b8006e
SW		 1923
1924 case CEPH_MSGR_TAG_RESETSESSION:
1925 /*
1926 * If we connected with a large connect_seq but the peer
1927 * has no record of a session with us (no connection, or
1928 * connect_seq == 0), they will send RESETSESION to indicate
1929 * that they must have reset their session, and may have
1930 * dropped messages.
1931 */
1932 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 1933 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 1934 pr_err("%s%lld %s connection reset\n",
1935 ENTITY_NAME(con->peer_name),
3d14c5d2 1936 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 1937 reset_connection(con);
6d4221b5 1938 con_out_kvec_reset(con);
5a0f8fdd
AE		 1939 ret = prepare_write_connect(con);
1940 if (ret < 0)
1941 return ret;
31b8006e
SW		 1942 prepare_read_connect(con);
1943
1944 /* Tell ceph about it. */
ec302645 1945 mutex_unlock(&con->mutex);
31b8006e
SW		 1946 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1947 if (con->ops->peer_reset)
1948 con->ops->peer_reset(con);
ec302645 1949 mutex_lock(&con->mutex);
8dacc7da 1950 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 1951 return -EAGAIN;
31b8006e
SW		 1952 break;
1953
1954 case CEPH_MSGR_TAG_RETRY_SESSION:
1955 /*
1956 * If we sent a smaller connect_seq than the peer has, try
1957 * again with a larger value.
1958 */
5bdca4e0 1959 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 1960 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 1961 le32_to_cpu(con->in_reply.connect_seq));
1962 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 1963 con_out_kvec_reset(con);
5a0f8fdd
AE		 1964 ret = prepare_write_connect(con);
1965 if (ret < 0)
1966 return ret;
31b8006e
SW		 1967 prepare_read_connect(con);
1968 break;
1969
1970 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1971 /*
1972 * If we sent a smaller global_seq than the peer has, try
1973 * again with a larger value.
1974 */
eed0ef2c 1975 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 1976 con->peer_global_seq,
5bdca4e0 1977 le32_to_cpu(con->in_reply.global_seq));
31b8006e 1978 get_global_seq(con->msgr,
5bdca4e0 1979 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 1980 con_out_kvec_reset(con);
5a0f8fdd
AE		 1981 ret = prepare_write_connect(con);
1982 if (ret < 0)
1983 return ret;
31b8006e
SW		 1984 prepare_read_connect(con);
1985 break;
1986
3a23083b 1987 case CEPH_MSGR_TAG_SEQ:
31b8006e 1988 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 1989 if (req_feat & ~server_feat) {
1990 pr_err("%s%lld %s protocol feature mismatch,"
1991 " my required %llx > server's %llx, need %llx\n",
1992 ENTITY_NAME(con->peer_name),
3d14c5d2 1993 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1994 req_feat, server_feat, req_feat & ~server_feat);
1995 con->error_msg = "missing required protocol features";
0fa6ebc6 1996 reset_connection(con);
04a419f9
SW		 1997 return -1;
1998 }
8dacc7da 1999
122070a2 2000 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2001 con->state = CON_STATE_OPEN;
20e55c4c 2002 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2003 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2004 con->connect_seq++;
aba558e2 2005 con->peer_features = server_feat;
31b8006e
SW		 2006 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2007 con->peer_global_seq,
2008 le32_to_cpu(con->in_reply.connect_seq),
2009 con->connect_seq);
2010 WARN_ON(con->connect_seq !=
2011 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2012
2013 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2014 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2015
85effe18 2016 con->delay = 0; /* reset backoff memory */
92ac41d0 2017
3a23083b
SW		 2018 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2019 prepare_write_seq(con);
2020 prepare_read_seq(con);
2021 } else {
2022 prepare_read_tag(con);
2023 }
31b8006e
SW		 2024 break;
2025
2026 case CEPH_MSGR_TAG_WAIT:
2027 /*
2028 * If there is a connection race (we are opening
2029 * connections to each other), one of us may just have
2030 * to WAIT. This shouldn't happen if we are the
2031 * client.
2032 */
04177882
SW		 2033 pr_err("process_connect got WAIT as client\n");
2034 con->error_msg = "protocol error, got WAIT as client";
2035 return -1;
31b8006e
SW		 2036
2037 default:
2038 pr_err("connect protocol error, will retry\n");
2039 con->error_msg = "protocol error, garbage tag during connect";
2040 return -1;
2041 }
2042 return 0;
2043}
2044
2045
2046/*
2047 * read (part of) an ack
2048 */
2049static int read_partial_ack(struct ceph_connection *con)
2050{
fd51653f
AE		 2051 int size = sizeof (con->in_temp_ack);
2052 int end = size;
31b8006e 2053
fd51653f 2054 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2055}
2056
31b8006e
SW		 2057/*
2058 * We can finally discard anything that's been acked.
2059 */
2060static void process_ack(struct ceph_connection *con)
2061{
2062 struct ceph_msg *m;
2063 u64 ack = le64_to_cpu(con->in_temp_ack);
2064 u64 seq;
2065
31b8006e
SW		 2066 while (!list_empty(&con->out_sent)) {
2067 m = list_first_entry(&con->out_sent, struct ceph_msg,
2068 list_head);
2069 seq = le64_to_cpu(m->hdr.seq);
2070 if (seq > ack)
2071 break;
2072 dout("got ack for seq %llu type %d at %p\n", seq,
2073 le16_to_cpu(m->hdr.type), m);
4cf9d544 2074 m->ack_stamp = jiffies;
31b8006e
SW		 2075 ceph_msg_remove(m);
2076 }
31b8006e
SW		 2077 prepare_read_tag(con);
2078}
2079
2080
2450418c 2081static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2082 struct kvec *section,
2083 unsigned int sec_len, u32 *crc)
2450418c 2084{
68b4476b 2085 int ret, left;
2450418c
YS		 2086
2087 BUG_ON(!section);
2088
2089 while (section->iov_len < sec_len) {
2090 BUG_ON(section->iov_base == NULL);
2091 left = sec_len - section->iov_len;
2092 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2093 section->iov_len, left);
2094 if (ret <= 0)
2095 return ret;
2096 section->iov_len += ret;
2450418c 2097 }
fe3ad593
AE		 2098 if (section->iov_len == sec_len)
2099 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2100
2450418c
YS		 2101 return 1;
2102}
31b8006e 2103
34d2d200
AE		 2104static int read_partial_msg_data(struct ceph_connection *con)
2105{
2106 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2107 struct ceph_msg_data_cursor *cursor = &msg->cursor;
34d2d200 2108 const bool do_datacrc = !con->msgr->nocrc;
686be208
AE		 2109 struct page *page;
2110 size_t page_offset;
2111 size_t length;
f5db90bc 2112 u32 crc = 0;
34d2d200
AE		 2113 int ret;
2114
2115 BUG_ON(!msg);
5240d9f9 2116 if (list_empty(&msg->data))
4c59b4a2 2117 return -EIO;
34d2d200 2118
f5db90bc
AE		 2119 if (do_datacrc)
2120 crc = con->in_data_crc;
643c68a4 2121 while (cursor->resid) {
8ae4f4f5 2122 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 2123 NULL);
686be208 2124 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2125 if (ret <= 0) {
2126 if (do_datacrc)
2127 con->in_data_crc = crc;
2128
686be208 2129 return ret;
f5db90bc 2130 }
686be208
AE		 2131
2132 if (do_datacrc)
f5db90bc 2133 crc = ceph_crc32c_page(crc, page, page_offset, ret);
8ae4f4f5 2134 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
34d2d200 2135 }
f5db90bc
AE		 2136 if (do_datacrc)
2137 con->in_data_crc = crc;
34d2d200
AE		 2138
2139 return 1; /* must return > 0 to indicate success */
2140}
2141
31b8006e
SW		 2142/*
2143 * read (part of) a message.
2144 */
686be208
AE		 2145static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2146
31b8006e
SW		 2147static int read_partial_message(struct ceph_connection *con)
2148{
2149 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2150 int size;
2151 int end;
31b8006e 2152 int ret;
95c96174 2153 unsigned int front_len, middle_len, data_len;
37675b0f 2154 bool do_datacrc = !con->msgr->nocrc;
ae18756b 2155 u64 seq;
fe3ad593 2156 u32 crc;
31b8006e
SW		 2157
2158 dout("read_partial_message con %p msg %p\n", con, m);
2159
2160 /* header */
fd51653f
AE		 2161 size = sizeof (con->in_hdr);
2162 end = size;
2163 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2164 if (ret <= 0)
2165 return ret;
fe3ad593
AE		 2166
2167 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2168 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2169 pr_err("read_partial_message bad hdr "
2170 " crc %u != expected %u\n",
2171 crc, con->in_hdr.crc);
2172 return -EBADMSG;
2173 }
2174
31b8006e
SW		 2175 front_len = le32_to_cpu(con->in_hdr.front_len);
2176 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2177 return -EIO;
2178 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2179 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2180 return -EIO;
2181 data_len = le32_to_cpu(con->in_hdr.data_len);
2182 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2183 return -EIO;
2184
ae18756b
SW		 2185 /* verify seq# */
2186 seq = le64_to_cpu(con->in_hdr.seq);
2187 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2188 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2189 ENTITY_NAME(con->peer_name),
3d14c5d2 2190 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2191 seq, con->in_seq + 1);
2192 con->in_base_pos = -front_len - middle_len - data_len -
2193 sizeof(m->footer);
2194 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2195 return 0;
2196 } else if ((s64)seq - (s64)con->in_seq > 1) {
2197 pr_err("read_partial_message bad seq %lld expected %lld\n",
2198 seq, con->in_seq + 1);
2199 con->error_msg = "bad message sequence # for incoming message";
2200 return -EBADMSG;
2201 }
2202
31b8006e
SW		 2203 /* allocate message? */
2204 if (!con->in_msg) {
4740a623
SW		 2205 int skip = 0;
2206
31b8006e 2207 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2208 front_len, data_len);
4740a623
SW		 2209 ret = ceph_con_in_msg_alloc(con, &skip);
2210 if (ret < 0)
2211 return ret;
f759ebb9
AE		 2212
2213 BUG_ON(!con->in_msg ^ skip);
2214 if (con->in_msg && data_len > con->in_msg->data_length) {
2215 pr_warning("%s skipping long message (%u > %zd)\n",
2216 __func__, data_len, con->in_msg->data_length);
2217 ceph_msg_put(con->in_msg);
2218 con->in_msg = NULL;
2219 skip = 1;
2220 }
2450418c 2221 if (skip) {
31b8006e 2222 /* skip this message */
a79832f2 2223 dout("alloc_msg said skip message\n");
31b8006e
SW		 2224 con->in_base_pos = -front_len - middle_len - data_len -
2225 sizeof(m->footer);
2226 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2227 con->in_seq++;
31b8006e
SW		 2228 return 0;
2229 }
38941f80 2230
4740a623 2231 BUG_ON(!con->in_msg);
38941f80 2232 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2233 m = con->in_msg;
2234 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2235 if (m->middle)
2236 m->middle->vec.iov_len = 0;
9d7f0f13 2237
78625051 2238 /* prepare for data payload, if any */
a4107026 2239
78625051 2240 if (data_len)
98fa5dd8 2241 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2242 }
2243
2244 /* front */
2450418c
YS		 2245 ret = read_partial_message_section(con, &m->front, front_len,
2246 &con->in_front_crc);
2247 if (ret <= 0)
2248 return ret;
31b8006e
SW		 2249
2250 /* middle */
2450418c 2251 if (m->middle) {
213c99ee
SW		 2252 ret = read_partial_message_section(con, &m->middle->vec,
2253 middle_len,
2450418c 2254 &con->in_middle_crc);
31b8006e
SW		 2255 if (ret <= 0)
2256 return ret;
31b8006e
SW		 2257 }
2258
2259 /* (page) data */
34d2d200
AE		 2260 if (data_len) {
2261 ret = read_partial_msg_data(con);
2262 if (ret <= 0)
2263 return ret;
31b8006e
SW		 2264 }
2265
31b8006e 2266 /* footer */
fd51653f
AE		 2267 size = sizeof (m->footer);
2268 end += size;
2269 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2270 if (ret <= 0)
2271 return ret;
2272
31b8006e
SW		 2273 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2274 m, front_len, m->footer.front_crc, middle_len,
2275 m->footer.middle_crc, data_len, m->footer.data_crc);
2276
2277 /* crc ok? */
2278 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2279 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2280 m, con->in_front_crc, m->footer.front_crc);
2281 return -EBADMSG;
2282 }
2283 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2284 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2285 m, con->in_middle_crc, m->footer.middle_crc);
2286 return -EBADMSG;
2287 }
bca064d2 2288 if (do_datacrc &&
31b8006e
SW		 2289 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2290 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2291 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2292 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2293 return -EBADMSG;
2294 }
2295
2296 return 1; /* done! */
2297}
2298
2299/*
2300 * Process message. This happens in the worker thread. The callback should
2301 * be careful not to do anything that waits on other incoming messages or it
2302 * may deadlock.
2303 */
2304static void process_message(struct ceph_connection *con)
2305{
5e095e8b 2306 struct ceph_msg *msg;
31b8006e 2307
38941f80
AE		 2308 BUG_ON(con->in_msg->con != con);
2309 con->in_msg->con = NULL;
5e095e8b 2310 msg = con->in_msg;
31b8006e 2311 con->in_msg = NULL;
36eb71aa 2312 con->ops->put(con);
31b8006e
SW		 2313
2314 /* if first message, set peer_name */
2315 if (con->peer_name.type == 0)
dbad185d 2316 con->peer_name = msg->hdr.src;
31b8006e 2317
31b8006e 2318 con->in_seq++;
ec302645 2319 mutex_unlock(&con->mutex);
31b8006e
SW		 2320
2321 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2322 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2323 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2324 le16_to_cpu(msg->hdr.type),
2325 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2326 le32_to_cpu(msg->hdr.front_len),
2327 le32_to_cpu(msg->hdr.data_len),
2328 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2329 con->ops->dispatch(con, msg);
ec302645
SW		 2330
2331 mutex_lock(&con->mutex);
31b8006e
SW		 2332}
2333
2334
2335/*
2336 * Write something to the socket. Called in a worker thread when the
2337 * socket appears to be writeable and we have something ready to send.
2338 */
2339static int try_write(struct ceph_connection *con)
2340{
31b8006e
SW		 2341 int ret = 1;
2342
d59315ca 2343 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2344
31b8006e
SW		 2345more:
2346 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2347
2348 /* open the socket first? */
8dacc7da
SW		 2349 if (con->state == CON_STATE_PREOPEN) {
2350 BUG_ON(con->sock);
2351 con->state = CON_STATE_CONNECTING;
a5988c49 2352
e2200423 2353 con_out_kvec_reset(con);
e825a66d 2354 prepare_write_banner(con);
eed0ef2c 2355 prepare_read_banner(con);
31b8006e 2356
cf3e5c40 2357 BUG_ON(con->in_msg);
31b8006e
SW		 2358 con->in_tag = CEPH_MSGR_TAG_READY;
2359 dout("try_write initiating connect on %p new state %lu\n",
2360 con, con->state);
41617d0c
AE		 2361 ret = ceph_tcp_connect(con);
2362 if (ret < 0) {
31b8006e 2363 con->error_msg = "connect error";
31b8006e
SW		 2364 goto out;
2365 }
2366 }
2367
2368more_kvec:
2369 /* kvec data queued? */
2370 if (con->out_skip) {
2371 ret = write_partial_skip(con);
2372 if (ret <= 0)
42961d23 2373 goto out;
31b8006e
SW		 2374 }
2375 if (con->out_kvec_left) {
2376 ret = write_partial_kvec(con);
2377 if (ret <= 0)
42961d23 2378 goto out;
31b8006e
SW		 2379 }
2380
2381 /* msg pages? */
2382 if (con->out_msg) {
c86a2930
SW		 2383 if (con->out_msg_done) {
2384 ceph_msg_put(con->out_msg);
2385 con->out_msg = NULL; /* we're done with this one */
2386 goto do_next;
2387 }
2388
34d2d200 2389 ret = write_partial_message_data(con);
31b8006e
SW		 2390 if (ret == 1)
2391 goto more_kvec; /* we need to send the footer, too! */
2392 if (ret == 0)
42961d23 2393 goto out;
31b8006e 2394 if (ret < 0) {
34d2d200 2395 dout("try_write write_partial_message_data err %d\n",
31b8006e 2396 ret);
42961d23 2397 goto out;
31b8006e
SW		 2398 }
2399 }
2400
c86a2930 2401do_next:
8dacc7da 2402 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2403 /* is anything else pending? */
2404 if (!list_empty(&con->out_queue)) {
2405 prepare_write_message(con);
2406 goto more;
2407 }
2408 if (con->in_seq > con->in_seq_acked) {
2409 prepare_write_ack(con);
2410 goto more;
2411 }
c9ffc77a 2412 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2413 prepare_write_keepalive(con);
2414 goto more;
2415 }
2416 }
2417
2418 /* Nothing to do! */
c9ffc77a 2419 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2420 dout("try_write nothing else to write.\n");
31b8006e
SW		 2421 ret = 0;
2422out:
42961d23 2423 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2424 return ret;
2425}
2426
2427
2428
2429/*
2430 * Read what we can from the socket.
2431 */
2432static int try_read(struct ceph_connection *con)
2433{
31b8006e
SW		 2434 int ret = -1;
2435
8dacc7da
SW		 2436more:
2437 dout("try_read start on %p state %lu\n", con, con->state);
2438 if (con->state != CON_STATE_CONNECTING &&
2439 con->state != CON_STATE_NEGOTIATING &&
2440 con->state != CON_STATE_OPEN)
31b8006e
SW		 2441 return 0;
2442
8dacc7da 2443 BUG_ON(!con->sock);
ec302645 2444
31b8006e
SW		 2445 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2446 con->in_base_pos);
0da5d703 2447
8dacc7da 2448 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2449 dout("try_read connecting\n");
2450 ret = read_partial_banner(con);
2451 if (ret <= 0)
ab166d5a 2452 goto out;
7593af92
AE		 2453 ret = process_banner(con);
2454 if (ret < 0)
2455 goto out;
2456
8dacc7da 2457 con->state = CON_STATE_NEGOTIATING;
7593af92 2458
6d4221b5
JS		 2459 /*
2460 * Received banner is good, exchange connection info.
2461 * Do not reset out_kvec, as sending our banner raced
2462 * with receiving peer banner after connect completed.
2463 */
7593af92
AE		 2464 ret = prepare_write_connect(con);
2465 if (ret < 0)
2466 goto out;
2467 prepare_read_connect(con);
2468
2469 /* Send connection info before awaiting response */
0da5d703
SW		 2470 goto out;
2471 }
2472
8dacc7da 2473 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2474 dout("try_read negotiating\n");
31b8006e
SW		 2475 ret = read_partial_connect(con);
2476 if (ret <= 0)
31b8006e 2477 goto out;
98bdb0aa
SW		 2478 ret = process_connect(con);
2479 if (ret < 0)
2480 goto out;
31b8006e
SW		 2481 goto more;
2482 }
2483
122070a2 2484 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2485
31b8006e
SW		 2486 if (con->in_base_pos < 0) {
2487 /*
2488 * skipping + discarding content.
2489 *
2490 * FIXME: there must be a better way to do this!
2491 */
84495f49
AE		 2492 static char buf[SKIP_BUF_SIZE];
2493 int skip = min((int) sizeof (buf), -con->in_base_pos);
2494
31b8006e
SW		 2495 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2496 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2497 if (ret <= 0)
98bdb0aa 2498 goto out;
31b8006e
SW		 2499 con->in_base_pos += ret;
2500 if (con->in_base_pos)
2501 goto more;
2502 }
2503 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2504 /*
2505 * what's next?
2506 */
2507 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2508 if (ret <= 0)
98bdb0aa 2509 goto out;
31b8006e
SW		 2510 dout("try_read got tag %d\n", (int)con->in_tag);
2511 switch (con->in_tag) {
2512 case CEPH_MSGR_TAG_MSG:
2513 prepare_read_message(con);
2514 break;
2515 case CEPH_MSGR_TAG_ACK:
2516 prepare_read_ack(con);
2517 break;
2518 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2519 con_close_socket(con);
2520 con->state = CON_STATE_CLOSED;
98bdb0aa 2521 goto out;
31b8006e
SW		 2522 default:
2523 goto bad_tag;
2524 }
2525 }
2526 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2527 ret = read_partial_message(con);
2528 if (ret <= 0) {
2529 switch (ret) {
2530 case -EBADMSG:
2531 con->error_msg = "bad crc";
2532 ret = -EIO;
98bdb0aa 2533 break;
31b8006e
SW		 2534 case -EIO:
2535 con->error_msg = "io error";
98bdb0aa 2536 break;
31b8006e 2537 }
98bdb0aa 2538 goto out;
31b8006e
SW		 2539 }
2540 if (con->in_tag == CEPH_MSGR_TAG_READY)
2541 goto more;
2542 process_message(con);
7b862e07
SW		 2543 if (con->state == CON_STATE_OPEN)
2544 prepare_read_tag(con);
31b8006e
SW		 2545 goto more;
2546 }
3a23083b
SW		 2547 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2548 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2549 /*
2550 * the final handshake seq exchange is semantically
2551 * equivalent to an ACK
2552 */
31b8006e
SW		 2553 ret = read_partial_ack(con);
2554 if (ret <= 0)
98bdb0aa 2555 goto out;
31b8006e
SW		 2556 process_ack(con);
2557 goto more;
2558 }
2559
31b8006e 2560out:
98bdb0aa 2561 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2562 return ret;
2563
2564bad_tag:
2565 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2566 con->error_msg = "protocol error, garbage tag";
2567 ret = -1;
2568 goto out;
2569}
2570
2571
2572/*
802c6d96
AE		 2573 * Atomically queue work on a connection after the specified delay.
2574 * Bump @con reference to avoid races with connection teardown.
2575 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2576 */
802c6d96 2577static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2578{
31b8006e 2579 if (!con->ops->get(con)) {
802c6d96
AE		 2580 dout("%s %p ref count 0\n", __func__, con);
2581
2582 return -ENOENT;
31b8006e
SW		 2583 }
2584
802c6d96
AE		 2585 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2586 dout("%s %p - already queued\n", __func__, con);
31b8006e 2587 con->ops->put(con);
802c6d96
AE		 2588
2589 return -EBUSY;
31b8006e 2590 }
802c6d96
AE		 2591
2592 dout("%s %p %lu\n", __func__, con, delay);
2593
2594 return 0;
2595}
2596
2597static void queue_con(struct ceph_connection *con)
2598{
2599 (void) queue_con_delay(con, 0);
31b8006e
SW		 2600}
2601
7bb21d68
AE		 2602static bool con_sock_closed(struct ceph_connection *con)
2603{
c9ffc77a 2604 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2605 return false;
2606
2607#define CASE(x) \
2608 case CON_STATE_ ## x: \
2609 con->error_msg = "socket closed (con state " #x ")"; \
2610 break;
2611
2612 switch (con->state) {
2613 CASE(CLOSED);
2614 CASE(PREOPEN);
2615 CASE(CONNECTING);
2616 CASE(NEGOTIATING);
2617 CASE(OPEN);
2618 CASE(STANDBY);
2619 default:
2620 pr_warning("%s con %p unrecognized state %lu\n",
2621 __func__, con, con->state);
2622 con->error_msg = "unrecognized con state";
2623 BUG();
2624 break;
2625 }
2626#undef CASE
2627
2628 return true;
2629}
2630
f20a39fd
AE		 2631static bool con_backoff(struct ceph_connection *con)
2632{
2633 int ret;
2634
2635 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2636 return false;
2637
2638 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2639 if (ret) {
2640 dout("%s: con %p FAILED to back off %lu\n", __func__,
2641 con, con->delay);
2642 BUG_ON(ret == -ENOENT);
2643 con_flag_set(con, CON_FLAG_BACKOFF);
2644 }
2645
2646 return true;
2647}
2648
93209264
AE		 2649/* Finish fault handling; con->mutex must *not* be held here */
2650
2651static void con_fault_finish(struct ceph_connection *con)
2652{
2653 /*
2654 * in case we faulted due to authentication, invalidate our
2655 * current tickets so that we can get new ones.
2656 */
2657 if (con->auth_retry && con->ops->invalidate_authorizer) {
2658 dout("calling invalidate_authorizer()\n");
2659 con->ops->invalidate_authorizer(con);
2660 }
2661
2662 if (con->ops->fault)
2663 con->ops->fault(con);
2664}
2665
31b8006e
SW		 2666/*
2667 * Do some work on a connection. Drop a connection ref when we're done.
2668 */
2669static void con_work(struct work_struct *work)
2670{
2671 struct ceph_connection *con = container_of(work, struct ceph_connection,
2672 work.work);
49659416 2673 bool fault;
31b8006e 2674
9dd4658d 2675 mutex_lock(&con->mutex);
49659416
AE		 2676 while (true) {
2677 int ret;
31b8006e 2678
49659416
AE		 2679 if ((fault = con_sock_closed(con))) {
2680 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2681 break;
2682 }
2683 if (con_backoff(con)) {
2684 dout("%s: con %p BACKOFF\n", __func__, con);
2685 break;
2686 }
2687 if (con->state == CON_STATE_STANDBY) {
2688 dout("%s: con %p STANDBY\n", __func__, con);
2689 break;
2690 }
2691 if (con->state == CON_STATE_CLOSED) {
2692 dout("%s: con %p CLOSED\n", __func__, con);
2693 BUG_ON(con->sock);
2694 break;
2695 }
2696 if (con->state == CON_STATE_PREOPEN) {
2697 dout("%s: con %p PREOPEN\n", __func__, con);
2698 BUG_ON(con->sock);
2699 }
0da5d703 2700
49659416
AE		 2701 ret = try_read(con);
2702 if (ret < 0) {
2703 if (ret == -EAGAIN)
2704 continue;
2705 con->error_msg = "socket error on read";
2706 fault = true;
2707 break;
2708 }
2709
2710 ret = try_write(con);
2711 if (ret < 0) {
2712 if (ret == -EAGAIN)
2713 continue;
2714 con->error_msg = "socket error on write";
2715 fault = true;
2716 }
2717
2718 break; /* If we make it to here, we're done */
3a140a0d 2719 }
b6e7b6a1
AE		 2720 if (fault)
2721 con_fault(con);
9dd4658d 2722 mutex_unlock(&con->mutex);
0da5d703 2723
b6e7b6a1
AE		 2724 if (fault)
2725 con_fault_finish(con);
2726
2727 con->ops->put(con);
31b8006e
SW		 2728}
2729
31b8006e
SW		 2730/*
2731 * Generic error/fault handler. A retry mechanism is used with
2732 * exponential backoff
2733 */
93209264 2734static void con_fault(struct ceph_connection *con)
31b8006e 2735{
28362986 2736 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
3d14c5d2 2737 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2738 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2739 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2740
122070a2 2741 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2742 con->state != CON_STATE_NEGOTIATING &&
2743 con->state != CON_STATE_OPEN);
ec302645 2744
31b8006e 2745 con_close_socket(con);
5e095e8b 2746
c9ffc77a 2747 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2748 dout("fault on LOSSYTX channel, marking CLOSED\n");
2749 con->state = CON_STATE_CLOSED;
93209264 2750 return;
3b5ede07
SW		 2751 }
2752
5e095e8b 2753 if (con->in_msg) {
38941f80
AE		 2754 BUG_ON(con->in_msg->con != con);
2755 con->in_msg->con = NULL;
5e095e8b
SW		 2756 ceph_msg_put(con->in_msg);
2757 con->in_msg = NULL;
36eb71aa 2758 con->ops->put(con);
5e095e8b 2759 }
31b8006e 2760
e80a52d1
SW		 2761 /* Requeue anything that hasn't been acked */
2762 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2763
e76661d0
SW		 2764 /* If there are no messages queued or keepalive pending, place
2765 * the connection in a STANDBY state */
2766 if (list_empty(&con->out_queue) &&
c9ffc77a 2767 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2768 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2769 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2770 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2771 } else {
2772 /* retry after a delay. */
8dacc7da 2773 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2774 if (con->delay == 0)
2775 con->delay = BASE_DELAY_INTERVAL;
2776 else if (con->delay < MAX_DELAY_INTERVAL)
2777 con->delay *= 2;
c9ffc77a 2778 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2779 queue_con(con);
31b8006e 2780 }
31b8006e
SW		 2781}
2782
2783
2784
2785/*
15d9882c 2786 * initialize a new messenger instance
31b8006e 2787 */
15d9882c
AE		 2788void ceph_messenger_init(struct ceph_messenger *msgr,
2789 struct ceph_entity_addr *myaddr,
2790 u32 supported_features,
2791 u32 required_features,
2792 bool nocrc)
31b8006e 2793{
3d14c5d2
YS		 2794 msgr->supported_features = supported_features;
2795 msgr->required_features = required_features;
2796
31b8006e
SW		 2797 spin_lock_init(&msgr->global_seq_lock);
2798
31b8006e
SW		 2799 if (myaddr)
2800 msgr->inst.addr = *myaddr;
2801
2802 /* select a random nonce */
ac8839d7 2803 msgr->inst.addr.type = 0;
103e2d3a 2804 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2805 encode_my_addr(msgr);
15d9882c 2806 msgr->nocrc = nocrc;
31b8006e 2807
a2a32584 2808 atomic_set(&msgr->stopping, 0);
31b8006e 2809
15d9882c 2810 dout("%s %p\n", __func__, msgr);
31b8006e 2811}
15d9882c 2812EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2813
e00de341
SW		 2814static void clear_standby(struct ceph_connection *con)
2815{
2816 /* come back from STANDBY? */
8dacc7da 2817 if (con->state == CON_STATE_STANDBY) {
e00de341 2818 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2819 con->state = CON_STATE_PREOPEN;
e00de341 2820 con->connect_seq++;
c9ffc77a
AE		 2821 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2822 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2823 }
2824}
2825
31b8006e
SW		 2826/*
2827 * Queue up an outgoing message on the given connection.
2828 */
2829void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2830{
31b8006e 2831 /* set src+dst */
dbad185d 2832 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2833 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2834 msg->needs_out_seq = true;
2835
ec302645 2836 mutex_lock(&con->mutex);
92ce034b 2837
8dacc7da 2838 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2839 dout("con_send %p closed, dropping %p\n", con, msg);
2840 ceph_msg_put(msg);
2841 mutex_unlock(&con->mutex);
2842 return;
2843 }
2844
38941f80 2845 BUG_ON(msg->con != NULL);
36eb71aa 2846 msg->con = con->ops->get(con);
92ce034b
AE		 2847 BUG_ON(msg->con == NULL);
2848
31b8006e
SW		 2849 BUG_ON(!list_empty(&msg->list_head));
2850 list_add_tail(&msg->list_head, &con->out_queue);
2851 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2852 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2853 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2854 le32_to_cpu(msg->hdr.front_len),
2855 le32_to_cpu(msg->hdr.middle_len),
2856 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 2857
2858 clear_standby(con);
ec302645 2859 mutex_unlock(&con->mutex);
31b8006e
SW		 2860
2861 /* if there wasn't anything waiting to send before, queue
2862 * new work */
c9ffc77a 2863 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2864 queue_con(con);
2865}
3d14c5d2 2866EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 2867
2868/*
2869 * Revoke a message that was previously queued for send
2870 */
6740a845 2871void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 2872{
6740a845
AE		 2873 struct ceph_connection *con = msg->con;
2874
2875 if (!con)
2876 return; /* Message not in our possession */
2877
ec302645 2878 mutex_lock(&con->mutex);
31b8006e 2879 if (!list_empty(&msg->list_head)) {
38941f80 2880 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 2881 list_del_init(&msg->list_head);
38941f80 2882 BUG_ON(msg->con == NULL);
36eb71aa 2883 msg->con->ops->put(msg->con);
38941f80 2884 msg->con = NULL;
31b8006e 2885 msg->hdr.seq = 0;
38941f80 2886
31b8006e 2887 ceph_msg_put(msg);
ed98adad
SW		 2888 }
2889 if (con->out_msg == msg) {
38941f80 2890 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 2891 con->out_msg = NULL;
31b8006e
SW		 2892 if (con->out_kvec_is_msg) {
2893 con->out_skip = con->out_kvec_bytes;
2894 con->out_kvec_is_msg = false;
2895 }
ed98adad 2896 msg->hdr.seq = 0;
92ce034b
AE		 2897
2898 ceph_msg_put(msg);
31b8006e 2899 }
ec302645 2900 mutex_unlock(&con->mutex);
31b8006e
SW		 2901}
2902
350b1c32 2903/*
0d59ab81 2904 * Revoke a message that we may be reading data into
350b1c32 2905 */
8921d114 2906void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 2907{
8921d114
AE		 2908 struct ceph_connection *con;
2909
2910 BUG_ON(msg == NULL);
2911 if (!msg->con) {
2912 dout("%s msg %p null con\n", __func__, msg);
2913
2914 return; /* Message not in our possession */
2915 }
2916
2917 con = msg->con;
350b1c32 2918 mutex_lock(&con->mutex);
8921d114 2919 if (con->in_msg == msg) {
95c96174
ED		 2920 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2921 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2922 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 2923
2924 /* skip rest of message */
8921d114
AE		 2925 dout("%s %p msg %p revoked\n", __func__, con, msg);
2926 con->in_base_pos = con->in_base_pos -
350b1c32 2927 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 2928 front_len -
2929 middle_len -
2930 data_len -
350b1c32 2931 sizeof(struct ceph_msg_footer);
350b1c32
SW		 2932 ceph_msg_put(con->in_msg);
2933 con->in_msg = NULL;
2934 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2935 con->in_seq++;
350b1c32 2936 } else {
8921d114
AE		 2937 dout("%s %p in_msg %p msg %p no-op\n",
2938 __func__, con, con->in_msg, msg);
350b1c32
SW		 2939 }
2940 mutex_unlock(&con->mutex);
2941}
2942
31b8006e
SW		 2943/*
2944 * Queue a keepalive byte to ensure the tcp connection is alive.
2945 */
2946void ceph_con_keepalive(struct ceph_connection *con)
2947{
e00de341 2948 dout("con_keepalive %p\n", con);
00650931 2949 mutex_lock(&con->mutex);
e00de341 2950 clear_standby(con);
00650931 2951 mutex_unlock(&con->mutex);
c9ffc77a
AE		 2952 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
2953 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2954 queue_con(con);
2955}
3d14c5d2 2956EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 2957
6644ed7b 2958static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 2959{
6644ed7b
AE		 2960 struct ceph_msg_data *data;
2961
2962 if (WARN_ON(!ceph_msg_data_type_valid(type)))
2963 return NULL;
2964
2965 data = kzalloc(sizeof (*data), GFP_NOFS);
2966 if (data)
2967 data->type = type;
5240d9f9 2968 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 2969
2970 return data;
2971}
2972
2973static void ceph_msg_data_destroy(struct ceph_msg_data *data)
2974{
2975 if (!data)
2976 return;
2977
5240d9f9 2978 WARN_ON(!list_empty(&data->links));
6644ed7b
AE		 2979 if (data->type == CEPH_MSG_DATA_PAGELIST) {
2980 ceph_pagelist_release(data->pagelist);
2981 kfree(data->pagelist);
2982 }
2983 kfree(data);
43794509
AE		 2984}
2985
02afca6c 2986void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 2987 size_t length, size_t alignment)
02afca6c 2988{
6644ed7b
AE		 2989 struct ceph_msg_data *data;
2990
07aa1558
AE		 2991 BUG_ON(!pages);
2992 BUG_ON(!length);
a1930804 2993 BUG_ON(msg->data_length);
5240d9f9 2994 BUG_ON(!list_empty(&msg->data));
6644ed7b
AE		 2995
2996 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
2997 BUG_ON(!data);
2998 data->pages = pages;
2999 data->length = length;
3000 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3001
5240d9f9
AE		 3002 BUG_ON(!list_empty(&msg->data));
3003 list_add_tail(&data->links, &msg->data);
3004 msg->data_length += length;
02afca6c
AE		 3005}
3006EXPORT_SYMBOL(ceph_msg_data_set_pages);
31b8006e 3007
27fa8385
AE		 3008void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3009 struct ceph_pagelist *pagelist)
3010{
6644ed7b
AE		 3011 struct ceph_msg_data *data;
3012
07aa1558
AE		 3013 BUG_ON(!pagelist);
3014 BUG_ON(!pagelist->length);
a1930804 3015 BUG_ON(msg->data_length);
5240d9f9 3016 BUG_ON(!list_empty(&msg->data));
27fa8385 3017
6644ed7b
AE		 3018 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3019 BUG_ON(!data);
3020 data->pagelist = pagelist;
3021
5240d9f9
AE		 3022 list_add_tail(&data->links, &msg->data);
3023 msg->data_length += pagelist->length;
27fa8385
AE		 3024}
3025EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3026
ea96571f 3027#ifdef CONFIG_BLOCK
a1930804
AE		 3028void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio,
3029 size_t length)
27fa8385 3030{
6644ed7b
AE		 3031 struct ceph_msg_data *data;
3032
07aa1558 3033 BUG_ON(!bio);
a1930804 3034 BUG_ON(msg->data_length);
5240d9f9 3035 BUG_ON(!list_empty(&msg->data));
27fa8385 3036
6644ed7b
AE		 3037 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3038 BUG_ON(!data);
3039 data->bio = bio;
c851c495 3040 data->bio_length = length;
6644ed7b 3041
5240d9f9
AE		 3042 list_add_tail(&data->links, &msg->data);
3043 msg->data_length += length;
27fa8385
AE		 3044}
3045EXPORT_SYMBOL(ceph_msg_data_set_bio);
ea96571f 3046#endif /* CONFIG_BLOCK */
27fa8385 3047
31b8006e
SW		 3048/*
3049 * construct a new message with given type, size
3050 * the new msg has a ref count of 1.
3051 */
b61c2763
SW		 3052struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3053 bool can_fail)
31b8006e
SW		 3054{
3055 struct ceph_msg *m;
3056
9516e45b 3057 m = kzalloc(sizeof(*m), flags);
31b8006e
SW		 3058 if (m == NULL)
3059 goto out;
31b8006e
SW		 3060
3061 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3062 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3063 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3064
9516e45b
AE		 3065 INIT_LIST_HEAD(&m->list_head);
3066 kref_init(&m->kref);
5240d9f9 3067 INIT_LIST_HEAD(&m->data);
ca20892d 3068
31b8006e 3069 /* front */
9516e45b 3070 m->front_max = front_len;
31b8006e
SW		 3071 if (front_len) {
3072 if (front_len > PAGE_CACHE_SIZE) {
34d23762 3073 m->front.iov_base = __vmalloc(front_len, flags,
31b8006e
SW		 3074 PAGE_KERNEL);
3075 m->front_is_vmalloc = true;
3076 } else {
34d23762 3077 m->front.iov_base = kmalloc(front_len, flags);
31b8006e
SW		 3078 }
3079 if (m->front.iov_base == NULL) {
b61c2763 3080 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3081 front_len);
3082 goto out2;
3083 }
3084 } else {
3085 m->front.iov_base = NULL;
3086 }
3087 m->front.iov_len = front_len;
3088
bb257664 3089 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3090 return m;
3091
3092out2:
3093 ceph_msg_put(m);
3094out:
b61c2763
SW		 3095 if (!can_fail) {
3096 pr_err("msg_new can't create type %d front %d\n", type,
3097 front_len);
f0ed1b7c 3098 WARN_ON(1);
b61c2763
SW		 3099 } else {
3100 dout("msg_new can't create type %d front %d\n", type,
3101 front_len);
3102 }
a79832f2 3103 return NULL;
31b8006e 3104}
3d14c5d2 3105EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3106
31b8006e
SW		 3107/*
3108 * Allocate "middle" portion of a message, if it is needed and wasn't
3109 * allocated by alloc_msg. This allows us to read a small fixed-size
3110 * per-type header in the front and then gracefully fail (i.e.,
3111 * propagate the error to the caller based on info in the front) when
3112 * the middle is too large.
3113 */
2450418c 3114static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3115{
3116 int type = le16_to_cpu(msg->hdr.type);
3117 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3118
3119 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3120 ceph_msg_type_name(type), middle_len);
3121 BUG_ON(!middle_len);
3122 BUG_ON(msg->middle);
3123
b6c1d5b8 3124 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3125 if (!msg->middle)
3126 return -ENOMEM;
3127 return 0;
3128}
3129
2450418c 3130/*
1c20f2d2
AE		 3131 * Allocate a message for receiving an incoming message on a
3132 * connection, and save the result in con->in_msg. Uses the
3133 * connection's private alloc_msg op if available.
3134 *
4740a623
SW		 3135 * Returns 0 on success, or a negative error code.
3136 *
3137 * On success, if we set *skip = 1:
3138 * - the next message should be skipped and ignored.
3139 * - con->in_msg == NULL
3140 * or if we set *skip = 0:
3141 * - con->in_msg is non-null.
3142 * On error (ENOMEM, EAGAIN, ...),
3143 * - con->in_msg == NULL
2450418c 3144 */
4740a623 3145static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3146{
4740a623 3147 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3148 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3149 struct ceph_msg *msg;
4740a623 3150 int ret = 0;
2450418c 3151
1c20f2d2 3152 BUG_ON(con->in_msg != NULL);
53ded495 3153 BUG_ON(!con->ops->alloc_msg);
2450418c 3154
53ded495
AE		 3155 mutex_unlock(&con->mutex);
3156 msg = con->ops->alloc_msg(con, hdr, skip);
3157 mutex_lock(&con->mutex);
3158 if (con->state != CON_STATE_OPEN) {
3159 if (msg)
1d866d1c 3160 ceph_msg_put(msg);
53ded495
AE		 3161 return -EAGAIN;
3162 }
4137577a
AE		 3163 if (msg) {
3164 BUG_ON(*skip);
3165 con->in_msg = msg;
36eb71aa 3166 con->in_msg->con = con->ops->get(con);
92ce034b 3167 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3168 } else {
3169 /*
3170 * Null message pointer means either we should skip
3171 * this message or we couldn't allocate memory. The
3172 * former is not an error.
3173 */
3174 if (*skip)
3175 return 0;
3176 con->error_msg = "error allocating memory for incoming message";
3177
53ded495 3178 return -ENOMEM;
2450418c 3179 }
1c20f2d2 3180 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3181
1c20f2d2
AE		 3182 if (middle_len && !con->in_msg->middle) {
3183 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3184 if (ret < 0) {
1c20f2d2
AE		 3185 ceph_msg_put(con->in_msg);
3186 con->in_msg = NULL;
2450418c
YS		 3187 }
3188 }
9d7f0f13 3189
4740a623 3190 return ret;
2450418c
YS		 3191}
3192
31b8006e
SW		 3193
3194/*
3195 * Free a generically kmalloc'd message.
3196 */
3197void ceph_msg_kfree(struct ceph_msg *m)
3198{
3199 dout("msg_kfree %p\n", m);
3200 if (m->front_is_vmalloc)
3201 vfree(m->front.iov_base);
3202 else
3203 kfree(m->front.iov_base);
3204 kfree(m);
3205}
3206
3207/*
3208 * Drop a msg ref. Destroy as needed.
3209 */
c2e552e7
SW		 3210void ceph_msg_last_put(struct kref *kref)
3211{
3212 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
5240d9f9
AE		 3213 LIST_HEAD(data);
3214 struct list_head *links;
3215 struct list_head *next;
31b8006e 3216
c2e552e7
SW		 3217 dout("ceph_msg_put last one on %p\n", m);
3218 WARN_ON(!list_empty(&m->list_head));
3219
3220 /* drop middle, data, if any */
3221 if (m->middle) {
3222 ceph_buffer_put(m->middle);
3223 m->middle = NULL;
31b8006e 3224 }
5240d9f9
AE		 3225
3226 list_splice_init(&m->data, &data);
3227 list_for_each_safe(links, next, &data) {
3228 struct ceph_msg_data *data;
3229
3230 data = list_entry(links, struct ceph_msg_data, links);
3231 list_del_init(links);
3232 ceph_msg_data_destroy(data);
3233 }
a1930804 3234 m->data_length = 0;
58bb3b37 3235
c2e552e7
SW		 3236 if (m->pool)
3237 ceph_msgpool_put(m->pool, m);
3238 else
3239 ceph_msg_kfree(m);
31b8006e 3240}
3d14c5d2 3241EXPORT_SYMBOL(ceph_msg_last_put);
9ec7cab1
SW		 3242
3243void ceph_msg_dump(struct ceph_msg *msg)
3244{
4a73ef27 3245 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
5240d9f9 3246 msg->front_max, msg->data_length);
9ec7cab1
SW		 3247 print_hex_dump(KERN_DEBUG, "header: ",
3248 DUMP_PREFIX_OFFSET, 16, 1,
3249 &msg->hdr, sizeof(msg->hdr), true);
3250 print_hex_dump(KERN_DEBUG, " front: ",
3251 DUMP_PREFIX_OFFSET, 16, 1,
3252 msg->front.iov_base, msg->front.iov_len, true);
3253 if (msg->middle)
3254 print_hex_dump(KERN_DEBUG, "middle: ",
3255 DUMP_PREFIX_OFFSET, 16, 1,
3256 msg->middle->vec.iov_base,
3257 msg->middle->vec.iov_len, true);
3258 print_hex_dump(KERN_DEBUG, "footer: ",
3259 DUMP_PREFIX_OFFSET, 16, 1,
3260 &msg->footer, sizeof(msg->footer), true);
3261}
3d14c5d2 3262EXPORT_SYMBOL(ceph_msg_dump);
Linux kernel - Deliverables
This page took 0.373162 seconds and 5 git commands to generate.