projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: don't add to crc unless data sent
[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 */
25aff7c5
AE		 725static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data *data,
726 size_t length)
6aaa4511
AE		 727{
728 struct ceph_msg_data_cursor *cursor = &data->cursor;
729 struct bio *bio;
730
731 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
732
733 bio = data->bio;
734 BUG_ON(!bio);
735 BUG_ON(!bio->bi_vcnt);
6aaa4511 736
25aff7c5 737 cursor->resid = length;
6aaa4511
AE		 738 cursor->bio = bio;
739 cursor->vector_index = 0;
740 cursor->vector_offset = 0;
25aff7c5 741 cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
6aaa4511
AE		 742}
743
744static struct page *ceph_msg_data_bio_next(struct ceph_msg_data *data,
745 size_t *page_offset,
746 size_t *length)
747{
748 struct ceph_msg_data_cursor *cursor = &data->cursor;
749 struct bio *bio;
750 struct bio_vec *bio_vec;
751 unsigned int index;
752
753 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
754
755 bio = cursor->bio;
756 BUG_ON(!bio);
757
758 index = cursor->vector_index;
759 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
760
761 bio_vec = &bio->bi_io_vec[index];
762 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
763 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
764 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 765 if (cursor->last_piece) /* pagelist offset is always 0 */
766 *length = cursor->resid;
767 else
768 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
6aaa4511 769 BUG_ON(*length > PAGE_SIZE);
25aff7c5 770 BUG_ON(*length > cursor->resid);
6aaa4511
AE		 771
772 return bio_vec->bv_page;
773}
774
775static bool ceph_msg_data_bio_advance(struct ceph_msg_data *data, size_t bytes)
776{
777 struct ceph_msg_data_cursor *cursor = &data->cursor;
778 struct bio *bio;
779 struct bio_vec *bio_vec;
780 unsigned int index;
781
782 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
783
784 bio = cursor->bio;
785 BUG_ON(!bio);
786
787 index = cursor->vector_index;
788 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
789 bio_vec = &bio->bi_io_vec[index];
6aaa4511
AE		 790
791 /* Advance the cursor offset */
792
25aff7c5
AE		 793 BUG_ON(cursor->resid < bytes);
794 cursor->resid -= bytes;
6aaa4511
AE		 795 cursor->vector_offset += bytes;
796 if (cursor->vector_offset < bio_vec->bv_len)
797 return false; /* more bytes to process in this segment */
25aff7c5 798 BUG_ON(cursor->vector_offset != bio_vec->bv_len);
6aaa4511
AE		 799
800 /* Move on to the next segment, and possibly the next bio */
801
25aff7c5 802 if (++index == (unsigned int) bio->bi_vcnt) {
6aaa4511 803 bio = bio->bi_next;
25aff7c5 804 index = 0;
6aaa4511 805 }
25aff7c5
AE		 806 cursor->bio = bio;
807 cursor->vector_index = index;
6aaa4511
AE		 808 cursor->vector_offset = 0;
809
25aff7c5
AE		 810 if (!cursor->last_piece) {
811 BUG_ON(!cursor->resid);
812 BUG_ON(!bio);
813 /* A short read is OK, so use <= rather than == */
814 if (cursor->resid <= bio->bi_io_vec[index].bv_len)
6aaa4511 815 cursor->last_piece = true;
25aff7c5 816 }
6aaa4511
AE		 817
818 return true;
819}
df6ad1f9
AE		 820#endif
821
e766d7b5
AE		 822/*
823 * For a page array, a piece comes from the first page in the array
824 * that has not already been fully consumed.
825 */
25aff7c5
AE		 826static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data *data,
827 size_t length)
e766d7b5
AE		 828{
829 struct ceph_msg_data_cursor *cursor = &data->cursor;
830 int page_count;
831
832 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
833
834 BUG_ON(!data->pages);
835 BUG_ON(!data->length);
25aff7c5 836 BUG_ON(length != data->length);
e766d7b5 837
25aff7c5 838 cursor->resid = length;
e766d7b5 839 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 840 cursor->page_offset = data->alignment & ~PAGE_MASK;
841 cursor->page_index = 0;
25aff7c5 842 BUG_ON(page_count > (int) USHRT_MAX);
e766d7b5 843 cursor->page_count = (unsigned short) page_count;
25aff7c5 844 cursor->last_piece = length <= PAGE_SIZE;
e766d7b5
AE		 845}
846
847static struct page *ceph_msg_data_pages_next(struct ceph_msg_data *data,
848 size_t *page_offset,
849 size_t *length)
850{
851 struct ceph_msg_data_cursor *cursor = &data->cursor;
852
853 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
854
855 BUG_ON(cursor->page_index >= cursor->page_count);
856 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 857
858 *page_offset = cursor->page_offset;
25aff7c5 859 if (cursor->last_piece)
e766d7b5 860 *length = cursor->resid;
25aff7c5 861 else
e766d7b5 862 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 863
864 return data->pages[cursor->page_index];
865}
866
867static bool ceph_msg_data_pages_advance(struct ceph_msg_data *data,
868 size_t bytes)
869{
870 struct ceph_msg_data_cursor *cursor = &data->cursor;
871
872 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
873
874 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 875
876 /* Advance the cursor page offset */
877
878 cursor->resid -= bytes;
879 cursor->page_offset += bytes;
880 if (!bytes || cursor->page_offset & ~PAGE_MASK)
881 return false; /* more bytes to process in the current page */
882
883 /* Move on to the next page */
884
885 BUG_ON(cursor->page_index >= cursor->page_count);
886 cursor->page_offset = 0;
887 cursor->page_index++;
25aff7c5 888 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 889
890 return true;
891}
892
fe38a2b6 893/*
dd236fcb
AE		 894 * For a pagelist, a piece is whatever remains to be consumed in the
895 * first page in the list, or the front of the next page.
fe38a2b6 896 */
25aff7c5
AE		 897static void ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data *data,
898 size_t length)
fe38a2b6
AE		 899{
900 struct ceph_msg_data_cursor *cursor = &data->cursor;
901 struct ceph_pagelist *pagelist;
902 struct page *page;
903
dd236fcb 904 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 905
906 pagelist = data->pagelist;
907 BUG_ON(!pagelist);
25aff7c5
AE		 908 BUG_ON(length != pagelist->length);
909
910 if (!length)
fe38a2b6
AE		 911 return; /* pagelist can be assigned but empty */
912
913 BUG_ON(list_empty(&pagelist->head));
914 page = list_first_entry(&pagelist->head, struct page, lru);
915
25aff7c5 916 cursor->resid = length;
fe38a2b6
AE		 917 cursor->page = page;
918 cursor->offset = 0;
25aff7c5 919 cursor->last_piece = length <= PAGE_SIZE;
fe38a2b6
AE		 920}
921
dd236fcb 922static struct page *ceph_msg_data_pagelist_next(struct ceph_msg_data *data,
fe38a2b6 923 size_t *page_offset,
dd236fcb 924 size_t *length)
fe38a2b6
AE		 925{
926 struct ceph_msg_data_cursor *cursor = &data->cursor;
927 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 928
929 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
930
931 pagelist = data->pagelist;
932 BUG_ON(!pagelist);
933
934 BUG_ON(!cursor->page);
25aff7c5 935 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 936
fe38a2b6 937 *page_offset = cursor->offset & ~PAGE_MASK;
25aff7c5
AE		 938 if (cursor->last_piece) /* pagelist offset is always 0 */
939 *length = cursor->resid;
940 else
941 *length = PAGE_SIZE - *page_offset;
fe38a2b6
AE		 942
943 return data->cursor.page;
944}
945
dd236fcb
AE		 946static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data *data,
947 size_t bytes)
fe38a2b6
AE		 948{
949 struct ceph_msg_data_cursor *cursor = &data->cursor;
950 struct ceph_pagelist *pagelist;
951
952 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
953
954 pagelist = data->pagelist;
955 BUG_ON(!pagelist);
25aff7c5
AE		 956
957 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 958 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
959
960 /* Advance the cursor offset */
961
25aff7c5 962 cursor->resid -= bytes;
fe38a2b6
AE		 963 cursor->offset += bytes;
964 /* pagelist offset is always 0 */
965 if (!bytes || cursor->offset & ~PAGE_MASK)
966 return false; /* more bytes to process in the current page */
967
968 /* Move on to the next page */
969
970 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
971 cursor->page = list_entry_next(cursor->page, lru);
25aff7c5 972 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 973
974 return true;
975}
976
dd236fcb
AE		 977/*
978 * Message data is handled (sent or received) in pieces, where each
979 * piece resides on a single page. The network layer might not
980 * consume an entire piece at once. A data item's cursor keeps
981 * track of which piece is next to process and how much remains to
982 * be processed in that piece. It also tracks whether the current
983 * piece is the last one in the data item.
984 */
25aff7c5
AE		 985static void ceph_msg_data_cursor_init(struct ceph_msg_data *data,
986 size_t length)
dd236fcb
AE		 987{
988 switch (data->type) {
989 case CEPH_MSG_DATA_PAGELIST:
25aff7c5 990 ceph_msg_data_pagelist_cursor_init(data, length);
dd236fcb 991 break;
e766d7b5 992 case CEPH_MSG_DATA_PAGES:
25aff7c5 993 ceph_msg_data_pages_cursor_init(data, length);
e766d7b5 994 break;
dd236fcb
AE		 995#ifdef CONFIG_BLOCK
996 case CEPH_MSG_DATA_BIO:
25aff7c5 997 ceph_msg_data_bio_cursor_init(data, length);
6aaa4511 998 break;
dd236fcb 999#endif /* CONFIG_BLOCK */
6aaa4511 1000 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1001 default:
1002 /* BUG(); */
1003 break;
1004 }
f5db90bc 1005 data->cursor.need_crc = true;
dd236fcb
AE		 1006}
1007
1008/*
1009 * Return the page containing the next piece to process for a given
1010 * data item, and supply the page offset and length of that piece.
1011 * Indicate whether this is the last piece in this data item.
1012 */
1013static struct page *ceph_msg_data_next(struct ceph_msg_data *data,
1014 size_t *page_offset,
1015 size_t *length,
1016 bool *last_piece)
1017{
1018 struct page *page;
1019
1020 switch (data->type) {
1021 case CEPH_MSG_DATA_PAGELIST:
1022 page = ceph_msg_data_pagelist_next(data, page_offset, length);
1023 break;
e766d7b5
AE		 1024 case CEPH_MSG_DATA_PAGES:
1025 page = ceph_msg_data_pages_next(data, page_offset, length);
1026 break;
dd236fcb
AE		 1027#ifdef CONFIG_BLOCK
1028 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1029 page = ceph_msg_data_bio_next(data, page_offset, length);
1030 break;
dd236fcb 1031#endif /* CONFIG_BLOCK */
6aaa4511 1032 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1033 default:
1034 page = NULL;
1035 break;
1036 }
1037 BUG_ON(!page);
1038 BUG_ON(*page_offset + *length > PAGE_SIZE);
1039 BUG_ON(!*length);
1040 if (last_piece)
1041 *last_piece = data->cursor.last_piece;
1042
1043 return page;
1044}
1045
1046/*
1047 * Returns true if the result moves the cursor on to the next piece
1048 * of the data item.
1049 */
1050static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes)
1051{
25aff7c5 1052 struct ceph_msg_data_cursor *cursor = &data->cursor;
dd236fcb
AE		 1053 bool new_piece;
1054
25aff7c5 1055 BUG_ON(bytes > cursor->resid);
dd236fcb
AE		 1056 switch (data->type) {
1057 case CEPH_MSG_DATA_PAGELIST:
1058 new_piece = ceph_msg_data_pagelist_advance(data, bytes);
1059 break;
e766d7b5
AE		 1060 case CEPH_MSG_DATA_PAGES:
1061 new_piece = ceph_msg_data_pages_advance(data, bytes);
1062 break;
dd236fcb
AE		 1063#ifdef CONFIG_BLOCK
1064 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1065 new_piece = ceph_msg_data_bio_advance(data, bytes);
1066 break;
dd236fcb 1067#endif /* CONFIG_BLOCK */
6aaa4511 1068 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1069 default:
1070 BUG();
1071 break;
1072 }
f5db90bc 1073 data->cursor.need_crc = new_piece;
dd236fcb
AE		 1074
1075 return new_piece;
1076}
1077
f5db90bc 1078static void prepare_message_data(struct ceph_msg *msg)
739c905b 1079{
25aff7c5
AE		 1080 size_t data_len;
1081
739c905b 1082 BUG_ON(!msg);
25aff7c5
AE		 1083
1084 data_len = le32_to_cpu(msg->hdr.data_len);
1085 BUG_ON(!data_len);
739c905b 1086
4c59b4a2 1087 /* Initialize data cursor */
fe38a2b6 1088
4c59b4a2 1089 ceph_msg_data_cursor_init(&msg->data, data_len);
739c905b
AE		 1090}
1091
31b8006e
SW		 1092/*
1093 * Prepare footer for currently outgoing message, and finish things
1094 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1095 */
859eb799 1096static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1097{
1098 struct ceph_msg *m = con->out_msg;
859eb799 1099 int v = con->out_kvec_left;
31b8006e 1100
fd154f3c
AE		 1101 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1102
31b8006e
SW		 1103 dout("prepare_write_message_footer %p\n", con);
1104 con->out_kvec_is_msg = true;
1105 con->out_kvec[v].iov_base = &m->footer;
1106 con->out_kvec[v].iov_len = sizeof(m->footer);
1107 con->out_kvec_bytes += sizeof(m->footer);
1108 con->out_kvec_left++;
1109 con->out_more = m->more_to_follow;
c86a2930 1110 con->out_msg_done = true;
31b8006e
SW		 1111}
1112
1113/*
1114 * Prepare headers for the next outgoing message.
1115 */
1116static void prepare_write_message(struct ceph_connection *con)
1117{
1118 struct ceph_msg *m;
a9a0c51a 1119 u32 crc;
31b8006e 1120
e2200423 1121 con_out_kvec_reset(con);
31b8006e 1122 con->out_kvec_is_msg = true;
c86a2930 1123 con->out_msg_done = false;
31b8006e
SW		 1124
1125 /* Sneak an ack in there first? If we can get it into the same
1126 * TCP packet that's a good thing. */
1127 if (con->in_seq > con->in_seq_acked) {
1128 con->in_seq_acked = con->in_seq;
e2200423 1129 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1130 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1131 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1132 &con->out_temp_ack);
31b8006e
SW		 1133 }
1134
38941f80 1135 BUG_ON(list_empty(&con->out_queue));
859eb799 1136 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1137 con->out_msg = m;
38941f80 1138 BUG_ON(m->con != con);
4cf9d544
SW		 1139
1140 /* put message on sent list */
1141 ceph_msg_get(m);
1142 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1143
e84346b7
SW		 1144 /*
1145 * only assign outgoing seq # if we haven't sent this message
1146 * yet. if it is requeued, resend with it's original seq.
1147 */
1148 if (m->needs_out_seq) {
1149 m->hdr.seq = cpu_to_le64(++con->out_seq);
1150 m->needs_out_seq = false;
1151 }
31b8006e 1152
4c59b4a2 1153 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d\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),
4c59b4a2 1156 le32_to_cpu(m->hdr.data_len));
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;
78625051 1187 if (m->hdr.data_len) {
f5db90bc 1188 prepare_message_data(con->out_msg);
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
84ca8fc8 1386static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
9d2a06c2 1387 size_t len, size_t sent)
68b4476b 1388{
84ca8fc8 1389 struct ceph_msg *msg = con->out_msg;
f5db90bc 1390 bool need_crc;
68b4476b 1391
84ca8fc8
AE		 1392 BUG_ON(!msg);
1393 BUG_ON(!sent);
68b4476b 1394
4c59b4a2 1395 need_crc = ceph_msg_data_advance(&msg->data, sent);
7fe1e5e5
AE		 1396 BUG_ON(need_crc && sent != len);
1397
5821bd8c
AE		 1398 if (sent < len)
1399 return;
68b4476b 1400
5821bd8c 1401 BUG_ON(sent != len);
84ca8fc8 1402}
68b4476b 1403
e788182f
AE		 1404static void in_msg_pos_next(struct ceph_connection *con, size_t len,
1405 size_t received)
1406{
1407 struct ceph_msg *msg = con->in_msg;
1408
1409 BUG_ON(!msg);
1410 BUG_ON(!received);
1411
4c59b4a2
AE		 1412 (void) ceph_msg_data_advance(&msg->data, received);
1413
e788182f
AE		 1414 if (received < len)
1415 return;
1416
1417 BUG_ON(received != len);
e788182f
AE		 1418}
1419
35b62808
AE		 1420static u32 ceph_crc32c_page(u32 crc, struct page *page,
1421 unsigned int page_offset,
1422 unsigned int length)
1423{
1424 char *kaddr;
1425
1426 kaddr = kmap(page);
1427 BUG_ON(kaddr == NULL);
1428 crc = crc32c(crc, kaddr + page_offset, length);
1429 kunmap(page);
1430
1431 return crc;
1432}
31b8006e
SW		 1433/*
1434 * Write as much message data payload as we can. If we finish, queue
1435 * up the footer.
1436 * 1 -> done, footer is now queued in out_kvec[].
1437 * 0 -> socket full, but more to do
1438 * <0 -> error
1439 */
34d2d200 1440static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1441{
1442 struct ceph_msg *msg = con->out_msg;
643c68a4 1443 struct ceph_msg_data_cursor *cursor = &msg->data.cursor;
37675b0f 1444 bool do_datacrc = !con->msgr->nocrc;
f5db90bc 1445 u32 crc;
31b8006e 1446
859a35d5 1447 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1448
4c59b4a2
AE		 1449 if (WARN_ON(!ceph_msg_has_data(msg)))
1450 return -EINVAL;
1451
5821bd8c
AE		 1452 /*
1453 * Iterate through each page that contains data to be
1454 * written, and send as much as possible for each.
1455 *
1456 * If we are calculating the data crc (the default), we will
1457 * need to map the page. If we have no pages, they have
1458 * been revoked, so use the zero page.
1459 */
f5db90bc 1460 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1461 while (cursor->resid) {
8a166d05 1462 struct page *page;
e387d525
AE		 1463 size_t page_offset;
1464 size_t length;
8a166d05 1465 bool last_piece;
f5db90bc 1466 int ret;
68b4476b 1467
4c59b4a2
AE		 1468 page = ceph_msg_data_next(&msg->data, &page_offset, &length,
1469 &last_piece);
e387d525 1470 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1471 length, last_piece);
f5db90bc
AE		 1472 if (ret <= 0) {
1473 if (do_datacrc)
1474 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1475
f5db90bc
AE		 1476 return ret;
1477 }
143334ff
AE		 1478 if (do_datacrc && cursor->need_crc)
1479 crc = ceph_crc32c_page(crc, page, page_offset, length);
9d2a06c2 1480 out_msg_pos_next(con, page, length, (size_t) ret);
31b8006e
SW		 1481 }
1482
34d2d200 1483 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1484
1485 /* prepare and queue up footer, too */
f5db90bc
AE		 1486 if (do_datacrc)
1487 msg->footer.data_crc = cpu_to_le32(crc);
1488 else
84ca8fc8 1489 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1490 con_out_kvec_reset(con);
859eb799 1491 prepare_write_message_footer(con);
f5db90bc
AE		 1492
1493 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1494}
1495
1496/*
1497 * write some zeros
1498 */
1499static int write_partial_skip(struct ceph_connection *con)
1500{
1501 int ret;
1502
1503 while (con->out_skip > 0) {
31739139 1504 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1505
e1dcb128 1506 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1507 if (ret <= 0)
1508 goto out;
1509 con->out_skip -= ret;
1510 }
1511 ret = 1;
1512out:
1513 return ret;
1514}
1515
1516/*
1517 * Prepare to read connection handshake, or an ack.
1518 */
eed0ef2c
SW		 1519static void prepare_read_banner(struct ceph_connection *con)
1520{
1521 dout("prepare_read_banner %p\n", con);
1522 con->in_base_pos = 0;
1523}
1524
31b8006e
SW		 1525static void prepare_read_connect(struct ceph_connection *con)
1526{
1527 dout("prepare_read_connect %p\n", con);
1528 con->in_base_pos = 0;
1529}
1530
1531static void prepare_read_ack(struct ceph_connection *con)
1532{
1533 dout("prepare_read_ack %p\n", con);
1534 con->in_base_pos = 0;
1535}
1536
3a23083b
SW		 1537static void prepare_read_seq(struct ceph_connection *con)
1538{
1539 dout("prepare_read_seq %p\n", con);
1540 con->in_base_pos = 0;
1541 con->in_tag = CEPH_MSGR_TAG_SEQ;
1542}
1543
31b8006e
SW		 1544static void prepare_read_tag(struct ceph_connection *con)
1545{
1546 dout("prepare_read_tag %p\n", con);
1547 con->in_base_pos = 0;
1548 con->in_tag = CEPH_MSGR_TAG_READY;
1549}
1550
1551/*
1552 * Prepare to read a message.
1553 */
1554static int prepare_read_message(struct ceph_connection *con)
1555{
1556 dout("prepare_read_message %p\n", con);
1557 BUG_ON(con->in_msg != NULL);
1558 con->in_base_pos = 0;
1559 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1560 return 0;
1561}
1562
1563
1564static int read_partial(struct ceph_connection *con,
fd51653f 1565 int end, int size, void *object)
31b8006e 1566{
e6cee71f
AE		 1567 while (con->in_base_pos < end) {
1568 int left = end - con->in_base_pos;
31b8006e
SW		 1569 int have = size - left;
1570 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1571 if (ret <= 0)
1572 return ret;
1573 con->in_base_pos += ret;
1574 }
1575 return 1;
1576}
1577
1578
1579/*
1580 * Read all or part of the connect-side handshake on a new connection
1581 */
eed0ef2c 1582static int read_partial_banner(struct ceph_connection *con)
31b8006e 1583{
fd51653f
AE		 1584 int size;
1585 int end;
1586 int ret;
31b8006e 1587
eed0ef2c 1588 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1589
1590 /* peer's banner */
fd51653f
AE		 1591 size = strlen(CEPH_BANNER);
1592 end = size;
1593 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1594 if (ret <= 0)
1595 goto out;
fd51653f
AE		 1596
1597 size = sizeof (con->actual_peer_addr);
1598 end += size;
1599 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1600 if (ret <= 0)
1601 goto out;
fd51653f
AE		 1602
1603 size = sizeof (con->peer_addr_for_me);
1604 end += size;
1605 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1606 if (ret <= 0)
1607 goto out;
fd51653f 1608
eed0ef2c
SW		 1609out:
1610 return ret;
1611}
1612
1613static int read_partial_connect(struct ceph_connection *con)
1614{
fd51653f
AE		 1615 int size;
1616 int end;
1617 int ret;
eed0ef2c
SW		 1618
1619 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1620
fd51653f
AE		 1621 size = sizeof (con->in_reply);
1622 end = size;
1623 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1624 if (ret <= 0)
1625 goto out;
fd51653f
AE		 1626
1627 size = le32_to_cpu(con->in_reply.authorizer_len);
1628 end += size;
1629 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1630 if (ret <= 0)
1631 goto out;
31b8006e 1632
4e7a5dcd
SW		 1633 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1634 con, (int)con->in_reply.tag,
1635 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1636 le32_to_cpu(con->in_reply.global_seq));
1637out:
1638 return ret;
eed0ef2c 1639
31b8006e
SW		 1640}
1641
1642/*
1643 * Verify the hello banner looks okay.
1644 */
1645static int verify_hello(struct ceph_connection *con)
1646{
1647 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1648 pr_err("connect to %s got bad banner\n",
3d14c5d2 1649 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1650 con->error_msg = "protocol error, bad banner";
1651 return -1;
1652 }
1653 return 0;
1654}
1655
1656static bool addr_is_blank(struct sockaddr_storage *ss)
1657{
1658 switch (ss->ss_family) {
1659 case AF_INET:
1660 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1661 case AF_INET6:
1662 return
1663 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1664 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1665 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1666 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1667 }
1668 return false;
1669}
1670
1671static int addr_port(struct sockaddr_storage *ss)
1672{
1673 switch (ss->ss_family) {
1674 case AF_INET:
f28bcfbe 1675 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1676 case AF_INET6:
f28bcfbe 1677 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1678 }
1679 return 0;
1680}
1681
1682static void addr_set_port(struct sockaddr_storage *ss, int p)
1683{
1684 switch (ss->ss_family) {
1685 case AF_INET:
1686 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1687 break;
31b8006e
SW		 1688 case AF_INET6:
1689 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1690 break;
31b8006e
SW		 1691 }
1692}
1693
ee3b56f2
NW		 1694/*
1695 * Unlike other *_pton function semantics, zero indicates success.
1696 */
1697static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1698 char delim, const char **ipend)
1699{
99f0f3b2
AE		 1700 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1701 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1702
1703 memset(ss, 0, sizeof(*ss));
1704
1705 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1706 ss->ss_family = AF_INET;
1707 return 0;
1708 }
1709
1710 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1711 ss->ss_family = AF_INET6;
1712 return 0;
1713 }
1714
1715 return -EINVAL;
1716}
1717
1718/*
1719 * Extract hostname string and resolve using kernel DNS facility.
1720 */
1721#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1722static int ceph_dns_resolve_name(const char *name, size_t namelen,
1723 struct sockaddr_storage *ss, char delim, const char **ipend)
1724{
1725 const char *end, *delim_p;
1726 char *colon_p, *ip_addr = NULL;
1727 int ip_len, ret;
1728
1729 /*
1730 * The end of the hostname occurs immediately preceding the delimiter or
1731 * the port marker (':') where the delimiter takes precedence.
1732 */
1733 delim_p = memchr(name, delim, namelen);
1734 colon_p = memchr(name, ':', namelen);
1735
1736 if (delim_p && colon_p)
1737 end = delim_p < colon_p ? delim_p : colon_p;
1738 else if (!delim_p && colon_p)
1739 end = colon_p;
1740 else {
1741 end = delim_p;
1742 if (!end) /* case: hostname:/ */
1743 end = name + namelen;
1744 }
1745
1746 if (end <= name)
1747 return -EINVAL;
1748
1749 /* do dns_resolve upcall */
1750 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1751 if (ip_len > 0)
1752 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1753 else
1754 ret = -ESRCH;
1755
1756 kfree(ip_addr);
1757
1758 *ipend = end;
1759
1760 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1761 ret, ret ? "failed" : ceph_pr_addr(ss));
1762
1763 return ret;
1764}
1765#else
1766static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1767 struct sockaddr_storage *ss, char delim, const char **ipend)
1768{
1769 return -EINVAL;
1770}
1771#endif
1772
1773/*
1774 * Parse a server name (IP or hostname). If a valid IP address is not found
1775 * then try to extract a hostname to resolve using userspace DNS upcall.
1776 */
1777static int ceph_parse_server_name(const char *name, size_t namelen,
1778 struct sockaddr_storage *ss, char delim, const char **ipend)
1779{
1780 int ret;
1781
1782 ret = ceph_pton(name, namelen, ss, delim, ipend);
1783 if (ret)
1784 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1785
1786 return ret;
1787}
1788
31b8006e
SW		 1789/*
1790 * Parse an ip[:port] list into an addr array. Use the default
1791 * monitor port if a port isn't specified.
1792 */
1793int ceph_parse_ips(const char *c, const char *end,
1794 struct ceph_entity_addr *addr,
1795 int max_count, int *count)
1796{
ee3b56f2 1797 int i, ret = -EINVAL;
31b8006e
SW		 1798 const char *p = c;
1799
1800 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1801 for (i = 0; i < max_count; i++) {
1802 const char *ipend;
1803 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1804 int port;
39139f64
SW		 1805 char delim = ',';
1806
1807 if (*p == '[') {
1808 delim = ']';
1809 p++;
1810 }
31b8006e 1811
ee3b56f2
NW		 1812 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1813 if (ret)
31b8006e 1814 goto bad;
ee3b56f2
NW		 1815 ret = -EINVAL;
1816
31b8006e
SW		 1817 p = ipend;
1818
39139f64
SW		 1819 if (delim == ']') {
1820 if (*p != ']') {
1821 dout("missing matching ']'\n");
1822 goto bad;
1823 }
1824 p++;
1825 }
1826
31b8006e
SW		 1827 /* port? */
1828 if (p < end && *p == ':') {
1829 port = 0;
1830 p++;
1831 while (p < end && *p >= '0' && *p <= '9') {
1832 port = (port * 10) + (*p - '0');
1833 p++;
1834 }
1835 if (port > 65535 || port == 0)
1836 goto bad;
1837 } else {
1838 port = CEPH_MON_PORT;
1839 }
1840
1841 addr_set_port(ss, port);
1842
3d14c5d2 1843 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1844
1845 if (p == end)
1846 break;
1847 if (*p != ',')
1848 goto bad;
1849 p++;
1850 }
1851
1852 if (p != end)
1853 goto bad;
1854
1855 if (count)
1856 *count = i + 1;
1857 return 0;
1858
1859bad:
39139f64 1860 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1861 return ret;
31b8006e 1862}
3d14c5d2 1863EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1864
eed0ef2c 1865static int process_banner(struct ceph_connection *con)
31b8006e 1866{
eed0ef2c 1867 dout("process_banner on %p\n", con);
31b8006e
SW		 1868
1869 if (verify_hello(con) < 0)
1870 return -1;
1871
63f2d211
SW		 1872 ceph_decode_addr(&con->actual_peer_addr);
1873 ceph_decode_addr(&con->peer_addr_for_me);
1874
31b8006e
SW		 1875 /*
1876 * Make sure the other end is who we wanted. note that the other
1877 * end may not yet know their ip address, so if it's 0.0.0.0, give
1878 * them the benefit of the doubt.
1879 */
103e2d3a
SW		 1880 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1881 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1882 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1883 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
cd84db6e 1884 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
3d14c5d2 1885 ceph_pr_addr(&con->peer_addr.in_addr),
cd84db6e 1886 (int)le32_to_cpu(con->peer_addr.nonce),
3d14c5d2 1887 ceph_pr_addr(&con->actual_peer_addr.in_addr),
cd84db6e 1888 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1889 con->error_msg = "wrong peer at address";
31b8006e
SW		 1890 return -1;
1891 }
1892
1893 /*
1894 * did we learn our address?
1895 */
1896 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1897 int port = addr_port(&con->msgr->inst.addr.in_addr);
1898
1899 memcpy(&con->msgr->inst.addr.in_addr,
1900 &con->peer_addr_for_me.in_addr,
1901 sizeof(con->peer_addr_for_me.in_addr));
1902 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1903 encode_my_addr(con->msgr);
eed0ef2c 1904 dout("process_banner learned my addr is %s\n",
3d14c5d2 1905 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1906 }
1907
eed0ef2c
SW		 1908 return 0;
1909}
1910
1911static int process_connect(struct ceph_connection *con)
1912{
3d14c5d2
YS		 1913 u64 sup_feat = con->msgr->supported_features;
1914 u64 req_feat = con->msgr->required_features;
04a419f9 1915 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 1916 int ret;
04a419f9 1917
eed0ef2c
SW		 1918 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1919
31b8006e 1920 switch (con->in_reply.tag) {
04a419f9
SW		 1921 case CEPH_MSGR_TAG_FEATURES:
1922 pr_err("%s%lld %s feature set mismatch,"
1923 " my %llx < server's %llx, missing %llx\n",
1924 ENTITY_NAME(con->peer_name),
3d14c5d2 1925 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1926 sup_feat, server_feat, server_feat & ~sup_feat);
1927 con->error_msg = "missing required protocol features";
0fa6ebc6 1928 reset_connection(con);
04a419f9
SW		 1929 return -1;
1930
31b8006e 1931 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 1932 pr_err("%s%lld %s protocol version mismatch,"
1933 " my %d != server's %d\n",
1934 ENTITY_NAME(con->peer_name),
3d14c5d2 1935 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 1936 le32_to_cpu(con->out_connect.protocol_version),
1937 le32_to_cpu(con->in_reply.protocol_version));
1938 con->error_msg = "protocol version mismatch";
0fa6ebc6 1939 reset_connection(con);
31b8006e
SW		 1940 return -1;
1941
4e7a5dcd
SW		 1942 case CEPH_MSGR_TAG_BADAUTHORIZER:
1943 con->auth_retry++;
1944 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1945 con->auth_retry);
1946 if (con->auth_retry == 2) {
1947 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 1948 return -1;
1949 }
6d4221b5 1950 con_out_kvec_reset(con);
e825a66d 1951 ret = prepare_write_connect(con);
0da5d703
SW		 1952 if (ret < 0)
1953 return ret;
63733a0f 1954 prepare_read_connect(con);
4e7a5dcd 1955 break;
31b8006e
SW		 1956
1957 case CEPH_MSGR_TAG_RESETSESSION:
1958 /*
1959 * If we connected with a large connect_seq but the peer
1960 * has no record of a session with us (no connection, or
1961 * connect_seq == 0), they will send RESETSESION to indicate
1962 * that they must have reset their session, and may have
1963 * dropped messages.
1964 */
1965 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 1966 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 1967 pr_err("%s%lld %s connection reset\n",
1968 ENTITY_NAME(con->peer_name),
3d14c5d2 1969 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 1970 reset_connection(con);
6d4221b5 1971 con_out_kvec_reset(con);
5a0f8fdd
AE		 1972 ret = prepare_write_connect(con);
1973 if (ret < 0)
1974 return ret;
31b8006e
SW		 1975 prepare_read_connect(con);
1976
1977 /* Tell ceph about it. */
ec302645 1978 mutex_unlock(&con->mutex);
31b8006e
SW		 1979 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1980 if (con->ops->peer_reset)
1981 con->ops->peer_reset(con);
ec302645 1982 mutex_lock(&con->mutex);
8dacc7da 1983 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 1984 return -EAGAIN;
31b8006e
SW		 1985 break;
1986
1987 case CEPH_MSGR_TAG_RETRY_SESSION:
1988 /*
1989 * If we sent a smaller connect_seq than the peer has, try
1990 * again with a larger value.
1991 */
5bdca4e0 1992 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 1993 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 1994 le32_to_cpu(con->in_reply.connect_seq));
1995 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 1996 con_out_kvec_reset(con);
5a0f8fdd
AE		 1997 ret = prepare_write_connect(con);
1998 if (ret < 0)
1999 return ret;
31b8006e
SW		 2000 prepare_read_connect(con);
2001 break;
2002
2003 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2004 /*
2005 * If we sent a smaller global_seq than the peer has, try
2006 * again with a larger value.
2007 */
eed0ef2c 2008 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2009 con->peer_global_seq,
5bdca4e0 2010 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2011 get_global_seq(con->msgr,
5bdca4e0 2012 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2013 con_out_kvec_reset(con);
5a0f8fdd
AE		 2014 ret = prepare_write_connect(con);
2015 if (ret < 0)
2016 return ret;
31b8006e
SW		 2017 prepare_read_connect(con);
2018 break;
2019
3a23083b 2020 case CEPH_MSGR_TAG_SEQ:
31b8006e 2021 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2022 if (req_feat & ~server_feat) {
2023 pr_err("%s%lld %s protocol feature mismatch,"
2024 " my required %llx > server's %llx, need %llx\n",
2025 ENTITY_NAME(con->peer_name),
3d14c5d2 2026 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2027 req_feat, server_feat, req_feat & ~server_feat);
2028 con->error_msg = "missing required protocol features";
0fa6ebc6 2029 reset_connection(con);
04a419f9
SW		 2030 return -1;
2031 }
8dacc7da 2032
122070a2 2033 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2034 con->state = CON_STATE_OPEN;
20e55c4c 2035 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2036 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2037 con->connect_seq++;
aba558e2 2038 con->peer_features = server_feat;
31b8006e
SW		 2039 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2040 con->peer_global_seq,
2041 le32_to_cpu(con->in_reply.connect_seq),
2042 con->connect_seq);
2043 WARN_ON(con->connect_seq !=
2044 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2045
2046 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2047 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2048
85effe18 2049 con->delay = 0; /* reset backoff memory */
92ac41d0 2050
3a23083b
SW		 2051 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2052 prepare_write_seq(con);
2053 prepare_read_seq(con);
2054 } else {
2055 prepare_read_tag(con);
2056 }
31b8006e
SW		 2057 break;
2058
2059 case CEPH_MSGR_TAG_WAIT:
2060 /*
2061 * If there is a connection race (we are opening
2062 * connections to each other), one of us may just have
2063 * to WAIT. This shouldn't happen if we are the
2064 * client.
2065 */
04177882
SW		 2066 pr_err("process_connect got WAIT as client\n");
2067 con->error_msg = "protocol error, got WAIT as client";
2068 return -1;
31b8006e
SW		 2069
2070 default:
2071 pr_err("connect protocol error, will retry\n");
2072 con->error_msg = "protocol error, garbage tag during connect";
2073 return -1;
2074 }
2075 return 0;
2076}
2077
2078
2079/*
2080 * read (part of) an ack
2081 */
2082static int read_partial_ack(struct ceph_connection *con)
2083{
fd51653f
AE		 2084 int size = sizeof (con->in_temp_ack);
2085 int end = size;
31b8006e 2086
fd51653f 2087 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2088}
2089
31b8006e
SW		 2090/*
2091 * We can finally discard anything that's been acked.
2092 */
2093static void process_ack(struct ceph_connection *con)
2094{
2095 struct ceph_msg *m;
2096 u64 ack = le64_to_cpu(con->in_temp_ack);
2097 u64 seq;
2098
31b8006e
SW		 2099 while (!list_empty(&con->out_sent)) {
2100 m = list_first_entry(&con->out_sent, struct ceph_msg,
2101 list_head);
2102 seq = le64_to_cpu(m->hdr.seq);
2103 if (seq > ack)
2104 break;
2105 dout("got ack for seq %llu type %d at %p\n", seq,
2106 le16_to_cpu(m->hdr.type), m);
4cf9d544 2107 m->ack_stamp = jiffies;
31b8006e
SW		 2108 ceph_msg_remove(m);
2109 }
31b8006e
SW		 2110 prepare_read_tag(con);
2111}
2112
2113
2450418c 2114static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2115 struct kvec *section,
2116 unsigned int sec_len, u32 *crc)
2450418c 2117{
68b4476b 2118 int ret, left;
2450418c
YS		 2119
2120 BUG_ON(!section);
2121
2122 while (section->iov_len < sec_len) {
2123 BUG_ON(section->iov_base == NULL);
2124 left = sec_len - section->iov_len;
2125 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2126 section->iov_len, left);
2127 if (ret <= 0)
2128 return ret;
2129 section->iov_len += ret;
2450418c 2130 }
fe3ad593
AE		 2131 if (section->iov_len == sec_len)
2132 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2133
2450418c
YS		 2134 return 1;
2135}
31b8006e 2136
34d2d200
AE		 2137static int read_partial_msg_data(struct ceph_connection *con)
2138{
2139 struct ceph_msg *msg = con->in_msg;
643c68a4 2140 struct ceph_msg_data_cursor *cursor = &msg->data.cursor;
34d2d200 2141 const bool do_datacrc = !con->msgr->nocrc;
686be208
AE		 2142 struct page *page;
2143 size_t page_offset;
2144 size_t length;
f5db90bc 2145 u32 crc = 0;
34d2d200
AE		 2146 int ret;
2147
2148 BUG_ON(!msg);
4c59b4a2
AE		 2149 if (WARN_ON(!ceph_msg_has_data(msg)))
2150 return -EIO;
34d2d200 2151
f5db90bc
AE		 2152 if (do_datacrc)
2153 crc = con->in_data_crc;
643c68a4 2154 while (cursor->resid) {
4c59b4a2
AE		 2155 page = ceph_msg_data_next(&msg->data, &page_offset, &length,
2156 NULL);
686be208 2157 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2158 if (ret <= 0) {
2159 if (do_datacrc)
2160 con->in_data_crc = crc;
2161
686be208 2162 return ret;
f5db90bc 2163 }
686be208
AE		 2164
2165 if (do_datacrc)
f5db90bc 2166 crc = ceph_crc32c_page(crc, page, page_offset, ret);
686be208 2167 in_msg_pos_next(con, length, ret);
34d2d200 2168 }
f5db90bc
AE		 2169 if (do_datacrc)
2170 con->in_data_crc = crc;
34d2d200
AE		 2171
2172 return 1; /* must return > 0 to indicate success */
2173}
2174
31b8006e
SW		 2175/*
2176 * read (part of) a message.
2177 */
686be208
AE		 2178static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2179
31b8006e
SW		 2180static int read_partial_message(struct ceph_connection *con)
2181{
2182 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2183 int size;
2184 int end;
31b8006e 2185 int ret;
95c96174 2186 unsigned int front_len, middle_len, data_len;
37675b0f 2187 bool do_datacrc = !con->msgr->nocrc;
ae18756b 2188 u64 seq;
fe3ad593 2189 u32 crc;
31b8006e
SW		 2190
2191 dout("read_partial_message con %p msg %p\n", con, m);
2192
2193 /* header */
fd51653f
AE		 2194 size = sizeof (con->in_hdr);
2195 end = size;
2196 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2197 if (ret <= 0)
2198 return ret;
fe3ad593
AE		 2199
2200 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2201 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2202 pr_err("read_partial_message bad hdr "
2203 " crc %u != expected %u\n",
2204 crc, con->in_hdr.crc);
2205 return -EBADMSG;
2206 }
2207
31b8006e
SW		 2208 front_len = le32_to_cpu(con->in_hdr.front_len);
2209 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2210 return -EIO;
2211 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2212 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2213 return -EIO;
2214 data_len = le32_to_cpu(con->in_hdr.data_len);
2215 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2216 return -EIO;
2217
ae18756b
SW		 2218 /* verify seq# */
2219 seq = le64_to_cpu(con->in_hdr.seq);
2220 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2221 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2222 ENTITY_NAME(con->peer_name),
3d14c5d2 2223 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2224 seq, con->in_seq + 1);
2225 con->in_base_pos = -front_len - middle_len - data_len -
2226 sizeof(m->footer);
2227 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2228 return 0;
2229 } else if ((s64)seq - (s64)con->in_seq > 1) {
2230 pr_err("read_partial_message bad seq %lld expected %lld\n",
2231 seq, con->in_seq + 1);
2232 con->error_msg = "bad message sequence # for incoming message";
2233 return -EBADMSG;
2234 }
2235
31b8006e
SW		 2236 /* allocate message? */
2237 if (!con->in_msg) {
4740a623
SW		 2238 int skip = 0;
2239
31b8006e 2240 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2241 front_len, data_len);
4740a623
SW		 2242 ret = ceph_con_in_msg_alloc(con, &skip);
2243 if (ret < 0)
2244 return ret;
2450418c 2245 if (skip) {
31b8006e 2246 /* skip this message */
a79832f2 2247 dout("alloc_msg said skip message\n");
ae32be31 2248 BUG_ON(con->in_msg);
31b8006e
SW		 2249 con->in_base_pos = -front_len - middle_len - data_len -
2250 sizeof(m->footer);
2251 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2252 con->in_seq++;
31b8006e
SW		 2253 return 0;
2254 }
38941f80 2255
4740a623 2256 BUG_ON(!con->in_msg);
38941f80 2257 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2258 m = con->in_msg;
2259 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2260 if (m->middle)
2261 m->middle->vec.iov_len = 0;
9d7f0f13 2262
78625051 2263 /* prepare for data payload, if any */
a4107026 2264
78625051 2265 if (data_len)
f5db90bc 2266 prepare_message_data(con->in_msg);
31b8006e
SW		 2267 }
2268
2269 /* front */
2450418c
YS		 2270 ret = read_partial_message_section(con, &m->front, front_len,
2271 &con->in_front_crc);
2272 if (ret <= 0)
2273 return ret;
31b8006e
SW		 2274
2275 /* middle */
2450418c 2276 if (m->middle) {
213c99ee
SW		 2277 ret = read_partial_message_section(con, &m->middle->vec,
2278 middle_len,
2450418c 2279 &con->in_middle_crc);
31b8006e
SW		 2280 if (ret <= 0)
2281 return ret;
31b8006e
SW		 2282 }
2283
2284 /* (page) data */
34d2d200
AE		 2285 if (data_len) {
2286 ret = read_partial_msg_data(con);
2287 if (ret <= 0)
2288 return ret;
31b8006e
SW		 2289 }
2290
31b8006e 2291 /* footer */
fd51653f
AE		 2292 size = sizeof (m->footer);
2293 end += size;
2294 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2295 if (ret <= 0)
2296 return ret;
2297
31b8006e
SW		 2298 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2299 m, front_len, m->footer.front_crc, middle_len,
2300 m->footer.middle_crc, data_len, m->footer.data_crc);
2301
2302 /* crc ok? */
2303 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2304 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2305 m, con->in_front_crc, m->footer.front_crc);
2306 return -EBADMSG;
2307 }
2308 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2309 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2310 m, con->in_middle_crc, m->footer.middle_crc);
2311 return -EBADMSG;
2312 }
bca064d2 2313 if (do_datacrc &&
31b8006e
SW		 2314 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2315 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2316 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2317 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2318 return -EBADMSG;
2319 }
2320
2321 return 1; /* done! */
2322}
2323
2324/*
2325 * Process message. This happens in the worker thread. The callback should
2326 * be careful not to do anything that waits on other incoming messages or it
2327 * may deadlock.
2328 */
2329static void process_message(struct ceph_connection *con)
2330{
5e095e8b 2331 struct ceph_msg *msg;
31b8006e 2332
38941f80
AE		 2333 BUG_ON(con->in_msg->con != con);
2334 con->in_msg->con = NULL;
5e095e8b 2335 msg = con->in_msg;
31b8006e 2336 con->in_msg = NULL;
36eb71aa 2337 con->ops->put(con);
31b8006e
SW		 2338
2339 /* if first message, set peer_name */
2340 if (con->peer_name.type == 0)
dbad185d 2341 con->peer_name = msg->hdr.src;
31b8006e 2342
31b8006e 2343 con->in_seq++;
ec302645 2344 mutex_unlock(&con->mutex);
31b8006e
SW		 2345
2346 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2347 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2348 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2349 le16_to_cpu(msg->hdr.type),
2350 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2351 le32_to_cpu(msg->hdr.front_len),
2352 le32_to_cpu(msg->hdr.data_len),
2353 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2354 con->ops->dispatch(con, msg);
ec302645
SW		 2355
2356 mutex_lock(&con->mutex);
31b8006e
SW		 2357}
2358
2359
2360/*
2361 * Write something to the socket. Called in a worker thread when the
2362 * socket appears to be writeable and we have something ready to send.
2363 */
2364static int try_write(struct ceph_connection *con)
2365{
31b8006e
SW		 2366 int ret = 1;
2367
d59315ca 2368 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2369
31b8006e
SW		 2370more:
2371 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2372
2373 /* open the socket first? */
8dacc7da
SW		 2374 if (con->state == CON_STATE_PREOPEN) {
2375 BUG_ON(con->sock);
2376 con->state = CON_STATE_CONNECTING;
a5988c49 2377
e2200423 2378 con_out_kvec_reset(con);
e825a66d 2379 prepare_write_banner(con);
eed0ef2c 2380 prepare_read_banner(con);
31b8006e 2381
cf3e5c40 2382 BUG_ON(con->in_msg);
31b8006e
SW		 2383 con->in_tag = CEPH_MSGR_TAG_READY;
2384 dout("try_write initiating connect on %p new state %lu\n",
2385 con, con->state);
41617d0c
AE		 2386 ret = ceph_tcp_connect(con);
2387 if (ret < 0) {
31b8006e 2388 con->error_msg = "connect error";
31b8006e
SW		 2389 goto out;
2390 }
2391 }
2392
2393more_kvec:
2394 /* kvec data queued? */
2395 if (con->out_skip) {
2396 ret = write_partial_skip(con);
2397 if (ret <= 0)
42961d23 2398 goto out;
31b8006e
SW		 2399 }
2400 if (con->out_kvec_left) {
2401 ret = write_partial_kvec(con);
2402 if (ret <= 0)
42961d23 2403 goto out;
31b8006e
SW		 2404 }
2405
2406 /* msg pages? */
2407 if (con->out_msg) {
c86a2930
SW		 2408 if (con->out_msg_done) {
2409 ceph_msg_put(con->out_msg);
2410 con->out_msg = NULL; /* we're done with this one */
2411 goto do_next;
2412 }
2413
34d2d200 2414 ret = write_partial_message_data(con);
31b8006e
SW		 2415 if (ret == 1)
2416 goto more_kvec; /* we need to send the footer, too! */
2417 if (ret == 0)
42961d23 2418 goto out;
31b8006e 2419 if (ret < 0) {
34d2d200 2420 dout("try_write write_partial_message_data err %d\n",
31b8006e 2421 ret);
42961d23 2422 goto out;
31b8006e
SW		 2423 }
2424 }
2425
c86a2930 2426do_next:
8dacc7da 2427 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2428 /* is anything else pending? */
2429 if (!list_empty(&con->out_queue)) {
2430 prepare_write_message(con);
2431 goto more;
2432 }
2433 if (con->in_seq > con->in_seq_acked) {
2434 prepare_write_ack(con);
2435 goto more;
2436 }
c9ffc77a 2437 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2438 prepare_write_keepalive(con);
2439 goto more;
2440 }
2441 }
2442
2443 /* Nothing to do! */
c9ffc77a 2444 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2445 dout("try_write nothing else to write.\n");
31b8006e
SW		 2446 ret = 0;
2447out:
42961d23 2448 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2449 return ret;
2450}
2451
2452
2453
2454/*
2455 * Read what we can from the socket.
2456 */
2457static int try_read(struct ceph_connection *con)
2458{
31b8006e
SW		 2459 int ret = -1;
2460
8dacc7da
SW		 2461more:
2462 dout("try_read start on %p state %lu\n", con, con->state);
2463 if (con->state != CON_STATE_CONNECTING &&
2464 con->state != CON_STATE_NEGOTIATING &&
2465 con->state != CON_STATE_OPEN)
31b8006e
SW		 2466 return 0;
2467
8dacc7da 2468 BUG_ON(!con->sock);
ec302645 2469
31b8006e
SW		 2470 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2471 con->in_base_pos);
0da5d703 2472
8dacc7da 2473 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2474 dout("try_read connecting\n");
2475 ret = read_partial_banner(con);
2476 if (ret <= 0)
ab166d5a 2477 goto out;
7593af92
AE		 2478 ret = process_banner(con);
2479 if (ret < 0)
2480 goto out;
2481
8dacc7da 2482 con->state = CON_STATE_NEGOTIATING;
7593af92 2483
6d4221b5
JS		 2484 /*
2485 * Received banner is good, exchange connection info.
2486 * Do not reset out_kvec, as sending our banner raced
2487 * with receiving peer banner after connect completed.
2488 */
7593af92
AE		 2489 ret = prepare_write_connect(con);
2490 if (ret < 0)
2491 goto out;
2492 prepare_read_connect(con);
2493
2494 /* Send connection info before awaiting response */
0da5d703
SW		 2495 goto out;
2496 }
2497
8dacc7da 2498 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2499 dout("try_read negotiating\n");
31b8006e
SW		 2500 ret = read_partial_connect(con);
2501 if (ret <= 0)
31b8006e 2502 goto out;
98bdb0aa
SW		 2503 ret = process_connect(con);
2504 if (ret < 0)
2505 goto out;
31b8006e
SW		 2506 goto more;
2507 }
2508
122070a2 2509 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2510
31b8006e
SW		 2511 if (con->in_base_pos < 0) {
2512 /*
2513 * skipping + discarding content.
2514 *
2515 * FIXME: there must be a better way to do this!
2516 */
84495f49
AE		 2517 static char buf[SKIP_BUF_SIZE];
2518 int skip = min((int) sizeof (buf), -con->in_base_pos);
2519
31b8006e
SW		 2520 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2521 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2522 if (ret <= 0)
98bdb0aa 2523 goto out;
31b8006e
SW		 2524 con->in_base_pos += ret;
2525 if (con->in_base_pos)
2526 goto more;
2527 }
2528 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2529 /*
2530 * what's next?
2531 */
2532 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2533 if (ret <= 0)
98bdb0aa 2534 goto out;
31b8006e
SW		 2535 dout("try_read got tag %d\n", (int)con->in_tag);
2536 switch (con->in_tag) {
2537 case CEPH_MSGR_TAG_MSG:
2538 prepare_read_message(con);
2539 break;
2540 case CEPH_MSGR_TAG_ACK:
2541 prepare_read_ack(con);
2542 break;
2543 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2544 con_close_socket(con);
2545 con->state = CON_STATE_CLOSED;
98bdb0aa 2546 goto out;
31b8006e
SW		 2547 default:
2548 goto bad_tag;
2549 }
2550 }
2551 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2552 ret = read_partial_message(con);
2553 if (ret <= 0) {
2554 switch (ret) {
2555 case -EBADMSG:
2556 con->error_msg = "bad crc";
2557 ret = -EIO;
98bdb0aa 2558 break;
31b8006e
SW		 2559 case -EIO:
2560 con->error_msg = "io error";
98bdb0aa 2561 break;
31b8006e 2562 }
98bdb0aa 2563 goto out;
31b8006e
SW		 2564 }
2565 if (con->in_tag == CEPH_MSGR_TAG_READY)
2566 goto more;
2567 process_message(con);
7b862e07
SW		 2568 if (con->state == CON_STATE_OPEN)
2569 prepare_read_tag(con);
31b8006e
SW		 2570 goto more;
2571 }
3a23083b
SW		 2572 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2573 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2574 /*
2575 * the final handshake seq exchange is semantically
2576 * equivalent to an ACK
2577 */
31b8006e
SW		 2578 ret = read_partial_ack(con);
2579 if (ret <= 0)
98bdb0aa 2580 goto out;
31b8006e
SW		 2581 process_ack(con);
2582 goto more;
2583 }
2584
31b8006e 2585out:
98bdb0aa 2586 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2587 return ret;
2588
2589bad_tag:
2590 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2591 con->error_msg = "protocol error, garbage tag";
2592 ret = -1;
2593 goto out;
2594}
2595
2596
2597/*
802c6d96
AE		 2598 * Atomically queue work on a connection after the specified delay.
2599 * Bump @con reference to avoid races with connection teardown.
2600 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2601 */
802c6d96 2602static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2603{
31b8006e 2604 if (!con->ops->get(con)) {
802c6d96
AE		 2605 dout("%s %p ref count 0\n", __func__, con);
2606
2607 return -ENOENT;
31b8006e
SW		 2608 }
2609
802c6d96
AE		 2610 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2611 dout("%s %p - already queued\n", __func__, con);
31b8006e 2612 con->ops->put(con);
802c6d96
AE		 2613
2614 return -EBUSY;
31b8006e 2615 }
802c6d96
AE		 2616
2617 dout("%s %p %lu\n", __func__, con, delay);
2618
2619 return 0;
2620}
2621
2622static void queue_con(struct ceph_connection *con)
2623{
2624 (void) queue_con_delay(con, 0);
31b8006e
SW		 2625}
2626
7bb21d68
AE		 2627static bool con_sock_closed(struct ceph_connection *con)
2628{
c9ffc77a 2629 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2630 return false;
2631
2632#define CASE(x) \
2633 case CON_STATE_ ## x: \
2634 con->error_msg = "socket closed (con state " #x ")"; \
2635 break;
2636
2637 switch (con->state) {
2638 CASE(CLOSED);
2639 CASE(PREOPEN);
2640 CASE(CONNECTING);
2641 CASE(NEGOTIATING);
2642 CASE(OPEN);
2643 CASE(STANDBY);
2644 default:
2645 pr_warning("%s con %p unrecognized state %lu\n",
2646 __func__, con, con->state);
2647 con->error_msg = "unrecognized con state";
2648 BUG();
2649 break;
2650 }
2651#undef CASE
2652
2653 return true;
2654}
2655
f20a39fd
AE		 2656static bool con_backoff(struct ceph_connection *con)
2657{
2658 int ret;
2659
2660 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2661 return false;
2662
2663 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2664 if (ret) {
2665 dout("%s: con %p FAILED to back off %lu\n", __func__,
2666 con, con->delay);
2667 BUG_ON(ret == -ENOENT);
2668 con_flag_set(con, CON_FLAG_BACKOFF);
2669 }
2670
2671 return true;
2672}
2673
93209264
AE		 2674/* Finish fault handling; con->mutex must *not* be held here */
2675
2676static void con_fault_finish(struct ceph_connection *con)
2677{
2678 /*
2679 * in case we faulted due to authentication, invalidate our
2680 * current tickets so that we can get new ones.
2681 */
2682 if (con->auth_retry && con->ops->invalidate_authorizer) {
2683 dout("calling invalidate_authorizer()\n");
2684 con->ops->invalidate_authorizer(con);
2685 }
2686
2687 if (con->ops->fault)
2688 con->ops->fault(con);
2689}
2690
31b8006e
SW		 2691/*
2692 * Do some work on a connection. Drop a connection ref when we're done.
2693 */
2694static void con_work(struct work_struct *work)
2695{
2696 struct ceph_connection *con = container_of(work, struct ceph_connection,
2697 work.work);
49659416 2698 bool fault;
31b8006e 2699
9dd4658d 2700 mutex_lock(&con->mutex);
49659416
AE		 2701 while (true) {
2702 int ret;
31b8006e 2703
49659416
AE		 2704 if ((fault = con_sock_closed(con))) {
2705 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2706 break;
2707 }
2708 if (con_backoff(con)) {
2709 dout("%s: con %p BACKOFF\n", __func__, con);
2710 break;
2711 }
2712 if (con->state == CON_STATE_STANDBY) {
2713 dout("%s: con %p STANDBY\n", __func__, con);
2714 break;
2715 }
2716 if (con->state == CON_STATE_CLOSED) {
2717 dout("%s: con %p CLOSED\n", __func__, con);
2718 BUG_ON(con->sock);
2719 break;
2720 }
2721 if (con->state == CON_STATE_PREOPEN) {
2722 dout("%s: con %p PREOPEN\n", __func__, con);
2723 BUG_ON(con->sock);
2724 }
0da5d703 2725
49659416
AE		 2726 ret = try_read(con);
2727 if (ret < 0) {
2728 if (ret == -EAGAIN)
2729 continue;
2730 con->error_msg = "socket error on read";
2731 fault = true;
2732 break;
2733 }
2734
2735 ret = try_write(con);
2736 if (ret < 0) {
2737 if (ret == -EAGAIN)
2738 continue;
2739 con->error_msg = "socket error on write";
2740 fault = true;
2741 }
2742
2743 break; /* If we make it to here, we're done */
3a140a0d 2744 }
b6e7b6a1
AE		 2745 if (fault)
2746 con_fault(con);
9dd4658d 2747 mutex_unlock(&con->mutex);
0da5d703 2748
b6e7b6a1
AE		 2749 if (fault)
2750 con_fault_finish(con);
2751
2752 con->ops->put(con);
31b8006e
SW		 2753}
2754
31b8006e
SW		 2755/*
2756 * Generic error/fault handler. A retry mechanism is used with
2757 * exponential backoff
2758 */
93209264 2759static void con_fault(struct ceph_connection *con)
31b8006e 2760{
28362986 2761 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
3d14c5d2 2762 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2763 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2764 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2765
122070a2 2766 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2767 con->state != CON_STATE_NEGOTIATING &&
2768 con->state != CON_STATE_OPEN);
ec302645 2769
31b8006e 2770 con_close_socket(con);
5e095e8b 2771
c9ffc77a 2772 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2773 dout("fault on LOSSYTX channel, marking CLOSED\n");
2774 con->state = CON_STATE_CLOSED;
93209264 2775 return;
3b5ede07
SW		 2776 }
2777
5e095e8b 2778 if (con->in_msg) {
38941f80
AE		 2779 BUG_ON(con->in_msg->con != con);
2780 con->in_msg->con = NULL;
5e095e8b
SW		 2781 ceph_msg_put(con->in_msg);
2782 con->in_msg = NULL;
36eb71aa 2783 con->ops->put(con);
5e095e8b 2784 }
31b8006e 2785
e80a52d1
SW		 2786 /* Requeue anything that hasn't been acked */
2787 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2788
e76661d0
SW		 2789 /* If there are no messages queued or keepalive pending, place
2790 * the connection in a STANDBY state */
2791 if (list_empty(&con->out_queue) &&
c9ffc77a 2792 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2793 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2794 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2795 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2796 } else {
2797 /* retry after a delay. */
8dacc7da 2798 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2799 if (con->delay == 0)
2800 con->delay = BASE_DELAY_INTERVAL;
2801 else if (con->delay < MAX_DELAY_INTERVAL)
2802 con->delay *= 2;
c9ffc77a 2803 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2804 queue_con(con);
31b8006e 2805 }
31b8006e
SW		 2806}
2807
2808
2809
2810/*
15d9882c 2811 * initialize a new messenger instance
31b8006e 2812 */
15d9882c
AE		 2813void ceph_messenger_init(struct ceph_messenger *msgr,
2814 struct ceph_entity_addr *myaddr,
2815 u32 supported_features,
2816 u32 required_features,
2817 bool nocrc)
31b8006e 2818{
3d14c5d2
YS		 2819 msgr->supported_features = supported_features;
2820 msgr->required_features = required_features;
2821
31b8006e
SW		 2822 spin_lock_init(&msgr->global_seq_lock);
2823
31b8006e
SW		 2824 if (myaddr)
2825 msgr->inst.addr = *myaddr;
2826
2827 /* select a random nonce */
ac8839d7 2828 msgr->inst.addr.type = 0;
103e2d3a 2829 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2830 encode_my_addr(msgr);
15d9882c 2831 msgr->nocrc = nocrc;
31b8006e 2832
a2a32584 2833 atomic_set(&msgr->stopping, 0);
31b8006e 2834
15d9882c 2835 dout("%s %p\n", __func__, msgr);
31b8006e 2836}
15d9882c 2837EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2838
e00de341
SW		 2839static void clear_standby(struct ceph_connection *con)
2840{
2841 /* come back from STANDBY? */
8dacc7da 2842 if (con->state == CON_STATE_STANDBY) {
e00de341 2843 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2844 con->state = CON_STATE_PREOPEN;
e00de341 2845 con->connect_seq++;
c9ffc77a
AE		 2846 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2847 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2848 }
2849}
2850
31b8006e
SW		 2851/*
2852 * Queue up an outgoing message on the given connection.
2853 */
2854void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2855{
31b8006e 2856 /* set src+dst */
dbad185d 2857 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2858 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2859 msg->needs_out_seq = true;
2860
ec302645 2861 mutex_lock(&con->mutex);
92ce034b 2862
8dacc7da 2863 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2864 dout("con_send %p closed, dropping %p\n", con, msg);
2865 ceph_msg_put(msg);
2866 mutex_unlock(&con->mutex);
2867 return;
2868 }
2869
38941f80 2870 BUG_ON(msg->con != NULL);
36eb71aa 2871 msg->con = con->ops->get(con);
92ce034b
AE		 2872 BUG_ON(msg->con == NULL);
2873
31b8006e
SW		 2874 BUG_ON(!list_empty(&msg->list_head));
2875 list_add_tail(&msg->list_head, &con->out_queue);
2876 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2877 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2878 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2879 le32_to_cpu(msg->hdr.front_len),
2880 le32_to_cpu(msg->hdr.middle_len),
2881 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 2882
2883 clear_standby(con);
ec302645 2884 mutex_unlock(&con->mutex);
31b8006e
SW		 2885
2886 /* if there wasn't anything waiting to send before, queue
2887 * new work */
c9ffc77a 2888 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2889 queue_con(con);
2890}
3d14c5d2 2891EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 2892
2893/*
2894 * Revoke a message that was previously queued for send
2895 */
6740a845 2896void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 2897{
6740a845
AE		 2898 struct ceph_connection *con = msg->con;
2899
2900 if (!con)
2901 return; /* Message not in our possession */
2902
ec302645 2903 mutex_lock(&con->mutex);
31b8006e 2904 if (!list_empty(&msg->list_head)) {
38941f80 2905 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 2906 list_del_init(&msg->list_head);
38941f80 2907 BUG_ON(msg->con == NULL);
36eb71aa 2908 msg->con->ops->put(msg->con);
38941f80 2909 msg->con = NULL;
31b8006e 2910 msg->hdr.seq = 0;
38941f80 2911
31b8006e 2912 ceph_msg_put(msg);
ed98adad
SW		 2913 }
2914 if (con->out_msg == msg) {
38941f80 2915 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 2916 con->out_msg = NULL;
31b8006e
SW		 2917 if (con->out_kvec_is_msg) {
2918 con->out_skip = con->out_kvec_bytes;
2919 con->out_kvec_is_msg = false;
2920 }
ed98adad 2921 msg->hdr.seq = 0;
92ce034b
AE		 2922
2923 ceph_msg_put(msg);
31b8006e 2924 }
ec302645 2925 mutex_unlock(&con->mutex);
31b8006e
SW		 2926}
2927
350b1c32 2928/*
0d59ab81 2929 * Revoke a message that we may be reading data into
350b1c32 2930 */
8921d114 2931void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 2932{
8921d114
AE		 2933 struct ceph_connection *con;
2934
2935 BUG_ON(msg == NULL);
2936 if (!msg->con) {
2937 dout("%s msg %p null con\n", __func__, msg);
2938
2939 return; /* Message not in our possession */
2940 }
2941
2942 con = msg->con;
350b1c32 2943 mutex_lock(&con->mutex);
8921d114 2944 if (con->in_msg == msg) {
95c96174
ED		 2945 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2946 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2947 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 2948
2949 /* skip rest of message */
8921d114
AE		 2950 dout("%s %p msg %p revoked\n", __func__, con, msg);
2951 con->in_base_pos = con->in_base_pos -
350b1c32 2952 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 2953 front_len -
2954 middle_len -
2955 data_len -
350b1c32 2956 sizeof(struct ceph_msg_footer);
350b1c32
SW		 2957 ceph_msg_put(con->in_msg);
2958 con->in_msg = NULL;
2959 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2960 con->in_seq++;
350b1c32 2961 } else {
8921d114
AE		 2962 dout("%s %p in_msg %p msg %p no-op\n",
2963 __func__, con, con->in_msg, msg);
350b1c32
SW		 2964 }
2965 mutex_unlock(&con->mutex);
2966}
2967
31b8006e
SW		 2968/*
2969 * Queue a keepalive byte to ensure the tcp connection is alive.
2970 */
2971void ceph_con_keepalive(struct ceph_connection *con)
2972{
e00de341 2973 dout("con_keepalive %p\n", con);
00650931 2974 mutex_lock(&con->mutex);
e00de341 2975 clear_standby(con);
00650931 2976 mutex_unlock(&con->mutex);
c9ffc77a
AE		 2977 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
2978 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2979 queue_con(con);
2980}
3d14c5d2 2981EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 2982
43794509
AE		 2983static void ceph_msg_data_init(struct ceph_msg_data *data)
2984{
2985 data->type = CEPH_MSG_DATA_NONE;
2986}
2987
02afca6c 2988void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 2989 size_t length, size_t alignment)
02afca6c 2990{
07aa1558
AE		 2991 BUG_ON(!pages);
2992 BUG_ON(!length);
4c59b4a2 2993 BUG_ON(msg->data.type != CEPH_MSG_DATA_NONE);
02afca6c 2994
4c59b4a2
AE		 2995 msg->data.type = CEPH_MSG_DATA_PAGES;
2996 msg->data.pages = pages;
2997 msg->data.length = length;
2998 msg->data.alignment = alignment & ~PAGE_MASK;
02afca6c
AE		 2999}
3000EXPORT_SYMBOL(ceph_msg_data_set_pages);
31b8006e 3001
27fa8385
AE		 3002void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3003 struct ceph_pagelist *pagelist)
3004{
07aa1558
AE		 3005 BUG_ON(!pagelist);
3006 BUG_ON(!pagelist->length);
4c59b4a2 3007 BUG_ON(msg->data.type != CEPH_MSG_DATA_NONE);
27fa8385 3008
4c59b4a2
AE		 3009 msg->data.type = CEPH_MSG_DATA_PAGELIST;
3010 msg->data.pagelist = pagelist;
27fa8385
AE		 3011}
3012EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3013
3014void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio)
3015{
07aa1558 3016 BUG_ON(!bio);
4c59b4a2 3017 BUG_ON(msg->data.type != CEPH_MSG_DATA_NONE);
27fa8385 3018
4c59b4a2
AE		 3019 msg->data.type = CEPH_MSG_DATA_BIO;
3020 msg->data.bio = bio;
27fa8385
AE		 3021}
3022EXPORT_SYMBOL(ceph_msg_data_set_bio);
3023
31b8006e
SW		 3024/*
3025 * construct a new message with given type, size
3026 * the new msg has a ref count of 1.
3027 */
b61c2763
SW		 3028struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3029 bool can_fail)
31b8006e
SW		 3030{
3031 struct ceph_msg *m;
3032
9516e45b 3033 m = kzalloc(sizeof(*m), flags);
31b8006e
SW		 3034 if (m == NULL)
3035 goto out;
31b8006e
SW		 3036
3037 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3038 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3039 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3040
9516e45b
AE		 3041 INIT_LIST_HEAD(&m->list_head);
3042 kref_init(&m->kref);
ca20892d 3043
4c59b4a2 3044 ceph_msg_data_init(&m->data);
43794509 3045
31b8006e 3046 /* front */
9516e45b 3047 m->front_max = front_len;
31b8006e
SW		 3048 if (front_len) {
3049 if (front_len > PAGE_CACHE_SIZE) {
34d23762 3050 m->front.iov_base = __vmalloc(front_len, flags,
31b8006e
SW		 3051 PAGE_KERNEL);
3052 m->front_is_vmalloc = true;
3053 } else {
34d23762 3054 m->front.iov_base = kmalloc(front_len, flags);
31b8006e
SW		 3055 }
3056 if (m->front.iov_base == NULL) {
b61c2763 3057 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3058 front_len);
3059 goto out2;
3060 }
3061 } else {
3062 m->front.iov_base = NULL;
3063 }
3064 m->front.iov_len = front_len;
3065
bb257664 3066 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3067 return m;
3068
3069out2:
3070 ceph_msg_put(m);
3071out:
b61c2763
SW		 3072 if (!can_fail) {
3073 pr_err("msg_new can't create type %d front %d\n", type,
3074 front_len);
f0ed1b7c 3075 WARN_ON(1);
b61c2763
SW		 3076 } else {
3077 dout("msg_new can't create type %d front %d\n", type,
3078 front_len);
3079 }
a79832f2 3080 return NULL;
31b8006e 3081}
3d14c5d2 3082EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3083
31b8006e
SW		 3084/*
3085 * Allocate "middle" portion of a message, if it is needed and wasn't
3086 * allocated by alloc_msg. This allows us to read a small fixed-size
3087 * per-type header in the front and then gracefully fail (i.e.,
3088 * propagate the error to the caller based on info in the front) when
3089 * the middle is too large.
3090 */
2450418c 3091static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3092{
3093 int type = le16_to_cpu(msg->hdr.type);
3094 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3095
3096 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3097 ceph_msg_type_name(type), middle_len);
3098 BUG_ON(!middle_len);
3099 BUG_ON(msg->middle);
3100
b6c1d5b8 3101 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3102 if (!msg->middle)
3103 return -ENOMEM;
3104 return 0;
3105}
3106
2450418c 3107/*
1c20f2d2
AE		 3108 * Allocate a message for receiving an incoming message on a
3109 * connection, and save the result in con->in_msg. Uses the
3110 * connection's private alloc_msg op if available.
3111 *
4740a623
SW		 3112 * Returns 0 on success, or a negative error code.
3113 *
3114 * On success, if we set *skip = 1:
3115 * - the next message should be skipped and ignored.
3116 * - con->in_msg == NULL
3117 * or if we set *skip = 0:
3118 * - con->in_msg is non-null.
3119 * On error (ENOMEM, EAGAIN, ...),
3120 * - con->in_msg == NULL
2450418c 3121 */
4740a623 3122static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3123{
4740a623 3124 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3125 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3126 struct ceph_msg *msg;
4740a623 3127 int ret = 0;
2450418c 3128
1c20f2d2 3129 BUG_ON(con->in_msg != NULL);
53ded495 3130 BUG_ON(!con->ops->alloc_msg);
2450418c 3131
53ded495
AE		 3132 mutex_unlock(&con->mutex);
3133 msg = con->ops->alloc_msg(con, hdr, skip);
3134 mutex_lock(&con->mutex);
3135 if (con->state != CON_STATE_OPEN) {
3136 if (msg)
1d866d1c 3137 ceph_msg_put(msg);
53ded495
AE		 3138 return -EAGAIN;
3139 }
4137577a
AE		 3140 if (msg) {
3141 BUG_ON(*skip);
3142 con->in_msg = msg;
36eb71aa 3143 con->in_msg->con = con->ops->get(con);
92ce034b 3144 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3145 } else {
3146 /*
3147 * Null message pointer means either we should skip
3148 * this message or we couldn't allocate memory. The
3149 * former is not an error.
3150 */
3151 if (*skip)
3152 return 0;
3153 con->error_msg = "error allocating memory for incoming message";
3154
53ded495 3155 return -ENOMEM;
2450418c 3156 }
1c20f2d2 3157 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3158
1c20f2d2
AE		 3159 if (middle_len && !con->in_msg->middle) {
3160 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3161 if (ret < 0) {
1c20f2d2
AE		 3162 ceph_msg_put(con->in_msg);
3163 con->in_msg = NULL;
2450418c
YS		 3164 }
3165 }
9d7f0f13 3166
4740a623 3167 return ret;
2450418c
YS		 3168}
3169
31b8006e
SW		 3170
3171/*
3172 * Free a generically kmalloc'd message.
3173 */
3174void ceph_msg_kfree(struct ceph_msg *m)
3175{
3176 dout("msg_kfree %p\n", m);
3177 if (m->front_is_vmalloc)
3178 vfree(m->front.iov_base);
3179 else
3180 kfree(m->front.iov_base);
3181 kfree(m);
3182}
3183
3184/*
3185 * Drop a msg ref. Destroy as needed.
3186 */
c2e552e7
SW		 3187void ceph_msg_last_put(struct kref *kref)
3188{
3189 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
31b8006e 3190
c2e552e7
SW		 3191 dout("ceph_msg_put last one on %p\n", m);
3192 WARN_ON(!list_empty(&m->list_head));
3193
3194 /* drop middle, data, if any */
3195 if (m->middle) {
3196 ceph_buffer_put(m->middle);
3197 m->middle = NULL;
31b8006e 3198 }
4c59b4a2
AE		 3199 if (ceph_msg_has_data(m)) {
3200 if (m->data.type == CEPH_MSG_DATA_PAGELIST) {
3201 ceph_pagelist_release(m->data.pagelist);
3202 kfree(m->data.pagelist);
3203 }
3204 memset(&m->data, 0, sizeof m->data);
3205 ceph_msg_data_init(&m->data);
58bb3b37
SW		 3206 }
3207
c2e552e7
SW		 3208 if (m->pool)
3209 ceph_msgpool_put(m->pool, m);
3210 else
3211 ceph_msg_kfree(m);
31b8006e 3212}
3d14c5d2 3213EXPORT_SYMBOL(ceph_msg_last_put);
9ec7cab1
SW		 3214
3215void ceph_msg_dump(struct ceph_msg *msg)
3216{
4a73ef27 3217 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
4c59b4a2 3218 msg->front_max, msg->data.length);
9ec7cab1
SW		 3219 print_hex_dump(KERN_DEBUG, "header: ",
3220 DUMP_PREFIX_OFFSET, 16, 1,
3221 &msg->hdr, sizeof(msg->hdr), true);
3222 print_hex_dump(KERN_DEBUG, " front: ",
3223 DUMP_PREFIX_OFFSET, 16, 1,
3224 msg->front.iov_base, msg->front.iov_len, true);
3225 if (msg->middle)
3226 print_hex_dump(KERN_DEBUG, "middle: ",
3227 DUMP_PREFIX_OFFSET, 16, 1,
3228 msg->middle->vec.iov_base,
3229 msg->middle->vec.iov_len, true);
3230 print_hex_dump(KERN_DEBUG, "footer: ",
3231 DUMP_PREFIX_OFFSET, 16, 1,
3232 &msg->footer, sizeof(msg->footer), true);
3233}
3d14c5d2 3234EXPORT_SYMBOL(ceph_msg_dump);
Linux kernel - Deliverables
This page took 0.554794 seconds and 5 git commands to generate.