projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: implement bio message data item cursor
[deliverable/linux.git] / net / ceph / messenger.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
5a0e3ad6 9#include <linux/slab.h>
31b8006e
SW		 10#include <linux/socket.h>
11#include <linux/string.h>
3ebc21f7 12#ifdef CONFIG_BLOCK
68b4476b 13#include <linux/bio.h>
3ebc21f7 14#endif /* CONFIG_BLOCK */
ee3b56f2 15#include <linux/dns_resolver.h>
31b8006e
SW		 16#include <net/tcp.h>
17
3d14c5d2
YS		 18#include <linux/ceph/libceph.h>
19#include <linux/ceph/messenger.h>
20#include <linux/ceph/decode.h>
21#include <linux/ceph/pagelist.h>
bc3b2d7f 22#include <linux/export.h>
31b8006e 23
fe38a2b6
AE		 24#define list_entry_next(pos, member) \
25 list_entry(pos->member.next, typeof(*pos), member)
26
31b8006e
SW		 27/*
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
33 * the sender.
34 */
35
bc18f4b1
AE		 36/*
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
40 * unexpected state.
41 *
42 * --------
43 * | NEW* | transient initial state
44 * --------
45 * | con_sock_state_init()
46 * v
47 * ----------
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
50 * ^ \
51 * | \ con_sock_state_connecting()
52 * | ----------------------
53 * | \
54 * + con_sock_state_closed() \
fbb85a47
SW		 55 * |+--------------------------- \
56 * | \ \ \
57 * | ----------- \ \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
60 * | ^ \ |
61 * | | \ |
62 * | + con_sock_state_closing() \ |
63 * | / \ | |
64 * | / --------------- | |
65 * | / \ v v
bc18f4b1
AE		 66 * | / --------------
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
70 * | | v
71 * -------------
72 * | CONNECTED | TCP connection established
73 * -------------
74 *
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
76 */
ce2c8903
AE		 77
78#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
83
8dacc7da
SW		 84/*
85 * connection states
86 */
87#define CON_STATE_CLOSED 1 /* -> PREOPEN */
88#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
93
4a861692
SW		 94/*
95 * ceph_connection flag bits
96 */
97#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 103
c9ffc77a
AE		 104static bool con_flag_valid(unsigned long con_flag)
105{
106 switch (con_flag) {
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
112 return true;
113 default:
114 return false;
115 }
116}
117
118static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
119{
120 BUG_ON(!con_flag_valid(con_flag));
121
122 clear_bit(con_flag, &con->flags);
123}
124
125static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
126{
127 BUG_ON(!con_flag_valid(con_flag));
128
129 set_bit(con_flag, &con->flags);
130}
131
132static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
133{
134 BUG_ON(!con_flag_valid(con_flag));
135
136 return test_bit(con_flag, &con->flags);
137}
138
139static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
141{
142 BUG_ON(!con_flag_valid(con_flag));
143
144 return test_and_clear_bit(con_flag, &con->flags);
145}
146
147static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
149{
150 BUG_ON(!con_flag_valid(con_flag));
151
152 return test_and_set_bit(con_flag, &con->flags);
153}
154
31b8006e
SW		 155/* static tag bytes (protocol control messages) */
156static char tag_msg = CEPH_MSGR_TAG_MSG;
157static char tag_ack = CEPH_MSGR_TAG_ACK;
158static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
159
a6a5349d
SW		 160#ifdef CONFIG_LOCKDEP
161static struct lock_class_key socket_class;
162#endif
163
84495f49
AE		 164/*
165 * When skipping (ignoring) a block of input we read it into a "skip
166 * buffer," which is this many bytes in size.
167 */
168#define SKIP_BUF_SIZE 1024
31b8006e
SW		 169
170static void queue_con(struct ceph_connection *con);
171static void con_work(struct work_struct *);
93209264 172static void con_fault(struct ceph_connection *con);
31b8006e 173
31b8006e 174/*
f64a9317
AE		 175 * Nicely render a sockaddr as a string. An array of formatted
176 * strings is used, to approximate reentrancy.
31b8006e 177 */
f64a9317
AE		 178#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
179#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
180#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
181#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
182
183static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
184static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 185
57666519 186static struct page *zero_page; /* used in certain error cases */
57666519 187
3d14c5d2 188const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 189{
190 int i;
191 char *s;
99f0f3b2
AE		 192 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
193 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 194
f64a9317 195 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 196 s = addr_str[i];
197
198 switch (ss->ss_family) {
199 case AF_INET:
bd406145
AE		 200 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
201 ntohs(in4->sin_port));
31b8006e
SW		 202 break;
203
204 case AF_INET6:
bd406145
AE		 205 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
206 ntohs(in6->sin6_port));
31b8006e
SW		 207 break;
208
209 default:
d3002b97
AE		 210 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
211 ss->ss_family);
31b8006e
SW		 212 }
213
214 return s;
215}
3d14c5d2 216EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 217
63f2d211
SW		 218static void encode_my_addr(struct ceph_messenger *msgr)
219{
220 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
221 ceph_encode_addr(&msgr->my_enc_addr);
222}
223
31b8006e
SW		 224/*
225 * work queue for all reading and writing to/from the socket.
226 */
e0f43c94 227static struct workqueue_struct *ceph_msgr_wq;
31b8006e 228
15417167 229static void _ceph_msgr_exit(void)
6173d1f0 230{
d3002b97 231 if (ceph_msgr_wq) {
6173d1f0 232 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 233 ceph_msgr_wq = NULL;
234 }
6173d1f0 235
6173d1f0
AE		 236 BUG_ON(zero_page == NULL);
237 kunmap(zero_page);
238 page_cache_release(zero_page);
239 zero_page = NULL;
240}
241
3d14c5d2 242int ceph_msgr_init(void)
31b8006e 243{
57666519
AE		 244 BUG_ON(zero_page != NULL);
245 zero_page = ZERO_PAGE(0);
246 page_cache_get(zero_page);
247
f363e45f 248 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
6173d1f0
AE		 249 if (ceph_msgr_wq)
250 return 0;
57666519 251
6173d1f0
AE		 252 pr_err("msgr_init failed to create workqueue\n");
253 _ceph_msgr_exit();
57666519 254
6173d1f0 255 return -ENOMEM;
31b8006e 256}
3d14c5d2 257EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 258
259void ceph_msgr_exit(void)
260{
57666519 261 BUG_ON(ceph_msgr_wq == NULL);
57666519 262
6173d1f0 263 _ceph_msgr_exit();
31b8006e 264}
3d14c5d2 265EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 266
cd84db6e 267void ceph_msgr_flush(void)
a922d38f
SW		 268{
269 flush_workqueue(ceph_msgr_wq);
270}
3d14c5d2 271EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 272
ce2c8903
AE		 273/* Connection socket state transition functions */
274
275static void con_sock_state_init(struct ceph_connection *con)
276{
277 int old_state;
278
279 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
280 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
281 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 282 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
283 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 284}
285
286static void con_sock_state_connecting(struct ceph_connection *con)
287{
288 int old_state;
289
290 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
291 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
292 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 293 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
294 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 295}
296
297static void con_sock_state_connected(struct ceph_connection *con)
298{
299 int old_state;
300
301 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
302 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
303 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 304 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
305 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 306}
307
308static void con_sock_state_closing(struct ceph_connection *con)
309{
310 int old_state;
311
312 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
313 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
314 old_state != CON_SOCK_STATE_CONNECTED &&
315 old_state != CON_SOCK_STATE_CLOSING))
316 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 317 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
318 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 319}
320
321static void con_sock_state_closed(struct ceph_connection *con)
322{
323 int old_state;
324
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 327 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 328 old_state != CON_SOCK_STATE_CONNECTING &&
329 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 330 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 331 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
332 CON_SOCK_STATE_CLOSED);
ce2c8903 333}
a922d38f 334
31b8006e
SW		 335/*
336 * socket callback functions
337 */
338
339/* data available on socket, or listen socket received a connect */
327800bd 340static void ceph_sock_data_ready(struct sock *sk, int count_unused)
31b8006e 341{
bd406145 342 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 343 if (atomic_read(&con->msgr->stopping)) {
344 return;
345 }
bd406145 346
31b8006e 347 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 348 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 349 con, con->state);
350 queue_con(con);
351 }
352}
353
354/* socket has buffer space for writing */
327800bd 355static void ceph_sock_write_space(struct sock *sk)
31b8006e 356{
d3002b97 357 struct ceph_connection *con = sk->sk_user_data;
31b8006e 358
182fac26
JS		 359 /* only queue to workqueue if there is data we want to write,
360 * and there is sufficient space in the socket buffer to accept
327800bd 361 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 362 * doesn't get called again until try_write() fills the socket
363 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
364 * and net/core/stream.c:sk_stream_write_space().
365 */
c9ffc77a 366 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
182fac26 367 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
327800bd 368 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 369 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
370 queue_con(con);
371 }
31b8006e 372 } else {
327800bd 373 dout("%s %p nothing to write\n", __func__, con);
31b8006e 374 }
31b8006e
SW		 375}
376
377/* socket's state has changed */
327800bd 378static void ceph_sock_state_change(struct sock *sk)
31b8006e 379{
bd406145 380 struct ceph_connection *con = sk->sk_user_data;
31b8006e 381
327800bd 382 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 383 con, con->state, sk->sk_state);
384
31b8006e
SW		 385 switch (sk->sk_state) {
386 case TCP_CLOSE:
327800bd 387 dout("%s TCP_CLOSE\n", __func__);
31b8006e 388 case TCP_CLOSE_WAIT:
327800bd 389 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 390 con_sock_state_closing(con);
c9ffc77a 391 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 392 queue_con(con);
31b8006e
SW		 393 break;
394 case TCP_ESTABLISHED:
327800bd 395 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 396 con_sock_state_connected(con);
31b8006e
SW		 397 queue_con(con);
398 break;
d3002b97
AE		 399 default: /* Everything else is uninteresting */
400 break;
31b8006e
SW		 401 }
402}
403
404/*
405 * set up socket callbacks
406 */
407static void set_sock_callbacks(struct socket *sock,
408 struct ceph_connection *con)
409{
410 struct sock *sk = sock->sk;
bd406145 411 sk->sk_user_data = con;
327800bd
AE		 412 sk->sk_data_ready = ceph_sock_data_ready;
413 sk->sk_write_space = ceph_sock_write_space;
414 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 415}
416
417
418/*
419 * socket helpers
420 */
421
422/*
423 * initiate connection to a remote socket.
424 */
41617d0c 425static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 426{
f91d3471 427 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e
SW		 428 struct socket *sock;
429 int ret;
430
431 BUG_ON(con->sock);
f91d3471
SW		 432 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
433 IPPROTO_TCP, &sock);
31b8006e 434 if (ret)
41617d0c 435 return ret;
31b8006e
SW		 436 sock->sk->sk_allocation = GFP_NOFS;
437
a6a5349d
SW		 438#ifdef CONFIG_LOCKDEP
439 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
440#endif
441
31b8006e
SW		 442 set_sock_callbacks(sock, con);
443
3d14c5d2 444 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 445
89a86be0 446 con_sock_state_connecting(con);
f91d3471
SW		 447 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
448 O_NONBLOCK);
31b8006e
SW		 449 if (ret == -EINPROGRESS) {
450 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 451 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 452 sock->sk->sk_state);
a5bc3129 453 } else if (ret < 0) {
31b8006e 454 pr_err("connect %s error %d\n",
3d14c5d2 455 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 456 sock_release(sock);
31b8006e 457 con->error_msg = "connect error";
31b8006e 458
41617d0c 459 return ret;
a5bc3129
AE		 460 }
461 con->sock = sock;
41617d0c 462 return 0;
31b8006e
SW		 463}
464
465static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
466{
467 struct kvec iov = {buf, len};
468 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 469 int r;
31b8006e 470
98bdb0aa
SW		 471 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
472 if (r == -EAGAIN)
473 r = 0;
474 return r;
31b8006e
SW		 475}
476
afb3d90e
AE		 477static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
478 int page_offset, size_t length)
479{
480 void *kaddr;
481 int ret;
482
483 BUG_ON(page_offset + length > PAGE_SIZE);
484
485 kaddr = kmap(page);
486 BUG_ON(!kaddr);
487 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
488 kunmap(page);
489
490 return ret;
491}
492
31b8006e
SW		 493/*
494 * write something. @more is true if caller will be sending more data
495 * shortly.
496 */
497static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
498 size_t kvlen, size_t len, int more)
499{
500 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 501 int r;
31b8006e
SW		 502
503 if (more)
504 msg.msg_flags |= MSG_MORE;
505 else
506 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
507
42961d23
SW		 508 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
509 if (r == -EAGAIN)
510 r = 0;
511 return r;
31b8006e
SW		 512}
513
31739139 514static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 515 int offset, size_t size, bool more)
31739139
AE		 516{
517 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
518 int ret;
519
520 ret = kernel_sendpage(sock, page, offset, size, flags);
521 if (ret == -EAGAIN)
522 ret = 0;
523
524 return ret;
525}
526
31b8006e
SW		 527
528/*
529 * Shutdown/close the socket for the given connection.
530 */
531static int con_close_socket(struct ceph_connection *con)
532{
8007b8d6 533 int rc = 0;
31b8006e
SW		 534
535 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 536 if (con->sock) {
537 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
538 sock_release(con->sock);
539 con->sock = NULL;
540 }
456ea468
AE		 541
542 /*
4a861692 543 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 544 * independent of the connection mutex, and we could have
545 * received a socket close event before we had the chance to
546 * shut the socket down.
547 */
c9ffc77a 548 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 549
ce2c8903 550 con_sock_state_closed(con);
31b8006e
SW		 551 return rc;
552}
553
554/*
555 * Reset a connection. Discard all incoming and outgoing messages
556 * and clear *_seq state.
557 */
558static void ceph_msg_remove(struct ceph_msg *msg)
559{
560 list_del_init(&msg->list_head);
38941f80 561 BUG_ON(msg->con == NULL);
36eb71aa 562 msg->con->ops->put(msg->con);
38941f80
AE		 563 msg->con = NULL;
564
31b8006e
SW		 565 ceph_msg_put(msg);
566}
567static void ceph_msg_remove_list(struct list_head *head)
568{
569 while (!list_empty(head)) {
570 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
571 list_head);
572 ceph_msg_remove(msg);
573 }
574}
575
576static void reset_connection(struct ceph_connection *con)
577{
578 /* reset connection, out_queue, msg_ and connect_seq */
579 /* discard existing out_queue and msg_seq */
0fa6ebc6 580 dout("reset_connection %p\n", con);
31b8006e
SW		 581 ceph_msg_remove_list(&con->out_queue);
582 ceph_msg_remove_list(&con->out_sent);
583
cf3e5c40 584 if (con->in_msg) {
38941f80
AE		 585 BUG_ON(con->in_msg->con != con);
586 con->in_msg->con = NULL;
cf3e5c40
SW		 587 ceph_msg_put(con->in_msg);
588 con->in_msg = NULL;
36eb71aa 589 con->ops->put(con);
cf3e5c40
SW		 590 }
591
31b8006e
SW		 592 con->connect_seq = 0;
593 con->out_seq = 0;
c86a2930
SW		 594 if (con->out_msg) {
595 ceph_msg_put(con->out_msg);
596 con->out_msg = NULL;
597 }
31b8006e 598 con->in_seq = 0;
0e0d5e0c 599 con->in_seq_acked = 0;
31b8006e
SW		 600}
601
602/*
603 * mark a peer down. drop any open connections.
604 */
605void ceph_con_close(struct ceph_connection *con)
606{
8c50c817 607 mutex_lock(&con->mutex);
3d14c5d2
YS		 608 dout("con_close %p peer %s\n", con,
609 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 610 con->state = CON_STATE_CLOSED;
a5988c49 611
c9ffc77a
AE		 612 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
613 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
614 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
615 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 616
31b8006e 617 reset_connection(con);
6f2bc3ff 618 con->peer_global_seq = 0;
91e45ce3 619 cancel_delayed_work(&con->work);
ee76e073 620 con_close_socket(con);
ec302645 621 mutex_unlock(&con->mutex);
31b8006e 622}
3d14c5d2 623EXPORT_SYMBOL(ceph_con_close);
31b8006e 624
31b8006e
SW		 625/*
626 * Reopen a closed connection, with a new peer address.
627 */
b7a9e5dd
SW		 628void ceph_con_open(struct ceph_connection *con,
629 __u8 entity_type, __u64 entity_num,
630 struct ceph_entity_addr *addr)
31b8006e 631{
5469155f 632 mutex_lock(&con->mutex);
3d14c5d2 633 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 634
122070a2 635 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 636 con->state = CON_STATE_PREOPEN;
a5988c49 637
b7a9e5dd
SW		 638 con->peer_name.type = (__u8) entity_type;
639 con->peer_name.num = cpu_to_le64(entity_num);
640
31b8006e 641 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 642 con->delay = 0; /* reset backoff memory */
5469155f 643 mutex_unlock(&con->mutex);
31b8006e
SW		 644 queue_con(con);
645}
3d14c5d2 646EXPORT_SYMBOL(ceph_con_open);
31b8006e 647
87b315a5
SW		 648/*
649 * return true if this connection ever successfully opened
650 */
651bool ceph_con_opened(struct ceph_connection *con)
652{
653 return con->connect_seq > 0;
654}
655
31b8006e
SW		 656/*
657 * initialize a new connection.
658 */
1bfd89f4
AE		 659void ceph_con_init(struct ceph_connection *con, void *private,
660 const struct ceph_connection_operations *ops,
b7a9e5dd 661 struct ceph_messenger *msgr)
31b8006e
SW		 662{
663 dout("con_init %p\n", con);
664 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 665 con->private = private;
666 con->ops = ops;
31b8006e 667 con->msgr = msgr;
ce2c8903
AE		 668
669 con_sock_state_init(con);
670
ec302645 671 mutex_init(&con->mutex);
31b8006e
SW		 672 INIT_LIST_HEAD(&con->out_queue);
673 INIT_LIST_HEAD(&con->out_sent);
674 INIT_DELAYED_WORK(&con->work, con_work);
a5988c49 675
8dacc7da 676 con->state = CON_STATE_CLOSED;
31b8006e 677}
3d14c5d2 678EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 679
680
681/*
682 * We maintain a global counter to order connection attempts. Get
683 * a unique seq greater than @gt.
684 */
685static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
686{
687 u32 ret;
688
689 spin_lock(&msgr->global_seq_lock);
690 if (msgr->global_seq < gt)
691 msgr->global_seq = gt;
692 ret = ++msgr->global_seq;
693 spin_unlock(&msgr->global_seq_lock);
694 return ret;
695}
696
e2200423 697static void con_out_kvec_reset(struct ceph_connection *con)
859eb799
AE		 698{
699 con->out_kvec_left = 0;
700 con->out_kvec_bytes = 0;
701 con->out_kvec_cur = &con->out_kvec[0];
702}
703
e2200423 704static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 705 size_t size, void *data)
706{
707 int index;
708
709 index = con->out_kvec_left;
710 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
711
712 con->out_kvec[index].iov_len = size;
713 con->out_kvec[index].iov_base = data;
714 con->out_kvec_left++;
715 con->out_kvec_bytes += size;
716}
31b8006e 717
df6ad1f9 718#ifdef CONFIG_BLOCK
07c09b72
AE		 719static void init_bio_iter(struct bio *bio, struct bio **bio_iter,
720 unsigned int *bio_seg)
df6ad1f9
AE		 721{
722 if (!bio) {
07c09b72
AE		 723 *bio_iter = NULL;
724 *bio_seg = 0;
df6ad1f9
AE		 725 return;
726 }
07c09b72
AE		 727 *bio_iter = bio;
728 *bio_seg = (unsigned int) bio->bi_idx;
df6ad1f9
AE		 729}
730
07c09b72 731static void iter_bio_next(struct bio **bio_iter, unsigned int *seg)
df6ad1f9
AE		 732{
733 if (*bio_iter == NULL)
734 return;
735
736 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
737
738 (*seg)++;
739 if (*seg == (*bio_iter)->bi_vcnt)
740 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
741}
6aaa4511
AE		 742
743/*
744 * For a bio data item, a piece is whatever remains of the next
745 * entry in the current bio iovec, or the first entry in the next
746 * bio in the list.
747 */
748static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data *data)
749{
750 struct ceph_msg_data_cursor *cursor = &data->cursor;
751 struct bio *bio;
752
753 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
754
755 bio = data->bio;
756 BUG_ON(!bio);
757 BUG_ON(!bio->bi_vcnt);
758 /* resid = bio->bi_size */
759
760 cursor->bio = bio;
761 cursor->vector_index = 0;
762 cursor->vector_offset = 0;
763 cursor->last_piece = !bio->bi_next && bio->bi_vcnt == 1;
764}
765
766static struct page *ceph_msg_data_bio_next(struct ceph_msg_data *data,
767 size_t *page_offset,
768 size_t *length)
769{
770 struct ceph_msg_data_cursor *cursor = &data->cursor;
771 struct bio *bio;
772 struct bio_vec *bio_vec;
773 unsigned int index;
774
775 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
776
777 bio = cursor->bio;
778 BUG_ON(!bio);
779
780 index = cursor->vector_index;
781 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
782
783 bio_vec = &bio->bi_io_vec[index];
784 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
785 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
786 BUG_ON(*page_offset >= PAGE_SIZE);
787 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
788 BUG_ON(*length > PAGE_SIZE);
789
790 return bio_vec->bv_page;
791}
792
793static bool ceph_msg_data_bio_advance(struct ceph_msg_data *data, size_t bytes)
794{
795 struct ceph_msg_data_cursor *cursor = &data->cursor;
796 struct bio *bio;
797 struct bio_vec *bio_vec;
798 unsigned int index;
799
800 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
801
802 bio = cursor->bio;
803 BUG_ON(!bio);
804
805 index = cursor->vector_index;
806 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
807 bio_vec = &bio->bi_io_vec[index];
808 BUG_ON(cursor->vector_offset + bytes > bio_vec->bv_len);
809
810 /* Advance the cursor offset */
811
812 cursor->vector_offset += bytes;
813 if (cursor->vector_offset < bio_vec->bv_len)
814 return false; /* more bytes to process in this segment */
815
816 /* Move on to the next segment, and possibly the next bio */
817
818 if (++cursor->vector_index == (unsigned int) bio->bi_vcnt) {
819 bio = bio->bi_next;
820 cursor->bio = bio;
821 cursor->vector_index = 0;
822 }
823 cursor->vector_offset = 0;
824
825 if (!cursor->last_piece && bio && !bio->bi_next)
826 if (cursor->vector_index == (unsigned int) bio->bi_vcnt - 1)
827 cursor->last_piece = true;
828
829 return true;
830}
df6ad1f9
AE		 831#endif
832
fe38a2b6 833/*
dd236fcb
AE		 834 * For a pagelist, a piece is whatever remains to be consumed in the
835 * first page in the list, or the front of the next page.
fe38a2b6 836 */
dd236fcb 837static void ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data *data)
fe38a2b6
AE		 838{
839 struct ceph_msg_data_cursor *cursor = &data->cursor;
840 struct ceph_pagelist *pagelist;
841 struct page *page;
842
dd236fcb 843 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 844
845 pagelist = data->pagelist;
846 BUG_ON(!pagelist);
847 if (!pagelist->length)
848 return; /* pagelist can be assigned but empty */
849
850 BUG_ON(list_empty(&pagelist->head));
851 page = list_first_entry(&pagelist->head, struct page, lru);
852
853 cursor->page = page;
854 cursor->offset = 0;
855 cursor->last_piece = pagelist->length <= PAGE_SIZE;
856}
857
dd236fcb 858static struct page *ceph_msg_data_pagelist_next(struct ceph_msg_data *data,
fe38a2b6 859 size_t *page_offset,
dd236fcb 860 size_t *length)
fe38a2b6
AE		 861{
862 struct ceph_msg_data_cursor *cursor = &data->cursor;
863 struct ceph_pagelist *pagelist;
864 size_t piece_end;
865
866 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
867
868 pagelist = data->pagelist;
869 BUG_ON(!pagelist);
870
871 BUG_ON(!cursor->page);
872 BUG_ON(cursor->offset >= pagelist->length);
873
dd236fcb 874 if (cursor->last_piece) {
fe38a2b6
AE		 875 /* pagelist offset is always 0 */
876 piece_end = pagelist->length & ~PAGE_MASK;
877 if (!piece_end)
878 piece_end = PAGE_SIZE;
879 } else {
880 piece_end = PAGE_SIZE;
881 }
882 *page_offset = cursor->offset & ~PAGE_MASK;
883 *length = piece_end - *page_offset;
884
885 return data->cursor.page;
886}
887
dd236fcb
AE		 888static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data *data,
889 size_t bytes)
fe38a2b6
AE		 890{
891 struct ceph_msg_data_cursor *cursor = &data->cursor;
892 struct ceph_pagelist *pagelist;
893
894 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
895
896 pagelist = data->pagelist;
897 BUG_ON(!pagelist);
898 BUG_ON(!cursor->page);
899 BUG_ON(cursor->offset + bytes > pagelist->length);
900 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
901
902 /* Advance the cursor offset */
903
904 cursor->offset += bytes;
905 /* pagelist offset is always 0 */
906 if (!bytes || cursor->offset & ~PAGE_MASK)
907 return false; /* more bytes to process in the current page */
908
909 /* Move on to the next page */
910
911 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
912 cursor->page = list_entry_next(cursor->page, lru);
913
914 /* cursor offset is at page boundary; pagelist offset is always 0 */
915 if (pagelist->length - cursor->offset <= PAGE_SIZE)
916 cursor->last_piece = true;
917
918 return true;
919}
920
dd236fcb
AE		 921/*
922 * Message data is handled (sent or received) in pieces, where each
923 * piece resides on a single page. The network layer might not
924 * consume an entire piece at once. A data item's cursor keeps
925 * track of which piece is next to process and how much remains to
926 * be processed in that piece. It also tracks whether the current
927 * piece is the last one in the data item.
928 */
929static void ceph_msg_data_cursor_init(struct ceph_msg_data *data)
930{
931 switch (data->type) {
932 case CEPH_MSG_DATA_PAGELIST:
933 ceph_msg_data_pagelist_cursor_init(data);
934 break;
dd236fcb
AE		 935#ifdef CONFIG_BLOCK
936 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 937 ceph_msg_data_bio_cursor_init(data);
938 break;
dd236fcb 939#endif /* CONFIG_BLOCK */
6aaa4511
AE		 940 case CEPH_MSG_DATA_NONE:
941 case CEPH_MSG_DATA_PAGES:
dd236fcb
AE		 942 default:
943 /* BUG(); */
944 break;
945 }
946}
947
948/*
949 * Return the page containing the next piece to process for a given
950 * data item, and supply the page offset and length of that piece.
951 * Indicate whether this is the last piece in this data item.
952 */
953static struct page *ceph_msg_data_next(struct ceph_msg_data *data,
954 size_t *page_offset,
955 size_t *length,
956 bool *last_piece)
957{
958 struct page *page;
959
960 switch (data->type) {
961 case CEPH_MSG_DATA_PAGELIST:
962 page = ceph_msg_data_pagelist_next(data, page_offset, length);
963 break;
dd236fcb
AE		 964#ifdef CONFIG_BLOCK
965 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 966 page = ceph_msg_data_bio_next(data, page_offset, length);
967 break;
dd236fcb 968#endif /* CONFIG_BLOCK */
6aaa4511
AE		 969 case CEPH_MSG_DATA_NONE:
970 case CEPH_MSG_DATA_PAGES:
dd236fcb
AE		 971 default:
972 page = NULL;
973 break;
974 }
975 BUG_ON(!page);
976 BUG_ON(*page_offset + *length > PAGE_SIZE);
977 BUG_ON(!*length);
978 if (last_piece)
979 *last_piece = data->cursor.last_piece;
980
981 return page;
982}
983
984/*
985 * Returns true if the result moves the cursor on to the next piece
986 * of the data item.
987 */
988static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes)
989{
990 bool new_piece;
991
992 switch (data->type) {
993 case CEPH_MSG_DATA_PAGELIST:
994 new_piece = ceph_msg_data_pagelist_advance(data, bytes);
995 break;
dd236fcb
AE		 996#ifdef CONFIG_BLOCK
997 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 998 new_piece = ceph_msg_data_bio_advance(data, bytes);
999 break;
dd236fcb 1000#endif /* CONFIG_BLOCK */
6aaa4511
AE		 1001 case CEPH_MSG_DATA_NONE:
1002 case CEPH_MSG_DATA_PAGES:
dd236fcb
AE		 1003 default:
1004 BUG();
1005 break;
1006 }
1007
1008 return new_piece;
1009}
1010
78625051
AE		 1011static void prepare_message_data(struct ceph_msg *msg,
1012 struct ceph_msg_pos *msg_pos)
739c905b 1013{
739c905b
AE		 1014 BUG_ON(!msg);
1015 BUG_ON(!msg->hdr.data_len);
1016
1017 /* initialize page iterator */
bae6acd9 1018 msg_pos->page = 0;
97fb1c7f 1019 if (ceph_msg_has_pages(msg))
f9e15777 1020 msg_pos->page_pos = msg->p.alignment;
739c905b 1021 else
bae6acd9 1022 msg_pos->page_pos = 0;
572c588e 1023#ifdef CONFIG_BLOCK
97fb1c7f 1024 if (ceph_msg_has_bio(msg))
f9e15777 1025 init_bio_iter(msg->b.bio, &msg->b.bio_iter, &msg->b.bio_seg);
572c588e 1026#endif
bae6acd9 1027 msg_pos->data_pos = 0;
fe38a2b6 1028
7fe1e5e5 1029 /* Initialize data cursors */
fe38a2b6 1030
6aaa4511
AE		 1031#ifdef CONFIG_BLOCK
1032 if (ceph_msg_has_bio(msg))
1033 ceph_msg_data_cursor_init(&msg->b);
1034#endif /* CONFIG_BLOCK */
7fe1e5e5
AE		 1035 if (ceph_msg_has_pagelist(msg))
1036 ceph_msg_data_cursor_init(&msg->l);
fe38a2b6
AE		 1037 if (ceph_msg_has_trail(msg))
1038 ceph_msg_data_cursor_init(&msg->t);
1039
bae6acd9 1040 msg_pos->did_page_crc = false;
739c905b
AE		 1041}
1042
31b8006e
SW		 1043/*
1044 * Prepare footer for currently outgoing message, and finish things
1045 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1046 */
859eb799 1047static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1048{
1049 struct ceph_msg *m = con->out_msg;
859eb799 1050 int v = con->out_kvec_left;
31b8006e 1051
fd154f3c
AE		 1052 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1053
31b8006e
SW		 1054 dout("prepare_write_message_footer %p\n", con);
1055 con->out_kvec_is_msg = true;
1056 con->out_kvec[v].iov_base = &m->footer;
1057 con->out_kvec[v].iov_len = sizeof(m->footer);
1058 con->out_kvec_bytes += sizeof(m->footer);
1059 con->out_kvec_left++;
1060 con->out_more = m->more_to_follow;
c86a2930 1061 con->out_msg_done = true;
31b8006e
SW		 1062}
1063
1064/*
1065 * Prepare headers for the next outgoing message.
1066 */
1067static void prepare_write_message(struct ceph_connection *con)
1068{
1069 struct ceph_msg *m;
a9a0c51a 1070 u32 crc;
31b8006e 1071
e2200423 1072 con_out_kvec_reset(con);
31b8006e 1073 con->out_kvec_is_msg = true;
c86a2930 1074 con->out_msg_done = false;
31b8006e
SW		 1075
1076 /* Sneak an ack in there first? If we can get it into the same
1077 * TCP packet that's a good thing. */
1078 if (con->in_seq > con->in_seq_acked) {
1079 con->in_seq_acked = con->in_seq;
e2200423 1080 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1081 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1082 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1083 &con->out_temp_ack);
31b8006e
SW		 1084 }
1085
38941f80 1086 BUG_ON(list_empty(&con->out_queue));
859eb799 1087 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1088 con->out_msg = m;
38941f80 1089 BUG_ON(m->con != con);
4cf9d544
SW		 1090
1091 /* put message on sent list */
1092 ceph_msg_get(m);
1093 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1094
e84346b7
SW		 1095 /*
1096 * only assign outgoing seq # if we haven't sent this message
1097 * yet. if it is requeued, resend with it's original seq.
1098 */
1099 if (m->needs_out_seq) {
1100 m->hdr.seq = cpu_to_le64(++con->out_seq);
1101 m->needs_out_seq = false;
1102 }
31b8006e 1103
4a73ef27 1104 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n",
31b8006e
SW		 1105 m, con->out_seq, le16_to_cpu(m->hdr.type),
1106 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
f9e15777 1107 le32_to_cpu(m->hdr.data_len), m->p.length);
31b8006e
SW		 1108 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1109
1110 /* tag + hdr + front + middle */
e2200423
AE		 1111 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1112 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1113 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1114
31b8006e 1115 if (m->middle)
e2200423 1116 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1117 m->middle->vec.iov_base);
31b8006e
SW		 1118
1119 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1120 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1121 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1122 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1123
1124 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1125 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1126 if (m->middle) {
1127 crc = crc32c(0, m->middle->vec.iov_base,
1128 m->middle->vec.iov_len);
1129 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1130 } else
31b8006e 1131 con->out_msg->footer.middle_crc = 0;
739c905b 1132 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1133 le32_to_cpu(con->out_msg->footer.front_crc),
1134 le32_to_cpu(con->out_msg->footer.middle_crc));
1135
1136 /* is there a data payload? */
739c905b 1137 con->out_msg->footer.data_crc = 0;
78625051
AE		 1138 if (m->hdr.data_len) {
1139 prepare_message_data(con->out_msg, &con->out_msg_pos);
1140 con->out_more = 1; /* data + footer will follow */
1141 } else {
31b8006e 1142 /* no, queue up footer too and be done */
859eb799 1143 prepare_write_message_footer(con);
78625051 1144 }
31b8006e 1145
c9ffc77a 1146 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1147}
1148
1149/*
1150 * Prepare an ack.
1151 */
1152static void prepare_write_ack(struct ceph_connection *con)
1153{
1154 dout("prepare_write_ack %p %llu -> %llu\n", con,
1155 con->in_seq_acked, con->in_seq);
1156 con->in_seq_acked = con->in_seq;
1157
e2200423 1158 con_out_kvec_reset(con);
859eb799 1159
e2200423 1160 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1161
31b8006e 1162 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1163 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1164 &con->out_temp_ack);
1165
31b8006e 1166 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1167 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1168}
1169
1170/*
1171 * Prepare to write keepalive byte.
1172 */
1173static void prepare_write_keepalive(struct ceph_connection *con)
1174{
1175 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1176 con_out_kvec_reset(con);
1177 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1178 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1179}
1180
1181/*
1182 * Connection negotiation.
1183 */
1184
dac1e716
AE		 1185static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1186 int *auth_proto)
4e7a5dcd 1187{
a3530df3 1188 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1189
1190 if (!con->ops->get_authorizer) {
1191 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1192 con->out_connect.authorizer_len = 0;
729796be 1193 return NULL;
b1c6b980
AE		 1194 }
1195
1196 /* Can't hold the mutex while getting authorizer */
ec302645 1197 mutex_unlock(&con->mutex);
dac1e716 1198 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1199 mutex_lock(&con->mutex);
4e7a5dcd 1200
a3530df3 1201 if (IS_ERR(auth))
729796be 1202 return auth;
8dacc7da 1203 if (con->state != CON_STATE_NEGOTIATING)
729796be 1204 return ERR_PTR(-EAGAIN);
0da5d703 1205
8f43fb53
AE		 1206 con->auth_reply_buf = auth->authorizer_reply_buf;
1207 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1208 return auth;
4e7a5dcd
SW		 1209}
1210
31b8006e
SW		 1211/*
1212 * We connected to a peer and are saying hello.
1213 */
e825a66d 1214static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1215{
e2200423
AE		 1216 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1217 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1218 &con->msgr->my_enc_addr);
eed0ef2c 1219
eed0ef2c 1220 con->out_more = 0;
c9ffc77a 1221 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1222}
1223
e825a66d 1224static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1225{
95c96174 1226 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1227 int proto;
dac1e716 1228 int auth_proto;
729796be 1229 struct ceph_auth_handshake *auth;
31b8006e
SW		 1230
1231 switch (con->peer_name.type) {
1232 case CEPH_ENTITY_TYPE_MON:
1233 proto = CEPH_MONC_PROTOCOL;
1234 break;
1235 case CEPH_ENTITY_TYPE_OSD:
1236 proto = CEPH_OSDC_PROTOCOL;
1237 break;
1238 case CEPH_ENTITY_TYPE_MDS:
1239 proto = CEPH_MDSC_PROTOCOL;
1240 break;
1241 default:
1242 BUG();
1243 }
1244
1245 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1246 con->connect_seq, global_seq, proto);
4e7a5dcd 1247
e825a66d 1248 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1249 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1250 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1251 con->out_connect.global_seq = cpu_to_le32(global_seq);
1252 con->out_connect.protocol_version = cpu_to_le32(proto);
1253 con->out_connect.flags = 0;
31b8006e 1254
dac1e716
AE		 1255 auth_proto = CEPH_AUTH_UNKNOWN;
1256 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1257 if (IS_ERR(auth))
1258 return PTR_ERR(auth);
3da54776 1259
dac1e716 1260 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1261 con->out_connect.authorizer_len = auth ?
1262 cpu_to_le32(auth->authorizer_buf_len) : 0;
1263
e2200423 1264 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1265 &con->out_connect);
1266 if (auth && auth->authorizer_buf_len)
e2200423 1267 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1268 auth->authorizer_buf);
859eb799 1269
31b8006e 1270 con->out_more = 0;
c9ffc77a 1271 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1272
e10c758e 1273 return 0;
31b8006e
SW		 1274}
1275
31b8006e
SW		 1276/*
1277 * write as much of pending kvecs to the socket as we can.
1278 * 1 -> done
1279 * 0 -> socket full, but more to do
1280 * <0 -> error
1281 */
1282static int write_partial_kvec(struct ceph_connection *con)
1283{
1284 int ret;
1285
1286 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1287 while (con->out_kvec_bytes > 0) {
1288 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1289 con->out_kvec_left, con->out_kvec_bytes,
1290 con->out_more);
1291 if (ret <= 0)
1292 goto out;
1293 con->out_kvec_bytes -= ret;
1294 if (con->out_kvec_bytes == 0)
1295 break; /* done */
f42299e6
AE		 1296
1297 /* account for full iov entries consumed */
1298 while (ret >= con->out_kvec_cur->iov_len) {
1299 BUG_ON(!con->out_kvec_left);
1300 ret -= con->out_kvec_cur->iov_len;
1301 con->out_kvec_cur++;
1302 con->out_kvec_left--;
1303 }
1304 /* and for a partially-consumed entry */
1305 if (ret) {
1306 con->out_kvec_cur->iov_len -= ret;
1307 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1308 }
1309 }
1310 con->out_kvec_left = 0;
1311 con->out_kvec_is_msg = false;
1312 ret = 1;
1313out:
1314 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1315 con->out_kvec_bytes, con->out_kvec_left, ret);
1316 return ret; /* done! */
1317}
1318
84ca8fc8
AE		 1319static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
1320 size_t len, size_t sent, bool in_trail)
68b4476b 1321{
84ca8fc8 1322 struct ceph_msg *msg = con->out_msg;
bae6acd9 1323 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
7fe1e5e5 1324 bool need_crc = false;
68b4476b 1325
84ca8fc8
AE		 1326 BUG_ON(!msg);
1327 BUG_ON(!sent);
68b4476b 1328
bae6acd9
AE		 1329 msg_pos->data_pos += sent;
1330 msg_pos->page_pos += sent;
7fe1e5e5 1331 if (in_trail)
fe38a2b6 1332 need_crc = ceph_msg_data_advance(&msg->t, sent);
7fe1e5e5
AE		 1333 else if (ceph_msg_has_pagelist(msg))
1334 need_crc = ceph_msg_data_advance(&msg->l, sent);
6aaa4511
AE		 1335#ifdef CONFIG_BLOCK
1336 else if (ceph_msg_has_bio(msg))
1337 need_crc = ceph_msg_data_advance(&msg->b, sent);
1338#endif /* CONFIG_BLOCK */
7fe1e5e5
AE		 1339 BUG_ON(need_crc && sent != len);
1340
5821bd8c
AE		 1341 if (sent < len)
1342 return;
68b4476b 1343
5821bd8c 1344 BUG_ON(sent != len);
bae6acd9
AE		 1345 msg_pos->page_pos = 0;
1346 msg_pos->page++;
1347 msg_pos->did_page_crc = false;
84ca8fc8 1348}
68b4476b 1349
e788182f
AE		 1350static void in_msg_pos_next(struct ceph_connection *con, size_t len,
1351 size_t received)
1352{
1353 struct ceph_msg *msg = con->in_msg;
bae6acd9 1354 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
e788182f
AE		 1355
1356 BUG_ON(!msg);
1357 BUG_ON(!received);
1358
bae6acd9
AE		 1359 msg_pos->data_pos += received;
1360 msg_pos->page_pos += received;
e788182f
AE		 1361 if (received < len)
1362 return;
1363
1364 BUG_ON(received != len);
bae6acd9
AE		 1365 msg_pos->page_pos = 0;
1366 msg_pos->page++;
e788182f 1367#ifdef CONFIG_BLOCK
f9e15777
AE		 1368 if (msg->b.bio)
1369 iter_bio_next(&msg->b.bio_iter, &msg->b.bio_seg);
e788182f
AE		 1370#endif /* CONFIG_BLOCK */
1371}
1372
35b62808
AE		 1373static u32 ceph_crc32c_page(u32 crc, struct page *page,
1374 unsigned int page_offset,
1375 unsigned int length)
1376{
1377 char *kaddr;
1378
1379 kaddr = kmap(page);
1380 BUG_ON(kaddr == NULL);
1381 crc = crc32c(crc, kaddr + page_offset, length);
1382 kunmap(page);
1383
1384 return crc;
1385}
31b8006e
SW		 1386/*
1387 * Write as much message data payload as we can. If we finish, queue
1388 * up the footer.
1389 * 1 -> done, footer is now queued in out_kvec[].
1390 * 0 -> socket full, but more to do
1391 * <0 -> error
1392 */
34d2d200 1393static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1394{
1395 struct ceph_msg *msg = con->out_msg;
bae6acd9 1396 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
95c96174 1397 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
37675b0f 1398 bool do_datacrc = !con->msgr->nocrc;
31b8006e 1399 int ret;
68b4476b 1400 int total_max_write;
84ca8fc8 1401 bool in_trail = false;
97fb1c7f
AE		 1402 size_t trail_len = 0;
1403 size_t trail_off = data_len;
1404
1405 if (ceph_msg_has_trail(msg)) {
43794509 1406 trail_len = msg->t.pagelist->length;
97fb1c7f
AE		 1407 trail_off -= trail_len;
1408 }
31b8006e 1409
34d2d200 1410 dout("%s %p msg %p page %d offset %d\n", __func__,
bae6acd9 1411 con, msg, msg_pos->page, msg_pos->page_pos);
31b8006e 1412
5821bd8c
AE		 1413 /*
1414 * Iterate through each page that contains data to be
1415 * written, and send as much as possible for each.
1416 *
1417 * If we are calculating the data crc (the default), we will
1418 * need to map the page. If we have no pages, they have
1419 * been revoked, so use the zero page.
1420 */
bae6acd9 1421 while (data_len > msg_pos->data_pos) {
31b8006e 1422 struct page *page = NULL;
e387d525
AE		 1423 size_t page_offset;
1424 size_t length;
fe38a2b6
AE		 1425 bool use_cursor = false;
1426 bool last_piece = true; /* preserve existing behavior */
68b4476b 1427
bae6acd9 1428 in_trail = in_trail || msg_pos->data_pos >= trail_off;
5821bd8c 1429 if (!in_trail)
bae6acd9 1430 total_max_write = trail_off - msg_pos->data_pos;
68b4476b 1431
5821bd8c 1432 if (in_trail) {
97fb1c7f 1433 BUG_ON(!ceph_msg_has_trail(msg));
fe38a2b6
AE		 1434 use_cursor = true;
1435 page = ceph_msg_data_next(&msg->t, &page_offset,
1436 &length, &last_piece);
97fb1c7f 1437 } else if (ceph_msg_has_pages(msg)) {
f9e15777 1438 page = msg->p.pages[msg_pos->page];
97fb1c7f 1439 } else if (ceph_msg_has_pagelist(msg)) {
7fe1e5e5
AE		 1440 use_cursor = true;
1441 page = ceph_msg_data_next(&msg->l, &page_offset,
1442 &length, &last_piece);
68b4476b 1443#ifdef CONFIG_BLOCK
97fb1c7f 1444 } else if (ceph_msg_has_bio(msg)) {
6aaa4511
AE		 1445 use_cursor = true;
1446 page = ceph_msg_data_next(&msg->b, &page_offset,
1447 &length, &last_piece);
68b4476b 1448#endif
31b8006e 1449 } else {
57666519 1450 page = zero_page;
31b8006e 1451 }
6aaa4511
AE		 1452 if (!use_cursor) {
1453 length = min_t(int, PAGE_SIZE - msg_pos->page_pos,
fe38a2b6 1454 total_max_write);
68b4476b 1455
6aaa4511
AE		 1456 page_offset = msg_pos->page_pos;
1457 }
bae6acd9 1458 if (do_datacrc && !msg_pos->did_page_crc) {
5821bd8c 1459 u32 crc = le32_to_cpu(msg->footer.data_crc);
31b8006e 1460
35b62808 1461 crc = ceph_crc32c_page(crc, page, page_offset, length);
84ca8fc8 1462 msg->footer.data_crc = cpu_to_le32(crc);
bae6acd9 1463 msg_pos->did_page_crc = true;
31b8006e 1464 }
e387d525 1465 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1466 length, last_piece);
31b8006e
SW		 1467 if (ret <= 0)
1468 goto out;
1469
e387d525 1470 out_msg_pos_next(con, page, length, (size_t) ret, in_trail);
31b8006e
SW		 1471 }
1472
34d2d200 1473 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1474
1475 /* prepare and queue up footer, too */
37675b0f 1476 if (!do_datacrc)
84ca8fc8 1477 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1478 con_out_kvec_reset(con);
859eb799 1479 prepare_write_message_footer(con);
31b8006e
SW		 1480 ret = 1;
1481out:
1482 return ret;
1483}
1484
1485/*
1486 * write some zeros
1487 */
1488static int write_partial_skip(struct ceph_connection *con)
1489{
1490 int ret;
1491
1492 while (con->out_skip > 0) {
31739139 1493 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1494
e1dcb128 1495 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1496 if (ret <= 0)
1497 goto out;
1498 con->out_skip -= ret;
1499 }
1500 ret = 1;
1501out:
1502 return ret;
1503}
1504
1505/*
1506 * Prepare to read connection handshake, or an ack.
1507 */
eed0ef2c
SW		 1508static void prepare_read_banner(struct ceph_connection *con)
1509{
1510 dout("prepare_read_banner %p\n", con);
1511 con->in_base_pos = 0;
1512}
1513
31b8006e
SW		 1514static void prepare_read_connect(struct ceph_connection *con)
1515{
1516 dout("prepare_read_connect %p\n", con);
1517 con->in_base_pos = 0;
1518}
1519
1520static void prepare_read_ack(struct ceph_connection *con)
1521{
1522 dout("prepare_read_ack %p\n", con);
1523 con->in_base_pos = 0;
1524}
1525
1526static void prepare_read_tag(struct ceph_connection *con)
1527{
1528 dout("prepare_read_tag %p\n", con);
1529 con->in_base_pos = 0;
1530 con->in_tag = CEPH_MSGR_TAG_READY;
1531}
1532
1533/*
1534 * Prepare to read a message.
1535 */
1536static int prepare_read_message(struct ceph_connection *con)
1537{
1538 dout("prepare_read_message %p\n", con);
1539 BUG_ON(con->in_msg != NULL);
1540 con->in_base_pos = 0;
1541 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1542 return 0;
1543}
1544
1545
1546static int read_partial(struct ceph_connection *con,
fd51653f 1547 int end, int size, void *object)
31b8006e 1548{
e6cee71f
AE		 1549 while (con->in_base_pos < end) {
1550 int left = end - con->in_base_pos;
31b8006e
SW		 1551 int have = size - left;
1552 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1553 if (ret <= 0)
1554 return ret;
1555 con->in_base_pos += ret;
1556 }
1557 return 1;
1558}
1559
1560
1561/*
1562 * Read all or part of the connect-side handshake on a new connection
1563 */
eed0ef2c 1564static int read_partial_banner(struct ceph_connection *con)
31b8006e 1565{
fd51653f
AE		 1566 int size;
1567 int end;
1568 int ret;
31b8006e 1569
eed0ef2c 1570 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1571
1572 /* peer's banner */
fd51653f
AE		 1573 size = strlen(CEPH_BANNER);
1574 end = size;
1575 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1576 if (ret <= 0)
1577 goto out;
fd51653f
AE		 1578
1579 size = sizeof (con->actual_peer_addr);
1580 end += size;
1581 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1582 if (ret <= 0)
1583 goto out;
fd51653f
AE		 1584
1585 size = sizeof (con->peer_addr_for_me);
1586 end += size;
1587 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1588 if (ret <= 0)
1589 goto out;
fd51653f 1590
eed0ef2c
SW		 1591out:
1592 return ret;
1593}
1594
1595static int read_partial_connect(struct ceph_connection *con)
1596{
fd51653f
AE		 1597 int size;
1598 int end;
1599 int ret;
eed0ef2c
SW		 1600
1601 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1602
fd51653f
AE		 1603 size = sizeof (con->in_reply);
1604 end = size;
1605 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1606 if (ret <= 0)
1607 goto out;
fd51653f
AE		 1608
1609 size = le32_to_cpu(con->in_reply.authorizer_len);
1610 end += size;
1611 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1612 if (ret <= 0)
1613 goto out;
31b8006e 1614
4e7a5dcd
SW		 1615 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1616 con, (int)con->in_reply.tag,
1617 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1618 le32_to_cpu(con->in_reply.global_seq));
1619out:
1620 return ret;
eed0ef2c 1621
31b8006e
SW		 1622}
1623
1624/*
1625 * Verify the hello banner looks okay.
1626 */
1627static int verify_hello(struct ceph_connection *con)
1628{
1629 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1630 pr_err("connect to %s got bad banner\n",
3d14c5d2 1631 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1632 con->error_msg = "protocol error, bad banner";
1633 return -1;
1634 }
1635 return 0;
1636}
1637
1638static bool addr_is_blank(struct sockaddr_storage *ss)
1639{
1640 switch (ss->ss_family) {
1641 case AF_INET:
1642 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1643 case AF_INET6:
1644 return
1645 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1646 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1647 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1648 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1649 }
1650 return false;
1651}
1652
1653static int addr_port(struct sockaddr_storage *ss)
1654{
1655 switch (ss->ss_family) {
1656 case AF_INET:
f28bcfbe 1657 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1658 case AF_INET6:
f28bcfbe 1659 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1660 }
1661 return 0;
1662}
1663
1664static void addr_set_port(struct sockaddr_storage *ss, int p)
1665{
1666 switch (ss->ss_family) {
1667 case AF_INET:
1668 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1669 break;
31b8006e
SW		 1670 case AF_INET6:
1671 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1672 break;
31b8006e
SW		 1673 }
1674}
1675
ee3b56f2
NW		 1676/*
1677 * Unlike other *_pton function semantics, zero indicates success.
1678 */
1679static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1680 char delim, const char **ipend)
1681{
99f0f3b2
AE		 1682 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1683 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1684
1685 memset(ss, 0, sizeof(*ss));
1686
1687 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1688 ss->ss_family = AF_INET;
1689 return 0;
1690 }
1691
1692 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1693 ss->ss_family = AF_INET6;
1694 return 0;
1695 }
1696
1697 return -EINVAL;
1698}
1699
1700/*
1701 * Extract hostname string and resolve using kernel DNS facility.
1702 */
1703#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1704static int ceph_dns_resolve_name(const char *name, size_t namelen,
1705 struct sockaddr_storage *ss, char delim, const char **ipend)
1706{
1707 const char *end, *delim_p;
1708 char *colon_p, *ip_addr = NULL;
1709 int ip_len, ret;
1710
1711 /*
1712 * The end of the hostname occurs immediately preceding the delimiter or
1713 * the port marker (':') where the delimiter takes precedence.
1714 */
1715 delim_p = memchr(name, delim, namelen);
1716 colon_p = memchr(name, ':', namelen);
1717
1718 if (delim_p && colon_p)
1719 end = delim_p < colon_p ? delim_p : colon_p;
1720 else if (!delim_p && colon_p)
1721 end = colon_p;
1722 else {
1723 end = delim_p;
1724 if (!end) /* case: hostname:/ */
1725 end = name + namelen;
1726 }
1727
1728 if (end <= name)
1729 return -EINVAL;
1730
1731 /* do dns_resolve upcall */
1732 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1733 if (ip_len > 0)
1734 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1735 else
1736 ret = -ESRCH;
1737
1738 kfree(ip_addr);
1739
1740 *ipend = end;
1741
1742 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1743 ret, ret ? "failed" : ceph_pr_addr(ss));
1744
1745 return ret;
1746}
1747#else
1748static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1749 struct sockaddr_storage *ss, char delim, const char **ipend)
1750{
1751 return -EINVAL;
1752}
1753#endif
1754
1755/*
1756 * Parse a server name (IP or hostname). If a valid IP address is not found
1757 * then try to extract a hostname to resolve using userspace DNS upcall.
1758 */
1759static int ceph_parse_server_name(const char *name, size_t namelen,
1760 struct sockaddr_storage *ss, char delim, const char **ipend)
1761{
1762 int ret;
1763
1764 ret = ceph_pton(name, namelen, ss, delim, ipend);
1765 if (ret)
1766 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1767
1768 return ret;
1769}
1770
31b8006e
SW		 1771/*
1772 * Parse an ip[:port] list into an addr array. Use the default
1773 * monitor port if a port isn't specified.
1774 */
1775int ceph_parse_ips(const char *c, const char *end,
1776 struct ceph_entity_addr *addr,
1777 int max_count, int *count)
1778{
ee3b56f2 1779 int i, ret = -EINVAL;
31b8006e
SW		 1780 const char *p = c;
1781
1782 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1783 for (i = 0; i < max_count; i++) {
1784 const char *ipend;
1785 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1786 int port;
39139f64
SW		 1787 char delim = ',';
1788
1789 if (*p == '[') {
1790 delim = ']';
1791 p++;
1792 }
31b8006e 1793
ee3b56f2
NW		 1794 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1795 if (ret)
31b8006e 1796 goto bad;
ee3b56f2
NW		 1797 ret = -EINVAL;
1798
31b8006e
SW		 1799 p = ipend;
1800
39139f64
SW		 1801 if (delim == ']') {
1802 if (*p != ']') {
1803 dout("missing matching ']'\n");
1804 goto bad;
1805 }
1806 p++;
1807 }
1808
31b8006e
SW		 1809 /* port? */
1810 if (p < end && *p == ':') {
1811 port = 0;
1812 p++;
1813 while (p < end && *p >= '0' && *p <= '9') {
1814 port = (port * 10) + (*p - '0');
1815 p++;
1816 }
1817 if (port > 65535 || port == 0)
1818 goto bad;
1819 } else {
1820 port = CEPH_MON_PORT;
1821 }
1822
1823 addr_set_port(ss, port);
1824
3d14c5d2 1825 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1826
1827 if (p == end)
1828 break;
1829 if (*p != ',')
1830 goto bad;
1831 p++;
1832 }
1833
1834 if (p != end)
1835 goto bad;
1836
1837 if (count)
1838 *count = i + 1;
1839 return 0;
1840
1841bad:
39139f64 1842 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1843 return ret;
31b8006e 1844}
3d14c5d2 1845EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1846
eed0ef2c 1847static int process_banner(struct ceph_connection *con)
31b8006e 1848{
eed0ef2c 1849 dout("process_banner on %p\n", con);
31b8006e
SW		 1850
1851 if (verify_hello(con) < 0)
1852 return -1;
1853
63f2d211
SW		 1854 ceph_decode_addr(&con->actual_peer_addr);
1855 ceph_decode_addr(&con->peer_addr_for_me);
1856
31b8006e
SW		 1857 /*
1858 * Make sure the other end is who we wanted. note that the other
1859 * end may not yet know their ip address, so if it's 0.0.0.0, give
1860 * them the benefit of the doubt.
1861 */
103e2d3a
SW		 1862 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1863 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1864 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1865 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
cd84db6e 1866 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
3d14c5d2 1867 ceph_pr_addr(&con->peer_addr.in_addr),
cd84db6e 1868 (int)le32_to_cpu(con->peer_addr.nonce),
3d14c5d2 1869 ceph_pr_addr(&con->actual_peer_addr.in_addr),
cd84db6e 1870 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1871 con->error_msg = "wrong peer at address";
31b8006e
SW		 1872 return -1;
1873 }
1874
1875 /*
1876 * did we learn our address?
1877 */
1878 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1879 int port = addr_port(&con->msgr->inst.addr.in_addr);
1880
1881 memcpy(&con->msgr->inst.addr.in_addr,
1882 &con->peer_addr_for_me.in_addr,
1883 sizeof(con->peer_addr_for_me.in_addr));
1884 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1885 encode_my_addr(con->msgr);
eed0ef2c 1886 dout("process_banner learned my addr is %s\n",
3d14c5d2 1887 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1888 }
1889
eed0ef2c
SW		 1890 return 0;
1891}
1892
1893static int process_connect(struct ceph_connection *con)
1894{
3d14c5d2
YS		 1895 u64 sup_feat = con->msgr->supported_features;
1896 u64 req_feat = con->msgr->required_features;
04a419f9 1897 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 1898 int ret;
04a419f9 1899
eed0ef2c
SW		 1900 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1901
31b8006e 1902 switch (con->in_reply.tag) {
04a419f9
SW		 1903 case CEPH_MSGR_TAG_FEATURES:
1904 pr_err("%s%lld %s feature set mismatch,"
1905 " my %llx < server's %llx, missing %llx\n",
1906 ENTITY_NAME(con->peer_name),
3d14c5d2 1907 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1908 sup_feat, server_feat, server_feat & ~sup_feat);
1909 con->error_msg = "missing required protocol features";
0fa6ebc6 1910 reset_connection(con);
04a419f9
SW		 1911 return -1;
1912
31b8006e 1913 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 1914 pr_err("%s%lld %s protocol version mismatch,"
1915 " my %d != server's %d\n",
1916 ENTITY_NAME(con->peer_name),
3d14c5d2 1917 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 1918 le32_to_cpu(con->out_connect.protocol_version),
1919 le32_to_cpu(con->in_reply.protocol_version));
1920 con->error_msg = "protocol version mismatch";
0fa6ebc6 1921 reset_connection(con);
31b8006e
SW		 1922 return -1;
1923
4e7a5dcd
SW		 1924 case CEPH_MSGR_TAG_BADAUTHORIZER:
1925 con->auth_retry++;
1926 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1927 con->auth_retry);
1928 if (con->auth_retry == 2) {
1929 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 1930 return -1;
1931 }
1932 con->auth_retry = 1;
6d4221b5 1933 con_out_kvec_reset(con);
e825a66d 1934 ret = prepare_write_connect(con);
0da5d703
SW		 1935 if (ret < 0)
1936 return ret;
63733a0f 1937 prepare_read_connect(con);
4e7a5dcd 1938 break;
31b8006e
SW		 1939
1940 case CEPH_MSGR_TAG_RESETSESSION:
1941 /*
1942 * If we connected with a large connect_seq but the peer
1943 * has no record of a session with us (no connection, or
1944 * connect_seq == 0), they will send RESETSESION to indicate
1945 * that they must have reset their session, and may have
1946 * dropped messages.
1947 */
1948 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 1949 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 1950 pr_err("%s%lld %s connection reset\n",
1951 ENTITY_NAME(con->peer_name),
3d14c5d2 1952 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 1953 reset_connection(con);
6d4221b5 1954 con_out_kvec_reset(con);
5a0f8fdd
AE		 1955 ret = prepare_write_connect(con);
1956 if (ret < 0)
1957 return ret;
31b8006e
SW		 1958 prepare_read_connect(con);
1959
1960 /* Tell ceph about it. */
ec302645 1961 mutex_unlock(&con->mutex);
31b8006e
SW		 1962 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1963 if (con->ops->peer_reset)
1964 con->ops->peer_reset(con);
ec302645 1965 mutex_lock(&con->mutex);
8dacc7da 1966 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 1967 return -EAGAIN;
31b8006e
SW		 1968 break;
1969
1970 case CEPH_MSGR_TAG_RETRY_SESSION:
1971 /*
1972 * If we sent a smaller connect_seq than the peer has, try
1973 * again with a larger value.
1974 */
5bdca4e0 1975 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 1976 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 1977 le32_to_cpu(con->in_reply.connect_seq));
1978 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 1979 con_out_kvec_reset(con);
5a0f8fdd
AE		 1980 ret = prepare_write_connect(con);
1981 if (ret < 0)
1982 return ret;
31b8006e
SW		 1983 prepare_read_connect(con);
1984 break;
1985
1986 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1987 /*
1988 * If we sent a smaller global_seq than the peer has, try
1989 * again with a larger value.
1990 */
eed0ef2c 1991 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 1992 con->peer_global_seq,
5bdca4e0 1993 le32_to_cpu(con->in_reply.global_seq));
31b8006e 1994 get_global_seq(con->msgr,
5bdca4e0 1995 le32_to_cpu(con->in_reply.global_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_READY:
04a419f9
SW		 2004 if (req_feat & ~server_feat) {
2005 pr_err("%s%lld %s protocol feature mismatch,"
2006 " my required %llx > server's %llx, need %llx\n",
2007 ENTITY_NAME(con->peer_name),
3d14c5d2 2008 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2009 req_feat, server_feat, req_feat & ~server_feat);
2010 con->error_msg = "missing required protocol features";
0fa6ebc6 2011 reset_connection(con);
04a419f9
SW		 2012 return -1;
2013 }
8dacc7da 2014
122070a2 2015 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da
SW		 2016 con->state = CON_STATE_OPEN;
2017
31b8006e
SW		 2018 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2019 con->connect_seq++;
aba558e2 2020 con->peer_features = server_feat;
31b8006e
SW		 2021 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2022 con->peer_global_seq,
2023 le32_to_cpu(con->in_reply.connect_seq),
2024 con->connect_seq);
2025 WARN_ON(con->connect_seq !=
2026 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2027
2028 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2029 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2030
85effe18 2031 con->delay = 0; /* reset backoff memory */
92ac41d0 2032
31b8006e
SW		 2033 prepare_read_tag(con);
2034 break;
2035
2036 case CEPH_MSGR_TAG_WAIT:
2037 /*
2038 * If there is a connection race (we are opening
2039 * connections to each other), one of us may just have
2040 * to WAIT. This shouldn't happen if we are the
2041 * client.
2042 */
04177882
SW		 2043 pr_err("process_connect got WAIT as client\n");
2044 con->error_msg = "protocol error, got WAIT as client";
2045 return -1;
31b8006e
SW		 2046
2047 default:
2048 pr_err("connect protocol error, will retry\n");
2049 con->error_msg = "protocol error, garbage tag during connect";
2050 return -1;
2051 }
2052 return 0;
2053}
2054
2055
2056/*
2057 * read (part of) an ack
2058 */
2059static int read_partial_ack(struct ceph_connection *con)
2060{
fd51653f
AE		 2061 int size = sizeof (con->in_temp_ack);
2062 int end = size;
31b8006e 2063
fd51653f 2064 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2065}
2066
2067
2068/*
2069 * We can finally discard anything that's been acked.
2070 */
2071static void process_ack(struct ceph_connection *con)
2072{
2073 struct ceph_msg *m;
2074 u64 ack = le64_to_cpu(con->in_temp_ack);
2075 u64 seq;
2076
31b8006e
SW		 2077 while (!list_empty(&con->out_sent)) {
2078 m = list_first_entry(&con->out_sent, struct ceph_msg,
2079 list_head);
2080 seq = le64_to_cpu(m->hdr.seq);
2081 if (seq > ack)
2082 break;
2083 dout("got ack for seq %llu type %d at %p\n", seq,
2084 le16_to_cpu(m->hdr.type), m);
4cf9d544 2085 m->ack_stamp = jiffies;
31b8006e
SW		 2086 ceph_msg_remove(m);
2087 }
31b8006e
SW		 2088 prepare_read_tag(con);
2089}
2090
2091
2092
2093
2450418c 2094static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2095 struct kvec *section,
2096 unsigned int sec_len, u32 *crc)
2450418c 2097{
68b4476b 2098 int ret, left;
2450418c
YS		 2099
2100 BUG_ON(!section);
2101
2102 while (section->iov_len < sec_len) {
2103 BUG_ON(section->iov_base == NULL);
2104 left = sec_len - section->iov_len;
2105 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2106 section->iov_len, left);
2107 if (ret <= 0)
2108 return ret;
2109 section->iov_len += ret;
2450418c 2110 }
fe3ad593
AE		 2111 if (section->iov_len == sec_len)
2112 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2113
2450418c
YS		 2114 return 1;
2115}
31b8006e 2116
4740a623 2117static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
68b4476b
YS		 2118
2119static int read_partial_message_pages(struct ceph_connection *con,
2120 struct page **pages,
95c96174 2121 unsigned int data_len, bool do_datacrc)
68b4476b 2122{
bae6acd9 2123 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
e788182f 2124 struct page *page;
afb3d90e
AE		 2125 size_t page_offset;
2126 size_t length;
2127 unsigned int left;
68b4476b 2128 int ret;
68b4476b 2129
68b4476b
YS		 2130 /* (page) data */
2131 BUG_ON(pages == NULL);
bae6acd9 2132 page = pages[msg_pos->page];
afb3d90e
AE		 2133 page_offset = msg_pos->page_pos;
2134 BUG_ON(msg_pos->data_pos >= data_len);
2135 left = data_len - msg_pos->data_pos;
2136 BUG_ON(page_offset >= PAGE_SIZE);
2137 length = min_t(unsigned int, PAGE_SIZE - page_offset, left);
2138
2139 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
68b4476b
YS		 2140 if (ret <= 0)
2141 return ret;
e788182f 2142
35b62808
AE		 2143 if (do_datacrc)
2144 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2145 page_offset, ret);
afb3d90e
AE		 2146
2147 in_msg_pos_next(con, length, ret);
68b4476b
YS		 2148
2149 return ret;
2150}
2151
2152#ifdef CONFIG_BLOCK
2153static int read_partial_message_bio(struct ceph_connection *con,
95c96174 2154 unsigned int data_len, bool do_datacrc)
68b4476b 2155{
b3d56fab 2156 struct ceph_msg *msg = con->in_msg;
bae6acd9 2157 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
b3d56fab 2158 struct bio_vec *bv;
e788182f 2159 struct page *page;
afb3d90e
AE		 2160 size_t page_offset;
2161 size_t length;
2162 unsigned int left;
2163 int ret;
68b4476b 2164
b3d56fab 2165 BUG_ON(!msg);
f9e15777
AE		 2166 BUG_ON(!msg->b.bio_iter);
2167 bv = bio_iovec_idx(msg->b.bio_iter, msg->b.bio_seg);
e788182f 2168 page = bv->bv_page;
afb3d90e
AE		 2169 page_offset = bv->bv_offset + msg_pos->page_pos;
2170 BUG_ON(msg_pos->data_pos >= data_len);
2171 left = data_len - msg_pos->data_pos;
2172 BUG_ON(msg_pos->page_pos >= bv->bv_len);
2173 length = min_t(unsigned int, bv->bv_len - msg_pos->page_pos, left);
68b4476b 2174
afb3d90e 2175 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
68b4476b
YS		 2176 if (ret <= 0)
2177 return ret;
e788182f 2178
35b62808
AE		 2179 if (do_datacrc)
2180 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2181 page_offset, ret);
afb3d90e
AE		 2182
2183 in_msg_pos_next(con, length, ret);
68b4476b
YS		 2184
2185 return ret;
2186}
2187#endif
2188
34d2d200
AE		 2189static int read_partial_msg_data(struct ceph_connection *con)
2190{
2191 struct ceph_msg *msg = con->in_msg;
2192 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2193 const bool do_datacrc = !con->msgr->nocrc;
2194 unsigned int data_len;
2195 int ret;
2196
2197 BUG_ON(!msg);
2198
2199 data_len = le32_to_cpu(con->in_hdr.data_len);
2200 while (msg_pos->data_pos < data_len) {
97fb1c7f 2201 if (ceph_msg_has_pages(msg)) {
f9e15777 2202 ret = read_partial_message_pages(con, msg->p.pages,
34d2d200
AE		 2203 data_len, do_datacrc);
2204 if (ret <= 0)
2205 return ret;
2206#ifdef CONFIG_BLOCK
97fb1c7f 2207 } else if (ceph_msg_has_bio(msg)) {
34d2d200
AE		 2208 ret = read_partial_message_bio(con,
2209 data_len, do_datacrc);
2210 if (ret <= 0)
2211 return ret;
2212#endif
2213 } else {
2214 BUG_ON(1);
2215 }
2216 }
2217
2218 return 1; /* must return > 0 to indicate success */
2219}
2220
31b8006e
SW		 2221/*
2222 * read (part of) a message.
2223 */
2224static int read_partial_message(struct ceph_connection *con)
2225{
2226 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2227 int size;
2228 int end;
31b8006e 2229 int ret;
95c96174 2230 unsigned int front_len, middle_len, data_len;
37675b0f 2231 bool do_datacrc = !con->msgr->nocrc;
ae18756b 2232 u64 seq;
fe3ad593 2233 u32 crc;
31b8006e
SW		 2234
2235 dout("read_partial_message con %p msg %p\n", con, m);
2236
2237 /* header */
fd51653f
AE		 2238 size = sizeof (con->in_hdr);
2239 end = size;
2240 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2241 if (ret <= 0)
2242 return ret;
fe3ad593
AE		 2243
2244 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2245 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2246 pr_err("read_partial_message bad hdr "
2247 " crc %u != expected %u\n",
2248 crc, con->in_hdr.crc);
2249 return -EBADMSG;
2250 }
2251
31b8006e
SW		 2252 front_len = le32_to_cpu(con->in_hdr.front_len);
2253 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2254 return -EIO;
2255 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2256 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2257 return -EIO;
2258 data_len = le32_to_cpu(con->in_hdr.data_len);
2259 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2260 return -EIO;
2261
ae18756b
SW		 2262 /* verify seq# */
2263 seq = le64_to_cpu(con->in_hdr.seq);
2264 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2265 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2266 ENTITY_NAME(con->peer_name),
3d14c5d2 2267 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2268 seq, con->in_seq + 1);
2269 con->in_base_pos = -front_len - middle_len - data_len -
2270 sizeof(m->footer);
2271 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2272 return 0;
2273 } else if ((s64)seq - (s64)con->in_seq > 1) {
2274 pr_err("read_partial_message bad seq %lld expected %lld\n",
2275 seq, con->in_seq + 1);
2276 con->error_msg = "bad message sequence # for incoming message";
2277 return -EBADMSG;
2278 }
2279
31b8006e
SW		 2280 /* allocate message? */
2281 if (!con->in_msg) {
4740a623
SW		 2282 int skip = 0;
2283
31b8006e 2284 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2285 front_len, data_len);
4740a623
SW		 2286 ret = ceph_con_in_msg_alloc(con, &skip);
2287 if (ret < 0)
2288 return ret;
2450418c 2289 if (skip) {
31b8006e 2290 /* skip this message */
a79832f2 2291 dout("alloc_msg said skip message\n");
ae32be31 2292 BUG_ON(con->in_msg);
31b8006e
SW		 2293 con->in_base_pos = -front_len - middle_len - data_len -
2294 sizeof(m->footer);
2295 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2296 con->in_seq++;
31b8006e
SW		 2297 return 0;
2298 }
38941f80 2299
4740a623 2300 BUG_ON(!con->in_msg);
38941f80 2301 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2302 m = con->in_msg;
2303 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2304 if (m->middle)
2305 m->middle->vec.iov_len = 0;
9d7f0f13 2306
78625051 2307 /* prepare for data payload, if any */
a4107026 2308
78625051
AE		 2309 if (data_len)
2310 prepare_message_data(con->in_msg, &con->in_msg_pos);
31b8006e
SW		 2311 }
2312
2313 /* front */
2450418c
YS		 2314 ret = read_partial_message_section(con, &m->front, front_len,
2315 &con->in_front_crc);
2316 if (ret <= 0)
2317 return ret;
31b8006e
SW		 2318
2319 /* middle */
2450418c 2320 if (m->middle) {
213c99ee
SW		 2321 ret = read_partial_message_section(con, &m->middle->vec,
2322 middle_len,
2450418c 2323 &con->in_middle_crc);
31b8006e
SW		 2324 if (ret <= 0)
2325 return ret;
31b8006e
SW		 2326 }
2327
2328 /* (page) data */
34d2d200
AE		 2329 if (data_len) {
2330 ret = read_partial_msg_data(con);
2331 if (ret <= 0)
2332 return ret;
31b8006e
SW		 2333 }
2334
31b8006e 2335 /* footer */
fd51653f
AE		 2336 size = sizeof (m->footer);
2337 end += size;
2338 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2339 if (ret <= 0)
2340 return ret;
2341
31b8006e
SW		 2342 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2343 m, front_len, m->footer.front_crc, middle_len,
2344 m->footer.middle_crc, data_len, m->footer.data_crc);
2345
2346 /* crc ok? */
2347 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2348 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2349 m, con->in_front_crc, m->footer.front_crc);
2350 return -EBADMSG;
2351 }
2352 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2353 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2354 m, con->in_middle_crc, m->footer.middle_crc);
2355 return -EBADMSG;
2356 }
bca064d2 2357 if (do_datacrc &&
31b8006e
SW		 2358 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2359 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2360 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2361 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2362 return -EBADMSG;
2363 }
2364
2365 return 1; /* done! */
2366}
2367
2368/*
2369 * Process message. This happens in the worker thread. The callback should
2370 * be careful not to do anything that waits on other incoming messages or it
2371 * may deadlock.
2372 */
2373static void process_message(struct ceph_connection *con)
2374{
5e095e8b 2375 struct ceph_msg *msg;
31b8006e 2376
38941f80
AE		 2377 BUG_ON(con->in_msg->con != con);
2378 con->in_msg->con = NULL;
5e095e8b 2379 msg = con->in_msg;
31b8006e 2380 con->in_msg = NULL;
36eb71aa 2381 con->ops->put(con);
31b8006e
SW		 2382
2383 /* if first message, set peer_name */
2384 if (con->peer_name.type == 0)
dbad185d 2385 con->peer_name = msg->hdr.src;
31b8006e 2386
31b8006e 2387 con->in_seq++;
ec302645 2388 mutex_unlock(&con->mutex);
31b8006e
SW		 2389
2390 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2391 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2392 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2393 le16_to_cpu(msg->hdr.type),
2394 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2395 le32_to_cpu(msg->hdr.front_len),
2396 le32_to_cpu(msg->hdr.data_len),
2397 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2398 con->ops->dispatch(con, msg);
ec302645
SW		 2399
2400 mutex_lock(&con->mutex);
31b8006e
SW		 2401}
2402
2403
2404/*
2405 * Write something to the socket. Called in a worker thread when the
2406 * socket appears to be writeable and we have something ready to send.
2407 */
2408static int try_write(struct ceph_connection *con)
2409{
31b8006e
SW		 2410 int ret = 1;
2411
d59315ca 2412 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2413
31b8006e
SW		 2414more:
2415 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2416
2417 /* open the socket first? */
8dacc7da
SW		 2418 if (con->state == CON_STATE_PREOPEN) {
2419 BUG_ON(con->sock);
2420 con->state = CON_STATE_CONNECTING;
a5988c49 2421
e2200423 2422 con_out_kvec_reset(con);
e825a66d 2423 prepare_write_banner(con);
eed0ef2c 2424 prepare_read_banner(con);
31b8006e 2425
cf3e5c40 2426 BUG_ON(con->in_msg);
31b8006e
SW		 2427 con->in_tag = CEPH_MSGR_TAG_READY;
2428 dout("try_write initiating connect on %p new state %lu\n",
2429 con, con->state);
41617d0c
AE		 2430 ret = ceph_tcp_connect(con);
2431 if (ret < 0) {
31b8006e 2432 con->error_msg = "connect error";
31b8006e
SW		 2433 goto out;
2434 }
2435 }
2436
2437more_kvec:
2438 /* kvec data queued? */
2439 if (con->out_skip) {
2440 ret = write_partial_skip(con);
2441 if (ret <= 0)
42961d23 2442 goto out;
31b8006e
SW		 2443 }
2444 if (con->out_kvec_left) {
2445 ret = write_partial_kvec(con);
2446 if (ret <= 0)
42961d23 2447 goto out;
31b8006e
SW		 2448 }
2449
2450 /* msg pages? */
2451 if (con->out_msg) {
c86a2930
SW		 2452 if (con->out_msg_done) {
2453 ceph_msg_put(con->out_msg);
2454 con->out_msg = NULL; /* we're done with this one */
2455 goto do_next;
2456 }
2457
34d2d200 2458 ret = write_partial_message_data(con);
31b8006e
SW		 2459 if (ret == 1)
2460 goto more_kvec; /* we need to send the footer, too! */
2461 if (ret == 0)
42961d23 2462 goto out;
31b8006e 2463 if (ret < 0) {
34d2d200 2464 dout("try_write write_partial_message_data err %d\n",
31b8006e 2465 ret);
42961d23 2466 goto out;
31b8006e
SW		 2467 }
2468 }
2469
c86a2930 2470do_next:
8dacc7da 2471 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2472 /* is anything else pending? */
2473 if (!list_empty(&con->out_queue)) {
2474 prepare_write_message(con);
2475 goto more;
2476 }
2477 if (con->in_seq > con->in_seq_acked) {
2478 prepare_write_ack(con);
2479 goto more;
2480 }
c9ffc77a 2481 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2482 prepare_write_keepalive(con);
2483 goto more;
2484 }
2485 }
2486
2487 /* Nothing to do! */
c9ffc77a 2488 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2489 dout("try_write nothing else to write.\n");
31b8006e
SW		 2490 ret = 0;
2491out:
42961d23 2492 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2493 return ret;
2494}
2495
2496
2497
2498/*
2499 * Read what we can from the socket.
2500 */
2501static int try_read(struct ceph_connection *con)
2502{
31b8006e
SW		 2503 int ret = -1;
2504
8dacc7da
SW		 2505more:
2506 dout("try_read start on %p state %lu\n", con, con->state);
2507 if (con->state != CON_STATE_CONNECTING &&
2508 con->state != CON_STATE_NEGOTIATING &&
2509 con->state != CON_STATE_OPEN)
31b8006e
SW		 2510 return 0;
2511
8dacc7da 2512 BUG_ON(!con->sock);
ec302645 2513
31b8006e
SW		 2514 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2515 con->in_base_pos);
0da5d703 2516
8dacc7da 2517 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2518 dout("try_read connecting\n");
2519 ret = read_partial_banner(con);
2520 if (ret <= 0)
ab166d5a 2521 goto out;
7593af92
AE		 2522 ret = process_banner(con);
2523 if (ret < 0)
2524 goto out;
2525
8dacc7da 2526 con->state = CON_STATE_NEGOTIATING;
7593af92 2527
6d4221b5
JS		 2528 /*
2529 * Received banner is good, exchange connection info.
2530 * Do not reset out_kvec, as sending our banner raced
2531 * with receiving peer banner after connect completed.
2532 */
7593af92
AE		 2533 ret = prepare_write_connect(con);
2534 if (ret < 0)
2535 goto out;
2536 prepare_read_connect(con);
2537
2538 /* Send connection info before awaiting response */
0da5d703
SW		 2539 goto out;
2540 }
2541
8dacc7da 2542 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2543 dout("try_read negotiating\n");
31b8006e
SW		 2544 ret = read_partial_connect(con);
2545 if (ret <= 0)
31b8006e 2546 goto out;
98bdb0aa
SW		 2547 ret = process_connect(con);
2548 if (ret < 0)
2549 goto out;
31b8006e
SW		 2550 goto more;
2551 }
2552
122070a2 2553 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2554
31b8006e
SW		 2555 if (con->in_base_pos < 0) {
2556 /*
2557 * skipping + discarding content.
2558 *
2559 * FIXME: there must be a better way to do this!
2560 */
84495f49
AE		 2561 static char buf[SKIP_BUF_SIZE];
2562 int skip = min((int) sizeof (buf), -con->in_base_pos);
2563
31b8006e
SW		 2564 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2565 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2566 if (ret <= 0)
98bdb0aa 2567 goto out;
31b8006e
SW		 2568 con->in_base_pos += ret;
2569 if (con->in_base_pos)
2570 goto more;
2571 }
2572 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2573 /*
2574 * what's next?
2575 */
2576 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2577 if (ret <= 0)
98bdb0aa 2578 goto out;
31b8006e
SW		 2579 dout("try_read got tag %d\n", (int)con->in_tag);
2580 switch (con->in_tag) {
2581 case CEPH_MSGR_TAG_MSG:
2582 prepare_read_message(con);
2583 break;
2584 case CEPH_MSGR_TAG_ACK:
2585 prepare_read_ack(con);
2586 break;
2587 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2588 con_close_socket(con);
2589 con->state = CON_STATE_CLOSED;
98bdb0aa 2590 goto out;
31b8006e
SW		 2591 default:
2592 goto bad_tag;
2593 }
2594 }
2595 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2596 ret = read_partial_message(con);
2597 if (ret <= 0) {
2598 switch (ret) {
2599 case -EBADMSG:
2600 con->error_msg = "bad crc";
2601 ret = -EIO;
98bdb0aa 2602 break;
31b8006e
SW		 2603 case -EIO:
2604 con->error_msg = "io error";
98bdb0aa 2605 break;
31b8006e 2606 }
98bdb0aa 2607 goto out;
31b8006e
SW		 2608 }
2609 if (con->in_tag == CEPH_MSGR_TAG_READY)
2610 goto more;
2611 process_message(con);
7b862e07
SW		 2612 if (con->state == CON_STATE_OPEN)
2613 prepare_read_tag(con);
31b8006e
SW		 2614 goto more;
2615 }
2616 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2617 ret = read_partial_ack(con);
2618 if (ret <= 0)
98bdb0aa 2619 goto out;
31b8006e
SW		 2620 process_ack(con);
2621 goto more;
2622 }
2623
31b8006e 2624out:
98bdb0aa 2625 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2626 return ret;
2627
2628bad_tag:
2629 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2630 con->error_msg = "protocol error, garbage tag";
2631 ret = -1;
2632 goto out;
2633}
2634
2635
2636/*
802c6d96
AE		 2637 * Atomically queue work on a connection after the specified delay.
2638 * Bump @con reference to avoid races with connection teardown.
2639 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2640 */
802c6d96 2641static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2642{
31b8006e 2643 if (!con->ops->get(con)) {
802c6d96
AE		 2644 dout("%s %p ref count 0\n", __func__, con);
2645
2646 return -ENOENT;
31b8006e
SW		 2647 }
2648
802c6d96
AE		 2649 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2650 dout("%s %p - already queued\n", __func__, con);
31b8006e 2651 con->ops->put(con);
802c6d96
AE		 2652
2653 return -EBUSY;
31b8006e 2654 }
802c6d96
AE		 2655
2656 dout("%s %p %lu\n", __func__, con, delay);
2657
2658 return 0;
2659}
2660
2661static void queue_con(struct ceph_connection *con)
2662{
2663 (void) queue_con_delay(con, 0);
31b8006e
SW		 2664}
2665
7bb21d68
AE		 2666static bool con_sock_closed(struct ceph_connection *con)
2667{
c9ffc77a 2668 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2669 return false;
2670
2671#define CASE(x) \
2672 case CON_STATE_ ## x: \
2673 con->error_msg = "socket closed (con state " #x ")"; \
2674 break;
2675
2676 switch (con->state) {
2677 CASE(CLOSED);
2678 CASE(PREOPEN);
2679 CASE(CONNECTING);
2680 CASE(NEGOTIATING);
2681 CASE(OPEN);
2682 CASE(STANDBY);
2683 default:
2684 pr_warning("%s con %p unrecognized state %lu\n",
2685 __func__, con, con->state);
2686 con->error_msg = "unrecognized con state";
2687 BUG();
2688 break;
2689 }
2690#undef CASE
2691
2692 return true;
2693}
2694
f20a39fd
AE		 2695static bool con_backoff(struct ceph_connection *con)
2696{
2697 int ret;
2698
2699 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2700 return false;
2701
2702 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2703 if (ret) {
2704 dout("%s: con %p FAILED to back off %lu\n", __func__,
2705 con, con->delay);
2706 BUG_ON(ret == -ENOENT);
2707 con_flag_set(con, CON_FLAG_BACKOFF);
2708 }
2709
2710 return true;
2711}
2712
93209264
AE		 2713/* Finish fault handling; con->mutex must *not* be held here */
2714
2715static void con_fault_finish(struct ceph_connection *con)
2716{
2717 /*
2718 * in case we faulted due to authentication, invalidate our
2719 * current tickets so that we can get new ones.
2720 */
2721 if (con->auth_retry && con->ops->invalidate_authorizer) {
2722 dout("calling invalidate_authorizer()\n");
2723 con->ops->invalidate_authorizer(con);
2724 }
2725
2726 if (con->ops->fault)
2727 con->ops->fault(con);
2728}
2729
31b8006e
SW		 2730/*
2731 * Do some work on a connection. Drop a connection ref when we're done.
2732 */
2733static void con_work(struct work_struct *work)
2734{
2735 struct ceph_connection *con = container_of(work, struct ceph_connection,
2736 work.work);
49659416 2737 bool fault;
31b8006e 2738
9dd4658d 2739 mutex_lock(&con->mutex);
49659416
AE		 2740 while (true) {
2741 int ret;
31b8006e 2742
49659416
AE		 2743 if ((fault = con_sock_closed(con))) {
2744 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2745 break;
2746 }
2747 if (con_backoff(con)) {
2748 dout("%s: con %p BACKOFF\n", __func__, con);
2749 break;
2750 }
2751 if (con->state == CON_STATE_STANDBY) {
2752 dout("%s: con %p STANDBY\n", __func__, con);
2753 break;
2754 }
2755 if (con->state == CON_STATE_CLOSED) {
2756 dout("%s: con %p CLOSED\n", __func__, con);
2757 BUG_ON(con->sock);
2758 break;
2759 }
2760 if (con->state == CON_STATE_PREOPEN) {
2761 dout("%s: con %p PREOPEN\n", __func__, con);
2762 BUG_ON(con->sock);
2763 }
0da5d703 2764
49659416
AE		 2765 ret = try_read(con);
2766 if (ret < 0) {
2767 if (ret == -EAGAIN)
2768 continue;
2769 con->error_msg = "socket error on read";
2770 fault = true;
2771 break;
2772 }
2773
2774 ret = try_write(con);
2775 if (ret < 0) {
2776 if (ret == -EAGAIN)
2777 continue;
2778 con->error_msg = "socket error on write";
2779 fault = true;
2780 }
2781
2782 break; /* If we make it to here, we're done */
3a140a0d 2783 }
b6e7b6a1
AE		 2784 if (fault)
2785 con_fault(con);
9dd4658d 2786 mutex_unlock(&con->mutex);
0da5d703 2787
b6e7b6a1
AE		 2788 if (fault)
2789 con_fault_finish(con);
2790
2791 con->ops->put(con);
31b8006e
SW		 2792}
2793
31b8006e
SW		 2794/*
2795 * Generic error/fault handler. A retry mechanism is used with
2796 * exponential backoff
2797 */
93209264 2798static void con_fault(struct ceph_connection *con)
31b8006e 2799{
28362986 2800 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
3d14c5d2 2801 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2802 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2803 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2804
122070a2 2805 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2806 con->state != CON_STATE_NEGOTIATING &&
2807 con->state != CON_STATE_OPEN);
ec302645 2808
31b8006e 2809 con_close_socket(con);
5e095e8b 2810
c9ffc77a 2811 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2812 dout("fault on LOSSYTX channel, marking CLOSED\n");
2813 con->state = CON_STATE_CLOSED;
93209264 2814 return;
3b5ede07
SW		 2815 }
2816
5e095e8b 2817 if (con->in_msg) {
38941f80
AE		 2818 BUG_ON(con->in_msg->con != con);
2819 con->in_msg->con = NULL;
5e095e8b
SW		 2820 ceph_msg_put(con->in_msg);
2821 con->in_msg = NULL;
36eb71aa 2822 con->ops->put(con);
5e095e8b 2823 }
31b8006e 2824
e80a52d1
SW		 2825 /* Requeue anything that hasn't been acked */
2826 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2827
e76661d0
SW		 2828 /* If there are no messages queued or keepalive pending, place
2829 * the connection in a STANDBY state */
2830 if (list_empty(&con->out_queue) &&
c9ffc77a 2831 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2832 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2833 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2834 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2835 } else {
2836 /* retry after a delay. */
8dacc7da 2837 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2838 if (con->delay == 0)
2839 con->delay = BASE_DELAY_INTERVAL;
2840 else if (con->delay < MAX_DELAY_INTERVAL)
2841 con->delay *= 2;
c9ffc77a 2842 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2843 queue_con(con);
31b8006e 2844 }
31b8006e
SW		 2845}
2846
2847
2848
2849/*
15d9882c 2850 * initialize a new messenger instance
31b8006e 2851 */
15d9882c
AE		 2852void ceph_messenger_init(struct ceph_messenger *msgr,
2853 struct ceph_entity_addr *myaddr,
2854 u32 supported_features,
2855 u32 required_features,
2856 bool nocrc)
31b8006e 2857{
3d14c5d2
YS		 2858 msgr->supported_features = supported_features;
2859 msgr->required_features = required_features;
2860
31b8006e
SW		 2861 spin_lock_init(&msgr->global_seq_lock);
2862
31b8006e
SW		 2863 if (myaddr)
2864 msgr->inst.addr = *myaddr;
2865
2866 /* select a random nonce */
ac8839d7 2867 msgr->inst.addr.type = 0;
103e2d3a 2868 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2869 encode_my_addr(msgr);
15d9882c 2870 msgr->nocrc = nocrc;
31b8006e 2871
a2a32584 2872 atomic_set(&msgr->stopping, 0);
31b8006e 2873
15d9882c 2874 dout("%s %p\n", __func__, msgr);
31b8006e 2875}
15d9882c 2876EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2877
e00de341
SW		 2878static void clear_standby(struct ceph_connection *con)
2879{
2880 /* come back from STANDBY? */
8dacc7da 2881 if (con->state == CON_STATE_STANDBY) {
e00de341 2882 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2883 con->state = CON_STATE_PREOPEN;
e00de341 2884 con->connect_seq++;
c9ffc77a
AE		 2885 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2886 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2887 }
2888}
2889
31b8006e
SW		 2890/*
2891 * Queue up an outgoing message on the given connection.
2892 */
2893void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2894{
31b8006e 2895 /* set src+dst */
dbad185d 2896 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2897 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2898 msg->needs_out_seq = true;
2899
ec302645 2900 mutex_lock(&con->mutex);
92ce034b 2901
8dacc7da 2902 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2903 dout("con_send %p closed, dropping %p\n", con, msg);
2904 ceph_msg_put(msg);
2905 mutex_unlock(&con->mutex);
2906 return;
2907 }
2908
38941f80 2909 BUG_ON(msg->con != NULL);
36eb71aa 2910 msg->con = con->ops->get(con);
92ce034b
AE		 2911 BUG_ON(msg->con == NULL);
2912
31b8006e
SW		 2913 BUG_ON(!list_empty(&msg->list_head));
2914 list_add_tail(&msg->list_head, &con->out_queue);
2915 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2916 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2917 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2918 le32_to_cpu(msg->hdr.front_len),
2919 le32_to_cpu(msg->hdr.middle_len),
2920 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 2921
2922 clear_standby(con);
ec302645 2923 mutex_unlock(&con->mutex);
31b8006e
SW		 2924
2925 /* if there wasn't anything waiting to send before, queue
2926 * new work */
c9ffc77a 2927 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2928 queue_con(con);
2929}
3d14c5d2 2930EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 2931
2932/*
2933 * Revoke a message that was previously queued for send
2934 */
6740a845 2935void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 2936{
6740a845
AE		 2937 struct ceph_connection *con = msg->con;
2938
2939 if (!con)
2940 return; /* Message not in our possession */
2941
ec302645 2942 mutex_lock(&con->mutex);
31b8006e 2943 if (!list_empty(&msg->list_head)) {
38941f80 2944 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 2945 list_del_init(&msg->list_head);
38941f80 2946 BUG_ON(msg->con == NULL);
36eb71aa 2947 msg->con->ops->put(msg->con);
38941f80 2948 msg->con = NULL;
31b8006e 2949 msg->hdr.seq = 0;
38941f80 2950
31b8006e 2951 ceph_msg_put(msg);
ed98adad
SW		 2952 }
2953 if (con->out_msg == msg) {
38941f80 2954 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 2955 con->out_msg = NULL;
31b8006e
SW		 2956 if (con->out_kvec_is_msg) {
2957 con->out_skip = con->out_kvec_bytes;
2958 con->out_kvec_is_msg = false;
2959 }
ed98adad 2960 msg->hdr.seq = 0;
92ce034b
AE		 2961
2962 ceph_msg_put(msg);
31b8006e 2963 }
ec302645 2964 mutex_unlock(&con->mutex);
31b8006e
SW		 2965}
2966
350b1c32 2967/*
0d59ab81 2968 * Revoke a message that we may be reading data into
350b1c32 2969 */
8921d114 2970void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 2971{
8921d114
AE		 2972 struct ceph_connection *con;
2973
2974 BUG_ON(msg == NULL);
2975 if (!msg->con) {
2976 dout("%s msg %p null con\n", __func__, msg);
2977
2978 return; /* Message not in our possession */
2979 }
2980
2981 con = msg->con;
350b1c32 2982 mutex_lock(&con->mutex);
8921d114 2983 if (con->in_msg == msg) {
95c96174
ED		 2984 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2985 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2986 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 2987
2988 /* skip rest of message */
8921d114
AE		 2989 dout("%s %p msg %p revoked\n", __func__, con, msg);
2990 con->in_base_pos = con->in_base_pos -
350b1c32 2991 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 2992 front_len -
2993 middle_len -
2994 data_len -
350b1c32 2995 sizeof(struct ceph_msg_footer);
350b1c32
SW		 2996 ceph_msg_put(con->in_msg);
2997 con->in_msg = NULL;
2998 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2999 con->in_seq++;
350b1c32 3000 } else {
8921d114
AE		 3001 dout("%s %p in_msg %p msg %p no-op\n",
3002 __func__, con, con->in_msg, msg);
350b1c32
SW		 3003 }
3004 mutex_unlock(&con->mutex);
3005}
3006
31b8006e
SW		 3007/*
3008 * Queue a keepalive byte to ensure the tcp connection is alive.
3009 */
3010void ceph_con_keepalive(struct ceph_connection *con)
3011{
e00de341 3012 dout("con_keepalive %p\n", con);
00650931 3013 mutex_lock(&con->mutex);
e00de341 3014 clear_standby(con);
00650931 3015 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3016 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3017 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3018 queue_con(con);
3019}
3d14c5d2 3020EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3021
43794509
AE		 3022static void ceph_msg_data_init(struct ceph_msg_data *data)
3023{
3024 data->type = CEPH_MSG_DATA_NONE;
3025}
3026
02afca6c 3027void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3028 size_t length, size_t alignment)
02afca6c 3029{
07aa1558
AE		 3030 BUG_ON(!pages);
3031 BUG_ON(!length);
43794509 3032 BUG_ON(msg->p.type != CEPH_MSG_DATA_NONE);
02afca6c 3033
43794509 3034 msg->p.type = CEPH_MSG_DATA_PAGES;
f9e15777
AE		 3035 msg->p.pages = pages;
3036 msg->p.length = length;
3037 msg->p.alignment = alignment & ~PAGE_MASK;
02afca6c
AE		 3038}
3039EXPORT_SYMBOL(ceph_msg_data_set_pages);
31b8006e 3040
27fa8385
AE		 3041void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3042 struct ceph_pagelist *pagelist)
3043{
07aa1558
AE		 3044 BUG_ON(!pagelist);
3045 BUG_ON(!pagelist->length);
43794509 3046 BUG_ON(msg->l.type != CEPH_MSG_DATA_NONE);
27fa8385 3047
43794509 3048 msg->l.type = CEPH_MSG_DATA_PAGELIST;
f9e15777 3049 msg->l.pagelist = pagelist;
27fa8385
AE		 3050}
3051EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3052
3053void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio)
3054{
07aa1558 3055 BUG_ON(!bio);
43794509 3056 BUG_ON(msg->b.type != CEPH_MSG_DATA_NONE);
27fa8385 3057
43794509 3058 msg->b.type = CEPH_MSG_DATA_BIO;
f9e15777 3059 msg->b.bio = bio;
27fa8385
AE		 3060}
3061EXPORT_SYMBOL(ceph_msg_data_set_bio);
3062
3063void ceph_msg_data_set_trail(struct ceph_msg *msg, struct ceph_pagelist *trail)
3064{
07aa1558
AE		 3065 BUG_ON(!trail);
3066 BUG_ON(!trail->length);
43794509 3067 BUG_ON(msg->b.type != CEPH_MSG_DATA_NONE);
27fa8385 3068
43794509
AE		 3069 msg->t.type = CEPH_MSG_DATA_PAGELIST;
3070 msg->t.pagelist = trail;
27fa8385
AE		 3071}
3072EXPORT_SYMBOL(ceph_msg_data_set_trail);
3073
31b8006e
SW		 3074/*
3075 * construct a new message with given type, size
3076 * the new msg has a ref count of 1.
3077 */
b61c2763
SW		 3078struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3079 bool can_fail)
31b8006e
SW		 3080{
3081 struct ceph_msg *m;
3082
9516e45b 3083 m = kzalloc(sizeof(*m), flags);
31b8006e
SW		 3084 if (m == NULL)
3085 goto out;
31b8006e
SW		 3086
3087 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3088 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3089 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3090
9516e45b
AE		 3091 INIT_LIST_HEAD(&m->list_head);
3092 kref_init(&m->kref);
ca20892d 3093
43794509
AE		 3094 ceph_msg_data_init(&m->p);
3095 ceph_msg_data_init(&m->l);
3096 ceph_msg_data_init(&m->b);
3097 ceph_msg_data_init(&m->t);
3098
31b8006e 3099 /* front */
9516e45b 3100 m->front_max = front_len;
31b8006e
SW		 3101 if (front_len) {
3102 if (front_len > PAGE_CACHE_SIZE) {
34d23762 3103 m->front.iov_base = __vmalloc(front_len, flags,
31b8006e
SW		 3104 PAGE_KERNEL);
3105 m->front_is_vmalloc = true;
3106 } else {
34d23762 3107 m->front.iov_base = kmalloc(front_len, flags);
31b8006e
SW		 3108 }
3109 if (m->front.iov_base == NULL) {
b61c2763 3110 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3111 front_len);
3112 goto out2;
3113 }
3114 } else {
3115 m->front.iov_base = NULL;
3116 }
3117 m->front.iov_len = front_len;
3118
bb257664 3119 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3120 return m;
3121
3122out2:
3123 ceph_msg_put(m);
3124out:
b61c2763
SW		 3125 if (!can_fail) {
3126 pr_err("msg_new can't create type %d front %d\n", type,
3127 front_len);
f0ed1b7c 3128 WARN_ON(1);
b61c2763
SW		 3129 } else {
3130 dout("msg_new can't create type %d front %d\n", type,
3131 front_len);
3132 }
a79832f2 3133 return NULL;
31b8006e 3134}
3d14c5d2 3135EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3136
31b8006e
SW		 3137/*
3138 * Allocate "middle" portion of a message, if it is needed and wasn't
3139 * allocated by alloc_msg. This allows us to read a small fixed-size
3140 * per-type header in the front and then gracefully fail (i.e.,
3141 * propagate the error to the caller based on info in the front) when
3142 * the middle is too large.
3143 */
2450418c 3144static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3145{
3146 int type = le16_to_cpu(msg->hdr.type);
3147 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3148
3149 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3150 ceph_msg_type_name(type), middle_len);
3151 BUG_ON(!middle_len);
3152 BUG_ON(msg->middle);
3153
b6c1d5b8 3154 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3155 if (!msg->middle)
3156 return -ENOMEM;
3157 return 0;
3158}
3159
2450418c 3160/*
1c20f2d2
AE		 3161 * Allocate a message for receiving an incoming message on a
3162 * connection, and save the result in con->in_msg. Uses the
3163 * connection's private alloc_msg op if available.
3164 *
4740a623
SW		 3165 * Returns 0 on success, or a negative error code.
3166 *
3167 * On success, if we set *skip = 1:
3168 * - the next message should be skipped and ignored.
3169 * - con->in_msg == NULL
3170 * or if we set *skip = 0:
3171 * - con->in_msg is non-null.
3172 * On error (ENOMEM, EAGAIN, ...),
3173 * - con->in_msg == NULL
2450418c 3174 */
4740a623 3175static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3176{
4740a623 3177 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3178 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3179 struct ceph_msg *msg;
4740a623 3180 int ret = 0;
2450418c 3181
1c20f2d2 3182 BUG_ON(con->in_msg != NULL);
53ded495 3183 BUG_ON(!con->ops->alloc_msg);
2450418c 3184
53ded495
AE		 3185 mutex_unlock(&con->mutex);
3186 msg = con->ops->alloc_msg(con, hdr, skip);
3187 mutex_lock(&con->mutex);
3188 if (con->state != CON_STATE_OPEN) {
3189 if (msg)
1d866d1c 3190 ceph_msg_put(msg);
53ded495
AE		 3191 return -EAGAIN;
3192 }
4137577a
AE		 3193 if (msg) {
3194 BUG_ON(*skip);
3195 con->in_msg = msg;
36eb71aa 3196 con->in_msg->con = con->ops->get(con);
92ce034b 3197 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3198 } else {
3199 /*
3200 * Null message pointer means either we should skip
3201 * this message or we couldn't allocate memory. The
3202 * former is not an error.
3203 */
3204 if (*skip)
3205 return 0;
3206 con->error_msg = "error allocating memory for incoming message";
3207
53ded495 3208 return -ENOMEM;
2450418c 3209 }
1c20f2d2 3210 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3211
1c20f2d2
AE		 3212 if (middle_len && !con->in_msg->middle) {
3213 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3214 if (ret < 0) {
1c20f2d2
AE		 3215 ceph_msg_put(con->in_msg);
3216 con->in_msg = NULL;
2450418c
YS		 3217 }
3218 }
9d7f0f13 3219
4740a623 3220 return ret;
2450418c
YS		 3221}
3222
31b8006e
SW		 3223
3224/*
3225 * Free a generically kmalloc'd message.
3226 */
3227void ceph_msg_kfree(struct ceph_msg *m)
3228{
3229 dout("msg_kfree %p\n", m);
3230 if (m->front_is_vmalloc)
3231 vfree(m->front.iov_base);
3232 else
3233 kfree(m->front.iov_base);
3234 kfree(m);
3235}
3236
3237/*
3238 * Drop a msg ref. Destroy as needed.
3239 */
c2e552e7
SW		 3240void ceph_msg_last_put(struct kref *kref)
3241{
3242 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
31b8006e 3243
c2e552e7
SW		 3244 dout("ceph_msg_put last one on %p\n", m);
3245 WARN_ON(!list_empty(&m->list_head));
3246
3247 /* drop middle, data, if any */
3248 if (m->middle) {
3249 ceph_buffer_put(m->middle);
3250 m->middle = NULL;
31b8006e 3251 }
97fb1c7f 3252 if (ceph_msg_has_pages(m)) {
f9e15777
AE		 3253 m->p.length = 0;
3254 m->p.pages = NULL;
97fb1c7f 3255 }
c2e552e7 3256
97fb1c7f 3257 if (ceph_msg_has_pagelist(m)) {
f9e15777
AE		 3258 ceph_pagelist_release(m->l.pagelist);
3259 kfree(m->l.pagelist);
3260 m->l.pagelist = NULL;
58bb3b37
SW		 3261 }
3262
97fb1c7f 3263 if (ceph_msg_has_trail(m))
43794509 3264 m->t.pagelist = NULL;
68b4476b 3265
c2e552e7
SW		 3266 if (m->pool)
3267 ceph_msgpool_put(m->pool, m);
3268 else
3269 ceph_msg_kfree(m);
31b8006e 3270}
3d14c5d2 3271EXPORT_SYMBOL(ceph_msg_last_put);
9ec7cab1
SW		 3272
3273void ceph_msg_dump(struct ceph_msg *msg)
3274{
4a73ef27 3275 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
f9e15777 3276 msg->front_max, msg->p.length);
9ec7cab1
SW		 3277 print_hex_dump(KERN_DEBUG, "header: ",
3278 DUMP_PREFIX_OFFSET, 16, 1,
3279 &msg->hdr, sizeof(msg->hdr), true);
3280 print_hex_dump(KERN_DEBUG, " front: ",
3281 DUMP_PREFIX_OFFSET, 16, 1,
3282 msg->front.iov_base, msg->front.iov_len, true);
3283 if (msg->middle)
3284 print_hex_dump(KERN_DEBUG, "middle: ",
3285 DUMP_PREFIX_OFFSET, 16, 1,
3286 msg->middle->vec.iov_base,
3287 msg->middle->vec.iov_len, true);
3288 print_hex_dump(KERN_DEBUG, "footer: ",
3289 DUMP_PREFIX_OFFSET, 16, 1,
3290 &msg->footer, sizeof(msg->footer), true);
3291}
3d14c5d2 3292EXPORT_SYMBOL(ceph_msg_dump);
Linux kernel - Deliverables
This page took 0.361693 seconds and 5 git commands to generate.