]> git.ipfire.org Git - people/ms/linux.git/blame - net/ceph/messenger.c
projects / people / ms / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: skip message if too big to receive
[people/ms/linux.git] / net / ceph / messenger.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
5a0e3ad6 9#include <linux/slab.h>
31b8006e
SW		 10#include <linux/socket.h>
11#include <linux/string.h>
3ebc21f7 12#ifdef CONFIG_BLOCK
68b4476b 13#include <linux/bio.h>
3ebc21f7 14#endif /* CONFIG_BLOCK */
ee3b56f2 15#include <linux/dns_resolver.h>
31b8006e
SW		 16#include <net/tcp.h>
17
3d14c5d2
YS		 18#include <linux/ceph/libceph.h>
19#include <linux/ceph/messenger.h>
20#include <linux/ceph/decode.h>
21#include <linux/ceph/pagelist.h>
bc3b2d7f 22#include <linux/export.h>
31b8006e 23
fe38a2b6
AE		 24#define list_entry_next(pos, member) \
25 list_entry(pos->member.next, typeof(*pos), member)
26
31b8006e
SW		 27/*
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
33 * the sender.
34 */
35
bc18f4b1
AE		 36/*
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
40 * unexpected state.
41 *
42 * --------
43 * | NEW* | transient initial state
44 * --------
45 * | con_sock_state_init()
46 * v
47 * ----------
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
50 * ^ \
51 * | \ con_sock_state_connecting()
52 * | ----------------------
53 * | \
54 * + con_sock_state_closed() \
fbb85a47
SW		 55 * |+--------------------------- \
56 * | \ \ \
57 * | ----------- \ \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
60 * | ^ \ |
61 * | | \ |
62 * | + con_sock_state_closing() \ |
63 * | / \ | |
64 * | / --------------- | |
65 * | / \ v v
bc18f4b1
AE		 66 * | / --------------
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
70 * | | v
71 * -------------
72 * | CONNECTED | TCP connection established
73 * -------------
74 *
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
76 */
ce2c8903
AE		 77
78#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
83
8dacc7da
SW		 84/*
85 * connection states
86 */
87#define CON_STATE_CLOSED 1 /* -> PREOPEN */
88#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
93
4a861692
SW		 94/*
95 * ceph_connection flag bits
96 */
97#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 103
c9ffc77a
AE		 104static bool con_flag_valid(unsigned long con_flag)
105{
106 switch (con_flag) {
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
112 return true;
113 default:
114 return false;
115 }
116}
117
118static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
119{
120 BUG_ON(!con_flag_valid(con_flag));
121
122 clear_bit(con_flag, &con->flags);
123}
124
125static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
126{
127 BUG_ON(!con_flag_valid(con_flag));
128
129 set_bit(con_flag, &con->flags);
130}
131
132static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
133{
134 BUG_ON(!con_flag_valid(con_flag));
135
136 return test_bit(con_flag, &con->flags);
137}
138
139static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
141{
142 BUG_ON(!con_flag_valid(con_flag));
143
144 return test_and_clear_bit(con_flag, &con->flags);
145}
146
147static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
149{
150 BUG_ON(!con_flag_valid(con_flag));
151
152 return test_and_set_bit(con_flag, &con->flags);
153}
154
31b8006e
SW		 155/* static tag bytes (protocol control messages) */
156static char tag_msg = CEPH_MSGR_TAG_MSG;
157static char tag_ack = CEPH_MSGR_TAG_ACK;
158static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
159
a6a5349d
SW		 160#ifdef CONFIG_LOCKDEP
161static struct lock_class_key socket_class;
162#endif
163
84495f49
AE		 164/*
165 * When skipping (ignoring) a block of input we read it into a "skip
166 * buffer," which is this many bytes in size.
167 */
168#define SKIP_BUF_SIZE 1024
31b8006e
SW		 169
170static void queue_con(struct ceph_connection *con);
171static void con_work(struct work_struct *);
93209264 172static void con_fault(struct ceph_connection *con);
31b8006e 173
31b8006e 174/*
f64a9317
AE		 175 * Nicely render a sockaddr as a string. An array of formatted
176 * strings is used, to approximate reentrancy.
31b8006e 177 */
f64a9317
AE		 178#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
179#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
180#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
181#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
182
183static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
184static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 185
57666519 186static struct page *zero_page; /* used in certain error cases */
57666519 187
3d14c5d2 188const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 189{
190 int i;
191 char *s;
99f0f3b2
AE		 192 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
193 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 194
f64a9317 195 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 196 s = addr_str[i];
197
198 switch (ss->ss_family) {
199 case AF_INET:
bd406145
AE		 200 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
201 ntohs(in4->sin_port));
31b8006e
SW		 202 break;
203
204 case AF_INET6:
bd406145
AE		 205 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
206 ntohs(in6->sin6_port));
31b8006e
SW		 207 break;
208
209 default:
d3002b97
AE		 210 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
211 ss->ss_family);
31b8006e
SW		 212 }
213
214 return s;
215}
3d14c5d2 216EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 217
63f2d211
SW		 218static void encode_my_addr(struct ceph_messenger *msgr)
219{
220 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
221 ceph_encode_addr(&msgr->my_enc_addr);
222}
223
31b8006e
SW		 224/*
225 * work queue for all reading and writing to/from the socket.
226 */
e0f43c94 227static struct workqueue_struct *ceph_msgr_wq;
31b8006e 228
15417167 229static void _ceph_msgr_exit(void)
6173d1f0 230{
d3002b97 231 if (ceph_msgr_wq) {
6173d1f0 232 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 233 ceph_msgr_wq = NULL;
234 }
6173d1f0 235
6173d1f0
AE		 236 BUG_ON(zero_page == NULL);
237 kunmap(zero_page);
238 page_cache_release(zero_page);
239 zero_page = NULL;
240}
241
3d14c5d2 242int ceph_msgr_init(void)
31b8006e 243{
57666519
AE		 244 BUG_ON(zero_page != NULL);
245 zero_page = ZERO_PAGE(0);
246 page_cache_get(zero_page);
247
f363e45f 248 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
6173d1f0
AE		 249 if (ceph_msgr_wq)
250 return 0;
57666519 251
6173d1f0
AE		 252 pr_err("msgr_init failed to create workqueue\n");
253 _ceph_msgr_exit();
57666519 254
6173d1f0 255 return -ENOMEM;
31b8006e 256}
3d14c5d2 257EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 258
259void ceph_msgr_exit(void)
260{
57666519 261 BUG_ON(ceph_msgr_wq == NULL);
57666519 262
6173d1f0 263 _ceph_msgr_exit();
31b8006e 264}
3d14c5d2 265EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 266
cd84db6e 267void ceph_msgr_flush(void)
a922d38f
SW		 268{
269 flush_workqueue(ceph_msgr_wq);
270}
3d14c5d2 271EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 272
ce2c8903
AE		 273/* Connection socket state transition functions */
274
275static void con_sock_state_init(struct ceph_connection *con)
276{
277 int old_state;
278
279 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
280 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
281 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 282 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
283 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 284}
285
286static void con_sock_state_connecting(struct ceph_connection *con)
287{
288 int old_state;
289
290 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
291 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
292 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 293 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
294 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 295}
296
297static void con_sock_state_connected(struct ceph_connection *con)
298{
299 int old_state;
300
301 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
302 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
303 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 304 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
305 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 306}
307
308static void con_sock_state_closing(struct ceph_connection *con)
309{
310 int old_state;
311
312 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
313 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
314 old_state != CON_SOCK_STATE_CONNECTED &&
315 old_state != CON_SOCK_STATE_CLOSING))
316 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 317 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
318 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 319}
320
321static void con_sock_state_closed(struct ceph_connection *con)
322{
323 int old_state;
324
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 327 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 328 old_state != CON_SOCK_STATE_CONNECTING &&
329 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 330 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 331 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
332 CON_SOCK_STATE_CLOSED);
ce2c8903 333}
a922d38f 334
31b8006e
SW		 335/*
336 * socket callback functions
337 */
338
339/* data available on socket, or listen socket received a connect */
327800bd 340static void ceph_sock_data_ready(struct sock *sk, int count_unused)
31b8006e 341{
bd406145 342 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 343 if (atomic_read(&con->msgr->stopping)) {
344 return;
345 }
bd406145 346
31b8006e 347 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 348 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 349 con, con->state);
350 queue_con(con);
351 }
352}
353
354/* socket has buffer space for writing */
327800bd 355static void ceph_sock_write_space(struct sock *sk)
31b8006e 356{
d3002b97 357 struct ceph_connection *con = sk->sk_user_data;
31b8006e 358
182fac26
JS		 359 /* only queue to workqueue if there is data we want to write,
360 * and there is sufficient space in the socket buffer to accept
327800bd 361 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 362 * doesn't get called again until try_write() fills the socket
363 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
364 * and net/core/stream.c:sk_stream_write_space().
365 */
c9ffc77a 366 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
182fac26 367 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
327800bd 368 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 369 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
370 queue_con(con);
371 }
31b8006e 372 } else {
327800bd 373 dout("%s %p nothing to write\n", __func__, con);
31b8006e 374 }
31b8006e
SW		 375}
376
377/* socket's state has changed */
327800bd 378static void ceph_sock_state_change(struct sock *sk)
31b8006e 379{
bd406145 380 struct ceph_connection *con = sk->sk_user_data;
31b8006e 381
327800bd 382 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 383 con, con->state, sk->sk_state);
384
31b8006e
SW		 385 switch (sk->sk_state) {
386 case TCP_CLOSE:
327800bd 387 dout("%s TCP_CLOSE\n", __func__);
31b8006e 388 case TCP_CLOSE_WAIT:
327800bd 389 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 390 con_sock_state_closing(con);
c9ffc77a 391 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 392 queue_con(con);
31b8006e
SW		 393 break;
394 case TCP_ESTABLISHED:
327800bd 395 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 396 con_sock_state_connected(con);
31b8006e
SW		 397 queue_con(con);
398 break;
d3002b97
AE		 399 default: /* Everything else is uninteresting */
400 break;
31b8006e
SW		 401 }
402}
403
404/*
405 * set up socket callbacks
406 */
407static void set_sock_callbacks(struct socket *sock,
408 struct ceph_connection *con)
409{
410 struct sock *sk = sock->sk;
bd406145 411 sk->sk_user_data = con;
327800bd
AE		 412 sk->sk_data_ready = ceph_sock_data_ready;
413 sk->sk_write_space = ceph_sock_write_space;
414 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 415}
416
417
418/*
419 * socket helpers
420 */
421
422/*
423 * initiate connection to a remote socket.
424 */
41617d0c 425static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 426{
f91d3471 427 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e
SW		 428 struct socket *sock;
429 int ret;
430
431 BUG_ON(con->sock);
f91d3471
SW		 432 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
433 IPPROTO_TCP, &sock);
31b8006e 434 if (ret)
41617d0c 435 return ret;
31b8006e
SW		 436 sock->sk->sk_allocation = GFP_NOFS;
437
a6a5349d
SW		 438#ifdef CONFIG_LOCKDEP
439 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
440#endif
441
31b8006e
SW		 442 set_sock_callbacks(sock, con);
443
3d14c5d2 444 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 445
89a86be0 446 con_sock_state_connecting(con);
f91d3471
SW		 447 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
448 O_NONBLOCK);
31b8006e
SW		 449 if (ret == -EINPROGRESS) {
450 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 451 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 452 sock->sk->sk_state);
a5bc3129 453 } else if (ret < 0) {
31b8006e 454 pr_err("connect %s error %d\n",
3d14c5d2 455 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 456 sock_release(sock);
31b8006e 457 con->error_msg = "connect error";
31b8006e 458
41617d0c 459 return ret;
a5bc3129
AE		 460 }
461 con->sock = sock;
41617d0c 462 return 0;
31b8006e
SW		 463}
464
465static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
466{
467 struct kvec iov = {buf, len};
468 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 469 int r;
31b8006e 470
98bdb0aa
SW		 471 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
472 if (r == -EAGAIN)
473 r = 0;
474 return r;
31b8006e
SW		 475}
476
afb3d90e
AE		 477static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
478 int page_offset, size_t length)
479{
480 void *kaddr;
481 int ret;
482
483 BUG_ON(page_offset + length > PAGE_SIZE);
484
485 kaddr = kmap(page);
486 BUG_ON(!kaddr);
487 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
488 kunmap(page);
489
490 return ret;
491}
492
31b8006e
SW		 493/*
494 * write something. @more is true if caller will be sending more data
495 * shortly.
496 */
497static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
498 size_t kvlen, size_t len, int more)
499{
500 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 501 int r;
31b8006e
SW		 502
503 if (more)
504 msg.msg_flags |= MSG_MORE;
505 else
506 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
507
42961d23
SW		 508 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
509 if (r == -EAGAIN)
510 r = 0;
511 return r;
31b8006e
SW		 512}
513
31739139 514static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 515 int offset, size_t size, bool more)
31739139
AE		 516{
517 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
518 int ret;
519
520 ret = kernel_sendpage(sock, page, offset, size, flags);
521 if (ret == -EAGAIN)
522 ret = 0;
523
524 return ret;
525}
526
31b8006e
SW		 527
528/*
529 * Shutdown/close the socket for the given connection.
530 */
531static int con_close_socket(struct ceph_connection *con)
532{
8007b8d6 533 int rc = 0;
31b8006e
SW		 534
535 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 536 if (con->sock) {
537 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
538 sock_release(con->sock);
539 con->sock = NULL;
540 }
456ea468
AE		 541
542 /*
4a861692 543 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 544 * independent of the connection mutex, and we could have
545 * received a socket close event before we had the chance to
546 * shut the socket down.
547 */
c9ffc77a 548 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 549
ce2c8903 550 con_sock_state_closed(con);
31b8006e
SW		 551 return rc;
552}
553
554/*
555 * Reset a connection. Discard all incoming and outgoing messages
556 * and clear *_seq state.
557 */
558static void ceph_msg_remove(struct ceph_msg *msg)
559{
560 list_del_init(&msg->list_head);
38941f80 561 BUG_ON(msg->con == NULL);
36eb71aa 562 msg->con->ops->put(msg->con);
38941f80
AE		 563 msg->con = NULL;
564
31b8006e
SW		 565 ceph_msg_put(msg);
566}
567static void ceph_msg_remove_list(struct list_head *head)
568{
569 while (!list_empty(head)) {
570 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
571 list_head);
572 ceph_msg_remove(msg);
573 }
574}
575
576static void reset_connection(struct ceph_connection *con)
577{
578 /* reset connection, out_queue, msg_ and connect_seq */
579 /* discard existing out_queue and msg_seq */
0fa6ebc6 580 dout("reset_connection %p\n", con);
31b8006e
SW		 581 ceph_msg_remove_list(&con->out_queue);
582 ceph_msg_remove_list(&con->out_sent);
583
cf3e5c40 584 if (con->in_msg) {
38941f80
AE		 585 BUG_ON(con->in_msg->con != con);
586 con->in_msg->con = NULL;
cf3e5c40
SW		 587 ceph_msg_put(con->in_msg);
588 con->in_msg = NULL;
36eb71aa 589 con->ops->put(con);
cf3e5c40
SW		 590 }
591
31b8006e
SW		 592 con->connect_seq = 0;
593 con->out_seq = 0;
c86a2930
SW		 594 if (con->out_msg) {
595 ceph_msg_put(con->out_msg);
596 con->out_msg = NULL;
597 }
31b8006e 598 con->in_seq = 0;
0e0d5e0c 599 con->in_seq_acked = 0;
31b8006e
SW		 600}
601
602/*
603 * mark a peer down. drop any open connections.
604 */
605void ceph_con_close(struct ceph_connection *con)
606{
8c50c817 607 mutex_lock(&con->mutex);
3d14c5d2
YS		 608 dout("con_close %p peer %s\n", con,
609 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 610 con->state = CON_STATE_CLOSED;
a5988c49 611
c9ffc77a
AE		 612 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
613 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
614 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
615 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 616
31b8006e 617 reset_connection(con);
6f2bc3ff 618 con->peer_global_seq = 0;
91e45ce3 619 cancel_delayed_work(&con->work);
ee76e073 620 con_close_socket(con);
ec302645 621 mutex_unlock(&con->mutex);
31b8006e 622}
3d14c5d2 623EXPORT_SYMBOL(ceph_con_close);
31b8006e 624
31b8006e
SW		 625/*
626 * Reopen a closed connection, with a new peer address.
627 */
b7a9e5dd
SW		 628void ceph_con_open(struct ceph_connection *con,
629 __u8 entity_type, __u64 entity_num,
630 struct ceph_entity_addr *addr)
31b8006e 631{
5469155f 632 mutex_lock(&con->mutex);
3d14c5d2 633 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 634
122070a2 635 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 636 con->state = CON_STATE_PREOPEN;
a5988c49 637
b7a9e5dd
SW		 638 con->peer_name.type = (__u8) entity_type;
639 con->peer_name.num = cpu_to_le64(entity_num);
640
31b8006e 641 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 642 con->delay = 0; /* reset backoff memory */
5469155f 643 mutex_unlock(&con->mutex);
31b8006e
SW		 644 queue_con(con);
645}
3d14c5d2 646EXPORT_SYMBOL(ceph_con_open);
31b8006e 647
87b315a5
SW		 648/*
649 * return true if this connection ever successfully opened
650 */
651bool ceph_con_opened(struct ceph_connection *con)
652{
653 return con->connect_seq > 0;
654}
655
31b8006e
SW		 656/*
657 * initialize a new connection.
658 */
1bfd89f4
AE		 659void ceph_con_init(struct ceph_connection *con, void *private,
660 const struct ceph_connection_operations *ops,
b7a9e5dd 661 struct ceph_messenger *msgr)
31b8006e
SW		 662{
663 dout("con_init %p\n", con);
664 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 665 con->private = private;
666 con->ops = ops;
31b8006e 667 con->msgr = msgr;
ce2c8903
AE		 668
669 con_sock_state_init(con);
670
ec302645 671 mutex_init(&con->mutex);
31b8006e
SW		 672 INIT_LIST_HEAD(&con->out_queue);
673 INIT_LIST_HEAD(&con->out_sent);
674 INIT_DELAYED_WORK(&con->work, con_work);
a5988c49 675
8dacc7da 676 con->state = CON_STATE_CLOSED;
31b8006e 677}
3d14c5d2 678EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 679
680
681/*
682 * We maintain a global counter to order connection attempts. Get
683 * a unique seq greater than @gt.
684 */
685static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
686{
687 u32 ret;
688
689 spin_lock(&msgr->global_seq_lock);
690 if (msgr->global_seq < gt)
691 msgr->global_seq = gt;
692 ret = ++msgr->global_seq;
693 spin_unlock(&msgr->global_seq_lock);
694 return ret;
695}
696
e2200423 697static void con_out_kvec_reset(struct ceph_connection *con)
859eb799
AE		 698{
699 con->out_kvec_left = 0;
700 con->out_kvec_bytes = 0;
701 con->out_kvec_cur = &con->out_kvec[0];
702}
703
e2200423 704static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 705 size_t size, void *data)
706{
707 int index;
708
709 index = con->out_kvec_left;
710 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
711
712 con->out_kvec[index].iov_len = size;
713 con->out_kvec[index].iov_base = data;
714 con->out_kvec_left++;
715 con->out_kvec_bytes += size;
716}
31b8006e 717
df6ad1f9 718#ifdef CONFIG_BLOCK
6aaa4511
AE		 719
720/*
721 * For a bio data item, a piece is whatever remains of the next
722 * entry in the current bio iovec, or the first entry in the next
723 * bio in the list.
724 */
25aff7c5
AE		 725static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data *data,
726 size_t length)
6aaa4511
AE		 727{
728 struct ceph_msg_data_cursor *cursor = &data->cursor;
729 struct bio *bio;
730
731 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
732
733 bio = data->bio;
734 BUG_ON(!bio);
735 BUG_ON(!bio->bi_vcnt);
6aaa4511 736
25aff7c5 737 cursor->resid = length;
6aaa4511
AE		 738 cursor->bio = bio;
739 cursor->vector_index = 0;
740 cursor->vector_offset = 0;
25aff7c5 741 cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
6aaa4511
AE		 742}
743
744static struct page *ceph_msg_data_bio_next(struct ceph_msg_data *data,
745 size_t *page_offset,
746 size_t *length)
747{
748 struct ceph_msg_data_cursor *cursor = &data->cursor;
749 struct bio *bio;
750 struct bio_vec *bio_vec;
751 unsigned int index;
752
753 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
754
755 bio = cursor->bio;
756 BUG_ON(!bio);
757
758 index = cursor->vector_index;
759 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
760
761 bio_vec = &bio->bi_io_vec[index];
762 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
763 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
764 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 765 if (cursor->last_piece) /* pagelist offset is always 0 */
766 *length = cursor->resid;
767 else
768 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
25aff7c5 769 BUG_ON(*length > cursor->resid);
5df521b1 770 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511
AE		 771
772 return bio_vec->bv_page;
773}
774
775static bool ceph_msg_data_bio_advance(struct ceph_msg_data *data, size_t bytes)
776{
777 struct ceph_msg_data_cursor *cursor = &data->cursor;
778 struct bio *bio;
779 struct bio_vec *bio_vec;
780 unsigned int index;
781
782 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
783
784 bio = cursor->bio;
785 BUG_ON(!bio);
786
787 index = cursor->vector_index;
788 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
789 bio_vec = &bio->bi_io_vec[index];
6aaa4511
AE		 790
791 /* Advance the cursor offset */
792
25aff7c5
AE		 793 BUG_ON(cursor->resid < bytes);
794 cursor->resid -= bytes;
6aaa4511
AE		 795 cursor->vector_offset += bytes;
796 if (cursor->vector_offset < bio_vec->bv_len)
797 return false; /* more bytes to process in this segment */
25aff7c5 798 BUG_ON(cursor->vector_offset != bio_vec->bv_len);
6aaa4511
AE		 799
800 /* Move on to the next segment, and possibly the next bio */
801
25aff7c5 802 if (++index == (unsigned int) bio->bi_vcnt) {
6aaa4511 803 bio = bio->bi_next;
25aff7c5 804 index = 0;
6aaa4511 805 }
25aff7c5
AE		 806 cursor->bio = bio;
807 cursor->vector_index = index;
6aaa4511
AE		 808 cursor->vector_offset = 0;
809
25aff7c5
AE		 810 if (!cursor->last_piece) {
811 BUG_ON(!cursor->resid);
812 BUG_ON(!bio);
813 /* A short read is OK, so use <= rather than == */
814 if (cursor->resid <= bio->bi_io_vec[index].bv_len)
6aaa4511 815 cursor->last_piece = true;
25aff7c5 816 }
6aaa4511
AE		 817
818 return true;
819}
ea96571f 820#endif /* CONFIG_BLOCK */
df6ad1f9 821
e766d7b5
AE		 822/*
823 * For a page array, a piece comes from the first page in the array
824 * that has not already been fully consumed.
825 */
25aff7c5
AE		 826static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data *data,
827 size_t length)
e766d7b5
AE		 828{
829 struct ceph_msg_data_cursor *cursor = &data->cursor;
830 int page_count;
831
832 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
833
834 BUG_ON(!data->pages);
835 BUG_ON(!data->length);
1190bf06 836 BUG_ON(length > data->length); /* short reads are OK */
e766d7b5 837
25aff7c5 838 cursor->resid = length;
e766d7b5 839 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 840 cursor->page_offset = data->alignment & ~PAGE_MASK;
841 cursor->page_index = 0;
56fc5659
AE		 842 BUG_ON(page_count > (int)USHRT_MAX);
843 cursor->page_count = (unsigned short)page_count;
844 BUG_ON(length > SIZE_MAX - cursor->page_offset);
845 cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE;
e766d7b5
AE		 846}
847
848static struct page *ceph_msg_data_pages_next(struct ceph_msg_data *data,
849 size_t *page_offset,
850 size_t *length)
851{
852 struct ceph_msg_data_cursor *cursor = &data->cursor;
853
854 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
855
856 BUG_ON(cursor->page_index >= cursor->page_count);
857 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 858
859 *page_offset = cursor->page_offset;
25aff7c5 860 if (cursor->last_piece)
e766d7b5 861 *length = cursor->resid;
25aff7c5 862 else
e766d7b5 863 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 864
865 return data->pages[cursor->page_index];
866}
867
868static bool ceph_msg_data_pages_advance(struct ceph_msg_data *data,
869 size_t bytes)
870{
871 struct ceph_msg_data_cursor *cursor = &data->cursor;
872
873 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
874
875 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 876
877 /* Advance the cursor page offset */
878
879 cursor->resid -= bytes;
5df521b1
AE		 880 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
881 if (!bytes || cursor->page_offset)
e766d7b5
AE		 882 return false; /* more bytes to process in the current page */
883
5df521b1 884 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 885
886 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 887 cursor->page_index++;
25aff7c5 888 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 889
890 return true;
891}
892
fe38a2b6 893/*
dd236fcb
AE		 894 * For a pagelist, a piece is whatever remains to be consumed in the
895 * first page in the list, or the front of the next page.
fe38a2b6 896 */
25aff7c5
AE		 897static void ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data *data,
898 size_t length)
fe38a2b6
AE		 899{
900 struct ceph_msg_data_cursor *cursor = &data->cursor;
901 struct ceph_pagelist *pagelist;
902 struct page *page;
903
dd236fcb 904 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 905
906 pagelist = data->pagelist;
907 BUG_ON(!pagelist);
1190bf06 908 BUG_ON(length > pagelist->length); /* short reads are OK */
25aff7c5
AE		 909
910 if (!length)
fe38a2b6
AE		 911 return; /* pagelist can be assigned but empty */
912
913 BUG_ON(list_empty(&pagelist->head));
914 page = list_first_entry(&pagelist->head, struct page, lru);
915
25aff7c5 916 cursor->resid = length;
fe38a2b6
AE		 917 cursor->page = page;
918 cursor->offset = 0;
25aff7c5 919 cursor->last_piece = length <= PAGE_SIZE;
fe38a2b6
AE		 920}
921
dd236fcb 922static struct page *ceph_msg_data_pagelist_next(struct ceph_msg_data *data,
fe38a2b6 923 size_t *page_offset,
dd236fcb 924 size_t *length)
fe38a2b6
AE		 925{
926 struct ceph_msg_data_cursor *cursor = &data->cursor;
927 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 928
929 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
930
931 pagelist = data->pagelist;
932 BUG_ON(!pagelist);
933
934 BUG_ON(!cursor->page);
25aff7c5 935 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 936
5df521b1 937 /* offset of first page in pagelist is always 0 */
fe38a2b6 938 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 939 if (cursor->last_piece)
25aff7c5
AE		 940 *length = cursor->resid;
941 else
942 *length = PAGE_SIZE - *page_offset;
fe38a2b6
AE		 943
944 return data->cursor.page;
945}
946
dd236fcb
AE		 947static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data *data,
948 size_t bytes)
fe38a2b6
AE		 949{
950 struct ceph_msg_data_cursor *cursor = &data->cursor;
951 struct ceph_pagelist *pagelist;
952
953 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
954
955 pagelist = data->pagelist;
956 BUG_ON(!pagelist);
25aff7c5
AE		 957
958 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 959 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
960
961 /* Advance the cursor offset */
962
25aff7c5 963 cursor->resid -= bytes;
fe38a2b6 964 cursor->offset += bytes;
5df521b1 965 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 966 if (!bytes || cursor->offset & ~PAGE_MASK)
967 return false; /* more bytes to process in the current page */
968
969 /* Move on to the next page */
970
971 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
972 cursor->page = list_entry_next(cursor->page, lru);
25aff7c5 973 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 974
975 return true;
976}
977
dd236fcb
AE		 978/*
979 * Message data is handled (sent or received) in pieces, where each
980 * piece resides on a single page. The network layer might not
981 * consume an entire piece at once. A data item's cursor keeps
982 * track of which piece is next to process and how much remains to
983 * be processed in that piece. It also tracks whether the current
984 * piece is the last one in the data item.
985 */
25aff7c5
AE		 986static void ceph_msg_data_cursor_init(struct ceph_msg_data *data,
987 size_t length)
dd236fcb
AE		 988{
989 switch (data->type) {
990 case CEPH_MSG_DATA_PAGELIST:
25aff7c5 991 ceph_msg_data_pagelist_cursor_init(data, length);
dd236fcb 992 break;
e766d7b5 993 case CEPH_MSG_DATA_PAGES:
25aff7c5 994 ceph_msg_data_pages_cursor_init(data, length);
e766d7b5 995 break;
dd236fcb
AE		 996#ifdef CONFIG_BLOCK
997 case CEPH_MSG_DATA_BIO:
25aff7c5 998 ceph_msg_data_bio_cursor_init(data, length);
6aaa4511 999 break;
dd236fcb 1000#endif /* CONFIG_BLOCK */
6aaa4511 1001 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1002 default:
1003 /* BUG(); */
1004 break;
1005 }
f5db90bc 1006 data->cursor.need_crc = true;
dd236fcb
AE		 1007}
1008
1009/*
1010 * Return the page containing the next piece to process for a given
1011 * data item, and supply the page offset and length of that piece.
1012 * Indicate whether this is the last piece in this data item.
1013 */
1014static struct page *ceph_msg_data_next(struct ceph_msg_data *data,
1015 size_t *page_offset,
1016 size_t *length,
1017 bool *last_piece)
1018{
1019 struct page *page;
1020
1021 switch (data->type) {
1022 case CEPH_MSG_DATA_PAGELIST:
1023 page = ceph_msg_data_pagelist_next(data, page_offset, length);
1024 break;
e766d7b5
AE		 1025 case CEPH_MSG_DATA_PAGES:
1026 page = ceph_msg_data_pages_next(data, page_offset, length);
1027 break;
dd236fcb
AE		 1028#ifdef CONFIG_BLOCK
1029 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1030 page = ceph_msg_data_bio_next(data, page_offset, length);
1031 break;
dd236fcb 1032#endif /* CONFIG_BLOCK */
6aaa4511 1033 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1034 default:
1035 page = NULL;
1036 break;
1037 }
1038 BUG_ON(!page);
1039 BUG_ON(*page_offset + *length > PAGE_SIZE);
1040 BUG_ON(!*length);
1041 if (last_piece)
1042 *last_piece = data->cursor.last_piece;
1043
1044 return page;
1045}
1046
1047/*
1048 * Returns true if the result moves the cursor on to the next piece
1049 * of the data item.
1050 */
1051static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes)
1052{
25aff7c5 1053 struct ceph_msg_data_cursor *cursor = &data->cursor;
dd236fcb
AE		 1054 bool new_piece;
1055
25aff7c5 1056 BUG_ON(bytes > cursor->resid);
dd236fcb
AE		 1057 switch (data->type) {
1058 case CEPH_MSG_DATA_PAGELIST:
1059 new_piece = ceph_msg_data_pagelist_advance(data, bytes);
1060 break;
e766d7b5
AE		 1061 case CEPH_MSG_DATA_PAGES:
1062 new_piece = ceph_msg_data_pages_advance(data, bytes);
1063 break;
dd236fcb
AE		 1064#ifdef CONFIG_BLOCK
1065 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1066 new_piece = ceph_msg_data_bio_advance(data, bytes);
1067 break;
dd236fcb 1068#endif /* CONFIG_BLOCK */
6aaa4511 1069 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1070 default:
1071 BUG();
1072 break;
1073 }
f5db90bc 1074 data->cursor.need_crc = new_piece;
dd236fcb
AE		 1075
1076 return new_piece;
1077}
1078
98fa5dd8 1079static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1080{
739c905b 1081 BUG_ON(!msg);
25aff7c5 1082 BUG_ON(!data_len);
739c905b 1083
4c59b4a2 1084 /* Initialize data cursor */
fe38a2b6 1085
98fa5dd8 1086 ceph_msg_data_cursor_init(msg->data, (size_t)data_len);
739c905b
AE		 1087}
1088
31b8006e
SW		 1089/*
1090 * Prepare footer for currently outgoing message, and finish things
1091 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1092 */
859eb799 1093static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1094{
1095 struct ceph_msg *m = con->out_msg;
859eb799 1096 int v = con->out_kvec_left;
31b8006e 1097
fd154f3c
AE		 1098 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1099
31b8006e
SW		 1100 dout("prepare_write_message_footer %p\n", con);
1101 con->out_kvec_is_msg = true;
1102 con->out_kvec[v].iov_base = &m->footer;
1103 con->out_kvec[v].iov_len = sizeof(m->footer);
1104 con->out_kvec_bytes += sizeof(m->footer);
1105 con->out_kvec_left++;
1106 con->out_more = m->more_to_follow;
c86a2930 1107 con->out_msg_done = true;
31b8006e
SW		 1108}
1109
1110/*
1111 * Prepare headers for the next outgoing message.
1112 */
1113static void prepare_write_message(struct ceph_connection *con)
1114{
1115 struct ceph_msg *m;
a9a0c51a 1116 u32 crc;
31b8006e 1117
e2200423 1118 con_out_kvec_reset(con);
31b8006e 1119 con->out_kvec_is_msg = true;
c86a2930 1120 con->out_msg_done = false;
31b8006e
SW		 1121
1122 /* Sneak an ack in there first? If we can get it into the same
1123 * TCP packet that's a good thing. */
1124 if (con->in_seq > con->in_seq_acked) {
1125 con->in_seq_acked = con->in_seq;
e2200423 1126 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1127 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1128 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1129 &con->out_temp_ack);
31b8006e
SW		 1130 }
1131
38941f80 1132 BUG_ON(list_empty(&con->out_queue));
859eb799 1133 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1134 con->out_msg = m;
38941f80 1135 BUG_ON(m->con != con);
4cf9d544
SW		 1136
1137 /* put message on sent list */
1138 ceph_msg_get(m);
1139 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1140
e84346b7
SW		 1141 /*
1142 * only assign outgoing seq # if we haven't sent this message
1143 * yet. if it is requeued, resend with it's original seq.
1144 */
1145 if (m->needs_out_seq) {
1146 m->hdr.seq = cpu_to_le64(++con->out_seq);
1147 m->needs_out_seq = false;
1148 }
98fa5dd8 1149 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1150
98fa5dd8 1151 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1152 m, con->out_seq, le16_to_cpu(m->hdr.type),
1153 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1154 m->data_length);
31b8006e
SW		 1155 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1156
1157 /* tag + hdr + front + middle */
e2200423
AE		 1158 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1159 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1160 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1161
31b8006e 1162 if (m->middle)
e2200423 1163 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1164 m->middle->vec.iov_base);
31b8006e
SW		 1165
1166 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1167 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1168 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1169 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1170
1171 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1172 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1173 if (m->middle) {
1174 crc = crc32c(0, m->middle->vec.iov_base,
1175 m->middle->vec.iov_len);
1176 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1177 } else
31b8006e 1178 con->out_msg->footer.middle_crc = 0;
739c905b 1179 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1180 le32_to_cpu(con->out_msg->footer.front_crc),
1181 le32_to_cpu(con->out_msg->footer.middle_crc));
1182
1183 /* is there a data payload? */
739c905b 1184 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1185 if (m->data_length) {
1186 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1187 con->out_more = 1; /* data + footer will follow */
1188 } else {
31b8006e 1189 /* no, queue up footer too and be done */
859eb799 1190 prepare_write_message_footer(con);
78625051 1191 }
31b8006e 1192
c9ffc77a 1193 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1194}
1195
1196/*
1197 * Prepare an ack.
1198 */
1199static void prepare_write_ack(struct ceph_connection *con)
1200{
1201 dout("prepare_write_ack %p %llu -> %llu\n", con,
1202 con->in_seq_acked, con->in_seq);
1203 con->in_seq_acked = con->in_seq;
1204
e2200423 1205 con_out_kvec_reset(con);
859eb799 1206
e2200423 1207 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1208
31b8006e 1209 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1210 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1211 &con->out_temp_ack);
1212
31b8006e 1213 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1214 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1215}
1216
3a23083b
SW		 1217/*
1218 * Prepare to share the seq during handshake
1219 */
1220static void prepare_write_seq(struct ceph_connection *con)
1221{
1222 dout("prepare_write_seq %p %llu -> %llu\n", con,
1223 con->in_seq_acked, con->in_seq);
1224 con->in_seq_acked = con->in_seq;
1225
1226 con_out_kvec_reset(con);
1227
1228 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1229 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1230 &con->out_temp_ack);
1231
1232 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1233}
1234
31b8006e
SW		 1235/*
1236 * Prepare to write keepalive byte.
1237 */
1238static void prepare_write_keepalive(struct ceph_connection *con)
1239{
1240 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1241 con_out_kvec_reset(con);
1242 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1243 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1244}
1245
1246/*
1247 * Connection negotiation.
1248 */
1249
dac1e716
AE		 1250static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1251 int *auth_proto)
4e7a5dcd 1252{
a3530df3 1253 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1254
1255 if (!con->ops->get_authorizer) {
1256 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1257 con->out_connect.authorizer_len = 0;
729796be 1258 return NULL;
b1c6b980
AE		 1259 }
1260
1261 /* Can't hold the mutex while getting authorizer */
ec302645 1262 mutex_unlock(&con->mutex);
dac1e716 1263 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1264 mutex_lock(&con->mutex);
4e7a5dcd 1265
a3530df3 1266 if (IS_ERR(auth))
729796be 1267 return auth;
8dacc7da 1268 if (con->state != CON_STATE_NEGOTIATING)
729796be 1269 return ERR_PTR(-EAGAIN);
0da5d703 1270
8f43fb53
AE		 1271 con->auth_reply_buf = auth->authorizer_reply_buf;
1272 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1273 return auth;
4e7a5dcd
SW		 1274}
1275
31b8006e
SW		 1276/*
1277 * We connected to a peer and are saying hello.
1278 */
e825a66d 1279static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1280{
e2200423
AE		 1281 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1282 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1283 &con->msgr->my_enc_addr);
eed0ef2c 1284
eed0ef2c 1285 con->out_more = 0;
c9ffc77a 1286 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1287}
1288
e825a66d 1289static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1290{
95c96174 1291 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1292 int proto;
dac1e716 1293 int auth_proto;
729796be 1294 struct ceph_auth_handshake *auth;
31b8006e
SW		 1295
1296 switch (con->peer_name.type) {
1297 case CEPH_ENTITY_TYPE_MON:
1298 proto = CEPH_MONC_PROTOCOL;
1299 break;
1300 case CEPH_ENTITY_TYPE_OSD:
1301 proto = CEPH_OSDC_PROTOCOL;
1302 break;
1303 case CEPH_ENTITY_TYPE_MDS:
1304 proto = CEPH_MDSC_PROTOCOL;
1305 break;
1306 default:
1307 BUG();
1308 }
1309
1310 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1311 con->connect_seq, global_seq, proto);
4e7a5dcd 1312
e825a66d 1313 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1314 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1315 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1316 con->out_connect.global_seq = cpu_to_le32(global_seq);
1317 con->out_connect.protocol_version = cpu_to_le32(proto);
1318 con->out_connect.flags = 0;
31b8006e 1319
dac1e716
AE		 1320 auth_proto = CEPH_AUTH_UNKNOWN;
1321 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1322 if (IS_ERR(auth))
1323 return PTR_ERR(auth);
3da54776 1324
dac1e716 1325 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1326 con->out_connect.authorizer_len = auth ?
1327 cpu_to_le32(auth->authorizer_buf_len) : 0;
1328
e2200423 1329 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1330 &con->out_connect);
1331 if (auth && auth->authorizer_buf_len)
e2200423 1332 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1333 auth->authorizer_buf);
859eb799 1334
31b8006e 1335 con->out_more = 0;
c9ffc77a 1336 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1337
e10c758e 1338 return 0;
31b8006e
SW		 1339}
1340
31b8006e
SW		 1341/*
1342 * write as much of pending kvecs to the socket as we can.
1343 * 1 -> done
1344 * 0 -> socket full, but more to do
1345 * <0 -> error
1346 */
1347static int write_partial_kvec(struct ceph_connection *con)
1348{
1349 int ret;
1350
1351 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1352 while (con->out_kvec_bytes > 0) {
1353 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1354 con->out_kvec_left, con->out_kvec_bytes,
1355 con->out_more);
1356 if (ret <= 0)
1357 goto out;
1358 con->out_kvec_bytes -= ret;
1359 if (con->out_kvec_bytes == 0)
1360 break; /* done */
f42299e6
AE		 1361
1362 /* account for full iov entries consumed */
1363 while (ret >= con->out_kvec_cur->iov_len) {
1364 BUG_ON(!con->out_kvec_left);
1365 ret -= con->out_kvec_cur->iov_len;
1366 con->out_kvec_cur++;
1367 con->out_kvec_left--;
1368 }
1369 /* and for a partially-consumed entry */
1370 if (ret) {
1371 con->out_kvec_cur->iov_len -= ret;
1372 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1373 }
1374 }
1375 con->out_kvec_left = 0;
1376 con->out_kvec_is_msg = false;
1377 ret = 1;
1378out:
1379 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1380 con->out_kvec_bytes, con->out_kvec_left, ret);
1381 return ret; /* done! */
1382}
1383
35b62808
AE		 1384static u32 ceph_crc32c_page(u32 crc, struct page *page,
1385 unsigned int page_offset,
1386 unsigned int length)
1387{
1388 char *kaddr;
1389
1390 kaddr = kmap(page);
1391 BUG_ON(kaddr == NULL);
1392 crc = crc32c(crc, kaddr + page_offset, length);
1393 kunmap(page);
1394
1395 return crc;
1396}
31b8006e
SW		 1397/*
1398 * Write as much message data payload as we can. If we finish, queue
1399 * up the footer.
1400 * 1 -> done, footer is now queued in out_kvec[].
1401 * 0 -> socket full, but more to do
1402 * <0 -> error
1403 */
34d2d200 1404static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1405{
1406 struct ceph_msg *msg = con->out_msg;
6644ed7b 1407 struct ceph_msg_data_cursor *cursor = &msg->data->cursor;
37675b0f 1408 bool do_datacrc = !con->msgr->nocrc;
f5db90bc 1409 u32 crc;
31b8006e 1410
859a35d5 1411 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1412
6644ed7b 1413 if (WARN_ON(!msg->data))
4c59b4a2
AE		 1414 return -EINVAL;
1415
5821bd8c
AE		 1416 /*
1417 * Iterate through each page that contains data to be
1418 * written, and send as much as possible for each.
1419 *
1420 * If we are calculating the data crc (the default), we will
1421 * need to map the page. If we have no pages, they have
1422 * been revoked, so use the zero page.
1423 */
f5db90bc 1424 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1425 while (cursor->resid) {
8a166d05 1426 struct page *page;
e387d525
AE		 1427 size_t page_offset;
1428 size_t length;
8a166d05 1429 bool last_piece;
8ea299bc 1430 bool need_crc;
f5db90bc 1431 int ret;
68b4476b 1432
6644ed7b 1433 page = ceph_msg_data_next(msg->data, &page_offset, &length,
4c59b4a2 1434 &last_piece);
e387d525 1435 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1436 length, last_piece);
f5db90bc
AE		 1437 if (ret <= 0) {
1438 if (do_datacrc)
1439 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1440
f5db90bc
AE		 1441 return ret;
1442 }
143334ff
AE		 1443 if (do_datacrc && cursor->need_crc)
1444 crc = ceph_crc32c_page(crc, page, page_offset, length);
6644ed7b 1445 need_crc = ceph_msg_data_advance(msg->data, (size_t)ret);
31b8006e
SW		 1446 }
1447
34d2d200 1448 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1449
1450 /* prepare and queue up footer, too */
f5db90bc
AE		 1451 if (do_datacrc)
1452 msg->footer.data_crc = cpu_to_le32(crc);
1453 else
84ca8fc8 1454 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1455 con_out_kvec_reset(con);
859eb799 1456 prepare_write_message_footer(con);
f5db90bc
AE		 1457
1458 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1459}
1460
1461/*
1462 * write some zeros
1463 */
1464static int write_partial_skip(struct ceph_connection *con)
1465{
1466 int ret;
1467
1468 while (con->out_skip > 0) {
31739139 1469 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1470
e1dcb128 1471 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1472 if (ret <= 0)
1473 goto out;
1474 con->out_skip -= ret;
1475 }
1476 ret = 1;
1477out:
1478 return ret;
1479}
1480
1481/*
1482 * Prepare to read connection handshake, or an ack.
1483 */
eed0ef2c
SW		 1484static void prepare_read_banner(struct ceph_connection *con)
1485{
1486 dout("prepare_read_banner %p\n", con);
1487 con->in_base_pos = 0;
1488}
1489
31b8006e
SW		 1490static void prepare_read_connect(struct ceph_connection *con)
1491{
1492 dout("prepare_read_connect %p\n", con);
1493 con->in_base_pos = 0;
1494}
1495
1496static void prepare_read_ack(struct ceph_connection *con)
1497{
1498 dout("prepare_read_ack %p\n", con);
1499 con->in_base_pos = 0;
1500}
1501
3a23083b
SW		 1502static void prepare_read_seq(struct ceph_connection *con)
1503{
1504 dout("prepare_read_seq %p\n", con);
1505 con->in_base_pos = 0;
1506 con->in_tag = CEPH_MSGR_TAG_SEQ;
1507}
1508
31b8006e
SW		 1509static void prepare_read_tag(struct ceph_connection *con)
1510{
1511 dout("prepare_read_tag %p\n", con);
1512 con->in_base_pos = 0;
1513 con->in_tag = CEPH_MSGR_TAG_READY;
1514}
1515
1516/*
1517 * Prepare to read a message.
1518 */
1519static int prepare_read_message(struct ceph_connection *con)
1520{
1521 dout("prepare_read_message %p\n", con);
1522 BUG_ON(con->in_msg != NULL);
1523 con->in_base_pos = 0;
1524 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1525 return 0;
1526}
1527
1528
1529static int read_partial(struct ceph_connection *con,
fd51653f 1530 int end, int size, void *object)
31b8006e 1531{
e6cee71f
AE		 1532 while (con->in_base_pos < end) {
1533 int left = end - con->in_base_pos;
31b8006e
SW		 1534 int have = size - left;
1535 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1536 if (ret <= 0)
1537 return ret;
1538 con->in_base_pos += ret;
1539 }
1540 return 1;
1541}
1542
1543
1544/*
1545 * Read all or part of the connect-side handshake on a new connection
1546 */
eed0ef2c 1547static int read_partial_banner(struct ceph_connection *con)
31b8006e 1548{
fd51653f
AE		 1549 int size;
1550 int end;
1551 int ret;
31b8006e 1552
eed0ef2c 1553 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1554
1555 /* peer's banner */
fd51653f
AE		 1556 size = strlen(CEPH_BANNER);
1557 end = size;
1558 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1559 if (ret <= 0)
1560 goto out;
fd51653f
AE		 1561
1562 size = sizeof (con->actual_peer_addr);
1563 end += size;
1564 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1565 if (ret <= 0)
1566 goto out;
fd51653f
AE		 1567
1568 size = sizeof (con->peer_addr_for_me);
1569 end += size;
1570 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1571 if (ret <= 0)
1572 goto out;
fd51653f 1573
eed0ef2c
SW		 1574out:
1575 return ret;
1576}
1577
1578static int read_partial_connect(struct ceph_connection *con)
1579{
fd51653f
AE		 1580 int size;
1581 int end;
1582 int ret;
eed0ef2c
SW		 1583
1584 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1585
fd51653f
AE		 1586 size = sizeof (con->in_reply);
1587 end = size;
1588 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1589 if (ret <= 0)
1590 goto out;
fd51653f
AE		 1591
1592 size = le32_to_cpu(con->in_reply.authorizer_len);
1593 end += size;
1594 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1595 if (ret <= 0)
1596 goto out;
31b8006e 1597
4e7a5dcd
SW		 1598 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1599 con, (int)con->in_reply.tag,
1600 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1601 le32_to_cpu(con->in_reply.global_seq));
1602out:
1603 return ret;
eed0ef2c 1604
31b8006e
SW		 1605}
1606
1607/*
1608 * Verify the hello banner looks okay.
1609 */
1610static int verify_hello(struct ceph_connection *con)
1611{
1612 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1613 pr_err("connect to %s got bad banner\n",
3d14c5d2 1614 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1615 con->error_msg = "protocol error, bad banner";
1616 return -1;
1617 }
1618 return 0;
1619}
1620
1621static bool addr_is_blank(struct sockaddr_storage *ss)
1622{
1623 switch (ss->ss_family) {
1624 case AF_INET:
1625 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1626 case AF_INET6:
1627 return
1628 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1629 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1630 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1631 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1632 }
1633 return false;
1634}
1635
1636static int addr_port(struct sockaddr_storage *ss)
1637{
1638 switch (ss->ss_family) {
1639 case AF_INET:
f28bcfbe 1640 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1641 case AF_INET6:
f28bcfbe 1642 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1643 }
1644 return 0;
1645}
1646
1647static void addr_set_port(struct sockaddr_storage *ss, int p)
1648{
1649 switch (ss->ss_family) {
1650 case AF_INET:
1651 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1652 break;
31b8006e
SW		 1653 case AF_INET6:
1654 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1655 break;
31b8006e
SW		 1656 }
1657}
1658
ee3b56f2
NW		 1659/*
1660 * Unlike other *_pton function semantics, zero indicates success.
1661 */
1662static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1663 char delim, const char **ipend)
1664{
99f0f3b2
AE		 1665 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1666 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1667
1668 memset(ss, 0, sizeof(*ss));
1669
1670 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1671 ss->ss_family = AF_INET;
1672 return 0;
1673 }
1674
1675 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1676 ss->ss_family = AF_INET6;
1677 return 0;
1678 }
1679
1680 return -EINVAL;
1681}
1682
1683/*
1684 * Extract hostname string and resolve using kernel DNS facility.
1685 */
1686#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1687static int ceph_dns_resolve_name(const char *name, size_t namelen,
1688 struct sockaddr_storage *ss, char delim, const char **ipend)
1689{
1690 const char *end, *delim_p;
1691 char *colon_p, *ip_addr = NULL;
1692 int ip_len, ret;
1693
1694 /*
1695 * The end of the hostname occurs immediately preceding the delimiter or
1696 * the port marker (':') where the delimiter takes precedence.
1697 */
1698 delim_p = memchr(name, delim, namelen);
1699 colon_p = memchr(name, ':', namelen);
1700
1701 if (delim_p && colon_p)
1702 end = delim_p < colon_p ? delim_p : colon_p;
1703 else if (!delim_p && colon_p)
1704 end = colon_p;
1705 else {
1706 end = delim_p;
1707 if (!end) /* case: hostname:/ */
1708 end = name + namelen;
1709 }
1710
1711 if (end <= name)
1712 return -EINVAL;
1713
1714 /* do dns_resolve upcall */
1715 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1716 if (ip_len > 0)
1717 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1718 else
1719 ret = -ESRCH;
1720
1721 kfree(ip_addr);
1722
1723 *ipend = end;
1724
1725 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1726 ret, ret ? "failed" : ceph_pr_addr(ss));
1727
1728 return ret;
1729}
1730#else
1731static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1732 struct sockaddr_storage *ss, char delim, const char **ipend)
1733{
1734 return -EINVAL;
1735}
1736#endif
1737
1738/*
1739 * Parse a server name (IP or hostname). If a valid IP address is not found
1740 * then try to extract a hostname to resolve using userspace DNS upcall.
1741 */
1742static int ceph_parse_server_name(const char *name, size_t namelen,
1743 struct sockaddr_storage *ss, char delim, const char **ipend)
1744{
1745 int ret;
1746
1747 ret = ceph_pton(name, namelen, ss, delim, ipend);
1748 if (ret)
1749 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1750
1751 return ret;
1752}
1753
31b8006e
SW		 1754/*
1755 * Parse an ip[:port] list into an addr array. Use the default
1756 * monitor port if a port isn't specified.
1757 */
1758int ceph_parse_ips(const char *c, const char *end,
1759 struct ceph_entity_addr *addr,
1760 int max_count, int *count)
1761{
ee3b56f2 1762 int i, ret = -EINVAL;
31b8006e
SW		 1763 const char *p = c;
1764
1765 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1766 for (i = 0; i < max_count; i++) {
1767 const char *ipend;
1768 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1769 int port;
39139f64
SW		 1770 char delim = ',';
1771
1772 if (*p == '[') {
1773 delim = ']';
1774 p++;
1775 }
31b8006e 1776
ee3b56f2
NW		 1777 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1778 if (ret)
31b8006e 1779 goto bad;
ee3b56f2
NW		 1780 ret = -EINVAL;
1781
31b8006e
SW		 1782 p = ipend;
1783
39139f64
SW		 1784 if (delim == ']') {
1785 if (*p != ']') {
1786 dout("missing matching ']'\n");
1787 goto bad;
1788 }
1789 p++;
1790 }
1791
31b8006e
SW		 1792 /* port? */
1793 if (p < end && *p == ':') {
1794 port = 0;
1795 p++;
1796 while (p < end && *p >= '0' && *p <= '9') {
1797 port = (port * 10) + (*p - '0');
1798 p++;
1799 }
1800 if (port > 65535 || port == 0)
1801 goto bad;
1802 } else {
1803 port = CEPH_MON_PORT;
1804 }
1805
1806 addr_set_port(ss, port);
1807
3d14c5d2 1808 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1809
1810 if (p == end)
1811 break;
1812 if (*p != ',')
1813 goto bad;
1814 p++;
1815 }
1816
1817 if (p != end)
1818 goto bad;
1819
1820 if (count)
1821 *count = i + 1;
1822 return 0;
1823
1824bad:
39139f64 1825 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1826 return ret;
31b8006e 1827}
3d14c5d2 1828EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1829
eed0ef2c 1830static int process_banner(struct ceph_connection *con)
31b8006e 1831{
eed0ef2c 1832 dout("process_banner on %p\n", con);
31b8006e
SW		 1833
1834 if (verify_hello(con) < 0)
1835 return -1;
1836
63f2d211
SW		 1837 ceph_decode_addr(&con->actual_peer_addr);
1838 ceph_decode_addr(&con->peer_addr_for_me);
1839
31b8006e
SW		 1840 /*
1841 * Make sure the other end is who we wanted. note that the other
1842 * end may not yet know their ip address, so if it's 0.0.0.0, give
1843 * them the benefit of the doubt.
1844 */
103e2d3a
SW		 1845 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1846 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1847 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1848 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
cd84db6e 1849 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
3d14c5d2 1850 ceph_pr_addr(&con->peer_addr.in_addr),
cd84db6e 1851 (int)le32_to_cpu(con->peer_addr.nonce),
3d14c5d2 1852 ceph_pr_addr(&con->actual_peer_addr.in_addr),
cd84db6e 1853 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1854 con->error_msg = "wrong peer at address";
31b8006e
SW		 1855 return -1;
1856 }
1857
1858 /*
1859 * did we learn our address?
1860 */
1861 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1862 int port = addr_port(&con->msgr->inst.addr.in_addr);
1863
1864 memcpy(&con->msgr->inst.addr.in_addr,
1865 &con->peer_addr_for_me.in_addr,
1866 sizeof(con->peer_addr_for_me.in_addr));
1867 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1868 encode_my_addr(con->msgr);
eed0ef2c 1869 dout("process_banner learned my addr is %s\n",
3d14c5d2 1870 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1871 }
1872
eed0ef2c
SW		 1873 return 0;
1874}
1875
1876static int process_connect(struct ceph_connection *con)
1877{
3d14c5d2
YS		 1878 u64 sup_feat = con->msgr->supported_features;
1879 u64 req_feat = con->msgr->required_features;
04a419f9 1880 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 1881 int ret;
04a419f9 1882
eed0ef2c
SW		 1883 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1884
31b8006e 1885 switch (con->in_reply.tag) {
04a419f9
SW		 1886 case CEPH_MSGR_TAG_FEATURES:
1887 pr_err("%s%lld %s feature set mismatch,"
1888 " my %llx < server's %llx, missing %llx\n",
1889 ENTITY_NAME(con->peer_name),
3d14c5d2 1890 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1891 sup_feat, server_feat, server_feat & ~sup_feat);
1892 con->error_msg = "missing required protocol features";
0fa6ebc6 1893 reset_connection(con);
04a419f9
SW		 1894 return -1;
1895
31b8006e 1896 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 1897 pr_err("%s%lld %s protocol version mismatch,"
1898 " my %d != server's %d\n",
1899 ENTITY_NAME(con->peer_name),
3d14c5d2 1900 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 1901 le32_to_cpu(con->out_connect.protocol_version),
1902 le32_to_cpu(con->in_reply.protocol_version));
1903 con->error_msg = "protocol version mismatch";
0fa6ebc6 1904 reset_connection(con);
31b8006e
SW		 1905 return -1;
1906
4e7a5dcd
SW		 1907 case CEPH_MSGR_TAG_BADAUTHORIZER:
1908 con->auth_retry++;
1909 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1910 con->auth_retry);
1911 if (con->auth_retry == 2) {
1912 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 1913 return -1;
1914 }
6d4221b5 1915 con_out_kvec_reset(con);
e825a66d 1916 ret = prepare_write_connect(con);
0da5d703
SW		 1917 if (ret < 0)
1918 return ret;
63733a0f 1919 prepare_read_connect(con);
4e7a5dcd 1920 break;
31b8006e
SW		 1921
1922 case CEPH_MSGR_TAG_RESETSESSION:
1923 /*
1924 * If we connected with a large connect_seq but the peer
1925 * has no record of a session with us (no connection, or
1926 * connect_seq == 0), they will send RESETSESION to indicate
1927 * that they must have reset their session, and may have
1928 * dropped messages.
1929 */
1930 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 1931 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 1932 pr_err("%s%lld %s connection reset\n",
1933 ENTITY_NAME(con->peer_name),
3d14c5d2 1934 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 1935 reset_connection(con);
6d4221b5 1936 con_out_kvec_reset(con);
5a0f8fdd
AE		 1937 ret = prepare_write_connect(con);
1938 if (ret < 0)
1939 return ret;
31b8006e
SW		 1940 prepare_read_connect(con);
1941
1942 /* Tell ceph about it. */
ec302645 1943 mutex_unlock(&con->mutex);
31b8006e
SW		 1944 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1945 if (con->ops->peer_reset)
1946 con->ops->peer_reset(con);
ec302645 1947 mutex_lock(&con->mutex);
8dacc7da 1948 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 1949 return -EAGAIN;
31b8006e
SW		 1950 break;
1951
1952 case CEPH_MSGR_TAG_RETRY_SESSION:
1953 /*
1954 * If we sent a smaller connect_seq than the peer has, try
1955 * again with a larger value.
1956 */
5bdca4e0 1957 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 1958 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 1959 le32_to_cpu(con->in_reply.connect_seq));
1960 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 1961 con_out_kvec_reset(con);
5a0f8fdd
AE		 1962 ret = prepare_write_connect(con);
1963 if (ret < 0)
1964 return ret;
31b8006e
SW		 1965 prepare_read_connect(con);
1966 break;
1967
1968 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1969 /*
1970 * If we sent a smaller global_seq than the peer has, try
1971 * again with a larger value.
1972 */
eed0ef2c 1973 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 1974 con->peer_global_seq,
5bdca4e0 1975 le32_to_cpu(con->in_reply.global_seq));
31b8006e 1976 get_global_seq(con->msgr,
5bdca4e0 1977 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 1978 con_out_kvec_reset(con);
5a0f8fdd
AE		 1979 ret = prepare_write_connect(con);
1980 if (ret < 0)
1981 return ret;
31b8006e
SW		 1982 prepare_read_connect(con);
1983 break;
1984
3a23083b 1985 case CEPH_MSGR_TAG_SEQ:
31b8006e 1986 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 1987 if (req_feat & ~server_feat) {
1988 pr_err("%s%lld %s protocol feature mismatch,"
1989 " my required %llx > server's %llx, need %llx\n",
1990 ENTITY_NAME(con->peer_name),
3d14c5d2 1991 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1992 req_feat, server_feat, req_feat & ~server_feat);
1993 con->error_msg = "missing required protocol features";
0fa6ebc6 1994 reset_connection(con);
04a419f9
SW		 1995 return -1;
1996 }
8dacc7da 1997
122070a2 1998 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 1999 con->state = CON_STATE_OPEN;
20e55c4c 2000 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2001 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2002 con->connect_seq++;
aba558e2 2003 con->peer_features = server_feat;
31b8006e
SW		 2004 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2005 con->peer_global_seq,
2006 le32_to_cpu(con->in_reply.connect_seq),
2007 con->connect_seq);
2008 WARN_ON(con->connect_seq !=
2009 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2010
2011 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2012 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2013
85effe18 2014 con->delay = 0; /* reset backoff memory */
92ac41d0 2015
3a23083b
SW		 2016 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2017 prepare_write_seq(con);
2018 prepare_read_seq(con);
2019 } else {
2020 prepare_read_tag(con);
2021 }
31b8006e
SW		 2022 break;
2023
2024 case CEPH_MSGR_TAG_WAIT:
2025 /*
2026 * If there is a connection race (we are opening
2027 * connections to each other), one of us may just have
2028 * to WAIT. This shouldn't happen if we are the
2029 * client.
2030 */
04177882
SW		 2031 pr_err("process_connect got WAIT as client\n");
2032 con->error_msg = "protocol error, got WAIT as client";
2033 return -1;
31b8006e
SW		 2034
2035 default:
2036 pr_err("connect protocol error, will retry\n");
2037 con->error_msg = "protocol error, garbage tag during connect";
2038 return -1;
2039 }
2040 return 0;
2041}
2042
2043
2044/*
2045 * read (part of) an ack
2046 */
2047static int read_partial_ack(struct ceph_connection *con)
2048{
fd51653f
AE		 2049 int size = sizeof (con->in_temp_ack);
2050 int end = size;
31b8006e 2051
fd51653f 2052 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2053}
2054
31b8006e
SW		 2055/*
2056 * We can finally discard anything that's been acked.
2057 */
2058static void process_ack(struct ceph_connection *con)
2059{
2060 struct ceph_msg *m;
2061 u64 ack = le64_to_cpu(con->in_temp_ack);
2062 u64 seq;
2063
31b8006e
SW		 2064 while (!list_empty(&con->out_sent)) {
2065 m = list_first_entry(&con->out_sent, struct ceph_msg,
2066 list_head);
2067 seq = le64_to_cpu(m->hdr.seq);
2068 if (seq > ack)
2069 break;
2070 dout("got ack for seq %llu type %d at %p\n", seq,
2071 le16_to_cpu(m->hdr.type), m);
4cf9d544 2072 m->ack_stamp = jiffies;
31b8006e
SW		 2073 ceph_msg_remove(m);
2074 }
31b8006e
SW		 2075 prepare_read_tag(con);
2076}
2077
2078
2450418c 2079static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2080 struct kvec *section,
2081 unsigned int sec_len, u32 *crc)
2450418c 2082{
68b4476b 2083 int ret, left;
2450418c
YS		 2084
2085 BUG_ON(!section);
2086
2087 while (section->iov_len < sec_len) {
2088 BUG_ON(section->iov_base == NULL);
2089 left = sec_len - section->iov_len;
2090 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2091 section->iov_len, left);
2092 if (ret <= 0)
2093 return ret;
2094 section->iov_len += ret;
2450418c 2095 }
fe3ad593
AE		 2096 if (section->iov_len == sec_len)
2097 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2098
2450418c
YS		 2099 return 1;
2100}
31b8006e 2101
34d2d200
AE		 2102static int read_partial_msg_data(struct ceph_connection *con)
2103{
2104 struct ceph_msg *msg = con->in_msg;
6644ed7b 2105 struct ceph_msg_data_cursor *cursor = &msg->data->cursor;
34d2d200 2106 const bool do_datacrc = !con->msgr->nocrc;
686be208
AE		 2107 struct page *page;
2108 size_t page_offset;
2109 size_t length;
f5db90bc 2110 u32 crc = 0;
34d2d200
AE		 2111 int ret;
2112
2113 BUG_ON(!msg);
6644ed7b 2114 if (!msg->data)
4c59b4a2 2115 return -EIO;
34d2d200 2116
f5db90bc
AE		 2117 if (do_datacrc)
2118 crc = con->in_data_crc;
643c68a4 2119 while (cursor->resid) {
6644ed7b 2120 page = ceph_msg_data_next(msg->data, &page_offset, &length,
4c59b4a2 2121 NULL);
686be208 2122 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2123 if (ret <= 0) {
2124 if (do_datacrc)
2125 con->in_data_crc = crc;
2126
686be208 2127 return ret;
f5db90bc 2128 }
686be208
AE		 2129
2130 if (do_datacrc)
f5db90bc 2131 crc = ceph_crc32c_page(crc, page, page_offset, ret);
6644ed7b 2132 (void) ceph_msg_data_advance(msg->data, (size_t)ret);
34d2d200 2133 }
f5db90bc
AE		 2134 if (do_datacrc)
2135 con->in_data_crc = crc;
34d2d200
AE		 2136
2137 return 1; /* must return > 0 to indicate success */
2138}
2139
31b8006e
SW		 2140/*
2141 * read (part of) a message.
2142 */
686be208
AE		 2143static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2144
31b8006e
SW		 2145static int read_partial_message(struct ceph_connection *con)
2146{
2147 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2148 int size;
2149 int end;
31b8006e 2150 int ret;
95c96174 2151 unsigned int front_len, middle_len, data_len;
37675b0f 2152 bool do_datacrc = !con->msgr->nocrc;
ae18756b 2153 u64 seq;
fe3ad593 2154 u32 crc;
31b8006e
SW		 2155
2156 dout("read_partial_message con %p msg %p\n", con, m);
2157
2158 /* header */
fd51653f
AE		 2159 size = sizeof (con->in_hdr);
2160 end = size;
2161 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2162 if (ret <= 0)
2163 return ret;
fe3ad593
AE		 2164
2165 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2166 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2167 pr_err("read_partial_message bad hdr "
2168 " crc %u != expected %u\n",
2169 crc, con->in_hdr.crc);
2170 return -EBADMSG;
2171 }
2172
31b8006e
SW		 2173 front_len = le32_to_cpu(con->in_hdr.front_len);
2174 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2175 return -EIO;
2176 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2177 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2178 return -EIO;
2179 data_len = le32_to_cpu(con->in_hdr.data_len);
2180 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2181 return -EIO;
2182
ae18756b
SW		 2183 /* verify seq# */
2184 seq = le64_to_cpu(con->in_hdr.seq);
2185 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2186 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2187 ENTITY_NAME(con->peer_name),
3d14c5d2 2188 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2189 seq, con->in_seq + 1);
2190 con->in_base_pos = -front_len - middle_len - data_len -
2191 sizeof(m->footer);
2192 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2193 return 0;
2194 } else if ((s64)seq - (s64)con->in_seq > 1) {
2195 pr_err("read_partial_message bad seq %lld expected %lld\n",
2196 seq, con->in_seq + 1);
2197 con->error_msg = "bad message sequence # for incoming message";
2198 return -EBADMSG;
2199 }
2200
31b8006e
SW		 2201 /* allocate message? */
2202 if (!con->in_msg) {
4740a623
SW		 2203 int skip = 0;
2204
31b8006e 2205 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2206 front_len, data_len);
4740a623
SW		 2207 ret = ceph_con_in_msg_alloc(con, &skip);
2208 if (ret < 0)
2209 return ret;
f759ebb9
AE		 2210
2211 BUG_ON(!con->in_msg ^ skip);
2212 if (con->in_msg && data_len > con->in_msg->data_length) {
2213 pr_warning("%s skipping long message (%u > %zd)\n",
2214 __func__, data_len, con->in_msg->data_length);
2215 ceph_msg_put(con->in_msg);
2216 con->in_msg = NULL;
2217 skip = 1;
2218 }
2450418c 2219 if (skip) {
31b8006e 2220 /* skip this message */
a79832f2 2221 dout("alloc_msg said skip message\n");
31b8006e
SW		 2222 con->in_base_pos = -front_len - middle_len - data_len -
2223 sizeof(m->footer);
2224 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2225 con->in_seq++;
31b8006e
SW		 2226 return 0;
2227 }
38941f80 2228
4740a623 2229 BUG_ON(!con->in_msg);
38941f80 2230 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2231 m = con->in_msg;
2232 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2233 if (m->middle)
2234 m->middle->vec.iov_len = 0;
9d7f0f13 2235
78625051 2236 /* prepare for data payload, if any */
a4107026 2237
78625051 2238 if (data_len)
98fa5dd8 2239 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2240 }
2241
2242 /* front */
2450418c
YS		 2243 ret = read_partial_message_section(con, &m->front, front_len,
2244 &con->in_front_crc);
2245 if (ret <= 0)
2246 return ret;
31b8006e
SW		 2247
2248 /* middle */
2450418c 2249 if (m->middle) {
213c99ee
SW		 2250 ret = read_partial_message_section(con, &m->middle->vec,
2251 middle_len,
2450418c 2252 &con->in_middle_crc);
31b8006e
SW		 2253 if (ret <= 0)
2254 return ret;
31b8006e
SW		 2255 }
2256
2257 /* (page) data */
34d2d200
AE		 2258 if (data_len) {
2259 ret = read_partial_msg_data(con);
2260 if (ret <= 0)
2261 return ret;
31b8006e
SW		 2262 }
2263
31b8006e 2264 /* footer */
fd51653f
AE		 2265 size = sizeof (m->footer);
2266 end += size;
2267 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2268 if (ret <= 0)
2269 return ret;
2270
31b8006e
SW		 2271 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2272 m, front_len, m->footer.front_crc, middle_len,
2273 m->footer.middle_crc, data_len, m->footer.data_crc);
2274
2275 /* crc ok? */
2276 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2277 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2278 m, con->in_front_crc, m->footer.front_crc);
2279 return -EBADMSG;
2280 }
2281 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2282 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2283 m, con->in_middle_crc, m->footer.middle_crc);
2284 return -EBADMSG;
2285 }
bca064d2 2286 if (do_datacrc &&
31b8006e
SW		 2287 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2288 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2289 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2290 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2291 return -EBADMSG;
2292 }
2293
2294 return 1; /* done! */
2295}
2296
2297/*
2298 * Process message. This happens in the worker thread. The callback should
2299 * be careful not to do anything that waits on other incoming messages or it
2300 * may deadlock.
2301 */
2302static void process_message(struct ceph_connection *con)
2303{
5e095e8b 2304 struct ceph_msg *msg;
31b8006e 2305
38941f80
AE		 2306 BUG_ON(con->in_msg->con != con);
2307 con->in_msg->con = NULL;
5e095e8b 2308 msg = con->in_msg;
31b8006e 2309 con->in_msg = NULL;
36eb71aa 2310 con->ops->put(con);
31b8006e
SW		 2311
2312 /* if first message, set peer_name */
2313 if (con->peer_name.type == 0)
dbad185d 2314 con->peer_name = msg->hdr.src;
31b8006e 2315
31b8006e 2316 con->in_seq++;
ec302645 2317 mutex_unlock(&con->mutex);
31b8006e
SW		 2318
2319 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2320 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2321 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2322 le16_to_cpu(msg->hdr.type),
2323 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2324 le32_to_cpu(msg->hdr.front_len),
2325 le32_to_cpu(msg->hdr.data_len),
2326 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2327 con->ops->dispatch(con, msg);
ec302645
SW		 2328
2329 mutex_lock(&con->mutex);
31b8006e
SW		 2330}
2331
2332
2333/*
2334 * Write something to the socket. Called in a worker thread when the
2335 * socket appears to be writeable and we have something ready to send.
2336 */
2337static int try_write(struct ceph_connection *con)
2338{
31b8006e
SW		 2339 int ret = 1;
2340
d59315ca 2341 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2342
31b8006e
SW		 2343more:
2344 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2345
2346 /* open the socket first? */
8dacc7da
SW		 2347 if (con->state == CON_STATE_PREOPEN) {
2348 BUG_ON(con->sock);
2349 con->state = CON_STATE_CONNECTING;
a5988c49 2350
e2200423 2351 con_out_kvec_reset(con);
e825a66d 2352 prepare_write_banner(con);
eed0ef2c 2353 prepare_read_banner(con);
31b8006e 2354
cf3e5c40 2355 BUG_ON(con->in_msg);
31b8006e
SW		 2356 con->in_tag = CEPH_MSGR_TAG_READY;
2357 dout("try_write initiating connect on %p new state %lu\n",
2358 con, con->state);
41617d0c
AE		 2359 ret = ceph_tcp_connect(con);
2360 if (ret < 0) {
31b8006e 2361 con->error_msg = "connect error";
31b8006e
SW		 2362 goto out;
2363 }
2364 }
2365
2366more_kvec:
2367 /* kvec data queued? */
2368 if (con->out_skip) {
2369 ret = write_partial_skip(con);
2370 if (ret <= 0)
42961d23 2371 goto out;
31b8006e
SW		 2372 }
2373 if (con->out_kvec_left) {
2374 ret = write_partial_kvec(con);
2375 if (ret <= 0)
42961d23 2376 goto out;
31b8006e
SW		 2377 }
2378
2379 /* msg pages? */
2380 if (con->out_msg) {
c86a2930
SW		 2381 if (con->out_msg_done) {
2382 ceph_msg_put(con->out_msg);
2383 con->out_msg = NULL; /* we're done with this one */
2384 goto do_next;
2385 }
2386
34d2d200 2387 ret = write_partial_message_data(con);
31b8006e
SW		 2388 if (ret == 1)
2389 goto more_kvec; /* we need to send the footer, too! */
2390 if (ret == 0)
42961d23 2391 goto out;
31b8006e 2392 if (ret < 0) {
34d2d200 2393 dout("try_write write_partial_message_data err %d\n",
31b8006e 2394 ret);
42961d23 2395 goto out;
31b8006e
SW		 2396 }
2397 }
2398
c86a2930 2399do_next:
8dacc7da 2400 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2401 /* is anything else pending? */
2402 if (!list_empty(&con->out_queue)) {
2403 prepare_write_message(con);
2404 goto more;
2405 }
2406 if (con->in_seq > con->in_seq_acked) {
2407 prepare_write_ack(con);
2408 goto more;
2409 }
c9ffc77a 2410 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2411 prepare_write_keepalive(con);
2412 goto more;
2413 }
2414 }
2415
2416 /* Nothing to do! */
c9ffc77a 2417 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2418 dout("try_write nothing else to write.\n");
31b8006e
SW		 2419 ret = 0;
2420out:
42961d23 2421 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2422 return ret;
2423}
2424
2425
2426
2427/*
2428 * Read what we can from the socket.
2429 */
2430static int try_read(struct ceph_connection *con)
2431{
31b8006e
SW		 2432 int ret = -1;
2433
8dacc7da
SW		 2434more:
2435 dout("try_read start on %p state %lu\n", con, con->state);
2436 if (con->state != CON_STATE_CONNECTING &&
2437 con->state != CON_STATE_NEGOTIATING &&
2438 con->state != CON_STATE_OPEN)
31b8006e
SW		 2439 return 0;
2440
8dacc7da 2441 BUG_ON(!con->sock);
ec302645 2442
31b8006e
SW		 2443 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2444 con->in_base_pos);
0da5d703 2445
8dacc7da 2446 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2447 dout("try_read connecting\n");
2448 ret = read_partial_banner(con);
2449 if (ret <= 0)
ab166d5a 2450 goto out;
7593af92
AE		 2451 ret = process_banner(con);
2452 if (ret < 0)
2453 goto out;
2454
8dacc7da 2455 con->state = CON_STATE_NEGOTIATING;
7593af92 2456
6d4221b5
JS		 2457 /*
2458 * Received banner is good, exchange connection info.
2459 * Do not reset out_kvec, as sending our banner raced
2460 * with receiving peer banner after connect completed.
2461 */
7593af92
AE		 2462 ret = prepare_write_connect(con);
2463 if (ret < 0)
2464 goto out;
2465 prepare_read_connect(con);
2466
2467 /* Send connection info before awaiting response */
0da5d703
SW		 2468 goto out;
2469 }
2470
8dacc7da 2471 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2472 dout("try_read negotiating\n");
31b8006e
SW		 2473 ret = read_partial_connect(con);
2474 if (ret <= 0)
31b8006e 2475 goto out;
98bdb0aa
SW		 2476 ret = process_connect(con);
2477 if (ret < 0)
2478 goto out;
31b8006e
SW		 2479 goto more;
2480 }
2481
122070a2 2482 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2483
31b8006e
SW		 2484 if (con->in_base_pos < 0) {
2485 /*
2486 * skipping + discarding content.
2487 *
2488 * FIXME: there must be a better way to do this!
2489 */
84495f49
AE		 2490 static char buf[SKIP_BUF_SIZE];
2491 int skip = min((int) sizeof (buf), -con->in_base_pos);
2492
31b8006e
SW		 2493 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2494 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2495 if (ret <= 0)
98bdb0aa 2496 goto out;
31b8006e
SW		 2497 con->in_base_pos += ret;
2498 if (con->in_base_pos)
2499 goto more;
2500 }
2501 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2502 /*
2503 * what's next?
2504 */
2505 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2506 if (ret <= 0)
98bdb0aa 2507 goto out;
31b8006e
SW		 2508 dout("try_read got tag %d\n", (int)con->in_tag);
2509 switch (con->in_tag) {
2510 case CEPH_MSGR_TAG_MSG:
2511 prepare_read_message(con);
2512 break;
2513 case CEPH_MSGR_TAG_ACK:
2514 prepare_read_ack(con);
2515 break;
2516 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2517 con_close_socket(con);
2518 con->state = CON_STATE_CLOSED;
98bdb0aa 2519 goto out;
31b8006e
SW		 2520 default:
2521 goto bad_tag;
2522 }
2523 }
2524 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2525 ret = read_partial_message(con);
2526 if (ret <= 0) {
2527 switch (ret) {
2528 case -EBADMSG:
2529 con->error_msg = "bad crc";
2530 ret = -EIO;
98bdb0aa 2531 break;
31b8006e
SW		 2532 case -EIO:
2533 con->error_msg = "io error";
98bdb0aa 2534 break;
31b8006e 2535 }
98bdb0aa 2536 goto out;
31b8006e
SW		 2537 }
2538 if (con->in_tag == CEPH_MSGR_TAG_READY)
2539 goto more;
2540 process_message(con);
7b862e07
SW		 2541 if (con->state == CON_STATE_OPEN)
2542 prepare_read_tag(con);
31b8006e
SW		 2543 goto more;
2544 }
3a23083b
SW		 2545 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2546 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2547 /*
2548 * the final handshake seq exchange is semantically
2549 * equivalent to an ACK
2550 */
31b8006e
SW		 2551 ret = read_partial_ack(con);
2552 if (ret <= 0)
98bdb0aa 2553 goto out;
31b8006e
SW		 2554 process_ack(con);
2555 goto more;
2556 }
2557
31b8006e 2558out:
98bdb0aa 2559 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2560 return ret;
2561
2562bad_tag:
2563 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2564 con->error_msg = "protocol error, garbage tag";
2565 ret = -1;
2566 goto out;
2567}
2568
2569
2570/*
802c6d96
AE		 2571 * Atomically queue work on a connection after the specified delay.
2572 * Bump @con reference to avoid races with connection teardown.
2573 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2574 */
802c6d96 2575static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2576{
31b8006e 2577 if (!con->ops->get(con)) {
802c6d96
AE		 2578 dout("%s %p ref count 0\n", __func__, con);
2579
2580 return -ENOENT;
31b8006e
SW		 2581 }
2582
802c6d96
AE		 2583 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2584 dout("%s %p - already queued\n", __func__, con);
31b8006e 2585 con->ops->put(con);
802c6d96
AE		 2586
2587 return -EBUSY;
31b8006e 2588 }
802c6d96
AE		 2589
2590 dout("%s %p %lu\n", __func__, con, delay);
2591
2592 return 0;
2593}
2594
2595static void queue_con(struct ceph_connection *con)
2596{
2597 (void) queue_con_delay(con, 0);
31b8006e
SW		 2598}
2599
7bb21d68
AE		 2600static bool con_sock_closed(struct ceph_connection *con)
2601{
c9ffc77a 2602 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2603 return false;
2604
2605#define CASE(x) \
2606 case CON_STATE_ ## x: \
2607 con->error_msg = "socket closed (con state " #x ")"; \
2608 break;
2609
2610 switch (con->state) {
2611 CASE(CLOSED);
2612 CASE(PREOPEN);
2613 CASE(CONNECTING);
2614 CASE(NEGOTIATING);
2615 CASE(OPEN);
2616 CASE(STANDBY);
2617 default:
2618 pr_warning("%s con %p unrecognized state %lu\n",
2619 __func__, con, con->state);
2620 con->error_msg = "unrecognized con state";
2621 BUG();
2622 break;
2623 }
2624#undef CASE
2625
2626 return true;
2627}
2628
f20a39fd
AE		 2629static bool con_backoff(struct ceph_connection *con)
2630{
2631 int ret;
2632
2633 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2634 return false;
2635
2636 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2637 if (ret) {
2638 dout("%s: con %p FAILED to back off %lu\n", __func__,
2639 con, con->delay);
2640 BUG_ON(ret == -ENOENT);
2641 con_flag_set(con, CON_FLAG_BACKOFF);
2642 }
2643
2644 return true;
2645}
2646
93209264
AE		 2647/* Finish fault handling; con->mutex must *not* be held here */
2648
2649static void con_fault_finish(struct ceph_connection *con)
2650{
2651 /*
2652 * in case we faulted due to authentication, invalidate our
2653 * current tickets so that we can get new ones.
2654 */
2655 if (con->auth_retry && con->ops->invalidate_authorizer) {
2656 dout("calling invalidate_authorizer()\n");
2657 con->ops->invalidate_authorizer(con);
2658 }
2659
2660 if (con->ops->fault)
2661 con->ops->fault(con);
2662}
2663
31b8006e
SW		 2664/*
2665 * Do some work on a connection. Drop a connection ref when we're done.
2666 */
2667static void con_work(struct work_struct *work)
2668{
2669 struct ceph_connection *con = container_of(work, struct ceph_connection,
2670 work.work);
49659416 2671 bool fault;
31b8006e 2672
9dd4658d 2673 mutex_lock(&con->mutex);
49659416
AE		 2674 while (true) {
2675 int ret;
31b8006e 2676
49659416
AE		 2677 if ((fault = con_sock_closed(con))) {
2678 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2679 break;
2680 }
2681 if (con_backoff(con)) {
2682 dout("%s: con %p BACKOFF\n", __func__, con);
2683 break;
2684 }
2685 if (con->state == CON_STATE_STANDBY) {
2686 dout("%s: con %p STANDBY\n", __func__, con);
2687 break;
2688 }
2689 if (con->state == CON_STATE_CLOSED) {
2690 dout("%s: con %p CLOSED\n", __func__, con);
2691 BUG_ON(con->sock);
2692 break;
2693 }
2694 if (con->state == CON_STATE_PREOPEN) {
2695 dout("%s: con %p PREOPEN\n", __func__, con);
2696 BUG_ON(con->sock);
2697 }
0da5d703 2698
49659416
AE		 2699 ret = try_read(con);
2700 if (ret < 0) {
2701 if (ret == -EAGAIN)
2702 continue;
2703 con->error_msg = "socket error on read";
2704 fault = true;
2705 break;
2706 }
2707
2708 ret = try_write(con);
2709 if (ret < 0) {
2710 if (ret == -EAGAIN)
2711 continue;
2712 con->error_msg = "socket error on write";
2713 fault = true;
2714 }
2715
2716 break; /* If we make it to here, we're done */
3a140a0d 2717 }
b6e7b6a1
AE		 2718 if (fault)
2719 con_fault(con);
9dd4658d 2720 mutex_unlock(&con->mutex);
0da5d703 2721
b6e7b6a1
AE		 2722 if (fault)
2723 con_fault_finish(con);
2724
2725 con->ops->put(con);
31b8006e
SW		 2726}
2727
31b8006e
SW		 2728/*
2729 * Generic error/fault handler. A retry mechanism is used with
2730 * exponential backoff
2731 */
93209264 2732static void con_fault(struct ceph_connection *con)
31b8006e 2733{
28362986 2734 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
3d14c5d2 2735 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2736 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2737 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2738
122070a2 2739 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2740 con->state != CON_STATE_NEGOTIATING &&
2741 con->state != CON_STATE_OPEN);
ec302645 2742
31b8006e 2743 con_close_socket(con);
5e095e8b 2744
c9ffc77a 2745 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2746 dout("fault on LOSSYTX channel, marking CLOSED\n");
2747 con->state = CON_STATE_CLOSED;
93209264 2748 return;
3b5ede07
SW		 2749 }
2750
5e095e8b 2751 if (con->in_msg) {
38941f80
AE		 2752 BUG_ON(con->in_msg->con != con);
2753 con->in_msg->con = NULL;
5e095e8b
SW		 2754 ceph_msg_put(con->in_msg);
2755 con->in_msg = NULL;
36eb71aa 2756 con->ops->put(con);
5e095e8b 2757 }
31b8006e 2758
e80a52d1
SW		 2759 /* Requeue anything that hasn't been acked */
2760 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2761
e76661d0
SW		 2762 /* If there are no messages queued or keepalive pending, place
2763 * the connection in a STANDBY state */
2764 if (list_empty(&con->out_queue) &&
c9ffc77a 2765 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2766 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2767 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2768 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2769 } else {
2770 /* retry after a delay. */
8dacc7da 2771 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2772 if (con->delay == 0)
2773 con->delay = BASE_DELAY_INTERVAL;
2774 else if (con->delay < MAX_DELAY_INTERVAL)
2775 con->delay *= 2;
c9ffc77a 2776 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2777 queue_con(con);
31b8006e 2778 }
31b8006e
SW		 2779}
2780
2781
2782
2783/*
15d9882c 2784 * initialize a new messenger instance
31b8006e 2785 */
15d9882c
AE		 2786void ceph_messenger_init(struct ceph_messenger *msgr,
2787 struct ceph_entity_addr *myaddr,
2788 u32 supported_features,
2789 u32 required_features,
2790 bool nocrc)
31b8006e 2791{
3d14c5d2
YS		 2792 msgr->supported_features = supported_features;
2793 msgr->required_features = required_features;
2794
31b8006e
SW		 2795 spin_lock_init(&msgr->global_seq_lock);
2796
31b8006e
SW		 2797 if (myaddr)
2798 msgr->inst.addr = *myaddr;
2799
2800 /* select a random nonce */
ac8839d7 2801 msgr->inst.addr.type = 0;
103e2d3a 2802 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2803 encode_my_addr(msgr);
15d9882c 2804 msgr->nocrc = nocrc;
31b8006e 2805
a2a32584 2806 atomic_set(&msgr->stopping, 0);
31b8006e 2807
15d9882c 2808 dout("%s %p\n", __func__, msgr);
31b8006e 2809}
15d9882c 2810EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2811
e00de341
SW		 2812static void clear_standby(struct ceph_connection *con)
2813{
2814 /* come back from STANDBY? */
8dacc7da 2815 if (con->state == CON_STATE_STANDBY) {
e00de341 2816 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2817 con->state = CON_STATE_PREOPEN;
e00de341 2818 con->connect_seq++;
c9ffc77a
AE		 2819 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2820 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2821 }
2822}
2823
31b8006e
SW		 2824/*
2825 * Queue up an outgoing message on the given connection.
2826 */
2827void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2828{
31b8006e 2829 /* set src+dst */
dbad185d 2830 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2831 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2832 msg->needs_out_seq = true;
2833
ec302645 2834 mutex_lock(&con->mutex);
92ce034b 2835
8dacc7da 2836 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2837 dout("con_send %p closed, dropping %p\n", con, msg);
2838 ceph_msg_put(msg);
2839 mutex_unlock(&con->mutex);
2840 return;
2841 }
2842
38941f80 2843 BUG_ON(msg->con != NULL);
36eb71aa 2844 msg->con = con->ops->get(con);
92ce034b
AE		 2845 BUG_ON(msg->con == NULL);
2846
31b8006e
SW		 2847 BUG_ON(!list_empty(&msg->list_head));
2848 list_add_tail(&msg->list_head, &con->out_queue);
2849 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2850 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2851 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2852 le32_to_cpu(msg->hdr.front_len),
2853 le32_to_cpu(msg->hdr.middle_len),
2854 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 2855
2856 clear_standby(con);
ec302645 2857 mutex_unlock(&con->mutex);
31b8006e
SW		 2858
2859 /* if there wasn't anything waiting to send before, queue
2860 * new work */
c9ffc77a 2861 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2862 queue_con(con);
2863}
3d14c5d2 2864EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 2865
2866/*
2867 * Revoke a message that was previously queued for send
2868 */
6740a845 2869void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 2870{
6740a845
AE		 2871 struct ceph_connection *con = msg->con;
2872
2873 if (!con)
2874 return; /* Message not in our possession */
2875
ec302645 2876 mutex_lock(&con->mutex);
31b8006e 2877 if (!list_empty(&msg->list_head)) {
38941f80 2878 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 2879 list_del_init(&msg->list_head);
38941f80 2880 BUG_ON(msg->con == NULL);
36eb71aa 2881 msg->con->ops->put(msg->con);
38941f80 2882 msg->con = NULL;
31b8006e 2883 msg->hdr.seq = 0;
38941f80 2884
31b8006e 2885 ceph_msg_put(msg);
ed98adad
SW		 2886 }
2887 if (con->out_msg == msg) {
38941f80 2888 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 2889 con->out_msg = NULL;
31b8006e
SW		 2890 if (con->out_kvec_is_msg) {
2891 con->out_skip = con->out_kvec_bytes;
2892 con->out_kvec_is_msg = false;
2893 }
ed98adad 2894 msg->hdr.seq = 0;
92ce034b
AE		 2895
2896 ceph_msg_put(msg);
31b8006e 2897 }
ec302645 2898 mutex_unlock(&con->mutex);
31b8006e
SW		 2899}
2900
350b1c32 2901/*
0d59ab81 2902 * Revoke a message that we may be reading data into
350b1c32 2903 */
8921d114 2904void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 2905{
8921d114
AE		 2906 struct ceph_connection *con;
2907
2908 BUG_ON(msg == NULL);
2909 if (!msg->con) {
2910 dout("%s msg %p null con\n", __func__, msg);
2911
2912 return; /* Message not in our possession */
2913 }
2914
2915 con = msg->con;
350b1c32 2916 mutex_lock(&con->mutex);
8921d114 2917 if (con->in_msg == msg) {
95c96174
ED		 2918 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2919 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2920 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 2921
2922 /* skip rest of message */
8921d114
AE		 2923 dout("%s %p msg %p revoked\n", __func__, con, msg);
2924 con->in_base_pos = con->in_base_pos -
350b1c32 2925 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 2926 front_len -
2927 middle_len -
2928 data_len -
350b1c32 2929 sizeof(struct ceph_msg_footer);
350b1c32
SW		 2930 ceph_msg_put(con->in_msg);
2931 con->in_msg = NULL;
2932 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2933 con->in_seq++;
350b1c32 2934 } else {
8921d114
AE		 2935 dout("%s %p in_msg %p msg %p no-op\n",
2936 __func__, con, con->in_msg, msg);
350b1c32
SW		 2937 }
2938 mutex_unlock(&con->mutex);
2939}
2940
31b8006e
SW		 2941/*
2942 * Queue a keepalive byte to ensure the tcp connection is alive.
2943 */
2944void ceph_con_keepalive(struct ceph_connection *con)
2945{
e00de341 2946 dout("con_keepalive %p\n", con);
00650931 2947 mutex_lock(&con->mutex);
e00de341 2948 clear_standby(con);
00650931 2949 mutex_unlock(&con->mutex);
c9ffc77a
AE		 2950 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
2951 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2952 queue_con(con);
2953}
3d14c5d2 2954EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 2955
6644ed7b 2956static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 2957{
6644ed7b
AE		 2958 struct ceph_msg_data *data;
2959
2960 if (WARN_ON(!ceph_msg_data_type_valid(type)))
2961 return NULL;
2962
2963 data = kzalloc(sizeof (*data), GFP_NOFS);
2964 if (data)
2965 data->type = type;
2966
2967 return data;
2968}
2969
2970static void ceph_msg_data_destroy(struct ceph_msg_data *data)
2971{
2972 if (!data)
2973 return;
2974
2975 if (data->type == CEPH_MSG_DATA_PAGELIST) {
2976 ceph_pagelist_release(data->pagelist);
2977 kfree(data->pagelist);
2978 }
2979 kfree(data);
43794509
AE		 2980}
2981
02afca6c 2982void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 2983 size_t length, size_t alignment)
02afca6c 2984{
6644ed7b
AE		 2985 struct ceph_msg_data *data;
2986
07aa1558
AE		 2987 BUG_ON(!pages);
2988 BUG_ON(!length);
a1930804 2989 BUG_ON(msg->data_length);
6644ed7b
AE		 2990 BUG_ON(msg->data != NULL);
2991
2992 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
2993 BUG_ON(!data);
2994 data->pages = pages;
2995 data->length = length;
2996 data->alignment = alignment & ~PAGE_MASK;
02afca6c 2997
6644ed7b 2998 msg->data = data;
a1930804 2999 msg->data_length = length;
02afca6c
AE		 3000}
3001EXPORT_SYMBOL(ceph_msg_data_set_pages);
31b8006e 3002
27fa8385
AE		 3003void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3004 struct ceph_pagelist *pagelist)
3005{
6644ed7b
AE		 3006 struct ceph_msg_data *data;
3007
07aa1558
AE		 3008 BUG_ON(!pagelist);
3009 BUG_ON(!pagelist->length);
a1930804 3010 BUG_ON(msg->data_length);
6644ed7b 3011 BUG_ON(msg->data != NULL);
27fa8385 3012
6644ed7b
AE		 3013 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3014 BUG_ON(!data);
3015 data->pagelist = pagelist;
3016
3017 msg->data = data;
a1930804 3018 msg->data_length = pagelist->length;
27fa8385
AE		 3019}
3020EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3021
ea96571f 3022#ifdef CONFIG_BLOCK
a1930804
AE		 3023void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio,
3024 size_t length)
27fa8385 3025{
6644ed7b
AE		 3026 struct ceph_msg_data *data;
3027
07aa1558 3028 BUG_ON(!bio);
a1930804 3029 BUG_ON(msg->data_length);
6644ed7b 3030 BUG_ON(msg->data != NULL);
27fa8385 3031
6644ed7b
AE		 3032 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3033 BUG_ON(!data);
3034 data->bio = bio;
3035
3036 msg->data = data;
a1930804 3037 msg->data_length = length;
27fa8385
AE		 3038}
3039EXPORT_SYMBOL(ceph_msg_data_set_bio);
ea96571f 3040#endif /* CONFIG_BLOCK */
27fa8385 3041
31b8006e
SW		 3042/*
3043 * construct a new message with given type, size
3044 * the new msg has a ref count of 1.
3045 */
b61c2763
SW		 3046struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3047 bool can_fail)
31b8006e
SW		 3048{
3049 struct ceph_msg *m;
3050
9516e45b 3051 m = kzalloc(sizeof(*m), flags);
31b8006e
SW		 3052 if (m == NULL)
3053 goto out;
31b8006e
SW		 3054
3055 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3056 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3057 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3058
9516e45b
AE		 3059 INIT_LIST_HEAD(&m->list_head);
3060 kref_init(&m->kref);
ca20892d 3061
31b8006e 3062 /* front */
9516e45b 3063 m->front_max = front_len;
31b8006e
SW		 3064 if (front_len) {
3065 if (front_len > PAGE_CACHE_SIZE) {
34d23762 3066 m->front.iov_base = __vmalloc(front_len, flags,
31b8006e
SW		 3067 PAGE_KERNEL);
3068 m->front_is_vmalloc = true;
3069 } else {
34d23762 3070 m->front.iov_base = kmalloc(front_len, flags);
31b8006e
SW		 3071 }
3072 if (m->front.iov_base == NULL) {
b61c2763 3073 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3074 front_len);
3075 goto out2;
3076 }
3077 } else {
3078 m->front.iov_base = NULL;
3079 }
3080 m->front.iov_len = front_len;
3081
bb257664 3082 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3083 return m;
3084
3085out2:
3086 ceph_msg_put(m);
3087out:
b61c2763
SW		 3088 if (!can_fail) {
3089 pr_err("msg_new can't create type %d front %d\n", type,
3090 front_len);
f0ed1b7c 3091 WARN_ON(1);
b61c2763
SW		 3092 } else {
3093 dout("msg_new can't create type %d front %d\n", type,
3094 front_len);
3095 }
a79832f2 3096 return NULL;
31b8006e 3097}
3d14c5d2 3098EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3099
31b8006e
SW		 3100/*
3101 * Allocate "middle" portion of a message, if it is needed and wasn't
3102 * allocated by alloc_msg. This allows us to read a small fixed-size
3103 * per-type header in the front and then gracefully fail (i.e.,
3104 * propagate the error to the caller based on info in the front) when
3105 * the middle is too large.
3106 */
2450418c 3107static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3108{
3109 int type = le16_to_cpu(msg->hdr.type);
3110 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3111
3112 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3113 ceph_msg_type_name(type), middle_len);
3114 BUG_ON(!middle_len);
3115 BUG_ON(msg->middle);
3116
b6c1d5b8 3117 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3118 if (!msg->middle)
3119 return -ENOMEM;
3120 return 0;
3121}
3122
2450418c 3123/*
1c20f2d2
AE		 3124 * Allocate a message for receiving an incoming message on a
3125 * connection, and save the result in con->in_msg. Uses the
3126 * connection's private alloc_msg op if available.
3127 *
4740a623
SW		 3128 * Returns 0 on success, or a negative error code.
3129 *
3130 * On success, if we set *skip = 1:
3131 * - the next message should be skipped and ignored.
3132 * - con->in_msg == NULL
3133 * or if we set *skip = 0:
3134 * - con->in_msg is non-null.
3135 * On error (ENOMEM, EAGAIN, ...),
3136 * - con->in_msg == NULL
2450418c 3137 */
4740a623 3138static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3139{
4740a623 3140 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3141 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3142 struct ceph_msg *msg;
4740a623 3143 int ret = 0;
2450418c 3144
1c20f2d2 3145 BUG_ON(con->in_msg != NULL);
53ded495 3146 BUG_ON(!con->ops->alloc_msg);
2450418c 3147
53ded495
AE		 3148 mutex_unlock(&con->mutex);
3149 msg = con->ops->alloc_msg(con, hdr, skip);
3150 mutex_lock(&con->mutex);
3151 if (con->state != CON_STATE_OPEN) {
3152 if (msg)
1d866d1c 3153 ceph_msg_put(msg);
53ded495
AE		 3154 return -EAGAIN;
3155 }
4137577a
AE		 3156 if (msg) {
3157 BUG_ON(*skip);
3158 con->in_msg = msg;
36eb71aa 3159 con->in_msg->con = con->ops->get(con);
92ce034b 3160 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3161 } else {
3162 /*
3163 * Null message pointer means either we should skip
3164 * this message or we couldn't allocate memory. The
3165 * former is not an error.
3166 */
3167 if (*skip)
3168 return 0;
3169 con->error_msg = "error allocating memory for incoming message";
3170
53ded495 3171 return -ENOMEM;
2450418c 3172 }
1c20f2d2 3173 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3174
1c20f2d2
AE		 3175 if (middle_len && !con->in_msg->middle) {
3176 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3177 if (ret < 0) {
1c20f2d2
AE		 3178 ceph_msg_put(con->in_msg);
3179 con->in_msg = NULL;
2450418c
YS		 3180 }
3181 }
9d7f0f13 3182
4740a623 3183 return ret;
2450418c
YS		 3184}
3185
31b8006e
SW		 3186
3187/*
3188 * Free a generically kmalloc'd message.
3189 */
3190void ceph_msg_kfree(struct ceph_msg *m)
3191{
3192 dout("msg_kfree %p\n", m);
3193 if (m->front_is_vmalloc)
3194 vfree(m->front.iov_base);
3195 else
3196 kfree(m->front.iov_base);
3197 kfree(m);
3198}
3199
3200/*
3201 * Drop a msg ref. Destroy as needed.
3202 */
c2e552e7
SW		 3203void ceph_msg_last_put(struct kref *kref)
3204{
3205 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
31b8006e 3206
c2e552e7
SW		 3207 dout("ceph_msg_put last one on %p\n", m);
3208 WARN_ON(!list_empty(&m->list_head));
3209
3210 /* drop middle, data, if any */
3211 if (m->middle) {
3212 ceph_buffer_put(m->middle);
3213 m->middle = NULL;
31b8006e 3214 }
6644ed7b
AE		 3215 ceph_msg_data_destroy(m->data);
3216 m->data = NULL;
a1930804 3217 m->data_length = 0;
58bb3b37 3218
c2e552e7
SW		 3219 if (m->pool)
3220 ceph_msgpool_put(m->pool, m);
3221 else
3222 ceph_msg_kfree(m);
31b8006e 3223}
3d14c5d2 3224EXPORT_SYMBOL(ceph_msg_last_put);
9ec7cab1
SW		 3225
3226void ceph_msg_dump(struct ceph_msg *msg)
3227{
4a73ef27 3228 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
6644ed7b 3229 msg->front_max, msg->data->length);
9ec7cab1
SW		 3230 print_hex_dump(KERN_DEBUG, "header: ",
3231 DUMP_PREFIX_OFFSET, 16, 1,
3232 &msg->hdr, sizeof(msg->hdr), true);
3233 print_hex_dump(KERN_DEBUG, " front: ",
3234 DUMP_PREFIX_OFFSET, 16, 1,
3235 msg->front.iov_base, msg->front.iov_len, true);
3236 if (msg->middle)
3237 print_hex_dump(KERN_DEBUG, "middle: ",
3238 DUMP_PREFIX_OFFSET, 16, 1,
3239 msg->middle->vec.iov_base,
3240 msg->middle->vec.iov_len, true);
3241 print_hex_dump(KERN_DEBUG, "footer: ",
3242 DUMP_PREFIX_OFFSET, 16, 1,
3243 &msg->footer, sizeof(msg->footer), true);
3244}
3d14c5d2 3245EXPORT_SYMBOL(ceph_msg_dump);
Michael's Linux tree.