projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'v3.10.55' into update
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / ceph / messenger.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
5a0e3ad6 9#include <linux/slab.h>
31b8006e
SW		 10#include <linux/socket.h>
11#include <linux/string.h>
3ebc21f7 12#ifdef CONFIG_BLOCK
68b4476b 13#include <linux/bio.h>
3ebc21f7 14#endif /* CONFIG_BLOCK */
ee3b56f2 15#include <linux/dns_resolver.h>
31b8006e
SW		 16#include <net/tcp.h>
17
3d14c5d2
YS		 18#include <linux/ceph/libceph.h>
19#include <linux/ceph/messenger.h>
20#include <linux/ceph/decode.h>
21#include <linux/ceph/pagelist.h>
bc3b2d7f 22#include <linux/export.h>
31b8006e 23
fe38a2b6
AE		 24#define list_entry_next(pos, member) \
25 list_entry(pos->member.next, typeof(*pos), member)
26
31b8006e
SW		 27/*
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
33 * the sender.
34 */
35
bc18f4b1
AE		 36/*
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
40 * unexpected state.
41 *
42 * --------
43 * | NEW* | transient initial state
44 * --------
45 * | con_sock_state_init()
46 * v
47 * ----------
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
50 * ^ \
51 * | \ con_sock_state_connecting()
52 * | ----------------------
53 * | \
54 * + con_sock_state_closed() \
fbb85a47
SW		 55 * |+--------------------------- \
56 * | \ \ \
57 * | ----------- \ \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
60 * | ^ \ |
61 * | | \ |
62 * | + con_sock_state_closing() \ |
63 * | / \ | |
64 * | / --------------- | |
65 * | / \ v v
bc18f4b1
AE		 66 * | / --------------
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
70 * | | v
71 * -------------
72 * | CONNECTED | TCP connection established
73 * -------------
74 *
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
76 */
ce2c8903
AE		 77
78#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
83
8dacc7da
SW		 84/*
85 * connection states
86 */
87#define CON_STATE_CLOSED 1 /* -> PREOPEN */
88#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
93
4a861692
SW		 94/*
95 * ceph_connection flag bits
96 */
97#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 103
c9ffc77a
AE		 104static bool con_flag_valid(unsigned long con_flag)
105{
106 switch (con_flag) {
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
112 return true;
113 default:
114 return false;
115 }
116}
117
118static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
119{
120 BUG_ON(!con_flag_valid(con_flag));
121
122 clear_bit(con_flag, &con->flags);
123}
124
125static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
126{
127 BUG_ON(!con_flag_valid(con_flag));
128
129 set_bit(con_flag, &con->flags);
130}
131
132static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
133{
134 BUG_ON(!con_flag_valid(con_flag));
135
136 return test_bit(con_flag, &con->flags);
137}
138
139static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
141{
142 BUG_ON(!con_flag_valid(con_flag));
143
144 return test_and_clear_bit(con_flag, &con->flags);
145}
146
147static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
149{
150 BUG_ON(!con_flag_valid(con_flag));
151
152 return test_and_set_bit(con_flag, &con->flags);
153}
154
e3d5d638
AE		 155/* Slab caches for frequently-allocated structures */
156
157static struct kmem_cache *ceph_msg_cache;
81b36be4 158static struct kmem_cache *ceph_msg_data_cache;
e3d5d638 159
31b8006e
SW		 160/* static tag bytes (protocol control messages) */
161static char tag_msg = CEPH_MSGR_TAG_MSG;
162static char tag_ack = CEPH_MSGR_TAG_ACK;
163static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
164
a6a5349d
SW		 165#ifdef CONFIG_LOCKDEP
166static struct lock_class_key socket_class;
167#endif
168
84495f49
AE		 169/*
170 * When skipping (ignoring) a block of input we read it into a "skip
171 * buffer," which is this many bytes in size.
172 */
173#define SKIP_BUF_SIZE 1024
31b8006e
SW		 174
175static void queue_con(struct ceph_connection *con);
176static void con_work(struct work_struct *);
93209264 177static void con_fault(struct ceph_connection *con);
31b8006e 178
31b8006e 179/*
f64a9317
AE		 180 * Nicely render a sockaddr as a string. An array of formatted
181 * strings is used, to approximate reentrancy.
31b8006e 182 */
f64a9317
AE		 183#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
184#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
185#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
186#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
187
188static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
189static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 190
57666519 191static struct page *zero_page; /* used in certain error cases */
57666519 192
3d14c5d2 193const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 194{
195 int i;
196 char *s;
99f0f3b2
AE		 197 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
198 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 199
f64a9317 200 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 201 s = addr_str[i];
202
203 switch (ss->ss_family) {
204 case AF_INET:
bd406145
AE		 205 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
206 ntohs(in4->sin_port));
31b8006e
SW		 207 break;
208
209 case AF_INET6:
bd406145
AE		 210 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
211 ntohs(in6->sin6_port));
31b8006e
SW		 212 break;
213
214 default:
d3002b97
AE		 215 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
216 ss->ss_family);
31b8006e
SW		 217 }
218
219 return s;
220}
3d14c5d2 221EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 222
63f2d211
SW		 223static void encode_my_addr(struct ceph_messenger *msgr)
224{
225 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
226 ceph_encode_addr(&msgr->my_enc_addr);
227}
228
31b8006e
SW		 229/*
230 * work queue for all reading and writing to/from the socket.
231 */
e0f43c94 232static struct workqueue_struct *ceph_msgr_wq;
31b8006e 233
e3d5d638
AE		 234static int ceph_msgr_slab_init(void)
235{
236 BUG_ON(ceph_msg_cache);
237 ceph_msg_cache = kmem_cache_create("ceph_msg",
238 sizeof (struct ceph_msg),
239 __alignof__(struct ceph_msg), 0, NULL);
81b36be4
AE		 240
241 if (!ceph_msg_cache)
242 return -ENOMEM;
243
244 BUG_ON(ceph_msg_data_cache);
245 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
246 sizeof (struct ceph_msg_data),
247 __alignof__(struct ceph_msg_data),
248 0, NULL);
249 if (ceph_msg_data_cache)
250 return 0;
251
252 kmem_cache_destroy(ceph_msg_cache);
253 ceph_msg_cache = NULL;
254
255 return -ENOMEM;
e3d5d638
AE		 256}
257
258static void ceph_msgr_slab_exit(void)
259{
81b36be4
AE		 260 BUG_ON(!ceph_msg_data_cache);
261 kmem_cache_destroy(ceph_msg_data_cache);
262 ceph_msg_data_cache = NULL;
263
e3d5d638
AE		 264 BUG_ON(!ceph_msg_cache);
265 kmem_cache_destroy(ceph_msg_cache);
266 ceph_msg_cache = NULL;
267}
268
15417167 269static void _ceph_msgr_exit(void)
6173d1f0 270{
d3002b97 271 if (ceph_msgr_wq) {
6173d1f0 272 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 273 ceph_msgr_wq = NULL;
274 }
6173d1f0 275
e3d5d638
AE		 276 ceph_msgr_slab_exit();
277
6173d1f0
AE		 278 BUG_ON(zero_page == NULL);
279 kunmap(zero_page);
280 page_cache_release(zero_page);
281 zero_page = NULL;
282}
283
3d14c5d2 284int ceph_msgr_init(void)
31b8006e 285{
57666519
AE		 286 BUG_ON(zero_page != NULL);
287 zero_page = ZERO_PAGE(0);
288 page_cache_get(zero_page);
289
e3d5d638
AE		 290 if (ceph_msgr_slab_init())
291 return -ENOMEM;
292
f363e45f 293 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
6173d1f0
AE		 294 if (ceph_msgr_wq)
295 return 0;
57666519 296
6173d1f0
AE		 297 pr_err("msgr_init failed to create workqueue\n");
298 _ceph_msgr_exit();
57666519 299
6173d1f0 300 return -ENOMEM;
31b8006e 301}
3d14c5d2 302EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 303
304void ceph_msgr_exit(void)
305{
57666519 306 BUG_ON(ceph_msgr_wq == NULL);
57666519 307
6173d1f0 308 _ceph_msgr_exit();
31b8006e 309}
3d14c5d2 310EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 311
cd84db6e 312void ceph_msgr_flush(void)
a922d38f
SW		 313{
314 flush_workqueue(ceph_msgr_wq);
315}
3d14c5d2 316EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 317
ce2c8903
AE		 318/* Connection socket state transition functions */
319
320static void con_sock_state_init(struct ceph_connection *con)
321{
322 int old_state;
323
324 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
325 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
326 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 327 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
328 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 329}
330
331static void con_sock_state_connecting(struct ceph_connection *con)
332{
333 int old_state;
334
335 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
336 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
337 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 338 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
339 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 340}
341
342static void con_sock_state_connected(struct ceph_connection *con)
343{
344 int old_state;
345
346 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
347 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
348 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 349 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
350 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 351}
352
353static void con_sock_state_closing(struct ceph_connection *con)
354{
355 int old_state;
356
357 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
358 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
359 old_state != CON_SOCK_STATE_CONNECTED &&
360 old_state != CON_SOCK_STATE_CLOSING))
361 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 362 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
363 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 364}
365
366static void con_sock_state_closed(struct ceph_connection *con)
367{
368 int old_state;
369
370 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
371 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 372 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 373 old_state != CON_SOCK_STATE_CONNECTING &&
374 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 375 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 376 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
377 CON_SOCK_STATE_CLOSED);
ce2c8903 378}
a922d38f 379
31b8006e
SW		 380/*
381 * socket callback functions
382 */
383
384/* data available on socket, or listen socket received a connect */
327800bd 385static void ceph_sock_data_ready(struct sock *sk, int count_unused)
31b8006e 386{
bd406145 387 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 388 if (atomic_read(&con->msgr->stopping)) {
389 return;
390 }
bd406145 391
31b8006e 392 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 393 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 394 con, con->state);
395 queue_con(con);
396 }
397}
398
399/* socket has buffer space for writing */
327800bd 400static void ceph_sock_write_space(struct sock *sk)
31b8006e 401{
d3002b97 402 struct ceph_connection *con = sk->sk_user_data;
31b8006e 403
182fac26
JS		 404 /* only queue to workqueue if there is data we want to write,
405 * and there is sufficient space in the socket buffer to accept
327800bd 406 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 407 * doesn't get called again until try_write() fills the socket
408 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
409 * and net/core/stream.c:sk_stream_write_space().
410 */
c9ffc77a 411 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
182fac26 412 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
327800bd 413 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 414 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
415 queue_con(con);
416 }
31b8006e 417 } else {
327800bd 418 dout("%s %p nothing to write\n", __func__, con);
31b8006e 419 }
31b8006e
SW		 420}
421
422/* socket's state has changed */
327800bd 423static void ceph_sock_state_change(struct sock *sk)
31b8006e 424{
bd406145 425 struct ceph_connection *con = sk->sk_user_data;
31b8006e 426
327800bd 427 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 428 con, con->state, sk->sk_state);
429
31b8006e
SW		 430 switch (sk->sk_state) {
431 case TCP_CLOSE:
327800bd 432 dout("%s TCP_CLOSE\n", __func__);
31b8006e 433 case TCP_CLOSE_WAIT:
327800bd 434 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 435 con_sock_state_closing(con);
c9ffc77a 436 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 437 queue_con(con);
31b8006e
SW		 438 break;
439 case TCP_ESTABLISHED:
327800bd 440 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 441 con_sock_state_connected(con);
31b8006e
SW		 442 queue_con(con);
443 break;
d3002b97
AE		 444 default: /* Everything else is uninteresting */
445 break;
31b8006e
SW		 446 }
447}
448
449/*
450 * set up socket callbacks
451 */
452static void set_sock_callbacks(struct socket *sock,
453 struct ceph_connection *con)
454{
455 struct sock *sk = sock->sk;
bd406145 456 sk->sk_user_data = con;
327800bd
AE		 457 sk->sk_data_ready = ceph_sock_data_ready;
458 sk->sk_write_space = ceph_sock_write_space;
459 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 460}
461
462
463/*
464 * socket helpers
465 */
466
467/*
468 * initiate connection to a remote socket.
469 */
41617d0c 470static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 471{
f91d3471 472 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e
SW		 473 struct socket *sock;
474 int ret;
475
476 BUG_ON(con->sock);
f91d3471
SW		 477 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
478 IPPROTO_TCP, &sock);
31b8006e 479 if (ret)
41617d0c 480 return ret;
31b8006e
SW		 481 sock->sk->sk_allocation = GFP_NOFS;
482
a6a5349d
SW		 483#ifdef CONFIG_LOCKDEP
484 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
485#endif
486
31b8006e
SW		 487 set_sock_callbacks(sock, con);
488
3d14c5d2 489 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 490
89a86be0 491 con_sock_state_connecting(con);
f91d3471
SW		 492 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
493 O_NONBLOCK);
31b8006e
SW		 494 if (ret == -EINPROGRESS) {
495 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 496 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 497 sock->sk->sk_state);
a5bc3129 498 } else if (ret < 0) {
31b8006e 499 pr_err("connect %s error %d\n",
3d14c5d2 500 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 501 sock_release(sock);
31b8006e 502 con->error_msg = "connect error";
31b8006e 503
41617d0c 504 return ret;
a5bc3129
AE		 505 }
506 con->sock = sock;
41617d0c 507 return 0;
31b8006e
SW		 508}
509
510static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
511{
512 struct kvec iov = {buf, len};
513 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 514 int r;
31b8006e 515
98bdb0aa
SW		 516 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
517 if (r == -EAGAIN)
518 r = 0;
519 return r;
31b8006e
SW		 520}
521
afb3d90e
AE		 522static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
523 int page_offset, size_t length)
524{
525 void *kaddr;
526 int ret;
527
528 BUG_ON(page_offset + length > PAGE_SIZE);
529
530 kaddr = kmap(page);
531 BUG_ON(!kaddr);
532 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
533 kunmap(page);
534
535 return ret;
536}
537
31b8006e
SW		 538/*
539 * write something. @more is true if caller will be sending more data
540 * shortly.
541 */
542static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
543 size_t kvlen, size_t len, int more)
544{
545 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 546 int r;
31b8006e
SW		 547
548 if (more)
549 msg.msg_flags |= MSG_MORE;
550 else
551 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
552
42961d23
SW		 553 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
554 if (r == -EAGAIN)
555 r = 0;
556 return r;
31b8006e
SW		 557}
558
a757a4e2 559static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 560 int offset, size_t size, bool more)
31739139
AE		 561{
562 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
563 int ret;
564
565 ret = kernel_sendpage(sock, page, offset, size, flags);
566 if (ret == -EAGAIN)
567 ret = 0;
568
569 return ret;
570}
571
a757a4e2
CC		 572static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
573 int offset, size_t size, bool more)
574{
575 int ret;
576 struct kvec iov;
577
578 /* sendpage cannot properly handle pages with page_count == 0,
579 * we need to fallback to sendmsg if that's the case */
580 if (page_count(page) >= 1)
581 return __ceph_tcp_sendpage(sock, page, offset, size, more);
582
583 iov.iov_base = kmap(page) + offset;
584 iov.iov_len = size;
585 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
586 kunmap(page);
587
588 return ret;
589}
31b8006e
SW		 590
591/*
592 * Shutdown/close the socket for the given connection.
593 */
594static int con_close_socket(struct ceph_connection *con)
595{
8007b8d6 596 int rc = 0;
31b8006e
SW		 597
598 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 599 if (con->sock) {
600 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
601 sock_release(con->sock);
602 con->sock = NULL;
603 }
456ea468
AE		 604
605 /*
4a861692 606 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 607 * independent of the connection mutex, and we could have
608 * received a socket close event before we had the chance to
609 * shut the socket down.
610 */
c9ffc77a 611 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 612
ce2c8903 613 con_sock_state_closed(con);
31b8006e
SW		 614 return rc;
615}
616
617/*
618 * Reset a connection. Discard all incoming and outgoing messages
619 * and clear *_seq state.
620 */
621static void ceph_msg_remove(struct ceph_msg *msg)
622{
623 list_del_init(&msg->list_head);
38941f80 624 BUG_ON(msg->con == NULL);
36eb71aa 625 msg->con->ops->put(msg->con);
38941f80
AE		 626 msg->con = NULL;
627
31b8006e
SW		 628 ceph_msg_put(msg);
629}
630static void ceph_msg_remove_list(struct list_head *head)
631{
632 while (!list_empty(head)) {
633 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
634 list_head);
635 ceph_msg_remove(msg);
636 }
637}
638
639static void reset_connection(struct ceph_connection *con)
640{
641 /* reset connection, out_queue, msg_ and connect_seq */
642 /* discard existing out_queue and msg_seq */
0fa6ebc6 643 dout("reset_connection %p\n", con);
31b8006e
SW		 644 ceph_msg_remove_list(&con->out_queue);
645 ceph_msg_remove_list(&con->out_sent);
646
cf3e5c40 647 if (con->in_msg) {
38941f80
AE		 648 BUG_ON(con->in_msg->con != con);
649 con->in_msg->con = NULL;
cf3e5c40
SW		 650 ceph_msg_put(con->in_msg);
651 con->in_msg = NULL;
36eb71aa 652 con->ops->put(con);
cf3e5c40
SW		 653 }
654
31b8006e
SW		 655 con->connect_seq = 0;
656 con->out_seq = 0;
c86a2930
SW		 657 if (con->out_msg) {
658 ceph_msg_put(con->out_msg);
659 con->out_msg = NULL;
660 }
31b8006e 661 con->in_seq = 0;
0e0d5e0c 662 con->in_seq_acked = 0;
31b8006e
SW		 663}
664
665/*
666 * mark a peer down. drop any open connections.
667 */
668void ceph_con_close(struct ceph_connection *con)
669{
8c50c817 670 mutex_lock(&con->mutex);
3d14c5d2
YS		 671 dout("con_close %p peer %s\n", con,
672 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 673 con->state = CON_STATE_CLOSED;
a5988c49 674
c9ffc77a
AE		 675 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
676 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
677 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
678 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 679
31b8006e 680 reset_connection(con);
6f2bc3ff 681 con->peer_global_seq = 0;
91e45ce3 682 cancel_delayed_work(&con->work);
ee76e073 683 con_close_socket(con);
ec302645 684 mutex_unlock(&con->mutex);
31b8006e 685}
3d14c5d2 686EXPORT_SYMBOL(ceph_con_close);
31b8006e 687
31b8006e
SW		 688/*
689 * Reopen a closed connection, with a new peer address.
690 */
b7a9e5dd
SW		 691void ceph_con_open(struct ceph_connection *con,
692 __u8 entity_type, __u64 entity_num,
693 struct ceph_entity_addr *addr)
31b8006e 694{
5469155f 695 mutex_lock(&con->mutex);
3d14c5d2 696 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 697
122070a2 698 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 699 con->state = CON_STATE_PREOPEN;
a5988c49 700
b7a9e5dd
SW		 701 con->peer_name.type = (__u8) entity_type;
702 con->peer_name.num = cpu_to_le64(entity_num);
703
31b8006e 704 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 705 con->delay = 0; /* reset backoff memory */
5469155f 706 mutex_unlock(&con->mutex);
31b8006e
SW		 707 queue_con(con);
708}
3d14c5d2 709EXPORT_SYMBOL(ceph_con_open);
31b8006e 710
87b315a5
SW		 711/*
712 * return true if this connection ever successfully opened
713 */
714bool ceph_con_opened(struct ceph_connection *con)
715{
716 return con->connect_seq > 0;
717}
718
31b8006e
SW		 719/*
720 * initialize a new connection.
721 */
1bfd89f4
AE		 722void ceph_con_init(struct ceph_connection *con, void *private,
723 const struct ceph_connection_operations *ops,
b7a9e5dd 724 struct ceph_messenger *msgr)
31b8006e
SW		 725{
726 dout("con_init %p\n", con);
727 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 728 con->private = private;
729 con->ops = ops;
31b8006e 730 con->msgr = msgr;
ce2c8903
AE		 731
732 con_sock_state_init(con);
733
ec302645 734 mutex_init(&con->mutex);
31b8006e
SW		 735 INIT_LIST_HEAD(&con->out_queue);
736 INIT_LIST_HEAD(&con->out_sent);
737 INIT_DELAYED_WORK(&con->work, con_work);
a5988c49 738
8dacc7da 739 con->state = CON_STATE_CLOSED;
31b8006e 740}
3d14c5d2 741EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 742
743
744/*
745 * We maintain a global counter to order connection attempts. Get
746 * a unique seq greater than @gt.
747 */
748static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
749{
750 u32 ret;
751
752 spin_lock(&msgr->global_seq_lock);
753 if (msgr->global_seq < gt)
754 msgr->global_seq = gt;
755 ret = ++msgr->global_seq;
756 spin_unlock(&msgr->global_seq_lock);
757 return ret;
758}
759
e2200423 760static void con_out_kvec_reset(struct ceph_connection *con)
859eb799
AE		 761{
762 con->out_kvec_left = 0;
763 con->out_kvec_bytes = 0;
764 con->out_kvec_cur = &con->out_kvec[0];
765}
766
e2200423 767static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 768 size_t size, void *data)
769{
770 int index;
771
772 index = con->out_kvec_left;
773 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
774
775 con->out_kvec[index].iov_len = size;
776 con->out_kvec[index].iov_base = data;
777 con->out_kvec_left++;
778 con->out_kvec_bytes += size;
779}
31b8006e 780
df6ad1f9 781#ifdef CONFIG_BLOCK
6aaa4511
AE		 782
783/*
784 * For a bio data item, a piece is whatever remains of the next
785 * entry in the current bio iovec, or the first entry in the next
786 * bio in the list.
787 */
8ae4f4f5 788static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 789 size_t length)
6aaa4511 790{
8ae4f4f5 791 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 792 struct bio *bio;
793
794 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
795
796 bio = data->bio;
797 BUG_ON(!bio);
798 BUG_ON(!bio->bi_vcnt);
6aaa4511 799
ca8b3a69 800 cursor->resid = min(length, data->bio_length);
6aaa4511
AE		 801 cursor->bio = bio;
802 cursor->vector_index = 0;
803 cursor->vector_offset = 0;
25aff7c5 804 cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
6aaa4511
AE		 805}
806
8ae4f4f5 807static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 808 size_t *page_offset,
809 size_t *length)
810{
8ae4f4f5 811 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 812 struct bio *bio;
813 struct bio_vec *bio_vec;
814 unsigned int index;
815
816 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
817
818 bio = cursor->bio;
819 BUG_ON(!bio);
820
821 index = cursor->vector_index;
822 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
823
824 bio_vec = &bio->bi_io_vec[index];
825 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
826 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
827 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 828 if (cursor->last_piece) /* pagelist offset is always 0 */
829 *length = cursor->resid;
830 else
831 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
25aff7c5 832 BUG_ON(*length > cursor->resid);
5df521b1 833 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511
AE		 834
835 return bio_vec->bv_page;
836}
837
8ae4f4f5
AE		 838static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
839 size_t bytes)
6aaa4511 840{
6aaa4511
AE		 841 struct bio *bio;
842 struct bio_vec *bio_vec;
843 unsigned int index;
844
8ae4f4f5 845 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
6aaa4511
AE		 846
847 bio = cursor->bio;
848 BUG_ON(!bio);
849
850 index = cursor->vector_index;
851 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
852 bio_vec = &bio->bi_io_vec[index];
6aaa4511
AE		 853
854 /* Advance the cursor offset */
855
25aff7c5
AE		 856 BUG_ON(cursor->resid < bytes);
857 cursor->resid -= bytes;
6aaa4511
AE		 858 cursor->vector_offset += bytes;
859 if (cursor->vector_offset < bio_vec->bv_len)
860 return false; /* more bytes to process in this segment */
25aff7c5 861 BUG_ON(cursor->vector_offset != bio_vec->bv_len);
6aaa4511
AE		 862
863 /* Move on to the next segment, and possibly the next bio */
864
25aff7c5 865 if (++index == (unsigned int) bio->bi_vcnt) {
6aaa4511 866 bio = bio->bi_next;
25aff7c5 867 index = 0;
6aaa4511 868 }
25aff7c5
AE		 869 cursor->bio = bio;
870 cursor->vector_index = index;
6aaa4511
AE		 871 cursor->vector_offset = 0;
872
25aff7c5
AE		 873 if (!cursor->last_piece) {
874 BUG_ON(!cursor->resid);
875 BUG_ON(!bio);
876 /* A short read is OK, so use <= rather than == */
877 if (cursor->resid <= bio->bi_io_vec[index].bv_len)
6aaa4511 878 cursor->last_piece = true;
25aff7c5 879 }
6aaa4511
AE		 880
881 return true;
882}
ea96571f 883#endif /* CONFIG_BLOCK */
df6ad1f9 884
e766d7b5
AE		 885/*
886 * For a page array, a piece comes from the first page in the array
887 * that has not already been fully consumed.
888 */
8ae4f4f5 889static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 890 size_t length)
e766d7b5 891{
8ae4f4f5 892 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 893 int page_count;
894
895 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
896
897 BUG_ON(!data->pages);
898 BUG_ON(!data->length);
899
ca8b3a69 900 cursor->resid = min(length, data->length);
e766d7b5 901 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 902 cursor->page_offset = data->alignment & ~PAGE_MASK;
903 cursor->page_index = 0;
56fc5659
AE		 904 BUG_ON(page_count > (int)USHRT_MAX);
905 cursor->page_count = (unsigned short)page_count;
906 BUG_ON(length > SIZE_MAX - cursor->page_offset);
a6489727 907 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 908}
909
8ae4f4f5
AE		 910static struct page *
911ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
912 size_t *page_offset, size_t *length)
e766d7b5 913{
8ae4f4f5 914 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 915
916 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
917
918 BUG_ON(cursor->page_index >= cursor->page_count);
919 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 920
921 *page_offset = cursor->page_offset;
25aff7c5 922 if (cursor->last_piece)
e766d7b5 923 *length = cursor->resid;
25aff7c5 924 else
e766d7b5 925 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 926
927 return data->pages[cursor->page_index];
928}
929
8ae4f4f5 930static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 931 size_t bytes)
932{
8ae4f4f5 933 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 934
935 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 936
937 /* Advance the cursor page offset */
938
939 cursor->resid -= bytes;
5df521b1
AE		 940 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
941 if (!bytes || cursor->page_offset)
e766d7b5
AE		 942 return false; /* more bytes to process in the current page */
943
5df521b1 944 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 945
946 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 947 cursor->page_index++;
25aff7c5 948 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 949
950 return true;
951}
952
fe38a2b6 953/*
dd236fcb
AE		 954 * For a pagelist, a piece is whatever remains to be consumed in the
955 * first page in the list, or the front of the next page.
fe38a2b6 956 */
8ae4f4f5
AE		 957static void
958ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 959 size_t length)
fe38a2b6 960{
8ae4f4f5 961 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 962 struct ceph_pagelist *pagelist;
963 struct page *page;
964
dd236fcb 965 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 966
967 pagelist = data->pagelist;
968 BUG_ON(!pagelist);
25aff7c5
AE		 969
970 if (!length)
fe38a2b6
AE		 971 return; /* pagelist can be assigned but empty */
972
973 BUG_ON(list_empty(&pagelist->head));
974 page = list_first_entry(&pagelist->head, struct page, lru);
975
ca8b3a69 976 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 977 cursor->page = page;
978 cursor->offset = 0;
a51b272e 979 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 980}
981
8ae4f4f5
AE		 982static struct page *
983ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
984 size_t *page_offset, size_t *length)
fe38a2b6 985{
8ae4f4f5 986 struct ceph_msg_data *data = cursor->data;
fe38a2b6 987 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 988
989 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
990
991 pagelist = data->pagelist;
992 BUG_ON(!pagelist);
993
994 BUG_ON(!cursor->page);
25aff7c5 995 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 996
5df521b1 997 /* offset of first page in pagelist is always 0 */
fe38a2b6 998 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 999 if (cursor->last_piece)
25aff7c5
AE		 1000 *length = cursor->resid;
1001 else
1002 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1003
8ae4f4f5 1004 return cursor->page;
fe38a2b6
AE		 1005}
1006
8ae4f4f5 1007static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1008 size_t bytes)
fe38a2b6 1009{
8ae4f4f5 1010 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1011 struct ceph_pagelist *pagelist;
1012
1013 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1014
1015 pagelist = data->pagelist;
1016 BUG_ON(!pagelist);
25aff7c5
AE		 1017
1018 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1019 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1020
1021 /* Advance the cursor offset */
1022
25aff7c5 1023 cursor->resid -= bytes;
fe38a2b6 1024 cursor->offset += bytes;
5df521b1 1025 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1026 if (!bytes || cursor->offset & ~PAGE_MASK)
1027 return false; /* more bytes to process in the current page */
1028
1029 /* Move on to the next page */
1030
1031 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1032 cursor->page = list_entry_next(cursor->page, lru);
25aff7c5 1033 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1034
1035 return true;
1036}
1037
dd236fcb
AE		 1038/*
1039 * Message data is handled (sent or received) in pieces, where each
1040 * piece resides on a single page. The network layer might not
1041 * consume an entire piece at once. A data item's cursor keeps
1042 * track of which piece is next to process and how much remains to
1043 * be processed in that piece. It also tracks whether the current
1044 * piece is the last one in the data item.
1045 */
ca8b3a69 1046static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1047{
ca8b3a69 1048 size_t length = cursor->total_resid;
8ae4f4f5 1049
8ae4f4f5 1050 switch (cursor->data->type) {
dd236fcb 1051 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1052 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1053 break;
e766d7b5 1054 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1055 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1056 break;
dd236fcb
AE		 1057#ifdef CONFIG_BLOCK
1058 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1059 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1060 break;
dd236fcb 1061#endif /* CONFIG_BLOCK */
6aaa4511 1062 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1063 default:
1064 /* BUG(); */
1065 break;
1066 }
8ae4f4f5 1067 cursor->need_crc = true;
dd236fcb
AE		 1068}
1069
ca8b3a69
AE		 1070static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1071{
1072 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1073 struct ceph_msg_data *data;
1074
1075 BUG_ON(!length);
1076 BUG_ON(length > msg->data_length);
1077 BUG_ON(list_empty(&msg->data));
1078
ca8b3a69
AE		 1079 cursor->data_head = &msg->data;
1080 cursor->total_resid = length;
1081 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1082 cursor->data = data;
1083
1084 __ceph_msg_data_cursor_init(cursor);
1085}
1086
dd236fcb
AE		 1087/*
1088 * Return the page containing the next piece to process for a given
1089 * data item, and supply the page offset and length of that piece.
1090 * Indicate whether this is the last piece in this data item.
1091 */
8ae4f4f5
AE		 1092static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1093 size_t *page_offset, size_t *length,
dd236fcb
AE		 1094 bool *last_piece)
1095{
1096 struct page *page;
1097
8ae4f4f5 1098 switch (cursor->data->type) {
dd236fcb 1099 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1100 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1101 break;
e766d7b5 1102 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1103 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1104 break;
dd236fcb
AE		 1105#ifdef CONFIG_BLOCK
1106 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1107 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1108 break;
dd236fcb 1109#endif /* CONFIG_BLOCK */
6aaa4511 1110 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1111 default:
1112 page = NULL;
1113 break;
1114 }
1115 BUG_ON(!page);
1116 BUG_ON(*page_offset + *length > PAGE_SIZE);
1117 BUG_ON(!*length);
1118 if (last_piece)
8ae4f4f5 1119 *last_piece = cursor->last_piece;
dd236fcb
AE		 1120
1121 return page;
1122}
1123
1124/*
1125 * Returns true if the result moves the cursor on to the next piece
1126 * of the data item.
1127 */
8ae4f4f5
AE		 1128static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1129 size_t bytes)
dd236fcb
AE		 1130{
1131 bool new_piece;
1132
25aff7c5 1133 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1134 switch (cursor->data->type) {
dd236fcb 1135 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1136 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1137 break;
e766d7b5 1138 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1139 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1140 break;
dd236fcb
AE		 1141#ifdef CONFIG_BLOCK
1142 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1143 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1144 break;
dd236fcb 1145#endif /* CONFIG_BLOCK */
6aaa4511 1146 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1147 default:
1148 BUG();
1149 break;
1150 }
ca8b3a69 1151 cursor->total_resid -= bytes;
dd236fcb 1152
ca8b3a69
AE		 1153 if (!cursor->resid && cursor->total_resid) {
1154 WARN_ON(!cursor->last_piece);
1155 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1156 cursor->data = list_entry_next(cursor->data, links);
1157 __ceph_msg_data_cursor_init(cursor);
a51b272e 1158 new_piece = true;
ca8b3a69 1159 }
a51b272e 1160 cursor->need_crc = new_piece;
ca8b3a69 1161
dd236fcb
AE		 1162 return new_piece;
1163}
1164
98fa5dd8 1165static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1166{
739c905b 1167 BUG_ON(!msg);
25aff7c5 1168 BUG_ON(!data_len);
739c905b 1169
4c59b4a2 1170 /* Initialize data cursor */
fe38a2b6 1171
8ae4f4f5 1172 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1173}
1174
31b8006e
SW		 1175/*
1176 * Prepare footer for currently outgoing message, and finish things
1177 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1178 */
859eb799 1179static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1180{
1181 struct ceph_msg *m = con->out_msg;
859eb799 1182 int v = con->out_kvec_left;
31b8006e 1183
fd154f3c
AE		 1184 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1185
31b8006e
SW		 1186 dout("prepare_write_message_footer %p\n", con);
1187 con->out_kvec_is_msg = true;
1188 con->out_kvec[v].iov_base = &m->footer;
1189 con->out_kvec[v].iov_len = sizeof(m->footer);
1190 con->out_kvec_bytes += sizeof(m->footer);
1191 con->out_kvec_left++;
1192 con->out_more = m->more_to_follow;
c86a2930 1193 con->out_msg_done = true;
31b8006e
SW		 1194}
1195
1196/*
1197 * Prepare headers for the next outgoing message.
1198 */
1199static void prepare_write_message(struct ceph_connection *con)
1200{
1201 struct ceph_msg *m;
a9a0c51a 1202 u32 crc;
31b8006e 1203
e2200423 1204 con_out_kvec_reset(con);
31b8006e 1205 con->out_kvec_is_msg = true;
c86a2930 1206 con->out_msg_done = false;
31b8006e
SW		 1207
1208 /* Sneak an ack in there first? If we can get it into the same
1209 * TCP packet that's a good thing. */
1210 if (con->in_seq > con->in_seq_acked) {
1211 con->in_seq_acked = con->in_seq;
e2200423 1212 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1213 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1214 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1215 &con->out_temp_ack);
31b8006e
SW		 1216 }
1217
38941f80 1218 BUG_ON(list_empty(&con->out_queue));
859eb799 1219 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1220 con->out_msg = m;
38941f80 1221 BUG_ON(m->con != con);
4cf9d544
SW		 1222
1223 /* put message on sent list */
1224 ceph_msg_get(m);
1225 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1226
e84346b7
SW		 1227 /*
1228 * only assign outgoing seq # if we haven't sent this message
1229 * yet. if it is requeued, resend with it's original seq.
1230 */
1231 if (m->needs_out_seq) {
1232 m->hdr.seq = cpu_to_le64(++con->out_seq);
1233 m->needs_out_seq = false;
1234 }
98fa5dd8 1235 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1236
98fa5dd8 1237 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1238 m, con->out_seq, le16_to_cpu(m->hdr.type),
1239 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1240 m->data_length);
31b8006e
SW		 1241 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1242
1243 /* tag + hdr + front + middle */
e2200423
AE		 1244 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1245 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1246 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1247
31b8006e 1248 if (m->middle)
e2200423 1249 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1250 m->middle->vec.iov_base);
31b8006e
SW		 1251
1252 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1253 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1254 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1255 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1256
1257 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1258 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1259 if (m->middle) {
1260 crc = crc32c(0, m->middle->vec.iov_base,
1261 m->middle->vec.iov_len);
1262 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1263 } else
31b8006e 1264 con->out_msg->footer.middle_crc = 0;
739c905b 1265 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1266 le32_to_cpu(con->out_msg->footer.front_crc),
1267 le32_to_cpu(con->out_msg->footer.middle_crc));
1268
1269 /* is there a data payload? */
739c905b 1270 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1271 if (m->data_length) {
1272 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1273 con->out_more = 1; /* data + footer will follow */
1274 } else {
31b8006e 1275 /* no, queue up footer too and be done */
859eb799 1276 prepare_write_message_footer(con);
78625051 1277 }
31b8006e 1278
c9ffc77a 1279 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1280}
1281
1282/*
1283 * Prepare an ack.
1284 */
1285static void prepare_write_ack(struct ceph_connection *con)
1286{
1287 dout("prepare_write_ack %p %llu -> %llu\n", con,
1288 con->in_seq_acked, con->in_seq);
1289 con->in_seq_acked = con->in_seq;
1290
e2200423 1291 con_out_kvec_reset(con);
859eb799 1292
e2200423 1293 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1294
31b8006e 1295 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1296 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1297 &con->out_temp_ack);
1298
31b8006e 1299 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1300 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1301}
1302
3a23083b
SW		 1303/*
1304 * Prepare to share the seq during handshake
1305 */
1306static void prepare_write_seq(struct ceph_connection *con)
1307{
1308 dout("prepare_write_seq %p %llu -> %llu\n", con,
1309 con->in_seq_acked, con->in_seq);
1310 con->in_seq_acked = con->in_seq;
1311
1312 con_out_kvec_reset(con);
1313
1314 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1315 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1316 &con->out_temp_ack);
1317
1318 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1319}
1320
31b8006e
SW		 1321/*
1322 * Prepare to write keepalive byte.
1323 */
1324static void prepare_write_keepalive(struct ceph_connection *con)
1325{
1326 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1327 con_out_kvec_reset(con);
1328 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1329 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1330}
1331
1332/*
1333 * Connection negotiation.
1334 */
1335
dac1e716
AE		 1336static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1337 int *auth_proto)
4e7a5dcd 1338{
a3530df3 1339 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1340
1341 if (!con->ops->get_authorizer) {
1342 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1343 con->out_connect.authorizer_len = 0;
729796be 1344 return NULL;
b1c6b980
AE		 1345 }
1346
1347 /* Can't hold the mutex while getting authorizer */
ec302645 1348 mutex_unlock(&con->mutex);
dac1e716 1349 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1350 mutex_lock(&con->mutex);
4e7a5dcd 1351
a3530df3 1352 if (IS_ERR(auth))
729796be 1353 return auth;
8dacc7da 1354 if (con->state != CON_STATE_NEGOTIATING)
729796be 1355 return ERR_PTR(-EAGAIN);
0da5d703 1356
8f43fb53
AE		 1357 con->auth_reply_buf = auth->authorizer_reply_buf;
1358 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1359 return auth;
4e7a5dcd
SW		 1360}
1361
31b8006e
SW		 1362/*
1363 * We connected to a peer and are saying hello.
1364 */
e825a66d 1365static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1366{
e2200423
AE		 1367 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1368 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1369 &con->msgr->my_enc_addr);
eed0ef2c 1370
eed0ef2c 1371 con->out_more = 0;
c9ffc77a 1372 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1373}
1374
e825a66d 1375static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1376{
95c96174 1377 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1378 int proto;
dac1e716 1379 int auth_proto;
729796be 1380 struct ceph_auth_handshake *auth;
31b8006e
SW		 1381
1382 switch (con->peer_name.type) {
1383 case CEPH_ENTITY_TYPE_MON:
1384 proto = CEPH_MONC_PROTOCOL;
1385 break;
1386 case CEPH_ENTITY_TYPE_OSD:
1387 proto = CEPH_OSDC_PROTOCOL;
1388 break;
1389 case CEPH_ENTITY_TYPE_MDS:
1390 proto = CEPH_MDSC_PROTOCOL;
1391 break;
1392 default:
1393 BUG();
1394 }
1395
1396 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1397 con->connect_seq, global_seq, proto);
4e7a5dcd 1398
e825a66d 1399 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1400 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1401 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1402 con->out_connect.global_seq = cpu_to_le32(global_seq);
1403 con->out_connect.protocol_version = cpu_to_le32(proto);
1404 con->out_connect.flags = 0;
31b8006e 1405
dac1e716
AE		 1406 auth_proto = CEPH_AUTH_UNKNOWN;
1407 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1408 if (IS_ERR(auth))
1409 return PTR_ERR(auth);
3da54776 1410
dac1e716 1411 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1412 con->out_connect.authorizer_len = auth ?
1413 cpu_to_le32(auth->authorizer_buf_len) : 0;
1414
e2200423 1415 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1416 &con->out_connect);
1417 if (auth && auth->authorizer_buf_len)
e2200423 1418 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1419 auth->authorizer_buf);
859eb799 1420
31b8006e 1421 con->out_more = 0;
c9ffc77a 1422 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1423
e10c758e 1424 return 0;
31b8006e
SW		 1425}
1426
31b8006e
SW		 1427/*
1428 * write as much of pending kvecs to the socket as we can.
1429 * 1 -> done
1430 * 0 -> socket full, but more to do
1431 * <0 -> error
1432 */
1433static int write_partial_kvec(struct ceph_connection *con)
1434{
1435 int ret;
1436
1437 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1438 while (con->out_kvec_bytes > 0) {
1439 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1440 con->out_kvec_left, con->out_kvec_bytes,
1441 con->out_more);
1442 if (ret <= 0)
1443 goto out;
1444 con->out_kvec_bytes -= ret;
1445 if (con->out_kvec_bytes == 0)
1446 break; /* done */
f42299e6
AE		 1447
1448 /* account for full iov entries consumed */
1449 while (ret >= con->out_kvec_cur->iov_len) {
1450 BUG_ON(!con->out_kvec_left);
1451 ret -= con->out_kvec_cur->iov_len;
1452 con->out_kvec_cur++;
1453 con->out_kvec_left--;
1454 }
1455 /* and for a partially-consumed entry */
1456 if (ret) {
1457 con->out_kvec_cur->iov_len -= ret;
1458 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1459 }
1460 }
1461 con->out_kvec_left = 0;
1462 con->out_kvec_is_msg = false;
1463 ret = 1;
1464out:
1465 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1466 con->out_kvec_bytes, con->out_kvec_left, ret);
1467 return ret; /* done! */
1468}
1469
35b62808
AE		 1470static u32 ceph_crc32c_page(u32 crc, struct page *page,
1471 unsigned int page_offset,
1472 unsigned int length)
1473{
1474 char *kaddr;
1475
1476 kaddr = kmap(page);
1477 BUG_ON(kaddr == NULL);
1478 crc = crc32c(crc, kaddr + page_offset, length);
1479 kunmap(page);
1480
1481 return crc;
1482}
31b8006e
SW		 1483/*
1484 * Write as much message data payload as we can. If we finish, queue
1485 * up the footer.
1486 * 1 -> done, footer is now queued in out_kvec[].
1487 * 0 -> socket full, but more to do
1488 * <0 -> error
1489 */
34d2d200 1490static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1491{
1492 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1493 struct ceph_msg_data_cursor *cursor = &msg->cursor;
37675b0f 1494 bool do_datacrc = !con->msgr->nocrc;
f5db90bc 1495 u32 crc;
31b8006e 1496
859a35d5 1497 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1498
5240d9f9 1499 if (list_empty(&msg->data))
4c59b4a2
AE		 1500 return -EINVAL;
1501
5821bd8c
AE		 1502 /*
1503 * Iterate through each page that contains data to be
1504 * written, and send as much as possible for each.
1505 *
1506 * If we are calculating the data crc (the default), we will
1507 * need to map the page. If we have no pages, they have
1508 * been revoked, so use the zero page.
1509 */
f5db90bc 1510 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1511 while (cursor->resid) {
8a166d05 1512 struct page *page;
e387d525
AE		 1513 size_t page_offset;
1514 size_t length;
8a166d05 1515 bool last_piece;
8ea299bc 1516 bool need_crc;
f5db90bc 1517 int ret;
68b4476b 1518
8ae4f4f5 1519 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 1520 &last_piece);
e387d525 1521 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1522 length, last_piece);
f5db90bc
AE		 1523 if (ret <= 0) {
1524 if (do_datacrc)
1525 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1526
f5db90bc
AE		 1527 return ret;
1528 }
143334ff
AE		 1529 if (do_datacrc && cursor->need_crc)
1530 crc = ceph_crc32c_page(crc, page, page_offset, length);
8ae4f4f5 1531 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
31b8006e
SW		 1532 }
1533
34d2d200 1534 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1535
1536 /* prepare and queue up footer, too */
f5db90bc
AE		 1537 if (do_datacrc)
1538 msg->footer.data_crc = cpu_to_le32(crc);
1539 else
84ca8fc8 1540 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1541 con_out_kvec_reset(con);
859eb799 1542 prepare_write_message_footer(con);
f5db90bc
AE		 1543
1544 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1545}
1546
1547/*
1548 * write some zeros
1549 */
1550static int write_partial_skip(struct ceph_connection *con)
1551{
1552 int ret;
1553
1554 while (con->out_skip > 0) {
31739139 1555 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1556
e1dcb128 1557 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1558 if (ret <= 0)
1559 goto out;
1560 con->out_skip -= ret;
1561 }
1562 ret = 1;
1563out:
1564 return ret;
1565}
1566
1567/*
1568 * Prepare to read connection handshake, or an ack.
1569 */
eed0ef2c
SW		 1570static void prepare_read_banner(struct ceph_connection *con)
1571{
1572 dout("prepare_read_banner %p\n", con);
1573 con->in_base_pos = 0;
1574}
1575
31b8006e
SW		 1576static void prepare_read_connect(struct ceph_connection *con)
1577{
1578 dout("prepare_read_connect %p\n", con);
1579 con->in_base_pos = 0;
1580}
1581
1582static void prepare_read_ack(struct ceph_connection *con)
1583{
1584 dout("prepare_read_ack %p\n", con);
1585 con->in_base_pos = 0;
1586}
1587
3a23083b
SW		 1588static void prepare_read_seq(struct ceph_connection *con)
1589{
1590 dout("prepare_read_seq %p\n", con);
1591 con->in_base_pos = 0;
1592 con->in_tag = CEPH_MSGR_TAG_SEQ;
1593}
1594
31b8006e
SW		 1595static void prepare_read_tag(struct ceph_connection *con)
1596{
1597 dout("prepare_read_tag %p\n", con);
1598 con->in_base_pos = 0;
1599 con->in_tag = CEPH_MSGR_TAG_READY;
1600}
1601
1602/*
1603 * Prepare to read a message.
1604 */
1605static int prepare_read_message(struct ceph_connection *con)
1606{
1607 dout("prepare_read_message %p\n", con);
1608 BUG_ON(con->in_msg != NULL);
1609 con->in_base_pos = 0;
1610 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1611 return 0;
1612}
1613
1614
1615static int read_partial(struct ceph_connection *con,
fd51653f 1616 int end, int size, void *object)
31b8006e 1617{
e6cee71f
AE		 1618 while (con->in_base_pos < end) {
1619 int left = end - con->in_base_pos;
31b8006e
SW		 1620 int have = size - left;
1621 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1622 if (ret <= 0)
1623 return ret;
1624 con->in_base_pos += ret;
1625 }
1626 return 1;
1627}
1628
1629
1630/*
1631 * Read all or part of the connect-side handshake on a new connection
1632 */
eed0ef2c 1633static int read_partial_banner(struct ceph_connection *con)
31b8006e 1634{
fd51653f
AE		 1635 int size;
1636 int end;
1637 int ret;
31b8006e 1638
eed0ef2c 1639 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1640
1641 /* peer's banner */
fd51653f
AE		 1642 size = strlen(CEPH_BANNER);
1643 end = size;
1644 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1645 if (ret <= 0)
1646 goto out;
fd51653f
AE		 1647
1648 size = sizeof (con->actual_peer_addr);
1649 end += size;
1650 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1651 if (ret <= 0)
1652 goto out;
fd51653f
AE		 1653
1654 size = sizeof (con->peer_addr_for_me);
1655 end += size;
1656 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1657 if (ret <= 0)
1658 goto out;
fd51653f 1659
eed0ef2c
SW		 1660out:
1661 return ret;
1662}
1663
1664static int read_partial_connect(struct ceph_connection *con)
1665{
fd51653f
AE		 1666 int size;
1667 int end;
1668 int ret;
eed0ef2c
SW		 1669
1670 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1671
fd51653f
AE		 1672 size = sizeof (con->in_reply);
1673 end = size;
1674 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1675 if (ret <= 0)
1676 goto out;
fd51653f
AE		 1677
1678 size = le32_to_cpu(con->in_reply.authorizer_len);
1679 end += size;
1680 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1681 if (ret <= 0)
1682 goto out;
31b8006e 1683
4e7a5dcd
SW		 1684 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1685 con, (int)con->in_reply.tag,
1686 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1687 le32_to_cpu(con->in_reply.global_seq));
1688out:
1689 return ret;
eed0ef2c 1690
31b8006e
SW		 1691}
1692
1693/*
1694 * Verify the hello banner looks okay.
1695 */
1696static int verify_hello(struct ceph_connection *con)
1697{
1698 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1699 pr_err("connect to %s got bad banner\n",
3d14c5d2 1700 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1701 con->error_msg = "protocol error, bad banner";
1702 return -1;
1703 }
1704 return 0;
1705}
1706
1707static bool addr_is_blank(struct sockaddr_storage *ss)
1708{
1709 switch (ss->ss_family) {
1710 case AF_INET:
1711 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1712 case AF_INET6:
1713 return
1714 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1715 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1716 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1717 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1718 }
1719 return false;
1720}
1721
1722static int addr_port(struct sockaddr_storage *ss)
1723{
1724 switch (ss->ss_family) {
1725 case AF_INET:
f28bcfbe 1726 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1727 case AF_INET6:
f28bcfbe 1728 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1729 }
1730 return 0;
1731}
1732
1733static void addr_set_port(struct sockaddr_storage *ss, int p)
1734{
1735 switch (ss->ss_family) {
1736 case AF_INET:
1737 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1738 break;
31b8006e
SW		 1739 case AF_INET6:
1740 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1741 break;
31b8006e
SW		 1742 }
1743}
1744
ee3b56f2
NW		 1745/*
1746 * Unlike other *_pton function semantics, zero indicates success.
1747 */
1748static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1749 char delim, const char **ipend)
1750{
99f0f3b2
AE		 1751 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1752 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1753
1754 memset(ss, 0, sizeof(*ss));
1755
1756 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1757 ss->ss_family = AF_INET;
1758 return 0;
1759 }
1760
1761 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1762 ss->ss_family = AF_INET6;
1763 return 0;
1764 }
1765
1766 return -EINVAL;
1767}
1768
1769/*
1770 * Extract hostname string and resolve using kernel DNS facility.
1771 */
1772#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1773static int ceph_dns_resolve_name(const char *name, size_t namelen,
1774 struct sockaddr_storage *ss, char delim, const char **ipend)
1775{
1776 const char *end, *delim_p;
1777 char *colon_p, *ip_addr = NULL;
1778 int ip_len, ret;
1779
1780 /*
1781 * The end of the hostname occurs immediately preceding the delimiter or
1782 * the port marker (':') where the delimiter takes precedence.
1783 */
1784 delim_p = memchr(name, delim, namelen);
1785 colon_p = memchr(name, ':', namelen);
1786
1787 if (delim_p && colon_p)
1788 end = delim_p < colon_p ? delim_p : colon_p;
1789 else if (!delim_p && colon_p)
1790 end = colon_p;
1791 else {
1792 end = delim_p;
1793 if (!end) /* case: hostname:/ */
1794 end = name + namelen;
1795 }
1796
1797 if (end <= name)
1798 return -EINVAL;
1799
1800 /* do dns_resolve upcall */
1801 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1802 if (ip_len > 0)
1803 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1804 else
1805 ret = -ESRCH;
1806
1807 kfree(ip_addr);
1808
1809 *ipend = end;
1810
1811 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1812 ret, ret ? "failed" : ceph_pr_addr(ss));
1813
1814 return ret;
1815}
1816#else
1817static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1818 struct sockaddr_storage *ss, char delim, const char **ipend)
1819{
1820 return -EINVAL;
1821}
1822#endif
1823
1824/*
1825 * Parse a server name (IP or hostname). If a valid IP address is not found
1826 * then try to extract a hostname to resolve using userspace DNS upcall.
1827 */
1828static int ceph_parse_server_name(const char *name, size_t namelen,
1829 struct sockaddr_storage *ss, char delim, const char **ipend)
1830{
1831 int ret;
1832
1833 ret = ceph_pton(name, namelen, ss, delim, ipend);
1834 if (ret)
1835 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1836
1837 return ret;
1838}
1839
31b8006e
SW		 1840/*
1841 * Parse an ip[:port] list into an addr array. Use the default
1842 * monitor port if a port isn't specified.
1843 */
1844int ceph_parse_ips(const char *c, const char *end,
1845 struct ceph_entity_addr *addr,
1846 int max_count, int *count)
1847{
ee3b56f2 1848 int i, ret = -EINVAL;
31b8006e
SW		 1849 const char *p = c;
1850
1851 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1852 for (i = 0; i < max_count; i++) {
1853 const char *ipend;
1854 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1855 int port;
39139f64
SW		 1856 char delim = ',';
1857
1858 if (*p == '[') {
1859 delim = ']';
1860 p++;
1861 }
31b8006e 1862
ee3b56f2
NW		 1863 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1864 if (ret)
31b8006e 1865 goto bad;
ee3b56f2
NW		 1866 ret = -EINVAL;
1867
31b8006e
SW		 1868 p = ipend;
1869
39139f64
SW		 1870 if (delim == ']') {
1871 if (*p != ']') {
1872 dout("missing matching ']'\n");
1873 goto bad;
1874 }
1875 p++;
1876 }
1877
31b8006e
SW		 1878 /* port? */
1879 if (p < end && *p == ':') {
1880 port = 0;
1881 p++;
1882 while (p < end && *p >= '0' && *p <= '9') {
1883 port = (port * 10) + (*p - '0');
1884 p++;
1885 }
1886 if (port > 65535 || port == 0)
1887 goto bad;
1888 } else {
1889 port = CEPH_MON_PORT;
1890 }
1891
1892 addr_set_port(ss, port);
1893
3d14c5d2 1894 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1895
1896 if (p == end)
1897 break;
1898 if (*p != ',')
1899 goto bad;
1900 p++;
1901 }
1902
1903 if (p != end)
1904 goto bad;
1905
1906 if (count)
1907 *count = i + 1;
1908 return 0;
1909
1910bad:
39139f64 1911 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1912 return ret;
31b8006e 1913}
3d14c5d2 1914EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1915
eed0ef2c 1916static int process_banner(struct ceph_connection *con)
31b8006e 1917{
eed0ef2c 1918 dout("process_banner on %p\n", con);
31b8006e
SW		 1919
1920 if (verify_hello(con) < 0)
1921 return -1;
1922
63f2d211
SW		 1923 ceph_decode_addr(&con->actual_peer_addr);
1924 ceph_decode_addr(&con->peer_addr_for_me);
1925
31b8006e
SW		 1926 /*
1927 * Make sure the other end is who we wanted. note that the other
1928 * end may not yet know their ip address, so if it's 0.0.0.0, give
1929 * them the benefit of the doubt.
1930 */
103e2d3a
SW		 1931 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1932 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1933 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1934 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
cd84db6e 1935 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
3d14c5d2 1936 ceph_pr_addr(&con->peer_addr.in_addr),
cd84db6e 1937 (int)le32_to_cpu(con->peer_addr.nonce),
3d14c5d2 1938 ceph_pr_addr(&con->actual_peer_addr.in_addr),
cd84db6e 1939 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1940 con->error_msg = "wrong peer at address";
31b8006e
SW		 1941 return -1;
1942 }
1943
1944 /*
1945 * did we learn our address?
1946 */
1947 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1948 int port = addr_port(&con->msgr->inst.addr.in_addr);
1949
1950 memcpy(&con->msgr->inst.addr.in_addr,
1951 &con->peer_addr_for_me.in_addr,
1952 sizeof(con->peer_addr_for_me.in_addr));
1953 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1954 encode_my_addr(con->msgr);
eed0ef2c 1955 dout("process_banner learned my addr is %s\n",
3d14c5d2 1956 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1957 }
1958
eed0ef2c
SW		 1959 return 0;
1960}
1961
1962static int process_connect(struct ceph_connection *con)
1963{
3d14c5d2
YS		 1964 u64 sup_feat = con->msgr->supported_features;
1965 u64 req_feat = con->msgr->required_features;
04a419f9 1966 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 1967 int ret;
04a419f9 1968
eed0ef2c
SW		 1969 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1970
31b8006e 1971 switch (con->in_reply.tag) {
04a419f9
SW		 1972 case CEPH_MSGR_TAG_FEATURES:
1973 pr_err("%s%lld %s feature set mismatch,"
1974 " my %llx < server's %llx, missing %llx\n",
1975 ENTITY_NAME(con->peer_name),
3d14c5d2 1976 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1977 sup_feat, server_feat, server_feat & ~sup_feat);
1978 con->error_msg = "missing required protocol features";
0fa6ebc6 1979 reset_connection(con);
04a419f9
SW		 1980 return -1;
1981
31b8006e 1982 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 1983 pr_err("%s%lld %s protocol version mismatch,"
1984 " my %d != server's %d\n",
1985 ENTITY_NAME(con->peer_name),
3d14c5d2 1986 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 1987 le32_to_cpu(con->out_connect.protocol_version),
1988 le32_to_cpu(con->in_reply.protocol_version));
1989 con->error_msg = "protocol version mismatch";
0fa6ebc6 1990 reset_connection(con);
31b8006e
SW		 1991 return -1;
1992
4e7a5dcd
SW		 1993 case CEPH_MSGR_TAG_BADAUTHORIZER:
1994 con->auth_retry++;
1995 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1996 con->auth_retry);
1997 if (con->auth_retry == 2) {
1998 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 1999 return -1;
2000 }
6d4221b5 2001 con_out_kvec_reset(con);
e825a66d 2002 ret = prepare_write_connect(con);
0da5d703
SW		 2003 if (ret < 0)
2004 return ret;
63733a0f 2005 prepare_read_connect(con);
4e7a5dcd 2006 break;
31b8006e
SW		 2007
2008 case CEPH_MSGR_TAG_RESETSESSION:
2009 /*
2010 * If we connected with a large connect_seq but the peer
2011 * has no record of a session with us (no connection, or
2012 * connect_seq == 0), they will send RESETSESION to indicate
2013 * that they must have reset their session, and may have
2014 * dropped messages.
2015 */
2016 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2017 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2018 pr_err("%s%lld %s connection reset\n",
2019 ENTITY_NAME(con->peer_name),
3d14c5d2 2020 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2021 reset_connection(con);
6d4221b5 2022 con_out_kvec_reset(con);
5a0f8fdd
AE		 2023 ret = prepare_write_connect(con);
2024 if (ret < 0)
2025 return ret;
31b8006e
SW		 2026 prepare_read_connect(con);
2027
2028 /* Tell ceph about it. */
ec302645 2029 mutex_unlock(&con->mutex);
31b8006e
SW		 2030 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2031 if (con->ops->peer_reset)
2032 con->ops->peer_reset(con);
ec302645 2033 mutex_lock(&con->mutex);
8dacc7da 2034 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2035 return -EAGAIN;
31b8006e
SW		 2036 break;
2037
2038 case CEPH_MSGR_TAG_RETRY_SESSION:
2039 /*
2040 * If we sent a smaller connect_seq than the peer has, try
2041 * again with a larger value.
2042 */
5bdca4e0 2043 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2044 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2045 le32_to_cpu(con->in_reply.connect_seq));
2046 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2047 con_out_kvec_reset(con);
5a0f8fdd
AE		 2048 ret = prepare_write_connect(con);
2049 if (ret < 0)
2050 return ret;
31b8006e
SW		 2051 prepare_read_connect(con);
2052 break;
2053
2054 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2055 /*
2056 * If we sent a smaller global_seq than the peer has, try
2057 * again with a larger value.
2058 */
eed0ef2c 2059 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2060 con->peer_global_seq,
5bdca4e0 2061 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2062 get_global_seq(con->msgr,
5bdca4e0 2063 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2064 con_out_kvec_reset(con);
5a0f8fdd
AE		 2065 ret = prepare_write_connect(con);
2066 if (ret < 0)
2067 return ret;
31b8006e
SW		 2068 prepare_read_connect(con);
2069 break;
2070
3a23083b 2071 case CEPH_MSGR_TAG_SEQ:
31b8006e 2072 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2073 if (req_feat & ~server_feat) {
2074 pr_err("%s%lld %s protocol feature mismatch,"
2075 " my required %llx > server's %llx, need %llx\n",
2076 ENTITY_NAME(con->peer_name),
3d14c5d2 2077 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2078 req_feat, server_feat, req_feat & ~server_feat);
2079 con->error_msg = "missing required protocol features";
0fa6ebc6 2080 reset_connection(con);
04a419f9
SW		 2081 return -1;
2082 }
8dacc7da 2083
122070a2 2084 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2085 con->state = CON_STATE_OPEN;
20e55c4c 2086 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2087 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2088 con->connect_seq++;
aba558e2 2089 con->peer_features = server_feat;
31b8006e
SW		 2090 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2091 con->peer_global_seq,
2092 le32_to_cpu(con->in_reply.connect_seq),
2093 con->connect_seq);
2094 WARN_ON(con->connect_seq !=
2095 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2096
2097 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2098 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2099
85effe18 2100 con->delay = 0; /* reset backoff memory */
92ac41d0 2101
3a23083b
SW		 2102 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2103 prepare_write_seq(con);
2104 prepare_read_seq(con);
2105 } else {
2106 prepare_read_tag(con);
2107 }
31b8006e
SW		 2108 break;
2109
2110 case CEPH_MSGR_TAG_WAIT:
2111 /*
2112 * If there is a connection race (we are opening
2113 * connections to each other), one of us may just have
2114 * to WAIT. This shouldn't happen if we are the
2115 * client.
2116 */
04177882
SW		 2117 pr_err("process_connect got WAIT as client\n");
2118 con->error_msg = "protocol error, got WAIT as client";
2119 return -1;
31b8006e
SW		 2120
2121 default:
2122 pr_err("connect protocol error, will retry\n");
2123 con->error_msg = "protocol error, garbage tag during connect";
2124 return -1;
2125 }
2126 return 0;
2127}
2128
2129
2130/*
2131 * read (part of) an ack
2132 */
2133static int read_partial_ack(struct ceph_connection *con)
2134{
fd51653f
AE		 2135 int size = sizeof (con->in_temp_ack);
2136 int end = size;
31b8006e 2137
fd51653f 2138 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2139}
2140
31b8006e
SW		 2141/*
2142 * We can finally discard anything that's been acked.
2143 */
2144static void process_ack(struct ceph_connection *con)
2145{
2146 struct ceph_msg *m;
2147 u64 ack = le64_to_cpu(con->in_temp_ack);
2148 u64 seq;
2149
31b8006e
SW		 2150 while (!list_empty(&con->out_sent)) {
2151 m = list_first_entry(&con->out_sent, struct ceph_msg,
2152 list_head);
2153 seq = le64_to_cpu(m->hdr.seq);
2154 if (seq > ack)
2155 break;
2156 dout("got ack for seq %llu type %d at %p\n", seq,
2157 le16_to_cpu(m->hdr.type), m);
4cf9d544 2158 m->ack_stamp = jiffies;
31b8006e
SW		 2159 ceph_msg_remove(m);
2160 }
31b8006e
SW		 2161 prepare_read_tag(con);
2162}
2163
2164
2450418c 2165static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2166 struct kvec *section,
2167 unsigned int sec_len, u32 *crc)
2450418c 2168{
68b4476b 2169 int ret, left;
2450418c
YS		 2170
2171 BUG_ON(!section);
2172
2173 while (section->iov_len < sec_len) {
2174 BUG_ON(section->iov_base == NULL);
2175 left = sec_len - section->iov_len;
2176 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2177 section->iov_len, left);
2178 if (ret <= 0)
2179 return ret;
2180 section->iov_len += ret;
2450418c 2181 }
fe3ad593
AE		 2182 if (section->iov_len == sec_len)
2183 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2184
2450418c
YS		 2185 return 1;
2186}
31b8006e 2187
34d2d200
AE		 2188static int read_partial_msg_data(struct ceph_connection *con)
2189{
2190 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2191 struct ceph_msg_data_cursor *cursor = &msg->cursor;
34d2d200 2192 const bool do_datacrc = !con->msgr->nocrc;
686be208
AE		 2193 struct page *page;
2194 size_t page_offset;
2195 size_t length;
f5db90bc 2196 u32 crc = 0;
34d2d200
AE		 2197 int ret;
2198
2199 BUG_ON(!msg);
5240d9f9 2200 if (list_empty(&msg->data))
4c59b4a2 2201 return -EIO;
34d2d200 2202
f5db90bc
AE		 2203 if (do_datacrc)
2204 crc = con->in_data_crc;
643c68a4 2205 while (cursor->resid) {
8ae4f4f5 2206 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 2207 NULL);
686be208 2208 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2209 if (ret <= 0) {
2210 if (do_datacrc)
2211 con->in_data_crc = crc;
2212
686be208 2213 return ret;
f5db90bc 2214 }
686be208
AE		 2215
2216 if (do_datacrc)
f5db90bc 2217 crc = ceph_crc32c_page(crc, page, page_offset, ret);
8ae4f4f5 2218 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
34d2d200 2219 }
f5db90bc
AE		 2220 if (do_datacrc)
2221 con->in_data_crc = crc;
34d2d200
AE		 2222
2223 return 1; /* must return > 0 to indicate success */
2224}
2225
31b8006e
SW		 2226/*
2227 * read (part of) a message.
2228 */
686be208
AE		 2229static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2230
31b8006e
SW		 2231static int read_partial_message(struct ceph_connection *con)
2232{
2233 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2234 int size;
2235 int end;
31b8006e 2236 int ret;
95c96174 2237 unsigned int front_len, middle_len, data_len;
37675b0f 2238 bool do_datacrc = !con->msgr->nocrc;
ae18756b 2239 u64 seq;
fe3ad593 2240 u32 crc;
31b8006e
SW		 2241
2242 dout("read_partial_message con %p msg %p\n", con, m);
2243
2244 /* header */
fd51653f
AE		 2245 size = sizeof (con->in_hdr);
2246 end = size;
2247 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2248 if (ret <= 0)
2249 return ret;
fe3ad593
AE		 2250
2251 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2252 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2253 pr_err("read_partial_message bad hdr "
2254 " crc %u != expected %u\n",
2255 crc, con->in_hdr.crc);
2256 return -EBADMSG;
2257 }
2258
31b8006e
SW		 2259 front_len = le32_to_cpu(con->in_hdr.front_len);
2260 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2261 return -EIO;
2262 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2263 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2264 return -EIO;
2265 data_len = le32_to_cpu(con->in_hdr.data_len);
2266 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2267 return -EIO;
2268
ae18756b
SW		 2269 /* verify seq# */
2270 seq = le64_to_cpu(con->in_hdr.seq);
2271 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2272 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2273 ENTITY_NAME(con->peer_name),
3d14c5d2 2274 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2275 seq, con->in_seq + 1);
2276 con->in_base_pos = -front_len - middle_len - data_len -
2277 sizeof(m->footer);
2278 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2279 return 0;
2280 } else if ((s64)seq - (s64)con->in_seq > 1) {
2281 pr_err("read_partial_message bad seq %lld expected %lld\n",
2282 seq, con->in_seq + 1);
2283 con->error_msg = "bad message sequence # for incoming message";
2284 return -EBADMSG;
2285 }
2286
31b8006e
SW		 2287 /* allocate message? */
2288 if (!con->in_msg) {
4740a623
SW		 2289 int skip = 0;
2290
31b8006e 2291 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2292 front_len, data_len);
4740a623
SW		 2293 ret = ceph_con_in_msg_alloc(con, &skip);
2294 if (ret < 0)
2295 return ret;
f759ebb9
AE		 2296
2297 BUG_ON(!con->in_msg ^ skip);
2298 if (con->in_msg && data_len > con->in_msg->data_length) {
2299 pr_warning("%s skipping long message (%u > %zd)\n",
2300 __func__, data_len, con->in_msg->data_length);
2301 ceph_msg_put(con->in_msg);
2302 con->in_msg = NULL;
2303 skip = 1;
2304 }
2450418c 2305 if (skip) {
31b8006e 2306 /* skip this message */
a79832f2 2307 dout("alloc_msg said skip message\n");
31b8006e
SW		 2308 con->in_base_pos = -front_len - middle_len - data_len -
2309 sizeof(m->footer);
2310 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2311 con->in_seq++;
31b8006e
SW		 2312 return 0;
2313 }
38941f80 2314
4740a623 2315 BUG_ON(!con->in_msg);
38941f80 2316 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2317 m = con->in_msg;
2318 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2319 if (m->middle)
2320 m->middle->vec.iov_len = 0;
9d7f0f13 2321
78625051 2322 /* prepare for data payload, if any */
a4107026 2323
78625051 2324 if (data_len)
98fa5dd8 2325 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2326 }
2327
2328 /* front */
2450418c
YS		 2329 ret = read_partial_message_section(con, &m->front, front_len,
2330 &con->in_front_crc);
2331 if (ret <= 0)
2332 return ret;
31b8006e
SW		 2333
2334 /* middle */
2450418c 2335 if (m->middle) {
213c99ee
SW		 2336 ret = read_partial_message_section(con, &m->middle->vec,
2337 middle_len,
2450418c 2338 &con->in_middle_crc);
31b8006e
SW		 2339 if (ret <= 0)
2340 return ret;
31b8006e
SW		 2341 }
2342
2343 /* (page) data */
34d2d200
AE		 2344 if (data_len) {
2345 ret = read_partial_msg_data(con);
2346 if (ret <= 0)
2347 return ret;
31b8006e
SW		 2348 }
2349
31b8006e 2350 /* footer */
fd51653f
AE		 2351 size = sizeof (m->footer);
2352 end += size;
2353 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2354 if (ret <= 0)
2355 return ret;
2356
31b8006e
SW		 2357 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2358 m, front_len, m->footer.front_crc, middle_len,
2359 m->footer.middle_crc, data_len, m->footer.data_crc);
2360
2361 /* crc ok? */
2362 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2363 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2364 m, con->in_front_crc, m->footer.front_crc);
2365 return -EBADMSG;
2366 }
2367 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2368 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2369 m, con->in_middle_crc, m->footer.middle_crc);
2370 return -EBADMSG;
2371 }
bca064d2 2372 if (do_datacrc &&
31b8006e
SW		 2373 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2374 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2375 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2376 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2377 return -EBADMSG;
2378 }
2379
2380 return 1; /* done! */
2381}
2382
2383/*
2384 * Process message. This happens in the worker thread. The callback should
2385 * be careful not to do anything that waits on other incoming messages or it
2386 * may deadlock.
2387 */
2388static void process_message(struct ceph_connection *con)
2389{
5e095e8b 2390 struct ceph_msg *msg;
31b8006e 2391
38941f80
AE		 2392 BUG_ON(con->in_msg->con != con);
2393 con->in_msg->con = NULL;
5e095e8b 2394 msg = con->in_msg;
31b8006e 2395 con->in_msg = NULL;
36eb71aa 2396 con->ops->put(con);
31b8006e
SW		 2397
2398 /* if first message, set peer_name */
2399 if (con->peer_name.type == 0)
dbad185d 2400 con->peer_name = msg->hdr.src;
31b8006e 2401
31b8006e 2402 con->in_seq++;
ec302645 2403 mutex_unlock(&con->mutex);
31b8006e
SW		 2404
2405 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2406 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2407 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2408 le16_to_cpu(msg->hdr.type),
2409 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2410 le32_to_cpu(msg->hdr.front_len),
2411 le32_to_cpu(msg->hdr.data_len),
2412 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2413 con->ops->dispatch(con, msg);
ec302645
SW		 2414
2415 mutex_lock(&con->mutex);
31b8006e
SW		 2416}
2417
2418
2419/*
2420 * Write something to the socket. Called in a worker thread when the
2421 * socket appears to be writeable and we have something ready to send.
2422 */
2423static int try_write(struct ceph_connection *con)
2424{
31b8006e
SW		 2425 int ret = 1;
2426
d59315ca 2427 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2428
31b8006e
SW		 2429more:
2430 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2431
2432 /* open the socket first? */
8dacc7da
SW		 2433 if (con->state == CON_STATE_PREOPEN) {
2434 BUG_ON(con->sock);
2435 con->state = CON_STATE_CONNECTING;
a5988c49 2436
e2200423 2437 con_out_kvec_reset(con);
e825a66d 2438 prepare_write_banner(con);
eed0ef2c 2439 prepare_read_banner(con);
31b8006e 2440
cf3e5c40 2441 BUG_ON(con->in_msg);
31b8006e
SW		 2442 con->in_tag = CEPH_MSGR_TAG_READY;
2443 dout("try_write initiating connect on %p new state %lu\n",
2444 con, con->state);
41617d0c
AE		 2445 ret = ceph_tcp_connect(con);
2446 if (ret < 0) {
31b8006e 2447 con->error_msg = "connect error";
31b8006e
SW		 2448 goto out;
2449 }
2450 }
2451
2452more_kvec:
2453 /* kvec data queued? */
2454 if (con->out_skip) {
2455 ret = write_partial_skip(con);
2456 if (ret <= 0)
42961d23 2457 goto out;
31b8006e
SW		 2458 }
2459 if (con->out_kvec_left) {
2460 ret = write_partial_kvec(con);
2461 if (ret <= 0)
42961d23 2462 goto out;
31b8006e
SW		 2463 }
2464
2465 /* msg pages? */
2466 if (con->out_msg) {
c86a2930
SW		 2467 if (con->out_msg_done) {
2468 ceph_msg_put(con->out_msg);
2469 con->out_msg = NULL; /* we're done with this one */
2470 goto do_next;
2471 }
2472
34d2d200 2473 ret = write_partial_message_data(con);
31b8006e
SW		 2474 if (ret == 1)
2475 goto more_kvec; /* we need to send the footer, too! */
2476 if (ret == 0)
42961d23 2477 goto out;
31b8006e 2478 if (ret < 0) {
34d2d200 2479 dout("try_write write_partial_message_data err %d\n",
31b8006e 2480 ret);
42961d23 2481 goto out;
31b8006e
SW		 2482 }
2483 }
2484
c86a2930 2485do_next:
8dacc7da 2486 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2487 /* is anything else pending? */
2488 if (!list_empty(&con->out_queue)) {
2489 prepare_write_message(con);
2490 goto more;
2491 }
2492 if (con->in_seq > con->in_seq_acked) {
2493 prepare_write_ack(con);
2494 goto more;
2495 }
c9ffc77a 2496 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2497 prepare_write_keepalive(con);
2498 goto more;
2499 }
2500 }
2501
2502 /* Nothing to do! */
c9ffc77a 2503 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2504 dout("try_write nothing else to write.\n");
31b8006e
SW		 2505 ret = 0;
2506out:
42961d23 2507 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2508 return ret;
2509}
2510
2511
2512
2513/*
2514 * Read what we can from the socket.
2515 */
2516static int try_read(struct ceph_connection *con)
2517{
31b8006e
SW		 2518 int ret = -1;
2519
8dacc7da
SW		 2520more:
2521 dout("try_read start on %p state %lu\n", con, con->state);
2522 if (con->state != CON_STATE_CONNECTING &&
2523 con->state != CON_STATE_NEGOTIATING &&
2524 con->state != CON_STATE_OPEN)
31b8006e
SW		 2525 return 0;
2526
8dacc7da 2527 BUG_ON(!con->sock);
ec302645 2528
31b8006e
SW		 2529 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2530 con->in_base_pos);
0da5d703 2531
8dacc7da 2532 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2533 dout("try_read connecting\n");
2534 ret = read_partial_banner(con);
2535 if (ret <= 0)
ab166d5a 2536 goto out;
7593af92
AE		 2537 ret = process_banner(con);
2538 if (ret < 0)
2539 goto out;
2540
8dacc7da 2541 con->state = CON_STATE_NEGOTIATING;
7593af92 2542
6d4221b5
JS		 2543 /*
2544 * Received banner is good, exchange connection info.
2545 * Do not reset out_kvec, as sending our banner raced
2546 * with receiving peer banner after connect completed.
2547 */
7593af92
AE		 2548 ret = prepare_write_connect(con);
2549 if (ret < 0)
2550 goto out;
2551 prepare_read_connect(con);
2552
2553 /* Send connection info before awaiting response */
0da5d703
SW		 2554 goto out;
2555 }
2556
8dacc7da 2557 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2558 dout("try_read negotiating\n");
31b8006e
SW		 2559 ret = read_partial_connect(con);
2560 if (ret <= 0)
31b8006e 2561 goto out;
98bdb0aa
SW		 2562 ret = process_connect(con);
2563 if (ret < 0)
2564 goto out;
31b8006e
SW		 2565 goto more;
2566 }
2567
122070a2 2568 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2569
31b8006e
SW		 2570 if (con->in_base_pos < 0) {
2571 /*
2572 * skipping + discarding content.
2573 *
2574 * FIXME: there must be a better way to do this!
2575 */
84495f49
AE		 2576 static char buf[SKIP_BUF_SIZE];
2577 int skip = min((int) sizeof (buf), -con->in_base_pos);
2578
31b8006e
SW		 2579 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2580 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2581 if (ret <= 0)
98bdb0aa 2582 goto out;
31b8006e
SW		 2583 con->in_base_pos += ret;
2584 if (con->in_base_pos)
2585 goto more;
2586 }
2587 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2588 /*
2589 * what's next?
2590 */
2591 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2592 if (ret <= 0)
98bdb0aa 2593 goto out;
31b8006e
SW		 2594 dout("try_read got tag %d\n", (int)con->in_tag);
2595 switch (con->in_tag) {
2596 case CEPH_MSGR_TAG_MSG:
2597 prepare_read_message(con);
2598 break;
2599 case CEPH_MSGR_TAG_ACK:
2600 prepare_read_ack(con);
2601 break;
2602 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2603 con_close_socket(con);
2604 con->state = CON_STATE_CLOSED;
98bdb0aa 2605 goto out;
31b8006e
SW		 2606 default:
2607 goto bad_tag;
2608 }
2609 }
2610 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2611 ret = read_partial_message(con);
2612 if (ret <= 0) {
2613 switch (ret) {
2614 case -EBADMSG:
2615 con->error_msg = "bad crc";
2616 ret = -EIO;
98bdb0aa 2617 break;
31b8006e
SW		 2618 case -EIO:
2619 con->error_msg = "io error";
98bdb0aa 2620 break;
31b8006e 2621 }
98bdb0aa 2622 goto out;
31b8006e
SW		 2623 }
2624 if (con->in_tag == CEPH_MSGR_TAG_READY)
2625 goto more;
2626 process_message(con);
7b862e07
SW		 2627 if (con->state == CON_STATE_OPEN)
2628 prepare_read_tag(con);
31b8006e
SW		 2629 goto more;
2630 }
3a23083b
SW		 2631 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2632 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2633 /*
2634 * the final handshake seq exchange is semantically
2635 * equivalent to an ACK
2636 */
31b8006e
SW		 2637 ret = read_partial_ack(con);
2638 if (ret <= 0)
98bdb0aa 2639 goto out;
31b8006e
SW		 2640 process_ack(con);
2641 goto more;
2642 }
2643
31b8006e 2644out:
98bdb0aa 2645 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2646 return ret;
2647
2648bad_tag:
2649 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2650 con->error_msg = "protocol error, garbage tag";
2651 ret = -1;
2652 goto out;
2653}
2654
2655
2656/*
802c6d96
AE		 2657 * Atomically queue work on a connection after the specified delay.
2658 * Bump @con reference to avoid races with connection teardown.
2659 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2660 */
802c6d96 2661static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2662{
31b8006e 2663 if (!con->ops->get(con)) {
802c6d96
AE		 2664 dout("%s %p ref count 0\n", __func__, con);
2665
2666 return -ENOENT;
31b8006e
SW		 2667 }
2668
802c6d96
AE		 2669 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2670 dout("%s %p - already queued\n", __func__, con);
31b8006e 2671 con->ops->put(con);
802c6d96
AE		 2672
2673 return -EBUSY;
31b8006e 2674 }
802c6d96
AE		 2675
2676 dout("%s %p %lu\n", __func__, con, delay);
2677
2678 return 0;
2679}
2680
2681static void queue_con(struct ceph_connection *con)
2682{
2683 (void) queue_con_delay(con, 0);
31b8006e
SW		 2684}
2685
7bb21d68
AE		 2686static bool con_sock_closed(struct ceph_connection *con)
2687{
c9ffc77a 2688 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2689 return false;
2690
2691#define CASE(x) \
2692 case CON_STATE_ ## x: \
2693 con->error_msg = "socket closed (con state " #x ")"; \
2694 break;
2695
2696 switch (con->state) {
2697 CASE(CLOSED);
2698 CASE(PREOPEN);
2699 CASE(CONNECTING);
2700 CASE(NEGOTIATING);
2701 CASE(OPEN);
2702 CASE(STANDBY);
2703 default:
2704 pr_warning("%s con %p unrecognized state %lu\n",
2705 __func__, con, con->state);
2706 con->error_msg = "unrecognized con state";
2707 BUG();
2708 break;
2709 }
2710#undef CASE
2711
2712 return true;
2713}
2714
f20a39fd
AE		 2715static bool con_backoff(struct ceph_connection *con)
2716{
2717 int ret;
2718
2719 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2720 return false;
2721
2722 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2723 if (ret) {
2724 dout("%s: con %p FAILED to back off %lu\n", __func__,
2725 con, con->delay);
2726 BUG_ON(ret == -ENOENT);
2727 con_flag_set(con, CON_FLAG_BACKOFF);
2728 }
2729
2730 return true;
2731}
2732
93209264
AE		 2733/* Finish fault handling; con->mutex must *not* be held here */
2734
2735static void con_fault_finish(struct ceph_connection *con)
2736{
2737 /*
2738 * in case we faulted due to authentication, invalidate our
2739 * current tickets so that we can get new ones.
2740 */
2741 if (con->auth_retry && con->ops->invalidate_authorizer) {
2742 dout("calling invalidate_authorizer()\n");
2743 con->ops->invalidate_authorizer(con);
2744 }
2745
2746 if (con->ops->fault)
2747 con->ops->fault(con);
2748}
2749
31b8006e
SW		 2750/*
2751 * Do some work on a connection. Drop a connection ref when we're done.
2752 */
2753static void con_work(struct work_struct *work)
2754{
2755 struct ceph_connection *con = container_of(work, struct ceph_connection,
2756 work.work);
49659416 2757 bool fault;
31b8006e 2758
9dd4658d 2759 mutex_lock(&con->mutex);
49659416
AE		 2760 while (true) {
2761 int ret;
31b8006e 2762
49659416
AE		 2763 if ((fault = con_sock_closed(con))) {
2764 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2765 break;
2766 }
2767 if (con_backoff(con)) {
2768 dout("%s: con %p BACKOFF\n", __func__, con);
2769 break;
2770 }
2771 if (con->state == CON_STATE_STANDBY) {
2772 dout("%s: con %p STANDBY\n", __func__, con);
2773 break;
2774 }
2775 if (con->state == CON_STATE_CLOSED) {
2776 dout("%s: con %p CLOSED\n", __func__, con);
2777 BUG_ON(con->sock);
2778 break;
2779 }
2780 if (con->state == CON_STATE_PREOPEN) {
2781 dout("%s: con %p PREOPEN\n", __func__, con);
2782 BUG_ON(con->sock);
2783 }
0da5d703 2784
49659416
AE		 2785 ret = try_read(con);
2786 if (ret < 0) {
2787 if (ret == -EAGAIN)
2788 continue;
2789 con->error_msg = "socket error on read";
2790 fault = true;
2791 break;
2792 }
2793
2794 ret = try_write(con);
2795 if (ret < 0) {
2796 if (ret == -EAGAIN)
2797 continue;
2798 con->error_msg = "socket error on write";
2799 fault = true;
2800 }
2801
2802 break; /* If we make it to here, we're done */
3a140a0d 2803 }
b6e7b6a1
AE		 2804 if (fault)
2805 con_fault(con);
9dd4658d 2806 mutex_unlock(&con->mutex);
0da5d703 2807
b6e7b6a1
AE		 2808 if (fault)
2809 con_fault_finish(con);
2810
2811 con->ops->put(con);
31b8006e
SW		 2812}
2813
31b8006e
SW		 2814/*
2815 * Generic error/fault handler. A retry mechanism is used with
2816 * exponential backoff
2817 */
93209264 2818static void con_fault(struct ceph_connection *con)
31b8006e 2819{
28362986 2820 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
3d14c5d2 2821 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2822 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2823 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2824
122070a2 2825 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2826 con->state != CON_STATE_NEGOTIATING &&
2827 con->state != CON_STATE_OPEN);
ec302645 2828
31b8006e 2829 con_close_socket(con);
5e095e8b 2830
c9ffc77a 2831 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2832 dout("fault on LOSSYTX channel, marking CLOSED\n");
2833 con->state = CON_STATE_CLOSED;
93209264 2834 return;
3b5ede07
SW		 2835 }
2836
5e095e8b 2837 if (con->in_msg) {
38941f80
AE		 2838 BUG_ON(con->in_msg->con != con);
2839 con->in_msg->con = NULL;
5e095e8b
SW		 2840 ceph_msg_put(con->in_msg);
2841 con->in_msg = NULL;
36eb71aa 2842 con->ops->put(con);
5e095e8b 2843 }
31b8006e 2844
e80a52d1
SW		 2845 /* Requeue anything that hasn't been acked */
2846 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2847
e76661d0
SW		 2848 /* If there are no messages queued or keepalive pending, place
2849 * the connection in a STANDBY state */
2850 if (list_empty(&con->out_queue) &&
c9ffc77a 2851 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2852 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2853 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2854 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2855 } else {
2856 /* retry after a delay. */
8dacc7da 2857 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2858 if (con->delay == 0)
2859 con->delay = BASE_DELAY_INTERVAL;
2860 else if (con->delay < MAX_DELAY_INTERVAL)
2861 con->delay *= 2;
c9ffc77a 2862 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2863 queue_con(con);
31b8006e 2864 }
31b8006e
SW		 2865}
2866
2867
2868
2869/*
15d9882c 2870 * initialize a new messenger instance
31b8006e 2871 */
15d9882c
AE		 2872void ceph_messenger_init(struct ceph_messenger *msgr,
2873 struct ceph_entity_addr *myaddr,
2874 u32 supported_features,
2875 u32 required_features,
2876 bool nocrc)
31b8006e 2877{
3d14c5d2
YS		 2878 msgr->supported_features = supported_features;
2879 msgr->required_features = required_features;
2880
31b8006e
SW		 2881 spin_lock_init(&msgr->global_seq_lock);
2882
31b8006e
SW		 2883 if (myaddr)
2884 msgr->inst.addr = *myaddr;
2885
2886 /* select a random nonce */
ac8839d7 2887 msgr->inst.addr.type = 0;
103e2d3a 2888 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2889 encode_my_addr(msgr);
15d9882c 2890 msgr->nocrc = nocrc;
31b8006e 2891
a2a32584 2892 atomic_set(&msgr->stopping, 0);
31b8006e 2893
15d9882c 2894 dout("%s %p\n", __func__, msgr);
31b8006e 2895}
15d9882c 2896EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2897
e00de341
SW		 2898static void clear_standby(struct ceph_connection *con)
2899{
2900 /* come back from STANDBY? */
8dacc7da 2901 if (con->state == CON_STATE_STANDBY) {
e00de341 2902 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2903 con->state = CON_STATE_PREOPEN;
e00de341 2904 con->connect_seq++;
c9ffc77a
AE		 2905 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2906 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2907 }
2908}
2909
31b8006e
SW		 2910/*
2911 * Queue up an outgoing message on the given connection.
2912 */
2913void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2914{
31b8006e 2915 /* set src+dst */
dbad185d 2916 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2917 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2918 msg->needs_out_seq = true;
2919
ec302645 2920 mutex_lock(&con->mutex);
92ce034b 2921
8dacc7da 2922 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2923 dout("con_send %p closed, dropping %p\n", con, msg);
2924 ceph_msg_put(msg);
2925 mutex_unlock(&con->mutex);
2926 return;
2927 }
2928
38941f80 2929 BUG_ON(msg->con != NULL);
36eb71aa 2930 msg->con = con->ops->get(con);
92ce034b
AE		 2931 BUG_ON(msg->con == NULL);
2932
31b8006e
SW		 2933 BUG_ON(!list_empty(&msg->list_head));
2934 list_add_tail(&msg->list_head, &con->out_queue);
2935 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2936 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2937 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2938 le32_to_cpu(msg->hdr.front_len),
2939 le32_to_cpu(msg->hdr.middle_len),
2940 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 2941
2942 clear_standby(con);
ec302645 2943 mutex_unlock(&con->mutex);
31b8006e
SW		 2944
2945 /* if there wasn't anything waiting to send before, queue
2946 * new work */
c9ffc77a 2947 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2948 queue_con(con);
2949}
3d14c5d2 2950EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 2951
2952/*
2953 * Revoke a message that was previously queued for send
2954 */
6740a845 2955void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 2956{
6740a845
AE		 2957 struct ceph_connection *con = msg->con;
2958
2959 if (!con)
2960 return; /* Message not in our possession */
2961
ec302645 2962 mutex_lock(&con->mutex);
31b8006e 2963 if (!list_empty(&msg->list_head)) {
38941f80 2964 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 2965 list_del_init(&msg->list_head);
38941f80 2966 BUG_ON(msg->con == NULL);
36eb71aa 2967 msg->con->ops->put(msg->con);
38941f80 2968 msg->con = NULL;
31b8006e 2969 msg->hdr.seq = 0;
38941f80 2970
31b8006e 2971 ceph_msg_put(msg);
ed98adad
SW		 2972 }
2973 if (con->out_msg == msg) {
38941f80 2974 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 2975 con->out_msg = NULL;
31b8006e
SW		 2976 if (con->out_kvec_is_msg) {
2977 con->out_skip = con->out_kvec_bytes;
2978 con->out_kvec_is_msg = false;
2979 }
ed98adad 2980 msg->hdr.seq = 0;
92ce034b
AE		 2981
2982 ceph_msg_put(msg);
31b8006e 2983 }
ec302645 2984 mutex_unlock(&con->mutex);
31b8006e
SW		 2985}
2986
350b1c32 2987/*
0d59ab81 2988 * Revoke a message that we may be reading data into
350b1c32 2989 */
8921d114 2990void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 2991{
8921d114
AE		 2992 struct ceph_connection *con;
2993
2994 BUG_ON(msg == NULL);
2995 if (!msg->con) {
2996 dout("%s msg %p null con\n", __func__, msg);
2997
2998 return; /* Message not in our possession */
2999 }
3000
3001 con = msg->con;
350b1c32 3002 mutex_lock(&con->mutex);
8921d114 3003 if (con->in_msg == msg) {
95c96174
ED		 3004 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3005 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3006 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3007
3008 /* skip rest of message */
8921d114
AE		 3009 dout("%s %p msg %p revoked\n", __func__, con, msg);
3010 con->in_base_pos = con->in_base_pos -
350b1c32 3011 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3012 front_len -
3013 middle_len -
3014 data_len -
350b1c32 3015 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3016 ceph_msg_put(con->in_msg);
3017 con->in_msg = NULL;
3018 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3019 con->in_seq++;
350b1c32 3020 } else {
8921d114
AE		 3021 dout("%s %p in_msg %p msg %p no-op\n",
3022 __func__, con, con->in_msg, msg);
350b1c32
SW		 3023 }
3024 mutex_unlock(&con->mutex);
3025}
3026
31b8006e
SW		 3027/*
3028 * Queue a keepalive byte to ensure the tcp connection is alive.
3029 */
3030void ceph_con_keepalive(struct ceph_connection *con)
3031{
e00de341 3032 dout("con_keepalive %p\n", con);
00650931 3033 mutex_lock(&con->mutex);
e00de341 3034 clear_standby(con);
00650931 3035 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3036 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3037 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3038 queue_con(con);
3039}
3d14c5d2 3040EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3041
6644ed7b 3042static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3043{
6644ed7b
AE		 3044 struct ceph_msg_data *data;
3045
3046 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3047 return NULL;
3048
81b36be4 3049 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
6644ed7b
AE		 3050 if (data)
3051 data->type = type;
5240d9f9 3052 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3053
3054 return data;
3055}
3056
3057static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3058{
3059 if (!data)
3060 return;
3061
5240d9f9 3062 WARN_ON(!list_empty(&data->links));
6644ed7b
AE		 3063 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3064 ceph_pagelist_release(data->pagelist);
3065 kfree(data->pagelist);
3066 }
81b36be4 3067 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3068}
3069
90af3602 3070void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3071 size_t length, size_t alignment)
02afca6c 3072{
6644ed7b
AE		 3073 struct ceph_msg_data *data;
3074
07aa1558
AE		 3075 BUG_ON(!pages);
3076 BUG_ON(!length);
6644ed7b
AE		 3077
3078 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3079 BUG_ON(!data);
3080 data->pages = pages;
3081 data->length = length;
3082 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3083
5240d9f9
AE		 3084 list_add_tail(&data->links, &msg->data);
3085 msg->data_length += length;
02afca6c 3086}
90af3602 3087EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3088
90af3602 3089void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3090 struct ceph_pagelist *pagelist)
3091{
6644ed7b
AE		 3092 struct ceph_msg_data *data;
3093
07aa1558
AE		 3094 BUG_ON(!pagelist);
3095 BUG_ON(!pagelist->length);
27fa8385 3096
6644ed7b
AE		 3097 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3098 BUG_ON(!data);
3099 data->pagelist = pagelist;
3100
5240d9f9
AE		 3101 list_add_tail(&data->links, &msg->data);
3102 msg->data_length += pagelist->length;
27fa8385 3103}
90af3602 3104EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3105
ea96571f 3106#ifdef CONFIG_BLOCK
90af3602 3107void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3108 size_t length)
27fa8385 3109{
6644ed7b
AE		 3110 struct ceph_msg_data *data;
3111
07aa1558 3112 BUG_ON(!bio);
27fa8385 3113
6644ed7b
AE		 3114 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3115 BUG_ON(!data);
3116 data->bio = bio;
c851c495 3117 data->bio_length = length;
6644ed7b 3118
5240d9f9
AE		 3119 list_add_tail(&data->links, &msg->data);
3120 msg->data_length += length;
27fa8385 3121}
90af3602 3122EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3123#endif /* CONFIG_BLOCK */
27fa8385 3124
31b8006e
SW		 3125/*
3126 * construct a new message with given type, size
3127 * the new msg has a ref count of 1.
3128 */
b61c2763
SW		 3129struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3130 bool can_fail)
31b8006e
SW		 3131{
3132 struct ceph_msg *m;
3133
e3d5d638 3134 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3135 if (m == NULL)
3136 goto out;
31b8006e
SW		 3137
3138 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3139 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3140 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3141
9516e45b
AE		 3142 INIT_LIST_HEAD(&m->list_head);
3143 kref_init(&m->kref);
5240d9f9 3144 INIT_LIST_HEAD(&m->data);
ca20892d 3145
31b8006e 3146 /* front */
842a5780 3147 m->front_alloc_len = front_len;
31b8006e
SW		 3148 if (front_len) {
3149 if (front_len > PAGE_CACHE_SIZE) {
34d23762 3150 m->front.iov_base = __vmalloc(front_len, flags,
31b8006e
SW		 3151 PAGE_KERNEL);
3152 m->front_is_vmalloc = true;
3153 } else {
34d23762 3154 m->front.iov_base = kmalloc(front_len, flags);
31b8006e
SW		 3155 }
3156 if (m->front.iov_base == NULL) {
b61c2763 3157 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3158 front_len);
3159 goto out2;
3160 }
3161 } else {
3162 m->front.iov_base = NULL;
3163 }
3164 m->front.iov_len = front_len;
3165
bb257664 3166 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3167 return m;
3168
3169out2:
3170 ceph_msg_put(m);
3171out:
b61c2763
SW		 3172 if (!can_fail) {
3173 pr_err("msg_new can't create type %d front %d\n", type,
3174 front_len);
f0ed1b7c 3175 WARN_ON(1);
b61c2763
SW		 3176 } else {
3177 dout("msg_new can't create type %d front %d\n", type,
3178 front_len);
3179 }
a79832f2 3180 return NULL;
31b8006e 3181}
3d14c5d2 3182EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3183
31b8006e
SW		 3184/*
3185 * Allocate "middle" portion of a message, if it is needed and wasn't
3186 * allocated by alloc_msg. This allows us to read a small fixed-size
3187 * per-type header in the front and then gracefully fail (i.e.,
3188 * propagate the error to the caller based on info in the front) when
3189 * the middle is too large.
3190 */
2450418c 3191static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3192{
3193 int type = le16_to_cpu(msg->hdr.type);
3194 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3195
3196 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3197 ceph_msg_type_name(type), middle_len);
3198 BUG_ON(!middle_len);
3199 BUG_ON(msg->middle);
3200
b6c1d5b8 3201 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3202 if (!msg->middle)
3203 return -ENOMEM;
3204 return 0;
3205}
3206
2450418c 3207/*
1c20f2d2
AE		 3208 * Allocate a message for receiving an incoming message on a
3209 * connection, and save the result in con->in_msg. Uses the
3210 * connection's private alloc_msg op if available.
3211 *
4740a623
SW		 3212 * Returns 0 on success, or a negative error code.
3213 *
3214 * On success, if we set *skip = 1:
3215 * - the next message should be skipped and ignored.
3216 * - con->in_msg == NULL
3217 * or if we set *skip = 0:
3218 * - con->in_msg is non-null.
3219 * On error (ENOMEM, EAGAIN, ...),
3220 * - con->in_msg == NULL
2450418c 3221 */
4740a623 3222static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3223{
4740a623 3224 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3225 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3226 struct ceph_msg *msg;
4740a623 3227 int ret = 0;
2450418c 3228
1c20f2d2 3229 BUG_ON(con->in_msg != NULL);
53ded495 3230 BUG_ON(!con->ops->alloc_msg);
2450418c 3231
53ded495
AE		 3232 mutex_unlock(&con->mutex);
3233 msg = con->ops->alloc_msg(con, hdr, skip);
3234 mutex_lock(&con->mutex);
3235 if (con->state != CON_STATE_OPEN) {
3236 if (msg)
1d866d1c 3237 ceph_msg_put(msg);
53ded495
AE		 3238 return -EAGAIN;
3239 }
4137577a
AE		 3240 if (msg) {
3241 BUG_ON(*skip);
3242 con->in_msg = msg;
36eb71aa 3243 con->in_msg->con = con->ops->get(con);
92ce034b 3244 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3245 } else {
3246 /*
3247 * Null message pointer means either we should skip
3248 * this message or we couldn't allocate memory. The
3249 * former is not an error.
3250 */
3251 if (*skip)
3252 return 0;
3253 con->error_msg = "error allocating memory for incoming message";
3254
53ded495 3255 return -ENOMEM;
2450418c 3256 }
1c20f2d2 3257 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3258
1c20f2d2
AE		 3259 if (middle_len && !con->in_msg->middle) {
3260 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3261 if (ret < 0) {
1c20f2d2
AE		 3262 ceph_msg_put(con->in_msg);
3263 con->in_msg = NULL;
2450418c
YS		 3264 }
3265 }
9d7f0f13 3266
4740a623 3267 return ret;
2450418c
YS		 3268}
3269
31b8006e
SW		 3270
3271/*
3272 * Free a generically kmalloc'd message.
3273 */
3274void ceph_msg_kfree(struct ceph_msg *m)
3275{
3276 dout("msg_kfree %p\n", m);
3277 if (m->front_is_vmalloc)
3278 vfree(m->front.iov_base);
3279 else
3280 kfree(m->front.iov_base);
e3d5d638 3281 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3282}
3283
3284/*
3285 * Drop a msg ref. Destroy as needed.
3286 */
c2e552e7
SW		 3287void ceph_msg_last_put(struct kref *kref)
3288{
3289 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
5240d9f9
AE		 3290 LIST_HEAD(data);
3291 struct list_head *links;
3292 struct list_head *next;
31b8006e 3293
c2e552e7
SW		 3294 dout("ceph_msg_put last one on %p\n", m);
3295 WARN_ON(!list_empty(&m->list_head));
3296
3297 /* drop middle, data, if any */
3298 if (m->middle) {
3299 ceph_buffer_put(m->middle);
3300 m->middle = NULL;
31b8006e 3301 }
5240d9f9
AE		 3302
3303 list_splice_init(&m->data, &data);
3304 list_for_each_safe(links, next, &data) {
3305 struct ceph_msg_data *data;
3306
3307 data = list_entry(links, struct ceph_msg_data, links);
3308 list_del_init(links);
3309 ceph_msg_data_destroy(data);
3310 }
a1930804 3311 m->data_length = 0;
58bb3b37 3312
c2e552e7
SW		 3313 if (m->pool)
3314 ceph_msgpool_put(m->pool, m);
3315 else
3316 ceph_msg_kfree(m);
31b8006e 3317}
3d14c5d2 3318EXPORT_SYMBOL(ceph_msg_last_put);
9ec7cab1
SW		 3319
3320void ceph_msg_dump(struct ceph_msg *msg)
3321{
842a5780
ID		 3322 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3323 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3324 print_hex_dump(KERN_DEBUG, "header: ",
3325 DUMP_PREFIX_OFFSET, 16, 1,
3326 &msg->hdr, sizeof(msg->hdr), true);
3327 print_hex_dump(KERN_DEBUG, " front: ",
3328 DUMP_PREFIX_OFFSET, 16, 1,
3329 msg->front.iov_base, msg->front.iov_len, true);
3330 if (msg->middle)
3331 print_hex_dump(KERN_DEBUG, "middle: ",
3332 DUMP_PREFIX_OFFSET, 16, 1,
3333 msg->middle->vec.iov_base,
3334 msg->middle->vec.iov_len, true);
3335 print_hex_dump(KERN_DEBUG, "footer: ",
3336 DUMP_PREFIX_OFFSET, 16, 1,
3337 &msg->footer, sizeof(msg->footer), true);
3338}
3d14c5d2 3339EXPORT_SYMBOL(ceph_msg_dump);
kernel source for telekom puls (alcatel/mediatek)