projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'v3.10.107' 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>
edaad97b 9#include <linux/sched.h>
5a0e3ad6 10#include <linux/slab.h>
31b8006e
SW		 11#include <linux/socket.h>
12#include <linux/string.h>
3ebc21f7 13#ifdef CONFIG_BLOCK
68b4476b 14#include <linux/bio.h>
3ebc21f7 15#endif /* CONFIG_BLOCK */
ee3b56f2 16#include <linux/dns_resolver.h>
31b8006e
SW		 17#include <net/tcp.h>
18
3d14c5d2
YS		 19#include <linux/ceph/libceph.h>
20#include <linux/ceph/messenger.h>
21#include <linux/ceph/decode.h>
22#include <linux/ceph/pagelist.h>
bc3b2d7f 23#include <linux/export.h>
31b8006e 24
fe38a2b6
AE		 25#define list_entry_next(pos, member) \
26 list_entry(pos->member.next, typeof(*pos), member)
27
31b8006e
SW		 28/*
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
34 * the sender.
35 */
36
bc18f4b1
AE		 37/*
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
41 * unexpected state.
42 *
43 * --------
44 * | NEW* | transient initial state
45 * --------
46 * | con_sock_state_init()
47 * v
48 * ----------
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
51 * ^ \
52 * | \ con_sock_state_connecting()
53 * | ----------------------
54 * | \
55 * + con_sock_state_closed() \
fbb85a47
SW		 56 * |+--------------------------- \
57 * | \ \ \
58 * | ----------- \ \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
61 * | ^ \ |
62 * | | \ |
63 * | + con_sock_state_closing() \ |
64 * | / \ | |
65 * | / --------------- | |
66 * | / \ v v
bc18f4b1
AE		 67 * | / --------------
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
71 * | | v
72 * -------------
73 * | CONNECTED | TCP connection established
74 * -------------
75 *
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77 */
ce2c8903
AE		 78
79#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
84
8dacc7da
SW		 85/*
86 * connection states
87 */
88#define CON_STATE_CLOSED 1 /* -> PREOPEN */
89#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
94
4a861692
SW		 95/*
96 * ceph_connection flag bits
97 */
98#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 104
c9ffc77a
AE		 105static bool con_flag_valid(unsigned long con_flag)
106{
107 switch (con_flag) {
108 case CON_FLAG_LOSSYTX:
109 case CON_FLAG_KEEPALIVE_PENDING:
110 case CON_FLAG_WRITE_PENDING:
111 case CON_FLAG_SOCK_CLOSED:
112 case CON_FLAG_BACKOFF:
113 return true;
114 default:
115 return false;
116 }
117}
118
119static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120{
121 BUG_ON(!con_flag_valid(con_flag));
122
123 clear_bit(con_flag, &con->flags);
124}
125
126static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127{
128 BUG_ON(!con_flag_valid(con_flag));
129
130 set_bit(con_flag, &con->flags);
131}
132
133static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134{
135 BUG_ON(!con_flag_valid(con_flag));
136
137 return test_bit(con_flag, &con->flags);
138}
139
140static bool con_flag_test_and_clear(struct ceph_connection *con,
141 unsigned long con_flag)
142{
143 BUG_ON(!con_flag_valid(con_flag));
144
145 return test_and_clear_bit(con_flag, &con->flags);
146}
147
148static bool con_flag_test_and_set(struct ceph_connection *con,
149 unsigned long con_flag)
150{
151 BUG_ON(!con_flag_valid(con_flag));
152
153 return test_and_set_bit(con_flag, &con->flags);
154}
155
e3d5d638
AE		 156/* Slab caches for frequently-allocated structures */
157
158static struct kmem_cache *ceph_msg_cache;
81b36be4 159static struct kmem_cache *ceph_msg_data_cache;
e3d5d638 160
31b8006e
SW		 161/* static tag bytes (protocol control messages) */
162static char tag_msg = CEPH_MSGR_TAG_MSG;
163static char tag_ack = CEPH_MSGR_TAG_ACK;
164static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
165
a6a5349d
SW		 166#ifdef CONFIG_LOCKDEP
167static struct lock_class_key socket_class;
168#endif
169
84495f49
AE		 170/*
171 * When skipping (ignoring) a block of input we read it into a "skip
172 * buffer," which is this many bytes in size.
173 */
174#define SKIP_BUF_SIZE 1024
31b8006e
SW		 175
176static void queue_con(struct ceph_connection *con);
177static void con_work(struct work_struct *);
93209264 178static void con_fault(struct ceph_connection *con);
31b8006e 179
31b8006e 180/*
f64a9317
AE		 181 * Nicely render a sockaddr as a string. An array of formatted
182 * strings is used, to approximate reentrancy.
31b8006e 183 */
f64a9317
AE		 184#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
185#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
186#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
187#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
188
189static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
190static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 191
57666519 192static struct page *zero_page; /* used in certain error cases */
57666519 193
3d14c5d2 194const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 195{
196 int i;
197 char *s;
99f0f3b2
AE		 198 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
199 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 200
f64a9317 201 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 202 s = addr_str[i];
203
204 switch (ss->ss_family) {
205 case AF_INET:
bd406145
AE		 206 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
207 ntohs(in4->sin_port));
31b8006e
SW		 208 break;
209
210 case AF_INET6:
bd406145
AE		 211 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
212 ntohs(in6->sin6_port));
31b8006e
SW		 213 break;
214
215 default:
d3002b97
AE		 216 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
217 ss->ss_family);
31b8006e
SW		 218 }
219
220 return s;
221}
3d14c5d2 222EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 223
63f2d211
SW		 224static void encode_my_addr(struct ceph_messenger *msgr)
225{
226 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
227 ceph_encode_addr(&msgr->my_enc_addr);
228}
229
31b8006e
SW		 230/*
231 * work queue for all reading and writing to/from the socket.
232 */
e0f43c94 233static struct workqueue_struct *ceph_msgr_wq;
31b8006e 234
e3d5d638
AE		 235static int ceph_msgr_slab_init(void)
236{
237 BUG_ON(ceph_msg_cache);
238 ceph_msg_cache = kmem_cache_create("ceph_msg",
239 sizeof (struct ceph_msg),
240 __alignof__(struct ceph_msg), 0, NULL);
81b36be4
AE		 241
242 if (!ceph_msg_cache)
243 return -ENOMEM;
244
245 BUG_ON(ceph_msg_data_cache);
246 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
247 sizeof (struct ceph_msg_data),
248 __alignof__(struct ceph_msg_data),
249 0, NULL);
250 if (ceph_msg_data_cache)
251 return 0;
252
253 kmem_cache_destroy(ceph_msg_cache);
254 ceph_msg_cache = NULL;
255
256 return -ENOMEM;
e3d5d638
AE		 257}
258
259static void ceph_msgr_slab_exit(void)
260{
81b36be4
AE		 261 BUG_ON(!ceph_msg_data_cache);
262 kmem_cache_destroy(ceph_msg_data_cache);
263 ceph_msg_data_cache = NULL;
264
e3d5d638
AE		 265 BUG_ON(!ceph_msg_cache);
266 kmem_cache_destroy(ceph_msg_cache);
267 ceph_msg_cache = NULL;
268}
269
15417167 270static void _ceph_msgr_exit(void)
6173d1f0 271{
d3002b97 272 if (ceph_msgr_wq) {
6173d1f0 273 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 274 ceph_msgr_wq = NULL;
275 }
6173d1f0 276
e3d5d638
AE		 277 ceph_msgr_slab_exit();
278
6173d1f0
AE		 279 BUG_ON(zero_page == NULL);
280 kunmap(zero_page);
281 page_cache_release(zero_page);
282 zero_page = NULL;
283}
284
3d14c5d2 285int ceph_msgr_init(void)
31b8006e 286{
57666519
AE		 287 BUG_ON(zero_page != NULL);
288 zero_page = ZERO_PAGE(0);
289 page_cache_get(zero_page);
290
e3d5d638
AE		 291 if (ceph_msgr_slab_init())
292 return -ENOMEM;
293
f05c0daa
ID		 294 ceph_msgr_wq = alloc_workqueue("ceph-msgr",
295 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
6173d1f0
AE		 296 if (ceph_msgr_wq)
297 return 0;
57666519 298
6173d1f0
AE		 299 pr_err("msgr_init failed to create workqueue\n");
300 _ceph_msgr_exit();
57666519 301
6173d1f0 302 return -ENOMEM;
31b8006e 303}
3d14c5d2 304EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 305
306void ceph_msgr_exit(void)
307{
57666519 308 BUG_ON(ceph_msgr_wq == NULL);
57666519 309
6173d1f0 310 _ceph_msgr_exit();
31b8006e 311}
3d14c5d2 312EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 313
cd84db6e 314void ceph_msgr_flush(void)
a922d38f
SW		 315{
316 flush_workqueue(ceph_msgr_wq);
317}
3d14c5d2 318EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 319
ce2c8903
AE		 320/* Connection socket state transition functions */
321
322static void con_sock_state_init(struct ceph_connection *con)
323{
324 int old_state;
325
326 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
327 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
328 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 329 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
330 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 331}
332
333static void con_sock_state_connecting(struct ceph_connection *con)
334{
335 int old_state;
336
337 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
338 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
339 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 340 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
341 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 342}
343
344static void con_sock_state_connected(struct ceph_connection *con)
345{
346 int old_state;
347
348 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
349 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
350 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 351 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
352 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 353}
354
355static void con_sock_state_closing(struct ceph_connection *con)
356{
357 int old_state;
358
359 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
360 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
361 old_state != CON_SOCK_STATE_CONNECTED &&
362 old_state != CON_SOCK_STATE_CLOSING))
363 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 364 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
365 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 366}
367
368static void con_sock_state_closed(struct ceph_connection *con)
369{
370 int old_state;
371
372 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
373 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 374 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 375 old_state != CON_SOCK_STATE_CONNECTING &&
376 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 377 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 378 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
379 CON_SOCK_STATE_CLOSED);
ce2c8903 380}
a922d38f 381
31b8006e
SW		 382/*
383 * socket callback functions
384 */
385
386/* data available on socket, or listen socket received a connect */
327800bd 387static void ceph_sock_data_ready(struct sock *sk, int count_unused)
31b8006e 388{
bd406145 389 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 390 if (atomic_read(&con->msgr->stopping)) {
391 return;
392 }
bd406145 393
31b8006e 394 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 395 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 396 con, con->state);
397 queue_con(con);
398 }
399}
400
401/* socket has buffer space for writing */
327800bd 402static void ceph_sock_write_space(struct sock *sk)
31b8006e 403{
d3002b97 404 struct ceph_connection *con = sk->sk_user_data;
31b8006e 405
182fac26
JS		 406 /* only queue to workqueue if there is data we want to write,
407 * and there is sufficient space in the socket buffer to accept
327800bd 408 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 409 * doesn't get called again until try_write() fills the socket
410 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
411 * and net/core/stream.c:sk_stream_write_space().
412 */
c9ffc77a 413 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
182fac26 414 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
327800bd 415 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 416 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
417 queue_con(con);
418 }
31b8006e 419 } else {
327800bd 420 dout("%s %p nothing to write\n", __func__, con);
31b8006e 421 }
31b8006e
SW		 422}
423
424/* socket's state has changed */
327800bd 425static void ceph_sock_state_change(struct sock *sk)
31b8006e 426{
bd406145 427 struct ceph_connection *con = sk->sk_user_data;
31b8006e 428
327800bd 429 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 430 con, con->state, sk->sk_state);
431
31b8006e
SW		 432 switch (sk->sk_state) {
433 case TCP_CLOSE:
327800bd 434 dout("%s TCP_CLOSE\n", __func__);
31b8006e 435 case TCP_CLOSE_WAIT:
327800bd 436 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 437 con_sock_state_closing(con);
c9ffc77a 438 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 439 queue_con(con);
31b8006e
SW		 440 break;
441 case TCP_ESTABLISHED:
327800bd 442 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 443 con_sock_state_connected(con);
31b8006e
SW		 444 queue_con(con);
445 break;
d3002b97
AE		 446 default: /* Everything else is uninteresting */
447 break;
31b8006e
SW		 448 }
449}
450
451/*
452 * set up socket callbacks
453 */
454static void set_sock_callbacks(struct socket *sock,
455 struct ceph_connection *con)
456{
457 struct sock *sk = sock->sk;
bd406145 458 sk->sk_user_data = con;
327800bd
AE		 459 sk->sk_data_ready = ceph_sock_data_ready;
460 sk->sk_write_space = ceph_sock_write_space;
461 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 462}
463
464
465/*
466 * socket helpers
467 */
468
469/*
470 * initiate connection to a remote socket.
471 */
41617d0c 472static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 473{
f91d3471 474 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e 475 struct socket *sock;
edaad97b 476 unsigned int noio_flag;
31b8006e
SW		 477 int ret;
478
479 BUG_ON(con->sock);
edaad97b
ID		 480
481 /* sock_create_kern() allocates with GFP_KERNEL */
482 noio_flag = memalloc_noio_save();
f91d3471
SW		 483 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
484 IPPROTO_TCP, &sock);
edaad97b 485 memalloc_noio_restore(noio_flag);
31b8006e 486 if (ret)
41617d0c 487 return ret;
31b8006e
SW		 488 sock->sk->sk_allocation = GFP_NOFS;
489
a6a5349d
SW		 490#ifdef CONFIG_LOCKDEP
491 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
492#endif
493
31b8006e
SW		 494 set_sock_callbacks(sock, con);
495
3d14c5d2 496 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 497
89a86be0 498 con_sock_state_connecting(con);
f91d3471
SW		 499 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
500 O_NONBLOCK);
31b8006e
SW		 501 if (ret == -EINPROGRESS) {
502 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 503 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 504 sock->sk->sk_state);
a5bc3129 505 } else if (ret < 0) {
31b8006e 506 pr_err("connect %s error %d\n",
3d14c5d2 507 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 508 sock_release(sock);
31b8006e 509 con->error_msg = "connect error";
31b8006e 510
41617d0c 511 return ret;
a5bc3129
AE		 512 }
513 con->sock = sock;
41617d0c 514 return 0;
31b8006e
SW		 515}
516
517static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
518{
519 struct kvec iov = {buf, len};
520 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 521 int r;
31b8006e 522
98bdb0aa
SW		 523 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
524 if (r == -EAGAIN)
525 r = 0;
526 return r;
31b8006e
SW		 527}
528
afb3d90e
AE		 529static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
530 int page_offset, size_t length)
531{
532 void *kaddr;
533 int ret;
534
535 BUG_ON(page_offset + length > PAGE_SIZE);
536
537 kaddr = kmap(page);
538 BUG_ON(!kaddr);
539 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
540 kunmap(page);
541
542 return ret;
543}
544
31b8006e
SW		 545/*
546 * write something. @more is true if caller will be sending more data
547 * shortly.
548 */
549static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
550 size_t kvlen, size_t len, int more)
551{
552 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 553 int r;
31b8006e
SW		 554
555 if (more)
556 msg.msg_flags |= MSG_MORE;
557 else
558 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
559
42961d23
SW		 560 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
561 if (r == -EAGAIN)
562 r = 0;
563 return r;
31b8006e
SW		 564}
565
a757a4e2 566static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 567 int offset, size_t size, bool more)
31739139
AE		 568{
569 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
570 int ret;
571
572 ret = kernel_sendpage(sock, page, offset, size, flags);
573 if (ret == -EAGAIN)
574 ret = 0;
575
576 return ret;
577}
578
a757a4e2
CC		 579static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
580 int offset, size_t size, bool more)
581{
582 int ret;
583 struct kvec iov;
584
585 /* sendpage cannot properly handle pages with page_count == 0,
586 * we need to fallback to sendmsg if that's the case */
587 if (page_count(page) >= 1)
588 return __ceph_tcp_sendpage(sock, page, offset, size, more);
589
590 iov.iov_base = kmap(page) + offset;
591 iov.iov_len = size;
592 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
593 kunmap(page);
594
595 return ret;
596}
31b8006e
SW		 597
598/*
599 * Shutdown/close the socket for the given connection.
600 */
601static int con_close_socket(struct ceph_connection *con)
602{
8007b8d6 603 int rc = 0;
31b8006e
SW		 604
605 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 606 if (con->sock) {
607 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
608 sock_release(con->sock);
609 con->sock = NULL;
610 }
456ea468
AE		 611
612 /*
4a861692 613 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 614 * independent of the connection mutex, and we could have
615 * received a socket close event before we had the chance to
616 * shut the socket down.
617 */
c9ffc77a 618 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 619
ce2c8903 620 con_sock_state_closed(con);
31b8006e
SW		 621 return rc;
622}
623
624/*
625 * Reset a connection. Discard all incoming and outgoing messages
626 * and clear *_seq state.
627 */
628static void ceph_msg_remove(struct ceph_msg *msg)
629{
630 list_del_init(&msg->list_head);
38941f80 631 BUG_ON(msg->con == NULL);
36eb71aa 632 msg->con->ops->put(msg->con);
38941f80
AE		 633 msg->con = NULL;
634
31b8006e
SW		 635 ceph_msg_put(msg);
636}
637static void ceph_msg_remove_list(struct list_head *head)
638{
639 while (!list_empty(head)) {
640 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
641 list_head);
642 ceph_msg_remove(msg);
643 }
644}
645
646static void reset_connection(struct ceph_connection *con)
647{
648 /* reset connection, out_queue, msg_ and connect_seq */
649 /* discard existing out_queue and msg_seq */
0fa6ebc6 650 dout("reset_connection %p\n", con);
31b8006e
SW		 651 ceph_msg_remove_list(&con->out_queue);
652 ceph_msg_remove_list(&con->out_sent);
653
cf3e5c40 654 if (con->in_msg) {
38941f80
AE		 655 BUG_ON(con->in_msg->con != con);
656 con->in_msg->con = NULL;
cf3e5c40
SW		 657 ceph_msg_put(con->in_msg);
658 con->in_msg = NULL;
36eb71aa 659 con->ops->put(con);
cf3e5c40
SW		 660 }
661
31b8006e
SW		 662 con->connect_seq = 0;
663 con->out_seq = 0;
c86a2930
SW		 664 if (con->out_msg) {
665 ceph_msg_put(con->out_msg);
666 con->out_msg = NULL;
667 }
31b8006e 668 con->in_seq = 0;
0e0d5e0c 669 con->in_seq_acked = 0;
31b8006e
SW		 670}
671
672/*
673 * mark a peer down. drop any open connections.
674 */
675void ceph_con_close(struct ceph_connection *con)
676{
8c50c817 677 mutex_lock(&con->mutex);
3d14c5d2
YS		 678 dout("con_close %p peer %s\n", con,
679 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 680 con->state = CON_STATE_CLOSED;
a5988c49 681
c9ffc77a
AE		 682 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
683 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
684 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
685 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 686
31b8006e 687 reset_connection(con);
6f2bc3ff 688 con->peer_global_seq = 0;
91e45ce3 689 cancel_delayed_work(&con->work);
ee76e073 690 con_close_socket(con);
ec302645 691 mutex_unlock(&con->mutex);
31b8006e 692}
3d14c5d2 693EXPORT_SYMBOL(ceph_con_close);
31b8006e 694
31b8006e
SW		 695/*
696 * Reopen a closed connection, with a new peer address.
697 */
b7a9e5dd
SW		 698void ceph_con_open(struct ceph_connection *con,
699 __u8 entity_type, __u64 entity_num,
700 struct ceph_entity_addr *addr)
31b8006e 701{
5469155f 702 mutex_lock(&con->mutex);
3d14c5d2 703 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 704
122070a2 705 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 706 con->state = CON_STATE_PREOPEN;
a5988c49 707
b7a9e5dd
SW		 708 con->peer_name.type = (__u8) entity_type;
709 con->peer_name.num = cpu_to_le64(entity_num);
710
31b8006e 711 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 712 con->delay = 0; /* reset backoff memory */
5469155f 713 mutex_unlock(&con->mutex);
31b8006e
SW		 714 queue_con(con);
715}
3d14c5d2 716EXPORT_SYMBOL(ceph_con_open);
31b8006e 717
87b315a5
SW		 718/*
719 * return true if this connection ever successfully opened
720 */
721bool ceph_con_opened(struct ceph_connection *con)
722{
723 return con->connect_seq > 0;
724}
725
31b8006e
SW		 726/*
727 * initialize a new connection.
728 */
1bfd89f4
AE		 729void ceph_con_init(struct ceph_connection *con, void *private,
730 const struct ceph_connection_operations *ops,
b7a9e5dd 731 struct ceph_messenger *msgr)
31b8006e
SW		 732{
733 dout("con_init %p\n", con);
734 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 735 con->private = private;
736 con->ops = ops;
31b8006e 737 con->msgr = msgr;
ce2c8903
AE		 738
739 con_sock_state_init(con);
740
ec302645 741 mutex_init(&con->mutex);
31b8006e
SW		 742 INIT_LIST_HEAD(&con->out_queue);
743 INIT_LIST_HEAD(&con->out_sent);
744 INIT_DELAYED_WORK(&con->work, con_work);
a5988c49 745
8dacc7da 746 con->state = CON_STATE_CLOSED;
31b8006e 747}
3d14c5d2 748EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 749
750
751/*
752 * We maintain a global counter to order connection attempts. Get
753 * a unique seq greater than @gt.
754 */
755static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
756{
757 u32 ret;
758
759 spin_lock(&msgr->global_seq_lock);
760 if (msgr->global_seq < gt)
761 msgr->global_seq = gt;
762 ret = ++msgr->global_seq;
763 spin_unlock(&msgr->global_seq_lock);
764 return ret;
765}
766
e2200423 767static void con_out_kvec_reset(struct ceph_connection *con)
859eb799
AE		 768{
769 con->out_kvec_left = 0;
770 con->out_kvec_bytes = 0;
771 con->out_kvec_cur = &con->out_kvec[0];
772}
773
e2200423 774static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 775 size_t size, void *data)
776{
777 int index;
778
779 index = con->out_kvec_left;
780 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
781
782 con->out_kvec[index].iov_len = size;
783 con->out_kvec[index].iov_base = data;
784 con->out_kvec_left++;
785 con->out_kvec_bytes += size;
786}
31b8006e 787
df6ad1f9 788#ifdef CONFIG_BLOCK
6aaa4511
AE		 789
790/*
791 * For a bio data item, a piece is whatever remains of the next
792 * entry in the current bio iovec, or the first entry in the next
793 * bio in the list.
794 */
8ae4f4f5 795static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 796 size_t length)
6aaa4511 797{
8ae4f4f5 798 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 799 struct bio *bio;
800
801 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
802
803 bio = data->bio;
804 BUG_ON(!bio);
805 BUG_ON(!bio->bi_vcnt);
6aaa4511 806
ca8b3a69 807 cursor->resid = min(length, data->bio_length);
6aaa4511
AE		 808 cursor->bio = bio;
809 cursor->vector_index = 0;
810 cursor->vector_offset = 0;
25aff7c5 811 cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
6aaa4511
AE		 812}
813
8ae4f4f5 814static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 815 size_t *page_offset,
816 size_t *length)
817{
8ae4f4f5 818 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 819 struct bio *bio;
820 struct bio_vec *bio_vec;
821 unsigned int index;
822
823 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
824
825 bio = cursor->bio;
826 BUG_ON(!bio);
827
828 index = cursor->vector_index;
829 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
830
831 bio_vec = &bio->bi_io_vec[index];
832 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
833 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
834 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 835 if (cursor->last_piece) /* pagelist offset is always 0 */
836 *length = cursor->resid;
837 else
838 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
25aff7c5 839 BUG_ON(*length > cursor->resid);
5df521b1 840 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511
AE		 841
842 return bio_vec->bv_page;
843}
844
8ae4f4f5
AE		 845static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
846 size_t bytes)
6aaa4511 847{
6aaa4511
AE		 848 struct bio *bio;
849 struct bio_vec *bio_vec;
850 unsigned int index;
851
8ae4f4f5 852 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
6aaa4511
AE		 853
854 bio = cursor->bio;
855 BUG_ON(!bio);
856
857 index = cursor->vector_index;
858 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
859 bio_vec = &bio->bi_io_vec[index];
6aaa4511
AE		 860
861 /* Advance the cursor offset */
862
25aff7c5
AE		 863 BUG_ON(cursor->resid < bytes);
864 cursor->resid -= bytes;
6aaa4511
AE		 865 cursor->vector_offset += bytes;
866 if (cursor->vector_offset < bio_vec->bv_len)
867 return false; /* more bytes to process in this segment */
25aff7c5 868 BUG_ON(cursor->vector_offset != bio_vec->bv_len);
6aaa4511
AE		 869
870 /* Move on to the next segment, and possibly the next bio */
871
25aff7c5 872 if (++index == (unsigned int) bio->bi_vcnt) {
6aaa4511 873 bio = bio->bi_next;
25aff7c5 874 index = 0;
6aaa4511 875 }
25aff7c5
AE		 876 cursor->bio = bio;
877 cursor->vector_index = index;
6aaa4511
AE		 878 cursor->vector_offset = 0;
879
25aff7c5
AE		 880 if (!cursor->last_piece) {
881 BUG_ON(!cursor->resid);
882 BUG_ON(!bio);
883 /* A short read is OK, so use <= rather than == */
884 if (cursor->resid <= bio->bi_io_vec[index].bv_len)
6aaa4511 885 cursor->last_piece = true;
25aff7c5 886 }
6aaa4511
AE		 887
888 return true;
889}
ea96571f 890#endif /* CONFIG_BLOCK */
df6ad1f9 891
e766d7b5
AE		 892/*
893 * For a page array, a piece comes from the first page in the array
894 * that has not already been fully consumed.
895 */
8ae4f4f5 896static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 897 size_t length)
e766d7b5 898{
8ae4f4f5 899 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 900 int page_count;
901
902 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
903
904 BUG_ON(!data->pages);
905 BUG_ON(!data->length);
906
ca8b3a69 907 cursor->resid = min(length, data->length);
e766d7b5 908 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 909 cursor->page_offset = data->alignment & ~PAGE_MASK;
910 cursor->page_index = 0;
56fc5659
AE		 911 BUG_ON(page_count > (int)USHRT_MAX);
912 cursor->page_count = (unsigned short)page_count;
913 BUG_ON(length > SIZE_MAX - cursor->page_offset);
a6489727 914 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 915}
916
8ae4f4f5
AE		 917static struct page *
918ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
919 size_t *page_offset, size_t *length)
e766d7b5 920{
8ae4f4f5 921 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 922
923 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
924
925 BUG_ON(cursor->page_index >= cursor->page_count);
926 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 927
928 *page_offset = cursor->page_offset;
25aff7c5 929 if (cursor->last_piece)
e766d7b5 930 *length = cursor->resid;
25aff7c5 931 else
e766d7b5 932 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 933
934 return data->pages[cursor->page_index];
935}
936
8ae4f4f5 937static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 938 size_t bytes)
939{
8ae4f4f5 940 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 941
942 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 943
944 /* Advance the cursor page offset */
945
946 cursor->resid -= bytes;
5df521b1
AE		 947 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
948 if (!bytes || cursor->page_offset)
e766d7b5
AE		 949 return false; /* more bytes to process in the current page */
950
5df521b1 951 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 952
953 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 954 cursor->page_index++;
25aff7c5 955 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 956
957 return true;
958}
959
fe38a2b6 960/*
dd236fcb
AE		 961 * For a pagelist, a piece is whatever remains to be consumed in the
962 * first page in the list, or the front of the next page.
fe38a2b6 963 */
8ae4f4f5
AE		 964static void
965ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 966 size_t length)
fe38a2b6 967{
8ae4f4f5 968 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 969 struct ceph_pagelist *pagelist;
970 struct page *page;
971
dd236fcb 972 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 973
974 pagelist = data->pagelist;
975 BUG_ON(!pagelist);
25aff7c5
AE		 976
977 if (!length)
fe38a2b6
AE		 978 return; /* pagelist can be assigned but empty */
979
980 BUG_ON(list_empty(&pagelist->head));
981 page = list_first_entry(&pagelist->head, struct page, lru);
982
ca8b3a69 983 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 984 cursor->page = page;
985 cursor->offset = 0;
a51b272e 986 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 987}
988
8ae4f4f5
AE		 989static struct page *
990ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
991 size_t *page_offset, size_t *length)
fe38a2b6 992{
8ae4f4f5 993 struct ceph_msg_data *data = cursor->data;
fe38a2b6 994 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 995
996 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
997
998 pagelist = data->pagelist;
999 BUG_ON(!pagelist);
1000
1001 BUG_ON(!cursor->page);
25aff7c5 1002 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 1003
5df521b1 1004 /* offset of first page in pagelist is always 0 */
fe38a2b6 1005 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 1006 if (cursor->last_piece)
25aff7c5
AE		 1007 *length = cursor->resid;
1008 else
1009 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1010
8ae4f4f5 1011 return cursor->page;
fe38a2b6
AE		 1012}
1013
8ae4f4f5 1014static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1015 size_t bytes)
fe38a2b6 1016{
8ae4f4f5 1017 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1018 struct ceph_pagelist *pagelist;
1019
1020 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1021
1022 pagelist = data->pagelist;
1023 BUG_ON(!pagelist);
25aff7c5
AE		 1024
1025 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1026 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1027
1028 /* Advance the cursor offset */
1029
25aff7c5 1030 cursor->resid -= bytes;
fe38a2b6 1031 cursor->offset += bytes;
5df521b1 1032 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1033 if (!bytes || cursor->offset & ~PAGE_MASK)
1034 return false; /* more bytes to process in the current page */
1035
1036 /* Move on to the next page */
1037
1038 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1039 cursor->page = list_entry_next(cursor->page, lru);
25aff7c5 1040 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1041
1042 return true;
1043}
1044
dd236fcb
AE		 1045/*
1046 * Message data is handled (sent or received) in pieces, where each
1047 * piece resides on a single page. The network layer might not
1048 * consume an entire piece at once. A data item's cursor keeps
1049 * track of which piece is next to process and how much remains to
1050 * be processed in that piece. It also tracks whether the current
1051 * piece is the last one in the data item.
1052 */
ca8b3a69 1053static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1054{
ca8b3a69 1055 size_t length = cursor->total_resid;
8ae4f4f5 1056
8ae4f4f5 1057 switch (cursor->data->type) {
dd236fcb 1058 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1059 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1060 break;
e766d7b5 1061 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1062 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1063 break;
dd236fcb
AE		 1064#ifdef CONFIG_BLOCK
1065 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1066 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1067 break;
dd236fcb 1068#endif /* CONFIG_BLOCK */
6aaa4511 1069 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1070 default:
1071 /* BUG(); */
1072 break;
1073 }
8ae4f4f5 1074 cursor->need_crc = true;
dd236fcb
AE		 1075}
1076
ca8b3a69
AE		 1077static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1078{
1079 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1080 struct ceph_msg_data *data;
1081
1082 BUG_ON(!length);
1083 BUG_ON(length > msg->data_length);
1084 BUG_ON(list_empty(&msg->data));
1085
ca8b3a69
AE		 1086 cursor->data_head = &msg->data;
1087 cursor->total_resid = length;
1088 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1089 cursor->data = data;
1090
1091 __ceph_msg_data_cursor_init(cursor);
1092}
1093
dd236fcb
AE		 1094/*
1095 * Return the page containing the next piece to process for a given
1096 * data item, and supply the page offset and length of that piece.
1097 * Indicate whether this is the last piece in this data item.
1098 */
8ae4f4f5
AE		 1099static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1100 size_t *page_offset, size_t *length,
dd236fcb
AE		 1101 bool *last_piece)
1102{
1103 struct page *page;
1104
8ae4f4f5 1105 switch (cursor->data->type) {
dd236fcb 1106 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1107 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1108 break;
e766d7b5 1109 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1110 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1111 break;
dd236fcb
AE		 1112#ifdef CONFIG_BLOCK
1113 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1114 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1115 break;
dd236fcb 1116#endif /* CONFIG_BLOCK */
6aaa4511 1117 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1118 default:
1119 page = NULL;
1120 break;
1121 }
1122 BUG_ON(!page);
1123 BUG_ON(*page_offset + *length > PAGE_SIZE);
1124 BUG_ON(!*length);
1125 if (last_piece)
8ae4f4f5 1126 *last_piece = cursor->last_piece;
dd236fcb
AE		 1127
1128 return page;
1129}
1130
1131/*
1132 * Returns true if the result moves the cursor on to the next piece
1133 * of the data item.
1134 */
8ae4f4f5
AE		 1135static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1136 size_t bytes)
dd236fcb
AE		 1137{
1138 bool new_piece;
1139
25aff7c5 1140 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1141 switch (cursor->data->type) {
dd236fcb 1142 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1143 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1144 break;
e766d7b5 1145 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1146 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1147 break;
dd236fcb
AE		 1148#ifdef CONFIG_BLOCK
1149 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1150 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1151 break;
dd236fcb 1152#endif /* CONFIG_BLOCK */
6aaa4511 1153 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1154 default:
1155 BUG();
1156 break;
1157 }
ca8b3a69 1158 cursor->total_resid -= bytes;
dd236fcb 1159
ca8b3a69
AE		 1160 if (!cursor->resid && cursor->total_resid) {
1161 WARN_ON(!cursor->last_piece);
1162 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1163 cursor->data = list_entry_next(cursor->data, links);
1164 __ceph_msg_data_cursor_init(cursor);
a51b272e 1165 new_piece = true;
ca8b3a69 1166 }
a51b272e 1167 cursor->need_crc = new_piece;
ca8b3a69 1168
dd236fcb
AE		 1169 return new_piece;
1170}
1171
98fa5dd8 1172static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1173{
739c905b 1174 BUG_ON(!msg);
25aff7c5 1175 BUG_ON(!data_len);
739c905b 1176
4c59b4a2 1177 /* Initialize data cursor */
fe38a2b6 1178
8ae4f4f5 1179 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1180}
1181
31b8006e
SW		 1182/*
1183 * Prepare footer for currently outgoing message, and finish things
1184 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1185 */
859eb799 1186static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1187{
1188 struct ceph_msg *m = con->out_msg;
859eb799 1189 int v = con->out_kvec_left;
31b8006e 1190
fd154f3c
AE		 1191 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1192
31b8006e
SW		 1193 dout("prepare_write_message_footer %p\n", con);
1194 con->out_kvec_is_msg = true;
1195 con->out_kvec[v].iov_base = &m->footer;
1196 con->out_kvec[v].iov_len = sizeof(m->footer);
1197 con->out_kvec_bytes += sizeof(m->footer);
1198 con->out_kvec_left++;
1199 con->out_more = m->more_to_follow;
c86a2930 1200 con->out_msg_done = true;
31b8006e
SW		 1201}
1202
1203/*
1204 * Prepare headers for the next outgoing message.
1205 */
1206static void prepare_write_message(struct ceph_connection *con)
1207{
1208 struct ceph_msg *m;
a9a0c51a 1209 u32 crc;
31b8006e 1210
e2200423 1211 con_out_kvec_reset(con);
31b8006e 1212 con->out_kvec_is_msg = true;
c86a2930 1213 con->out_msg_done = false;
31b8006e
SW		 1214
1215 /* Sneak an ack in there first? If we can get it into the same
1216 * TCP packet that's a good thing. */
1217 if (con->in_seq > con->in_seq_acked) {
1218 con->in_seq_acked = con->in_seq;
e2200423 1219 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1220 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1221 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1222 &con->out_temp_ack);
31b8006e
SW		 1223 }
1224
38941f80 1225 BUG_ON(list_empty(&con->out_queue));
859eb799 1226 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1227 con->out_msg = m;
38941f80 1228 BUG_ON(m->con != con);
4cf9d544
SW		 1229
1230 /* put message on sent list */
1231 ceph_msg_get(m);
1232 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1233
e84346b7
SW		 1234 /*
1235 * only assign outgoing seq # if we haven't sent this message
1236 * yet. if it is requeued, resend with it's original seq.
1237 */
1238 if (m->needs_out_seq) {
1239 m->hdr.seq = cpu_to_le64(++con->out_seq);
1240 m->needs_out_seq = false;
1241 }
98fa5dd8 1242 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1243
98fa5dd8 1244 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1245 m, con->out_seq, le16_to_cpu(m->hdr.type),
1246 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1247 m->data_length);
31b8006e
SW		 1248 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1249
1250 /* tag + hdr + front + middle */
e2200423
AE		 1251 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1252 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1253 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1254
31b8006e 1255 if (m->middle)
e2200423 1256 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1257 m->middle->vec.iov_base);
31b8006e
SW		 1258
1259 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1260 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1261 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1262 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1263
1264 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1265 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1266 if (m->middle) {
1267 crc = crc32c(0, m->middle->vec.iov_base,
1268 m->middle->vec.iov_len);
1269 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1270 } else
31b8006e 1271 con->out_msg->footer.middle_crc = 0;
739c905b 1272 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1273 le32_to_cpu(con->out_msg->footer.front_crc),
1274 le32_to_cpu(con->out_msg->footer.middle_crc));
1275
1276 /* is there a data payload? */
739c905b 1277 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1278 if (m->data_length) {
1279 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1280 con->out_more = 1; /* data + footer will follow */
1281 } else {
31b8006e 1282 /* no, queue up footer too and be done */
859eb799 1283 prepare_write_message_footer(con);
78625051 1284 }
31b8006e 1285
c9ffc77a 1286 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1287}
1288
1289/*
1290 * Prepare an ack.
1291 */
1292static void prepare_write_ack(struct ceph_connection *con)
1293{
1294 dout("prepare_write_ack %p %llu -> %llu\n", con,
1295 con->in_seq_acked, con->in_seq);
1296 con->in_seq_acked = con->in_seq;
1297
e2200423 1298 con_out_kvec_reset(con);
859eb799 1299
e2200423 1300 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1301
31b8006e 1302 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1303 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1304 &con->out_temp_ack);
1305
31b8006e 1306 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1307 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1308}
1309
3a23083b
SW		 1310/*
1311 * Prepare to share the seq during handshake
1312 */
1313static void prepare_write_seq(struct ceph_connection *con)
1314{
1315 dout("prepare_write_seq %p %llu -> %llu\n", con,
1316 con->in_seq_acked, con->in_seq);
1317 con->in_seq_acked = con->in_seq;
1318
1319 con_out_kvec_reset(con);
1320
1321 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1322 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1323 &con->out_temp_ack);
1324
1325 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1326}
1327
31b8006e
SW		 1328/*
1329 * Prepare to write keepalive byte.
1330 */
1331static void prepare_write_keepalive(struct ceph_connection *con)
1332{
1333 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1334 con_out_kvec_reset(con);
1335 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1336 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1337}
1338
1339/*
1340 * Connection negotiation.
1341 */
1342
dac1e716
AE		 1343static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1344 int *auth_proto)
4e7a5dcd 1345{
a3530df3 1346 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1347
1348 if (!con->ops->get_authorizer) {
1349 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1350 con->out_connect.authorizer_len = 0;
729796be 1351 return NULL;
b1c6b980
AE		 1352 }
1353
1354 /* Can't hold the mutex while getting authorizer */
ec302645 1355 mutex_unlock(&con->mutex);
dac1e716 1356 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1357 mutex_lock(&con->mutex);
4e7a5dcd 1358
a3530df3 1359 if (IS_ERR(auth))
729796be 1360 return auth;
8dacc7da 1361 if (con->state != CON_STATE_NEGOTIATING)
729796be 1362 return ERR_PTR(-EAGAIN);
0da5d703 1363
8f43fb53
AE		 1364 con->auth_reply_buf = auth->authorizer_reply_buf;
1365 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1366 return auth;
4e7a5dcd
SW		 1367}
1368
31b8006e
SW		 1369/*
1370 * We connected to a peer and are saying hello.
1371 */
e825a66d 1372static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1373{
e2200423
AE		 1374 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1375 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1376 &con->msgr->my_enc_addr);
eed0ef2c 1377
eed0ef2c 1378 con->out_more = 0;
c9ffc77a 1379 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1380}
1381
e825a66d 1382static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1383{
95c96174 1384 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1385 int proto;
dac1e716 1386 int auth_proto;
729796be 1387 struct ceph_auth_handshake *auth;
31b8006e
SW		 1388
1389 switch (con->peer_name.type) {
1390 case CEPH_ENTITY_TYPE_MON:
1391 proto = CEPH_MONC_PROTOCOL;
1392 break;
1393 case CEPH_ENTITY_TYPE_OSD:
1394 proto = CEPH_OSDC_PROTOCOL;
1395 break;
1396 case CEPH_ENTITY_TYPE_MDS:
1397 proto = CEPH_MDSC_PROTOCOL;
1398 break;
1399 default:
1400 BUG();
1401 }
1402
1403 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1404 con->connect_seq, global_seq, proto);
4e7a5dcd 1405
e825a66d 1406 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1407 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1408 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1409 con->out_connect.global_seq = cpu_to_le32(global_seq);
1410 con->out_connect.protocol_version = cpu_to_le32(proto);
1411 con->out_connect.flags = 0;
31b8006e 1412
dac1e716
AE		 1413 auth_proto = CEPH_AUTH_UNKNOWN;
1414 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1415 if (IS_ERR(auth))
1416 return PTR_ERR(auth);
3da54776 1417
dac1e716 1418 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1419 con->out_connect.authorizer_len = auth ?
1420 cpu_to_le32(auth->authorizer_buf_len) : 0;
1421
e2200423 1422 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1423 &con->out_connect);
1424 if (auth && auth->authorizer_buf_len)
e2200423 1425 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1426 auth->authorizer_buf);
859eb799 1427
31b8006e 1428 con->out_more = 0;
c9ffc77a 1429 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1430
e10c758e 1431 return 0;
31b8006e
SW		 1432}
1433
31b8006e
SW		 1434/*
1435 * write as much of pending kvecs to the socket as we can.
1436 * 1 -> done
1437 * 0 -> socket full, but more to do
1438 * <0 -> error
1439 */
1440static int write_partial_kvec(struct ceph_connection *con)
1441{
1442 int ret;
1443
1444 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1445 while (con->out_kvec_bytes > 0) {
1446 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1447 con->out_kvec_left, con->out_kvec_bytes,
1448 con->out_more);
1449 if (ret <= 0)
1450 goto out;
1451 con->out_kvec_bytes -= ret;
1452 if (con->out_kvec_bytes == 0)
1453 break; /* done */
f42299e6
AE		 1454
1455 /* account for full iov entries consumed */
1456 while (ret >= con->out_kvec_cur->iov_len) {
1457 BUG_ON(!con->out_kvec_left);
1458 ret -= con->out_kvec_cur->iov_len;
1459 con->out_kvec_cur++;
1460 con->out_kvec_left--;
1461 }
1462 /* and for a partially-consumed entry */
1463 if (ret) {
1464 con->out_kvec_cur->iov_len -= ret;
1465 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1466 }
1467 }
1468 con->out_kvec_left = 0;
1469 con->out_kvec_is_msg = false;
1470 ret = 1;
1471out:
1472 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1473 con->out_kvec_bytes, con->out_kvec_left, ret);
1474 return ret; /* done! */
1475}
1476
35b62808
AE		 1477static u32 ceph_crc32c_page(u32 crc, struct page *page,
1478 unsigned int page_offset,
1479 unsigned int length)
1480{
1481 char *kaddr;
1482
1483 kaddr = kmap(page);
1484 BUG_ON(kaddr == NULL);
1485 crc = crc32c(crc, kaddr + page_offset, length);
1486 kunmap(page);
1487
1488 return crc;
1489}
31b8006e
SW		 1490/*
1491 * Write as much message data payload as we can. If we finish, queue
1492 * up the footer.
1493 * 1 -> done, footer is now queued in out_kvec[].
1494 * 0 -> socket full, but more to do
1495 * <0 -> error
1496 */
34d2d200 1497static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1498{
1499 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1500 struct ceph_msg_data_cursor *cursor = &msg->cursor;
37675b0f 1501 bool do_datacrc = !con->msgr->nocrc;
f5db90bc 1502 u32 crc;
31b8006e 1503
859a35d5 1504 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1505
5240d9f9 1506 if (list_empty(&msg->data))
4c59b4a2
AE		 1507 return -EINVAL;
1508
5821bd8c
AE		 1509 /*
1510 * Iterate through each page that contains data to be
1511 * written, and send as much as possible for each.
1512 *
1513 * If we are calculating the data crc (the default), we will
1514 * need to map the page. If we have no pages, they have
1515 * been revoked, so use the zero page.
1516 */
f5db90bc 1517 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1518 while (cursor->resid) {
8a166d05 1519 struct page *page;
e387d525
AE		 1520 size_t page_offset;
1521 size_t length;
8a166d05 1522 bool last_piece;
8ea299bc 1523 bool need_crc;
f5db90bc 1524 int ret;
68b4476b 1525
8ae4f4f5 1526 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 1527 &last_piece);
e387d525 1528 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1529 length, last_piece);
f5db90bc
AE		 1530 if (ret <= 0) {
1531 if (do_datacrc)
1532 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1533
f5db90bc
AE		 1534 return ret;
1535 }
143334ff
AE		 1536 if (do_datacrc && cursor->need_crc)
1537 crc = ceph_crc32c_page(crc, page, page_offset, length);
8ae4f4f5 1538 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
31b8006e
SW		 1539 }
1540
34d2d200 1541 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1542
1543 /* prepare and queue up footer, too */
f5db90bc
AE		 1544 if (do_datacrc)
1545 msg->footer.data_crc = cpu_to_le32(crc);
1546 else
84ca8fc8 1547 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1548 con_out_kvec_reset(con);
859eb799 1549 prepare_write_message_footer(con);
f5db90bc
AE		 1550
1551 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1552}
1553
1554/*
1555 * write some zeros
1556 */
1557static int write_partial_skip(struct ceph_connection *con)
1558{
1559 int ret;
1560
1561 while (con->out_skip > 0) {
31739139 1562 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1563
e1dcb128 1564 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1565 if (ret <= 0)
1566 goto out;
1567 con->out_skip -= ret;
1568 }
1569 ret = 1;
1570out:
1571 return ret;
1572}
1573
1574/*
1575 * Prepare to read connection handshake, or an ack.
1576 */
eed0ef2c
SW		 1577static void prepare_read_banner(struct ceph_connection *con)
1578{
1579 dout("prepare_read_banner %p\n", con);
1580 con->in_base_pos = 0;
1581}
1582
31b8006e
SW		 1583static void prepare_read_connect(struct ceph_connection *con)
1584{
1585 dout("prepare_read_connect %p\n", con);
1586 con->in_base_pos = 0;
1587}
1588
1589static void prepare_read_ack(struct ceph_connection *con)
1590{
1591 dout("prepare_read_ack %p\n", con);
1592 con->in_base_pos = 0;
1593}
1594
3a23083b
SW		 1595static void prepare_read_seq(struct ceph_connection *con)
1596{
1597 dout("prepare_read_seq %p\n", con);
1598 con->in_base_pos = 0;
1599 con->in_tag = CEPH_MSGR_TAG_SEQ;
1600}
1601
31b8006e
SW		 1602static void prepare_read_tag(struct ceph_connection *con)
1603{
1604 dout("prepare_read_tag %p\n", con);
1605 con->in_base_pos = 0;
1606 con->in_tag = CEPH_MSGR_TAG_READY;
1607}
1608
1609/*
1610 * Prepare to read a message.
1611 */
1612static int prepare_read_message(struct ceph_connection *con)
1613{
1614 dout("prepare_read_message %p\n", con);
1615 BUG_ON(con->in_msg != NULL);
1616 con->in_base_pos = 0;
1617 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1618 return 0;
1619}
1620
1621
1622static int read_partial(struct ceph_connection *con,
fd51653f 1623 int end, int size, void *object)
31b8006e 1624{
e6cee71f
AE		 1625 while (con->in_base_pos < end) {
1626 int left = end - con->in_base_pos;
31b8006e
SW		 1627 int have = size - left;
1628 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1629 if (ret <= 0)
1630 return ret;
1631 con->in_base_pos += ret;
1632 }
1633 return 1;
1634}
1635
1636
1637/*
1638 * Read all or part of the connect-side handshake on a new connection
1639 */
eed0ef2c 1640static int read_partial_banner(struct ceph_connection *con)
31b8006e 1641{
fd51653f
AE		 1642 int size;
1643 int end;
1644 int ret;
31b8006e 1645
eed0ef2c 1646 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1647
1648 /* peer's banner */
fd51653f
AE		 1649 size = strlen(CEPH_BANNER);
1650 end = size;
1651 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1652 if (ret <= 0)
1653 goto out;
fd51653f
AE		 1654
1655 size = sizeof (con->actual_peer_addr);
1656 end += size;
1657 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1658 if (ret <= 0)
1659 goto out;
fd51653f
AE		 1660
1661 size = sizeof (con->peer_addr_for_me);
1662 end += size;
1663 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1664 if (ret <= 0)
1665 goto out;
fd51653f 1666
eed0ef2c
SW		 1667out:
1668 return ret;
1669}
1670
1671static int read_partial_connect(struct ceph_connection *con)
1672{
fd51653f
AE		 1673 int size;
1674 int end;
1675 int ret;
eed0ef2c
SW		 1676
1677 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1678
fd51653f
AE		 1679 size = sizeof (con->in_reply);
1680 end = size;
1681 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1682 if (ret <= 0)
1683 goto out;
fd51653f
AE		 1684
1685 size = le32_to_cpu(con->in_reply.authorizer_len);
1686 end += size;
1687 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1688 if (ret <= 0)
1689 goto out;
31b8006e 1690
4e7a5dcd
SW		 1691 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1692 con, (int)con->in_reply.tag,
1693 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1694 le32_to_cpu(con->in_reply.global_seq));
1695out:
1696 return ret;
eed0ef2c 1697
31b8006e
SW		 1698}
1699
1700/*
1701 * Verify the hello banner looks okay.
1702 */
1703static int verify_hello(struct ceph_connection *con)
1704{
1705 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1706 pr_err("connect to %s got bad banner\n",
3d14c5d2 1707 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1708 con->error_msg = "protocol error, bad banner";
1709 return -1;
1710 }
1711 return 0;
1712}
1713
1714static bool addr_is_blank(struct sockaddr_storage *ss)
1715{
1716 switch (ss->ss_family) {
1717 case AF_INET:
1718 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1719 case AF_INET6:
1720 return
1721 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1722 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1723 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1724 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1725 }
1726 return false;
1727}
1728
1729static int addr_port(struct sockaddr_storage *ss)
1730{
1731 switch (ss->ss_family) {
1732 case AF_INET:
f28bcfbe 1733 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1734 case AF_INET6:
f28bcfbe 1735 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1736 }
1737 return 0;
1738}
1739
1740static void addr_set_port(struct sockaddr_storage *ss, int p)
1741{
1742 switch (ss->ss_family) {
1743 case AF_INET:
1744 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1745 break;
31b8006e
SW		 1746 case AF_INET6:
1747 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1748 break;
31b8006e
SW		 1749 }
1750}
1751
ee3b56f2
NW		 1752/*
1753 * Unlike other *_pton function semantics, zero indicates success.
1754 */
1755static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1756 char delim, const char **ipend)
1757{
99f0f3b2
AE		 1758 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1759 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1760
1761 memset(ss, 0, sizeof(*ss));
1762
1763 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1764 ss->ss_family = AF_INET;
1765 return 0;
1766 }
1767
1768 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1769 ss->ss_family = AF_INET6;
1770 return 0;
1771 }
1772
1773 return -EINVAL;
1774}
1775
1776/*
1777 * Extract hostname string and resolve using kernel DNS facility.
1778 */
1779#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1780static int ceph_dns_resolve_name(const char *name, size_t namelen,
1781 struct sockaddr_storage *ss, char delim, const char **ipend)
1782{
1783 const char *end, *delim_p;
1784 char *colon_p, *ip_addr = NULL;
1785 int ip_len, ret;
1786
1787 /*
1788 * The end of the hostname occurs immediately preceding the delimiter or
1789 * the port marker (':') where the delimiter takes precedence.
1790 */
1791 delim_p = memchr(name, delim, namelen);
1792 colon_p = memchr(name, ':', namelen);
1793
1794 if (delim_p && colon_p)
1795 end = delim_p < colon_p ? delim_p : colon_p;
1796 else if (!delim_p && colon_p)
1797 end = colon_p;
1798 else {
1799 end = delim_p;
1800 if (!end) /* case: hostname:/ */
1801 end = name + namelen;
1802 }
1803
1804 if (end <= name)
1805 return -EINVAL;
1806
1807 /* do dns_resolve upcall */
1808 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1809 if (ip_len > 0)
1810 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1811 else
1812 ret = -ESRCH;
1813
1814 kfree(ip_addr);
1815
1816 *ipend = end;
1817
1818 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1819 ret, ret ? "failed" : ceph_pr_addr(ss));
1820
1821 return ret;
1822}
1823#else
1824static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1825 struct sockaddr_storage *ss, char delim, const char **ipend)
1826{
1827 return -EINVAL;
1828}
1829#endif
1830
1831/*
1832 * Parse a server name (IP or hostname). If a valid IP address is not found
1833 * then try to extract a hostname to resolve using userspace DNS upcall.
1834 */
1835static int ceph_parse_server_name(const char *name, size_t namelen,
1836 struct sockaddr_storage *ss, char delim, const char **ipend)
1837{
1838 int ret;
1839
1840 ret = ceph_pton(name, namelen, ss, delim, ipend);
1841 if (ret)
1842 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1843
1844 return ret;
1845}
1846
31b8006e
SW		 1847/*
1848 * Parse an ip[:port] list into an addr array. Use the default
1849 * monitor port if a port isn't specified.
1850 */
1851int ceph_parse_ips(const char *c, const char *end,
1852 struct ceph_entity_addr *addr,
1853 int max_count, int *count)
1854{
ee3b56f2 1855 int i, ret = -EINVAL;
31b8006e
SW		 1856 const char *p = c;
1857
1858 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1859 for (i = 0; i < max_count; i++) {
1860 const char *ipend;
1861 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1862 int port;
39139f64
SW		 1863 char delim = ',';
1864
1865 if (*p == '[') {
1866 delim = ']';
1867 p++;
1868 }
31b8006e 1869
ee3b56f2
NW		 1870 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1871 if (ret)
31b8006e 1872 goto bad;
ee3b56f2
NW		 1873 ret = -EINVAL;
1874
31b8006e
SW		 1875 p = ipend;
1876
39139f64
SW		 1877 if (delim == ']') {
1878 if (*p != ']') {
1879 dout("missing matching ']'\n");
1880 goto bad;
1881 }
1882 p++;
1883 }
1884
31b8006e
SW		 1885 /* port? */
1886 if (p < end && *p == ':') {
1887 port = 0;
1888 p++;
1889 while (p < end && *p >= '0' && *p <= '9') {
1890 port = (port * 10) + (*p - '0');
1891 p++;
1892 }
1893 if (port > 65535 || port == 0)
1894 goto bad;
1895 } else {
1896 port = CEPH_MON_PORT;
1897 }
1898
1899 addr_set_port(ss, port);
1900
3d14c5d2 1901 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1902
1903 if (p == end)
1904 break;
1905 if (*p != ',')
1906 goto bad;
1907 p++;
1908 }
1909
1910 if (p != end)
1911 goto bad;
1912
1913 if (count)
1914 *count = i + 1;
1915 return 0;
1916
1917bad:
39139f64 1918 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1919 return ret;
31b8006e 1920}
3d14c5d2 1921EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1922
eed0ef2c 1923static int process_banner(struct ceph_connection *con)
31b8006e 1924{
eed0ef2c 1925 dout("process_banner on %p\n", con);
31b8006e
SW		 1926
1927 if (verify_hello(con) < 0)
1928 return -1;
1929
63f2d211
SW		 1930 ceph_decode_addr(&con->actual_peer_addr);
1931 ceph_decode_addr(&con->peer_addr_for_me);
1932
31b8006e
SW		 1933 /*
1934 * Make sure the other end is who we wanted. note that the other
1935 * end may not yet know their ip address, so if it's 0.0.0.0, give
1936 * them the benefit of the doubt.
1937 */
103e2d3a
SW		 1938 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1939 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1940 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1941 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
cd84db6e 1942 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
3d14c5d2 1943 ceph_pr_addr(&con->peer_addr.in_addr),
cd84db6e 1944 (int)le32_to_cpu(con->peer_addr.nonce),
3d14c5d2 1945 ceph_pr_addr(&con->actual_peer_addr.in_addr),
cd84db6e 1946 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1947 con->error_msg = "wrong peer at address";
31b8006e
SW		 1948 return -1;
1949 }
1950
1951 /*
1952 * did we learn our address?
1953 */
1954 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1955 int port = addr_port(&con->msgr->inst.addr.in_addr);
1956
1957 memcpy(&con->msgr->inst.addr.in_addr,
1958 &con->peer_addr_for_me.in_addr,
1959 sizeof(con->peer_addr_for_me.in_addr));
1960 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1961 encode_my_addr(con->msgr);
eed0ef2c 1962 dout("process_banner learned my addr is %s\n",
3d14c5d2 1963 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1964 }
1965
eed0ef2c
SW		 1966 return 0;
1967}
1968
1969static int process_connect(struct ceph_connection *con)
1970{
3d14c5d2
YS		 1971 u64 sup_feat = con->msgr->supported_features;
1972 u64 req_feat = con->msgr->required_features;
04a419f9 1973 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 1974 int ret;
04a419f9 1975
eed0ef2c
SW		 1976 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1977
93c83e37
ID		 1978 if (con->auth_reply_buf) {
1979 /*
1980 * Any connection that defines ->get_authorizer()
1981 * should also define ->verify_authorizer_reply().
1982 * See get_connect_authorizer().
1983 */
1984 ret = con->ops->verify_authorizer_reply(con, 0);
1985 if (ret < 0) {
1986 con->error_msg = "bad authorize reply";
1987 return ret;
1988 }
1989 }
1990
31b8006e 1991 switch (con->in_reply.tag) {
04a419f9
SW		 1992 case CEPH_MSGR_TAG_FEATURES:
1993 pr_err("%s%lld %s feature set mismatch,"
1994 " my %llx < server's %llx, missing %llx\n",
1995 ENTITY_NAME(con->peer_name),
3d14c5d2 1996 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1997 sup_feat, server_feat, server_feat & ~sup_feat);
1998 con->error_msg = "missing required protocol features";
0fa6ebc6 1999 reset_connection(con);
04a419f9
SW		 2000 return -1;
2001
31b8006e 2002 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2003 pr_err("%s%lld %s protocol version mismatch,"
2004 " my %d != server's %d\n",
2005 ENTITY_NAME(con->peer_name),
3d14c5d2 2006 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2007 le32_to_cpu(con->out_connect.protocol_version),
2008 le32_to_cpu(con->in_reply.protocol_version));
2009 con->error_msg = "protocol version mismatch";
0fa6ebc6 2010 reset_connection(con);
31b8006e
SW		 2011 return -1;
2012
4e7a5dcd
SW		 2013 case CEPH_MSGR_TAG_BADAUTHORIZER:
2014 con->auth_retry++;
2015 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2016 con->auth_retry);
2017 if (con->auth_retry == 2) {
2018 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2019 return -1;
2020 }
6d4221b5 2021 con_out_kvec_reset(con);
e825a66d 2022 ret = prepare_write_connect(con);
0da5d703
SW		 2023 if (ret < 0)
2024 return ret;
63733a0f 2025 prepare_read_connect(con);
4e7a5dcd 2026 break;
31b8006e
SW		 2027
2028 case CEPH_MSGR_TAG_RESETSESSION:
2029 /*
2030 * If we connected with a large connect_seq but the peer
2031 * has no record of a session with us (no connection, or
2032 * connect_seq == 0), they will send RESETSESION to indicate
2033 * that they must have reset their session, and may have
2034 * dropped messages.
2035 */
2036 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2037 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2038 pr_err("%s%lld %s connection reset\n",
2039 ENTITY_NAME(con->peer_name),
3d14c5d2 2040 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2041 reset_connection(con);
6d4221b5 2042 con_out_kvec_reset(con);
5a0f8fdd
AE		 2043 ret = prepare_write_connect(con);
2044 if (ret < 0)
2045 return ret;
31b8006e
SW		 2046 prepare_read_connect(con);
2047
2048 /* Tell ceph about it. */
ec302645 2049 mutex_unlock(&con->mutex);
31b8006e
SW		 2050 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2051 if (con->ops->peer_reset)
2052 con->ops->peer_reset(con);
ec302645 2053 mutex_lock(&con->mutex);
8dacc7da 2054 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2055 return -EAGAIN;
31b8006e
SW		 2056 break;
2057
2058 case CEPH_MSGR_TAG_RETRY_SESSION:
2059 /*
2060 * If we sent a smaller connect_seq than the peer has, try
2061 * again with a larger value.
2062 */
5bdca4e0 2063 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2064 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2065 le32_to_cpu(con->in_reply.connect_seq));
2066 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2067 con_out_kvec_reset(con);
5a0f8fdd
AE		 2068 ret = prepare_write_connect(con);
2069 if (ret < 0)
2070 return ret;
31b8006e
SW		 2071 prepare_read_connect(con);
2072 break;
2073
2074 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2075 /*
2076 * If we sent a smaller global_seq than the peer has, try
2077 * again with a larger value.
2078 */
eed0ef2c 2079 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2080 con->peer_global_seq,
5bdca4e0 2081 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2082 get_global_seq(con->msgr,
5bdca4e0 2083 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2084 con_out_kvec_reset(con);
5a0f8fdd
AE		 2085 ret = prepare_write_connect(con);
2086 if (ret < 0)
2087 return ret;
31b8006e
SW		 2088 prepare_read_connect(con);
2089 break;
2090
3a23083b 2091 case CEPH_MSGR_TAG_SEQ:
31b8006e 2092 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2093 if (req_feat & ~server_feat) {
2094 pr_err("%s%lld %s protocol feature mismatch,"
2095 " my required %llx > server's %llx, need %llx\n",
2096 ENTITY_NAME(con->peer_name),
3d14c5d2 2097 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2098 req_feat, server_feat, req_feat & ~server_feat);
2099 con->error_msg = "missing required protocol features";
0fa6ebc6 2100 reset_connection(con);
04a419f9
SW		 2101 return -1;
2102 }
8dacc7da 2103
122070a2 2104 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2105 con->state = CON_STATE_OPEN;
20e55c4c 2106 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2107 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2108 con->connect_seq++;
aba558e2 2109 con->peer_features = server_feat;
31b8006e
SW		 2110 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2111 con->peer_global_seq,
2112 le32_to_cpu(con->in_reply.connect_seq),
2113 con->connect_seq);
2114 WARN_ON(con->connect_seq !=
2115 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2116
2117 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2118 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2119
85effe18 2120 con->delay = 0; /* reset backoff memory */
92ac41d0 2121
3a23083b
SW		 2122 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2123 prepare_write_seq(con);
2124 prepare_read_seq(con);
2125 } else {
2126 prepare_read_tag(con);
2127 }
31b8006e
SW		 2128 break;
2129
2130 case CEPH_MSGR_TAG_WAIT:
2131 /*
2132 * If there is a connection race (we are opening
2133 * connections to each other), one of us may just have
2134 * to WAIT. This shouldn't happen if we are the
2135 * client.
2136 */
04177882
SW		 2137 pr_err("process_connect got WAIT as client\n");
2138 con->error_msg = "protocol error, got WAIT as client";
2139 return -1;
31b8006e
SW		 2140
2141 default:
2142 pr_err("connect protocol error, will retry\n");
2143 con->error_msg = "protocol error, garbage tag during connect";
2144 return -1;
2145 }
2146 return 0;
2147}
2148
2149
2150/*
2151 * read (part of) an ack
2152 */
2153static int read_partial_ack(struct ceph_connection *con)
2154{
fd51653f
AE		 2155 int size = sizeof (con->in_temp_ack);
2156 int end = size;
31b8006e 2157
fd51653f 2158 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2159}
2160
31b8006e
SW		 2161/*
2162 * We can finally discard anything that's been acked.
2163 */
2164static void process_ack(struct ceph_connection *con)
2165{
2166 struct ceph_msg *m;
2167 u64 ack = le64_to_cpu(con->in_temp_ack);
2168 u64 seq;
2169
31b8006e
SW		 2170 while (!list_empty(&con->out_sent)) {
2171 m = list_first_entry(&con->out_sent, struct ceph_msg,
2172 list_head);
2173 seq = le64_to_cpu(m->hdr.seq);
2174 if (seq > ack)
2175 break;
2176 dout("got ack for seq %llu type %d at %p\n", seq,
2177 le16_to_cpu(m->hdr.type), m);
4cf9d544 2178 m->ack_stamp = jiffies;
31b8006e
SW		 2179 ceph_msg_remove(m);
2180 }
31b8006e
SW		 2181 prepare_read_tag(con);
2182}
2183
2184
2450418c 2185static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2186 struct kvec *section,
2187 unsigned int sec_len, u32 *crc)
2450418c 2188{
68b4476b 2189 int ret, left;
2450418c
YS		 2190
2191 BUG_ON(!section);
2192
2193 while (section->iov_len < sec_len) {
2194 BUG_ON(section->iov_base == NULL);
2195 left = sec_len - section->iov_len;
2196 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2197 section->iov_len, left);
2198 if (ret <= 0)
2199 return ret;
2200 section->iov_len += ret;
2450418c 2201 }
fe3ad593
AE		 2202 if (section->iov_len == sec_len)
2203 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2204
2450418c
YS		 2205 return 1;
2206}
31b8006e 2207
34d2d200
AE		 2208static int read_partial_msg_data(struct ceph_connection *con)
2209{
2210 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2211 struct ceph_msg_data_cursor *cursor = &msg->cursor;
34d2d200 2212 const bool do_datacrc = !con->msgr->nocrc;
686be208
AE		 2213 struct page *page;
2214 size_t page_offset;
2215 size_t length;
f5db90bc 2216 u32 crc = 0;
34d2d200
AE		 2217 int ret;
2218
2219 BUG_ON(!msg);
5240d9f9 2220 if (list_empty(&msg->data))
4c59b4a2 2221 return -EIO;
34d2d200 2222
f5db90bc
AE		 2223 if (do_datacrc)
2224 crc = con->in_data_crc;
643c68a4 2225 while (cursor->resid) {
8ae4f4f5 2226 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 2227 NULL);
686be208 2228 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2229 if (ret <= 0) {
2230 if (do_datacrc)
2231 con->in_data_crc = crc;
2232
686be208 2233 return ret;
f5db90bc 2234 }
686be208
AE		 2235
2236 if (do_datacrc)
f5db90bc 2237 crc = ceph_crc32c_page(crc, page, page_offset, ret);
8ae4f4f5 2238 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
34d2d200 2239 }
f5db90bc
AE		 2240 if (do_datacrc)
2241 con->in_data_crc = crc;
34d2d200
AE		 2242
2243 return 1; /* must return > 0 to indicate success */
2244}
2245
31b8006e
SW		 2246/*
2247 * read (part of) a message.
2248 */
686be208
AE		 2249static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2250
31b8006e
SW		 2251static int read_partial_message(struct ceph_connection *con)
2252{
2253 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2254 int size;
2255 int end;
31b8006e 2256 int ret;
95c96174 2257 unsigned int front_len, middle_len, data_len;
37675b0f 2258 bool do_datacrc = !con->msgr->nocrc;
ae18756b 2259 u64 seq;
fe3ad593 2260 u32 crc;
31b8006e
SW		 2261
2262 dout("read_partial_message con %p msg %p\n", con, m);
2263
2264 /* header */
fd51653f
AE		 2265 size = sizeof (con->in_hdr);
2266 end = size;
2267 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2268 if (ret <= 0)
2269 return ret;
fe3ad593
AE		 2270
2271 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2272 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2273 pr_err("read_partial_message bad hdr "
2274 " crc %u != expected %u\n",
2275 crc, con->in_hdr.crc);
2276 return -EBADMSG;
2277 }
2278
31b8006e
SW		 2279 front_len = le32_to_cpu(con->in_hdr.front_len);
2280 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2281 return -EIO;
2282 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2283 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2284 return -EIO;
2285 data_len = le32_to_cpu(con->in_hdr.data_len);
2286 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2287 return -EIO;
2288
ae18756b
SW		 2289 /* verify seq# */
2290 seq = le64_to_cpu(con->in_hdr.seq);
2291 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2292 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2293 ENTITY_NAME(con->peer_name),
3d14c5d2 2294 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2295 seq, con->in_seq + 1);
2296 con->in_base_pos = -front_len - middle_len - data_len -
2297 sizeof(m->footer);
2298 con->in_tag = CEPH_MSGR_TAG_READY;
636a9c8a 2299 return 1;
ae18756b
SW		 2300 } else if ((s64)seq - (s64)con->in_seq > 1) {
2301 pr_err("read_partial_message bad seq %lld expected %lld\n",
2302 seq, con->in_seq + 1);
2303 con->error_msg = "bad message sequence # for incoming message";
2304 return -EBADMSG;
2305 }
2306
31b8006e
SW		 2307 /* allocate message? */
2308 if (!con->in_msg) {
4740a623
SW		 2309 int skip = 0;
2310
31b8006e 2311 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2312 front_len, data_len);
4740a623
SW		 2313 ret = ceph_con_in_msg_alloc(con, &skip);
2314 if (ret < 0)
2315 return ret;
f759ebb9
AE		 2316
2317 BUG_ON(!con->in_msg ^ skip);
2318 if (con->in_msg && data_len > con->in_msg->data_length) {
2319 pr_warning("%s skipping long message (%u > %zd)\n",
2320 __func__, data_len, con->in_msg->data_length);
2321 ceph_msg_put(con->in_msg);
2322 con->in_msg = NULL;
2323 skip = 1;
2324 }
2450418c 2325 if (skip) {
31b8006e 2326 /* skip this message */
a79832f2 2327 dout("alloc_msg said skip message\n");
31b8006e
SW		 2328 con->in_base_pos = -front_len - middle_len - data_len -
2329 sizeof(m->footer);
2330 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2331 con->in_seq++;
636a9c8a 2332 return 1;
31b8006e 2333 }
38941f80 2334
4740a623 2335 BUG_ON(!con->in_msg);
38941f80 2336 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2337 m = con->in_msg;
2338 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2339 if (m->middle)
2340 m->middle->vec.iov_len = 0;
9d7f0f13 2341
78625051 2342 /* prepare for data payload, if any */
a4107026 2343
78625051 2344 if (data_len)
98fa5dd8 2345 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2346 }
2347
2348 /* front */
2450418c
YS		 2349 ret = read_partial_message_section(con, &m->front, front_len,
2350 &con->in_front_crc);
2351 if (ret <= 0)
2352 return ret;
31b8006e
SW		 2353
2354 /* middle */
2450418c 2355 if (m->middle) {
213c99ee
SW		 2356 ret = read_partial_message_section(con, &m->middle->vec,
2357 middle_len,
2450418c 2358 &con->in_middle_crc);
31b8006e
SW		 2359 if (ret <= 0)
2360 return ret;
31b8006e
SW		 2361 }
2362
2363 /* (page) data */
34d2d200
AE		 2364 if (data_len) {
2365 ret = read_partial_msg_data(con);
2366 if (ret <= 0)
2367 return ret;
31b8006e
SW		 2368 }
2369
31b8006e 2370 /* footer */
fd51653f
AE		 2371 size = sizeof (m->footer);
2372 end += size;
2373 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2374 if (ret <= 0)
2375 return ret;
2376
31b8006e
SW		 2377 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2378 m, front_len, m->footer.front_crc, middle_len,
2379 m->footer.middle_crc, data_len, m->footer.data_crc);
2380
2381 /* crc ok? */
2382 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2383 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2384 m, con->in_front_crc, m->footer.front_crc);
2385 return -EBADMSG;
2386 }
2387 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2388 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2389 m, con->in_middle_crc, m->footer.middle_crc);
2390 return -EBADMSG;
2391 }
bca064d2 2392 if (do_datacrc &&
31b8006e
SW		 2393 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2394 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2395 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2396 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2397 return -EBADMSG;
2398 }
2399
2400 return 1; /* done! */
2401}
2402
2403/*
2404 * Process message. This happens in the worker thread. The callback should
2405 * be careful not to do anything that waits on other incoming messages or it
2406 * may deadlock.
2407 */
2408static void process_message(struct ceph_connection *con)
2409{
5e095e8b 2410 struct ceph_msg *msg;
31b8006e 2411
38941f80
AE		 2412 BUG_ON(con->in_msg->con != con);
2413 con->in_msg->con = NULL;
5e095e8b 2414 msg = con->in_msg;
31b8006e 2415 con->in_msg = NULL;
36eb71aa 2416 con->ops->put(con);
31b8006e
SW		 2417
2418 /* if first message, set peer_name */
2419 if (con->peer_name.type == 0)
dbad185d 2420 con->peer_name = msg->hdr.src;
31b8006e 2421
31b8006e 2422 con->in_seq++;
ec302645 2423 mutex_unlock(&con->mutex);
31b8006e
SW		 2424
2425 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2426 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2427 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2428 le16_to_cpu(msg->hdr.type),
2429 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2430 le32_to_cpu(msg->hdr.front_len),
2431 le32_to_cpu(msg->hdr.data_len),
2432 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2433 con->ops->dispatch(con, msg);
ec302645
SW		 2434
2435 mutex_lock(&con->mutex);
31b8006e
SW		 2436}
2437
2438
2439/*
2440 * Write something to the socket. Called in a worker thread when the
2441 * socket appears to be writeable and we have something ready to send.
2442 */
2443static int try_write(struct ceph_connection *con)
2444{
31b8006e
SW		 2445 int ret = 1;
2446
d59315ca 2447 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2448
31b8006e
SW		 2449more:
2450 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2451
2452 /* open the socket first? */
8dacc7da
SW		 2453 if (con->state == CON_STATE_PREOPEN) {
2454 BUG_ON(con->sock);
2455 con->state = CON_STATE_CONNECTING;
a5988c49 2456
e2200423 2457 con_out_kvec_reset(con);
e825a66d 2458 prepare_write_banner(con);
eed0ef2c 2459 prepare_read_banner(con);
31b8006e 2460
cf3e5c40 2461 BUG_ON(con->in_msg);
31b8006e
SW		 2462 con->in_tag = CEPH_MSGR_TAG_READY;
2463 dout("try_write initiating connect on %p new state %lu\n",
2464 con, con->state);
41617d0c
AE		 2465 ret = ceph_tcp_connect(con);
2466 if (ret < 0) {
31b8006e 2467 con->error_msg = "connect error";
31b8006e
SW		 2468 goto out;
2469 }
2470 }
2471
2472more_kvec:
2473 /* kvec data queued? */
2474 if (con->out_skip) {
2475 ret = write_partial_skip(con);
2476 if (ret <= 0)
42961d23 2477 goto out;
31b8006e
SW		 2478 }
2479 if (con->out_kvec_left) {
2480 ret = write_partial_kvec(con);
2481 if (ret <= 0)
42961d23 2482 goto out;
31b8006e
SW		 2483 }
2484
2485 /* msg pages? */
2486 if (con->out_msg) {
c86a2930
SW		 2487 if (con->out_msg_done) {
2488 ceph_msg_put(con->out_msg);
2489 con->out_msg = NULL; /* we're done with this one */
2490 goto do_next;
2491 }
2492
34d2d200 2493 ret = write_partial_message_data(con);
31b8006e
SW		 2494 if (ret == 1)
2495 goto more_kvec; /* we need to send the footer, too! */
2496 if (ret == 0)
42961d23 2497 goto out;
31b8006e 2498 if (ret < 0) {
34d2d200 2499 dout("try_write write_partial_message_data err %d\n",
31b8006e 2500 ret);
42961d23 2501 goto out;
31b8006e
SW		 2502 }
2503 }
2504
c86a2930 2505do_next:
8dacc7da 2506 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2507 /* is anything else pending? */
2508 if (!list_empty(&con->out_queue)) {
2509 prepare_write_message(con);
2510 goto more;
2511 }
2512 if (con->in_seq > con->in_seq_acked) {
2513 prepare_write_ack(con);
2514 goto more;
2515 }
c9ffc77a 2516 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2517 prepare_write_keepalive(con);
2518 goto more;
2519 }
2520 }
2521
2522 /* Nothing to do! */
c9ffc77a 2523 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2524 dout("try_write nothing else to write.\n");
31b8006e
SW		 2525 ret = 0;
2526out:
42961d23 2527 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2528 return ret;
2529}
2530
2531
2532
2533/*
2534 * Read what we can from the socket.
2535 */
2536static int try_read(struct ceph_connection *con)
2537{
31b8006e
SW		 2538 int ret = -1;
2539
8dacc7da
SW		 2540more:
2541 dout("try_read start on %p state %lu\n", con, con->state);
2542 if (con->state != CON_STATE_CONNECTING &&
2543 con->state != CON_STATE_NEGOTIATING &&
2544 con->state != CON_STATE_OPEN)
31b8006e
SW		 2545 return 0;
2546
8dacc7da 2547 BUG_ON(!con->sock);
ec302645 2548
31b8006e
SW		 2549 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2550 con->in_base_pos);
0da5d703 2551
8dacc7da 2552 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2553 dout("try_read connecting\n");
2554 ret = read_partial_banner(con);
2555 if (ret <= 0)
ab166d5a 2556 goto out;
7593af92
AE		 2557 ret = process_banner(con);
2558 if (ret < 0)
2559 goto out;
2560
8dacc7da 2561 con->state = CON_STATE_NEGOTIATING;
7593af92 2562
6d4221b5
JS		 2563 /*
2564 * Received banner is good, exchange connection info.
2565 * Do not reset out_kvec, as sending our banner raced
2566 * with receiving peer banner after connect completed.
2567 */
7593af92
AE		 2568 ret = prepare_write_connect(con);
2569 if (ret < 0)
2570 goto out;
2571 prepare_read_connect(con);
2572
2573 /* Send connection info before awaiting response */
0da5d703
SW		 2574 goto out;
2575 }
2576
8dacc7da 2577 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2578 dout("try_read negotiating\n");
31b8006e
SW		 2579 ret = read_partial_connect(con);
2580 if (ret <= 0)
31b8006e 2581 goto out;
98bdb0aa
SW		 2582 ret = process_connect(con);
2583 if (ret < 0)
2584 goto out;
31b8006e
SW		 2585 goto more;
2586 }
2587
122070a2 2588 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2589
31b8006e
SW		 2590 if (con->in_base_pos < 0) {
2591 /*
2592 * skipping + discarding content.
2593 *
2594 * FIXME: there must be a better way to do this!
2595 */
84495f49
AE		 2596 static char buf[SKIP_BUF_SIZE];
2597 int skip = min((int) sizeof (buf), -con->in_base_pos);
2598
31b8006e
SW		 2599 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2600 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2601 if (ret <= 0)
98bdb0aa 2602 goto out;
31b8006e
SW		 2603 con->in_base_pos += ret;
2604 if (con->in_base_pos)
2605 goto more;
2606 }
2607 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2608 /*
2609 * what's next?
2610 */
2611 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2612 if (ret <= 0)
98bdb0aa 2613 goto out;
31b8006e
SW		 2614 dout("try_read got tag %d\n", (int)con->in_tag);
2615 switch (con->in_tag) {
2616 case CEPH_MSGR_TAG_MSG:
2617 prepare_read_message(con);
2618 break;
2619 case CEPH_MSGR_TAG_ACK:
2620 prepare_read_ack(con);
2621 break;
2622 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2623 con_close_socket(con);
2624 con->state = CON_STATE_CLOSED;
98bdb0aa 2625 goto out;
31b8006e
SW		 2626 default:
2627 goto bad_tag;
2628 }
2629 }
2630 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2631 ret = read_partial_message(con);
2632 if (ret <= 0) {
2633 switch (ret) {
2634 case -EBADMSG:
2635 con->error_msg = "bad crc";
2636 ret = -EIO;
98bdb0aa 2637 break;
31b8006e
SW		 2638 case -EIO:
2639 con->error_msg = "io error";
98bdb0aa 2640 break;
31b8006e 2641 }
98bdb0aa 2642 goto out;
31b8006e
SW		 2643 }
2644 if (con->in_tag == CEPH_MSGR_TAG_READY)
2645 goto more;
2646 process_message(con);
7b862e07
SW		 2647 if (con->state == CON_STATE_OPEN)
2648 prepare_read_tag(con);
31b8006e
SW		 2649 goto more;
2650 }
3a23083b
SW		 2651 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2652 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2653 /*
2654 * the final handshake seq exchange is semantically
2655 * equivalent to an ACK
2656 */
31b8006e
SW		 2657 ret = read_partial_ack(con);
2658 if (ret <= 0)
98bdb0aa 2659 goto out;
31b8006e
SW		 2660 process_ack(con);
2661 goto more;
2662 }
2663
31b8006e 2664out:
98bdb0aa 2665 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2666 return ret;
2667
2668bad_tag:
2669 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2670 con->error_msg = "protocol error, garbage tag";
2671 ret = -1;
2672 goto out;
2673}
2674
2675
2676/*
802c6d96
AE		 2677 * Atomically queue work on a connection after the specified delay.
2678 * Bump @con reference to avoid races with connection teardown.
2679 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2680 */
802c6d96 2681static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2682{
31b8006e 2683 if (!con->ops->get(con)) {
802c6d96
AE		 2684 dout("%s %p ref count 0\n", __func__, con);
2685
2686 return -ENOENT;
31b8006e
SW		 2687 }
2688
802c6d96
AE		 2689 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2690 dout("%s %p - already queued\n", __func__, con);
31b8006e 2691 con->ops->put(con);
802c6d96
AE		 2692
2693 return -EBUSY;
31b8006e 2694 }
802c6d96
AE		 2695
2696 dout("%s %p %lu\n", __func__, con, delay);
2697
2698 return 0;
2699}
2700
2701static void queue_con(struct ceph_connection *con)
2702{
2703 (void) queue_con_delay(con, 0);
31b8006e
SW		 2704}
2705
7bb21d68
AE		 2706static bool con_sock_closed(struct ceph_connection *con)
2707{
c9ffc77a 2708 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2709 return false;
2710
2711#define CASE(x) \
2712 case CON_STATE_ ## x: \
2713 con->error_msg = "socket closed (con state " #x ")"; \
2714 break;
2715
2716 switch (con->state) {
2717 CASE(CLOSED);
2718 CASE(PREOPEN);
2719 CASE(CONNECTING);
2720 CASE(NEGOTIATING);
2721 CASE(OPEN);
2722 CASE(STANDBY);
2723 default:
2724 pr_warning("%s con %p unrecognized state %lu\n",
2725 __func__, con, con->state);
2726 con->error_msg = "unrecognized con state";
2727 BUG();
2728 break;
2729 }
2730#undef CASE
2731
2732 return true;
2733}
2734
f20a39fd
AE		 2735static bool con_backoff(struct ceph_connection *con)
2736{
2737 int ret;
2738
2739 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2740 return false;
2741
2742 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2743 if (ret) {
2744 dout("%s: con %p FAILED to back off %lu\n", __func__,
2745 con, con->delay);
2746 BUG_ON(ret == -ENOENT);
2747 con_flag_set(con, CON_FLAG_BACKOFF);
2748 }
2749
2750 return true;
2751}
2752
93209264
AE		 2753/* Finish fault handling; con->mutex must *not* be held here */
2754
2755static void con_fault_finish(struct ceph_connection *con)
2756{
2757 /*
2758 * in case we faulted due to authentication, invalidate our
2759 * current tickets so that we can get new ones.
2760 */
2761 if (con->auth_retry && con->ops->invalidate_authorizer) {
2762 dout("calling invalidate_authorizer()\n");
2763 con->ops->invalidate_authorizer(con);
2764 }
2765
2766 if (con->ops->fault)
2767 con->ops->fault(con);
2768}
2769
31b8006e
SW		 2770/*
2771 * Do some work on a connection. Drop a connection ref when we're done.
2772 */
2773static void con_work(struct work_struct *work)
2774{
2775 struct ceph_connection *con = container_of(work, struct ceph_connection,
2776 work.work);
49659416 2777 bool fault;
31b8006e 2778
9dd4658d 2779 mutex_lock(&con->mutex);
49659416
AE		 2780 while (true) {
2781 int ret;
31b8006e 2782
49659416
AE		 2783 if ((fault = con_sock_closed(con))) {
2784 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2785 break;
2786 }
2787 if (con_backoff(con)) {
2788 dout("%s: con %p BACKOFF\n", __func__, con);
2789 break;
2790 }
2791 if (con->state == CON_STATE_STANDBY) {
2792 dout("%s: con %p STANDBY\n", __func__, con);
2793 break;
2794 }
2795 if (con->state == CON_STATE_CLOSED) {
2796 dout("%s: con %p CLOSED\n", __func__, con);
2797 BUG_ON(con->sock);
2798 break;
2799 }
2800 if (con->state == CON_STATE_PREOPEN) {
2801 dout("%s: con %p PREOPEN\n", __func__, con);
2802 BUG_ON(con->sock);
2803 }
0da5d703 2804
49659416
AE		 2805 ret = try_read(con);
2806 if (ret < 0) {
2807 if (ret == -EAGAIN)
2808 continue;
2809 con->error_msg = "socket error on read";
2810 fault = true;
2811 break;
2812 }
2813
2814 ret = try_write(con);
2815 if (ret < 0) {
2816 if (ret == -EAGAIN)
2817 continue;
2818 con->error_msg = "socket error on write";
2819 fault = true;
2820 }
2821
2822 break; /* If we make it to here, we're done */
3a140a0d 2823 }
b6e7b6a1
AE		 2824 if (fault)
2825 con_fault(con);
9dd4658d 2826 mutex_unlock(&con->mutex);
0da5d703 2827
b6e7b6a1
AE		 2828 if (fault)
2829 con_fault_finish(con);
2830
2831 con->ops->put(con);
31b8006e
SW		 2832}
2833
31b8006e
SW		 2834/*
2835 * Generic error/fault handler. A retry mechanism is used with
2836 * exponential backoff
2837 */
93209264 2838static void con_fault(struct ceph_connection *con)
31b8006e 2839{
28362986 2840 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
3d14c5d2 2841 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2842 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2843 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2844
122070a2 2845 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2846 con->state != CON_STATE_NEGOTIATING &&
2847 con->state != CON_STATE_OPEN);
ec302645 2848
31b8006e 2849 con_close_socket(con);
5e095e8b 2850
c9ffc77a 2851 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2852 dout("fault on LOSSYTX channel, marking CLOSED\n");
2853 con->state = CON_STATE_CLOSED;
93209264 2854 return;
3b5ede07
SW		 2855 }
2856
5e095e8b 2857 if (con->in_msg) {
38941f80
AE		 2858 BUG_ON(con->in_msg->con != con);
2859 con->in_msg->con = NULL;
5e095e8b
SW		 2860 ceph_msg_put(con->in_msg);
2861 con->in_msg = NULL;
36eb71aa 2862 con->ops->put(con);
5e095e8b 2863 }
31b8006e 2864
e80a52d1
SW		 2865 /* Requeue anything that hasn't been acked */
2866 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2867
e76661d0
SW		 2868 /* If there are no messages queued or keepalive pending, place
2869 * the connection in a STANDBY state */
2870 if (list_empty(&con->out_queue) &&
c9ffc77a 2871 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2872 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2873 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2874 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2875 } else {
2876 /* retry after a delay. */
8dacc7da 2877 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2878 if (con->delay == 0)
2879 con->delay = BASE_DELAY_INTERVAL;
2880 else if (con->delay < MAX_DELAY_INTERVAL)
2881 con->delay *= 2;
c9ffc77a 2882 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2883 queue_con(con);
31b8006e 2884 }
31b8006e
SW		 2885}
2886
2887
2888
2889/*
15d9882c 2890 * initialize a new messenger instance
31b8006e 2891 */
15d9882c
AE		 2892void ceph_messenger_init(struct ceph_messenger *msgr,
2893 struct ceph_entity_addr *myaddr,
2894 u32 supported_features,
2895 u32 required_features,
2896 bool nocrc)
31b8006e 2897{
3d14c5d2
YS		 2898 msgr->supported_features = supported_features;
2899 msgr->required_features = required_features;
2900
31b8006e
SW		 2901 spin_lock_init(&msgr->global_seq_lock);
2902
31b8006e
SW		 2903 if (myaddr)
2904 msgr->inst.addr = *myaddr;
2905
2906 /* select a random nonce */
ac8839d7 2907 msgr->inst.addr.type = 0;
103e2d3a 2908 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2909 encode_my_addr(msgr);
15d9882c 2910 msgr->nocrc = nocrc;
31b8006e 2911
a2a32584 2912 atomic_set(&msgr->stopping, 0);
31b8006e 2913
15d9882c 2914 dout("%s %p\n", __func__, msgr);
31b8006e 2915}
15d9882c 2916EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2917
e00de341
SW		 2918static void clear_standby(struct ceph_connection *con)
2919{
2920 /* come back from STANDBY? */
8dacc7da 2921 if (con->state == CON_STATE_STANDBY) {
e00de341 2922 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2923 con->state = CON_STATE_PREOPEN;
e00de341 2924 con->connect_seq++;
c9ffc77a
AE		 2925 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2926 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2927 }
2928}
2929
31b8006e
SW		 2930/*
2931 * Queue up an outgoing message on the given connection.
2932 */
2933void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2934{
31b8006e 2935 /* set src+dst */
dbad185d 2936 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2937 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2938 msg->needs_out_seq = true;
2939
ec302645 2940 mutex_lock(&con->mutex);
92ce034b 2941
8dacc7da 2942 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2943 dout("con_send %p closed, dropping %p\n", con, msg);
2944 ceph_msg_put(msg);
2945 mutex_unlock(&con->mutex);
2946 return;
2947 }
2948
38941f80 2949 BUG_ON(msg->con != NULL);
36eb71aa 2950 msg->con = con->ops->get(con);
92ce034b
AE		 2951 BUG_ON(msg->con == NULL);
2952
31b8006e
SW		 2953 BUG_ON(!list_empty(&msg->list_head));
2954 list_add_tail(&msg->list_head, &con->out_queue);
2955 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2956 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2957 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2958 le32_to_cpu(msg->hdr.front_len),
2959 le32_to_cpu(msg->hdr.middle_len),
2960 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 2961
2962 clear_standby(con);
ec302645 2963 mutex_unlock(&con->mutex);
31b8006e
SW		 2964
2965 /* if there wasn't anything waiting to send before, queue
2966 * new work */
c9ffc77a 2967 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2968 queue_con(con);
2969}
3d14c5d2 2970EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 2971
2972/*
2973 * Revoke a message that was previously queued for send
2974 */
6740a845 2975void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 2976{
6740a845
AE		 2977 struct ceph_connection *con = msg->con;
2978
2979 if (!con)
2980 return; /* Message not in our possession */
2981
ec302645 2982 mutex_lock(&con->mutex);
31b8006e 2983 if (!list_empty(&msg->list_head)) {
38941f80 2984 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 2985 list_del_init(&msg->list_head);
38941f80 2986 BUG_ON(msg->con == NULL);
36eb71aa 2987 msg->con->ops->put(msg->con);
38941f80 2988 msg->con = NULL;
31b8006e 2989 msg->hdr.seq = 0;
38941f80 2990
31b8006e 2991 ceph_msg_put(msg);
ed98adad
SW		 2992 }
2993 if (con->out_msg == msg) {
38941f80 2994 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 2995 con->out_msg = NULL;
31b8006e
SW		 2996 if (con->out_kvec_is_msg) {
2997 con->out_skip = con->out_kvec_bytes;
2998 con->out_kvec_is_msg = false;
2999 }
ed98adad 3000 msg->hdr.seq = 0;
92ce034b
AE		 3001
3002 ceph_msg_put(msg);
31b8006e 3003 }
ec302645 3004 mutex_unlock(&con->mutex);
31b8006e
SW		 3005}
3006
350b1c32 3007/*
0d59ab81 3008 * Revoke a message that we may be reading data into
350b1c32 3009 */
8921d114 3010void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3011{
8921d114
AE		 3012 struct ceph_connection *con;
3013
3014 BUG_ON(msg == NULL);
3015 if (!msg->con) {
3016 dout("%s msg %p null con\n", __func__, msg);
3017
3018 return; /* Message not in our possession */
3019 }
3020
3021 con = msg->con;
350b1c32 3022 mutex_lock(&con->mutex);
8921d114 3023 if (con->in_msg == msg) {
95c96174
ED		 3024 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3025 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3026 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3027
3028 /* skip rest of message */
8921d114
AE		 3029 dout("%s %p msg %p revoked\n", __func__, con, msg);
3030 con->in_base_pos = con->in_base_pos -
350b1c32 3031 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3032 front_len -
3033 middle_len -
3034 data_len -
350b1c32 3035 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3036 ceph_msg_put(con->in_msg);
3037 con->in_msg = NULL;
3038 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3039 con->in_seq++;
350b1c32 3040 } else {
8921d114
AE		 3041 dout("%s %p in_msg %p msg %p no-op\n",
3042 __func__, con, con->in_msg, msg);
350b1c32
SW		 3043 }
3044 mutex_unlock(&con->mutex);
3045}
3046
31b8006e
SW		 3047/*
3048 * Queue a keepalive byte to ensure the tcp connection is alive.
3049 */
3050void ceph_con_keepalive(struct ceph_connection *con)
3051{
e00de341 3052 dout("con_keepalive %p\n", con);
00650931 3053 mutex_lock(&con->mutex);
e00de341 3054 clear_standby(con);
00650931 3055 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3056 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3057 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3058 queue_con(con);
3059}
3d14c5d2 3060EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3061
6644ed7b 3062static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3063{
6644ed7b
AE		 3064 struct ceph_msg_data *data;
3065
3066 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3067 return NULL;
3068
81b36be4 3069 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
6644ed7b
AE		 3070 if (data)
3071 data->type = type;
5240d9f9 3072 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3073
3074 return data;
3075}
3076
3077static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3078{
3079 if (!data)
3080 return;
3081
5240d9f9 3082 WARN_ON(!list_empty(&data->links));
6644ed7b
AE		 3083 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3084 ceph_pagelist_release(data->pagelist);
3085 kfree(data->pagelist);
3086 }
81b36be4 3087 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3088}
3089
90af3602 3090void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3091 size_t length, size_t alignment)
02afca6c 3092{
6644ed7b
AE		 3093 struct ceph_msg_data *data;
3094
07aa1558
AE		 3095 BUG_ON(!pages);
3096 BUG_ON(!length);
6644ed7b
AE		 3097
3098 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3099 BUG_ON(!data);
3100 data->pages = pages;
3101 data->length = length;
3102 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3103
5240d9f9
AE		 3104 list_add_tail(&data->links, &msg->data);
3105 msg->data_length += length;
02afca6c 3106}
90af3602 3107EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3108
90af3602 3109void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3110 struct ceph_pagelist *pagelist)
3111{
6644ed7b
AE		 3112 struct ceph_msg_data *data;
3113
07aa1558
AE		 3114 BUG_ON(!pagelist);
3115 BUG_ON(!pagelist->length);
27fa8385 3116
6644ed7b
AE		 3117 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3118 BUG_ON(!data);
3119 data->pagelist = pagelist;
3120
5240d9f9
AE		 3121 list_add_tail(&data->links, &msg->data);
3122 msg->data_length += pagelist->length;
27fa8385 3123}
90af3602 3124EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3125
ea96571f 3126#ifdef CONFIG_BLOCK
90af3602 3127void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3128 size_t length)
27fa8385 3129{
6644ed7b
AE		 3130 struct ceph_msg_data *data;
3131
07aa1558 3132 BUG_ON(!bio);
27fa8385 3133
6644ed7b
AE		 3134 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3135 BUG_ON(!data);
3136 data->bio = bio;
c851c495 3137 data->bio_length = length;
6644ed7b 3138
5240d9f9
AE		 3139 list_add_tail(&data->links, &msg->data);
3140 msg->data_length += length;
27fa8385 3141}
90af3602 3142EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3143#endif /* CONFIG_BLOCK */
27fa8385 3144
31b8006e
SW		 3145/*
3146 * construct a new message with given type, size
3147 * the new msg has a ref count of 1.
3148 */
b61c2763
SW		 3149struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3150 bool can_fail)
31b8006e
SW		 3151{
3152 struct ceph_msg *m;
3153
e3d5d638 3154 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3155 if (m == NULL)
3156 goto out;
31b8006e
SW		 3157
3158 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3159 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3160 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3161
9516e45b
AE		 3162 INIT_LIST_HEAD(&m->list_head);
3163 kref_init(&m->kref);
5240d9f9 3164 INIT_LIST_HEAD(&m->data);
ca20892d 3165
31b8006e 3166 /* front */
842a5780 3167 m->front_alloc_len = front_len;
31b8006e
SW		 3168 if (front_len) {
3169 if (front_len > PAGE_CACHE_SIZE) {
34d23762 3170 m->front.iov_base = __vmalloc(front_len, flags,
31b8006e
SW		 3171 PAGE_KERNEL);
3172 m->front_is_vmalloc = true;
3173 } else {
34d23762 3174 m->front.iov_base = kmalloc(front_len, flags);
31b8006e
SW		 3175 }
3176 if (m->front.iov_base == NULL) {
b61c2763 3177 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3178 front_len);
3179 goto out2;
3180 }
3181 } else {
3182 m->front.iov_base = NULL;
3183 }
3184 m->front.iov_len = front_len;
3185
bb257664 3186 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3187 return m;
3188
3189out2:
3190 ceph_msg_put(m);
3191out:
b61c2763
SW		 3192 if (!can_fail) {
3193 pr_err("msg_new can't create type %d front %d\n", type,
3194 front_len);
f0ed1b7c 3195 WARN_ON(1);
b61c2763
SW		 3196 } else {
3197 dout("msg_new can't create type %d front %d\n", type,
3198 front_len);
3199 }
a79832f2 3200 return NULL;
31b8006e 3201}
3d14c5d2 3202EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3203
31b8006e
SW		 3204/*
3205 * Allocate "middle" portion of a message, if it is needed and wasn't
3206 * allocated by alloc_msg. This allows us to read a small fixed-size
3207 * per-type header in the front and then gracefully fail (i.e.,
3208 * propagate the error to the caller based on info in the front) when
3209 * the middle is too large.
3210 */
2450418c 3211static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3212{
3213 int type = le16_to_cpu(msg->hdr.type);
3214 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3215
3216 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3217 ceph_msg_type_name(type), middle_len);
3218 BUG_ON(!middle_len);
3219 BUG_ON(msg->middle);
3220
b6c1d5b8 3221 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3222 if (!msg->middle)
3223 return -ENOMEM;
3224 return 0;
3225}
3226
2450418c 3227/*
1c20f2d2
AE		 3228 * Allocate a message for receiving an incoming message on a
3229 * connection, and save the result in con->in_msg. Uses the
3230 * connection's private alloc_msg op if available.
3231 *
4740a623
SW		 3232 * Returns 0 on success, or a negative error code.
3233 *
3234 * On success, if we set *skip = 1:
3235 * - the next message should be skipped and ignored.
3236 * - con->in_msg == NULL
3237 * or if we set *skip = 0:
3238 * - con->in_msg is non-null.
3239 * On error (ENOMEM, EAGAIN, ...),
3240 * - con->in_msg == NULL
2450418c 3241 */
4740a623 3242static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3243{
4740a623 3244 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3245 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3246 struct ceph_msg *msg;
4740a623 3247 int ret = 0;
2450418c 3248
1c20f2d2 3249 BUG_ON(con->in_msg != NULL);
53ded495 3250 BUG_ON(!con->ops->alloc_msg);
2450418c 3251
53ded495
AE		 3252 mutex_unlock(&con->mutex);
3253 msg = con->ops->alloc_msg(con, hdr, skip);
3254 mutex_lock(&con->mutex);
3255 if (con->state != CON_STATE_OPEN) {
3256 if (msg)
1d866d1c 3257 ceph_msg_put(msg);
53ded495
AE		 3258 return -EAGAIN;
3259 }
4137577a
AE		 3260 if (msg) {
3261 BUG_ON(*skip);
3262 con->in_msg = msg;
36eb71aa 3263 con->in_msg->con = con->ops->get(con);
92ce034b 3264 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3265 } else {
3266 /*
3267 * Null message pointer means either we should skip
3268 * this message or we couldn't allocate memory. The
3269 * former is not an error.
3270 */
3271 if (*skip)
3272 return 0;
3273 con->error_msg = "error allocating memory for incoming message";
3274
53ded495 3275 return -ENOMEM;
2450418c 3276 }
1c20f2d2 3277 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3278
1c20f2d2
AE		 3279 if (middle_len && !con->in_msg->middle) {
3280 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3281 if (ret < 0) {
1c20f2d2
AE		 3282 ceph_msg_put(con->in_msg);
3283 con->in_msg = NULL;
2450418c
YS		 3284 }
3285 }
9d7f0f13 3286
4740a623 3287 return ret;
2450418c
YS		 3288}
3289
31b8006e
SW		 3290
3291/*
3292 * Free a generically kmalloc'd message.
3293 */
3294void ceph_msg_kfree(struct ceph_msg *m)
3295{
3296 dout("msg_kfree %p\n", m);
3297 if (m->front_is_vmalloc)
3298 vfree(m->front.iov_base);
3299 else
3300 kfree(m->front.iov_base);
e3d5d638 3301 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3302}
3303
3304/*
3305 * Drop a msg ref. Destroy as needed.
3306 */
c2e552e7
SW		 3307void ceph_msg_last_put(struct kref *kref)
3308{
3309 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
5240d9f9
AE		 3310 LIST_HEAD(data);
3311 struct list_head *links;
3312 struct list_head *next;
31b8006e 3313
c2e552e7
SW		 3314 dout("ceph_msg_put last one on %p\n", m);
3315 WARN_ON(!list_empty(&m->list_head));
3316
3317 /* drop middle, data, if any */
3318 if (m->middle) {
3319 ceph_buffer_put(m->middle);
3320 m->middle = NULL;
31b8006e 3321 }
5240d9f9
AE		 3322
3323 list_splice_init(&m->data, &data);
3324 list_for_each_safe(links, next, &data) {
3325 struct ceph_msg_data *data;
3326
3327 data = list_entry(links, struct ceph_msg_data, links);
3328 list_del_init(links);
3329 ceph_msg_data_destroy(data);
3330 }
a1930804 3331 m->data_length = 0;
58bb3b37 3332
c2e552e7
SW		 3333 if (m->pool)
3334 ceph_msgpool_put(m->pool, m);
3335 else
3336 ceph_msg_kfree(m);
31b8006e 3337}
3d14c5d2 3338EXPORT_SYMBOL(ceph_msg_last_put);
9ec7cab1
SW		 3339
3340void ceph_msg_dump(struct ceph_msg *msg)
3341{
842a5780
ID		 3342 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3343 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3344 print_hex_dump(KERN_DEBUG, "header: ",
3345 DUMP_PREFIX_OFFSET, 16, 1,
3346 &msg->hdr, sizeof(msg->hdr), true);
3347 print_hex_dump(KERN_DEBUG, " front: ",
3348 DUMP_PREFIX_OFFSET, 16, 1,
3349 msg->front.iov_base, msg->front.iov_len, true);
3350 if (msg->middle)
3351 print_hex_dump(KERN_DEBUG, "middle: ",
3352 DUMP_PREFIX_OFFSET, 16, 1,
3353 msg->middle->vec.iov_base,
3354 msg->middle->vec.iov_len, true);
3355 print_hex_dump(KERN_DEBUG, "footer: ",
3356 DUMP_PREFIX_OFFSET, 16, 1,
3357 &msg->footer, sizeof(msg->footer), true);
3358}
3d14c5d2 3359EXPORT_SYMBOL(ceph_msg_dump);
kernel source for telekom puls (alcatel/mediatek)