projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: msg signing callouts don't need con argument
[linux-2.6-block.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>
757856d2 9#include <linux/nsproxy.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
2b3e0c90 19#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 20#include <linux/ceph/libceph.h>
21#include <linux/ceph/messenger.h>
22#include <linux/ceph/decode.h>
23#include <linux/ceph/pagelist.h>
bc3b2d7f 24#include <linux/export.h>
31b8006e 25
fe38a2b6
AE		 26#define list_entry_next(pos, member) \
27 list_entry(pos->member.next, typeof(*pos), member)
28
31b8006e
SW		 29/*
30 * Ceph uses the messenger to exchange ceph_msg messages with other
31 * hosts in the system. The messenger provides ordered and reliable
32 * delivery. We tolerate TCP disconnects by reconnecting (with
33 * exponential backoff) in the case of a fault (disconnection, bad
34 * crc, protocol error). Acks allow sent messages to be discarded by
35 * the sender.
36 */
37
bc18f4b1
AE		 38/*
39 * We track the state of the socket on a given connection using
40 * values defined below. The transition to a new socket state is
41 * handled by a function which verifies we aren't coming from an
42 * unexpected state.
43 *
44 * --------
45 * | NEW* | transient initial state
46 * --------
47 * | con_sock_state_init()
48 * v
49 * ----------
50 * | CLOSED | initialized, but no socket (and no
51 * ---------- TCP connection)
52 * ^ \
53 * | \ con_sock_state_connecting()
54 * | ----------------------
55 * | \
56 * + con_sock_state_closed() \
fbb85a47
SW		 57 * |+--------------------------- \
58 * | \ \ \
59 * | ----------- \ \
60 * | | CLOSING | socket event; \ \
61 * | ----------- await close \ \
62 * | ^ \ |
63 * | | \ |
64 * | + con_sock_state_closing() \ |
65 * | / \ | |
66 * | / --------------- | |
67 * | / \ v v
bc18f4b1
AE		 68 * | / --------------
69 * | / -----------------| CONNECTING | socket created, TCP
70 * | | / -------------- connect initiated
71 * | | | con_sock_state_connected()
72 * | | v
73 * -------------
74 * | CONNECTED | TCP connection established
75 * -------------
76 *
77 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
78 */
ce2c8903
AE		 79
80#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
81#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
82#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
83#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
84#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
85
8dacc7da
SW		 86/*
87 * connection states
88 */
89#define CON_STATE_CLOSED 1 /* -> PREOPEN */
90#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
91#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
92#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
93#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
94#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
95
4a861692
SW		 96/*
97 * ceph_connection flag bits
98 */
99#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
100 * messages on errors */
101#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
102#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
103#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
104#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 105
c9ffc77a
AE		 106static bool con_flag_valid(unsigned long con_flag)
107{
108 switch (con_flag) {
109 case CON_FLAG_LOSSYTX:
110 case CON_FLAG_KEEPALIVE_PENDING:
111 case CON_FLAG_WRITE_PENDING:
112 case CON_FLAG_SOCK_CLOSED:
113 case CON_FLAG_BACKOFF:
114 return true;
115 default:
116 return false;
117 }
118}
119
120static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
121{
122 BUG_ON(!con_flag_valid(con_flag));
123
124 clear_bit(con_flag, &con->flags);
125}
126
127static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
128{
129 BUG_ON(!con_flag_valid(con_flag));
130
131 set_bit(con_flag, &con->flags);
132}
133
134static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
135{
136 BUG_ON(!con_flag_valid(con_flag));
137
138 return test_bit(con_flag, &con->flags);
139}
140
141static bool con_flag_test_and_clear(struct ceph_connection *con,
142 unsigned long con_flag)
143{
144 BUG_ON(!con_flag_valid(con_flag));
145
146 return test_and_clear_bit(con_flag, &con->flags);
147}
148
149static bool con_flag_test_and_set(struct ceph_connection *con,
150 unsigned long con_flag)
151{
152 BUG_ON(!con_flag_valid(con_flag));
153
154 return test_and_set_bit(con_flag, &con->flags);
155}
156
e3d5d638
AE		 157/* Slab caches for frequently-allocated structures */
158
159static struct kmem_cache *ceph_msg_cache;
81b36be4 160static struct kmem_cache *ceph_msg_data_cache;
e3d5d638 161
31b8006e
SW		 162/* static tag bytes (protocol control messages) */
163static char tag_msg = CEPH_MSGR_TAG_MSG;
164static char tag_ack = CEPH_MSGR_TAG_ACK;
165static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
8b9558aa 166static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
31b8006e 167
a6a5349d
SW		 168#ifdef CONFIG_LOCKDEP
169static struct lock_class_key socket_class;
170#endif
171
84495f49
AE		 172/*
173 * When skipping (ignoring) a block of input we read it into a "skip
174 * buffer," which is this many bytes in size.
175 */
176#define SKIP_BUF_SIZE 1024
31b8006e
SW		 177
178static void queue_con(struct ceph_connection *con);
37ab77ac 179static void cancel_con(struct ceph_connection *con);
68931622 180static void ceph_con_workfn(struct work_struct *);
93209264 181static void con_fault(struct ceph_connection *con);
31b8006e 182
31b8006e 183/*
f64a9317
AE		 184 * Nicely render a sockaddr as a string. An array of formatted
185 * strings is used, to approximate reentrancy.
31b8006e 186 */
f64a9317
AE		 187#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
188#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
189#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
190#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
191
192static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
193static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 194
57666519 195static struct page *zero_page; /* used in certain error cases */
57666519 196
3d14c5d2 197const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 198{
199 int i;
200 char *s;
99f0f3b2
AE		 201 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
202 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 203
f64a9317 204 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 205 s = addr_str[i];
206
207 switch (ss->ss_family) {
208 case AF_INET:
bd406145
AE		 209 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
210 ntohs(in4->sin_port));
31b8006e
SW		 211 break;
212
213 case AF_INET6:
bd406145
AE		 214 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
215 ntohs(in6->sin6_port));
31b8006e
SW		 216 break;
217
218 default:
d3002b97
AE		 219 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
220 ss->ss_family);
31b8006e
SW		 221 }
222
223 return s;
224}
3d14c5d2 225EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 226
63f2d211
SW		 227static void encode_my_addr(struct ceph_messenger *msgr)
228{
229 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
230 ceph_encode_addr(&msgr->my_enc_addr);
231}
232
31b8006e
SW		 233/*
234 * work queue for all reading and writing to/from the socket.
235 */
e0f43c94 236static struct workqueue_struct *ceph_msgr_wq;
31b8006e 237
e3d5d638
AE		 238static int ceph_msgr_slab_init(void)
239{
240 BUG_ON(ceph_msg_cache);
241 ceph_msg_cache = kmem_cache_create("ceph_msg",
242 sizeof (struct ceph_msg),
243 __alignof__(struct ceph_msg), 0, NULL);
81b36be4
AE		 244
245 if (!ceph_msg_cache)
246 return -ENOMEM;
247
248 BUG_ON(ceph_msg_data_cache);
249 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
250 sizeof (struct ceph_msg_data),
251 __alignof__(struct ceph_msg_data),
252 0, NULL);
253 if (ceph_msg_data_cache)
254 return 0;
255
256 kmem_cache_destroy(ceph_msg_cache);
257 ceph_msg_cache = NULL;
258
259 return -ENOMEM;
e3d5d638
AE		 260}
261
262static void ceph_msgr_slab_exit(void)
263{
81b36be4
AE		 264 BUG_ON(!ceph_msg_data_cache);
265 kmem_cache_destroy(ceph_msg_data_cache);
266 ceph_msg_data_cache = NULL;
267
e3d5d638
AE		 268 BUG_ON(!ceph_msg_cache);
269 kmem_cache_destroy(ceph_msg_cache);
270 ceph_msg_cache = NULL;
271}
272
15417167 273static void _ceph_msgr_exit(void)
6173d1f0 274{
d3002b97 275 if (ceph_msgr_wq) {
6173d1f0 276 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 277 ceph_msgr_wq = NULL;
278 }
6173d1f0 279
6173d1f0 280 BUG_ON(zero_page == NULL);
6173d1f0
AE		 281 page_cache_release(zero_page);
282 zero_page = NULL;
d920ff6f
BC		 283
284 ceph_msgr_slab_exit();
6173d1f0
AE		 285}
286
3d14c5d2 287int ceph_msgr_init(void)
31b8006e 288{
d920ff6f
BC		 289 if (ceph_msgr_slab_init())
290 return -ENOMEM;
291
57666519
AE		 292 BUG_ON(zero_page != NULL);
293 zero_page = ZERO_PAGE(0);
294 page_cache_get(zero_page);
295
f9865f06
ID		 296 /*
297 * The number of active work items is limited by the number of
298 * connections, so leave @max_active at default.
299 */
300 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
6173d1f0
AE		 301 if (ceph_msgr_wq)
302 return 0;
57666519 303
6173d1f0
AE		 304 pr_err("msgr_init failed to create workqueue\n");
305 _ceph_msgr_exit();
57666519 306
6173d1f0 307 return -ENOMEM;
31b8006e 308}
3d14c5d2 309EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 310
311void ceph_msgr_exit(void)
312{
57666519 313 BUG_ON(ceph_msgr_wq == NULL);
57666519 314
6173d1f0 315 _ceph_msgr_exit();
31b8006e 316}
3d14c5d2 317EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 318
cd84db6e 319void ceph_msgr_flush(void)
a922d38f
SW		 320{
321 flush_workqueue(ceph_msgr_wq);
322}
3d14c5d2 323EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 324
ce2c8903
AE		 325/* Connection socket state transition functions */
326
327static void con_sock_state_init(struct ceph_connection *con)
328{
329 int old_state;
330
331 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
332 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
333 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 334 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
335 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 336}
337
338static void con_sock_state_connecting(struct ceph_connection *con)
339{
340 int old_state;
341
342 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
343 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
344 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 345 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
346 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 347}
348
349static void con_sock_state_connected(struct ceph_connection *con)
350{
351 int old_state;
352
353 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
354 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
355 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 356 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
357 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 358}
359
360static void con_sock_state_closing(struct ceph_connection *con)
361{
362 int old_state;
363
364 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
365 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
366 old_state != CON_SOCK_STATE_CONNECTED &&
367 old_state != CON_SOCK_STATE_CLOSING))
368 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 369 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
370 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 371}
372
373static void con_sock_state_closed(struct ceph_connection *con)
374{
375 int old_state;
376
377 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
378 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 379 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 380 old_state != CON_SOCK_STATE_CONNECTING &&
381 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 382 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 383 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
384 CON_SOCK_STATE_CLOSED);
ce2c8903 385}
a922d38f 386
31b8006e
SW		 387/*
388 * socket callback functions
389 */
390
391/* data available on socket, or listen socket received a connect */
676d2369 392static void ceph_sock_data_ready(struct sock *sk)
31b8006e 393{
bd406145 394 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 395 if (atomic_read(&con->msgr->stopping)) {
396 return;
397 }
bd406145 398
31b8006e 399 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 400 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 401 con, con->state);
402 queue_con(con);
403 }
404}
405
406/* socket has buffer space for writing */
327800bd 407static void ceph_sock_write_space(struct sock *sk)
31b8006e 408{
d3002b97 409 struct ceph_connection *con = sk->sk_user_data;
31b8006e 410
182fac26
JS		 411 /* only queue to workqueue if there is data we want to write,
412 * and there is sufficient space in the socket buffer to accept
327800bd 413 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 414 * doesn't get called again until try_write() fills the socket
415 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
416 * and net/core/stream.c:sk_stream_write_space().
417 */
c9ffc77a 418 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
64dc6130 419 if (sk_stream_is_writeable(sk)) {
327800bd 420 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 421 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
422 queue_con(con);
423 }
31b8006e 424 } else {
327800bd 425 dout("%s %p nothing to write\n", __func__, con);
31b8006e 426 }
31b8006e
SW		 427}
428
429/* socket's state has changed */
327800bd 430static void ceph_sock_state_change(struct sock *sk)
31b8006e 431{
bd406145 432 struct ceph_connection *con = sk->sk_user_data;
31b8006e 433
327800bd 434 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 435 con, con->state, sk->sk_state);
436
31b8006e
SW		 437 switch (sk->sk_state) {
438 case TCP_CLOSE:
327800bd 439 dout("%s TCP_CLOSE\n", __func__);
31b8006e 440 case TCP_CLOSE_WAIT:
327800bd 441 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 442 con_sock_state_closing(con);
c9ffc77a 443 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 444 queue_con(con);
31b8006e
SW		 445 break;
446 case TCP_ESTABLISHED:
327800bd 447 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 448 con_sock_state_connected(con);
31b8006e
SW		 449 queue_con(con);
450 break;
d3002b97
AE		 451 default: /* Everything else is uninteresting */
452 break;
31b8006e
SW		 453 }
454}
455
456/*
457 * set up socket callbacks
458 */
459static void set_sock_callbacks(struct socket *sock,
460 struct ceph_connection *con)
461{
462 struct sock *sk = sock->sk;
bd406145 463 sk->sk_user_data = con;
327800bd
AE		 464 sk->sk_data_ready = ceph_sock_data_ready;
465 sk->sk_write_space = ceph_sock_write_space;
466 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 467}
468
469
470/*
471 * socket helpers
472 */
473
474/*
475 * initiate connection to a remote socket.
476 */
41617d0c 477static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 478{
f91d3471 479 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e
SW		 480 struct socket *sock;
481 int ret;
482
483 BUG_ON(con->sock);
757856d2 484 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
eeb1bd5c 485 SOCK_STREAM, IPPROTO_TCP, &sock);
31b8006e 486 if (ret)
41617d0c 487 return ret;
6d7fdb0a 488 sock->sk->sk_allocation = GFP_NOFS;
31b8006e 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);
41617d0c 509 return ret;
a5bc3129 510 }
89baaa57 511
ba988f87
CH		 512 if (con->msgr->tcp_nodelay) {
513 int optval = 1;
514
515 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
516 (char *)&optval, sizeof(optval));
517 if (ret)
518 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
519 ret);
520 }
521
a5bc3129 522 con->sock = sock;
41617d0c 523 return 0;
31b8006e
SW		 524}
525
526static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
527{
528 struct kvec iov = {buf, len};
529 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 530 int r;
31b8006e 531
98bdb0aa
SW		 532 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
533 if (r == -EAGAIN)
534 r = 0;
535 return r;
31b8006e
SW		 536}
537
afb3d90e
AE		 538static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
539 int page_offset, size_t length)
540{
541 void *kaddr;
542 int ret;
543
544 BUG_ON(page_offset + length > PAGE_SIZE);
545
546 kaddr = kmap(page);
547 BUG_ON(!kaddr);
548 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
549 kunmap(page);
550
551 return ret;
552}
553
31b8006e
SW		 554/*
555 * write something. @more is true if caller will be sending more data
556 * shortly.
557 */
558static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
559 size_t kvlen, size_t len, int more)
560{
561 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 562 int r;
31b8006e
SW		 563
564 if (more)
565 msg.msg_flags |= MSG_MORE;
566 else
567 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
568
42961d23
SW		 569 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
570 if (r == -EAGAIN)
571 r = 0;
572 return r;
31b8006e
SW		 573}
574
178eda29 575static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 576 int offset, size_t size, bool more)
31739139
AE		 577{
578 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
579 int ret;
580
581 ret = kernel_sendpage(sock, page, offset, size, flags);
582 if (ret == -EAGAIN)
583 ret = 0;
584
585 return ret;
586}
587
178eda29
CC		 588static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
589 int offset, size_t size, bool more)
590{
591 int ret;
592 struct kvec iov;
593
594 /* sendpage cannot properly handle pages with page_count == 0,
595 * we need to fallback to sendmsg if that's the case */
596 if (page_count(page) >= 1)
597 return __ceph_tcp_sendpage(sock, page, offset, size, more);
598
599 iov.iov_base = kmap(page) + offset;
600 iov.iov_len = size;
601 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
602 kunmap(page);
603
604 return ret;
605}
31b8006e
SW		 606
607/*
608 * Shutdown/close the socket for the given connection.
609 */
610static int con_close_socket(struct ceph_connection *con)
611{
8007b8d6 612 int rc = 0;
31b8006e
SW		 613
614 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 615 if (con->sock) {
616 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
617 sock_release(con->sock);
618 con->sock = NULL;
619 }
456ea468
AE		 620
621 /*
4a861692 622 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 623 * independent of the connection mutex, and we could have
624 * received a socket close event before we had the chance to
625 * shut the socket down.
626 */
c9ffc77a 627 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 628
ce2c8903 629 con_sock_state_closed(con);
31b8006e
SW		 630 return rc;
631}
632
633/*
634 * Reset a connection. Discard all incoming and outgoing messages
635 * and clear *_seq state.
636 */
637static void ceph_msg_remove(struct ceph_msg *msg)
638{
639 list_del_init(&msg->list_head);
38941f80 640 BUG_ON(msg->con == NULL);
36eb71aa 641 msg->con->ops->put(msg->con);
38941f80
AE		 642 msg->con = NULL;
643
31b8006e
SW		 644 ceph_msg_put(msg);
645}
646static void ceph_msg_remove_list(struct list_head *head)
647{
648 while (!list_empty(head)) {
649 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
650 list_head);
651 ceph_msg_remove(msg);
652 }
653}
654
655static void reset_connection(struct ceph_connection *con)
656{
657 /* reset connection, out_queue, msg_ and connect_seq */
658 /* discard existing out_queue and msg_seq */
0fa6ebc6 659 dout("reset_connection %p\n", con);
31b8006e
SW		 660 ceph_msg_remove_list(&con->out_queue);
661 ceph_msg_remove_list(&con->out_sent);
662
cf3e5c40 663 if (con->in_msg) {
38941f80
AE		 664 BUG_ON(con->in_msg->con != con);
665 con->in_msg->con = NULL;
cf3e5c40
SW		 666 ceph_msg_put(con->in_msg);
667 con->in_msg = NULL;
36eb71aa 668 con->ops->put(con);
cf3e5c40
SW		 669 }
670
31b8006e
SW		 671 con->connect_seq = 0;
672 con->out_seq = 0;
c86a2930
SW		 673 if (con->out_msg) {
674 ceph_msg_put(con->out_msg);
675 con->out_msg = NULL;
676 }
31b8006e 677 con->in_seq = 0;
0e0d5e0c 678 con->in_seq_acked = 0;
31b8006e
SW		 679}
680
681/*
682 * mark a peer down. drop any open connections.
683 */
684void ceph_con_close(struct ceph_connection *con)
685{
8c50c817 686 mutex_lock(&con->mutex);
3d14c5d2
YS		 687 dout("con_close %p peer %s\n", con,
688 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 689 con->state = CON_STATE_CLOSED;
a5988c49 690
c9ffc77a
AE		 691 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
692 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
693 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
694 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 695
31b8006e 696 reset_connection(con);
6f2bc3ff 697 con->peer_global_seq = 0;
37ab77ac 698 cancel_con(con);
ee76e073 699 con_close_socket(con);
ec302645 700 mutex_unlock(&con->mutex);
31b8006e 701}
3d14c5d2 702EXPORT_SYMBOL(ceph_con_close);
31b8006e 703
31b8006e
SW		 704/*
705 * Reopen a closed connection, with a new peer address.
706 */
b7a9e5dd
SW		 707void ceph_con_open(struct ceph_connection *con,
708 __u8 entity_type, __u64 entity_num,
709 struct ceph_entity_addr *addr)
31b8006e 710{
5469155f 711 mutex_lock(&con->mutex);
3d14c5d2 712 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 713
122070a2 714 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 715 con->state = CON_STATE_PREOPEN;
a5988c49 716
b7a9e5dd
SW		 717 con->peer_name.type = (__u8) entity_type;
718 con->peer_name.num = cpu_to_le64(entity_num);
719
31b8006e 720 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 721 con->delay = 0; /* reset backoff memory */
5469155f 722 mutex_unlock(&con->mutex);
31b8006e
SW		 723 queue_con(con);
724}
3d14c5d2 725EXPORT_SYMBOL(ceph_con_open);
31b8006e 726
87b315a5
SW		 727/*
728 * return true if this connection ever successfully opened
729 */
730bool ceph_con_opened(struct ceph_connection *con)
731{
732 return con->connect_seq > 0;
733}
734
31b8006e
SW		 735/*
736 * initialize a new connection.
737 */
1bfd89f4
AE		 738void ceph_con_init(struct ceph_connection *con, void *private,
739 const struct ceph_connection_operations *ops,
b7a9e5dd 740 struct ceph_messenger *msgr)
31b8006e
SW		 741{
742 dout("con_init %p\n", con);
743 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 744 con->private = private;
745 con->ops = ops;
31b8006e 746 con->msgr = msgr;
ce2c8903
AE		 747
748 con_sock_state_init(con);
749
ec302645 750 mutex_init(&con->mutex);
31b8006e
SW		 751 INIT_LIST_HEAD(&con->out_queue);
752 INIT_LIST_HEAD(&con->out_sent);
68931622 753 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
a5988c49 754
8dacc7da 755 con->state = CON_STATE_CLOSED;
31b8006e 756}
3d14c5d2 757EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 758
759
760/*
761 * We maintain a global counter to order connection attempts. Get
762 * a unique seq greater than @gt.
763 */
764static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
765{
766 u32 ret;
767
768 spin_lock(&msgr->global_seq_lock);
769 if (msgr->global_seq < gt)
770 msgr->global_seq = gt;
771 ret = ++msgr->global_seq;
772 spin_unlock(&msgr->global_seq_lock);
773 return ret;
774}
775
e2200423 776static void con_out_kvec_reset(struct ceph_connection *con)
859eb799
AE		 777{
778 con->out_kvec_left = 0;
779 con->out_kvec_bytes = 0;
780 con->out_kvec_cur = &con->out_kvec[0];
781}
782
e2200423 783static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 784 size_t size, void *data)
785{
786 int index;
787
788 index = con->out_kvec_left;
789 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
790
791 con->out_kvec[index].iov_len = size;
792 con->out_kvec[index].iov_base = data;
793 con->out_kvec_left++;
794 con->out_kvec_bytes += size;
795}
31b8006e 796
df6ad1f9 797#ifdef CONFIG_BLOCK
6aaa4511
AE		 798
799/*
800 * For a bio data item, a piece is whatever remains of the next
801 * entry in the current bio iovec, or the first entry in the next
802 * bio in the list.
803 */
8ae4f4f5 804static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 805 size_t length)
6aaa4511 806{
8ae4f4f5 807 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 808 struct bio *bio;
809
810 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
811
812 bio = data->bio;
813 BUG_ON(!bio);
6aaa4511 814
ca8b3a69 815 cursor->resid = min(length, data->bio_length);
6aaa4511 816 cursor->bio = bio;
f38a5181
KO		 817 cursor->bvec_iter = bio->bi_iter;
818 cursor->last_piece =
819 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
6aaa4511
AE		 820}
821
8ae4f4f5 822static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 823 size_t *page_offset,
824 size_t *length)
825{
8ae4f4f5 826 struct ceph_msg_data *data = cursor->data;
6aaa4511 827 struct bio *bio;
f38a5181 828 struct bio_vec bio_vec;
6aaa4511
AE		 829
830 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
831
832 bio = cursor->bio;
833 BUG_ON(!bio);
834
f38a5181 835 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511 836
f38a5181 837 *page_offset = (size_t) bio_vec.bv_offset;
6aaa4511 838 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 839 if (cursor->last_piece) /* pagelist offset is always 0 */
840 *length = cursor->resid;
841 else
f38a5181 842 *length = (size_t) bio_vec.bv_len;
25aff7c5 843 BUG_ON(*length > cursor->resid);
5df521b1 844 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511 845
f38a5181 846 return bio_vec.bv_page;
6aaa4511
AE		 847}
848
8ae4f4f5
AE		 849static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
850 size_t bytes)
6aaa4511 851{
6aaa4511 852 struct bio *bio;
f38a5181 853 struct bio_vec bio_vec;
6aaa4511 854
8ae4f4f5 855 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
6aaa4511
AE		 856
857 bio = cursor->bio;
858 BUG_ON(!bio);
859
f38a5181 860 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511
AE		 861
862 /* Advance the cursor offset */
863
25aff7c5
AE		 864 BUG_ON(cursor->resid < bytes);
865 cursor->resid -= bytes;
f38a5181
KO		 866
867 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
868
869 if (bytes < bio_vec.bv_len)
6aaa4511
AE		 870 return false; /* more bytes to process in this segment */
871
872 /* Move on to the next segment, and possibly the next bio */
873
f38a5181 874 if (!cursor->bvec_iter.bi_size) {
6aaa4511 875 bio = bio->bi_next;
0ec1d15e
ID		 876 cursor->bio = bio;
877 if (bio)
878 cursor->bvec_iter = bio->bi_iter;
879 else
880 memset(&cursor->bvec_iter, 0,
881 sizeof(cursor->bvec_iter));
6aaa4511 882 }
6aaa4511 883
25aff7c5
AE		 884 if (!cursor->last_piece) {
885 BUG_ON(!cursor->resid);
886 BUG_ON(!bio);
887 /* A short read is OK, so use <= rather than == */
f38a5181 888 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
6aaa4511 889 cursor->last_piece = true;
25aff7c5 890 }
6aaa4511
AE		 891
892 return true;
893}
ea96571f 894#endif /* CONFIG_BLOCK */
df6ad1f9 895
e766d7b5
AE		 896/*
897 * For a page array, a piece comes from the first page in the array
898 * that has not already been fully consumed.
899 */
8ae4f4f5 900static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 901 size_t length)
e766d7b5 902{
8ae4f4f5 903 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 904 int page_count;
905
906 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
907
908 BUG_ON(!data->pages);
909 BUG_ON(!data->length);
910
ca8b3a69 911 cursor->resid = min(length, data->length);
e766d7b5 912 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 913 cursor->page_offset = data->alignment & ~PAGE_MASK;
914 cursor->page_index = 0;
56fc5659
AE		 915 BUG_ON(page_count > (int)USHRT_MAX);
916 cursor->page_count = (unsigned short)page_count;
917 BUG_ON(length > SIZE_MAX - cursor->page_offset);
5f740d7e 918 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 919}
920
8ae4f4f5
AE		 921static struct page *
922ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
923 size_t *page_offset, size_t *length)
e766d7b5 924{
8ae4f4f5 925 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 926
927 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
928
929 BUG_ON(cursor->page_index >= cursor->page_count);
930 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 931
932 *page_offset = cursor->page_offset;
25aff7c5 933 if (cursor->last_piece)
e766d7b5 934 *length = cursor->resid;
25aff7c5 935 else
e766d7b5 936 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 937
938 return data->pages[cursor->page_index];
939}
940
8ae4f4f5 941static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 942 size_t bytes)
943{
8ae4f4f5 944 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 945
946 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 947
948 /* Advance the cursor page offset */
949
950 cursor->resid -= bytes;
5df521b1
AE		 951 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
952 if (!bytes || cursor->page_offset)
e766d7b5
AE		 953 return false; /* more bytes to process in the current page */
954
d90deda6
YZ		 955 if (!cursor->resid)
956 return false; /* no more data */
957
5df521b1 958 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 959
960 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 961 cursor->page_index++;
25aff7c5 962 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 963
964 return true;
965}
966
fe38a2b6 967/*
dd236fcb
AE		 968 * For a pagelist, a piece is whatever remains to be consumed in the
969 * first page in the list, or the front of the next page.
fe38a2b6 970 */
8ae4f4f5
AE		 971static void
972ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 973 size_t length)
fe38a2b6 974{
8ae4f4f5 975 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 976 struct ceph_pagelist *pagelist;
977 struct page *page;
978
dd236fcb 979 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 980
981 pagelist = data->pagelist;
982 BUG_ON(!pagelist);
25aff7c5
AE		 983
984 if (!length)
fe38a2b6
AE		 985 return; /* pagelist can be assigned but empty */
986
987 BUG_ON(list_empty(&pagelist->head));
988 page = list_first_entry(&pagelist->head, struct page, lru);
989
ca8b3a69 990 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 991 cursor->page = page;
992 cursor->offset = 0;
a51b272e 993 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 994}
995
8ae4f4f5
AE		 996static struct page *
997ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
998 size_t *page_offset, size_t *length)
fe38a2b6 999{
8ae4f4f5 1000 struct ceph_msg_data *data = cursor->data;
fe38a2b6 1001 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 1002
1003 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1004
1005 pagelist = data->pagelist;
1006 BUG_ON(!pagelist);
1007
1008 BUG_ON(!cursor->page);
25aff7c5 1009 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 1010
5df521b1 1011 /* offset of first page in pagelist is always 0 */
fe38a2b6 1012 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 1013 if (cursor->last_piece)
25aff7c5
AE		 1014 *length = cursor->resid;
1015 else
1016 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1017
8ae4f4f5 1018 return cursor->page;
fe38a2b6
AE		 1019}
1020
8ae4f4f5 1021static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1022 size_t bytes)
fe38a2b6 1023{
8ae4f4f5 1024 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1025 struct ceph_pagelist *pagelist;
1026
1027 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1028
1029 pagelist = data->pagelist;
1030 BUG_ON(!pagelist);
25aff7c5
AE		 1031
1032 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1033 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1034
1035 /* Advance the cursor offset */
1036
25aff7c5 1037 cursor->resid -= bytes;
fe38a2b6 1038 cursor->offset += bytes;
5df521b1 1039 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1040 if (!bytes || cursor->offset & ~PAGE_MASK)
1041 return false; /* more bytes to process in the current page */
1042
d90deda6
YZ		 1043 if (!cursor->resid)
1044 return false; /* no more data */
1045
fe38a2b6
AE		 1046 /* Move on to the next page */
1047
1048 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1049 cursor->page = list_entry_next(cursor->page, lru);
25aff7c5 1050 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1051
1052 return true;
1053}
1054
dd236fcb
AE		 1055/*
1056 * Message data is handled (sent or received) in pieces, where each
1057 * piece resides on a single page. The network layer might not
1058 * consume an entire piece at once. A data item's cursor keeps
1059 * track of which piece is next to process and how much remains to
1060 * be processed in that piece. It also tracks whether the current
1061 * piece is the last one in the data item.
1062 */
ca8b3a69 1063static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1064{
ca8b3a69 1065 size_t length = cursor->total_resid;
8ae4f4f5 1066
8ae4f4f5 1067 switch (cursor->data->type) {
dd236fcb 1068 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1069 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1070 break;
e766d7b5 1071 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1072 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1073 break;
dd236fcb
AE		 1074#ifdef CONFIG_BLOCK
1075 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1076 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1077 break;
dd236fcb 1078#endif /* CONFIG_BLOCK */
6aaa4511 1079 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1080 default:
1081 /* BUG(); */
1082 break;
1083 }
8ae4f4f5 1084 cursor->need_crc = true;
dd236fcb
AE		 1085}
1086
ca8b3a69
AE		 1087static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1088{
1089 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1090 struct ceph_msg_data *data;
1091
1092 BUG_ON(!length);
1093 BUG_ON(length > msg->data_length);
1094 BUG_ON(list_empty(&msg->data));
1095
ca8b3a69
AE		 1096 cursor->data_head = &msg->data;
1097 cursor->total_resid = length;
1098 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1099 cursor->data = data;
1100
1101 __ceph_msg_data_cursor_init(cursor);
1102}
1103
dd236fcb
AE		 1104/*
1105 * Return the page containing the next piece to process for a given
1106 * data item, and supply the page offset and length of that piece.
1107 * Indicate whether this is the last piece in this data item.
1108 */
8ae4f4f5
AE		 1109static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1110 size_t *page_offset, size_t *length,
dd236fcb
AE		 1111 bool *last_piece)
1112{
1113 struct page *page;
1114
8ae4f4f5 1115 switch (cursor->data->type) {
dd236fcb 1116 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1117 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1118 break;
e766d7b5 1119 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1120 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1121 break;
dd236fcb
AE		 1122#ifdef CONFIG_BLOCK
1123 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1124 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1125 break;
dd236fcb 1126#endif /* CONFIG_BLOCK */
6aaa4511 1127 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1128 default:
1129 page = NULL;
1130 break;
1131 }
1132 BUG_ON(!page);
1133 BUG_ON(*page_offset + *length > PAGE_SIZE);
1134 BUG_ON(!*length);
1135 if (last_piece)
8ae4f4f5 1136 *last_piece = cursor->last_piece;
dd236fcb
AE		 1137
1138 return page;
1139}
1140
1141/*
1142 * Returns true if the result moves the cursor on to the next piece
1143 * of the data item.
1144 */
8ae4f4f5
AE		 1145static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1146 size_t bytes)
dd236fcb
AE		 1147{
1148 bool new_piece;
1149
25aff7c5 1150 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1151 switch (cursor->data->type) {
dd236fcb 1152 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1153 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1154 break;
e766d7b5 1155 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1156 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1157 break;
dd236fcb
AE		 1158#ifdef CONFIG_BLOCK
1159 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1160 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1161 break;
dd236fcb 1162#endif /* CONFIG_BLOCK */
6aaa4511 1163 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1164 default:
1165 BUG();
1166 break;
1167 }
ca8b3a69 1168 cursor->total_resid -= bytes;
dd236fcb 1169
ca8b3a69
AE		 1170 if (!cursor->resid && cursor->total_resid) {
1171 WARN_ON(!cursor->last_piece);
1172 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1173 cursor->data = list_entry_next(cursor->data, links);
1174 __ceph_msg_data_cursor_init(cursor);
a51b272e 1175 new_piece = true;
ca8b3a69 1176 }
a51b272e 1177 cursor->need_crc = new_piece;
ca8b3a69 1178
dd236fcb
AE		 1179 return new_piece;
1180}
1181
98fa5dd8 1182static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1183{
739c905b 1184 BUG_ON(!msg);
25aff7c5 1185 BUG_ON(!data_len);
739c905b 1186
4c59b4a2 1187 /* Initialize data cursor */
fe38a2b6 1188
8ae4f4f5 1189 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1190}
1191
31b8006e
SW		 1192/*
1193 * Prepare footer for currently outgoing message, and finish things
1194 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1195 */
859eb799 1196static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1197{
1198 struct ceph_msg *m = con->out_msg;
859eb799 1199 int v = con->out_kvec_left;
31b8006e 1200
fd154f3c
AE		 1201 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1202
31b8006e
SW		 1203 dout("prepare_write_message_footer %p\n", con);
1204 con->out_kvec_is_msg = true;
1205 con->out_kvec[v].iov_base = &m->footer;
33d07337
YZ		 1206 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1207 if (con->ops->sign_message)
79dbd1ba 1208 con->ops->sign_message(m);
33d07337
YZ		 1209 else
1210 m->footer.sig = 0;
1211 con->out_kvec[v].iov_len = sizeof(m->footer);
1212 con->out_kvec_bytes += sizeof(m->footer);
1213 } else {
1214 m->old_footer.flags = m->footer.flags;
1215 con->out_kvec[v].iov_len = sizeof(m->old_footer);
1216 con->out_kvec_bytes += sizeof(m->old_footer);
1217 }
31b8006e
SW		 1218 con->out_kvec_left++;
1219 con->out_more = m->more_to_follow;
c86a2930 1220 con->out_msg_done = true;
31b8006e
SW		 1221}
1222
1223/*
1224 * Prepare headers for the next outgoing message.
1225 */
1226static void prepare_write_message(struct ceph_connection *con)
1227{
1228 struct ceph_msg *m;
a9a0c51a 1229 u32 crc;
31b8006e 1230
e2200423 1231 con_out_kvec_reset(con);
31b8006e 1232 con->out_kvec_is_msg = true;
c86a2930 1233 con->out_msg_done = false;
31b8006e
SW		 1234
1235 /* Sneak an ack in there first? If we can get it into the same
1236 * TCP packet that's a good thing. */
1237 if (con->in_seq > con->in_seq_acked) {
1238 con->in_seq_acked = con->in_seq;
e2200423 1239 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1240 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1241 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1242 &con->out_temp_ack);
31b8006e
SW		 1243 }
1244
38941f80 1245 BUG_ON(list_empty(&con->out_queue));
859eb799 1246 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1247 con->out_msg = m;
38941f80 1248 BUG_ON(m->con != con);
4cf9d544
SW		 1249
1250 /* put message on sent list */
1251 ceph_msg_get(m);
1252 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1253
e84346b7
SW		 1254 /*
1255 * only assign outgoing seq # if we haven't sent this message
1256 * yet. if it is requeued, resend with it's original seq.
1257 */
1258 if (m->needs_out_seq) {
1259 m->hdr.seq = cpu_to_le64(++con->out_seq);
1260 m->needs_out_seq = false;
1261 }
98fa5dd8 1262 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1263
98fa5dd8 1264 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1265 m, con->out_seq, le16_to_cpu(m->hdr.type),
1266 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1267 m->data_length);
31b8006e
SW		 1268 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1269
1270 /* tag + hdr + front + middle */
e2200423
AE		 1271 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1272 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1273 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1274
31b8006e 1275 if (m->middle)
e2200423 1276 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1277 m->middle->vec.iov_base);
31b8006e
SW		 1278
1279 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1280 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1281 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1282 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1283
1284 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1285 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1286 if (m->middle) {
1287 crc = crc32c(0, m->middle->vec.iov_base,
1288 m->middle->vec.iov_len);
1289 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1290 } else
31b8006e 1291 con->out_msg->footer.middle_crc = 0;
739c905b 1292 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1293 le32_to_cpu(con->out_msg->footer.front_crc),
1294 le32_to_cpu(con->out_msg->footer.middle_crc));
1295
1296 /* is there a data payload? */
739c905b 1297 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1298 if (m->data_length) {
1299 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1300 con->out_more = 1; /* data + footer will follow */
1301 } else {
31b8006e 1302 /* no, queue up footer too and be done */
859eb799 1303 prepare_write_message_footer(con);
78625051 1304 }
31b8006e 1305
c9ffc77a 1306 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1307}
1308
1309/*
1310 * Prepare an ack.
1311 */
1312static void prepare_write_ack(struct ceph_connection *con)
1313{
1314 dout("prepare_write_ack %p %llu -> %llu\n", con,
1315 con->in_seq_acked, con->in_seq);
1316 con->in_seq_acked = con->in_seq;
1317
e2200423 1318 con_out_kvec_reset(con);
859eb799 1319
e2200423 1320 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1321
31b8006e 1322 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1323 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1324 &con->out_temp_ack);
1325
31b8006e 1326 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1327 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1328}
1329
3a23083b
SW		 1330/*
1331 * Prepare to share the seq during handshake
1332 */
1333static void prepare_write_seq(struct ceph_connection *con)
1334{
1335 dout("prepare_write_seq %p %llu -> %llu\n", con,
1336 con->in_seq_acked, con->in_seq);
1337 con->in_seq_acked = con->in_seq;
1338
1339 con_out_kvec_reset(con);
1340
1341 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1342 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1343 &con->out_temp_ack);
1344
1345 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1346}
1347
31b8006e
SW		 1348/*
1349 * Prepare to write keepalive byte.
1350 */
1351static void prepare_write_keepalive(struct ceph_connection *con)
1352{
1353 dout("prepare_write_keepalive %p\n", con);
e2200423 1354 con_out_kvec_reset(con);
8b9558aa 1355 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
7f61f545
ID		 1356 struct timespec now = CURRENT_TIME;
1357
8b9558aa 1358 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
7f61f545
ID		 1359 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1360 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1361 &con->out_temp_keepalive2);
8b9558aa
YZ		 1362 } else {
1363 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1364 }
c9ffc77a 1365 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1366}
1367
1368/*
1369 * Connection negotiation.
1370 */
1371
dac1e716
AE		 1372static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1373 int *auth_proto)
4e7a5dcd 1374{
a3530df3 1375 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1376
1377 if (!con->ops->get_authorizer) {
1378 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1379 con->out_connect.authorizer_len = 0;
729796be 1380 return NULL;
b1c6b980
AE		 1381 }
1382
1383 /* Can't hold the mutex while getting authorizer */
ec302645 1384 mutex_unlock(&con->mutex);
dac1e716 1385 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1386 mutex_lock(&con->mutex);
4e7a5dcd 1387
a3530df3 1388 if (IS_ERR(auth))
729796be 1389 return auth;
8dacc7da 1390 if (con->state != CON_STATE_NEGOTIATING)
729796be 1391 return ERR_PTR(-EAGAIN);
0da5d703 1392
8f43fb53
AE		 1393 con->auth_reply_buf = auth->authorizer_reply_buf;
1394 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1395 return auth;
4e7a5dcd
SW		 1396}
1397
31b8006e
SW		 1398/*
1399 * We connected to a peer and are saying hello.
1400 */
e825a66d 1401static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1402{
e2200423
AE		 1403 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1404 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1405 &con->msgr->my_enc_addr);
eed0ef2c 1406
eed0ef2c 1407 con->out_more = 0;
c9ffc77a 1408 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1409}
1410
e825a66d 1411static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1412{
95c96174 1413 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1414 int proto;
dac1e716 1415 int auth_proto;
729796be 1416 struct ceph_auth_handshake *auth;
31b8006e
SW		 1417
1418 switch (con->peer_name.type) {
1419 case CEPH_ENTITY_TYPE_MON:
1420 proto = CEPH_MONC_PROTOCOL;
1421 break;
1422 case CEPH_ENTITY_TYPE_OSD:
1423 proto = CEPH_OSDC_PROTOCOL;
1424 break;
1425 case CEPH_ENTITY_TYPE_MDS:
1426 proto = CEPH_MDSC_PROTOCOL;
1427 break;
1428 default:
1429 BUG();
1430 }
1431
1432 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1433 con->connect_seq, global_seq, proto);
4e7a5dcd 1434
e825a66d 1435 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1436 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1437 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1438 con->out_connect.global_seq = cpu_to_le32(global_seq);
1439 con->out_connect.protocol_version = cpu_to_le32(proto);
1440 con->out_connect.flags = 0;
31b8006e 1441
dac1e716
AE		 1442 auth_proto = CEPH_AUTH_UNKNOWN;
1443 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1444 if (IS_ERR(auth))
1445 return PTR_ERR(auth);
3da54776 1446
dac1e716 1447 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1448 con->out_connect.authorizer_len = auth ?
1449 cpu_to_le32(auth->authorizer_buf_len) : 0;
1450
e2200423 1451 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1452 &con->out_connect);
1453 if (auth && auth->authorizer_buf_len)
e2200423 1454 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1455 auth->authorizer_buf);
859eb799 1456
31b8006e 1457 con->out_more = 0;
c9ffc77a 1458 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1459
e10c758e 1460 return 0;
31b8006e
SW		 1461}
1462
31b8006e
SW		 1463/*
1464 * write as much of pending kvecs to the socket as we can.
1465 * 1 -> done
1466 * 0 -> socket full, but more to do
1467 * <0 -> error
1468 */
1469static int write_partial_kvec(struct ceph_connection *con)
1470{
1471 int ret;
1472
1473 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1474 while (con->out_kvec_bytes > 0) {
1475 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1476 con->out_kvec_left, con->out_kvec_bytes,
1477 con->out_more);
1478 if (ret <= 0)
1479 goto out;
1480 con->out_kvec_bytes -= ret;
1481 if (con->out_kvec_bytes == 0)
1482 break; /* done */
f42299e6
AE		 1483
1484 /* account for full iov entries consumed */
1485 while (ret >= con->out_kvec_cur->iov_len) {
1486 BUG_ON(!con->out_kvec_left);
1487 ret -= con->out_kvec_cur->iov_len;
1488 con->out_kvec_cur++;
1489 con->out_kvec_left--;
1490 }
1491 /* and for a partially-consumed entry */
1492 if (ret) {
1493 con->out_kvec_cur->iov_len -= ret;
1494 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1495 }
1496 }
1497 con->out_kvec_left = 0;
1498 con->out_kvec_is_msg = false;
1499 ret = 1;
1500out:
1501 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1502 con->out_kvec_bytes, con->out_kvec_left, ret);
1503 return ret; /* done! */
1504}
1505
35b62808
AE		 1506static u32 ceph_crc32c_page(u32 crc, struct page *page,
1507 unsigned int page_offset,
1508 unsigned int length)
1509{
1510 char *kaddr;
1511
1512 kaddr = kmap(page);
1513 BUG_ON(kaddr == NULL);
1514 crc = crc32c(crc, kaddr + page_offset, length);
1515 kunmap(page);
1516
1517 return crc;
1518}
31b8006e
SW		 1519/*
1520 * Write as much message data payload as we can. If we finish, queue
1521 * up the footer.
1522 * 1 -> done, footer is now queued in out_kvec[].
1523 * 0 -> socket full, but more to do
1524 * <0 -> error
1525 */
34d2d200 1526static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1527{
1528 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1529 struct ceph_msg_data_cursor *cursor = &msg->cursor;
37675b0f 1530 bool do_datacrc = !con->msgr->nocrc;
f5db90bc 1531 u32 crc;
31b8006e 1532
859a35d5 1533 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1534
5240d9f9 1535 if (list_empty(&msg->data))
4c59b4a2
AE		 1536 return -EINVAL;
1537
5821bd8c
AE		 1538 /*
1539 * Iterate through each page that contains data to be
1540 * written, and send as much as possible for each.
1541 *
1542 * If we are calculating the data crc (the default), we will
1543 * need to map the page. If we have no pages, they have
1544 * been revoked, so use the zero page.
1545 */
f5db90bc 1546 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1547 while (cursor->resid) {
8a166d05 1548 struct page *page;
e387d525
AE		 1549 size_t page_offset;
1550 size_t length;
8a166d05 1551 bool last_piece;
8ea299bc 1552 bool need_crc;
f5db90bc 1553 int ret;
68b4476b 1554
343128ce
SB		 1555 page = ceph_msg_data_next(cursor, &page_offset, &length,
1556 &last_piece);
e387d525 1557 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
c2cfa194 1558 length, !last_piece);
f5db90bc
AE		 1559 if (ret <= 0) {
1560 if (do_datacrc)
1561 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1562
f5db90bc
AE		 1563 return ret;
1564 }
143334ff
AE		 1565 if (do_datacrc && cursor->need_crc)
1566 crc = ceph_crc32c_page(crc, page, page_offset, length);
343128ce 1567 need_crc = ceph_msg_data_advance(cursor, (size_t)ret);
31b8006e
SW		 1568 }
1569
34d2d200 1570 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1571
1572 /* prepare and queue up footer, too */
f5db90bc
AE		 1573 if (do_datacrc)
1574 msg->footer.data_crc = cpu_to_le32(crc);
1575 else
84ca8fc8 1576 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1577 con_out_kvec_reset(con);
859eb799 1578 prepare_write_message_footer(con);
f5db90bc
AE		 1579
1580 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1581}
1582
1583/*
1584 * write some zeros
1585 */
1586static int write_partial_skip(struct ceph_connection *con)
1587{
1588 int ret;
1589
1590 while (con->out_skip > 0) {
31739139 1591 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1592
e1dcb128 1593 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1594 if (ret <= 0)
1595 goto out;
1596 con->out_skip -= ret;
1597 }
1598 ret = 1;
1599out:
1600 return ret;
1601}
1602
1603/*
1604 * Prepare to read connection handshake, or an ack.
1605 */
eed0ef2c
SW		 1606static void prepare_read_banner(struct ceph_connection *con)
1607{
1608 dout("prepare_read_banner %p\n", con);
1609 con->in_base_pos = 0;
1610}
1611
31b8006e
SW		 1612static void prepare_read_connect(struct ceph_connection *con)
1613{
1614 dout("prepare_read_connect %p\n", con);
1615 con->in_base_pos = 0;
1616}
1617
1618static void prepare_read_ack(struct ceph_connection *con)
1619{
1620 dout("prepare_read_ack %p\n", con);
1621 con->in_base_pos = 0;
1622}
1623
3a23083b
SW		 1624static void prepare_read_seq(struct ceph_connection *con)
1625{
1626 dout("prepare_read_seq %p\n", con);
1627 con->in_base_pos = 0;
1628 con->in_tag = CEPH_MSGR_TAG_SEQ;
1629}
1630
31b8006e
SW		 1631static void prepare_read_tag(struct ceph_connection *con)
1632{
1633 dout("prepare_read_tag %p\n", con);
1634 con->in_base_pos = 0;
1635 con->in_tag = CEPH_MSGR_TAG_READY;
1636}
1637
8b9558aa
YZ		 1638static void prepare_read_keepalive_ack(struct ceph_connection *con)
1639{
1640 dout("prepare_read_keepalive_ack %p\n", con);
1641 con->in_base_pos = 0;
1642}
1643
31b8006e
SW		 1644/*
1645 * Prepare to read a message.
1646 */
1647static int prepare_read_message(struct ceph_connection *con)
1648{
1649 dout("prepare_read_message %p\n", con);
1650 BUG_ON(con->in_msg != NULL);
1651 con->in_base_pos = 0;
1652 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1653 return 0;
1654}
1655
1656
1657static int read_partial(struct ceph_connection *con,
fd51653f 1658 int end, int size, void *object)
31b8006e 1659{
e6cee71f
AE		 1660 while (con->in_base_pos < end) {
1661 int left = end - con->in_base_pos;
31b8006e
SW		 1662 int have = size - left;
1663 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1664 if (ret <= 0)
1665 return ret;
1666 con->in_base_pos += ret;
1667 }
1668 return 1;
1669}
1670
1671
1672/*
1673 * Read all or part of the connect-side handshake on a new connection
1674 */
eed0ef2c 1675static int read_partial_banner(struct ceph_connection *con)
31b8006e 1676{
fd51653f
AE		 1677 int size;
1678 int end;
1679 int ret;
31b8006e 1680
eed0ef2c 1681 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1682
1683 /* peer's banner */
fd51653f
AE		 1684 size = strlen(CEPH_BANNER);
1685 end = size;
1686 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1687 if (ret <= 0)
1688 goto out;
fd51653f
AE		 1689
1690 size = sizeof (con->actual_peer_addr);
1691 end += size;
1692 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1693 if (ret <= 0)
1694 goto out;
fd51653f
AE		 1695
1696 size = sizeof (con->peer_addr_for_me);
1697 end += size;
1698 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1699 if (ret <= 0)
1700 goto out;
fd51653f 1701
eed0ef2c
SW		 1702out:
1703 return ret;
1704}
1705
1706static int read_partial_connect(struct ceph_connection *con)
1707{
fd51653f
AE		 1708 int size;
1709 int end;
1710 int ret;
eed0ef2c
SW		 1711
1712 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1713
fd51653f
AE		 1714 size = sizeof (con->in_reply);
1715 end = size;
1716 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1717 if (ret <= 0)
1718 goto out;
fd51653f
AE		 1719
1720 size = le32_to_cpu(con->in_reply.authorizer_len);
1721 end += size;
1722 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1723 if (ret <= 0)
1724 goto out;
31b8006e 1725
4e7a5dcd
SW		 1726 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1727 con, (int)con->in_reply.tag,
1728 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1729 le32_to_cpu(con->in_reply.global_seq));
1730out:
1731 return ret;
eed0ef2c 1732
31b8006e
SW		 1733}
1734
1735/*
1736 * Verify the hello banner looks okay.
1737 */
1738static int verify_hello(struct ceph_connection *con)
1739{
1740 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1741 pr_err("connect to %s got bad banner\n",
3d14c5d2 1742 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1743 con->error_msg = "protocol error, bad banner";
1744 return -1;
1745 }
1746 return 0;
1747}
1748
1749static bool addr_is_blank(struct sockaddr_storage *ss)
1750{
c44bd69c
ID		 1751 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1752 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1753
31b8006e
SW		 1754 switch (ss->ss_family) {
1755 case AF_INET:
c44bd69c 1756 return addr->s_addr == htonl(INADDR_ANY);
31b8006e 1757 case AF_INET6:
c44bd69c
ID		 1758 return ipv6_addr_any(addr6);
1759 default:
1760 return true;
31b8006e 1761 }
31b8006e
SW		 1762}
1763
1764static int addr_port(struct sockaddr_storage *ss)
1765{
1766 switch (ss->ss_family) {
1767 case AF_INET:
f28bcfbe 1768 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1769 case AF_INET6:
f28bcfbe 1770 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1771 }
1772 return 0;
1773}
1774
1775static void addr_set_port(struct sockaddr_storage *ss, int p)
1776{
1777 switch (ss->ss_family) {
1778 case AF_INET:
1779 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1780 break;
31b8006e
SW		 1781 case AF_INET6:
1782 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1783 break;
31b8006e
SW		 1784 }
1785}
1786
ee3b56f2
NW		 1787/*
1788 * Unlike other *_pton function semantics, zero indicates success.
1789 */
1790static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1791 char delim, const char **ipend)
1792{
99f0f3b2
AE		 1793 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1794 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1795
1796 memset(ss, 0, sizeof(*ss));
1797
1798 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1799 ss->ss_family = AF_INET;
1800 return 0;
1801 }
1802
1803 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1804 ss->ss_family = AF_INET6;
1805 return 0;
1806 }
1807
1808 return -EINVAL;
1809}
1810
1811/*
1812 * Extract hostname string and resolve using kernel DNS facility.
1813 */
1814#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1815static int ceph_dns_resolve_name(const char *name, size_t namelen,
1816 struct sockaddr_storage *ss, char delim, const char **ipend)
1817{
1818 const char *end, *delim_p;
1819 char *colon_p, *ip_addr = NULL;
1820 int ip_len, ret;
1821
1822 /*
1823 * The end of the hostname occurs immediately preceding the delimiter or
1824 * the port marker (':') where the delimiter takes precedence.
1825 */
1826 delim_p = memchr(name, delim, namelen);
1827 colon_p = memchr(name, ':', namelen);
1828
1829 if (delim_p && colon_p)
1830 end = delim_p < colon_p ? delim_p : colon_p;
1831 else if (!delim_p && colon_p)
1832 end = colon_p;
1833 else {
1834 end = delim_p;
1835 if (!end) /* case: hostname:/ */
1836 end = name + namelen;
1837 }
1838
1839 if (end <= name)
1840 return -EINVAL;
1841
1842 /* do dns_resolve upcall */
1843 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1844 if (ip_len > 0)
1845 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1846 else
1847 ret = -ESRCH;
1848
1849 kfree(ip_addr);
1850
1851 *ipend = end;
1852
1853 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1854 ret, ret ? "failed" : ceph_pr_addr(ss));
1855
1856 return ret;
1857}
1858#else
1859static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1860 struct sockaddr_storage *ss, char delim, const char **ipend)
1861{
1862 return -EINVAL;
1863}
1864#endif
1865
1866/*
1867 * Parse a server name (IP or hostname). If a valid IP address is not found
1868 * then try to extract a hostname to resolve using userspace DNS upcall.
1869 */
1870static int ceph_parse_server_name(const char *name, size_t namelen,
1871 struct sockaddr_storage *ss, char delim, const char **ipend)
1872{
1873 int ret;
1874
1875 ret = ceph_pton(name, namelen, ss, delim, ipend);
1876 if (ret)
1877 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1878
1879 return ret;
1880}
1881
31b8006e
SW		 1882/*
1883 * Parse an ip[:port] list into an addr array. Use the default
1884 * monitor port if a port isn't specified.
1885 */
1886int ceph_parse_ips(const char *c, const char *end,
1887 struct ceph_entity_addr *addr,
1888 int max_count, int *count)
1889{
ee3b56f2 1890 int i, ret = -EINVAL;
31b8006e
SW		 1891 const char *p = c;
1892
1893 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1894 for (i = 0; i < max_count; i++) {
1895 const char *ipend;
1896 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1897 int port;
39139f64
SW		 1898 char delim = ',';
1899
1900 if (*p == '[') {
1901 delim = ']';
1902 p++;
1903 }
31b8006e 1904
ee3b56f2
NW		 1905 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1906 if (ret)
31b8006e 1907 goto bad;
ee3b56f2
NW		 1908 ret = -EINVAL;
1909
31b8006e
SW		 1910 p = ipend;
1911
39139f64
SW		 1912 if (delim == ']') {
1913 if (*p != ']') {
1914 dout("missing matching ']'\n");
1915 goto bad;
1916 }
1917 p++;
1918 }
1919
31b8006e
SW		 1920 /* port? */
1921 if (p < end && *p == ':') {
1922 port = 0;
1923 p++;
1924 while (p < end && *p >= '0' && *p <= '9') {
1925 port = (port * 10) + (*p - '0');
1926 p++;
1927 }
f48db1e9
ID		 1928 if (port == 0)
1929 port = CEPH_MON_PORT;
1930 else if (port > 65535)
31b8006e
SW		 1931 goto bad;
1932 } else {
1933 port = CEPH_MON_PORT;
1934 }
1935
1936 addr_set_port(ss, port);
1937
3d14c5d2 1938 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1939
1940 if (p == end)
1941 break;
1942 if (*p != ',')
1943 goto bad;
1944 p++;
1945 }
1946
1947 if (p != end)
1948 goto bad;
1949
1950 if (count)
1951 *count = i + 1;
1952 return 0;
1953
1954bad:
39139f64 1955 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1956 return ret;
31b8006e 1957}
3d14c5d2 1958EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1959
eed0ef2c 1960static int process_banner(struct ceph_connection *con)
31b8006e 1961{
eed0ef2c 1962 dout("process_banner on %p\n", con);
31b8006e
SW		 1963
1964 if (verify_hello(con) < 0)
1965 return -1;
1966
63f2d211
SW		 1967 ceph_decode_addr(&con->actual_peer_addr);
1968 ceph_decode_addr(&con->peer_addr_for_me);
1969
31b8006e
SW		 1970 /*
1971 * Make sure the other end is who we wanted. note that the other
1972 * end may not yet know their ip address, so if it's 0.0.0.0, give
1973 * them the benefit of the doubt.
1974 */
103e2d3a
SW		 1975 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1976 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1977 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1978 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 1979 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1980 ceph_pr_addr(&con->peer_addr.in_addr),
1981 (int)le32_to_cpu(con->peer_addr.nonce),
1982 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1983 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1984 con->error_msg = "wrong peer at address";
31b8006e
SW		 1985 return -1;
1986 }
1987
1988 /*
1989 * did we learn our address?
1990 */
1991 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1992 int port = addr_port(&con->msgr->inst.addr.in_addr);
1993
1994 memcpy(&con->msgr->inst.addr.in_addr,
1995 &con->peer_addr_for_me.in_addr,
1996 sizeof(con->peer_addr_for_me.in_addr));
1997 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1998 encode_my_addr(con->msgr);
eed0ef2c 1999 dout("process_banner learned my addr is %s\n",
3d14c5d2 2000 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 2001 }
2002
eed0ef2c
SW		 2003 return 0;
2004}
2005
2006static int process_connect(struct ceph_connection *con)
2007{
3d14c5d2
YS		 2008 u64 sup_feat = con->msgr->supported_features;
2009 u64 req_feat = con->msgr->required_features;
2b3e0c90
ID		 2010 u64 server_feat = ceph_sanitize_features(
2011 le64_to_cpu(con->in_reply.features));
0da5d703 2012 int ret;
04a419f9 2013
eed0ef2c
SW		 2014 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2015
31b8006e 2016 switch (con->in_reply.tag) {
04a419f9
SW		 2017 case CEPH_MSGR_TAG_FEATURES:
2018 pr_err("%s%lld %s feature set mismatch,"
2019 " my %llx < server's %llx, missing %llx\n",
2020 ENTITY_NAME(con->peer_name),
3d14c5d2 2021 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2022 sup_feat, server_feat, server_feat & ~sup_feat);
2023 con->error_msg = "missing required protocol features";
0fa6ebc6 2024 reset_connection(con);
04a419f9
SW		 2025 return -1;
2026
31b8006e 2027 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2028 pr_err("%s%lld %s protocol version mismatch,"
2029 " my %d != server's %d\n",
2030 ENTITY_NAME(con->peer_name),
3d14c5d2 2031 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2032 le32_to_cpu(con->out_connect.protocol_version),
2033 le32_to_cpu(con->in_reply.protocol_version));
2034 con->error_msg = "protocol version mismatch";
0fa6ebc6 2035 reset_connection(con);
31b8006e
SW		 2036 return -1;
2037
4e7a5dcd
SW		 2038 case CEPH_MSGR_TAG_BADAUTHORIZER:
2039 con->auth_retry++;
2040 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2041 con->auth_retry);
2042 if (con->auth_retry == 2) {
2043 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2044 return -1;
2045 }
6d4221b5 2046 con_out_kvec_reset(con);
e825a66d 2047 ret = prepare_write_connect(con);
0da5d703
SW		 2048 if (ret < 0)
2049 return ret;
63733a0f 2050 prepare_read_connect(con);
4e7a5dcd 2051 break;
31b8006e
SW		 2052
2053 case CEPH_MSGR_TAG_RESETSESSION:
2054 /*
2055 * If we connected with a large connect_seq but the peer
2056 * has no record of a session with us (no connection, or
2057 * connect_seq == 0), they will send RESETSESION to indicate
2058 * that they must have reset their session, and may have
2059 * dropped messages.
2060 */
2061 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2062 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2063 pr_err("%s%lld %s connection reset\n",
2064 ENTITY_NAME(con->peer_name),
3d14c5d2 2065 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2066 reset_connection(con);
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
2073 /* Tell ceph about it. */
ec302645 2074 mutex_unlock(&con->mutex);
31b8006e
SW		 2075 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2076 if (con->ops->peer_reset)
2077 con->ops->peer_reset(con);
ec302645 2078 mutex_lock(&con->mutex);
8dacc7da 2079 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2080 return -EAGAIN;
31b8006e
SW		 2081 break;
2082
2083 case CEPH_MSGR_TAG_RETRY_SESSION:
2084 /*
2085 * If we sent a smaller connect_seq than the peer has, try
2086 * again with a larger value.
2087 */
5bdca4e0 2088 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2089 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2090 le32_to_cpu(con->in_reply.connect_seq));
2091 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2092 con_out_kvec_reset(con);
5a0f8fdd
AE		 2093 ret = prepare_write_connect(con);
2094 if (ret < 0)
2095 return ret;
31b8006e
SW		 2096 prepare_read_connect(con);
2097 break;
2098
2099 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2100 /*
2101 * If we sent a smaller global_seq than the peer has, try
2102 * again with a larger value.
2103 */
eed0ef2c 2104 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2105 con->peer_global_seq,
5bdca4e0 2106 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2107 get_global_seq(con->msgr,
5bdca4e0 2108 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2109 con_out_kvec_reset(con);
5a0f8fdd
AE		 2110 ret = prepare_write_connect(con);
2111 if (ret < 0)
2112 return ret;
31b8006e
SW		 2113 prepare_read_connect(con);
2114 break;
2115
3a23083b 2116 case CEPH_MSGR_TAG_SEQ:
31b8006e 2117 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2118 if (req_feat & ~server_feat) {
2119 pr_err("%s%lld %s protocol feature mismatch,"
2120 " my required %llx > server's %llx, need %llx\n",
2121 ENTITY_NAME(con->peer_name),
3d14c5d2 2122 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2123 req_feat, server_feat, req_feat & ~server_feat);
2124 con->error_msg = "missing required protocol features";
0fa6ebc6 2125 reset_connection(con);
04a419f9
SW		 2126 return -1;
2127 }
8dacc7da 2128
122070a2 2129 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2130 con->state = CON_STATE_OPEN;
20e55c4c 2131 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2132 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2133 con->connect_seq++;
aba558e2 2134 con->peer_features = server_feat;
31b8006e
SW		 2135 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2136 con->peer_global_seq,
2137 le32_to_cpu(con->in_reply.connect_seq),
2138 con->connect_seq);
2139 WARN_ON(con->connect_seq !=
2140 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2141
2142 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2143 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2144
85effe18 2145 con->delay = 0; /* reset backoff memory */
92ac41d0 2146
3a23083b
SW		 2147 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2148 prepare_write_seq(con);
2149 prepare_read_seq(con);
2150 } else {
2151 prepare_read_tag(con);
2152 }
31b8006e
SW		 2153 break;
2154
2155 case CEPH_MSGR_TAG_WAIT:
2156 /*
2157 * If there is a connection race (we are opening
2158 * connections to each other), one of us may just have
2159 * to WAIT. This shouldn't happen if we are the
2160 * client.
2161 */
04177882
SW		 2162 con->error_msg = "protocol error, got WAIT as client";
2163 return -1;
31b8006e
SW		 2164
2165 default:
31b8006e
SW		 2166 con->error_msg = "protocol error, garbage tag during connect";
2167 return -1;
2168 }
2169 return 0;
2170}
2171
2172
2173/*
2174 * read (part of) an ack
2175 */
2176static int read_partial_ack(struct ceph_connection *con)
2177{
fd51653f
AE		 2178 int size = sizeof (con->in_temp_ack);
2179 int end = size;
31b8006e 2180
fd51653f 2181 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2182}
2183
31b8006e
SW		 2184/*
2185 * We can finally discard anything that's been acked.
2186 */
2187static void process_ack(struct ceph_connection *con)
2188{
2189 struct ceph_msg *m;
2190 u64 ack = le64_to_cpu(con->in_temp_ack);
2191 u64 seq;
2192
31b8006e
SW		 2193 while (!list_empty(&con->out_sent)) {
2194 m = list_first_entry(&con->out_sent, struct ceph_msg,
2195 list_head);
2196 seq = le64_to_cpu(m->hdr.seq);
2197 if (seq > ack)
2198 break;
2199 dout("got ack for seq %llu type %d at %p\n", seq,
2200 le16_to_cpu(m->hdr.type), m);
4cf9d544 2201 m->ack_stamp = jiffies;
31b8006e
SW		 2202 ceph_msg_remove(m);
2203 }
31b8006e
SW		 2204 prepare_read_tag(con);
2205}
2206
2207
2450418c 2208static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2209 struct kvec *section,
2210 unsigned int sec_len, u32 *crc)
2450418c 2211{
68b4476b 2212 int ret, left;
2450418c
YS		 2213
2214 BUG_ON(!section);
2215
2216 while (section->iov_len < sec_len) {
2217 BUG_ON(section->iov_base == NULL);
2218 left = sec_len - section->iov_len;
2219 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2220 section->iov_len, left);
2221 if (ret <= 0)
2222 return ret;
2223 section->iov_len += ret;
2450418c 2224 }
fe3ad593
AE		 2225 if (section->iov_len == sec_len)
2226 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2227
2450418c
YS		 2228 return 1;
2229}
31b8006e 2230
34d2d200
AE		 2231static int read_partial_msg_data(struct ceph_connection *con)
2232{
2233 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2234 struct ceph_msg_data_cursor *cursor = &msg->cursor;
34d2d200 2235 const bool do_datacrc = !con->msgr->nocrc;
686be208
AE		 2236 struct page *page;
2237 size_t page_offset;
2238 size_t length;
f5db90bc 2239 u32 crc = 0;
34d2d200
AE		 2240 int ret;
2241
2242 BUG_ON(!msg);
5240d9f9 2243 if (list_empty(&msg->data))
4c59b4a2 2244 return -EIO;
34d2d200 2245
f5db90bc
AE		 2246 if (do_datacrc)
2247 crc = con->in_data_crc;
643c68a4 2248 while (cursor->resid) {
343128ce 2249 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
686be208 2250 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2251 if (ret <= 0) {
2252 if (do_datacrc)
2253 con->in_data_crc = crc;
2254
686be208 2255 return ret;
f5db90bc 2256 }
686be208
AE		 2257
2258 if (do_datacrc)
f5db90bc 2259 crc = ceph_crc32c_page(crc, page, page_offset, ret);
343128ce 2260 (void) ceph_msg_data_advance(cursor, (size_t)ret);
34d2d200 2261 }
f5db90bc
AE		 2262 if (do_datacrc)
2263 con->in_data_crc = crc;
34d2d200
AE		 2264
2265 return 1; /* must return > 0 to indicate success */
2266}
2267
31b8006e
SW		 2268/*
2269 * read (part of) a message.
2270 */
686be208
AE		 2271static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2272
31b8006e
SW		 2273static int read_partial_message(struct ceph_connection *con)
2274{
2275 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2276 int size;
2277 int end;
31b8006e 2278 int ret;
95c96174 2279 unsigned int front_len, middle_len, data_len;
37675b0f 2280 bool do_datacrc = !con->msgr->nocrc;
33d07337 2281 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
ae18756b 2282 u64 seq;
fe3ad593 2283 u32 crc;
31b8006e
SW		 2284
2285 dout("read_partial_message con %p msg %p\n", con, m);
2286
2287 /* header */
fd51653f
AE		 2288 size = sizeof (con->in_hdr);
2289 end = size;
2290 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2291 if (ret <= 0)
2292 return ret;
fe3ad593
AE		 2293
2294 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2295 if (cpu_to_le32(crc) != con->in_hdr.crc) {
67c64eb7 2296 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
fe3ad593
AE		 2297 crc, con->in_hdr.crc);
2298 return -EBADMSG;
2299 }
2300
31b8006e
SW		 2301 front_len = le32_to_cpu(con->in_hdr.front_len);
2302 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2303 return -EIO;
2304 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2305 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2306 return -EIO;
2307 data_len = le32_to_cpu(con->in_hdr.data_len);
2308 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2309 return -EIO;
2310
ae18756b
SW		 2311 /* verify seq# */
2312 seq = le64_to_cpu(con->in_hdr.seq);
2313 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2314 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2315 ENTITY_NAME(con->peer_name),
3d14c5d2 2316 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2317 seq, con->in_seq + 1);
2318 con->in_base_pos = -front_len - middle_len - data_len -
2319 sizeof(m->footer);
2320 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2321 return 0;
2322 } else if ((s64)seq - (s64)con->in_seq > 1) {
2323 pr_err("read_partial_message bad seq %lld expected %lld\n",
2324 seq, con->in_seq + 1);
2325 con->error_msg = "bad message sequence # for incoming message";
67c64eb7 2326 return -EBADE;
ae18756b
SW		 2327 }
2328
31b8006e
SW		 2329 /* allocate message? */
2330 if (!con->in_msg) {
4740a623
SW		 2331 int skip = 0;
2332
31b8006e 2333 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2334 front_len, data_len);
4740a623
SW		 2335 ret = ceph_con_in_msg_alloc(con, &skip);
2336 if (ret < 0)
2337 return ret;
f759ebb9
AE		 2338
2339 BUG_ON(!con->in_msg ^ skip);
2450418c 2340 if (skip) {
31b8006e 2341 /* skip this message */
a79832f2 2342 dout("alloc_msg said skip message\n");
31b8006e
SW		 2343 con->in_base_pos = -front_len - middle_len - data_len -
2344 sizeof(m->footer);
2345 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2346 con->in_seq++;
31b8006e
SW		 2347 return 0;
2348 }
38941f80 2349
4740a623 2350 BUG_ON(!con->in_msg);
38941f80 2351 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2352 m = con->in_msg;
2353 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2354 if (m->middle)
2355 m->middle->vec.iov_len = 0;
9d7f0f13 2356
78625051 2357 /* prepare for data payload, if any */
a4107026 2358
78625051 2359 if (data_len)
98fa5dd8 2360 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2361 }
2362
2363 /* front */
2450418c
YS		 2364 ret = read_partial_message_section(con, &m->front, front_len,
2365 &con->in_front_crc);
2366 if (ret <= 0)
2367 return ret;
31b8006e
SW		 2368
2369 /* middle */
2450418c 2370 if (m->middle) {
213c99ee
SW		 2371 ret = read_partial_message_section(con, &m->middle->vec,
2372 middle_len,
2450418c 2373 &con->in_middle_crc);
31b8006e
SW		 2374 if (ret <= 0)
2375 return ret;
31b8006e
SW		 2376 }
2377
2378 /* (page) data */
34d2d200
AE		 2379 if (data_len) {
2380 ret = read_partial_msg_data(con);
2381 if (ret <= 0)
2382 return ret;
31b8006e
SW		 2383 }
2384
31b8006e 2385 /* footer */
33d07337
YZ		 2386 if (need_sign)
2387 size = sizeof(m->footer);
2388 else
2389 size = sizeof(m->old_footer);
2390
fd51653f
AE		 2391 end += size;
2392 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2393 if (ret <= 0)
2394 return ret;
2395
33d07337
YZ		 2396 if (!need_sign) {
2397 m->footer.flags = m->old_footer.flags;
2398 m->footer.sig = 0;
2399 }
2400
31b8006e
SW		 2401 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2402 m, front_len, m->footer.front_crc, middle_len,
2403 m->footer.middle_crc, data_len, m->footer.data_crc);
2404
2405 /* crc ok? */
2406 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2407 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2408 m, con->in_front_crc, m->footer.front_crc);
2409 return -EBADMSG;
2410 }
2411 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2412 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2413 m, con->in_middle_crc, m->footer.middle_crc);
2414 return -EBADMSG;
2415 }
bca064d2 2416 if (do_datacrc &&
31b8006e
SW		 2417 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2418 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2419 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2420 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2421 return -EBADMSG;
2422 }
2423
33d07337 2424 if (need_sign && con->ops->check_message_signature &&
79dbd1ba 2425 con->ops->check_message_signature(m)) {
33d07337
YZ		 2426 pr_err("read_partial_message %p signature check failed\n", m);
2427 return -EBADMSG;
2428 }
2429
31b8006e
SW		 2430 return 1; /* done! */
2431}
2432
2433/*
2434 * Process message. This happens in the worker thread. The callback should
2435 * be careful not to do anything that waits on other incoming messages or it
2436 * may deadlock.
2437 */
2438static void process_message(struct ceph_connection *con)
2439{
5e095e8b 2440 struct ceph_msg *msg;
31b8006e 2441
38941f80
AE		 2442 BUG_ON(con->in_msg->con != con);
2443 con->in_msg->con = NULL;
5e095e8b 2444 msg = con->in_msg;
31b8006e 2445 con->in_msg = NULL;
36eb71aa 2446 con->ops->put(con);
31b8006e
SW		 2447
2448 /* if first message, set peer_name */
2449 if (con->peer_name.type == 0)
dbad185d 2450 con->peer_name = msg->hdr.src;
31b8006e 2451
31b8006e 2452 con->in_seq++;
ec302645 2453 mutex_unlock(&con->mutex);
31b8006e
SW		 2454
2455 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2456 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2457 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2458 le16_to_cpu(msg->hdr.type),
2459 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2460 le32_to_cpu(msg->hdr.front_len),
2461 le32_to_cpu(msg->hdr.data_len),
2462 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2463 con->ops->dispatch(con, msg);
ec302645
SW		 2464
2465 mutex_lock(&con->mutex);
31b8006e
SW		 2466}
2467
8b9558aa
YZ		 2468static int read_keepalive_ack(struct ceph_connection *con)
2469{
2470 struct ceph_timespec ceph_ts;
2471 size_t size = sizeof(ceph_ts);
2472 int ret = read_partial(con, size, size, &ceph_ts);
2473 if (ret <= 0)
2474 return ret;
2475 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2476 prepare_read_tag(con);
2477 return 1;
2478}
31b8006e
SW		 2479
2480/*
2481 * Write something to the socket. Called in a worker thread when the
2482 * socket appears to be writeable and we have something ready to send.
2483 */
2484static int try_write(struct ceph_connection *con)
2485{
31b8006e
SW		 2486 int ret = 1;
2487
d59315ca 2488 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2489
31b8006e
SW		 2490more:
2491 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2492
2493 /* open the socket first? */
8dacc7da
SW		 2494 if (con->state == CON_STATE_PREOPEN) {
2495 BUG_ON(con->sock);
2496 con->state = CON_STATE_CONNECTING;
a5988c49 2497
e2200423 2498 con_out_kvec_reset(con);
e825a66d 2499 prepare_write_banner(con);
eed0ef2c 2500 prepare_read_banner(con);
31b8006e 2501
cf3e5c40 2502 BUG_ON(con->in_msg);
31b8006e
SW		 2503 con->in_tag = CEPH_MSGR_TAG_READY;
2504 dout("try_write initiating connect on %p new state %lu\n",
2505 con, con->state);
41617d0c
AE		 2506 ret = ceph_tcp_connect(con);
2507 if (ret < 0) {
31b8006e 2508 con->error_msg = "connect error";
31b8006e
SW		 2509 goto out;
2510 }
2511 }
2512
2513more_kvec:
2514 /* kvec data queued? */
2515 if (con->out_skip) {
2516 ret = write_partial_skip(con);
2517 if (ret <= 0)
42961d23 2518 goto out;
31b8006e
SW		 2519 }
2520 if (con->out_kvec_left) {
2521 ret = write_partial_kvec(con);
2522 if (ret <= 0)
42961d23 2523 goto out;
31b8006e
SW		 2524 }
2525
2526 /* msg pages? */
2527 if (con->out_msg) {
c86a2930
SW		 2528 if (con->out_msg_done) {
2529 ceph_msg_put(con->out_msg);
2530 con->out_msg = NULL; /* we're done with this one */
2531 goto do_next;
2532 }
2533
34d2d200 2534 ret = write_partial_message_data(con);
31b8006e
SW		 2535 if (ret == 1)
2536 goto more_kvec; /* we need to send the footer, too! */
2537 if (ret == 0)
42961d23 2538 goto out;
31b8006e 2539 if (ret < 0) {
34d2d200 2540 dout("try_write write_partial_message_data err %d\n",
31b8006e 2541 ret);
42961d23 2542 goto out;
31b8006e
SW		 2543 }
2544 }
2545
c86a2930 2546do_next:
8dacc7da 2547 if (con->state == CON_STATE_OPEN) {
8b9558aa
YZ		 2548 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2549 prepare_write_keepalive(con);
2550 goto more;
2551 }
31b8006e
SW		 2552 /* is anything else pending? */
2553 if (!list_empty(&con->out_queue)) {
2554 prepare_write_message(con);
2555 goto more;
2556 }
2557 if (con->in_seq > con->in_seq_acked) {
2558 prepare_write_ack(con);
2559 goto more;
2560 }
31b8006e
SW		 2561 }
2562
2563 /* Nothing to do! */
c9ffc77a 2564 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2565 dout("try_write nothing else to write.\n");
31b8006e
SW		 2566 ret = 0;
2567out:
42961d23 2568 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2569 return ret;
2570}
2571
2572
2573
2574/*
2575 * Read what we can from the socket.
2576 */
2577static int try_read(struct ceph_connection *con)
2578{
31b8006e
SW		 2579 int ret = -1;
2580
8dacc7da
SW		 2581more:
2582 dout("try_read start on %p state %lu\n", con, con->state);
2583 if (con->state != CON_STATE_CONNECTING &&
2584 con->state != CON_STATE_NEGOTIATING &&
2585 con->state != CON_STATE_OPEN)
31b8006e
SW		 2586 return 0;
2587
8dacc7da 2588 BUG_ON(!con->sock);
ec302645 2589
31b8006e
SW		 2590 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2591 con->in_base_pos);
0da5d703 2592
8dacc7da 2593 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2594 dout("try_read connecting\n");
2595 ret = read_partial_banner(con);
2596 if (ret <= 0)
ab166d5a 2597 goto out;
7593af92
AE		 2598 ret = process_banner(con);
2599 if (ret < 0)
2600 goto out;
2601
8dacc7da 2602 con->state = CON_STATE_NEGOTIATING;
7593af92 2603
6d4221b5
JS		 2604 /*
2605 * Received banner is good, exchange connection info.
2606 * Do not reset out_kvec, as sending our banner raced
2607 * with receiving peer banner after connect completed.
2608 */
7593af92
AE		 2609 ret = prepare_write_connect(con);
2610 if (ret < 0)
2611 goto out;
2612 prepare_read_connect(con);
2613
2614 /* Send connection info before awaiting response */
0da5d703
SW		 2615 goto out;
2616 }
2617
8dacc7da 2618 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2619 dout("try_read negotiating\n");
31b8006e
SW		 2620 ret = read_partial_connect(con);
2621 if (ret <= 0)
31b8006e 2622 goto out;
98bdb0aa
SW		 2623 ret = process_connect(con);
2624 if (ret < 0)
2625 goto out;
31b8006e
SW		 2626 goto more;
2627 }
2628
122070a2 2629 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2630
31b8006e
SW		 2631 if (con->in_base_pos < 0) {
2632 /*
2633 * skipping + discarding content.
2634 *
2635 * FIXME: there must be a better way to do this!
2636 */
84495f49
AE		 2637 static char buf[SKIP_BUF_SIZE];
2638 int skip = min((int) sizeof (buf), -con->in_base_pos);
2639
31b8006e
SW		 2640 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2641 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2642 if (ret <= 0)
98bdb0aa 2643 goto out;
31b8006e
SW		 2644 con->in_base_pos += ret;
2645 if (con->in_base_pos)
2646 goto more;
2647 }
2648 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2649 /*
2650 * what's next?
2651 */
2652 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2653 if (ret <= 0)
98bdb0aa 2654 goto out;
31b8006e
SW		 2655 dout("try_read got tag %d\n", (int)con->in_tag);
2656 switch (con->in_tag) {
2657 case CEPH_MSGR_TAG_MSG:
2658 prepare_read_message(con);
2659 break;
2660 case CEPH_MSGR_TAG_ACK:
2661 prepare_read_ack(con);
2662 break;
8b9558aa
YZ		 2663 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2664 prepare_read_keepalive_ack(con);
2665 break;
31b8006e 2666 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2667 con_close_socket(con);
2668 con->state = CON_STATE_CLOSED;
98bdb0aa 2669 goto out;
31b8006e
SW		 2670 default:
2671 goto bad_tag;
2672 }
2673 }
2674 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2675 ret = read_partial_message(con);
2676 if (ret <= 0) {
2677 switch (ret) {
2678 case -EBADMSG:
2679 con->error_msg = "bad crc";
67c64eb7
ID		 2680 /* fall through */
2681 case -EBADE:
31b8006e 2682 ret = -EIO;
98bdb0aa 2683 break;
31b8006e
SW		 2684 case -EIO:
2685 con->error_msg = "io error";
98bdb0aa 2686 break;
31b8006e 2687 }
98bdb0aa 2688 goto out;
31b8006e
SW		 2689 }
2690 if (con->in_tag == CEPH_MSGR_TAG_READY)
2691 goto more;
2692 process_message(con);
7b862e07
SW		 2693 if (con->state == CON_STATE_OPEN)
2694 prepare_read_tag(con);
31b8006e
SW		 2695 goto more;
2696 }
3a23083b
SW		 2697 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2698 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2699 /*
2700 * the final handshake seq exchange is semantically
2701 * equivalent to an ACK
2702 */
31b8006e
SW		 2703 ret = read_partial_ack(con);
2704 if (ret <= 0)
98bdb0aa 2705 goto out;
31b8006e
SW		 2706 process_ack(con);
2707 goto more;
2708 }
8b9558aa
YZ		 2709 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2710 ret = read_keepalive_ack(con);
2711 if (ret <= 0)
2712 goto out;
2713 goto more;
2714 }
31b8006e 2715
31b8006e 2716out:
98bdb0aa 2717 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2718 return ret;
2719
2720bad_tag:
2721 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2722 con->error_msg = "protocol error, garbage tag";
2723 ret = -1;
2724 goto out;
2725}
2726
2727
2728/*
802c6d96
AE		 2729 * Atomically queue work on a connection after the specified delay.
2730 * Bump @con reference to avoid races with connection teardown.
2731 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2732 */
802c6d96 2733static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2734{
31b8006e 2735 if (!con->ops->get(con)) {
802c6d96 2736 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2737 return -ENOENT;
31b8006e
SW		 2738 }
2739
802c6d96
AE		 2740 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2741 dout("%s %p - already queued\n", __func__, con);
31b8006e 2742 con->ops->put(con);
802c6d96 2743 return -EBUSY;
31b8006e 2744 }
802c6d96
AE		 2745
2746 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2747 return 0;
2748}
2749
2750static void queue_con(struct ceph_connection *con)
2751{
2752 (void) queue_con_delay(con, 0);
31b8006e
SW		 2753}
2754
37ab77ac
ID		 2755static void cancel_con(struct ceph_connection *con)
2756{
2757 if (cancel_delayed_work(&con->work)) {
2758 dout("%s %p\n", __func__, con);
2759 con->ops->put(con);
2760 }
2761}
2762
7bb21d68
AE		 2763static bool con_sock_closed(struct ceph_connection *con)
2764{
c9ffc77a 2765 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2766 return false;
2767
2768#define CASE(x) \
2769 case CON_STATE_ ## x: \
2770 con->error_msg = "socket closed (con state " #x ")"; \
2771 break;
2772
2773 switch (con->state) {
2774 CASE(CLOSED);
2775 CASE(PREOPEN);
2776 CASE(CONNECTING);
2777 CASE(NEGOTIATING);
2778 CASE(OPEN);
2779 CASE(STANDBY);
2780 default:
b9a67899 2781 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2782 __func__, con, con->state);
2783 con->error_msg = "unrecognized con state";
2784 BUG();
2785 break;
2786 }
2787#undef CASE
2788
2789 return true;
2790}
2791
f20a39fd
AE		 2792static bool con_backoff(struct ceph_connection *con)
2793{
2794 int ret;
2795
2796 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2797 return false;
2798
2799 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2800 if (ret) {
2801 dout("%s: con %p FAILED to back off %lu\n", __func__,
2802 con, con->delay);
2803 BUG_ON(ret == -ENOENT);
2804 con_flag_set(con, CON_FLAG_BACKOFF);
2805 }
2806
2807 return true;
2808}
2809
93209264
AE		 2810/* Finish fault handling; con->mutex must *not* be held here */
2811
2812static void con_fault_finish(struct ceph_connection *con)
2813{
2814 /*
2815 * in case we faulted due to authentication, invalidate our
2816 * current tickets so that we can get new ones.
2817 */
2818 if (con->auth_retry && con->ops->invalidate_authorizer) {
2819 dout("calling invalidate_authorizer()\n");
2820 con->ops->invalidate_authorizer(con);
2821 }
2822
2823 if (con->ops->fault)
2824 con->ops->fault(con);
2825}
2826
31b8006e
SW		 2827/*
2828 * Do some work on a connection. Drop a connection ref when we're done.
2829 */
68931622 2830static void ceph_con_workfn(struct work_struct *work)
31b8006e
SW		 2831{
2832 struct ceph_connection *con = container_of(work, struct ceph_connection,
2833 work.work);
49659416 2834 bool fault;
31b8006e 2835
9dd4658d 2836 mutex_lock(&con->mutex);
49659416
AE		 2837 while (true) {
2838 int ret;
31b8006e 2839
49659416
AE		 2840 if ((fault = con_sock_closed(con))) {
2841 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2842 break;
2843 }
2844 if (con_backoff(con)) {
2845 dout("%s: con %p BACKOFF\n", __func__, con);
2846 break;
2847 }
2848 if (con->state == CON_STATE_STANDBY) {
2849 dout("%s: con %p STANDBY\n", __func__, con);
2850 break;
2851 }
2852 if (con->state == CON_STATE_CLOSED) {
2853 dout("%s: con %p CLOSED\n", __func__, con);
2854 BUG_ON(con->sock);
2855 break;
2856 }
2857 if (con->state == CON_STATE_PREOPEN) {
2858 dout("%s: con %p PREOPEN\n", __func__, con);
2859 BUG_ON(con->sock);
2860 }
0da5d703 2861
49659416
AE		 2862 ret = try_read(con);
2863 if (ret < 0) {
2864 if (ret == -EAGAIN)
2865 continue;
67c64eb7
ID		 2866 if (!con->error_msg)
2867 con->error_msg = "socket error on read";
49659416
AE		 2868 fault = true;
2869 break;
2870 }
2871
2872 ret = try_write(con);
2873 if (ret < 0) {
2874 if (ret == -EAGAIN)
2875 continue;
67c64eb7
ID		 2876 if (!con->error_msg)
2877 con->error_msg = "socket error on write";
49659416
AE		 2878 fault = true;
2879 }
2880
2881 break; /* If we make it to here, we're done */
3a140a0d 2882 }
b6e7b6a1
AE		 2883 if (fault)
2884 con_fault(con);
9dd4658d 2885 mutex_unlock(&con->mutex);
0da5d703 2886
b6e7b6a1
AE		 2887 if (fault)
2888 con_fault_finish(con);
2889
2890 con->ops->put(con);
31b8006e
SW		 2891}
2892
31b8006e
SW		 2893/*
2894 * Generic error/fault handler. A retry mechanism is used with
2895 * exponential backoff
2896 */
93209264 2897static void con_fault(struct ceph_connection *con)
31b8006e 2898{
31b8006e 2899 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2900 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2901
67c64eb7
ID		 2902 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2903 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2904 con->error_msg = NULL;
2905
122070a2 2906 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2907 con->state != CON_STATE_NEGOTIATING &&
2908 con->state != CON_STATE_OPEN);
ec302645 2909
31b8006e 2910 con_close_socket(con);
5e095e8b 2911
c9ffc77a 2912 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2913 dout("fault on LOSSYTX channel, marking CLOSED\n");
2914 con->state = CON_STATE_CLOSED;
93209264 2915 return;
3b5ede07
SW		 2916 }
2917
5e095e8b 2918 if (con->in_msg) {
38941f80
AE		 2919 BUG_ON(con->in_msg->con != con);
2920 con->in_msg->con = NULL;
5e095e8b
SW		 2921 ceph_msg_put(con->in_msg);
2922 con->in_msg = NULL;
36eb71aa 2923 con->ops->put(con);
5e095e8b 2924 }
31b8006e 2925
e80a52d1
SW		 2926 /* Requeue anything that hasn't been acked */
2927 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2928
e76661d0
SW		 2929 /* If there are no messages queued or keepalive pending, place
2930 * the connection in a STANDBY state */
2931 if (list_empty(&con->out_queue) &&
c9ffc77a 2932 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2933 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2934 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2935 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2936 } else {
2937 /* retry after a delay. */
8dacc7da 2938 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2939 if (con->delay == 0)
2940 con->delay = BASE_DELAY_INTERVAL;
2941 else if (con->delay < MAX_DELAY_INTERVAL)
2942 con->delay *= 2;
c9ffc77a 2943 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2944 queue_con(con);
31b8006e 2945 }
31b8006e
SW		 2946}
2947
2948
2949
2950/*
15d9882c 2951 * initialize a new messenger instance
31b8006e 2952 */
15d9882c
AE		 2953void ceph_messenger_init(struct ceph_messenger *msgr,
2954 struct ceph_entity_addr *myaddr,
12b4629a
ID		 2955 u64 supported_features,
2956 u64 required_features,
ba988f87
CH		 2957 bool nocrc,
2958 bool tcp_nodelay)
31b8006e 2959{
3d14c5d2
YS		 2960 msgr->supported_features = supported_features;
2961 msgr->required_features = required_features;
2962
31b8006e
SW		 2963 spin_lock_init(&msgr->global_seq_lock);
2964
31b8006e
SW		 2965 if (myaddr)
2966 msgr->inst.addr = *myaddr;
2967
2968 /* select a random nonce */
ac8839d7 2969 msgr->inst.addr.type = 0;
103e2d3a 2970 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2971 encode_my_addr(msgr);
15d9882c 2972 msgr->nocrc = nocrc;
ba988f87 2973 msgr->tcp_nodelay = tcp_nodelay;
31b8006e 2974
a2a32584 2975 atomic_set(&msgr->stopping, 0);
757856d2 2976 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
31b8006e 2977
15d9882c 2978 dout("%s %p\n", __func__, msgr);
31b8006e 2979}
15d9882c 2980EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2981
757856d2
ID		 2982void ceph_messenger_fini(struct ceph_messenger *msgr)
2983{
2984 put_net(read_pnet(&msgr->net));
2985}
2986EXPORT_SYMBOL(ceph_messenger_fini);
2987
e00de341
SW		 2988static void clear_standby(struct ceph_connection *con)
2989{
2990 /* come back from STANDBY? */
8dacc7da 2991 if (con->state == CON_STATE_STANDBY) {
e00de341 2992 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2993 con->state = CON_STATE_PREOPEN;
e00de341 2994 con->connect_seq++;
c9ffc77a
AE		 2995 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2996 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2997 }
2998}
2999
31b8006e
SW		 3000/*
3001 * Queue up an outgoing message on the given connection.
3002 */
3003void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3004{
31b8006e 3005 /* set src+dst */
dbad185d 3006 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 3007 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 3008 msg->needs_out_seq = true;
3009
ec302645 3010 mutex_lock(&con->mutex);
92ce034b 3011
8dacc7da 3012 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 3013 dout("con_send %p closed, dropping %p\n", con, msg);
3014 ceph_msg_put(msg);
3015 mutex_unlock(&con->mutex);
3016 return;
3017 }
3018
38941f80 3019 BUG_ON(msg->con != NULL);
36eb71aa 3020 msg->con = con->ops->get(con);
92ce034b
AE		 3021 BUG_ON(msg->con == NULL);
3022
31b8006e
SW		 3023 BUG_ON(!list_empty(&msg->list_head));
3024 list_add_tail(&msg->list_head, &con->out_queue);
3025 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3026 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3027 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3028 le32_to_cpu(msg->hdr.front_len),
3029 le32_to_cpu(msg->hdr.middle_len),
3030 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 3031
3032 clear_standby(con);
ec302645 3033 mutex_unlock(&con->mutex);
31b8006e
SW		 3034
3035 /* if there wasn't anything waiting to send before, queue
3036 * new work */
c9ffc77a 3037 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3038 queue_con(con);
3039}
3d14c5d2 3040EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3041
3042/*
3043 * Revoke a message that was previously queued for send
3044 */
6740a845 3045void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3046{
6740a845
AE		 3047 struct ceph_connection *con = msg->con;
3048
3049 if (!con)
3050 return; /* Message not in our possession */
3051
ec302645 3052 mutex_lock(&con->mutex);
31b8006e 3053 if (!list_empty(&msg->list_head)) {
38941f80 3054 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3055 list_del_init(&msg->list_head);
38941f80 3056 BUG_ON(msg->con == NULL);
36eb71aa 3057 msg->con->ops->put(msg->con);
38941f80 3058 msg->con = NULL;
31b8006e 3059 msg->hdr.seq = 0;
38941f80 3060
31b8006e 3061 ceph_msg_put(msg);
ed98adad
SW		 3062 }
3063 if (con->out_msg == msg) {
38941f80 3064 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 3065 con->out_msg = NULL;
31b8006e
SW		 3066 if (con->out_kvec_is_msg) {
3067 con->out_skip = con->out_kvec_bytes;
3068 con->out_kvec_is_msg = false;
3069 }
ed98adad 3070 msg->hdr.seq = 0;
92ce034b
AE		 3071
3072 ceph_msg_put(msg);
31b8006e 3073 }
ec302645 3074 mutex_unlock(&con->mutex);
31b8006e
SW		 3075}
3076
350b1c32 3077/*
0d59ab81 3078 * Revoke a message that we may be reading data into
350b1c32 3079 */
8921d114 3080void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3081{
8921d114
AE		 3082 struct ceph_connection *con;
3083
3084 BUG_ON(msg == NULL);
3085 if (!msg->con) {
3086 dout("%s msg %p null con\n", __func__, msg);
3087
3088 return; /* Message not in our possession */
3089 }
3090
3091 con = msg->con;
350b1c32 3092 mutex_lock(&con->mutex);
8921d114 3093 if (con->in_msg == msg) {
95c96174
ED		 3094 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3095 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3096 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3097
3098 /* skip rest of message */
8921d114
AE		 3099 dout("%s %p msg %p revoked\n", __func__, con, msg);
3100 con->in_base_pos = con->in_base_pos -
350b1c32 3101 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3102 front_len -
3103 middle_len -
3104 data_len -
350b1c32 3105 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3106 ceph_msg_put(con->in_msg);
3107 con->in_msg = NULL;
3108 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3109 con->in_seq++;
350b1c32 3110 } else {
8921d114
AE		 3111 dout("%s %p in_msg %p msg %p no-op\n",
3112 __func__, con, con->in_msg, msg);
350b1c32
SW		 3113 }
3114 mutex_unlock(&con->mutex);
3115}
3116
31b8006e
SW		 3117/*
3118 * Queue a keepalive byte to ensure the tcp connection is alive.
3119 */
3120void ceph_con_keepalive(struct ceph_connection *con)
3121{
e00de341 3122 dout("con_keepalive %p\n", con);
00650931 3123 mutex_lock(&con->mutex);
e00de341 3124 clear_standby(con);
00650931 3125 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3126 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3127 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3128 queue_con(con);
3129}
3d14c5d2 3130EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3131
8b9558aa
YZ		 3132bool ceph_con_keepalive_expired(struct ceph_connection *con,
3133 unsigned long interval)
3134{
3135 if (interval > 0 &&
3136 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3137 struct timespec now = CURRENT_TIME;
3138 struct timespec ts;
3139 jiffies_to_timespec(interval, &ts);
3140 ts = timespec_add(con->last_keepalive_ack, ts);
3141 return timespec_compare(&now, &ts) >= 0;
3142 }
3143 return false;
3144}
3145
6644ed7b 3146static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3147{
6644ed7b
AE		 3148 struct ceph_msg_data *data;
3149
3150 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3151 return NULL;
3152
81b36be4 3153 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
6644ed7b
AE		 3154 if (data)
3155 data->type = type;
5240d9f9 3156 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3157
3158 return data;
3159}
3160
3161static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3162{
3163 if (!data)
3164 return;
3165
5240d9f9 3166 WARN_ON(!list_empty(&data->links));
e4339d28 3167 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3168 ceph_pagelist_release(data->pagelist);
81b36be4 3169 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3170}
3171
90af3602 3172void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3173 size_t length, size_t alignment)
02afca6c 3174{
6644ed7b
AE		 3175 struct ceph_msg_data *data;
3176
07aa1558
AE		 3177 BUG_ON(!pages);
3178 BUG_ON(!length);
6644ed7b
AE		 3179
3180 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3181 BUG_ON(!data);
3182 data->pages = pages;
3183 data->length = length;
3184 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3185
5240d9f9
AE		 3186 list_add_tail(&data->links, &msg->data);
3187 msg->data_length += length;
02afca6c 3188}
90af3602 3189EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3190
90af3602 3191void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3192 struct ceph_pagelist *pagelist)
3193{
6644ed7b
AE		 3194 struct ceph_msg_data *data;
3195
07aa1558
AE		 3196 BUG_ON(!pagelist);
3197 BUG_ON(!pagelist->length);
27fa8385 3198
6644ed7b
AE		 3199 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3200 BUG_ON(!data);
3201 data->pagelist = pagelist;
3202
5240d9f9
AE		 3203 list_add_tail(&data->links, &msg->data);
3204 msg->data_length += pagelist->length;
27fa8385 3205}
90af3602 3206EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3207
ea96571f 3208#ifdef CONFIG_BLOCK
90af3602 3209void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3210 size_t length)
27fa8385 3211{
6644ed7b
AE		 3212 struct ceph_msg_data *data;
3213
07aa1558 3214 BUG_ON(!bio);
27fa8385 3215
6644ed7b
AE		 3216 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3217 BUG_ON(!data);
3218 data->bio = bio;
c851c495 3219 data->bio_length = length;
6644ed7b 3220
5240d9f9
AE		 3221 list_add_tail(&data->links, &msg->data);
3222 msg->data_length += length;
27fa8385 3223}
90af3602 3224EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3225#endif /* CONFIG_BLOCK */
27fa8385 3226
31b8006e
SW		 3227/*
3228 * construct a new message with given type, size
3229 * the new msg has a ref count of 1.
3230 */
b61c2763
SW		 3231struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3232 bool can_fail)
31b8006e
SW		 3233{
3234 struct ceph_msg *m;
3235
e3d5d638 3236 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3237 if (m == NULL)
3238 goto out;
31b8006e
SW		 3239
3240 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3241 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3242 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3243
9516e45b
AE		 3244 INIT_LIST_HEAD(&m->list_head);
3245 kref_init(&m->kref);
5240d9f9 3246 INIT_LIST_HEAD(&m->data);
ca20892d 3247
31b8006e
SW		 3248 /* front */
3249 if (front_len) {
eeb0bed5 3250 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3251 if (m->front.iov_base == NULL) {
b61c2763 3252 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3253 front_len);
3254 goto out2;
3255 }
3256 } else {
3257 m->front.iov_base = NULL;
3258 }
f2be82b0 3259 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3260
bb257664 3261 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3262 return m;
3263
3264out2:
3265 ceph_msg_put(m);
3266out:
b61c2763
SW		 3267 if (!can_fail) {
3268 pr_err("msg_new can't create type %d front %d\n", type,
3269 front_len);
f0ed1b7c 3270 WARN_ON(1);
b61c2763
SW		 3271 } else {
3272 dout("msg_new can't create type %d front %d\n", type,
3273 front_len);
3274 }
a79832f2 3275 return NULL;
31b8006e 3276}
3d14c5d2 3277EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3278
31b8006e
SW		 3279/*
3280 * Allocate "middle" portion of a message, if it is needed and wasn't
3281 * allocated by alloc_msg. This allows us to read a small fixed-size
3282 * per-type header in the front and then gracefully fail (i.e.,
3283 * propagate the error to the caller based on info in the front) when
3284 * the middle is too large.
3285 */
2450418c 3286static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3287{
3288 int type = le16_to_cpu(msg->hdr.type);
3289 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3290
3291 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3292 ceph_msg_type_name(type), middle_len);
3293 BUG_ON(!middle_len);
3294 BUG_ON(msg->middle);
3295
b6c1d5b8 3296 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3297 if (!msg->middle)
3298 return -ENOMEM;
3299 return 0;
3300}
3301
2450418c 3302/*
1c20f2d2
AE		 3303 * Allocate a message for receiving an incoming message on a
3304 * connection, and save the result in con->in_msg. Uses the
3305 * connection's private alloc_msg op if available.
3306 *
4740a623
SW		 3307 * Returns 0 on success, or a negative error code.
3308 *
3309 * On success, if we set *skip = 1:
3310 * - the next message should be skipped and ignored.
3311 * - con->in_msg == NULL
3312 * or if we set *skip = 0:
3313 * - con->in_msg is non-null.
3314 * On error (ENOMEM, EAGAIN, ...),
3315 * - con->in_msg == NULL
2450418c 3316 */
4740a623 3317static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3318{
4740a623 3319 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3320 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3321 struct ceph_msg *msg;
4740a623 3322 int ret = 0;
2450418c 3323
1c20f2d2 3324 BUG_ON(con->in_msg != NULL);
53ded495 3325 BUG_ON(!con->ops->alloc_msg);
2450418c 3326
53ded495
AE		 3327 mutex_unlock(&con->mutex);
3328 msg = con->ops->alloc_msg(con, hdr, skip);
3329 mutex_lock(&con->mutex);
3330 if (con->state != CON_STATE_OPEN) {
3331 if (msg)
1d866d1c 3332 ceph_msg_put(msg);
53ded495
AE		 3333 return -EAGAIN;
3334 }
4137577a
AE		 3335 if (msg) {
3336 BUG_ON(*skip);
3337 con->in_msg = msg;
36eb71aa 3338 con->in_msg->con = con->ops->get(con);
92ce034b 3339 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3340 } else {
3341 /*
3342 * Null message pointer means either we should skip
3343 * this message or we couldn't allocate memory. The
3344 * former is not an error.
3345 */
3346 if (*skip)
3347 return 0;
4137577a 3348
67c64eb7 3349 con->error_msg = "error allocating memory for incoming message";
53ded495 3350 return -ENOMEM;
2450418c 3351 }
1c20f2d2 3352 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3353
1c20f2d2
AE		 3354 if (middle_len && !con->in_msg->middle) {
3355 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3356 if (ret < 0) {
1c20f2d2
AE		 3357 ceph_msg_put(con->in_msg);
3358 con->in_msg = NULL;
2450418c
YS		 3359 }
3360 }
9d7f0f13 3361
4740a623 3362 return ret;
2450418c
YS		 3363}
3364
31b8006e
SW		 3365
3366/*
3367 * Free a generically kmalloc'd message.
3368 */
0215e44b 3369static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3370{
0215e44b 3371 dout("%s %p\n", __func__, m);
4965fc38 3372 kvfree(m->front.iov_base);
e3d5d638 3373 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3374}
3375
0215e44b 3376static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3377{
3378 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
5240d9f9
AE		 3379 LIST_HEAD(data);
3380 struct list_head *links;
3381 struct list_head *next;
31b8006e 3382
0215e44b 3383 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3384 WARN_ON(!list_empty(&m->list_head));
3385
3386 /* drop middle, data, if any */
3387 if (m->middle) {
3388 ceph_buffer_put(m->middle);
3389 m->middle = NULL;
31b8006e 3390 }
5240d9f9
AE		 3391
3392 list_splice_init(&m->data, &data);
3393 list_for_each_safe(links, next, &data) {
3394 struct ceph_msg_data *data;
3395
3396 data = list_entry(links, struct ceph_msg_data, links);
3397 list_del_init(links);
3398 ceph_msg_data_destroy(data);
3399 }
a1930804 3400 m->data_length = 0;
58bb3b37 3401
c2e552e7
SW		 3402 if (m->pool)
3403 ceph_msgpool_put(m->pool, m);
3404 else
0215e44b
ID		 3405 ceph_msg_free(m);
3406}
3407
3408struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3409{
3410 dout("%s %p (was %d)\n", __func__, msg,
3411 atomic_read(&msg->kref.refcount));
3412 kref_get(&msg->kref);
3413 return msg;
3414}
3415EXPORT_SYMBOL(ceph_msg_get);
3416
3417void ceph_msg_put(struct ceph_msg *msg)
3418{
3419 dout("%s %p (was %d)\n", __func__, msg,
3420 atomic_read(&msg->kref.refcount));
3421 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3422}
0215e44b 3423EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3424
3425void ceph_msg_dump(struct ceph_msg *msg)
3426{
3cea4c30
ID		 3427 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3428 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3429 print_hex_dump(KERN_DEBUG, "header: ",
3430 DUMP_PREFIX_OFFSET, 16, 1,
3431 &msg->hdr, sizeof(msg->hdr), true);
3432 print_hex_dump(KERN_DEBUG, " front: ",
3433 DUMP_PREFIX_OFFSET, 16, 1,
3434 msg->front.iov_base, msg->front.iov_len, true);
3435 if (msg->middle)
3436 print_hex_dump(KERN_DEBUG, "middle: ",
3437 DUMP_PREFIX_OFFSET, 16, 1,
3438 msg->middle->vec.iov_base,
3439 msg->middle->vec.iov_len, true);
3440 print_hex_dump(KERN_DEBUG, "footer: ",
3441 DUMP_PREFIX_OFFSET, 16, 1,
3442 &msg->footer, sizeof(msg->footer), true);
3443}
3d14c5d2 3444EXPORT_SYMBOL(ceph_msg_dump);
Linux 6.x block layer and io_uring tree(s)