projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: add nocephx_sign_messages option
[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
859bff51 512 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
ba988f87
CH		 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
859bff51
ID		 1435 con->out_connect.features =
1436 cpu_to_le64(from_msgr(con->msgr)->supported_features);
31b8006e
SW		 1437 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1438 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1439 con->out_connect.global_seq = cpu_to_le32(global_seq);
1440 con->out_connect.protocol_version = cpu_to_le32(proto);
1441 con->out_connect.flags = 0;
31b8006e 1442
dac1e716
AE		 1443 auth_proto = CEPH_AUTH_UNKNOWN;
1444 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1445 if (IS_ERR(auth))
1446 return PTR_ERR(auth);
3da54776 1447
dac1e716 1448 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1449 con->out_connect.authorizer_len = auth ?
1450 cpu_to_le32(auth->authorizer_buf_len) : 0;
1451
e2200423 1452 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1453 &con->out_connect);
1454 if (auth && auth->authorizer_buf_len)
e2200423 1455 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1456 auth->authorizer_buf);
859eb799 1457
31b8006e 1458 con->out_more = 0;
c9ffc77a 1459 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1460
e10c758e 1461 return 0;
31b8006e
SW		 1462}
1463
31b8006e
SW		 1464/*
1465 * write as much of pending kvecs to the socket as we can.
1466 * 1 -> done
1467 * 0 -> socket full, but more to do
1468 * <0 -> error
1469 */
1470static int write_partial_kvec(struct ceph_connection *con)
1471{
1472 int ret;
1473
1474 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1475 while (con->out_kvec_bytes > 0) {
1476 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1477 con->out_kvec_left, con->out_kvec_bytes,
1478 con->out_more);
1479 if (ret <= 0)
1480 goto out;
1481 con->out_kvec_bytes -= ret;
1482 if (con->out_kvec_bytes == 0)
1483 break; /* done */
f42299e6
AE		 1484
1485 /* account for full iov entries consumed */
1486 while (ret >= con->out_kvec_cur->iov_len) {
1487 BUG_ON(!con->out_kvec_left);
1488 ret -= con->out_kvec_cur->iov_len;
1489 con->out_kvec_cur++;
1490 con->out_kvec_left--;
1491 }
1492 /* and for a partially-consumed entry */
1493 if (ret) {
1494 con->out_kvec_cur->iov_len -= ret;
1495 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1496 }
1497 }
1498 con->out_kvec_left = 0;
1499 con->out_kvec_is_msg = false;
1500 ret = 1;
1501out:
1502 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1503 con->out_kvec_bytes, con->out_kvec_left, ret);
1504 return ret; /* done! */
1505}
1506
35b62808
AE		 1507static u32 ceph_crc32c_page(u32 crc, struct page *page,
1508 unsigned int page_offset,
1509 unsigned int length)
1510{
1511 char *kaddr;
1512
1513 kaddr = kmap(page);
1514 BUG_ON(kaddr == NULL);
1515 crc = crc32c(crc, kaddr + page_offset, length);
1516 kunmap(page);
1517
1518 return crc;
1519}
31b8006e
SW		 1520/*
1521 * Write as much message data payload as we can. If we finish, queue
1522 * up the footer.
1523 * 1 -> done, footer is now queued in out_kvec[].
1524 * 0 -> socket full, but more to do
1525 * <0 -> error
1526 */
34d2d200 1527static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1528{
1529 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1530 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 1531 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
f5db90bc 1532 u32 crc;
31b8006e 1533
859a35d5 1534 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1535
5240d9f9 1536 if (list_empty(&msg->data))
4c59b4a2
AE		 1537 return -EINVAL;
1538
5821bd8c
AE		 1539 /*
1540 * Iterate through each page that contains data to be
1541 * written, and send as much as possible for each.
1542 *
1543 * If we are calculating the data crc (the default), we will
1544 * need to map the page. If we have no pages, they have
1545 * been revoked, so use the zero page.
1546 */
f5db90bc 1547 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1548 while (cursor->resid) {
8a166d05 1549 struct page *page;
e387d525
AE		 1550 size_t page_offset;
1551 size_t length;
8a166d05 1552 bool last_piece;
8ea299bc 1553 bool need_crc;
f5db90bc 1554 int ret;
68b4476b 1555
343128ce
SB		 1556 page = ceph_msg_data_next(cursor, &page_offset, &length,
1557 &last_piece);
e387d525 1558 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
c2cfa194 1559 length, !last_piece);
f5db90bc
AE		 1560 if (ret <= 0) {
1561 if (do_datacrc)
1562 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1563
f5db90bc
AE		 1564 return ret;
1565 }
143334ff
AE		 1566 if (do_datacrc && cursor->need_crc)
1567 crc = ceph_crc32c_page(crc, page, page_offset, length);
343128ce 1568 need_crc = ceph_msg_data_advance(cursor, (size_t)ret);
31b8006e
SW		 1569 }
1570
34d2d200 1571 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1572
1573 /* prepare and queue up footer, too */
f5db90bc
AE		 1574 if (do_datacrc)
1575 msg->footer.data_crc = cpu_to_le32(crc);
1576 else
84ca8fc8 1577 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1578 con_out_kvec_reset(con);
859eb799 1579 prepare_write_message_footer(con);
f5db90bc
AE		 1580
1581 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1582}
1583
1584/*
1585 * write some zeros
1586 */
1587static int write_partial_skip(struct ceph_connection *con)
1588{
1589 int ret;
1590
1591 while (con->out_skip > 0) {
31739139 1592 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1593
e1dcb128 1594 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1595 if (ret <= 0)
1596 goto out;
1597 con->out_skip -= ret;
1598 }
1599 ret = 1;
1600out:
1601 return ret;
1602}
1603
1604/*
1605 * Prepare to read connection handshake, or an ack.
1606 */
eed0ef2c
SW		 1607static void prepare_read_banner(struct ceph_connection *con)
1608{
1609 dout("prepare_read_banner %p\n", con);
1610 con->in_base_pos = 0;
1611}
1612
31b8006e
SW		 1613static void prepare_read_connect(struct ceph_connection *con)
1614{
1615 dout("prepare_read_connect %p\n", con);
1616 con->in_base_pos = 0;
1617}
1618
1619static void prepare_read_ack(struct ceph_connection *con)
1620{
1621 dout("prepare_read_ack %p\n", con);
1622 con->in_base_pos = 0;
1623}
1624
3a23083b
SW		 1625static void prepare_read_seq(struct ceph_connection *con)
1626{
1627 dout("prepare_read_seq %p\n", con);
1628 con->in_base_pos = 0;
1629 con->in_tag = CEPH_MSGR_TAG_SEQ;
1630}
1631
31b8006e
SW		 1632static void prepare_read_tag(struct ceph_connection *con)
1633{
1634 dout("prepare_read_tag %p\n", con);
1635 con->in_base_pos = 0;
1636 con->in_tag = CEPH_MSGR_TAG_READY;
1637}
1638
8b9558aa
YZ		 1639static void prepare_read_keepalive_ack(struct ceph_connection *con)
1640{
1641 dout("prepare_read_keepalive_ack %p\n", con);
1642 con->in_base_pos = 0;
1643}
1644
31b8006e
SW		 1645/*
1646 * Prepare to read a message.
1647 */
1648static int prepare_read_message(struct ceph_connection *con)
1649{
1650 dout("prepare_read_message %p\n", con);
1651 BUG_ON(con->in_msg != NULL);
1652 con->in_base_pos = 0;
1653 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1654 return 0;
1655}
1656
1657
1658static int read_partial(struct ceph_connection *con,
fd51653f 1659 int end, int size, void *object)
31b8006e 1660{
e6cee71f
AE		 1661 while (con->in_base_pos < end) {
1662 int left = end - con->in_base_pos;
31b8006e
SW		 1663 int have = size - left;
1664 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1665 if (ret <= 0)
1666 return ret;
1667 con->in_base_pos += ret;
1668 }
1669 return 1;
1670}
1671
1672
1673/*
1674 * Read all or part of the connect-side handshake on a new connection
1675 */
eed0ef2c 1676static int read_partial_banner(struct ceph_connection *con)
31b8006e 1677{
fd51653f
AE		 1678 int size;
1679 int end;
1680 int ret;
31b8006e 1681
eed0ef2c 1682 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1683
1684 /* peer's banner */
fd51653f
AE		 1685 size = strlen(CEPH_BANNER);
1686 end = size;
1687 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1688 if (ret <= 0)
1689 goto out;
fd51653f
AE		 1690
1691 size = sizeof (con->actual_peer_addr);
1692 end += size;
1693 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1694 if (ret <= 0)
1695 goto out;
fd51653f
AE		 1696
1697 size = sizeof (con->peer_addr_for_me);
1698 end += size;
1699 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1700 if (ret <= 0)
1701 goto out;
fd51653f 1702
eed0ef2c
SW		 1703out:
1704 return ret;
1705}
1706
1707static int read_partial_connect(struct ceph_connection *con)
1708{
fd51653f
AE		 1709 int size;
1710 int end;
1711 int ret;
eed0ef2c
SW		 1712
1713 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1714
fd51653f
AE		 1715 size = sizeof (con->in_reply);
1716 end = size;
1717 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1718 if (ret <= 0)
1719 goto out;
fd51653f
AE		 1720
1721 size = le32_to_cpu(con->in_reply.authorizer_len);
1722 end += size;
1723 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1724 if (ret <= 0)
1725 goto out;
31b8006e 1726
4e7a5dcd
SW		 1727 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1728 con, (int)con->in_reply.tag,
1729 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1730 le32_to_cpu(con->in_reply.global_seq));
1731out:
1732 return ret;
eed0ef2c 1733
31b8006e
SW		 1734}
1735
1736/*
1737 * Verify the hello banner looks okay.
1738 */
1739static int verify_hello(struct ceph_connection *con)
1740{
1741 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1742 pr_err("connect to %s got bad banner\n",
3d14c5d2 1743 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1744 con->error_msg = "protocol error, bad banner";
1745 return -1;
1746 }
1747 return 0;
1748}
1749
1750static bool addr_is_blank(struct sockaddr_storage *ss)
1751{
c44bd69c
ID		 1752 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1753 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1754
31b8006e
SW		 1755 switch (ss->ss_family) {
1756 case AF_INET:
c44bd69c 1757 return addr->s_addr == htonl(INADDR_ANY);
31b8006e 1758 case AF_INET6:
c44bd69c
ID		 1759 return ipv6_addr_any(addr6);
1760 default:
1761 return true;
31b8006e 1762 }
31b8006e
SW		 1763}
1764
1765static int addr_port(struct sockaddr_storage *ss)
1766{
1767 switch (ss->ss_family) {
1768 case AF_INET:
f28bcfbe 1769 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1770 case AF_INET6:
f28bcfbe 1771 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1772 }
1773 return 0;
1774}
1775
1776static void addr_set_port(struct sockaddr_storage *ss, int p)
1777{
1778 switch (ss->ss_family) {
1779 case AF_INET:
1780 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1781 break;
31b8006e
SW		 1782 case AF_INET6:
1783 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1784 break;
31b8006e
SW		 1785 }
1786}
1787
ee3b56f2
NW		 1788/*
1789 * Unlike other *_pton function semantics, zero indicates success.
1790 */
1791static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1792 char delim, const char **ipend)
1793{
99f0f3b2
AE		 1794 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1795 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1796
1797 memset(ss, 0, sizeof(*ss));
1798
1799 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1800 ss->ss_family = AF_INET;
1801 return 0;
1802 }
1803
1804 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1805 ss->ss_family = AF_INET6;
1806 return 0;
1807 }
1808
1809 return -EINVAL;
1810}
1811
1812/*
1813 * Extract hostname string and resolve using kernel DNS facility.
1814 */
1815#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1816static int ceph_dns_resolve_name(const char *name, size_t namelen,
1817 struct sockaddr_storage *ss, char delim, const char **ipend)
1818{
1819 const char *end, *delim_p;
1820 char *colon_p, *ip_addr = NULL;
1821 int ip_len, ret;
1822
1823 /*
1824 * The end of the hostname occurs immediately preceding the delimiter or
1825 * the port marker (':') where the delimiter takes precedence.
1826 */
1827 delim_p = memchr(name, delim, namelen);
1828 colon_p = memchr(name, ':', namelen);
1829
1830 if (delim_p && colon_p)
1831 end = delim_p < colon_p ? delim_p : colon_p;
1832 else if (!delim_p && colon_p)
1833 end = colon_p;
1834 else {
1835 end = delim_p;
1836 if (!end) /* case: hostname:/ */
1837 end = name + namelen;
1838 }
1839
1840 if (end <= name)
1841 return -EINVAL;
1842
1843 /* do dns_resolve upcall */
1844 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1845 if (ip_len > 0)
1846 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1847 else
1848 ret = -ESRCH;
1849
1850 kfree(ip_addr);
1851
1852 *ipend = end;
1853
1854 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1855 ret, ret ? "failed" : ceph_pr_addr(ss));
1856
1857 return ret;
1858}
1859#else
1860static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1861 struct sockaddr_storage *ss, char delim, const char **ipend)
1862{
1863 return -EINVAL;
1864}
1865#endif
1866
1867/*
1868 * Parse a server name (IP or hostname). If a valid IP address is not found
1869 * then try to extract a hostname to resolve using userspace DNS upcall.
1870 */
1871static int ceph_parse_server_name(const char *name, size_t namelen,
1872 struct sockaddr_storage *ss, char delim, const char **ipend)
1873{
1874 int ret;
1875
1876 ret = ceph_pton(name, namelen, ss, delim, ipend);
1877 if (ret)
1878 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1879
1880 return ret;
1881}
1882
31b8006e
SW		 1883/*
1884 * Parse an ip[:port] list into an addr array. Use the default
1885 * monitor port if a port isn't specified.
1886 */
1887int ceph_parse_ips(const char *c, const char *end,
1888 struct ceph_entity_addr *addr,
1889 int max_count, int *count)
1890{
ee3b56f2 1891 int i, ret = -EINVAL;
31b8006e
SW		 1892 const char *p = c;
1893
1894 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1895 for (i = 0; i < max_count; i++) {
1896 const char *ipend;
1897 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1898 int port;
39139f64
SW		 1899 char delim = ',';
1900
1901 if (*p == '[') {
1902 delim = ']';
1903 p++;
1904 }
31b8006e 1905
ee3b56f2
NW		 1906 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1907 if (ret)
31b8006e 1908 goto bad;
ee3b56f2
NW		 1909 ret = -EINVAL;
1910
31b8006e
SW		 1911 p = ipend;
1912
39139f64
SW		 1913 if (delim == ']') {
1914 if (*p != ']') {
1915 dout("missing matching ']'\n");
1916 goto bad;
1917 }
1918 p++;
1919 }
1920
31b8006e
SW		 1921 /* port? */
1922 if (p < end && *p == ':') {
1923 port = 0;
1924 p++;
1925 while (p < end && *p >= '0' && *p <= '9') {
1926 port = (port * 10) + (*p - '0');
1927 p++;
1928 }
f48db1e9
ID		 1929 if (port == 0)
1930 port = CEPH_MON_PORT;
1931 else if (port > 65535)
31b8006e
SW		 1932 goto bad;
1933 } else {
1934 port = CEPH_MON_PORT;
1935 }
1936
1937 addr_set_port(ss, port);
1938
3d14c5d2 1939 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1940
1941 if (p == end)
1942 break;
1943 if (*p != ',')
1944 goto bad;
1945 p++;
1946 }
1947
1948 if (p != end)
1949 goto bad;
1950
1951 if (count)
1952 *count = i + 1;
1953 return 0;
1954
1955bad:
39139f64 1956 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1957 return ret;
31b8006e 1958}
3d14c5d2 1959EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1960
eed0ef2c 1961static int process_banner(struct ceph_connection *con)
31b8006e 1962{
eed0ef2c 1963 dout("process_banner on %p\n", con);
31b8006e
SW		 1964
1965 if (verify_hello(con) < 0)
1966 return -1;
1967
63f2d211
SW		 1968 ceph_decode_addr(&con->actual_peer_addr);
1969 ceph_decode_addr(&con->peer_addr_for_me);
1970
31b8006e
SW		 1971 /*
1972 * Make sure the other end is who we wanted. note that the other
1973 * end may not yet know their ip address, so if it's 0.0.0.0, give
1974 * them the benefit of the doubt.
1975 */
103e2d3a
SW		 1976 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1977 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1978 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1979 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 1980 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1981 ceph_pr_addr(&con->peer_addr.in_addr),
1982 (int)le32_to_cpu(con->peer_addr.nonce),
1983 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1984 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1985 con->error_msg = "wrong peer at address";
31b8006e
SW		 1986 return -1;
1987 }
1988
1989 /*
1990 * did we learn our address?
1991 */
1992 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1993 int port = addr_port(&con->msgr->inst.addr.in_addr);
1994
1995 memcpy(&con->msgr->inst.addr.in_addr,
1996 &con->peer_addr_for_me.in_addr,
1997 sizeof(con->peer_addr_for_me.in_addr));
1998 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1999 encode_my_addr(con->msgr);
eed0ef2c 2000 dout("process_banner learned my addr is %s\n",
3d14c5d2 2001 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 2002 }
2003
eed0ef2c
SW		 2004 return 0;
2005}
2006
2007static int process_connect(struct ceph_connection *con)
2008{
859bff51
ID		 2009 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2010 u64 req_feat = from_msgr(con->msgr)->required_features;
2b3e0c90
ID		 2011 u64 server_feat = ceph_sanitize_features(
2012 le64_to_cpu(con->in_reply.features));
0da5d703 2013 int ret;
04a419f9 2014
eed0ef2c
SW		 2015 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2016
31b8006e 2017 switch (con->in_reply.tag) {
04a419f9
SW		 2018 case CEPH_MSGR_TAG_FEATURES:
2019 pr_err("%s%lld %s feature set mismatch,"
2020 " my %llx < server's %llx, missing %llx\n",
2021 ENTITY_NAME(con->peer_name),
3d14c5d2 2022 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2023 sup_feat, server_feat, server_feat & ~sup_feat);
2024 con->error_msg = "missing required protocol features";
0fa6ebc6 2025 reset_connection(con);
04a419f9
SW		 2026 return -1;
2027
31b8006e 2028 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2029 pr_err("%s%lld %s protocol version mismatch,"
2030 " my %d != server's %d\n",
2031 ENTITY_NAME(con->peer_name),
3d14c5d2 2032 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2033 le32_to_cpu(con->out_connect.protocol_version),
2034 le32_to_cpu(con->in_reply.protocol_version));
2035 con->error_msg = "protocol version mismatch";
0fa6ebc6 2036 reset_connection(con);
31b8006e
SW		 2037 return -1;
2038
4e7a5dcd
SW		 2039 case CEPH_MSGR_TAG_BADAUTHORIZER:
2040 con->auth_retry++;
2041 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2042 con->auth_retry);
2043 if (con->auth_retry == 2) {
2044 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2045 return -1;
2046 }
6d4221b5 2047 con_out_kvec_reset(con);
e825a66d 2048 ret = prepare_write_connect(con);
0da5d703
SW		 2049 if (ret < 0)
2050 return ret;
63733a0f 2051 prepare_read_connect(con);
4e7a5dcd 2052 break;
31b8006e
SW		 2053
2054 case CEPH_MSGR_TAG_RESETSESSION:
2055 /*
2056 * If we connected with a large connect_seq but the peer
2057 * has no record of a session with us (no connection, or
2058 * connect_seq == 0), they will send RESETSESION to indicate
2059 * that they must have reset their session, and may have
2060 * dropped messages.
2061 */
2062 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2063 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2064 pr_err("%s%lld %s connection reset\n",
2065 ENTITY_NAME(con->peer_name),
3d14c5d2 2066 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2067 reset_connection(con);
6d4221b5 2068 con_out_kvec_reset(con);
5a0f8fdd
AE		 2069 ret = prepare_write_connect(con);
2070 if (ret < 0)
2071 return ret;
31b8006e
SW		 2072 prepare_read_connect(con);
2073
2074 /* Tell ceph about it. */
ec302645 2075 mutex_unlock(&con->mutex);
31b8006e
SW		 2076 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2077 if (con->ops->peer_reset)
2078 con->ops->peer_reset(con);
ec302645 2079 mutex_lock(&con->mutex);
8dacc7da 2080 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2081 return -EAGAIN;
31b8006e
SW		 2082 break;
2083
2084 case CEPH_MSGR_TAG_RETRY_SESSION:
2085 /*
2086 * If we sent a smaller connect_seq than the peer has, try
2087 * again with a larger value.
2088 */
5bdca4e0 2089 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2090 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2091 le32_to_cpu(con->in_reply.connect_seq));
2092 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2093 con_out_kvec_reset(con);
5a0f8fdd
AE		 2094 ret = prepare_write_connect(con);
2095 if (ret < 0)
2096 return ret;
31b8006e
SW		 2097 prepare_read_connect(con);
2098 break;
2099
2100 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2101 /*
2102 * If we sent a smaller global_seq than the peer has, try
2103 * again with a larger value.
2104 */
eed0ef2c 2105 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2106 con->peer_global_seq,
5bdca4e0 2107 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2108 get_global_seq(con->msgr,
5bdca4e0 2109 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2110 con_out_kvec_reset(con);
5a0f8fdd
AE		 2111 ret = prepare_write_connect(con);
2112 if (ret < 0)
2113 return ret;
31b8006e
SW		 2114 prepare_read_connect(con);
2115 break;
2116
3a23083b 2117 case CEPH_MSGR_TAG_SEQ:
31b8006e 2118 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2119 if (req_feat & ~server_feat) {
2120 pr_err("%s%lld %s protocol feature mismatch,"
2121 " my required %llx > server's %llx, need %llx\n",
2122 ENTITY_NAME(con->peer_name),
3d14c5d2 2123 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2124 req_feat, server_feat, req_feat & ~server_feat);
2125 con->error_msg = "missing required protocol features";
0fa6ebc6 2126 reset_connection(con);
04a419f9
SW		 2127 return -1;
2128 }
8dacc7da 2129
122070a2 2130 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2131 con->state = CON_STATE_OPEN;
20e55c4c 2132 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2133 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2134 con->connect_seq++;
aba558e2 2135 con->peer_features = server_feat;
31b8006e
SW		 2136 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2137 con->peer_global_seq,
2138 le32_to_cpu(con->in_reply.connect_seq),
2139 con->connect_seq);
2140 WARN_ON(con->connect_seq !=
2141 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2142
2143 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2144 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2145
85effe18 2146 con->delay = 0; /* reset backoff memory */
92ac41d0 2147
3a23083b
SW		 2148 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2149 prepare_write_seq(con);
2150 prepare_read_seq(con);
2151 } else {
2152 prepare_read_tag(con);
2153 }
31b8006e
SW		 2154 break;
2155
2156 case CEPH_MSGR_TAG_WAIT:
2157 /*
2158 * If there is a connection race (we are opening
2159 * connections to each other), one of us may just have
2160 * to WAIT. This shouldn't happen if we are the
2161 * client.
2162 */
04177882
SW		 2163 con->error_msg = "protocol error, got WAIT as client";
2164 return -1;
31b8006e
SW		 2165
2166 default:
31b8006e
SW		 2167 con->error_msg = "protocol error, garbage tag during connect";
2168 return -1;
2169 }
2170 return 0;
2171}
2172
2173
2174/*
2175 * read (part of) an ack
2176 */
2177static int read_partial_ack(struct ceph_connection *con)
2178{
fd51653f
AE		 2179 int size = sizeof (con->in_temp_ack);
2180 int end = size;
31b8006e 2181
fd51653f 2182 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2183}
2184
31b8006e
SW		 2185/*
2186 * We can finally discard anything that's been acked.
2187 */
2188static void process_ack(struct ceph_connection *con)
2189{
2190 struct ceph_msg *m;
2191 u64 ack = le64_to_cpu(con->in_temp_ack);
2192 u64 seq;
2193
31b8006e
SW		 2194 while (!list_empty(&con->out_sent)) {
2195 m = list_first_entry(&con->out_sent, struct ceph_msg,
2196 list_head);
2197 seq = le64_to_cpu(m->hdr.seq);
2198 if (seq > ack)
2199 break;
2200 dout("got ack for seq %llu type %d at %p\n", seq,
2201 le16_to_cpu(m->hdr.type), m);
4cf9d544 2202 m->ack_stamp = jiffies;
31b8006e
SW		 2203 ceph_msg_remove(m);
2204 }
31b8006e
SW		 2205 prepare_read_tag(con);
2206}
2207
2208
2450418c 2209static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2210 struct kvec *section,
2211 unsigned int sec_len, u32 *crc)
2450418c 2212{
68b4476b 2213 int ret, left;
2450418c
YS		 2214
2215 BUG_ON(!section);
2216
2217 while (section->iov_len < sec_len) {
2218 BUG_ON(section->iov_base == NULL);
2219 left = sec_len - section->iov_len;
2220 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2221 section->iov_len, left);
2222 if (ret <= 0)
2223 return ret;
2224 section->iov_len += ret;
2450418c 2225 }
fe3ad593
AE		 2226 if (section->iov_len == sec_len)
2227 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2228
2450418c
YS		 2229 return 1;
2230}
31b8006e 2231
34d2d200
AE		 2232static int read_partial_msg_data(struct ceph_connection *con)
2233{
2234 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2235 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 2236 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
686be208
AE		 2237 struct page *page;
2238 size_t page_offset;
2239 size_t length;
f5db90bc 2240 u32 crc = 0;
34d2d200
AE		 2241 int ret;
2242
2243 BUG_ON(!msg);
5240d9f9 2244 if (list_empty(&msg->data))
4c59b4a2 2245 return -EIO;
34d2d200 2246
f5db90bc
AE		 2247 if (do_datacrc)
2248 crc = con->in_data_crc;
643c68a4 2249 while (cursor->resid) {
343128ce 2250 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
686be208 2251 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2252 if (ret <= 0) {
2253 if (do_datacrc)
2254 con->in_data_crc = crc;
2255
686be208 2256 return ret;
f5db90bc 2257 }
686be208
AE		 2258
2259 if (do_datacrc)
f5db90bc 2260 crc = ceph_crc32c_page(crc, page, page_offset, ret);
343128ce 2261 (void) ceph_msg_data_advance(cursor, (size_t)ret);
34d2d200 2262 }
f5db90bc
AE		 2263 if (do_datacrc)
2264 con->in_data_crc = crc;
34d2d200
AE		 2265
2266 return 1; /* must return > 0 to indicate success */
2267}
2268
31b8006e
SW		 2269/*
2270 * read (part of) a message.
2271 */
686be208
AE		 2272static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2273
31b8006e
SW		 2274static int read_partial_message(struct ceph_connection *con)
2275{
2276 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2277 int size;
2278 int end;
31b8006e 2279 int ret;
95c96174 2280 unsigned int front_len, middle_len, data_len;
859bff51 2281 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
33d07337 2282 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
ae18756b 2283 u64 seq;
fe3ad593 2284 u32 crc;
31b8006e
SW		 2285
2286 dout("read_partial_message con %p msg %p\n", con, m);
2287
2288 /* header */
fd51653f
AE		 2289 size = sizeof (con->in_hdr);
2290 end = size;
2291 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2292 if (ret <= 0)
2293 return ret;
fe3ad593
AE		 2294
2295 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2296 if (cpu_to_le32(crc) != con->in_hdr.crc) {
67c64eb7 2297 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
fe3ad593
AE		 2298 crc, con->in_hdr.crc);
2299 return -EBADMSG;
2300 }
2301
31b8006e
SW		 2302 front_len = le32_to_cpu(con->in_hdr.front_len);
2303 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2304 return -EIO;
2305 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2306 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2307 return -EIO;
2308 data_len = le32_to_cpu(con->in_hdr.data_len);
2309 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2310 return -EIO;
2311
ae18756b
SW		 2312 /* verify seq# */
2313 seq = le64_to_cpu(con->in_hdr.seq);
2314 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2315 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2316 ENTITY_NAME(con->peer_name),
3d14c5d2 2317 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2318 seq, con->in_seq + 1);
2319 con->in_base_pos = -front_len - middle_len - data_len -
2320 sizeof(m->footer);
2321 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2322 return 0;
2323 } else if ((s64)seq - (s64)con->in_seq > 1) {
2324 pr_err("read_partial_message bad seq %lld expected %lld\n",
2325 seq, con->in_seq + 1);
2326 con->error_msg = "bad message sequence # for incoming message";
67c64eb7 2327 return -EBADE;
ae18756b
SW		 2328 }
2329
31b8006e
SW		 2330 /* allocate message? */
2331 if (!con->in_msg) {
4740a623
SW		 2332 int skip = 0;
2333
31b8006e 2334 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2335 front_len, data_len);
4740a623
SW		 2336 ret = ceph_con_in_msg_alloc(con, &skip);
2337 if (ret < 0)
2338 return ret;
f759ebb9
AE		 2339
2340 BUG_ON(!con->in_msg ^ skip);
2450418c 2341 if (skip) {
31b8006e 2342 /* skip this message */
a79832f2 2343 dout("alloc_msg said skip message\n");
31b8006e
SW		 2344 con->in_base_pos = -front_len - middle_len - data_len -
2345 sizeof(m->footer);
2346 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2347 con->in_seq++;
31b8006e
SW		 2348 return 0;
2349 }
38941f80 2350
4740a623 2351 BUG_ON(!con->in_msg);
38941f80 2352 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2353 m = con->in_msg;
2354 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2355 if (m->middle)
2356 m->middle->vec.iov_len = 0;
9d7f0f13 2357
78625051 2358 /* prepare for data payload, if any */
a4107026 2359
78625051 2360 if (data_len)
98fa5dd8 2361 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2362 }
2363
2364 /* front */
2450418c
YS		 2365 ret = read_partial_message_section(con, &m->front, front_len,
2366 &con->in_front_crc);
2367 if (ret <= 0)
2368 return ret;
31b8006e
SW		 2369
2370 /* middle */
2450418c 2371 if (m->middle) {
213c99ee
SW		 2372 ret = read_partial_message_section(con, &m->middle->vec,
2373 middle_len,
2450418c 2374 &con->in_middle_crc);
31b8006e
SW		 2375 if (ret <= 0)
2376 return ret;
31b8006e
SW		 2377 }
2378
2379 /* (page) data */
34d2d200
AE		 2380 if (data_len) {
2381 ret = read_partial_msg_data(con);
2382 if (ret <= 0)
2383 return ret;
31b8006e
SW		 2384 }
2385
31b8006e 2386 /* footer */
33d07337
YZ		 2387 if (need_sign)
2388 size = sizeof(m->footer);
2389 else
2390 size = sizeof(m->old_footer);
2391
fd51653f
AE		 2392 end += size;
2393 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2394 if (ret <= 0)
2395 return ret;
2396
33d07337
YZ		 2397 if (!need_sign) {
2398 m->footer.flags = m->old_footer.flags;
2399 m->footer.sig = 0;
2400 }
2401
31b8006e
SW		 2402 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2403 m, front_len, m->footer.front_crc, middle_len,
2404 m->footer.middle_crc, data_len, m->footer.data_crc);
2405
2406 /* crc ok? */
2407 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2408 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2409 m, con->in_front_crc, m->footer.front_crc);
2410 return -EBADMSG;
2411 }
2412 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2413 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2414 m, con->in_middle_crc, m->footer.middle_crc);
2415 return -EBADMSG;
2416 }
bca064d2 2417 if (do_datacrc &&
31b8006e
SW		 2418 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2419 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2420 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2421 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2422 return -EBADMSG;
2423 }
2424
33d07337 2425 if (need_sign && con->ops->check_message_signature &&
79dbd1ba 2426 con->ops->check_message_signature(m)) {
33d07337
YZ		 2427 pr_err("read_partial_message %p signature check failed\n", m);
2428 return -EBADMSG;
2429 }
2430
31b8006e
SW		 2431 return 1; /* done! */
2432}
2433
2434/*
2435 * Process message. This happens in the worker thread. The callback should
2436 * be careful not to do anything that waits on other incoming messages or it
2437 * may deadlock.
2438 */
2439static void process_message(struct ceph_connection *con)
2440{
5e095e8b 2441 struct ceph_msg *msg;
31b8006e 2442
38941f80
AE		 2443 BUG_ON(con->in_msg->con != con);
2444 con->in_msg->con = NULL;
5e095e8b 2445 msg = con->in_msg;
31b8006e 2446 con->in_msg = NULL;
36eb71aa 2447 con->ops->put(con);
31b8006e
SW		 2448
2449 /* if first message, set peer_name */
2450 if (con->peer_name.type == 0)
dbad185d 2451 con->peer_name = msg->hdr.src;
31b8006e 2452
31b8006e 2453 con->in_seq++;
ec302645 2454 mutex_unlock(&con->mutex);
31b8006e
SW		 2455
2456 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2457 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2458 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2459 le16_to_cpu(msg->hdr.type),
2460 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2461 le32_to_cpu(msg->hdr.front_len),
2462 le32_to_cpu(msg->hdr.data_len),
2463 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2464 con->ops->dispatch(con, msg);
ec302645
SW		 2465
2466 mutex_lock(&con->mutex);
31b8006e
SW		 2467}
2468
8b9558aa
YZ		 2469static int read_keepalive_ack(struct ceph_connection *con)
2470{
2471 struct ceph_timespec ceph_ts;
2472 size_t size = sizeof(ceph_ts);
2473 int ret = read_partial(con, size, size, &ceph_ts);
2474 if (ret <= 0)
2475 return ret;
2476 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2477 prepare_read_tag(con);
2478 return 1;
2479}
31b8006e
SW		 2480
2481/*
2482 * Write something to the socket. Called in a worker thread when the
2483 * socket appears to be writeable and we have something ready to send.
2484 */
2485static int try_write(struct ceph_connection *con)
2486{
31b8006e
SW		 2487 int ret = 1;
2488
d59315ca 2489 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2490
31b8006e
SW		 2491more:
2492 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2493
2494 /* open the socket first? */
8dacc7da
SW		 2495 if (con->state == CON_STATE_PREOPEN) {
2496 BUG_ON(con->sock);
2497 con->state = CON_STATE_CONNECTING;
a5988c49 2498
e2200423 2499 con_out_kvec_reset(con);
e825a66d 2500 prepare_write_banner(con);
eed0ef2c 2501 prepare_read_banner(con);
31b8006e 2502
cf3e5c40 2503 BUG_ON(con->in_msg);
31b8006e
SW		 2504 con->in_tag = CEPH_MSGR_TAG_READY;
2505 dout("try_write initiating connect on %p new state %lu\n",
2506 con, con->state);
41617d0c
AE		 2507 ret = ceph_tcp_connect(con);
2508 if (ret < 0) {
31b8006e 2509 con->error_msg = "connect error";
31b8006e
SW		 2510 goto out;
2511 }
2512 }
2513
2514more_kvec:
2515 /* kvec data queued? */
2516 if (con->out_skip) {
2517 ret = write_partial_skip(con);
2518 if (ret <= 0)
42961d23 2519 goto out;
31b8006e
SW		 2520 }
2521 if (con->out_kvec_left) {
2522 ret = write_partial_kvec(con);
2523 if (ret <= 0)
42961d23 2524 goto out;
31b8006e
SW		 2525 }
2526
2527 /* msg pages? */
2528 if (con->out_msg) {
c86a2930
SW		 2529 if (con->out_msg_done) {
2530 ceph_msg_put(con->out_msg);
2531 con->out_msg = NULL; /* we're done with this one */
2532 goto do_next;
2533 }
2534
34d2d200 2535 ret = write_partial_message_data(con);
31b8006e
SW		 2536 if (ret == 1)
2537 goto more_kvec; /* we need to send the footer, too! */
2538 if (ret == 0)
42961d23 2539 goto out;
31b8006e 2540 if (ret < 0) {
34d2d200 2541 dout("try_write write_partial_message_data err %d\n",
31b8006e 2542 ret);
42961d23 2543 goto out;
31b8006e
SW		 2544 }
2545 }
2546
c86a2930 2547do_next:
8dacc7da 2548 if (con->state == CON_STATE_OPEN) {
8b9558aa
YZ		 2549 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2550 prepare_write_keepalive(con);
2551 goto more;
2552 }
31b8006e
SW		 2553 /* is anything else pending? */
2554 if (!list_empty(&con->out_queue)) {
2555 prepare_write_message(con);
2556 goto more;
2557 }
2558 if (con->in_seq > con->in_seq_acked) {
2559 prepare_write_ack(con);
2560 goto more;
2561 }
31b8006e
SW		 2562 }
2563
2564 /* Nothing to do! */
c9ffc77a 2565 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2566 dout("try_write nothing else to write.\n");
31b8006e
SW		 2567 ret = 0;
2568out:
42961d23 2569 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2570 return ret;
2571}
2572
2573
2574
2575/*
2576 * Read what we can from the socket.
2577 */
2578static int try_read(struct ceph_connection *con)
2579{
31b8006e
SW		 2580 int ret = -1;
2581
8dacc7da
SW		 2582more:
2583 dout("try_read start on %p state %lu\n", con, con->state);
2584 if (con->state != CON_STATE_CONNECTING &&
2585 con->state != CON_STATE_NEGOTIATING &&
2586 con->state != CON_STATE_OPEN)
31b8006e
SW		 2587 return 0;
2588
8dacc7da 2589 BUG_ON(!con->sock);
ec302645 2590
31b8006e
SW		 2591 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2592 con->in_base_pos);
0da5d703 2593
8dacc7da 2594 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2595 dout("try_read connecting\n");
2596 ret = read_partial_banner(con);
2597 if (ret <= 0)
ab166d5a 2598 goto out;
7593af92
AE		 2599 ret = process_banner(con);
2600 if (ret < 0)
2601 goto out;
2602
8dacc7da 2603 con->state = CON_STATE_NEGOTIATING;
7593af92 2604
6d4221b5
JS		 2605 /*
2606 * Received banner is good, exchange connection info.
2607 * Do not reset out_kvec, as sending our banner raced
2608 * with receiving peer banner after connect completed.
2609 */
7593af92
AE		 2610 ret = prepare_write_connect(con);
2611 if (ret < 0)
2612 goto out;
2613 prepare_read_connect(con);
2614
2615 /* Send connection info before awaiting response */
0da5d703
SW		 2616 goto out;
2617 }
2618
8dacc7da 2619 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2620 dout("try_read negotiating\n");
31b8006e
SW		 2621 ret = read_partial_connect(con);
2622 if (ret <= 0)
31b8006e 2623 goto out;
98bdb0aa
SW		 2624 ret = process_connect(con);
2625 if (ret < 0)
2626 goto out;
31b8006e
SW		 2627 goto more;
2628 }
2629
122070a2 2630 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2631
31b8006e
SW		 2632 if (con->in_base_pos < 0) {
2633 /*
2634 * skipping + discarding content.
2635 *
2636 * FIXME: there must be a better way to do this!
2637 */
84495f49
AE		 2638 static char buf[SKIP_BUF_SIZE];
2639 int skip = min((int) sizeof (buf), -con->in_base_pos);
2640
31b8006e
SW		 2641 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2642 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2643 if (ret <= 0)
98bdb0aa 2644 goto out;
31b8006e
SW		 2645 con->in_base_pos += ret;
2646 if (con->in_base_pos)
2647 goto more;
2648 }
2649 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2650 /*
2651 * what's next?
2652 */
2653 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2654 if (ret <= 0)
98bdb0aa 2655 goto out;
31b8006e
SW		 2656 dout("try_read got tag %d\n", (int)con->in_tag);
2657 switch (con->in_tag) {
2658 case CEPH_MSGR_TAG_MSG:
2659 prepare_read_message(con);
2660 break;
2661 case CEPH_MSGR_TAG_ACK:
2662 prepare_read_ack(con);
2663 break;
8b9558aa
YZ		 2664 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2665 prepare_read_keepalive_ack(con);
2666 break;
31b8006e 2667 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2668 con_close_socket(con);
2669 con->state = CON_STATE_CLOSED;
98bdb0aa 2670 goto out;
31b8006e
SW		 2671 default:
2672 goto bad_tag;
2673 }
2674 }
2675 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2676 ret = read_partial_message(con);
2677 if (ret <= 0) {
2678 switch (ret) {
2679 case -EBADMSG:
a51983e4 2680 con->error_msg = "bad crc/signature";
67c64eb7
ID		 2681 /* fall through */
2682 case -EBADE:
31b8006e 2683 ret = -EIO;
98bdb0aa 2684 break;
31b8006e
SW		 2685 case -EIO:
2686 con->error_msg = "io error";
98bdb0aa 2687 break;
31b8006e 2688 }
98bdb0aa 2689 goto out;
31b8006e
SW		 2690 }
2691 if (con->in_tag == CEPH_MSGR_TAG_READY)
2692 goto more;
2693 process_message(con);
7b862e07
SW		 2694 if (con->state == CON_STATE_OPEN)
2695 prepare_read_tag(con);
31b8006e
SW		 2696 goto more;
2697 }
3a23083b
SW		 2698 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2699 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2700 /*
2701 * the final handshake seq exchange is semantically
2702 * equivalent to an ACK
2703 */
31b8006e
SW		 2704 ret = read_partial_ack(con);
2705 if (ret <= 0)
98bdb0aa 2706 goto out;
31b8006e
SW		 2707 process_ack(con);
2708 goto more;
2709 }
8b9558aa
YZ		 2710 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2711 ret = read_keepalive_ack(con);
2712 if (ret <= 0)
2713 goto out;
2714 goto more;
2715 }
31b8006e 2716
31b8006e 2717out:
98bdb0aa 2718 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2719 return ret;
2720
2721bad_tag:
2722 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2723 con->error_msg = "protocol error, garbage tag";
2724 ret = -1;
2725 goto out;
2726}
2727
2728
2729/*
802c6d96
AE		 2730 * Atomically queue work on a connection after the specified delay.
2731 * Bump @con reference to avoid races with connection teardown.
2732 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2733 */
802c6d96 2734static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2735{
31b8006e 2736 if (!con->ops->get(con)) {
802c6d96 2737 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2738 return -ENOENT;
31b8006e
SW		 2739 }
2740
802c6d96
AE		 2741 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2742 dout("%s %p - already queued\n", __func__, con);
31b8006e 2743 con->ops->put(con);
802c6d96 2744 return -EBUSY;
31b8006e 2745 }
802c6d96
AE		 2746
2747 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2748 return 0;
2749}
2750
2751static void queue_con(struct ceph_connection *con)
2752{
2753 (void) queue_con_delay(con, 0);
31b8006e
SW		 2754}
2755
37ab77ac
ID		 2756static void cancel_con(struct ceph_connection *con)
2757{
2758 if (cancel_delayed_work(&con->work)) {
2759 dout("%s %p\n", __func__, con);
2760 con->ops->put(con);
2761 }
2762}
2763
7bb21d68
AE		 2764static bool con_sock_closed(struct ceph_connection *con)
2765{
c9ffc77a 2766 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2767 return false;
2768
2769#define CASE(x) \
2770 case CON_STATE_ ## x: \
2771 con->error_msg = "socket closed (con state " #x ")"; \
2772 break;
2773
2774 switch (con->state) {
2775 CASE(CLOSED);
2776 CASE(PREOPEN);
2777 CASE(CONNECTING);
2778 CASE(NEGOTIATING);
2779 CASE(OPEN);
2780 CASE(STANDBY);
2781 default:
b9a67899 2782 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2783 __func__, con, con->state);
2784 con->error_msg = "unrecognized con state";
2785 BUG();
2786 break;
2787 }
2788#undef CASE
2789
2790 return true;
2791}
2792
f20a39fd
AE		 2793static bool con_backoff(struct ceph_connection *con)
2794{
2795 int ret;
2796
2797 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2798 return false;
2799
2800 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2801 if (ret) {
2802 dout("%s: con %p FAILED to back off %lu\n", __func__,
2803 con, con->delay);
2804 BUG_ON(ret == -ENOENT);
2805 con_flag_set(con, CON_FLAG_BACKOFF);
2806 }
2807
2808 return true;
2809}
2810
93209264
AE		 2811/* Finish fault handling; con->mutex must *not* be held here */
2812
2813static void con_fault_finish(struct ceph_connection *con)
2814{
2815 /*
2816 * in case we faulted due to authentication, invalidate our
2817 * current tickets so that we can get new ones.
2818 */
2819 if (con->auth_retry && con->ops->invalidate_authorizer) {
2820 dout("calling invalidate_authorizer()\n");
2821 con->ops->invalidate_authorizer(con);
2822 }
2823
2824 if (con->ops->fault)
2825 con->ops->fault(con);
2826}
2827
31b8006e
SW		 2828/*
2829 * Do some work on a connection. Drop a connection ref when we're done.
2830 */
68931622 2831static void ceph_con_workfn(struct work_struct *work)
31b8006e
SW		 2832{
2833 struct ceph_connection *con = container_of(work, struct ceph_connection,
2834 work.work);
49659416 2835 bool fault;
31b8006e 2836
9dd4658d 2837 mutex_lock(&con->mutex);
49659416
AE		 2838 while (true) {
2839 int ret;
31b8006e 2840
49659416
AE		 2841 if ((fault = con_sock_closed(con))) {
2842 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2843 break;
2844 }
2845 if (con_backoff(con)) {
2846 dout("%s: con %p BACKOFF\n", __func__, con);
2847 break;
2848 }
2849 if (con->state == CON_STATE_STANDBY) {
2850 dout("%s: con %p STANDBY\n", __func__, con);
2851 break;
2852 }
2853 if (con->state == CON_STATE_CLOSED) {
2854 dout("%s: con %p CLOSED\n", __func__, con);
2855 BUG_ON(con->sock);
2856 break;
2857 }
2858 if (con->state == CON_STATE_PREOPEN) {
2859 dout("%s: con %p PREOPEN\n", __func__, con);
2860 BUG_ON(con->sock);
2861 }
0da5d703 2862
49659416
AE		 2863 ret = try_read(con);
2864 if (ret < 0) {
2865 if (ret == -EAGAIN)
2866 continue;
67c64eb7
ID		 2867 if (!con->error_msg)
2868 con->error_msg = "socket error on read";
49659416
AE		 2869 fault = true;
2870 break;
2871 }
2872
2873 ret = try_write(con);
2874 if (ret < 0) {
2875 if (ret == -EAGAIN)
2876 continue;
67c64eb7
ID		 2877 if (!con->error_msg)
2878 con->error_msg = "socket error on write";
49659416
AE		 2879 fault = true;
2880 }
2881
2882 break; /* If we make it to here, we're done */
3a140a0d 2883 }
b6e7b6a1
AE		 2884 if (fault)
2885 con_fault(con);
9dd4658d 2886 mutex_unlock(&con->mutex);
0da5d703 2887
b6e7b6a1
AE		 2888 if (fault)
2889 con_fault_finish(con);
2890
2891 con->ops->put(con);
31b8006e
SW		 2892}
2893
31b8006e
SW		 2894/*
2895 * Generic error/fault handler. A retry mechanism is used with
2896 * exponential backoff
2897 */
93209264 2898static void con_fault(struct ceph_connection *con)
31b8006e 2899{
31b8006e 2900 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2901 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2902
67c64eb7
ID		 2903 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2904 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2905 con->error_msg = NULL;
2906
122070a2 2907 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2908 con->state != CON_STATE_NEGOTIATING &&
2909 con->state != CON_STATE_OPEN);
ec302645 2910
31b8006e 2911 con_close_socket(con);
5e095e8b 2912
c9ffc77a 2913 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2914 dout("fault on LOSSYTX channel, marking CLOSED\n");
2915 con->state = CON_STATE_CLOSED;
93209264 2916 return;
3b5ede07
SW		 2917 }
2918
5e095e8b 2919 if (con->in_msg) {
38941f80
AE		 2920 BUG_ON(con->in_msg->con != con);
2921 con->in_msg->con = NULL;
5e095e8b
SW		 2922 ceph_msg_put(con->in_msg);
2923 con->in_msg = NULL;
36eb71aa 2924 con->ops->put(con);
5e095e8b 2925 }
31b8006e 2926
e80a52d1
SW		 2927 /* Requeue anything that hasn't been acked */
2928 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2929
e76661d0
SW		 2930 /* If there are no messages queued or keepalive pending, place
2931 * the connection in a STANDBY state */
2932 if (list_empty(&con->out_queue) &&
c9ffc77a 2933 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2934 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2935 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2936 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2937 } else {
2938 /* retry after a delay. */
8dacc7da 2939 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2940 if (con->delay == 0)
2941 con->delay = BASE_DELAY_INTERVAL;
2942 else if (con->delay < MAX_DELAY_INTERVAL)
2943 con->delay *= 2;
c9ffc77a 2944 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2945 queue_con(con);
31b8006e 2946 }
31b8006e
SW		 2947}
2948
2949
2950
2951/*
15d9882c 2952 * initialize a new messenger instance
31b8006e 2953 */
15d9882c 2954void ceph_messenger_init(struct ceph_messenger *msgr,
859bff51 2955 struct ceph_entity_addr *myaddr)
31b8006e 2956{
31b8006e
SW		 2957 spin_lock_init(&msgr->global_seq_lock);
2958
31b8006e
SW		 2959 if (myaddr)
2960 msgr->inst.addr = *myaddr;
2961
2962 /* select a random nonce */
ac8839d7 2963 msgr->inst.addr.type = 0;
103e2d3a 2964 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2965 encode_my_addr(msgr);
31b8006e 2966
a2a32584 2967 atomic_set(&msgr->stopping, 0);
757856d2 2968 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
31b8006e 2969
15d9882c 2970 dout("%s %p\n", __func__, msgr);
31b8006e 2971}
15d9882c 2972EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2973
757856d2
ID		 2974void ceph_messenger_fini(struct ceph_messenger *msgr)
2975{
2976 put_net(read_pnet(&msgr->net));
2977}
2978EXPORT_SYMBOL(ceph_messenger_fini);
2979
e00de341
SW		 2980static void clear_standby(struct ceph_connection *con)
2981{
2982 /* come back from STANDBY? */
8dacc7da 2983 if (con->state == CON_STATE_STANDBY) {
e00de341 2984 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2985 con->state = CON_STATE_PREOPEN;
e00de341 2986 con->connect_seq++;
c9ffc77a
AE		 2987 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2988 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2989 }
2990}
2991
31b8006e
SW		 2992/*
2993 * Queue up an outgoing message on the given connection.
2994 */
2995void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2996{
31b8006e 2997 /* set src+dst */
dbad185d 2998 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2999 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 3000 msg->needs_out_seq = true;
3001
ec302645 3002 mutex_lock(&con->mutex);
92ce034b 3003
8dacc7da 3004 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 3005 dout("con_send %p closed, dropping %p\n", con, msg);
3006 ceph_msg_put(msg);
3007 mutex_unlock(&con->mutex);
3008 return;
3009 }
3010
38941f80 3011 BUG_ON(msg->con != NULL);
36eb71aa 3012 msg->con = con->ops->get(con);
92ce034b
AE		 3013 BUG_ON(msg->con == NULL);
3014
31b8006e
SW		 3015 BUG_ON(!list_empty(&msg->list_head));
3016 list_add_tail(&msg->list_head, &con->out_queue);
3017 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3018 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3019 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3020 le32_to_cpu(msg->hdr.front_len),
3021 le32_to_cpu(msg->hdr.middle_len),
3022 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 3023
3024 clear_standby(con);
ec302645 3025 mutex_unlock(&con->mutex);
31b8006e
SW		 3026
3027 /* if there wasn't anything waiting to send before, queue
3028 * new work */
c9ffc77a 3029 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3030 queue_con(con);
3031}
3d14c5d2 3032EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3033
3034/*
3035 * Revoke a message that was previously queued for send
3036 */
6740a845 3037void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3038{
6740a845
AE		 3039 struct ceph_connection *con = msg->con;
3040
3041 if (!con)
3042 return; /* Message not in our possession */
3043
ec302645 3044 mutex_lock(&con->mutex);
31b8006e 3045 if (!list_empty(&msg->list_head)) {
38941f80 3046 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3047 list_del_init(&msg->list_head);
38941f80 3048 BUG_ON(msg->con == NULL);
36eb71aa 3049 msg->con->ops->put(msg->con);
38941f80 3050 msg->con = NULL;
31b8006e 3051 msg->hdr.seq = 0;
38941f80 3052
31b8006e 3053 ceph_msg_put(msg);
ed98adad
SW		 3054 }
3055 if (con->out_msg == msg) {
38941f80 3056 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 3057 con->out_msg = NULL;
31b8006e
SW		 3058 if (con->out_kvec_is_msg) {
3059 con->out_skip = con->out_kvec_bytes;
3060 con->out_kvec_is_msg = false;
3061 }
ed98adad 3062 msg->hdr.seq = 0;
92ce034b
AE		 3063
3064 ceph_msg_put(msg);
31b8006e 3065 }
ec302645 3066 mutex_unlock(&con->mutex);
31b8006e
SW		 3067}
3068
350b1c32 3069/*
0d59ab81 3070 * Revoke a message that we may be reading data into
350b1c32 3071 */
8921d114 3072void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3073{
8921d114
AE		 3074 struct ceph_connection *con;
3075
3076 BUG_ON(msg == NULL);
3077 if (!msg->con) {
3078 dout("%s msg %p null con\n", __func__, msg);
3079
3080 return; /* Message not in our possession */
3081 }
3082
3083 con = msg->con;
350b1c32 3084 mutex_lock(&con->mutex);
8921d114 3085 if (con->in_msg == msg) {
95c96174
ED		 3086 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3087 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3088 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3089
3090 /* skip rest of message */
8921d114
AE		 3091 dout("%s %p msg %p revoked\n", __func__, con, msg);
3092 con->in_base_pos = con->in_base_pos -
350b1c32 3093 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3094 front_len -
3095 middle_len -
3096 data_len -
350b1c32 3097 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3098 ceph_msg_put(con->in_msg);
3099 con->in_msg = NULL;
3100 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3101 con->in_seq++;
350b1c32 3102 } else {
8921d114
AE		 3103 dout("%s %p in_msg %p msg %p no-op\n",
3104 __func__, con, con->in_msg, msg);
350b1c32
SW		 3105 }
3106 mutex_unlock(&con->mutex);
3107}
3108
31b8006e
SW		 3109/*
3110 * Queue a keepalive byte to ensure the tcp connection is alive.
3111 */
3112void ceph_con_keepalive(struct ceph_connection *con)
3113{
e00de341 3114 dout("con_keepalive %p\n", con);
00650931 3115 mutex_lock(&con->mutex);
e00de341 3116 clear_standby(con);
00650931 3117 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3118 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3119 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3120 queue_con(con);
3121}
3d14c5d2 3122EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3123
8b9558aa
YZ		 3124bool ceph_con_keepalive_expired(struct ceph_connection *con,
3125 unsigned long interval)
3126{
3127 if (interval > 0 &&
3128 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3129 struct timespec now = CURRENT_TIME;
3130 struct timespec ts;
3131 jiffies_to_timespec(interval, &ts);
3132 ts = timespec_add(con->last_keepalive_ack, ts);
3133 return timespec_compare(&now, &ts) >= 0;
3134 }
3135 return false;
3136}
3137
6644ed7b 3138static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3139{
6644ed7b
AE		 3140 struct ceph_msg_data *data;
3141
3142 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3143 return NULL;
3144
81b36be4 3145 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
6644ed7b
AE		 3146 if (data)
3147 data->type = type;
5240d9f9 3148 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3149
3150 return data;
3151}
3152
3153static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3154{
3155 if (!data)
3156 return;
3157
5240d9f9 3158 WARN_ON(!list_empty(&data->links));
e4339d28 3159 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3160 ceph_pagelist_release(data->pagelist);
81b36be4 3161 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3162}
3163
90af3602 3164void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3165 size_t length, size_t alignment)
02afca6c 3166{
6644ed7b
AE		 3167 struct ceph_msg_data *data;
3168
07aa1558
AE		 3169 BUG_ON(!pages);
3170 BUG_ON(!length);
6644ed7b
AE		 3171
3172 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3173 BUG_ON(!data);
3174 data->pages = pages;
3175 data->length = length;
3176 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3177
5240d9f9
AE		 3178 list_add_tail(&data->links, &msg->data);
3179 msg->data_length += length;
02afca6c 3180}
90af3602 3181EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3182
90af3602 3183void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3184 struct ceph_pagelist *pagelist)
3185{
6644ed7b
AE		 3186 struct ceph_msg_data *data;
3187
07aa1558
AE		 3188 BUG_ON(!pagelist);
3189 BUG_ON(!pagelist->length);
27fa8385 3190
6644ed7b
AE		 3191 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3192 BUG_ON(!data);
3193 data->pagelist = pagelist;
3194
5240d9f9
AE		 3195 list_add_tail(&data->links, &msg->data);
3196 msg->data_length += pagelist->length;
27fa8385 3197}
90af3602 3198EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3199
ea96571f 3200#ifdef CONFIG_BLOCK
90af3602 3201void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3202 size_t length)
27fa8385 3203{
6644ed7b
AE		 3204 struct ceph_msg_data *data;
3205
07aa1558 3206 BUG_ON(!bio);
27fa8385 3207
6644ed7b
AE		 3208 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3209 BUG_ON(!data);
3210 data->bio = bio;
c851c495 3211 data->bio_length = length;
6644ed7b 3212
5240d9f9
AE		 3213 list_add_tail(&data->links, &msg->data);
3214 msg->data_length += length;
27fa8385 3215}
90af3602 3216EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3217#endif /* CONFIG_BLOCK */
27fa8385 3218
31b8006e
SW		 3219/*
3220 * construct a new message with given type, size
3221 * the new msg has a ref count of 1.
3222 */
b61c2763
SW		 3223struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3224 bool can_fail)
31b8006e
SW		 3225{
3226 struct ceph_msg *m;
3227
e3d5d638 3228 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3229 if (m == NULL)
3230 goto out;
31b8006e
SW		 3231
3232 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3233 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3234 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3235
9516e45b
AE		 3236 INIT_LIST_HEAD(&m->list_head);
3237 kref_init(&m->kref);
5240d9f9 3238 INIT_LIST_HEAD(&m->data);
ca20892d 3239
31b8006e
SW		 3240 /* front */
3241 if (front_len) {
eeb0bed5 3242 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3243 if (m->front.iov_base == NULL) {
b61c2763 3244 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3245 front_len);
3246 goto out2;
3247 }
3248 } else {
3249 m->front.iov_base = NULL;
3250 }
f2be82b0 3251 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3252
bb257664 3253 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3254 return m;
3255
3256out2:
3257 ceph_msg_put(m);
3258out:
b61c2763
SW		 3259 if (!can_fail) {
3260 pr_err("msg_new can't create type %d front %d\n", type,
3261 front_len);
f0ed1b7c 3262 WARN_ON(1);
b61c2763
SW		 3263 } else {
3264 dout("msg_new can't create type %d front %d\n", type,
3265 front_len);
3266 }
a79832f2 3267 return NULL;
31b8006e 3268}
3d14c5d2 3269EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3270
31b8006e
SW		 3271/*
3272 * Allocate "middle" portion of a message, if it is needed and wasn't
3273 * allocated by alloc_msg. This allows us to read a small fixed-size
3274 * per-type header in the front and then gracefully fail (i.e.,
3275 * propagate the error to the caller based on info in the front) when
3276 * the middle is too large.
3277 */
2450418c 3278static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3279{
3280 int type = le16_to_cpu(msg->hdr.type);
3281 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3282
3283 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3284 ceph_msg_type_name(type), middle_len);
3285 BUG_ON(!middle_len);
3286 BUG_ON(msg->middle);
3287
b6c1d5b8 3288 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3289 if (!msg->middle)
3290 return -ENOMEM;
3291 return 0;
3292}
3293
2450418c 3294/*
1c20f2d2
AE		 3295 * Allocate a message for receiving an incoming message on a
3296 * connection, and save the result in con->in_msg. Uses the
3297 * connection's private alloc_msg op if available.
3298 *
4740a623
SW		 3299 * Returns 0 on success, or a negative error code.
3300 *
3301 * On success, if we set *skip = 1:
3302 * - the next message should be skipped and ignored.
3303 * - con->in_msg == NULL
3304 * or if we set *skip = 0:
3305 * - con->in_msg is non-null.
3306 * On error (ENOMEM, EAGAIN, ...),
3307 * - con->in_msg == NULL
2450418c 3308 */
4740a623 3309static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3310{
4740a623 3311 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3312 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3313 struct ceph_msg *msg;
4740a623 3314 int ret = 0;
2450418c 3315
1c20f2d2 3316 BUG_ON(con->in_msg != NULL);
53ded495 3317 BUG_ON(!con->ops->alloc_msg);
2450418c 3318
53ded495
AE		 3319 mutex_unlock(&con->mutex);
3320 msg = con->ops->alloc_msg(con, hdr, skip);
3321 mutex_lock(&con->mutex);
3322 if (con->state != CON_STATE_OPEN) {
3323 if (msg)
1d866d1c 3324 ceph_msg_put(msg);
53ded495
AE		 3325 return -EAGAIN;
3326 }
4137577a
AE		 3327 if (msg) {
3328 BUG_ON(*skip);
3329 con->in_msg = msg;
36eb71aa 3330 con->in_msg->con = con->ops->get(con);
92ce034b 3331 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3332 } else {
3333 /*
3334 * Null message pointer means either we should skip
3335 * this message or we couldn't allocate memory. The
3336 * former is not an error.
3337 */
3338 if (*skip)
3339 return 0;
4137577a 3340
67c64eb7 3341 con->error_msg = "error allocating memory for incoming message";
53ded495 3342 return -ENOMEM;
2450418c 3343 }
1c20f2d2 3344 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3345
1c20f2d2
AE		 3346 if (middle_len && !con->in_msg->middle) {
3347 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3348 if (ret < 0) {
1c20f2d2
AE		 3349 ceph_msg_put(con->in_msg);
3350 con->in_msg = NULL;
2450418c
YS		 3351 }
3352 }
9d7f0f13 3353
4740a623 3354 return ret;
2450418c
YS		 3355}
3356
31b8006e
SW		 3357
3358/*
3359 * Free a generically kmalloc'd message.
3360 */
0215e44b 3361static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3362{
0215e44b 3363 dout("%s %p\n", __func__, m);
4965fc38 3364 kvfree(m->front.iov_base);
e3d5d638 3365 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3366}
3367
0215e44b 3368static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3369{
3370 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
5240d9f9
AE		 3371 LIST_HEAD(data);
3372 struct list_head *links;
3373 struct list_head *next;
31b8006e 3374
0215e44b 3375 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3376 WARN_ON(!list_empty(&m->list_head));
3377
3378 /* drop middle, data, if any */
3379 if (m->middle) {
3380 ceph_buffer_put(m->middle);
3381 m->middle = NULL;
31b8006e 3382 }
5240d9f9
AE		 3383
3384 list_splice_init(&m->data, &data);
3385 list_for_each_safe(links, next, &data) {
3386 struct ceph_msg_data *data;
3387
3388 data = list_entry(links, struct ceph_msg_data, links);
3389 list_del_init(links);
3390 ceph_msg_data_destroy(data);
3391 }
a1930804 3392 m->data_length = 0;
58bb3b37 3393
c2e552e7
SW		 3394 if (m->pool)
3395 ceph_msgpool_put(m->pool, m);
3396 else
0215e44b
ID		 3397 ceph_msg_free(m);
3398}
3399
3400struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3401{
3402 dout("%s %p (was %d)\n", __func__, msg,
3403 atomic_read(&msg->kref.refcount));
3404 kref_get(&msg->kref);
3405 return msg;
3406}
3407EXPORT_SYMBOL(ceph_msg_get);
3408
3409void ceph_msg_put(struct ceph_msg *msg)
3410{
3411 dout("%s %p (was %d)\n", __func__, msg,
3412 atomic_read(&msg->kref.refcount));
3413 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3414}
0215e44b 3415EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3416
3417void ceph_msg_dump(struct ceph_msg *msg)
3418{
3cea4c30
ID		 3419 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3420 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3421 print_hex_dump(KERN_DEBUG, "header: ",
3422 DUMP_PREFIX_OFFSET, 16, 1,
3423 &msg->hdr, sizeof(msg->hdr), true);
3424 print_hex_dump(KERN_DEBUG, " front: ",
3425 DUMP_PREFIX_OFFSET, 16, 1,
3426 msg->front.iov_base, msg->front.iov_len, true);
3427 if (msg->middle)
3428 print_hex_dump(KERN_DEBUG, "middle: ",
3429 DUMP_PREFIX_OFFSET, 16, 1,
3430 msg->middle->vec.iov_base,
3431 msg->middle->vec.iov_len, true);
3432 print_hex_dump(KERN_DEBUG, "footer: ",
3433 DUMP_PREFIX_OFFSET, 16, 1,
3434 &msg->footer, sizeof(msg->footer), true);
3435}
3d14c5d2 3436EXPORT_SYMBOL(ceph_msg_dump);
Linux 6.x block layer and io_uring tree(s)