]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/ceph/messenger.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: drop pages parameter
[mirror_ubuntu-bionic-kernel.git] / net / ceph / messenger.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
5a0e3ad6 9#include <linux/slab.h>
31b8006e
SW		 10#include <linux/socket.h>
11#include <linux/string.h>
3ebc21f7 12#ifdef CONFIG_BLOCK
68b4476b 13#include <linux/bio.h>
3ebc21f7 14#endif /* CONFIG_BLOCK */
ee3b56f2 15#include <linux/dns_resolver.h>
31b8006e
SW		 16#include <net/tcp.h>
17
3d14c5d2
YS		 18#include <linux/ceph/libceph.h>
19#include <linux/ceph/messenger.h>
20#include <linux/ceph/decode.h>
21#include <linux/ceph/pagelist.h>
bc3b2d7f 22#include <linux/export.h>
31b8006e 23
fe38a2b6
AE		 24#define list_entry_next(pos, member) \
25 list_entry(pos->member.next, typeof(*pos), member)
26
31b8006e
SW		 27/*
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
33 * the sender.
34 */
35
bc18f4b1
AE		 36/*
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
40 * unexpected state.
41 *
42 * --------
43 * | NEW* | transient initial state
44 * --------
45 * | con_sock_state_init()
46 * v
47 * ----------
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
50 * ^ \
51 * | \ con_sock_state_connecting()
52 * | ----------------------
53 * | \
54 * + con_sock_state_closed() \
fbb85a47
SW		 55 * |+--------------------------- \
56 * | \ \ \
57 * | ----------- \ \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
60 * | ^ \ |
61 * | | \ |
62 * | + con_sock_state_closing() \ |
63 * | / \ | |
64 * | / --------------- | |
65 * | / \ v v
bc18f4b1
AE		 66 * | / --------------
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
70 * | | v
71 * -------------
72 * | CONNECTED | TCP connection established
73 * -------------
74 *
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
76 */
ce2c8903
AE		 77
78#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
83
8dacc7da
SW		 84/*
85 * connection states
86 */
87#define CON_STATE_CLOSED 1 /* -> PREOPEN */
88#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
93
4a861692
SW		 94/*
95 * ceph_connection flag bits
96 */
97#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 103
c9ffc77a
AE		 104static bool con_flag_valid(unsigned long con_flag)
105{
106 switch (con_flag) {
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
112 return true;
113 default:
114 return false;
115 }
116}
117
118static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
119{
120 BUG_ON(!con_flag_valid(con_flag));
121
122 clear_bit(con_flag, &con->flags);
123}
124
125static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
126{
127 BUG_ON(!con_flag_valid(con_flag));
128
129 set_bit(con_flag, &con->flags);
130}
131
132static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
133{
134 BUG_ON(!con_flag_valid(con_flag));
135
136 return test_bit(con_flag, &con->flags);
137}
138
139static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
141{
142 BUG_ON(!con_flag_valid(con_flag));
143
144 return test_and_clear_bit(con_flag, &con->flags);
145}
146
147static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
149{
150 BUG_ON(!con_flag_valid(con_flag));
151
152 return test_and_set_bit(con_flag, &con->flags);
153}
154
31b8006e
SW		 155/* static tag bytes (protocol control messages) */
156static char tag_msg = CEPH_MSGR_TAG_MSG;
157static char tag_ack = CEPH_MSGR_TAG_ACK;
158static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
159
a6a5349d
SW		 160#ifdef CONFIG_LOCKDEP
161static struct lock_class_key socket_class;
162#endif
163
84495f49
AE		 164/*
165 * When skipping (ignoring) a block of input we read it into a "skip
166 * buffer," which is this many bytes in size.
167 */
168#define SKIP_BUF_SIZE 1024
31b8006e
SW		 169
170static void queue_con(struct ceph_connection *con);
171static void con_work(struct work_struct *);
93209264 172static void con_fault(struct ceph_connection *con);
31b8006e 173
31b8006e 174/*
f64a9317
AE		 175 * Nicely render a sockaddr as a string. An array of formatted
176 * strings is used, to approximate reentrancy.
31b8006e 177 */
f64a9317
AE		 178#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
179#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
180#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
181#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
182
183static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
184static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 185
57666519 186static struct page *zero_page; /* used in certain error cases */
57666519 187
3d14c5d2 188const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 189{
190 int i;
191 char *s;
99f0f3b2
AE		 192 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
193 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 194
f64a9317 195 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 196 s = addr_str[i];
197
198 switch (ss->ss_family) {
199 case AF_INET:
bd406145
AE		 200 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
201 ntohs(in4->sin_port));
31b8006e
SW		 202 break;
203
204 case AF_INET6:
bd406145
AE		 205 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
206 ntohs(in6->sin6_port));
31b8006e
SW		 207 break;
208
209 default:
d3002b97
AE		 210 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
211 ss->ss_family);
31b8006e
SW		 212 }
213
214 return s;
215}
3d14c5d2 216EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 217
63f2d211
SW		 218static void encode_my_addr(struct ceph_messenger *msgr)
219{
220 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
221 ceph_encode_addr(&msgr->my_enc_addr);
222}
223
31b8006e
SW		 224/*
225 * work queue for all reading and writing to/from the socket.
226 */
e0f43c94 227static struct workqueue_struct *ceph_msgr_wq;
31b8006e 228
15417167 229static void _ceph_msgr_exit(void)
6173d1f0 230{
d3002b97 231 if (ceph_msgr_wq) {
6173d1f0 232 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 233 ceph_msgr_wq = NULL;
234 }
6173d1f0 235
6173d1f0
AE		 236 BUG_ON(zero_page == NULL);
237 kunmap(zero_page);
238 page_cache_release(zero_page);
239 zero_page = NULL;
240}
241
3d14c5d2 242int ceph_msgr_init(void)
31b8006e 243{
57666519
AE		 244 BUG_ON(zero_page != NULL);
245 zero_page = ZERO_PAGE(0);
246 page_cache_get(zero_page);
247
f363e45f 248 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
6173d1f0
AE		 249 if (ceph_msgr_wq)
250 return 0;
57666519 251
6173d1f0
AE		 252 pr_err("msgr_init failed to create workqueue\n");
253 _ceph_msgr_exit();
57666519 254
6173d1f0 255 return -ENOMEM;
31b8006e 256}
3d14c5d2 257EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 258
259void ceph_msgr_exit(void)
260{
57666519 261 BUG_ON(ceph_msgr_wq == NULL);
57666519 262
6173d1f0 263 _ceph_msgr_exit();
31b8006e 264}
3d14c5d2 265EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 266
cd84db6e 267void ceph_msgr_flush(void)
a922d38f
SW		 268{
269 flush_workqueue(ceph_msgr_wq);
270}
3d14c5d2 271EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 272
ce2c8903
AE		 273/* Connection socket state transition functions */
274
275static void con_sock_state_init(struct ceph_connection *con)
276{
277 int old_state;
278
279 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
280 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
281 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 282 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
283 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 284}
285
286static void con_sock_state_connecting(struct ceph_connection *con)
287{
288 int old_state;
289
290 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
291 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
292 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 293 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
294 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 295}
296
297static void con_sock_state_connected(struct ceph_connection *con)
298{
299 int old_state;
300
301 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
302 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
303 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 304 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
305 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 306}
307
308static void con_sock_state_closing(struct ceph_connection *con)
309{
310 int old_state;
311
312 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
313 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
314 old_state != CON_SOCK_STATE_CONNECTED &&
315 old_state != CON_SOCK_STATE_CLOSING))
316 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 317 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
318 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 319}
320
321static void con_sock_state_closed(struct ceph_connection *con)
322{
323 int old_state;
324
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 327 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 328 old_state != CON_SOCK_STATE_CONNECTING &&
329 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 330 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 331 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
332 CON_SOCK_STATE_CLOSED);
ce2c8903 333}
a922d38f 334
31b8006e
SW		 335/*
336 * socket callback functions
337 */
338
339/* data available on socket, or listen socket received a connect */
327800bd 340static void ceph_sock_data_ready(struct sock *sk, int count_unused)
31b8006e 341{
bd406145 342 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 343 if (atomic_read(&con->msgr->stopping)) {
344 return;
345 }
bd406145 346
31b8006e 347 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 348 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 349 con, con->state);
350 queue_con(con);
351 }
352}
353
354/* socket has buffer space for writing */
327800bd 355static void ceph_sock_write_space(struct sock *sk)
31b8006e 356{
d3002b97 357 struct ceph_connection *con = sk->sk_user_data;
31b8006e 358
182fac26
JS		 359 /* only queue to workqueue if there is data we want to write,
360 * and there is sufficient space in the socket buffer to accept
327800bd 361 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 362 * doesn't get called again until try_write() fills the socket
363 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
364 * and net/core/stream.c:sk_stream_write_space().
365 */
c9ffc77a 366 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
182fac26 367 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
327800bd 368 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 369 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
370 queue_con(con);
371 }
31b8006e 372 } else {
327800bd 373 dout("%s %p nothing to write\n", __func__, con);
31b8006e 374 }
31b8006e
SW		 375}
376
377/* socket's state has changed */
327800bd 378static void ceph_sock_state_change(struct sock *sk)
31b8006e 379{
bd406145 380 struct ceph_connection *con = sk->sk_user_data;
31b8006e 381
327800bd 382 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 383 con, con->state, sk->sk_state);
384
31b8006e
SW		 385 switch (sk->sk_state) {
386 case TCP_CLOSE:
327800bd 387 dout("%s TCP_CLOSE\n", __func__);
31b8006e 388 case TCP_CLOSE_WAIT:
327800bd 389 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 390 con_sock_state_closing(con);
c9ffc77a 391 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 392 queue_con(con);
31b8006e
SW		 393 break;
394 case TCP_ESTABLISHED:
327800bd 395 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 396 con_sock_state_connected(con);
31b8006e
SW		 397 queue_con(con);
398 break;
d3002b97
AE		 399 default: /* Everything else is uninteresting */
400 break;
31b8006e
SW		 401 }
402}
403
404/*
405 * set up socket callbacks
406 */
407static void set_sock_callbacks(struct socket *sock,
408 struct ceph_connection *con)
409{
410 struct sock *sk = sock->sk;
bd406145 411 sk->sk_user_data = con;
327800bd
AE		 412 sk->sk_data_ready = ceph_sock_data_ready;
413 sk->sk_write_space = ceph_sock_write_space;
414 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 415}
416
417
418/*
419 * socket helpers
420 */
421
422/*
423 * initiate connection to a remote socket.
424 */
41617d0c 425static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 426{
f91d3471 427 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e
SW		 428 struct socket *sock;
429 int ret;
430
431 BUG_ON(con->sock);
f91d3471
SW		 432 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
433 IPPROTO_TCP, &sock);
31b8006e 434 if (ret)
41617d0c 435 return ret;
31b8006e
SW		 436 sock->sk->sk_allocation = GFP_NOFS;
437
a6a5349d
SW		 438#ifdef CONFIG_LOCKDEP
439 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
440#endif
441
31b8006e
SW		 442 set_sock_callbacks(sock, con);
443
3d14c5d2 444 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 445
89a86be0 446 con_sock_state_connecting(con);
f91d3471
SW		 447 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
448 O_NONBLOCK);
31b8006e
SW		 449 if (ret == -EINPROGRESS) {
450 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 451 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 452 sock->sk->sk_state);
a5bc3129 453 } else if (ret < 0) {
31b8006e 454 pr_err("connect %s error %d\n",
3d14c5d2 455 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 456 sock_release(sock);
31b8006e 457 con->error_msg = "connect error";
31b8006e 458
41617d0c 459 return ret;
a5bc3129
AE		 460 }
461 con->sock = sock;
41617d0c 462 return 0;
31b8006e
SW		 463}
464
465static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
466{
467 struct kvec iov = {buf, len};
468 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 469 int r;
31b8006e 470
98bdb0aa
SW		 471 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
472 if (r == -EAGAIN)
473 r = 0;
474 return r;
31b8006e
SW		 475}
476
afb3d90e
AE		 477static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
478 int page_offset, size_t length)
479{
480 void *kaddr;
481 int ret;
482
483 BUG_ON(page_offset + length > PAGE_SIZE);
484
485 kaddr = kmap(page);
486 BUG_ON(!kaddr);
487 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
488 kunmap(page);
489
490 return ret;
491}
492
31b8006e
SW		 493/*
494 * write something. @more is true if caller will be sending more data
495 * shortly.
496 */
497static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
498 size_t kvlen, size_t len, int more)
499{
500 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 501 int r;
31b8006e
SW		 502
503 if (more)
504 msg.msg_flags |= MSG_MORE;
505 else
506 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
507
42961d23
SW		 508 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
509 if (r == -EAGAIN)
510 r = 0;
511 return r;
31b8006e
SW		 512}
513
31739139 514static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 515 int offset, size_t size, bool more)
31739139
AE		 516{
517 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
518 int ret;
519
520 ret = kernel_sendpage(sock, page, offset, size, flags);
521 if (ret == -EAGAIN)
522 ret = 0;
523
524 return ret;
525}
526
31b8006e
SW		 527
528/*
529 * Shutdown/close the socket for the given connection.
530 */
531static int con_close_socket(struct ceph_connection *con)
532{
8007b8d6 533 int rc = 0;
31b8006e
SW		 534
535 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 536 if (con->sock) {
537 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
538 sock_release(con->sock);
539 con->sock = NULL;
540 }
456ea468
AE		 541
542 /*
4a861692 543 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 544 * independent of the connection mutex, and we could have
545 * received a socket close event before we had the chance to
546 * shut the socket down.
547 */
c9ffc77a 548 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 549
ce2c8903 550 con_sock_state_closed(con);
31b8006e
SW		 551 return rc;
552}
553
554/*
555 * Reset a connection. Discard all incoming and outgoing messages
556 * and clear *_seq state.
557 */
558static void ceph_msg_remove(struct ceph_msg *msg)
559{
560 list_del_init(&msg->list_head);
38941f80 561 BUG_ON(msg->con == NULL);
36eb71aa 562 msg->con->ops->put(msg->con);
38941f80
AE		 563 msg->con = NULL;
564
31b8006e
SW		 565 ceph_msg_put(msg);
566}
567static void ceph_msg_remove_list(struct list_head *head)
568{
569 while (!list_empty(head)) {
570 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
571 list_head);
572 ceph_msg_remove(msg);
573 }
574}
575
576static void reset_connection(struct ceph_connection *con)
577{
578 /* reset connection, out_queue, msg_ and connect_seq */
579 /* discard existing out_queue and msg_seq */
0fa6ebc6 580 dout("reset_connection %p\n", con);
31b8006e
SW		 581 ceph_msg_remove_list(&con->out_queue);
582 ceph_msg_remove_list(&con->out_sent);
583
cf3e5c40 584 if (con->in_msg) {
38941f80
AE		 585 BUG_ON(con->in_msg->con != con);
586 con->in_msg->con = NULL;
cf3e5c40
SW		 587 ceph_msg_put(con->in_msg);
588 con->in_msg = NULL;
36eb71aa 589 con->ops->put(con);
cf3e5c40
SW		 590 }
591
31b8006e
SW		 592 con->connect_seq = 0;
593 con->out_seq = 0;
c86a2930
SW		 594 if (con->out_msg) {
595 ceph_msg_put(con->out_msg);
596 con->out_msg = NULL;
597 }
31b8006e 598 con->in_seq = 0;
0e0d5e0c 599 con->in_seq_acked = 0;
31b8006e
SW		 600}
601
602/*
603 * mark a peer down. drop any open connections.
604 */
605void ceph_con_close(struct ceph_connection *con)
606{
8c50c817 607 mutex_lock(&con->mutex);
3d14c5d2
YS		 608 dout("con_close %p peer %s\n", con,
609 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 610 con->state = CON_STATE_CLOSED;
a5988c49 611
c9ffc77a
AE		 612 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
613 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
614 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
615 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 616
31b8006e 617 reset_connection(con);
6f2bc3ff 618 con->peer_global_seq = 0;
91e45ce3 619 cancel_delayed_work(&con->work);
ee76e073 620 con_close_socket(con);
ec302645 621 mutex_unlock(&con->mutex);
31b8006e 622}
3d14c5d2 623EXPORT_SYMBOL(ceph_con_close);
31b8006e 624
31b8006e
SW		 625/*
626 * Reopen a closed connection, with a new peer address.
627 */
b7a9e5dd
SW		 628void ceph_con_open(struct ceph_connection *con,
629 __u8 entity_type, __u64 entity_num,
630 struct ceph_entity_addr *addr)
31b8006e 631{
5469155f 632 mutex_lock(&con->mutex);
3d14c5d2 633 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 634
122070a2 635 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 636 con->state = CON_STATE_PREOPEN;
a5988c49 637
b7a9e5dd
SW		 638 con->peer_name.type = (__u8) entity_type;
639 con->peer_name.num = cpu_to_le64(entity_num);
640
31b8006e 641 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 642 con->delay = 0; /* reset backoff memory */
5469155f 643 mutex_unlock(&con->mutex);
31b8006e
SW		 644 queue_con(con);
645}
3d14c5d2 646EXPORT_SYMBOL(ceph_con_open);
31b8006e 647
87b315a5
SW		 648/*
649 * return true if this connection ever successfully opened
650 */
651bool ceph_con_opened(struct ceph_connection *con)
652{
653 return con->connect_seq > 0;
654}
655
31b8006e
SW		 656/*
657 * initialize a new connection.
658 */
1bfd89f4
AE		 659void ceph_con_init(struct ceph_connection *con, void *private,
660 const struct ceph_connection_operations *ops,
b7a9e5dd 661 struct ceph_messenger *msgr)
31b8006e
SW		 662{
663 dout("con_init %p\n", con);
664 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 665 con->private = private;
666 con->ops = ops;
31b8006e 667 con->msgr = msgr;
ce2c8903
AE		 668
669 con_sock_state_init(con);
670
ec302645 671 mutex_init(&con->mutex);
31b8006e
SW		 672 INIT_LIST_HEAD(&con->out_queue);
673 INIT_LIST_HEAD(&con->out_sent);
674 INIT_DELAYED_WORK(&con->work, con_work);
a5988c49 675
8dacc7da 676 con->state = CON_STATE_CLOSED;
31b8006e 677}
3d14c5d2 678EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 679
680
681/*
682 * We maintain a global counter to order connection attempts. Get
683 * a unique seq greater than @gt.
684 */
685static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
686{
687 u32 ret;
688
689 spin_lock(&msgr->global_seq_lock);
690 if (msgr->global_seq < gt)
691 msgr->global_seq = gt;
692 ret = ++msgr->global_seq;
693 spin_unlock(&msgr->global_seq_lock);
694 return ret;
695}
696
e2200423 697static void con_out_kvec_reset(struct ceph_connection *con)
859eb799
AE		 698{
699 con->out_kvec_left = 0;
700 con->out_kvec_bytes = 0;
701 con->out_kvec_cur = &con->out_kvec[0];
702}
703
e2200423 704static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 705 size_t size, void *data)
706{
707 int index;
708
709 index = con->out_kvec_left;
710 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
711
712 con->out_kvec[index].iov_len = size;
713 con->out_kvec[index].iov_base = data;
714 con->out_kvec_left++;
715 con->out_kvec_bytes += size;
716}
31b8006e 717
df6ad1f9 718#ifdef CONFIG_BLOCK
07c09b72
AE		 719static void init_bio_iter(struct bio *bio, struct bio **bio_iter,
720 unsigned int *bio_seg)
df6ad1f9
AE		 721{
722 if (!bio) {
07c09b72
AE		 723 *bio_iter = NULL;
724 *bio_seg = 0;
df6ad1f9
AE		 725 return;
726 }
07c09b72
AE		 727 *bio_iter = bio;
728 *bio_seg = (unsigned int) bio->bi_idx;
df6ad1f9
AE		 729}
730
07c09b72 731static void iter_bio_next(struct bio **bio_iter, unsigned int *seg)
df6ad1f9
AE		 732{
733 if (*bio_iter == NULL)
734 return;
735
736 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
737
738 (*seg)++;
739 if (*seg == (*bio_iter)->bi_vcnt)
740 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
741}
6aaa4511
AE		 742
743/*
744 * For a bio data item, a piece is whatever remains of the next
745 * entry in the current bio iovec, or the first entry in the next
746 * bio in the list.
747 */
748static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data *data)
749{
750 struct ceph_msg_data_cursor *cursor = &data->cursor;
751 struct bio *bio;
752
753 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
754
755 bio = data->bio;
756 BUG_ON(!bio);
757 BUG_ON(!bio->bi_vcnt);
758 /* resid = bio->bi_size */
759
760 cursor->bio = bio;
761 cursor->vector_index = 0;
762 cursor->vector_offset = 0;
763 cursor->last_piece = !bio->bi_next && bio->bi_vcnt == 1;
764}
765
766static struct page *ceph_msg_data_bio_next(struct ceph_msg_data *data,
767 size_t *page_offset,
768 size_t *length)
769{
770 struct ceph_msg_data_cursor *cursor = &data->cursor;
771 struct bio *bio;
772 struct bio_vec *bio_vec;
773 unsigned int index;
774
775 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
776
777 bio = cursor->bio;
778 BUG_ON(!bio);
779
780 index = cursor->vector_index;
781 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
782
783 bio_vec = &bio->bi_io_vec[index];
784 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
785 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
786 BUG_ON(*page_offset >= PAGE_SIZE);
787 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
788 BUG_ON(*length > PAGE_SIZE);
789
790 return bio_vec->bv_page;
791}
792
793static bool ceph_msg_data_bio_advance(struct ceph_msg_data *data, size_t bytes)
794{
795 struct ceph_msg_data_cursor *cursor = &data->cursor;
796 struct bio *bio;
797 struct bio_vec *bio_vec;
798 unsigned int index;
799
800 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
801
802 bio = cursor->bio;
803 BUG_ON(!bio);
804
805 index = cursor->vector_index;
806 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
807 bio_vec = &bio->bi_io_vec[index];
808 BUG_ON(cursor->vector_offset + bytes > bio_vec->bv_len);
809
810 /* Advance the cursor offset */
811
812 cursor->vector_offset += bytes;
813 if (cursor->vector_offset < bio_vec->bv_len)
814 return false; /* more bytes to process in this segment */
815
816 /* Move on to the next segment, and possibly the next bio */
817
818 if (++cursor->vector_index == (unsigned int) bio->bi_vcnt) {
819 bio = bio->bi_next;
820 cursor->bio = bio;
821 cursor->vector_index = 0;
822 }
823 cursor->vector_offset = 0;
824
825 if (!cursor->last_piece && bio && !bio->bi_next)
826 if (cursor->vector_index == (unsigned int) bio->bi_vcnt - 1)
827 cursor->last_piece = true;
828
829 return true;
830}
df6ad1f9
AE		 831#endif
832
e766d7b5
AE		 833/*
834 * For a page array, a piece comes from the first page in the array
835 * that has not already been fully consumed.
836 */
837static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data *data)
838{
839 struct ceph_msg_data_cursor *cursor = &data->cursor;
840 int page_count;
841
842 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
843
844 BUG_ON(!data->pages);
845 BUG_ON(!data->length);
846
847 page_count = calc_pages_for(data->alignment, (u64)data->length);
848 BUG_ON(page_count > (int) USHRT_MAX);
849 cursor->resid = data->length;
850 cursor->page_offset = data->alignment & ~PAGE_MASK;
851 cursor->page_index = 0;
852 cursor->page_count = (unsigned short) page_count;
853 cursor->last_piece = cursor->page_count == 1;
854}
855
856static struct page *ceph_msg_data_pages_next(struct ceph_msg_data *data,
857 size_t *page_offset,
858 size_t *length)
859{
860 struct ceph_msg_data_cursor *cursor = &data->cursor;
861
862 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
863
864 BUG_ON(cursor->page_index >= cursor->page_count);
865 BUG_ON(cursor->page_offset >= PAGE_SIZE);
866 BUG_ON(!cursor->resid);
867
868 *page_offset = cursor->page_offset;
869 if (cursor->last_piece) {
870 BUG_ON(*page_offset + cursor->resid > PAGE_SIZE);
871 *length = cursor->resid;
872 } else {
873 *length = PAGE_SIZE - *page_offset;
874 }
875
876 return data->pages[cursor->page_index];
877}
878
879static bool ceph_msg_data_pages_advance(struct ceph_msg_data *data,
880 size_t bytes)
881{
882 struct ceph_msg_data_cursor *cursor = &data->cursor;
883
884 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
885
886 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
887 BUG_ON(bytes > cursor->resid);
888
889 /* Advance the cursor page offset */
890
891 cursor->resid -= bytes;
892 cursor->page_offset += bytes;
893 if (!bytes || cursor->page_offset & ~PAGE_MASK)
894 return false; /* more bytes to process in the current page */
895
896 /* Move on to the next page */
897
898 BUG_ON(cursor->page_index >= cursor->page_count);
899 cursor->page_offset = 0;
900 cursor->page_index++;
901 cursor->last_piece = cursor->page_index == cursor->page_count - 1;
902
903 return true;
904}
905
fe38a2b6 906/*
dd236fcb
AE		 907 * For a pagelist, a piece is whatever remains to be consumed in the
908 * first page in the list, or the front of the next page.
fe38a2b6 909 */
dd236fcb 910static void ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data *data)
fe38a2b6
AE		 911{
912 struct ceph_msg_data_cursor *cursor = &data->cursor;
913 struct ceph_pagelist *pagelist;
914 struct page *page;
915
dd236fcb 916 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 917
918 pagelist = data->pagelist;
919 BUG_ON(!pagelist);
920 if (!pagelist->length)
921 return; /* pagelist can be assigned but empty */
922
923 BUG_ON(list_empty(&pagelist->head));
924 page = list_first_entry(&pagelist->head, struct page, lru);
925
926 cursor->page = page;
927 cursor->offset = 0;
928 cursor->last_piece = pagelist->length <= PAGE_SIZE;
929}
930
dd236fcb 931static struct page *ceph_msg_data_pagelist_next(struct ceph_msg_data *data,
fe38a2b6 932 size_t *page_offset,
dd236fcb 933 size_t *length)
fe38a2b6
AE		 934{
935 struct ceph_msg_data_cursor *cursor = &data->cursor;
936 struct ceph_pagelist *pagelist;
937 size_t piece_end;
938
939 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
940
941 pagelist = data->pagelist;
942 BUG_ON(!pagelist);
943
944 BUG_ON(!cursor->page);
945 BUG_ON(cursor->offset >= pagelist->length);
946
dd236fcb 947 if (cursor->last_piece) {
fe38a2b6
AE		 948 /* pagelist offset is always 0 */
949 piece_end = pagelist->length & ~PAGE_MASK;
950 if (!piece_end)
951 piece_end = PAGE_SIZE;
952 } else {
953 piece_end = PAGE_SIZE;
954 }
955 *page_offset = cursor->offset & ~PAGE_MASK;
956 *length = piece_end - *page_offset;
957
958 return data->cursor.page;
959}
960
dd236fcb
AE		 961static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data *data,
962 size_t bytes)
fe38a2b6
AE		 963{
964 struct ceph_msg_data_cursor *cursor = &data->cursor;
965 struct ceph_pagelist *pagelist;
966
967 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
968
969 pagelist = data->pagelist;
970 BUG_ON(!pagelist);
971 BUG_ON(!cursor->page);
972 BUG_ON(cursor->offset + bytes > pagelist->length);
973 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
974
975 /* Advance the cursor offset */
976
977 cursor->offset += bytes;
978 /* pagelist offset is always 0 */
979 if (!bytes || cursor->offset & ~PAGE_MASK)
980 return false; /* more bytes to process in the current page */
981
982 /* Move on to the next page */
983
984 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
985 cursor->page = list_entry_next(cursor->page, lru);
986
987 /* cursor offset is at page boundary; pagelist offset is always 0 */
988 if (pagelist->length - cursor->offset <= PAGE_SIZE)
989 cursor->last_piece = true;
990
991 return true;
992}
993
dd236fcb
AE		 994/*
995 * Message data is handled (sent or received) in pieces, where each
996 * piece resides on a single page. The network layer might not
997 * consume an entire piece at once. A data item's cursor keeps
998 * track of which piece is next to process and how much remains to
999 * be processed in that piece. It also tracks whether the current
1000 * piece is the last one in the data item.
1001 */
1002static void ceph_msg_data_cursor_init(struct ceph_msg_data *data)
1003{
1004 switch (data->type) {
1005 case CEPH_MSG_DATA_PAGELIST:
1006 ceph_msg_data_pagelist_cursor_init(data);
1007 break;
e766d7b5
AE		 1008 case CEPH_MSG_DATA_PAGES:
1009 ceph_msg_data_pages_cursor_init(data);
1010 break;
dd236fcb
AE		 1011#ifdef CONFIG_BLOCK
1012 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1013 ceph_msg_data_bio_cursor_init(data);
1014 break;
dd236fcb 1015#endif /* CONFIG_BLOCK */
6aaa4511 1016 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1017 default:
1018 /* BUG(); */
1019 break;
1020 }
1021}
1022
1023/*
1024 * Return the page containing the next piece to process for a given
1025 * data item, and supply the page offset and length of that piece.
1026 * Indicate whether this is the last piece in this data item.
1027 */
1028static struct page *ceph_msg_data_next(struct ceph_msg_data *data,
1029 size_t *page_offset,
1030 size_t *length,
1031 bool *last_piece)
1032{
1033 struct page *page;
1034
1035 switch (data->type) {
1036 case CEPH_MSG_DATA_PAGELIST:
1037 page = ceph_msg_data_pagelist_next(data, page_offset, length);
1038 break;
e766d7b5
AE		 1039 case CEPH_MSG_DATA_PAGES:
1040 page = ceph_msg_data_pages_next(data, page_offset, length);
1041 break;
dd236fcb
AE		 1042#ifdef CONFIG_BLOCK
1043 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1044 page = ceph_msg_data_bio_next(data, page_offset, length);
1045 break;
dd236fcb 1046#endif /* CONFIG_BLOCK */
6aaa4511 1047 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1048 default:
1049 page = NULL;
1050 break;
1051 }
1052 BUG_ON(!page);
1053 BUG_ON(*page_offset + *length > PAGE_SIZE);
1054 BUG_ON(!*length);
1055 if (last_piece)
1056 *last_piece = data->cursor.last_piece;
1057
1058 return page;
1059}
1060
1061/*
1062 * Returns true if the result moves the cursor on to the next piece
1063 * of the data item.
1064 */
1065static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes)
1066{
1067 bool new_piece;
1068
1069 switch (data->type) {
1070 case CEPH_MSG_DATA_PAGELIST:
1071 new_piece = ceph_msg_data_pagelist_advance(data, bytes);
1072 break;
e766d7b5
AE		 1073 case CEPH_MSG_DATA_PAGES:
1074 new_piece = ceph_msg_data_pages_advance(data, bytes);
1075 break;
dd236fcb
AE		 1076#ifdef CONFIG_BLOCK
1077 case CEPH_MSG_DATA_BIO:
6aaa4511
AE		 1078 new_piece = ceph_msg_data_bio_advance(data, bytes);
1079 break;
dd236fcb 1080#endif /* CONFIG_BLOCK */
6aaa4511 1081 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1082 default:
1083 BUG();
1084 break;
1085 }
1086
1087 return new_piece;
1088}
1089
78625051
AE		 1090static void prepare_message_data(struct ceph_msg *msg,
1091 struct ceph_msg_pos *msg_pos)
739c905b 1092{
739c905b
AE		 1093 BUG_ON(!msg);
1094 BUG_ON(!msg->hdr.data_len);
1095
1096 /* initialize page iterator */
bae6acd9 1097 msg_pos->page = 0;
97fb1c7f 1098 if (ceph_msg_has_pages(msg))
f9e15777 1099 msg_pos->page_pos = msg->p.alignment;
739c905b 1100 else
bae6acd9 1101 msg_pos->page_pos = 0;
572c588e 1102#ifdef CONFIG_BLOCK
97fb1c7f 1103 if (ceph_msg_has_bio(msg))
f9e15777 1104 init_bio_iter(msg->b.bio, &msg->b.bio_iter, &msg->b.bio_seg);
572c588e 1105#endif
bae6acd9 1106 msg_pos->data_pos = 0;
fe38a2b6 1107
7fe1e5e5 1108 /* Initialize data cursors */
fe38a2b6 1109
6aaa4511
AE		 1110#ifdef CONFIG_BLOCK
1111 if (ceph_msg_has_bio(msg))
1112 ceph_msg_data_cursor_init(&msg->b);
1113#endif /* CONFIG_BLOCK */
e766d7b5
AE		 1114 if (ceph_msg_has_pages(msg))
1115 ceph_msg_data_cursor_init(&msg->p);
7fe1e5e5
AE		 1116 if (ceph_msg_has_pagelist(msg))
1117 ceph_msg_data_cursor_init(&msg->l);
fe38a2b6 1118
bae6acd9 1119 msg_pos->did_page_crc = false;
739c905b
AE		 1120}
1121
31b8006e
SW		 1122/*
1123 * Prepare footer for currently outgoing message, and finish things
1124 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1125 */
859eb799 1126static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1127{
1128 struct ceph_msg *m = con->out_msg;
859eb799 1129 int v = con->out_kvec_left;
31b8006e 1130
fd154f3c
AE		 1131 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1132
31b8006e
SW		 1133 dout("prepare_write_message_footer %p\n", con);
1134 con->out_kvec_is_msg = true;
1135 con->out_kvec[v].iov_base = &m->footer;
1136 con->out_kvec[v].iov_len = sizeof(m->footer);
1137 con->out_kvec_bytes += sizeof(m->footer);
1138 con->out_kvec_left++;
1139 con->out_more = m->more_to_follow;
c86a2930 1140 con->out_msg_done = true;
31b8006e
SW		 1141}
1142
1143/*
1144 * Prepare headers for the next outgoing message.
1145 */
1146static void prepare_write_message(struct ceph_connection *con)
1147{
1148 struct ceph_msg *m;
a9a0c51a 1149 u32 crc;
31b8006e 1150
e2200423 1151 con_out_kvec_reset(con);
31b8006e 1152 con->out_kvec_is_msg = true;
c86a2930 1153 con->out_msg_done = false;
31b8006e
SW		 1154
1155 /* Sneak an ack in there first? If we can get it into the same
1156 * TCP packet that's a good thing. */
1157 if (con->in_seq > con->in_seq_acked) {
1158 con->in_seq_acked = con->in_seq;
e2200423 1159 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1160 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1161 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1162 &con->out_temp_ack);
31b8006e
SW		 1163 }
1164
38941f80 1165 BUG_ON(list_empty(&con->out_queue));
859eb799 1166 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1167 con->out_msg = m;
38941f80 1168 BUG_ON(m->con != con);
4cf9d544
SW		 1169
1170 /* put message on sent list */
1171 ceph_msg_get(m);
1172 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1173
e84346b7
SW		 1174 /*
1175 * only assign outgoing seq # if we haven't sent this message
1176 * yet. if it is requeued, resend with it's original seq.
1177 */
1178 if (m->needs_out_seq) {
1179 m->hdr.seq = cpu_to_le64(++con->out_seq);
1180 m->needs_out_seq = false;
1181 }
31b8006e 1182
4a73ef27 1183 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n",
31b8006e
SW		 1184 m, con->out_seq, le16_to_cpu(m->hdr.type),
1185 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
f9e15777 1186 le32_to_cpu(m->hdr.data_len), m->p.length);
31b8006e
SW		 1187 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1188
1189 /* tag + hdr + front + middle */
e2200423
AE		 1190 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1191 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1192 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1193
31b8006e 1194 if (m->middle)
e2200423 1195 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1196 m->middle->vec.iov_base);
31b8006e
SW		 1197
1198 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1199 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1200 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1201 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1202
1203 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1204 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1205 if (m->middle) {
1206 crc = crc32c(0, m->middle->vec.iov_base,
1207 m->middle->vec.iov_len);
1208 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1209 } else
31b8006e 1210 con->out_msg->footer.middle_crc = 0;
739c905b 1211 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1212 le32_to_cpu(con->out_msg->footer.front_crc),
1213 le32_to_cpu(con->out_msg->footer.middle_crc));
1214
1215 /* is there a data payload? */
739c905b 1216 con->out_msg->footer.data_crc = 0;
78625051
AE		 1217 if (m->hdr.data_len) {
1218 prepare_message_data(con->out_msg, &con->out_msg_pos);
1219 con->out_more = 1; /* data + footer will follow */
1220 } else {
31b8006e 1221 /* no, queue up footer too and be done */
859eb799 1222 prepare_write_message_footer(con);
78625051 1223 }
31b8006e 1224
c9ffc77a 1225 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1226}
1227
1228/*
1229 * Prepare an ack.
1230 */
1231static void prepare_write_ack(struct ceph_connection *con)
1232{
1233 dout("prepare_write_ack %p %llu -> %llu\n", con,
1234 con->in_seq_acked, con->in_seq);
1235 con->in_seq_acked = con->in_seq;
1236
e2200423 1237 con_out_kvec_reset(con);
859eb799 1238
e2200423 1239 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1240
31b8006e 1241 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1242 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1243 &con->out_temp_ack);
1244
31b8006e 1245 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1246 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1247}
1248
3a23083b
SW		 1249/*
1250 * Prepare to share the seq during handshake
1251 */
1252static void prepare_write_seq(struct ceph_connection *con)
1253{
1254 dout("prepare_write_seq %p %llu -> %llu\n", con,
1255 con->in_seq_acked, con->in_seq);
1256 con->in_seq_acked = con->in_seq;
1257
1258 con_out_kvec_reset(con);
1259
1260 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1261 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1262 &con->out_temp_ack);
1263
1264 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1265}
1266
31b8006e
SW		 1267/*
1268 * Prepare to write keepalive byte.
1269 */
1270static void prepare_write_keepalive(struct ceph_connection *con)
1271{
1272 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1273 con_out_kvec_reset(con);
1274 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1275 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1276}
1277
1278/*
1279 * Connection negotiation.
1280 */
1281
dac1e716
AE		 1282static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1283 int *auth_proto)
4e7a5dcd 1284{
a3530df3 1285 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1286
1287 if (!con->ops->get_authorizer) {
1288 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1289 con->out_connect.authorizer_len = 0;
729796be 1290 return NULL;
b1c6b980
AE		 1291 }
1292
1293 /* Can't hold the mutex while getting authorizer */
ec302645 1294 mutex_unlock(&con->mutex);
dac1e716 1295 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1296 mutex_lock(&con->mutex);
4e7a5dcd 1297
a3530df3 1298 if (IS_ERR(auth))
729796be 1299 return auth;
8dacc7da 1300 if (con->state != CON_STATE_NEGOTIATING)
729796be 1301 return ERR_PTR(-EAGAIN);
0da5d703 1302
8f43fb53
AE		 1303 con->auth_reply_buf = auth->authorizer_reply_buf;
1304 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1305 return auth;
4e7a5dcd
SW		 1306}
1307
31b8006e
SW		 1308/*
1309 * We connected to a peer and are saying hello.
1310 */
e825a66d 1311static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1312{
e2200423
AE		 1313 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1314 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1315 &con->msgr->my_enc_addr);
eed0ef2c 1316
eed0ef2c 1317 con->out_more = 0;
c9ffc77a 1318 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1319}
1320
e825a66d 1321static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1322{
95c96174 1323 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1324 int proto;
dac1e716 1325 int auth_proto;
729796be 1326 struct ceph_auth_handshake *auth;
31b8006e
SW		 1327
1328 switch (con->peer_name.type) {
1329 case CEPH_ENTITY_TYPE_MON:
1330 proto = CEPH_MONC_PROTOCOL;
1331 break;
1332 case CEPH_ENTITY_TYPE_OSD:
1333 proto = CEPH_OSDC_PROTOCOL;
1334 break;
1335 case CEPH_ENTITY_TYPE_MDS:
1336 proto = CEPH_MDSC_PROTOCOL;
1337 break;
1338 default:
1339 BUG();
1340 }
1341
1342 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1343 con->connect_seq, global_seq, proto);
4e7a5dcd 1344
e825a66d 1345 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1346 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1347 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1348 con->out_connect.global_seq = cpu_to_le32(global_seq);
1349 con->out_connect.protocol_version = cpu_to_le32(proto);
1350 con->out_connect.flags = 0;
31b8006e 1351
dac1e716
AE		 1352 auth_proto = CEPH_AUTH_UNKNOWN;
1353 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1354 if (IS_ERR(auth))
1355 return PTR_ERR(auth);
3da54776 1356
dac1e716 1357 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1358 con->out_connect.authorizer_len = auth ?
1359 cpu_to_le32(auth->authorizer_buf_len) : 0;
1360
e2200423 1361 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1362 &con->out_connect);
1363 if (auth && auth->authorizer_buf_len)
e2200423 1364 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1365 auth->authorizer_buf);
859eb799 1366
31b8006e 1367 con->out_more = 0;
c9ffc77a 1368 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1369
e10c758e 1370 return 0;
31b8006e
SW		 1371}
1372
31b8006e
SW		 1373/*
1374 * write as much of pending kvecs to the socket as we can.
1375 * 1 -> done
1376 * 0 -> socket full, but more to do
1377 * <0 -> error
1378 */
1379static int write_partial_kvec(struct ceph_connection *con)
1380{
1381 int ret;
1382
1383 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1384 while (con->out_kvec_bytes > 0) {
1385 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1386 con->out_kvec_left, con->out_kvec_bytes,
1387 con->out_more);
1388 if (ret <= 0)
1389 goto out;
1390 con->out_kvec_bytes -= ret;
1391 if (con->out_kvec_bytes == 0)
1392 break; /* done */
f42299e6
AE		 1393
1394 /* account for full iov entries consumed */
1395 while (ret >= con->out_kvec_cur->iov_len) {
1396 BUG_ON(!con->out_kvec_left);
1397 ret -= con->out_kvec_cur->iov_len;
1398 con->out_kvec_cur++;
1399 con->out_kvec_left--;
1400 }
1401 /* and for a partially-consumed entry */
1402 if (ret) {
1403 con->out_kvec_cur->iov_len -= ret;
1404 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1405 }
1406 }
1407 con->out_kvec_left = 0;
1408 con->out_kvec_is_msg = false;
1409 ret = 1;
1410out:
1411 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1412 con->out_kvec_bytes, con->out_kvec_left, ret);
1413 return ret; /* done! */
1414}
1415
84ca8fc8 1416static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
9d2a06c2 1417 size_t len, size_t sent)
68b4476b 1418{
84ca8fc8 1419 struct ceph_msg *msg = con->out_msg;
bae6acd9 1420 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
7fe1e5e5 1421 bool need_crc = false;
68b4476b 1422
84ca8fc8
AE		 1423 BUG_ON(!msg);
1424 BUG_ON(!sent);
68b4476b 1425
bae6acd9
AE		 1426 msg_pos->data_pos += sent;
1427 msg_pos->page_pos += sent;
9d2a06c2 1428 if (ceph_msg_has_pages(msg))
e766d7b5 1429 need_crc = ceph_msg_data_advance(&msg->p, sent);
7fe1e5e5
AE		 1430 else if (ceph_msg_has_pagelist(msg))
1431 need_crc = ceph_msg_data_advance(&msg->l, sent);
6aaa4511
AE		 1432#ifdef CONFIG_BLOCK
1433 else if (ceph_msg_has_bio(msg))
1434 need_crc = ceph_msg_data_advance(&msg->b, sent);
1435#endif /* CONFIG_BLOCK */
7fe1e5e5
AE		 1436 BUG_ON(need_crc && sent != len);
1437
5821bd8c
AE		 1438 if (sent < len)
1439 return;
68b4476b 1440
5821bd8c 1441 BUG_ON(sent != len);
bae6acd9
AE		 1442 msg_pos->page_pos = 0;
1443 msg_pos->page++;
1444 msg_pos->did_page_crc = false;
84ca8fc8 1445}
68b4476b 1446
e788182f
AE		 1447static void in_msg_pos_next(struct ceph_connection *con, size_t len,
1448 size_t received)
1449{
1450 struct ceph_msg *msg = con->in_msg;
bae6acd9 1451 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
e788182f
AE		 1452
1453 BUG_ON(!msg);
1454 BUG_ON(!received);
1455
bae6acd9
AE		 1456 msg_pos->data_pos += received;
1457 msg_pos->page_pos += received;
e788182f
AE		 1458 if (received < len)
1459 return;
1460
1461 BUG_ON(received != len);
bae6acd9
AE		 1462 msg_pos->page_pos = 0;
1463 msg_pos->page++;
e788182f 1464#ifdef CONFIG_BLOCK
f9e15777
AE		 1465 if (msg->b.bio)
1466 iter_bio_next(&msg->b.bio_iter, &msg->b.bio_seg);
e788182f
AE		 1467#endif /* CONFIG_BLOCK */
1468}
1469
35b62808
AE		 1470static u32 ceph_crc32c_page(u32 crc, struct page *page,
1471 unsigned int page_offset,
1472 unsigned int length)
1473{
1474 char *kaddr;
1475
1476 kaddr = kmap(page);
1477 BUG_ON(kaddr == NULL);
1478 crc = crc32c(crc, kaddr + page_offset, length);
1479 kunmap(page);
1480
1481 return crc;
1482}
31b8006e
SW		 1483/*
1484 * Write as much message data payload as we can. If we finish, queue
1485 * up the footer.
1486 * 1 -> done, footer is now queued in out_kvec[].
1487 * 0 -> socket full, but more to do
1488 * <0 -> error
1489 */
34d2d200 1490static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1491{
1492 struct ceph_msg *msg = con->out_msg;
bae6acd9 1493 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
95c96174 1494 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
37675b0f 1495 bool do_datacrc = !con->msgr->nocrc;
31b8006e
SW		 1496 int ret;
1497
34d2d200 1498 dout("%s %p msg %p page %d offset %d\n", __func__,
bae6acd9 1499 con, msg, msg_pos->page, msg_pos->page_pos);
31b8006e 1500
5821bd8c
AE		 1501 /*
1502 * Iterate through each page that contains data to be
1503 * written, and send as much as possible for each.
1504 *
1505 * If we are calculating the data crc (the default), we will
1506 * need to map the page. If we have no pages, they have
1507 * been revoked, so use the zero page.
1508 */
bae6acd9 1509 while (data_len > msg_pos->data_pos) {
8a166d05 1510 struct page *page;
e387d525
AE		 1511 size_t page_offset;
1512 size_t length;
8a166d05 1513 bool last_piece;
68b4476b 1514
9d2a06c2 1515 if (ceph_msg_has_pages(msg)) {
e766d7b5
AE		 1516 page = ceph_msg_data_next(&msg->p, &page_offset,
1517 &length, &last_piece);
97fb1c7f 1518 } else if (ceph_msg_has_pagelist(msg)) {
7fe1e5e5
AE		 1519 page = ceph_msg_data_next(&msg->l, &page_offset,
1520 &length, &last_piece);
68b4476b 1521#ifdef CONFIG_BLOCK
97fb1c7f 1522 } else if (ceph_msg_has_bio(msg)) {
6aaa4511
AE		 1523 page = ceph_msg_data_next(&msg->b, &page_offset,
1524 &length, &last_piece);
68b4476b 1525#endif
31b8006e 1526 } else {
8a166d05 1527 size_t resid = data_len - msg_pos->data_pos;
68b4476b 1528
8a166d05 1529 page = zero_page;
6aaa4511 1530 page_offset = msg_pos->page_pos;
8a166d05
AE		 1531 length = PAGE_SIZE - page_offset;
1532 length = min(resid, length);
1533 last_piece = length == resid;
6aaa4511 1534 }
bae6acd9 1535 if (do_datacrc && !msg_pos->did_page_crc) {
5821bd8c 1536 u32 crc = le32_to_cpu(msg->footer.data_crc);
31b8006e 1537
35b62808 1538 crc = ceph_crc32c_page(crc, page, page_offset, length);
84ca8fc8 1539 msg->footer.data_crc = cpu_to_le32(crc);
bae6acd9 1540 msg_pos->did_page_crc = true;
31b8006e 1541 }
e387d525 1542 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1543 length, last_piece);
31b8006e
SW		 1544 if (ret <= 0)
1545 goto out;
1546
9d2a06c2 1547 out_msg_pos_next(con, page, length, (size_t) ret);
31b8006e
SW		 1548 }
1549
34d2d200 1550 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1551
1552 /* prepare and queue up footer, too */
37675b0f 1553 if (!do_datacrc)
84ca8fc8 1554 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1555 con_out_kvec_reset(con);
859eb799 1556 prepare_write_message_footer(con);
31b8006e
SW		 1557 ret = 1;
1558out:
1559 return ret;
1560}
1561
1562/*
1563 * write some zeros
1564 */
1565static int write_partial_skip(struct ceph_connection *con)
1566{
1567 int ret;
1568
1569 while (con->out_skip > 0) {
31739139 1570 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1571
e1dcb128 1572 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1573 if (ret <= 0)
1574 goto out;
1575 con->out_skip -= ret;
1576 }
1577 ret = 1;
1578out:
1579 return ret;
1580}
1581
1582/*
1583 * Prepare to read connection handshake, or an ack.
1584 */
eed0ef2c
SW		 1585static void prepare_read_banner(struct ceph_connection *con)
1586{
1587 dout("prepare_read_banner %p\n", con);
1588 con->in_base_pos = 0;
1589}
1590
31b8006e
SW		 1591static void prepare_read_connect(struct ceph_connection *con)
1592{
1593 dout("prepare_read_connect %p\n", con);
1594 con->in_base_pos = 0;
1595}
1596
1597static void prepare_read_ack(struct ceph_connection *con)
1598{
1599 dout("prepare_read_ack %p\n", con);
1600 con->in_base_pos = 0;
1601}
1602
3a23083b
SW		 1603static void prepare_read_seq(struct ceph_connection *con)
1604{
1605 dout("prepare_read_seq %p\n", con);
1606 con->in_base_pos = 0;
1607 con->in_tag = CEPH_MSGR_TAG_SEQ;
1608}
1609
31b8006e
SW		 1610static void prepare_read_tag(struct ceph_connection *con)
1611{
1612 dout("prepare_read_tag %p\n", con);
1613 con->in_base_pos = 0;
1614 con->in_tag = CEPH_MSGR_TAG_READY;
1615}
1616
1617/*
1618 * Prepare to read a message.
1619 */
1620static int prepare_read_message(struct ceph_connection *con)
1621{
1622 dout("prepare_read_message %p\n", con);
1623 BUG_ON(con->in_msg != NULL);
1624 con->in_base_pos = 0;
1625 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1626 return 0;
1627}
1628
1629
1630static int read_partial(struct ceph_connection *con,
fd51653f 1631 int end, int size, void *object)
31b8006e 1632{
e6cee71f
AE		 1633 while (con->in_base_pos < end) {
1634 int left = end - con->in_base_pos;
31b8006e
SW		 1635 int have = size - left;
1636 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1637 if (ret <= 0)
1638 return ret;
1639 con->in_base_pos += ret;
1640 }
1641 return 1;
1642}
1643
1644
1645/*
1646 * Read all or part of the connect-side handshake on a new connection
1647 */
eed0ef2c 1648static int read_partial_banner(struct ceph_connection *con)
31b8006e 1649{
fd51653f
AE		 1650 int size;
1651 int end;
1652 int ret;
31b8006e 1653
eed0ef2c 1654 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1655
1656 /* peer's banner */
fd51653f
AE		 1657 size = strlen(CEPH_BANNER);
1658 end = size;
1659 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1660 if (ret <= 0)
1661 goto out;
fd51653f
AE		 1662
1663 size = sizeof (con->actual_peer_addr);
1664 end += size;
1665 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1666 if (ret <= 0)
1667 goto out;
fd51653f
AE		 1668
1669 size = sizeof (con->peer_addr_for_me);
1670 end += size;
1671 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1672 if (ret <= 0)
1673 goto out;
fd51653f 1674
eed0ef2c
SW		 1675out:
1676 return ret;
1677}
1678
1679static int read_partial_connect(struct ceph_connection *con)
1680{
fd51653f
AE		 1681 int size;
1682 int end;
1683 int ret;
eed0ef2c
SW		 1684
1685 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1686
fd51653f
AE		 1687 size = sizeof (con->in_reply);
1688 end = size;
1689 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1690 if (ret <= 0)
1691 goto out;
fd51653f
AE		 1692
1693 size = le32_to_cpu(con->in_reply.authorizer_len);
1694 end += size;
1695 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1696 if (ret <= 0)
1697 goto out;
31b8006e 1698
4e7a5dcd
SW		 1699 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1700 con, (int)con->in_reply.tag,
1701 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1702 le32_to_cpu(con->in_reply.global_seq));
1703out:
1704 return ret;
eed0ef2c 1705
31b8006e
SW		 1706}
1707
1708/*
1709 * Verify the hello banner looks okay.
1710 */
1711static int verify_hello(struct ceph_connection *con)
1712{
1713 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1714 pr_err("connect to %s got bad banner\n",
3d14c5d2 1715 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1716 con->error_msg = "protocol error, bad banner";
1717 return -1;
1718 }
1719 return 0;
1720}
1721
1722static bool addr_is_blank(struct sockaddr_storage *ss)
1723{
1724 switch (ss->ss_family) {
1725 case AF_INET:
1726 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1727 case AF_INET6:
1728 return
1729 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1730 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1731 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1732 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1733 }
1734 return false;
1735}
1736
1737static int addr_port(struct sockaddr_storage *ss)
1738{
1739 switch (ss->ss_family) {
1740 case AF_INET:
f28bcfbe 1741 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1742 case AF_INET6:
f28bcfbe 1743 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1744 }
1745 return 0;
1746}
1747
1748static void addr_set_port(struct sockaddr_storage *ss, int p)
1749{
1750 switch (ss->ss_family) {
1751 case AF_INET:
1752 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1753 break;
31b8006e
SW		 1754 case AF_INET6:
1755 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1756 break;
31b8006e
SW		 1757 }
1758}
1759
ee3b56f2
NW		 1760/*
1761 * Unlike other *_pton function semantics, zero indicates success.
1762 */
1763static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1764 char delim, const char **ipend)
1765{
99f0f3b2
AE		 1766 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1767 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1768
1769 memset(ss, 0, sizeof(*ss));
1770
1771 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1772 ss->ss_family = AF_INET;
1773 return 0;
1774 }
1775
1776 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1777 ss->ss_family = AF_INET6;
1778 return 0;
1779 }
1780
1781 return -EINVAL;
1782}
1783
1784/*
1785 * Extract hostname string and resolve using kernel DNS facility.
1786 */
1787#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1788static int ceph_dns_resolve_name(const char *name, size_t namelen,
1789 struct sockaddr_storage *ss, char delim, const char **ipend)
1790{
1791 const char *end, *delim_p;
1792 char *colon_p, *ip_addr = NULL;
1793 int ip_len, ret;
1794
1795 /*
1796 * The end of the hostname occurs immediately preceding the delimiter or
1797 * the port marker (':') where the delimiter takes precedence.
1798 */
1799 delim_p = memchr(name, delim, namelen);
1800 colon_p = memchr(name, ':', namelen);
1801
1802 if (delim_p && colon_p)
1803 end = delim_p < colon_p ? delim_p : colon_p;
1804 else if (!delim_p && colon_p)
1805 end = colon_p;
1806 else {
1807 end = delim_p;
1808 if (!end) /* case: hostname:/ */
1809 end = name + namelen;
1810 }
1811
1812 if (end <= name)
1813 return -EINVAL;
1814
1815 /* do dns_resolve upcall */
1816 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1817 if (ip_len > 0)
1818 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1819 else
1820 ret = -ESRCH;
1821
1822 kfree(ip_addr);
1823
1824 *ipend = end;
1825
1826 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1827 ret, ret ? "failed" : ceph_pr_addr(ss));
1828
1829 return ret;
1830}
1831#else
1832static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1833 struct sockaddr_storage *ss, char delim, const char **ipend)
1834{
1835 return -EINVAL;
1836}
1837#endif
1838
1839/*
1840 * Parse a server name (IP or hostname). If a valid IP address is not found
1841 * then try to extract a hostname to resolve using userspace DNS upcall.
1842 */
1843static int ceph_parse_server_name(const char *name, size_t namelen,
1844 struct sockaddr_storage *ss, char delim, const char **ipend)
1845{
1846 int ret;
1847
1848 ret = ceph_pton(name, namelen, ss, delim, ipend);
1849 if (ret)
1850 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1851
1852 return ret;
1853}
1854
31b8006e
SW		 1855/*
1856 * Parse an ip[:port] list into an addr array. Use the default
1857 * monitor port if a port isn't specified.
1858 */
1859int ceph_parse_ips(const char *c, const char *end,
1860 struct ceph_entity_addr *addr,
1861 int max_count, int *count)
1862{
ee3b56f2 1863 int i, ret = -EINVAL;
31b8006e
SW		 1864 const char *p = c;
1865
1866 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1867 for (i = 0; i < max_count; i++) {
1868 const char *ipend;
1869 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1870 int port;
39139f64
SW		 1871 char delim = ',';
1872
1873 if (*p == '[') {
1874 delim = ']';
1875 p++;
1876 }
31b8006e 1877
ee3b56f2
NW		 1878 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1879 if (ret)
31b8006e 1880 goto bad;
ee3b56f2
NW		 1881 ret = -EINVAL;
1882
31b8006e
SW		 1883 p = ipend;
1884
39139f64
SW		 1885 if (delim == ']') {
1886 if (*p != ']') {
1887 dout("missing matching ']'\n");
1888 goto bad;
1889 }
1890 p++;
1891 }
1892
31b8006e
SW		 1893 /* port? */
1894 if (p < end && *p == ':') {
1895 port = 0;
1896 p++;
1897 while (p < end && *p >= '0' && *p <= '9') {
1898 port = (port * 10) + (*p - '0');
1899 p++;
1900 }
1901 if (port > 65535 || port == 0)
1902 goto bad;
1903 } else {
1904 port = CEPH_MON_PORT;
1905 }
1906
1907 addr_set_port(ss, port);
1908
3d14c5d2 1909 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1910
1911 if (p == end)
1912 break;
1913 if (*p != ',')
1914 goto bad;
1915 p++;
1916 }
1917
1918 if (p != end)
1919 goto bad;
1920
1921 if (count)
1922 *count = i + 1;
1923 return 0;
1924
1925bad:
39139f64 1926 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1927 return ret;
31b8006e 1928}
3d14c5d2 1929EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1930
eed0ef2c 1931static int process_banner(struct ceph_connection *con)
31b8006e 1932{
eed0ef2c 1933 dout("process_banner on %p\n", con);
31b8006e
SW		 1934
1935 if (verify_hello(con) < 0)
1936 return -1;
1937
63f2d211
SW		 1938 ceph_decode_addr(&con->actual_peer_addr);
1939 ceph_decode_addr(&con->peer_addr_for_me);
1940
31b8006e
SW		 1941 /*
1942 * Make sure the other end is who we wanted. note that the other
1943 * end may not yet know their ip address, so if it's 0.0.0.0, give
1944 * them the benefit of the doubt.
1945 */
103e2d3a
SW		 1946 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1947 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1948 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1949 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
cd84db6e 1950 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
3d14c5d2 1951 ceph_pr_addr(&con->peer_addr.in_addr),
cd84db6e 1952 (int)le32_to_cpu(con->peer_addr.nonce),
3d14c5d2 1953 ceph_pr_addr(&con->actual_peer_addr.in_addr),
cd84db6e 1954 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1955 con->error_msg = "wrong peer at address";
31b8006e
SW		 1956 return -1;
1957 }
1958
1959 /*
1960 * did we learn our address?
1961 */
1962 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1963 int port = addr_port(&con->msgr->inst.addr.in_addr);
1964
1965 memcpy(&con->msgr->inst.addr.in_addr,
1966 &con->peer_addr_for_me.in_addr,
1967 sizeof(con->peer_addr_for_me.in_addr));
1968 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1969 encode_my_addr(con->msgr);
eed0ef2c 1970 dout("process_banner learned my addr is %s\n",
3d14c5d2 1971 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1972 }
1973
eed0ef2c
SW		 1974 return 0;
1975}
1976
1977static int process_connect(struct ceph_connection *con)
1978{
3d14c5d2
YS		 1979 u64 sup_feat = con->msgr->supported_features;
1980 u64 req_feat = con->msgr->required_features;
04a419f9 1981 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 1982 int ret;
04a419f9 1983
eed0ef2c
SW		 1984 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1985
31b8006e 1986 switch (con->in_reply.tag) {
04a419f9
SW		 1987 case CEPH_MSGR_TAG_FEATURES:
1988 pr_err("%s%lld %s feature set mismatch,"
1989 " my %llx < server's %llx, missing %llx\n",
1990 ENTITY_NAME(con->peer_name),
3d14c5d2 1991 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1992 sup_feat, server_feat, server_feat & ~sup_feat);
1993 con->error_msg = "missing required protocol features";
0fa6ebc6 1994 reset_connection(con);
04a419f9
SW		 1995 return -1;
1996
31b8006e 1997 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 1998 pr_err("%s%lld %s protocol version mismatch,"
1999 " my %d != server's %d\n",
2000 ENTITY_NAME(con->peer_name),
3d14c5d2 2001 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2002 le32_to_cpu(con->out_connect.protocol_version),
2003 le32_to_cpu(con->in_reply.protocol_version));
2004 con->error_msg = "protocol version mismatch";
0fa6ebc6 2005 reset_connection(con);
31b8006e
SW		 2006 return -1;
2007
4e7a5dcd
SW		 2008 case CEPH_MSGR_TAG_BADAUTHORIZER:
2009 con->auth_retry++;
2010 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2011 con->auth_retry);
2012 if (con->auth_retry == 2) {
2013 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2014 return -1;
2015 }
6d4221b5 2016 con_out_kvec_reset(con);
e825a66d 2017 ret = prepare_write_connect(con);
0da5d703
SW		 2018 if (ret < 0)
2019 return ret;
63733a0f 2020 prepare_read_connect(con);
4e7a5dcd 2021 break;
31b8006e
SW		 2022
2023 case CEPH_MSGR_TAG_RESETSESSION:
2024 /*
2025 * If we connected with a large connect_seq but the peer
2026 * has no record of a session with us (no connection, or
2027 * connect_seq == 0), they will send RESETSESION to indicate
2028 * that they must have reset their session, and may have
2029 * dropped messages.
2030 */
2031 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2032 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2033 pr_err("%s%lld %s connection reset\n",
2034 ENTITY_NAME(con->peer_name),
3d14c5d2 2035 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2036 reset_connection(con);
6d4221b5 2037 con_out_kvec_reset(con);
5a0f8fdd
AE		 2038 ret = prepare_write_connect(con);
2039 if (ret < 0)
2040 return ret;
31b8006e
SW		 2041 prepare_read_connect(con);
2042
2043 /* Tell ceph about it. */
ec302645 2044 mutex_unlock(&con->mutex);
31b8006e
SW		 2045 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2046 if (con->ops->peer_reset)
2047 con->ops->peer_reset(con);
ec302645 2048 mutex_lock(&con->mutex);
8dacc7da 2049 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2050 return -EAGAIN;
31b8006e
SW		 2051 break;
2052
2053 case CEPH_MSGR_TAG_RETRY_SESSION:
2054 /*
2055 * If we sent a smaller connect_seq than the peer has, try
2056 * again with a larger value.
2057 */
5bdca4e0 2058 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2059 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2060 le32_to_cpu(con->in_reply.connect_seq));
2061 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2062 con_out_kvec_reset(con);
5a0f8fdd
AE		 2063 ret = prepare_write_connect(con);
2064 if (ret < 0)
2065 return ret;
31b8006e
SW		 2066 prepare_read_connect(con);
2067 break;
2068
2069 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2070 /*
2071 * If we sent a smaller global_seq than the peer has, try
2072 * again with a larger value.
2073 */
eed0ef2c 2074 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2075 con->peer_global_seq,
5bdca4e0 2076 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2077 get_global_seq(con->msgr,
5bdca4e0 2078 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2079 con_out_kvec_reset(con);
5a0f8fdd
AE		 2080 ret = prepare_write_connect(con);
2081 if (ret < 0)
2082 return ret;
31b8006e
SW		 2083 prepare_read_connect(con);
2084 break;
2085
3a23083b 2086 case CEPH_MSGR_TAG_SEQ:
31b8006e 2087 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2088 if (req_feat & ~server_feat) {
2089 pr_err("%s%lld %s protocol feature mismatch,"
2090 " my required %llx > server's %llx, need %llx\n",
2091 ENTITY_NAME(con->peer_name),
3d14c5d2 2092 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2093 req_feat, server_feat, req_feat & ~server_feat);
2094 con->error_msg = "missing required protocol features";
0fa6ebc6 2095 reset_connection(con);
04a419f9
SW		 2096 return -1;
2097 }
8dacc7da 2098
122070a2 2099 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2100 con->state = CON_STATE_OPEN;
20e55c4c 2101 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2102 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2103 con->connect_seq++;
aba558e2 2104 con->peer_features = server_feat;
31b8006e
SW		 2105 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2106 con->peer_global_seq,
2107 le32_to_cpu(con->in_reply.connect_seq),
2108 con->connect_seq);
2109 WARN_ON(con->connect_seq !=
2110 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2111
2112 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2113 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2114
85effe18 2115 con->delay = 0; /* reset backoff memory */
92ac41d0 2116
3a23083b
SW		 2117 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2118 prepare_write_seq(con);
2119 prepare_read_seq(con);
2120 } else {
2121 prepare_read_tag(con);
2122 }
31b8006e
SW		 2123 break;
2124
2125 case CEPH_MSGR_TAG_WAIT:
2126 /*
2127 * If there is a connection race (we are opening
2128 * connections to each other), one of us may just have
2129 * to WAIT. This shouldn't happen if we are the
2130 * client.
2131 */
04177882
SW		 2132 pr_err("process_connect got WAIT as client\n");
2133 con->error_msg = "protocol error, got WAIT as client";
2134 return -1;
31b8006e
SW		 2135
2136 default:
2137 pr_err("connect protocol error, will retry\n");
2138 con->error_msg = "protocol error, garbage tag during connect";
2139 return -1;
2140 }
2141 return 0;
2142}
2143
2144
2145/*
2146 * read (part of) an ack
2147 */
2148static int read_partial_ack(struct ceph_connection *con)
2149{
fd51653f
AE		 2150 int size = sizeof (con->in_temp_ack);
2151 int end = size;
31b8006e 2152
fd51653f 2153 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2154}
2155
31b8006e
SW		 2156/*
2157 * We can finally discard anything that's been acked.
2158 */
2159static void process_ack(struct ceph_connection *con)
2160{
2161 struct ceph_msg *m;
2162 u64 ack = le64_to_cpu(con->in_temp_ack);
2163 u64 seq;
2164
31b8006e
SW		 2165 while (!list_empty(&con->out_sent)) {
2166 m = list_first_entry(&con->out_sent, struct ceph_msg,
2167 list_head);
2168 seq = le64_to_cpu(m->hdr.seq);
2169 if (seq > ack)
2170 break;
2171 dout("got ack for seq %llu type %d at %p\n", seq,
2172 le16_to_cpu(m->hdr.type), m);
4cf9d544 2173 m->ack_stamp = jiffies;
31b8006e
SW		 2174 ceph_msg_remove(m);
2175 }
31b8006e
SW		 2176 prepare_read_tag(con);
2177}
2178
2179
2450418c 2180static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2181 struct kvec *section,
2182 unsigned int sec_len, u32 *crc)
2450418c 2183{
68b4476b 2184 int ret, left;
2450418c
YS		 2185
2186 BUG_ON(!section);
2187
2188 while (section->iov_len < sec_len) {
2189 BUG_ON(section->iov_base == NULL);
2190 left = sec_len - section->iov_len;
2191 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2192 section->iov_len, left);
2193 if (ret <= 0)
2194 return ret;
2195 section->iov_len += ret;
2450418c 2196 }
fe3ad593
AE		 2197 if (section->iov_len == sec_len)
2198 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2199
2450418c
YS		 2200 return 1;
2201}
31b8006e 2202
4740a623 2203static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
68b4476b
YS		 2204
2205static int read_partial_message_pages(struct ceph_connection *con,
95c96174 2206 unsigned int data_len, bool do_datacrc)
68b4476b 2207{
28a89dde 2208 struct ceph_msg *msg = con->in_msg;
bae6acd9 2209 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
28a89dde 2210 struct page **pages;
e788182f 2211 struct page *page;
afb3d90e
AE		 2212 size_t page_offset;
2213 size_t length;
2214 unsigned int left;
68b4476b 2215 int ret;
68b4476b 2216
68b4476b 2217 /* (page) data */
28a89dde 2218 pages = msg->p.pages;
68b4476b 2219 BUG_ON(pages == NULL);
bae6acd9 2220 page = pages[msg_pos->page];
afb3d90e
AE		 2221 page_offset = msg_pos->page_pos;
2222 BUG_ON(msg_pos->data_pos >= data_len);
2223 left = data_len - msg_pos->data_pos;
2224 BUG_ON(page_offset >= PAGE_SIZE);
2225 length = min_t(unsigned int, PAGE_SIZE - page_offset, left);
2226
2227 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
68b4476b
YS		 2228 if (ret <= 0)
2229 return ret;
e788182f 2230
35b62808
AE		 2231 if (do_datacrc)
2232 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2233 page_offset, ret);
afb3d90e
AE		 2234
2235 in_msg_pos_next(con, length, ret);
68b4476b
YS		 2236
2237 return ret;
2238}
2239
2240#ifdef CONFIG_BLOCK
2241static int read_partial_message_bio(struct ceph_connection *con,
95c96174 2242 unsigned int data_len, bool do_datacrc)
68b4476b 2243{
b3d56fab 2244 struct ceph_msg *msg = con->in_msg;
bae6acd9 2245 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
b3d56fab 2246 struct bio_vec *bv;
e788182f 2247 struct page *page;
afb3d90e
AE		 2248 size_t page_offset;
2249 size_t length;
2250 unsigned int left;
2251 int ret;
68b4476b 2252
b3d56fab 2253 BUG_ON(!msg);
f9e15777
AE		 2254 BUG_ON(!msg->b.bio_iter);
2255 bv = bio_iovec_idx(msg->b.bio_iter, msg->b.bio_seg);
e788182f 2256 page = bv->bv_page;
afb3d90e
AE		 2257 page_offset = bv->bv_offset + msg_pos->page_pos;
2258 BUG_ON(msg_pos->data_pos >= data_len);
2259 left = data_len - msg_pos->data_pos;
2260 BUG_ON(msg_pos->page_pos >= bv->bv_len);
2261 length = min_t(unsigned int, bv->bv_len - msg_pos->page_pos, left);
68b4476b 2262
afb3d90e 2263 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
68b4476b
YS		 2264 if (ret <= 0)
2265 return ret;
e788182f 2266
35b62808
AE		 2267 if (do_datacrc)
2268 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2269 page_offset, ret);
afb3d90e
AE		 2270
2271 in_msg_pos_next(con, length, ret);
68b4476b
YS		 2272
2273 return ret;
2274}
2275#endif
2276
34d2d200
AE		 2277static int read_partial_msg_data(struct ceph_connection *con)
2278{
2279 struct ceph_msg *msg = con->in_msg;
2280 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2281 const bool do_datacrc = !con->msgr->nocrc;
2282 unsigned int data_len;
2283 int ret;
2284
2285 BUG_ON(!msg);
2286
2287 data_len = le32_to_cpu(con->in_hdr.data_len);
2288 while (msg_pos->data_pos < data_len) {
97fb1c7f 2289 if (ceph_msg_has_pages(msg)) {
28a89dde
AE		 2290 ret = read_partial_message_pages(con, data_len,
2291 do_datacrc);
34d2d200
AE		 2292 if (ret <= 0)
2293 return ret;
2294#ifdef CONFIG_BLOCK
97fb1c7f 2295 } else if (ceph_msg_has_bio(msg)) {
34d2d200
AE		 2296 ret = read_partial_message_bio(con,
2297 data_len, do_datacrc);
2298 if (ret <= 0)
2299 return ret;
2300#endif
2301 } else {
2302 BUG_ON(1);
2303 }
2304 }
2305
2306 return 1; /* must return > 0 to indicate success */
2307}
2308
31b8006e
SW		 2309/*
2310 * read (part of) a message.
2311 */
2312static int read_partial_message(struct ceph_connection *con)
2313{
2314 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2315 int size;
2316 int end;
31b8006e 2317 int ret;
95c96174 2318 unsigned int front_len, middle_len, data_len;
37675b0f 2319 bool do_datacrc = !con->msgr->nocrc;
ae18756b 2320 u64 seq;
fe3ad593 2321 u32 crc;
31b8006e
SW		 2322
2323 dout("read_partial_message con %p msg %p\n", con, m);
2324
2325 /* header */
fd51653f
AE		 2326 size = sizeof (con->in_hdr);
2327 end = size;
2328 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2329 if (ret <= 0)
2330 return ret;
fe3ad593
AE		 2331
2332 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2333 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2334 pr_err("read_partial_message bad hdr "
2335 " crc %u != expected %u\n",
2336 crc, con->in_hdr.crc);
2337 return -EBADMSG;
2338 }
2339
31b8006e
SW		 2340 front_len = le32_to_cpu(con->in_hdr.front_len);
2341 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2342 return -EIO;
2343 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2344 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2345 return -EIO;
2346 data_len = le32_to_cpu(con->in_hdr.data_len);
2347 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2348 return -EIO;
2349
ae18756b
SW		 2350 /* verify seq# */
2351 seq = le64_to_cpu(con->in_hdr.seq);
2352 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2353 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2354 ENTITY_NAME(con->peer_name),
3d14c5d2 2355 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2356 seq, con->in_seq + 1);
2357 con->in_base_pos = -front_len - middle_len - data_len -
2358 sizeof(m->footer);
2359 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2360 return 0;
2361 } else if ((s64)seq - (s64)con->in_seq > 1) {
2362 pr_err("read_partial_message bad seq %lld expected %lld\n",
2363 seq, con->in_seq + 1);
2364 con->error_msg = "bad message sequence # for incoming message";
2365 return -EBADMSG;
2366 }
2367
31b8006e
SW		 2368 /* allocate message? */
2369 if (!con->in_msg) {
4740a623
SW		 2370 int skip = 0;
2371
31b8006e 2372 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2373 front_len, data_len);
4740a623
SW		 2374 ret = ceph_con_in_msg_alloc(con, &skip);
2375 if (ret < 0)
2376 return ret;
2450418c 2377 if (skip) {
31b8006e 2378 /* skip this message */
a79832f2 2379 dout("alloc_msg said skip message\n");
ae32be31 2380 BUG_ON(con->in_msg);
31b8006e
SW		 2381 con->in_base_pos = -front_len - middle_len - data_len -
2382 sizeof(m->footer);
2383 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2384 con->in_seq++;
31b8006e
SW		 2385 return 0;
2386 }
38941f80 2387
4740a623 2388 BUG_ON(!con->in_msg);
38941f80 2389 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2390 m = con->in_msg;
2391 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2392 if (m->middle)
2393 m->middle->vec.iov_len = 0;
9d7f0f13 2394
78625051 2395 /* prepare for data payload, if any */
a4107026 2396
78625051
AE		 2397 if (data_len)
2398 prepare_message_data(con->in_msg, &con->in_msg_pos);
31b8006e
SW		 2399 }
2400
2401 /* front */
2450418c
YS		 2402 ret = read_partial_message_section(con, &m->front, front_len,
2403 &con->in_front_crc);
2404 if (ret <= 0)
2405 return ret;
31b8006e
SW		 2406
2407 /* middle */
2450418c 2408 if (m->middle) {
213c99ee
SW		 2409 ret = read_partial_message_section(con, &m->middle->vec,
2410 middle_len,
2450418c 2411 &con->in_middle_crc);
31b8006e
SW		 2412 if (ret <= 0)
2413 return ret;
31b8006e
SW		 2414 }
2415
2416 /* (page) data */
34d2d200
AE		 2417 if (data_len) {
2418 ret = read_partial_msg_data(con);
2419 if (ret <= 0)
2420 return ret;
31b8006e
SW		 2421 }
2422
31b8006e 2423 /* footer */
fd51653f
AE		 2424 size = sizeof (m->footer);
2425 end += size;
2426 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2427 if (ret <= 0)
2428 return ret;
2429
31b8006e
SW		 2430 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2431 m, front_len, m->footer.front_crc, middle_len,
2432 m->footer.middle_crc, data_len, m->footer.data_crc);
2433
2434 /* crc ok? */
2435 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2436 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2437 m, con->in_front_crc, m->footer.front_crc);
2438 return -EBADMSG;
2439 }
2440 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2441 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2442 m, con->in_middle_crc, m->footer.middle_crc);
2443 return -EBADMSG;
2444 }
bca064d2 2445 if (do_datacrc &&
31b8006e
SW		 2446 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2447 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2448 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2449 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2450 return -EBADMSG;
2451 }
2452
2453 return 1; /* done! */
2454}
2455
2456/*
2457 * Process message. This happens in the worker thread. The callback should
2458 * be careful not to do anything that waits on other incoming messages or it
2459 * may deadlock.
2460 */
2461static void process_message(struct ceph_connection *con)
2462{
5e095e8b 2463 struct ceph_msg *msg;
31b8006e 2464
38941f80
AE		 2465 BUG_ON(con->in_msg->con != con);
2466 con->in_msg->con = NULL;
5e095e8b 2467 msg = con->in_msg;
31b8006e 2468 con->in_msg = NULL;
36eb71aa 2469 con->ops->put(con);
31b8006e
SW		 2470
2471 /* if first message, set peer_name */
2472 if (con->peer_name.type == 0)
dbad185d 2473 con->peer_name = msg->hdr.src;
31b8006e 2474
31b8006e 2475 con->in_seq++;
ec302645 2476 mutex_unlock(&con->mutex);
31b8006e
SW		 2477
2478 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2479 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2480 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2481 le16_to_cpu(msg->hdr.type),
2482 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2483 le32_to_cpu(msg->hdr.front_len),
2484 le32_to_cpu(msg->hdr.data_len),
2485 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2486 con->ops->dispatch(con, msg);
ec302645
SW		 2487
2488 mutex_lock(&con->mutex);
31b8006e
SW		 2489}
2490
2491
2492/*
2493 * Write something to the socket. Called in a worker thread when the
2494 * socket appears to be writeable and we have something ready to send.
2495 */
2496static int try_write(struct ceph_connection *con)
2497{
31b8006e
SW		 2498 int ret = 1;
2499
d59315ca 2500 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2501
31b8006e
SW		 2502more:
2503 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2504
2505 /* open the socket first? */
8dacc7da
SW		 2506 if (con->state == CON_STATE_PREOPEN) {
2507 BUG_ON(con->sock);
2508 con->state = CON_STATE_CONNECTING;
a5988c49 2509
e2200423 2510 con_out_kvec_reset(con);
e825a66d 2511 prepare_write_banner(con);
eed0ef2c 2512 prepare_read_banner(con);
31b8006e 2513
cf3e5c40 2514 BUG_ON(con->in_msg);
31b8006e
SW		 2515 con->in_tag = CEPH_MSGR_TAG_READY;
2516 dout("try_write initiating connect on %p new state %lu\n",
2517 con, con->state);
41617d0c
AE		 2518 ret = ceph_tcp_connect(con);
2519 if (ret < 0) {
31b8006e 2520 con->error_msg = "connect error";
31b8006e
SW		 2521 goto out;
2522 }
2523 }
2524
2525more_kvec:
2526 /* kvec data queued? */
2527 if (con->out_skip) {
2528 ret = write_partial_skip(con);
2529 if (ret <= 0)
42961d23 2530 goto out;
31b8006e
SW		 2531 }
2532 if (con->out_kvec_left) {
2533 ret = write_partial_kvec(con);
2534 if (ret <= 0)
42961d23 2535 goto out;
31b8006e
SW		 2536 }
2537
2538 /* msg pages? */
2539 if (con->out_msg) {
c86a2930
SW		 2540 if (con->out_msg_done) {
2541 ceph_msg_put(con->out_msg);
2542 con->out_msg = NULL; /* we're done with this one */
2543 goto do_next;
2544 }
2545
34d2d200 2546 ret = write_partial_message_data(con);
31b8006e
SW		 2547 if (ret == 1)
2548 goto more_kvec; /* we need to send the footer, too! */
2549 if (ret == 0)
42961d23 2550 goto out;
31b8006e 2551 if (ret < 0) {
34d2d200 2552 dout("try_write write_partial_message_data err %d\n",
31b8006e 2553 ret);
42961d23 2554 goto out;
31b8006e
SW		 2555 }
2556 }
2557
c86a2930 2558do_next:
8dacc7da 2559 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2560 /* is anything else pending? */
2561 if (!list_empty(&con->out_queue)) {
2562 prepare_write_message(con);
2563 goto more;
2564 }
2565 if (con->in_seq > con->in_seq_acked) {
2566 prepare_write_ack(con);
2567 goto more;
2568 }
c9ffc77a 2569 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2570 prepare_write_keepalive(con);
2571 goto more;
2572 }
2573 }
2574
2575 /* Nothing to do! */
c9ffc77a 2576 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2577 dout("try_write nothing else to write.\n");
31b8006e
SW		 2578 ret = 0;
2579out:
42961d23 2580 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2581 return ret;
2582}
2583
2584
2585
2586/*
2587 * Read what we can from the socket.
2588 */
2589static int try_read(struct ceph_connection *con)
2590{
31b8006e
SW		 2591 int ret = -1;
2592
8dacc7da
SW		 2593more:
2594 dout("try_read start on %p state %lu\n", con, con->state);
2595 if (con->state != CON_STATE_CONNECTING &&
2596 con->state != CON_STATE_NEGOTIATING &&
2597 con->state != CON_STATE_OPEN)
31b8006e
SW		 2598 return 0;
2599
8dacc7da 2600 BUG_ON(!con->sock);
ec302645 2601
31b8006e
SW		 2602 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2603 con->in_base_pos);
0da5d703 2604
8dacc7da 2605 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2606 dout("try_read connecting\n");
2607 ret = read_partial_banner(con);
2608 if (ret <= 0)
ab166d5a 2609 goto out;
7593af92
AE		 2610 ret = process_banner(con);
2611 if (ret < 0)
2612 goto out;
2613
8dacc7da 2614 con->state = CON_STATE_NEGOTIATING;
7593af92 2615
6d4221b5
JS		 2616 /*
2617 * Received banner is good, exchange connection info.
2618 * Do not reset out_kvec, as sending our banner raced
2619 * with receiving peer banner after connect completed.
2620 */
7593af92
AE		 2621 ret = prepare_write_connect(con);
2622 if (ret < 0)
2623 goto out;
2624 prepare_read_connect(con);
2625
2626 /* Send connection info before awaiting response */
0da5d703
SW		 2627 goto out;
2628 }
2629
8dacc7da 2630 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2631 dout("try_read negotiating\n");
31b8006e
SW		 2632 ret = read_partial_connect(con);
2633 if (ret <= 0)
31b8006e 2634 goto out;
98bdb0aa
SW		 2635 ret = process_connect(con);
2636 if (ret < 0)
2637 goto out;
31b8006e
SW		 2638 goto more;
2639 }
2640
122070a2 2641 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2642
31b8006e
SW		 2643 if (con->in_base_pos < 0) {
2644 /*
2645 * skipping + discarding content.
2646 *
2647 * FIXME: there must be a better way to do this!
2648 */
84495f49
AE		 2649 static char buf[SKIP_BUF_SIZE];
2650 int skip = min((int) sizeof (buf), -con->in_base_pos);
2651
31b8006e
SW		 2652 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2653 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2654 if (ret <= 0)
98bdb0aa 2655 goto out;
31b8006e
SW		 2656 con->in_base_pos += ret;
2657 if (con->in_base_pos)
2658 goto more;
2659 }
2660 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2661 /*
2662 * what's next?
2663 */
2664 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2665 if (ret <= 0)
98bdb0aa 2666 goto out;
31b8006e
SW		 2667 dout("try_read got tag %d\n", (int)con->in_tag);
2668 switch (con->in_tag) {
2669 case CEPH_MSGR_TAG_MSG:
2670 prepare_read_message(con);
2671 break;
2672 case CEPH_MSGR_TAG_ACK:
2673 prepare_read_ack(con);
2674 break;
2675 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2676 con_close_socket(con);
2677 con->state = CON_STATE_CLOSED;
98bdb0aa 2678 goto out;
31b8006e
SW		 2679 default:
2680 goto bad_tag;
2681 }
2682 }
2683 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2684 ret = read_partial_message(con);
2685 if (ret <= 0) {
2686 switch (ret) {
2687 case -EBADMSG:
2688 con->error_msg = "bad crc";
2689 ret = -EIO;
98bdb0aa 2690 break;
31b8006e
SW		 2691 case -EIO:
2692 con->error_msg = "io error";
98bdb0aa 2693 break;
31b8006e 2694 }
98bdb0aa 2695 goto out;
31b8006e
SW		 2696 }
2697 if (con->in_tag == CEPH_MSGR_TAG_READY)
2698 goto more;
2699 process_message(con);
7b862e07
SW		 2700 if (con->state == CON_STATE_OPEN)
2701 prepare_read_tag(con);
31b8006e
SW		 2702 goto more;
2703 }
3a23083b
SW		 2704 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2705 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2706 /*
2707 * the final handshake seq exchange is semantically
2708 * equivalent to an ACK
2709 */
31b8006e
SW		 2710 ret = read_partial_ack(con);
2711 if (ret <= 0)
98bdb0aa 2712 goto out;
31b8006e
SW		 2713 process_ack(con);
2714 goto more;
2715 }
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
AE		 2737 dout("%s %p ref count 0\n", __func__, con);
2738
2739 return -ENOENT;
31b8006e
SW		 2740 }
2741
802c6d96
AE		 2742 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2743 dout("%s %p - already queued\n", __func__, con);
31b8006e 2744 con->ops->put(con);
802c6d96
AE		 2745
2746 return -EBUSY;
31b8006e 2747 }
802c6d96
AE		 2748
2749 dout("%s %p %lu\n", __func__, con, delay);
2750
2751 return 0;
2752}
2753
2754static void queue_con(struct ceph_connection *con)
2755{
2756 (void) queue_con_delay(con, 0);
31b8006e
SW		 2757}
2758
7bb21d68
AE		 2759static bool con_sock_closed(struct ceph_connection *con)
2760{
c9ffc77a 2761 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2762 return false;
2763
2764#define CASE(x) \
2765 case CON_STATE_ ## x: \
2766 con->error_msg = "socket closed (con state " #x ")"; \
2767 break;
2768
2769 switch (con->state) {
2770 CASE(CLOSED);
2771 CASE(PREOPEN);
2772 CASE(CONNECTING);
2773 CASE(NEGOTIATING);
2774 CASE(OPEN);
2775 CASE(STANDBY);
2776 default:
2777 pr_warning("%s con %p unrecognized state %lu\n",
2778 __func__, con, con->state);
2779 con->error_msg = "unrecognized con state";
2780 BUG();
2781 break;
2782 }
2783#undef CASE
2784
2785 return true;
2786}
2787
f20a39fd
AE		 2788static bool con_backoff(struct ceph_connection *con)
2789{
2790 int ret;
2791
2792 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2793 return false;
2794
2795 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2796 if (ret) {
2797 dout("%s: con %p FAILED to back off %lu\n", __func__,
2798 con, con->delay);
2799 BUG_ON(ret == -ENOENT);
2800 con_flag_set(con, CON_FLAG_BACKOFF);
2801 }
2802
2803 return true;
2804}
2805
93209264
AE		 2806/* Finish fault handling; con->mutex must *not* be held here */
2807
2808static void con_fault_finish(struct ceph_connection *con)
2809{
2810 /*
2811 * in case we faulted due to authentication, invalidate our
2812 * current tickets so that we can get new ones.
2813 */
2814 if (con->auth_retry && con->ops->invalidate_authorizer) {
2815 dout("calling invalidate_authorizer()\n");
2816 con->ops->invalidate_authorizer(con);
2817 }
2818
2819 if (con->ops->fault)
2820 con->ops->fault(con);
2821}
2822
31b8006e
SW		 2823/*
2824 * Do some work on a connection. Drop a connection ref when we're done.
2825 */
2826static void con_work(struct work_struct *work)
2827{
2828 struct ceph_connection *con = container_of(work, struct ceph_connection,
2829 work.work);
49659416 2830 bool fault;
31b8006e 2831
9dd4658d 2832 mutex_lock(&con->mutex);
49659416
AE		 2833 while (true) {
2834 int ret;
31b8006e 2835
49659416
AE		 2836 if ((fault = con_sock_closed(con))) {
2837 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2838 break;
2839 }
2840 if (con_backoff(con)) {
2841 dout("%s: con %p BACKOFF\n", __func__, con);
2842 break;
2843 }
2844 if (con->state == CON_STATE_STANDBY) {
2845 dout("%s: con %p STANDBY\n", __func__, con);
2846 break;
2847 }
2848 if (con->state == CON_STATE_CLOSED) {
2849 dout("%s: con %p CLOSED\n", __func__, con);
2850 BUG_ON(con->sock);
2851 break;
2852 }
2853 if (con->state == CON_STATE_PREOPEN) {
2854 dout("%s: con %p PREOPEN\n", __func__, con);
2855 BUG_ON(con->sock);
2856 }
0da5d703 2857
49659416
AE		 2858 ret = try_read(con);
2859 if (ret < 0) {
2860 if (ret == -EAGAIN)
2861 continue;
2862 con->error_msg = "socket error on read";
2863 fault = true;
2864 break;
2865 }
2866
2867 ret = try_write(con);
2868 if (ret < 0) {
2869 if (ret == -EAGAIN)
2870 continue;
2871 con->error_msg = "socket error on write";
2872 fault = true;
2873 }
2874
2875 break; /* If we make it to here, we're done */
3a140a0d 2876 }
b6e7b6a1
AE		 2877 if (fault)
2878 con_fault(con);
9dd4658d 2879 mutex_unlock(&con->mutex);
0da5d703 2880
b6e7b6a1
AE		 2881 if (fault)
2882 con_fault_finish(con);
2883
2884 con->ops->put(con);
31b8006e
SW		 2885}
2886
31b8006e
SW		 2887/*
2888 * Generic error/fault handler. A retry mechanism is used with
2889 * exponential backoff
2890 */
93209264 2891static void con_fault(struct ceph_connection *con)
31b8006e 2892{
28362986 2893 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
3d14c5d2 2894 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2895 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2896 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2897
122070a2 2898 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2899 con->state != CON_STATE_NEGOTIATING &&
2900 con->state != CON_STATE_OPEN);
ec302645 2901
31b8006e 2902 con_close_socket(con);
5e095e8b 2903
c9ffc77a 2904 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2905 dout("fault on LOSSYTX channel, marking CLOSED\n");
2906 con->state = CON_STATE_CLOSED;
93209264 2907 return;
3b5ede07
SW		 2908 }
2909
5e095e8b 2910 if (con->in_msg) {
38941f80
AE		 2911 BUG_ON(con->in_msg->con != con);
2912 con->in_msg->con = NULL;
5e095e8b
SW		 2913 ceph_msg_put(con->in_msg);
2914 con->in_msg = NULL;
36eb71aa 2915 con->ops->put(con);
5e095e8b 2916 }
31b8006e 2917
e80a52d1
SW		 2918 /* Requeue anything that hasn't been acked */
2919 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2920
e76661d0
SW		 2921 /* If there are no messages queued or keepalive pending, place
2922 * the connection in a STANDBY state */
2923 if (list_empty(&con->out_queue) &&
c9ffc77a 2924 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2925 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2926 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2927 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2928 } else {
2929 /* retry after a delay. */
8dacc7da 2930 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2931 if (con->delay == 0)
2932 con->delay = BASE_DELAY_INTERVAL;
2933 else if (con->delay < MAX_DELAY_INTERVAL)
2934 con->delay *= 2;
c9ffc77a 2935 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2936 queue_con(con);
31b8006e 2937 }
31b8006e
SW		 2938}
2939
2940
2941
2942/*
15d9882c 2943 * initialize a new messenger instance
31b8006e 2944 */
15d9882c
AE		 2945void ceph_messenger_init(struct ceph_messenger *msgr,
2946 struct ceph_entity_addr *myaddr,
2947 u32 supported_features,
2948 u32 required_features,
2949 bool nocrc)
31b8006e 2950{
3d14c5d2
YS		 2951 msgr->supported_features = supported_features;
2952 msgr->required_features = required_features;
2953
31b8006e
SW		 2954 spin_lock_init(&msgr->global_seq_lock);
2955
31b8006e
SW		 2956 if (myaddr)
2957 msgr->inst.addr = *myaddr;
2958
2959 /* select a random nonce */
ac8839d7 2960 msgr->inst.addr.type = 0;
103e2d3a 2961 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2962 encode_my_addr(msgr);
15d9882c 2963 msgr->nocrc = nocrc;
31b8006e 2964
a2a32584 2965 atomic_set(&msgr->stopping, 0);
31b8006e 2966
15d9882c 2967 dout("%s %p\n", __func__, msgr);
31b8006e 2968}
15d9882c 2969EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2970
e00de341
SW		 2971static void clear_standby(struct ceph_connection *con)
2972{
2973 /* come back from STANDBY? */
8dacc7da 2974 if (con->state == CON_STATE_STANDBY) {
e00de341 2975 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2976 con->state = CON_STATE_PREOPEN;
e00de341 2977 con->connect_seq++;
c9ffc77a
AE		 2978 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2979 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2980 }
2981}
2982
31b8006e
SW		 2983/*
2984 * Queue up an outgoing message on the given connection.
2985 */
2986void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2987{
31b8006e 2988 /* set src+dst */
dbad185d 2989 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2990 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2991 msg->needs_out_seq = true;
2992
ec302645 2993 mutex_lock(&con->mutex);
92ce034b 2994
8dacc7da 2995 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2996 dout("con_send %p closed, dropping %p\n", con, msg);
2997 ceph_msg_put(msg);
2998 mutex_unlock(&con->mutex);
2999 return;
3000 }
3001
38941f80 3002 BUG_ON(msg->con != NULL);
36eb71aa 3003 msg->con = con->ops->get(con);
92ce034b
AE		 3004 BUG_ON(msg->con == NULL);
3005
31b8006e
SW		 3006 BUG_ON(!list_empty(&msg->list_head));
3007 list_add_tail(&msg->list_head, &con->out_queue);
3008 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3009 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3010 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3011 le32_to_cpu(msg->hdr.front_len),
3012 le32_to_cpu(msg->hdr.middle_len),
3013 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 3014
3015 clear_standby(con);
ec302645 3016 mutex_unlock(&con->mutex);
31b8006e
SW		 3017
3018 /* if there wasn't anything waiting to send before, queue
3019 * new work */
c9ffc77a 3020 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3021 queue_con(con);
3022}
3d14c5d2 3023EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3024
3025/*
3026 * Revoke a message that was previously queued for send
3027 */
6740a845 3028void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3029{
6740a845
AE		 3030 struct ceph_connection *con = msg->con;
3031
3032 if (!con)
3033 return; /* Message not in our possession */
3034
ec302645 3035 mutex_lock(&con->mutex);
31b8006e 3036 if (!list_empty(&msg->list_head)) {
38941f80 3037 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3038 list_del_init(&msg->list_head);
38941f80 3039 BUG_ON(msg->con == NULL);
36eb71aa 3040 msg->con->ops->put(msg->con);
38941f80 3041 msg->con = NULL;
31b8006e 3042 msg->hdr.seq = 0;
38941f80 3043
31b8006e 3044 ceph_msg_put(msg);
ed98adad
SW		 3045 }
3046 if (con->out_msg == msg) {
38941f80 3047 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 3048 con->out_msg = NULL;
31b8006e
SW		 3049 if (con->out_kvec_is_msg) {
3050 con->out_skip = con->out_kvec_bytes;
3051 con->out_kvec_is_msg = false;
3052 }
ed98adad 3053 msg->hdr.seq = 0;
92ce034b
AE		 3054
3055 ceph_msg_put(msg);
31b8006e 3056 }
ec302645 3057 mutex_unlock(&con->mutex);
31b8006e
SW		 3058}
3059
350b1c32 3060/*
0d59ab81 3061 * Revoke a message that we may be reading data into
350b1c32 3062 */
8921d114 3063void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3064{
8921d114
AE		 3065 struct ceph_connection *con;
3066
3067 BUG_ON(msg == NULL);
3068 if (!msg->con) {
3069 dout("%s msg %p null con\n", __func__, msg);
3070
3071 return; /* Message not in our possession */
3072 }
3073
3074 con = msg->con;
350b1c32 3075 mutex_lock(&con->mutex);
8921d114 3076 if (con->in_msg == msg) {
95c96174
ED		 3077 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3078 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3079 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3080
3081 /* skip rest of message */
8921d114
AE		 3082 dout("%s %p msg %p revoked\n", __func__, con, msg);
3083 con->in_base_pos = con->in_base_pos -
350b1c32 3084 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3085 front_len -
3086 middle_len -
3087 data_len -
350b1c32 3088 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3089 ceph_msg_put(con->in_msg);
3090 con->in_msg = NULL;
3091 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3092 con->in_seq++;
350b1c32 3093 } else {
8921d114
AE		 3094 dout("%s %p in_msg %p msg %p no-op\n",
3095 __func__, con, con->in_msg, msg);
350b1c32
SW		 3096 }
3097 mutex_unlock(&con->mutex);
3098}
3099
31b8006e
SW		 3100/*
3101 * Queue a keepalive byte to ensure the tcp connection is alive.
3102 */
3103void ceph_con_keepalive(struct ceph_connection *con)
3104{
e00de341 3105 dout("con_keepalive %p\n", con);
00650931 3106 mutex_lock(&con->mutex);
e00de341 3107 clear_standby(con);
00650931 3108 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3109 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3110 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3111 queue_con(con);
3112}
3d14c5d2 3113EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3114
43794509
AE		 3115static void ceph_msg_data_init(struct ceph_msg_data *data)
3116{
3117 data->type = CEPH_MSG_DATA_NONE;
3118}
3119
02afca6c 3120void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3121 size_t length, size_t alignment)
02afca6c 3122{
07aa1558
AE		 3123 BUG_ON(!pages);
3124 BUG_ON(!length);
43794509 3125 BUG_ON(msg->p.type != CEPH_MSG_DATA_NONE);
02afca6c 3126
43794509 3127 msg->p.type = CEPH_MSG_DATA_PAGES;
f9e15777
AE		 3128 msg->p.pages = pages;
3129 msg->p.length = length;
3130 msg->p.alignment = alignment & ~PAGE_MASK;
02afca6c
AE		 3131}
3132EXPORT_SYMBOL(ceph_msg_data_set_pages);
31b8006e 3133
27fa8385
AE		 3134void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3135 struct ceph_pagelist *pagelist)
3136{
07aa1558
AE		 3137 BUG_ON(!pagelist);
3138 BUG_ON(!pagelist->length);
43794509 3139 BUG_ON(msg->l.type != CEPH_MSG_DATA_NONE);
27fa8385 3140
43794509 3141 msg->l.type = CEPH_MSG_DATA_PAGELIST;
f9e15777 3142 msg->l.pagelist = pagelist;
27fa8385
AE		 3143}
3144EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3145
3146void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio)
3147{
07aa1558 3148 BUG_ON(!bio);
43794509 3149 BUG_ON(msg->b.type != CEPH_MSG_DATA_NONE);
27fa8385 3150
43794509 3151 msg->b.type = CEPH_MSG_DATA_BIO;
f9e15777 3152 msg->b.bio = bio;
27fa8385
AE		 3153}
3154EXPORT_SYMBOL(ceph_msg_data_set_bio);
3155
31b8006e
SW		 3156/*
3157 * construct a new message with given type, size
3158 * the new msg has a ref count of 1.
3159 */
b61c2763
SW		 3160struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3161 bool can_fail)
31b8006e
SW		 3162{
3163 struct ceph_msg *m;
3164
9516e45b 3165 m = kzalloc(sizeof(*m), flags);
31b8006e
SW		 3166 if (m == NULL)
3167 goto out;
31b8006e
SW		 3168
3169 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3170 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3171 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3172
9516e45b
AE		 3173 INIT_LIST_HEAD(&m->list_head);
3174 kref_init(&m->kref);
ca20892d 3175
43794509
AE		 3176 ceph_msg_data_init(&m->p);
3177 ceph_msg_data_init(&m->l);
3178 ceph_msg_data_init(&m->b);
43794509 3179
31b8006e 3180 /* front */
9516e45b 3181 m->front_max = front_len;
31b8006e
SW		 3182 if (front_len) {
3183 if (front_len > PAGE_CACHE_SIZE) {
34d23762 3184 m->front.iov_base = __vmalloc(front_len, flags,
31b8006e
SW		 3185 PAGE_KERNEL);
3186 m->front_is_vmalloc = true;
3187 } else {
34d23762 3188 m->front.iov_base = kmalloc(front_len, flags);
31b8006e
SW		 3189 }
3190 if (m->front.iov_base == NULL) {
b61c2763 3191 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3192 front_len);
3193 goto out2;
3194 }
3195 } else {
3196 m->front.iov_base = NULL;
3197 }
3198 m->front.iov_len = front_len;
3199
bb257664 3200 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3201 return m;
3202
3203out2:
3204 ceph_msg_put(m);
3205out:
b61c2763
SW		 3206 if (!can_fail) {
3207 pr_err("msg_new can't create type %d front %d\n", type,
3208 front_len);
f0ed1b7c 3209 WARN_ON(1);
b61c2763
SW		 3210 } else {
3211 dout("msg_new can't create type %d front %d\n", type,
3212 front_len);
3213 }
a79832f2 3214 return NULL;
31b8006e 3215}
3d14c5d2 3216EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3217
31b8006e
SW		 3218/*
3219 * Allocate "middle" portion of a message, if it is needed and wasn't
3220 * allocated by alloc_msg. This allows us to read a small fixed-size
3221 * per-type header in the front and then gracefully fail (i.e.,
3222 * propagate the error to the caller based on info in the front) when
3223 * the middle is too large.
3224 */
2450418c 3225static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3226{
3227 int type = le16_to_cpu(msg->hdr.type);
3228 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3229
3230 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3231 ceph_msg_type_name(type), middle_len);
3232 BUG_ON(!middle_len);
3233 BUG_ON(msg->middle);
3234
b6c1d5b8 3235 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3236 if (!msg->middle)
3237 return -ENOMEM;
3238 return 0;
3239}
3240
2450418c 3241/*
1c20f2d2
AE		 3242 * Allocate a message for receiving an incoming message on a
3243 * connection, and save the result in con->in_msg. Uses the
3244 * connection's private alloc_msg op if available.
3245 *
4740a623
SW		 3246 * Returns 0 on success, or a negative error code.
3247 *
3248 * On success, if we set *skip = 1:
3249 * - the next message should be skipped and ignored.
3250 * - con->in_msg == NULL
3251 * or if we set *skip = 0:
3252 * - con->in_msg is non-null.
3253 * On error (ENOMEM, EAGAIN, ...),
3254 * - con->in_msg == NULL
2450418c 3255 */
4740a623 3256static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3257{
4740a623 3258 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3259 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3260 struct ceph_msg *msg;
4740a623 3261 int ret = 0;
2450418c 3262
1c20f2d2 3263 BUG_ON(con->in_msg != NULL);
53ded495 3264 BUG_ON(!con->ops->alloc_msg);
2450418c 3265
53ded495
AE		 3266 mutex_unlock(&con->mutex);
3267 msg = con->ops->alloc_msg(con, hdr, skip);
3268 mutex_lock(&con->mutex);
3269 if (con->state != CON_STATE_OPEN) {
3270 if (msg)
1d866d1c 3271 ceph_msg_put(msg);
53ded495
AE		 3272 return -EAGAIN;
3273 }
4137577a
AE		 3274 if (msg) {
3275 BUG_ON(*skip);
3276 con->in_msg = msg;
36eb71aa 3277 con->in_msg->con = con->ops->get(con);
92ce034b 3278 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3279 } else {
3280 /*
3281 * Null message pointer means either we should skip
3282 * this message or we couldn't allocate memory. The
3283 * former is not an error.
3284 */
3285 if (*skip)
3286 return 0;
3287 con->error_msg = "error allocating memory for incoming message";
3288
53ded495 3289 return -ENOMEM;
2450418c 3290 }
1c20f2d2 3291 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3292
1c20f2d2
AE		 3293 if (middle_len && !con->in_msg->middle) {
3294 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3295 if (ret < 0) {
1c20f2d2
AE		 3296 ceph_msg_put(con->in_msg);
3297 con->in_msg = NULL;
2450418c
YS		 3298 }
3299 }
9d7f0f13 3300
4740a623 3301 return ret;
2450418c
YS		 3302}
3303
31b8006e
SW		 3304
3305/*
3306 * Free a generically kmalloc'd message.
3307 */
3308void ceph_msg_kfree(struct ceph_msg *m)
3309{
3310 dout("msg_kfree %p\n", m);
3311 if (m->front_is_vmalloc)
3312 vfree(m->front.iov_base);
3313 else
3314 kfree(m->front.iov_base);
3315 kfree(m);
3316}
3317
3318/*
3319 * Drop a msg ref. Destroy as needed.
3320 */
c2e552e7
SW		 3321void ceph_msg_last_put(struct kref *kref)
3322{
3323 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
31b8006e 3324
c2e552e7
SW		 3325 dout("ceph_msg_put last one on %p\n", m);
3326 WARN_ON(!list_empty(&m->list_head));
3327
3328 /* drop middle, data, if any */
3329 if (m->middle) {
3330 ceph_buffer_put(m->middle);
3331 m->middle = NULL;
31b8006e 3332 }
97fb1c7f 3333 if (ceph_msg_has_pages(m)) {
f9e15777
AE		 3334 m->p.length = 0;
3335 m->p.pages = NULL;
888334f9 3336 m->p.type = CEPH_OSD_DATA_TYPE_NONE;
97fb1c7f 3337 }
97fb1c7f 3338 if (ceph_msg_has_pagelist(m)) {
f9e15777
AE		 3339 ceph_pagelist_release(m->l.pagelist);
3340 kfree(m->l.pagelist);
3341 m->l.pagelist = NULL;
888334f9
AE		 3342 m->l.type = CEPH_OSD_DATA_TYPE_NONE;
3343 }
3344 if (ceph_msg_has_bio(m)) {
3345 m->b.bio = NULL;
3346 m->b.type = CEPH_OSD_DATA_TYPE_NONE;
58bb3b37
SW		 3347 }
3348
c2e552e7
SW		 3349 if (m->pool)
3350 ceph_msgpool_put(m->pool, m);
3351 else
3352 ceph_msg_kfree(m);
31b8006e 3353}
3d14c5d2 3354EXPORT_SYMBOL(ceph_msg_last_put);
9ec7cab1
SW		 3355
3356void ceph_msg_dump(struct ceph_msg *msg)
3357{
4a73ef27 3358 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
f9e15777 3359 msg->front_max, msg->p.length);
9ec7cab1
SW		 3360 print_hex_dump(KERN_DEBUG, "header: ",
3361 DUMP_PREFIX_OFFSET, 16, 1,
3362 &msg->hdr, sizeof(msg->hdr), true);
3363 print_hex_dump(KERN_DEBUG, " front: ",
3364 DUMP_PREFIX_OFFSET, 16, 1,
3365 msg->front.iov_base, msg->front.iov_len, true);
3366 if (msg->middle)
3367 print_hex_dump(KERN_DEBUG, "middle: ",
3368 DUMP_PREFIX_OFFSET, 16, 1,
3369 msg->middle->vec.iov_base,
3370 msg->middle->vec.iov_len, true);
3371 print_hex_dump(KERN_DEBUG, "footer: ",
3372 DUMP_PREFIX_OFFSET, 16, 1,
3373 &msg->footer, sizeof(msg->footer), true);
3374}
3d14c5d2 3375EXPORT_SYMBOL(ceph_msg_dump);
ubuntu-bionic-kernel mirror