projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
xen-blkfront: don't add indirect pages to list when !feature_persistent
[linux-2.6-block.git] / net / ceph / messenger.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
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
2b3e0c90 18#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 19#include <linux/ceph/libceph.h>
20#include <linux/ceph/messenger.h>
21#include <linux/ceph/decode.h>
22#include <linux/ceph/pagelist.h>
bc3b2d7f 23#include <linux/export.h>
31b8006e 24
fe38a2b6
AE		 25#define list_entry_next(pos, member) \
26 list_entry(pos->member.next, typeof(*pos), member)
27
31b8006e
SW		 28/*
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
34 * the sender.
35 */
36
bc18f4b1
AE		 37/*
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
41 * unexpected state.
42 *
43 * --------
44 * | NEW* | transient initial state
45 * --------
46 * | con_sock_state_init()
47 * v
48 * ----------
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
51 * ^ \
52 * | \ con_sock_state_connecting()
53 * | ----------------------
54 * | \
55 * + con_sock_state_closed() \
fbb85a47
SW		 56 * |+--------------------------- \
57 * | \ \ \
58 * | ----------- \ \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
61 * | ^ \ |
62 * | | \ |
63 * | + con_sock_state_closing() \ |
64 * | / \ | |
65 * | / --------------- | |
66 * | / \ v v
bc18f4b1
AE		 67 * | / --------------
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
71 * | | v
72 * -------------
73 * | CONNECTED | TCP connection established
74 * -------------
75 *
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77 */
ce2c8903
AE		 78
79#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
84
8dacc7da
SW		 85/*
86 * connection states
87 */
88#define CON_STATE_CLOSED 1 /* -> PREOPEN */
89#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
94
4a861692
SW		 95/*
96 * ceph_connection flag bits
97 */
98#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 104
c9ffc77a
AE		 105static bool con_flag_valid(unsigned long con_flag)
106{
107 switch (con_flag) {
108 case CON_FLAG_LOSSYTX:
109 case CON_FLAG_KEEPALIVE_PENDING:
110 case CON_FLAG_WRITE_PENDING:
111 case CON_FLAG_SOCK_CLOSED:
112 case CON_FLAG_BACKOFF:
113 return true;
114 default:
115 return false;
116 }
117}
118
119static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120{
121 BUG_ON(!con_flag_valid(con_flag));
122
123 clear_bit(con_flag, &con->flags);
124}
125
126static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127{
128 BUG_ON(!con_flag_valid(con_flag));
129
130 set_bit(con_flag, &con->flags);
131}
132
133static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134{
135 BUG_ON(!con_flag_valid(con_flag));
136
137 return test_bit(con_flag, &con->flags);
138}
139
140static bool con_flag_test_and_clear(struct ceph_connection *con,
141 unsigned long con_flag)
142{
143 BUG_ON(!con_flag_valid(con_flag));
144
145 return test_and_clear_bit(con_flag, &con->flags);
146}
147
148static bool con_flag_test_and_set(struct ceph_connection *con,
149 unsigned long con_flag)
150{
151 BUG_ON(!con_flag_valid(con_flag));
152
153 return test_and_set_bit(con_flag, &con->flags);
154}
155
e3d5d638
AE		 156/* Slab caches for frequently-allocated structures */
157
158static struct kmem_cache *ceph_msg_cache;
81b36be4 159static struct kmem_cache *ceph_msg_data_cache;
e3d5d638 160
31b8006e
SW		 161/* static tag bytes (protocol control messages) */
162static char tag_msg = CEPH_MSGR_TAG_MSG;
163static char tag_ack = CEPH_MSGR_TAG_ACK;
164static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
165
a6a5349d
SW		 166#ifdef CONFIG_LOCKDEP
167static struct lock_class_key socket_class;
168#endif
169
84495f49
AE		 170/*
171 * When skipping (ignoring) a block of input we read it into a "skip
172 * buffer," which is this many bytes in size.
173 */
174#define SKIP_BUF_SIZE 1024
31b8006e
SW		 175
176static void queue_con(struct ceph_connection *con);
37ab77ac 177static void cancel_con(struct ceph_connection *con);
31b8006e 178static void con_work(struct work_struct *);
93209264 179static void con_fault(struct ceph_connection *con);
31b8006e 180
31b8006e 181/*
f64a9317
AE		 182 * Nicely render a sockaddr as a string. An array of formatted
183 * strings is used, to approximate reentrancy.
31b8006e 184 */
f64a9317
AE		 185#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
186#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
187#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
188#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
189
190static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
191static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 192
57666519 193static struct page *zero_page; /* used in certain error cases */
57666519 194
3d14c5d2 195const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 196{
197 int i;
198 char *s;
99f0f3b2
AE		 199 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
200 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 201
f64a9317 202 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 203 s = addr_str[i];
204
205 switch (ss->ss_family) {
206 case AF_INET:
bd406145
AE		 207 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
208 ntohs(in4->sin_port));
31b8006e
SW		 209 break;
210
211 case AF_INET6:
bd406145
AE		 212 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
213 ntohs(in6->sin6_port));
31b8006e
SW		 214 break;
215
216 default:
d3002b97
AE		 217 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
218 ss->ss_family);
31b8006e
SW		 219 }
220
221 return s;
222}
3d14c5d2 223EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 224
63f2d211
SW		 225static void encode_my_addr(struct ceph_messenger *msgr)
226{
227 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
228 ceph_encode_addr(&msgr->my_enc_addr);
229}
230
31b8006e
SW		 231/*
232 * work queue for all reading and writing to/from the socket.
233 */
e0f43c94 234static struct workqueue_struct *ceph_msgr_wq;
31b8006e 235
e3d5d638
AE		 236static int ceph_msgr_slab_init(void)
237{
238 BUG_ON(ceph_msg_cache);
239 ceph_msg_cache = kmem_cache_create("ceph_msg",
240 sizeof (struct ceph_msg),
241 __alignof__(struct ceph_msg), 0, NULL);
81b36be4
AE		 242
243 if (!ceph_msg_cache)
244 return -ENOMEM;
245
246 BUG_ON(ceph_msg_data_cache);
247 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
248 sizeof (struct ceph_msg_data),
249 __alignof__(struct ceph_msg_data),
250 0, NULL);
251 if (ceph_msg_data_cache)
252 return 0;
253
254 kmem_cache_destroy(ceph_msg_cache);
255 ceph_msg_cache = NULL;
256
257 return -ENOMEM;
e3d5d638
AE		 258}
259
260static void ceph_msgr_slab_exit(void)
261{
81b36be4
AE		 262 BUG_ON(!ceph_msg_data_cache);
263 kmem_cache_destroy(ceph_msg_data_cache);
264 ceph_msg_data_cache = NULL;
265
e3d5d638
AE		 266 BUG_ON(!ceph_msg_cache);
267 kmem_cache_destroy(ceph_msg_cache);
268 ceph_msg_cache = NULL;
269}
270
15417167 271static void _ceph_msgr_exit(void)
6173d1f0 272{
d3002b97 273 if (ceph_msgr_wq) {
6173d1f0 274 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 275 ceph_msgr_wq = NULL;
276 }
6173d1f0 277
e3d5d638
AE		 278 ceph_msgr_slab_exit();
279
6173d1f0
AE		 280 BUG_ON(zero_page == NULL);
281 kunmap(zero_page);
282 page_cache_release(zero_page);
283 zero_page = NULL;
284}
285
3d14c5d2 286int ceph_msgr_init(void)
31b8006e 287{
57666519
AE		 288 BUG_ON(zero_page != NULL);
289 zero_page = ZERO_PAGE(0);
290 page_cache_get(zero_page);
291
e3d5d638
AE		 292 if (ceph_msgr_slab_init())
293 return -ENOMEM;
294
f9865f06
ID		 295 /*
296 * The number of active work items is limited by the number of
297 * connections, so leave @max_active at default.
298 */
299 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
6173d1f0
AE		 300 if (ceph_msgr_wq)
301 return 0;
57666519 302
6173d1f0
AE		 303 pr_err("msgr_init failed to create workqueue\n");
304 _ceph_msgr_exit();
57666519 305
6173d1f0 306 return -ENOMEM;
31b8006e 307}
3d14c5d2 308EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 309
310void ceph_msgr_exit(void)
311{
57666519 312 BUG_ON(ceph_msgr_wq == NULL);
57666519 313
6173d1f0 314 _ceph_msgr_exit();
31b8006e 315}
3d14c5d2 316EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 317
cd84db6e 318void ceph_msgr_flush(void)
a922d38f
SW		 319{
320 flush_workqueue(ceph_msgr_wq);
321}
3d14c5d2 322EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 323
ce2c8903
AE		 324/* Connection socket state transition functions */
325
326static void con_sock_state_init(struct ceph_connection *con)
327{
328 int old_state;
329
330 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
331 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
332 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 333 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
334 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 335}
336
337static void con_sock_state_connecting(struct ceph_connection *con)
338{
339 int old_state;
340
341 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
342 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
343 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 344 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
345 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 346}
347
348static void con_sock_state_connected(struct ceph_connection *con)
349{
350 int old_state;
351
352 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
353 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
354 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 355 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
356 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 357}
358
359static void con_sock_state_closing(struct ceph_connection *con)
360{
361 int old_state;
362
363 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
364 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
365 old_state != CON_SOCK_STATE_CONNECTED &&
366 old_state != CON_SOCK_STATE_CLOSING))
367 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 368 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
369 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 370}
371
372static void con_sock_state_closed(struct ceph_connection *con)
373{
374 int old_state;
375
376 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
377 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 378 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 379 old_state != CON_SOCK_STATE_CONNECTING &&
380 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 381 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 382 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
383 CON_SOCK_STATE_CLOSED);
ce2c8903 384}
a922d38f 385
31b8006e
SW		 386/*
387 * socket callback functions
388 */
389
390/* data available on socket, or listen socket received a connect */
676d2369 391static void ceph_sock_data_ready(struct sock *sk)
31b8006e 392{
bd406145 393 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 394 if (atomic_read(&con->msgr->stopping)) {
395 return;
396 }
bd406145 397
31b8006e 398 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 399 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 400 con, con->state);
401 queue_con(con);
402 }
403}
404
405/* socket has buffer space for writing */
327800bd 406static void ceph_sock_write_space(struct sock *sk)
31b8006e 407{
d3002b97 408 struct ceph_connection *con = sk->sk_user_data;
31b8006e 409
182fac26
JS		 410 /* only queue to workqueue if there is data we want to write,
411 * and there is sufficient space in the socket buffer to accept
327800bd 412 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 413 * doesn't get called again until try_write() fills the socket
414 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
415 * and net/core/stream.c:sk_stream_write_space().
416 */
c9ffc77a 417 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
64dc6130 418 if (sk_stream_is_writeable(sk)) {
327800bd 419 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 420 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
421 queue_con(con);
422 }
31b8006e 423 } else {
327800bd 424 dout("%s %p nothing to write\n", __func__, con);
31b8006e 425 }
31b8006e
SW		 426}
427
428/* socket's state has changed */
327800bd 429static void ceph_sock_state_change(struct sock *sk)
31b8006e 430{
bd406145 431 struct ceph_connection *con = sk->sk_user_data;
31b8006e 432
327800bd 433 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 434 con, con->state, sk->sk_state);
435
31b8006e
SW		 436 switch (sk->sk_state) {
437 case TCP_CLOSE:
327800bd 438 dout("%s TCP_CLOSE\n", __func__);
31b8006e 439 case TCP_CLOSE_WAIT:
327800bd 440 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 441 con_sock_state_closing(con);
c9ffc77a 442 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 443 queue_con(con);
31b8006e
SW		 444 break;
445 case TCP_ESTABLISHED:
327800bd 446 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 447 con_sock_state_connected(con);
31b8006e
SW		 448 queue_con(con);
449 break;
d3002b97
AE		 450 default: /* Everything else is uninteresting */
451 break;
31b8006e
SW		 452 }
453}
454
455/*
456 * set up socket callbacks
457 */
458static void set_sock_callbacks(struct socket *sock,
459 struct ceph_connection *con)
460{
461 struct sock *sk = sock->sk;
bd406145 462 sk->sk_user_data = con;
327800bd
AE		 463 sk->sk_data_ready = ceph_sock_data_ready;
464 sk->sk_write_space = ceph_sock_write_space;
465 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 466}
467
468
469/*
470 * socket helpers
471 */
472
473/*
474 * initiate connection to a remote socket.
475 */
41617d0c 476static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 477{
f91d3471 478 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e
SW		 479 struct socket *sock;
480 int ret;
481
482 BUG_ON(con->sock);
f91d3471
SW		 483 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
484 IPPROTO_TCP, &sock);
31b8006e 485 if (ret)
41617d0c 486 return ret;
6d7fdb0a 487 sock->sk->sk_allocation = GFP_NOFS;
31b8006e 488
a6a5349d
SW		 489#ifdef CONFIG_LOCKDEP
490 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
491#endif
492
31b8006e
SW		 493 set_sock_callbacks(sock, con);
494
3d14c5d2 495 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 496
89a86be0 497 con_sock_state_connecting(con);
f91d3471
SW		 498 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
499 O_NONBLOCK);
31b8006e
SW		 500 if (ret == -EINPROGRESS) {
501 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 502 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 503 sock->sk->sk_state);
a5bc3129 504 } else if (ret < 0) {
31b8006e 505 pr_err("connect %s error %d\n",
3d14c5d2 506 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 507 sock_release(sock);
41617d0c 508 return ret;
a5bc3129 509 }
89baaa57 510
ba988f87
CH		 511 if (con->msgr->tcp_nodelay) {
512 int optval = 1;
513
514 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
515 (char *)&optval, sizeof(optval));
516 if (ret)
517 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
518 ret);
519 }
520
a5bc3129 521 con->sock = sock;
41617d0c 522 return 0;
31b8006e
SW		 523}
524
525static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
526{
527 struct kvec iov = {buf, len};
528 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 529 int r;
31b8006e 530
98bdb0aa
SW		 531 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
532 if (r == -EAGAIN)
533 r = 0;
534 return r;
31b8006e
SW		 535}
536
afb3d90e
AE		 537static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
538 int page_offset, size_t length)
539{
540 void *kaddr;
541 int ret;
542
543 BUG_ON(page_offset + length > PAGE_SIZE);
544
545 kaddr = kmap(page);
546 BUG_ON(!kaddr);
547 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
548 kunmap(page);
549
550 return ret;
551}
552
31b8006e
SW		 553/*
554 * write something. @more is true if caller will be sending more data
555 * shortly.
556 */
557static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
558 size_t kvlen, size_t len, int more)
559{
560 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 561 int r;
31b8006e
SW		 562
563 if (more)
564 msg.msg_flags |= MSG_MORE;
565 else
566 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
567
42961d23
SW		 568 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
569 if (r == -EAGAIN)
570 r = 0;
571 return r;
31b8006e
SW		 572}
573
178eda29 574static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 575 int offset, size_t size, bool more)
31739139
AE		 576{
577 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
578 int ret;
579
580 ret = kernel_sendpage(sock, page, offset, size, flags);
581 if (ret == -EAGAIN)
582 ret = 0;
583
584 return ret;
585}
586
178eda29
CC		 587static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
588 int offset, size_t size, bool more)
589{
590 int ret;
591 struct kvec iov;
592
593 /* sendpage cannot properly handle pages with page_count == 0,
594 * we need to fallback to sendmsg if that's the case */
595 if (page_count(page) >= 1)
596 return __ceph_tcp_sendpage(sock, page, offset, size, more);
597
598 iov.iov_base = kmap(page) + offset;
599 iov.iov_len = size;
600 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
601 kunmap(page);
602
603 return ret;
604}
31b8006e
SW		 605
606/*
607 * Shutdown/close the socket for the given connection.
608 */
609static int con_close_socket(struct ceph_connection *con)
610{
8007b8d6 611 int rc = 0;
31b8006e
SW		 612
613 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 614 if (con->sock) {
615 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
616 sock_release(con->sock);
617 con->sock = NULL;
618 }
456ea468
AE		 619
620 /*
4a861692 621 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 622 * independent of the connection mutex, and we could have
623 * received a socket close event before we had the chance to
624 * shut the socket down.
625 */
c9ffc77a 626 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 627
ce2c8903 628 con_sock_state_closed(con);
31b8006e
SW		 629 return rc;
630}
631
632/*
633 * Reset a connection. Discard all incoming and outgoing messages
634 * and clear *_seq state.
635 */
636static void ceph_msg_remove(struct ceph_msg *msg)
637{
638 list_del_init(&msg->list_head);
38941f80 639 BUG_ON(msg->con == NULL);
36eb71aa 640 msg->con->ops->put(msg->con);
38941f80
AE		 641 msg->con = NULL;
642
31b8006e
SW		 643 ceph_msg_put(msg);
644}
645static void ceph_msg_remove_list(struct list_head *head)
646{
647 while (!list_empty(head)) {
648 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
649 list_head);
650 ceph_msg_remove(msg);
651 }
652}
653
654static void reset_connection(struct ceph_connection *con)
655{
656 /* reset connection, out_queue, msg_ and connect_seq */
657 /* discard existing out_queue and msg_seq */
0fa6ebc6 658 dout("reset_connection %p\n", con);
31b8006e
SW		 659 ceph_msg_remove_list(&con->out_queue);
660 ceph_msg_remove_list(&con->out_sent);
661
cf3e5c40 662 if (con->in_msg) {
38941f80
AE		 663 BUG_ON(con->in_msg->con != con);
664 con->in_msg->con = NULL;
cf3e5c40
SW		 665 ceph_msg_put(con->in_msg);
666 con->in_msg = NULL;
36eb71aa 667 con->ops->put(con);
cf3e5c40
SW		 668 }
669
31b8006e
SW		 670 con->connect_seq = 0;
671 con->out_seq = 0;
c86a2930
SW		 672 if (con->out_msg) {
673 ceph_msg_put(con->out_msg);
674 con->out_msg = NULL;
675 }
31b8006e 676 con->in_seq = 0;
0e0d5e0c 677 con->in_seq_acked = 0;
31b8006e
SW		 678}
679
680/*
681 * mark a peer down. drop any open connections.
682 */
683void ceph_con_close(struct ceph_connection *con)
684{
8c50c817 685 mutex_lock(&con->mutex);
3d14c5d2
YS		 686 dout("con_close %p peer %s\n", con,
687 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 688 con->state = CON_STATE_CLOSED;
a5988c49 689
c9ffc77a
AE		 690 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
691 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
692 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
693 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 694
31b8006e 695 reset_connection(con);
6f2bc3ff 696 con->peer_global_seq = 0;
37ab77ac 697 cancel_con(con);
ee76e073 698 con_close_socket(con);
ec302645 699 mutex_unlock(&con->mutex);
31b8006e 700}
3d14c5d2 701EXPORT_SYMBOL(ceph_con_close);
31b8006e 702
31b8006e
SW		 703/*
704 * Reopen a closed connection, with a new peer address.
705 */
b7a9e5dd
SW		 706void ceph_con_open(struct ceph_connection *con,
707 __u8 entity_type, __u64 entity_num,
708 struct ceph_entity_addr *addr)
31b8006e 709{
5469155f 710 mutex_lock(&con->mutex);
3d14c5d2 711 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 712
122070a2 713 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 714 con->state = CON_STATE_PREOPEN;
a5988c49 715
b7a9e5dd
SW		 716 con->peer_name.type = (__u8) entity_type;
717 con->peer_name.num = cpu_to_le64(entity_num);
718
31b8006e 719 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 720 con->delay = 0; /* reset backoff memory */
5469155f 721 mutex_unlock(&con->mutex);
31b8006e
SW		 722 queue_con(con);
723}
3d14c5d2 724EXPORT_SYMBOL(ceph_con_open);
31b8006e 725
87b315a5
SW		 726/*
727 * return true if this connection ever successfully opened
728 */
729bool ceph_con_opened(struct ceph_connection *con)
730{
731 return con->connect_seq > 0;
732}
733
31b8006e
SW		 734/*
735 * initialize a new connection.
736 */
1bfd89f4
AE		 737void ceph_con_init(struct ceph_connection *con, void *private,
738 const struct ceph_connection_operations *ops,
b7a9e5dd 739 struct ceph_messenger *msgr)
31b8006e
SW		 740{
741 dout("con_init %p\n", con);
742 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 743 con->private = private;
744 con->ops = ops;
31b8006e 745 con->msgr = msgr;
ce2c8903
AE		 746
747 con_sock_state_init(con);
748
ec302645 749 mutex_init(&con->mutex);
31b8006e
SW		 750 INIT_LIST_HEAD(&con->out_queue);
751 INIT_LIST_HEAD(&con->out_sent);
752 INIT_DELAYED_WORK(&con->work, con_work);
a5988c49 753
8dacc7da 754 con->state = CON_STATE_CLOSED;
31b8006e 755}
3d14c5d2 756EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 757
758
759/*
760 * We maintain a global counter to order connection attempts. Get
761 * a unique seq greater than @gt.
762 */
763static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
764{
765 u32 ret;
766
767 spin_lock(&msgr->global_seq_lock);
768 if (msgr->global_seq < gt)
769 msgr->global_seq = gt;
770 ret = ++msgr->global_seq;
771 spin_unlock(&msgr->global_seq_lock);
772 return ret;
773}
774
e2200423 775static void con_out_kvec_reset(struct ceph_connection *con)
859eb799
AE		 776{
777 con->out_kvec_left = 0;
778 con->out_kvec_bytes = 0;
779 con->out_kvec_cur = &con->out_kvec[0];
780}
781
e2200423 782static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 783 size_t size, void *data)
784{
785 int index;
786
787 index = con->out_kvec_left;
788 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
789
790 con->out_kvec[index].iov_len = size;
791 con->out_kvec[index].iov_base = data;
792 con->out_kvec_left++;
793 con->out_kvec_bytes += size;
794}
31b8006e 795
df6ad1f9 796#ifdef CONFIG_BLOCK
6aaa4511
AE		 797
798/*
799 * For a bio data item, a piece is whatever remains of the next
800 * entry in the current bio iovec, or the first entry in the next
801 * bio in the list.
802 */
8ae4f4f5 803static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 804 size_t length)
6aaa4511 805{
8ae4f4f5 806 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 807 struct bio *bio;
808
809 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
810
811 bio = data->bio;
812 BUG_ON(!bio);
6aaa4511 813
ca8b3a69 814 cursor->resid = min(length, data->bio_length);
6aaa4511 815 cursor->bio = bio;
f38a5181
KO		 816 cursor->bvec_iter = bio->bi_iter;
817 cursor->last_piece =
818 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
6aaa4511
AE		 819}
820
8ae4f4f5 821static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 822 size_t *page_offset,
823 size_t *length)
824{
8ae4f4f5 825 struct ceph_msg_data *data = cursor->data;
6aaa4511 826 struct bio *bio;
f38a5181 827 struct bio_vec bio_vec;
6aaa4511
AE		 828
829 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
830
831 bio = cursor->bio;
832 BUG_ON(!bio);
833
f38a5181 834 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511 835
f38a5181 836 *page_offset = (size_t) bio_vec.bv_offset;
6aaa4511 837 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 838 if (cursor->last_piece) /* pagelist offset is always 0 */
839 *length = cursor->resid;
840 else
f38a5181 841 *length = (size_t) bio_vec.bv_len;
25aff7c5 842 BUG_ON(*length > cursor->resid);
5df521b1 843 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511 844
f38a5181 845 return bio_vec.bv_page;
6aaa4511
AE		 846}
847
8ae4f4f5
AE		 848static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
849 size_t bytes)
6aaa4511 850{
6aaa4511 851 struct bio *bio;
f38a5181 852 struct bio_vec bio_vec;
6aaa4511 853
8ae4f4f5 854 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
6aaa4511
AE		 855
856 bio = cursor->bio;
857 BUG_ON(!bio);
858
f38a5181 859 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511
AE		 860
861 /* Advance the cursor offset */
862
25aff7c5
AE		 863 BUG_ON(cursor->resid < bytes);
864 cursor->resid -= bytes;
f38a5181
KO		 865
866 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
867
868 if (bytes < bio_vec.bv_len)
6aaa4511
AE		 869 return false; /* more bytes to process in this segment */
870
871 /* Move on to the next segment, and possibly the next bio */
872
f38a5181 873 if (!cursor->bvec_iter.bi_size) {
6aaa4511 874 bio = bio->bi_next;
0ec1d15e
ID		 875 cursor->bio = bio;
876 if (bio)
877 cursor->bvec_iter = bio->bi_iter;
878 else
879 memset(&cursor->bvec_iter, 0,
880 sizeof(cursor->bvec_iter));
6aaa4511 881 }
6aaa4511 882
25aff7c5
AE		 883 if (!cursor->last_piece) {
884 BUG_ON(!cursor->resid);
885 BUG_ON(!bio);
886 /* A short read is OK, so use <= rather than == */
f38a5181 887 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
6aaa4511 888 cursor->last_piece = true;
25aff7c5 889 }
6aaa4511
AE		 890
891 return true;
892}
ea96571f 893#endif /* CONFIG_BLOCK */
df6ad1f9 894
e766d7b5
AE		 895/*
896 * For a page array, a piece comes from the first page in the array
897 * that has not already been fully consumed.
898 */
8ae4f4f5 899static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 900 size_t length)
e766d7b5 901{
8ae4f4f5 902 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 903 int page_count;
904
905 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
906
907 BUG_ON(!data->pages);
908 BUG_ON(!data->length);
909
ca8b3a69 910 cursor->resid = min(length, data->length);
e766d7b5 911 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 912 cursor->page_offset = data->alignment & ~PAGE_MASK;
913 cursor->page_index = 0;
56fc5659
AE		 914 BUG_ON(page_count > (int)USHRT_MAX);
915 cursor->page_count = (unsigned short)page_count;
916 BUG_ON(length > SIZE_MAX - cursor->page_offset);
5f740d7e 917 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 918}
919
8ae4f4f5
AE		 920static struct page *
921ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
922 size_t *page_offset, size_t *length)
e766d7b5 923{
8ae4f4f5 924 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 925
926 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
927
928 BUG_ON(cursor->page_index >= cursor->page_count);
929 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 930
931 *page_offset = cursor->page_offset;
25aff7c5 932 if (cursor->last_piece)
e766d7b5 933 *length = cursor->resid;
25aff7c5 934 else
e766d7b5 935 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 936
937 return data->pages[cursor->page_index];
938}
939
8ae4f4f5 940static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 941 size_t bytes)
942{
8ae4f4f5 943 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 944
945 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 946
947 /* Advance the cursor page offset */
948
949 cursor->resid -= bytes;
5df521b1
AE		 950 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
951 if (!bytes || cursor->page_offset)
e766d7b5
AE		 952 return false; /* more bytes to process in the current page */
953
d90deda6
YZ		 954 if (!cursor->resid)
955 return false; /* no more data */
956
5df521b1 957 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 958
959 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 960 cursor->page_index++;
25aff7c5 961 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 962
963 return true;
964}
965
fe38a2b6 966/*
dd236fcb
AE		 967 * For a pagelist, a piece is whatever remains to be consumed in the
968 * first page in the list, or the front of the next page.
fe38a2b6 969 */
8ae4f4f5
AE		 970static void
971ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 972 size_t length)
fe38a2b6 973{
8ae4f4f5 974 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 975 struct ceph_pagelist *pagelist;
976 struct page *page;
977
dd236fcb 978 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 979
980 pagelist = data->pagelist;
981 BUG_ON(!pagelist);
25aff7c5
AE		 982
983 if (!length)
fe38a2b6
AE		 984 return; /* pagelist can be assigned but empty */
985
986 BUG_ON(list_empty(&pagelist->head));
987 page = list_first_entry(&pagelist->head, struct page, lru);
988
ca8b3a69 989 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 990 cursor->page = page;
991 cursor->offset = 0;
a51b272e 992 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 993}
994
8ae4f4f5
AE		 995static struct page *
996ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
997 size_t *page_offset, size_t *length)
fe38a2b6 998{
8ae4f4f5 999 struct ceph_msg_data *data = cursor->data;
fe38a2b6 1000 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 1001
1002 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1003
1004 pagelist = data->pagelist;
1005 BUG_ON(!pagelist);
1006
1007 BUG_ON(!cursor->page);
25aff7c5 1008 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 1009
5df521b1 1010 /* offset of first page in pagelist is always 0 */
fe38a2b6 1011 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 1012 if (cursor->last_piece)
25aff7c5
AE		 1013 *length = cursor->resid;
1014 else
1015 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1016
8ae4f4f5 1017 return cursor->page;
fe38a2b6
AE		 1018}
1019
8ae4f4f5 1020static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1021 size_t bytes)
fe38a2b6 1022{
8ae4f4f5 1023 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1024 struct ceph_pagelist *pagelist;
1025
1026 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1027
1028 pagelist = data->pagelist;
1029 BUG_ON(!pagelist);
25aff7c5
AE		 1030
1031 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1032 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1033
1034 /* Advance the cursor offset */
1035
25aff7c5 1036 cursor->resid -= bytes;
fe38a2b6 1037 cursor->offset += bytes;
5df521b1 1038 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1039 if (!bytes || cursor->offset & ~PAGE_MASK)
1040 return false; /* more bytes to process in the current page */
1041
d90deda6
YZ		 1042 if (!cursor->resid)
1043 return false; /* no more data */
1044
fe38a2b6
AE		 1045 /* Move on to the next page */
1046
1047 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1048 cursor->page = list_entry_next(cursor->page, lru);
25aff7c5 1049 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1050
1051 return true;
1052}
1053
dd236fcb
AE		 1054/*
1055 * Message data is handled (sent or received) in pieces, where each
1056 * piece resides on a single page. The network layer might not
1057 * consume an entire piece at once. A data item's cursor keeps
1058 * track of which piece is next to process and how much remains to
1059 * be processed in that piece. It also tracks whether the current
1060 * piece is the last one in the data item.
1061 */
ca8b3a69 1062static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1063{
ca8b3a69 1064 size_t length = cursor->total_resid;
8ae4f4f5 1065
8ae4f4f5 1066 switch (cursor->data->type) {
dd236fcb 1067 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1068 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1069 break;
e766d7b5 1070 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1071 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1072 break;
dd236fcb
AE		 1073#ifdef CONFIG_BLOCK
1074 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1075 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1076 break;
dd236fcb 1077#endif /* CONFIG_BLOCK */
6aaa4511 1078 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1079 default:
1080 /* BUG(); */
1081 break;
1082 }
8ae4f4f5 1083 cursor->need_crc = true;
dd236fcb
AE		 1084}
1085
ca8b3a69
AE		 1086static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1087{
1088 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1089 struct ceph_msg_data *data;
1090
1091 BUG_ON(!length);
1092 BUG_ON(length > msg->data_length);
1093 BUG_ON(list_empty(&msg->data));
1094
ca8b3a69
AE		 1095 cursor->data_head = &msg->data;
1096 cursor->total_resid = length;
1097 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1098 cursor->data = data;
1099
1100 __ceph_msg_data_cursor_init(cursor);
1101}
1102
dd236fcb
AE		 1103/*
1104 * Return the page containing the next piece to process for a given
1105 * data item, and supply the page offset and length of that piece.
1106 * Indicate whether this is the last piece in this data item.
1107 */
8ae4f4f5
AE		 1108static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1109 size_t *page_offset, size_t *length,
dd236fcb
AE		 1110 bool *last_piece)
1111{
1112 struct page *page;
1113
8ae4f4f5 1114 switch (cursor->data->type) {
dd236fcb 1115 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1116 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1117 break;
e766d7b5 1118 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1119 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1120 break;
dd236fcb
AE		 1121#ifdef CONFIG_BLOCK
1122 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1123 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1124 break;
dd236fcb 1125#endif /* CONFIG_BLOCK */
6aaa4511 1126 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1127 default:
1128 page = NULL;
1129 break;
1130 }
1131 BUG_ON(!page);
1132 BUG_ON(*page_offset + *length > PAGE_SIZE);
1133 BUG_ON(!*length);
1134 if (last_piece)
8ae4f4f5 1135 *last_piece = cursor->last_piece;
dd236fcb
AE		 1136
1137 return page;
1138}
1139
1140/*
1141 * Returns true if the result moves the cursor on to the next piece
1142 * of the data item.
1143 */
8ae4f4f5
AE		 1144static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1145 size_t bytes)
dd236fcb
AE		 1146{
1147 bool new_piece;
1148
25aff7c5 1149 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1150 switch (cursor->data->type) {
dd236fcb 1151 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1152 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1153 break;
e766d7b5 1154 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1155 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1156 break;
dd236fcb
AE		 1157#ifdef CONFIG_BLOCK
1158 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1159 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1160 break;
dd236fcb 1161#endif /* CONFIG_BLOCK */
6aaa4511 1162 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1163 default:
1164 BUG();
1165 break;
1166 }
ca8b3a69 1167 cursor->total_resid -= bytes;
dd236fcb 1168
ca8b3a69
AE		 1169 if (!cursor->resid && cursor->total_resid) {
1170 WARN_ON(!cursor->last_piece);
1171 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1172 cursor->data = list_entry_next(cursor->data, links);
1173 __ceph_msg_data_cursor_init(cursor);
a51b272e 1174 new_piece = true;
ca8b3a69 1175 }
a51b272e 1176 cursor->need_crc = new_piece;
ca8b3a69 1177
dd236fcb
AE		 1178 return new_piece;
1179}
1180
98fa5dd8 1181static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1182{
739c905b 1183 BUG_ON(!msg);
25aff7c5 1184 BUG_ON(!data_len);
739c905b 1185
4c59b4a2 1186 /* Initialize data cursor */
fe38a2b6 1187
8ae4f4f5 1188 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1189}
1190
31b8006e
SW		 1191/*
1192 * Prepare footer for currently outgoing message, and finish things
1193 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1194 */
859eb799 1195static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1196{
1197 struct ceph_msg *m = con->out_msg;
859eb799 1198 int v = con->out_kvec_left;
31b8006e 1199
fd154f3c
AE		 1200 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1201
31b8006e
SW		 1202 dout("prepare_write_message_footer %p\n", con);
1203 con->out_kvec_is_msg = true;
1204 con->out_kvec[v].iov_base = &m->footer;
33d07337
YZ		 1205 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1206 if (con->ops->sign_message)
1207 con->ops->sign_message(con, m);
1208 else
1209 m->footer.sig = 0;
1210 con->out_kvec[v].iov_len = sizeof(m->footer);
1211 con->out_kvec_bytes += sizeof(m->footer);
1212 } else {
1213 m->old_footer.flags = m->footer.flags;
1214 con->out_kvec[v].iov_len = sizeof(m->old_footer);
1215 con->out_kvec_bytes += sizeof(m->old_footer);
1216 }
31b8006e
SW		 1217 con->out_kvec_left++;
1218 con->out_more = m->more_to_follow;
c86a2930 1219 con->out_msg_done = true;
31b8006e
SW		 1220}
1221
1222/*
1223 * Prepare headers for the next outgoing message.
1224 */
1225static void prepare_write_message(struct ceph_connection *con)
1226{
1227 struct ceph_msg *m;
a9a0c51a 1228 u32 crc;
31b8006e 1229
e2200423 1230 con_out_kvec_reset(con);
31b8006e 1231 con->out_kvec_is_msg = true;
c86a2930 1232 con->out_msg_done = false;
31b8006e
SW		 1233
1234 /* Sneak an ack in there first? If we can get it into the same
1235 * TCP packet that's a good thing. */
1236 if (con->in_seq > con->in_seq_acked) {
1237 con->in_seq_acked = con->in_seq;
e2200423 1238 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1239 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1240 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1241 &con->out_temp_ack);
31b8006e
SW		 1242 }
1243
38941f80 1244 BUG_ON(list_empty(&con->out_queue));
859eb799 1245 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1246 con->out_msg = m;
38941f80 1247 BUG_ON(m->con != con);
4cf9d544
SW		 1248
1249 /* put message on sent list */
1250 ceph_msg_get(m);
1251 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1252
e84346b7
SW		 1253 /*
1254 * only assign outgoing seq # if we haven't sent this message
1255 * yet. if it is requeued, resend with it's original seq.
1256 */
1257 if (m->needs_out_seq) {
1258 m->hdr.seq = cpu_to_le64(++con->out_seq);
1259 m->needs_out_seq = false;
1260 }
98fa5dd8 1261 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1262
98fa5dd8 1263 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1264 m, con->out_seq, le16_to_cpu(m->hdr.type),
1265 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1266 m->data_length);
31b8006e
SW		 1267 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1268
1269 /* tag + hdr + front + middle */
e2200423
AE		 1270 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1271 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1272 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1273
31b8006e 1274 if (m->middle)
e2200423 1275 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1276 m->middle->vec.iov_base);
31b8006e
SW		 1277
1278 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1279 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1280 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1281 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1282
1283 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1284 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1285 if (m->middle) {
1286 crc = crc32c(0, m->middle->vec.iov_base,
1287 m->middle->vec.iov_len);
1288 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1289 } else
31b8006e 1290 con->out_msg->footer.middle_crc = 0;
739c905b 1291 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1292 le32_to_cpu(con->out_msg->footer.front_crc),
1293 le32_to_cpu(con->out_msg->footer.middle_crc));
1294
1295 /* is there a data payload? */
739c905b 1296 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1297 if (m->data_length) {
1298 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1299 con->out_more = 1; /* data + footer will follow */
1300 } else {
31b8006e 1301 /* no, queue up footer too and be done */
859eb799 1302 prepare_write_message_footer(con);
78625051 1303 }
31b8006e 1304
c9ffc77a 1305 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1306}
1307
1308/*
1309 * Prepare an ack.
1310 */
1311static void prepare_write_ack(struct ceph_connection *con)
1312{
1313 dout("prepare_write_ack %p %llu -> %llu\n", con,
1314 con->in_seq_acked, con->in_seq);
1315 con->in_seq_acked = con->in_seq;
1316
e2200423 1317 con_out_kvec_reset(con);
859eb799 1318
e2200423 1319 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1320
31b8006e 1321 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1322 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1323 &con->out_temp_ack);
1324
31b8006e 1325 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1326 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1327}
1328
3a23083b
SW		 1329/*
1330 * Prepare to share the seq during handshake
1331 */
1332static void prepare_write_seq(struct ceph_connection *con)
1333{
1334 dout("prepare_write_seq %p %llu -> %llu\n", con,
1335 con->in_seq_acked, con->in_seq);
1336 con->in_seq_acked = con->in_seq;
1337
1338 con_out_kvec_reset(con);
1339
1340 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1341 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1342 &con->out_temp_ack);
1343
1344 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1345}
1346
31b8006e
SW		 1347/*
1348 * Prepare to write keepalive byte.
1349 */
1350static void prepare_write_keepalive(struct ceph_connection *con)
1351{
1352 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1353 con_out_kvec_reset(con);
1354 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1355 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1356}
1357
1358/*
1359 * Connection negotiation.
1360 */
1361
dac1e716
AE		 1362static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1363 int *auth_proto)
4e7a5dcd 1364{
a3530df3 1365 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1366
1367 if (!con->ops->get_authorizer) {
1368 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1369 con->out_connect.authorizer_len = 0;
729796be 1370 return NULL;
b1c6b980
AE		 1371 }
1372
1373 /* Can't hold the mutex while getting authorizer */
ec302645 1374 mutex_unlock(&con->mutex);
dac1e716 1375 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1376 mutex_lock(&con->mutex);
4e7a5dcd 1377
a3530df3 1378 if (IS_ERR(auth))
729796be 1379 return auth;
8dacc7da 1380 if (con->state != CON_STATE_NEGOTIATING)
729796be 1381 return ERR_PTR(-EAGAIN);
0da5d703 1382
8f43fb53
AE		 1383 con->auth_reply_buf = auth->authorizer_reply_buf;
1384 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1385 return auth;
4e7a5dcd
SW		 1386}
1387
31b8006e
SW		 1388/*
1389 * We connected to a peer and are saying hello.
1390 */
e825a66d 1391static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1392{
e2200423
AE		 1393 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1394 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1395 &con->msgr->my_enc_addr);
eed0ef2c 1396
eed0ef2c 1397 con->out_more = 0;
c9ffc77a 1398 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1399}
1400
e825a66d 1401static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1402{
95c96174 1403 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1404 int proto;
dac1e716 1405 int auth_proto;
729796be 1406 struct ceph_auth_handshake *auth;
31b8006e
SW		 1407
1408 switch (con->peer_name.type) {
1409 case CEPH_ENTITY_TYPE_MON:
1410 proto = CEPH_MONC_PROTOCOL;
1411 break;
1412 case CEPH_ENTITY_TYPE_OSD:
1413 proto = CEPH_OSDC_PROTOCOL;
1414 break;
1415 case CEPH_ENTITY_TYPE_MDS:
1416 proto = CEPH_MDSC_PROTOCOL;
1417 break;
1418 default:
1419 BUG();
1420 }
1421
1422 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1423 con->connect_seq, global_seq, proto);
4e7a5dcd 1424
e825a66d 1425 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1426 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1427 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1428 con->out_connect.global_seq = cpu_to_le32(global_seq);
1429 con->out_connect.protocol_version = cpu_to_le32(proto);
1430 con->out_connect.flags = 0;
31b8006e 1431
dac1e716
AE		 1432 auth_proto = CEPH_AUTH_UNKNOWN;
1433 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1434 if (IS_ERR(auth))
1435 return PTR_ERR(auth);
3da54776 1436
dac1e716 1437 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1438 con->out_connect.authorizer_len = auth ?
1439 cpu_to_le32(auth->authorizer_buf_len) : 0;
1440
e2200423 1441 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1442 &con->out_connect);
1443 if (auth && auth->authorizer_buf_len)
e2200423 1444 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1445 auth->authorizer_buf);
859eb799 1446
31b8006e 1447 con->out_more = 0;
c9ffc77a 1448 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1449
e10c758e 1450 return 0;
31b8006e
SW		 1451}
1452
31b8006e
SW		 1453/*
1454 * write as much of pending kvecs to the socket as we can.
1455 * 1 -> done
1456 * 0 -> socket full, but more to do
1457 * <0 -> error
1458 */
1459static int write_partial_kvec(struct ceph_connection *con)
1460{
1461 int ret;
1462
1463 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1464 while (con->out_kvec_bytes > 0) {
1465 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1466 con->out_kvec_left, con->out_kvec_bytes,
1467 con->out_more);
1468 if (ret <= 0)
1469 goto out;
1470 con->out_kvec_bytes -= ret;
1471 if (con->out_kvec_bytes == 0)
1472 break; /* done */
f42299e6
AE		 1473
1474 /* account for full iov entries consumed */
1475 while (ret >= con->out_kvec_cur->iov_len) {
1476 BUG_ON(!con->out_kvec_left);
1477 ret -= con->out_kvec_cur->iov_len;
1478 con->out_kvec_cur++;
1479 con->out_kvec_left--;
1480 }
1481 /* and for a partially-consumed entry */
1482 if (ret) {
1483 con->out_kvec_cur->iov_len -= ret;
1484 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1485 }
1486 }
1487 con->out_kvec_left = 0;
1488 con->out_kvec_is_msg = false;
1489 ret = 1;
1490out:
1491 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1492 con->out_kvec_bytes, con->out_kvec_left, ret);
1493 return ret; /* done! */
1494}
1495
35b62808
AE		 1496static u32 ceph_crc32c_page(u32 crc, struct page *page,
1497 unsigned int page_offset,
1498 unsigned int length)
1499{
1500 char *kaddr;
1501
1502 kaddr = kmap(page);
1503 BUG_ON(kaddr == NULL);
1504 crc = crc32c(crc, kaddr + page_offset, length);
1505 kunmap(page);
1506
1507 return crc;
1508}
31b8006e
SW		 1509/*
1510 * Write as much message data payload as we can. If we finish, queue
1511 * up the footer.
1512 * 1 -> done, footer is now queued in out_kvec[].
1513 * 0 -> socket full, but more to do
1514 * <0 -> error
1515 */
34d2d200 1516static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1517{
1518 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1519 struct ceph_msg_data_cursor *cursor = &msg->cursor;
37675b0f 1520 bool do_datacrc = !con->msgr->nocrc;
f5db90bc 1521 u32 crc;
31b8006e 1522
859a35d5 1523 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1524
5240d9f9 1525 if (list_empty(&msg->data))
4c59b4a2
AE		 1526 return -EINVAL;
1527
5821bd8c
AE		 1528 /*
1529 * Iterate through each page that contains data to be
1530 * written, and send as much as possible for each.
1531 *
1532 * If we are calculating the data crc (the default), we will
1533 * need to map the page. If we have no pages, they have
1534 * been revoked, so use the zero page.
1535 */
f5db90bc 1536 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1537 while (cursor->resid) {
8a166d05 1538 struct page *page;
e387d525
AE		 1539 size_t page_offset;
1540 size_t length;
8a166d05 1541 bool last_piece;
8ea299bc 1542 bool need_crc;
f5db90bc 1543 int ret;
68b4476b 1544
8ae4f4f5 1545 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 1546 &last_piece);
e387d525 1547 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1548 length, last_piece);
f5db90bc
AE		 1549 if (ret <= 0) {
1550 if (do_datacrc)
1551 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1552
f5db90bc
AE		 1553 return ret;
1554 }
143334ff
AE		 1555 if (do_datacrc && cursor->need_crc)
1556 crc = ceph_crc32c_page(crc, page, page_offset, length);
8ae4f4f5 1557 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
31b8006e
SW		 1558 }
1559
34d2d200 1560 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1561
1562 /* prepare and queue up footer, too */
f5db90bc
AE		 1563 if (do_datacrc)
1564 msg->footer.data_crc = cpu_to_le32(crc);
1565 else
84ca8fc8 1566 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1567 con_out_kvec_reset(con);
859eb799 1568 prepare_write_message_footer(con);
f5db90bc
AE		 1569
1570 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1571}
1572
1573/*
1574 * write some zeros
1575 */
1576static int write_partial_skip(struct ceph_connection *con)
1577{
1578 int ret;
1579
1580 while (con->out_skip > 0) {
31739139 1581 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1582
e1dcb128 1583 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1584 if (ret <= 0)
1585 goto out;
1586 con->out_skip -= ret;
1587 }
1588 ret = 1;
1589out:
1590 return ret;
1591}
1592
1593/*
1594 * Prepare to read connection handshake, or an ack.
1595 */
eed0ef2c
SW		 1596static void prepare_read_banner(struct ceph_connection *con)
1597{
1598 dout("prepare_read_banner %p\n", con);
1599 con->in_base_pos = 0;
1600}
1601
31b8006e
SW		 1602static void prepare_read_connect(struct ceph_connection *con)
1603{
1604 dout("prepare_read_connect %p\n", con);
1605 con->in_base_pos = 0;
1606}
1607
1608static void prepare_read_ack(struct ceph_connection *con)
1609{
1610 dout("prepare_read_ack %p\n", con);
1611 con->in_base_pos = 0;
1612}
1613
3a23083b
SW		 1614static void prepare_read_seq(struct ceph_connection *con)
1615{
1616 dout("prepare_read_seq %p\n", con);
1617 con->in_base_pos = 0;
1618 con->in_tag = CEPH_MSGR_TAG_SEQ;
1619}
1620
31b8006e
SW		 1621static void prepare_read_tag(struct ceph_connection *con)
1622{
1623 dout("prepare_read_tag %p\n", con);
1624 con->in_base_pos = 0;
1625 con->in_tag = CEPH_MSGR_TAG_READY;
1626}
1627
1628/*
1629 * Prepare to read a message.
1630 */
1631static int prepare_read_message(struct ceph_connection *con)
1632{
1633 dout("prepare_read_message %p\n", con);
1634 BUG_ON(con->in_msg != NULL);
1635 con->in_base_pos = 0;
1636 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1637 return 0;
1638}
1639
1640
1641static int read_partial(struct ceph_connection *con,
fd51653f 1642 int end, int size, void *object)
31b8006e 1643{
e6cee71f
AE		 1644 while (con->in_base_pos < end) {
1645 int left = end - con->in_base_pos;
31b8006e
SW		 1646 int have = size - left;
1647 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1648 if (ret <= 0)
1649 return ret;
1650 con->in_base_pos += ret;
1651 }
1652 return 1;
1653}
1654
1655
1656/*
1657 * Read all or part of the connect-side handshake on a new connection
1658 */
eed0ef2c 1659static int read_partial_banner(struct ceph_connection *con)
31b8006e 1660{
fd51653f
AE		 1661 int size;
1662 int end;
1663 int ret;
31b8006e 1664
eed0ef2c 1665 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1666
1667 /* peer's banner */
fd51653f
AE		 1668 size = strlen(CEPH_BANNER);
1669 end = size;
1670 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1671 if (ret <= 0)
1672 goto out;
fd51653f
AE		 1673
1674 size = sizeof (con->actual_peer_addr);
1675 end += size;
1676 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1677 if (ret <= 0)
1678 goto out;
fd51653f
AE		 1679
1680 size = sizeof (con->peer_addr_for_me);
1681 end += size;
1682 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1683 if (ret <= 0)
1684 goto out;
fd51653f 1685
eed0ef2c
SW		 1686out:
1687 return ret;
1688}
1689
1690static int read_partial_connect(struct ceph_connection *con)
1691{
fd51653f
AE		 1692 int size;
1693 int end;
1694 int ret;
eed0ef2c
SW		 1695
1696 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1697
fd51653f
AE		 1698 size = sizeof (con->in_reply);
1699 end = size;
1700 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1701 if (ret <= 0)
1702 goto out;
fd51653f
AE		 1703
1704 size = le32_to_cpu(con->in_reply.authorizer_len);
1705 end += size;
1706 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1707 if (ret <= 0)
1708 goto out;
31b8006e 1709
4e7a5dcd
SW		 1710 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1711 con, (int)con->in_reply.tag,
1712 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1713 le32_to_cpu(con->in_reply.global_seq));
1714out:
1715 return ret;
eed0ef2c 1716
31b8006e
SW		 1717}
1718
1719/*
1720 * Verify the hello banner looks okay.
1721 */
1722static int verify_hello(struct ceph_connection *con)
1723{
1724 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1725 pr_err("connect to %s got bad banner\n",
3d14c5d2 1726 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1727 con->error_msg = "protocol error, bad banner";
1728 return -1;
1729 }
1730 return 0;
1731}
1732
1733static bool addr_is_blank(struct sockaddr_storage *ss)
1734{
1735 switch (ss->ss_family) {
1736 case AF_INET:
1737 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1738 case AF_INET6:
1739 return
1740 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1741 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1742 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1743 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1744 }
1745 return false;
1746}
1747
1748static int addr_port(struct sockaddr_storage *ss)
1749{
1750 switch (ss->ss_family) {
1751 case AF_INET:
f28bcfbe 1752 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1753 case AF_INET6:
f28bcfbe 1754 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1755 }
1756 return 0;
1757}
1758
1759static void addr_set_port(struct sockaddr_storage *ss, int p)
1760{
1761 switch (ss->ss_family) {
1762 case AF_INET:
1763 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1764 break;
31b8006e
SW		 1765 case AF_INET6:
1766 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1767 break;
31b8006e
SW		 1768 }
1769}
1770
ee3b56f2
NW		 1771/*
1772 * Unlike other *_pton function semantics, zero indicates success.
1773 */
1774static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1775 char delim, const char **ipend)
1776{
99f0f3b2
AE		 1777 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1778 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1779
1780 memset(ss, 0, sizeof(*ss));
1781
1782 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1783 ss->ss_family = AF_INET;
1784 return 0;
1785 }
1786
1787 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1788 ss->ss_family = AF_INET6;
1789 return 0;
1790 }
1791
1792 return -EINVAL;
1793}
1794
1795/*
1796 * Extract hostname string and resolve using kernel DNS facility.
1797 */
1798#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1799static int ceph_dns_resolve_name(const char *name, size_t namelen,
1800 struct sockaddr_storage *ss, char delim, const char **ipend)
1801{
1802 const char *end, *delim_p;
1803 char *colon_p, *ip_addr = NULL;
1804 int ip_len, ret;
1805
1806 /*
1807 * The end of the hostname occurs immediately preceding the delimiter or
1808 * the port marker (':') where the delimiter takes precedence.
1809 */
1810 delim_p = memchr(name, delim, namelen);
1811 colon_p = memchr(name, ':', namelen);
1812
1813 if (delim_p && colon_p)
1814 end = delim_p < colon_p ? delim_p : colon_p;
1815 else if (!delim_p && colon_p)
1816 end = colon_p;
1817 else {
1818 end = delim_p;
1819 if (!end) /* case: hostname:/ */
1820 end = name + namelen;
1821 }
1822
1823 if (end <= name)
1824 return -EINVAL;
1825
1826 /* do dns_resolve upcall */
1827 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1828 if (ip_len > 0)
1829 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1830 else
1831 ret = -ESRCH;
1832
1833 kfree(ip_addr);
1834
1835 *ipend = end;
1836
1837 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1838 ret, ret ? "failed" : ceph_pr_addr(ss));
1839
1840 return ret;
1841}
1842#else
1843static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1844 struct sockaddr_storage *ss, char delim, const char **ipend)
1845{
1846 return -EINVAL;
1847}
1848#endif
1849
1850/*
1851 * Parse a server name (IP or hostname). If a valid IP address is not found
1852 * then try to extract a hostname to resolve using userspace DNS upcall.
1853 */
1854static int ceph_parse_server_name(const char *name, size_t namelen,
1855 struct sockaddr_storage *ss, char delim, const char **ipend)
1856{
1857 int ret;
1858
1859 ret = ceph_pton(name, namelen, ss, delim, ipend);
1860 if (ret)
1861 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1862
1863 return ret;
1864}
1865
31b8006e
SW		 1866/*
1867 * Parse an ip[:port] list into an addr array. Use the default
1868 * monitor port if a port isn't specified.
1869 */
1870int ceph_parse_ips(const char *c, const char *end,
1871 struct ceph_entity_addr *addr,
1872 int max_count, int *count)
1873{
ee3b56f2 1874 int i, ret = -EINVAL;
31b8006e
SW		 1875 const char *p = c;
1876
1877 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1878 for (i = 0; i < max_count; i++) {
1879 const char *ipend;
1880 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1881 int port;
39139f64
SW		 1882 char delim = ',';
1883
1884 if (*p == '[') {
1885 delim = ']';
1886 p++;
1887 }
31b8006e 1888
ee3b56f2
NW		 1889 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1890 if (ret)
31b8006e 1891 goto bad;
ee3b56f2
NW		 1892 ret = -EINVAL;
1893
31b8006e
SW		 1894 p = ipend;
1895
39139f64
SW		 1896 if (delim == ']') {
1897 if (*p != ']') {
1898 dout("missing matching ']'\n");
1899 goto bad;
1900 }
1901 p++;
1902 }
1903
31b8006e
SW		 1904 /* port? */
1905 if (p < end && *p == ':') {
1906 port = 0;
1907 p++;
1908 while (p < end && *p >= '0' && *p <= '9') {
1909 port = (port * 10) + (*p - '0');
1910 p++;
1911 }
f48db1e9
ID		 1912 if (port == 0)
1913 port = CEPH_MON_PORT;
1914 else if (port > 65535)
31b8006e
SW		 1915 goto bad;
1916 } else {
1917 port = CEPH_MON_PORT;
1918 }
1919
1920 addr_set_port(ss, port);
1921
3d14c5d2 1922 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1923
1924 if (p == end)
1925 break;
1926 if (*p != ',')
1927 goto bad;
1928 p++;
1929 }
1930
1931 if (p != end)
1932 goto bad;
1933
1934 if (count)
1935 *count = i + 1;
1936 return 0;
1937
1938bad:
39139f64 1939 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1940 return ret;
31b8006e 1941}
3d14c5d2 1942EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1943
eed0ef2c 1944static int process_banner(struct ceph_connection *con)
31b8006e 1945{
eed0ef2c 1946 dout("process_banner on %p\n", con);
31b8006e
SW		 1947
1948 if (verify_hello(con) < 0)
1949 return -1;
1950
63f2d211
SW		 1951 ceph_decode_addr(&con->actual_peer_addr);
1952 ceph_decode_addr(&con->peer_addr_for_me);
1953
31b8006e
SW		 1954 /*
1955 * Make sure the other end is who we wanted. note that the other
1956 * end may not yet know their ip address, so if it's 0.0.0.0, give
1957 * them the benefit of the doubt.
1958 */
103e2d3a
SW		 1959 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1960 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1961 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1962 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 1963 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1964 ceph_pr_addr(&con->peer_addr.in_addr),
1965 (int)le32_to_cpu(con->peer_addr.nonce),
1966 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1967 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1968 con->error_msg = "wrong peer at address";
31b8006e
SW		 1969 return -1;
1970 }
1971
1972 /*
1973 * did we learn our address?
1974 */
1975 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1976 int port = addr_port(&con->msgr->inst.addr.in_addr);
1977
1978 memcpy(&con->msgr->inst.addr.in_addr,
1979 &con->peer_addr_for_me.in_addr,
1980 sizeof(con->peer_addr_for_me.in_addr));
1981 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1982 encode_my_addr(con->msgr);
eed0ef2c 1983 dout("process_banner learned my addr is %s\n",
3d14c5d2 1984 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1985 }
1986
eed0ef2c
SW		 1987 return 0;
1988}
1989
1990static int process_connect(struct ceph_connection *con)
1991{
3d14c5d2
YS		 1992 u64 sup_feat = con->msgr->supported_features;
1993 u64 req_feat = con->msgr->required_features;
2b3e0c90
ID		 1994 u64 server_feat = ceph_sanitize_features(
1995 le64_to_cpu(con->in_reply.features));
0da5d703 1996 int ret;
04a419f9 1997
eed0ef2c
SW		 1998 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1999
31b8006e 2000 switch (con->in_reply.tag) {
04a419f9
SW		 2001 case CEPH_MSGR_TAG_FEATURES:
2002 pr_err("%s%lld %s feature set mismatch,"
2003 " my %llx < server's %llx, missing %llx\n",
2004 ENTITY_NAME(con->peer_name),
3d14c5d2 2005 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2006 sup_feat, server_feat, server_feat & ~sup_feat);
2007 con->error_msg = "missing required protocol features";
0fa6ebc6 2008 reset_connection(con);
04a419f9
SW		 2009 return -1;
2010
31b8006e 2011 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2012 pr_err("%s%lld %s protocol version mismatch,"
2013 " my %d != server's %d\n",
2014 ENTITY_NAME(con->peer_name),
3d14c5d2 2015 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2016 le32_to_cpu(con->out_connect.protocol_version),
2017 le32_to_cpu(con->in_reply.protocol_version));
2018 con->error_msg = "protocol version mismatch";
0fa6ebc6 2019 reset_connection(con);
31b8006e
SW		 2020 return -1;
2021
4e7a5dcd
SW		 2022 case CEPH_MSGR_TAG_BADAUTHORIZER:
2023 con->auth_retry++;
2024 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2025 con->auth_retry);
2026 if (con->auth_retry == 2) {
2027 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2028 return -1;
2029 }
6d4221b5 2030 con_out_kvec_reset(con);
e825a66d 2031 ret = prepare_write_connect(con);
0da5d703
SW		 2032 if (ret < 0)
2033 return ret;
63733a0f 2034 prepare_read_connect(con);
4e7a5dcd 2035 break;
31b8006e
SW		 2036
2037 case CEPH_MSGR_TAG_RESETSESSION:
2038 /*
2039 * If we connected with a large connect_seq but the peer
2040 * has no record of a session with us (no connection, or
2041 * connect_seq == 0), they will send RESETSESION to indicate
2042 * that they must have reset their session, and may have
2043 * dropped messages.
2044 */
2045 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2046 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2047 pr_err("%s%lld %s connection reset\n",
2048 ENTITY_NAME(con->peer_name),
3d14c5d2 2049 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2050 reset_connection(con);
6d4221b5 2051 con_out_kvec_reset(con);
5a0f8fdd
AE		 2052 ret = prepare_write_connect(con);
2053 if (ret < 0)
2054 return ret;
31b8006e
SW		 2055 prepare_read_connect(con);
2056
2057 /* Tell ceph about it. */
ec302645 2058 mutex_unlock(&con->mutex);
31b8006e
SW		 2059 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2060 if (con->ops->peer_reset)
2061 con->ops->peer_reset(con);
ec302645 2062 mutex_lock(&con->mutex);
8dacc7da 2063 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2064 return -EAGAIN;
31b8006e
SW		 2065 break;
2066
2067 case CEPH_MSGR_TAG_RETRY_SESSION:
2068 /*
2069 * If we sent a smaller connect_seq than the peer has, try
2070 * again with a larger value.
2071 */
5bdca4e0 2072 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2073 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2074 le32_to_cpu(con->in_reply.connect_seq));
2075 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2076 con_out_kvec_reset(con);
5a0f8fdd
AE		 2077 ret = prepare_write_connect(con);
2078 if (ret < 0)
2079 return ret;
31b8006e
SW		 2080 prepare_read_connect(con);
2081 break;
2082
2083 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2084 /*
2085 * If we sent a smaller global_seq than the peer has, try
2086 * again with a larger value.
2087 */
eed0ef2c 2088 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2089 con->peer_global_seq,
5bdca4e0 2090 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2091 get_global_seq(con->msgr,
5bdca4e0 2092 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2093 con_out_kvec_reset(con);
5a0f8fdd
AE		 2094 ret = prepare_write_connect(con);
2095 if (ret < 0)
2096 return ret;
31b8006e
SW		 2097 prepare_read_connect(con);
2098 break;
2099
3a23083b 2100 case CEPH_MSGR_TAG_SEQ:
31b8006e 2101 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2102 if (req_feat & ~server_feat) {
2103 pr_err("%s%lld %s protocol feature mismatch,"
2104 " my required %llx > server's %llx, need %llx\n",
2105 ENTITY_NAME(con->peer_name),
3d14c5d2 2106 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2107 req_feat, server_feat, req_feat & ~server_feat);
2108 con->error_msg = "missing required protocol features";
0fa6ebc6 2109 reset_connection(con);
04a419f9
SW		 2110 return -1;
2111 }
8dacc7da 2112
122070a2 2113 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2114 con->state = CON_STATE_OPEN;
20e55c4c 2115 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2116 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2117 con->connect_seq++;
aba558e2 2118 con->peer_features = server_feat;
31b8006e
SW		 2119 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2120 con->peer_global_seq,
2121 le32_to_cpu(con->in_reply.connect_seq),
2122 con->connect_seq);
2123 WARN_ON(con->connect_seq !=
2124 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2125
2126 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2127 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2128
85effe18 2129 con->delay = 0; /* reset backoff memory */
92ac41d0 2130
3a23083b
SW		 2131 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2132 prepare_write_seq(con);
2133 prepare_read_seq(con);
2134 } else {
2135 prepare_read_tag(con);
2136 }
31b8006e
SW		 2137 break;
2138
2139 case CEPH_MSGR_TAG_WAIT:
2140 /*
2141 * If there is a connection race (we are opening
2142 * connections to each other), one of us may just have
2143 * to WAIT. This shouldn't happen if we are the
2144 * client.
2145 */
04177882
SW		 2146 con->error_msg = "protocol error, got WAIT as client";
2147 return -1;
31b8006e
SW		 2148
2149 default:
31b8006e
SW		 2150 con->error_msg = "protocol error, garbage tag during connect";
2151 return -1;
2152 }
2153 return 0;
2154}
2155
2156
2157/*
2158 * read (part of) an ack
2159 */
2160static int read_partial_ack(struct ceph_connection *con)
2161{
fd51653f
AE		 2162 int size = sizeof (con->in_temp_ack);
2163 int end = size;
31b8006e 2164
fd51653f 2165 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2166}
2167
31b8006e
SW		 2168/*
2169 * We can finally discard anything that's been acked.
2170 */
2171static void process_ack(struct ceph_connection *con)
2172{
2173 struct ceph_msg *m;
2174 u64 ack = le64_to_cpu(con->in_temp_ack);
2175 u64 seq;
2176
31b8006e
SW		 2177 while (!list_empty(&con->out_sent)) {
2178 m = list_first_entry(&con->out_sent, struct ceph_msg,
2179 list_head);
2180 seq = le64_to_cpu(m->hdr.seq);
2181 if (seq > ack)
2182 break;
2183 dout("got ack for seq %llu type %d at %p\n", seq,
2184 le16_to_cpu(m->hdr.type), m);
4cf9d544 2185 m->ack_stamp = jiffies;
31b8006e
SW		 2186 ceph_msg_remove(m);
2187 }
31b8006e
SW		 2188 prepare_read_tag(con);
2189}
2190
2191
2450418c 2192static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2193 struct kvec *section,
2194 unsigned int sec_len, u32 *crc)
2450418c 2195{
68b4476b 2196 int ret, left;
2450418c
YS		 2197
2198 BUG_ON(!section);
2199
2200 while (section->iov_len < sec_len) {
2201 BUG_ON(section->iov_base == NULL);
2202 left = sec_len - section->iov_len;
2203 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2204 section->iov_len, left);
2205 if (ret <= 0)
2206 return ret;
2207 section->iov_len += ret;
2450418c 2208 }
fe3ad593
AE		 2209 if (section->iov_len == sec_len)
2210 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2211
2450418c
YS		 2212 return 1;
2213}
31b8006e 2214
34d2d200
AE		 2215static int read_partial_msg_data(struct ceph_connection *con)
2216{
2217 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2218 struct ceph_msg_data_cursor *cursor = &msg->cursor;
34d2d200 2219 const bool do_datacrc = !con->msgr->nocrc;
686be208
AE		 2220 struct page *page;
2221 size_t page_offset;
2222 size_t length;
f5db90bc 2223 u32 crc = 0;
34d2d200
AE		 2224 int ret;
2225
2226 BUG_ON(!msg);
5240d9f9 2227 if (list_empty(&msg->data))
4c59b4a2 2228 return -EIO;
34d2d200 2229
f5db90bc
AE		 2230 if (do_datacrc)
2231 crc = con->in_data_crc;
643c68a4 2232 while (cursor->resid) {
8ae4f4f5 2233 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 2234 NULL);
686be208 2235 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2236 if (ret <= 0) {
2237 if (do_datacrc)
2238 con->in_data_crc = crc;
2239
686be208 2240 return ret;
f5db90bc 2241 }
686be208
AE		 2242
2243 if (do_datacrc)
f5db90bc 2244 crc = ceph_crc32c_page(crc, page, page_offset, ret);
8ae4f4f5 2245 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
34d2d200 2246 }
f5db90bc
AE		 2247 if (do_datacrc)
2248 con->in_data_crc = crc;
34d2d200
AE		 2249
2250 return 1; /* must return > 0 to indicate success */
2251}
2252
31b8006e
SW		 2253/*
2254 * read (part of) a message.
2255 */
686be208
AE		 2256static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2257
31b8006e
SW		 2258static int read_partial_message(struct ceph_connection *con)
2259{
2260 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2261 int size;
2262 int end;
31b8006e 2263 int ret;
95c96174 2264 unsigned int front_len, middle_len, data_len;
37675b0f 2265 bool do_datacrc = !con->msgr->nocrc;
33d07337 2266 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
ae18756b 2267 u64 seq;
fe3ad593 2268 u32 crc;
31b8006e
SW		 2269
2270 dout("read_partial_message con %p msg %p\n", con, m);
2271
2272 /* header */
fd51653f
AE		 2273 size = sizeof (con->in_hdr);
2274 end = size;
2275 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2276 if (ret <= 0)
2277 return ret;
fe3ad593
AE		 2278
2279 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2280 if (cpu_to_le32(crc) != con->in_hdr.crc) {
67c64eb7 2281 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
fe3ad593
AE		 2282 crc, con->in_hdr.crc);
2283 return -EBADMSG;
2284 }
2285
31b8006e
SW		 2286 front_len = le32_to_cpu(con->in_hdr.front_len);
2287 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2288 return -EIO;
2289 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2290 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2291 return -EIO;
2292 data_len = le32_to_cpu(con->in_hdr.data_len);
2293 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2294 return -EIO;
2295
ae18756b
SW		 2296 /* verify seq# */
2297 seq = le64_to_cpu(con->in_hdr.seq);
2298 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2299 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2300 ENTITY_NAME(con->peer_name),
3d14c5d2 2301 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2302 seq, con->in_seq + 1);
2303 con->in_base_pos = -front_len - middle_len - data_len -
2304 sizeof(m->footer);
2305 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2306 return 0;
2307 } else if ((s64)seq - (s64)con->in_seq > 1) {
2308 pr_err("read_partial_message bad seq %lld expected %lld\n",
2309 seq, con->in_seq + 1);
2310 con->error_msg = "bad message sequence # for incoming message";
67c64eb7 2311 return -EBADE;
ae18756b
SW		 2312 }
2313
31b8006e
SW		 2314 /* allocate message? */
2315 if (!con->in_msg) {
4740a623
SW		 2316 int skip = 0;
2317
31b8006e 2318 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2319 front_len, data_len);
4740a623
SW		 2320 ret = ceph_con_in_msg_alloc(con, &skip);
2321 if (ret < 0)
2322 return ret;
f759ebb9
AE		 2323
2324 BUG_ON(!con->in_msg ^ skip);
2325 if (con->in_msg && data_len > con->in_msg->data_length) {
b9a67899 2326 pr_warn("%s skipping long message (%u > %zd)\n",
f759ebb9
AE		 2327 __func__, data_len, con->in_msg->data_length);
2328 ceph_msg_put(con->in_msg);
2329 con->in_msg = NULL;
2330 skip = 1;
2331 }
2450418c 2332 if (skip) {
31b8006e 2333 /* skip this message */
a79832f2 2334 dout("alloc_msg said skip message\n");
31b8006e
SW		 2335 con->in_base_pos = -front_len - middle_len - data_len -
2336 sizeof(m->footer);
2337 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2338 con->in_seq++;
31b8006e
SW		 2339 return 0;
2340 }
38941f80 2341
4740a623 2342 BUG_ON(!con->in_msg);
38941f80 2343 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2344 m = con->in_msg;
2345 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2346 if (m->middle)
2347 m->middle->vec.iov_len = 0;
9d7f0f13 2348
78625051 2349 /* prepare for data payload, if any */
a4107026 2350
78625051 2351 if (data_len)
98fa5dd8 2352 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2353 }
2354
2355 /* front */
2450418c
YS		 2356 ret = read_partial_message_section(con, &m->front, front_len,
2357 &con->in_front_crc);
2358 if (ret <= 0)
2359 return ret;
31b8006e
SW		 2360
2361 /* middle */
2450418c 2362 if (m->middle) {
213c99ee
SW		 2363 ret = read_partial_message_section(con, &m->middle->vec,
2364 middle_len,
2450418c 2365 &con->in_middle_crc);
31b8006e
SW		 2366 if (ret <= 0)
2367 return ret;
31b8006e
SW		 2368 }
2369
2370 /* (page) data */
34d2d200
AE		 2371 if (data_len) {
2372 ret = read_partial_msg_data(con);
2373 if (ret <= 0)
2374 return ret;
31b8006e
SW		 2375 }
2376
31b8006e 2377 /* footer */
33d07337
YZ		 2378 if (need_sign)
2379 size = sizeof(m->footer);
2380 else
2381 size = sizeof(m->old_footer);
2382
fd51653f
AE		 2383 end += size;
2384 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2385 if (ret <= 0)
2386 return ret;
2387
33d07337
YZ		 2388 if (!need_sign) {
2389 m->footer.flags = m->old_footer.flags;
2390 m->footer.sig = 0;
2391 }
2392
31b8006e
SW		 2393 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2394 m, front_len, m->footer.front_crc, middle_len,
2395 m->footer.middle_crc, data_len, m->footer.data_crc);
2396
2397 /* crc ok? */
2398 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2399 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2400 m, con->in_front_crc, m->footer.front_crc);
2401 return -EBADMSG;
2402 }
2403 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2404 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2405 m, con->in_middle_crc, m->footer.middle_crc);
2406 return -EBADMSG;
2407 }
bca064d2 2408 if (do_datacrc &&
31b8006e
SW		 2409 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2410 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2411 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2412 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2413 return -EBADMSG;
2414 }
2415
33d07337
YZ		 2416 if (need_sign && con->ops->check_message_signature &&
2417 con->ops->check_message_signature(con, m)) {
2418 pr_err("read_partial_message %p signature check failed\n", m);
2419 return -EBADMSG;
2420 }
2421
31b8006e
SW		 2422 return 1; /* done! */
2423}
2424
2425/*
2426 * Process message. This happens in the worker thread. The callback should
2427 * be careful not to do anything that waits on other incoming messages or it
2428 * may deadlock.
2429 */
2430static void process_message(struct ceph_connection *con)
2431{
5e095e8b 2432 struct ceph_msg *msg;
31b8006e 2433
38941f80
AE		 2434 BUG_ON(con->in_msg->con != con);
2435 con->in_msg->con = NULL;
5e095e8b 2436 msg = con->in_msg;
31b8006e 2437 con->in_msg = NULL;
36eb71aa 2438 con->ops->put(con);
31b8006e
SW		 2439
2440 /* if first message, set peer_name */
2441 if (con->peer_name.type == 0)
dbad185d 2442 con->peer_name = msg->hdr.src;
31b8006e 2443
31b8006e 2444 con->in_seq++;
ec302645 2445 mutex_unlock(&con->mutex);
31b8006e
SW		 2446
2447 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2448 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2449 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2450 le16_to_cpu(msg->hdr.type),
2451 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2452 le32_to_cpu(msg->hdr.front_len),
2453 le32_to_cpu(msg->hdr.data_len),
2454 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2455 con->ops->dispatch(con, msg);
ec302645
SW		 2456
2457 mutex_lock(&con->mutex);
31b8006e
SW		 2458}
2459
2460
2461/*
2462 * Write something to the socket. Called in a worker thread when the
2463 * socket appears to be writeable and we have something ready to send.
2464 */
2465static int try_write(struct ceph_connection *con)
2466{
31b8006e
SW		 2467 int ret = 1;
2468
d59315ca 2469 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2470
31b8006e
SW		 2471more:
2472 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2473
2474 /* open the socket first? */
8dacc7da
SW		 2475 if (con->state == CON_STATE_PREOPEN) {
2476 BUG_ON(con->sock);
2477 con->state = CON_STATE_CONNECTING;
a5988c49 2478
e2200423 2479 con_out_kvec_reset(con);
e825a66d 2480 prepare_write_banner(con);
eed0ef2c 2481 prepare_read_banner(con);
31b8006e 2482
cf3e5c40 2483 BUG_ON(con->in_msg);
31b8006e
SW		 2484 con->in_tag = CEPH_MSGR_TAG_READY;
2485 dout("try_write initiating connect on %p new state %lu\n",
2486 con, con->state);
41617d0c
AE		 2487 ret = ceph_tcp_connect(con);
2488 if (ret < 0) {
31b8006e 2489 con->error_msg = "connect error";
31b8006e
SW		 2490 goto out;
2491 }
2492 }
2493
2494more_kvec:
2495 /* kvec data queued? */
2496 if (con->out_skip) {
2497 ret = write_partial_skip(con);
2498 if (ret <= 0)
42961d23 2499 goto out;
31b8006e
SW		 2500 }
2501 if (con->out_kvec_left) {
2502 ret = write_partial_kvec(con);
2503 if (ret <= 0)
42961d23 2504 goto out;
31b8006e
SW		 2505 }
2506
2507 /* msg pages? */
2508 if (con->out_msg) {
c86a2930
SW		 2509 if (con->out_msg_done) {
2510 ceph_msg_put(con->out_msg);
2511 con->out_msg = NULL; /* we're done with this one */
2512 goto do_next;
2513 }
2514
34d2d200 2515 ret = write_partial_message_data(con);
31b8006e
SW		 2516 if (ret == 1)
2517 goto more_kvec; /* we need to send the footer, too! */
2518 if (ret == 0)
42961d23 2519 goto out;
31b8006e 2520 if (ret < 0) {
34d2d200 2521 dout("try_write write_partial_message_data err %d\n",
31b8006e 2522 ret);
42961d23 2523 goto out;
31b8006e
SW		 2524 }
2525 }
2526
c86a2930 2527do_next:
8dacc7da 2528 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2529 /* is anything else pending? */
2530 if (!list_empty(&con->out_queue)) {
2531 prepare_write_message(con);
2532 goto more;
2533 }
2534 if (con->in_seq > con->in_seq_acked) {
2535 prepare_write_ack(con);
2536 goto more;
2537 }
c9ffc77a 2538 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2539 prepare_write_keepalive(con);
2540 goto more;
2541 }
2542 }
2543
2544 /* Nothing to do! */
c9ffc77a 2545 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2546 dout("try_write nothing else to write.\n");
31b8006e
SW		 2547 ret = 0;
2548out:
42961d23 2549 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2550 return ret;
2551}
2552
2553
2554
2555/*
2556 * Read what we can from the socket.
2557 */
2558static int try_read(struct ceph_connection *con)
2559{
31b8006e
SW		 2560 int ret = -1;
2561
8dacc7da
SW		 2562more:
2563 dout("try_read start on %p state %lu\n", con, con->state);
2564 if (con->state != CON_STATE_CONNECTING &&
2565 con->state != CON_STATE_NEGOTIATING &&
2566 con->state != CON_STATE_OPEN)
31b8006e
SW		 2567 return 0;
2568
8dacc7da 2569 BUG_ON(!con->sock);
ec302645 2570
31b8006e
SW		 2571 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2572 con->in_base_pos);
0da5d703 2573
8dacc7da 2574 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2575 dout("try_read connecting\n");
2576 ret = read_partial_banner(con);
2577 if (ret <= 0)
ab166d5a 2578 goto out;
7593af92
AE		 2579 ret = process_banner(con);
2580 if (ret < 0)
2581 goto out;
2582
8dacc7da 2583 con->state = CON_STATE_NEGOTIATING;
7593af92 2584
6d4221b5
JS		 2585 /*
2586 * Received banner is good, exchange connection info.
2587 * Do not reset out_kvec, as sending our banner raced
2588 * with receiving peer banner after connect completed.
2589 */
7593af92
AE		 2590 ret = prepare_write_connect(con);
2591 if (ret < 0)
2592 goto out;
2593 prepare_read_connect(con);
2594
2595 /* Send connection info before awaiting response */
0da5d703
SW		 2596 goto out;
2597 }
2598
8dacc7da 2599 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2600 dout("try_read negotiating\n");
31b8006e
SW		 2601 ret = read_partial_connect(con);
2602 if (ret <= 0)
31b8006e 2603 goto out;
98bdb0aa
SW		 2604 ret = process_connect(con);
2605 if (ret < 0)
2606 goto out;
31b8006e
SW		 2607 goto more;
2608 }
2609
122070a2 2610 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2611
31b8006e
SW		 2612 if (con->in_base_pos < 0) {
2613 /*
2614 * skipping + discarding content.
2615 *
2616 * FIXME: there must be a better way to do this!
2617 */
84495f49
AE		 2618 static char buf[SKIP_BUF_SIZE];
2619 int skip = min((int) sizeof (buf), -con->in_base_pos);
2620
31b8006e
SW		 2621 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2622 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2623 if (ret <= 0)
98bdb0aa 2624 goto out;
31b8006e
SW		 2625 con->in_base_pos += ret;
2626 if (con->in_base_pos)
2627 goto more;
2628 }
2629 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2630 /*
2631 * what's next?
2632 */
2633 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2634 if (ret <= 0)
98bdb0aa 2635 goto out;
31b8006e
SW		 2636 dout("try_read got tag %d\n", (int)con->in_tag);
2637 switch (con->in_tag) {
2638 case CEPH_MSGR_TAG_MSG:
2639 prepare_read_message(con);
2640 break;
2641 case CEPH_MSGR_TAG_ACK:
2642 prepare_read_ack(con);
2643 break;
2644 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2645 con_close_socket(con);
2646 con->state = CON_STATE_CLOSED;
98bdb0aa 2647 goto out;
31b8006e
SW		 2648 default:
2649 goto bad_tag;
2650 }
2651 }
2652 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2653 ret = read_partial_message(con);
2654 if (ret <= 0) {
2655 switch (ret) {
2656 case -EBADMSG:
2657 con->error_msg = "bad crc";
67c64eb7
ID		 2658 /* fall through */
2659 case -EBADE:
31b8006e 2660 ret = -EIO;
98bdb0aa 2661 break;
31b8006e
SW		 2662 case -EIO:
2663 con->error_msg = "io error";
98bdb0aa 2664 break;
31b8006e 2665 }
98bdb0aa 2666 goto out;
31b8006e
SW		 2667 }
2668 if (con->in_tag == CEPH_MSGR_TAG_READY)
2669 goto more;
2670 process_message(con);
7b862e07
SW		 2671 if (con->state == CON_STATE_OPEN)
2672 prepare_read_tag(con);
31b8006e
SW		 2673 goto more;
2674 }
3a23083b
SW		 2675 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2676 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2677 /*
2678 * the final handshake seq exchange is semantically
2679 * equivalent to an ACK
2680 */
31b8006e
SW		 2681 ret = read_partial_ack(con);
2682 if (ret <= 0)
98bdb0aa 2683 goto out;
31b8006e
SW		 2684 process_ack(con);
2685 goto more;
2686 }
2687
31b8006e 2688out:
98bdb0aa 2689 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2690 return ret;
2691
2692bad_tag:
2693 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2694 con->error_msg = "protocol error, garbage tag";
2695 ret = -1;
2696 goto out;
2697}
2698
2699
2700/*
802c6d96
AE		 2701 * Atomically queue work on a connection after the specified delay.
2702 * Bump @con reference to avoid races with connection teardown.
2703 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2704 */
802c6d96 2705static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2706{
31b8006e 2707 if (!con->ops->get(con)) {
802c6d96 2708 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2709 return -ENOENT;
31b8006e
SW		 2710 }
2711
802c6d96
AE		 2712 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2713 dout("%s %p - already queued\n", __func__, con);
31b8006e 2714 con->ops->put(con);
802c6d96 2715 return -EBUSY;
31b8006e 2716 }
802c6d96
AE		 2717
2718 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2719 return 0;
2720}
2721
2722static void queue_con(struct ceph_connection *con)
2723{
2724 (void) queue_con_delay(con, 0);
31b8006e
SW		 2725}
2726
37ab77ac
ID		 2727static void cancel_con(struct ceph_connection *con)
2728{
2729 if (cancel_delayed_work(&con->work)) {
2730 dout("%s %p\n", __func__, con);
2731 con->ops->put(con);
2732 }
2733}
2734
7bb21d68
AE		 2735static bool con_sock_closed(struct ceph_connection *con)
2736{
c9ffc77a 2737 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2738 return false;
2739
2740#define CASE(x) \
2741 case CON_STATE_ ## x: \
2742 con->error_msg = "socket closed (con state " #x ")"; \
2743 break;
2744
2745 switch (con->state) {
2746 CASE(CLOSED);
2747 CASE(PREOPEN);
2748 CASE(CONNECTING);
2749 CASE(NEGOTIATING);
2750 CASE(OPEN);
2751 CASE(STANDBY);
2752 default:
b9a67899 2753 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2754 __func__, con, con->state);
2755 con->error_msg = "unrecognized con state";
2756 BUG();
2757 break;
2758 }
2759#undef CASE
2760
2761 return true;
2762}
2763
f20a39fd
AE		 2764static bool con_backoff(struct ceph_connection *con)
2765{
2766 int ret;
2767
2768 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2769 return false;
2770
2771 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2772 if (ret) {
2773 dout("%s: con %p FAILED to back off %lu\n", __func__,
2774 con, con->delay);
2775 BUG_ON(ret == -ENOENT);
2776 con_flag_set(con, CON_FLAG_BACKOFF);
2777 }
2778
2779 return true;
2780}
2781
93209264
AE		 2782/* Finish fault handling; con->mutex must *not* be held here */
2783
2784static void con_fault_finish(struct ceph_connection *con)
2785{
2786 /*
2787 * in case we faulted due to authentication, invalidate our
2788 * current tickets so that we can get new ones.
2789 */
2790 if (con->auth_retry && con->ops->invalidate_authorizer) {
2791 dout("calling invalidate_authorizer()\n");
2792 con->ops->invalidate_authorizer(con);
2793 }
2794
2795 if (con->ops->fault)
2796 con->ops->fault(con);
2797}
2798
31b8006e
SW		 2799/*
2800 * Do some work on a connection. Drop a connection ref when we're done.
2801 */
2802static void con_work(struct work_struct *work)
2803{
2804 struct ceph_connection *con = container_of(work, struct ceph_connection,
2805 work.work);
49659416 2806 bool fault;
31b8006e 2807
9dd4658d 2808 mutex_lock(&con->mutex);
49659416
AE		 2809 while (true) {
2810 int ret;
31b8006e 2811
49659416
AE		 2812 if ((fault = con_sock_closed(con))) {
2813 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2814 break;
2815 }
2816 if (con_backoff(con)) {
2817 dout("%s: con %p BACKOFF\n", __func__, con);
2818 break;
2819 }
2820 if (con->state == CON_STATE_STANDBY) {
2821 dout("%s: con %p STANDBY\n", __func__, con);
2822 break;
2823 }
2824 if (con->state == CON_STATE_CLOSED) {
2825 dout("%s: con %p CLOSED\n", __func__, con);
2826 BUG_ON(con->sock);
2827 break;
2828 }
2829 if (con->state == CON_STATE_PREOPEN) {
2830 dout("%s: con %p PREOPEN\n", __func__, con);
2831 BUG_ON(con->sock);
2832 }
0da5d703 2833
49659416
AE		 2834 ret = try_read(con);
2835 if (ret < 0) {
2836 if (ret == -EAGAIN)
2837 continue;
67c64eb7
ID		 2838 if (!con->error_msg)
2839 con->error_msg = "socket error on read";
49659416
AE		 2840 fault = true;
2841 break;
2842 }
2843
2844 ret = try_write(con);
2845 if (ret < 0) {
2846 if (ret == -EAGAIN)
2847 continue;
67c64eb7
ID		 2848 if (!con->error_msg)
2849 con->error_msg = "socket error on write";
49659416
AE		 2850 fault = true;
2851 }
2852
2853 break; /* If we make it to here, we're done */
3a140a0d 2854 }
b6e7b6a1
AE		 2855 if (fault)
2856 con_fault(con);
9dd4658d 2857 mutex_unlock(&con->mutex);
0da5d703 2858
b6e7b6a1
AE		 2859 if (fault)
2860 con_fault_finish(con);
2861
2862 con->ops->put(con);
31b8006e
SW		 2863}
2864
31b8006e
SW		 2865/*
2866 * Generic error/fault handler. A retry mechanism is used with
2867 * exponential backoff
2868 */
93209264 2869static void con_fault(struct ceph_connection *con)
31b8006e 2870{
31b8006e 2871 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2872 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2873
67c64eb7
ID		 2874 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2875 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2876 con->error_msg = NULL;
2877
122070a2 2878 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2879 con->state != CON_STATE_NEGOTIATING &&
2880 con->state != CON_STATE_OPEN);
ec302645 2881
31b8006e 2882 con_close_socket(con);
5e095e8b 2883
c9ffc77a 2884 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2885 dout("fault on LOSSYTX channel, marking CLOSED\n");
2886 con->state = CON_STATE_CLOSED;
93209264 2887 return;
3b5ede07
SW		 2888 }
2889
5e095e8b 2890 if (con->in_msg) {
38941f80
AE		 2891 BUG_ON(con->in_msg->con != con);
2892 con->in_msg->con = NULL;
5e095e8b
SW		 2893 ceph_msg_put(con->in_msg);
2894 con->in_msg = NULL;
36eb71aa 2895 con->ops->put(con);
5e095e8b 2896 }
31b8006e 2897
e80a52d1
SW		 2898 /* Requeue anything that hasn't been acked */
2899 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2900
e76661d0
SW		 2901 /* If there are no messages queued or keepalive pending, place
2902 * the connection in a STANDBY state */
2903 if (list_empty(&con->out_queue) &&
c9ffc77a 2904 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2905 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2906 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2907 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2908 } else {
2909 /* retry after a delay. */
8dacc7da 2910 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2911 if (con->delay == 0)
2912 con->delay = BASE_DELAY_INTERVAL;
2913 else if (con->delay < MAX_DELAY_INTERVAL)
2914 con->delay *= 2;
c9ffc77a 2915 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2916 queue_con(con);
31b8006e 2917 }
31b8006e
SW		 2918}
2919
2920
2921
2922/*
15d9882c 2923 * initialize a new messenger instance
31b8006e 2924 */
15d9882c
AE		 2925void ceph_messenger_init(struct ceph_messenger *msgr,
2926 struct ceph_entity_addr *myaddr,
12b4629a
ID		 2927 u64 supported_features,
2928 u64 required_features,
ba988f87
CH		 2929 bool nocrc,
2930 bool tcp_nodelay)
31b8006e 2931{
3d14c5d2
YS		 2932 msgr->supported_features = supported_features;
2933 msgr->required_features = required_features;
2934
31b8006e
SW		 2935 spin_lock_init(&msgr->global_seq_lock);
2936
31b8006e
SW		 2937 if (myaddr)
2938 msgr->inst.addr = *myaddr;
2939
2940 /* select a random nonce */
ac8839d7 2941 msgr->inst.addr.type = 0;
103e2d3a 2942 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2943 encode_my_addr(msgr);
15d9882c 2944 msgr->nocrc = nocrc;
ba988f87 2945 msgr->tcp_nodelay = tcp_nodelay;
31b8006e 2946
a2a32584 2947 atomic_set(&msgr->stopping, 0);
31b8006e 2948
15d9882c 2949 dout("%s %p\n", __func__, msgr);
31b8006e 2950}
15d9882c 2951EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2952
e00de341
SW		 2953static void clear_standby(struct ceph_connection *con)
2954{
2955 /* come back from STANDBY? */
8dacc7da 2956 if (con->state == CON_STATE_STANDBY) {
e00de341 2957 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2958 con->state = CON_STATE_PREOPEN;
e00de341 2959 con->connect_seq++;
c9ffc77a
AE		 2960 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2961 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2962 }
2963}
2964
31b8006e
SW		 2965/*
2966 * Queue up an outgoing message on the given connection.
2967 */
2968void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2969{
31b8006e 2970 /* set src+dst */
dbad185d 2971 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2972 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2973 msg->needs_out_seq = true;
2974
ec302645 2975 mutex_lock(&con->mutex);
92ce034b 2976
8dacc7da 2977 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2978 dout("con_send %p closed, dropping %p\n", con, msg);
2979 ceph_msg_put(msg);
2980 mutex_unlock(&con->mutex);
2981 return;
2982 }
2983
38941f80 2984 BUG_ON(msg->con != NULL);
36eb71aa 2985 msg->con = con->ops->get(con);
92ce034b
AE		 2986 BUG_ON(msg->con == NULL);
2987
31b8006e
SW		 2988 BUG_ON(!list_empty(&msg->list_head));
2989 list_add_tail(&msg->list_head, &con->out_queue);
2990 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2991 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2992 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2993 le32_to_cpu(msg->hdr.front_len),
2994 le32_to_cpu(msg->hdr.middle_len),
2995 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 2996
2997 clear_standby(con);
ec302645 2998 mutex_unlock(&con->mutex);
31b8006e
SW		 2999
3000 /* if there wasn't anything waiting to send before, queue
3001 * new work */
c9ffc77a 3002 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3003 queue_con(con);
3004}
3d14c5d2 3005EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3006
3007/*
3008 * Revoke a message that was previously queued for send
3009 */
6740a845 3010void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3011{
6740a845
AE		 3012 struct ceph_connection *con = msg->con;
3013
3014 if (!con)
3015 return; /* Message not in our possession */
3016
ec302645 3017 mutex_lock(&con->mutex);
31b8006e 3018 if (!list_empty(&msg->list_head)) {
38941f80 3019 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3020 list_del_init(&msg->list_head);
38941f80 3021 BUG_ON(msg->con == NULL);
36eb71aa 3022 msg->con->ops->put(msg->con);
38941f80 3023 msg->con = NULL;
31b8006e 3024 msg->hdr.seq = 0;
38941f80 3025
31b8006e 3026 ceph_msg_put(msg);
ed98adad
SW		 3027 }
3028 if (con->out_msg == msg) {
38941f80 3029 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 3030 con->out_msg = NULL;
31b8006e
SW		 3031 if (con->out_kvec_is_msg) {
3032 con->out_skip = con->out_kvec_bytes;
3033 con->out_kvec_is_msg = false;
3034 }
ed98adad 3035 msg->hdr.seq = 0;
92ce034b
AE		 3036
3037 ceph_msg_put(msg);
31b8006e 3038 }
ec302645 3039 mutex_unlock(&con->mutex);
31b8006e
SW		 3040}
3041
350b1c32 3042/*
0d59ab81 3043 * Revoke a message that we may be reading data into
350b1c32 3044 */
8921d114 3045void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3046{
8921d114
AE		 3047 struct ceph_connection *con;
3048
3049 BUG_ON(msg == NULL);
3050 if (!msg->con) {
3051 dout("%s msg %p null con\n", __func__, msg);
3052
3053 return; /* Message not in our possession */
3054 }
3055
3056 con = msg->con;
350b1c32 3057 mutex_lock(&con->mutex);
8921d114 3058 if (con->in_msg == msg) {
95c96174
ED		 3059 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3060 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3061 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3062
3063 /* skip rest of message */
8921d114
AE		 3064 dout("%s %p msg %p revoked\n", __func__, con, msg);
3065 con->in_base_pos = con->in_base_pos -
350b1c32 3066 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3067 front_len -
3068 middle_len -
3069 data_len -
350b1c32 3070 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3071 ceph_msg_put(con->in_msg);
3072 con->in_msg = NULL;
3073 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3074 con->in_seq++;
350b1c32 3075 } else {
8921d114
AE		 3076 dout("%s %p in_msg %p msg %p no-op\n",
3077 __func__, con, con->in_msg, msg);
350b1c32
SW		 3078 }
3079 mutex_unlock(&con->mutex);
3080}
3081
31b8006e
SW		 3082/*
3083 * Queue a keepalive byte to ensure the tcp connection is alive.
3084 */
3085void ceph_con_keepalive(struct ceph_connection *con)
3086{
e00de341 3087 dout("con_keepalive %p\n", con);
00650931 3088 mutex_lock(&con->mutex);
e00de341 3089 clear_standby(con);
00650931 3090 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3091 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3092 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3093 queue_con(con);
3094}
3d14c5d2 3095EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3096
6644ed7b 3097static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3098{
6644ed7b
AE		 3099 struct ceph_msg_data *data;
3100
3101 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3102 return NULL;
3103
81b36be4 3104 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
6644ed7b
AE		 3105 if (data)
3106 data->type = type;
5240d9f9 3107 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3108
3109 return data;
3110}
3111
3112static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3113{
3114 if (!data)
3115 return;
3116
5240d9f9 3117 WARN_ON(!list_empty(&data->links));
e4339d28 3118 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3119 ceph_pagelist_release(data->pagelist);
81b36be4 3120 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3121}
3122
90af3602 3123void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3124 size_t length, size_t alignment)
02afca6c 3125{
6644ed7b
AE		 3126 struct ceph_msg_data *data;
3127
07aa1558
AE		 3128 BUG_ON(!pages);
3129 BUG_ON(!length);
6644ed7b
AE		 3130
3131 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3132 BUG_ON(!data);
3133 data->pages = pages;
3134 data->length = length;
3135 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3136
5240d9f9
AE		 3137 list_add_tail(&data->links, &msg->data);
3138 msg->data_length += length;
02afca6c 3139}
90af3602 3140EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3141
90af3602 3142void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3143 struct ceph_pagelist *pagelist)
3144{
6644ed7b
AE		 3145 struct ceph_msg_data *data;
3146
07aa1558
AE		 3147 BUG_ON(!pagelist);
3148 BUG_ON(!pagelist->length);
27fa8385 3149
6644ed7b
AE		 3150 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3151 BUG_ON(!data);
3152 data->pagelist = pagelist;
3153
5240d9f9
AE		 3154 list_add_tail(&data->links, &msg->data);
3155 msg->data_length += pagelist->length;
27fa8385 3156}
90af3602 3157EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3158
ea96571f 3159#ifdef CONFIG_BLOCK
90af3602 3160void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3161 size_t length)
27fa8385 3162{
6644ed7b
AE		 3163 struct ceph_msg_data *data;
3164
07aa1558 3165 BUG_ON(!bio);
27fa8385 3166
6644ed7b
AE		 3167 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3168 BUG_ON(!data);
3169 data->bio = bio;
c851c495 3170 data->bio_length = length;
6644ed7b 3171
5240d9f9
AE		 3172 list_add_tail(&data->links, &msg->data);
3173 msg->data_length += length;
27fa8385 3174}
90af3602 3175EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3176#endif /* CONFIG_BLOCK */
27fa8385 3177
31b8006e
SW		 3178/*
3179 * construct a new message with given type, size
3180 * the new msg has a ref count of 1.
3181 */
b61c2763
SW		 3182struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3183 bool can_fail)
31b8006e
SW		 3184{
3185 struct ceph_msg *m;
3186
e3d5d638 3187 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3188 if (m == NULL)
3189 goto out;
31b8006e
SW		 3190
3191 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3192 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3193 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3194
9516e45b
AE		 3195 INIT_LIST_HEAD(&m->list_head);
3196 kref_init(&m->kref);
5240d9f9 3197 INIT_LIST_HEAD(&m->data);
ca20892d 3198
31b8006e
SW		 3199 /* front */
3200 if (front_len) {
eeb0bed5 3201 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3202 if (m->front.iov_base == NULL) {
b61c2763 3203 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3204 front_len);
3205 goto out2;
3206 }
3207 } else {
3208 m->front.iov_base = NULL;
3209 }
f2be82b0 3210 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3211
bb257664 3212 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3213 return m;
3214
3215out2:
3216 ceph_msg_put(m);
3217out:
b61c2763
SW		 3218 if (!can_fail) {
3219 pr_err("msg_new can't create type %d front %d\n", type,
3220 front_len);
f0ed1b7c 3221 WARN_ON(1);
b61c2763
SW		 3222 } else {
3223 dout("msg_new can't create type %d front %d\n", type,
3224 front_len);
3225 }
a79832f2 3226 return NULL;
31b8006e 3227}
3d14c5d2 3228EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3229
31b8006e
SW		 3230/*
3231 * Allocate "middle" portion of a message, if it is needed and wasn't
3232 * allocated by alloc_msg. This allows us to read a small fixed-size
3233 * per-type header in the front and then gracefully fail (i.e.,
3234 * propagate the error to the caller based on info in the front) when
3235 * the middle is too large.
3236 */
2450418c 3237static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3238{
3239 int type = le16_to_cpu(msg->hdr.type);
3240 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3241
3242 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3243 ceph_msg_type_name(type), middle_len);
3244 BUG_ON(!middle_len);
3245 BUG_ON(msg->middle);
3246
b6c1d5b8 3247 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3248 if (!msg->middle)
3249 return -ENOMEM;
3250 return 0;
3251}
3252
2450418c 3253/*
1c20f2d2
AE		 3254 * Allocate a message for receiving an incoming message on a
3255 * connection, and save the result in con->in_msg. Uses the
3256 * connection's private alloc_msg op if available.
3257 *
4740a623
SW		 3258 * Returns 0 on success, or a negative error code.
3259 *
3260 * On success, if we set *skip = 1:
3261 * - the next message should be skipped and ignored.
3262 * - con->in_msg == NULL
3263 * or if we set *skip = 0:
3264 * - con->in_msg is non-null.
3265 * On error (ENOMEM, EAGAIN, ...),
3266 * - con->in_msg == NULL
2450418c 3267 */
4740a623 3268static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3269{
4740a623 3270 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3271 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3272 struct ceph_msg *msg;
4740a623 3273 int ret = 0;
2450418c 3274
1c20f2d2 3275 BUG_ON(con->in_msg != NULL);
53ded495 3276 BUG_ON(!con->ops->alloc_msg);
2450418c 3277
53ded495
AE		 3278 mutex_unlock(&con->mutex);
3279 msg = con->ops->alloc_msg(con, hdr, skip);
3280 mutex_lock(&con->mutex);
3281 if (con->state != CON_STATE_OPEN) {
3282 if (msg)
1d866d1c 3283 ceph_msg_put(msg);
53ded495
AE		 3284 return -EAGAIN;
3285 }
4137577a
AE		 3286 if (msg) {
3287 BUG_ON(*skip);
3288 con->in_msg = msg;
36eb71aa 3289 con->in_msg->con = con->ops->get(con);
92ce034b 3290 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3291 } else {
3292 /*
3293 * Null message pointer means either we should skip
3294 * this message or we couldn't allocate memory. The
3295 * former is not an error.
3296 */
3297 if (*skip)
3298 return 0;
4137577a 3299
67c64eb7 3300 con->error_msg = "error allocating memory for incoming message";
53ded495 3301 return -ENOMEM;
2450418c 3302 }
1c20f2d2 3303 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3304
1c20f2d2
AE		 3305 if (middle_len && !con->in_msg->middle) {
3306 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3307 if (ret < 0) {
1c20f2d2
AE		 3308 ceph_msg_put(con->in_msg);
3309 con->in_msg = NULL;
2450418c
YS		 3310 }
3311 }
9d7f0f13 3312
4740a623 3313 return ret;
2450418c
YS		 3314}
3315
31b8006e
SW		 3316
3317/*
3318 * Free a generically kmalloc'd message.
3319 */
0215e44b 3320static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3321{
0215e44b 3322 dout("%s %p\n", __func__, m);
4965fc38 3323 kvfree(m->front.iov_base);
e3d5d638 3324 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3325}
3326
0215e44b 3327static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3328{
3329 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
5240d9f9
AE		 3330 LIST_HEAD(data);
3331 struct list_head *links;
3332 struct list_head *next;
31b8006e 3333
0215e44b 3334 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3335 WARN_ON(!list_empty(&m->list_head));
3336
3337 /* drop middle, data, if any */
3338 if (m->middle) {
3339 ceph_buffer_put(m->middle);
3340 m->middle = NULL;
31b8006e 3341 }
5240d9f9
AE		 3342
3343 list_splice_init(&m->data, &data);
3344 list_for_each_safe(links, next, &data) {
3345 struct ceph_msg_data *data;
3346
3347 data = list_entry(links, struct ceph_msg_data, links);
3348 list_del_init(links);
3349 ceph_msg_data_destroy(data);
3350 }
a1930804 3351 m->data_length = 0;
58bb3b37 3352
c2e552e7
SW		 3353 if (m->pool)
3354 ceph_msgpool_put(m->pool, m);
3355 else
0215e44b
ID		 3356 ceph_msg_free(m);
3357}
3358
3359struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3360{
3361 dout("%s %p (was %d)\n", __func__, msg,
3362 atomic_read(&msg->kref.refcount));
3363 kref_get(&msg->kref);
3364 return msg;
3365}
3366EXPORT_SYMBOL(ceph_msg_get);
3367
3368void ceph_msg_put(struct ceph_msg *msg)
3369{
3370 dout("%s %p (was %d)\n", __func__, msg,
3371 atomic_read(&msg->kref.refcount));
3372 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3373}
0215e44b 3374EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3375
3376void ceph_msg_dump(struct ceph_msg *msg)
3377{
3cea4c30
ID		 3378 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3379 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3380 print_hex_dump(KERN_DEBUG, "header: ",
3381 DUMP_PREFIX_OFFSET, 16, 1,
3382 &msg->hdr, sizeof(msg->hdr), true);
3383 print_hex_dump(KERN_DEBUG, " front: ",
3384 DUMP_PREFIX_OFFSET, 16, 1,
3385 msg->front.iov_base, msg->front.iov_len, true);
3386 if (msg->middle)
3387 print_hex_dump(KERN_DEBUG, "middle: ",
3388 DUMP_PREFIX_OFFSET, 16, 1,
3389 msg->middle->vec.iov_base,
3390 msg->middle->vec.iov_len, true);
3391 print_hex_dump(KERN_DEBUG, "footer: ",
3392 DUMP_PREFIX_OFFSET, 16, 1,
3393 &msg->footer, sizeof(msg->footer), true);
3394}
3d14c5d2 3395EXPORT_SYMBOL(ceph_msg_dump);
Linux 6.x block layer and io_uring tree(s)