projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: store session key in cephx authorizer
[deliverable/linux.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;
89baaa57 487 sock->sk->sk_allocation = GFP_NOFS | __GFP_MEMALLOC;
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);
31b8006e 508 con->error_msg = "connect error";
31b8006e 509
41617d0c 510 return ret;
a5bc3129 511 }
89baaa57
MC		 512
513 sk_set_memalloc(sock->sk);
514
a5bc3129 515 con->sock = sock;
41617d0c 516 return 0;
31b8006e
SW		 517}
518
519static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
520{
521 struct kvec iov = {buf, len};
522 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 523 int r;
31b8006e 524
98bdb0aa
SW		 525 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
526 if (r == -EAGAIN)
527 r = 0;
528 return r;
31b8006e
SW		 529}
530
afb3d90e
AE		 531static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
532 int page_offset, size_t length)
533{
534 void *kaddr;
535 int ret;
536
537 BUG_ON(page_offset + length > PAGE_SIZE);
538
539 kaddr = kmap(page);
540 BUG_ON(!kaddr);
541 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
542 kunmap(page);
543
544 return ret;
545}
546
31b8006e
SW		 547/*
548 * write something. @more is true if caller will be sending more data
549 * shortly.
550 */
551static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
552 size_t kvlen, size_t len, int more)
553{
554 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 555 int r;
31b8006e
SW		 556
557 if (more)
558 msg.msg_flags |= MSG_MORE;
559 else
560 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
561
42961d23
SW		 562 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
563 if (r == -EAGAIN)
564 r = 0;
565 return r;
31b8006e
SW		 566}
567
178eda29 568static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 569 int offset, size_t size, bool more)
31739139
AE		 570{
571 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
572 int ret;
573
574 ret = kernel_sendpage(sock, page, offset, size, flags);
575 if (ret == -EAGAIN)
576 ret = 0;
577
578 return ret;
579}
580
178eda29
CC		 581static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
582 int offset, size_t size, bool more)
583{
584 int ret;
585 struct kvec iov;
586
587 /* sendpage cannot properly handle pages with page_count == 0,
588 * we need to fallback to sendmsg if that's the case */
589 if (page_count(page) >= 1)
590 return __ceph_tcp_sendpage(sock, page, offset, size, more);
591
592 iov.iov_base = kmap(page) + offset;
593 iov.iov_len = size;
594 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
595 kunmap(page);
596
597 return ret;
598}
31b8006e
SW		 599
600/*
601 * Shutdown/close the socket for the given connection.
602 */
603static int con_close_socket(struct ceph_connection *con)
604{
8007b8d6 605 int rc = 0;
31b8006e
SW		 606
607 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 608 if (con->sock) {
609 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
610 sock_release(con->sock);
611 con->sock = NULL;
612 }
456ea468
AE		 613
614 /*
4a861692 615 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 616 * independent of the connection mutex, and we could have
617 * received a socket close event before we had the chance to
618 * shut the socket down.
619 */
c9ffc77a 620 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 621
ce2c8903 622 con_sock_state_closed(con);
31b8006e
SW		 623 return rc;
624}
625
626/*
627 * Reset a connection. Discard all incoming and outgoing messages
628 * and clear *_seq state.
629 */
630static void ceph_msg_remove(struct ceph_msg *msg)
631{
632 list_del_init(&msg->list_head);
38941f80 633 BUG_ON(msg->con == NULL);
36eb71aa 634 msg->con->ops->put(msg->con);
38941f80
AE		 635 msg->con = NULL;
636
31b8006e
SW		 637 ceph_msg_put(msg);
638}
639static void ceph_msg_remove_list(struct list_head *head)
640{
641 while (!list_empty(head)) {
642 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
643 list_head);
644 ceph_msg_remove(msg);
645 }
646}
647
648static void reset_connection(struct ceph_connection *con)
649{
650 /* reset connection, out_queue, msg_ and connect_seq */
651 /* discard existing out_queue and msg_seq */
0fa6ebc6 652 dout("reset_connection %p\n", con);
31b8006e
SW		 653 ceph_msg_remove_list(&con->out_queue);
654 ceph_msg_remove_list(&con->out_sent);
655
cf3e5c40 656 if (con->in_msg) {
38941f80
AE		 657 BUG_ON(con->in_msg->con != con);
658 con->in_msg->con = NULL;
cf3e5c40
SW		 659 ceph_msg_put(con->in_msg);
660 con->in_msg = NULL;
36eb71aa 661 con->ops->put(con);
cf3e5c40
SW		 662 }
663
31b8006e
SW		 664 con->connect_seq = 0;
665 con->out_seq = 0;
c86a2930
SW		 666 if (con->out_msg) {
667 ceph_msg_put(con->out_msg);
668 con->out_msg = NULL;
669 }
31b8006e 670 con->in_seq = 0;
0e0d5e0c 671 con->in_seq_acked = 0;
31b8006e
SW		 672}
673
674/*
675 * mark a peer down. drop any open connections.
676 */
677void ceph_con_close(struct ceph_connection *con)
678{
8c50c817 679 mutex_lock(&con->mutex);
3d14c5d2
YS		 680 dout("con_close %p peer %s\n", con,
681 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 682 con->state = CON_STATE_CLOSED;
a5988c49 683
c9ffc77a
AE		 684 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
685 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
686 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
687 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 688
31b8006e 689 reset_connection(con);
6f2bc3ff 690 con->peer_global_seq = 0;
37ab77ac 691 cancel_con(con);
ee76e073 692 con_close_socket(con);
ec302645 693 mutex_unlock(&con->mutex);
31b8006e 694}
3d14c5d2 695EXPORT_SYMBOL(ceph_con_close);
31b8006e 696
31b8006e
SW		 697/*
698 * Reopen a closed connection, with a new peer address.
699 */
b7a9e5dd
SW		 700void ceph_con_open(struct ceph_connection *con,
701 __u8 entity_type, __u64 entity_num,
702 struct ceph_entity_addr *addr)
31b8006e 703{
5469155f 704 mutex_lock(&con->mutex);
3d14c5d2 705 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 706
122070a2 707 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 708 con->state = CON_STATE_PREOPEN;
a5988c49 709
b7a9e5dd
SW		 710 con->peer_name.type = (__u8) entity_type;
711 con->peer_name.num = cpu_to_le64(entity_num);
712
31b8006e 713 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 714 con->delay = 0; /* reset backoff memory */
5469155f 715 mutex_unlock(&con->mutex);
31b8006e
SW		 716 queue_con(con);
717}
3d14c5d2 718EXPORT_SYMBOL(ceph_con_open);
31b8006e 719
87b315a5
SW		 720/*
721 * return true if this connection ever successfully opened
722 */
723bool ceph_con_opened(struct ceph_connection *con)
724{
725 return con->connect_seq > 0;
726}
727
31b8006e
SW		 728/*
729 * initialize a new connection.
730 */
1bfd89f4
AE		 731void ceph_con_init(struct ceph_connection *con, void *private,
732 const struct ceph_connection_operations *ops,
b7a9e5dd 733 struct ceph_messenger *msgr)
31b8006e
SW		 734{
735 dout("con_init %p\n", con);
736 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 737 con->private = private;
738 con->ops = ops;
31b8006e 739 con->msgr = msgr;
ce2c8903
AE		 740
741 con_sock_state_init(con);
742
ec302645 743 mutex_init(&con->mutex);
31b8006e
SW		 744 INIT_LIST_HEAD(&con->out_queue);
745 INIT_LIST_HEAD(&con->out_sent);
746 INIT_DELAYED_WORK(&con->work, con_work);
a5988c49 747
8dacc7da 748 con->state = CON_STATE_CLOSED;
31b8006e 749}
3d14c5d2 750EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 751
752
753/*
754 * We maintain a global counter to order connection attempts. Get
755 * a unique seq greater than @gt.
756 */
757static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
758{
759 u32 ret;
760
761 spin_lock(&msgr->global_seq_lock);
762 if (msgr->global_seq < gt)
763 msgr->global_seq = gt;
764 ret = ++msgr->global_seq;
765 spin_unlock(&msgr->global_seq_lock);
766 return ret;
767}
768
e2200423 769static void con_out_kvec_reset(struct ceph_connection *con)
859eb799
AE		 770{
771 con->out_kvec_left = 0;
772 con->out_kvec_bytes = 0;
773 con->out_kvec_cur = &con->out_kvec[0];
774}
775
e2200423 776static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 777 size_t size, void *data)
778{
779 int index;
780
781 index = con->out_kvec_left;
782 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
783
784 con->out_kvec[index].iov_len = size;
785 con->out_kvec[index].iov_base = data;
786 con->out_kvec_left++;
787 con->out_kvec_bytes += size;
788}
31b8006e 789
df6ad1f9 790#ifdef CONFIG_BLOCK
6aaa4511
AE		 791
792/*
793 * For a bio data item, a piece is whatever remains of the next
794 * entry in the current bio iovec, or the first entry in the next
795 * bio in the list.
796 */
8ae4f4f5 797static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 798 size_t length)
6aaa4511 799{
8ae4f4f5 800 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 801 struct bio *bio;
802
803 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
804
805 bio = data->bio;
806 BUG_ON(!bio);
6aaa4511 807
ca8b3a69 808 cursor->resid = min(length, data->bio_length);
6aaa4511 809 cursor->bio = bio;
f38a5181
KO		 810 cursor->bvec_iter = bio->bi_iter;
811 cursor->last_piece =
812 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
6aaa4511
AE		 813}
814
8ae4f4f5 815static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 816 size_t *page_offset,
817 size_t *length)
818{
8ae4f4f5 819 struct ceph_msg_data *data = cursor->data;
6aaa4511 820 struct bio *bio;
f38a5181 821 struct bio_vec bio_vec;
6aaa4511
AE		 822
823 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
824
825 bio = cursor->bio;
826 BUG_ON(!bio);
827
f38a5181 828 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511 829
f38a5181 830 *page_offset = (size_t) bio_vec.bv_offset;
6aaa4511 831 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 832 if (cursor->last_piece) /* pagelist offset is always 0 */
833 *length = cursor->resid;
834 else
f38a5181 835 *length = (size_t) bio_vec.bv_len;
25aff7c5 836 BUG_ON(*length > cursor->resid);
5df521b1 837 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511 838
f38a5181 839 return bio_vec.bv_page;
6aaa4511
AE		 840}
841
8ae4f4f5
AE		 842static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
843 size_t bytes)
6aaa4511 844{
6aaa4511 845 struct bio *bio;
f38a5181 846 struct bio_vec bio_vec;
6aaa4511 847
8ae4f4f5 848 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
6aaa4511
AE		 849
850 bio = cursor->bio;
851 BUG_ON(!bio);
852
f38a5181 853 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511
AE		 854
855 /* Advance the cursor offset */
856
25aff7c5
AE		 857 BUG_ON(cursor->resid < bytes);
858 cursor->resid -= bytes;
f38a5181
KO		 859
860 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
861
862 if (bytes < bio_vec.bv_len)
6aaa4511
AE		 863 return false; /* more bytes to process in this segment */
864
865 /* Move on to the next segment, and possibly the next bio */
866
f38a5181 867 if (!cursor->bvec_iter.bi_size) {
6aaa4511 868 bio = bio->bi_next;
0ec1d15e
ID		 869 cursor->bio = bio;
870 if (bio)
871 cursor->bvec_iter = bio->bi_iter;
872 else
873 memset(&cursor->bvec_iter, 0,
874 sizeof(cursor->bvec_iter));
6aaa4511 875 }
6aaa4511 876
25aff7c5
AE		 877 if (!cursor->last_piece) {
878 BUG_ON(!cursor->resid);
879 BUG_ON(!bio);
880 /* A short read is OK, so use <= rather than == */
f38a5181 881 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
6aaa4511 882 cursor->last_piece = true;
25aff7c5 883 }
6aaa4511
AE		 884
885 return true;
886}
ea96571f 887#endif /* CONFIG_BLOCK */
df6ad1f9 888
e766d7b5
AE		 889/*
890 * For a page array, a piece comes from the first page in the array
891 * that has not already been fully consumed.
892 */
8ae4f4f5 893static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 894 size_t length)
e766d7b5 895{
8ae4f4f5 896 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 897 int page_count;
898
899 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
900
901 BUG_ON(!data->pages);
902 BUG_ON(!data->length);
903
ca8b3a69 904 cursor->resid = min(length, data->length);
e766d7b5 905 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 906 cursor->page_offset = data->alignment & ~PAGE_MASK;
907 cursor->page_index = 0;
56fc5659
AE		 908 BUG_ON(page_count > (int)USHRT_MAX);
909 cursor->page_count = (unsigned short)page_count;
910 BUG_ON(length > SIZE_MAX - cursor->page_offset);
5f740d7e 911 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 912}
913
8ae4f4f5
AE		 914static struct page *
915ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
916 size_t *page_offset, size_t *length)
e766d7b5 917{
8ae4f4f5 918 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 919
920 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
921
922 BUG_ON(cursor->page_index >= cursor->page_count);
923 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 924
925 *page_offset = cursor->page_offset;
25aff7c5 926 if (cursor->last_piece)
e766d7b5 927 *length = cursor->resid;
25aff7c5 928 else
e766d7b5 929 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 930
931 return data->pages[cursor->page_index];
932}
933
8ae4f4f5 934static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 935 size_t bytes)
936{
8ae4f4f5 937 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 938
939 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 940
941 /* Advance the cursor page offset */
942
943 cursor->resid -= bytes;
5df521b1
AE		 944 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
945 if (!bytes || cursor->page_offset)
e766d7b5
AE		 946 return false; /* more bytes to process in the current page */
947
d90deda6
YZ		 948 if (!cursor->resid)
949 return false; /* no more data */
950
5df521b1 951 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 952
953 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 954 cursor->page_index++;
25aff7c5 955 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 956
957 return true;
958}
959
fe38a2b6 960/*
dd236fcb
AE		 961 * For a pagelist, a piece is whatever remains to be consumed in the
962 * first page in the list, or the front of the next page.
fe38a2b6 963 */
8ae4f4f5
AE		 964static void
965ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 966 size_t length)
fe38a2b6 967{
8ae4f4f5 968 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 969 struct ceph_pagelist *pagelist;
970 struct page *page;
971
dd236fcb 972 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 973
974 pagelist = data->pagelist;
975 BUG_ON(!pagelist);
25aff7c5
AE		 976
977 if (!length)
fe38a2b6
AE		 978 return; /* pagelist can be assigned but empty */
979
980 BUG_ON(list_empty(&pagelist->head));
981 page = list_first_entry(&pagelist->head, struct page, lru);
982
ca8b3a69 983 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 984 cursor->page = page;
985 cursor->offset = 0;
a51b272e 986 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 987}
988
8ae4f4f5
AE		 989static struct page *
990ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
991 size_t *page_offset, size_t *length)
fe38a2b6 992{
8ae4f4f5 993 struct ceph_msg_data *data = cursor->data;
fe38a2b6 994 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 995
996 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
997
998 pagelist = data->pagelist;
999 BUG_ON(!pagelist);
1000
1001 BUG_ON(!cursor->page);
25aff7c5 1002 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 1003
5df521b1 1004 /* offset of first page in pagelist is always 0 */
fe38a2b6 1005 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 1006 if (cursor->last_piece)
25aff7c5
AE		 1007 *length = cursor->resid;
1008 else
1009 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1010
8ae4f4f5 1011 return cursor->page;
fe38a2b6
AE		 1012}
1013
8ae4f4f5 1014static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1015 size_t bytes)
fe38a2b6 1016{
8ae4f4f5 1017 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1018 struct ceph_pagelist *pagelist;
1019
1020 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1021
1022 pagelist = data->pagelist;
1023 BUG_ON(!pagelist);
25aff7c5
AE		 1024
1025 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1026 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1027
1028 /* Advance the cursor offset */
1029
25aff7c5 1030 cursor->resid -= bytes;
fe38a2b6 1031 cursor->offset += bytes;
5df521b1 1032 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1033 if (!bytes || cursor->offset & ~PAGE_MASK)
1034 return false; /* more bytes to process in the current page */
1035
d90deda6
YZ		 1036 if (!cursor->resid)
1037 return false; /* no more data */
1038
fe38a2b6
AE		 1039 /* Move on to the next page */
1040
1041 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1042 cursor->page = list_entry_next(cursor->page, lru);
25aff7c5 1043 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1044
1045 return true;
1046}
1047
dd236fcb
AE		 1048/*
1049 * Message data is handled (sent or received) in pieces, where each
1050 * piece resides on a single page. The network layer might not
1051 * consume an entire piece at once. A data item's cursor keeps
1052 * track of which piece is next to process and how much remains to
1053 * be processed in that piece. It also tracks whether the current
1054 * piece is the last one in the data item.
1055 */
ca8b3a69 1056static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1057{
ca8b3a69 1058 size_t length = cursor->total_resid;
8ae4f4f5 1059
8ae4f4f5 1060 switch (cursor->data->type) {
dd236fcb 1061 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1062 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1063 break;
e766d7b5 1064 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1065 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1066 break;
dd236fcb
AE		 1067#ifdef CONFIG_BLOCK
1068 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1069 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1070 break;
dd236fcb 1071#endif /* CONFIG_BLOCK */
6aaa4511 1072 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1073 default:
1074 /* BUG(); */
1075 break;
1076 }
8ae4f4f5 1077 cursor->need_crc = true;
dd236fcb
AE		 1078}
1079
ca8b3a69
AE		 1080static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1081{
1082 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1083 struct ceph_msg_data *data;
1084
1085 BUG_ON(!length);
1086 BUG_ON(length > msg->data_length);
1087 BUG_ON(list_empty(&msg->data));
1088
ca8b3a69
AE		 1089 cursor->data_head = &msg->data;
1090 cursor->total_resid = length;
1091 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1092 cursor->data = data;
1093
1094 __ceph_msg_data_cursor_init(cursor);
1095}
1096
dd236fcb
AE		 1097/*
1098 * Return the page containing the next piece to process for a given
1099 * data item, and supply the page offset and length of that piece.
1100 * Indicate whether this is the last piece in this data item.
1101 */
8ae4f4f5
AE		 1102static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1103 size_t *page_offset, size_t *length,
dd236fcb
AE		 1104 bool *last_piece)
1105{
1106 struct page *page;
1107
8ae4f4f5 1108 switch (cursor->data->type) {
dd236fcb 1109 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1110 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1111 break;
e766d7b5 1112 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1113 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1114 break;
dd236fcb
AE		 1115#ifdef CONFIG_BLOCK
1116 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1117 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1118 break;
dd236fcb 1119#endif /* CONFIG_BLOCK */
6aaa4511 1120 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1121 default:
1122 page = NULL;
1123 break;
1124 }
1125 BUG_ON(!page);
1126 BUG_ON(*page_offset + *length > PAGE_SIZE);
1127 BUG_ON(!*length);
1128 if (last_piece)
8ae4f4f5 1129 *last_piece = cursor->last_piece;
dd236fcb
AE		 1130
1131 return page;
1132}
1133
1134/*
1135 * Returns true if the result moves the cursor on to the next piece
1136 * of the data item.
1137 */
8ae4f4f5
AE		 1138static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1139 size_t bytes)
dd236fcb
AE		 1140{
1141 bool new_piece;
1142
25aff7c5 1143 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1144 switch (cursor->data->type) {
dd236fcb 1145 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1146 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1147 break;
e766d7b5 1148 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1149 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1150 break;
dd236fcb
AE		 1151#ifdef CONFIG_BLOCK
1152 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1153 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1154 break;
dd236fcb 1155#endif /* CONFIG_BLOCK */
6aaa4511 1156 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1157 default:
1158 BUG();
1159 break;
1160 }
ca8b3a69 1161 cursor->total_resid -= bytes;
dd236fcb 1162
ca8b3a69
AE		 1163 if (!cursor->resid && cursor->total_resid) {
1164 WARN_ON(!cursor->last_piece);
1165 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1166 cursor->data = list_entry_next(cursor->data, links);
1167 __ceph_msg_data_cursor_init(cursor);
a51b272e 1168 new_piece = true;
ca8b3a69 1169 }
a51b272e 1170 cursor->need_crc = new_piece;
ca8b3a69 1171
dd236fcb
AE		 1172 return new_piece;
1173}
1174
98fa5dd8 1175static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1176{
739c905b 1177 BUG_ON(!msg);
25aff7c5 1178 BUG_ON(!data_len);
739c905b 1179
4c59b4a2 1180 /* Initialize data cursor */
fe38a2b6 1181
8ae4f4f5 1182 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1183}
1184
31b8006e
SW		 1185/*
1186 * Prepare footer for currently outgoing message, and finish things
1187 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1188 */
859eb799 1189static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1190{
1191 struct ceph_msg *m = con->out_msg;
859eb799 1192 int v = con->out_kvec_left;
31b8006e 1193
fd154f3c
AE		 1194 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1195
31b8006e
SW		 1196 dout("prepare_write_message_footer %p\n", con);
1197 con->out_kvec_is_msg = true;
1198 con->out_kvec[v].iov_base = &m->footer;
1199 con->out_kvec[v].iov_len = sizeof(m->footer);
1200 con->out_kvec_bytes += sizeof(m->footer);
1201 con->out_kvec_left++;
1202 con->out_more = m->more_to_follow;
c86a2930 1203 con->out_msg_done = true;
31b8006e
SW		 1204}
1205
1206/*
1207 * Prepare headers for the next outgoing message.
1208 */
1209static void prepare_write_message(struct ceph_connection *con)
1210{
1211 struct ceph_msg *m;
a9a0c51a 1212 u32 crc;
31b8006e 1213
e2200423 1214 con_out_kvec_reset(con);
31b8006e 1215 con->out_kvec_is_msg = true;
c86a2930 1216 con->out_msg_done = false;
31b8006e
SW		 1217
1218 /* Sneak an ack in there first? If we can get it into the same
1219 * TCP packet that's a good thing. */
1220 if (con->in_seq > con->in_seq_acked) {
1221 con->in_seq_acked = con->in_seq;
e2200423 1222 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1223 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1224 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1225 &con->out_temp_ack);
31b8006e
SW		 1226 }
1227
38941f80 1228 BUG_ON(list_empty(&con->out_queue));
859eb799 1229 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1230 con->out_msg = m;
38941f80 1231 BUG_ON(m->con != con);
4cf9d544
SW		 1232
1233 /* put message on sent list */
1234 ceph_msg_get(m);
1235 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1236
e84346b7
SW		 1237 /*
1238 * only assign outgoing seq # if we haven't sent this message
1239 * yet. if it is requeued, resend with it's original seq.
1240 */
1241 if (m->needs_out_seq) {
1242 m->hdr.seq = cpu_to_le64(++con->out_seq);
1243 m->needs_out_seq = false;
1244 }
98fa5dd8 1245 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1246
98fa5dd8 1247 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1248 m, con->out_seq, le16_to_cpu(m->hdr.type),
1249 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1250 m->data_length);
31b8006e
SW		 1251 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1252
1253 /* tag + hdr + front + middle */
e2200423
AE		 1254 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1255 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1256 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1257
31b8006e 1258 if (m->middle)
e2200423 1259 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1260 m->middle->vec.iov_base);
31b8006e
SW		 1261
1262 /* fill in crc (except data pages), footer */
a9a0c51a
AE		 1263 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1264 con->out_msg->hdr.crc = cpu_to_le32(crc);
fd154f3c 1265 con->out_msg->footer.flags = 0;
a9a0c51a
AE		 1266
1267 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1268 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1269 if (m->middle) {
1270 crc = crc32c(0, m->middle->vec.iov_base,
1271 m->middle->vec.iov_len);
1272 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1273 } else
31b8006e 1274 con->out_msg->footer.middle_crc = 0;
739c905b 1275 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1276 le32_to_cpu(con->out_msg->footer.front_crc),
1277 le32_to_cpu(con->out_msg->footer.middle_crc));
1278
1279 /* is there a data payload? */
739c905b 1280 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1281 if (m->data_length) {
1282 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1283 con->out_more = 1; /* data + footer will follow */
1284 } else {
31b8006e 1285 /* no, queue up footer too and be done */
859eb799 1286 prepare_write_message_footer(con);
78625051 1287 }
31b8006e 1288
c9ffc77a 1289 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1290}
1291
1292/*
1293 * Prepare an ack.
1294 */
1295static void prepare_write_ack(struct ceph_connection *con)
1296{
1297 dout("prepare_write_ack %p %llu -> %llu\n", con,
1298 con->in_seq_acked, con->in_seq);
1299 con->in_seq_acked = con->in_seq;
1300
e2200423 1301 con_out_kvec_reset(con);
859eb799 1302
e2200423 1303 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1304
31b8006e 1305 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1306 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1307 &con->out_temp_ack);
1308
31b8006e 1309 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1310 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1311}
1312
3a23083b
SW		 1313/*
1314 * Prepare to share the seq during handshake
1315 */
1316static void prepare_write_seq(struct ceph_connection *con)
1317{
1318 dout("prepare_write_seq %p %llu -> %llu\n", con,
1319 con->in_seq_acked, con->in_seq);
1320 con->in_seq_acked = con->in_seq;
1321
1322 con_out_kvec_reset(con);
1323
1324 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1325 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1326 &con->out_temp_ack);
1327
1328 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1329}
1330
31b8006e
SW		 1331/*
1332 * Prepare to write keepalive byte.
1333 */
1334static void prepare_write_keepalive(struct ceph_connection *con)
1335{
1336 dout("prepare_write_keepalive %p\n", con);
e2200423
AE		 1337 con_out_kvec_reset(con);
1338 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
c9ffc77a 1339 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1340}
1341
1342/*
1343 * Connection negotiation.
1344 */
1345
dac1e716
AE		 1346static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1347 int *auth_proto)
4e7a5dcd 1348{
a3530df3 1349 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1350
1351 if (!con->ops->get_authorizer) {
1352 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1353 con->out_connect.authorizer_len = 0;
729796be 1354 return NULL;
b1c6b980
AE		 1355 }
1356
1357 /* Can't hold the mutex while getting authorizer */
ec302645 1358 mutex_unlock(&con->mutex);
dac1e716 1359 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1360 mutex_lock(&con->mutex);
4e7a5dcd 1361
a3530df3 1362 if (IS_ERR(auth))
729796be 1363 return auth;
8dacc7da 1364 if (con->state != CON_STATE_NEGOTIATING)
729796be 1365 return ERR_PTR(-EAGAIN);
0da5d703 1366
8f43fb53
AE		 1367 con->auth_reply_buf = auth->authorizer_reply_buf;
1368 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1369 return auth;
4e7a5dcd
SW		 1370}
1371
31b8006e
SW		 1372/*
1373 * We connected to a peer and are saying hello.
1374 */
e825a66d 1375static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1376{
e2200423
AE		 1377 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1378 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1379 &con->msgr->my_enc_addr);
eed0ef2c 1380
eed0ef2c 1381 con->out_more = 0;
c9ffc77a 1382 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1383}
1384
e825a66d 1385static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1386{
95c96174 1387 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1388 int proto;
dac1e716 1389 int auth_proto;
729796be 1390 struct ceph_auth_handshake *auth;
31b8006e
SW		 1391
1392 switch (con->peer_name.type) {
1393 case CEPH_ENTITY_TYPE_MON:
1394 proto = CEPH_MONC_PROTOCOL;
1395 break;
1396 case CEPH_ENTITY_TYPE_OSD:
1397 proto = CEPH_OSDC_PROTOCOL;
1398 break;
1399 case CEPH_ENTITY_TYPE_MDS:
1400 proto = CEPH_MDSC_PROTOCOL;
1401 break;
1402 default:
1403 BUG();
1404 }
1405
1406 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1407 con->connect_seq, global_seq, proto);
4e7a5dcd 1408
e825a66d 1409 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
31b8006e
SW		 1410 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1411 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1412 con->out_connect.global_seq = cpu_to_le32(global_seq);
1413 con->out_connect.protocol_version = cpu_to_le32(proto);
1414 con->out_connect.flags = 0;
31b8006e 1415
dac1e716
AE		 1416 auth_proto = CEPH_AUTH_UNKNOWN;
1417 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1418 if (IS_ERR(auth))
1419 return PTR_ERR(auth);
3da54776 1420
dac1e716 1421 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1422 con->out_connect.authorizer_len = auth ?
1423 cpu_to_le32(auth->authorizer_buf_len) : 0;
1424
e2200423 1425 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1426 &con->out_connect);
1427 if (auth && auth->authorizer_buf_len)
e2200423 1428 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1429 auth->authorizer_buf);
859eb799 1430
31b8006e 1431 con->out_more = 0;
c9ffc77a 1432 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1433
e10c758e 1434 return 0;
31b8006e
SW		 1435}
1436
31b8006e
SW		 1437/*
1438 * write as much of pending kvecs to the socket as we can.
1439 * 1 -> done
1440 * 0 -> socket full, but more to do
1441 * <0 -> error
1442 */
1443static int write_partial_kvec(struct ceph_connection *con)
1444{
1445 int ret;
1446
1447 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1448 while (con->out_kvec_bytes > 0) {
1449 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1450 con->out_kvec_left, con->out_kvec_bytes,
1451 con->out_more);
1452 if (ret <= 0)
1453 goto out;
1454 con->out_kvec_bytes -= ret;
1455 if (con->out_kvec_bytes == 0)
1456 break; /* done */
f42299e6
AE		 1457
1458 /* account for full iov entries consumed */
1459 while (ret >= con->out_kvec_cur->iov_len) {
1460 BUG_ON(!con->out_kvec_left);
1461 ret -= con->out_kvec_cur->iov_len;
1462 con->out_kvec_cur++;
1463 con->out_kvec_left--;
1464 }
1465 /* and for a partially-consumed entry */
1466 if (ret) {
1467 con->out_kvec_cur->iov_len -= ret;
1468 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1469 }
1470 }
1471 con->out_kvec_left = 0;
1472 con->out_kvec_is_msg = false;
1473 ret = 1;
1474out:
1475 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1476 con->out_kvec_bytes, con->out_kvec_left, ret);
1477 return ret; /* done! */
1478}
1479
35b62808
AE		 1480static u32 ceph_crc32c_page(u32 crc, struct page *page,
1481 unsigned int page_offset,
1482 unsigned int length)
1483{
1484 char *kaddr;
1485
1486 kaddr = kmap(page);
1487 BUG_ON(kaddr == NULL);
1488 crc = crc32c(crc, kaddr + page_offset, length);
1489 kunmap(page);
1490
1491 return crc;
1492}
31b8006e
SW		 1493/*
1494 * Write as much message data payload as we can. If we finish, queue
1495 * up the footer.
1496 * 1 -> done, footer is now queued in out_kvec[].
1497 * 0 -> socket full, but more to do
1498 * <0 -> error
1499 */
34d2d200 1500static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1501{
1502 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1503 struct ceph_msg_data_cursor *cursor = &msg->cursor;
37675b0f 1504 bool do_datacrc = !con->msgr->nocrc;
f5db90bc 1505 u32 crc;
31b8006e 1506
859a35d5 1507 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1508
5240d9f9 1509 if (list_empty(&msg->data))
4c59b4a2
AE		 1510 return -EINVAL;
1511
5821bd8c
AE		 1512 /*
1513 * Iterate through each page that contains data to be
1514 * written, and send as much as possible for each.
1515 *
1516 * If we are calculating the data crc (the default), we will
1517 * need to map the page. If we have no pages, they have
1518 * been revoked, so use the zero page.
1519 */
f5db90bc 1520 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1521 while (cursor->resid) {
8a166d05 1522 struct page *page;
e387d525
AE		 1523 size_t page_offset;
1524 size_t length;
8a166d05 1525 bool last_piece;
8ea299bc 1526 bool need_crc;
f5db90bc 1527 int ret;
68b4476b 1528
8ae4f4f5 1529 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 1530 &last_piece);
e387d525 1531 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
fe38a2b6 1532 length, last_piece);
f5db90bc
AE		 1533 if (ret <= 0) {
1534 if (do_datacrc)
1535 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1536
f5db90bc
AE		 1537 return ret;
1538 }
143334ff
AE		 1539 if (do_datacrc && cursor->need_crc)
1540 crc = ceph_crc32c_page(crc, page, page_offset, length);
8ae4f4f5 1541 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
31b8006e
SW		 1542 }
1543
34d2d200 1544 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1545
1546 /* prepare and queue up footer, too */
f5db90bc
AE		 1547 if (do_datacrc)
1548 msg->footer.data_crc = cpu_to_le32(crc);
1549 else
84ca8fc8 1550 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1551 con_out_kvec_reset(con);
859eb799 1552 prepare_write_message_footer(con);
f5db90bc
AE		 1553
1554 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1555}
1556
1557/*
1558 * write some zeros
1559 */
1560static int write_partial_skip(struct ceph_connection *con)
1561{
1562 int ret;
1563
1564 while (con->out_skip > 0) {
31739139 1565 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1566
e1dcb128 1567 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1568 if (ret <= 0)
1569 goto out;
1570 con->out_skip -= ret;
1571 }
1572 ret = 1;
1573out:
1574 return ret;
1575}
1576
1577/*
1578 * Prepare to read connection handshake, or an ack.
1579 */
eed0ef2c
SW		 1580static void prepare_read_banner(struct ceph_connection *con)
1581{
1582 dout("prepare_read_banner %p\n", con);
1583 con->in_base_pos = 0;
1584}
1585
31b8006e
SW		 1586static void prepare_read_connect(struct ceph_connection *con)
1587{
1588 dout("prepare_read_connect %p\n", con);
1589 con->in_base_pos = 0;
1590}
1591
1592static void prepare_read_ack(struct ceph_connection *con)
1593{
1594 dout("prepare_read_ack %p\n", con);
1595 con->in_base_pos = 0;
1596}
1597
3a23083b
SW		 1598static void prepare_read_seq(struct ceph_connection *con)
1599{
1600 dout("prepare_read_seq %p\n", con);
1601 con->in_base_pos = 0;
1602 con->in_tag = CEPH_MSGR_TAG_SEQ;
1603}
1604
31b8006e
SW		 1605static void prepare_read_tag(struct ceph_connection *con)
1606{
1607 dout("prepare_read_tag %p\n", con);
1608 con->in_base_pos = 0;
1609 con->in_tag = CEPH_MSGR_TAG_READY;
1610}
1611
1612/*
1613 * Prepare to read a message.
1614 */
1615static int prepare_read_message(struct ceph_connection *con)
1616{
1617 dout("prepare_read_message %p\n", con);
1618 BUG_ON(con->in_msg != NULL);
1619 con->in_base_pos = 0;
1620 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1621 return 0;
1622}
1623
1624
1625static int read_partial(struct ceph_connection *con,
fd51653f 1626 int end, int size, void *object)
31b8006e 1627{
e6cee71f
AE		 1628 while (con->in_base_pos < end) {
1629 int left = end - con->in_base_pos;
31b8006e
SW		 1630 int have = size - left;
1631 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1632 if (ret <= 0)
1633 return ret;
1634 con->in_base_pos += ret;
1635 }
1636 return 1;
1637}
1638
1639
1640/*
1641 * Read all or part of the connect-side handshake on a new connection
1642 */
eed0ef2c 1643static int read_partial_banner(struct ceph_connection *con)
31b8006e 1644{
fd51653f
AE		 1645 int size;
1646 int end;
1647 int ret;
31b8006e 1648
eed0ef2c 1649 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1650
1651 /* peer's banner */
fd51653f
AE		 1652 size = strlen(CEPH_BANNER);
1653 end = size;
1654 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1655 if (ret <= 0)
1656 goto out;
fd51653f
AE		 1657
1658 size = sizeof (con->actual_peer_addr);
1659 end += size;
1660 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1661 if (ret <= 0)
1662 goto out;
fd51653f
AE		 1663
1664 size = sizeof (con->peer_addr_for_me);
1665 end += size;
1666 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1667 if (ret <= 0)
1668 goto out;
fd51653f 1669
eed0ef2c
SW		 1670out:
1671 return ret;
1672}
1673
1674static int read_partial_connect(struct ceph_connection *con)
1675{
fd51653f
AE		 1676 int size;
1677 int end;
1678 int ret;
eed0ef2c
SW		 1679
1680 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1681
fd51653f
AE		 1682 size = sizeof (con->in_reply);
1683 end = size;
1684 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1685 if (ret <= 0)
1686 goto out;
fd51653f
AE		 1687
1688 size = le32_to_cpu(con->in_reply.authorizer_len);
1689 end += size;
1690 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1691 if (ret <= 0)
1692 goto out;
31b8006e 1693
4e7a5dcd
SW		 1694 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1695 con, (int)con->in_reply.tag,
1696 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1697 le32_to_cpu(con->in_reply.global_seq));
1698out:
1699 return ret;
eed0ef2c 1700
31b8006e
SW		 1701}
1702
1703/*
1704 * Verify the hello banner looks okay.
1705 */
1706static int verify_hello(struct ceph_connection *con)
1707{
1708 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1709 pr_err("connect to %s got bad banner\n",
3d14c5d2 1710 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1711 con->error_msg = "protocol error, bad banner";
1712 return -1;
1713 }
1714 return 0;
1715}
1716
1717static bool addr_is_blank(struct sockaddr_storage *ss)
1718{
1719 switch (ss->ss_family) {
1720 case AF_INET:
1721 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1722 case AF_INET6:
1723 return
1724 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1725 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1726 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1727 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1728 }
1729 return false;
1730}
1731
1732static int addr_port(struct sockaddr_storage *ss)
1733{
1734 switch (ss->ss_family) {
1735 case AF_INET:
f28bcfbe 1736 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1737 case AF_INET6:
f28bcfbe 1738 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1739 }
1740 return 0;
1741}
1742
1743static void addr_set_port(struct sockaddr_storage *ss, int p)
1744{
1745 switch (ss->ss_family) {
1746 case AF_INET:
1747 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1748 break;
31b8006e
SW		 1749 case AF_INET6:
1750 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1751 break;
31b8006e
SW		 1752 }
1753}
1754
ee3b56f2
NW		 1755/*
1756 * Unlike other *_pton function semantics, zero indicates success.
1757 */
1758static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1759 char delim, const char **ipend)
1760{
99f0f3b2
AE		 1761 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1762 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1763
1764 memset(ss, 0, sizeof(*ss));
1765
1766 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1767 ss->ss_family = AF_INET;
1768 return 0;
1769 }
1770
1771 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1772 ss->ss_family = AF_INET6;
1773 return 0;
1774 }
1775
1776 return -EINVAL;
1777}
1778
1779/*
1780 * Extract hostname string and resolve using kernel DNS facility.
1781 */
1782#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1783static int ceph_dns_resolve_name(const char *name, size_t namelen,
1784 struct sockaddr_storage *ss, char delim, const char **ipend)
1785{
1786 const char *end, *delim_p;
1787 char *colon_p, *ip_addr = NULL;
1788 int ip_len, ret;
1789
1790 /*
1791 * The end of the hostname occurs immediately preceding the delimiter or
1792 * the port marker (':') where the delimiter takes precedence.
1793 */
1794 delim_p = memchr(name, delim, namelen);
1795 colon_p = memchr(name, ':', namelen);
1796
1797 if (delim_p && colon_p)
1798 end = delim_p < colon_p ? delim_p : colon_p;
1799 else if (!delim_p && colon_p)
1800 end = colon_p;
1801 else {
1802 end = delim_p;
1803 if (!end) /* case: hostname:/ */
1804 end = name + namelen;
1805 }
1806
1807 if (end <= name)
1808 return -EINVAL;
1809
1810 /* do dns_resolve upcall */
1811 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1812 if (ip_len > 0)
1813 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1814 else
1815 ret = -ESRCH;
1816
1817 kfree(ip_addr);
1818
1819 *ipend = end;
1820
1821 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1822 ret, ret ? "failed" : ceph_pr_addr(ss));
1823
1824 return ret;
1825}
1826#else
1827static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1828 struct sockaddr_storage *ss, char delim, const char **ipend)
1829{
1830 return -EINVAL;
1831}
1832#endif
1833
1834/*
1835 * Parse a server name (IP or hostname). If a valid IP address is not found
1836 * then try to extract a hostname to resolve using userspace DNS upcall.
1837 */
1838static int ceph_parse_server_name(const char *name, size_t namelen,
1839 struct sockaddr_storage *ss, char delim, const char **ipend)
1840{
1841 int ret;
1842
1843 ret = ceph_pton(name, namelen, ss, delim, ipend);
1844 if (ret)
1845 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1846
1847 return ret;
1848}
1849
31b8006e
SW		 1850/*
1851 * Parse an ip[:port] list into an addr array. Use the default
1852 * monitor port if a port isn't specified.
1853 */
1854int ceph_parse_ips(const char *c, const char *end,
1855 struct ceph_entity_addr *addr,
1856 int max_count, int *count)
1857{
ee3b56f2 1858 int i, ret = -EINVAL;
31b8006e
SW		 1859 const char *p = c;
1860
1861 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1862 for (i = 0; i < max_count; i++) {
1863 const char *ipend;
1864 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1865 int port;
39139f64
SW		 1866 char delim = ',';
1867
1868 if (*p == '[') {
1869 delim = ']';
1870 p++;
1871 }
31b8006e 1872
ee3b56f2
NW		 1873 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1874 if (ret)
31b8006e 1875 goto bad;
ee3b56f2
NW		 1876 ret = -EINVAL;
1877
31b8006e
SW		 1878 p = ipend;
1879
39139f64
SW		 1880 if (delim == ']') {
1881 if (*p != ']') {
1882 dout("missing matching ']'\n");
1883 goto bad;
1884 }
1885 p++;
1886 }
1887
31b8006e
SW		 1888 /* port? */
1889 if (p < end && *p == ':') {
1890 port = 0;
1891 p++;
1892 while (p < end && *p >= '0' && *p <= '9') {
1893 port = (port * 10) + (*p - '0');
1894 p++;
1895 }
f48db1e9
ID		 1896 if (port == 0)
1897 port = CEPH_MON_PORT;
1898 else if (port > 65535)
31b8006e
SW		 1899 goto bad;
1900 } else {
1901 port = CEPH_MON_PORT;
1902 }
1903
1904 addr_set_port(ss, port);
1905
3d14c5d2 1906 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1907
1908 if (p == end)
1909 break;
1910 if (*p != ',')
1911 goto bad;
1912 p++;
1913 }
1914
1915 if (p != end)
1916 goto bad;
1917
1918 if (count)
1919 *count = i + 1;
1920 return 0;
1921
1922bad:
39139f64 1923 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1924 return ret;
31b8006e 1925}
3d14c5d2 1926EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1927
eed0ef2c 1928static int process_banner(struct ceph_connection *con)
31b8006e 1929{
eed0ef2c 1930 dout("process_banner on %p\n", con);
31b8006e
SW		 1931
1932 if (verify_hello(con) < 0)
1933 return -1;
1934
63f2d211
SW		 1935 ceph_decode_addr(&con->actual_peer_addr);
1936 ceph_decode_addr(&con->peer_addr_for_me);
1937
31b8006e
SW		 1938 /*
1939 * Make sure the other end is who we wanted. note that the other
1940 * end may not yet know their ip address, so if it's 0.0.0.0, give
1941 * them the benefit of the doubt.
1942 */
103e2d3a
SW		 1943 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1944 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1945 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1946 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 1947 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1948 ceph_pr_addr(&con->peer_addr.in_addr),
1949 (int)le32_to_cpu(con->peer_addr.nonce),
1950 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1951 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1952 con->error_msg = "wrong peer at address";
31b8006e
SW		 1953 return -1;
1954 }
1955
1956 /*
1957 * did we learn our address?
1958 */
1959 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1960 int port = addr_port(&con->msgr->inst.addr.in_addr);
1961
1962 memcpy(&con->msgr->inst.addr.in_addr,
1963 &con->peer_addr_for_me.in_addr,
1964 sizeof(con->peer_addr_for_me.in_addr));
1965 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1966 encode_my_addr(con->msgr);
eed0ef2c 1967 dout("process_banner learned my addr is %s\n",
3d14c5d2 1968 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 1969 }
1970
eed0ef2c
SW		 1971 return 0;
1972}
1973
1974static int process_connect(struct ceph_connection *con)
1975{
3d14c5d2
YS		 1976 u64 sup_feat = con->msgr->supported_features;
1977 u64 req_feat = con->msgr->required_features;
2b3e0c90
ID		 1978 u64 server_feat = ceph_sanitize_features(
1979 le64_to_cpu(con->in_reply.features));
0da5d703 1980 int ret;
04a419f9 1981
eed0ef2c
SW		 1982 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1983
31b8006e 1984 switch (con->in_reply.tag) {
04a419f9
SW		 1985 case CEPH_MSGR_TAG_FEATURES:
1986 pr_err("%s%lld %s feature set mismatch,"
1987 " my %llx < server's %llx, missing %llx\n",
1988 ENTITY_NAME(con->peer_name),
3d14c5d2 1989 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 1990 sup_feat, server_feat, server_feat & ~sup_feat);
1991 con->error_msg = "missing required protocol features";
0fa6ebc6 1992 reset_connection(con);
04a419f9
SW		 1993 return -1;
1994
31b8006e 1995 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 1996 pr_err("%s%lld %s protocol version mismatch,"
1997 " my %d != server's %d\n",
1998 ENTITY_NAME(con->peer_name),
3d14c5d2 1999 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2000 le32_to_cpu(con->out_connect.protocol_version),
2001 le32_to_cpu(con->in_reply.protocol_version));
2002 con->error_msg = "protocol version mismatch";
0fa6ebc6 2003 reset_connection(con);
31b8006e
SW		 2004 return -1;
2005
4e7a5dcd
SW		 2006 case CEPH_MSGR_TAG_BADAUTHORIZER:
2007 con->auth_retry++;
2008 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2009 con->auth_retry);
2010 if (con->auth_retry == 2) {
2011 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2012 return -1;
2013 }
6d4221b5 2014 con_out_kvec_reset(con);
e825a66d 2015 ret = prepare_write_connect(con);
0da5d703
SW		 2016 if (ret < 0)
2017 return ret;
63733a0f 2018 prepare_read_connect(con);
4e7a5dcd 2019 break;
31b8006e
SW		 2020
2021 case CEPH_MSGR_TAG_RESETSESSION:
2022 /*
2023 * If we connected with a large connect_seq but the peer
2024 * has no record of a session with us (no connection, or
2025 * connect_seq == 0), they will send RESETSESION to indicate
2026 * that they must have reset their session, and may have
2027 * dropped messages.
2028 */
2029 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2030 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2031 pr_err("%s%lld %s connection reset\n",
2032 ENTITY_NAME(con->peer_name),
3d14c5d2 2033 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2034 reset_connection(con);
6d4221b5 2035 con_out_kvec_reset(con);
5a0f8fdd
AE		 2036 ret = prepare_write_connect(con);
2037 if (ret < 0)
2038 return ret;
31b8006e
SW		 2039 prepare_read_connect(con);
2040
2041 /* Tell ceph about it. */
ec302645 2042 mutex_unlock(&con->mutex);
31b8006e
SW		 2043 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2044 if (con->ops->peer_reset)
2045 con->ops->peer_reset(con);
ec302645 2046 mutex_lock(&con->mutex);
8dacc7da 2047 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2048 return -EAGAIN;
31b8006e
SW		 2049 break;
2050
2051 case CEPH_MSGR_TAG_RETRY_SESSION:
2052 /*
2053 * If we sent a smaller connect_seq than the peer has, try
2054 * again with a larger value.
2055 */
5bdca4e0 2056 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2057 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2058 le32_to_cpu(con->in_reply.connect_seq));
2059 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2060 con_out_kvec_reset(con);
5a0f8fdd
AE		 2061 ret = prepare_write_connect(con);
2062 if (ret < 0)
2063 return ret;
31b8006e
SW		 2064 prepare_read_connect(con);
2065 break;
2066
2067 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2068 /*
2069 * If we sent a smaller global_seq than the peer has, try
2070 * again with a larger value.
2071 */
eed0ef2c 2072 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2073 con->peer_global_seq,
5bdca4e0 2074 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2075 get_global_seq(con->msgr,
5bdca4e0 2076 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2077 con_out_kvec_reset(con);
5a0f8fdd
AE		 2078 ret = prepare_write_connect(con);
2079 if (ret < 0)
2080 return ret;
31b8006e
SW		 2081 prepare_read_connect(con);
2082 break;
2083
3a23083b 2084 case CEPH_MSGR_TAG_SEQ:
31b8006e 2085 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2086 if (req_feat & ~server_feat) {
2087 pr_err("%s%lld %s protocol feature mismatch,"
2088 " my required %llx > server's %llx, need %llx\n",
2089 ENTITY_NAME(con->peer_name),
3d14c5d2 2090 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2091 req_feat, server_feat, req_feat & ~server_feat);
2092 con->error_msg = "missing required protocol features";
0fa6ebc6 2093 reset_connection(con);
04a419f9
SW		 2094 return -1;
2095 }
8dacc7da 2096
122070a2 2097 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2098 con->state = CON_STATE_OPEN;
20e55c4c 2099 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2100 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2101 con->connect_seq++;
aba558e2 2102 con->peer_features = server_feat;
31b8006e
SW		 2103 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2104 con->peer_global_seq,
2105 le32_to_cpu(con->in_reply.connect_seq),
2106 con->connect_seq);
2107 WARN_ON(con->connect_seq !=
2108 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2109
2110 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2111 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2112
85effe18 2113 con->delay = 0; /* reset backoff memory */
92ac41d0 2114
3a23083b
SW		 2115 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2116 prepare_write_seq(con);
2117 prepare_read_seq(con);
2118 } else {
2119 prepare_read_tag(con);
2120 }
31b8006e
SW		 2121 break;
2122
2123 case CEPH_MSGR_TAG_WAIT:
2124 /*
2125 * If there is a connection race (we are opening
2126 * connections to each other), one of us may just have
2127 * to WAIT. This shouldn't happen if we are the
2128 * client.
2129 */
04177882
SW		 2130 pr_err("process_connect got WAIT as client\n");
2131 con->error_msg = "protocol error, got WAIT as client";
2132 return -1;
31b8006e
SW		 2133
2134 default:
2135 pr_err("connect protocol error, will retry\n");
2136 con->error_msg = "protocol error, garbage tag during connect";
2137 return -1;
2138 }
2139 return 0;
2140}
2141
2142
2143/*
2144 * read (part of) an ack
2145 */
2146static int read_partial_ack(struct ceph_connection *con)
2147{
fd51653f
AE		 2148 int size = sizeof (con->in_temp_ack);
2149 int end = size;
31b8006e 2150
fd51653f 2151 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2152}
2153
31b8006e
SW		 2154/*
2155 * We can finally discard anything that's been acked.
2156 */
2157static void process_ack(struct ceph_connection *con)
2158{
2159 struct ceph_msg *m;
2160 u64 ack = le64_to_cpu(con->in_temp_ack);
2161 u64 seq;
2162
31b8006e
SW		 2163 while (!list_empty(&con->out_sent)) {
2164 m = list_first_entry(&con->out_sent, struct ceph_msg,
2165 list_head);
2166 seq = le64_to_cpu(m->hdr.seq);
2167 if (seq > ack)
2168 break;
2169 dout("got ack for seq %llu type %d at %p\n", seq,
2170 le16_to_cpu(m->hdr.type), m);
4cf9d544 2171 m->ack_stamp = jiffies;
31b8006e
SW		 2172 ceph_msg_remove(m);
2173 }
31b8006e
SW		 2174 prepare_read_tag(con);
2175}
2176
2177
2450418c 2178static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2179 struct kvec *section,
2180 unsigned int sec_len, u32 *crc)
2450418c 2181{
68b4476b 2182 int ret, left;
2450418c
YS		 2183
2184 BUG_ON(!section);
2185
2186 while (section->iov_len < sec_len) {
2187 BUG_ON(section->iov_base == NULL);
2188 left = sec_len - section->iov_len;
2189 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2190 section->iov_len, left);
2191 if (ret <= 0)
2192 return ret;
2193 section->iov_len += ret;
2450418c 2194 }
fe3ad593
AE		 2195 if (section->iov_len == sec_len)
2196 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2197
2450418c
YS		 2198 return 1;
2199}
31b8006e 2200
34d2d200
AE		 2201static int read_partial_msg_data(struct ceph_connection *con)
2202{
2203 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2204 struct ceph_msg_data_cursor *cursor = &msg->cursor;
34d2d200 2205 const bool do_datacrc = !con->msgr->nocrc;
686be208
AE		 2206 struct page *page;
2207 size_t page_offset;
2208 size_t length;
f5db90bc 2209 u32 crc = 0;
34d2d200
AE		 2210 int ret;
2211
2212 BUG_ON(!msg);
5240d9f9 2213 if (list_empty(&msg->data))
4c59b4a2 2214 return -EIO;
34d2d200 2215
f5db90bc
AE		 2216 if (do_datacrc)
2217 crc = con->in_data_crc;
643c68a4 2218 while (cursor->resid) {
8ae4f4f5 2219 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
4c59b4a2 2220 NULL);
686be208 2221 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2222 if (ret <= 0) {
2223 if (do_datacrc)
2224 con->in_data_crc = crc;
2225
686be208 2226 return ret;
f5db90bc 2227 }
686be208
AE		 2228
2229 if (do_datacrc)
f5db90bc 2230 crc = ceph_crc32c_page(crc, page, page_offset, ret);
8ae4f4f5 2231 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
34d2d200 2232 }
f5db90bc
AE		 2233 if (do_datacrc)
2234 con->in_data_crc = crc;
34d2d200
AE		 2235
2236 return 1; /* must return > 0 to indicate success */
2237}
2238
31b8006e
SW		 2239/*
2240 * read (part of) a message.
2241 */
686be208
AE		 2242static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2243
31b8006e
SW		 2244static int read_partial_message(struct ceph_connection *con)
2245{
2246 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2247 int size;
2248 int end;
31b8006e 2249 int ret;
95c96174 2250 unsigned int front_len, middle_len, data_len;
37675b0f 2251 bool do_datacrc = !con->msgr->nocrc;
ae18756b 2252 u64 seq;
fe3ad593 2253 u32 crc;
31b8006e
SW		 2254
2255 dout("read_partial_message con %p msg %p\n", con, m);
2256
2257 /* header */
fd51653f
AE		 2258 size = sizeof (con->in_hdr);
2259 end = size;
2260 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2261 if (ret <= 0)
2262 return ret;
fe3ad593
AE		 2263
2264 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2265 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2266 pr_err("read_partial_message bad hdr "
2267 " crc %u != expected %u\n",
2268 crc, con->in_hdr.crc);
2269 return -EBADMSG;
2270 }
2271
31b8006e
SW		 2272 front_len = le32_to_cpu(con->in_hdr.front_len);
2273 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2274 return -EIO;
2275 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2276 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2277 return -EIO;
2278 data_len = le32_to_cpu(con->in_hdr.data_len);
2279 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2280 return -EIO;
2281
ae18756b
SW		 2282 /* verify seq# */
2283 seq = le64_to_cpu(con->in_hdr.seq);
2284 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2285 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2286 ENTITY_NAME(con->peer_name),
3d14c5d2 2287 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2288 seq, con->in_seq + 1);
2289 con->in_base_pos = -front_len - middle_len - data_len -
2290 sizeof(m->footer);
2291 con->in_tag = CEPH_MSGR_TAG_READY;
ae18756b
SW		 2292 return 0;
2293 } else if ((s64)seq - (s64)con->in_seq > 1) {
2294 pr_err("read_partial_message bad seq %lld expected %lld\n",
2295 seq, con->in_seq + 1);
2296 con->error_msg = "bad message sequence # for incoming message";
2297 return -EBADMSG;
2298 }
2299
31b8006e
SW		 2300 /* allocate message? */
2301 if (!con->in_msg) {
4740a623
SW		 2302 int skip = 0;
2303
31b8006e 2304 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2305 front_len, data_len);
4740a623
SW		 2306 ret = ceph_con_in_msg_alloc(con, &skip);
2307 if (ret < 0)
2308 return ret;
f759ebb9
AE		 2309
2310 BUG_ON(!con->in_msg ^ skip);
2311 if (con->in_msg && data_len > con->in_msg->data_length) {
b9a67899 2312 pr_warn("%s skipping long message (%u > %zd)\n",
f759ebb9
AE		 2313 __func__, data_len, con->in_msg->data_length);
2314 ceph_msg_put(con->in_msg);
2315 con->in_msg = NULL;
2316 skip = 1;
2317 }
2450418c 2318 if (skip) {
31b8006e 2319 /* skip this message */
a79832f2 2320 dout("alloc_msg said skip message\n");
31b8006e
SW		 2321 con->in_base_pos = -front_len - middle_len - data_len -
2322 sizeof(m->footer);
2323 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2324 con->in_seq++;
31b8006e
SW		 2325 return 0;
2326 }
38941f80 2327
4740a623 2328 BUG_ON(!con->in_msg);
38941f80 2329 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2330 m = con->in_msg;
2331 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2332 if (m->middle)
2333 m->middle->vec.iov_len = 0;
9d7f0f13 2334
78625051 2335 /* prepare for data payload, if any */
a4107026 2336
78625051 2337 if (data_len)
98fa5dd8 2338 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2339 }
2340
2341 /* front */
2450418c
YS		 2342 ret = read_partial_message_section(con, &m->front, front_len,
2343 &con->in_front_crc);
2344 if (ret <= 0)
2345 return ret;
31b8006e
SW		 2346
2347 /* middle */
2450418c 2348 if (m->middle) {
213c99ee
SW		 2349 ret = read_partial_message_section(con, &m->middle->vec,
2350 middle_len,
2450418c 2351 &con->in_middle_crc);
31b8006e
SW		 2352 if (ret <= 0)
2353 return ret;
31b8006e
SW		 2354 }
2355
2356 /* (page) data */
34d2d200
AE		 2357 if (data_len) {
2358 ret = read_partial_msg_data(con);
2359 if (ret <= 0)
2360 return ret;
31b8006e
SW		 2361 }
2362
31b8006e 2363 /* footer */
fd51653f
AE		 2364 size = sizeof (m->footer);
2365 end += size;
2366 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2367 if (ret <= 0)
2368 return ret;
2369
31b8006e
SW		 2370 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2371 m, front_len, m->footer.front_crc, middle_len,
2372 m->footer.middle_crc, data_len, m->footer.data_crc);
2373
2374 /* crc ok? */
2375 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2376 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2377 m, con->in_front_crc, m->footer.front_crc);
2378 return -EBADMSG;
2379 }
2380 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2381 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2382 m, con->in_middle_crc, m->footer.middle_crc);
2383 return -EBADMSG;
2384 }
bca064d2 2385 if (do_datacrc &&
31b8006e
SW		 2386 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2387 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2388 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2389 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2390 return -EBADMSG;
2391 }
2392
2393 return 1; /* done! */
2394}
2395
2396/*
2397 * Process message. This happens in the worker thread. The callback should
2398 * be careful not to do anything that waits on other incoming messages or it
2399 * may deadlock.
2400 */
2401static void process_message(struct ceph_connection *con)
2402{
5e095e8b 2403 struct ceph_msg *msg;
31b8006e 2404
38941f80
AE		 2405 BUG_ON(con->in_msg->con != con);
2406 con->in_msg->con = NULL;
5e095e8b 2407 msg = con->in_msg;
31b8006e 2408 con->in_msg = NULL;
36eb71aa 2409 con->ops->put(con);
31b8006e
SW		 2410
2411 /* if first message, set peer_name */
2412 if (con->peer_name.type == 0)
dbad185d 2413 con->peer_name = msg->hdr.src;
31b8006e 2414
31b8006e 2415 con->in_seq++;
ec302645 2416 mutex_unlock(&con->mutex);
31b8006e
SW		 2417
2418 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2419 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2420 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2421 le16_to_cpu(msg->hdr.type),
2422 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2423 le32_to_cpu(msg->hdr.front_len),
2424 le32_to_cpu(msg->hdr.data_len),
2425 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2426 con->ops->dispatch(con, msg);
ec302645
SW		 2427
2428 mutex_lock(&con->mutex);
31b8006e
SW		 2429}
2430
2431
2432/*
2433 * Write something to the socket. Called in a worker thread when the
2434 * socket appears to be writeable and we have something ready to send.
2435 */
2436static int try_write(struct ceph_connection *con)
2437{
31b8006e
SW		 2438 int ret = 1;
2439
d59315ca 2440 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2441
31b8006e
SW		 2442more:
2443 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2444
2445 /* open the socket first? */
8dacc7da
SW		 2446 if (con->state == CON_STATE_PREOPEN) {
2447 BUG_ON(con->sock);
2448 con->state = CON_STATE_CONNECTING;
a5988c49 2449
e2200423 2450 con_out_kvec_reset(con);
e825a66d 2451 prepare_write_banner(con);
eed0ef2c 2452 prepare_read_banner(con);
31b8006e 2453
cf3e5c40 2454 BUG_ON(con->in_msg);
31b8006e
SW		 2455 con->in_tag = CEPH_MSGR_TAG_READY;
2456 dout("try_write initiating connect on %p new state %lu\n",
2457 con, con->state);
41617d0c
AE		 2458 ret = ceph_tcp_connect(con);
2459 if (ret < 0) {
31b8006e 2460 con->error_msg = "connect error";
31b8006e
SW		 2461 goto out;
2462 }
2463 }
2464
2465more_kvec:
2466 /* kvec data queued? */
2467 if (con->out_skip) {
2468 ret = write_partial_skip(con);
2469 if (ret <= 0)
42961d23 2470 goto out;
31b8006e
SW		 2471 }
2472 if (con->out_kvec_left) {
2473 ret = write_partial_kvec(con);
2474 if (ret <= 0)
42961d23 2475 goto out;
31b8006e
SW		 2476 }
2477
2478 /* msg pages? */
2479 if (con->out_msg) {
c86a2930
SW		 2480 if (con->out_msg_done) {
2481 ceph_msg_put(con->out_msg);
2482 con->out_msg = NULL; /* we're done with this one */
2483 goto do_next;
2484 }
2485
34d2d200 2486 ret = write_partial_message_data(con);
31b8006e
SW		 2487 if (ret == 1)
2488 goto more_kvec; /* we need to send the footer, too! */
2489 if (ret == 0)
42961d23 2490 goto out;
31b8006e 2491 if (ret < 0) {
34d2d200 2492 dout("try_write write_partial_message_data err %d\n",
31b8006e 2493 ret);
42961d23 2494 goto out;
31b8006e
SW		 2495 }
2496 }
2497
c86a2930 2498do_next:
8dacc7da 2499 if (con->state == CON_STATE_OPEN) {
31b8006e
SW		 2500 /* is anything else pending? */
2501 if (!list_empty(&con->out_queue)) {
2502 prepare_write_message(con);
2503 goto more;
2504 }
2505 if (con->in_seq > con->in_seq_acked) {
2506 prepare_write_ack(con);
2507 goto more;
2508 }
c9ffc77a 2509 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
31b8006e
SW		 2510 prepare_write_keepalive(con);
2511 goto more;
2512 }
2513 }
2514
2515 /* Nothing to do! */
c9ffc77a 2516 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2517 dout("try_write nothing else to write.\n");
31b8006e
SW		 2518 ret = 0;
2519out:
42961d23 2520 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2521 return ret;
2522}
2523
2524
2525
2526/*
2527 * Read what we can from the socket.
2528 */
2529static int try_read(struct ceph_connection *con)
2530{
31b8006e
SW		 2531 int ret = -1;
2532
8dacc7da
SW		 2533more:
2534 dout("try_read start on %p state %lu\n", con, con->state);
2535 if (con->state != CON_STATE_CONNECTING &&
2536 con->state != CON_STATE_NEGOTIATING &&
2537 con->state != CON_STATE_OPEN)
31b8006e
SW		 2538 return 0;
2539
8dacc7da 2540 BUG_ON(!con->sock);
ec302645 2541
31b8006e
SW		 2542 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2543 con->in_base_pos);
0da5d703 2544
8dacc7da 2545 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2546 dout("try_read connecting\n");
2547 ret = read_partial_banner(con);
2548 if (ret <= 0)
ab166d5a 2549 goto out;
7593af92
AE		 2550 ret = process_banner(con);
2551 if (ret < 0)
2552 goto out;
2553
8dacc7da 2554 con->state = CON_STATE_NEGOTIATING;
7593af92 2555
6d4221b5
JS		 2556 /*
2557 * Received banner is good, exchange connection info.
2558 * Do not reset out_kvec, as sending our banner raced
2559 * with receiving peer banner after connect completed.
2560 */
7593af92
AE		 2561 ret = prepare_write_connect(con);
2562 if (ret < 0)
2563 goto out;
2564 prepare_read_connect(con);
2565
2566 /* Send connection info before awaiting response */
0da5d703
SW		 2567 goto out;
2568 }
2569
8dacc7da 2570 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2571 dout("try_read negotiating\n");
31b8006e
SW		 2572 ret = read_partial_connect(con);
2573 if (ret <= 0)
31b8006e 2574 goto out;
98bdb0aa
SW		 2575 ret = process_connect(con);
2576 if (ret < 0)
2577 goto out;
31b8006e
SW		 2578 goto more;
2579 }
2580
122070a2 2581 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2582
31b8006e
SW		 2583 if (con->in_base_pos < 0) {
2584 /*
2585 * skipping + discarding content.
2586 *
2587 * FIXME: there must be a better way to do this!
2588 */
84495f49
AE		 2589 static char buf[SKIP_BUF_SIZE];
2590 int skip = min((int) sizeof (buf), -con->in_base_pos);
2591
31b8006e
SW		 2592 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2593 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2594 if (ret <= 0)
98bdb0aa 2595 goto out;
31b8006e
SW		 2596 con->in_base_pos += ret;
2597 if (con->in_base_pos)
2598 goto more;
2599 }
2600 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2601 /*
2602 * what's next?
2603 */
2604 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2605 if (ret <= 0)
98bdb0aa 2606 goto out;
31b8006e
SW		 2607 dout("try_read got tag %d\n", (int)con->in_tag);
2608 switch (con->in_tag) {
2609 case CEPH_MSGR_TAG_MSG:
2610 prepare_read_message(con);
2611 break;
2612 case CEPH_MSGR_TAG_ACK:
2613 prepare_read_ack(con);
2614 break;
2615 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2616 con_close_socket(con);
2617 con->state = CON_STATE_CLOSED;
98bdb0aa 2618 goto out;
31b8006e
SW		 2619 default:
2620 goto bad_tag;
2621 }
2622 }
2623 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2624 ret = read_partial_message(con);
2625 if (ret <= 0) {
2626 switch (ret) {
2627 case -EBADMSG:
2628 con->error_msg = "bad crc";
2629 ret = -EIO;
98bdb0aa 2630 break;
31b8006e
SW		 2631 case -EIO:
2632 con->error_msg = "io error";
98bdb0aa 2633 break;
31b8006e 2634 }
98bdb0aa 2635 goto out;
31b8006e
SW		 2636 }
2637 if (con->in_tag == CEPH_MSGR_TAG_READY)
2638 goto more;
2639 process_message(con);
7b862e07
SW		 2640 if (con->state == CON_STATE_OPEN)
2641 prepare_read_tag(con);
31b8006e
SW		 2642 goto more;
2643 }
3a23083b
SW		 2644 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2645 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2646 /*
2647 * the final handshake seq exchange is semantically
2648 * equivalent to an ACK
2649 */
31b8006e
SW		 2650 ret = read_partial_ack(con);
2651 if (ret <= 0)
98bdb0aa 2652 goto out;
31b8006e
SW		 2653 process_ack(con);
2654 goto more;
2655 }
2656
31b8006e 2657out:
98bdb0aa 2658 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2659 return ret;
2660
2661bad_tag:
2662 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2663 con->error_msg = "protocol error, garbage tag";
2664 ret = -1;
2665 goto out;
2666}
2667
2668
2669/*
802c6d96
AE		 2670 * Atomically queue work on a connection after the specified delay.
2671 * Bump @con reference to avoid races with connection teardown.
2672 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2673 */
802c6d96 2674static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2675{
31b8006e 2676 if (!con->ops->get(con)) {
802c6d96 2677 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2678 return -ENOENT;
31b8006e
SW		 2679 }
2680
802c6d96
AE		 2681 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2682 dout("%s %p - already queued\n", __func__, con);
31b8006e 2683 con->ops->put(con);
802c6d96 2684 return -EBUSY;
31b8006e 2685 }
802c6d96
AE		 2686
2687 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2688 return 0;
2689}
2690
2691static void queue_con(struct ceph_connection *con)
2692{
2693 (void) queue_con_delay(con, 0);
31b8006e
SW		 2694}
2695
37ab77ac
ID		 2696static void cancel_con(struct ceph_connection *con)
2697{
2698 if (cancel_delayed_work(&con->work)) {
2699 dout("%s %p\n", __func__, con);
2700 con->ops->put(con);
2701 }
2702}
2703
7bb21d68
AE		 2704static bool con_sock_closed(struct ceph_connection *con)
2705{
c9ffc77a 2706 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2707 return false;
2708
2709#define CASE(x) \
2710 case CON_STATE_ ## x: \
2711 con->error_msg = "socket closed (con state " #x ")"; \
2712 break;
2713
2714 switch (con->state) {
2715 CASE(CLOSED);
2716 CASE(PREOPEN);
2717 CASE(CONNECTING);
2718 CASE(NEGOTIATING);
2719 CASE(OPEN);
2720 CASE(STANDBY);
2721 default:
b9a67899 2722 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2723 __func__, con, con->state);
2724 con->error_msg = "unrecognized con state";
2725 BUG();
2726 break;
2727 }
2728#undef CASE
2729
2730 return true;
2731}
2732
f20a39fd
AE		 2733static bool con_backoff(struct ceph_connection *con)
2734{
2735 int ret;
2736
2737 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2738 return false;
2739
2740 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2741 if (ret) {
2742 dout("%s: con %p FAILED to back off %lu\n", __func__,
2743 con, con->delay);
2744 BUG_ON(ret == -ENOENT);
2745 con_flag_set(con, CON_FLAG_BACKOFF);
2746 }
2747
2748 return true;
2749}
2750
93209264
AE		 2751/* Finish fault handling; con->mutex must *not* be held here */
2752
2753static void con_fault_finish(struct ceph_connection *con)
2754{
2755 /*
2756 * in case we faulted due to authentication, invalidate our
2757 * current tickets so that we can get new ones.
2758 */
2759 if (con->auth_retry && con->ops->invalidate_authorizer) {
2760 dout("calling invalidate_authorizer()\n");
2761 con->ops->invalidate_authorizer(con);
2762 }
2763
2764 if (con->ops->fault)
2765 con->ops->fault(con);
2766}
2767
31b8006e
SW		 2768/*
2769 * Do some work on a connection. Drop a connection ref when we're done.
2770 */
2771static void con_work(struct work_struct *work)
2772{
2773 struct ceph_connection *con = container_of(work, struct ceph_connection,
2774 work.work);
89baaa57 2775 unsigned long pflags = current->flags;
49659416 2776 bool fault;
31b8006e 2777
89baaa57
MC		 2778 current->flags |= PF_MEMALLOC;
2779
9dd4658d 2780 mutex_lock(&con->mutex);
49659416
AE		 2781 while (true) {
2782 int ret;
31b8006e 2783
49659416
AE		 2784 if ((fault = con_sock_closed(con))) {
2785 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2786 break;
2787 }
2788 if (con_backoff(con)) {
2789 dout("%s: con %p BACKOFF\n", __func__, con);
2790 break;
2791 }
2792 if (con->state == CON_STATE_STANDBY) {
2793 dout("%s: con %p STANDBY\n", __func__, con);
2794 break;
2795 }
2796 if (con->state == CON_STATE_CLOSED) {
2797 dout("%s: con %p CLOSED\n", __func__, con);
2798 BUG_ON(con->sock);
2799 break;
2800 }
2801 if (con->state == CON_STATE_PREOPEN) {
2802 dout("%s: con %p PREOPEN\n", __func__, con);
2803 BUG_ON(con->sock);
2804 }
0da5d703 2805
49659416
AE		 2806 ret = try_read(con);
2807 if (ret < 0) {
2808 if (ret == -EAGAIN)
2809 continue;
2810 con->error_msg = "socket error on read";
2811 fault = true;
2812 break;
2813 }
2814
2815 ret = try_write(con);
2816 if (ret < 0) {
2817 if (ret == -EAGAIN)
2818 continue;
2819 con->error_msg = "socket error on write";
2820 fault = true;
2821 }
2822
2823 break; /* If we make it to here, we're done */
3a140a0d 2824 }
b6e7b6a1
AE		 2825 if (fault)
2826 con_fault(con);
9dd4658d 2827 mutex_unlock(&con->mutex);
0da5d703 2828
b6e7b6a1
AE		 2829 if (fault)
2830 con_fault_finish(con);
2831
2832 con->ops->put(con);
89baaa57
MC		 2833
2834 tsk_restore_flags(current, pflags, PF_MEMALLOC);
31b8006e
SW		 2835}
2836
31b8006e
SW		 2837/*
2838 * Generic error/fault handler. A retry mechanism is used with
2839 * exponential backoff
2840 */
93209264 2841static void con_fault(struct ceph_connection *con)
31b8006e 2842{
b9a67899
JP		 2843 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2844 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
31b8006e 2845 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2846 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2847
122070a2 2848 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2849 con->state != CON_STATE_NEGOTIATING &&
2850 con->state != CON_STATE_OPEN);
ec302645 2851
31b8006e 2852 con_close_socket(con);
5e095e8b 2853
c9ffc77a 2854 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2855 dout("fault on LOSSYTX channel, marking CLOSED\n");
2856 con->state = CON_STATE_CLOSED;
93209264 2857 return;
3b5ede07
SW		 2858 }
2859
5e095e8b 2860 if (con->in_msg) {
38941f80
AE		 2861 BUG_ON(con->in_msg->con != con);
2862 con->in_msg->con = NULL;
5e095e8b
SW		 2863 ceph_msg_put(con->in_msg);
2864 con->in_msg = NULL;
36eb71aa 2865 con->ops->put(con);
5e095e8b 2866 }
31b8006e 2867
e80a52d1
SW		 2868 /* Requeue anything that hasn't been acked */
2869 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2870
e76661d0
SW		 2871 /* If there are no messages queued or keepalive pending, place
2872 * the connection in a STANDBY state */
2873 if (list_empty(&con->out_queue) &&
c9ffc77a 2874 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2875 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2876 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2877 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2878 } else {
2879 /* retry after a delay. */
8dacc7da 2880 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2881 if (con->delay == 0)
2882 con->delay = BASE_DELAY_INTERVAL;
2883 else if (con->delay < MAX_DELAY_INTERVAL)
2884 con->delay *= 2;
c9ffc77a 2885 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2886 queue_con(con);
31b8006e 2887 }
31b8006e
SW		 2888}
2889
2890
2891
2892/*
15d9882c 2893 * initialize a new messenger instance
31b8006e 2894 */
15d9882c
AE		 2895void ceph_messenger_init(struct ceph_messenger *msgr,
2896 struct ceph_entity_addr *myaddr,
12b4629a
ID		 2897 u64 supported_features,
2898 u64 required_features,
15d9882c 2899 bool nocrc)
31b8006e 2900{
3d14c5d2
YS		 2901 msgr->supported_features = supported_features;
2902 msgr->required_features = required_features;
2903
31b8006e
SW		 2904 spin_lock_init(&msgr->global_seq_lock);
2905
31b8006e
SW		 2906 if (myaddr)
2907 msgr->inst.addr = *myaddr;
2908
2909 /* select a random nonce */
ac8839d7 2910 msgr->inst.addr.type = 0;
103e2d3a 2911 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2912 encode_my_addr(msgr);
15d9882c 2913 msgr->nocrc = nocrc;
31b8006e 2914
a2a32584 2915 atomic_set(&msgr->stopping, 0);
31b8006e 2916
15d9882c 2917 dout("%s %p\n", __func__, msgr);
31b8006e 2918}
15d9882c 2919EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2920
e00de341
SW		 2921static void clear_standby(struct ceph_connection *con)
2922{
2923 /* come back from STANDBY? */
8dacc7da 2924 if (con->state == CON_STATE_STANDBY) {
e00de341 2925 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 2926 con->state = CON_STATE_PREOPEN;
e00de341 2927 con->connect_seq++;
c9ffc77a
AE		 2928 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2929 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 2930 }
2931}
2932
31b8006e
SW		 2933/*
2934 * Queue up an outgoing message on the given connection.
2935 */
2936void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2937{
31b8006e 2938 /* set src+dst */
dbad185d 2939 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 2940 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 2941 msg->needs_out_seq = true;
2942
ec302645 2943 mutex_lock(&con->mutex);
92ce034b 2944
8dacc7da 2945 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 2946 dout("con_send %p closed, dropping %p\n", con, msg);
2947 ceph_msg_put(msg);
2948 mutex_unlock(&con->mutex);
2949 return;
2950 }
2951
38941f80 2952 BUG_ON(msg->con != NULL);
36eb71aa 2953 msg->con = con->ops->get(con);
92ce034b
AE		 2954 BUG_ON(msg->con == NULL);
2955
31b8006e
SW		 2956 BUG_ON(!list_empty(&msg->list_head));
2957 list_add_tail(&msg->list_head, &con->out_queue);
2958 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2959 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2960 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2961 le32_to_cpu(msg->hdr.front_len),
2962 le32_to_cpu(msg->hdr.middle_len),
2963 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 2964
2965 clear_standby(con);
ec302645 2966 mutex_unlock(&con->mutex);
31b8006e
SW		 2967
2968 /* if there wasn't anything waiting to send before, queue
2969 * new work */
c9ffc77a 2970 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 2971 queue_con(con);
2972}
3d14c5d2 2973EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 2974
2975/*
2976 * Revoke a message that was previously queued for send
2977 */
6740a845 2978void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 2979{
6740a845
AE		 2980 struct ceph_connection *con = msg->con;
2981
2982 if (!con)
2983 return; /* Message not in our possession */
2984
ec302645 2985 mutex_lock(&con->mutex);
31b8006e 2986 if (!list_empty(&msg->list_head)) {
38941f80 2987 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 2988 list_del_init(&msg->list_head);
38941f80 2989 BUG_ON(msg->con == NULL);
36eb71aa 2990 msg->con->ops->put(msg->con);
38941f80 2991 msg->con = NULL;
31b8006e 2992 msg->hdr.seq = 0;
38941f80 2993
31b8006e 2994 ceph_msg_put(msg);
ed98adad
SW		 2995 }
2996 if (con->out_msg == msg) {
38941f80 2997 dout("%s %p msg %p - was sending\n", __func__, con, msg);
ed98adad 2998 con->out_msg = NULL;
31b8006e
SW		 2999 if (con->out_kvec_is_msg) {
3000 con->out_skip = con->out_kvec_bytes;
3001 con->out_kvec_is_msg = false;
3002 }
ed98adad 3003 msg->hdr.seq = 0;
92ce034b
AE		 3004
3005 ceph_msg_put(msg);
31b8006e 3006 }
ec302645 3007 mutex_unlock(&con->mutex);
31b8006e
SW		 3008}
3009
350b1c32 3010/*
0d59ab81 3011 * Revoke a message that we may be reading data into
350b1c32 3012 */
8921d114 3013void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3014{
8921d114
AE		 3015 struct ceph_connection *con;
3016
3017 BUG_ON(msg == NULL);
3018 if (!msg->con) {
3019 dout("%s msg %p null con\n", __func__, msg);
3020
3021 return; /* Message not in our possession */
3022 }
3023
3024 con = msg->con;
350b1c32 3025 mutex_lock(&con->mutex);
8921d114 3026 if (con->in_msg == msg) {
95c96174
ED		 3027 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3028 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3029 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3030
3031 /* skip rest of message */
8921d114
AE		 3032 dout("%s %p msg %p revoked\n", __func__, con, msg);
3033 con->in_base_pos = con->in_base_pos -
350b1c32 3034 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3035 front_len -
3036 middle_len -
3037 data_len -
350b1c32 3038 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3039 ceph_msg_put(con->in_msg);
3040 con->in_msg = NULL;
3041 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3042 con->in_seq++;
350b1c32 3043 } else {
8921d114
AE		 3044 dout("%s %p in_msg %p msg %p no-op\n",
3045 __func__, con, con->in_msg, msg);
350b1c32
SW		 3046 }
3047 mutex_unlock(&con->mutex);
3048}
3049
31b8006e
SW		 3050/*
3051 * Queue a keepalive byte to ensure the tcp connection is alive.
3052 */
3053void ceph_con_keepalive(struct ceph_connection *con)
3054{
e00de341 3055 dout("con_keepalive %p\n", con);
00650931 3056 mutex_lock(&con->mutex);
e00de341 3057 clear_standby(con);
00650931 3058 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3059 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3060 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3061 queue_con(con);
3062}
3d14c5d2 3063EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3064
6644ed7b 3065static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3066{
6644ed7b
AE		 3067 struct ceph_msg_data *data;
3068
3069 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3070 return NULL;
3071
81b36be4 3072 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
6644ed7b
AE		 3073 if (data)
3074 data->type = type;
5240d9f9 3075 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3076
3077 return data;
3078}
3079
3080static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3081{
3082 if (!data)
3083 return;
3084
5240d9f9 3085 WARN_ON(!list_empty(&data->links));
e4339d28 3086 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3087 ceph_pagelist_release(data->pagelist);
81b36be4 3088 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3089}
3090
90af3602 3091void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3092 size_t length, size_t alignment)
02afca6c 3093{
6644ed7b
AE		 3094 struct ceph_msg_data *data;
3095
07aa1558
AE		 3096 BUG_ON(!pages);
3097 BUG_ON(!length);
6644ed7b
AE		 3098
3099 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3100 BUG_ON(!data);
3101 data->pages = pages;
3102 data->length = length;
3103 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3104
5240d9f9
AE		 3105 list_add_tail(&data->links, &msg->data);
3106 msg->data_length += length;
02afca6c 3107}
90af3602 3108EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3109
90af3602 3110void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3111 struct ceph_pagelist *pagelist)
3112{
6644ed7b
AE		 3113 struct ceph_msg_data *data;
3114
07aa1558
AE		 3115 BUG_ON(!pagelist);
3116 BUG_ON(!pagelist->length);
27fa8385 3117
6644ed7b
AE		 3118 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3119 BUG_ON(!data);
3120 data->pagelist = pagelist;
3121
5240d9f9
AE		 3122 list_add_tail(&data->links, &msg->data);
3123 msg->data_length += pagelist->length;
27fa8385 3124}
90af3602 3125EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3126
ea96571f 3127#ifdef CONFIG_BLOCK
90af3602 3128void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3129 size_t length)
27fa8385 3130{
6644ed7b
AE		 3131 struct ceph_msg_data *data;
3132
07aa1558 3133 BUG_ON(!bio);
27fa8385 3134
6644ed7b
AE		 3135 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3136 BUG_ON(!data);
3137 data->bio = bio;
c851c495 3138 data->bio_length = length;
6644ed7b 3139
5240d9f9
AE		 3140 list_add_tail(&data->links, &msg->data);
3141 msg->data_length += length;
27fa8385 3142}
90af3602 3143EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3144#endif /* CONFIG_BLOCK */
27fa8385 3145
31b8006e
SW		 3146/*
3147 * construct a new message with given type, size
3148 * the new msg has a ref count of 1.
3149 */
b61c2763
SW		 3150struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3151 bool can_fail)
31b8006e
SW		 3152{
3153 struct ceph_msg *m;
3154
e3d5d638 3155 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3156 if (m == NULL)
3157 goto out;
31b8006e
SW		 3158
3159 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3160 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3161 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3162
9516e45b
AE		 3163 INIT_LIST_HEAD(&m->list_head);
3164 kref_init(&m->kref);
5240d9f9 3165 INIT_LIST_HEAD(&m->data);
ca20892d 3166
31b8006e
SW		 3167 /* front */
3168 if (front_len) {
eeb0bed5 3169 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3170 if (m->front.iov_base == NULL) {
b61c2763 3171 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3172 front_len);
3173 goto out2;
3174 }
3175 } else {
3176 m->front.iov_base = NULL;
3177 }
f2be82b0 3178 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3179
bb257664 3180 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3181 return m;
3182
3183out2:
3184 ceph_msg_put(m);
3185out:
b61c2763
SW		 3186 if (!can_fail) {
3187 pr_err("msg_new can't create type %d front %d\n", type,
3188 front_len);
f0ed1b7c 3189 WARN_ON(1);
b61c2763
SW		 3190 } else {
3191 dout("msg_new can't create type %d front %d\n", type,
3192 front_len);
3193 }
a79832f2 3194 return NULL;
31b8006e 3195}
3d14c5d2 3196EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3197
31b8006e
SW		 3198/*
3199 * Allocate "middle" portion of a message, if it is needed and wasn't
3200 * allocated by alloc_msg. This allows us to read a small fixed-size
3201 * per-type header in the front and then gracefully fail (i.e.,
3202 * propagate the error to the caller based on info in the front) when
3203 * the middle is too large.
3204 */
2450418c 3205static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3206{
3207 int type = le16_to_cpu(msg->hdr.type);
3208 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3209
3210 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3211 ceph_msg_type_name(type), middle_len);
3212 BUG_ON(!middle_len);
3213 BUG_ON(msg->middle);
3214
b6c1d5b8 3215 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3216 if (!msg->middle)
3217 return -ENOMEM;
3218 return 0;
3219}
3220
2450418c 3221/*
1c20f2d2
AE		 3222 * Allocate a message for receiving an incoming message on a
3223 * connection, and save the result in con->in_msg. Uses the
3224 * connection's private alloc_msg op if available.
3225 *
4740a623
SW		 3226 * Returns 0 on success, or a negative error code.
3227 *
3228 * On success, if we set *skip = 1:
3229 * - the next message should be skipped and ignored.
3230 * - con->in_msg == NULL
3231 * or if we set *skip = 0:
3232 * - con->in_msg is non-null.
3233 * On error (ENOMEM, EAGAIN, ...),
3234 * - con->in_msg == NULL
2450418c 3235 */
4740a623 3236static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3237{
4740a623 3238 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3239 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3240 struct ceph_msg *msg;
4740a623 3241 int ret = 0;
2450418c 3242
1c20f2d2 3243 BUG_ON(con->in_msg != NULL);
53ded495 3244 BUG_ON(!con->ops->alloc_msg);
2450418c 3245
53ded495
AE		 3246 mutex_unlock(&con->mutex);
3247 msg = con->ops->alloc_msg(con, hdr, skip);
3248 mutex_lock(&con->mutex);
3249 if (con->state != CON_STATE_OPEN) {
3250 if (msg)
1d866d1c 3251 ceph_msg_put(msg);
53ded495
AE		 3252 return -EAGAIN;
3253 }
4137577a
AE		 3254 if (msg) {
3255 BUG_ON(*skip);
3256 con->in_msg = msg;
36eb71aa 3257 con->in_msg->con = con->ops->get(con);
92ce034b 3258 BUG_ON(con->in_msg->con == NULL);
4137577a
AE		 3259 } else {
3260 /*
3261 * Null message pointer means either we should skip
3262 * this message or we couldn't allocate memory. The
3263 * former is not an error.
3264 */
3265 if (*skip)
3266 return 0;
3267 con->error_msg = "error allocating memory for incoming message";
3268
53ded495 3269 return -ENOMEM;
2450418c 3270 }
1c20f2d2 3271 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3272
1c20f2d2
AE		 3273 if (middle_len && !con->in_msg->middle) {
3274 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3275 if (ret < 0) {
1c20f2d2
AE		 3276 ceph_msg_put(con->in_msg);
3277 con->in_msg = NULL;
2450418c
YS		 3278 }
3279 }
9d7f0f13 3280
4740a623 3281 return ret;
2450418c
YS		 3282}
3283
31b8006e
SW		 3284
3285/*
3286 * Free a generically kmalloc'd message.
3287 */
0215e44b 3288static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3289{
0215e44b 3290 dout("%s %p\n", __func__, m);
4965fc38 3291 kvfree(m->front.iov_base);
e3d5d638 3292 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3293}
3294
0215e44b 3295static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3296{
3297 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
5240d9f9
AE		 3298 LIST_HEAD(data);
3299 struct list_head *links;
3300 struct list_head *next;
31b8006e 3301
0215e44b 3302 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3303 WARN_ON(!list_empty(&m->list_head));
3304
3305 /* drop middle, data, if any */
3306 if (m->middle) {
3307 ceph_buffer_put(m->middle);
3308 m->middle = NULL;
31b8006e 3309 }
5240d9f9
AE		 3310
3311 list_splice_init(&m->data, &data);
3312 list_for_each_safe(links, next, &data) {
3313 struct ceph_msg_data *data;
3314
3315 data = list_entry(links, struct ceph_msg_data, links);
3316 list_del_init(links);
3317 ceph_msg_data_destroy(data);
3318 }
a1930804 3319 m->data_length = 0;
58bb3b37 3320
c2e552e7
SW		 3321 if (m->pool)
3322 ceph_msgpool_put(m->pool, m);
3323 else
0215e44b
ID		 3324 ceph_msg_free(m);
3325}
3326
3327struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3328{
3329 dout("%s %p (was %d)\n", __func__, msg,
3330 atomic_read(&msg->kref.refcount));
3331 kref_get(&msg->kref);
3332 return msg;
3333}
3334EXPORT_SYMBOL(ceph_msg_get);
3335
3336void ceph_msg_put(struct ceph_msg *msg)
3337{
3338 dout("%s %p (was %d)\n", __func__, msg,
3339 atomic_read(&msg->kref.refcount));
3340 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3341}
0215e44b 3342EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3343
3344void ceph_msg_dump(struct ceph_msg *msg)
3345{
3cea4c30
ID		 3346 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3347 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3348 print_hex_dump(KERN_DEBUG, "header: ",
3349 DUMP_PREFIX_OFFSET, 16, 1,
3350 &msg->hdr, sizeof(msg->hdr), true);
3351 print_hex_dump(KERN_DEBUG, " front: ",
3352 DUMP_PREFIX_OFFSET, 16, 1,
3353 msg->front.iov_base, msg->front.iov_len, true);
3354 if (msg->middle)
3355 print_hex_dump(KERN_DEBUG, "middle: ",
3356 DUMP_PREFIX_OFFSET, 16, 1,
3357 msg->middle->vec.iov_base,
3358 msg->middle->vec.iov_len, true);
3359 print_hex_dump(KERN_DEBUG, "footer: ",
3360 DUMP_PREFIX_OFFSET, 16, 1,
3361 &msg->footer, sizeof(msg->footer), true);
3362}
3d14c5d2 3363EXPORT_SYMBOL(ceph_msg_dump);
Linux kernel - Deliverables
This page took 0.415826 seconds and 5 git commands to generate.