]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/ceph/messenger.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: drop len argument of *verify_authorizer_reply()
[mirror_ubuntu-bionic-kernel.git] / net / ceph / messenger.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
757856d2 9#include <linux/nsproxy.h>
5a0e3ad6 10#include <linux/slab.h>
31b8006e
SW		 11#include <linux/socket.h>
12#include <linux/string.h>
3ebc21f7 13#ifdef CONFIG_BLOCK
68b4476b 14#include <linux/bio.h>
3ebc21f7 15#endif /* CONFIG_BLOCK */
ee3b56f2 16#include <linux/dns_resolver.h>
31b8006e
SW		 17#include <net/tcp.h>
18
2b3e0c90 19#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 20#include <linux/ceph/libceph.h>
21#include <linux/ceph/messenger.h>
22#include <linux/ceph/decode.h>
23#include <linux/ceph/pagelist.h>
bc3b2d7f 24#include <linux/export.h>
31b8006e
SW		 25
26/*
27 * Ceph uses the messenger to exchange ceph_msg messages with other
28 * hosts in the system. The messenger provides ordered and reliable
29 * delivery. We tolerate TCP disconnects by reconnecting (with
30 * exponential backoff) in the case of a fault (disconnection, bad
31 * crc, protocol error). Acks allow sent messages to be discarded by
32 * the sender.
33 */
34
bc18f4b1
AE		 35/*
36 * We track the state of the socket on a given connection using
37 * values defined below. The transition to a new socket state is
38 * handled by a function which verifies we aren't coming from an
39 * unexpected state.
40 *
41 * --------
42 * | NEW* | transient initial state
43 * --------
44 * | con_sock_state_init()
45 * v
46 * ----------
47 * | CLOSED | initialized, but no socket (and no
48 * ---------- TCP connection)
49 * ^ \
50 * | \ con_sock_state_connecting()
51 * | ----------------------
52 * | \
53 * + con_sock_state_closed() \
fbb85a47
SW		 54 * |+--------------------------- \
55 * | \ \ \
56 * | ----------- \ \
57 * | | CLOSING | socket event; \ \
58 * | ----------- await close \ \
59 * | ^ \ |
60 * | | \ |
61 * | + con_sock_state_closing() \ |
62 * | / \ | |
63 * | / --------------- | |
64 * | / \ v v
bc18f4b1
AE		 65 * | / --------------
66 * | / -----------------| CONNECTING | socket created, TCP
67 * | | / -------------- connect initiated
68 * | | | con_sock_state_connected()
69 * | | v
70 * -------------
71 * | CONNECTED | TCP connection established
72 * -------------
73 *
74 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
75 */
ce2c8903
AE		 76
77#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
78#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
79#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
80#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
81#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
82
8dacc7da
SW		 83/*
84 * connection states
85 */
86#define CON_STATE_CLOSED 1 /* -> PREOPEN */
87#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
88#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
89#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
90#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
91#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
92
4a861692
SW		 93/*
94 * ceph_connection flag bits
95 */
96#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
97 * messages on errors */
98#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
99#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
100#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
101#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 102
c9ffc77a
AE		 103static bool con_flag_valid(unsigned long con_flag)
104{
105 switch (con_flag) {
106 case CON_FLAG_LOSSYTX:
107 case CON_FLAG_KEEPALIVE_PENDING:
108 case CON_FLAG_WRITE_PENDING:
109 case CON_FLAG_SOCK_CLOSED:
110 case CON_FLAG_BACKOFF:
111 return true;
112 default:
113 return false;
114 }
115}
116
117static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
118{
119 BUG_ON(!con_flag_valid(con_flag));
120
121 clear_bit(con_flag, &con->flags);
122}
123
124static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
125{
126 BUG_ON(!con_flag_valid(con_flag));
127
128 set_bit(con_flag, &con->flags);
129}
130
131static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
132{
133 BUG_ON(!con_flag_valid(con_flag));
134
135 return test_bit(con_flag, &con->flags);
136}
137
138static bool con_flag_test_and_clear(struct ceph_connection *con,
139 unsigned long con_flag)
140{
141 BUG_ON(!con_flag_valid(con_flag));
142
143 return test_and_clear_bit(con_flag, &con->flags);
144}
145
146static bool con_flag_test_and_set(struct ceph_connection *con,
147 unsigned long con_flag)
148{
149 BUG_ON(!con_flag_valid(con_flag));
150
151 return test_and_set_bit(con_flag, &con->flags);
152}
153
e3d5d638
AE		 154/* Slab caches for frequently-allocated structures */
155
156static struct kmem_cache *ceph_msg_cache;
81b36be4 157static struct kmem_cache *ceph_msg_data_cache;
e3d5d638 158
31b8006e
SW		 159/* static tag bytes (protocol control messages) */
160static char tag_msg = CEPH_MSGR_TAG_MSG;
161static char tag_ack = CEPH_MSGR_TAG_ACK;
162static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
8b9558aa 163static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
31b8006e 164
a6a5349d
SW		 165#ifdef CONFIG_LOCKDEP
166static struct lock_class_key socket_class;
167#endif
168
84495f49
AE		 169/*
170 * When skipping (ignoring) a block of input we read it into a "skip
171 * buffer," which is this many bytes in size.
172 */
173#define SKIP_BUF_SIZE 1024
31b8006e
SW		 174
175static void queue_con(struct ceph_connection *con);
37ab77ac 176static void cancel_con(struct ceph_connection *con);
68931622 177static void ceph_con_workfn(struct work_struct *);
93209264 178static void con_fault(struct ceph_connection *con);
31b8006e 179
31b8006e 180/*
f64a9317
AE		 181 * Nicely render a sockaddr as a string. An array of formatted
182 * strings is used, to approximate reentrancy.
31b8006e 183 */
f64a9317
AE		 184#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
185#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
186#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
187#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
188
189static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
190static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 191
57666519 192static struct page *zero_page; /* used in certain error cases */
57666519 193
3d14c5d2 194const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 195{
196 int i;
197 char *s;
99f0f3b2
AE		 198 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
199 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 200
f64a9317 201 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 202 s = addr_str[i];
203
204 switch (ss->ss_family) {
205 case AF_INET:
bd406145
AE		 206 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
207 ntohs(in4->sin_port));
31b8006e
SW		 208 break;
209
210 case AF_INET6:
bd406145
AE		 211 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
212 ntohs(in6->sin6_port));
31b8006e
SW		 213 break;
214
215 default:
d3002b97
AE		 216 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
217 ss->ss_family);
31b8006e
SW		 218 }
219
220 return s;
221}
3d14c5d2 222EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 223
63f2d211
SW		 224static void encode_my_addr(struct ceph_messenger *msgr)
225{
226 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
227 ceph_encode_addr(&msgr->my_enc_addr);
228}
229
31b8006e
SW		 230/*
231 * work queue for all reading and writing to/from the socket.
232 */
e0f43c94 233static struct workqueue_struct *ceph_msgr_wq;
31b8006e 234
e3d5d638
AE		 235static int ceph_msgr_slab_init(void)
236{
237 BUG_ON(ceph_msg_cache);
5ee61e95 238 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
81b36be4
AE		 239 if (!ceph_msg_cache)
240 return -ENOMEM;
241
242 BUG_ON(ceph_msg_data_cache);
5ee61e95 243 ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
81b36be4
AE		 244 if (ceph_msg_data_cache)
245 return 0;
246
247 kmem_cache_destroy(ceph_msg_cache);
248 ceph_msg_cache = NULL;
249
250 return -ENOMEM;
e3d5d638
AE		 251}
252
253static void ceph_msgr_slab_exit(void)
254{
81b36be4
AE		 255 BUG_ON(!ceph_msg_data_cache);
256 kmem_cache_destroy(ceph_msg_data_cache);
257 ceph_msg_data_cache = NULL;
258
e3d5d638
AE		 259 BUG_ON(!ceph_msg_cache);
260 kmem_cache_destroy(ceph_msg_cache);
261 ceph_msg_cache = NULL;
262}
263
15417167 264static void _ceph_msgr_exit(void)
6173d1f0 265{
d3002b97 266 if (ceph_msgr_wq) {
6173d1f0 267 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 268 ceph_msgr_wq = NULL;
269 }
6173d1f0 270
6173d1f0 271 BUG_ON(zero_page == NULL);
09cbfeaf 272 put_page(zero_page);
6173d1f0 273 zero_page = NULL;
d920ff6f
BC		 274
275 ceph_msgr_slab_exit();
6173d1f0
AE		 276}
277
3d14c5d2 278int ceph_msgr_init(void)
31b8006e 279{
d920ff6f
BC		 280 if (ceph_msgr_slab_init())
281 return -ENOMEM;
282
57666519
AE		 283 BUG_ON(zero_page != NULL);
284 zero_page = ZERO_PAGE(0);
09cbfeaf 285 get_page(zero_page);
57666519 286
f9865f06
ID		 287 /*
288 * The number of active work items is limited by the number of
289 * connections, so leave @max_active at default.
290 */
291 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
6173d1f0
AE		 292 if (ceph_msgr_wq)
293 return 0;
57666519 294
6173d1f0
AE		 295 pr_err("msgr_init failed to create workqueue\n");
296 _ceph_msgr_exit();
57666519 297
6173d1f0 298 return -ENOMEM;
31b8006e 299}
3d14c5d2 300EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 301
302void ceph_msgr_exit(void)
303{
57666519 304 BUG_ON(ceph_msgr_wq == NULL);
57666519 305
6173d1f0 306 _ceph_msgr_exit();
31b8006e 307}
3d14c5d2 308EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 309
cd84db6e 310void ceph_msgr_flush(void)
a922d38f
SW		 311{
312 flush_workqueue(ceph_msgr_wq);
313}
3d14c5d2 314EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 315
ce2c8903
AE		 316/* Connection socket state transition functions */
317
318static void con_sock_state_init(struct ceph_connection *con)
319{
320 int old_state;
321
322 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
323 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
324 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 325 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
326 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 327}
328
329static void con_sock_state_connecting(struct ceph_connection *con)
330{
331 int old_state;
332
333 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
334 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
335 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 336 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
337 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 338}
339
340static void con_sock_state_connected(struct ceph_connection *con)
341{
342 int old_state;
343
344 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
345 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
346 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 347 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
348 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 349}
350
351static void con_sock_state_closing(struct ceph_connection *con)
352{
353 int old_state;
354
355 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
356 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
357 old_state != CON_SOCK_STATE_CONNECTED &&
358 old_state != CON_SOCK_STATE_CLOSING))
359 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 360 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
361 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 362}
363
364static void con_sock_state_closed(struct ceph_connection *con)
365{
366 int old_state;
367
368 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
369 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 370 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 371 old_state != CON_SOCK_STATE_CONNECTING &&
372 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 373 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 374 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
375 CON_SOCK_STATE_CLOSED);
ce2c8903 376}
a922d38f 377
31b8006e
SW		 378/*
379 * socket callback functions
380 */
381
382/* data available on socket, or listen socket received a connect */
676d2369 383static void ceph_sock_data_ready(struct sock *sk)
31b8006e 384{
bd406145 385 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 386 if (atomic_read(&con->msgr->stopping)) {
387 return;
388 }
bd406145 389
31b8006e 390 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 391 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 392 con, con->state);
393 queue_con(con);
394 }
395}
396
397/* socket has buffer space for writing */
327800bd 398static void ceph_sock_write_space(struct sock *sk)
31b8006e 399{
d3002b97 400 struct ceph_connection *con = sk->sk_user_data;
31b8006e 401
182fac26
JS		 402 /* only queue to workqueue if there is data we want to write,
403 * and there is sufficient space in the socket buffer to accept
327800bd 404 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 405 * doesn't get called again until try_write() fills the socket
406 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
407 * and net/core/stream.c:sk_stream_write_space().
408 */
c9ffc77a 409 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
64dc6130 410 if (sk_stream_is_writeable(sk)) {
327800bd 411 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 412 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
413 queue_con(con);
414 }
31b8006e 415 } else {
327800bd 416 dout("%s %p nothing to write\n", __func__, con);
31b8006e 417 }
31b8006e
SW		 418}
419
420/* socket's state has changed */
327800bd 421static void ceph_sock_state_change(struct sock *sk)
31b8006e 422{
bd406145 423 struct ceph_connection *con = sk->sk_user_data;
31b8006e 424
327800bd 425 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 426 con, con->state, sk->sk_state);
427
31b8006e
SW		 428 switch (sk->sk_state) {
429 case TCP_CLOSE:
327800bd 430 dout("%s TCP_CLOSE\n", __func__);
31b8006e 431 case TCP_CLOSE_WAIT:
327800bd 432 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 433 con_sock_state_closing(con);
c9ffc77a 434 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 435 queue_con(con);
31b8006e
SW		 436 break;
437 case TCP_ESTABLISHED:
327800bd 438 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 439 con_sock_state_connected(con);
31b8006e
SW		 440 queue_con(con);
441 break;
d3002b97
AE		 442 default: /* Everything else is uninteresting */
443 break;
31b8006e
SW		 444 }
445}
446
447/*
448 * set up socket callbacks
449 */
450static void set_sock_callbacks(struct socket *sock,
451 struct ceph_connection *con)
452{
453 struct sock *sk = sock->sk;
bd406145 454 sk->sk_user_data = con;
327800bd
AE		 455 sk->sk_data_ready = ceph_sock_data_ready;
456 sk->sk_write_space = ceph_sock_write_space;
457 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 458}
459
460
461/*
462 * socket helpers
463 */
464
465/*
466 * initiate connection to a remote socket.
467 */
41617d0c 468static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 469{
f91d3471 470 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e
SW		 471 struct socket *sock;
472 int ret;
473
474 BUG_ON(con->sock);
757856d2 475 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
eeb1bd5c 476 SOCK_STREAM, IPPROTO_TCP, &sock);
31b8006e 477 if (ret)
41617d0c 478 return ret;
6d7fdb0a 479 sock->sk->sk_allocation = GFP_NOFS;
31b8006e 480
a6a5349d
SW		 481#ifdef CONFIG_LOCKDEP
482 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
483#endif
484
31b8006e
SW		 485 set_sock_callbacks(sock, con);
486
3d14c5d2 487 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 488
89a86be0 489 con_sock_state_connecting(con);
f91d3471
SW		 490 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
491 O_NONBLOCK);
31b8006e
SW		 492 if (ret == -EINPROGRESS) {
493 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 494 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 495 sock->sk->sk_state);
a5bc3129 496 } else if (ret < 0) {
31b8006e 497 pr_err("connect %s error %d\n",
3d14c5d2 498 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 499 sock_release(sock);
41617d0c 500 return ret;
a5bc3129 501 }
89baaa57 502
859bff51 503 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
ba988f87
CH		 504 int optval = 1;
505
506 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
507 (char *)&optval, sizeof(optval));
508 if (ret)
509 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
510 ret);
511 }
512
a5bc3129 513 con->sock = sock;
41617d0c 514 return 0;
31b8006e
SW		 515}
516
517static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
518{
519 struct kvec iov = {buf, len};
520 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 521 int r;
31b8006e 522
98bdb0aa
SW		 523 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
524 if (r == -EAGAIN)
525 r = 0;
526 return r;
31b8006e
SW		 527}
528
afb3d90e
AE		 529static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
530 int page_offset, size_t length)
531{
532 void *kaddr;
533 int ret;
534
535 BUG_ON(page_offset + length > PAGE_SIZE);
536
537 kaddr = kmap(page);
538 BUG_ON(!kaddr);
539 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
540 kunmap(page);
541
542 return ret;
543}
544
31b8006e
SW		 545/*
546 * write something. @more is true if caller will be sending more data
547 * shortly.
548 */
549static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
550 size_t kvlen, size_t len, int more)
551{
552 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 553 int r;
31b8006e
SW		 554
555 if (more)
556 msg.msg_flags |= MSG_MORE;
557 else
558 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
559
42961d23
SW		 560 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
561 if (r == -EAGAIN)
562 r = 0;
563 return r;
31b8006e
SW		 564}
565
178eda29 566static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 567 int offset, size_t size, bool more)
31739139
AE		 568{
569 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
570 int ret;
571
572 ret = kernel_sendpage(sock, page, offset, size, flags);
573 if (ret == -EAGAIN)
574 ret = 0;
575
576 return ret;
577}
578
178eda29
CC		 579static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
580 int offset, size_t size, bool more)
581{
582 int ret;
583 struct kvec iov;
584
585 /* sendpage cannot properly handle pages with page_count == 0,
586 * we need to fallback to sendmsg if that's the case */
587 if (page_count(page) >= 1)
588 return __ceph_tcp_sendpage(sock, page, offset, size, more);
589
590 iov.iov_base = kmap(page) + offset;
591 iov.iov_len = size;
592 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
593 kunmap(page);
594
595 return ret;
596}
31b8006e
SW		 597
598/*
599 * Shutdown/close the socket for the given connection.
600 */
601static int con_close_socket(struct ceph_connection *con)
602{
8007b8d6 603 int rc = 0;
31b8006e
SW		 604
605 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 606 if (con->sock) {
607 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
608 sock_release(con->sock);
609 con->sock = NULL;
610 }
456ea468
AE		 611
612 /*
4a861692 613 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 614 * independent of the connection mutex, and we could have
615 * received a socket close event before we had the chance to
616 * shut the socket down.
617 */
c9ffc77a 618 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 619
ce2c8903 620 con_sock_state_closed(con);
31b8006e
SW		 621 return rc;
622}
623
624/*
625 * Reset a connection. Discard all incoming and outgoing messages
626 * and clear *_seq state.
627 */
628static void ceph_msg_remove(struct ceph_msg *msg)
629{
630 list_del_init(&msg->list_head);
38941f80 631
31b8006e
SW		 632 ceph_msg_put(msg);
633}
634static void ceph_msg_remove_list(struct list_head *head)
635{
636 while (!list_empty(head)) {
637 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
638 list_head);
639 ceph_msg_remove(msg);
640 }
641}
642
643static void reset_connection(struct ceph_connection *con)
644{
645 /* reset connection, out_queue, msg_ and connect_seq */
646 /* discard existing out_queue and msg_seq */
0fa6ebc6 647 dout("reset_connection %p\n", con);
31b8006e
SW		 648 ceph_msg_remove_list(&con->out_queue);
649 ceph_msg_remove_list(&con->out_sent);
650
cf3e5c40 651 if (con->in_msg) {
38941f80 652 BUG_ON(con->in_msg->con != con);
cf3e5c40
SW		 653 ceph_msg_put(con->in_msg);
654 con->in_msg = NULL;
655 }
656
31b8006e
SW		 657 con->connect_seq = 0;
658 con->out_seq = 0;
c86a2930 659 if (con->out_msg) {
583d0fef 660 BUG_ON(con->out_msg->con != con);
c86a2930
SW		 661 ceph_msg_put(con->out_msg);
662 con->out_msg = NULL;
663 }
31b8006e 664 con->in_seq = 0;
0e0d5e0c 665 con->in_seq_acked = 0;
67645d76
ID		 666
667 con->out_skip = 0;
31b8006e
SW		 668}
669
670/*
671 * mark a peer down. drop any open connections.
672 */
673void ceph_con_close(struct ceph_connection *con)
674{
8c50c817 675 mutex_lock(&con->mutex);
3d14c5d2
YS		 676 dout("con_close %p peer %s\n", con,
677 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 678 con->state = CON_STATE_CLOSED;
a5988c49 679
c9ffc77a
AE		 680 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
681 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
682 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
683 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 684
31b8006e 685 reset_connection(con);
6f2bc3ff 686 con->peer_global_seq = 0;
37ab77ac 687 cancel_con(con);
ee76e073 688 con_close_socket(con);
ec302645 689 mutex_unlock(&con->mutex);
31b8006e 690}
3d14c5d2 691EXPORT_SYMBOL(ceph_con_close);
31b8006e 692
31b8006e
SW		 693/*
694 * Reopen a closed connection, with a new peer address.
695 */
b7a9e5dd
SW		 696void ceph_con_open(struct ceph_connection *con,
697 __u8 entity_type, __u64 entity_num,
698 struct ceph_entity_addr *addr)
31b8006e 699{
5469155f 700 mutex_lock(&con->mutex);
3d14c5d2 701 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 702
122070a2 703 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 704 con->state = CON_STATE_PREOPEN;
a5988c49 705
b7a9e5dd
SW		 706 con->peer_name.type = (__u8) entity_type;
707 con->peer_name.num = cpu_to_le64(entity_num);
708
31b8006e 709 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 710 con->delay = 0; /* reset backoff memory */
5469155f 711 mutex_unlock(&con->mutex);
31b8006e
SW		 712 queue_con(con);
713}
3d14c5d2 714EXPORT_SYMBOL(ceph_con_open);
31b8006e 715
87b315a5
SW		 716/*
717 * return true if this connection ever successfully opened
718 */
719bool ceph_con_opened(struct ceph_connection *con)
720{
721 return con->connect_seq > 0;
722}
723
31b8006e
SW		 724/*
725 * initialize a new connection.
726 */
1bfd89f4
AE		 727void ceph_con_init(struct ceph_connection *con, void *private,
728 const struct ceph_connection_operations *ops,
b7a9e5dd 729 struct ceph_messenger *msgr)
31b8006e
SW		 730{
731 dout("con_init %p\n", con);
732 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 733 con->private = private;
734 con->ops = ops;
31b8006e 735 con->msgr = msgr;
ce2c8903
AE		 736
737 con_sock_state_init(con);
738
ec302645 739 mutex_init(&con->mutex);
31b8006e
SW		 740 INIT_LIST_HEAD(&con->out_queue);
741 INIT_LIST_HEAD(&con->out_sent);
68931622 742 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
a5988c49 743
8dacc7da 744 con->state = CON_STATE_CLOSED;
31b8006e 745}
3d14c5d2 746EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 747
748
749/*
750 * We maintain a global counter to order connection attempts. Get
751 * a unique seq greater than @gt.
752 */
753static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
754{
755 u32 ret;
756
757 spin_lock(&msgr->global_seq_lock);
758 if (msgr->global_seq < gt)
759 msgr->global_seq = gt;
760 ret = ++msgr->global_seq;
761 spin_unlock(&msgr->global_seq_lock);
762 return ret;
763}
764
e2200423 765static void con_out_kvec_reset(struct ceph_connection *con)
859eb799 766{
67645d76
ID		 767 BUG_ON(con->out_skip);
768
859eb799
AE		 769 con->out_kvec_left = 0;
770 con->out_kvec_bytes = 0;
771 con->out_kvec_cur = &con->out_kvec[0];
772}
773
e2200423 774static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 775 size_t size, void *data)
776{
67645d76 777 int index = con->out_kvec_left;
859eb799 778
67645d76 779 BUG_ON(con->out_skip);
859eb799
AE		 780 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
781
782 con->out_kvec[index].iov_len = size;
783 con->out_kvec[index].iov_base = data;
784 con->out_kvec_left++;
785 con->out_kvec_bytes += size;
786}
31b8006e 787
67645d76
ID		 788/*
789 * Chop off a kvec from the end. Return residual number of bytes for
790 * that kvec, i.e. how many bytes would have been written if the kvec
791 * hadn't been nuked.
792 */
793static int con_out_kvec_skip(struct ceph_connection *con)
794{
795 int off = con->out_kvec_cur - con->out_kvec;
796 int skip = 0;
797
798 if (con->out_kvec_bytes > 0) {
799 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
800 BUG_ON(con->out_kvec_bytes < skip);
801 BUG_ON(!con->out_kvec_left);
802 con->out_kvec_bytes -= skip;
803 con->out_kvec_left--;
804 }
805
806 return skip;
807}
808
df6ad1f9 809#ifdef CONFIG_BLOCK
6aaa4511
AE		 810
811/*
812 * For a bio data item, a piece is whatever remains of the next
813 * entry in the current bio iovec, or the first entry in the next
814 * bio in the list.
815 */
8ae4f4f5 816static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 817 size_t length)
6aaa4511 818{
8ae4f4f5 819 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 820 struct bio *bio;
821
822 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
823
824 bio = data->bio;
825 BUG_ON(!bio);
6aaa4511 826
ca8b3a69 827 cursor->resid = min(length, data->bio_length);
6aaa4511 828 cursor->bio = bio;
f38a5181
KO		 829 cursor->bvec_iter = bio->bi_iter;
830 cursor->last_piece =
831 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
6aaa4511
AE		 832}
833
8ae4f4f5 834static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 835 size_t *page_offset,
836 size_t *length)
837{
8ae4f4f5 838 struct ceph_msg_data *data = cursor->data;
6aaa4511 839 struct bio *bio;
f38a5181 840 struct bio_vec bio_vec;
6aaa4511
AE		 841
842 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
843
844 bio = cursor->bio;
845 BUG_ON(!bio);
846
f38a5181 847 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511 848
f38a5181 849 *page_offset = (size_t) bio_vec.bv_offset;
6aaa4511 850 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 851 if (cursor->last_piece) /* pagelist offset is always 0 */
852 *length = cursor->resid;
853 else
f38a5181 854 *length = (size_t) bio_vec.bv_len;
25aff7c5 855 BUG_ON(*length > cursor->resid);
5df521b1 856 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511 857
f38a5181 858 return bio_vec.bv_page;
6aaa4511
AE		 859}
860
8ae4f4f5
AE		 861static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
862 size_t bytes)
6aaa4511 863{
6aaa4511 864 struct bio *bio;
f38a5181 865 struct bio_vec bio_vec;
6aaa4511 866
8ae4f4f5 867 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
6aaa4511
AE		 868
869 bio = cursor->bio;
870 BUG_ON(!bio);
871
f38a5181 872 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511
AE		 873
874 /* Advance the cursor offset */
875
25aff7c5
AE		 876 BUG_ON(cursor->resid < bytes);
877 cursor->resid -= bytes;
f38a5181
KO		 878
879 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
880
881 if (bytes < bio_vec.bv_len)
6aaa4511
AE		 882 return false; /* more bytes to process in this segment */
883
884 /* Move on to the next segment, and possibly the next bio */
885
f38a5181 886 if (!cursor->bvec_iter.bi_size) {
6aaa4511 887 bio = bio->bi_next;
0ec1d15e
ID		 888 cursor->bio = bio;
889 if (bio)
890 cursor->bvec_iter = bio->bi_iter;
891 else
892 memset(&cursor->bvec_iter, 0,
893 sizeof(cursor->bvec_iter));
6aaa4511 894 }
6aaa4511 895
25aff7c5
AE		 896 if (!cursor->last_piece) {
897 BUG_ON(!cursor->resid);
898 BUG_ON(!bio);
899 /* A short read is OK, so use <= rather than == */
f38a5181 900 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
6aaa4511 901 cursor->last_piece = true;
25aff7c5 902 }
6aaa4511
AE		 903
904 return true;
905}
ea96571f 906#endif /* CONFIG_BLOCK */
df6ad1f9 907
e766d7b5
AE		 908/*
909 * For a page array, a piece comes from the first page in the array
910 * that has not already been fully consumed.
911 */
8ae4f4f5 912static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 913 size_t length)
e766d7b5 914{
8ae4f4f5 915 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 916 int page_count;
917
918 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
919
920 BUG_ON(!data->pages);
921 BUG_ON(!data->length);
922
ca8b3a69 923 cursor->resid = min(length, data->length);
e766d7b5 924 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 925 cursor->page_offset = data->alignment & ~PAGE_MASK;
926 cursor->page_index = 0;
56fc5659
AE		 927 BUG_ON(page_count > (int)USHRT_MAX);
928 cursor->page_count = (unsigned short)page_count;
929 BUG_ON(length > SIZE_MAX - cursor->page_offset);
5f740d7e 930 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 931}
932
8ae4f4f5
AE		 933static struct page *
934ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
935 size_t *page_offset, size_t *length)
e766d7b5 936{
8ae4f4f5 937 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 938
939 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
940
941 BUG_ON(cursor->page_index >= cursor->page_count);
942 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 943
944 *page_offset = cursor->page_offset;
25aff7c5 945 if (cursor->last_piece)
e766d7b5 946 *length = cursor->resid;
25aff7c5 947 else
e766d7b5 948 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 949
950 return data->pages[cursor->page_index];
951}
952
8ae4f4f5 953static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 954 size_t bytes)
955{
8ae4f4f5 956 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 957
958 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 959
960 /* Advance the cursor page offset */
961
962 cursor->resid -= bytes;
5df521b1
AE		 963 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
964 if (!bytes || cursor->page_offset)
e766d7b5
AE		 965 return false; /* more bytes to process in the current page */
966
d90deda6
YZ		 967 if (!cursor->resid)
968 return false; /* no more data */
969
5df521b1 970 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 971
972 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 973 cursor->page_index++;
25aff7c5 974 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 975
976 return true;
977}
978
fe38a2b6 979/*
dd236fcb
AE		 980 * For a pagelist, a piece is whatever remains to be consumed in the
981 * first page in the list, or the front of the next page.
fe38a2b6 982 */
8ae4f4f5
AE		 983static void
984ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 985 size_t length)
fe38a2b6 986{
8ae4f4f5 987 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 988 struct ceph_pagelist *pagelist;
989 struct page *page;
990
dd236fcb 991 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 992
993 pagelist = data->pagelist;
994 BUG_ON(!pagelist);
25aff7c5
AE		 995
996 if (!length)
fe38a2b6
AE		 997 return; /* pagelist can be assigned but empty */
998
999 BUG_ON(list_empty(&pagelist->head));
1000 page = list_first_entry(&pagelist->head, struct page, lru);
1001
ca8b3a69 1002 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 1003 cursor->page = page;
1004 cursor->offset = 0;
a51b272e 1005 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1006}
1007
8ae4f4f5
AE		 1008static struct page *
1009ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1010 size_t *page_offset, size_t *length)
fe38a2b6 1011{
8ae4f4f5 1012 struct ceph_msg_data *data = cursor->data;
fe38a2b6 1013 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 1014
1015 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1016
1017 pagelist = data->pagelist;
1018 BUG_ON(!pagelist);
1019
1020 BUG_ON(!cursor->page);
25aff7c5 1021 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 1022
5df521b1 1023 /* offset of first page in pagelist is always 0 */
fe38a2b6 1024 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 1025 if (cursor->last_piece)
25aff7c5
AE		 1026 *length = cursor->resid;
1027 else
1028 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1029
8ae4f4f5 1030 return cursor->page;
fe38a2b6
AE		 1031}
1032
8ae4f4f5 1033static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1034 size_t bytes)
fe38a2b6 1035{
8ae4f4f5 1036 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1037 struct ceph_pagelist *pagelist;
1038
1039 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1040
1041 pagelist = data->pagelist;
1042 BUG_ON(!pagelist);
25aff7c5
AE		 1043
1044 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1045 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1046
1047 /* Advance the cursor offset */
1048
25aff7c5 1049 cursor->resid -= bytes;
fe38a2b6 1050 cursor->offset += bytes;
5df521b1 1051 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1052 if (!bytes || cursor->offset & ~PAGE_MASK)
1053 return false; /* more bytes to process in the current page */
1054
d90deda6
YZ		 1055 if (!cursor->resid)
1056 return false; /* no more data */
1057
fe38a2b6
AE		 1058 /* Move on to the next page */
1059
1060 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
17ddc49b 1061 cursor->page = list_next_entry(cursor->page, lru);
25aff7c5 1062 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1063
1064 return true;
1065}
1066
dd236fcb
AE		 1067/*
1068 * Message data is handled (sent or received) in pieces, where each
1069 * piece resides on a single page. The network layer might not
1070 * consume an entire piece at once. A data item's cursor keeps
1071 * track of which piece is next to process and how much remains to
1072 * be processed in that piece. It also tracks whether the current
1073 * piece is the last one in the data item.
1074 */
ca8b3a69 1075static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1076{
ca8b3a69 1077 size_t length = cursor->total_resid;
8ae4f4f5 1078
8ae4f4f5 1079 switch (cursor->data->type) {
dd236fcb 1080 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1081 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1082 break;
e766d7b5 1083 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1084 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1085 break;
dd236fcb
AE		 1086#ifdef CONFIG_BLOCK
1087 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1088 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1089 break;
dd236fcb 1090#endif /* CONFIG_BLOCK */
6aaa4511 1091 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1092 default:
1093 /* BUG(); */
1094 break;
1095 }
8ae4f4f5 1096 cursor->need_crc = true;
dd236fcb
AE		 1097}
1098
ca8b3a69
AE		 1099static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1100{
1101 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1102 struct ceph_msg_data *data;
1103
1104 BUG_ON(!length);
1105 BUG_ON(length > msg->data_length);
1106 BUG_ON(list_empty(&msg->data));
1107
ca8b3a69
AE		 1108 cursor->data_head = &msg->data;
1109 cursor->total_resid = length;
1110 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1111 cursor->data = data;
1112
1113 __ceph_msg_data_cursor_init(cursor);
1114}
1115
dd236fcb
AE		 1116/*
1117 * Return the page containing the next piece to process for a given
1118 * data item, and supply the page offset and length of that piece.
1119 * Indicate whether this is the last piece in this data item.
1120 */
8ae4f4f5
AE		 1121static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1122 size_t *page_offset, size_t *length,
dd236fcb
AE		 1123 bool *last_piece)
1124{
1125 struct page *page;
1126
8ae4f4f5 1127 switch (cursor->data->type) {
dd236fcb 1128 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1129 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1130 break;
e766d7b5 1131 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1132 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1133 break;
dd236fcb
AE		 1134#ifdef CONFIG_BLOCK
1135 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1136 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1137 break;
dd236fcb 1138#endif /* CONFIG_BLOCK */
6aaa4511 1139 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1140 default:
1141 page = NULL;
1142 break;
1143 }
1144 BUG_ON(!page);
1145 BUG_ON(*page_offset + *length > PAGE_SIZE);
1146 BUG_ON(!*length);
1147 if (last_piece)
8ae4f4f5 1148 *last_piece = cursor->last_piece;
dd236fcb
AE		 1149
1150 return page;
1151}
1152
1153/*
1154 * Returns true if the result moves the cursor on to the next piece
1155 * of the data item.
1156 */
8ae4f4f5
AE		 1157static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1158 size_t bytes)
dd236fcb
AE		 1159{
1160 bool new_piece;
1161
25aff7c5 1162 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1163 switch (cursor->data->type) {
dd236fcb 1164 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1165 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1166 break;
e766d7b5 1167 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1168 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1169 break;
dd236fcb
AE		 1170#ifdef CONFIG_BLOCK
1171 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1172 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1173 break;
dd236fcb 1174#endif /* CONFIG_BLOCK */
6aaa4511 1175 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1176 default:
1177 BUG();
1178 break;
1179 }
ca8b3a69 1180 cursor->total_resid -= bytes;
dd236fcb 1181
ca8b3a69
AE		 1182 if (!cursor->resid && cursor->total_resid) {
1183 WARN_ON(!cursor->last_piece);
1184 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
17ddc49b 1185 cursor->data = list_next_entry(cursor->data, links);
ca8b3a69 1186 __ceph_msg_data_cursor_init(cursor);
a51b272e 1187 new_piece = true;
ca8b3a69 1188 }
a51b272e 1189 cursor->need_crc = new_piece;
ca8b3a69 1190
dd236fcb
AE		 1191 return new_piece;
1192}
1193
dbc0d3ca
ID		 1194static size_t sizeof_footer(struct ceph_connection *con)
1195{
1196 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1197 sizeof(struct ceph_msg_footer) :
1198 sizeof(struct ceph_msg_footer_old);
1199}
1200
98fa5dd8 1201static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1202{
739c905b 1203 BUG_ON(!msg);
25aff7c5 1204 BUG_ON(!data_len);
739c905b 1205
4c59b4a2 1206 /* Initialize data cursor */
fe38a2b6 1207
8ae4f4f5 1208 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1209}
1210
31b8006e
SW		 1211/*
1212 * Prepare footer for currently outgoing message, and finish things
1213 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1214 */
859eb799 1215static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1216{
1217 struct ceph_msg *m = con->out_msg;
1218
fd154f3c
AE		 1219 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1220
31b8006e 1221 dout("prepare_write_message_footer %p\n", con);
89f08173 1222 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
33d07337
YZ		 1223 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1224 if (con->ops->sign_message)
79dbd1ba 1225 con->ops->sign_message(m);
33d07337
YZ		 1226 else
1227 m->footer.sig = 0;
33d07337
YZ		 1228 } else {
1229 m->old_footer.flags = m->footer.flags;
33d07337 1230 }
31b8006e 1231 con->out_more = m->more_to_follow;
c86a2930 1232 con->out_msg_done = true;
31b8006e
SW		 1233}
1234
1235/*
1236 * Prepare headers for the next outgoing message.
1237 */
1238static void prepare_write_message(struct ceph_connection *con)
1239{
1240 struct ceph_msg *m;
a9a0c51a 1241 u32 crc;
31b8006e 1242
e2200423 1243 con_out_kvec_reset(con);
c86a2930 1244 con->out_msg_done = false;
31b8006e
SW		 1245
1246 /* Sneak an ack in there first? If we can get it into the same
1247 * TCP packet that's a good thing. */
1248 if (con->in_seq > con->in_seq_acked) {
1249 con->in_seq_acked = con->in_seq;
e2200423 1250 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1251 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1252 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1253 &con->out_temp_ack);
31b8006e
SW		 1254 }
1255
38941f80 1256 BUG_ON(list_empty(&con->out_queue));
859eb799 1257 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1258 con->out_msg = m;
38941f80 1259 BUG_ON(m->con != con);
4cf9d544
SW		 1260
1261 /* put message on sent list */
1262 ceph_msg_get(m);
1263 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1264
e84346b7
SW		 1265 /*
1266 * only assign outgoing seq # if we haven't sent this message
1267 * yet. if it is requeued, resend with it's original seq.
1268 */
1269 if (m->needs_out_seq) {
1270 m->hdr.seq = cpu_to_le64(++con->out_seq);
1271 m->needs_out_seq = false;
1272 }
98fa5dd8 1273 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1274
98fa5dd8 1275 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1276 m, con->out_seq, le16_to_cpu(m->hdr.type),
1277 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1278 m->data_length);
31b8006e
SW		 1279 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1280
1281 /* tag + hdr + front + middle */
e2200423 1282 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
67645d76 1283 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
e2200423 1284 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1285
31b8006e 1286 if (m->middle)
e2200423 1287 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1288 m->middle->vec.iov_base);
31b8006e 1289
67645d76 1290 /* fill in hdr crc and finalize hdr */
a9a0c51a
AE		 1291 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1292 con->out_msg->hdr.crc = cpu_to_le32(crc);
67645d76 1293 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
a9a0c51a 1294
67645d76 1295 /* fill in front and middle crc, footer */
a9a0c51a
AE		 1296 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1297 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1298 if (m->middle) {
1299 crc = crc32c(0, m->middle->vec.iov_base,
1300 m->middle->vec.iov_len);
1301 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1302 } else
31b8006e 1303 con->out_msg->footer.middle_crc = 0;
739c905b 1304 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1305 le32_to_cpu(con->out_msg->footer.front_crc),
1306 le32_to_cpu(con->out_msg->footer.middle_crc));
67645d76 1307 con->out_msg->footer.flags = 0;
31b8006e
SW		 1308
1309 /* is there a data payload? */
739c905b 1310 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1311 if (m->data_length) {
1312 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1313 con->out_more = 1; /* data + footer will follow */
1314 } else {
31b8006e 1315 /* no, queue up footer too and be done */
859eb799 1316 prepare_write_message_footer(con);
78625051 1317 }
31b8006e 1318
c9ffc77a 1319 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1320}
1321
1322/*
1323 * Prepare an ack.
1324 */
1325static void prepare_write_ack(struct ceph_connection *con)
1326{
1327 dout("prepare_write_ack %p %llu -> %llu\n", con,
1328 con->in_seq_acked, con->in_seq);
1329 con->in_seq_acked = con->in_seq;
1330
e2200423 1331 con_out_kvec_reset(con);
859eb799 1332
e2200423 1333 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1334
31b8006e 1335 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1336 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1337 &con->out_temp_ack);
1338
31b8006e 1339 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1340 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1341}
1342
3a23083b
SW		 1343/*
1344 * Prepare to share the seq during handshake
1345 */
1346static void prepare_write_seq(struct ceph_connection *con)
1347{
1348 dout("prepare_write_seq %p %llu -> %llu\n", con,
1349 con->in_seq_acked, con->in_seq);
1350 con->in_seq_acked = con->in_seq;
1351
1352 con_out_kvec_reset(con);
1353
1354 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1355 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1356 &con->out_temp_ack);
1357
1358 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1359}
1360
31b8006e
SW		 1361/*
1362 * Prepare to write keepalive byte.
1363 */
1364static void prepare_write_keepalive(struct ceph_connection *con)
1365{
1366 dout("prepare_write_keepalive %p\n", con);
e2200423 1367 con_out_kvec_reset(con);
8b9558aa 1368 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
7f61f545
ID		 1369 struct timespec now = CURRENT_TIME;
1370
8b9558aa 1371 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
7f61f545
ID		 1372 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1373 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1374 &con->out_temp_keepalive2);
8b9558aa
YZ		 1375 } else {
1376 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1377 }
c9ffc77a 1378 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1379}
1380
1381/*
1382 * Connection negotiation.
1383 */
1384
dac1e716
AE		 1385static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1386 int *auth_proto)
4e7a5dcd 1387{
a3530df3 1388 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1389
1390 if (!con->ops->get_authorizer) {
1391 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1392 con->out_connect.authorizer_len = 0;
729796be 1393 return NULL;
b1c6b980
AE		 1394 }
1395
1396 /* Can't hold the mutex while getting authorizer */
ec302645 1397 mutex_unlock(&con->mutex);
dac1e716 1398 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1399 mutex_lock(&con->mutex);
4e7a5dcd 1400
a3530df3 1401 if (IS_ERR(auth))
729796be 1402 return auth;
8dacc7da 1403 if (con->state != CON_STATE_NEGOTIATING)
729796be 1404 return ERR_PTR(-EAGAIN);
0da5d703 1405
8f43fb53
AE		 1406 con->auth_reply_buf = auth->authorizer_reply_buf;
1407 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1408 return auth;
4e7a5dcd
SW		 1409}
1410
31b8006e
SW		 1411/*
1412 * We connected to a peer and are saying hello.
1413 */
e825a66d 1414static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1415{
e2200423
AE		 1416 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1417 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1418 &con->msgr->my_enc_addr);
eed0ef2c 1419
eed0ef2c 1420 con->out_more = 0;
c9ffc77a 1421 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1422}
1423
e825a66d 1424static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1425{
95c96174 1426 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1427 int proto;
dac1e716 1428 int auth_proto;
729796be 1429 struct ceph_auth_handshake *auth;
31b8006e
SW		 1430
1431 switch (con->peer_name.type) {
1432 case CEPH_ENTITY_TYPE_MON:
1433 proto = CEPH_MONC_PROTOCOL;
1434 break;
1435 case CEPH_ENTITY_TYPE_OSD:
1436 proto = CEPH_OSDC_PROTOCOL;
1437 break;
1438 case CEPH_ENTITY_TYPE_MDS:
1439 proto = CEPH_MDSC_PROTOCOL;
1440 break;
1441 default:
1442 BUG();
1443 }
1444
1445 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1446 con->connect_seq, global_seq, proto);
4e7a5dcd 1447
859bff51
ID		 1448 con->out_connect.features =
1449 cpu_to_le64(from_msgr(con->msgr)->supported_features);
31b8006e
SW		 1450 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1451 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1452 con->out_connect.global_seq = cpu_to_le32(global_seq);
1453 con->out_connect.protocol_version = cpu_to_le32(proto);
1454 con->out_connect.flags = 0;
31b8006e 1455
dac1e716
AE		 1456 auth_proto = CEPH_AUTH_UNKNOWN;
1457 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1458 if (IS_ERR(auth))
1459 return PTR_ERR(auth);
3da54776 1460
dac1e716 1461 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1462 con->out_connect.authorizer_len = auth ?
1463 cpu_to_le32(auth->authorizer_buf_len) : 0;
1464
e2200423 1465 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1466 &con->out_connect);
1467 if (auth && auth->authorizer_buf_len)
e2200423 1468 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1469 auth->authorizer_buf);
859eb799 1470
31b8006e 1471 con->out_more = 0;
c9ffc77a 1472 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1473
e10c758e 1474 return 0;
31b8006e
SW		 1475}
1476
31b8006e
SW		 1477/*
1478 * write as much of pending kvecs to the socket as we can.
1479 * 1 -> done
1480 * 0 -> socket full, but more to do
1481 * <0 -> error
1482 */
1483static int write_partial_kvec(struct ceph_connection *con)
1484{
1485 int ret;
1486
1487 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1488 while (con->out_kvec_bytes > 0) {
1489 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1490 con->out_kvec_left, con->out_kvec_bytes,
1491 con->out_more);
1492 if (ret <= 0)
1493 goto out;
1494 con->out_kvec_bytes -= ret;
1495 if (con->out_kvec_bytes == 0)
1496 break; /* done */
f42299e6
AE		 1497
1498 /* account for full iov entries consumed */
1499 while (ret >= con->out_kvec_cur->iov_len) {
1500 BUG_ON(!con->out_kvec_left);
1501 ret -= con->out_kvec_cur->iov_len;
1502 con->out_kvec_cur++;
1503 con->out_kvec_left--;
1504 }
1505 /* and for a partially-consumed entry */
1506 if (ret) {
1507 con->out_kvec_cur->iov_len -= ret;
1508 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1509 }
1510 }
1511 con->out_kvec_left = 0;
31b8006e
SW		 1512 ret = 1;
1513out:
1514 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1515 con->out_kvec_bytes, con->out_kvec_left, ret);
1516 return ret; /* done! */
1517}
1518
35b62808
AE		 1519static u32 ceph_crc32c_page(u32 crc, struct page *page,
1520 unsigned int page_offset,
1521 unsigned int length)
1522{
1523 char *kaddr;
1524
1525 kaddr = kmap(page);
1526 BUG_ON(kaddr == NULL);
1527 crc = crc32c(crc, kaddr + page_offset, length);
1528 kunmap(page);
1529
1530 return crc;
1531}
31b8006e
SW		 1532/*
1533 * Write as much message data payload as we can. If we finish, queue
1534 * up the footer.
1535 * 1 -> done, footer is now queued in out_kvec[].
1536 * 0 -> socket full, but more to do
1537 * <0 -> error
1538 */
34d2d200 1539static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1540{
1541 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1542 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 1543 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
f5db90bc 1544 u32 crc;
31b8006e 1545
859a35d5 1546 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1547
5240d9f9 1548 if (list_empty(&msg->data))
4c59b4a2
AE		 1549 return -EINVAL;
1550
5821bd8c
AE		 1551 /*
1552 * Iterate through each page that contains data to be
1553 * written, and send as much as possible for each.
1554 *
1555 * If we are calculating the data crc (the default), we will
1556 * need to map the page. If we have no pages, they have
1557 * been revoked, so use the zero page.
1558 */
f5db90bc 1559 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1560 while (cursor->resid) {
8a166d05 1561 struct page *page;
e387d525
AE		 1562 size_t page_offset;
1563 size_t length;
8a166d05 1564 bool last_piece;
8ea299bc 1565 bool need_crc;
f5db90bc 1566 int ret;
68b4476b 1567
343128ce
SB		 1568 page = ceph_msg_data_next(cursor, &page_offset, &length,
1569 &last_piece);
e387d525 1570 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
c2cfa194 1571 length, !last_piece);
f5db90bc
AE		 1572 if (ret <= 0) {
1573 if (do_datacrc)
1574 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1575
f5db90bc
AE		 1576 return ret;
1577 }
143334ff
AE		 1578 if (do_datacrc && cursor->need_crc)
1579 crc = ceph_crc32c_page(crc, page, page_offset, length);
343128ce 1580 need_crc = ceph_msg_data_advance(cursor, (size_t)ret);
31b8006e
SW		 1581 }
1582
34d2d200 1583 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1584
1585 /* prepare and queue up footer, too */
f5db90bc
AE		 1586 if (do_datacrc)
1587 msg->footer.data_crc = cpu_to_le32(crc);
1588 else
84ca8fc8 1589 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1590 con_out_kvec_reset(con);
859eb799 1591 prepare_write_message_footer(con);
f5db90bc
AE		 1592
1593 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1594}
1595
1596/*
1597 * write some zeros
1598 */
1599static int write_partial_skip(struct ceph_connection *con)
1600{
1601 int ret;
1602
67645d76 1603 dout("%s %p %d left\n", __func__, con, con->out_skip);
31b8006e 1604 while (con->out_skip > 0) {
09cbfeaf 1605 size_t size = min(con->out_skip, (int) PAGE_SIZE);
31b8006e 1606
e1dcb128 1607 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1608 if (ret <= 0)
1609 goto out;
1610 con->out_skip -= ret;
1611 }
1612 ret = 1;
1613out:
1614 return ret;
1615}
1616
1617/*
1618 * Prepare to read connection handshake, or an ack.
1619 */
eed0ef2c
SW		 1620static void prepare_read_banner(struct ceph_connection *con)
1621{
1622 dout("prepare_read_banner %p\n", con);
1623 con->in_base_pos = 0;
1624}
1625
31b8006e
SW		 1626static void prepare_read_connect(struct ceph_connection *con)
1627{
1628 dout("prepare_read_connect %p\n", con);
1629 con->in_base_pos = 0;
1630}
1631
1632static void prepare_read_ack(struct ceph_connection *con)
1633{
1634 dout("prepare_read_ack %p\n", con);
1635 con->in_base_pos = 0;
1636}
1637
3a23083b
SW		 1638static void prepare_read_seq(struct ceph_connection *con)
1639{
1640 dout("prepare_read_seq %p\n", con);
1641 con->in_base_pos = 0;
1642 con->in_tag = CEPH_MSGR_TAG_SEQ;
1643}
1644
31b8006e
SW		 1645static void prepare_read_tag(struct ceph_connection *con)
1646{
1647 dout("prepare_read_tag %p\n", con);
1648 con->in_base_pos = 0;
1649 con->in_tag = CEPH_MSGR_TAG_READY;
1650}
1651
8b9558aa
YZ		 1652static void prepare_read_keepalive_ack(struct ceph_connection *con)
1653{
1654 dout("prepare_read_keepalive_ack %p\n", con);
1655 con->in_base_pos = 0;
1656}
1657
31b8006e
SW		 1658/*
1659 * Prepare to read a message.
1660 */
1661static int prepare_read_message(struct ceph_connection *con)
1662{
1663 dout("prepare_read_message %p\n", con);
1664 BUG_ON(con->in_msg != NULL);
1665 con->in_base_pos = 0;
1666 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1667 return 0;
1668}
1669
1670
1671static int read_partial(struct ceph_connection *con,
fd51653f 1672 int end, int size, void *object)
31b8006e 1673{
e6cee71f
AE		 1674 while (con->in_base_pos < end) {
1675 int left = end - con->in_base_pos;
31b8006e
SW		 1676 int have = size - left;
1677 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1678 if (ret <= 0)
1679 return ret;
1680 con->in_base_pos += ret;
1681 }
1682 return 1;
1683}
1684
1685
1686/*
1687 * Read all or part of the connect-side handshake on a new connection
1688 */
eed0ef2c 1689static int read_partial_banner(struct ceph_connection *con)
31b8006e 1690{
fd51653f
AE		 1691 int size;
1692 int end;
1693 int ret;
31b8006e 1694
eed0ef2c 1695 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1696
1697 /* peer's banner */
fd51653f
AE		 1698 size = strlen(CEPH_BANNER);
1699 end = size;
1700 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1701 if (ret <= 0)
1702 goto out;
fd51653f
AE		 1703
1704 size = sizeof (con->actual_peer_addr);
1705 end += size;
1706 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1707 if (ret <= 0)
1708 goto out;
fd51653f
AE		 1709
1710 size = sizeof (con->peer_addr_for_me);
1711 end += size;
1712 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1713 if (ret <= 0)
1714 goto out;
fd51653f 1715
eed0ef2c
SW		 1716out:
1717 return ret;
1718}
1719
1720static int read_partial_connect(struct ceph_connection *con)
1721{
fd51653f
AE		 1722 int size;
1723 int end;
1724 int ret;
eed0ef2c
SW		 1725
1726 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1727
fd51653f
AE		 1728 size = sizeof (con->in_reply);
1729 end = size;
1730 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1731 if (ret <= 0)
1732 goto out;
fd51653f
AE		 1733
1734 size = le32_to_cpu(con->in_reply.authorizer_len);
1735 end += size;
1736 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1737 if (ret <= 0)
1738 goto out;
31b8006e 1739
4e7a5dcd
SW		 1740 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1741 con, (int)con->in_reply.tag,
1742 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1743 le32_to_cpu(con->in_reply.global_seq));
1744out:
1745 return ret;
eed0ef2c 1746
31b8006e
SW		 1747}
1748
1749/*
1750 * Verify the hello banner looks okay.
1751 */
1752static int verify_hello(struct ceph_connection *con)
1753{
1754 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1755 pr_err("connect to %s got bad banner\n",
3d14c5d2 1756 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1757 con->error_msg = "protocol error, bad banner";
1758 return -1;
1759 }
1760 return 0;
1761}
1762
1763static bool addr_is_blank(struct sockaddr_storage *ss)
1764{
c44bd69c
ID		 1765 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1766 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1767
31b8006e
SW		 1768 switch (ss->ss_family) {
1769 case AF_INET:
c44bd69c 1770 return addr->s_addr == htonl(INADDR_ANY);
31b8006e 1771 case AF_INET6:
c44bd69c
ID		 1772 return ipv6_addr_any(addr6);
1773 default:
1774 return true;
31b8006e 1775 }
31b8006e
SW		 1776}
1777
1778static int addr_port(struct sockaddr_storage *ss)
1779{
1780 switch (ss->ss_family) {
1781 case AF_INET:
f28bcfbe 1782 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1783 case AF_INET6:
f28bcfbe 1784 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1785 }
1786 return 0;
1787}
1788
1789static void addr_set_port(struct sockaddr_storage *ss, int p)
1790{
1791 switch (ss->ss_family) {
1792 case AF_INET:
1793 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1794 break;
31b8006e
SW		 1795 case AF_INET6:
1796 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1797 break;
31b8006e
SW		 1798 }
1799}
1800
ee3b56f2
NW		 1801/*
1802 * Unlike other *_pton function semantics, zero indicates success.
1803 */
1804static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1805 char delim, const char **ipend)
1806{
99f0f3b2
AE		 1807 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1808 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1809
1810 memset(ss, 0, sizeof(*ss));
1811
1812 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1813 ss->ss_family = AF_INET;
1814 return 0;
1815 }
1816
1817 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1818 ss->ss_family = AF_INET6;
1819 return 0;
1820 }
1821
1822 return -EINVAL;
1823}
1824
1825/*
1826 * Extract hostname string and resolve using kernel DNS facility.
1827 */
1828#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1829static int ceph_dns_resolve_name(const char *name, size_t namelen,
1830 struct sockaddr_storage *ss, char delim, const char **ipend)
1831{
1832 const char *end, *delim_p;
1833 char *colon_p, *ip_addr = NULL;
1834 int ip_len, ret;
1835
1836 /*
1837 * The end of the hostname occurs immediately preceding the delimiter or
1838 * the port marker (':') where the delimiter takes precedence.
1839 */
1840 delim_p = memchr(name, delim, namelen);
1841 colon_p = memchr(name, ':', namelen);
1842
1843 if (delim_p && colon_p)
1844 end = delim_p < colon_p ? delim_p : colon_p;
1845 else if (!delim_p && colon_p)
1846 end = colon_p;
1847 else {
1848 end = delim_p;
1849 if (!end) /* case: hostname:/ */
1850 end = name + namelen;
1851 }
1852
1853 if (end <= name)
1854 return -EINVAL;
1855
1856 /* do dns_resolve upcall */
1857 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1858 if (ip_len > 0)
1859 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1860 else
1861 ret = -ESRCH;
1862
1863 kfree(ip_addr);
1864
1865 *ipend = end;
1866
1867 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1868 ret, ret ? "failed" : ceph_pr_addr(ss));
1869
1870 return ret;
1871}
1872#else
1873static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1874 struct sockaddr_storage *ss, char delim, const char **ipend)
1875{
1876 return -EINVAL;
1877}
1878#endif
1879
1880/*
1881 * Parse a server name (IP or hostname). If a valid IP address is not found
1882 * then try to extract a hostname to resolve using userspace DNS upcall.
1883 */
1884static int ceph_parse_server_name(const char *name, size_t namelen,
1885 struct sockaddr_storage *ss, char delim, const char **ipend)
1886{
1887 int ret;
1888
1889 ret = ceph_pton(name, namelen, ss, delim, ipend);
1890 if (ret)
1891 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1892
1893 return ret;
1894}
1895
31b8006e
SW		 1896/*
1897 * Parse an ip[:port] list into an addr array. Use the default
1898 * monitor port if a port isn't specified.
1899 */
1900int ceph_parse_ips(const char *c, const char *end,
1901 struct ceph_entity_addr *addr,
1902 int max_count, int *count)
1903{
ee3b56f2 1904 int i, ret = -EINVAL;
31b8006e
SW		 1905 const char *p = c;
1906
1907 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1908 for (i = 0; i < max_count; i++) {
1909 const char *ipend;
1910 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1911 int port;
39139f64
SW		 1912 char delim = ',';
1913
1914 if (*p == '[') {
1915 delim = ']';
1916 p++;
1917 }
31b8006e 1918
ee3b56f2
NW		 1919 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1920 if (ret)
31b8006e 1921 goto bad;
ee3b56f2
NW		 1922 ret = -EINVAL;
1923
31b8006e
SW		 1924 p = ipend;
1925
39139f64
SW		 1926 if (delim == ']') {
1927 if (*p != ']') {
1928 dout("missing matching ']'\n");
1929 goto bad;
1930 }
1931 p++;
1932 }
1933
31b8006e
SW		 1934 /* port? */
1935 if (p < end && *p == ':') {
1936 port = 0;
1937 p++;
1938 while (p < end && *p >= '0' && *p <= '9') {
1939 port = (port * 10) + (*p - '0');
1940 p++;
1941 }
f48db1e9
ID		 1942 if (port == 0)
1943 port = CEPH_MON_PORT;
1944 else if (port > 65535)
31b8006e
SW		 1945 goto bad;
1946 } else {
1947 port = CEPH_MON_PORT;
1948 }
1949
1950 addr_set_port(ss, port);
1951
3d14c5d2 1952 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1953
1954 if (p == end)
1955 break;
1956 if (*p != ',')
1957 goto bad;
1958 p++;
1959 }
1960
1961 if (p != end)
1962 goto bad;
1963
1964 if (count)
1965 *count = i + 1;
1966 return 0;
1967
1968bad:
39139f64 1969 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1970 return ret;
31b8006e 1971}
3d14c5d2 1972EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1973
eed0ef2c 1974static int process_banner(struct ceph_connection *con)
31b8006e 1975{
eed0ef2c 1976 dout("process_banner on %p\n", con);
31b8006e
SW		 1977
1978 if (verify_hello(con) < 0)
1979 return -1;
1980
63f2d211
SW		 1981 ceph_decode_addr(&con->actual_peer_addr);
1982 ceph_decode_addr(&con->peer_addr_for_me);
1983
31b8006e
SW		 1984 /*
1985 * Make sure the other end is who we wanted. note that the other
1986 * end may not yet know their ip address, so if it's 0.0.0.0, give
1987 * them the benefit of the doubt.
1988 */
103e2d3a
SW		 1989 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1990 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1991 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1992 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 1993 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1994 ceph_pr_addr(&con->peer_addr.in_addr),
1995 (int)le32_to_cpu(con->peer_addr.nonce),
1996 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1997 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 1998 con->error_msg = "wrong peer at address";
31b8006e
SW		 1999 return -1;
2000 }
2001
2002 /*
2003 * did we learn our address?
2004 */
2005 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2006 int port = addr_port(&con->msgr->inst.addr.in_addr);
2007
2008 memcpy(&con->msgr->inst.addr.in_addr,
2009 &con->peer_addr_for_me.in_addr,
2010 sizeof(con->peer_addr_for_me.in_addr));
2011 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 2012 encode_my_addr(con->msgr);
eed0ef2c 2013 dout("process_banner learned my addr is %s\n",
3d14c5d2 2014 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 2015 }
2016
eed0ef2c
SW		 2017 return 0;
2018}
2019
2020static int process_connect(struct ceph_connection *con)
2021{
859bff51
ID		 2022 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2023 u64 req_feat = from_msgr(con->msgr)->required_features;
2b3e0c90
ID		 2024 u64 server_feat = ceph_sanitize_features(
2025 le64_to_cpu(con->in_reply.features));
0da5d703 2026 int ret;
04a419f9 2027
eed0ef2c
SW		 2028 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2029
5c056fdc
ID		 2030 if (con->auth_reply_buf) {
2031 /*
2032 * Any connection that defines ->get_authorizer()
2033 * should also define ->verify_authorizer_reply().
2034 * See get_connect_authorizer().
2035 */
0dde5848 2036 ret = con->ops->verify_authorizer_reply(con);
5c056fdc
ID		 2037 if (ret < 0) {
2038 con->error_msg = "bad authorize reply";
2039 return ret;
2040 }
2041 }
2042
31b8006e 2043 switch (con->in_reply.tag) {
04a419f9
SW		 2044 case CEPH_MSGR_TAG_FEATURES:
2045 pr_err("%s%lld %s feature set mismatch,"
2046 " my %llx < server's %llx, missing %llx\n",
2047 ENTITY_NAME(con->peer_name),
3d14c5d2 2048 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2049 sup_feat, server_feat, server_feat & ~sup_feat);
2050 con->error_msg = "missing required protocol features";
0fa6ebc6 2051 reset_connection(con);
04a419f9
SW		 2052 return -1;
2053
31b8006e 2054 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2055 pr_err("%s%lld %s protocol version mismatch,"
2056 " my %d != server's %d\n",
2057 ENTITY_NAME(con->peer_name),
3d14c5d2 2058 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2059 le32_to_cpu(con->out_connect.protocol_version),
2060 le32_to_cpu(con->in_reply.protocol_version));
2061 con->error_msg = "protocol version mismatch";
0fa6ebc6 2062 reset_connection(con);
31b8006e
SW		 2063 return -1;
2064
4e7a5dcd
SW		 2065 case CEPH_MSGR_TAG_BADAUTHORIZER:
2066 con->auth_retry++;
2067 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2068 con->auth_retry);
2069 if (con->auth_retry == 2) {
2070 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2071 return -1;
2072 }
6d4221b5 2073 con_out_kvec_reset(con);
e825a66d 2074 ret = prepare_write_connect(con);
0da5d703
SW		 2075 if (ret < 0)
2076 return ret;
63733a0f 2077 prepare_read_connect(con);
4e7a5dcd 2078 break;
31b8006e
SW		 2079
2080 case CEPH_MSGR_TAG_RESETSESSION:
2081 /*
2082 * If we connected with a large connect_seq but the peer
2083 * has no record of a session with us (no connection, or
2084 * connect_seq == 0), they will send RESETSESION to indicate
2085 * that they must have reset their session, and may have
2086 * dropped messages.
2087 */
2088 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2089 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2090 pr_err("%s%lld %s connection reset\n",
2091 ENTITY_NAME(con->peer_name),
3d14c5d2 2092 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2093 reset_connection(con);
6d4221b5 2094 con_out_kvec_reset(con);
5a0f8fdd
AE		 2095 ret = prepare_write_connect(con);
2096 if (ret < 0)
2097 return ret;
31b8006e
SW		 2098 prepare_read_connect(con);
2099
2100 /* Tell ceph about it. */
ec302645 2101 mutex_unlock(&con->mutex);
31b8006e
SW		 2102 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2103 if (con->ops->peer_reset)
2104 con->ops->peer_reset(con);
ec302645 2105 mutex_lock(&con->mutex);
8dacc7da 2106 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2107 return -EAGAIN;
31b8006e
SW		 2108 break;
2109
2110 case CEPH_MSGR_TAG_RETRY_SESSION:
2111 /*
2112 * If we sent a smaller connect_seq than the peer has, try
2113 * again with a larger value.
2114 */
5bdca4e0 2115 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2116 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2117 le32_to_cpu(con->in_reply.connect_seq));
2118 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2119 con_out_kvec_reset(con);
5a0f8fdd
AE		 2120 ret = prepare_write_connect(con);
2121 if (ret < 0)
2122 return ret;
31b8006e
SW		 2123 prepare_read_connect(con);
2124 break;
2125
2126 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2127 /*
2128 * If we sent a smaller global_seq than the peer has, try
2129 * again with a larger value.
2130 */
eed0ef2c 2131 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2132 con->peer_global_seq,
5bdca4e0 2133 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2134 get_global_seq(con->msgr,
5bdca4e0 2135 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2136 con_out_kvec_reset(con);
5a0f8fdd
AE		 2137 ret = prepare_write_connect(con);
2138 if (ret < 0)
2139 return ret;
31b8006e
SW		 2140 prepare_read_connect(con);
2141 break;
2142
3a23083b 2143 case CEPH_MSGR_TAG_SEQ:
31b8006e 2144 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2145 if (req_feat & ~server_feat) {
2146 pr_err("%s%lld %s protocol feature mismatch,"
2147 " my required %llx > server's %llx, need %llx\n",
2148 ENTITY_NAME(con->peer_name),
3d14c5d2 2149 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2150 req_feat, server_feat, req_feat & ~server_feat);
2151 con->error_msg = "missing required protocol features";
0fa6ebc6 2152 reset_connection(con);
04a419f9
SW		 2153 return -1;
2154 }
8dacc7da 2155
122070a2 2156 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2157 con->state = CON_STATE_OPEN;
20e55c4c 2158 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2159 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2160 con->connect_seq++;
aba558e2 2161 con->peer_features = server_feat;
31b8006e
SW		 2162 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2163 con->peer_global_seq,
2164 le32_to_cpu(con->in_reply.connect_seq),
2165 con->connect_seq);
2166 WARN_ON(con->connect_seq !=
2167 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2168
2169 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2170 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2171
85effe18 2172 con->delay = 0; /* reset backoff memory */
92ac41d0 2173
3a23083b
SW		 2174 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2175 prepare_write_seq(con);
2176 prepare_read_seq(con);
2177 } else {
2178 prepare_read_tag(con);
2179 }
31b8006e
SW		 2180 break;
2181
2182 case CEPH_MSGR_TAG_WAIT:
2183 /*
2184 * If there is a connection race (we are opening
2185 * connections to each other), one of us may just have
2186 * to WAIT. This shouldn't happen if we are the
2187 * client.
2188 */
04177882
SW		 2189 con->error_msg = "protocol error, got WAIT as client";
2190 return -1;
31b8006e
SW		 2191
2192 default:
31b8006e
SW		 2193 con->error_msg = "protocol error, garbage tag during connect";
2194 return -1;
2195 }
2196 return 0;
2197}
2198
2199
2200/*
2201 * read (part of) an ack
2202 */
2203static int read_partial_ack(struct ceph_connection *con)
2204{
fd51653f
AE		 2205 int size = sizeof (con->in_temp_ack);
2206 int end = size;
31b8006e 2207
fd51653f 2208 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2209}
2210
31b8006e
SW		 2211/*
2212 * We can finally discard anything that's been acked.
2213 */
2214static void process_ack(struct ceph_connection *con)
2215{
2216 struct ceph_msg *m;
2217 u64 ack = le64_to_cpu(con->in_temp_ack);
2218 u64 seq;
2219
31b8006e
SW		 2220 while (!list_empty(&con->out_sent)) {
2221 m = list_first_entry(&con->out_sent, struct ceph_msg,
2222 list_head);
2223 seq = le64_to_cpu(m->hdr.seq);
2224 if (seq > ack)
2225 break;
2226 dout("got ack for seq %llu type %d at %p\n", seq,
2227 le16_to_cpu(m->hdr.type), m);
4cf9d544 2228 m->ack_stamp = jiffies;
31b8006e
SW		 2229 ceph_msg_remove(m);
2230 }
31b8006e
SW		 2231 prepare_read_tag(con);
2232}
2233
2234
2450418c 2235static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2236 struct kvec *section,
2237 unsigned int sec_len, u32 *crc)
2450418c 2238{
68b4476b 2239 int ret, left;
2450418c
YS		 2240
2241 BUG_ON(!section);
2242
2243 while (section->iov_len < sec_len) {
2244 BUG_ON(section->iov_base == NULL);
2245 left = sec_len - section->iov_len;
2246 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2247 section->iov_len, left);
2248 if (ret <= 0)
2249 return ret;
2250 section->iov_len += ret;
2450418c 2251 }
fe3ad593
AE		 2252 if (section->iov_len == sec_len)
2253 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2254
2450418c
YS		 2255 return 1;
2256}
31b8006e 2257
34d2d200
AE		 2258static int read_partial_msg_data(struct ceph_connection *con)
2259{
2260 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2261 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 2262 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
686be208
AE		 2263 struct page *page;
2264 size_t page_offset;
2265 size_t length;
f5db90bc 2266 u32 crc = 0;
34d2d200
AE		 2267 int ret;
2268
2269 BUG_ON(!msg);
5240d9f9 2270 if (list_empty(&msg->data))
4c59b4a2 2271 return -EIO;
34d2d200 2272
f5db90bc
AE		 2273 if (do_datacrc)
2274 crc = con->in_data_crc;
643c68a4 2275 while (cursor->resid) {
343128ce 2276 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
686be208 2277 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2278 if (ret <= 0) {
2279 if (do_datacrc)
2280 con->in_data_crc = crc;
2281
686be208 2282 return ret;
f5db90bc 2283 }
686be208
AE		 2284
2285 if (do_datacrc)
f5db90bc 2286 crc = ceph_crc32c_page(crc, page, page_offset, ret);
343128ce 2287 (void) ceph_msg_data_advance(cursor, (size_t)ret);
34d2d200 2288 }
f5db90bc
AE		 2289 if (do_datacrc)
2290 con->in_data_crc = crc;
34d2d200
AE		 2291
2292 return 1; /* must return > 0 to indicate success */
2293}
2294
31b8006e
SW		 2295/*
2296 * read (part of) a message.
2297 */
686be208
AE		 2298static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2299
31b8006e
SW		 2300static int read_partial_message(struct ceph_connection *con)
2301{
2302 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2303 int size;
2304 int end;
31b8006e 2305 int ret;
95c96174 2306 unsigned int front_len, middle_len, data_len;
859bff51 2307 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
33d07337 2308 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
ae18756b 2309 u64 seq;
fe3ad593 2310 u32 crc;
31b8006e
SW		 2311
2312 dout("read_partial_message con %p msg %p\n", con, m);
2313
2314 /* header */
fd51653f
AE		 2315 size = sizeof (con->in_hdr);
2316 end = size;
2317 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2318 if (ret <= 0)
2319 return ret;
fe3ad593
AE		 2320
2321 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2322 if (cpu_to_le32(crc) != con->in_hdr.crc) {
67c64eb7 2323 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
fe3ad593
AE		 2324 crc, con->in_hdr.crc);
2325 return -EBADMSG;
2326 }
2327
31b8006e
SW		 2328 front_len = le32_to_cpu(con->in_hdr.front_len);
2329 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2330 return -EIO;
2331 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2332 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2333 return -EIO;
2334 data_len = le32_to_cpu(con->in_hdr.data_len);
2335 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2336 return -EIO;
2337
ae18756b
SW		 2338 /* verify seq# */
2339 seq = le64_to_cpu(con->in_hdr.seq);
2340 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2341 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2342 ENTITY_NAME(con->peer_name),
3d14c5d2 2343 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2344 seq, con->in_seq + 1);
2345 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2346 sizeof_footer(con);
ae18756b 2347 con->in_tag = CEPH_MSGR_TAG_READY;
e7a88e82 2348 return 1;
ae18756b
SW		 2349 } else if ((s64)seq - (s64)con->in_seq > 1) {
2350 pr_err("read_partial_message bad seq %lld expected %lld\n",
2351 seq, con->in_seq + 1);
2352 con->error_msg = "bad message sequence # for incoming message";
67c64eb7 2353 return -EBADE;
ae18756b
SW		 2354 }
2355
31b8006e
SW		 2356 /* allocate message? */
2357 if (!con->in_msg) {
4740a623
SW		 2358 int skip = 0;
2359
31b8006e 2360 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2361 front_len, data_len);
4740a623
SW		 2362 ret = ceph_con_in_msg_alloc(con, &skip);
2363 if (ret < 0)
2364 return ret;
f759ebb9
AE		 2365
2366 BUG_ON(!con->in_msg ^ skip);
2450418c 2367 if (skip) {
31b8006e 2368 /* skip this message */
a79832f2 2369 dout("alloc_msg said skip message\n");
31b8006e 2370 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2371 sizeof_footer(con);
31b8006e 2372 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2373 con->in_seq++;
e7a88e82 2374 return 1;
31b8006e 2375 }
38941f80 2376
4740a623 2377 BUG_ON(!con->in_msg);
38941f80 2378 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2379 m = con->in_msg;
2380 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2381 if (m->middle)
2382 m->middle->vec.iov_len = 0;
9d7f0f13 2383
78625051 2384 /* prepare for data payload, if any */
a4107026 2385
78625051 2386 if (data_len)
98fa5dd8 2387 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2388 }
2389
2390 /* front */
2450418c
YS		 2391 ret = read_partial_message_section(con, &m->front, front_len,
2392 &con->in_front_crc);
2393 if (ret <= 0)
2394 return ret;
31b8006e
SW		 2395
2396 /* middle */
2450418c 2397 if (m->middle) {
213c99ee
SW		 2398 ret = read_partial_message_section(con, &m->middle->vec,
2399 middle_len,
2450418c 2400 &con->in_middle_crc);
31b8006e
SW		 2401 if (ret <= 0)
2402 return ret;
31b8006e
SW		 2403 }
2404
2405 /* (page) data */
34d2d200
AE		 2406 if (data_len) {
2407 ret = read_partial_msg_data(con);
2408 if (ret <= 0)
2409 return ret;
31b8006e
SW		 2410 }
2411
31b8006e 2412 /* footer */
89f08173 2413 size = sizeof_footer(con);
fd51653f
AE		 2414 end += size;
2415 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2416 if (ret <= 0)
2417 return ret;
2418
33d07337
YZ		 2419 if (!need_sign) {
2420 m->footer.flags = m->old_footer.flags;
2421 m->footer.sig = 0;
2422 }
2423
31b8006e
SW		 2424 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2425 m, front_len, m->footer.front_crc, middle_len,
2426 m->footer.middle_crc, data_len, m->footer.data_crc);
2427
2428 /* crc ok? */
2429 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2430 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2431 m, con->in_front_crc, m->footer.front_crc);
2432 return -EBADMSG;
2433 }
2434 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2435 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2436 m, con->in_middle_crc, m->footer.middle_crc);
2437 return -EBADMSG;
2438 }
bca064d2 2439 if (do_datacrc &&
31b8006e
SW		 2440 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2441 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2442 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2443 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2444 return -EBADMSG;
2445 }
2446
33d07337 2447 if (need_sign && con->ops->check_message_signature &&
79dbd1ba 2448 con->ops->check_message_signature(m)) {
33d07337
YZ		 2449 pr_err("read_partial_message %p signature check failed\n", m);
2450 return -EBADMSG;
2451 }
2452
31b8006e
SW		 2453 return 1; /* done! */
2454}
2455
2456/*
2457 * Process message. This happens in the worker thread. The callback should
2458 * be careful not to do anything that waits on other incoming messages or it
2459 * may deadlock.
2460 */
2461static void process_message(struct ceph_connection *con)
2462{
583d0fef 2463 struct ceph_msg *msg = con->in_msg;
31b8006e 2464
38941f80 2465 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2466 con->in_msg = NULL;
2467
2468 /* if first message, set peer_name */
2469 if (con->peer_name.type == 0)
dbad185d 2470 con->peer_name = msg->hdr.src;
31b8006e 2471
31b8006e 2472 con->in_seq++;
ec302645 2473 mutex_unlock(&con->mutex);
31b8006e
SW		 2474
2475 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2476 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2477 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2478 le16_to_cpu(msg->hdr.type),
2479 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2480 le32_to_cpu(msg->hdr.front_len),
2481 le32_to_cpu(msg->hdr.data_len),
2482 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2483 con->ops->dispatch(con, msg);
ec302645
SW		 2484
2485 mutex_lock(&con->mutex);
31b8006e
SW		 2486}
2487
8b9558aa
YZ		 2488static int read_keepalive_ack(struct ceph_connection *con)
2489{
2490 struct ceph_timespec ceph_ts;
2491 size_t size = sizeof(ceph_ts);
2492 int ret = read_partial(con, size, size, &ceph_ts);
2493 if (ret <= 0)
2494 return ret;
2495 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2496 prepare_read_tag(con);
2497 return 1;
2498}
31b8006e
SW		 2499
2500/*
2501 * Write something to the socket. Called in a worker thread when the
2502 * socket appears to be writeable and we have something ready to send.
2503 */
2504static int try_write(struct ceph_connection *con)
2505{
31b8006e
SW		 2506 int ret = 1;
2507
d59315ca 2508 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2509
31b8006e
SW		 2510more:
2511 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2512
2513 /* open the socket first? */
8dacc7da
SW		 2514 if (con->state == CON_STATE_PREOPEN) {
2515 BUG_ON(con->sock);
2516 con->state = CON_STATE_CONNECTING;
a5988c49 2517
e2200423 2518 con_out_kvec_reset(con);
e825a66d 2519 prepare_write_banner(con);
eed0ef2c 2520 prepare_read_banner(con);
31b8006e 2521
cf3e5c40 2522 BUG_ON(con->in_msg);
31b8006e
SW		 2523 con->in_tag = CEPH_MSGR_TAG_READY;
2524 dout("try_write initiating connect on %p new state %lu\n",
2525 con, con->state);
41617d0c
AE		 2526 ret = ceph_tcp_connect(con);
2527 if (ret < 0) {
31b8006e 2528 con->error_msg = "connect error";
31b8006e
SW		 2529 goto out;
2530 }
2531 }
2532
2533more_kvec:
2534 /* kvec data queued? */
67645d76
ID		 2535 if (con->out_kvec_left) {
2536 ret = write_partial_kvec(con);
31b8006e 2537 if (ret <= 0)
42961d23 2538 goto out;
31b8006e 2539 }
67645d76
ID		 2540 if (con->out_skip) {
2541 ret = write_partial_skip(con);
31b8006e 2542 if (ret <= 0)
42961d23 2543 goto out;
31b8006e
SW		 2544 }
2545
2546 /* msg pages? */
2547 if (con->out_msg) {
c86a2930
SW		 2548 if (con->out_msg_done) {
2549 ceph_msg_put(con->out_msg);
2550 con->out_msg = NULL; /* we're done with this one */
2551 goto do_next;
2552 }
2553
34d2d200 2554 ret = write_partial_message_data(con);
31b8006e
SW		 2555 if (ret == 1)
2556 goto more_kvec; /* we need to send the footer, too! */
2557 if (ret == 0)
42961d23 2558 goto out;
31b8006e 2559 if (ret < 0) {
34d2d200 2560 dout("try_write write_partial_message_data err %d\n",
31b8006e 2561 ret);
42961d23 2562 goto out;
31b8006e
SW		 2563 }
2564 }
2565
c86a2930 2566do_next:
8dacc7da 2567 if (con->state == CON_STATE_OPEN) {
8b9558aa
YZ		 2568 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2569 prepare_write_keepalive(con);
2570 goto more;
2571 }
31b8006e
SW		 2572 /* is anything else pending? */
2573 if (!list_empty(&con->out_queue)) {
2574 prepare_write_message(con);
2575 goto more;
2576 }
2577 if (con->in_seq > con->in_seq_acked) {
2578 prepare_write_ack(con);
2579 goto more;
2580 }
31b8006e
SW		 2581 }
2582
2583 /* Nothing to do! */
c9ffc77a 2584 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2585 dout("try_write nothing else to write.\n");
31b8006e
SW		 2586 ret = 0;
2587out:
42961d23 2588 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2589 return ret;
2590}
2591
2592
2593
2594/*
2595 * Read what we can from the socket.
2596 */
2597static int try_read(struct ceph_connection *con)
2598{
31b8006e
SW		 2599 int ret = -1;
2600
8dacc7da
SW		 2601more:
2602 dout("try_read start on %p state %lu\n", con, con->state);
2603 if (con->state != CON_STATE_CONNECTING &&
2604 con->state != CON_STATE_NEGOTIATING &&
2605 con->state != CON_STATE_OPEN)
31b8006e
SW		 2606 return 0;
2607
8dacc7da 2608 BUG_ON(!con->sock);
ec302645 2609
31b8006e
SW		 2610 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2611 con->in_base_pos);
0da5d703 2612
8dacc7da 2613 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2614 dout("try_read connecting\n");
2615 ret = read_partial_banner(con);
2616 if (ret <= 0)
ab166d5a 2617 goto out;
7593af92
AE		 2618 ret = process_banner(con);
2619 if (ret < 0)
2620 goto out;
2621
8dacc7da 2622 con->state = CON_STATE_NEGOTIATING;
7593af92 2623
6d4221b5
JS		 2624 /*
2625 * Received banner is good, exchange connection info.
2626 * Do not reset out_kvec, as sending our banner raced
2627 * with receiving peer banner after connect completed.
2628 */
7593af92
AE		 2629 ret = prepare_write_connect(con);
2630 if (ret < 0)
2631 goto out;
2632 prepare_read_connect(con);
2633
2634 /* Send connection info before awaiting response */
0da5d703
SW		 2635 goto out;
2636 }
2637
8dacc7da 2638 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2639 dout("try_read negotiating\n");
31b8006e
SW		 2640 ret = read_partial_connect(con);
2641 if (ret <= 0)
31b8006e 2642 goto out;
98bdb0aa
SW		 2643 ret = process_connect(con);
2644 if (ret < 0)
2645 goto out;
31b8006e
SW		 2646 goto more;
2647 }
2648
122070a2 2649 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2650
31b8006e
SW		 2651 if (con->in_base_pos < 0) {
2652 /*
2653 * skipping + discarding content.
2654 *
2655 * FIXME: there must be a better way to do this!
2656 */
84495f49
AE		 2657 static char buf[SKIP_BUF_SIZE];
2658 int skip = min((int) sizeof (buf), -con->in_base_pos);
2659
31b8006e
SW		 2660 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2661 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2662 if (ret <= 0)
98bdb0aa 2663 goto out;
31b8006e
SW		 2664 con->in_base_pos += ret;
2665 if (con->in_base_pos)
2666 goto more;
2667 }
2668 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2669 /*
2670 * what's next?
2671 */
2672 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2673 if (ret <= 0)
98bdb0aa 2674 goto out;
31b8006e
SW		 2675 dout("try_read got tag %d\n", (int)con->in_tag);
2676 switch (con->in_tag) {
2677 case CEPH_MSGR_TAG_MSG:
2678 prepare_read_message(con);
2679 break;
2680 case CEPH_MSGR_TAG_ACK:
2681 prepare_read_ack(con);
2682 break;
8b9558aa
YZ		 2683 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2684 prepare_read_keepalive_ack(con);
2685 break;
31b8006e 2686 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2687 con_close_socket(con);
2688 con->state = CON_STATE_CLOSED;
98bdb0aa 2689 goto out;
31b8006e
SW		 2690 default:
2691 goto bad_tag;
2692 }
2693 }
2694 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2695 ret = read_partial_message(con);
2696 if (ret <= 0) {
2697 switch (ret) {
2698 case -EBADMSG:
a51983e4 2699 con->error_msg = "bad crc/signature";
67c64eb7
ID		 2700 /* fall through */
2701 case -EBADE:
31b8006e 2702 ret = -EIO;
98bdb0aa 2703 break;
31b8006e
SW		 2704 case -EIO:
2705 con->error_msg = "io error";
98bdb0aa 2706 break;
31b8006e 2707 }
98bdb0aa 2708 goto out;
31b8006e
SW		 2709 }
2710 if (con->in_tag == CEPH_MSGR_TAG_READY)
2711 goto more;
2712 process_message(con);
7b862e07
SW		 2713 if (con->state == CON_STATE_OPEN)
2714 prepare_read_tag(con);
31b8006e
SW		 2715 goto more;
2716 }
3a23083b
SW		 2717 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2718 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2719 /*
2720 * the final handshake seq exchange is semantically
2721 * equivalent to an ACK
2722 */
31b8006e
SW		 2723 ret = read_partial_ack(con);
2724 if (ret <= 0)
98bdb0aa 2725 goto out;
31b8006e
SW		 2726 process_ack(con);
2727 goto more;
2728 }
8b9558aa
YZ		 2729 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2730 ret = read_keepalive_ack(con);
2731 if (ret <= 0)
2732 goto out;
2733 goto more;
2734 }
31b8006e 2735
31b8006e 2736out:
98bdb0aa 2737 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2738 return ret;
2739
2740bad_tag:
2741 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2742 con->error_msg = "protocol error, garbage tag";
2743 ret = -1;
2744 goto out;
2745}
2746
2747
2748/*
802c6d96
AE		 2749 * Atomically queue work on a connection after the specified delay.
2750 * Bump @con reference to avoid races with connection teardown.
2751 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2752 */
802c6d96 2753static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2754{
31b8006e 2755 if (!con->ops->get(con)) {
802c6d96 2756 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2757 return -ENOENT;
31b8006e
SW		 2758 }
2759
802c6d96
AE		 2760 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2761 dout("%s %p - already queued\n", __func__, con);
31b8006e 2762 con->ops->put(con);
802c6d96 2763 return -EBUSY;
31b8006e 2764 }
802c6d96
AE		 2765
2766 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2767 return 0;
2768}
2769
2770static void queue_con(struct ceph_connection *con)
2771{
2772 (void) queue_con_delay(con, 0);
31b8006e
SW		 2773}
2774
37ab77ac
ID		 2775static void cancel_con(struct ceph_connection *con)
2776{
2777 if (cancel_delayed_work(&con->work)) {
2778 dout("%s %p\n", __func__, con);
2779 con->ops->put(con);
2780 }
2781}
2782
7bb21d68
AE		 2783static bool con_sock_closed(struct ceph_connection *con)
2784{
c9ffc77a 2785 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2786 return false;
2787
2788#define CASE(x) \
2789 case CON_STATE_ ## x: \
2790 con->error_msg = "socket closed (con state " #x ")"; \
2791 break;
2792
2793 switch (con->state) {
2794 CASE(CLOSED);
2795 CASE(PREOPEN);
2796 CASE(CONNECTING);
2797 CASE(NEGOTIATING);
2798 CASE(OPEN);
2799 CASE(STANDBY);
2800 default:
b9a67899 2801 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2802 __func__, con, con->state);
2803 con->error_msg = "unrecognized con state";
2804 BUG();
2805 break;
2806 }
2807#undef CASE
2808
2809 return true;
2810}
2811
f20a39fd
AE		 2812static bool con_backoff(struct ceph_connection *con)
2813{
2814 int ret;
2815
2816 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2817 return false;
2818
2819 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2820 if (ret) {
2821 dout("%s: con %p FAILED to back off %lu\n", __func__,
2822 con, con->delay);
2823 BUG_ON(ret == -ENOENT);
2824 con_flag_set(con, CON_FLAG_BACKOFF);
2825 }
2826
2827 return true;
2828}
2829
93209264
AE		 2830/* Finish fault handling; con->mutex must *not* be held here */
2831
2832static void con_fault_finish(struct ceph_connection *con)
2833{
f6330cc1
ID		 2834 dout("%s %p\n", __func__, con);
2835
93209264
AE		 2836 /*
2837 * in case we faulted due to authentication, invalidate our
2838 * current tickets so that we can get new ones.
2839 */
f6330cc1
ID		 2840 if (con->auth_retry) {
2841 dout("auth_retry %d, invalidating\n", con->auth_retry);
2842 if (con->ops->invalidate_authorizer)
2843 con->ops->invalidate_authorizer(con);
2844 con->auth_retry = 0;
93209264
AE		 2845 }
2846
2847 if (con->ops->fault)
2848 con->ops->fault(con);
2849}
2850
31b8006e
SW		 2851/*
2852 * Do some work on a connection. Drop a connection ref when we're done.
2853 */
68931622 2854static void ceph_con_workfn(struct work_struct *work)
31b8006e
SW		 2855{
2856 struct ceph_connection *con = container_of(work, struct ceph_connection,
2857 work.work);
49659416 2858 bool fault;
31b8006e 2859
9dd4658d 2860 mutex_lock(&con->mutex);
49659416
AE		 2861 while (true) {
2862 int ret;
31b8006e 2863
49659416
AE		 2864 if ((fault = con_sock_closed(con))) {
2865 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2866 break;
2867 }
2868 if (con_backoff(con)) {
2869 dout("%s: con %p BACKOFF\n", __func__, con);
2870 break;
2871 }
2872 if (con->state == CON_STATE_STANDBY) {
2873 dout("%s: con %p STANDBY\n", __func__, con);
2874 break;
2875 }
2876 if (con->state == CON_STATE_CLOSED) {
2877 dout("%s: con %p CLOSED\n", __func__, con);
2878 BUG_ON(con->sock);
2879 break;
2880 }
2881 if (con->state == CON_STATE_PREOPEN) {
2882 dout("%s: con %p PREOPEN\n", __func__, con);
2883 BUG_ON(con->sock);
2884 }
0da5d703 2885
49659416
AE		 2886 ret = try_read(con);
2887 if (ret < 0) {
2888 if (ret == -EAGAIN)
2889 continue;
67c64eb7
ID		 2890 if (!con->error_msg)
2891 con->error_msg = "socket error on read";
49659416
AE		 2892 fault = true;
2893 break;
2894 }
2895
2896 ret = try_write(con);
2897 if (ret < 0) {
2898 if (ret == -EAGAIN)
2899 continue;
67c64eb7
ID		 2900 if (!con->error_msg)
2901 con->error_msg = "socket error on write";
49659416
AE		 2902 fault = true;
2903 }
2904
2905 break; /* If we make it to here, we're done */
3a140a0d 2906 }
b6e7b6a1
AE		 2907 if (fault)
2908 con_fault(con);
9dd4658d 2909 mutex_unlock(&con->mutex);
0da5d703 2910
b6e7b6a1
AE		 2911 if (fault)
2912 con_fault_finish(con);
2913
2914 con->ops->put(con);
31b8006e
SW		 2915}
2916
31b8006e
SW		 2917/*
2918 * Generic error/fault handler. A retry mechanism is used with
2919 * exponential backoff
2920 */
93209264 2921static void con_fault(struct ceph_connection *con)
31b8006e 2922{
31b8006e 2923 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2924 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2925
67c64eb7
ID		 2926 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2927 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2928 con->error_msg = NULL;
2929
122070a2 2930 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2931 con->state != CON_STATE_NEGOTIATING &&
2932 con->state != CON_STATE_OPEN);
ec302645 2933
31b8006e 2934 con_close_socket(con);
5e095e8b 2935
c9ffc77a 2936 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2937 dout("fault on LOSSYTX channel, marking CLOSED\n");
2938 con->state = CON_STATE_CLOSED;
93209264 2939 return;
3b5ede07
SW		 2940 }
2941
5e095e8b 2942 if (con->in_msg) {
38941f80 2943 BUG_ON(con->in_msg->con != con);
5e095e8b
SW		 2944 ceph_msg_put(con->in_msg);
2945 con->in_msg = NULL;
2946 }
31b8006e 2947
e80a52d1
SW		 2948 /* Requeue anything that hasn't been acked */
2949 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2950
e76661d0
SW		 2951 /* If there are no messages queued or keepalive pending, place
2952 * the connection in a STANDBY state */
2953 if (list_empty(&con->out_queue) &&
c9ffc77a 2954 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2955 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2956 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2957 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2958 } else {
2959 /* retry after a delay. */
8dacc7da 2960 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2961 if (con->delay == 0)
2962 con->delay = BASE_DELAY_INTERVAL;
2963 else if (con->delay < MAX_DELAY_INTERVAL)
2964 con->delay *= 2;
c9ffc77a 2965 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2966 queue_con(con);
31b8006e 2967 }
31b8006e
SW		 2968}
2969
2970
2971
2972/*
15d9882c 2973 * initialize a new messenger instance
31b8006e 2974 */
15d9882c 2975void ceph_messenger_init(struct ceph_messenger *msgr,
859bff51 2976 struct ceph_entity_addr *myaddr)
31b8006e 2977{
31b8006e
SW		 2978 spin_lock_init(&msgr->global_seq_lock);
2979
31b8006e
SW		 2980 if (myaddr)
2981 msgr->inst.addr = *myaddr;
2982
2983 /* select a random nonce */
ac8839d7 2984 msgr->inst.addr.type = 0;
103e2d3a 2985 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2986 encode_my_addr(msgr);
31b8006e 2987
a2a32584 2988 atomic_set(&msgr->stopping, 0);
757856d2 2989 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
31b8006e 2990
15d9882c 2991 dout("%s %p\n", __func__, msgr);
31b8006e 2992}
15d9882c 2993EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2994
757856d2
ID		 2995void ceph_messenger_fini(struct ceph_messenger *msgr)
2996{
2997 put_net(read_pnet(&msgr->net));
2998}
2999EXPORT_SYMBOL(ceph_messenger_fini);
3000
583d0fef
ID		 3001static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3002{
3003 if (msg->con)
3004 msg->con->ops->put(msg->con);
3005
3006 msg->con = con ? con->ops->get(con) : NULL;
3007 BUG_ON(msg->con != con);
3008}
3009
e00de341
SW		 3010static void clear_standby(struct ceph_connection *con)
3011{
3012 /* come back from STANDBY? */
8dacc7da 3013 if (con->state == CON_STATE_STANDBY) {
e00de341 3014 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 3015 con->state = CON_STATE_PREOPEN;
e00de341 3016 con->connect_seq++;
c9ffc77a
AE		 3017 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3018 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 3019 }
3020}
3021
31b8006e
SW		 3022/*
3023 * Queue up an outgoing message on the given connection.
3024 */
3025void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3026{
31b8006e 3027 /* set src+dst */
dbad185d 3028 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 3029 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 3030 msg->needs_out_seq = true;
3031
ec302645 3032 mutex_lock(&con->mutex);
92ce034b 3033
8dacc7da 3034 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 3035 dout("con_send %p closed, dropping %p\n", con, msg);
3036 ceph_msg_put(msg);
3037 mutex_unlock(&con->mutex);
3038 return;
3039 }
3040
583d0fef 3041 msg_con_set(msg, con);
92ce034b 3042
31b8006e
SW		 3043 BUG_ON(!list_empty(&msg->list_head));
3044 list_add_tail(&msg->list_head, &con->out_queue);
3045 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3046 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3047 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3048 le32_to_cpu(msg->hdr.front_len),
3049 le32_to_cpu(msg->hdr.middle_len),
3050 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 3051
3052 clear_standby(con);
ec302645 3053 mutex_unlock(&con->mutex);
31b8006e
SW		 3054
3055 /* if there wasn't anything waiting to send before, queue
3056 * new work */
c9ffc77a 3057 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3058 queue_con(con);
3059}
3d14c5d2 3060EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3061
3062/*
3063 * Revoke a message that was previously queued for send
3064 */
6740a845 3065void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3066{
6740a845
AE		 3067 struct ceph_connection *con = msg->con;
3068
583d0fef
ID		 3069 if (!con) {
3070 dout("%s msg %p null con\n", __func__, msg);
6740a845 3071 return; /* Message not in our possession */
583d0fef 3072 }
6740a845 3073
ec302645 3074 mutex_lock(&con->mutex);
31b8006e 3075 if (!list_empty(&msg->list_head)) {
38941f80 3076 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3077 list_del_init(&msg->list_head);
31b8006e 3078 msg->hdr.seq = 0;
38941f80 3079
31b8006e 3080 ceph_msg_put(msg);
ed98adad
SW		 3081 }
3082 if (con->out_msg == msg) {
67645d76
ID		 3083 BUG_ON(con->out_skip);
3084 /* footer */
3085 if (con->out_msg_done) {
3086 con->out_skip += con_out_kvec_skip(con);
3087 } else {
3088 BUG_ON(!msg->data_length);
89f08173 3089 con->out_skip += sizeof_footer(con);
31b8006e 3090 }
67645d76
ID		 3091 /* data, middle, front */
3092 if (msg->data_length)
3093 con->out_skip += msg->cursor.total_resid;
3094 if (msg->middle)
3095 con->out_skip += con_out_kvec_skip(con);
3096 con->out_skip += con_out_kvec_skip(con);
3097
3098 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3099 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
ed98adad 3100 msg->hdr.seq = 0;
67645d76 3101 con->out_msg = NULL;
92ce034b 3102 ceph_msg_put(msg);
31b8006e 3103 }
67645d76 3104
ec302645 3105 mutex_unlock(&con->mutex);
31b8006e
SW		 3106}
3107
350b1c32 3108/*
0d59ab81 3109 * Revoke a message that we may be reading data into
350b1c32 3110 */
8921d114 3111void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3112{
583d0fef 3113 struct ceph_connection *con = msg->con;
8921d114 3114
583d0fef 3115 if (!con) {
8921d114 3116 dout("%s msg %p null con\n", __func__, msg);
8921d114
AE		 3117 return; /* Message not in our possession */
3118 }
3119
350b1c32 3120 mutex_lock(&con->mutex);
8921d114 3121 if (con->in_msg == msg) {
95c96174
ED		 3122 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3123 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3124 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3125
3126 /* skip rest of message */
8921d114
AE		 3127 dout("%s %p msg %p revoked\n", __func__, con, msg);
3128 con->in_base_pos = con->in_base_pos -
350b1c32 3129 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3130 front_len -
3131 middle_len -
3132 data_len -
350b1c32 3133 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3134 ceph_msg_put(con->in_msg);
3135 con->in_msg = NULL;
3136 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3137 con->in_seq++;
350b1c32 3138 } else {
8921d114
AE		 3139 dout("%s %p in_msg %p msg %p no-op\n",
3140 __func__, con, con->in_msg, msg);
350b1c32
SW		 3141 }
3142 mutex_unlock(&con->mutex);
3143}
3144
31b8006e
SW		 3145/*
3146 * Queue a keepalive byte to ensure the tcp connection is alive.
3147 */
3148void ceph_con_keepalive(struct ceph_connection *con)
3149{
e00de341 3150 dout("con_keepalive %p\n", con);
00650931 3151 mutex_lock(&con->mutex);
e00de341 3152 clear_standby(con);
00650931 3153 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3154 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3155 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3156 queue_con(con);
3157}
3d14c5d2 3158EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3159
8b9558aa
YZ		 3160bool ceph_con_keepalive_expired(struct ceph_connection *con,
3161 unsigned long interval)
3162{
3163 if (interval > 0 &&
3164 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3165 struct timespec now = CURRENT_TIME;
3166 struct timespec ts;
3167 jiffies_to_timespec(interval, &ts);
3168 ts = timespec_add(con->last_keepalive_ack, ts);
3169 return timespec_compare(&now, &ts) >= 0;
3170 }
3171 return false;
3172}
3173
6644ed7b 3174static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3175{
6644ed7b
AE		 3176 struct ceph_msg_data *data;
3177
3178 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3179 return NULL;
3180
81b36be4 3181 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
6644ed7b
AE		 3182 if (data)
3183 data->type = type;
5240d9f9 3184 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3185
3186 return data;
3187}
3188
3189static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3190{
3191 if (!data)
3192 return;
3193
5240d9f9 3194 WARN_ON(!list_empty(&data->links));
e4339d28 3195 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3196 ceph_pagelist_release(data->pagelist);
81b36be4 3197 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3198}
3199
90af3602 3200void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3201 size_t length, size_t alignment)
02afca6c 3202{
6644ed7b
AE		 3203 struct ceph_msg_data *data;
3204
07aa1558
AE		 3205 BUG_ON(!pages);
3206 BUG_ON(!length);
6644ed7b
AE		 3207
3208 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3209 BUG_ON(!data);
3210 data->pages = pages;
3211 data->length = length;
3212 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3213
5240d9f9
AE		 3214 list_add_tail(&data->links, &msg->data);
3215 msg->data_length += length;
02afca6c 3216}
90af3602 3217EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3218
90af3602 3219void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3220 struct ceph_pagelist *pagelist)
3221{
6644ed7b
AE		 3222 struct ceph_msg_data *data;
3223
07aa1558
AE		 3224 BUG_ON(!pagelist);
3225 BUG_ON(!pagelist->length);
27fa8385 3226
6644ed7b
AE		 3227 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3228 BUG_ON(!data);
3229 data->pagelist = pagelist;
3230
5240d9f9
AE		 3231 list_add_tail(&data->links, &msg->data);
3232 msg->data_length += pagelist->length;
27fa8385 3233}
90af3602 3234EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3235
ea96571f 3236#ifdef CONFIG_BLOCK
90af3602 3237void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3238 size_t length)
27fa8385 3239{
6644ed7b
AE		 3240 struct ceph_msg_data *data;
3241
07aa1558 3242 BUG_ON(!bio);
27fa8385 3243
6644ed7b
AE		 3244 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3245 BUG_ON(!data);
3246 data->bio = bio;
c851c495 3247 data->bio_length = length;
6644ed7b 3248
5240d9f9
AE		 3249 list_add_tail(&data->links, &msg->data);
3250 msg->data_length += length;
27fa8385 3251}
90af3602 3252EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3253#endif /* CONFIG_BLOCK */
27fa8385 3254
31b8006e
SW		 3255/*
3256 * construct a new message with given type, size
3257 * the new msg has a ref count of 1.
3258 */
b61c2763
SW		 3259struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3260 bool can_fail)
31b8006e
SW		 3261{
3262 struct ceph_msg *m;
3263
e3d5d638 3264 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3265 if (m == NULL)
3266 goto out;
31b8006e
SW		 3267
3268 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3269 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3270 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3271
9516e45b
AE		 3272 INIT_LIST_HEAD(&m->list_head);
3273 kref_init(&m->kref);
5240d9f9 3274 INIT_LIST_HEAD(&m->data);
ca20892d 3275
31b8006e
SW		 3276 /* front */
3277 if (front_len) {
eeb0bed5 3278 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3279 if (m->front.iov_base == NULL) {
b61c2763 3280 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3281 front_len);
3282 goto out2;
3283 }
3284 } else {
3285 m->front.iov_base = NULL;
3286 }
f2be82b0 3287 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3288
bb257664 3289 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3290 return m;
3291
3292out2:
3293 ceph_msg_put(m);
3294out:
b61c2763
SW		 3295 if (!can_fail) {
3296 pr_err("msg_new can't create type %d front %d\n", type,
3297 front_len);
f0ed1b7c 3298 WARN_ON(1);
b61c2763
SW		 3299 } else {
3300 dout("msg_new can't create type %d front %d\n", type,
3301 front_len);
3302 }
a79832f2 3303 return NULL;
31b8006e 3304}
3d14c5d2 3305EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3306
31b8006e
SW		 3307/*
3308 * Allocate "middle" portion of a message, if it is needed and wasn't
3309 * allocated by alloc_msg. This allows us to read a small fixed-size
3310 * per-type header in the front and then gracefully fail (i.e.,
3311 * propagate the error to the caller based on info in the front) when
3312 * the middle is too large.
3313 */
2450418c 3314static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3315{
3316 int type = le16_to_cpu(msg->hdr.type);
3317 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3318
3319 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3320 ceph_msg_type_name(type), middle_len);
3321 BUG_ON(!middle_len);
3322 BUG_ON(msg->middle);
3323
b6c1d5b8 3324 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3325 if (!msg->middle)
3326 return -ENOMEM;
3327 return 0;
3328}
3329
2450418c 3330/*
1c20f2d2
AE		 3331 * Allocate a message for receiving an incoming message on a
3332 * connection, and save the result in con->in_msg. Uses the
3333 * connection's private alloc_msg op if available.
3334 *
4740a623
SW		 3335 * Returns 0 on success, or a negative error code.
3336 *
3337 * On success, if we set *skip = 1:
3338 * - the next message should be skipped and ignored.
3339 * - con->in_msg == NULL
3340 * or if we set *skip = 0:
3341 * - con->in_msg is non-null.
3342 * On error (ENOMEM, EAGAIN, ...),
3343 * - con->in_msg == NULL
2450418c 3344 */
4740a623 3345static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3346{
4740a623 3347 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3348 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3349 struct ceph_msg *msg;
4740a623 3350 int ret = 0;
2450418c 3351
1c20f2d2 3352 BUG_ON(con->in_msg != NULL);
53ded495 3353 BUG_ON(!con->ops->alloc_msg);
2450418c 3354
53ded495
AE		 3355 mutex_unlock(&con->mutex);
3356 msg = con->ops->alloc_msg(con, hdr, skip);
3357 mutex_lock(&con->mutex);
3358 if (con->state != CON_STATE_OPEN) {
3359 if (msg)
1d866d1c 3360 ceph_msg_put(msg);
53ded495
AE		 3361 return -EAGAIN;
3362 }
4137577a
AE		 3363 if (msg) {
3364 BUG_ON(*skip);
583d0fef 3365 msg_con_set(msg, con);
4137577a 3366 con->in_msg = msg;
4137577a
AE		 3367 } else {
3368 /*
3369 * Null message pointer means either we should skip
3370 * this message or we couldn't allocate memory. The
3371 * former is not an error.
3372 */
3373 if (*skip)
3374 return 0;
4137577a 3375
67c64eb7 3376 con->error_msg = "error allocating memory for incoming message";
53ded495 3377 return -ENOMEM;
2450418c 3378 }
1c20f2d2 3379 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3380
1c20f2d2
AE		 3381 if (middle_len && !con->in_msg->middle) {
3382 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3383 if (ret < 0) {
1c20f2d2
AE		 3384 ceph_msg_put(con->in_msg);
3385 con->in_msg = NULL;
2450418c
YS		 3386 }
3387 }
9d7f0f13 3388
4740a623 3389 return ret;
2450418c
YS		 3390}
3391
31b8006e
SW		 3392
3393/*
3394 * Free a generically kmalloc'd message.
3395 */
0215e44b 3396static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3397{
0215e44b 3398 dout("%s %p\n", __func__, m);
4965fc38 3399 kvfree(m->front.iov_base);
e3d5d638 3400 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3401}
3402
0215e44b 3403static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3404{
3405 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
10bcee14 3406 struct ceph_msg_data *data, *next;
31b8006e 3407
0215e44b 3408 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3409 WARN_ON(!list_empty(&m->list_head));
3410
583d0fef
ID		 3411 msg_con_set(m, NULL);
3412
c2e552e7
SW		 3413 /* drop middle, data, if any */
3414 if (m->middle) {
3415 ceph_buffer_put(m->middle);
3416 m->middle = NULL;
31b8006e 3417 }
5240d9f9 3418
10bcee14
GT		 3419 list_for_each_entry_safe(data, next, &m->data, links) {
3420 list_del_init(&data->links);
5240d9f9
AE		 3421 ceph_msg_data_destroy(data);
3422 }
a1930804 3423 m->data_length = 0;
58bb3b37 3424
c2e552e7
SW		 3425 if (m->pool)
3426 ceph_msgpool_put(m->pool, m);
3427 else
0215e44b
ID		 3428 ceph_msg_free(m);
3429}
3430
3431struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3432{
3433 dout("%s %p (was %d)\n", __func__, msg,
3434 atomic_read(&msg->kref.refcount));
3435 kref_get(&msg->kref);
3436 return msg;
3437}
3438EXPORT_SYMBOL(ceph_msg_get);
3439
3440void ceph_msg_put(struct ceph_msg *msg)
3441{
3442 dout("%s %p (was %d)\n", __func__, msg,
3443 atomic_read(&msg->kref.refcount));
3444 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3445}
0215e44b 3446EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3447
3448void ceph_msg_dump(struct ceph_msg *msg)
3449{
3cea4c30
ID		 3450 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3451 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3452 print_hex_dump(KERN_DEBUG, "header: ",
3453 DUMP_PREFIX_OFFSET, 16, 1,
3454 &msg->hdr, sizeof(msg->hdr), true);
3455 print_hex_dump(KERN_DEBUG, " front: ",
3456 DUMP_PREFIX_OFFSET, 16, 1,
3457 msg->front.iov_base, msg->front.iov_len, true);
3458 if (msg->middle)
3459 print_hex_dump(KERN_DEBUG, "middle: ",
3460 DUMP_PREFIX_OFFSET, 16, 1,
3461 msg->middle->vec.iov_base,
3462 msg->middle->vec.iov_len, true);
3463 print_hex_dump(KERN_DEBUG, "footer: ",
3464 DUMP_PREFIX_OFFSET, 16, 1,
3465 &msg->footer, sizeof(msg->footer), true);
3466}
3d14c5d2 3467EXPORT_SYMBOL(ceph_msg_dump);
ubuntu-bionic-kernel mirror