]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - net/ceph/messenger.c
projects / mirror_ubuntu-zesty-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[mirror_ubuntu-zesty-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);
238 ceph_msg_cache = kmem_cache_create("ceph_msg",
239 sizeof (struct ceph_msg),
240 __alignof__(struct ceph_msg), 0, NULL);
81b36be4
AE		 241
242 if (!ceph_msg_cache)
243 return -ENOMEM;
244
245 BUG_ON(ceph_msg_data_cache);
246 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
247 sizeof (struct ceph_msg_data),
248 __alignof__(struct ceph_msg_data),
249 0, NULL);
250 if (ceph_msg_data_cache)
251 return 0;
252
253 kmem_cache_destroy(ceph_msg_cache);
254 ceph_msg_cache = NULL;
255
256 return -ENOMEM;
e3d5d638
AE		 257}
258
259static void ceph_msgr_slab_exit(void)
260{
81b36be4
AE		 261 BUG_ON(!ceph_msg_data_cache);
262 kmem_cache_destroy(ceph_msg_data_cache);
263 ceph_msg_data_cache = NULL;
264
e3d5d638
AE		 265 BUG_ON(!ceph_msg_cache);
266 kmem_cache_destroy(ceph_msg_cache);
267 ceph_msg_cache = NULL;
268}
269
15417167 270static void _ceph_msgr_exit(void)
6173d1f0 271{
d3002b97 272 if (ceph_msgr_wq) {
6173d1f0 273 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 274 ceph_msgr_wq = NULL;
275 }
6173d1f0 276
6173d1f0 277 BUG_ON(zero_page == NULL);
6173d1f0
AE		 278 page_cache_release(zero_page);
279 zero_page = NULL;
d920ff6f
BC		 280
281 ceph_msgr_slab_exit();
6173d1f0
AE		 282}
283
3d14c5d2 284int ceph_msgr_init(void)
31b8006e 285{
d920ff6f
BC		 286 if (ceph_msgr_slab_init())
287 return -ENOMEM;
288
57666519
AE		 289 BUG_ON(zero_page != NULL);
290 zero_page = ZERO_PAGE(0);
291 page_cache_get(zero_page);
292
f9865f06
ID		 293 /*
294 * The number of active work items is limited by the number of
295 * connections, so leave @max_active at default.
296 */
297 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
6173d1f0
AE		 298 if (ceph_msgr_wq)
299 return 0;
57666519 300
6173d1f0
AE		 301 pr_err("msgr_init failed to create workqueue\n");
302 _ceph_msgr_exit();
57666519 303
6173d1f0 304 return -ENOMEM;
31b8006e 305}
3d14c5d2 306EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 307
308void ceph_msgr_exit(void)
309{
57666519 310 BUG_ON(ceph_msgr_wq == NULL);
57666519 311
6173d1f0 312 _ceph_msgr_exit();
31b8006e 313}
3d14c5d2 314EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 315
cd84db6e 316void ceph_msgr_flush(void)
a922d38f
SW		 317{
318 flush_workqueue(ceph_msgr_wq);
319}
3d14c5d2 320EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 321
ce2c8903
AE		 322/* Connection socket state transition functions */
323
324static void con_sock_state_init(struct ceph_connection *con)
325{
326 int old_state;
327
328 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
329 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
330 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 331 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
332 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 333}
334
335static void con_sock_state_connecting(struct ceph_connection *con)
336{
337 int old_state;
338
339 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
340 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
341 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 342 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
343 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 344}
345
346static void con_sock_state_connected(struct ceph_connection *con)
347{
348 int old_state;
349
350 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
351 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
352 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 353 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
354 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 355}
356
357static void con_sock_state_closing(struct ceph_connection *con)
358{
359 int old_state;
360
361 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
362 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
363 old_state != CON_SOCK_STATE_CONNECTED &&
364 old_state != CON_SOCK_STATE_CLOSING))
365 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 366 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
367 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 368}
369
370static void con_sock_state_closed(struct ceph_connection *con)
371{
372 int old_state;
373
374 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
375 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 376 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 377 old_state != CON_SOCK_STATE_CONNECTING &&
378 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 379 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 380 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
381 CON_SOCK_STATE_CLOSED);
ce2c8903 382}
a922d38f 383
31b8006e
SW		 384/*
385 * socket callback functions
386 */
387
388/* data available on socket, or listen socket received a connect */
676d2369 389static void ceph_sock_data_ready(struct sock *sk)
31b8006e 390{
bd406145 391 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 392 if (atomic_read(&con->msgr->stopping)) {
393 return;
394 }
bd406145 395
31b8006e 396 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 397 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 398 con, con->state);
399 queue_con(con);
400 }
401}
402
403/* socket has buffer space for writing */
327800bd 404static void ceph_sock_write_space(struct sock *sk)
31b8006e 405{
d3002b97 406 struct ceph_connection *con = sk->sk_user_data;
31b8006e 407
182fac26
JS		 408 /* only queue to workqueue if there is data we want to write,
409 * and there is sufficient space in the socket buffer to accept
327800bd 410 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 411 * doesn't get called again until try_write() fills the socket
412 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
413 * and net/core/stream.c:sk_stream_write_space().
414 */
c9ffc77a 415 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
64dc6130 416 if (sk_stream_is_writeable(sk)) {
327800bd 417 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 418 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
419 queue_con(con);
420 }
31b8006e 421 } else {
327800bd 422 dout("%s %p nothing to write\n", __func__, con);
31b8006e 423 }
31b8006e
SW		 424}
425
426/* socket's state has changed */
327800bd 427static void ceph_sock_state_change(struct sock *sk)
31b8006e 428{
bd406145 429 struct ceph_connection *con = sk->sk_user_data;
31b8006e 430
327800bd 431 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 432 con, con->state, sk->sk_state);
433
31b8006e
SW		 434 switch (sk->sk_state) {
435 case TCP_CLOSE:
327800bd 436 dout("%s TCP_CLOSE\n", __func__);
31b8006e 437 case TCP_CLOSE_WAIT:
327800bd 438 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 439 con_sock_state_closing(con);
c9ffc77a 440 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 441 queue_con(con);
31b8006e
SW		 442 break;
443 case TCP_ESTABLISHED:
327800bd 444 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 445 con_sock_state_connected(con);
31b8006e
SW		 446 queue_con(con);
447 break;
d3002b97
AE		 448 default: /* Everything else is uninteresting */
449 break;
31b8006e
SW		 450 }
451}
452
453/*
454 * set up socket callbacks
455 */
456static void set_sock_callbacks(struct socket *sock,
457 struct ceph_connection *con)
458{
459 struct sock *sk = sock->sk;
bd406145 460 sk->sk_user_data = con;
327800bd
AE		 461 sk->sk_data_ready = ceph_sock_data_ready;
462 sk->sk_write_space = ceph_sock_write_space;
463 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 464}
465
466
467/*
468 * socket helpers
469 */
470
471/*
472 * initiate connection to a remote socket.
473 */
41617d0c 474static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 475{
f91d3471 476 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e
SW		 477 struct socket *sock;
478 int ret;
479
480 BUG_ON(con->sock);
757856d2 481 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
eeb1bd5c 482 SOCK_STREAM, IPPROTO_TCP, &sock);
31b8006e 483 if (ret)
41617d0c 484 return ret;
6d7fdb0a 485 sock->sk->sk_allocation = GFP_NOFS;
31b8006e 486
a6a5349d
SW		 487#ifdef CONFIG_LOCKDEP
488 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
489#endif
490
31b8006e
SW		 491 set_sock_callbacks(sock, con);
492
3d14c5d2 493 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 494
89a86be0 495 con_sock_state_connecting(con);
f91d3471
SW		 496 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
497 O_NONBLOCK);
31b8006e
SW		 498 if (ret == -EINPROGRESS) {
499 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 500 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 501 sock->sk->sk_state);
a5bc3129 502 } else if (ret < 0) {
31b8006e 503 pr_err("connect %s error %d\n",
3d14c5d2 504 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 505 sock_release(sock);
41617d0c 506 return ret;
a5bc3129 507 }
89baaa57 508
859bff51 509 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
ba988f87
CH		 510 int optval = 1;
511
512 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
513 (char *)&optval, sizeof(optval));
514 if (ret)
515 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
516 ret);
517 }
518
a5bc3129 519 con->sock = sock;
41617d0c 520 return 0;
31b8006e
SW		 521}
522
523static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
524{
525 struct kvec iov = {buf, len};
526 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 527 int r;
31b8006e 528
98bdb0aa
SW		 529 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
530 if (r == -EAGAIN)
531 r = 0;
532 return r;
31b8006e
SW		 533}
534
afb3d90e
AE		 535static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
536 int page_offset, size_t length)
537{
538 void *kaddr;
539 int ret;
540
541 BUG_ON(page_offset + length > PAGE_SIZE);
542
543 kaddr = kmap(page);
544 BUG_ON(!kaddr);
545 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
546 kunmap(page);
547
548 return ret;
549}
550
31b8006e
SW		 551/*
552 * write something. @more is true if caller will be sending more data
553 * shortly.
554 */
555static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
556 size_t kvlen, size_t len, int more)
557{
558 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 559 int r;
31b8006e
SW		 560
561 if (more)
562 msg.msg_flags |= MSG_MORE;
563 else
564 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
565
42961d23
SW		 566 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
567 if (r == -EAGAIN)
568 r = 0;
569 return r;
31b8006e
SW		 570}
571
178eda29 572static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 573 int offset, size_t size, bool more)
31739139
AE		 574{
575 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
576 int ret;
577
578 ret = kernel_sendpage(sock, page, offset, size, flags);
579 if (ret == -EAGAIN)
580 ret = 0;
581
582 return ret;
583}
584
178eda29
CC		 585static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
586 int offset, size_t size, bool more)
587{
588 int ret;
589 struct kvec iov;
590
591 /* sendpage cannot properly handle pages with page_count == 0,
592 * we need to fallback to sendmsg if that's the case */
593 if (page_count(page) >= 1)
594 return __ceph_tcp_sendpage(sock, page, offset, size, more);
595
596 iov.iov_base = kmap(page) + offset;
597 iov.iov_len = size;
598 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
599 kunmap(page);
600
601 return ret;
602}
31b8006e
SW		 603
604/*
605 * Shutdown/close the socket for the given connection.
606 */
607static int con_close_socket(struct ceph_connection *con)
608{
8007b8d6 609 int rc = 0;
31b8006e
SW		 610
611 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 612 if (con->sock) {
613 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
614 sock_release(con->sock);
615 con->sock = NULL;
616 }
456ea468
AE		 617
618 /*
4a861692 619 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 620 * independent of the connection mutex, and we could have
621 * received a socket close event before we had the chance to
622 * shut the socket down.
623 */
c9ffc77a 624 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 625
ce2c8903 626 con_sock_state_closed(con);
31b8006e
SW		 627 return rc;
628}
629
630/*
631 * Reset a connection. Discard all incoming and outgoing messages
632 * and clear *_seq state.
633 */
634static void ceph_msg_remove(struct ceph_msg *msg)
635{
636 list_del_init(&msg->list_head);
38941f80 637
31b8006e
SW		 638 ceph_msg_put(msg);
639}
640static void ceph_msg_remove_list(struct list_head *head)
641{
642 while (!list_empty(head)) {
643 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
644 list_head);
645 ceph_msg_remove(msg);
646 }
647}
648
649static void reset_connection(struct ceph_connection *con)
650{
651 /* reset connection, out_queue, msg_ and connect_seq */
652 /* discard existing out_queue and msg_seq */
0fa6ebc6 653 dout("reset_connection %p\n", con);
31b8006e
SW		 654 ceph_msg_remove_list(&con->out_queue);
655 ceph_msg_remove_list(&con->out_sent);
656
cf3e5c40 657 if (con->in_msg) {
38941f80 658 BUG_ON(con->in_msg->con != con);
cf3e5c40
SW		 659 ceph_msg_put(con->in_msg);
660 con->in_msg = NULL;
661 }
662
31b8006e
SW		 663 con->connect_seq = 0;
664 con->out_seq = 0;
c86a2930 665 if (con->out_msg) {
583d0fef 666 BUG_ON(con->out_msg->con != con);
c86a2930
SW		 667 ceph_msg_put(con->out_msg);
668 con->out_msg = NULL;
669 }
31b8006e 670 con->in_seq = 0;
0e0d5e0c 671 con->in_seq_acked = 0;
67645d76
ID		 672
673 con->out_skip = 0;
31b8006e
SW		 674}
675
676/*
677 * mark a peer down. drop any open connections.
678 */
679void ceph_con_close(struct ceph_connection *con)
680{
8c50c817 681 mutex_lock(&con->mutex);
3d14c5d2
YS		 682 dout("con_close %p peer %s\n", con,
683 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 684 con->state = CON_STATE_CLOSED;
a5988c49 685
c9ffc77a
AE		 686 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
687 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
688 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
689 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 690
31b8006e 691 reset_connection(con);
6f2bc3ff 692 con->peer_global_seq = 0;
37ab77ac 693 cancel_con(con);
ee76e073 694 con_close_socket(con);
ec302645 695 mutex_unlock(&con->mutex);
31b8006e 696}
3d14c5d2 697EXPORT_SYMBOL(ceph_con_close);
31b8006e 698
31b8006e
SW		 699/*
700 * Reopen a closed connection, with a new peer address.
701 */
b7a9e5dd
SW		 702void ceph_con_open(struct ceph_connection *con,
703 __u8 entity_type, __u64 entity_num,
704 struct ceph_entity_addr *addr)
31b8006e 705{
5469155f 706 mutex_lock(&con->mutex);
3d14c5d2 707 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 708
122070a2 709 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 710 con->state = CON_STATE_PREOPEN;
a5988c49 711
b7a9e5dd
SW		 712 con->peer_name.type = (__u8) entity_type;
713 con->peer_name.num = cpu_to_le64(entity_num);
714
31b8006e 715 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 716 con->delay = 0; /* reset backoff memory */
5469155f 717 mutex_unlock(&con->mutex);
31b8006e
SW		 718 queue_con(con);
719}
3d14c5d2 720EXPORT_SYMBOL(ceph_con_open);
31b8006e 721
87b315a5
SW		 722/*
723 * return true if this connection ever successfully opened
724 */
725bool ceph_con_opened(struct ceph_connection *con)
726{
727 return con->connect_seq > 0;
728}
729
31b8006e
SW		 730/*
731 * initialize a new connection.
732 */
1bfd89f4
AE		 733void ceph_con_init(struct ceph_connection *con, void *private,
734 const struct ceph_connection_operations *ops,
b7a9e5dd 735 struct ceph_messenger *msgr)
31b8006e
SW		 736{
737 dout("con_init %p\n", con);
738 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 739 con->private = private;
740 con->ops = ops;
31b8006e 741 con->msgr = msgr;
ce2c8903
AE		 742
743 con_sock_state_init(con);
744
ec302645 745 mutex_init(&con->mutex);
31b8006e
SW		 746 INIT_LIST_HEAD(&con->out_queue);
747 INIT_LIST_HEAD(&con->out_sent);
68931622 748 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
a5988c49 749
8dacc7da 750 con->state = CON_STATE_CLOSED;
31b8006e 751}
3d14c5d2 752EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 753
754
755/*
756 * We maintain a global counter to order connection attempts. Get
757 * a unique seq greater than @gt.
758 */
759static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
760{
761 u32 ret;
762
763 spin_lock(&msgr->global_seq_lock);
764 if (msgr->global_seq < gt)
765 msgr->global_seq = gt;
766 ret = ++msgr->global_seq;
767 spin_unlock(&msgr->global_seq_lock);
768 return ret;
769}
770
e2200423 771static void con_out_kvec_reset(struct ceph_connection *con)
859eb799 772{
67645d76
ID		 773 BUG_ON(con->out_skip);
774
859eb799
AE		 775 con->out_kvec_left = 0;
776 con->out_kvec_bytes = 0;
777 con->out_kvec_cur = &con->out_kvec[0];
778}
779
e2200423 780static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 781 size_t size, void *data)
782{
67645d76 783 int index = con->out_kvec_left;
859eb799 784
67645d76 785 BUG_ON(con->out_skip);
859eb799
AE		 786 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
787
788 con->out_kvec[index].iov_len = size;
789 con->out_kvec[index].iov_base = data;
790 con->out_kvec_left++;
791 con->out_kvec_bytes += size;
792}
31b8006e 793
67645d76
ID		 794/*
795 * Chop off a kvec from the end. Return residual number of bytes for
796 * that kvec, i.e. how many bytes would have been written if the kvec
797 * hadn't been nuked.
798 */
799static int con_out_kvec_skip(struct ceph_connection *con)
800{
801 int off = con->out_kvec_cur - con->out_kvec;
802 int skip = 0;
803
804 if (con->out_kvec_bytes > 0) {
805 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
806 BUG_ON(con->out_kvec_bytes < skip);
807 BUG_ON(!con->out_kvec_left);
808 con->out_kvec_bytes -= skip;
809 con->out_kvec_left--;
810 }
811
812 return skip;
813}
814
df6ad1f9 815#ifdef CONFIG_BLOCK
6aaa4511
AE		 816
817/*
818 * For a bio data item, a piece is whatever remains of the next
819 * entry in the current bio iovec, or the first entry in the next
820 * bio in the list.
821 */
8ae4f4f5 822static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 823 size_t length)
6aaa4511 824{
8ae4f4f5 825 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 826 struct bio *bio;
827
828 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
829
830 bio = data->bio;
831 BUG_ON(!bio);
6aaa4511 832
ca8b3a69 833 cursor->resid = min(length, data->bio_length);
6aaa4511 834 cursor->bio = bio;
f38a5181
KO		 835 cursor->bvec_iter = bio->bi_iter;
836 cursor->last_piece =
837 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
6aaa4511
AE		 838}
839
8ae4f4f5 840static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 841 size_t *page_offset,
842 size_t *length)
843{
8ae4f4f5 844 struct ceph_msg_data *data = cursor->data;
6aaa4511 845 struct bio *bio;
f38a5181 846 struct bio_vec bio_vec;
6aaa4511
AE		 847
848 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
849
850 bio = cursor->bio;
851 BUG_ON(!bio);
852
f38a5181 853 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511 854
f38a5181 855 *page_offset = (size_t) bio_vec.bv_offset;
6aaa4511 856 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 857 if (cursor->last_piece) /* pagelist offset is always 0 */
858 *length = cursor->resid;
859 else
f38a5181 860 *length = (size_t) bio_vec.bv_len;
25aff7c5 861 BUG_ON(*length > cursor->resid);
5df521b1 862 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511 863
f38a5181 864 return bio_vec.bv_page;
6aaa4511
AE		 865}
866
8ae4f4f5
AE		 867static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
868 size_t bytes)
6aaa4511 869{
6aaa4511 870 struct bio *bio;
f38a5181 871 struct bio_vec bio_vec;
6aaa4511 872
8ae4f4f5 873 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
6aaa4511
AE		 874
875 bio = cursor->bio;
876 BUG_ON(!bio);
877
f38a5181 878 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511
AE		 879
880 /* Advance the cursor offset */
881
25aff7c5
AE		 882 BUG_ON(cursor->resid < bytes);
883 cursor->resid -= bytes;
f38a5181
KO		 884
885 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
886
887 if (bytes < bio_vec.bv_len)
6aaa4511
AE		 888 return false; /* more bytes to process in this segment */
889
890 /* Move on to the next segment, and possibly the next bio */
891
f38a5181 892 if (!cursor->bvec_iter.bi_size) {
6aaa4511 893 bio = bio->bi_next;
0ec1d15e
ID		 894 cursor->bio = bio;
895 if (bio)
896 cursor->bvec_iter = bio->bi_iter;
897 else
898 memset(&cursor->bvec_iter, 0,
899 sizeof(cursor->bvec_iter));
6aaa4511 900 }
6aaa4511 901
25aff7c5
AE		 902 if (!cursor->last_piece) {
903 BUG_ON(!cursor->resid);
904 BUG_ON(!bio);
905 /* A short read is OK, so use <= rather than == */
f38a5181 906 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
6aaa4511 907 cursor->last_piece = true;
25aff7c5 908 }
6aaa4511
AE		 909
910 return true;
911}
ea96571f 912#endif /* CONFIG_BLOCK */
df6ad1f9 913
e766d7b5
AE		 914/*
915 * For a page array, a piece comes from the first page in the array
916 * that has not already been fully consumed.
917 */
8ae4f4f5 918static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 919 size_t length)
e766d7b5 920{
8ae4f4f5 921 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 922 int page_count;
923
924 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
925
926 BUG_ON(!data->pages);
927 BUG_ON(!data->length);
928
ca8b3a69 929 cursor->resid = min(length, data->length);
e766d7b5 930 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 931 cursor->page_offset = data->alignment & ~PAGE_MASK;
932 cursor->page_index = 0;
56fc5659
AE		 933 BUG_ON(page_count > (int)USHRT_MAX);
934 cursor->page_count = (unsigned short)page_count;
935 BUG_ON(length > SIZE_MAX - cursor->page_offset);
5f740d7e 936 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 937}
938
8ae4f4f5
AE		 939static struct page *
940ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
941 size_t *page_offset, size_t *length)
e766d7b5 942{
8ae4f4f5 943 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 944
945 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
946
947 BUG_ON(cursor->page_index >= cursor->page_count);
948 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 949
950 *page_offset = cursor->page_offset;
25aff7c5 951 if (cursor->last_piece)
e766d7b5 952 *length = cursor->resid;
25aff7c5 953 else
e766d7b5 954 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 955
956 return data->pages[cursor->page_index];
957}
958
8ae4f4f5 959static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 960 size_t bytes)
961{
8ae4f4f5 962 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 963
964 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 965
966 /* Advance the cursor page offset */
967
968 cursor->resid -= bytes;
5df521b1
AE		 969 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
970 if (!bytes || cursor->page_offset)
e766d7b5
AE		 971 return false; /* more bytes to process in the current page */
972
d90deda6
YZ		 973 if (!cursor->resid)
974 return false; /* no more data */
975
5df521b1 976 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 977
978 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 979 cursor->page_index++;
25aff7c5 980 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 981
982 return true;
983}
984
fe38a2b6 985/*
dd236fcb
AE		 986 * For a pagelist, a piece is whatever remains to be consumed in the
987 * first page in the list, or the front of the next page.
fe38a2b6 988 */
8ae4f4f5
AE		 989static void
990ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 991 size_t length)
fe38a2b6 992{
8ae4f4f5 993 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 994 struct ceph_pagelist *pagelist;
995 struct page *page;
996
dd236fcb 997 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 998
999 pagelist = data->pagelist;
1000 BUG_ON(!pagelist);
25aff7c5
AE		 1001
1002 if (!length)
fe38a2b6
AE		 1003 return; /* pagelist can be assigned but empty */
1004
1005 BUG_ON(list_empty(&pagelist->head));
1006 page = list_first_entry(&pagelist->head, struct page, lru);
1007
ca8b3a69 1008 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 1009 cursor->page = page;
1010 cursor->offset = 0;
a51b272e 1011 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1012}
1013
8ae4f4f5
AE		 1014static struct page *
1015ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1016 size_t *page_offset, size_t *length)
fe38a2b6 1017{
8ae4f4f5 1018 struct ceph_msg_data *data = cursor->data;
fe38a2b6 1019 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 1020
1021 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1022
1023 pagelist = data->pagelist;
1024 BUG_ON(!pagelist);
1025
1026 BUG_ON(!cursor->page);
25aff7c5 1027 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 1028
5df521b1 1029 /* offset of first page in pagelist is always 0 */
fe38a2b6 1030 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 1031 if (cursor->last_piece)
25aff7c5
AE		 1032 *length = cursor->resid;
1033 else
1034 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1035
8ae4f4f5 1036 return cursor->page;
fe38a2b6
AE		 1037}
1038
8ae4f4f5 1039static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1040 size_t bytes)
fe38a2b6 1041{
8ae4f4f5 1042 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1043 struct ceph_pagelist *pagelist;
1044
1045 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1046
1047 pagelist = data->pagelist;
1048 BUG_ON(!pagelist);
25aff7c5
AE		 1049
1050 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1051 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1052
1053 /* Advance the cursor offset */
1054
25aff7c5 1055 cursor->resid -= bytes;
fe38a2b6 1056 cursor->offset += bytes;
5df521b1 1057 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1058 if (!bytes || cursor->offset & ~PAGE_MASK)
1059 return false; /* more bytes to process in the current page */
1060
d90deda6
YZ		 1061 if (!cursor->resid)
1062 return false; /* no more data */
1063
fe38a2b6
AE		 1064 /* Move on to the next page */
1065
1066 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
17ddc49b 1067 cursor->page = list_next_entry(cursor->page, lru);
25aff7c5 1068 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1069
1070 return true;
1071}
1072
dd236fcb
AE		 1073/*
1074 * Message data is handled (sent or received) in pieces, where each
1075 * piece resides on a single page. The network layer might not
1076 * consume an entire piece at once. A data item's cursor keeps
1077 * track of which piece is next to process and how much remains to
1078 * be processed in that piece. It also tracks whether the current
1079 * piece is the last one in the data item.
1080 */
ca8b3a69 1081static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1082{
ca8b3a69 1083 size_t length = cursor->total_resid;
8ae4f4f5 1084
8ae4f4f5 1085 switch (cursor->data->type) {
dd236fcb 1086 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1087 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1088 break;
e766d7b5 1089 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1090 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1091 break;
dd236fcb
AE		 1092#ifdef CONFIG_BLOCK
1093 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1094 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1095 break;
dd236fcb 1096#endif /* CONFIG_BLOCK */
6aaa4511 1097 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1098 default:
1099 /* BUG(); */
1100 break;
1101 }
8ae4f4f5 1102 cursor->need_crc = true;
dd236fcb
AE		 1103}
1104
ca8b3a69
AE		 1105static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1106{
1107 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1108 struct ceph_msg_data *data;
1109
1110 BUG_ON(!length);
1111 BUG_ON(length > msg->data_length);
1112 BUG_ON(list_empty(&msg->data));
1113
ca8b3a69
AE		 1114 cursor->data_head = &msg->data;
1115 cursor->total_resid = length;
1116 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1117 cursor->data = data;
1118
1119 __ceph_msg_data_cursor_init(cursor);
1120}
1121
dd236fcb
AE		 1122/*
1123 * Return the page containing the next piece to process for a given
1124 * data item, and supply the page offset and length of that piece.
1125 * Indicate whether this is the last piece in this data item.
1126 */
8ae4f4f5
AE		 1127static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1128 size_t *page_offset, size_t *length,
dd236fcb
AE		 1129 bool *last_piece)
1130{
1131 struct page *page;
1132
8ae4f4f5 1133 switch (cursor->data->type) {
dd236fcb 1134 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1135 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1136 break;
e766d7b5 1137 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1138 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1139 break;
dd236fcb
AE		 1140#ifdef CONFIG_BLOCK
1141 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1142 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1143 break;
dd236fcb 1144#endif /* CONFIG_BLOCK */
6aaa4511 1145 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1146 default:
1147 page = NULL;
1148 break;
1149 }
1150 BUG_ON(!page);
1151 BUG_ON(*page_offset + *length > PAGE_SIZE);
1152 BUG_ON(!*length);
1153 if (last_piece)
8ae4f4f5 1154 *last_piece = cursor->last_piece;
dd236fcb
AE		 1155
1156 return page;
1157}
1158
1159/*
1160 * Returns true if the result moves the cursor on to the next piece
1161 * of the data item.
1162 */
8ae4f4f5
AE		 1163static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1164 size_t bytes)
dd236fcb
AE		 1165{
1166 bool new_piece;
1167
25aff7c5 1168 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1169 switch (cursor->data->type) {
dd236fcb 1170 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1171 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1172 break;
e766d7b5 1173 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1174 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1175 break;
dd236fcb
AE		 1176#ifdef CONFIG_BLOCK
1177 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1178 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1179 break;
dd236fcb 1180#endif /* CONFIG_BLOCK */
6aaa4511 1181 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1182 default:
1183 BUG();
1184 break;
1185 }
ca8b3a69 1186 cursor->total_resid -= bytes;
dd236fcb 1187
ca8b3a69
AE		 1188 if (!cursor->resid && cursor->total_resid) {
1189 WARN_ON(!cursor->last_piece);
1190 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
17ddc49b 1191 cursor->data = list_next_entry(cursor->data, links);
ca8b3a69 1192 __ceph_msg_data_cursor_init(cursor);
a51b272e 1193 new_piece = true;
ca8b3a69 1194 }
a51b272e 1195 cursor->need_crc = new_piece;
ca8b3a69 1196
dd236fcb
AE		 1197 return new_piece;
1198}
1199
dbc0d3ca
ID		 1200static size_t sizeof_footer(struct ceph_connection *con)
1201{
1202 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1203 sizeof(struct ceph_msg_footer) :
1204 sizeof(struct ceph_msg_footer_old);
1205}
1206
98fa5dd8 1207static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1208{
739c905b 1209 BUG_ON(!msg);
25aff7c5 1210 BUG_ON(!data_len);
739c905b 1211
4c59b4a2 1212 /* Initialize data cursor */
fe38a2b6 1213
8ae4f4f5 1214 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1215}
1216
31b8006e
SW		 1217/*
1218 * Prepare footer for currently outgoing message, and finish things
1219 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1220 */
859eb799 1221static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1222{
1223 struct ceph_msg *m = con->out_msg;
859eb799 1224 int v = con->out_kvec_left;
31b8006e 1225
fd154f3c
AE		 1226 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1227
31b8006e 1228 dout("prepare_write_message_footer %p\n", con);
31b8006e 1229 con->out_kvec[v].iov_base = &m->footer;
33d07337
YZ		 1230 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1231 if (con->ops->sign_message)
79dbd1ba 1232 con->ops->sign_message(m);
33d07337
YZ		 1233 else
1234 m->footer.sig = 0;
1235 con->out_kvec[v].iov_len = sizeof(m->footer);
1236 con->out_kvec_bytes += sizeof(m->footer);
1237 } else {
1238 m->old_footer.flags = m->footer.flags;
1239 con->out_kvec[v].iov_len = sizeof(m->old_footer);
1240 con->out_kvec_bytes += sizeof(m->old_footer);
1241 }
31b8006e
SW		 1242 con->out_kvec_left++;
1243 con->out_more = m->more_to_follow;
c86a2930 1244 con->out_msg_done = true;
31b8006e
SW		 1245}
1246
1247/*
1248 * Prepare headers for the next outgoing message.
1249 */
1250static void prepare_write_message(struct ceph_connection *con)
1251{
1252 struct ceph_msg *m;
a9a0c51a 1253 u32 crc;
31b8006e 1254
e2200423 1255 con_out_kvec_reset(con);
c86a2930 1256 con->out_msg_done = false;
31b8006e
SW		 1257
1258 /* Sneak an ack in there first? If we can get it into the same
1259 * TCP packet that's a good thing. */
1260 if (con->in_seq > con->in_seq_acked) {
1261 con->in_seq_acked = con->in_seq;
e2200423 1262 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1263 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1264 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1265 &con->out_temp_ack);
31b8006e
SW		 1266 }
1267
38941f80 1268 BUG_ON(list_empty(&con->out_queue));
859eb799 1269 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1270 con->out_msg = m;
38941f80 1271 BUG_ON(m->con != con);
4cf9d544
SW		 1272
1273 /* put message on sent list */
1274 ceph_msg_get(m);
1275 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1276
e84346b7
SW		 1277 /*
1278 * only assign outgoing seq # if we haven't sent this message
1279 * yet. if it is requeued, resend with it's original seq.
1280 */
1281 if (m->needs_out_seq) {
1282 m->hdr.seq = cpu_to_le64(++con->out_seq);
1283 m->needs_out_seq = false;
1284 }
98fa5dd8 1285 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1286
98fa5dd8 1287 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1288 m, con->out_seq, le16_to_cpu(m->hdr.type),
1289 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1290 m->data_length);
31b8006e
SW		 1291 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1292
1293 /* tag + hdr + front + middle */
e2200423 1294 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
67645d76 1295 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
e2200423 1296 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1297
31b8006e 1298 if (m->middle)
e2200423 1299 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1300 m->middle->vec.iov_base);
31b8006e 1301
67645d76 1302 /* fill in hdr crc and finalize hdr */
a9a0c51a
AE		 1303 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1304 con->out_msg->hdr.crc = cpu_to_le32(crc);
67645d76 1305 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
a9a0c51a 1306
67645d76 1307 /* fill in front and middle crc, footer */
a9a0c51a
AE		 1308 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1309 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1310 if (m->middle) {
1311 crc = crc32c(0, m->middle->vec.iov_base,
1312 m->middle->vec.iov_len);
1313 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1314 } else
31b8006e 1315 con->out_msg->footer.middle_crc = 0;
739c905b 1316 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1317 le32_to_cpu(con->out_msg->footer.front_crc),
1318 le32_to_cpu(con->out_msg->footer.middle_crc));
67645d76 1319 con->out_msg->footer.flags = 0;
31b8006e
SW		 1320
1321 /* is there a data payload? */
739c905b 1322 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1323 if (m->data_length) {
1324 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1325 con->out_more = 1; /* data + footer will follow */
1326 } else {
31b8006e 1327 /* no, queue up footer too and be done */
859eb799 1328 prepare_write_message_footer(con);
78625051 1329 }
31b8006e 1330
c9ffc77a 1331 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1332}
1333
1334/*
1335 * Prepare an ack.
1336 */
1337static void prepare_write_ack(struct ceph_connection *con)
1338{
1339 dout("prepare_write_ack %p %llu -> %llu\n", con,
1340 con->in_seq_acked, con->in_seq);
1341 con->in_seq_acked = con->in_seq;
1342
e2200423 1343 con_out_kvec_reset(con);
859eb799 1344
e2200423 1345 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1346
31b8006e 1347 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1348 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1349 &con->out_temp_ack);
1350
31b8006e 1351 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1352 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1353}
1354
3a23083b
SW		 1355/*
1356 * Prepare to share the seq during handshake
1357 */
1358static void prepare_write_seq(struct ceph_connection *con)
1359{
1360 dout("prepare_write_seq %p %llu -> %llu\n", con,
1361 con->in_seq_acked, con->in_seq);
1362 con->in_seq_acked = con->in_seq;
1363
1364 con_out_kvec_reset(con);
1365
1366 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1367 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1368 &con->out_temp_ack);
1369
1370 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1371}
1372
31b8006e
SW		 1373/*
1374 * Prepare to write keepalive byte.
1375 */
1376static void prepare_write_keepalive(struct ceph_connection *con)
1377{
1378 dout("prepare_write_keepalive %p\n", con);
e2200423 1379 con_out_kvec_reset(con);
8b9558aa 1380 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
7f61f545
ID		 1381 struct timespec now = CURRENT_TIME;
1382
8b9558aa 1383 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
7f61f545
ID		 1384 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1385 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1386 &con->out_temp_keepalive2);
8b9558aa
YZ		 1387 } else {
1388 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1389 }
c9ffc77a 1390 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1391}
1392
1393/*
1394 * Connection negotiation.
1395 */
1396
dac1e716
AE		 1397static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1398 int *auth_proto)
4e7a5dcd 1399{
a3530df3 1400 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1401
1402 if (!con->ops->get_authorizer) {
1403 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1404 con->out_connect.authorizer_len = 0;
729796be 1405 return NULL;
b1c6b980
AE		 1406 }
1407
1408 /* Can't hold the mutex while getting authorizer */
ec302645 1409 mutex_unlock(&con->mutex);
dac1e716 1410 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1411 mutex_lock(&con->mutex);
4e7a5dcd 1412
a3530df3 1413 if (IS_ERR(auth))
729796be 1414 return auth;
8dacc7da 1415 if (con->state != CON_STATE_NEGOTIATING)
729796be 1416 return ERR_PTR(-EAGAIN);
0da5d703 1417
8f43fb53
AE		 1418 con->auth_reply_buf = auth->authorizer_reply_buf;
1419 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1420 return auth;
4e7a5dcd
SW		 1421}
1422
31b8006e
SW		 1423/*
1424 * We connected to a peer and are saying hello.
1425 */
e825a66d 1426static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1427{
e2200423
AE		 1428 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1429 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1430 &con->msgr->my_enc_addr);
eed0ef2c 1431
eed0ef2c 1432 con->out_more = 0;
c9ffc77a 1433 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1434}
1435
e825a66d 1436static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1437{
95c96174 1438 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1439 int proto;
dac1e716 1440 int auth_proto;
729796be 1441 struct ceph_auth_handshake *auth;
31b8006e
SW		 1442
1443 switch (con->peer_name.type) {
1444 case CEPH_ENTITY_TYPE_MON:
1445 proto = CEPH_MONC_PROTOCOL;
1446 break;
1447 case CEPH_ENTITY_TYPE_OSD:
1448 proto = CEPH_OSDC_PROTOCOL;
1449 break;
1450 case CEPH_ENTITY_TYPE_MDS:
1451 proto = CEPH_MDSC_PROTOCOL;
1452 break;
1453 default:
1454 BUG();
1455 }
1456
1457 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1458 con->connect_seq, global_seq, proto);
4e7a5dcd 1459
859bff51
ID		 1460 con->out_connect.features =
1461 cpu_to_le64(from_msgr(con->msgr)->supported_features);
31b8006e
SW		 1462 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1463 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1464 con->out_connect.global_seq = cpu_to_le32(global_seq);
1465 con->out_connect.protocol_version = cpu_to_le32(proto);
1466 con->out_connect.flags = 0;
31b8006e 1467
dac1e716
AE		 1468 auth_proto = CEPH_AUTH_UNKNOWN;
1469 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1470 if (IS_ERR(auth))
1471 return PTR_ERR(auth);
3da54776 1472
dac1e716 1473 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1474 con->out_connect.authorizer_len = auth ?
1475 cpu_to_le32(auth->authorizer_buf_len) : 0;
1476
e2200423 1477 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1478 &con->out_connect);
1479 if (auth && auth->authorizer_buf_len)
e2200423 1480 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1481 auth->authorizer_buf);
859eb799 1482
31b8006e 1483 con->out_more = 0;
c9ffc77a 1484 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1485
e10c758e 1486 return 0;
31b8006e
SW		 1487}
1488
31b8006e
SW		 1489/*
1490 * write as much of pending kvecs to the socket as we can.
1491 * 1 -> done
1492 * 0 -> socket full, but more to do
1493 * <0 -> error
1494 */
1495static int write_partial_kvec(struct ceph_connection *con)
1496{
1497 int ret;
1498
1499 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1500 while (con->out_kvec_bytes > 0) {
1501 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1502 con->out_kvec_left, con->out_kvec_bytes,
1503 con->out_more);
1504 if (ret <= 0)
1505 goto out;
1506 con->out_kvec_bytes -= ret;
1507 if (con->out_kvec_bytes == 0)
1508 break; /* done */
f42299e6
AE		 1509
1510 /* account for full iov entries consumed */
1511 while (ret >= con->out_kvec_cur->iov_len) {
1512 BUG_ON(!con->out_kvec_left);
1513 ret -= con->out_kvec_cur->iov_len;
1514 con->out_kvec_cur++;
1515 con->out_kvec_left--;
1516 }
1517 /* and for a partially-consumed entry */
1518 if (ret) {
1519 con->out_kvec_cur->iov_len -= ret;
1520 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1521 }
1522 }
1523 con->out_kvec_left = 0;
31b8006e
SW		 1524 ret = 1;
1525out:
1526 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1527 con->out_kvec_bytes, con->out_kvec_left, ret);
1528 return ret; /* done! */
1529}
1530
35b62808
AE		 1531static u32 ceph_crc32c_page(u32 crc, struct page *page,
1532 unsigned int page_offset,
1533 unsigned int length)
1534{
1535 char *kaddr;
1536
1537 kaddr = kmap(page);
1538 BUG_ON(kaddr == NULL);
1539 crc = crc32c(crc, kaddr + page_offset, length);
1540 kunmap(page);
1541
1542 return crc;
1543}
31b8006e
SW		 1544/*
1545 * Write as much message data payload as we can. If we finish, queue
1546 * up the footer.
1547 * 1 -> done, footer is now queued in out_kvec[].
1548 * 0 -> socket full, but more to do
1549 * <0 -> error
1550 */
34d2d200 1551static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1552{
1553 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1554 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 1555 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
f5db90bc 1556 u32 crc;
31b8006e 1557
859a35d5 1558 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1559
5240d9f9 1560 if (list_empty(&msg->data))
4c59b4a2
AE		 1561 return -EINVAL;
1562
5821bd8c
AE		 1563 /*
1564 * Iterate through each page that contains data to be
1565 * written, and send as much as possible for each.
1566 *
1567 * If we are calculating the data crc (the default), we will
1568 * need to map the page. If we have no pages, they have
1569 * been revoked, so use the zero page.
1570 */
f5db90bc 1571 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1572 while (cursor->resid) {
8a166d05 1573 struct page *page;
e387d525
AE		 1574 size_t page_offset;
1575 size_t length;
8a166d05 1576 bool last_piece;
8ea299bc 1577 bool need_crc;
f5db90bc 1578 int ret;
68b4476b 1579
343128ce
SB		 1580 page = ceph_msg_data_next(cursor, &page_offset, &length,
1581 &last_piece);
e387d525 1582 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
c2cfa194 1583 length, !last_piece);
f5db90bc
AE		 1584 if (ret <= 0) {
1585 if (do_datacrc)
1586 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1587
f5db90bc
AE		 1588 return ret;
1589 }
143334ff
AE		 1590 if (do_datacrc && cursor->need_crc)
1591 crc = ceph_crc32c_page(crc, page, page_offset, length);
343128ce 1592 need_crc = ceph_msg_data_advance(cursor, (size_t)ret);
31b8006e
SW		 1593 }
1594
34d2d200 1595 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1596
1597 /* prepare and queue up footer, too */
f5db90bc
AE		 1598 if (do_datacrc)
1599 msg->footer.data_crc = cpu_to_le32(crc);
1600 else
84ca8fc8 1601 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1602 con_out_kvec_reset(con);
859eb799 1603 prepare_write_message_footer(con);
f5db90bc
AE		 1604
1605 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1606}
1607
1608/*
1609 * write some zeros
1610 */
1611static int write_partial_skip(struct ceph_connection *con)
1612{
1613 int ret;
1614
67645d76 1615 dout("%s %p %d left\n", __func__, con, con->out_skip);
31b8006e 1616 while (con->out_skip > 0) {
31739139 1617 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1618
e1dcb128 1619 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1620 if (ret <= 0)
1621 goto out;
1622 con->out_skip -= ret;
1623 }
1624 ret = 1;
1625out:
1626 return ret;
1627}
1628
1629/*
1630 * Prepare to read connection handshake, or an ack.
1631 */
eed0ef2c
SW		 1632static void prepare_read_banner(struct ceph_connection *con)
1633{
1634 dout("prepare_read_banner %p\n", con);
1635 con->in_base_pos = 0;
1636}
1637
31b8006e
SW		 1638static void prepare_read_connect(struct ceph_connection *con)
1639{
1640 dout("prepare_read_connect %p\n", con);
1641 con->in_base_pos = 0;
1642}
1643
1644static void prepare_read_ack(struct ceph_connection *con)
1645{
1646 dout("prepare_read_ack %p\n", con);
1647 con->in_base_pos = 0;
1648}
1649
3a23083b
SW		 1650static void prepare_read_seq(struct ceph_connection *con)
1651{
1652 dout("prepare_read_seq %p\n", con);
1653 con->in_base_pos = 0;
1654 con->in_tag = CEPH_MSGR_TAG_SEQ;
1655}
1656
31b8006e
SW		 1657static void prepare_read_tag(struct ceph_connection *con)
1658{
1659 dout("prepare_read_tag %p\n", con);
1660 con->in_base_pos = 0;
1661 con->in_tag = CEPH_MSGR_TAG_READY;
1662}
1663
8b9558aa
YZ		 1664static void prepare_read_keepalive_ack(struct ceph_connection *con)
1665{
1666 dout("prepare_read_keepalive_ack %p\n", con);
1667 con->in_base_pos = 0;
1668}
1669
31b8006e
SW		 1670/*
1671 * Prepare to read a message.
1672 */
1673static int prepare_read_message(struct ceph_connection *con)
1674{
1675 dout("prepare_read_message %p\n", con);
1676 BUG_ON(con->in_msg != NULL);
1677 con->in_base_pos = 0;
1678 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1679 return 0;
1680}
1681
1682
1683static int read_partial(struct ceph_connection *con,
fd51653f 1684 int end, int size, void *object)
31b8006e 1685{
e6cee71f
AE		 1686 while (con->in_base_pos < end) {
1687 int left = end - con->in_base_pos;
31b8006e
SW		 1688 int have = size - left;
1689 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1690 if (ret <= 0)
1691 return ret;
1692 con->in_base_pos += ret;
1693 }
1694 return 1;
1695}
1696
1697
1698/*
1699 * Read all or part of the connect-side handshake on a new connection
1700 */
eed0ef2c 1701static int read_partial_banner(struct ceph_connection *con)
31b8006e 1702{
fd51653f
AE		 1703 int size;
1704 int end;
1705 int ret;
31b8006e 1706
eed0ef2c 1707 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1708
1709 /* peer's banner */
fd51653f
AE		 1710 size = strlen(CEPH_BANNER);
1711 end = size;
1712 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1713 if (ret <= 0)
1714 goto out;
fd51653f
AE		 1715
1716 size = sizeof (con->actual_peer_addr);
1717 end += size;
1718 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1719 if (ret <= 0)
1720 goto out;
fd51653f
AE		 1721
1722 size = sizeof (con->peer_addr_for_me);
1723 end += size;
1724 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1725 if (ret <= 0)
1726 goto out;
fd51653f 1727
eed0ef2c
SW		 1728out:
1729 return ret;
1730}
1731
1732static int read_partial_connect(struct ceph_connection *con)
1733{
fd51653f
AE		 1734 int size;
1735 int end;
1736 int ret;
eed0ef2c
SW		 1737
1738 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1739
fd51653f
AE		 1740 size = sizeof (con->in_reply);
1741 end = size;
1742 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1743 if (ret <= 0)
1744 goto out;
fd51653f
AE		 1745
1746 size = le32_to_cpu(con->in_reply.authorizer_len);
1747 end += size;
1748 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1749 if (ret <= 0)
1750 goto out;
31b8006e 1751
4e7a5dcd
SW		 1752 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1753 con, (int)con->in_reply.tag,
1754 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1755 le32_to_cpu(con->in_reply.global_seq));
1756out:
1757 return ret;
eed0ef2c 1758
31b8006e
SW		 1759}
1760
1761/*
1762 * Verify the hello banner looks okay.
1763 */
1764static int verify_hello(struct ceph_connection *con)
1765{
1766 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1767 pr_err("connect to %s got bad banner\n",
3d14c5d2 1768 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1769 con->error_msg = "protocol error, bad banner";
1770 return -1;
1771 }
1772 return 0;
1773}
1774
1775static bool addr_is_blank(struct sockaddr_storage *ss)
1776{
c44bd69c
ID		 1777 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1778 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1779
31b8006e
SW		 1780 switch (ss->ss_family) {
1781 case AF_INET:
c44bd69c 1782 return addr->s_addr == htonl(INADDR_ANY);
31b8006e 1783 case AF_INET6:
c44bd69c
ID		 1784 return ipv6_addr_any(addr6);
1785 default:
1786 return true;
31b8006e 1787 }
31b8006e
SW		 1788}
1789
1790static int addr_port(struct sockaddr_storage *ss)
1791{
1792 switch (ss->ss_family) {
1793 case AF_INET:
f28bcfbe 1794 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1795 case AF_INET6:
f28bcfbe 1796 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1797 }
1798 return 0;
1799}
1800
1801static void addr_set_port(struct sockaddr_storage *ss, int p)
1802{
1803 switch (ss->ss_family) {
1804 case AF_INET:
1805 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1806 break;
31b8006e
SW		 1807 case AF_INET6:
1808 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1809 break;
31b8006e
SW		 1810 }
1811}
1812
ee3b56f2
NW		 1813/*
1814 * Unlike other *_pton function semantics, zero indicates success.
1815 */
1816static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1817 char delim, const char **ipend)
1818{
99f0f3b2
AE		 1819 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1820 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1821
1822 memset(ss, 0, sizeof(*ss));
1823
1824 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1825 ss->ss_family = AF_INET;
1826 return 0;
1827 }
1828
1829 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1830 ss->ss_family = AF_INET6;
1831 return 0;
1832 }
1833
1834 return -EINVAL;
1835}
1836
1837/*
1838 * Extract hostname string and resolve using kernel DNS facility.
1839 */
1840#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1841static int ceph_dns_resolve_name(const char *name, size_t namelen,
1842 struct sockaddr_storage *ss, char delim, const char **ipend)
1843{
1844 const char *end, *delim_p;
1845 char *colon_p, *ip_addr = NULL;
1846 int ip_len, ret;
1847
1848 /*
1849 * The end of the hostname occurs immediately preceding the delimiter or
1850 * the port marker (':') where the delimiter takes precedence.
1851 */
1852 delim_p = memchr(name, delim, namelen);
1853 colon_p = memchr(name, ':', namelen);
1854
1855 if (delim_p && colon_p)
1856 end = delim_p < colon_p ? delim_p : colon_p;
1857 else if (!delim_p && colon_p)
1858 end = colon_p;
1859 else {
1860 end = delim_p;
1861 if (!end) /* case: hostname:/ */
1862 end = name + namelen;
1863 }
1864
1865 if (end <= name)
1866 return -EINVAL;
1867
1868 /* do dns_resolve upcall */
1869 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1870 if (ip_len > 0)
1871 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1872 else
1873 ret = -ESRCH;
1874
1875 kfree(ip_addr);
1876
1877 *ipend = end;
1878
1879 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1880 ret, ret ? "failed" : ceph_pr_addr(ss));
1881
1882 return ret;
1883}
1884#else
1885static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1886 struct sockaddr_storage *ss, char delim, const char **ipend)
1887{
1888 return -EINVAL;
1889}
1890#endif
1891
1892/*
1893 * Parse a server name (IP or hostname). If a valid IP address is not found
1894 * then try to extract a hostname to resolve using userspace DNS upcall.
1895 */
1896static int ceph_parse_server_name(const char *name, size_t namelen,
1897 struct sockaddr_storage *ss, char delim, const char **ipend)
1898{
1899 int ret;
1900
1901 ret = ceph_pton(name, namelen, ss, delim, ipend);
1902 if (ret)
1903 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1904
1905 return ret;
1906}
1907
31b8006e
SW		 1908/*
1909 * Parse an ip[:port] list into an addr array. Use the default
1910 * monitor port if a port isn't specified.
1911 */
1912int ceph_parse_ips(const char *c, const char *end,
1913 struct ceph_entity_addr *addr,
1914 int max_count, int *count)
1915{
ee3b56f2 1916 int i, ret = -EINVAL;
31b8006e
SW		 1917 const char *p = c;
1918
1919 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1920 for (i = 0; i < max_count; i++) {
1921 const char *ipend;
1922 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1923 int port;
39139f64
SW		 1924 char delim = ',';
1925
1926 if (*p == '[') {
1927 delim = ']';
1928 p++;
1929 }
31b8006e 1930
ee3b56f2
NW		 1931 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1932 if (ret)
31b8006e 1933 goto bad;
ee3b56f2
NW		 1934 ret = -EINVAL;
1935
31b8006e
SW		 1936 p = ipend;
1937
39139f64
SW		 1938 if (delim == ']') {
1939 if (*p != ']') {
1940 dout("missing matching ']'\n");
1941 goto bad;
1942 }
1943 p++;
1944 }
1945
31b8006e
SW		 1946 /* port? */
1947 if (p < end && *p == ':') {
1948 port = 0;
1949 p++;
1950 while (p < end && *p >= '0' && *p <= '9') {
1951 port = (port * 10) + (*p - '0');
1952 p++;
1953 }
f48db1e9
ID		 1954 if (port == 0)
1955 port = CEPH_MON_PORT;
1956 else if (port > 65535)
31b8006e
SW		 1957 goto bad;
1958 } else {
1959 port = CEPH_MON_PORT;
1960 }
1961
1962 addr_set_port(ss, port);
1963
3d14c5d2 1964 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1965
1966 if (p == end)
1967 break;
1968 if (*p != ',')
1969 goto bad;
1970 p++;
1971 }
1972
1973 if (p != end)
1974 goto bad;
1975
1976 if (count)
1977 *count = i + 1;
1978 return 0;
1979
1980bad:
39139f64 1981 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1982 return ret;
31b8006e 1983}
3d14c5d2 1984EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1985
eed0ef2c 1986static int process_banner(struct ceph_connection *con)
31b8006e 1987{
eed0ef2c 1988 dout("process_banner on %p\n", con);
31b8006e
SW		 1989
1990 if (verify_hello(con) < 0)
1991 return -1;
1992
63f2d211
SW		 1993 ceph_decode_addr(&con->actual_peer_addr);
1994 ceph_decode_addr(&con->peer_addr_for_me);
1995
31b8006e
SW		 1996 /*
1997 * Make sure the other end is who we wanted. note that the other
1998 * end may not yet know their ip address, so if it's 0.0.0.0, give
1999 * them the benefit of the doubt.
2000 */
103e2d3a
SW		 2001 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2002 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 2003 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2004 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 2005 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2006 ceph_pr_addr(&con->peer_addr.in_addr),
2007 (int)le32_to_cpu(con->peer_addr.nonce),
2008 ceph_pr_addr(&con->actual_peer_addr.in_addr),
2009 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 2010 con->error_msg = "wrong peer at address";
31b8006e
SW		 2011 return -1;
2012 }
2013
2014 /*
2015 * did we learn our address?
2016 */
2017 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2018 int port = addr_port(&con->msgr->inst.addr.in_addr);
2019
2020 memcpy(&con->msgr->inst.addr.in_addr,
2021 &con->peer_addr_for_me.in_addr,
2022 sizeof(con->peer_addr_for_me.in_addr));
2023 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 2024 encode_my_addr(con->msgr);
eed0ef2c 2025 dout("process_banner learned my addr is %s\n",
3d14c5d2 2026 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 2027 }
2028
eed0ef2c
SW		 2029 return 0;
2030}
2031
2032static int process_connect(struct ceph_connection *con)
2033{
859bff51
ID		 2034 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2035 u64 req_feat = from_msgr(con->msgr)->required_features;
2b3e0c90
ID		 2036 u64 server_feat = ceph_sanitize_features(
2037 le64_to_cpu(con->in_reply.features));
0da5d703 2038 int ret;
04a419f9 2039
eed0ef2c
SW		 2040 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2041
31b8006e 2042 switch (con->in_reply.tag) {
04a419f9
SW		 2043 case CEPH_MSGR_TAG_FEATURES:
2044 pr_err("%s%lld %s feature set mismatch,"
2045 " my %llx < server's %llx, missing %llx\n",
2046 ENTITY_NAME(con->peer_name),
3d14c5d2 2047 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2048 sup_feat, server_feat, server_feat & ~sup_feat);
2049 con->error_msg = "missing required protocol features";
0fa6ebc6 2050 reset_connection(con);
04a419f9
SW		 2051 return -1;
2052
31b8006e 2053 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2054 pr_err("%s%lld %s protocol version mismatch,"
2055 " my %d != server's %d\n",
2056 ENTITY_NAME(con->peer_name),
3d14c5d2 2057 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2058 le32_to_cpu(con->out_connect.protocol_version),
2059 le32_to_cpu(con->in_reply.protocol_version));
2060 con->error_msg = "protocol version mismatch";
0fa6ebc6 2061 reset_connection(con);
31b8006e
SW		 2062 return -1;
2063
4e7a5dcd
SW		 2064 case CEPH_MSGR_TAG_BADAUTHORIZER:
2065 con->auth_retry++;
2066 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2067 con->auth_retry);
2068 if (con->auth_retry == 2) {
2069 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2070 return -1;
2071 }
6d4221b5 2072 con_out_kvec_reset(con);
e825a66d 2073 ret = prepare_write_connect(con);
0da5d703
SW		 2074 if (ret < 0)
2075 return ret;
63733a0f 2076 prepare_read_connect(con);
4e7a5dcd 2077 break;
31b8006e
SW		 2078
2079 case CEPH_MSGR_TAG_RESETSESSION:
2080 /*
2081 * If we connected with a large connect_seq but the peer
2082 * has no record of a session with us (no connection, or
2083 * connect_seq == 0), they will send RESETSESION to indicate
2084 * that they must have reset their session, and may have
2085 * dropped messages.
2086 */
2087 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2088 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2089 pr_err("%s%lld %s connection reset\n",
2090 ENTITY_NAME(con->peer_name),
3d14c5d2 2091 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2092 reset_connection(con);
6d4221b5 2093 con_out_kvec_reset(con);
5a0f8fdd
AE		 2094 ret = prepare_write_connect(con);
2095 if (ret < 0)
2096 return ret;
31b8006e
SW		 2097 prepare_read_connect(con);
2098
2099 /* Tell ceph about it. */
ec302645 2100 mutex_unlock(&con->mutex);
31b8006e
SW		 2101 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2102 if (con->ops->peer_reset)
2103 con->ops->peer_reset(con);
ec302645 2104 mutex_lock(&con->mutex);
8dacc7da 2105 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2106 return -EAGAIN;
31b8006e
SW		 2107 break;
2108
2109 case CEPH_MSGR_TAG_RETRY_SESSION:
2110 /*
2111 * If we sent a smaller connect_seq than the peer has, try
2112 * again with a larger value.
2113 */
5bdca4e0 2114 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2115 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2116 le32_to_cpu(con->in_reply.connect_seq));
2117 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2118 con_out_kvec_reset(con);
5a0f8fdd
AE		 2119 ret = prepare_write_connect(con);
2120 if (ret < 0)
2121 return ret;
31b8006e
SW		 2122 prepare_read_connect(con);
2123 break;
2124
2125 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2126 /*
2127 * If we sent a smaller global_seq than the peer has, try
2128 * again with a larger value.
2129 */
eed0ef2c 2130 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2131 con->peer_global_seq,
5bdca4e0 2132 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2133 get_global_seq(con->msgr,
5bdca4e0 2134 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2135 con_out_kvec_reset(con);
5a0f8fdd
AE		 2136 ret = prepare_write_connect(con);
2137 if (ret < 0)
2138 return ret;
31b8006e
SW		 2139 prepare_read_connect(con);
2140 break;
2141
3a23083b 2142 case CEPH_MSGR_TAG_SEQ:
31b8006e 2143 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2144 if (req_feat & ~server_feat) {
2145 pr_err("%s%lld %s protocol feature mismatch,"
2146 " my required %llx > server's %llx, need %llx\n",
2147 ENTITY_NAME(con->peer_name),
3d14c5d2 2148 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2149 req_feat, server_feat, req_feat & ~server_feat);
2150 con->error_msg = "missing required protocol features";
0fa6ebc6 2151 reset_connection(con);
04a419f9
SW		 2152 return -1;
2153 }
8dacc7da 2154
122070a2 2155 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2156 con->state = CON_STATE_OPEN;
20e55c4c 2157 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2158 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2159 con->connect_seq++;
aba558e2 2160 con->peer_features = server_feat;
31b8006e
SW		 2161 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2162 con->peer_global_seq,
2163 le32_to_cpu(con->in_reply.connect_seq),
2164 con->connect_seq);
2165 WARN_ON(con->connect_seq !=
2166 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2167
2168 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2169 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2170
85effe18 2171 con->delay = 0; /* reset backoff memory */
92ac41d0 2172
3a23083b
SW		 2173 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2174 prepare_write_seq(con);
2175 prepare_read_seq(con);
2176 } else {
2177 prepare_read_tag(con);
2178 }
31b8006e
SW		 2179 break;
2180
2181 case CEPH_MSGR_TAG_WAIT:
2182 /*
2183 * If there is a connection race (we are opening
2184 * connections to each other), one of us may just have
2185 * to WAIT. This shouldn't happen if we are the
2186 * client.
2187 */
04177882
SW		 2188 con->error_msg = "protocol error, got WAIT as client";
2189 return -1;
31b8006e
SW		 2190
2191 default:
31b8006e
SW		 2192 con->error_msg = "protocol error, garbage tag during connect";
2193 return -1;
2194 }
2195 return 0;
2196}
2197
2198
2199/*
2200 * read (part of) an ack
2201 */
2202static int read_partial_ack(struct ceph_connection *con)
2203{
fd51653f
AE		 2204 int size = sizeof (con->in_temp_ack);
2205 int end = size;
31b8006e 2206
fd51653f 2207 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2208}
2209
31b8006e
SW		 2210/*
2211 * We can finally discard anything that's been acked.
2212 */
2213static void process_ack(struct ceph_connection *con)
2214{
2215 struct ceph_msg *m;
2216 u64 ack = le64_to_cpu(con->in_temp_ack);
2217 u64 seq;
2218
31b8006e
SW		 2219 while (!list_empty(&con->out_sent)) {
2220 m = list_first_entry(&con->out_sent, struct ceph_msg,
2221 list_head);
2222 seq = le64_to_cpu(m->hdr.seq);
2223 if (seq > ack)
2224 break;
2225 dout("got ack for seq %llu type %d at %p\n", seq,
2226 le16_to_cpu(m->hdr.type), m);
4cf9d544 2227 m->ack_stamp = jiffies;
31b8006e
SW		 2228 ceph_msg_remove(m);
2229 }
31b8006e
SW		 2230 prepare_read_tag(con);
2231}
2232
2233
2450418c 2234static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2235 struct kvec *section,
2236 unsigned int sec_len, u32 *crc)
2450418c 2237{
68b4476b 2238 int ret, left;
2450418c
YS		 2239
2240 BUG_ON(!section);
2241
2242 while (section->iov_len < sec_len) {
2243 BUG_ON(section->iov_base == NULL);
2244 left = sec_len - section->iov_len;
2245 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2246 section->iov_len, left);
2247 if (ret <= 0)
2248 return ret;
2249 section->iov_len += ret;
2450418c 2250 }
fe3ad593
AE		 2251 if (section->iov_len == sec_len)
2252 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2253
2450418c
YS		 2254 return 1;
2255}
31b8006e 2256
34d2d200
AE		 2257static int read_partial_msg_data(struct ceph_connection *con)
2258{
2259 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2260 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 2261 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
686be208
AE		 2262 struct page *page;
2263 size_t page_offset;
2264 size_t length;
f5db90bc 2265 u32 crc = 0;
34d2d200
AE		 2266 int ret;
2267
2268 BUG_ON(!msg);
5240d9f9 2269 if (list_empty(&msg->data))
4c59b4a2 2270 return -EIO;
34d2d200 2271
f5db90bc
AE		 2272 if (do_datacrc)
2273 crc = con->in_data_crc;
643c68a4 2274 while (cursor->resid) {
343128ce 2275 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
686be208 2276 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2277 if (ret <= 0) {
2278 if (do_datacrc)
2279 con->in_data_crc = crc;
2280
686be208 2281 return ret;
f5db90bc 2282 }
686be208
AE		 2283
2284 if (do_datacrc)
f5db90bc 2285 crc = ceph_crc32c_page(crc, page, page_offset, ret);
343128ce 2286 (void) ceph_msg_data_advance(cursor, (size_t)ret);
34d2d200 2287 }
f5db90bc
AE		 2288 if (do_datacrc)
2289 con->in_data_crc = crc;
34d2d200
AE		 2290
2291 return 1; /* must return > 0 to indicate success */
2292}
2293
31b8006e
SW		 2294/*
2295 * read (part of) a message.
2296 */
686be208
AE		 2297static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2298
31b8006e
SW		 2299static int read_partial_message(struct ceph_connection *con)
2300{
2301 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2302 int size;
2303 int end;
31b8006e 2304 int ret;
95c96174 2305 unsigned int front_len, middle_len, data_len;
859bff51 2306 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
33d07337 2307 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
ae18756b 2308 u64 seq;
fe3ad593 2309 u32 crc;
31b8006e
SW		 2310
2311 dout("read_partial_message con %p msg %p\n", con, m);
2312
2313 /* header */
fd51653f
AE		 2314 size = sizeof (con->in_hdr);
2315 end = size;
2316 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2317 if (ret <= 0)
2318 return ret;
fe3ad593
AE		 2319
2320 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2321 if (cpu_to_le32(crc) != con->in_hdr.crc) {
67c64eb7 2322 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
fe3ad593
AE		 2323 crc, con->in_hdr.crc);
2324 return -EBADMSG;
2325 }
2326
31b8006e
SW		 2327 front_len = le32_to_cpu(con->in_hdr.front_len);
2328 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2329 return -EIO;
2330 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2331 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2332 return -EIO;
2333 data_len = le32_to_cpu(con->in_hdr.data_len);
2334 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2335 return -EIO;
2336
ae18756b
SW		 2337 /* verify seq# */
2338 seq = le64_to_cpu(con->in_hdr.seq);
2339 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2340 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2341 ENTITY_NAME(con->peer_name),
3d14c5d2 2342 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2343 seq, con->in_seq + 1);
2344 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2345 sizeof_footer(con);
ae18756b 2346 con->in_tag = CEPH_MSGR_TAG_READY;
e7a88e82 2347 return 1;
ae18756b
SW		 2348 } else if ((s64)seq - (s64)con->in_seq > 1) {
2349 pr_err("read_partial_message bad seq %lld expected %lld\n",
2350 seq, con->in_seq + 1);
2351 con->error_msg = "bad message sequence # for incoming message";
67c64eb7 2352 return -EBADE;
ae18756b
SW		 2353 }
2354
31b8006e
SW		 2355 /* allocate message? */
2356 if (!con->in_msg) {
4740a623
SW		 2357 int skip = 0;
2358
31b8006e 2359 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2360 front_len, data_len);
4740a623
SW		 2361 ret = ceph_con_in_msg_alloc(con, &skip);
2362 if (ret < 0)
2363 return ret;
f759ebb9
AE		 2364
2365 BUG_ON(!con->in_msg ^ skip);
2450418c 2366 if (skip) {
31b8006e 2367 /* skip this message */
a79832f2 2368 dout("alloc_msg said skip message\n");
31b8006e 2369 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2370 sizeof_footer(con);
31b8006e 2371 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2372 con->in_seq++;
e7a88e82 2373 return 1;
31b8006e 2374 }
38941f80 2375
4740a623 2376 BUG_ON(!con->in_msg);
38941f80 2377 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2378 m = con->in_msg;
2379 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2380 if (m->middle)
2381 m->middle->vec.iov_len = 0;
9d7f0f13 2382
78625051 2383 /* prepare for data payload, if any */
a4107026 2384
78625051 2385 if (data_len)
98fa5dd8 2386 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2387 }
2388
2389 /* front */
2450418c
YS		 2390 ret = read_partial_message_section(con, &m->front, front_len,
2391 &con->in_front_crc);
2392 if (ret <= 0)
2393 return ret;
31b8006e
SW		 2394
2395 /* middle */
2450418c 2396 if (m->middle) {
213c99ee
SW		 2397 ret = read_partial_message_section(con, &m->middle->vec,
2398 middle_len,
2450418c 2399 &con->in_middle_crc);
31b8006e
SW		 2400 if (ret <= 0)
2401 return ret;
31b8006e
SW		 2402 }
2403
2404 /* (page) data */
34d2d200
AE		 2405 if (data_len) {
2406 ret = read_partial_msg_data(con);
2407 if (ret <= 0)
2408 return ret;
31b8006e
SW		 2409 }
2410
31b8006e 2411 /* footer */
33d07337
YZ		 2412 if (need_sign)
2413 size = sizeof(m->footer);
2414 else
2415 size = sizeof(m->old_footer);
2416
fd51653f
AE		 2417 end += size;
2418 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2419 if (ret <= 0)
2420 return ret;
2421
33d07337
YZ		 2422 if (!need_sign) {
2423 m->footer.flags = m->old_footer.flags;
2424 m->footer.sig = 0;
2425 }
2426
31b8006e
SW		 2427 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2428 m, front_len, m->footer.front_crc, middle_len,
2429 m->footer.middle_crc, data_len, m->footer.data_crc);
2430
2431 /* crc ok? */
2432 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2433 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2434 m, con->in_front_crc, m->footer.front_crc);
2435 return -EBADMSG;
2436 }
2437 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2438 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2439 m, con->in_middle_crc, m->footer.middle_crc);
2440 return -EBADMSG;
2441 }
bca064d2 2442 if (do_datacrc &&
31b8006e
SW		 2443 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2444 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2445 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2446 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2447 return -EBADMSG;
2448 }
2449
33d07337 2450 if (need_sign && con->ops->check_message_signature &&
79dbd1ba 2451 con->ops->check_message_signature(m)) {
33d07337
YZ		 2452 pr_err("read_partial_message %p signature check failed\n", m);
2453 return -EBADMSG;
2454 }
2455
31b8006e
SW		 2456 return 1; /* done! */
2457}
2458
2459/*
2460 * Process message. This happens in the worker thread. The callback should
2461 * be careful not to do anything that waits on other incoming messages or it
2462 * may deadlock.
2463 */
2464static void process_message(struct ceph_connection *con)
2465{
583d0fef 2466 struct ceph_msg *msg = con->in_msg;
31b8006e 2467
38941f80 2468 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2469 con->in_msg = NULL;
2470
2471 /* if first message, set peer_name */
2472 if (con->peer_name.type == 0)
dbad185d 2473 con->peer_name = msg->hdr.src;
31b8006e 2474
31b8006e 2475 con->in_seq++;
ec302645 2476 mutex_unlock(&con->mutex);
31b8006e
SW		 2477
2478 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2479 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2480 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2481 le16_to_cpu(msg->hdr.type),
2482 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2483 le32_to_cpu(msg->hdr.front_len),
2484 le32_to_cpu(msg->hdr.data_len),
2485 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2486 con->ops->dispatch(con, msg);
ec302645
SW		 2487
2488 mutex_lock(&con->mutex);
31b8006e
SW		 2489}
2490
8b9558aa
YZ		 2491static int read_keepalive_ack(struct ceph_connection *con)
2492{
2493 struct ceph_timespec ceph_ts;
2494 size_t size = sizeof(ceph_ts);
2495 int ret = read_partial(con, size, size, &ceph_ts);
2496 if (ret <= 0)
2497 return ret;
2498 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2499 prepare_read_tag(con);
2500 return 1;
2501}
31b8006e
SW		 2502
2503/*
2504 * Write something to the socket. Called in a worker thread when the
2505 * socket appears to be writeable and we have something ready to send.
2506 */
2507static int try_write(struct ceph_connection *con)
2508{
31b8006e
SW		 2509 int ret = 1;
2510
d59315ca 2511 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2512
31b8006e
SW		 2513more:
2514 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2515
2516 /* open the socket first? */
8dacc7da
SW		 2517 if (con->state == CON_STATE_PREOPEN) {
2518 BUG_ON(con->sock);
2519 con->state = CON_STATE_CONNECTING;
a5988c49 2520
e2200423 2521 con_out_kvec_reset(con);
e825a66d 2522 prepare_write_banner(con);
eed0ef2c 2523 prepare_read_banner(con);
31b8006e 2524
cf3e5c40 2525 BUG_ON(con->in_msg);
31b8006e
SW		 2526 con->in_tag = CEPH_MSGR_TAG_READY;
2527 dout("try_write initiating connect on %p new state %lu\n",
2528 con, con->state);
41617d0c
AE		 2529 ret = ceph_tcp_connect(con);
2530 if (ret < 0) {
31b8006e 2531 con->error_msg = "connect error";
31b8006e
SW		 2532 goto out;
2533 }
2534 }
2535
2536more_kvec:
2537 /* kvec data queued? */
67645d76
ID		 2538 if (con->out_kvec_left) {
2539 ret = write_partial_kvec(con);
31b8006e 2540 if (ret <= 0)
42961d23 2541 goto out;
31b8006e 2542 }
67645d76
ID		 2543 if (con->out_skip) {
2544 ret = write_partial_skip(con);
31b8006e 2545 if (ret <= 0)
42961d23 2546 goto out;
31b8006e
SW		 2547 }
2548
2549 /* msg pages? */
2550 if (con->out_msg) {
c86a2930
SW		 2551 if (con->out_msg_done) {
2552 ceph_msg_put(con->out_msg);
2553 con->out_msg = NULL; /* we're done with this one */
2554 goto do_next;
2555 }
2556
34d2d200 2557 ret = write_partial_message_data(con);
31b8006e
SW		 2558 if (ret == 1)
2559 goto more_kvec; /* we need to send the footer, too! */
2560 if (ret == 0)
42961d23 2561 goto out;
31b8006e 2562 if (ret < 0) {
34d2d200 2563 dout("try_write write_partial_message_data err %d\n",
31b8006e 2564 ret);
42961d23 2565 goto out;
31b8006e
SW		 2566 }
2567 }
2568
c86a2930 2569do_next:
8dacc7da 2570 if (con->state == CON_STATE_OPEN) {
8b9558aa
YZ		 2571 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2572 prepare_write_keepalive(con);
2573 goto more;
2574 }
31b8006e
SW		 2575 /* is anything else pending? */
2576 if (!list_empty(&con->out_queue)) {
2577 prepare_write_message(con);
2578 goto more;
2579 }
2580 if (con->in_seq > con->in_seq_acked) {
2581 prepare_write_ack(con);
2582 goto more;
2583 }
31b8006e
SW		 2584 }
2585
2586 /* Nothing to do! */
c9ffc77a 2587 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2588 dout("try_write nothing else to write.\n");
31b8006e
SW		 2589 ret = 0;
2590out:
42961d23 2591 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2592 return ret;
2593}
2594
2595
2596
2597/*
2598 * Read what we can from the socket.
2599 */
2600static int try_read(struct ceph_connection *con)
2601{
31b8006e
SW		 2602 int ret = -1;
2603
8dacc7da
SW		 2604more:
2605 dout("try_read start on %p state %lu\n", con, con->state);
2606 if (con->state != CON_STATE_CONNECTING &&
2607 con->state != CON_STATE_NEGOTIATING &&
2608 con->state != CON_STATE_OPEN)
31b8006e
SW		 2609 return 0;
2610
8dacc7da 2611 BUG_ON(!con->sock);
ec302645 2612
31b8006e
SW		 2613 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2614 con->in_base_pos);
0da5d703 2615
8dacc7da 2616 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2617 dout("try_read connecting\n");
2618 ret = read_partial_banner(con);
2619 if (ret <= 0)
ab166d5a 2620 goto out;
7593af92
AE		 2621 ret = process_banner(con);
2622 if (ret < 0)
2623 goto out;
2624
8dacc7da 2625 con->state = CON_STATE_NEGOTIATING;
7593af92 2626
6d4221b5
JS		 2627 /*
2628 * Received banner is good, exchange connection info.
2629 * Do not reset out_kvec, as sending our banner raced
2630 * with receiving peer banner after connect completed.
2631 */
7593af92
AE		 2632 ret = prepare_write_connect(con);
2633 if (ret < 0)
2634 goto out;
2635 prepare_read_connect(con);
2636
2637 /* Send connection info before awaiting response */
0da5d703
SW		 2638 goto out;
2639 }
2640
8dacc7da 2641 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2642 dout("try_read negotiating\n");
31b8006e
SW		 2643 ret = read_partial_connect(con);
2644 if (ret <= 0)
31b8006e 2645 goto out;
98bdb0aa
SW		 2646 ret = process_connect(con);
2647 if (ret < 0)
2648 goto out;
31b8006e
SW		 2649 goto more;
2650 }
2651
122070a2 2652 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2653
31b8006e
SW		 2654 if (con->in_base_pos < 0) {
2655 /*
2656 * skipping + discarding content.
2657 *
2658 * FIXME: there must be a better way to do this!
2659 */
84495f49
AE		 2660 static char buf[SKIP_BUF_SIZE];
2661 int skip = min((int) sizeof (buf), -con->in_base_pos);
2662
31b8006e
SW		 2663 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2664 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2665 if (ret <= 0)
98bdb0aa 2666 goto out;
31b8006e
SW		 2667 con->in_base_pos += ret;
2668 if (con->in_base_pos)
2669 goto more;
2670 }
2671 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2672 /*
2673 * what's next?
2674 */
2675 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2676 if (ret <= 0)
98bdb0aa 2677 goto out;
31b8006e
SW		 2678 dout("try_read got tag %d\n", (int)con->in_tag);
2679 switch (con->in_tag) {
2680 case CEPH_MSGR_TAG_MSG:
2681 prepare_read_message(con);
2682 break;
2683 case CEPH_MSGR_TAG_ACK:
2684 prepare_read_ack(con);
2685 break;
8b9558aa
YZ		 2686 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2687 prepare_read_keepalive_ack(con);
2688 break;
31b8006e 2689 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2690 con_close_socket(con);
2691 con->state = CON_STATE_CLOSED;
98bdb0aa 2692 goto out;
31b8006e
SW		 2693 default:
2694 goto bad_tag;
2695 }
2696 }
2697 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2698 ret = read_partial_message(con);
2699 if (ret <= 0) {
2700 switch (ret) {
2701 case -EBADMSG:
a51983e4 2702 con->error_msg = "bad crc/signature";
67c64eb7
ID		 2703 /* fall through */
2704 case -EBADE:
31b8006e 2705 ret = -EIO;
98bdb0aa 2706 break;
31b8006e
SW		 2707 case -EIO:
2708 con->error_msg = "io error";
98bdb0aa 2709 break;
31b8006e 2710 }
98bdb0aa 2711 goto out;
31b8006e
SW		 2712 }
2713 if (con->in_tag == CEPH_MSGR_TAG_READY)
2714 goto more;
2715 process_message(con);
7b862e07
SW		 2716 if (con->state == CON_STATE_OPEN)
2717 prepare_read_tag(con);
31b8006e
SW		 2718 goto more;
2719 }
3a23083b
SW		 2720 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2721 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2722 /*
2723 * the final handshake seq exchange is semantically
2724 * equivalent to an ACK
2725 */
31b8006e
SW		 2726 ret = read_partial_ack(con);
2727 if (ret <= 0)
98bdb0aa 2728 goto out;
31b8006e
SW		 2729 process_ack(con);
2730 goto more;
2731 }
8b9558aa
YZ		 2732 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2733 ret = read_keepalive_ack(con);
2734 if (ret <= 0)
2735 goto out;
2736 goto more;
2737 }
31b8006e 2738
31b8006e 2739out:
98bdb0aa 2740 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2741 return ret;
2742
2743bad_tag:
2744 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2745 con->error_msg = "protocol error, garbage tag";
2746 ret = -1;
2747 goto out;
2748}
2749
2750
2751/*
802c6d96
AE		 2752 * Atomically queue work on a connection after the specified delay.
2753 * Bump @con reference to avoid races with connection teardown.
2754 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2755 */
802c6d96 2756static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2757{
31b8006e 2758 if (!con->ops->get(con)) {
802c6d96 2759 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2760 return -ENOENT;
31b8006e
SW		 2761 }
2762
802c6d96
AE		 2763 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2764 dout("%s %p - already queued\n", __func__, con);
31b8006e 2765 con->ops->put(con);
802c6d96 2766 return -EBUSY;
31b8006e 2767 }
802c6d96
AE		 2768
2769 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2770 return 0;
2771}
2772
2773static void queue_con(struct ceph_connection *con)
2774{
2775 (void) queue_con_delay(con, 0);
31b8006e
SW		 2776}
2777
37ab77ac
ID		 2778static void cancel_con(struct ceph_connection *con)
2779{
2780 if (cancel_delayed_work(&con->work)) {
2781 dout("%s %p\n", __func__, con);
2782 con->ops->put(con);
2783 }
2784}
2785
7bb21d68
AE		 2786static bool con_sock_closed(struct ceph_connection *con)
2787{
c9ffc77a 2788 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2789 return false;
2790
2791#define CASE(x) \
2792 case CON_STATE_ ## x: \
2793 con->error_msg = "socket closed (con state " #x ")"; \
2794 break;
2795
2796 switch (con->state) {
2797 CASE(CLOSED);
2798 CASE(PREOPEN);
2799 CASE(CONNECTING);
2800 CASE(NEGOTIATING);
2801 CASE(OPEN);
2802 CASE(STANDBY);
2803 default:
b9a67899 2804 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2805 __func__, con, con->state);
2806 con->error_msg = "unrecognized con state";
2807 BUG();
2808 break;
2809 }
2810#undef CASE
2811
2812 return true;
2813}
2814
f20a39fd
AE		 2815static bool con_backoff(struct ceph_connection *con)
2816{
2817 int ret;
2818
2819 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2820 return false;
2821
2822 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2823 if (ret) {
2824 dout("%s: con %p FAILED to back off %lu\n", __func__,
2825 con, con->delay);
2826 BUG_ON(ret == -ENOENT);
2827 con_flag_set(con, CON_FLAG_BACKOFF);
2828 }
2829
2830 return true;
2831}
2832
93209264
AE		 2833/* Finish fault handling; con->mutex must *not* be held here */
2834
2835static void con_fault_finish(struct ceph_connection *con)
2836{
f6330cc1
ID		 2837 dout("%s %p\n", __func__, con);
2838
93209264
AE		 2839 /*
2840 * in case we faulted due to authentication, invalidate our
2841 * current tickets so that we can get new ones.
2842 */
f6330cc1
ID		 2843 if (con->auth_retry) {
2844 dout("auth_retry %d, invalidating\n", con->auth_retry);
2845 if (con->ops->invalidate_authorizer)
2846 con->ops->invalidate_authorizer(con);
2847 con->auth_retry = 0;
93209264
AE		 2848 }
2849
2850 if (con->ops->fault)
2851 con->ops->fault(con);
2852}
2853
31b8006e
SW		 2854/*
2855 * Do some work on a connection. Drop a connection ref when we're done.
2856 */
68931622 2857static void ceph_con_workfn(struct work_struct *work)
31b8006e
SW		 2858{
2859 struct ceph_connection *con = container_of(work, struct ceph_connection,
2860 work.work);
49659416 2861 bool fault;
31b8006e 2862
9dd4658d 2863 mutex_lock(&con->mutex);
49659416
AE		 2864 while (true) {
2865 int ret;
31b8006e 2866
49659416
AE		 2867 if ((fault = con_sock_closed(con))) {
2868 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2869 break;
2870 }
2871 if (con_backoff(con)) {
2872 dout("%s: con %p BACKOFF\n", __func__, con);
2873 break;
2874 }
2875 if (con->state == CON_STATE_STANDBY) {
2876 dout("%s: con %p STANDBY\n", __func__, con);
2877 break;
2878 }
2879 if (con->state == CON_STATE_CLOSED) {
2880 dout("%s: con %p CLOSED\n", __func__, con);
2881 BUG_ON(con->sock);
2882 break;
2883 }
2884 if (con->state == CON_STATE_PREOPEN) {
2885 dout("%s: con %p PREOPEN\n", __func__, con);
2886 BUG_ON(con->sock);
2887 }
0da5d703 2888
49659416
AE		 2889 ret = try_read(con);
2890 if (ret < 0) {
2891 if (ret == -EAGAIN)
2892 continue;
67c64eb7
ID		 2893 if (!con->error_msg)
2894 con->error_msg = "socket error on read";
49659416
AE		 2895 fault = true;
2896 break;
2897 }
2898
2899 ret = try_write(con);
2900 if (ret < 0) {
2901 if (ret == -EAGAIN)
2902 continue;
67c64eb7
ID		 2903 if (!con->error_msg)
2904 con->error_msg = "socket error on write";
49659416
AE		 2905 fault = true;
2906 }
2907
2908 break; /* If we make it to here, we're done */
3a140a0d 2909 }
b6e7b6a1
AE		 2910 if (fault)
2911 con_fault(con);
9dd4658d 2912 mutex_unlock(&con->mutex);
0da5d703 2913
b6e7b6a1
AE		 2914 if (fault)
2915 con_fault_finish(con);
2916
2917 con->ops->put(con);
31b8006e
SW		 2918}
2919
31b8006e
SW		 2920/*
2921 * Generic error/fault handler. A retry mechanism is used with
2922 * exponential backoff
2923 */
93209264 2924static void con_fault(struct ceph_connection *con)
31b8006e 2925{
31b8006e 2926 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2927 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2928
67c64eb7
ID		 2929 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2930 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2931 con->error_msg = NULL;
2932
122070a2 2933 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2934 con->state != CON_STATE_NEGOTIATING &&
2935 con->state != CON_STATE_OPEN);
ec302645 2936
31b8006e 2937 con_close_socket(con);
5e095e8b 2938
c9ffc77a 2939 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2940 dout("fault on LOSSYTX channel, marking CLOSED\n");
2941 con->state = CON_STATE_CLOSED;
93209264 2942 return;
3b5ede07
SW		 2943 }
2944
5e095e8b 2945 if (con->in_msg) {
38941f80 2946 BUG_ON(con->in_msg->con != con);
5e095e8b
SW		 2947 ceph_msg_put(con->in_msg);
2948 con->in_msg = NULL;
2949 }
31b8006e 2950
e80a52d1
SW		 2951 /* Requeue anything that hasn't been acked */
2952 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2953
e76661d0
SW		 2954 /* If there are no messages queued or keepalive pending, place
2955 * the connection in a STANDBY state */
2956 if (list_empty(&con->out_queue) &&
c9ffc77a 2957 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2958 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2959 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2960 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2961 } else {
2962 /* retry after a delay. */
8dacc7da 2963 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2964 if (con->delay == 0)
2965 con->delay = BASE_DELAY_INTERVAL;
2966 else if (con->delay < MAX_DELAY_INTERVAL)
2967 con->delay *= 2;
c9ffc77a 2968 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2969 queue_con(con);
31b8006e 2970 }
31b8006e
SW		 2971}
2972
2973
2974
2975/*
15d9882c 2976 * initialize a new messenger instance
31b8006e 2977 */
15d9882c 2978void ceph_messenger_init(struct ceph_messenger *msgr,
859bff51 2979 struct ceph_entity_addr *myaddr)
31b8006e 2980{
31b8006e
SW		 2981 spin_lock_init(&msgr->global_seq_lock);
2982
31b8006e
SW		 2983 if (myaddr)
2984 msgr->inst.addr = *myaddr;
2985
2986 /* select a random nonce */
ac8839d7 2987 msgr->inst.addr.type = 0;
103e2d3a 2988 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2989 encode_my_addr(msgr);
31b8006e 2990
a2a32584 2991 atomic_set(&msgr->stopping, 0);
757856d2 2992 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
31b8006e 2993
15d9882c 2994 dout("%s %p\n", __func__, msgr);
31b8006e 2995}
15d9882c 2996EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2997
757856d2
ID		 2998void ceph_messenger_fini(struct ceph_messenger *msgr)
2999{
3000 put_net(read_pnet(&msgr->net));
3001}
3002EXPORT_SYMBOL(ceph_messenger_fini);
3003
583d0fef
ID		 3004static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3005{
3006 if (msg->con)
3007 msg->con->ops->put(msg->con);
3008
3009 msg->con = con ? con->ops->get(con) : NULL;
3010 BUG_ON(msg->con != con);
3011}
3012
e00de341
SW		 3013static void clear_standby(struct ceph_connection *con)
3014{
3015 /* come back from STANDBY? */
8dacc7da 3016 if (con->state == CON_STATE_STANDBY) {
e00de341 3017 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 3018 con->state = CON_STATE_PREOPEN;
e00de341 3019 con->connect_seq++;
c9ffc77a
AE		 3020 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3021 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 3022 }
3023}
3024
31b8006e
SW		 3025/*
3026 * Queue up an outgoing message on the given connection.
3027 */
3028void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3029{
31b8006e 3030 /* set src+dst */
dbad185d 3031 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 3032 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 3033 msg->needs_out_seq = true;
3034
ec302645 3035 mutex_lock(&con->mutex);
92ce034b 3036
8dacc7da 3037 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 3038 dout("con_send %p closed, dropping %p\n", con, msg);
3039 ceph_msg_put(msg);
3040 mutex_unlock(&con->mutex);
3041 return;
3042 }
3043
583d0fef 3044 msg_con_set(msg, con);
92ce034b 3045
31b8006e
SW		 3046 BUG_ON(!list_empty(&msg->list_head));
3047 list_add_tail(&msg->list_head, &con->out_queue);
3048 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3049 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3050 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3051 le32_to_cpu(msg->hdr.front_len),
3052 le32_to_cpu(msg->hdr.middle_len),
3053 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 3054
3055 clear_standby(con);
ec302645 3056 mutex_unlock(&con->mutex);
31b8006e
SW		 3057
3058 /* if there wasn't anything waiting to send before, queue
3059 * new work */
c9ffc77a 3060 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3061 queue_con(con);
3062}
3d14c5d2 3063EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3064
3065/*
3066 * Revoke a message that was previously queued for send
3067 */
6740a845 3068void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3069{
6740a845
AE		 3070 struct ceph_connection *con = msg->con;
3071
583d0fef
ID		 3072 if (!con) {
3073 dout("%s msg %p null con\n", __func__, msg);
6740a845 3074 return; /* Message not in our possession */
583d0fef 3075 }
6740a845 3076
ec302645 3077 mutex_lock(&con->mutex);
31b8006e 3078 if (!list_empty(&msg->list_head)) {
38941f80 3079 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3080 list_del_init(&msg->list_head);
31b8006e 3081 msg->hdr.seq = 0;
38941f80 3082
31b8006e 3083 ceph_msg_put(msg);
ed98adad
SW		 3084 }
3085 if (con->out_msg == msg) {
67645d76
ID		 3086 BUG_ON(con->out_skip);
3087 /* footer */
3088 if (con->out_msg_done) {
3089 con->out_skip += con_out_kvec_skip(con);
3090 } else {
3091 BUG_ON(!msg->data_length);
3092 if (con->peer_features & CEPH_FEATURE_MSG_AUTH)
3093 con->out_skip += sizeof(msg->footer);
3094 else
3095 con->out_skip += sizeof(msg->old_footer);
31b8006e 3096 }
67645d76
ID		 3097 /* data, middle, front */
3098 if (msg->data_length)
3099 con->out_skip += msg->cursor.total_resid;
3100 if (msg->middle)
3101 con->out_skip += con_out_kvec_skip(con);
3102 con->out_skip += con_out_kvec_skip(con);
3103
3104 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3105 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
ed98adad 3106 msg->hdr.seq = 0;
67645d76 3107 con->out_msg = NULL;
92ce034b 3108 ceph_msg_put(msg);
31b8006e 3109 }
67645d76 3110
ec302645 3111 mutex_unlock(&con->mutex);
31b8006e
SW		 3112}
3113
350b1c32 3114/*
0d59ab81 3115 * Revoke a message that we may be reading data into
350b1c32 3116 */
8921d114 3117void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3118{
583d0fef 3119 struct ceph_connection *con = msg->con;
8921d114 3120
583d0fef 3121 if (!con) {
8921d114 3122 dout("%s msg %p null con\n", __func__, msg);
8921d114
AE		 3123 return; /* Message not in our possession */
3124 }
3125
350b1c32 3126 mutex_lock(&con->mutex);
8921d114 3127 if (con->in_msg == msg) {
95c96174
ED		 3128 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3129 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3130 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3131
3132 /* skip rest of message */
8921d114
AE		 3133 dout("%s %p msg %p revoked\n", __func__, con, msg);
3134 con->in_base_pos = con->in_base_pos -
350b1c32 3135 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3136 front_len -
3137 middle_len -
3138 data_len -
350b1c32 3139 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3140 ceph_msg_put(con->in_msg);
3141 con->in_msg = NULL;
3142 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3143 con->in_seq++;
350b1c32 3144 } else {
8921d114
AE		 3145 dout("%s %p in_msg %p msg %p no-op\n",
3146 __func__, con, con->in_msg, msg);
350b1c32
SW		 3147 }
3148 mutex_unlock(&con->mutex);
3149}
3150
31b8006e
SW		 3151/*
3152 * Queue a keepalive byte to ensure the tcp connection is alive.
3153 */
3154void ceph_con_keepalive(struct ceph_connection *con)
3155{
e00de341 3156 dout("con_keepalive %p\n", con);
00650931 3157 mutex_lock(&con->mutex);
e00de341 3158 clear_standby(con);
00650931 3159 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3160 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3161 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3162 queue_con(con);
3163}
3d14c5d2 3164EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3165
8b9558aa
YZ		 3166bool ceph_con_keepalive_expired(struct ceph_connection *con,
3167 unsigned long interval)
3168{
3169 if (interval > 0 &&
3170 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3171 struct timespec now = CURRENT_TIME;
3172 struct timespec ts;
3173 jiffies_to_timespec(interval, &ts);
3174 ts = timespec_add(con->last_keepalive_ack, ts);
3175 return timespec_compare(&now, &ts) >= 0;
3176 }
3177 return false;
3178}
3179
6644ed7b 3180static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3181{
6644ed7b
AE		 3182 struct ceph_msg_data *data;
3183
3184 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3185 return NULL;
3186
81b36be4 3187 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
6644ed7b
AE		 3188 if (data)
3189 data->type = type;
5240d9f9 3190 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3191
3192 return data;
3193}
3194
3195static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3196{
3197 if (!data)
3198 return;
3199
5240d9f9 3200 WARN_ON(!list_empty(&data->links));
e4339d28 3201 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3202 ceph_pagelist_release(data->pagelist);
81b36be4 3203 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3204}
3205
90af3602 3206void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3207 size_t length, size_t alignment)
02afca6c 3208{
6644ed7b
AE		 3209 struct ceph_msg_data *data;
3210
07aa1558
AE		 3211 BUG_ON(!pages);
3212 BUG_ON(!length);
6644ed7b
AE		 3213
3214 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3215 BUG_ON(!data);
3216 data->pages = pages;
3217 data->length = length;
3218 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3219
5240d9f9
AE		 3220 list_add_tail(&data->links, &msg->data);
3221 msg->data_length += length;
02afca6c 3222}
90af3602 3223EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3224
90af3602 3225void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3226 struct ceph_pagelist *pagelist)
3227{
6644ed7b
AE		 3228 struct ceph_msg_data *data;
3229
07aa1558
AE		 3230 BUG_ON(!pagelist);
3231 BUG_ON(!pagelist->length);
27fa8385 3232
6644ed7b
AE		 3233 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3234 BUG_ON(!data);
3235 data->pagelist = pagelist;
3236
5240d9f9
AE		 3237 list_add_tail(&data->links, &msg->data);
3238 msg->data_length += pagelist->length;
27fa8385 3239}
90af3602 3240EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3241
ea96571f 3242#ifdef CONFIG_BLOCK
90af3602 3243void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3244 size_t length)
27fa8385 3245{
6644ed7b
AE		 3246 struct ceph_msg_data *data;
3247
07aa1558 3248 BUG_ON(!bio);
27fa8385 3249
6644ed7b
AE		 3250 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3251 BUG_ON(!data);
3252 data->bio = bio;
c851c495 3253 data->bio_length = length;
6644ed7b 3254
5240d9f9
AE		 3255 list_add_tail(&data->links, &msg->data);
3256 msg->data_length += length;
27fa8385 3257}
90af3602 3258EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3259#endif /* CONFIG_BLOCK */
27fa8385 3260
31b8006e
SW		 3261/*
3262 * construct a new message with given type, size
3263 * the new msg has a ref count of 1.
3264 */
b61c2763
SW		 3265struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3266 bool can_fail)
31b8006e
SW		 3267{
3268 struct ceph_msg *m;
3269
e3d5d638 3270 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3271 if (m == NULL)
3272 goto out;
31b8006e
SW		 3273
3274 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3275 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3276 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3277
9516e45b
AE		 3278 INIT_LIST_HEAD(&m->list_head);
3279 kref_init(&m->kref);
5240d9f9 3280 INIT_LIST_HEAD(&m->data);
ca20892d 3281
31b8006e
SW		 3282 /* front */
3283 if (front_len) {
eeb0bed5 3284 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3285 if (m->front.iov_base == NULL) {
b61c2763 3286 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3287 front_len);
3288 goto out2;
3289 }
3290 } else {
3291 m->front.iov_base = NULL;
3292 }
f2be82b0 3293 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3294
bb257664 3295 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3296 return m;
3297
3298out2:
3299 ceph_msg_put(m);
3300out:
b61c2763
SW		 3301 if (!can_fail) {
3302 pr_err("msg_new can't create type %d front %d\n", type,
3303 front_len);
f0ed1b7c 3304 WARN_ON(1);
b61c2763
SW		 3305 } else {
3306 dout("msg_new can't create type %d front %d\n", type,
3307 front_len);
3308 }
a79832f2 3309 return NULL;
31b8006e 3310}
3d14c5d2 3311EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3312
31b8006e
SW		 3313/*
3314 * Allocate "middle" portion of a message, if it is needed and wasn't
3315 * allocated by alloc_msg. This allows us to read a small fixed-size
3316 * per-type header in the front and then gracefully fail (i.e.,
3317 * propagate the error to the caller based on info in the front) when
3318 * the middle is too large.
3319 */
2450418c 3320static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3321{
3322 int type = le16_to_cpu(msg->hdr.type);
3323 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3324
3325 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3326 ceph_msg_type_name(type), middle_len);
3327 BUG_ON(!middle_len);
3328 BUG_ON(msg->middle);
3329
b6c1d5b8 3330 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3331 if (!msg->middle)
3332 return -ENOMEM;
3333 return 0;
3334}
3335
2450418c 3336/*
1c20f2d2
AE		 3337 * Allocate a message for receiving an incoming message on a
3338 * connection, and save the result in con->in_msg. Uses the
3339 * connection's private alloc_msg op if available.
3340 *
4740a623
SW		 3341 * Returns 0 on success, or a negative error code.
3342 *
3343 * On success, if we set *skip = 1:
3344 * - the next message should be skipped and ignored.
3345 * - con->in_msg == NULL
3346 * or if we set *skip = 0:
3347 * - con->in_msg is non-null.
3348 * On error (ENOMEM, EAGAIN, ...),
3349 * - con->in_msg == NULL
2450418c 3350 */
4740a623 3351static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3352{
4740a623 3353 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3354 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3355 struct ceph_msg *msg;
4740a623 3356 int ret = 0;
2450418c 3357
1c20f2d2 3358 BUG_ON(con->in_msg != NULL);
53ded495 3359 BUG_ON(!con->ops->alloc_msg);
2450418c 3360
53ded495
AE		 3361 mutex_unlock(&con->mutex);
3362 msg = con->ops->alloc_msg(con, hdr, skip);
3363 mutex_lock(&con->mutex);
3364 if (con->state != CON_STATE_OPEN) {
3365 if (msg)
1d866d1c 3366 ceph_msg_put(msg);
53ded495
AE		 3367 return -EAGAIN;
3368 }
4137577a
AE		 3369 if (msg) {
3370 BUG_ON(*skip);
583d0fef 3371 msg_con_set(msg, con);
4137577a 3372 con->in_msg = msg;
4137577a
AE		 3373 } else {
3374 /*
3375 * Null message pointer means either we should skip
3376 * this message or we couldn't allocate memory. The
3377 * former is not an error.
3378 */
3379 if (*skip)
3380 return 0;
4137577a 3381
67c64eb7 3382 con->error_msg = "error allocating memory for incoming message";
53ded495 3383 return -ENOMEM;
2450418c 3384 }
1c20f2d2 3385 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3386
1c20f2d2
AE		 3387 if (middle_len && !con->in_msg->middle) {
3388 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3389 if (ret < 0) {
1c20f2d2
AE		 3390 ceph_msg_put(con->in_msg);
3391 con->in_msg = NULL;
2450418c
YS		 3392 }
3393 }
9d7f0f13 3394
4740a623 3395 return ret;
2450418c
YS		 3396}
3397
31b8006e
SW		 3398
3399/*
3400 * Free a generically kmalloc'd message.
3401 */
0215e44b 3402static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3403{
0215e44b 3404 dout("%s %p\n", __func__, m);
4965fc38 3405 kvfree(m->front.iov_base);
e3d5d638 3406 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3407}
3408
0215e44b 3409static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3410{
3411 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
10bcee14 3412 struct ceph_msg_data *data, *next;
31b8006e 3413
0215e44b 3414 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3415 WARN_ON(!list_empty(&m->list_head));
3416
583d0fef
ID		 3417 msg_con_set(m, NULL);
3418
c2e552e7
SW		 3419 /* drop middle, data, if any */
3420 if (m->middle) {
3421 ceph_buffer_put(m->middle);
3422 m->middle = NULL;
31b8006e 3423 }
5240d9f9 3424
10bcee14
GT		 3425 list_for_each_entry_safe(data, next, &m->data, links) {
3426 list_del_init(&data->links);
5240d9f9
AE		 3427 ceph_msg_data_destroy(data);
3428 }
a1930804 3429 m->data_length = 0;
58bb3b37 3430
c2e552e7
SW		 3431 if (m->pool)
3432 ceph_msgpool_put(m->pool, m);
3433 else
0215e44b
ID		 3434 ceph_msg_free(m);
3435}
3436
3437struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3438{
3439 dout("%s %p (was %d)\n", __func__, msg,
3440 atomic_read(&msg->kref.refcount));
3441 kref_get(&msg->kref);
3442 return msg;
3443}
3444EXPORT_SYMBOL(ceph_msg_get);
3445
3446void ceph_msg_put(struct ceph_msg *msg)
3447{
3448 dout("%s %p (was %d)\n", __func__, msg,
3449 atomic_read(&msg->kref.refcount));
3450 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3451}
0215e44b 3452EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3453
3454void ceph_msg_dump(struct ceph_msg *msg)
3455{
3cea4c30
ID		 3456 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3457 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3458 print_hex_dump(KERN_DEBUG, "header: ",
3459 DUMP_PREFIX_OFFSET, 16, 1,
3460 &msg->hdr, sizeof(msg->hdr), true);
3461 print_hex_dump(KERN_DEBUG, " front: ",
3462 DUMP_PREFIX_OFFSET, 16, 1,
3463 msg->front.iov_base, msg->front.iov_len, true);
3464 if (msg->middle)
3465 print_hex_dump(KERN_DEBUG, "middle: ",
3466 DUMP_PREFIX_OFFSET, 16, 1,
3467 msg->middle->vec.iov_base,
3468 msg->middle->vec.iov_len, true);
3469 print_hex_dump(KERN_DEBUG, "footer: ",
3470 DUMP_PREFIX_OFFSET, 16, 1,
3471 &msg->footer, sizeof(msg->footer), true);
3472}
3d14c5d2 3473EXPORT_SYMBOL(ceph_msg_dump);
ubuntu-zesty-kernel mirror