]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/ceph/messenger.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: don't bail early from try_read() when skipping a message
[mirror_ubuntu-bionic-kernel.git] / net / ceph / messenger.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
757856d2 9#include <linux/nsproxy.h>
5a0e3ad6 10#include <linux/slab.h>
31b8006e
SW		 11#include <linux/socket.h>
12#include <linux/string.h>
3ebc21f7 13#ifdef CONFIG_BLOCK
68b4476b 14#include <linux/bio.h>
3ebc21f7 15#endif /* CONFIG_BLOCK */
ee3b56f2 16#include <linux/dns_resolver.h>
31b8006e
SW		 17#include <net/tcp.h>
18
2b3e0c90 19#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 20#include <linux/ceph/libceph.h>
21#include <linux/ceph/messenger.h>
22#include <linux/ceph/decode.h>
23#include <linux/ceph/pagelist.h>
bc3b2d7f 24#include <linux/export.h>
31b8006e
SW		 25
26/*
27 * Ceph uses the messenger to exchange ceph_msg messages with other
28 * hosts in the system. The messenger provides ordered and reliable
29 * delivery. We tolerate TCP disconnects by reconnecting (with
30 * exponential backoff) in the case of a fault (disconnection, bad
31 * crc, protocol error). Acks allow sent messages to be discarded by
32 * the sender.
33 */
34
bc18f4b1
AE		 35/*
36 * We track the state of the socket on a given connection using
37 * values defined below. The transition to a new socket state is
38 * handled by a function which verifies we aren't coming from an
39 * unexpected state.
40 *
41 * --------
42 * | NEW* | transient initial state
43 * --------
44 * | con_sock_state_init()
45 * v
46 * ----------
47 * | CLOSED | initialized, but no socket (and no
48 * ---------- TCP connection)
49 * ^ \
50 * | \ con_sock_state_connecting()
51 * | ----------------------
52 * | \
53 * + con_sock_state_closed() \
fbb85a47
SW		 54 * |+--------------------------- \
55 * | \ \ \
56 * | ----------- \ \
57 * | | CLOSING | socket event; \ \
58 * | ----------- await close \ \
59 * | ^ \ |
60 * | | \ |
61 * | + con_sock_state_closing() \ |
62 * | / \ | |
63 * | / --------------- | |
64 * | / \ v v
bc18f4b1
AE		 65 * | / --------------
66 * | / -----------------| CONNECTING | socket created, TCP
67 * | | / -------------- connect initiated
68 * | | | con_sock_state_connected()
69 * | | v
70 * -------------
71 * | CONNECTED | TCP connection established
72 * -------------
73 *
74 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
75 */
ce2c8903
AE		 76
77#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
78#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
79#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
80#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
81#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
82
8dacc7da
SW		 83/*
84 * connection states
85 */
86#define CON_STATE_CLOSED 1 /* -> PREOPEN */
87#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
88#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
89#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
90#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
91#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
92
4a861692
SW		 93/*
94 * ceph_connection flag bits
95 */
96#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
97 * messages on errors */
98#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
99#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
100#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
101#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 102
c9ffc77a
AE		 103static bool con_flag_valid(unsigned long con_flag)
104{
105 switch (con_flag) {
106 case CON_FLAG_LOSSYTX:
107 case CON_FLAG_KEEPALIVE_PENDING:
108 case CON_FLAG_WRITE_PENDING:
109 case CON_FLAG_SOCK_CLOSED:
110 case CON_FLAG_BACKOFF:
111 return true;
112 default:
113 return false;
114 }
115}
116
117static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
118{
119 BUG_ON(!con_flag_valid(con_flag));
120
121 clear_bit(con_flag, &con->flags);
122}
123
124static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
125{
126 BUG_ON(!con_flag_valid(con_flag));
127
128 set_bit(con_flag, &con->flags);
129}
130
131static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
132{
133 BUG_ON(!con_flag_valid(con_flag));
134
135 return test_bit(con_flag, &con->flags);
136}
137
138static bool con_flag_test_and_clear(struct ceph_connection *con,
139 unsigned long con_flag)
140{
141 BUG_ON(!con_flag_valid(con_flag));
142
143 return test_and_clear_bit(con_flag, &con->flags);
144}
145
146static bool con_flag_test_and_set(struct ceph_connection *con,
147 unsigned long con_flag)
148{
149 BUG_ON(!con_flag_valid(con_flag));
150
151 return test_and_set_bit(con_flag, &con->flags);
152}
153
e3d5d638
AE		 154/* Slab caches for frequently-allocated structures */
155
156static struct kmem_cache *ceph_msg_cache;
81b36be4 157static struct kmem_cache *ceph_msg_data_cache;
e3d5d638 158
31b8006e
SW		 159/* static tag bytes (protocol control messages) */
160static char tag_msg = CEPH_MSGR_TAG_MSG;
161static char tag_ack = CEPH_MSGR_TAG_ACK;
162static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
8b9558aa 163static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
31b8006e 164
a6a5349d
SW		 165#ifdef CONFIG_LOCKDEP
166static struct lock_class_key socket_class;
167#endif
168
84495f49
AE		 169/*
170 * When skipping (ignoring) a block of input we read it into a "skip
171 * buffer," which is this many bytes in size.
172 */
173#define SKIP_BUF_SIZE 1024
31b8006e
SW		 174
175static void queue_con(struct ceph_connection *con);
37ab77ac 176static void cancel_con(struct ceph_connection *con);
68931622 177static void ceph_con_workfn(struct work_struct *);
93209264 178static void con_fault(struct ceph_connection *con);
31b8006e 179
31b8006e 180/*
f64a9317
AE		 181 * Nicely render a sockaddr as a string. An array of formatted
182 * strings is used, to approximate reentrancy.
31b8006e 183 */
f64a9317
AE		 184#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
185#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
186#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
187#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
188
189static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
190static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 191
57666519 192static struct page *zero_page; /* used in certain error cases */
57666519 193
3d14c5d2 194const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 195{
196 int i;
197 char *s;
99f0f3b2
AE		 198 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
199 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 200
f64a9317 201 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 202 s = addr_str[i];
203
204 switch (ss->ss_family) {
205 case AF_INET:
bd406145
AE		 206 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
207 ntohs(in4->sin_port));
31b8006e
SW		 208 break;
209
210 case AF_INET6:
bd406145
AE		 211 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
212 ntohs(in6->sin6_port));
31b8006e
SW		 213 break;
214
215 default:
d3002b97
AE		 216 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
217 ss->ss_family);
31b8006e
SW		 218 }
219
220 return s;
221}
3d14c5d2 222EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 223
63f2d211
SW		 224static void encode_my_addr(struct ceph_messenger *msgr)
225{
226 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
227 ceph_encode_addr(&msgr->my_enc_addr);
228}
229
31b8006e
SW		 230/*
231 * work queue for all reading and writing to/from the socket.
232 */
e0f43c94 233static struct workqueue_struct *ceph_msgr_wq;
31b8006e 234
e3d5d638
AE		 235static int ceph_msgr_slab_init(void)
236{
237 BUG_ON(ceph_msg_cache);
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
98fa5dd8 1200static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1201{
739c905b 1202 BUG_ON(!msg);
25aff7c5 1203 BUG_ON(!data_len);
739c905b 1204
4c59b4a2 1205 /* Initialize data cursor */
fe38a2b6 1206
8ae4f4f5 1207 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1208}
1209
31b8006e
SW		 1210/*
1211 * Prepare footer for currently outgoing message, and finish things
1212 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1213 */
859eb799 1214static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1215{
1216 struct ceph_msg *m = con->out_msg;
859eb799 1217 int v = con->out_kvec_left;
31b8006e 1218
fd154f3c
AE		 1219 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1220
31b8006e 1221 dout("prepare_write_message_footer %p\n", con);
31b8006e 1222 con->out_kvec[v].iov_base = &m->footer;
33d07337
YZ		 1223 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1224 if (con->ops->sign_message)
79dbd1ba 1225 con->ops->sign_message(m);
33d07337
YZ		 1226 else
1227 m->footer.sig = 0;
1228 con->out_kvec[v].iov_len = sizeof(m->footer);
1229 con->out_kvec_bytes += sizeof(m->footer);
1230 } else {
1231 m->old_footer.flags = m->footer.flags;
1232 con->out_kvec[v].iov_len = sizeof(m->old_footer);
1233 con->out_kvec_bytes += sizeof(m->old_footer);
1234 }
31b8006e
SW		 1235 con->out_kvec_left++;
1236 con->out_more = m->more_to_follow;
c86a2930 1237 con->out_msg_done = true;
31b8006e
SW		 1238}
1239
1240/*
1241 * Prepare headers for the next outgoing message.
1242 */
1243static void prepare_write_message(struct ceph_connection *con)
1244{
1245 struct ceph_msg *m;
a9a0c51a 1246 u32 crc;
31b8006e 1247
e2200423 1248 con_out_kvec_reset(con);
c86a2930 1249 con->out_msg_done = false;
31b8006e
SW		 1250
1251 /* Sneak an ack in there first? If we can get it into the same
1252 * TCP packet that's a good thing. */
1253 if (con->in_seq > con->in_seq_acked) {
1254 con->in_seq_acked = con->in_seq;
e2200423 1255 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1256 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1257 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1258 &con->out_temp_ack);
31b8006e
SW		 1259 }
1260
38941f80 1261 BUG_ON(list_empty(&con->out_queue));
859eb799 1262 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1263 con->out_msg = m;
38941f80 1264 BUG_ON(m->con != con);
4cf9d544
SW		 1265
1266 /* put message on sent list */
1267 ceph_msg_get(m);
1268 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1269
e84346b7
SW		 1270 /*
1271 * only assign outgoing seq # if we haven't sent this message
1272 * yet. if it is requeued, resend with it's original seq.
1273 */
1274 if (m->needs_out_seq) {
1275 m->hdr.seq = cpu_to_le64(++con->out_seq);
1276 m->needs_out_seq = false;
1277 }
98fa5dd8 1278 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1279
98fa5dd8 1280 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1281 m, con->out_seq, le16_to_cpu(m->hdr.type),
1282 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1283 m->data_length);
31b8006e
SW		 1284 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1285
1286 /* tag + hdr + front + middle */
e2200423 1287 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
67645d76 1288 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
e2200423 1289 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1290
31b8006e 1291 if (m->middle)
e2200423 1292 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1293 m->middle->vec.iov_base);
31b8006e 1294
67645d76 1295 /* fill in hdr crc and finalize hdr */
a9a0c51a
AE		 1296 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1297 con->out_msg->hdr.crc = cpu_to_le32(crc);
67645d76 1298 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
a9a0c51a 1299
67645d76 1300 /* fill in front and middle crc, footer */
a9a0c51a
AE		 1301 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1302 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1303 if (m->middle) {
1304 crc = crc32c(0, m->middle->vec.iov_base,
1305 m->middle->vec.iov_len);
1306 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1307 } else
31b8006e 1308 con->out_msg->footer.middle_crc = 0;
739c905b 1309 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1310 le32_to_cpu(con->out_msg->footer.front_crc),
1311 le32_to_cpu(con->out_msg->footer.middle_crc));
67645d76 1312 con->out_msg->footer.flags = 0;
31b8006e
SW		 1313
1314 /* is there a data payload? */
739c905b 1315 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1316 if (m->data_length) {
1317 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1318 con->out_more = 1; /* data + footer will follow */
1319 } else {
31b8006e 1320 /* no, queue up footer too and be done */
859eb799 1321 prepare_write_message_footer(con);
78625051 1322 }
31b8006e 1323
c9ffc77a 1324 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1325}
1326
1327/*
1328 * Prepare an ack.
1329 */
1330static void prepare_write_ack(struct ceph_connection *con)
1331{
1332 dout("prepare_write_ack %p %llu -> %llu\n", con,
1333 con->in_seq_acked, con->in_seq);
1334 con->in_seq_acked = con->in_seq;
1335
e2200423 1336 con_out_kvec_reset(con);
859eb799 1337
e2200423 1338 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1339
31b8006e 1340 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1341 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1342 &con->out_temp_ack);
1343
31b8006e 1344 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1345 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1346}
1347
3a23083b
SW		 1348/*
1349 * Prepare to share the seq during handshake
1350 */
1351static void prepare_write_seq(struct ceph_connection *con)
1352{
1353 dout("prepare_write_seq %p %llu -> %llu\n", con,
1354 con->in_seq_acked, con->in_seq);
1355 con->in_seq_acked = con->in_seq;
1356
1357 con_out_kvec_reset(con);
1358
1359 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1360 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1361 &con->out_temp_ack);
1362
1363 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1364}
1365
31b8006e
SW		 1366/*
1367 * Prepare to write keepalive byte.
1368 */
1369static void prepare_write_keepalive(struct ceph_connection *con)
1370{
1371 dout("prepare_write_keepalive %p\n", con);
e2200423 1372 con_out_kvec_reset(con);
8b9558aa 1373 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
7f61f545
ID		 1374 struct timespec now = CURRENT_TIME;
1375
8b9558aa 1376 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
7f61f545
ID		 1377 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1378 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1379 &con->out_temp_keepalive2);
8b9558aa
YZ		 1380 } else {
1381 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1382 }
c9ffc77a 1383 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1384}
1385
1386/*
1387 * Connection negotiation.
1388 */
1389
dac1e716
AE		 1390static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1391 int *auth_proto)
4e7a5dcd 1392{
a3530df3 1393 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1394
1395 if (!con->ops->get_authorizer) {
1396 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1397 con->out_connect.authorizer_len = 0;
729796be 1398 return NULL;
b1c6b980
AE		 1399 }
1400
1401 /* Can't hold the mutex while getting authorizer */
ec302645 1402 mutex_unlock(&con->mutex);
dac1e716 1403 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
ec302645 1404 mutex_lock(&con->mutex);
4e7a5dcd 1405
a3530df3 1406 if (IS_ERR(auth))
729796be 1407 return auth;
8dacc7da 1408 if (con->state != CON_STATE_NEGOTIATING)
729796be 1409 return ERR_PTR(-EAGAIN);
0da5d703 1410
8f43fb53
AE		 1411 con->auth_reply_buf = auth->authorizer_reply_buf;
1412 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1413 return auth;
4e7a5dcd
SW		 1414}
1415
31b8006e
SW		 1416/*
1417 * We connected to a peer and are saying hello.
1418 */
e825a66d 1419static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1420{
e2200423
AE		 1421 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1422 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1423 &con->msgr->my_enc_addr);
eed0ef2c 1424
eed0ef2c 1425 con->out_more = 0;
c9ffc77a 1426 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1427}
1428
e825a66d 1429static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1430{
95c96174 1431 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1432 int proto;
dac1e716 1433 int auth_proto;
729796be 1434 struct ceph_auth_handshake *auth;
31b8006e
SW		 1435
1436 switch (con->peer_name.type) {
1437 case CEPH_ENTITY_TYPE_MON:
1438 proto = CEPH_MONC_PROTOCOL;
1439 break;
1440 case CEPH_ENTITY_TYPE_OSD:
1441 proto = CEPH_OSDC_PROTOCOL;
1442 break;
1443 case CEPH_ENTITY_TYPE_MDS:
1444 proto = CEPH_MDSC_PROTOCOL;
1445 break;
1446 default:
1447 BUG();
1448 }
1449
1450 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1451 con->connect_seq, global_seq, proto);
4e7a5dcd 1452
859bff51
ID		 1453 con->out_connect.features =
1454 cpu_to_le64(from_msgr(con->msgr)->supported_features);
31b8006e
SW		 1455 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1456 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1457 con->out_connect.global_seq = cpu_to_le32(global_seq);
1458 con->out_connect.protocol_version = cpu_to_le32(proto);
1459 con->out_connect.flags = 0;
31b8006e 1460
dac1e716
AE		 1461 auth_proto = CEPH_AUTH_UNKNOWN;
1462 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1463 if (IS_ERR(auth))
1464 return PTR_ERR(auth);
3da54776 1465
dac1e716 1466 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1467 con->out_connect.authorizer_len = auth ?
1468 cpu_to_le32(auth->authorizer_buf_len) : 0;
1469
e2200423 1470 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1471 &con->out_connect);
1472 if (auth && auth->authorizer_buf_len)
e2200423 1473 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1474 auth->authorizer_buf);
859eb799 1475
31b8006e 1476 con->out_more = 0;
c9ffc77a 1477 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1478
e10c758e 1479 return 0;
31b8006e
SW		 1480}
1481
31b8006e
SW		 1482/*
1483 * write as much of pending kvecs to the socket as we can.
1484 * 1 -> done
1485 * 0 -> socket full, but more to do
1486 * <0 -> error
1487 */
1488static int write_partial_kvec(struct ceph_connection *con)
1489{
1490 int ret;
1491
1492 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1493 while (con->out_kvec_bytes > 0) {
1494 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1495 con->out_kvec_left, con->out_kvec_bytes,
1496 con->out_more);
1497 if (ret <= 0)
1498 goto out;
1499 con->out_kvec_bytes -= ret;
1500 if (con->out_kvec_bytes == 0)
1501 break; /* done */
f42299e6
AE		 1502
1503 /* account for full iov entries consumed */
1504 while (ret >= con->out_kvec_cur->iov_len) {
1505 BUG_ON(!con->out_kvec_left);
1506 ret -= con->out_kvec_cur->iov_len;
1507 con->out_kvec_cur++;
1508 con->out_kvec_left--;
1509 }
1510 /* and for a partially-consumed entry */
1511 if (ret) {
1512 con->out_kvec_cur->iov_len -= ret;
1513 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1514 }
1515 }
1516 con->out_kvec_left = 0;
31b8006e
SW		 1517 ret = 1;
1518out:
1519 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1520 con->out_kvec_bytes, con->out_kvec_left, ret);
1521 return ret; /* done! */
1522}
1523
35b62808
AE		 1524static u32 ceph_crc32c_page(u32 crc, struct page *page,
1525 unsigned int page_offset,
1526 unsigned int length)
1527{
1528 char *kaddr;
1529
1530 kaddr = kmap(page);
1531 BUG_ON(kaddr == NULL);
1532 crc = crc32c(crc, kaddr + page_offset, length);
1533 kunmap(page);
1534
1535 return crc;
1536}
31b8006e
SW		 1537/*
1538 * Write as much message data payload as we can. If we finish, queue
1539 * up the footer.
1540 * 1 -> done, footer is now queued in out_kvec[].
1541 * 0 -> socket full, but more to do
1542 * <0 -> error
1543 */
34d2d200 1544static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1545{
1546 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1547 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 1548 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
f5db90bc 1549 u32 crc;
31b8006e 1550
859a35d5 1551 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1552
5240d9f9 1553 if (list_empty(&msg->data))
4c59b4a2
AE		 1554 return -EINVAL;
1555
5821bd8c
AE		 1556 /*
1557 * Iterate through each page that contains data to be
1558 * written, and send as much as possible for each.
1559 *
1560 * If we are calculating the data crc (the default), we will
1561 * need to map the page. If we have no pages, they have
1562 * been revoked, so use the zero page.
1563 */
f5db90bc 1564 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1565 while (cursor->resid) {
8a166d05 1566 struct page *page;
e387d525
AE		 1567 size_t page_offset;
1568 size_t length;
8a166d05 1569 bool last_piece;
8ea299bc 1570 bool need_crc;
f5db90bc 1571 int ret;
68b4476b 1572
343128ce
SB		 1573 page = ceph_msg_data_next(cursor, &page_offset, &length,
1574 &last_piece);
e387d525 1575 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
c2cfa194 1576 length, !last_piece);
f5db90bc
AE		 1577 if (ret <= 0) {
1578 if (do_datacrc)
1579 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1580
f5db90bc
AE		 1581 return ret;
1582 }
143334ff
AE		 1583 if (do_datacrc && cursor->need_crc)
1584 crc = ceph_crc32c_page(crc, page, page_offset, length);
343128ce 1585 need_crc = ceph_msg_data_advance(cursor, (size_t)ret);
31b8006e
SW		 1586 }
1587
34d2d200 1588 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1589
1590 /* prepare and queue up footer, too */
f5db90bc
AE		 1591 if (do_datacrc)
1592 msg->footer.data_crc = cpu_to_le32(crc);
1593 else
84ca8fc8 1594 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1595 con_out_kvec_reset(con);
859eb799 1596 prepare_write_message_footer(con);
f5db90bc
AE		 1597
1598 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1599}
1600
1601/*
1602 * write some zeros
1603 */
1604static int write_partial_skip(struct ceph_connection *con)
1605{
1606 int ret;
1607
67645d76 1608 dout("%s %p %d left\n", __func__, con, con->out_skip);
31b8006e 1609 while (con->out_skip > 0) {
31739139 1610 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
31b8006e 1611
e1dcb128 1612 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1613 if (ret <= 0)
1614 goto out;
1615 con->out_skip -= ret;
1616 }
1617 ret = 1;
1618out:
1619 return ret;
1620}
1621
1622/*
1623 * Prepare to read connection handshake, or an ack.
1624 */
eed0ef2c
SW		 1625static void prepare_read_banner(struct ceph_connection *con)
1626{
1627 dout("prepare_read_banner %p\n", con);
1628 con->in_base_pos = 0;
1629}
1630
31b8006e
SW		 1631static void prepare_read_connect(struct ceph_connection *con)
1632{
1633 dout("prepare_read_connect %p\n", con);
1634 con->in_base_pos = 0;
1635}
1636
1637static void prepare_read_ack(struct ceph_connection *con)
1638{
1639 dout("prepare_read_ack %p\n", con);
1640 con->in_base_pos = 0;
1641}
1642
3a23083b
SW		 1643static void prepare_read_seq(struct ceph_connection *con)
1644{
1645 dout("prepare_read_seq %p\n", con);
1646 con->in_base_pos = 0;
1647 con->in_tag = CEPH_MSGR_TAG_SEQ;
1648}
1649
31b8006e
SW		 1650static void prepare_read_tag(struct ceph_connection *con)
1651{
1652 dout("prepare_read_tag %p\n", con);
1653 con->in_base_pos = 0;
1654 con->in_tag = CEPH_MSGR_TAG_READY;
1655}
1656
8b9558aa
YZ		 1657static void prepare_read_keepalive_ack(struct ceph_connection *con)
1658{
1659 dout("prepare_read_keepalive_ack %p\n", con);
1660 con->in_base_pos = 0;
1661}
1662
31b8006e
SW		 1663/*
1664 * Prepare to read a message.
1665 */
1666static int prepare_read_message(struct ceph_connection *con)
1667{
1668 dout("prepare_read_message %p\n", con);
1669 BUG_ON(con->in_msg != NULL);
1670 con->in_base_pos = 0;
1671 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1672 return 0;
1673}
1674
1675
1676static int read_partial(struct ceph_connection *con,
fd51653f 1677 int end, int size, void *object)
31b8006e 1678{
e6cee71f
AE		 1679 while (con->in_base_pos < end) {
1680 int left = end - con->in_base_pos;
31b8006e
SW		 1681 int have = size - left;
1682 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1683 if (ret <= 0)
1684 return ret;
1685 con->in_base_pos += ret;
1686 }
1687 return 1;
1688}
1689
1690
1691/*
1692 * Read all or part of the connect-side handshake on a new connection
1693 */
eed0ef2c 1694static int read_partial_banner(struct ceph_connection *con)
31b8006e 1695{
fd51653f
AE		 1696 int size;
1697 int end;
1698 int ret;
31b8006e 1699
eed0ef2c 1700 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1701
1702 /* peer's banner */
fd51653f
AE		 1703 size = strlen(CEPH_BANNER);
1704 end = size;
1705 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1706 if (ret <= 0)
1707 goto out;
fd51653f
AE		 1708
1709 size = sizeof (con->actual_peer_addr);
1710 end += size;
1711 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1712 if (ret <= 0)
1713 goto out;
fd51653f
AE		 1714
1715 size = sizeof (con->peer_addr_for_me);
1716 end += size;
1717 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1718 if (ret <= 0)
1719 goto out;
fd51653f 1720
eed0ef2c
SW		 1721out:
1722 return ret;
1723}
1724
1725static int read_partial_connect(struct ceph_connection *con)
1726{
fd51653f
AE		 1727 int size;
1728 int end;
1729 int ret;
eed0ef2c
SW		 1730
1731 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1732
fd51653f
AE		 1733 size = sizeof (con->in_reply);
1734 end = size;
1735 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1736 if (ret <= 0)
1737 goto out;
fd51653f
AE		 1738
1739 size = le32_to_cpu(con->in_reply.authorizer_len);
1740 end += size;
1741 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1742 if (ret <= 0)
1743 goto out;
31b8006e 1744
4e7a5dcd
SW		 1745 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1746 con, (int)con->in_reply.tag,
1747 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1748 le32_to_cpu(con->in_reply.global_seq));
1749out:
1750 return ret;
eed0ef2c 1751
31b8006e
SW		 1752}
1753
1754/*
1755 * Verify the hello banner looks okay.
1756 */
1757static int verify_hello(struct ceph_connection *con)
1758{
1759 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1760 pr_err("connect to %s got bad banner\n",
3d14c5d2 1761 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1762 con->error_msg = "protocol error, bad banner";
1763 return -1;
1764 }
1765 return 0;
1766}
1767
1768static bool addr_is_blank(struct sockaddr_storage *ss)
1769{
c44bd69c
ID		 1770 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1771 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1772
31b8006e
SW		 1773 switch (ss->ss_family) {
1774 case AF_INET:
c44bd69c 1775 return addr->s_addr == htonl(INADDR_ANY);
31b8006e 1776 case AF_INET6:
c44bd69c
ID		 1777 return ipv6_addr_any(addr6);
1778 default:
1779 return true;
31b8006e 1780 }
31b8006e
SW		 1781}
1782
1783static int addr_port(struct sockaddr_storage *ss)
1784{
1785 switch (ss->ss_family) {
1786 case AF_INET:
f28bcfbe 1787 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1788 case AF_INET6:
f28bcfbe 1789 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1790 }
1791 return 0;
1792}
1793
1794static void addr_set_port(struct sockaddr_storage *ss, int p)
1795{
1796 switch (ss->ss_family) {
1797 case AF_INET:
1798 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1799 break;
31b8006e
SW		 1800 case AF_INET6:
1801 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1802 break;
31b8006e
SW		 1803 }
1804}
1805
ee3b56f2
NW		 1806/*
1807 * Unlike other *_pton function semantics, zero indicates success.
1808 */
1809static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1810 char delim, const char **ipend)
1811{
99f0f3b2
AE		 1812 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1813 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1814
1815 memset(ss, 0, sizeof(*ss));
1816
1817 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1818 ss->ss_family = AF_INET;
1819 return 0;
1820 }
1821
1822 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1823 ss->ss_family = AF_INET6;
1824 return 0;
1825 }
1826
1827 return -EINVAL;
1828}
1829
1830/*
1831 * Extract hostname string and resolve using kernel DNS facility.
1832 */
1833#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1834static int ceph_dns_resolve_name(const char *name, size_t namelen,
1835 struct sockaddr_storage *ss, char delim, const char **ipend)
1836{
1837 const char *end, *delim_p;
1838 char *colon_p, *ip_addr = NULL;
1839 int ip_len, ret;
1840
1841 /*
1842 * The end of the hostname occurs immediately preceding the delimiter or
1843 * the port marker (':') where the delimiter takes precedence.
1844 */
1845 delim_p = memchr(name, delim, namelen);
1846 colon_p = memchr(name, ':', namelen);
1847
1848 if (delim_p && colon_p)
1849 end = delim_p < colon_p ? delim_p : colon_p;
1850 else if (!delim_p && colon_p)
1851 end = colon_p;
1852 else {
1853 end = delim_p;
1854 if (!end) /* case: hostname:/ */
1855 end = name + namelen;
1856 }
1857
1858 if (end <= name)
1859 return -EINVAL;
1860
1861 /* do dns_resolve upcall */
1862 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1863 if (ip_len > 0)
1864 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1865 else
1866 ret = -ESRCH;
1867
1868 kfree(ip_addr);
1869
1870 *ipend = end;
1871
1872 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1873 ret, ret ? "failed" : ceph_pr_addr(ss));
1874
1875 return ret;
1876}
1877#else
1878static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1879 struct sockaddr_storage *ss, char delim, const char **ipend)
1880{
1881 return -EINVAL;
1882}
1883#endif
1884
1885/*
1886 * Parse a server name (IP or hostname). If a valid IP address is not found
1887 * then try to extract a hostname to resolve using userspace DNS upcall.
1888 */
1889static int ceph_parse_server_name(const char *name, size_t namelen,
1890 struct sockaddr_storage *ss, char delim, const char **ipend)
1891{
1892 int ret;
1893
1894 ret = ceph_pton(name, namelen, ss, delim, ipend);
1895 if (ret)
1896 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1897
1898 return ret;
1899}
1900
31b8006e
SW		 1901/*
1902 * Parse an ip[:port] list into an addr array. Use the default
1903 * monitor port if a port isn't specified.
1904 */
1905int ceph_parse_ips(const char *c, const char *end,
1906 struct ceph_entity_addr *addr,
1907 int max_count, int *count)
1908{
ee3b56f2 1909 int i, ret = -EINVAL;
31b8006e
SW		 1910 const char *p = c;
1911
1912 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1913 for (i = 0; i < max_count; i++) {
1914 const char *ipend;
1915 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1916 int port;
39139f64
SW		 1917 char delim = ',';
1918
1919 if (*p == '[') {
1920 delim = ']';
1921 p++;
1922 }
31b8006e 1923
ee3b56f2
NW		 1924 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1925 if (ret)
31b8006e 1926 goto bad;
ee3b56f2
NW		 1927 ret = -EINVAL;
1928
31b8006e
SW		 1929 p = ipend;
1930
39139f64
SW		 1931 if (delim == ']') {
1932 if (*p != ']') {
1933 dout("missing matching ']'\n");
1934 goto bad;
1935 }
1936 p++;
1937 }
1938
31b8006e
SW		 1939 /* port? */
1940 if (p < end && *p == ':') {
1941 port = 0;
1942 p++;
1943 while (p < end && *p >= '0' && *p <= '9') {
1944 port = (port * 10) + (*p - '0');
1945 p++;
1946 }
f48db1e9
ID		 1947 if (port == 0)
1948 port = CEPH_MON_PORT;
1949 else if (port > 65535)
31b8006e
SW		 1950 goto bad;
1951 } else {
1952 port = CEPH_MON_PORT;
1953 }
1954
1955 addr_set_port(ss, port);
1956
3d14c5d2 1957 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1958
1959 if (p == end)
1960 break;
1961 if (*p != ',')
1962 goto bad;
1963 p++;
1964 }
1965
1966 if (p != end)
1967 goto bad;
1968
1969 if (count)
1970 *count = i + 1;
1971 return 0;
1972
1973bad:
39139f64 1974 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1975 return ret;
31b8006e 1976}
3d14c5d2 1977EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1978
eed0ef2c 1979static int process_banner(struct ceph_connection *con)
31b8006e 1980{
eed0ef2c 1981 dout("process_banner on %p\n", con);
31b8006e
SW		 1982
1983 if (verify_hello(con) < 0)
1984 return -1;
1985
63f2d211
SW		 1986 ceph_decode_addr(&con->actual_peer_addr);
1987 ceph_decode_addr(&con->peer_addr_for_me);
1988
31b8006e
SW		 1989 /*
1990 * Make sure the other end is who we wanted. note that the other
1991 * end may not yet know their ip address, so if it's 0.0.0.0, give
1992 * them the benefit of the doubt.
1993 */
103e2d3a
SW		 1994 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1995 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 1996 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1997 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 1998 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1999 ceph_pr_addr(&con->peer_addr.in_addr),
2000 (int)le32_to_cpu(con->peer_addr.nonce),
2001 ceph_pr_addr(&con->actual_peer_addr.in_addr),
2002 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 2003 con->error_msg = "wrong peer at address";
31b8006e
SW		 2004 return -1;
2005 }
2006
2007 /*
2008 * did we learn our address?
2009 */
2010 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2011 int port = addr_port(&con->msgr->inst.addr.in_addr);
2012
2013 memcpy(&con->msgr->inst.addr.in_addr,
2014 &con->peer_addr_for_me.in_addr,
2015 sizeof(con->peer_addr_for_me.in_addr));
2016 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 2017 encode_my_addr(con->msgr);
eed0ef2c 2018 dout("process_banner learned my addr is %s\n",
3d14c5d2 2019 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 2020 }
2021
eed0ef2c
SW		 2022 return 0;
2023}
2024
2025static int process_connect(struct ceph_connection *con)
2026{
859bff51
ID		 2027 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2028 u64 req_feat = from_msgr(con->msgr)->required_features;
2b3e0c90
ID		 2029 u64 server_feat = ceph_sanitize_features(
2030 le64_to_cpu(con->in_reply.features));
0da5d703 2031 int ret;
04a419f9 2032
eed0ef2c
SW		 2033 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2034
31b8006e 2035 switch (con->in_reply.tag) {
04a419f9
SW		 2036 case CEPH_MSGR_TAG_FEATURES:
2037 pr_err("%s%lld %s feature set mismatch,"
2038 " my %llx < server's %llx, missing %llx\n",
2039 ENTITY_NAME(con->peer_name),
3d14c5d2 2040 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2041 sup_feat, server_feat, server_feat & ~sup_feat);
2042 con->error_msg = "missing required protocol features";
0fa6ebc6 2043 reset_connection(con);
04a419f9
SW		 2044 return -1;
2045
31b8006e 2046 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2047 pr_err("%s%lld %s protocol version mismatch,"
2048 " my %d != server's %d\n",
2049 ENTITY_NAME(con->peer_name),
3d14c5d2 2050 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2051 le32_to_cpu(con->out_connect.protocol_version),
2052 le32_to_cpu(con->in_reply.protocol_version));
2053 con->error_msg = "protocol version mismatch";
0fa6ebc6 2054 reset_connection(con);
31b8006e
SW		 2055 return -1;
2056
4e7a5dcd
SW		 2057 case CEPH_MSGR_TAG_BADAUTHORIZER:
2058 con->auth_retry++;
2059 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2060 con->auth_retry);
2061 if (con->auth_retry == 2) {
2062 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2063 return -1;
2064 }
6d4221b5 2065 con_out_kvec_reset(con);
e825a66d 2066 ret = prepare_write_connect(con);
0da5d703
SW		 2067 if (ret < 0)
2068 return ret;
63733a0f 2069 prepare_read_connect(con);
4e7a5dcd 2070 break;
31b8006e
SW		 2071
2072 case CEPH_MSGR_TAG_RESETSESSION:
2073 /*
2074 * If we connected with a large connect_seq but the peer
2075 * has no record of a session with us (no connection, or
2076 * connect_seq == 0), they will send RESETSESION to indicate
2077 * that they must have reset their session, and may have
2078 * dropped messages.
2079 */
2080 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2081 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2082 pr_err("%s%lld %s connection reset\n",
2083 ENTITY_NAME(con->peer_name),
3d14c5d2 2084 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2085 reset_connection(con);
6d4221b5 2086 con_out_kvec_reset(con);
5a0f8fdd
AE		 2087 ret = prepare_write_connect(con);
2088 if (ret < 0)
2089 return ret;
31b8006e
SW		 2090 prepare_read_connect(con);
2091
2092 /* Tell ceph about it. */
ec302645 2093 mutex_unlock(&con->mutex);
31b8006e
SW		 2094 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2095 if (con->ops->peer_reset)
2096 con->ops->peer_reset(con);
ec302645 2097 mutex_lock(&con->mutex);
8dacc7da 2098 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2099 return -EAGAIN;
31b8006e
SW		 2100 break;
2101
2102 case CEPH_MSGR_TAG_RETRY_SESSION:
2103 /*
2104 * If we sent a smaller connect_seq than the peer has, try
2105 * again with a larger value.
2106 */
5bdca4e0 2107 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2108 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2109 le32_to_cpu(con->in_reply.connect_seq));
2110 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2111 con_out_kvec_reset(con);
5a0f8fdd
AE		 2112 ret = prepare_write_connect(con);
2113 if (ret < 0)
2114 return ret;
31b8006e
SW		 2115 prepare_read_connect(con);
2116 break;
2117
2118 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2119 /*
2120 * If we sent a smaller global_seq than the peer has, try
2121 * again with a larger value.
2122 */
eed0ef2c 2123 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2124 con->peer_global_seq,
5bdca4e0 2125 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2126 get_global_seq(con->msgr,
5bdca4e0 2127 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2128 con_out_kvec_reset(con);
5a0f8fdd
AE		 2129 ret = prepare_write_connect(con);
2130 if (ret < 0)
2131 return ret;
31b8006e
SW		 2132 prepare_read_connect(con);
2133 break;
2134
3a23083b 2135 case CEPH_MSGR_TAG_SEQ:
31b8006e 2136 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2137 if (req_feat & ~server_feat) {
2138 pr_err("%s%lld %s protocol feature mismatch,"
2139 " my required %llx > server's %llx, need %llx\n",
2140 ENTITY_NAME(con->peer_name),
3d14c5d2 2141 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2142 req_feat, server_feat, req_feat & ~server_feat);
2143 con->error_msg = "missing required protocol features";
0fa6ebc6 2144 reset_connection(con);
04a419f9
SW		 2145 return -1;
2146 }
8dacc7da 2147
122070a2 2148 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2149 con->state = CON_STATE_OPEN;
20e55c4c 2150 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2151 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2152 con->connect_seq++;
aba558e2 2153 con->peer_features = server_feat;
31b8006e
SW		 2154 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2155 con->peer_global_seq,
2156 le32_to_cpu(con->in_reply.connect_seq),
2157 con->connect_seq);
2158 WARN_ON(con->connect_seq !=
2159 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2160
2161 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2162 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2163
85effe18 2164 con->delay = 0; /* reset backoff memory */
92ac41d0 2165
3a23083b
SW		 2166 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2167 prepare_write_seq(con);
2168 prepare_read_seq(con);
2169 } else {
2170 prepare_read_tag(con);
2171 }
31b8006e
SW		 2172 break;
2173
2174 case CEPH_MSGR_TAG_WAIT:
2175 /*
2176 * If there is a connection race (we are opening
2177 * connections to each other), one of us may just have
2178 * to WAIT. This shouldn't happen if we are the
2179 * client.
2180 */
04177882
SW		 2181 con->error_msg = "protocol error, got WAIT as client";
2182 return -1;
31b8006e
SW		 2183
2184 default:
31b8006e
SW		 2185 con->error_msg = "protocol error, garbage tag during connect";
2186 return -1;
2187 }
2188 return 0;
2189}
2190
2191
2192/*
2193 * read (part of) an ack
2194 */
2195static int read_partial_ack(struct ceph_connection *con)
2196{
fd51653f
AE		 2197 int size = sizeof (con->in_temp_ack);
2198 int end = size;
31b8006e 2199
fd51653f 2200 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2201}
2202
31b8006e
SW		 2203/*
2204 * We can finally discard anything that's been acked.
2205 */
2206static void process_ack(struct ceph_connection *con)
2207{
2208 struct ceph_msg *m;
2209 u64 ack = le64_to_cpu(con->in_temp_ack);
2210 u64 seq;
2211
31b8006e
SW		 2212 while (!list_empty(&con->out_sent)) {
2213 m = list_first_entry(&con->out_sent, struct ceph_msg,
2214 list_head);
2215 seq = le64_to_cpu(m->hdr.seq);
2216 if (seq > ack)
2217 break;
2218 dout("got ack for seq %llu type %d at %p\n", seq,
2219 le16_to_cpu(m->hdr.type), m);
4cf9d544 2220 m->ack_stamp = jiffies;
31b8006e
SW		 2221 ceph_msg_remove(m);
2222 }
31b8006e
SW		 2223 prepare_read_tag(con);
2224}
2225
2226
2450418c 2227static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2228 struct kvec *section,
2229 unsigned int sec_len, u32 *crc)
2450418c 2230{
68b4476b 2231 int ret, left;
2450418c
YS		 2232
2233 BUG_ON(!section);
2234
2235 while (section->iov_len < sec_len) {
2236 BUG_ON(section->iov_base == NULL);
2237 left = sec_len - section->iov_len;
2238 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2239 section->iov_len, left);
2240 if (ret <= 0)
2241 return ret;
2242 section->iov_len += ret;
2450418c 2243 }
fe3ad593
AE		 2244 if (section->iov_len == sec_len)
2245 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2246
2450418c
YS		 2247 return 1;
2248}
31b8006e 2249
34d2d200
AE		 2250static int read_partial_msg_data(struct ceph_connection *con)
2251{
2252 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2253 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 2254 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
686be208
AE		 2255 struct page *page;
2256 size_t page_offset;
2257 size_t length;
f5db90bc 2258 u32 crc = 0;
34d2d200
AE		 2259 int ret;
2260
2261 BUG_ON(!msg);
5240d9f9 2262 if (list_empty(&msg->data))
4c59b4a2 2263 return -EIO;
34d2d200 2264
f5db90bc
AE		 2265 if (do_datacrc)
2266 crc = con->in_data_crc;
643c68a4 2267 while (cursor->resid) {
343128ce 2268 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
686be208 2269 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2270 if (ret <= 0) {
2271 if (do_datacrc)
2272 con->in_data_crc = crc;
2273
686be208 2274 return ret;
f5db90bc 2275 }
686be208
AE		 2276
2277 if (do_datacrc)
f5db90bc 2278 crc = ceph_crc32c_page(crc, page, page_offset, ret);
343128ce 2279 (void) ceph_msg_data_advance(cursor, (size_t)ret);
34d2d200 2280 }
f5db90bc
AE		 2281 if (do_datacrc)
2282 con->in_data_crc = crc;
34d2d200
AE		 2283
2284 return 1; /* must return > 0 to indicate success */
2285}
2286
31b8006e
SW		 2287/*
2288 * read (part of) a message.
2289 */
686be208
AE		 2290static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2291
31b8006e
SW		 2292static int read_partial_message(struct ceph_connection *con)
2293{
2294 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2295 int size;
2296 int end;
31b8006e 2297 int ret;
95c96174 2298 unsigned int front_len, middle_len, data_len;
859bff51 2299 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
33d07337 2300 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
ae18756b 2301 u64 seq;
fe3ad593 2302 u32 crc;
31b8006e
SW		 2303
2304 dout("read_partial_message con %p msg %p\n", con, m);
2305
2306 /* header */
fd51653f
AE		 2307 size = sizeof (con->in_hdr);
2308 end = size;
2309 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2310 if (ret <= 0)
2311 return ret;
fe3ad593
AE		 2312
2313 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2314 if (cpu_to_le32(crc) != con->in_hdr.crc) {
67c64eb7 2315 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
fe3ad593
AE		 2316 crc, con->in_hdr.crc);
2317 return -EBADMSG;
2318 }
2319
31b8006e
SW		 2320 front_len = le32_to_cpu(con->in_hdr.front_len);
2321 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2322 return -EIO;
2323 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2324 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2325 return -EIO;
2326 data_len = le32_to_cpu(con->in_hdr.data_len);
2327 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2328 return -EIO;
2329
ae18756b
SW		 2330 /* verify seq# */
2331 seq = le64_to_cpu(con->in_hdr.seq);
2332 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2333 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2334 ENTITY_NAME(con->peer_name),
3d14c5d2 2335 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2336 seq, con->in_seq + 1);
2337 con->in_base_pos = -front_len - middle_len - data_len -
2338 sizeof(m->footer);
2339 con->in_tag = CEPH_MSGR_TAG_READY;
e7a88e82 2340 return 1;
ae18756b
SW		 2341 } else if ((s64)seq - (s64)con->in_seq > 1) {
2342 pr_err("read_partial_message bad seq %lld expected %lld\n",
2343 seq, con->in_seq + 1);
2344 con->error_msg = "bad message sequence # for incoming message";
67c64eb7 2345 return -EBADE;
ae18756b
SW		 2346 }
2347
31b8006e
SW		 2348 /* allocate message? */
2349 if (!con->in_msg) {
4740a623
SW		 2350 int skip = 0;
2351
31b8006e 2352 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2353 front_len, data_len);
4740a623
SW		 2354 ret = ceph_con_in_msg_alloc(con, &skip);
2355 if (ret < 0)
2356 return ret;
f759ebb9
AE		 2357
2358 BUG_ON(!con->in_msg ^ skip);
2450418c 2359 if (skip) {
31b8006e 2360 /* skip this message */
a79832f2 2361 dout("alloc_msg said skip message\n");
31b8006e
SW		 2362 con->in_base_pos = -front_len - middle_len - data_len -
2363 sizeof(m->footer);
2364 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2365 con->in_seq++;
e7a88e82 2366 return 1;
31b8006e 2367 }
38941f80 2368
4740a623 2369 BUG_ON(!con->in_msg);
38941f80 2370 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2371 m = con->in_msg;
2372 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2373 if (m->middle)
2374 m->middle->vec.iov_len = 0;
9d7f0f13 2375
78625051 2376 /* prepare for data payload, if any */
a4107026 2377
78625051 2378 if (data_len)
98fa5dd8 2379 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2380 }
2381
2382 /* front */
2450418c
YS		 2383 ret = read_partial_message_section(con, &m->front, front_len,
2384 &con->in_front_crc);
2385 if (ret <= 0)
2386 return ret;
31b8006e
SW		 2387
2388 /* middle */
2450418c 2389 if (m->middle) {
213c99ee
SW		 2390 ret = read_partial_message_section(con, &m->middle->vec,
2391 middle_len,
2450418c 2392 &con->in_middle_crc);
31b8006e
SW		 2393 if (ret <= 0)
2394 return ret;
31b8006e
SW		 2395 }
2396
2397 /* (page) data */
34d2d200
AE		 2398 if (data_len) {
2399 ret = read_partial_msg_data(con);
2400 if (ret <= 0)
2401 return ret;
31b8006e
SW		 2402 }
2403
31b8006e 2404 /* footer */
33d07337
YZ		 2405 if (need_sign)
2406 size = sizeof(m->footer);
2407 else
2408 size = sizeof(m->old_footer);
2409
fd51653f
AE		 2410 end += size;
2411 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2412 if (ret <= 0)
2413 return ret;
2414
33d07337
YZ		 2415 if (!need_sign) {
2416 m->footer.flags = m->old_footer.flags;
2417 m->footer.sig = 0;
2418 }
2419
31b8006e
SW		 2420 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2421 m, front_len, m->footer.front_crc, middle_len,
2422 m->footer.middle_crc, data_len, m->footer.data_crc);
2423
2424 /* crc ok? */
2425 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2426 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2427 m, con->in_front_crc, m->footer.front_crc);
2428 return -EBADMSG;
2429 }
2430 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2431 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2432 m, con->in_middle_crc, m->footer.middle_crc);
2433 return -EBADMSG;
2434 }
bca064d2 2435 if (do_datacrc &&
31b8006e
SW		 2436 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2437 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2438 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2439 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2440 return -EBADMSG;
2441 }
2442
33d07337 2443 if (need_sign && con->ops->check_message_signature &&
79dbd1ba 2444 con->ops->check_message_signature(m)) {
33d07337
YZ		 2445 pr_err("read_partial_message %p signature check failed\n", m);
2446 return -EBADMSG;
2447 }
2448
31b8006e
SW		 2449 return 1; /* done! */
2450}
2451
2452/*
2453 * Process message. This happens in the worker thread. The callback should
2454 * be careful not to do anything that waits on other incoming messages or it
2455 * may deadlock.
2456 */
2457static void process_message(struct ceph_connection *con)
2458{
583d0fef 2459 struct ceph_msg *msg = con->in_msg;
31b8006e 2460
38941f80 2461 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2462 con->in_msg = NULL;
2463
2464 /* if first message, set peer_name */
2465 if (con->peer_name.type == 0)
dbad185d 2466 con->peer_name = msg->hdr.src;
31b8006e 2467
31b8006e 2468 con->in_seq++;
ec302645 2469 mutex_unlock(&con->mutex);
31b8006e
SW		 2470
2471 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2472 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2473 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2474 le16_to_cpu(msg->hdr.type),
2475 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2476 le32_to_cpu(msg->hdr.front_len),
2477 le32_to_cpu(msg->hdr.data_len),
2478 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2479 con->ops->dispatch(con, msg);
ec302645
SW		 2480
2481 mutex_lock(&con->mutex);
31b8006e
SW		 2482}
2483
8b9558aa
YZ		 2484static int read_keepalive_ack(struct ceph_connection *con)
2485{
2486 struct ceph_timespec ceph_ts;
2487 size_t size = sizeof(ceph_ts);
2488 int ret = read_partial(con, size, size, &ceph_ts);
2489 if (ret <= 0)
2490 return ret;
2491 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2492 prepare_read_tag(con);
2493 return 1;
2494}
31b8006e
SW		 2495
2496/*
2497 * Write something to the socket. Called in a worker thread when the
2498 * socket appears to be writeable and we have something ready to send.
2499 */
2500static int try_write(struct ceph_connection *con)
2501{
31b8006e
SW		 2502 int ret = 1;
2503
d59315ca 2504 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2505
31b8006e
SW		 2506more:
2507 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2508
2509 /* open the socket first? */
8dacc7da
SW		 2510 if (con->state == CON_STATE_PREOPEN) {
2511 BUG_ON(con->sock);
2512 con->state = CON_STATE_CONNECTING;
a5988c49 2513
e2200423 2514 con_out_kvec_reset(con);
e825a66d 2515 prepare_write_banner(con);
eed0ef2c 2516 prepare_read_banner(con);
31b8006e 2517
cf3e5c40 2518 BUG_ON(con->in_msg);
31b8006e
SW		 2519 con->in_tag = CEPH_MSGR_TAG_READY;
2520 dout("try_write initiating connect on %p new state %lu\n",
2521 con, con->state);
41617d0c
AE		 2522 ret = ceph_tcp_connect(con);
2523 if (ret < 0) {
31b8006e 2524 con->error_msg = "connect error";
31b8006e
SW		 2525 goto out;
2526 }
2527 }
2528
2529more_kvec:
2530 /* kvec data queued? */
67645d76
ID		 2531 if (con->out_kvec_left) {
2532 ret = write_partial_kvec(con);
31b8006e 2533 if (ret <= 0)
42961d23 2534 goto out;
31b8006e 2535 }
67645d76
ID		 2536 if (con->out_skip) {
2537 ret = write_partial_skip(con);
31b8006e 2538 if (ret <= 0)
42961d23 2539 goto out;
31b8006e
SW		 2540 }
2541
2542 /* msg pages? */
2543 if (con->out_msg) {
c86a2930
SW		 2544 if (con->out_msg_done) {
2545 ceph_msg_put(con->out_msg);
2546 con->out_msg = NULL; /* we're done with this one */
2547 goto do_next;
2548 }
2549
34d2d200 2550 ret = write_partial_message_data(con);
31b8006e
SW		 2551 if (ret == 1)
2552 goto more_kvec; /* we need to send the footer, too! */
2553 if (ret == 0)
42961d23 2554 goto out;
31b8006e 2555 if (ret < 0) {
34d2d200 2556 dout("try_write write_partial_message_data err %d\n",
31b8006e 2557 ret);
42961d23 2558 goto out;
31b8006e
SW		 2559 }
2560 }
2561
c86a2930 2562do_next:
8dacc7da 2563 if (con->state == CON_STATE_OPEN) {
8b9558aa
YZ		 2564 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2565 prepare_write_keepalive(con);
2566 goto more;
2567 }
31b8006e
SW		 2568 /* is anything else pending? */
2569 if (!list_empty(&con->out_queue)) {
2570 prepare_write_message(con);
2571 goto more;
2572 }
2573 if (con->in_seq > con->in_seq_acked) {
2574 prepare_write_ack(con);
2575 goto more;
2576 }
31b8006e
SW		 2577 }
2578
2579 /* Nothing to do! */
c9ffc77a 2580 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2581 dout("try_write nothing else to write.\n");
31b8006e
SW		 2582 ret = 0;
2583out:
42961d23 2584 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2585 return ret;
2586}
2587
2588
2589
2590/*
2591 * Read what we can from the socket.
2592 */
2593static int try_read(struct ceph_connection *con)
2594{
31b8006e
SW		 2595 int ret = -1;
2596
8dacc7da
SW		 2597more:
2598 dout("try_read start on %p state %lu\n", con, con->state);
2599 if (con->state != CON_STATE_CONNECTING &&
2600 con->state != CON_STATE_NEGOTIATING &&
2601 con->state != CON_STATE_OPEN)
31b8006e
SW		 2602 return 0;
2603
8dacc7da 2604 BUG_ON(!con->sock);
ec302645 2605
31b8006e
SW		 2606 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2607 con->in_base_pos);
0da5d703 2608
8dacc7da 2609 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2610 dout("try_read connecting\n");
2611 ret = read_partial_banner(con);
2612 if (ret <= 0)
ab166d5a 2613 goto out;
7593af92
AE		 2614 ret = process_banner(con);
2615 if (ret < 0)
2616 goto out;
2617
8dacc7da 2618 con->state = CON_STATE_NEGOTIATING;
7593af92 2619
6d4221b5
JS		 2620 /*
2621 * Received banner is good, exchange connection info.
2622 * Do not reset out_kvec, as sending our banner raced
2623 * with receiving peer banner after connect completed.
2624 */
7593af92
AE		 2625 ret = prepare_write_connect(con);
2626 if (ret < 0)
2627 goto out;
2628 prepare_read_connect(con);
2629
2630 /* Send connection info before awaiting response */
0da5d703
SW		 2631 goto out;
2632 }
2633
8dacc7da 2634 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2635 dout("try_read negotiating\n");
31b8006e
SW		 2636 ret = read_partial_connect(con);
2637 if (ret <= 0)
31b8006e 2638 goto out;
98bdb0aa
SW		 2639 ret = process_connect(con);
2640 if (ret < 0)
2641 goto out;
31b8006e
SW		 2642 goto more;
2643 }
2644
122070a2 2645 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2646
31b8006e
SW		 2647 if (con->in_base_pos < 0) {
2648 /*
2649 * skipping + discarding content.
2650 *
2651 * FIXME: there must be a better way to do this!
2652 */
84495f49
AE		 2653 static char buf[SKIP_BUF_SIZE];
2654 int skip = min((int) sizeof (buf), -con->in_base_pos);
2655
31b8006e
SW		 2656 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2657 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2658 if (ret <= 0)
98bdb0aa 2659 goto out;
31b8006e
SW		 2660 con->in_base_pos += ret;
2661 if (con->in_base_pos)
2662 goto more;
2663 }
2664 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2665 /*
2666 * what's next?
2667 */
2668 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2669 if (ret <= 0)
98bdb0aa 2670 goto out;
31b8006e
SW		 2671 dout("try_read got tag %d\n", (int)con->in_tag);
2672 switch (con->in_tag) {
2673 case CEPH_MSGR_TAG_MSG:
2674 prepare_read_message(con);
2675 break;
2676 case CEPH_MSGR_TAG_ACK:
2677 prepare_read_ack(con);
2678 break;
8b9558aa
YZ		 2679 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2680 prepare_read_keepalive_ack(con);
2681 break;
31b8006e 2682 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2683 con_close_socket(con);
2684 con->state = CON_STATE_CLOSED;
98bdb0aa 2685 goto out;
31b8006e
SW		 2686 default:
2687 goto bad_tag;
2688 }
2689 }
2690 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2691 ret = read_partial_message(con);
2692 if (ret <= 0) {
2693 switch (ret) {
2694 case -EBADMSG:
a51983e4 2695 con->error_msg = "bad crc/signature";
67c64eb7
ID		 2696 /* fall through */
2697 case -EBADE:
31b8006e 2698 ret = -EIO;
98bdb0aa 2699 break;
31b8006e
SW		 2700 case -EIO:
2701 con->error_msg = "io error";
98bdb0aa 2702 break;
31b8006e 2703 }
98bdb0aa 2704 goto out;
31b8006e
SW		 2705 }
2706 if (con->in_tag == CEPH_MSGR_TAG_READY)
2707 goto more;
2708 process_message(con);
7b862e07
SW		 2709 if (con->state == CON_STATE_OPEN)
2710 prepare_read_tag(con);
31b8006e
SW		 2711 goto more;
2712 }
3a23083b
SW		 2713 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2714 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2715 /*
2716 * the final handshake seq exchange is semantically
2717 * equivalent to an ACK
2718 */
31b8006e
SW		 2719 ret = read_partial_ack(con);
2720 if (ret <= 0)
98bdb0aa 2721 goto out;
31b8006e
SW		 2722 process_ack(con);
2723 goto more;
2724 }
8b9558aa
YZ		 2725 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2726 ret = read_keepalive_ack(con);
2727 if (ret <= 0)
2728 goto out;
2729 goto more;
2730 }
31b8006e 2731
31b8006e 2732out:
98bdb0aa 2733 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2734 return ret;
2735
2736bad_tag:
2737 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2738 con->error_msg = "protocol error, garbage tag";
2739 ret = -1;
2740 goto out;
2741}
2742
2743
2744/*
802c6d96
AE		 2745 * Atomically queue work on a connection after the specified delay.
2746 * Bump @con reference to avoid races with connection teardown.
2747 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2748 */
802c6d96 2749static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2750{
31b8006e 2751 if (!con->ops->get(con)) {
802c6d96 2752 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2753 return -ENOENT;
31b8006e
SW		 2754 }
2755
802c6d96
AE		 2756 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2757 dout("%s %p - already queued\n", __func__, con);
31b8006e 2758 con->ops->put(con);
802c6d96 2759 return -EBUSY;
31b8006e 2760 }
802c6d96
AE		 2761
2762 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2763 return 0;
2764}
2765
2766static void queue_con(struct ceph_connection *con)
2767{
2768 (void) queue_con_delay(con, 0);
31b8006e
SW		 2769}
2770
37ab77ac
ID		 2771static void cancel_con(struct ceph_connection *con)
2772{
2773 if (cancel_delayed_work(&con->work)) {
2774 dout("%s %p\n", __func__, con);
2775 con->ops->put(con);
2776 }
2777}
2778
7bb21d68
AE		 2779static bool con_sock_closed(struct ceph_connection *con)
2780{
c9ffc77a 2781 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2782 return false;
2783
2784#define CASE(x) \
2785 case CON_STATE_ ## x: \
2786 con->error_msg = "socket closed (con state " #x ")"; \
2787 break;
2788
2789 switch (con->state) {
2790 CASE(CLOSED);
2791 CASE(PREOPEN);
2792 CASE(CONNECTING);
2793 CASE(NEGOTIATING);
2794 CASE(OPEN);
2795 CASE(STANDBY);
2796 default:
b9a67899 2797 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2798 __func__, con, con->state);
2799 con->error_msg = "unrecognized con state";
2800 BUG();
2801 break;
2802 }
2803#undef CASE
2804
2805 return true;
2806}
2807
f20a39fd
AE		 2808static bool con_backoff(struct ceph_connection *con)
2809{
2810 int ret;
2811
2812 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2813 return false;
2814
2815 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2816 if (ret) {
2817 dout("%s: con %p FAILED to back off %lu\n", __func__,
2818 con, con->delay);
2819 BUG_ON(ret == -ENOENT);
2820 con_flag_set(con, CON_FLAG_BACKOFF);
2821 }
2822
2823 return true;
2824}
2825
93209264
AE		 2826/* Finish fault handling; con->mutex must *not* be held here */
2827
2828static void con_fault_finish(struct ceph_connection *con)
2829{
f6330cc1
ID		 2830 dout("%s %p\n", __func__, con);
2831
93209264
AE		 2832 /*
2833 * in case we faulted due to authentication, invalidate our
2834 * current tickets so that we can get new ones.
2835 */
f6330cc1
ID		 2836 if (con->auth_retry) {
2837 dout("auth_retry %d, invalidating\n", con->auth_retry);
2838 if (con->ops->invalidate_authorizer)
2839 con->ops->invalidate_authorizer(con);
2840 con->auth_retry = 0;
93209264
AE		 2841 }
2842
2843 if (con->ops->fault)
2844 con->ops->fault(con);
2845}
2846
31b8006e
SW		 2847/*
2848 * Do some work on a connection. Drop a connection ref when we're done.
2849 */
68931622 2850static void ceph_con_workfn(struct work_struct *work)
31b8006e
SW		 2851{
2852 struct ceph_connection *con = container_of(work, struct ceph_connection,
2853 work.work);
49659416 2854 bool fault;
31b8006e 2855
9dd4658d 2856 mutex_lock(&con->mutex);
49659416
AE		 2857 while (true) {
2858 int ret;
31b8006e 2859
49659416
AE		 2860 if ((fault = con_sock_closed(con))) {
2861 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2862 break;
2863 }
2864 if (con_backoff(con)) {
2865 dout("%s: con %p BACKOFF\n", __func__, con);
2866 break;
2867 }
2868 if (con->state == CON_STATE_STANDBY) {
2869 dout("%s: con %p STANDBY\n", __func__, con);
2870 break;
2871 }
2872 if (con->state == CON_STATE_CLOSED) {
2873 dout("%s: con %p CLOSED\n", __func__, con);
2874 BUG_ON(con->sock);
2875 break;
2876 }
2877 if (con->state == CON_STATE_PREOPEN) {
2878 dout("%s: con %p PREOPEN\n", __func__, con);
2879 BUG_ON(con->sock);
2880 }
0da5d703 2881
49659416
AE		 2882 ret = try_read(con);
2883 if (ret < 0) {
2884 if (ret == -EAGAIN)
2885 continue;
67c64eb7
ID		 2886 if (!con->error_msg)
2887 con->error_msg = "socket error on read";
49659416
AE		 2888 fault = true;
2889 break;
2890 }
2891
2892 ret = try_write(con);
2893 if (ret < 0) {
2894 if (ret == -EAGAIN)
2895 continue;
67c64eb7
ID		 2896 if (!con->error_msg)
2897 con->error_msg = "socket error on write";
49659416
AE		 2898 fault = true;
2899 }
2900
2901 break; /* If we make it to here, we're done */
3a140a0d 2902 }
b6e7b6a1
AE		 2903 if (fault)
2904 con_fault(con);
9dd4658d 2905 mutex_unlock(&con->mutex);
0da5d703 2906
b6e7b6a1
AE		 2907 if (fault)
2908 con_fault_finish(con);
2909
2910 con->ops->put(con);
31b8006e
SW		 2911}
2912
31b8006e
SW		 2913/*
2914 * Generic error/fault handler. A retry mechanism is used with
2915 * exponential backoff
2916 */
93209264 2917static void con_fault(struct ceph_connection *con)
31b8006e 2918{
31b8006e 2919 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2920 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2921
67c64eb7
ID		 2922 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2923 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2924 con->error_msg = NULL;
2925
122070a2 2926 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2927 con->state != CON_STATE_NEGOTIATING &&
2928 con->state != CON_STATE_OPEN);
ec302645 2929
31b8006e 2930 con_close_socket(con);
5e095e8b 2931
c9ffc77a 2932 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2933 dout("fault on LOSSYTX channel, marking CLOSED\n");
2934 con->state = CON_STATE_CLOSED;
93209264 2935 return;
3b5ede07
SW		 2936 }
2937
5e095e8b 2938 if (con->in_msg) {
38941f80 2939 BUG_ON(con->in_msg->con != con);
5e095e8b
SW		 2940 ceph_msg_put(con->in_msg);
2941 con->in_msg = NULL;
2942 }
31b8006e 2943
e80a52d1
SW		 2944 /* Requeue anything that hasn't been acked */
2945 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2946
e76661d0
SW		 2947 /* If there are no messages queued or keepalive pending, place
2948 * the connection in a STANDBY state */
2949 if (list_empty(&con->out_queue) &&
c9ffc77a 2950 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2951 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2952 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2953 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2954 } else {
2955 /* retry after a delay. */
8dacc7da 2956 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2957 if (con->delay == 0)
2958 con->delay = BASE_DELAY_INTERVAL;
2959 else if (con->delay < MAX_DELAY_INTERVAL)
2960 con->delay *= 2;
c9ffc77a 2961 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2962 queue_con(con);
31b8006e 2963 }
31b8006e
SW		 2964}
2965
2966
2967
2968/*
15d9882c 2969 * initialize a new messenger instance
31b8006e 2970 */
15d9882c 2971void ceph_messenger_init(struct ceph_messenger *msgr,
859bff51 2972 struct ceph_entity_addr *myaddr)
31b8006e 2973{
31b8006e
SW		 2974 spin_lock_init(&msgr->global_seq_lock);
2975
31b8006e
SW		 2976 if (myaddr)
2977 msgr->inst.addr = *myaddr;
2978
2979 /* select a random nonce */
ac8839d7 2980 msgr->inst.addr.type = 0;
103e2d3a 2981 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 2982 encode_my_addr(msgr);
31b8006e 2983
a2a32584 2984 atomic_set(&msgr->stopping, 0);
757856d2 2985 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
31b8006e 2986
15d9882c 2987 dout("%s %p\n", __func__, msgr);
31b8006e 2988}
15d9882c 2989EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 2990
757856d2
ID		 2991void ceph_messenger_fini(struct ceph_messenger *msgr)
2992{
2993 put_net(read_pnet(&msgr->net));
2994}
2995EXPORT_SYMBOL(ceph_messenger_fini);
2996
583d0fef
ID		 2997static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
2998{
2999 if (msg->con)
3000 msg->con->ops->put(msg->con);
3001
3002 msg->con = con ? con->ops->get(con) : NULL;
3003 BUG_ON(msg->con != con);
3004}
3005
e00de341
SW		 3006static void clear_standby(struct ceph_connection *con)
3007{
3008 /* come back from STANDBY? */
8dacc7da 3009 if (con->state == CON_STATE_STANDBY) {
e00de341 3010 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 3011 con->state = CON_STATE_PREOPEN;
e00de341 3012 con->connect_seq++;
c9ffc77a
AE		 3013 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3014 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 3015 }
3016}
3017
31b8006e
SW		 3018/*
3019 * Queue up an outgoing message on the given connection.
3020 */
3021void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3022{
31b8006e 3023 /* set src+dst */
dbad185d 3024 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 3025 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 3026 msg->needs_out_seq = true;
3027
ec302645 3028 mutex_lock(&con->mutex);
92ce034b 3029
8dacc7da 3030 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 3031 dout("con_send %p closed, dropping %p\n", con, msg);
3032 ceph_msg_put(msg);
3033 mutex_unlock(&con->mutex);
3034 return;
3035 }
3036
583d0fef 3037 msg_con_set(msg, con);
92ce034b 3038
31b8006e
SW		 3039 BUG_ON(!list_empty(&msg->list_head));
3040 list_add_tail(&msg->list_head, &con->out_queue);
3041 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3042 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3043 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3044 le32_to_cpu(msg->hdr.front_len),
3045 le32_to_cpu(msg->hdr.middle_len),
3046 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 3047
3048 clear_standby(con);
ec302645 3049 mutex_unlock(&con->mutex);
31b8006e
SW		 3050
3051 /* if there wasn't anything waiting to send before, queue
3052 * new work */
c9ffc77a 3053 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3054 queue_con(con);
3055}
3d14c5d2 3056EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3057
3058/*
3059 * Revoke a message that was previously queued for send
3060 */
6740a845 3061void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3062{
6740a845
AE		 3063 struct ceph_connection *con = msg->con;
3064
583d0fef
ID		 3065 if (!con) {
3066 dout("%s msg %p null con\n", __func__, msg);
6740a845 3067 return; /* Message not in our possession */
583d0fef 3068 }
6740a845 3069
ec302645 3070 mutex_lock(&con->mutex);
31b8006e 3071 if (!list_empty(&msg->list_head)) {
38941f80 3072 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3073 list_del_init(&msg->list_head);
31b8006e 3074 msg->hdr.seq = 0;
38941f80 3075
31b8006e 3076 ceph_msg_put(msg);
ed98adad
SW		 3077 }
3078 if (con->out_msg == msg) {
67645d76
ID		 3079 BUG_ON(con->out_skip);
3080 /* footer */
3081 if (con->out_msg_done) {
3082 con->out_skip += con_out_kvec_skip(con);
3083 } else {
3084 BUG_ON(!msg->data_length);
3085 if (con->peer_features & CEPH_FEATURE_MSG_AUTH)
3086 con->out_skip += sizeof(msg->footer);
3087 else
3088 con->out_skip += sizeof(msg->old_footer);
31b8006e 3089 }
67645d76
ID		 3090 /* data, middle, front */
3091 if (msg->data_length)
3092 con->out_skip += msg->cursor.total_resid;
3093 if (msg->middle)
3094 con->out_skip += con_out_kvec_skip(con);
3095 con->out_skip += con_out_kvec_skip(con);
3096
3097 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3098 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
ed98adad 3099 msg->hdr.seq = 0;
67645d76 3100 con->out_msg = NULL;
92ce034b 3101 ceph_msg_put(msg);
31b8006e 3102 }
67645d76 3103
ec302645 3104 mutex_unlock(&con->mutex);
31b8006e
SW		 3105}
3106
350b1c32 3107/*
0d59ab81 3108 * Revoke a message that we may be reading data into
350b1c32 3109 */
8921d114 3110void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3111{
583d0fef 3112 struct ceph_connection *con = msg->con;
8921d114 3113
583d0fef 3114 if (!con) {
8921d114 3115 dout("%s msg %p null con\n", __func__, msg);
8921d114
AE		 3116 return; /* Message not in our possession */
3117 }
3118
350b1c32 3119 mutex_lock(&con->mutex);
8921d114 3120 if (con->in_msg == msg) {
95c96174
ED		 3121 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3122 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3123 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3124
3125 /* skip rest of message */
8921d114
AE		 3126 dout("%s %p msg %p revoked\n", __func__, con, msg);
3127 con->in_base_pos = con->in_base_pos -
350b1c32 3128 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3129 front_len -
3130 middle_len -
3131 data_len -
350b1c32 3132 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3133 ceph_msg_put(con->in_msg);
3134 con->in_msg = NULL;
3135 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3136 con->in_seq++;
350b1c32 3137 } else {
8921d114
AE		 3138 dout("%s %p in_msg %p msg %p no-op\n",
3139 __func__, con, con->in_msg, msg);
350b1c32
SW		 3140 }
3141 mutex_unlock(&con->mutex);
3142}
3143
31b8006e
SW		 3144/*
3145 * Queue a keepalive byte to ensure the tcp connection is alive.
3146 */
3147void ceph_con_keepalive(struct ceph_connection *con)
3148{
e00de341 3149 dout("con_keepalive %p\n", con);
00650931 3150 mutex_lock(&con->mutex);
e00de341 3151 clear_standby(con);
00650931 3152 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3153 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3154 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3155 queue_con(con);
3156}
3d14c5d2 3157EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3158
8b9558aa
YZ		 3159bool ceph_con_keepalive_expired(struct ceph_connection *con,
3160 unsigned long interval)
3161{
3162 if (interval > 0 &&
3163 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3164 struct timespec now = CURRENT_TIME;
3165 struct timespec ts;
3166 jiffies_to_timespec(interval, &ts);
3167 ts = timespec_add(con->last_keepalive_ack, ts);
3168 return timespec_compare(&now, &ts) >= 0;
3169 }
3170 return false;
3171}
3172
6644ed7b 3173static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3174{
6644ed7b
AE		 3175 struct ceph_msg_data *data;
3176
3177 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3178 return NULL;
3179
81b36be4 3180 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
6644ed7b
AE		 3181 if (data)
3182 data->type = type;
5240d9f9 3183 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3184
3185 return data;
3186}
3187
3188static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3189{
3190 if (!data)
3191 return;
3192
5240d9f9 3193 WARN_ON(!list_empty(&data->links));
e4339d28 3194 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3195 ceph_pagelist_release(data->pagelist);
81b36be4 3196 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3197}
3198
90af3602 3199void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3200 size_t length, size_t alignment)
02afca6c 3201{
6644ed7b
AE		 3202 struct ceph_msg_data *data;
3203
07aa1558
AE		 3204 BUG_ON(!pages);
3205 BUG_ON(!length);
6644ed7b
AE		 3206
3207 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3208 BUG_ON(!data);
3209 data->pages = pages;
3210 data->length = length;
3211 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3212
5240d9f9
AE		 3213 list_add_tail(&data->links, &msg->data);
3214 msg->data_length += length;
02afca6c 3215}
90af3602 3216EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3217
90af3602 3218void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3219 struct ceph_pagelist *pagelist)
3220{
6644ed7b
AE		 3221 struct ceph_msg_data *data;
3222
07aa1558
AE		 3223 BUG_ON(!pagelist);
3224 BUG_ON(!pagelist->length);
27fa8385 3225
6644ed7b
AE		 3226 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3227 BUG_ON(!data);
3228 data->pagelist = pagelist;
3229
5240d9f9
AE		 3230 list_add_tail(&data->links, &msg->data);
3231 msg->data_length += pagelist->length;
27fa8385 3232}
90af3602 3233EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3234
ea96571f 3235#ifdef CONFIG_BLOCK
90af3602 3236void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3237 size_t length)
27fa8385 3238{
6644ed7b
AE		 3239 struct ceph_msg_data *data;
3240
07aa1558 3241 BUG_ON(!bio);
27fa8385 3242
6644ed7b
AE		 3243 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3244 BUG_ON(!data);
3245 data->bio = bio;
c851c495 3246 data->bio_length = length;
6644ed7b 3247
5240d9f9
AE		 3248 list_add_tail(&data->links, &msg->data);
3249 msg->data_length += length;
27fa8385 3250}
90af3602 3251EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3252#endif /* CONFIG_BLOCK */
27fa8385 3253
31b8006e
SW		 3254/*
3255 * construct a new message with given type, size
3256 * the new msg has a ref count of 1.
3257 */
b61c2763
SW		 3258struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3259 bool can_fail)
31b8006e
SW		 3260{
3261 struct ceph_msg *m;
3262
e3d5d638 3263 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3264 if (m == NULL)
3265 goto out;
31b8006e
SW		 3266
3267 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3268 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3269 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3270
9516e45b
AE		 3271 INIT_LIST_HEAD(&m->list_head);
3272 kref_init(&m->kref);
5240d9f9 3273 INIT_LIST_HEAD(&m->data);
ca20892d 3274
31b8006e
SW		 3275 /* front */
3276 if (front_len) {
eeb0bed5 3277 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3278 if (m->front.iov_base == NULL) {
b61c2763 3279 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3280 front_len);
3281 goto out2;
3282 }
3283 } else {
3284 m->front.iov_base = NULL;
3285 }
f2be82b0 3286 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3287
bb257664 3288 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3289 return m;
3290
3291out2:
3292 ceph_msg_put(m);
3293out:
b61c2763
SW		 3294 if (!can_fail) {
3295 pr_err("msg_new can't create type %d front %d\n", type,
3296 front_len);
f0ed1b7c 3297 WARN_ON(1);
b61c2763
SW		 3298 } else {
3299 dout("msg_new can't create type %d front %d\n", type,
3300 front_len);
3301 }
a79832f2 3302 return NULL;
31b8006e 3303}
3d14c5d2 3304EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3305
31b8006e
SW		 3306/*
3307 * Allocate "middle" portion of a message, if it is needed and wasn't
3308 * allocated by alloc_msg. This allows us to read a small fixed-size
3309 * per-type header in the front and then gracefully fail (i.e.,
3310 * propagate the error to the caller based on info in the front) when
3311 * the middle is too large.
3312 */
2450418c 3313static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3314{
3315 int type = le16_to_cpu(msg->hdr.type);
3316 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3317
3318 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3319 ceph_msg_type_name(type), middle_len);
3320 BUG_ON(!middle_len);
3321 BUG_ON(msg->middle);
3322
b6c1d5b8 3323 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3324 if (!msg->middle)
3325 return -ENOMEM;
3326 return 0;
3327}
3328
2450418c 3329/*
1c20f2d2
AE		 3330 * Allocate a message for receiving an incoming message on a
3331 * connection, and save the result in con->in_msg. Uses the
3332 * connection's private alloc_msg op if available.
3333 *
4740a623
SW		 3334 * Returns 0 on success, or a negative error code.
3335 *
3336 * On success, if we set *skip = 1:
3337 * - the next message should be skipped and ignored.
3338 * - con->in_msg == NULL
3339 * or if we set *skip = 0:
3340 * - con->in_msg is non-null.
3341 * On error (ENOMEM, EAGAIN, ...),
3342 * - con->in_msg == NULL
2450418c 3343 */
4740a623 3344static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3345{
4740a623 3346 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3347 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3348 struct ceph_msg *msg;
4740a623 3349 int ret = 0;
2450418c 3350
1c20f2d2 3351 BUG_ON(con->in_msg != NULL);
53ded495 3352 BUG_ON(!con->ops->alloc_msg);
2450418c 3353
53ded495
AE		 3354 mutex_unlock(&con->mutex);
3355 msg = con->ops->alloc_msg(con, hdr, skip);
3356 mutex_lock(&con->mutex);
3357 if (con->state != CON_STATE_OPEN) {
3358 if (msg)
1d866d1c 3359 ceph_msg_put(msg);
53ded495
AE		 3360 return -EAGAIN;
3361 }
4137577a
AE		 3362 if (msg) {
3363 BUG_ON(*skip);
583d0fef 3364 msg_con_set(msg, con);
4137577a 3365 con->in_msg = msg;
4137577a
AE		 3366 } else {
3367 /*
3368 * Null message pointer means either we should skip
3369 * this message or we couldn't allocate memory. The
3370 * former is not an error.
3371 */
3372 if (*skip)
3373 return 0;
4137577a 3374
67c64eb7 3375 con->error_msg = "error allocating memory for incoming message";
53ded495 3376 return -ENOMEM;
2450418c 3377 }
1c20f2d2 3378 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3379
1c20f2d2
AE		 3380 if (middle_len && !con->in_msg->middle) {
3381 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3382 if (ret < 0) {
1c20f2d2
AE		 3383 ceph_msg_put(con->in_msg);
3384 con->in_msg = NULL;
2450418c
YS		 3385 }
3386 }
9d7f0f13 3387
4740a623 3388 return ret;
2450418c
YS		 3389}
3390
31b8006e
SW		 3391
3392/*
3393 * Free a generically kmalloc'd message.
3394 */
0215e44b 3395static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3396{
0215e44b 3397 dout("%s %p\n", __func__, m);
4965fc38 3398 kvfree(m->front.iov_base);
e3d5d638 3399 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3400}
3401
0215e44b 3402static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3403{
3404 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
10bcee14 3405 struct ceph_msg_data *data, *next;
31b8006e 3406
0215e44b 3407 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3408 WARN_ON(!list_empty(&m->list_head));
3409
583d0fef
ID		 3410 msg_con_set(m, NULL);
3411
c2e552e7
SW		 3412 /* drop middle, data, if any */
3413 if (m->middle) {
3414 ceph_buffer_put(m->middle);
3415 m->middle = NULL;
31b8006e 3416 }
5240d9f9 3417
10bcee14
GT		 3418 list_for_each_entry_safe(data, next, &m->data, links) {
3419 list_del_init(&data->links);
5240d9f9
AE		 3420 ceph_msg_data_destroy(data);
3421 }
a1930804 3422 m->data_length = 0;
58bb3b37 3423
c2e552e7
SW		 3424 if (m->pool)
3425 ceph_msgpool_put(m->pool, m);
3426 else
0215e44b
ID		 3427 ceph_msg_free(m);
3428}
3429
3430struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3431{
3432 dout("%s %p (was %d)\n", __func__, msg,
3433 atomic_read(&msg->kref.refcount));
3434 kref_get(&msg->kref);
3435 return msg;
3436}
3437EXPORT_SYMBOL(ceph_msg_get);
3438
3439void ceph_msg_put(struct ceph_msg *msg)
3440{
3441 dout("%s %p (was %d)\n", __func__, msg,
3442 atomic_read(&msg->kref.refcount));
3443 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3444}
0215e44b 3445EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3446
3447void ceph_msg_dump(struct ceph_msg *msg)
3448{
3cea4c30
ID		 3449 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3450 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3451 print_hex_dump(KERN_DEBUG, "header: ",
3452 DUMP_PREFIX_OFFSET, 16, 1,
3453 &msg->hdr, sizeof(msg->hdr), true);
3454 print_hex_dump(KERN_DEBUG, " front: ",
3455 DUMP_PREFIX_OFFSET, 16, 1,
3456 msg->front.iov_base, msg->front.iov_len, true);
3457 if (msg->middle)
3458 print_hex_dump(KERN_DEBUG, "middle: ",
3459 DUMP_PREFIX_OFFSET, 16, 1,
3460 msg->middle->vec.iov_base,
3461 msg->middle->vec.iov_len, true);
3462 print_hex_dump(KERN_DEBUG, "footer: ",
3463 DUMP_PREFIX_OFFSET, 16, 1,
3464 &msg->footer, sizeof(msg->footer), true);
3465}
3d14c5d2 3466EXPORT_SYMBOL(ceph_msg_dump);
ubuntu-bionic-kernel mirror