]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blame - net/ceph/messenger.c
projects / mirror_ubuntu-hirsute-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Linux 4.20-rc3
[mirror_ubuntu-hirsute-kernel.git] / net / ceph / messenger.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
3d14c5d2 2#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 3
4#include <linux/crc32c.h>
5#include <linux/ctype.h>
6#include <linux/highmem.h>
7#include <linux/inet.h>
8#include <linux/kthread.h>
9#include <linux/net.h>
757856d2 10#include <linux/nsproxy.h>
633ee407 11#include <linux/sched/mm.h>
5a0e3ad6 12#include <linux/slab.h>
31b8006e
SW		 13#include <linux/socket.h>
14#include <linux/string.h>
3ebc21f7 15#ifdef CONFIG_BLOCK
68b4476b 16#include <linux/bio.h>
3ebc21f7 17#endif /* CONFIG_BLOCK */
ee3b56f2 18#include <linux/dns_resolver.h>
31b8006e
SW		 19#include <net/tcp.h>
20
2b3e0c90 21#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 22#include <linux/ceph/libceph.h>
23#include <linux/ceph/messenger.h>
24#include <linux/ceph/decode.h>
25#include <linux/ceph/pagelist.h>
bc3b2d7f 26#include <linux/export.h>
31b8006e
SW		 27
28/*
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
34 * the sender.
35 */
36
bc18f4b1
AE		 37/*
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
41 * unexpected state.
42 *
43 * --------
44 * | NEW* | transient initial state
45 * --------
46 * | con_sock_state_init()
47 * v
48 * ----------
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
51 * ^ \
52 * | \ con_sock_state_connecting()
53 * | ----------------------
54 * | \
55 * + con_sock_state_closed() \
fbb85a47
SW		 56 * |+--------------------------- \
57 * | \ \ \
58 * | ----------- \ \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
61 * | ^ \ |
62 * | | \ |
63 * | + con_sock_state_closing() \ |
64 * | / \ | |
65 * | / --------------- | |
66 * | / \ v v
bc18f4b1
AE		 67 * | / --------------
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
71 * | | v
72 * -------------
73 * | CONNECTED | TCP connection established
74 * -------------
75 *
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77 */
ce2c8903
AE		 78
79#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
84
8dacc7da
SW		 85/*
86 * connection states
87 */
88#define CON_STATE_CLOSED 1 /* -> PREOPEN */
89#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
94
4a861692
SW		 95/*
96 * ceph_connection flag bits
97 */
98#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 104
c9ffc77a
AE		 105static bool con_flag_valid(unsigned long con_flag)
106{
107 switch (con_flag) {
108 case CON_FLAG_LOSSYTX:
109 case CON_FLAG_KEEPALIVE_PENDING:
110 case CON_FLAG_WRITE_PENDING:
111 case CON_FLAG_SOCK_CLOSED:
112 case CON_FLAG_BACKOFF:
113 return true;
114 default:
115 return false;
116 }
117}
118
119static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120{
121 BUG_ON(!con_flag_valid(con_flag));
122
123 clear_bit(con_flag, &con->flags);
124}
125
126static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127{
128 BUG_ON(!con_flag_valid(con_flag));
129
130 set_bit(con_flag, &con->flags);
131}
132
133static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134{
135 BUG_ON(!con_flag_valid(con_flag));
136
137 return test_bit(con_flag, &con->flags);
138}
139
140static bool con_flag_test_and_clear(struct ceph_connection *con,
141 unsigned long con_flag)
142{
143 BUG_ON(!con_flag_valid(con_flag));
144
145 return test_and_clear_bit(con_flag, &con->flags);
146}
147
148static bool con_flag_test_and_set(struct ceph_connection *con,
149 unsigned long con_flag)
150{
151 BUG_ON(!con_flag_valid(con_flag));
152
153 return test_and_set_bit(con_flag, &con->flags);
154}
155
e3d5d638
AE		 156/* Slab caches for frequently-allocated structures */
157
158static struct kmem_cache *ceph_msg_cache;
159
31b8006e
SW		 160/* static tag bytes (protocol control messages) */
161static char tag_msg = CEPH_MSGR_TAG_MSG;
162static char tag_ack = CEPH_MSGR_TAG_ACK;
163static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
8b9558aa 164static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
31b8006e 165
a6a5349d
SW		 166#ifdef CONFIG_LOCKDEP
167static struct lock_class_key socket_class;
168#endif
169
31b8006e 170static void queue_con(struct ceph_connection *con);
37ab77ac 171static void cancel_con(struct ceph_connection *con);
68931622 172static void ceph_con_workfn(struct work_struct *);
93209264 173static void con_fault(struct ceph_connection *con);
31b8006e 174
31b8006e 175/*
f64a9317
AE		 176 * Nicely render a sockaddr as a string. An array of formatted
177 * strings is used, to approximate reentrancy.
31b8006e 178 */
f64a9317
AE		 179#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
180#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
181#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
182#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
183
184static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
185static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 186
57666519 187static struct page *zero_page; /* used in certain error cases */
57666519 188
3d14c5d2 189const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 190{
191 int i;
192 char *s;
99f0f3b2
AE		 193 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
194 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 195
f64a9317 196 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 197 s = addr_str[i];
198
199 switch (ss->ss_family) {
200 case AF_INET:
bd406145
AE		 201 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
202 ntohs(in4->sin_port));
31b8006e
SW		 203 break;
204
205 case AF_INET6:
bd406145
AE		 206 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
207 ntohs(in6->sin6_port));
31b8006e
SW		 208 break;
209
210 default:
d3002b97
AE		 211 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
212 ss->ss_family);
31b8006e
SW		 213 }
214
215 return s;
216}
3d14c5d2 217EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 218
63f2d211
SW		 219static void encode_my_addr(struct ceph_messenger *msgr)
220{
221 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
222 ceph_encode_addr(&msgr->my_enc_addr);
223}
224
31b8006e
SW		 225/*
226 * work queue for all reading and writing to/from the socket.
227 */
e0f43c94 228static struct workqueue_struct *ceph_msgr_wq;
31b8006e 229
e3d5d638
AE		 230static int ceph_msgr_slab_init(void)
231{
232 BUG_ON(ceph_msg_cache);
5ee61e95 233 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
81b36be4
AE		 234 if (!ceph_msg_cache)
235 return -ENOMEM;
236
0d9c1ab3 237 return 0;
e3d5d638
AE		 238}
239
240static void ceph_msgr_slab_exit(void)
241{
242 BUG_ON(!ceph_msg_cache);
243 kmem_cache_destroy(ceph_msg_cache);
244 ceph_msg_cache = NULL;
245}
246
15417167 247static void _ceph_msgr_exit(void)
6173d1f0 248{
d3002b97 249 if (ceph_msgr_wq) {
6173d1f0 250 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 251 ceph_msgr_wq = NULL;
252 }
6173d1f0 253
6173d1f0 254 BUG_ON(zero_page == NULL);
09cbfeaf 255 put_page(zero_page);
6173d1f0 256 zero_page = NULL;
d920ff6f
BC		 257
258 ceph_msgr_slab_exit();
6173d1f0
AE		 259}
260
57a35dfb 261int __init ceph_msgr_init(void)
31b8006e 262{
d920ff6f
BC		 263 if (ceph_msgr_slab_init())
264 return -ENOMEM;
265
57666519
AE		 266 BUG_ON(zero_page != NULL);
267 zero_page = ZERO_PAGE(0);
09cbfeaf 268 get_page(zero_page);
57666519 269
f9865f06
ID		 270 /*
271 * The number of active work items is limited by the number of
272 * connections, so leave @max_active at default.
273 */
274 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
6173d1f0
AE		 275 if (ceph_msgr_wq)
276 return 0;
57666519 277
6173d1f0
AE		 278 pr_err("msgr_init failed to create workqueue\n");
279 _ceph_msgr_exit();
57666519 280
6173d1f0 281 return -ENOMEM;
31b8006e
SW		 282}
283
284void ceph_msgr_exit(void)
285{
57666519 286 BUG_ON(ceph_msgr_wq == NULL);
57666519 287
6173d1f0 288 _ceph_msgr_exit();
31b8006e
SW		 289}
290
cd84db6e 291void ceph_msgr_flush(void)
a922d38f
SW		 292{
293 flush_workqueue(ceph_msgr_wq);
294}
3d14c5d2 295EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 296
ce2c8903
AE		 297/* Connection socket state transition functions */
298
299static void con_sock_state_init(struct ceph_connection *con)
300{
301 int old_state;
302
303 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
304 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
305 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 306 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
307 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 308}
309
310static void con_sock_state_connecting(struct ceph_connection *con)
311{
312 int old_state;
313
314 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
315 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
316 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 317 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
318 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 319}
320
321static void con_sock_state_connected(struct ceph_connection *con)
322{
323 int old_state;
324
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
327 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 328 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
329 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 330}
331
332static void con_sock_state_closing(struct ceph_connection *con)
333{
334 int old_state;
335
336 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
337 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
338 old_state != CON_SOCK_STATE_CONNECTED &&
339 old_state != CON_SOCK_STATE_CLOSING))
340 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 341 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
342 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 343}
344
345static void con_sock_state_closed(struct ceph_connection *con)
346{
347 int old_state;
348
349 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
350 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 351 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 352 old_state != CON_SOCK_STATE_CONNECTING &&
353 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 354 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 355 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
356 CON_SOCK_STATE_CLOSED);
ce2c8903 357}
a922d38f 358
31b8006e
SW		 359/*
360 * socket callback functions
361 */
362
363/* data available on socket, or listen socket received a connect */
676d2369 364static void ceph_sock_data_ready(struct sock *sk)
31b8006e 365{
bd406145 366 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 367 if (atomic_read(&con->msgr->stopping)) {
368 return;
369 }
bd406145 370
31b8006e 371 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 372 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 373 con, con->state);
374 queue_con(con);
375 }
376}
377
378/* socket has buffer space for writing */
327800bd 379static void ceph_sock_write_space(struct sock *sk)
31b8006e 380{
d3002b97 381 struct ceph_connection *con = sk->sk_user_data;
31b8006e 382
182fac26
JS		 383 /* only queue to workqueue if there is data we want to write,
384 * and there is sufficient space in the socket buffer to accept
327800bd 385 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 386 * doesn't get called again until try_write() fills the socket
387 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
388 * and net/core/stream.c:sk_stream_write_space().
389 */
c9ffc77a 390 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
64dc6130 391 if (sk_stream_is_writeable(sk)) {
327800bd 392 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 393 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
394 queue_con(con);
395 }
31b8006e 396 } else {
327800bd 397 dout("%s %p nothing to write\n", __func__, con);
31b8006e 398 }
31b8006e
SW		 399}
400
401/* socket's state has changed */
327800bd 402static void ceph_sock_state_change(struct sock *sk)
31b8006e 403{
bd406145 404 struct ceph_connection *con = sk->sk_user_data;
31b8006e 405
327800bd 406 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 407 con, con->state, sk->sk_state);
408
31b8006e
SW		 409 switch (sk->sk_state) {
410 case TCP_CLOSE:
327800bd 411 dout("%s TCP_CLOSE\n", __func__);
18370b36 412 /* fall through */
31b8006e 413 case TCP_CLOSE_WAIT:
327800bd 414 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 415 con_sock_state_closing(con);
c9ffc77a 416 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 417 queue_con(con);
31b8006e
SW		 418 break;
419 case TCP_ESTABLISHED:
327800bd 420 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 421 con_sock_state_connected(con);
31b8006e
SW		 422 queue_con(con);
423 break;
d3002b97
AE		 424 default: /* Everything else is uninteresting */
425 break;
31b8006e
SW		 426 }
427}
428
429/*
430 * set up socket callbacks
431 */
432static void set_sock_callbacks(struct socket *sock,
433 struct ceph_connection *con)
434{
435 struct sock *sk = sock->sk;
bd406145 436 sk->sk_user_data = con;
327800bd
AE		 437 sk->sk_data_ready = ceph_sock_data_ready;
438 sk->sk_write_space = ceph_sock_write_space;
439 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 440}
441
442
443/*
444 * socket helpers
445 */
446
447/*
448 * initiate connection to a remote socket.
449 */
41617d0c 450static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 451{
f91d3471 452 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e 453 struct socket *sock;
633ee407 454 unsigned int noio_flag;
31b8006e
SW		 455 int ret;
456
457 BUG_ON(con->sock);
633ee407
ID		 458
459 /* sock_create_kern() allocates with GFP_KERNEL */
460 noio_flag = memalloc_noio_save();
757856d2 461 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
eeb1bd5c 462 SOCK_STREAM, IPPROTO_TCP, &sock);
633ee407 463 memalloc_noio_restore(noio_flag);
31b8006e 464 if (ret)
41617d0c 465 return ret;
6d7fdb0a 466 sock->sk->sk_allocation = GFP_NOFS;
31b8006e 467
a6a5349d
SW		 468#ifdef CONFIG_LOCKDEP
469 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
470#endif
471
31b8006e
SW		 472 set_sock_callbacks(sock, con);
473
3d14c5d2 474 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 475
89a86be0 476 con_sock_state_connecting(con);
f91d3471
SW		 477 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
478 O_NONBLOCK);
31b8006e
SW		 479 if (ret == -EINPROGRESS) {
480 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 481 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 482 sock->sk->sk_state);
a5bc3129 483 } else if (ret < 0) {
31b8006e 484 pr_err("connect %s error %d\n",
3d14c5d2 485 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 486 sock_release(sock);
41617d0c 487 return ret;
a5bc3129 488 }
89baaa57 489
859bff51 490 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
ba988f87
CH		 491 int optval = 1;
492
493 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
494 (char *)&optval, sizeof(optval));
495 if (ret)
496 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
497 ret);
498 }
499
a5bc3129 500 con->sock = sock;
41617d0c 501 return 0;
31b8006e
SW		 502}
503
e5c93883
ID		 504/*
505 * If @buf is NULL, discard up to @len bytes.
506 */
31b8006e
SW		 507static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
508{
509 struct kvec iov = {buf, len};
510 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 511 int r;
31b8006e 512
e5c93883
ID		 513 if (!buf)
514 msg.msg_flags |= MSG_TRUNC;
515
aa563d7b 516 iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, len);
100803a8 517 r = sock_recvmsg(sock, &msg, msg.msg_flags);
98bdb0aa
SW		 518 if (r == -EAGAIN)
519 r = 0;
520 return r;
31b8006e
SW		 521}
522
afb3d90e
AE		 523static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
524 int page_offset, size_t length)
525{
100803a8
AV		 526 struct bio_vec bvec = {
527 .bv_page = page,
528 .bv_offset = page_offset,
529 .bv_len = length
530 };
531 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
532 int r;
afb3d90e
AE		 533
534 BUG_ON(page_offset + length > PAGE_SIZE);
aa563d7b 535 iov_iter_bvec(&msg.msg_iter, READ, &bvec, 1, length);
100803a8
AV		 536 r = sock_recvmsg(sock, &msg, msg.msg_flags);
537 if (r == -EAGAIN)
538 r = 0;
539 return r;
afb3d90e
AE		 540}
541
31b8006e
SW		 542/*
543 * write something. @more is true if caller will be sending more data
544 * shortly.
545 */
546static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
547 size_t kvlen, size_t len, int more)
548{
549 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 550 int r;
31b8006e
SW		 551
552 if (more)
553 msg.msg_flags |= MSG_MORE;
554 else
555 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
556
42961d23
SW		 557 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
558 if (r == -EAGAIN)
559 r = 0;
560 return r;
31b8006e
SW		 561}
562
178eda29 563static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 564 int offset, size_t size, bool more)
31739139
AE		 565{
566 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
567 int ret;
568
569 ret = kernel_sendpage(sock, page, offset, size, flags);
570 if (ret == -EAGAIN)
571 ret = 0;
572
573 return ret;
574}
575
178eda29
CC		 576static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
577 int offset, size_t size, bool more)
578{
61ff6e9b
AV		 579 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
580 struct bio_vec bvec;
178eda29 581 int ret;
178eda29
CC		 582
583 /* sendpage cannot properly handle pages with page_count == 0,
584 * we need to fallback to sendmsg if that's the case */
585 if (page_count(page) >= 1)
586 return __ceph_tcp_sendpage(sock, page, offset, size, more);
587
61ff6e9b
AV		 588 bvec.bv_page = page;
589 bvec.bv_offset = offset;
590 bvec.bv_len = size;
591
592 if (more)
593 msg.msg_flags |= MSG_MORE;
594 else
595 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
596
aa563d7b 597 iov_iter_bvec(&msg.msg_iter, WRITE, &bvec, 1, size);
61ff6e9b
AV		 598 ret = sock_sendmsg(sock, &msg);
599 if (ret == -EAGAIN)
600 ret = 0;
178eda29
CC		 601
602 return ret;
603}
31b8006e
SW		 604
605/*
606 * Shutdown/close the socket for the given connection.
607 */
608static int con_close_socket(struct ceph_connection *con)
609{
8007b8d6 610 int rc = 0;
31b8006e
SW		 611
612 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 613 if (con->sock) {
614 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
615 sock_release(con->sock);
616 con->sock = NULL;
617 }
456ea468
AE		 618
619 /*
4a861692 620 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 621 * independent of the connection mutex, and we could have
622 * received a socket close event before we had the chance to
623 * shut the socket down.
624 */
c9ffc77a 625 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 626
ce2c8903 627 con_sock_state_closed(con);
31b8006e
SW		 628 return rc;
629}
630
631/*
632 * Reset a connection. Discard all incoming and outgoing messages
633 * and clear *_seq state.
634 */
635static void ceph_msg_remove(struct ceph_msg *msg)
636{
637 list_del_init(&msg->list_head);
38941f80 638
31b8006e
SW		 639 ceph_msg_put(msg);
640}
641static void ceph_msg_remove_list(struct list_head *head)
642{
643 while (!list_empty(head)) {
644 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
645 list_head);
646 ceph_msg_remove(msg);
647 }
648}
649
650static void reset_connection(struct ceph_connection *con)
651{
652 /* reset connection, out_queue, msg_ and connect_seq */
653 /* discard existing out_queue and msg_seq */
0fa6ebc6 654 dout("reset_connection %p\n", con);
31b8006e
SW		 655 ceph_msg_remove_list(&con->out_queue);
656 ceph_msg_remove_list(&con->out_sent);
657
cf3e5c40 658 if (con->in_msg) {
38941f80 659 BUG_ON(con->in_msg->con != con);
cf3e5c40
SW		 660 ceph_msg_put(con->in_msg);
661 con->in_msg = NULL;
662 }
663
31b8006e
SW		 664 con->connect_seq = 0;
665 con->out_seq = 0;
c86a2930 666 if (con->out_msg) {
583d0fef 667 BUG_ON(con->out_msg->con != con);
c86a2930
SW		 668 ceph_msg_put(con->out_msg);
669 con->out_msg = NULL;
670 }
31b8006e 671 con->in_seq = 0;
0e0d5e0c 672 con->in_seq_acked = 0;
67645d76
ID		 673
674 con->out_skip = 0;
31b8006e
SW		 675}
676
677/*
678 * mark a peer down. drop any open connections.
679 */
680void ceph_con_close(struct ceph_connection *con)
681{
8c50c817 682 mutex_lock(&con->mutex);
3d14c5d2
YS		 683 dout("con_close %p peer %s\n", con,
684 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 685 con->state = CON_STATE_CLOSED;
a5988c49 686
c9ffc77a
AE		 687 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
688 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
689 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
690 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 691
31b8006e 692 reset_connection(con);
6f2bc3ff 693 con->peer_global_seq = 0;
37ab77ac 694 cancel_con(con);
ee76e073 695 con_close_socket(con);
ec302645 696 mutex_unlock(&con->mutex);
31b8006e 697}
3d14c5d2 698EXPORT_SYMBOL(ceph_con_close);
31b8006e 699
31b8006e
SW		 700/*
701 * Reopen a closed connection, with a new peer address.
702 */
b7a9e5dd
SW		 703void ceph_con_open(struct ceph_connection *con,
704 __u8 entity_type, __u64 entity_num,
705 struct ceph_entity_addr *addr)
31b8006e 706{
5469155f 707 mutex_lock(&con->mutex);
3d14c5d2 708 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 709
122070a2 710 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 711 con->state = CON_STATE_PREOPEN;
a5988c49 712
b7a9e5dd
SW		 713 con->peer_name.type = (__u8) entity_type;
714 con->peer_name.num = cpu_to_le64(entity_num);
715
31b8006e 716 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 717 con->delay = 0; /* reset backoff memory */
5469155f 718 mutex_unlock(&con->mutex);
31b8006e
SW		 719 queue_con(con);
720}
3d14c5d2 721EXPORT_SYMBOL(ceph_con_open);
31b8006e 722
87b315a5
SW		 723/*
724 * return true if this connection ever successfully opened
725 */
726bool ceph_con_opened(struct ceph_connection *con)
727{
728 return con->connect_seq > 0;
729}
730
31b8006e
SW		 731/*
732 * initialize a new connection.
733 */
1bfd89f4
AE		 734void ceph_con_init(struct ceph_connection *con, void *private,
735 const struct ceph_connection_operations *ops,
b7a9e5dd 736 struct ceph_messenger *msgr)
31b8006e
SW		 737{
738 dout("con_init %p\n", con);
739 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 740 con->private = private;
741 con->ops = ops;
31b8006e 742 con->msgr = msgr;
ce2c8903
AE		 743
744 con_sock_state_init(con);
745
ec302645 746 mutex_init(&con->mutex);
31b8006e
SW		 747 INIT_LIST_HEAD(&con->out_queue);
748 INIT_LIST_HEAD(&con->out_sent);
68931622 749 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
a5988c49 750
8dacc7da 751 con->state = CON_STATE_CLOSED;
31b8006e 752}
3d14c5d2 753EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 754
755
756/*
757 * We maintain a global counter to order connection attempts. Get
758 * a unique seq greater than @gt.
759 */
760static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
761{
762 u32 ret;
763
764 spin_lock(&msgr->global_seq_lock);
765 if (msgr->global_seq < gt)
766 msgr->global_seq = gt;
767 ret = ++msgr->global_seq;
768 spin_unlock(&msgr->global_seq_lock);
769 return ret;
770}
771
e2200423 772static void con_out_kvec_reset(struct ceph_connection *con)
859eb799 773{
67645d76
ID		 774 BUG_ON(con->out_skip);
775
859eb799
AE		 776 con->out_kvec_left = 0;
777 con->out_kvec_bytes = 0;
778 con->out_kvec_cur = &con->out_kvec[0];
779}
780
e2200423 781static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 782 size_t size, void *data)
783{
67645d76 784 int index = con->out_kvec_left;
859eb799 785
67645d76 786 BUG_ON(con->out_skip);
859eb799
AE		 787 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
788
789 con->out_kvec[index].iov_len = size;
790 con->out_kvec[index].iov_base = data;
791 con->out_kvec_left++;
792 con->out_kvec_bytes += size;
793}
31b8006e 794
67645d76
ID		 795/*
796 * Chop off a kvec from the end. Return residual number of bytes for
797 * that kvec, i.e. how many bytes would have been written if the kvec
798 * hadn't been nuked.
799 */
800static int con_out_kvec_skip(struct ceph_connection *con)
801{
802 int off = con->out_kvec_cur - con->out_kvec;
803 int skip = 0;
804
805 if (con->out_kvec_bytes > 0) {
806 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
807 BUG_ON(con->out_kvec_bytes < skip);
808 BUG_ON(!con->out_kvec_left);
809 con->out_kvec_bytes -= skip;
810 con->out_kvec_left--;
811 }
812
813 return skip;
814}
815
df6ad1f9 816#ifdef CONFIG_BLOCK
6aaa4511
AE		 817
818/*
819 * For a bio data item, a piece is whatever remains of the next
820 * entry in the current bio iovec, or the first entry in the next
821 * bio in the list.
822 */
8ae4f4f5 823static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 824 size_t length)
6aaa4511 825{
8ae4f4f5 826 struct ceph_msg_data *data = cursor->data;
5359a17d 827 struct ceph_bio_iter *it = &cursor->bio_iter;
6aaa4511 828
5359a17d
ID		 829 cursor->resid = min_t(size_t, length, data->bio_length);
830 *it = data->bio_pos;
831 if (cursor->resid < it->iter.bi_size)
832 it->iter.bi_size = cursor->resid;
6aaa4511 833
5359a17d
ID		 834 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
835 cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
6aaa4511
AE		 836}
837
8ae4f4f5 838static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 839 size_t *page_offset,
840 size_t *length)
841{
5359a17d
ID		 842 struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
843 cursor->bio_iter.iter);
6aaa4511 844
5359a17d
ID		 845 *page_offset = bv.bv_offset;
846 *length = bv.bv_len;
847 return bv.bv_page;
6aaa4511
AE		 848}
849
8ae4f4f5
AE		 850static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
851 size_t bytes)
6aaa4511 852{
5359a17d 853 struct ceph_bio_iter *it = &cursor->bio_iter;
6aaa4511 854
5359a17d
ID		 855 BUG_ON(bytes > cursor->resid);
856 BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
25aff7c5 857 cursor->resid -= bytes;
5359a17d 858 bio_advance_iter(it->bio, &it->iter, bytes);
f38a5181 859
5359a17d
ID		 860 if (!cursor->resid) {
861 BUG_ON(!cursor->last_piece);
862 return false; /* no more data */
863 }
f38a5181 864
5359a17d 865 if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done))
6aaa4511
AE		 866 return false; /* more bytes to process in this segment */
867
5359a17d
ID		 868 if (!it->iter.bi_size) {
869 it->bio = it->bio->bi_next;
870 it->iter = it->bio->bi_iter;
871 if (cursor->resid < it->iter.bi_size)
872 it->iter.bi_size = cursor->resid;
25aff7c5 873 }
6aaa4511 874
5359a17d
ID		 875 BUG_ON(cursor->last_piece);
876 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
877 cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
6aaa4511
AE		 878 return true;
879}
ea96571f 880#endif /* CONFIG_BLOCK */
df6ad1f9 881
b9e281c2
ID		 882static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
883 size_t length)
884{
885 struct ceph_msg_data *data = cursor->data;
886 struct bio_vec *bvecs = data->bvec_pos.bvecs;
887
888 cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
889 cursor->bvec_iter = data->bvec_pos.iter;
890 cursor->bvec_iter.bi_size = cursor->resid;
891
892 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
893 cursor->last_piece =
894 cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
895}
896
897static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
898 size_t *page_offset,
899 size_t *length)
900{
901 struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
902 cursor->bvec_iter);
903
904 *page_offset = bv.bv_offset;
905 *length = bv.bv_len;
906 return bv.bv_page;
907}
908
909static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
910 size_t bytes)
911{
912 struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
913
914 BUG_ON(bytes > cursor->resid);
915 BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
916 cursor->resid -= bytes;
917 bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
918
919 if (!cursor->resid) {
920 BUG_ON(!cursor->last_piece);
921 return false; /* no more data */
922 }
923
924 if (!bytes || cursor->bvec_iter.bi_bvec_done)
925 return false; /* more bytes to process in this segment */
926
927 BUG_ON(cursor->last_piece);
928 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
929 cursor->last_piece =
930 cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
931 return true;
932}
933
e766d7b5
AE		 934/*
935 * For a page array, a piece comes from the first page in the array
936 * that has not already been fully consumed.
937 */
8ae4f4f5 938static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 939 size_t length)
e766d7b5 940{
8ae4f4f5 941 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 942 int page_count;
943
944 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
945
946 BUG_ON(!data->pages);
947 BUG_ON(!data->length);
948
ca8b3a69 949 cursor->resid = min(length, data->length);
e766d7b5 950 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 951 cursor->page_offset = data->alignment & ~PAGE_MASK;
952 cursor->page_index = 0;
56fc5659
AE		 953 BUG_ON(page_count > (int)USHRT_MAX);
954 cursor->page_count = (unsigned short)page_count;
955 BUG_ON(length > SIZE_MAX - cursor->page_offset);
5f740d7e 956 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 957}
958
8ae4f4f5
AE		 959static struct page *
960ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
961 size_t *page_offset, size_t *length)
e766d7b5 962{
8ae4f4f5 963 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 964
965 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
966
967 BUG_ON(cursor->page_index >= cursor->page_count);
968 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 969
970 *page_offset = cursor->page_offset;
25aff7c5 971 if (cursor->last_piece)
e766d7b5 972 *length = cursor->resid;
25aff7c5 973 else
e766d7b5 974 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 975
976 return data->pages[cursor->page_index];
977}
978
8ae4f4f5 979static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 980 size_t bytes)
981{
8ae4f4f5 982 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 983
984 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 985
986 /* Advance the cursor page offset */
987
988 cursor->resid -= bytes;
5df521b1
AE		 989 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
990 if (!bytes || cursor->page_offset)
e766d7b5
AE		 991 return false; /* more bytes to process in the current page */
992
d90deda6
YZ		 993 if (!cursor->resid)
994 return false; /* no more data */
995
5df521b1 996 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 997
998 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 999 cursor->page_index++;
25aff7c5 1000 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 1001
1002 return true;
1003}
1004
fe38a2b6 1005/*
dd236fcb
AE		 1006 * For a pagelist, a piece is whatever remains to be consumed in the
1007 * first page in the list, or the front of the next page.
fe38a2b6 1008 */
8ae4f4f5
AE		 1009static void
1010ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 1011 size_t length)
fe38a2b6 1012{
8ae4f4f5 1013 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1014 struct ceph_pagelist *pagelist;
1015 struct page *page;
1016
dd236fcb 1017 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 1018
1019 pagelist = data->pagelist;
1020 BUG_ON(!pagelist);
25aff7c5
AE		 1021
1022 if (!length)
fe38a2b6
AE		 1023 return; /* pagelist can be assigned but empty */
1024
1025 BUG_ON(list_empty(&pagelist->head));
1026 page = list_first_entry(&pagelist->head, struct page, lru);
1027
ca8b3a69 1028 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 1029 cursor->page = page;
1030 cursor->offset = 0;
a51b272e 1031 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1032}
1033
8ae4f4f5
AE		 1034static struct page *
1035ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1036 size_t *page_offset, size_t *length)
fe38a2b6 1037{
8ae4f4f5 1038 struct ceph_msg_data *data = cursor->data;
fe38a2b6 1039 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 1040
1041 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1042
1043 pagelist = data->pagelist;
1044 BUG_ON(!pagelist);
1045
1046 BUG_ON(!cursor->page);
25aff7c5 1047 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 1048
5df521b1 1049 /* offset of first page in pagelist is always 0 */
fe38a2b6 1050 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 1051 if (cursor->last_piece)
25aff7c5
AE		 1052 *length = cursor->resid;
1053 else
1054 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1055
8ae4f4f5 1056 return cursor->page;
fe38a2b6
AE		 1057}
1058
8ae4f4f5 1059static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1060 size_t bytes)
fe38a2b6 1061{
8ae4f4f5 1062 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1063 struct ceph_pagelist *pagelist;
1064
1065 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1066
1067 pagelist = data->pagelist;
1068 BUG_ON(!pagelist);
25aff7c5
AE		 1069
1070 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1071 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1072
1073 /* Advance the cursor offset */
1074
25aff7c5 1075 cursor->resid -= bytes;
fe38a2b6 1076 cursor->offset += bytes;
5df521b1 1077 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1078 if (!bytes || cursor->offset & ~PAGE_MASK)
1079 return false; /* more bytes to process in the current page */
1080
d90deda6
YZ		 1081 if (!cursor->resid)
1082 return false; /* no more data */
1083
fe38a2b6
AE		 1084 /* Move on to the next page */
1085
1086 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
17ddc49b 1087 cursor->page = list_next_entry(cursor->page, lru);
25aff7c5 1088 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1089
1090 return true;
1091}
1092
dd236fcb
AE		 1093/*
1094 * Message data is handled (sent or received) in pieces, where each
1095 * piece resides on a single page. The network layer might not
1096 * consume an entire piece at once. A data item's cursor keeps
1097 * track of which piece is next to process and how much remains to
1098 * be processed in that piece. It also tracks whether the current
1099 * piece is the last one in the data item.
1100 */
ca8b3a69 1101static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1102{
ca8b3a69 1103 size_t length = cursor->total_resid;
8ae4f4f5 1104
8ae4f4f5 1105 switch (cursor->data->type) {
dd236fcb 1106 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1107 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1108 break;
e766d7b5 1109 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1110 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1111 break;
dd236fcb
AE		 1112#ifdef CONFIG_BLOCK
1113 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1114 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1115 break;
dd236fcb 1116#endif /* CONFIG_BLOCK */
b9e281c2
ID		 1117 case CEPH_MSG_DATA_BVECS:
1118 ceph_msg_data_bvecs_cursor_init(cursor, length);
1119 break;
6aaa4511 1120 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1121 default:
1122 /* BUG(); */
1123 break;
1124 }
8ae4f4f5 1125 cursor->need_crc = true;
dd236fcb
AE		 1126}
1127
ca8b3a69
AE		 1128static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1129{
1130 struct ceph_msg_data_cursor *cursor = &msg->cursor;
ca8b3a69
AE		 1131
1132 BUG_ON(!length);
1133 BUG_ON(length > msg->data_length);
0d9c1ab3 1134 BUG_ON(!msg->num_data_items);
ca8b3a69 1135
ca8b3a69 1136 cursor->total_resid = length;
0d9c1ab3 1137 cursor->data = msg->data;
ca8b3a69
AE		 1138
1139 __ceph_msg_data_cursor_init(cursor);
1140}
1141
dd236fcb
AE		 1142/*
1143 * Return the page containing the next piece to process for a given
1144 * data item, and supply the page offset and length of that piece.
1145 * Indicate whether this is the last piece in this data item.
1146 */
8ae4f4f5
AE		 1147static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1148 size_t *page_offset, size_t *length,
dd236fcb
AE		 1149 bool *last_piece)
1150{
1151 struct page *page;
1152
8ae4f4f5 1153 switch (cursor->data->type) {
dd236fcb 1154 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1155 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1156 break;
e766d7b5 1157 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1158 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1159 break;
dd236fcb
AE		 1160#ifdef CONFIG_BLOCK
1161 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1162 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1163 break;
dd236fcb 1164#endif /* CONFIG_BLOCK */
b9e281c2
ID		 1165 case CEPH_MSG_DATA_BVECS:
1166 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1167 break;
6aaa4511 1168 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1169 default:
1170 page = NULL;
1171 break;
1172 }
5359a17d 1173
dd236fcb
AE		 1174 BUG_ON(!page);
1175 BUG_ON(*page_offset + *length > PAGE_SIZE);
1176 BUG_ON(!*length);
5359a17d 1177 BUG_ON(*length > cursor->resid);
dd236fcb 1178 if (last_piece)
8ae4f4f5 1179 *last_piece = cursor->last_piece;
dd236fcb
AE		 1180
1181 return page;
1182}
1183
1184/*
1185 * Returns true if the result moves the cursor on to the next piece
1186 * of the data item.
1187 */
1759f7b0
ID		 1188static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1189 size_t bytes)
dd236fcb
AE		 1190{
1191 bool new_piece;
1192
25aff7c5 1193 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1194 switch (cursor->data->type) {
dd236fcb 1195 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1196 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1197 break;
e766d7b5 1198 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1199 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1200 break;
dd236fcb
AE		 1201#ifdef CONFIG_BLOCK
1202 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1203 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1204 break;
dd236fcb 1205#endif /* CONFIG_BLOCK */
b9e281c2
ID		 1206 case CEPH_MSG_DATA_BVECS:
1207 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1208 break;
6aaa4511 1209 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1210 default:
1211 BUG();
1212 break;
1213 }
ca8b3a69 1214 cursor->total_resid -= bytes;
dd236fcb 1215
ca8b3a69
AE		 1216 if (!cursor->resid && cursor->total_resid) {
1217 WARN_ON(!cursor->last_piece);
0d9c1ab3 1218 cursor->data++;
ca8b3a69 1219 __ceph_msg_data_cursor_init(cursor);
a51b272e 1220 new_piece = true;
ca8b3a69 1221 }
a51b272e 1222 cursor->need_crc = new_piece;
dd236fcb
AE		 1223}
1224
dbc0d3ca
ID		 1225static size_t sizeof_footer(struct ceph_connection *con)
1226{
1227 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1228 sizeof(struct ceph_msg_footer) :
1229 sizeof(struct ceph_msg_footer_old);
1230}
1231
98fa5dd8 1232static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1233{
4c59b4a2 1234 /* Initialize data cursor */
fe38a2b6 1235
8ae4f4f5 1236 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1237}
1238
31b8006e
SW		 1239/*
1240 * Prepare footer for currently outgoing message, and finish things
1241 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1242 */
859eb799 1243static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1244{
1245 struct ceph_msg *m = con->out_msg;
1246
fd154f3c
AE		 1247 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1248
31b8006e 1249 dout("prepare_write_message_footer %p\n", con);
89f08173 1250 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
33d07337
YZ		 1251 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1252 if (con->ops->sign_message)
79dbd1ba 1253 con->ops->sign_message(m);
33d07337
YZ		 1254 else
1255 m->footer.sig = 0;
33d07337
YZ		 1256 } else {
1257 m->old_footer.flags = m->footer.flags;
33d07337 1258 }
31b8006e 1259 con->out_more = m->more_to_follow;
c86a2930 1260 con->out_msg_done = true;
31b8006e
SW		 1261}
1262
1263/*
1264 * Prepare headers for the next outgoing message.
1265 */
1266static void prepare_write_message(struct ceph_connection *con)
1267{
1268 struct ceph_msg *m;
a9a0c51a 1269 u32 crc;
31b8006e 1270
e2200423 1271 con_out_kvec_reset(con);
c86a2930 1272 con->out_msg_done = false;
31b8006e
SW		 1273
1274 /* Sneak an ack in there first? If we can get it into the same
1275 * TCP packet that's a good thing. */
1276 if (con->in_seq > con->in_seq_acked) {
1277 con->in_seq_acked = con->in_seq;
e2200423 1278 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1279 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1280 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1281 &con->out_temp_ack);
31b8006e
SW		 1282 }
1283
38941f80 1284 BUG_ON(list_empty(&con->out_queue));
859eb799 1285 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1286 con->out_msg = m;
38941f80 1287 BUG_ON(m->con != con);
4cf9d544
SW		 1288
1289 /* put message on sent list */
1290 ceph_msg_get(m);
1291 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1292
e84346b7
SW		 1293 /*
1294 * only assign outgoing seq # if we haven't sent this message
1295 * yet. if it is requeued, resend with it's original seq.
1296 */
1297 if (m->needs_out_seq) {
1298 m->hdr.seq = cpu_to_le64(++con->out_seq);
1299 m->needs_out_seq = false;
98ad5ebd 1300
4690faf0
ID		 1301 if (con->ops->reencode_message)
1302 con->ops->reencode_message(m);
1303 }
31b8006e 1304
98fa5dd8 1305 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1306 m, con->out_seq, le16_to_cpu(m->hdr.type),
1307 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1308 m->data_length);
98ad5ebd
ID		 1309 WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1310 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e
SW		 1311
1312 /* tag + hdr + front + middle */
e2200423 1313 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
67645d76 1314 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
e2200423 1315 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1316
31b8006e 1317 if (m->middle)
e2200423 1318 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1319 m->middle->vec.iov_base);
31b8006e 1320
67645d76 1321 /* fill in hdr crc and finalize hdr */
a9a0c51a
AE		 1322 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1323 con->out_msg->hdr.crc = cpu_to_le32(crc);
67645d76 1324 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
a9a0c51a 1325
67645d76 1326 /* fill in front and middle crc, footer */
a9a0c51a
AE		 1327 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1328 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1329 if (m->middle) {
1330 crc = crc32c(0, m->middle->vec.iov_base,
1331 m->middle->vec.iov_len);
1332 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1333 } else
31b8006e 1334 con->out_msg->footer.middle_crc = 0;
739c905b 1335 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1336 le32_to_cpu(con->out_msg->footer.front_crc),
1337 le32_to_cpu(con->out_msg->footer.middle_crc));
67645d76 1338 con->out_msg->footer.flags = 0;
31b8006e
SW		 1339
1340 /* is there a data payload? */
739c905b 1341 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1342 if (m->data_length) {
1343 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1344 con->out_more = 1; /* data + footer will follow */
1345 } else {
31b8006e 1346 /* no, queue up footer too and be done */
859eb799 1347 prepare_write_message_footer(con);
78625051 1348 }
31b8006e 1349
c9ffc77a 1350 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1351}
1352
1353/*
1354 * Prepare an ack.
1355 */
1356static void prepare_write_ack(struct ceph_connection *con)
1357{
1358 dout("prepare_write_ack %p %llu -> %llu\n", con,
1359 con->in_seq_acked, con->in_seq);
1360 con->in_seq_acked = con->in_seq;
1361
e2200423 1362 con_out_kvec_reset(con);
859eb799 1363
e2200423 1364 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1365
31b8006e 1366 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1367 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1368 &con->out_temp_ack);
1369
31b8006e 1370 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1371 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1372}
1373
3a23083b
SW		 1374/*
1375 * Prepare to share the seq during handshake
1376 */
1377static void prepare_write_seq(struct ceph_connection *con)
1378{
1379 dout("prepare_write_seq %p %llu -> %llu\n", con,
1380 con->in_seq_acked, con->in_seq);
1381 con->in_seq_acked = con->in_seq;
1382
1383 con_out_kvec_reset(con);
1384
1385 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1386 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1387 &con->out_temp_ack);
1388
1389 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1390}
1391
31b8006e
SW		 1392/*
1393 * Prepare to write keepalive byte.
1394 */
1395static void prepare_write_keepalive(struct ceph_connection *con)
1396{
1397 dout("prepare_write_keepalive %p\n", con);
e2200423 1398 con_out_kvec_reset(con);
8b9558aa 1399 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
473bd2d7 1400 struct timespec64 now;
7f61f545 1401
473bd2d7 1402 ktime_get_real_ts64(&now);
8b9558aa 1403 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
473bd2d7 1404 ceph_encode_timespec64(&con->out_temp_keepalive2, &now);
7f61f545
ID		 1405 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1406 &con->out_temp_keepalive2);
8b9558aa
YZ		 1407 } else {
1408 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1409 }
c9ffc77a 1410 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1411}
1412
1413/*
1414 * Connection negotiation.
1415 */
1416
262614c4 1417static int get_connect_authorizer(struct ceph_connection *con)
4e7a5dcd 1418{
a3530df3 1419 struct ceph_auth_handshake *auth;
262614c4 1420 int auth_proto;
b1c6b980
AE		 1421
1422 if (!con->ops->get_authorizer) {
262614c4 1423 con->auth = NULL;
b1c6b980
AE		 1424 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1425 con->out_connect.authorizer_len = 0;
262614c4 1426 return 0;
b1c6b980
AE		 1427 }
1428
262614c4 1429 auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
a3530df3 1430 if (IS_ERR(auth))
262614c4 1431 return PTR_ERR(auth);
0da5d703 1432
262614c4
ID		 1433 con->auth = auth;
1434 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1435 con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1436 return 0;
4e7a5dcd
SW		 1437}
1438
31b8006e
SW		 1439/*
1440 * We connected to a peer and are saying hello.
1441 */
e825a66d 1442static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1443{
e2200423
AE		 1444 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1445 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1446 &con->msgr->my_enc_addr);
eed0ef2c 1447
eed0ef2c 1448 con->out_more = 0;
c9ffc77a 1449 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1450}
1451
c0f56b48
ID		 1452static void __prepare_write_connect(struct ceph_connection *con)
1453{
1454 con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1455 if (con->auth)
1456 con_out_kvec_add(con, con->auth->authorizer_buf_len,
1457 con->auth->authorizer_buf);
1458
1459 con->out_more = 0;
1460 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1461}
1462
e825a66d 1463static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1464{
95c96174 1465 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1466 int proto;
262614c4 1467 int ret;
31b8006e
SW		 1468
1469 switch (con->peer_name.type) {
1470 case CEPH_ENTITY_TYPE_MON:
1471 proto = CEPH_MONC_PROTOCOL;
1472 break;
1473 case CEPH_ENTITY_TYPE_OSD:
1474 proto = CEPH_OSDC_PROTOCOL;
1475 break;
1476 case CEPH_ENTITY_TYPE_MDS:
1477 proto = CEPH_MDSC_PROTOCOL;
1478 break;
1479 default:
1480 BUG();
1481 }
1482
1483 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1484 con->connect_seq, global_seq, proto);
4e7a5dcd 1485
859bff51
ID		 1486 con->out_connect.features =
1487 cpu_to_le64(from_msgr(con->msgr)->supported_features);
31b8006e
SW		 1488 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1489 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1490 con->out_connect.global_seq = cpu_to_le32(global_seq);
1491 con->out_connect.protocol_version = cpu_to_le32(proto);
1492 con->out_connect.flags = 0;
31b8006e 1493
262614c4
ID		 1494 ret = get_connect_authorizer(con);
1495 if (ret)
1496 return ret;
3da54776 1497
c0f56b48 1498 __prepare_write_connect(con);
e10c758e 1499 return 0;
31b8006e
SW		 1500}
1501
31b8006e
SW		 1502/*
1503 * write as much of pending kvecs to the socket as we can.
1504 * 1 -> done
1505 * 0 -> socket full, but more to do
1506 * <0 -> error
1507 */
1508static int write_partial_kvec(struct ceph_connection *con)
1509{
1510 int ret;
1511
1512 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1513 while (con->out_kvec_bytes > 0) {
1514 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1515 con->out_kvec_left, con->out_kvec_bytes,
1516 con->out_more);
1517 if (ret <= 0)
1518 goto out;
1519 con->out_kvec_bytes -= ret;
1520 if (con->out_kvec_bytes == 0)
1521 break; /* done */
f42299e6
AE		 1522
1523 /* account for full iov entries consumed */
1524 while (ret >= con->out_kvec_cur->iov_len) {
1525 BUG_ON(!con->out_kvec_left);
1526 ret -= con->out_kvec_cur->iov_len;
1527 con->out_kvec_cur++;
1528 con->out_kvec_left--;
1529 }
1530 /* and for a partially-consumed entry */
1531 if (ret) {
1532 con->out_kvec_cur->iov_len -= ret;
1533 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1534 }
1535 }
1536 con->out_kvec_left = 0;
31b8006e
SW		 1537 ret = 1;
1538out:
1539 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1540 con->out_kvec_bytes, con->out_kvec_left, ret);
1541 return ret; /* done! */
1542}
1543
35b62808
AE		 1544static u32 ceph_crc32c_page(u32 crc, struct page *page,
1545 unsigned int page_offset,
1546 unsigned int length)
1547{
1548 char *kaddr;
1549
1550 kaddr = kmap(page);
1551 BUG_ON(kaddr == NULL);
1552 crc = crc32c(crc, kaddr + page_offset, length);
1553 kunmap(page);
1554
1555 return crc;
1556}
31b8006e
SW		 1557/*
1558 * Write as much message data payload as we can. If we finish, queue
1559 * up the footer.
1560 * 1 -> done, footer is now queued in out_kvec[].
1561 * 0 -> socket full, but more to do
1562 * <0 -> error
1563 */
34d2d200 1564static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1565{
1566 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1567 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 1568 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
f5db90bc 1569 u32 crc;
31b8006e 1570
859a35d5 1571 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1572
0d9c1ab3 1573 if (!msg->num_data_items)
4c59b4a2
AE		 1574 return -EINVAL;
1575
5821bd8c
AE		 1576 /*
1577 * Iterate through each page that contains data to be
1578 * written, and send as much as possible for each.
1579 *
1580 * If we are calculating the data crc (the default), we will
1581 * need to map the page. If we have no pages, they have
1582 * been revoked, so use the zero page.
1583 */
f5db90bc 1584 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
45a267db 1585 while (cursor->total_resid) {
8a166d05 1586 struct page *page;
e387d525
AE		 1587 size_t page_offset;
1588 size_t length;
8a166d05 1589 bool last_piece;
f5db90bc 1590 int ret;
68b4476b 1591
45a267db
ID		 1592 if (!cursor->resid) {
1593 ceph_msg_data_advance(cursor, 0);
1594 continue;
1595 }
1596
343128ce
SB		 1597 page = ceph_msg_data_next(cursor, &page_offset, &length,
1598 &last_piece);
e387d525 1599 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
c2cfa194 1600 length, !last_piece);
f5db90bc
AE		 1601 if (ret <= 0) {
1602 if (do_datacrc)
1603 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1604
f5db90bc
AE		 1605 return ret;
1606 }
143334ff
AE		 1607 if (do_datacrc && cursor->need_crc)
1608 crc = ceph_crc32c_page(crc, page, page_offset, length);
1759f7b0 1609 ceph_msg_data_advance(cursor, (size_t)ret);
31b8006e
SW		 1610 }
1611
34d2d200 1612 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1613
1614 /* prepare and queue up footer, too */
f5db90bc
AE		 1615 if (do_datacrc)
1616 msg->footer.data_crc = cpu_to_le32(crc);
1617 else
84ca8fc8 1618 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1619 con_out_kvec_reset(con);
859eb799 1620 prepare_write_message_footer(con);
f5db90bc
AE		 1621
1622 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1623}
1624
1625/*
1626 * write some zeros
1627 */
1628static int write_partial_skip(struct ceph_connection *con)
1629{
1630 int ret;
1631
67645d76 1632 dout("%s %p %d left\n", __func__, con, con->out_skip);
31b8006e 1633 while (con->out_skip > 0) {
09cbfeaf 1634 size_t size = min(con->out_skip, (int) PAGE_SIZE);
31b8006e 1635
e1dcb128 1636 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1637 if (ret <= 0)
1638 goto out;
1639 con->out_skip -= ret;
1640 }
1641 ret = 1;
1642out:
1643 return ret;
1644}
1645
1646/*
1647 * Prepare to read connection handshake, or an ack.
1648 */
eed0ef2c
SW		 1649static void prepare_read_banner(struct ceph_connection *con)
1650{
1651 dout("prepare_read_banner %p\n", con);
1652 con->in_base_pos = 0;
1653}
1654
31b8006e
SW		 1655static void prepare_read_connect(struct ceph_connection *con)
1656{
1657 dout("prepare_read_connect %p\n", con);
1658 con->in_base_pos = 0;
1659}
1660
1661static void prepare_read_ack(struct ceph_connection *con)
1662{
1663 dout("prepare_read_ack %p\n", con);
1664 con->in_base_pos = 0;
1665}
1666
3a23083b
SW		 1667static void prepare_read_seq(struct ceph_connection *con)
1668{
1669 dout("prepare_read_seq %p\n", con);
1670 con->in_base_pos = 0;
1671 con->in_tag = CEPH_MSGR_TAG_SEQ;
1672}
1673
31b8006e
SW		 1674static void prepare_read_tag(struct ceph_connection *con)
1675{
1676 dout("prepare_read_tag %p\n", con);
1677 con->in_base_pos = 0;
1678 con->in_tag = CEPH_MSGR_TAG_READY;
1679}
1680
8b9558aa
YZ		 1681static void prepare_read_keepalive_ack(struct ceph_connection *con)
1682{
1683 dout("prepare_read_keepalive_ack %p\n", con);
1684 con->in_base_pos = 0;
1685}
1686
31b8006e
SW		 1687/*
1688 * Prepare to read a message.
1689 */
1690static int prepare_read_message(struct ceph_connection *con)
1691{
1692 dout("prepare_read_message %p\n", con);
1693 BUG_ON(con->in_msg != NULL);
1694 con->in_base_pos = 0;
1695 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1696 return 0;
1697}
1698
1699
1700static int read_partial(struct ceph_connection *con,
fd51653f 1701 int end, int size, void *object)
31b8006e 1702{
e6cee71f
AE		 1703 while (con->in_base_pos < end) {
1704 int left = end - con->in_base_pos;
31b8006e
SW		 1705 int have = size - left;
1706 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1707 if (ret <= 0)
1708 return ret;
1709 con->in_base_pos += ret;
1710 }
1711 return 1;
1712}
1713
1714
1715/*
1716 * Read all or part of the connect-side handshake on a new connection
1717 */
eed0ef2c 1718static int read_partial_banner(struct ceph_connection *con)
31b8006e 1719{
fd51653f
AE		 1720 int size;
1721 int end;
1722 int ret;
31b8006e 1723
eed0ef2c 1724 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1725
1726 /* peer's banner */
fd51653f
AE		 1727 size = strlen(CEPH_BANNER);
1728 end = size;
1729 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1730 if (ret <= 0)
1731 goto out;
fd51653f
AE		 1732
1733 size = sizeof (con->actual_peer_addr);
1734 end += size;
1735 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1736 if (ret <= 0)
1737 goto out;
fd51653f
AE		 1738
1739 size = sizeof (con->peer_addr_for_me);
1740 end += size;
1741 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1742 if (ret <= 0)
1743 goto out;
fd51653f 1744
eed0ef2c
SW		 1745out:
1746 return ret;
1747}
1748
1749static int read_partial_connect(struct ceph_connection *con)
1750{
fd51653f
AE		 1751 int size;
1752 int end;
1753 int ret;
eed0ef2c
SW		 1754
1755 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1756
fd51653f
AE		 1757 size = sizeof (con->in_reply);
1758 end = size;
1759 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1760 if (ret <= 0)
1761 goto out;
fd51653f 1762
262614c4
ID		 1763 if (con->auth) {
1764 size = le32_to_cpu(con->in_reply.authorizer_len);
130f52f2
ID		 1765 if (size > con->auth->authorizer_reply_buf_len) {
1766 pr_err("authorizer reply too big: %d > %zu\n", size,
1767 con->auth->authorizer_reply_buf_len);
1768 ret = -EINVAL;
1769 goto out;
1770 }
1771
262614c4
ID		 1772 end += size;
1773 ret = read_partial(con, end, size,
1774 con->auth->authorizer_reply_buf);
1775 if (ret <= 0)
1776 goto out;
1777 }
31b8006e 1778
4e7a5dcd
SW		 1779 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1780 con, (int)con->in_reply.tag,
1781 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1782 le32_to_cpu(con->in_reply.global_seq));
1783out:
1784 return ret;
1785}
1786
1787/*
1788 * Verify the hello banner looks okay.
1789 */
1790static int verify_hello(struct ceph_connection *con)
1791{
1792 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1793 pr_err("connect to %s got bad banner\n",
3d14c5d2 1794 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1795 con->error_msg = "protocol error, bad banner";
1796 return -1;
1797 }
1798 return 0;
1799}
1800
1801static bool addr_is_blank(struct sockaddr_storage *ss)
1802{
c44bd69c
ID		 1803 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1804 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1805
31b8006e
SW		 1806 switch (ss->ss_family) {
1807 case AF_INET:
c44bd69c 1808 return addr->s_addr == htonl(INADDR_ANY);
31b8006e 1809 case AF_INET6:
c44bd69c
ID		 1810 return ipv6_addr_any(addr6);
1811 default:
1812 return true;
31b8006e 1813 }
31b8006e
SW		 1814}
1815
1816static int addr_port(struct sockaddr_storage *ss)
1817{
1818 switch (ss->ss_family) {
1819 case AF_INET:
f28bcfbe 1820 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1821 case AF_INET6:
f28bcfbe 1822 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1823 }
1824 return 0;
1825}
1826
1827static void addr_set_port(struct sockaddr_storage *ss, int p)
1828{
1829 switch (ss->ss_family) {
1830 case AF_INET:
1831 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1832 break;
31b8006e
SW		 1833 case AF_INET6:
1834 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1835 break;
31b8006e
SW		 1836 }
1837}
1838
ee3b56f2
NW		 1839/*
1840 * Unlike other *_pton function semantics, zero indicates success.
1841 */
1842static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1843 char delim, const char **ipend)
1844{
99f0f3b2
AE		 1845 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1846 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1847
1848 memset(ss, 0, sizeof(*ss));
1849
1850 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1851 ss->ss_family = AF_INET;
1852 return 0;
1853 }
1854
1855 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1856 ss->ss_family = AF_INET6;
1857 return 0;
1858 }
1859
1860 return -EINVAL;
1861}
1862
1863/*
1864 * Extract hostname string and resolve using kernel DNS facility.
1865 */
1866#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1867static int ceph_dns_resolve_name(const char *name, size_t namelen,
1868 struct sockaddr_storage *ss, char delim, const char **ipend)
1869{
1870 const char *end, *delim_p;
1871 char *colon_p, *ip_addr = NULL;
1872 int ip_len, ret;
1873
1874 /*
1875 * The end of the hostname occurs immediately preceding the delimiter or
1876 * the port marker (':') where the delimiter takes precedence.
1877 */
1878 delim_p = memchr(name, delim, namelen);
1879 colon_p = memchr(name, ':', namelen);
1880
1881 if (delim_p && colon_p)
1882 end = delim_p < colon_p ? delim_p : colon_p;
1883 else if (!delim_p && colon_p)
1884 end = colon_p;
1885 else {
1886 end = delim_p;
1887 if (!end) /* case: hostname:/ */
1888 end = name + namelen;
1889 }
1890
1891 if (end <= name)
1892 return -EINVAL;
1893
1894 /* do dns_resolve upcall */
1895 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1896 if (ip_len > 0)
1897 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1898 else
1899 ret = -ESRCH;
1900
1901 kfree(ip_addr);
1902
1903 *ipend = end;
1904
1905 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1906 ret, ret ? "failed" : ceph_pr_addr(ss));
1907
1908 return ret;
1909}
1910#else
1911static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1912 struct sockaddr_storage *ss, char delim, const char **ipend)
1913{
1914 return -EINVAL;
1915}
1916#endif
1917
1918/*
1919 * Parse a server name (IP or hostname). If a valid IP address is not found
1920 * then try to extract a hostname to resolve using userspace DNS upcall.
1921 */
1922static int ceph_parse_server_name(const char *name, size_t namelen,
1923 struct sockaddr_storage *ss, char delim, const char **ipend)
1924{
1925 int ret;
1926
1927 ret = ceph_pton(name, namelen, ss, delim, ipend);
1928 if (ret)
1929 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1930
1931 return ret;
1932}
1933
31b8006e
SW		 1934/*
1935 * Parse an ip[:port] list into an addr array. Use the default
1936 * monitor port if a port isn't specified.
1937 */
1938int ceph_parse_ips(const char *c, const char *end,
1939 struct ceph_entity_addr *addr,
1940 int max_count, int *count)
1941{
ee3b56f2 1942 int i, ret = -EINVAL;
31b8006e
SW		 1943 const char *p = c;
1944
1945 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1946 for (i = 0; i < max_count; i++) {
1947 const char *ipend;
1948 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1949 int port;
39139f64
SW		 1950 char delim = ',';
1951
1952 if (*p == '[') {
1953 delim = ']';
1954 p++;
1955 }
31b8006e 1956
ee3b56f2
NW		 1957 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1958 if (ret)
31b8006e 1959 goto bad;
ee3b56f2
NW		 1960 ret = -EINVAL;
1961
31b8006e
SW		 1962 p = ipend;
1963
39139f64
SW		 1964 if (delim == ']') {
1965 if (*p != ']') {
1966 dout("missing matching ']'\n");
1967 goto bad;
1968 }
1969 p++;
1970 }
1971
31b8006e
SW		 1972 /* port? */
1973 if (p < end && *p == ':') {
1974 port = 0;
1975 p++;
1976 while (p < end && *p >= '0' && *p <= '9') {
1977 port = (port * 10) + (*p - '0');
1978 p++;
1979 }
f48db1e9
ID		 1980 if (port == 0)
1981 port = CEPH_MON_PORT;
1982 else if (port > 65535)
31b8006e
SW		 1983 goto bad;
1984 } else {
1985 port = CEPH_MON_PORT;
1986 }
1987
1988 addr_set_port(ss, port);
1989
3d14c5d2 1990 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1991
1992 if (p == end)
1993 break;
1994 if (*p != ',')
1995 goto bad;
1996 p++;
1997 }
1998
1999 if (p != end)
2000 goto bad;
2001
2002 if (count)
2003 *count = i + 1;
2004 return 0;
2005
2006bad:
39139f64 2007 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 2008 return ret;
31b8006e 2009}
3d14c5d2 2010EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 2011
eed0ef2c 2012static int process_banner(struct ceph_connection *con)
31b8006e 2013{
eed0ef2c 2014 dout("process_banner on %p\n", con);
31b8006e
SW		 2015
2016 if (verify_hello(con) < 0)
2017 return -1;
2018
63f2d211
SW		 2019 ceph_decode_addr(&con->actual_peer_addr);
2020 ceph_decode_addr(&con->peer_addr_for_me);
2021
31b8006e
SW		 2022 /*
2023 * Make sure the other end is who we wanted. note that the other
2024 * end may not yet know their ip address, so if it's 0.0.0.0, give
2025 * them the benefit of the doubt.
2026 */
103e2d3a
SW		 2027 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2028 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 2029 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2030 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 2031 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2032 ceph_pr_addr(&con->peer_addr.in_addr),
2033 (int)le32_to_cpu(con->peer_addr.nonce),
2034 ceph_pr_addr(&con->actual_peer_addr.in_addr),
2035 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 2036 con->error_msg = "wrong peer at address";
31b8006e
SW		 2037 return -1;
2038 }
2039
2040 /*
2041 * did we learn our address?
2042 */
2043 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2044 int port = addr_port(&con->msgr->inst.addr.in_addr);
2045
2046 memcpy(&con->msgr->inst.addr.in_addr,
2047 &con->peer_addr_for_me.in_addr,
2048 sizeof(con->peer_addr_for_me.in_addr));
2049 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 2050 encode_my_addr(con->msgr);
eed0ef2c 2051 dout("process_banner learned my addr is %s\n",
3d14c5d2 2052 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 2053 }
2054
eed0ef2c
SW		 2055 return 0;
2056}
2057
2058static int process_connect(struct ceph_connection *con)
2059{
859bff51
ID		 2060 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2061 u64 req_feat = from_msgr(con->msgr)->required_features;
dcbbd97c 2062 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 2063 int ret;
04a419f9 2064
eed0ef2c
SW		 2065 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2066
262614c4 2067 if (con->auth) {
5c056fdc
ID		 2068 /*
2069 * Any connection that defines ->get_authorizer()
6daca13d
ID		 2070 * should also define ->add_authorizer_challenge() and
2071 * ->verify_authorizer_reply().
2072 *
5c056fdc
ID		 2073 * See get_connect_authorizer().
2074 */
6daca13d
ID		 2075 if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
2076 ret = con->ops->add_authorizer_challenge(
2077 con, con->auth->authorizer_reply_buf,
2078 le32_to_cpu(con->in_reply.authorizer_len));
2079 if (ret < 0)
2080 return ret;
2081
2082 con_out_kvec_reset(con);
2083 __prepare_write_connect(con);
2084 prepare_read_connect(con);
2085 return 0;
2086 }
2087
0dde5848 2088 ret = con->ops->verify_authorizer_reply(con);
5c056fdc
ID		 2089 if (ret < 0) {
2090 con->error_msg = "bad authorize reply";
2091 return ret;
2092 }
2093 }
2094
31b8006e 2095 switch (con->in_reply.tag) {
04a419f9
SW		 2096 case CEPH_MSGR_TAG_FEATURES:
2097 pr_err("%s%lld %s feature set mismatch,"
2098 " my %llx < server's %llx, missing %llx\n",
2099 ENTITY_NAME(con->peer_name),
3d14c5d2 2100 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2101 sup_feat, server_feat, server_feat & ~sup_feat);
2102 con->error_msg = "missing required protocol features";
0fa6ebc6 2103 reset_connection(con);
04a419f9
SW		 2104 return -1;
2105
31b8006e 2106 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2107 pr_err("%s%lld %s protocol version mismatch,"
2108 " my %d != server's %d\n",
2109 ENTITY_NAME(con->peer_name),
3d14c5d2 2110 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2111 le32_to_cpu(con->out_connect.protocol_version),
2112 le32_to_cpu(con->in_reply.protocol_version));
2113 con->error_msg = "protocol version mismatch";
0fa6ebc6 2114 reset_connection(con);
31b8006e
SW		 2115 return -1;
2116
4e7a5dcd
SW		 2117 case CEPH_MSGR_TAG_BADAUTHORIZER:
2118 con->auth_retry++;
2119 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2120 con->auth_retry);
2121 if (con->auth_retry == 2) {
2122 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2123 return -1;
2124 }
6d4221b5 2125 con_out_kvec_reset(con);
e825a66d 2126 ret = prepare_write_connect(con);
0da5d703
SW		 2127 if (ret < 0)
2128 return ret;
63733a0f 2129 prepare_read_connect(con);
4e7a5dcd 2130 break;
31b8006e
SW		 2131
2132 case CEPH_MSGR_TAG_RESETSESSION:
2133 /*
2134 * If we connected with a large connect_seq but the peer
2135 * has no record of a session with us (no connection, or
2136 * connect_seq == 0), they will send RESETSESION to indicate
2137 * that they must have reset their session, and may have
2138 * dropped messages.
2139 */
2140 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2141 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2142 pr_err("%s%lld %s connection reset\n",
2143 ENTITY_NAME(con->peer_name),
3d14c5d2 2144 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2145 reset_connection(con);
6d4221b5 2146 con_out_kvec_reset(con);
5a0f8fdd
AE		 2147 ret = prepare_write_connect(con);
2148 if (ret < 0)
2149 return ret;
31b8006e
SW		 2150 prepare_read_connect(con);
2151
2152 /* Tell ceph about it. */
ec302645 2153 mutex_unlock(&con->mutex);
31b8006e
SW		 2154 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2155 if (con->ops->peer_reset)
2156 con->ops->peer_reset(con);
ec302645 2157 mutex_lock(&con->mutex);
8dacc7da 2158 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2159 return -EAGAIN;
31b8006e
SW		 2160 break;
2161
2162 case CEPH_MSGR_TAG_RETRY_SESSION:
2163 /*
2164 * If we sent a smaller connect_seq than the peer has, try
2165 * again with a larger value.
2166 */
5bdca4e0 2167 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2168 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2169 le32_to_cpu(con->in_reply.connect_seq));
2170 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2171 con_out_kvec_reset(con);
5a0f8fdd
AE		 2172 ret = prepare_write_connect(con);
2173 if (ret < 0)
2174 return ret;
31b8006e
SW		 2175 prepare_read_connect(con);
2176 break;
2177
2178 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2179 /*
2180 * If we sent a smaller global_seq than the peer has, try
2181 * again with a larger value.
2182 */
eed0ef2c 2183 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2184 con->peer_global_seq,
5bdca4e0 2185 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2186 get_global_seq(con->msgr,
5bdca4e0 2187 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2188 con_out_kvec_reset(con);
5a0f8fdd
AE		 2189 ret = prepare_write_connect(con);
2190 if (ret < 0)
2191 return ret;
31b8006e
SW		 2192 prepare_read_connect(con);
2193 break;
2194
3a23083b 2195 case CEPH_MSGR_TAG_SEQ:
31b8006e 2196 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2197 if (req_feat & ~server_feat) {
2198 pr_err("%s%lld %s protocol feature mismatch,"
2199 " my required %llx > server's %llx, need %llx\n",
2200 ENTITY_NAME(con->peer_name),
3d14c5d2 2201 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2202 req_feat, server_feat, req_feat & ~server_feat);
2203 con->error_msg = "missing required protocol features";
0fa6ebc6 2204 reset_connection(con);
04a419f9
SW		 2205 return -1;
2206 }
8dacc7da 2207
122070a2 2208 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2209 con->state = CON_STATE_OPEN;
20e55c4c 2210 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2211 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2212 con->connect_seq++;
aba558e2 2213 con->peer_features = server_feat;
31b8006e
SW		 2214 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2215 con->peer_global_seq,
2216 le32_to_cpu(con->in_reply.connect_seq),
2217 con->connect_seq);
2218 WARN_ON(con->connect_seq !=
2219 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2220
2221 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2222 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2223
85effe18 2224 con->delay = 0; /* reset backoff memory */
92ac41d0 2225
3a23083b
SW		 2226 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2227 prepare_write_seq(con);
2228 prepare_read_seq(con);
2229 } else {
2230 prepare_read_tag(con);
2231 }
31b8006e
SW		 2232 break;
2233
2234 case CEPH_MSGR_TAG_WAIT:
2235 /*
2236 * If there is a connection race (we are opening
2237 * connections to each other), one of us may just have
2238 * to WAIT. This shouldn't happen if we are the
2239 * client.
2240 */
04177882
SW		 2241 con->error_msg = "protocol error, got WAIT as client";
2242 return -1;
31b8006e
SW		 2243
2244 default:
31b8006e
SW		 2245 con->error_msg = "protocol error, garbage tag during connect";
2246 return -1;
2247 }
2248 return 0;
2249}
2250
2251
2252/*
2253 * read (part of) an ack
2254 */
2255static int read_partial_ack(struct ceph_connection *con)
2256{
fd51653f
AE		 2257 int size = sizeof (con->in_temp_ack);
2258 int end = size;
31b8006e 2259
fd51653f 2260 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2261}
2262
31b8006e
SW		 2263/*
2264 * We can finally discard anything that's been acked.
2265 */
2266static void process_ack(struct ceph_connection *con)
2267{
2268 struct ceph_msg *m;
2269 u64 ack = le64_to_cpu(con->in_temp_ack);
2270 u64 seq;
0a2ad541
YZ		 2271 bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2272 struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
31b8006e 2273
0a2ad541
YZ		 2274 /*
2275 * In the reconnect case, con_fault() has requeued messages
2276 * in out_sent. We should cleanup old messages according to
2277 * the reconnect seq.
2278 */
2279 while (!list_empty(list)) {
2280 m = list_first_entry(list, struct ceph_msg, list_head);
2281 if (reconnect && m->needs_out_seq)
2282 break;
31b8006e
SW		 2283 seq = le64_to_cpu(m->hdr.seq);
2284 if (seq > ack)
2285 break;
2286 dout("got ack for seq %llu type %d at %p\n", seq,
2287 le16_to_cpu(m->hdr.type), m);
4cf9d544 2288 m->ack_stamp = jiffies;
31b8006e
SW		 2289 ceph_msg_remove(m);
2290 }
0a2ad541 2291
31b8006e
SW		 2292 prepare_read_tag(con);
2293}
2294
2295
2450418c 2296static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2297 struct kvec *section,
2298 unsigned int sec_len, u32 *crc)
2450418c 2299{
68b4476b 2300 int ret, left;
2450418c
YS		 2301
2302 BUG_ON(!section);
2303
2304 while (section->iov_len < sec_len) {
2305 BUG_ON(section->iov_base == NULL);
2306 left = sec_len - section->iov_len;
2307 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2308 section->iov_len, left);
2309 if (ret <= 0)
2310 return ret;
2311 section->iov_len += ret;
2450418c 2312 }
fe3ad593
AE		 2313 if (section->iov_len == sec_len)
2314 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2315
2450418c
YS		 2316 return 1;
2317}
31b8006e 2318
34d2d200
AE		 2319static int read_partial_msg_data(struct ceph_connection *con)
2320{
2321 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2322 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 2323 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
686be208
AE		 2324 struct page *page;
2325 size_t page_offset;
2326 size_t length;
f5db90bc 2327 u32 crc = 0;
34d2d200
AE		 2328 int ret;
2329
0d9c1ab3 2330 if (!msg->num_data_items)
4c59b4a2 2331 return -EIO;
34d2d200 2332
f5db90bc
AE		 2333 if (do_datacrc)
2334 crc = con->in_data_crc;
45a267db
ID		 2335 while (cursor->total_resid) {
2336 if (!cursor->resid) {
2337 ceph_msg_data_advance(cursor, 0);
2338 continue;
2339 }
2340
343128ce 2341 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
686be208 2342 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2343 if (ret <= 0) {
2344 if (do_datacrc)
2345 con->in_data_crc = crc;
2346
686be208 2347 return ret;
f5db90bc 2348 }
686be208
AE		 2349
2350 if (do_datacrc)
f5db90bc 2351 crc = ceph_crc32c_page(crc, page, page_offset, ret);
1759f7b0 2352 ceph_msg_data_advance(cursor, (size_t)ret);
34d2d200 2353 }
f5db90bc
AE		 2354 if (do_datacrc)
2355 con->in_data_crc = crc;
34d2d200
AE		 2356
2357 return 1; /* must return > 0 to indicate success */
2358}
2359
31b8006e
SW		 2360/*
2361 * read (part of) a message.
2362 */
686be208
AE		 2363static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2364
31b8006e
SW		 2365static int read_partial_message(struct ceph_connection *con)
2366{
2367 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2368 int size;
2369 int end;
31b8006e 2370 int ret;
95c96174 2371 unsigned int front_len, middle_len, data_len;
859bff51 2372 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
33d07337 2373 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
ae18756b 2374 u64 seq;
fe3ad593 2375 u32 crc;
31b8006e
SW		 2376
2377 dout("read_partial_message con %p msg %p\n", con, m);
2378
2379 /* header */
fd51653f
AE		 2380 size = sizeof (con->in_hdr);
2381 end = size;
2382 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2383 if (ret <= 0)
2384 return ret;
fe3ad593
AE		 2385
2386 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2387 if (cpu_to_le32(crc) != con->in_hdr.crc) {
67c64eb7 2388 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
fe3ad593
AE		 2389 crc, con->in_hdr.crc);
2390 return -EBADMSG;
2391 }
2392
31b8006e
SW		 2393 front_len = le32_to_cpu(con->in_hdr.front_len);
2394 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2395 return -EIO;
2396 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2397 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2398 return -EIO;
2399 data_len = le32_to_cpu(con->in_hdr.data_len);
2400 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2401 return -EIO;
2402
ae18756b
SW		 2403 /* verify seq# */
2404 seq = le64_to_cpu(con->in_hdr.seq);
2405 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2406 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2407 ENTITY_NAME(con->peer_name),
3d14c5d2 2408 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2409 seq, con->in_seq + 1);
2410 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2411 sizeof_footer(con);
ae18756b 2412 con->in_tag = CEPH_MSGR_TAG_READY;
e7a88e82 2413 return 1;
ae18756b
SW		 2414 } else if ((s64)seq - (s64)con->in_seq > 1) {
2415 pr_err("read_partial_message bad seq %lld expected %lld\n",
2416 seq, con->in_seq + 1);
2417 con->error_msg = "bad message sequence # for incoming message";
67c64eb7 2418 return -EBADE;
ae18756b
SW		 2419 }
2420
31b8006e
SW		 2421 /* allocate message? */
2422 if (!con->in_msg) {
4740a623
SW		 2423 int skip = 0;
2424
31b8006e 2425 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2426 front_len, data_len);
4740a623
SW		 2427 ret = ceph_con_in_msg_alloc(con, &skip);
2428 if (ret < 0)
2429 return ret;
f759ebb9
AE		 2430
2431 BUG_ON(!con->in_msg ^ skip);
2450418c 2432 if (skip) {
31b8006e 2433 /* skip this message */
a79832f2 2434 dout("alloc_msg said skip message\n");
31b8006e 2435 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2436 sizeof_footer(con);
31b8006e 2437 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2438 con->in_seq++;
e7a88e82 2439 return 1;
31b8006e 2440 }
38941f80 2441
4740a623 2442 BUG_ON(!con->in_msg);
38941f80 2443 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2444 m = con->in_msg;
2445 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2446 if (m->middle)
2447 m->middle->vec.iov_len = 0;
9d7f0f13 2448
78625051 2449 /* prepare for data payload, if any */
a4107026 2450
78625051 2451 if (data_len)
98fa5dd8 2452 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2453 }
2454
2455 /* front */
2450418c
YS		 2456 ret = read_partial_message_section(con, &m->front, front_len,
2457 &con->in_front_crc);
2458 if (ret <= 0)
2459 return ret;
31b8006e
SW		 2460
2461 /* middle */
2450418c 2462 if (m->middle) {
213c99ee
SW		 2463 ret = read_partial_message_section(con, &m->middle->vec,
2464 middle_len,
2450418c 2465 &con->in_middle_crc);
31b8006e
SW		 2466 if (ret <= 0)
2467 return ret;
31b8006e
SW		 2468 }
2469
2470 /* (page) data */
34d2d200
AE		 2471 if (data_len) {
2472 ret = read_partial_msg_data(con);
2473 if (ret <= 0)
2474 return ret;
31b8006e
SW		 2475 }
2476
31b8006e 2477 /* footer */
89f08173 2478 size = sizeof_footer(con);
fd51653f
AE		 2479 end += size;
2480 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2481 if (ret <= 0)
2482 return ret;
2483
33d07337
YZ		 2484 if (!need_sign) {
2485 m->footer.flags = m->old_footer.flags;
2486 m->footer.sig = 0;
2487 }
2488
31b8006e
SW		 2489 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2490 m, front_len, m->footer.front_crc, middle_len,
2491 m->footer.middle_crc, data_len, m->footer.data_crc);
2492
2493 /* crc ok? */
2494 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2495 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2496 m, con->in_front_crc, m->footer.front_crc);
2497 return -EBADMSG;
2498 }
2499 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2500 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2501 m, con->in_middle_crc, m->footer.middle_crc);
2502 return -EBADMSG;
2503 }
bca064d2 2504 if (do_datacrc &&
31b8006e
SW		 2505 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2506 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2507 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2508 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2509 return -EBADMSG;
2510 }
2511
33d07337 2512 if (need_sign && con->ops->check_message_signature &&
79dbd1ba 2513 con->ops->check_message_signature(m)) {
33d07337
YZ		 2514 pr_err("read_partial_message %p signature check failed\n", m);
2515 return -EBADMSG;
2516 }
2517
31b8006e
SW		 2518 return 1; /* done! */
2519}
2520
2521/*
2522 * Process message. This happens in the worker thread. The callback should
2523 * be careful not to do anything that waits on other incoming messages or it
2524 * may deadlock.
2525 */
2526static void process_message(struct ceph_connection *con)
2527{
583d0fef 2528 struct ceph_msg *msg = con->in_msg;
31b8006e 2529
38941f80 2530 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2531 con->in_msg = NULL;
2532
2533 /* if first message, set peer_name */
2534 if (con->peer_name.type == 0)
dbad185d 2535 con->peer_name = msg->hdr.src;
31b8006e 2536
31b8006e 2537 con->in_seq++;
ec302645 2538 mutex_unlock(&con->mutex);
31b8006e
SW		 2539
2540 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2541 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2542 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2543 le16_to_cpu(msg->hdr.type),
2544 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2545 le32_to_cpu(msg->hdr.front_len),
2546 le32_to_cpu(msg->hdr.data_len),
2547 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2548 con->ops->dispatch(con, msg);
ec302645
SW		 2549
2550 mutex_lock(&con->mutex);
31b8006e
SW		 2551}
2552
8b9558aa
YZ		 2553static int read_keepalive_ack(struct ceph_connection *con)
2554{
2555 struct ceph_timespec ceph_ts;
2556 size_t size = sizeof(ceph_ts);
2557 int ret = read_partial(con, size, size, &ceph_ts);
2558 if (ret <= 0)
2559 return ret;
473bd2d7 2560 ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
8b9558aa
YZ		 2561 prepare_read_tag(con);
2562 return 1;
2563}
31b8006e
SW		 2564
2565/*
2566 * Write something to the socket. Called in a worker thread when the
2567 * socket appears to be writeable and we have something ready to send.
2568 */
2569static int try_write(struct ceph_connection *con)
2570{
31b8006e
SW		 2571 int ret = 1;
2572
d59315ca 2573 dout("try_write start %p state %lu\n", con, con->state);
9c55ad1c
ID		 2574 if (con->state != CON_STATE_PREOPEN &&
2575 con->state != CON_STATE_CONNECTING &&
2576 con->state != CON_STATE_NEGOTIATING &&
2577 con->state != CON_STATE_OPEN)
2578 return 0;
31b8006e 2579
31b8006e 2580 /* open the socket first? */
8dacc7da
SW		 2581 if (con->state == CON_STATE_PREOPEN) {
2582 BUG_ON(con->sock);
2583 con->state = CON_STATE_CONNECTING;
a5988c49 2584
e2200423 2585 con_out_kvec_reset(con);
e825a66d 2586 prepare_write_banner(con);
eed0ef2c 2587 prepare_read_banner(con);
31b8006e 2588
cf3e5c40 2589 BUG_ON(con->in_msg);
31b8006e
SW		 2590 con->in_tag = CEPH_MSGR_TAG_READY;
2591 dout("try_write initiating connect on %p new state %lu\n",
2592 con, con->state);
41617d0c
AE		 2593 ret = ceph_tcp_connect(con);
2594 if (ret < 0) {
31b8006e 2595 con->error_msg = "connect error";
31b8006e
SW		 2596 goto out;
2597 }
2598 }
2599
d2935d6f
ID		 2600more:
2601 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
9c55ad1c
ID		 2602 BUG_ON(!con->sock);
2603
31b8006e 2604 /* kvec data queued? */
67645d76
ID		 2605 if (con->out_kvec_left) {
2606 ret = write_partial_kvec(con);
31b8006e 2607 if (ret <= 0)
42961d23 2608 goto out;
31b8006e 2609 }
67645d76
ID		 2610 if (con->out_skip) {
2611 ret = write_partial_skip(con);
31b8006e 2612 if (ret <= 0)
42961d23 2613 goto out;
31b8006e
SW		 2614 }
2615
2616 /* msg pages? */
2617 if (con->out_msg) {
c86a2930
SW		 2618 if (con->out_msg_done) {
2619 ceph_msg_put(con->out_msg);
2620 con->out_msg = NULL; /* we're done with this one */
2621 goto do_next;
2622 }
2623
34d2d200 2624 ret = write_partial_message_data(con);
31b8006e 2625 if (ret == 1)
d2935d6f 2626 goto more; /* we need to send the footer, too! */
31b8006e 2627 if (ret == 0)
42961d23 2628 goto out;
31b8006e 2629 if (ret < 0) {
34d2d200 2630 dout("try_write write_partial_message_data err %d\n",
31b8006e 2631 ret);
42961d23 2632 goto out;
31b8006e
SW		 2633 }
2634 }
2635
c86a2930 2636do_next:
8dacc7da 2637 if (con->state == CON_STATE_OPEN) {
8b9558aa
YZ		 2638 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2639 prepare_write_keepalive(con);
2640 goto more;
2641 }
31b8006e
SW		 2642 /* is anything else pending? */
2643 if (!list_empty(&con->out_queue)) {
2644 prepare_write_message(con);
2645 goto more;
2646 }
2647 if (con->in_seq > con->in_seq_acked) {
2648 prepare_write_ack(con);
2649 goto more;
2650 }
31b8006e
SW		 2651 }
2652
2653 /* Nothing to do! */
c9ffc77a 2654 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2655 dout("try_write nothing else to write.\n");
31b8006e
SW		 2656 ret = 0;
2657out:
42961d23 2658 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2659 return ret;
2660}
2661
31b8006e
SW		 2662/*
2663 * Read what we can from the socket.
2664 */
2665static int try_read(struct ceph_connection *con)
2666{
31b8006e
SW		 2667 int ret = -1;
2668
8dacc7da
SW		 2669more:
2670 dout("try_read start on %p state %lu\n", con, con->state);
2671 if (con->state != CON_STATE_CONNECTING &&
2672 con->state != CON_STATE_NEGOTIATING &&
2673 con->state != CON_STATE_OPEN)
31b8006e
SW		 2674 return 0;
2675
8dacc7da 2676 BUG_ON(!con->sock);
ec302645 2677
31b8006e
SW		 2678 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2679 con->in_base_pos);
0da5d703 2680
8dacc7da 2681 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2682 dout("try_read connecting\n");
2683 ret = read_partial_banner(con);
2684 if (ret <= 0)
ab166d5a 2685 goto out;
7593af92
AE		 2686 ret = process_banner(con);
2687 if (ret < 0)
2688 goto out;
2689
8dacc7da 2690 con->state = CON_STATE_NEGOTIATING;
7593af92 2691
6d4221b5
JS		 2692 /*
2693 * Received banner is good, exchange connection info.
2694 * Do not reset out_kvec, as sending our banner raced
2695 * with receiving peer banner after connect completed.
2696 */
7593af92
AE		 2697 ret = prepare_write_connect(con);
2698 if (ret < 0)
2699 goto out;
2700 prepare_read_connect(con);
2701
2702 /* Send connection info before awaiting response */
0da5d703
SW		 2703 goto out;
2704 }
2705
8dacc7da 2706 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2707 dout("try_read negotiating\n");
31b8006e
SW		 2708 ret = read_partial_connect(con);
2709 if (ret <= 0)
31b8006e 2710 goto out;
98bdb0aa
SW		 2711 ret = process_connect(con);
2712 if (ret < 0)
2713 goto out;
31b8006e
SW		 2714 goto more;
2715 }
2716
122070a2 2717 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2718
31b8006e
SW		 2719 if (con->in_base_pos < 0) {
2720 /*
2721 * skipping + discarding content.
31b8006e 2722 */
e5c93883 2723 ret = ceph_tcp_recvmsg(con->sock, NULL, -con->in_base_pos);
31b8006e 2724 if (ret <= 0)
98bdb0aa 2725 goto out;
e5c93883 2726 dout("skipped %d / %d bytes\n", ret, -con->in_base_pos);
31b8006e
SW		 2727 con->in_base_pos += ret;
2728 if (con->in_base_pos)
2729 goto more;
2730 }
2731 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2732 /*
2733 * what's next?
2734 */
2735 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2736 if (ret <= 0)
98bdb0aa 2737 goto out;
31b8006e
SW		 2738 dout("try_read got tag %d\n", (int)con->in_tag);
2739 switch (con->in_tag) {
2740 case CEPH_MSGR_TAG_MSG:
2741 prepare_read_message(con);
2742 break;
2743 case CEPH_MSGR_TAG_ACK:
2744 prepare_read_ack(con);
2745 break;
8b9558aa
YZ		 2746 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2747 prepare_read_keepalive_ack(con);
2748 break;
31b8006e 2749 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2750 con_close_socket(con);
2751 con->state = CON_STATE_CLOSED;
98bdb0aa 2752 goto out;
31b8006e
SW		 2753 default:
2754 goto bad_tag;
2755 }
2756 }
2757 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2758 ret = read_partial_message(con);
2759 if (ret <= 0) {
2760 switch (ret) {
2761 case -EBADMSG:
a51983e4 2762 con->error_msg = "bad crc/signature";
67c64eb7
ID		 2763 /* fall through */
2764 case -EBADE:
31b8006e 2765 ret = -EIO;
98bdb0aa 2766 break;
31b8006e
SW		 2767 case -EIO:
2768 con->error_msg = "io error";
98bdb0aa 2769 break;
31b8006e 2770 }
98bdb0aa 2771 goto out;
31b8006e
SW		 2772 }
2773 if (con->in_tag == CEPH_MSGR_TAG_READY)
2774 goto more;
2775 process_message(con);
7b862e07
SW		 2776 if (con->state == CON_STATE_OPEN)
2777 prepare_read_tag(con);
31b8006e
SW		 2778 goto more;
2779 }
3a23083b
SW		 2780 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2781 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2782 /*
2783 * the final handshake seq exchange is semantically
2784 * equivalent to an ACK
2785 */
31b8006e
SW		 2786 ret = read_partial_ack(con);
2787 if (ret <= 0)
98bdb0aa 2788 goto out;
31b8006e
SW		 2789 process_ack(con);
2790 goto more;
2791 }
8b9558aa
YZ		 2792 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2793 ret = read_keepalive_ack(con);
2794 if (ret <= 0)
2795 goto out;
2796 goto more;
2797 }
31b8006e 2798
31b8006e 2799out:
98bdb0aa 2800 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2801 return ret;
2802
2803bad_tag:
2804 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2805 con->error_msg = "protocol error, garbage tag";
2806 ret = -1;
2807 goto out;
2808}
2809
2810
2811/*
802c6d96
AE		 2812 * Atomically queue work on a connection after the specified delay.
2813 * Bump @con reference to avoid races with connection teardown.
2814 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2815 */
802c6d96 2816static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2817{
31b8006e 2818 if (!con->ops->get(con)) {
802c6d96 2819 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2820 return -ENOENT;
31b8006e
SW		 2821 }
2822
802c6d96
AE		 2823 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2824 dout("%s %p - already queued\n", __func__, con);
31b8006e 2825 con->ops->put(con);
802c6d96 2826 return -EBUSY;
31b8006e 2827 }
802c6d96
AE		 2828
2829 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2830 return 0;
2831}
2832
2833static void queue_con(struct ceph_connection *con)
2834{
2835 (void) queue_con_delay(con, 0);
31b8006e
SW		 2836}
2837
37ab77ac
ID		 2838static void cancel_con(struct ceph_connection *con)
2839{
2840 if (cancel_delayed_work(&con->work)) {
2841 dout("%s %p\n", __func__, con);
2842 con->ops->put(con);
2843 }
2844}
2845
7bb21d68
AE		 2846static bool con_sock_closed(struct ceph_connection *con)
2847{
c9ffc77a 2848 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2849 return false;
2850
2851#define CASE(x) \
2852 case CON_STATE_ ## x: \
2853 con->error_msg = "socket closed (con state " #x ")"; \
2854 break;
2855
2856 switch (con->state) {
2857 CASE(CLOSED);
2858 CASE(PREOPEN);
2859 CASE(CONNECTING);
2860 CASE(NEGOTIATING);
2861 CASE(OPEN);
2862 CASE(STANDBY);
2863 default:
b9a67899 2864 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2865 __func__, con, con->state);
2866 con->error_msg = "unrecognized con state";
2867 BUG();
2868 break;
2869 }
2870#undef CASE
2871
2872 return true;
2873}
2874
f20a39fd
AE		 2875static bool con_backoff(struct ceph_connection *con)
2876{
2877 int ret;
2878
2879 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2880 return false;
2881
2882 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2883 if (ret) {
2884 dout("%s: con %p FAILED to back off %lu\n", __func__,
2885 con, con->delay);
2886 BUG_ON(ret == -ENOENT);
2887 con_flag_set(con, CON_FLAG_BACKOFF);
2888 }
2889
2890 return true;
2891}
2892
93209264
AE		 2893/* Finish fault handling; con->mutex must *not* be held here */
2894
2895static void con_fault_finish(struct ceph_connection *con)
2896{
f6330cc1
ID		 2897 dout("%s %p\n", __func__, con);
2898
93209264
AE		 2899 /*
2900 * in case we faulted due to authentication, invalidate our
2901 * current tickets so that we can get new ones.
2902 */
f6330cc1
ID		 2903 if (con->auth_retry) {
2904 dout("auth_retry %d, invalidating\n", con->auth_retry);
2905 if (con->ops->invalidate_authorizer)
2906 con->ops->invalidate_authorizer(con);
2907 con->auth_retry = 0;
93209264
AE		 2908 }
2909
2910 if (con->ops->fault)
2911 con->ops->fault(con);
2912}
2913
31b8006e
SW		 2914/*
2915 * Do some work on a connection. Drop a connection ref when we're done.
2916 */
68931622 2917static void ceph_con_workfn(struct work_struct *work)
31b8006e
SW		 2918{
2919 struct ceph_connection *con = container_of(work, struct ceph_connection,
2920 work.work);
49659416 2921 bool fault;
31b8006e 2922
9dd4658d 2923 mutex_lock(&con->mutex);
49659416
AE		 2924 while (true) {
2925 int ret;
31b8006e 2926
49659416
AE		 2927 if ((fault = con_sock_closed(con))) {
2928 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2929 break;
2930 }
2931 if (con_backoff(con)) {
2932 dout("%s: con %p BACKOFF\n", __func__, con);
2933 break;
2934 }
2935 if (con->state == CON_STATE_STANDBY) {
2936 dout("%s: con %p STANDBY\n", __func__, con);
2937 break;
2938 }
2939 if (con->state == CON_STATE_CLOSED) {
2940 dout("%s: con %p CLOSED\n", __func__, con);
2941 BUG_ON(con->sock);
2942 break;
2943 }
2944 if (con->state == CON_STATE_PREOPEN) {
2945 dout("%s: con %p PREOPEN\n", __func__, con);
2946 BUG_ON(con->sock);
2947 }
0da5d703 2948
49659416
AE		 2949 ret = try_read(con);
2950 if (ret < 0) {
2951 if (ret == -EAGAIN)
2952 continue;
67c64eb7
ID		 2953 if (!con->error_msg)
2954 con->error_msg = "socket error on read";
49659416
AE		 2955 fault = true;
2956 break;
2957 }
2958
2959 ret = try_write(con);
2960 if (ret < 0) {
2961 if (ret == -EAGAIN)
2962 continue;
67c64eb7
ID		 2963 if (!con->error_msg)
2964 con->error_msg = "socket error on write";
49659416
AE		 2965 fault = true;
2966 }
2967
2968 break; /* If we make it to here, we're done */
3a140a0d 2969 }
b6e7b6a1
AE		 2970 if (fault)
2971 con_fault(con);
9dd4658d 2972 mutex_unlock(&con->mutex);
0da5d703 2973
b6e7b6a1
AE		 2974 if (fault)
2975 con_fault_finish(con);
2976
2977 con->ops->put(con);
31b8006e
SW		 2978}
2979
31b8006e
SW		 2980/*
2981 * Generic error/fault handler. A retry mechanism is used with
2982 * exponential backoff
2983 */
93209264 2984static void con_fault(struct ceph_connection *con)
31b8006e 2985{
31b8006e 2986 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2987 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2988
67c64eb7
ID		 2989 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2990 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2991 con->error_msg = NULL;
2992
122070a2 2993 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2994 con->state != CON_STATE_NEGOTIATING &&
2995 con->state != CON_STATE_OPEN);
ec302645 2996
31b8006e 2997 con_close_socket(con);
5e095e8b 2998
c9ffc77a 2999 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 3000 dout("fault on LOSSYTX channel, marking CLOSED\n");
3001 con->state = CON_STATE_CLOSED;
93209264 3002 return;
3b5ede07
SW		 3003 }
3004
5e095e8b 3005 if (con->in_msg) {
38941f80 3006 BUG_ON(con->in_msg->con != con);
5e095e8b
SW		 3007 ceph_msg_put(con->in_msg);
3008 con->in_msg = NULL;
3009 }
31b8006e 3010
e80a52d1
SW		 3011 /* Requeue anything that hasn't been acked */
3012 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 3013
e76661d0
SW		 3014 /* If there are no messages queued or keepalive pending, place
3015 * the connection in a STANDBY state */
3016 if (list_empty(&con->out_queue) &&
c9ffc77a 3017 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 3018 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 3019 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 3020 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 3021 } else {
3022 /* retry after a delay. */
8dacc7da 3023 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 3024 if (con->delay == 0)
3025 con->delay = BASE_DELAY_INTERVAL;
3026 else if (con->delay < MAX_DELAY_INTERVAL)
3027 con->delay *= 2;
c9ffc77a 3028 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 3029 queue_con(con);
31b8006e 3030 }
31b8006e
SW		 3031}
3032
3033
3034
3035/*
15d9882c 3036 * initialize a new messenger instance
31b8006e 3037 */
15d9882c 3038void ceph_messenger_init(struct ceph_messenger *msgr,
859bff51 3039 struct ceph_entity_addr *myaddr)
31b8006e 3040{
31b8006e
SW		 3041 spin_lock_init(&msgr->global_seq_lock);
3042
31b8006e
SW		 3043 if (myaddr)
3044 msgr->inst.addr = *myaddr;
3045
3046 /* select a random nonce */
ac8839d7 3047 msgr->inst.addr.type = 0;
103e2d3a 3048 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 3049 encode_my_addr(msgr);
31b8006e 3050
a2a32584 3051 atomic_set(&msgr->stopping, 0);
757856d2 3052 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
31b8006e 3053
15d9882c 3054 dout("%s %p\n", __func__, msgr);
31b8006e 3055}
15d9882c 3056EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 3057
757856d2
ID		 3058void ceph_messenger_fini(struct ceph_messenger *msgr)
3059{
3060 put_net(read_pnet(&msgr->net));
3061}
3062EXPORT_SYMBOL(ceph_messenger_fini);
3063
583d0fef
ID		 3064static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3065{
3066 if (msg->con)
3067 msg->con->ops->put(msg->con);
3068
3069 msg->con = con ? con->ops->get(con) : NULL;
3070 BUG_ON(msg->con != con);
3071}
3072
e00de341
SW		 3073static void clear_standby(struct ceph_connection *con)
3074{
3075 /* come back from STANDBY? */
8dacc7da 3076 if (con->state == CON_STATE_STANDBY) {
e00de341 3077 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 3078 con->state = CON_STATE_PREOPEN;
e00de341 3079 con->connect_seq++;
c9ffc77a
AE		 3080 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3081 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 3082 }
3083}
3084
31b8006e
SW		 3085/*
3086 * Queue up an outgoing message on the given connection.
3087 */
3088void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3089{
31b8006e 3090 /* set src+dst */
dbad185d 3091 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 3092 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 3093 msg->needs_out_seq = true;
3094
ec302645 3095 mutex_lock(&con->mutex);
92ce034b 3096
8dacc7da 3097 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 3098 dout("con_send %p closed, dropping %p\n", con, msg);
3099 ceph_msg_put(msg);
3100 mutex_unlock(&con->mutex);
3101 return;
3102 }
3103
583d0fef 3104 msg_con_set(msg, con);
92ce034b 3105
31b8006e
SW		 3106 BUG_ON(!list_empty(&msg->list_head));
3107 list_add_tail(&msg->list_head, &con->out_queue);
3108 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3109 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3110 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3111 le32_to_cpu(msg->hdr.front_len),
3112 le32_to_cpu(msg->hdr.middle_len),
3113 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 3114
3115 clear_standby(con);
ec302645 3116 mutex_unlock(&con->mutex);
31b8006e
SW		 3117
3118 /* if there wasn't anything waiting to send before, queue
3119 * new work */
c9ffc77a 3120 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3121 queue_con(con);
3122}
3d14c5d2 3123EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3124
3125/*
3126 * Revoke a message that was previously queued for send
3127 */
6740a845 3128void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3129{
6740a845
AE		 3130 struct ceph_connection *con = msg->con;
3131
583d0fef
ID		 3132 if (!con) {
3133 dout("%s msg %p null con\n", __func__, msg);
6740a845 3134 return; /* Message not in our possession */
583d0fef 3135 }
6740a845 3136
ec302645 3137 mutex_lock(&con->mutex);
31b8006e 3138 if (!list_empty(&msg->list_head)) {
38941f80 3139 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3140 list_del_init(&msg->list_head);
31b8006e 3141 msg->hdr.seq = 0;
38941f80 3142
31b8006e 3143 ceph_msg_put(msg);
ed98adad
SW		 3144 }
3145 if (con->out_msg == msg) {
67645d76
ID		 3146 BUG_ON(con->out_skip);
3147 /* footer */
3148 if (con->out_msg_done) {
3149 con->out_skip += con_out_kvec_skip(con);
3150 } else {
3151 BUG_ON(!msg->data_length);
89f08173 3152 con->out_skip += sizeof_footer(con);
31b8006e 3153 }
67645d76
ID		 3154 /* data, middle, front */
3155 if (msg->data_length)
3156 con->out_skip += msg->cursor.total_resid;
3157 if (msg->middle)
3158 con->out_skip += con_out_kvec_skip(con);
3159 con->out_skip += con_out_kvec_skip(con);
3160
3161 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3162 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
ed98adad 3163 msg->hdr.seq = 0;
67645d76 3164 con->out_msg = NULL;
92ce034b 3165 ceph_msg_put(msg);
31b8006e 3166 }
67645d76 3167
ec302645 3168 mutex_unlock(&con->mutex);
31b8006e
SW		 3169}
3170
350b1c32 3171/*
0d59ab81 3172 * Revoke a message that we may be reading data into
350b1c32 3173 */
8921d114 3174void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3175{
583d0fef 3176 struct ceph_connection *con = msg->con;
8921d114 3177
583d0fef 3178 if (!con) {
8921d114 3179 dout("%s msg %p null con\n", __func__, msg);
8921d114
AE		 3180 return; /* Message not in our possession */
3181 }
3182
350b1c32 3183 mutex_lock(&con->mutex);
8921d114 3184 if (con->in_msg == msg) {
95c96174
ED		 3185 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3186 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3187 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3188
3189 /* skip rest of message */
8921d114
AE		 3190 dout("%s %p msg %p revoked\n", __func__, con, msg);
3191 con->in_base_pos = con->in_base_pos -
350b1c32 3192 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3193 front_len -
3194 middle_len -
3195 data_len -
350b1c32 3196 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3197 ceph_msg_put(con->in_msg);
3198 con->in_msg = NULL;
3199 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3200 con->in_seq++;
350b1c32 3201 } else {
8921d114
AE		 3202 dout("%s %p in_msg %p msg %p no-op\n",
3203 __func__, con, con->in_msg, msg);
350b1c32
SW		 3204 }
3205 mutex_unlock(&con->mutex);
3206}
3207
31b8006e
SW		 3208/*
3209 * Queue a keepalive byte to ensure the tcp connection is alive.
3210 */
3211void ceph_con_keepalive(struct ceph_connection *con)
3212{
e00de341 3213 dout("con_keepalive %p\n", con);
00650931 3214 mutex_lock(&con->mutex);
e00de341 3215 clear_standby(con);
00650931 3216 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3217 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3218 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3219 queue_con(con);
3220}
3d14c5d2 3221EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3222
8b9558aa
YZ		 3223bool ceph_con_keepalive_expired(struct ceph_connection *con,
3224 unsigned long interval)
3225{
3226 if (interval > 0 &&
3227 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
473bd2d7
AB		 3228 struct timespec64 now;
3229 struct timespec64 ts;
3230 ktime_get_real_ts64(&now);
3231 jiffies_to_timespec64(interval, &ts);
3232 ts = timespec64_add(con->last_keepalive_ack, ts);
3233 return timespec64_compare(&now, &ts) >= 0;
8b9558aa
YZ		 3234 }
3235 return false;
3236}
3237
0d9c1ab3 3238static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
43794509 3239{
0d9c1ab3
ID		 3240 BUG_ON(msg->num_data_items >= msg->max_data_items);
3241 return &msg->data[msg->num_data_items++];
6644ed7b
AE		 3242}
3243
3244static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3245{
e4339d28 3246 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3247 ceph_pagelist_release(data->pagelist);
43794509
AE		 3248}
3249
90af3602 3250void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3251 size_t length, size_t alignment)
02afca6c 3252{
6644ed7b
AE		 3253 struct ceph_msg_data *data;
3254
07aa1558
AE		 3255 BUG_ON(!pages);
3256 BUG_ON(!length);
6644ed7b 3257
0d9c1ab3
ID		 3258 data = ceph_msg_data_add(msg);
3259 data->type = CEPH_MSG_DATA_PAGES;
6644ed7b
AE		 3260 data->pages = pages;
3261 data->length = length;
3262 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3263
5240d9f9 3264 msg->data_length += length;
02afca6c 3265}
90af3602 3266EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3267
90af3602 3268void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3269 struct ceph_pagelist *pagelist)
3270{
6644ed7b
AE		 3271 struct ceph_msg_data *data;
3272
07aa1558
AE		 3273 BUG_ON(!pagelist);
3274 BUG_ON(!pagelist->length);
27fa8385 3275
0d9c1ab3
ID		 3276 data = ceph_msg_data_add(msg);
3277 data->type = CEPH_MSG_DATA_PAGELIST;
89486833 3278 refcount_inc(&pagelist->refcnt);
6644ed7b
AE		 3279 data->pagelist = pagelist;
3280
5240d9f9 3281 msg->data_length += pagelist->length;
27fa8385 3282}
90af3602 3283EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3284
ea96571f 3285#ifdef CONFIG_BLOCK
5359a17d
ID		 3286void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
3287 u32 length)
27fa8385 3288{
6644ed7b
AE		 3289 struct ceph_msg_data *data;
3290
0d9c1ab3
ID		 3291 data = ceph_msg_data_add(msg);
3292 data->type = CEPH_MSG_DATA_BIO;
5359a17d 3293 data->bio_pos = *bio_pos;
c851c495 3294 data->bio_length = length;
6644ed7b 3295
5240d9f9 3296 msg->data_length += length;
27fa8385 3297}
90af3602 3298EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3299#endif /* CONFIG_BLOCK */
27fa8385 3300
b9e281c2
ID		 3301void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
3302 struct ceph_bvec_iter *bvec_pos)
3303{
3304 struct ceph_msg_data *data;
3305
0d9c1ab3
ID		 3306 data = ceph_msg_data_add(msg);
3307 data->type = CEPH_MSG_DATA_BVECS;
b9e281c2
ID		 3308 data->bvec_pos = *bvec_pos;
3309
b9e281c2
ID		 3310 msg->data_length += bvec_pos->iter.bi_size;
3311}
3312EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
3313
31b8006e
SW		 3314/*
3315 * construct a new message with given type, size
3316 * the new msg has a ref count of 1.
3317 */
0d9c1ab3
ID		 3318struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
3319 gfp_t flags, bool can_fail)
31b8006e
SW		 3320{
3321 struct ceph_msg *m;
3322
e3d5d638 3323 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3324 if (m == NULL)
3325 goto out;
31b8006e
SW		 3326
3327 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3328 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3329 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3330
9516e45b
AE		 3331 INIT_LIST_HEAD(&m->list_head);
3332 kref_init(&m->kref);
ca20892d 3333
31b8006e
SW		 3334 /* front */
3335 if (front_len) {
eeb0bed5 3336 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3337 if (m->front.iov_base == NULL) {
b61c2763 3338 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3339 front_len);
3340 goto out2;
3341 }
3342 } else {
3343 m->front.iov_base = NULL;
3344 }
f2be82b0 3345 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3346
0d9c1ab3
ID		 3347 if (max_data_items) {
3348 m->data = kmalloc_array(max_data_items, sizeof(*m->data),
3349 flags);
3350 if (!m->data)
3351 goto out2;
3352
3353 m->max_data_items = max_data_items;
3354 }
3355
bb257664 3356 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3357 return m;
3358
3359out2:
3360 ceph_msg_put(m);
3361out:
b61c2763
SW		 3362 if (!can_fail) {
3363 pr_err("msg_new can't create type %d front %d\n", type,
3364 front_len);
f0ed1b7c 3365 WARN_ON(1);
b61c2763
SW		 3366 } else {
3367 dout("msg_new can't create type %d front %d\n", type,
3368 front_len);
3369 }
a79832f2 3370 return NULL;
31b8006e 3371}
0d9c1ab3
ID		 3372EXPORT_SYMBOL(ceph_msg_new2);
3373
3374struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3375 bool can_fail)
3376{
3377 return ceph_msg_new2(type, front_len, 0, flags, can_fail);
3378}
3d14c5d2 3379EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3380
31b8006e
SW		 3381/*
3382 * Allocate "middle" portion of a message, if it is needed and wasn't
3383 * allocated by alloc_msg. This allows us to read a small fixed-size
3384 * per-type header in the front and then gracefully fail (i.e.,
3385 * propagate the error to the caller based on info in the front) when
3386 * the middle is too large.
3387 */
2450418c 3388static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3389{
3390 int type = le16_to_cpu(msg->hdr.type);
3391 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3392
3393 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3394 ceph_msg_type_name(type), middle_len);
3395 BUG_ON(!middle_len);
3396 BUG_ON(msg->middle);
3397
b6c1d5b8 3398 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3399 if (!msg->middle)
3400 return -ENOMEM;
3401 return 0;
3402}
3403
2450418c 3404/*
1c20f2d2
AE		 3405 * Allocate a message for receiving an incoming message on a
3406 * connection, and save the result in con->in_msg. Uses the
3407 * connection's private alloc_msg op if available.
3408 *
4740a623
SW		 3409 * Returns 0 on success, or a negative error code.
3410 *
3411 * On success, if we set *skip = 1:
3412 * - the next message should be skipped and ignored.
3413 * - con->in_msg == NULL
3414 * or if we set *skip = 0:
3415 * - con->in_msg is non-null.
3416 * On error (ENOMEM, EAGAIN, ...),
3417 * - con->in_msg == NULL
2450418c 3418 */
4740a623 3419static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3420{
4740a623 3421 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3422 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3423 struct ceph_msg *msg;
4740a623 3424 int ret = 0;
2450418c 3425
1c20f2d2 3426 BUG_ON(con->in_msg != NULL);
53ded495 3427 BUG_ON(!con->ops->alloc_msg);
2450418c 3428
53ded495
AE		 3429 mutex_unlock(&con->mutex);
3430 msg = con->ops->alloc_msg(con, hdr, skip);
3431 mutex_lock(&con->mutex);
3432 if (con->state != CON_STATE_OPEN) {
3433 if (msg)
1d866d1c 3434 ceph_msg_put(msg);
53ded495
AE		 3435 return -EAGAIN;
3436 }
4137577a
AE		 3437 if (msg) {
3438 BUG_ON(*skip);
583d0fef 3439 msg_con_set(msg, con);
4137577a 3440 con->in_msg = msg;
4137577a
AE		 3441 } else {
3442 /*
3443 * Null message pointer means either we should skip
3444 * this message or we couldn't allocate memory. The
3445 * former is not an error.
3446 */
3447 if (*skip)
3448 return 0;
4137577a 3449
67c64eb7 3450 con->error_msg = "error allocating memory for incoming message";
53ded495 3451 return -ENOMEM;
2450418c 3452 }
1c20f2d2 3453 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3454
1c20f2d2
AE		 3455 if (middle_len && !con->in_msg->middle) {
3456 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3457 if (ret < 0) {
1c20f2d2
AE		 3458 ceph_msg_put(con->in_msg);
3459 con->in_msg = NULL;
2450418c
YS		 3460 }
3461 }
9d7f0f13 3462
4740a623 3463 return ret;
2450418c
YS		 3464}
3465
31b8006e
SW		 3466
3467/*
3468 * Free a generically kmalloc'd message.
3469 */
0215e44b 3470static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3471{
0215e44b 3472 dout("%s %p\n", __func__, m);
4965fc38 3473 kvfree(m->front.iov_base);
0d9c1ab3 3474 kfree(m->data);
e3d5d638 3475 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3476}
3477
0215e44b 3478static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3479{
3480 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
0d9c1ab3 3481 int i;
31b8006e 3482
0215e44b 3483 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3484 WARN_ON(!list_empty(&m->list_head));
3485
583d0fef
ID		 3486 msg_con_set(m, NULL);
3487
c2e552e7
SW		 3488 /* drop middle, data, if any */
3489 if (m->middle) {
3490 ceph_buffer_put(m->middle);
3491 m->middle = NULL;
31b8006e 3492 }
5240d9f9 3493
0d9c1ab3
ID		 3494 for (i = 0; i < m->num_data_items; i++)
3495 ceph_msg_data_destroy(&m->data[i]);
58bb3b37 3496
c2e552e7
SW		 3497 if (m->pool)
3498 ceph_msgpool_put(m->pool, m);
3499 else
0215e44b
ID		 3500 ceph_msg_free(m);
3501}
3502
3503struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3504{
3505 dout("%s %p (was %d)\n", __func__, msg,
2c935bc5 3506 kref_read(&msg->kref));
0215e44b
ID		 3507 kref_get(&msg->kref);
3508 return msg;
3509}
3510EXPORT_SYMBOL(ceph_msg_get);
3511
3512void ceph_msg_put(struct ceph_msg *msg)
3513{
3514 dout("%s %p (was %d)\n", __func__, msg,
2c935bc5 3515 kref_read(&msg->kref));
0215e44b 3516 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3517}
0215e44b 3518EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3519
3520void ceph_msg_dump(struct ceph_msg *msg)
3521{
3cea4c30
ID		 3522 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3523 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3524 print_hex_dump(KERN_DEBUG, "header: ",
3525 DUMP_PREFIX_OFFSET, 16, 1,
3526 &msg->hdr, sizeof(msg->hdr), true);
3527 print_hex_dump(KERN_DEBUG, " front: ",
3528 DUMP_PREFIX_OFFSET, 16, 1,
3529 msg->front.iov_base, msg->front.iov_len, true);
3530 if (msg->middle)
3531 print_hex_dump(KERN_DEBUG, "middle: ",
3532 DUMP_PREFIX_OFFSET, 16, 1,
3533 msg->middle->vec.iov_base,
3534 msg->middle->vec.iov_len, true);
3535 print_hex_dump(KERN_DEBUG, "footer: ",
3536 DUMP_PREFIX_OFFSET, 16, 1,
3537 &msg->footer, sizeof(msg->footer), true);
3538}
3d14c5d2 3539EXPORT_SYMBOL(ceph_msg_dump);
Ubuntu Hirsute Linux kernel mirror