]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/ceph/messenger.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
UBUNTU: Ubuntu-4.15.0-96.97
[mirror_ubuntu-bionic-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;
81b36be4 159static struct kmem_cache *ceph_msg_data_cache;
e3d5d638 160
31b8006e
SW		 161/* static tag bytes (protocol control messages) */
162static char tag_msg = CEPH_MSGR_TAG_MSG;
163static char tag_ack = CEPH_MSGR_TAG_ACK;
164static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
8b9558aa 165static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
31b8006e 166
a6a5349d
SW		 167#ifdef CONFIG_LOCKDEP
168static struct lock_class_key socket_class;
169#endif
170
84495f49
AE		 171/*
172 * When skipping (ignoring) a block of input we read it into a "skip
173 * buffer," which is this many bytes in size.
174 */
175#define SKIP_BUF_SIZE 1024
31b8006e
SW		 176
177static void queue_con(struct ceph_connection *con);
37ab77ac 178static void cancel_con(struct ceph_connection *con);
68931622 179static void ceph_con_workfn(struct work_struct *);
93209264 180static void con_fault(struct ceph_connection *con);
31b8006e 181
31b8006e 182/*
f64a9317
AE		 183 * Nicely render a sockaddr as a string. An array of formatted
184 * strings is used, to approximate reentrancy.
31b8006e 185 */
f64a9317
AE		 186#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
187#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
188#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
189#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
190
191static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
192static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 193
57666519 194static struct page *zero_page; /* used in certain error cases */
57666519 195
3d14c5d2 196const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 197{
198 int i;
199 char *s;
99f0f3b2
AE		 200 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
201 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 202
f64a9317 203 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 204 s = addr_str[i];
205
206 switch (ss->ss_family) {
207 case AF_INET:
bd406145
AE		 208 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
209 ntohs(in4->sin_port));
31b8006e
SW		 210 break;
211
212 case AF_INET6:
bd406145
AE		 213 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
214 ntohs(in6->sin6_port));
31b8006e
SW		 215 break;
216
217 default:
d3002b97
AE		 218 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
219 ss->ss_family);
31b8006e
SW		 220 }
221
222 return s;
223}
3d14c5d2 224EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 225
63f2d211
SW		 226static void encode_my_addr(struct ceph_messenger *msgr)
227{
228 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
229 ceph_encode_addr(&msgr->my_enc_addr);
230}
231
31b8006e
SW		 232/*
233 * work queue for all reading and writing to/from the socket.
234 */
e0f43c94 235static struct workqueue_struct *ceph_msgr_wq;
31b8006e 236
e3d5d638
AE		 237static int ceph_msgr_slab_init(void)
238{
239 BUG_ON(ceph_msg_cache);
5ee61e95 240 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
81b36be4
AE		 241 if (!ceph_msg_cache)
242 return -ENOMEM;
243
244 BUG_ON(ceph_msg_data_cache);
5ee61e95 245 ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
81b36be4
AE		 246 if (ceph_msg_data_cache)
247 return 0;
248
249 kmem_cache_destroy(ceph_msg_cache);
250 ceph_msg_cache = NULL;
251
252 return -ENOMEM;
e3d5d638
AE		 253}
254
255static void ceph_msgr_slab_exit(void)
256{
81b36be4
AE		 257 BUG_ON(!ceph_msg_data_cache);
258 kmem_cache_destroy(ceph_msg_data_cache);
259 ceph_msg_data_cache = NULL;
260
e3d5d638
AE		 261 BUG_ON(!ceph_msg_cache);
262 kmem_cache_destroy(ceph_msg_cache);
263 ceph_msg_cache = NULL;
264}
265
15417167 266static void _ceph_msgr_exit(void)
6173d1f0 267{
d3002b97 268 if (ceph_msgr_wq) {
6173d1f0 269 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 270 ceph_msgr_wq = NULL;
271 }
6173d1f0 272
6173d1f0 273 BUG_ON(zero_page == NULL);
09cbfeaf 274 put_page(zero_page);
6173d1f0 275 zero_page = NULL;
d920ff6f
BC		 276
277 ceph_msgr_slab_exit();
6173d1f0
AE		 278}
279
3d14c5d2 280int ceph_msgr_init(void)
31b8006e 281{
d920ff6f
BC		 282 if (ceph_msgr_slab_init())
283 return -ENOMEM;
284
57666519
AE		 285 BUG_ON(zero_page != NULL);
286 zero_page = ZERO_PAGE(0);
09cbfeaf 287 get_page(zero_page);
57666519 288
f9865f06
ID		 289 /*
290 * The number of active work items is limited by the number of
291 * connections, so leave @max_active at default.
292 */
293 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
6173d1f0
AE		 294 if (ceph_msgr_wq)
295 return 0;
57666519 296
6173d1f0
AE		 297 pr_err("msgr_init failed to create workqueue\n");
298 _ceph_msgr_exit();
57666519 299
6173d1f0 300 return -ENOMEM;
31b8006e 301}
3d14c5d2 302EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 303
304void ceph_msgr_exit(void)
305{
57666519 306 BUG_ON(ceph_msgr_wq == NULL);
57666519 307
6173d1f0 308 _ceph_msgr_exit();
31b8006e 309}
3d14c5d2 310EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 311
cd84db6e 312void ceph_msgr_flush(void)
a922d38f
SW		 313{
314 flush_workqueue(ceph_msgr_wq);
315}
3d14c5d2 316EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 317
ce2c8903
AE		 318/* Connection socket state transition functions */
319
320static void con_sock_state_init(struct ceph_connection *con)
321{
322 int old_state;
323
324 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
325 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
326 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 327 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
328 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 329}
330
331static void con_sock_state_connecting(struct ceph_connection *con)
332{
333 int old_state;
334
335 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
336 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
337 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 338 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
339 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 340}
341
342static void con_sock_state_connected(struct ceph_connection *con)
343{
344 int old_state;
345
346 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
347 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
348 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 349 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
350 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 351}
352
353static void con_sock_state_closing(struct ceph_connection *con)
354{
355 int old_state;
356
357 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
358 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
359 old_state != CON_SOCK_STATE_CONNECTED &&
360 old_state != CON_SOCK_STATE_CLOSING))
361 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 362 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
363 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 364}
365
366static void con_sock_state_closed(struct ceph_connection *con)
367{
368 int old_state;
369
370 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
371 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 372 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 373 old_state != CON_SOCK_STATE_CONNECTING &&
374 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 375 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 376 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
377 CON_SOCK_STATE_CLOSED);
ce2c8903 378}
a922d38f 379
31b8006e
SW		 380/*
381 * socket callback functions
382 */
383
384/* data available on socket, or listen socket received a connect */
676d2369 385static void ceph_sock_data_ready(struct sock *sk)
31b8006e 386{
bd406145 387 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 388 if (atomic_read(&con->msgr->stopping)) {
389 return;
390 }
bd406145 391
31b8006e 392 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 393 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 394 con, con->state);
395 queue_con(con);
396 }
397}
398
399/* socket has buffer space for writing */
327800bd 400static void ceph_sock_write_space(struct sock *sk)
31b8006e 401{
d3002b97 402 struct ceph_connection *con = sk->sk_user_data;
31b8006e 403
182fac26
JS		 404 /* only queue to workqueue if there is data we want to write,
405 * and there is sufficient space in the socket buffer to accept
327800bd 406 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 407 * doesn't get called again until try_write() fills the socket
408 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
409 * and net/core/stream.c:sk_stream_write_space().
410 */
c9ffc77a 411 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
64dc6130 412 if (sk_stream_is_writeable(sk)) {
327800bd 413 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 414 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
415 queue_con(con);
416 }
31b8006e 417 } else {
327800bd 418 dout("%s %p nothing to write\n", __func__, con);
31b8006e 419 }
31b8006e
SW		 420}
421
422/* socket's state has changed */
327800bd 423static void ceph_sock_state_change(struct sock *sk)
31b8006e 424{
bd406145 425 struct ceph_connection *con = sk->sk_user_data;
31b8006e 426
327800bd 427 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 428 con, con->state, sk->sk_state);
429
31b8006e
SW		 430 switch (sk->sk_state) {
431 case TCP_CLOSE:
327800bd 432 dout("%s TCP_CLOSE\n", __func__);
18370b36 433 /* fall through */
31b8006e 434 case TCP_CLOSE_WAIT:
327800bd 435 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 436 con_sock_state_closing(con);
c9ffc77a 437 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 438 queue_con(con);
31b8006e
SW		 439 break;
440 case TCP_ESTABLISHED:
327800bd 441 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 442 con_sock_state_connected(con);
31b8006e
SW		 443 queue_con(con);
444 break;
d3002b97
AE		 445 default: /* Everything else is uninteresting */
446 break;
31b8006e
SW		 447 }
448}
449
450/*
451 * set up socket callbacks
452 */
453static void set_sock_callbacks(struct socket *sock,
454 struct ceph_connection *con)
455{
456 struct sock *sk = sock->sk;
bd406145 457 sk->sk_user_data = con;
327800bd
AE		 458 sk->sk_data_ready = ceph_sock_data_ready;
459 sk->sk_write_space = ceph_sock_write_space;
460 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 461}
462
463
464/*
465 * socket helpers
466 */
467
468/*
469 * initiate connection to a remote socket.
470 */
41617d0c 471static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 472{
f91d3471 473 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e 474 struct socket *sock;
633ee407 475 unsigned int noio_flag;
31b8006e
SW		 476 int ret;
477
478 BUG_ON(con->sock);
633ee407
ID		 479
480 /* sock_create_kern() allocates with GFP_KERNEL */
481 noio_flag = memalloc_noio_save();
757856d2 482 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
eeb1bd5c 483 SOCK_STREAM, IPPROTO_TCP, &sock);
633ee407 484 memalloc_noio_restore(noio_flag);
31b8006e 485 if (ret)
41617d0c 486 return ret;
6d7fdb0a 487 sock->sk->sk_allocation = GFP_NOFS;
31b8006e 488
a6a5349d
SW		 489#ifdef CONFIG_LOCKDEP
490 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
491#endif
492
31b8006e
SW		 493 set_sock_callbacks(sock, con);
494
3d14c5d2 495 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 496
89a86be0 497 con_sock_state_connecting(con);
f91d3471
SW		 498 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
499 O_NONBLOCK);
31b8006e
SW		 500 if (ret == -EINPROGRESS) {
501 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 502 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 503 sock->sk->sk_state);
a5bc3129 504 } else if (ret < 0) {
31b8006e 505 pr_err("connect %s error %d\n",
3d14c5d2 506 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 507 sock_release(sock);
41617d0c 508 return ret;
a5bc3129 509 }
89baaa57 510
859bff51 511 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
ba988f87
CH		 512 int optval = 1;
513
514 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
515 (char *)&optval, sizeof(optval));
516 if (ret)
517 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
518 ret);
519 }
520
a5bc3129 521 con->sock = sock;
41617d0c 522 return 0;
31b8006e
SW		 523}
524
525static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
526{
527 struct kvec iov = {buf, len};
528 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 529 int r;
31b8006e 530
100803a8
AV		 531 iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, &iov, 1, len);
532 r = sock_recvmsg(sock, &msg, msg.msg_flags);
98bdb0aa
SW		 533 if (r == -EAGAIN)
534 r = 0;
535 return r;
31b8006e
SW		 536}
537
afb3d90e
AE		 538static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
539 int page_offset, size_t length)
540{
100803a8
AV		 541 struct bio_vec bvec = {
542 .bv_page = page,
543 .bv_offset = page_offset,
544 .bv_len = length
545 };
546 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
547 int r;
afb3d90e
AE		 548
549 BUG_ON(page_offset + length > PAGE_SIZE);
100803a8
AV		 550 iov_iter_bvec(&msg.msg_iter, READ | ITER_BVEC, &bvec, 1, length);
551 r = sock_recvmsg(sock, &msg, msg.msg_flags);
552 if (r == -EAGAIN)
553 r = 0;
554 return r;
afb3d90e
AE		 555}
556
31b8006e
SW		 557/*
558 * write something. @more is true if caller will be sending more data
559 * shortly.
560 */
561static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
562 size_t kvlen, size_t len, int more)
563{
564 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 565 int r;
31b8006e
SW		 566
567 if (more)
568 msg.msg_flags |= MSG_MORE;
569 else
570 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
571
42961d23
SW		 572 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
573 if (r == -EAGAIN)
574 r = 0;
575 return r;
31b8006e
SW		 576}
577
178eda29 578static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 579 int offset, size_t size, bool more)
31739139
AE		 580{
581 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
582 int ret;
583
584 ret = kernel_sendpage(sock, page, offset, size, flags);
585 if (ret == -EAGAIN)
586 ret = 0;
587
588 return ret;
589}
590
178eda29
CC		 591static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
592 int offset, size_t size, bool more)
593{
61ff6e9b
AV		 594 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
595 struct bio_vec bvec;
178eda29 596 int ret;
178eda29 597
05c51633
ID		 598 /*
599 * sendpage cannot properly handle pages with page_count == 0,
600 * we need to fall back to sendmsg if that's the case.
601 *
602 * Same goes for slab pages: skb_can_coalesce() allows
603 * coalescing neighboring slab objects into a single frag which
604 * triggers one of hardened usercopy checks.
605 */
606 if (page_count(page) >= 1 && !PageSlab(page))
178eda29
CC		 607 return __ceph_tcp_sendpage(sock, page, offset, size, more);
608
61ff6e9b
AV		 609 bvec.bv_page = page;
610 bvec.bv_offset = offset;
611 bvec.bv_len = size;
612
613 if (more)
614 msg.msg_flags |= MSG_MORE;
615 else
616 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
617
618 iov_iter_bvec(&msg.msg_iter, WRITE | ITER_BVEC, &bvec, 1, size);
619 ret = sock_sendmsg(sock, &msg);
620 if (ret == -EAGAIN)
621 ret = 0;
178eda29
CC		 622
623 return ret;
624}
31b8006e
SW		 625
626/*
627 * Shutdown/close the socket for the given connection.
628 */
629static int con_close_socket(struct ceph_connection *con)
630{
8007b8d6 631 int rc = 0;
31b8006e
SW		 632
633 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 634 if (con->sock) {
635 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
636 sock_release(con->sock);
637 con->sock = NULL;
638 }
456ea468
AE		 639
640 /*
4a861692 641 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 642 * independent of the connection mutex, and we could have
643 * received a socket close event before we had the chance to
644 * shut the socket down.
645 */
c9ffc77a 646 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 647
ce2c8903 648 con_sock_state_closed(con);
31b8006e
SW		 649 return rc;
650}
651
652/*
653 * Reset a connection. Discard all incoming and outgoing messages
654 * and clear *_seq state.
655 */
656static void ceph_msg_remove(struct ceph_msg *msg)
657{
658 list_del_init(&msg->list_head);
38941f80 659
31b8006e
SW		 660 ceph_msg_put(msg);
661}
662static void ceph_msg_remove_list(struct list_head *head)
663{
664 while (!list_empty(head)) {
665 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
666 list_head);
667 ceph_msg_remove(msg);
668 }
669}
670
671static void reset_connection(struct ceph_connection *con)
672{
673 /* reset connection, out_queue, msg_ and connect_seq */
674 /* discard existing out_queue and msg_seq */
0fa6ebc6 675 dout("reset_connection %p\n", con);
31b8006e
SW		 676 ceph_msg_remove_list(&con->out_queue);
677 ceph_msg_remove_list(&con->out_sent);
678
cf3e5c40 679 if (con->in_msg) {
38941f80 680 BUG_ON(con->in_msg->con != con);
cf3e5c40
SW		 681 ceph_msg_put(con->in_msg);
682 con->in_msg = NULL;
683 }
684
31b8006e
SW		 685 con->connect_seq = 0;
686 con->out_seq = 0;
c86a2930 687 if (con->out_msg) {
583d0fef 688 BUG_ON(con->out_msg->con != con);
c86a2930
SW		 689 ceph_msg_put(con->out_msg);
690 con->out_msg = NULL;
691 }
31b8006e 692 con->in_seq = 0;
0e0d5e0c 693 con->in_seq_acked = 0;
67645d76
ID		 694
695 con->out_skip = 0;
31b8006e
SW		 696}
697
698/*
699 * mark a peer down. drop any open connections.
700 */
701void ceph_con_close(struct ceph_connection *con)
702{
8c50c817 703 mutex_lock(&con->mutex);
3d14c5d2
YS		 704 dout("con_close %p peer %s\n", con,
705 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 706 con->state = CON_STATE_CLOSED;
a5988c49 707
c9ffc77a
AE		 708 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
709 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
710 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
711 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 712
31b8006e 713 reset_connection(con);
6f2bc3ff 714 con->peer_global_seq = 0;
37ab77ac 715 cancel_con(con);
ee76e073 716 con_close_socket(con);
ec302645 717 mutex_unlock(&con->mutex);
31b8006e 718}
3d14c5d2 719EXPORT_SYMBOL(ceph_con_close);
31b8006e 720
31b8006e
SW		 721/*
722 * Reopen a closed connection, with a new peer address.
723 */
b7a9e5dd
SW		 724void ceph_con_open(struct ceph_connection *con,
725 __u8 entity_type, __u64 entity_num,
726 struct ceph_entity_addr *addr)
31b8006e 727{
5469155f 728 mutex_lock(&con->mutex);
3d14c5d2 729 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 730
122070a2 731 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 732 con->state = CON_STATE_PREOPEN;
a5988c49 733
b7a9e5dd
SW		 734 con->peer_name.type = (__u8) entity_type;
735 con->peer_name.num = cpu_to_le64(entity_num);
736
31b8006e 737 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 738 con->delay = 0; /* reset backoff memory */
5469155f 739 mutex_unlock(&con->mutex);
31b8006e
SW		 740 queue_con(con);
741}
3d14c5d2 742EXPORT_SYMBOL(ceph_con_open);
31b8006e 743
87b315a5
SW		 744/*
745 * return true if this connection ever successfully opened
746 */
747bool ceph_con_opened(struct ceph_connection *con)
748{
749 return con->connect_seq > 0;
750}
751
31b8006e
SW		 752/*
753 * initialize a new connection.
754 */
1bfd89f4
AE		 755void ceph_con_init(struct ceph_connection *con, void *private,
756 const struct ceph_connection_operations *ops,
b7a9e5dd 757 struct ceph_messenger *msgr)
31b8006e
SW		 758{
759 dout("con_init %p\n", con);
760 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 761 con->private = private;
762 con->ops = ops;
31b8006e 763 con->msgr = msgr;
ce2c8903
AE		 764
765 con_sock_state_init(con);
766
ec302645 767 mutex_init(&con->mutex);
31b8006e
SW		 768 INIT_LIST_HEAD(&con->out_queue);
769 INIT_LIST_HEAD(&con->out_sent);
68931622 770 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
a5988c49 771
8dacc7da 772 con->state = CON_STATE_CLOSED;
31b8006e 773}
3d14c5d2 774EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 775
776
777/*
778 * We maintain a global counter to order connection attempts. Get
779 * a unique seq greater than @gt.
780 */
781static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
782{
783 u32 ret;
784
785 spin_lock(&msgr->global_seq_lock);
786 if (msgr->global_seq < gt)
787 msgr->global_seq = gt;
788 ret = ++msgr->global_seq;
789 spin_unlock(&msgr->global_seq_lock);
790 return ret;
791}
792
e2200423 793static void con_out_kvec_reset(struct ceph_connection *con)
859eb799 794{
67645d76
ID		 795 BUG_ON(con->out_skip);
796
859eb799
AE		 797 con->out_kvec_left = 0;
798 con->out_kvec_bytes = 0;
799 con->out_kvec_cur = &con->out_kvec[0];
800}
801
e2200423 802static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 803 size_t size, void *data)
804{
67645d76 805 int index = con->out_kvec_left;
859eb799 806
67645d76 807 BUG_ON(con->out_skip);
859eb799
AE		 808 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
809
810 con->out_kvec[index].iov_len = size;
811 con->out_kvec[index].iov_base = data;
812 con->out_kvec_left++;
813 con->out_kvec_bytes += size;
814}
31b8006e 815
67645d76
ID		 816/*
817 * Chop off a kvec from the end. Return residual number of bytes for
818 * that kvec, i.e. how many bytes would have been written if the kvec
819 * hadn't been nuked.
820 */
821static int con_out_kvec_skip(struct ceph_connection *con)
822{
823 int off = con->out_kvec_cur - con->out_kvec;
824 int skip = 0;
825
826 if (con->out_kvec_bytes > 0) {
827 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
828 BUG_ON(con->out_kvec_bytes < skip);
829 BUG_ON(!con->out_kvec_left);
830 con->out_kvec_bytes -= skip;
831 con->out_kvec_left--;
832 }
833
834 return skip;
835}
836
df6ad1f9 837#ifdef CONFIG_BLOCK
6aaa4511
AE		 838
839/*
840 * For a bio data item, a piece is whatever remains of the next
841 * entry in the current bio iovec, or the first entry in the next
842 * bio in the list.
843 */
8ae4f4f5 844static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 845 size_t length)
6aaa4511 846{
8ae4f4f5 847 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 848 struct bio *bio;
849
850 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
851
852 bio = data->bio;
853 BUG_ON(!bio);
6aaa4511 854
ca8b3a69 855 cursor->resid = min(length, data->bio_length);
6aaa4511 856 cursor->bio = bio;
f38a5181
KO		 857 cursor->bvec_iter = bio->bi_iter;
858 cursor->last_piece =
859 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
6aaa4511
AE		 860}
861
8ae4f4f5 862static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 863 size_t *page_offset,
864 size_t *length)
865{
8ae4f4f5 866 struct ceph_msg_data *data = cursor->data;
6aaa4511 867 struct bio *bio;
f38a5181 868 struct bio_vec bio_vec;
6aaa4511
AE		 869
870 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
871
872 bio = cursor->bio;
873 BUG_ON(!bio);
874
f38a5181 875 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511 876
f38a5181 877 *page_offset = (size_t) bio_vec.bv_offset;
6aaa4511 878 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 879 if (cursor->last_piece) /* pagelist offset is always 0 */
880 *length = cursor->resid;
881 else
f38a5181 882 *length = (size_t) bio_vec.bv_len;
25aff7c5 883 BUG_ON(*length > cursor->resid);
5df521b1 884 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511 885
f38a5181 886 return bio_vec.bv_page;
6aaa4511
AE		 887}
888
8ae4f4f5
AE		 889static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
890 size_t bytes)
6aaa4511 891{
6aaa4511 892 struct bio *bio;
f38a5181 893 struct bio_vec bio_vec;
6aaa4511 894
8ae4f4f5 895 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
6aaa4511
AE		 896
897 bio = cursor->bio;
898 BUG_ON(!bio);
899
f38a5181 900 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511
AE		 901
902 /* Advance the cursor offset */
903
25aff7c5
AE		 904 BUG_ON(cursor->resid < bytes);
905 cursor->resid -= bytes;
f38a5181
KO		 906
907 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
908
909 if (bytes < bio_vec.bv_len)
6aaa4511
AE		 910 return false; /* more bytes to process in this segment */
911
912 /* Move on to the next segment, and possibly the next bio */
913
f38a5181 914 if (!cursor->bvec_iter.bi_size) {
6aaa4511 915 bio = bio->bi_next;
0ec1d15e
ID		 916 cursor->bio = bio;
917 if (bio)
918 cursor->bvec_iter = bio->bi_iter;
919 else
920 memset(&cursor->bvec_iter, 0,
921 sizeof(cursor->bvec_iter));
6aaa4511 922 }
6aaa4511 923
25aff7c5
AE		 924 if (!cursor->last_piece) {
925 BUG_ON(!cursor->resid);
926 BUG_ON(!bio);
927 /* A short read is OK, so use <= rather than == */
f38a5181 928 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
6aaa4511 929 cursor->last_piece = true;
25aff7c5 930 }
6aaa4511
AE		 931
932 return true;
933}
ea96571f 934#endif /* CONFIG_BLOCK */
df6ad1f9 935
e766d7b5
AE		 936/*
937 * For a page array, a piece comes from the first page in the array
938 * that has not already been fully consumed.
939 */
8ae4f4f5 940static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 941 size_t length)
e766d7b5 942{
8ae4f4f5 943 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 944 int page_count;
945
946 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
947
948 BUG_ON(!data->pages);
949 BUG_ON(!data->length);
950
ca8b3a69 951 cursor->resid = min(length, data->length);
e766d7b5 952 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 953 cursor->page_offset = data->alignment & ~PAGE_MASK;
954 cursor->page_index = 0;
56fc5659
AE		 955 BUG_ON(page_count > (int)USHRT_MAX);
956 cursor->page_count = (unsigned short)page_count;
957 BUG_ON(length > SIZE_MAX - cursor->page_offset);
5f740d7e 958 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 959}
960
8ae4f4f5
AE		 961static struct page *
962ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
963 size_t *page_offset, size_t *length)
e766d7b5 964{
8ae4f4f5 965 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 966
967 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
968
969 BUG_ON(cursor->page_index >= cursor->page_count);
970 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 971
972 *page_offset = cursor->page_offset;
25aff7c5 973 if (cursor->last_piece)
e766d7b5 974 *length = cursor->resid;
25aff7c5 975 else
e766d7b5 976 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 977
978 return data->pages[cursor->page_index];
979}
980
8ae4f4f5 981static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 982 size_t bytes)
983{
8ae4f4f5 984 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 985
986 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 987
988 /* Advance the cursor page offset */
989
990 cursor->resid -= bytes;
5df521b1
AE		 991 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
992 if (!bytes || cursor->page_offset)
e766d7b5
AE		 993 return false; /* more bytes to process in the current page */
994
d90deda6
YZ		 995 if (!cursor->resid)
996 return false; /* no more data */
997
5df521b1 998 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 999
1000 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 1001 cursor->page_index++;
25aff7c5 1002 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 1003
1004 return true;
1005}
1006
fe38a2b6 1007/*
dd236fcb
AE		 1008 * For a pagelist, a piece is whatever remains to be consumed in the
1009 * first page in the list, or the front of the next page.
fe38a2b6 1010 */
8ae4f4f5
AE		 1011static void
1012ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 1013 size_t length)
fe38a2b6 1014{
8ae4f4f5 1015 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1016 struct ceph_pagelist *pagelist;
1017 struct page *page;
1018
dd236fcb 1019 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 1020
1021 pagelist = data->pagelist;
1022 BUG_ON(!pagelist);
25aff7c5
AE		 1023
1024 if (!length)
fe38a2b6
AE		 1025 return; /* pagelist can be assigned but empty */
1026
1027 BUG_ON(list_empty(&pagelist->head));
1028 page = list_first_entry(&pagelist->head, struct page, lru);
1029
ca8b3a69 1030 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 1031 cursor->page = page;
1032 cursor->offset = 0;
a51b272e 1033 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1034}
1035
8ae4f4f5
AE		 1036static struct page *
1037ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1038 size_t *page_offset, size_t *length)
fe38a2b6 1039{
8ae4f4f5 1040 struct ceph_msg_data *data = cursor->data;
fe38a2b6 1041 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 1042
1043 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1044
1045 pagelist = data->pagelist;
1046 BUG_ON(!pagelist);
1047
1048 BUG_ON(!cursor->page);
25aff7c5 1049 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 1050
5df521b1 1051 /* offset of first page in pagelist is always 0 */
fe38a2b6 1052 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 1053 if (cursor->last_piece)
25aff7c5
AE		 1054 *length = cursor->resid;
1055 else
1056 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1057
8ae4f4f5 1058 return cursor->page;
fe38a2b6
AE		 1059}
1060
8ae4f4f5 1061static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1062 size_t bytes)
fe38a2b6 1063{
8ae4f4f5 1064 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1065 struct ceph_pagelist *pagelist;
1066
1067 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1068
1069 pagelist = data->pagelist;
1070 BUG_ON(!pagelist);
25aff7c5
AE		 1071
1072 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1073 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1074
1075 /* Advance the cursor offset */
1076
25aff7c5 1077 cursor->resid -= bytes;
fe38a2b6 1078 cursor->offset += bytes;
5df521b1 1079 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1080 if (!bytes || cursor->offset & ~PAGE_MASK)
1081 return false; /* more bytes to process in the current page */
1082
d90deda6
YZ		 1083 if (!cursor->resid)
1084 return false; /* no more data */
1085
fe38a2b6
AE		 1086 /* Move on to the next page */
1087
1088 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
17ddc49b 1089 cursor->page = list_next_entry(cursor->page, lru);
25aff7c5 1090 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1091
1092 return true;
1093}
1094
dd236fcb
AE		 1095/*
1096 * Message data is handled (sent or received) in pieces, where each
1097 * piece resides on a single page. The network layer might not
1098 * consume an entire piece at once. A data item's cursor keeps
1099 * track of which piece is next to process and how much remains to
1100 * be processed in that piece. It also tracks whether the current
1101 * piece is the last one in the data item.
1102 */
ca8b3a69 1103static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1104{
ca8b3a69 1105 size_t length = cursor->total_resid;
8ae4f4f5 1106
8ae4f4f5 1107 switch (cursor->data->type) {
dd236fcb 1108 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1109 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1110 break;
e766d7b5 1111 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1112 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1113 break;
dd236fcb
AE		 1114#ifdef CONFIG_BLOCK
1115 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1116 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1117 break;
dd236fcb 1118#endif /* CONFIG_BLOCK */
6aaa4511 1119 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1120 default:
1121 /* BUG(); */
1122 break;
1123 }
8ae4f4f5 1124 cursor->need_crc = true;
dd236fcb
AE		 1125}
1126
ca8b3a69
AE		 1127static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1128{
1129 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1130 struct ceph_msg_data *data;
1131
1132 BUG_ON(!length);
1133 BUG_ON(length > msg->data_length);
1134 BUG_ON(list_empty(&msg->data));
1135
ca8b3a69
AE		 1136 cursor->data_head = &msg->data;
1137 cursor->total_resid = length;
1138 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1139 cursor->data = data;
1140
1141 __ceph_msg_data_cursor_init(cursor);
1142}
1143
dd236fcb
AE		 1144/*
1145 * Return the page containing the next piece to process for a given
1146 * data item, and supply the page offset and length of that piece.
1147 * Indicate whether this is the last piece in this data item.
1148 */
8ae4f4f5
AE		 1149static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1150 size_t *page_offset, size_t *length,
dd236fcb
AE		 1151 bool *last_piece)
1152{
1153 struct page *page;
1154
8ae4f4f5 1155 switch (cursor->data->type) {
dd236fcb 1156 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1157 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1158 break;
e766d7b5 1159 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1160 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1161 break;
dd236fcb
AE		 1162#ifdef CONFIG_BLOCK
1163 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1164 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1165 break;
dd236fcb 1166#endif /* CONFIG_BLOCK */
6aaa4511 1167 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1168 default:
1169 page = NULL;
1170 break;
1171 }
1172 BUG_ON(!page);
1173 BUG_ON(*page_offset + *length > PAGE_SIZE);
1174 BUG_ON(!*length);
1175 if (last_piece)
8ae4f4f5 1176 *last_piece = cursor->last_piece;
dd236fcb
AE		 1177
1178 return page;
1179}
1180
1181/*
1182 * Returns true if the result moves the cursor on to the next piece
1183 * of the data item.
1184 */
1759f7b0
ID		 1185static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1186 size_t bytes)
dd236fcb
AE		 1187{
1188 bool new_piece;
1189
25aff7c5 1190 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1191 switch (cursor->data->type) {
dd236fcb 1192 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1193 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1194 break;
e766d7b5 1195 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1196 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1197 break;
dd236fcb
AE		 1198#ifdef CONFIG_BLOCK
1199 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1200 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1201 break;
dd236fcb 1202#endif /* CONFIG_BLOCK */
6aaa4511 1203 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1204 default:
1205 BUG();
1206 break;
1207 }
ca8b3a69 1208 cursor->total_resid -= bytes;
dd236fcb 1209
ca8b3a69
AE		 1210 if (!cursor->resid && cursor->total_resid) {
1211 WARN_ON(!cursor->last_piece);
1212 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
17ddc49b 1213 cursor->data = list_next_entry(cursor->data, links);
ca8b3a69 1214 __ceph_msg_data_cursor_init(cursor);
a51b272e 1215 new_piece = true;
ca8b3a69 1216 }
a51b272e 1217 cursor->need_crc = new_piece;
dd236fcb
AE		 1218}
1219
dbc0d3ca
ID		 1220static size_t sizeof_footer(struct ceph_connection *con)
1221{
1222 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1223 sizeof(struct ceph_msg_footer) :
1224 sizeof(struct ceph_msg_footer_old);
1225}
1226
98fa5dd8 1227static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1228{
739c905b 1229 BUG_ON(!msg);
25aff7c5 1230 BUG_ON(!data_len);
739c905b 1231
4c59b4a2 1232 /* Initialize data cursor */
fe38a2b6 1233
8ae4f4f5 1234 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1235}
1236
31b8006e
SW		 1237/*
1238 * Prepare footer for currently outgoing message, and finish things
1239 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1240 */
859eb799 1241static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1242{
1243 struct ceph_msg *m = con->out_msg;
1244
fd154f3c
AE		 1245 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1246
31b8006e 1247 dout("prepare_write_message_footer %p\n", con);
89f08173 1248 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
33d07337
YZ		 1249 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1250 if (con->ops->sign_message)
79dbd1ba 1251 con->ops->sign_message(m);
33d07337
YZ		 1252 else
1253 m->footer.sig = 0;
33d07337
YZ		 1254 } else {
1255 m->old_footer.flags = m->footer.flags;
33d07337 1256 }
31b8006e 1257 con->out_more = m->more_to_follow;
c86a2930 1258 con->out_msg_done = true;
31b8006e
SW		 1259}
1260
1261/*
1262 * Prepare headers for the next outgoing message.
1263 */
1264static void prepare_write_message(struct ceph_connection *con)
1265{
1266 struct ceph_msg *m;
a9a0c51a 1267 u32 crc;
31b8006e 1268
e2200423 1269 con_out_kvec_reset(con);
c86a2930 1270 con->out_msg_done = false;
31b8006e
SW		 1271
1272 /* Sneak an ack in there first? If we can get it into the same
1273 * TCP packet that's a good thing. */
1274 if (con->in_seq > con->in_seq_acked) {
1275 con->in_seq_acked = con->in_seq;
e2200423 1276 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1277 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1278 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1279 &con->out_temp_ack);
31b8006e
SW		 1280 }
1281
38941f80 1282 BUG_ON(list_empty(&con->out_queue));
859eb799 1283 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1284 con->out_msg = m;
38941f80 1285 BUG_ON(m->con != con);
4cf9d544
SW		 1286
1287 /* put message on sent list */
1288 ceph_msg_get(m);
1289 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1290
e84346b7
SW		 1291 /*
1292 * only assign outgoing seq # if we haven't sent this message
1293 * yet. if it is requeued, resend with it's original seq.
1294 */
1295 if (m->needs_out_seq) {
1296 m->hdr.seq = cpu_to_le64(++con->out_seq);
1297 m->needs_out_seq = false;
98ad5ebd 1298
4690faf0
ID		 1299 if (con->ops->reencode_message)
1300 con->ops->reencode_message(m);
1301 }
31b8006e 1302
98fa5dd8 1303 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1304 m, con->out_seq, le16_to_cpu(m->hdr.type),
1305 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1306 m->data_length);
98ad5ebd
ID		 1307 WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1308 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e
SW		 1309
1310 /* tag + hdr + front + middle */
e2200423 1311 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
67645d76 1312 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
e2200423 1313 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1314
31b8006e 1315 if (m->middle)
e2200423 1316 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1317 m->middle->vec.iov_base);
31b8006e 1318
67645d76 1319 /* fill in hdr crc and finalize hdr */
a9a0c51a
AE		 1320 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1321 con->out_msg->hdr.crc = cpu_to_le32(crc);
67645d76 1322 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
a9a0c51a 1323
67645d76 1324 /* fill in front and middle crc, footer */
a9a0c51a
AE		 1325 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1326 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1327 if (m->middle) {
1328 crc = crc32c(0, m->middle->vec.iov_base,
1329 m->middle->vec.iov_len);
1330 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1331 } else
31b8006e 1332 con->out_msg->footer.middle_crc = 0;
739c905b 1333 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1334 le32_to_cpu(con->out_msg->footer.front_crc),
1335 le32_to_cpu(con->out_msg->footer.middle_crc));
67645d76 1336 con->out_msg->footer.flags = 0;
31b8006e
SW		 1337
1338 /* is there a data payload? */
739c905b 1339 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1340 if (m->data_length) {
1341 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1342 con->out_more = 1; /* data + footer will follow */
1343 } else {
31b8006e 1344 /* no, queue up footer too and be done */
859eb799 1345 prepare_write_message_footer(con);
78625051 1346 }
31b8006e 1347
c9ffc77a 1348 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1349}
1350
1351/*
1352 * Prepare an ack.
1353 */
1354static void prepare_write_ack(struct ceph_connection *con)
1355{
1356 dout("prepare_write_ack %p %llu -> %llu\n", con,
1357 con->in_seq_acked, con->in_seq);
1358 con->in_seq_acked = con->in_seq;
1359
e2200423 1360 con_out_kvec_reset(con);
859eb799 1361
e2200423 1362 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1363
31b8006e 1364 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1365 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1366 &con->out_temp_ack);
1367
31b8006e 1368 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1369 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1370}
1371
3a23083b
SW		 1372/*
1373 * Prepare to share the seq during handshake
1374 */
1375static void prepare_write_seq(struct ceph_connection *con)
1376{
1377 dout("prepare_write_seq %p %llu -> %llu\n", con,
1378 con->in_seq_acked, con->in_seq);
1379 con->in_seq_acked = con->in_seq;
1380
1381 con_out_kvec_reset(con);
1382
1383 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1384 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1385 &con->out_temp_ack);
1386
1387 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1388}
1389
31b8006e
SW		 1390/*
1391 * Prepare to write keepalive byte.
1392 */
1393static void prepare_write_keepalive(struct ceph_connection *con)
1394{
1395 dout("prepare_write_keepalive %p\n", con);
e2200423 1396 con_out_kvec_reset(con);
8b9558aa 1397 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1134e091 1398 struct timespec now;
7f61f545 1399
1134e091 1400 ktime_get_real_ts(&now);
8b9558aa 1401 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
7f61f545
ID		 1402 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1403 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1404 &con->out_temp_keepalive2);
8b9558aa
YZ		 1405 } else {
1406 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1407 }
c9ffc77a 1408 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1409}
1410
1411/*
1412 * Connection negotiation.
1413 */
1414
38e26535 1415static int get_connect_authorizer(struct ceph_connection *con)
4e7a5dcd 1416{
a3530df3 1417 struct ceph_auth_handshake *auth;
38e26535 1418 int auth_proto;
b1c6b980
AE		 1419
1420 if (!con->ops->get_authorizer) {
38e26535 1421 con->auth = NULL;
b1c6b980
AE		 1422 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1423 con->out_connect.authorizer_len = 0;
38e26535 1424 return 0;
b1c6b980
AE		 1425 }
1426
38e26535 1427 auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
a3530df3 1428 if (IS_ERR(auth))
38e26535 1429 return PTR_ERR(auth);
0da5d703 1430
38e26535
ID		 1431 con->auth = auth;
1432 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1433 con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1434 return 0;
4e7a5dcd
SW		 1435}
1436
31b8006e
SW		 1437/*
1438 * We connected to a peer and are saying hello.
1439 */
e825a66d 1440static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1441{
e2200423
AE		 1442 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1443 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1444 &con->msgr->my_enc_addr);
eed0ef2c 1445
eed0ef2c 1446 con->out_more = 0;
c9ffc77a 1447 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1448}
1449
e06bb8c9
ID		 1450static void __prepare_write_connect(struct ceph_connection *con)
1451{
1452 con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1453 if (con->auth)
1454 con_out_kvec_add(con, con->auth->authorizer_buf_len,
1455 con->auth->authorizer_buf);
1456
1457 con->out_more = 0;
1458 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1459}
1460
e825a66d 1461static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1462{
95c96174 1463 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1464 int proto;
38e26535 1465 int ret;
31b8006e
SW		 1466
1467 switch (con->peer_name.type) {
1468 case CEPH_ENTITY_TYPE_MON:
1469 proto = CEPH_MONC_PROTOCOL;
1470 break;
1471 case CEPH_ENTITY_TYPE_OSD:
1472 proto = CEPH_OSDC_PROTOCOL;
1473 break;
1474 case CEPH_ENTITY_TYPE_MDS:
1475 proto = CEPH_MDSC_PROTOCOL;
1476 break;
1477 default:
1478 BUG();
1479 }
1480
1481 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1482 con->connect_seq, global_seq, proto);
4e7a5dcd 1483
859bff51
ID		 1484 con->out_connect.features =
1485 cpu_to_le64(from_msgr(con->msgr)->supported_features);
31b8006e
SW		 1486 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1487 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1488 con->out_connect.global_seq = cpu_to_le32(global_seq);
1489 con->out_connect.protocol_version = cpu_to_le32(proto);
1490 con->out_connect.flags = 0;
31b8006e 1491
38e26535
ID		 1492 ret = get_connect_authorizer(con);
1493 if (ret)
1494 return ret;
3da54776 1495
e06bb8c9 1496 __prepare_write_connect(con);
e10c758e 1497 return 0;
31b8006e
SW		 1498}
1499
31b8006e
SW		 1500/*
1501 * write as much of pending kvecs to the socket as we can.
1502 * 1 -> done
1503 * 0 -> socket full, but more to do
1504 * <0 -> error
1505 */
1506static int write_partial_kvec(struct ceph_connection *con)
1507{
1508 int ret;
1509
1510 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1511 while (con->out_kvec_bytes > 0) {
1512 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1513 con->out_kvec_left, con->out_kvec_bytes,
1514 con->out_more);
1515 if (ret <= 0)
1516 goto out;
1517 con->out_kvec_bytes -= ret;
1518 if (con->out_kvec_bytes == 0)
1519 break; /* done */
f42299e6
AE		 1520
1521 /* account for full iov entries consumed */
1522 while (ret >= con->out_kvec_cur->iov_len) {
1523 BUG_ON(!con->out_kvec_left);
1524 ret -= con->out_kvec_cur->iov_len;
1525 con->out_kvec_cur++;
1526 con->out_kvec_left--;
1527 }
1528 /* and for a partially-consumed entry */
1529 if (ret) {
1530 con->out_kvec_cur->iov_len -= ret;
1531 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1532 }
1533 }
1534 con->out_kvec_left = 0;
31b8006e
SW		 1535 ret = 1;
1536out:
1537 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1538 con->out_kvec_bytes, con->out_kvec_left, ret);
1539 return ret; /* done! */
1540}
1541
35b62808
AE		 1542static u32 ceph_crc32c_page(u32 crc, struct page *page,
1543 unsigned int page_offset,
1544 unsigned int length)
1545{
1546 char *kaddr;
1547
1548 kaddr = kmap(page);
1549 BUG_ON(kaddr == NULL);
1550 crc = crc32c(crc, kaddr + page_offset, length);
1551 kunmap(page);
1552
1553 return crc;
1554}
31b8006e
SW		 1555/*
1556 * Write as much message data payload as we can. If we finish, queue
1557 * up the footer.
1558 * 1 -> done, footer is now queued in out_kvec[].
1559 * 0 -> socket full, but more to do
1560 * <0 -> error
1561 */
34d2d200 1562static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1563{
1564 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1565 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 1566 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
f5db90bc 1567 u32 crc;
31b8006e 1568
859a35d5 1569 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1570
5240d9f9 1571 if (list_empty(&msg->data))
4c59b4a2
AE		 1572 return -EINVAL;
1573
5821bd8c
AE		 1574 /*
1575 * Iterate through each page that contains data to be
1576 * written, and send as much as possible for each.
1577 *
1578 * If we are calculating the data crc (the default), we will
1579 * need to map the page. If we have no pages, they have
1580 * been revoked, so use the zero page.
1581 */
f5db90bc 1582 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1583 while (cursor->resid) {
8a166d05 1584 struct page *page;
e387d525
AE		 1585 size_t page_offset;
1586 size_t length;
8a166d05 1587 bool last_piece;
f5db90bc 1588 int ret;
68b4476b 1589
343128ce
SB		 1590 page = ceph_msg_data_next(cursor, &page_offset, &length,
1591 &last_piece);
e387d525 1592 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
c2cfa194 1593 length, !last_piece);
f5db90bc
AE		 1594 if (ret <= 0) {
1595 if (do_datacrc)
1596 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1597
f5db90bc
AE		 1598 return ret;
1599 }
143334ff
AE		 1600 if (do_datacrc && cursor->need_crc)
1601 crc = ceph_crc32c_page(crc, page, page_offset, length);
1759f7b0 1602 ceph_msg_data_advance(cursor, (size_t)ret);
31b8006e
SW		 1603 }
1604
34d2d200 1605 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1606
1607 /* prepare and queue up footer, too */
f5db90bc
AE		 1608 if (do_datacrc)
1609 msg->footer.data_crc = cpu_to_le32(crc);
1610 else
84ca8fc8 1611 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1612 con_out_kvec_reset(con);
859eb799 1613 prepare_write_message_footer(con);
f5db90bc
AE		 1614
1615 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1616}
1617
1618/*
1619 * write some zeros
1620 */
1621static int write_partial_skip(struct ceph_connection *con)
1622{
1623 int ret;
1624
67645d76 1625 dout("%s %p %d left\n", __func__, con, con->out_skip);
31b8006e 1626 while (con->out_skip > 0) {
09cbfeaf 1627 size_t size = min(con->out_skip, (int) PAGE_SIZE);
31b8006e 1628
e1dcb128 1629 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1630 if (ret <= 0)
1631 goto out;
1632 con->out_skip -= ret;
1633 }
1634 ret = 1;
1635out:
1636 return ret;
1637}
1638
1639/*
1640 * Prepare to read connection handshake, or an ack.
1641 */
eed0ef2c
SW		 1642static void prepare_read_banner(struct ceph_connection *con)
1643{
1644 dout("prepare_read_banner %p\n", con);
1645 con->in_base_pos = 0;
1646}
1647
31b8006e
SW		 1648static void prepare_read_connect(struct ceph_connection *con)
1649{
1650 dout("prepare_read_connect %p\n", con);
1651 con->in_base_pos = 0;
1652}
1653
1654static void prepare_read_ack(struct ceph_connection *con)
1655{
1656 dout("prepare_read_ack %p\n", con);
1657 con->in_base_pos = 0;
1658}
1659
3a23083b
SW		 1660static void prepare_read_seq(struct ceph_connection *con)
1661{
1662 dout("prepare_read_seq %p\n", con);
1663 con->in_base_pos = 0;
1664 con->in_tag = CEPH_MSGR_TAG_SEQ;
1665}
1666
31b8006e
SW		 1667static void prepare_read_tag(struct ceph_connection *con)
1668{
1669 dout("prepare_read_tag %p\n", con);
1670 con->in_base_pos = 0;
1671 con->in_tag = CEPH_MSGR_TAG_READY;
1672}
1673
8b9558aa
YZ		 1674static void prepare_read_keepalive_ack(struct ceph_connection *con)
1675{
1676 dout("prepare_read_keepalive_ack %p\n", con);
1677 con->in_base_pos = 0;
1678}
1679
31b8006e
SW		 1680/*
1681 * Prepare to read a message.
1682 */
1683static int prepare_read_message(struct ceph_connection *con)
1684{
1685 dout("prepare_read_message %p\n", con);
1686 BUG_ON(con->in_msg != NULL);
1687 con->in_base_pos = 0;
1688 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1689 return 0;
1690}
1691
1692
1693static int read_partial(struct ceph_connection *con,
fd51653f 1694 int end, int size, void *object)
31b8006e 1695{
e6cee71f
AE		 1696 while (con->in_base_pos < end) {
1697 int left = end - con->in_base_pos;
31b8006e
SW		 1698 int have = size - left;
1699 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1700 if (ret <= 0)
1701 return ret;
1702 con->in_base_pos += ret;
1703 }
1704 return 1;
1705}
1706
1707
1708/*
1709 * Read all or part of the connect-side handshake on a new connection
1710 */
eed0ef2c 1711static int read_partial_banner(struct ceph_connection *con)
31b8006e 1712{
fd51653f
AE		 1713 int size;
1714 int end;
1715 int ret;
31b8006e 1716
eed0ef2c 1717 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1718
1719 /* peer's banner */
fd51653f
AE		 1720 size = strlen(CEPH_BANNER);
1721 end = size;
1722 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1723 if (ret <= 0)
1724 goto out;
fd51653f
AE		 1725
1726 size = sizeof (con->actual_peer_addr);
1727 end += size;
1728 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1729 if (ret <= 0)
1730 goto out;
fd51653f
AE		 1731
1732 size = sizeof (con->peer_addr_for_me);
1733 end += size;
1734 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1735 if (ret <= 0)
1736 goto out;
fd51653f 1737
eed0ef2c
SW		 1738out:
1739 return ret;
1740}
1741
1742static int read_partial_connect(struct ceph_connection *con)
1743{
fd51653f
AE		 1744 int size;
1745 int end;
1746 int ret;
eed0ef2c
SW		 1747
1748 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1749
fd51653f
AE		 1750 size = sizeof (con->in_reply);
1751 end = size;
1752 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1753 if (ret <= 0)
1754 goto out;
fd51653f 1755
38e26535
ID		 1756 if (con->auth) {
1757 size = le32_to_cpu(con->in_reply.authorizer_len);
bb6e5e4c
ID		 1758 if (size > con->auth->authorizer_reply_buf_len) {
1759 pr_err("authorizer reply too big: %d > %zu\n", size,
1760 con->auth->authorizer_reply_buf_len);
1761 ret = -EINVAL;
1762 goto out;
1763 }
1764
38e26535
ID		 1765 end += size;
1766 ret = read_partial(con, end, size,
1767 con->auth->authorizer_reply_buf);
1768 if (ret <= 0)
1769 goto out;
1770 }
31b8006e 1771
4e7a5dcd
SW		 1772 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1773 con, (int)con->in_reply.tag,
1774 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1775 le32_to_cpu(con->in_reply.global_seq));
1776out:
1777 return ret;
1778}
1779
1780/*
1781 * Verify the hello banner looks okay.
1782 */
1783static int verify_hello(struct ceph_connection *con)
1784{
1785 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1786 pr_err("connect to %s got bad banner\n",
3d14c5d2 1787 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1788 con->error_msg = "protocol error, bad banner";
1789 return -1;
1790 }
1791 return 0;
1792}
1793
1794static bool addr_is_blank(struct sockaddr_storage *ss)
1795{
c44bd69c
ID		 1796 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1797 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1798
31b8006e
SW		 1799 switch (ss->ss_family) {
1800 case AF_INET:
c44bd69c 1801 return addr->s_addr == htonl(INADDR_ANY);
31b8006e 1802 case AF_INET6:
c44bd69c
ID		 1803 return ipv6_addr_any(addr6);
1804 default:
1805 return true;
31b8006e 1806 }
31b8006e
SW		 1807}
1808
1809static int addr_port(struct sockaddr_storage *ss)
1810{
1811 switch (ss->ss_family) {
1812 case AF_INET:
f28bcfbe 1813 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1814 case AF_INET6:
f28bcfbe 1815 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1816 }
1817 return 0;
1818}
1819
1820static void addr_set_port(struct sockaddr_storage *ss, int p)
1821{
1822 switch (ss->ss_family) {
1823 case AF_INET:
1824 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1825 break;
31b8006e
SW		 1826 case AF_INET6:
1827 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1828 break;
31b8006e
SW		 1829 }
1830}
1831
ee3b56f2
NW		 1832/*
1833 * Unlike other *_pton function semantics, zero indicates success.
1834 */
1835static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1836 char delim, const char **ipend)
1837{
99f0f3b2
AE		 1838 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1839 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1840
1841 memset(ss, 0, sizeof(*ss));
1842
1843 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1844 ss->ss_family = AF_INET;
1845 return 0;
1846 }
1847
1848 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1849 ss->ss_family = AF_INET6;
1850 return 0;
1851 }
1852
1853 return -EINVAL;
1854}
1855
1856/*
1857 * Extract hostname string and resolve using kernel DNS facility.
1858 */
1859#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1860static int ceph_dns_resolve_name(const char *name, size_t namelen,
1861 struct sockaddr_storage *ss, char delim, const char **ipend)
1862{
1863 const char *end, *delim_p;
1864 char *colon_p, *ip_addr = NULL;
1865 int ip_len, ret;
1866
1867 /*
1868 * The end of the hostname occurs immediately preceding the delimiter or
1869 * the port marker (':') where the delimiter takes precedence.
1870 */
1871 delim_p = memchr(name, delim, namelen);
1872 colon_p = memchr(name, ':', namelen);
1873
1874 if (delim_p && colon_p)
1875 end = delim_p < colon_p ? delim_p : colon_p;
1876 else if (!delim_p && colon_p)
1877 end = colon_p;
1878 else {
1879 end = delim_p;
1880 if (!end) /* case: hostname:/ */
1881 end = name + namelen;
1882 }
1883
1884 if (end <= name)
1885 return -EINVAL;
1886
1887 /* do dns_resolve upcall */
1888 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1889 if (ip_len > 0)
1890 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1891 else
1892 ret = -ESRCH;
1893
1894 kfree(ip_addr);
1895
1896 *ipend = end;
1897
1898 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1899 ret, ret ? "failed" : ceph_pr_addr(ss));
1900
1901 return ret;
1902}
1903#else
1904static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1905 struct sockaddr_storage *ss, char delim, const char **ipend)
1906{
1907 return -EINVAL;
1908}
1909#endif
1910
1911/*
1912 * Parse a server name (IP or hostname). If a valid IP address is not found
1913 * then try to extract a hostname to resolve using userspace DNS upcall.
1914 */
1915static int ceph_parse_server_name(const char *name, size_t namelen,
1916 struct sockaddr_storage *ss, char delim, const char **ipend)
1917{
1918 int ret;
1919
1920 ret = ceph_pton(name, namelen, ss, delim, ipend);
1921 if (ret)
1922 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1923
1924 return ret;
1925}
1926
31b8006e
SW		 1927/*
1928 * Parse an ip[:port] list into an addr array. Use the default
1929 * monitor port if a port isn't specified.
1930 */
1931int ceph_parse_ips(const char *c, const char *end,
1932 struct ceph_entity_addr *addr,
1933 int max_count, int *count)
1934{
ee3b56f2 1935 int i, ret = -EINVAL;
31b8006e
SW		 1936 const char *p = c;
1937
1938 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1939 for (i = 0; i < max_count; i++) {
1940 const char *ipend;
1941 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1942 int port;
39139f64
SW		 1943 char delim = ',';
1944
1945 if (*p == '[') {
1946 delim = ']';
1947 p++;
1948 }
31b8006e 1949
ee3b56f2
NW		 1950 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1951 if (ret)
31b8006e 1952 goto bad;
ee3b56f2
NW		 1953 ret = -EINVAL;
1954
31b8006e
SW		 1955 p = ipend;
1956
39139f64
SW		 1957 if (delim == ']') {
1958 if (*p != ']') {
1959 dout("missing matching ']'\n");
1960 goto bad;
1961 }
1962 p++;
1963 }
1964
31b8006e
SW		 1965 /* port? */
1966 if (p < end && *p == ':') {
1967 port = 0;
1968 p++;
1969 while (p < end && *p >= '0' && *p <= '9') {
1970 port = (port * 10) + (*p - '0');
1971 p++;
1972 }
f48db1e9
ID		 1973 if (port == 0)
1974 port = CEPH_MON_PORT;
1975 else if (port > 65535)
31b8006e
SW		 1976 goto bad;
1977 } else {
1978 port = CEPH_MON_PORT;
1979 }
1980
1981 addr_set_port(ss, port);
1982
3d14c5d2 1983 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1984
1985 if (p == end)
1986 break;
1987 if (*p != ',')
1988 goto bad;
1989 p++;
1990 }
1991
1992 if (p != end)
1993 goto bad;
1994
1995 if (count)
1996 *count = i + 1;
1997 return 0;
1998
1999bad:
39139f64 2000 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 2001 return ret;
31b8006e 2002}
3d14c5d2 2003EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 2004
eed0ef2c 2005static int process_banner(struct ceph_connection *con)
31b8006e 2006{
eed0ef2c 2007 dout("process_banner on %p\n", con);
31b8006e
SW		 2008
2009 if (verify_hello(con) < 0)
2010 return -1;
2011
63f2d211
SW		 2012 ceph_decode_addr(&con->actual_peer_addr);
2013 ceph_decode_addr(&con->peer_addr_for_me);
2014
31b8006e
SW		 2015 /*
2016 * Make sure the other end is who we wanted. note that the other
2017 * end may not yet know their ip address, so if it's 0.0.0.0, give
2018 * them the benefit of the doubt.
2019 */
103e2d3a
SW		 2020 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2021 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 2022 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2023 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 2024 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2025 ceph_pr_addr(&con->peer_addr.in_addr),
2026 (int)le32_to_cpu(con->peer_addr.nonce),
2027 ceph_pr_addr(&con->actual_peer_addr.in_addr),
2028 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 2029 con->error_msg = "wrong peer at address";
31b8006e
SW		 2030 return -1;
2031 }
2032
2033 /*
2034 * did we learn our address?
2035 */
2036 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2037 int port = addr_port(&con->msgr->inst.addr.in_addr);
2038
2039 memcpy(&con->msgr->inst.addr.in_addr,
2040 &con->peer_addr_for_me.in_addr,
2041 sizeof(con->peer_addr_for_me.in_addr));
2042 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 2043 encode_my_addr(con->msgr);
eed0ef2c 2044 dout("process_banner learned my addr is %s\n",
3d14c5d2 2045 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 2046 }
2047
eed0ef2c
SW		 2048 return 0;
2049}
2050
2051static int process_connect(struct ceph_connection *con)
2052{
859bff51
ID		 2053 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2054 u64 req_feat = from_msgr(con->msgr)->required_features;
dcbbd97c 2055 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 2056 int ret;
04a419f9 2057
eed0ef2c
SW		 2058 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2059
38e26535 2060 if (con->auth) {
ff1e6015
ID		 2061 int len = le32_to_cpu(con->in_reply.authorizer_len);
2062
5c056fdc
ID		 2063 /*
2064 * Any connection that defines ->get_authorizer()
8549d879
ID		 2065 * should also define ->add_authorizer_challenge() and
2066 * ->verify_authorizer_reply().
2067 *
5c056fdc
ID		 2068 * See get_connect_authorizer().
2069 */
8549d879
ID		 2070 if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
2071 ret = con->ops->add_authorizer_challenge(
ff1e6015 2072 con, con->auth->authorizer_reply_buf, len);
8549d879
ID		 2073 if (ret < 0)
2074 return ret;
2075
2076 con_out_kvec_reset(con);
2077 __prepare_write_connect(con);
2078 prepare_read_connect(con);
2079 return 0;
2080 }
2081
ff1e6015
ID		 2082 if (len) {
2083 ret = con->ops->verify_authorizer_reply(con);
2084 if (ret < 0) {
2085 con->error_msg = "bad authorize reply";
2086 return ret;
2087 }
5c056fdc
ID		 2088 }
2089 }
2090
31b8006e 2091 switch (con->in_reply.tag) {
04a419f9
SW		 2092 case CEPH_MSGR_TAG_FEATURES:
2093 pr_err("%s%lld %s feature set mismatch,"
2094 " my %llx < server's %llx, missing %llx\n",
2095 ENTITY_NAME(con->peer_name),
3d14c5d2 2096 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2097 sup_feat, server_feat, server_feat & ~sup_feat);
2098 con->error_msg = "missing required protocol features";
0fa6ebc6 2099 reset_connection(con);
04a419f9
SW		 2100 return -1;
2101
31b8006e 2102 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2103 pr_err("%s%lld %s protocol version mismatch,"
2104 " my %d != server's %d\n",
2105 ENTITY_NAME(con->peer_name),
3d14c5d2 2106 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2107 le32_to_cpu(con->out_connect.protocol_version),
2108 le32_to_cpu(con->in_reply.protocol_version));
2109 con->error_msg = "protocol version mismatch";
0fa6ebc6 2110 reset_connection(con);
31b8006e
SW		 2111 return -1;
2112
4e7a5dcd
SW		 2113 case CEPH_MSGR_TAG_BADAUTHORIZER:
2114 con->auth_retry++;
2115 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2116 con->auth_retry);
2117 if (con->auth_retry == 2) {
2118 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2119 return -1;
2120 }
6d4221b5 2121 con_out_kvec_reset(con);
e825a66d 2122 ret = prepare_write_connect(con);
0da5d703
SW		 2123 if (ret < 0)
2124 return ret;
63733a0f 2125 prepare_read_connect(con);
4e7a5dcd 2126 break;
31b8006e
SW		 2127
2128 case CEPH_MSGR_TAG_RESETSESSION:
2129 /*
2130 * If we connected with a large connect_seq but the peer
2131 * has no record of a session with us (no connection, or
2132 * connect_seq == 0), they will send RESETSESION to indicate
2133 * that they must have reset their session, and may have
2134 * dropped messages.
2135 */
2136 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2137 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2138 pr_err("%s%lld %s connection reset\n",
2139 ENTITY_NAME(con->peer_name),
3d14c5d2 2140 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2141 reset_connection(con);
6d4221b5 2142 con_out_kvec_reset(con);
5a0f8fdd
AE		 2143 ret = prepare_write_connect(con);
2144 if (ret < 0)
2145 return ret;
31b8006e
SW		 2146 prepare_read_connect(con);
2147
2148 /* Tell ceph about it. */
ec302645 2149 mutex_unlock(&con->mutex);
31b8006e
SW		 2150 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2151 if (con->ops->peer_reset)
2152 con->ops->peer_reset(con);
ec302645 2153 mutex_lock(&con->mutex);
8dacc7da 2154 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2155 return -EAGAIN;
31b8006e
SW		 2156 break;
2157
2158 case CEPH_MSGR_TAG_RETRY_SESSION:
2159 /*
2160 * If we sent a smaller connect_seq than the peer has, try
2161 * again with a larger value.
2162 */
5bdca4e0 2163 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2164 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2165 le32_to_cpu(con->in_reply.connect_seq));
2166 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2167 con_out_kvec_reset(con);
5a0f8fdd
AE		 2168 ret = prepare_write_connect(con);
2169 if (ret < 0)
2170 return ret;
31b8006e
SW		 2171 prepare_read_connect(con);
2172 break;
2173
2174 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2175 /*
2176 * If we sent a smaller global_seq than the peer has, try
2177 * again with a larger value.
2178 */
eed0ef2c 2179 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2180 con->peer_global_seq,
5bdca4e0 2181 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2182 get_global_seq(con->msgr,
5bdca4e0 2183 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2184 con_out_kvec_reset(con);
5a0f8fdd
AE		 2185 ret = prepare_write_connect(con);
2186 if (ret < 0)
2187 return ret;
31b8006e
SW		 2188 prepare_read_connect(con);
2189 break;
2190
3a23083b 2191 case CEPH_MSGR_TAG_SEQ:
31b8006e 2192 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2193 if (req_feat & ~server_feat) {
2194 pr_err("%s%lld %s protocol feature mismatch,"
2195 " my required %llx > server's %llx, need %llx\n",
2196 ENTITY_NAME(con->peer_name),
3d14c5d2 2197 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2198 req_feat, server_feat, req_feat & ~server_feat);
2199 con->error_msg = "missing required protocol features";
0fa6ebc6 2200 reset_connection(con);
04a419f9
SW		 2201 return -1;
2202 }
8dacc7da 2203
122070a2 2204 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2205 con->state = CON_STATE_OPEN;
20e55c4c 2206 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2207 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2208 con->connect_seq++;
aba558e2 2209 con->peer_features = server_feat;
31b8006e
SW		 2210 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2211 con->peer_global_seq,
2212 le32_to_cpu(con->in_reply.connect_seq),
2213 con->connect_seq);
2214 WARN_ON(con->connect_seq !=
2215 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2216
2217 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2218 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2219
85effe18 2220 con->delay = 0; /* reset backoff memory */
92ac41d0 2221
3a23083b
SW		 2222 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2223 prepare_write_seq(con);
2224 prepare_read_seq(con);
2225 } else {
2226 prepare_read_tag(con);
2227 }
31b8006e
SW		 2228 break;
2229
2230 case CEPH_MSGR_TAG_WAIT:
2231 /*
2232 * If there is a connection race (we are opening
2233 * connections to each other), one of us may just have
2234 * to WAIT. This shouldn't happen if we are the
2235 * client.
2236 */
04177882
SW		 2237 con->error_msg = "protocol error, got WAIT as client";
2238 return -1;
31b8006e
SW		 2239
2240 default:
31b8006e
SW		 2241 con->error_msg = "protocol error, garbage tag during connect";
2242 return -1;
2243 }
2244 return 0;
2245}
2246
2247
2248/*
2249 * read (part of) an ack
2250 */
2251static int read_partial_ack(struct ceph_connection *con)
2252{
fd51653f
AE		 2253 int size = sizeof (con->in_temp_ack);
2254 int end = size;
31b8006e 2255
fd51653f 2256 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2257}
2258
31b8006e
SW		 2259/*
2260 * We can finally discard anything that's been acked.
2261 */
2262static void process_ack(struct ceph_connection *con)
2263{
2264 struct ceph_msg *m;
2265 u64 ack = le64_to_cpu(con->in_temp_ack);
2266 u64 seq;
0a2ad541
YZ		 2267 bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2268 struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
31b8006e 2269
0a2ad541
YZ		 2270 /*
2271 * In the reconnect case, con_fault() has requeued messages
2272 * in out_sent. We should cleanup old messages according to
2273 * the reconnect seq.
2274 */
2275 while (!list_empty(list)) {
2276 m = list_first_entry(list, struct ceph_msg, list_head);
2277 if (reconnect && m->needs_out_seq)
2278 break;
31b8006e
SW		 2279 seq = le64_to_cpu(m->hdr.seq);
2280 if (seq > ack)
2281 break;
2282 dout("got ack for seq %llu type %d at %p\n", seq,
2283 le16_to_cpu(m->hdr.type), m);
4cf9d544 2284 m->ack_stamp = jiffies;
31b8006e
SW		 2285 ceph_msg_remove(m);
2286 }
0a2ad541 2287
31b8006e
SW		 2288 prepare_read_tag(con);
2289}
2290
2291
2450418c 2292static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2293 struct kvec *section,
2294 unsigned int sec_len, u32 *crc)
2450418c 2295{
68b4476b 2296 int ret, left;
2450418c
YS		 2297
2298 BUG_ON(!section);
2299
2300 while (section->iov_len < sec_len) {
2301 BUG_ON(section->iov_base == NULL);
2302 left = sec_len - section->iov_len;
2303 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2304 section->iov_len, left);
2305 if (ret <= 0)
2306 return ret;
2307 section->iov_len += ret;
2450418c 2308 }
fe3ad593
AE		 2309 if (section->iov_len == sec_len)
2310 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2311
2450418c
YS		 2312 return 1;
2313}
31b8006e 2314
34d2d200
AE		 2315static int read_partial_msg_data(struct ceph_connection *con)
2316{
2317 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2318 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 2319 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
686be208
AE		 2320 struct page *page;
2321 size_t page_offset;
2322 size_t length;
f5db90bc 2323 u32 crc = 0;
34d2d200
AE		 2324 int ret;
2325
2326 BUG_ON(!msg);
5240d9f9 2327 if (list_empty(&msg->data))
4c59b4a2 2328 return -EIO;
34d2d200 2329
f5db90bc
AE		 2330 if (do_datacrc)
2331 crc = con->in_data_crc;
643c68a4 2332 while (cursor->resid) {
343128ce 2333 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
686be208 2334 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2335 if (ret <= 0) {
2336 if (do_datacrc)
2337 con->in_data_crc = crc;
2338
686be208 2339 return ret;
f5db90bc 2340 }
686be208
AE		 2341
2342 if (do_datacrc)
f5db90bc 2343 crc = ceph_crc32c_page(crc, page, page_offset, ret);
1759f7b0 2344 ceph_msg_data_advance(cursor, (size_t)ret);
34d2d200 2345 }
f5db90bc
AE		 2346 if (do_datacrc)
2347 con->in_data_crc = crc;
34d2d200
AE		 2348
2349 return 1; /* must return > 0 to indicate success */
2350}
2351
31b8006e
SW		 2352/*
2353 * read (part of) a message.
2354 */
686be208
AE		 2355static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2356
31b8006e
SW		 2357static int read_partial_message(struct ceph_connection *con)
2358{
2359 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2360 int size;
2361 int end;
31b8006e 2362 int ret;
95c96174 2363 unsigned int front_len, middle_len, data_len;
859bff51 2364 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
33d07337 2365 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
ae18756b 2366 u64 seq;
fe3ad593 2367 u32 crc;
31b8006e
SW		 2368
2369 dout("read_partial_message con %p msg %p\n", con, m);
2370
2371 /* header */
fd51653f
AE		 2372 size = sizeof (con->in_hdr);
2373 end = size;
2374 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2375 if (ret <= 0)
2376 return ret;
fe3ad593
AE		 2377
2378 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2379 if (cpu_to_le32(crc) != con->in_hdr.crc) {
67c64eb7 2380 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
fe3ad593
AE		 2381 crc, con->in_hdr.crc);
2382 return -EBADMSG;
2383 }
2384
31b8006e
SW		 2385 front_len = le32_to_cpu(con->in_hdr.front_len);
2386 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2387 return -EIO;
2388 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2389 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2390 return -EIO;
2391 data_len = le32_to_cpu(con->in_hdr.data_len);
2392 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2393 return -EIO;
2394
ae18756b
SW		 2395 /* verify seq# */
2396 seq = le64_to_cpu(con->in_hdr.seq);
2397 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2398 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2399 ENTITY_NAME(con->peer_name),
3d14c5d2 2400 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2401 seq, con->in_seq + 1);
2402 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2403 sizeof_footer(con);
ae18756b 2404 con->in_tag = CEPH_MSGR_TAG_READY;
e7a88e82 2405 return 1;
ae18756b
SW		 2406 } else if ((s64)seq - (s64)con->in_seq > 1) {
2407 pr_err("read_partial_message bad seq %lld expected %lld\n",
2408 seq, con->in_seq + 1);
2409 con->error_msg = "bad message sequence # for incoming message";
67c64eb7 2410 return -EBADE;
ae18756b
SW		 2411 }
2412
31b8006e
SW		 2413 /* allocate message? */
2414 if (!con->in_msg) {
4740a623
SW		 2415 int skip = 0;
2416
31b8006e 2417 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2418 front_len, data_len);
4740a623
SW		 2419 ret = ceph_con_in_msg_alloc(con, &skip);
2420 if (ret < 0)
2421 return ret;
f759ebb9
AE		 2422
2423 BUG_ON(!con->in_msg ^ skip);
2450418c 2424 if (skip) {
31b8006e 2425 /* skip this message */
a79832f2 2426 dout("alloc_msg said skip message\n");
31b8006e 2427 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2428 sizeof_footer(con);
31b8006e 2429 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2430 con->in_seq++;
e7a88e82 2431 return 1;
31b8006e 2432 }
38941f80 2433
4740a623 2434 BUG_ON(!con->in_msg);
38941f80 2435 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2436 m = con->in_msg;
2437 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2438 if (m->middle)
2439 m->middle->vec.iov_len = 0;
9d7f0f13 2440
78625051 2441 /* prepare for data payload, if any */
a4107026 2442
78625051 2443 if (data_len)
98fa5dd8 2444 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2445 }
2446
2447 /* front */
2450418c
YS		 2448 ret = read_partial_message_section(con, &m->front, front_len,
2449 &con->in_front_crc);
2450 if (ret <= 0)
2451 return ret;
31b8006e
SW		 2452
2453 /* middle */
2450418c 2454 if (m->middle) {
213c99ee
SW		 2455 ret = read_partial_message_section(con, &m->middle->vec,
2456 middle_len,
2450418c 2457 &con->in_middle_crc);
31b8006e
SW		 2458 if (ret <= 0)
2459 return ret;
31b8006e
SW		 2460 }
2461
2462 /* (page) data */
34d2d200
AE		 2463 if (data_len) {
2464 ret = read_partial_msg_data(con);
2465 if (ret <= 0)
2466 return ret;
31b8006e
SW		 2467 }
2468
31b8006e 2469 /* footer */
89f08173 2470 size = sizeof_footer(con);
fd51653f
AE		 2471 end += size;
2472 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2473 if (ret <= 0)
2474 return ret;
2475
33d07337
YZ		 2476 if (!need_sign) {
2477 m->footer.flags = m->old_footer.flags;
2478 m->footer.sig = 0;
2479 }
2480
31b8006e
SW		 2481 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2482 m, front_len, m->footer.front_crc, middle_len,
2483 m->footer.middle_crc, data_len, m->footer.data_crc);
2484
2485 /* crc ok? */
2486 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2487 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2488 m, con->in_front_crc, m->footer.front_crc);
2489 return -EBADMSG;
2490 }
2491 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2492 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2493 m, con->in_middle_crc, m->footer.middle_crc);
2494 return -EBADMSG;
2495 }
bca064d2 2496 if (do_datacrc &&
31b8006e
SW		 2497 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2498 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2499 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2500 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2501 return -EBADMSG;
2502 }
2503
33d07337 2504 if (need_sign && con->ops->check_message_signature &&
79dbd1ba 2505 con->ops->check_message_signature(m)) {
33d07337
YZ		 2506 pr_err("read_partial_message %p signature check failed\n", m);
2507 return -EBADMSG;
2508 }
2509
31b8006e
SW		 2510 return 1; /* done! */
2511}
2512
2513/*
2514 * Process message. This happens in the worker thread. The callback should
2515 * be careful not to do anything that waits on other incoming messages or it
2516 * may deadlock.
2517 */
2518static void process_message(struct ceph_connection *con)
2519{
583d0fef 2520 struct ceph_msg *msg = con->in_msg;
31b8006e 2521
38941f80 2522 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2523 con->in_msg = NULL;
2524
2525 /* if first message, set peer_name */
2526 if (con->peer_name.type == 0)
dbad185d 2527 con->peer_name = msg->hdr.src;
31b8006e 2528
31b8006e 2529 con->in_seq++;
ec302645 2530 mutex_unlock(&con->mutex);
31b8006e
SW		 2531
2532 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2533 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2534 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2535 le16_to_cpu(msg->hdr.type),
2536 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2537 le32_to_cpu(msg->hdr.front_len),
2538 le32_to_cpu(msg->hdr.data_len),
2539 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2540 con->ops->dispatch(con, msg);
ec302645
SW		 2541
2542 mutex_lock(&con->mutex);
31b8006e
SW		 2543}
2544
8b9558aa
YZ		 2545static int read_keepalive_ack(struct ceph_connection *con)
2546{
2547 struct ceph_timespec ceph_ts;
2548 size_t size = sizeof(ceph_ts);
2549 int ret = read_partial(con, size, size, &ceph_ts);
2550 if (ret <= 0)
2551 return ret;
2552 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2553 prepare_read_tag(con);
2554 return 1;
2555}
31b8006e
SW		 2556
2557/*
2558 * Write something to the socket. Called in a worker thread when the
2559 * socket appears to be writeable and we have something ready to send.
2560 */
2561static int try_write(struct ceph_connection *con)
2562{
31b8006e
SW		 2563 int ret = 1;
2564
d59315ca 2565 dout("try_write start %p state %lu\n", con, con->state);
b252bc18
ID		 2566 if (con->state != CON_STATE_PREOPEN &&
2567 con->state != CON_STATE_CONNECTING &&
2568 con->state != CON_STATE_NEGOTIATING &&
2569 con->state != CON_STATE_OPEN)
2570 return 0;
31b8006e 2571
31b8006e
SW		 2572more:
2573 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2574
2575 /* open the socket first? */
8dacc7da
SW		 2576 if (con->state == CON_STATE_PREOPEN) {
2577 BUG_ON(con->sock);
2578 con->state = CON_STATE_CONNECTING;
a5988c49 2579
e2200423 2580 con_out_kvec_reset(con);
e825a66d 2581 prepare_write_banner(con);
eed0ef2c 2582 prepare_read_banner(con);
31b8006e 2583
cf3e5c40 2584 BUG_ON(con->in_msg);
31b8006e
SW		 2585 con->in_tag = CEPH_MSGR_TAG_READY;
2586 dout("try_write initiating connect on %p new state %lu\n",
2587 con, con->state);
41617d0c
AE		 2588 ret = ceph_tcp_connect(con);
2589 if (ret < 0) {
31b8006e 2590 con->error_msg = "connect error";
31b8006e
SW		 2591 goto out;
2592 }
2593 }
2594
2595more_kvec:
b252bc18
ID		 2596 BUG_ON(!con->sock);
2597
31b8006e 2598 /* kvec data queued? */
67645d76
ID		 2599 if (con->out_kvec_left) {
2600 ret = write_partial_kvec(con);
31b8006e 2601 if (ret <= 0)
42961d23 2602 goto out;
31b8006e 2603 }
67645d76
ID		 2604 if (con->out_skip) {
2605 ret = write_partial_skip(con);
31b8006e 2606 if (ret <= 0)
42961d23 2607 goto out;
31b8006e
SW		 2608 }
2609
2610 /* msg pages? */
2611 if (con->out_msg) {
c86a2930
SW		 2612 if (con->out_msg_done) {
2613 ceph_msg_put(con->out_msg);
2614 con->out_msg = NULL; /* we're done with this one */
2615 goto do_next;
2616 }
2617
34d2d200 2618 ret = write_partial_message_data(con);
31b8006e
SW		 2619 if (ret == 1)
2620 goto more_kvec; /* we need to send the footer, too! */
2621 if (ret == 0)
42961d23 2622 goto out;
31b8006e 2623 if (ret < 0) {
34d2d200 2624 dout("try_write write_partial_message_data err %d\n",
31b8006e 2625 ret);
42961d23 2626 goto out;
31b8006e
SW		 2627 }
2628 }
2629
c86a2930 2630do_next:
8dacc7da 2631 if (con->state == CON_STATE_OPEN) {
8b9558aa
YZ		 2632 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2633 prepare_write_keepalive(con);
2634 goto more;
2635 }
31b8006e
SW		 2636 /* is anything else pending? */
2637 if (!list_empty(&con->out_queue)) {
2638 prepare_write_message(con);
2639 goto more;
2640 }
2641 if (con->in_seq > con->in_seq_acked) {
2642 prepare_write_ack(con);
2643 goto more;
2644 }
31b8006e
SW		 2645 }
2646
2647 /* Nothing to do! */
c9ffc77a 2648 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2649 dout("try_write nothing else to write.\n");
31b8006e
SW		 2650 ret = 0;
2651out:
42961d23 2652 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2653 return ret;
2654}
2655
2656
2657
2658/*
2659 * Read what we can from the socket.
2660 */
2661static int try_read(struct ceph_connection *con)
2662{
31b8006e
SW		 2663 int ret = -1;
2664
8dacc7da
SW		 2665more:
2666 dout("try_read start on %p state %lu\n", con, con->state);
2667 if (con->state != CON_STATE_CONNECTING &&
2668 con->state != CON_STATE_NEGOTIATING &&
2669 con->state != CON_STATE_OPEN)
31b8006e
SW		 2670 return 0;
2671
8dacc7da 2672 BUG_ON(!con->sock);
ec302645 2673
31b8006e
SW		 2674 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2675 con->in_base_pos);
0da5d703 2676
8dacc7da 2677 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2678 dout("try_read connecting\n");
2679 ret = read_partial_banner(con);
2680 if (ret <= 0)
ab166d5a 2681 goto out;
7593af92
AE		 2682 ret = process_banner(con);
2683 if (ret < 0)
2684 goto out;
2685
8dacc7da 2686 con->state = CON_STATE_NEGOTIATING;
7593af92 2687
6d4221b5
JS		 2688 /*
2689 * Received banner is good, exchange connection info.
2690 * Do not reset out_kvec, as sending our banner raced
2691 * with receiving peer banner after connect completed.
2692 */
7593af92
AE		 2693 ret = prepare_write_connect(con);
2694 if (ret < 0)
2695 goto out;
2696 prepare_read_connect(con);
2697
2698 /* Send connection info before awaiting response */
0da5d703
SW		 2699 goto out;
2700 }
2701
8dacc7da 2702 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2703 dout("try_read negotiating\n");
31b8006e
SW		 2704 ret = read_partial_connect(con);
2705 if (ret <= 0)
31b8006e 2706 goto out;
98bdb0aa
SW		 2707 ret = process_connect(con);
2708 if (ret < 0)
2709 goto out;
31b8006e
SW		 2710 goto more;
2711 }
2712
122070a2 2713 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2714
31b8006e
SW		 2715 if (con->in_base_pos < 0) {
2716 /*
2717 * skipping + discarding content.
2718 *
2719 * FIXME: there must be a better way to do this!
2720 */
84495f49
AE		 2721 static char buf[SKIP_BUF_SIZE];
2722 int skip = min((int) sizeof (buf), -con->in_base_pos);
2723
31b8006e
SW		 2724 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2725 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2726 if (ret <= 0)
98bdb0aa 2727 goto out;
31b8006e
SW		 2728 con->in_base_pos += ret;
2729 if (con->in_base_pos)
2730 goto more;
2731 }
2732 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2733 /*
2734 * what's next?
2735 */
2736 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2737 if (ret <= 0)
98bdb0aa 2738 goto out;
31b8006e
SW		 2739 dout("try_read got tag %d\n", (int)con->in_tag);
2740 switch (con->in_tag) {
2741 case CEPH_MSGR_TAG_MSG:
2742 prepare_read_message(con);
2743 break;
2744 case CEPH_MSGR_TAG_ACK:
2745 prepare_read_ack(con);
2746 break;
8b9558aa
YZ		 2747 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2748 prepare_read_keepalive_ack(con);
2749 break;
31b8006e 2750 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2751 con_close_socket(con);
2752 con->state = CON_STATE_CLOSED;
98bdb0aa 2753 goto out;
31b8006e
SW		 2754 default:
2755 goto bad_tag;
2756 }
2757 }
2758 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2759 ret = read_partial_message(con);
2760 if (ret <= 0) {
2761 switch (ret) {
2762 case -EBADMSG:
a51983e4 2763 con->error_msg = "bad crc/signature";
67c64eb7
ID		 2764 /* fall through */
2765 case -EBADE:
31b8006e 2766 ret = -EIO;
98bdb0aa 2767 break;
31b8006e
SW		 2768 case -EIO:
2769 con->error_msg = "io error";
98bdb0aa 2770 break;
31b8006e 2771 }
98bdb0aa 2772 goto out;
31b8006e
SW		 2773 }
2774 if (con->in_tag == CEPH_MSGR_TAG_READY)
2775 goto more;
2776 process_message(con);
7b862e07
SW		 2777 if (con->state == CON_STATE_OPEN)
2778 prepare_read_tag(con);
31b8006e
SW		 2779 goto more;
2780 }
3a23083b
SW		 2781 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2782 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2783 /*
2784 * the final handshake seq exchange is semantically
2785 * equivalent to an ACK
2786 */
31b8006e
SW		 2787 ret = read_partial_ack(con);
2788 if (ret <= 0)
98bdb0aa 2789 goto out;
31b8006e
SW		 2790 process_ack(con);
2791 goto more;
2792 }
8b9558aa
YZ		 2793 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2794 ret = read_keepalive_ack(con);
2795 if (ret <= 0)
2796 goto out;
2797 goto more;
2798 }
31b8006e 2799
31b8006e 2800out:
98bdb0aa 2801 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2802 return ret;
2803
2804bad_tag:
2805 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2806 con->error_msg = "protocol error, garbage tag";
2807 ret = -1;
2808 goto out;
2809}
2810
2811
2812/*
802c6d96
AE		 2813 * Atomically queue work on a connection after the specified delay.
2814 * Bump @con reference to avoid races with connection teardown.
2815 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2816 */
802c6d96 2817static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2818{
31b8006e 2819 if (!con->ops->get(con)) {
802c6d96 2820 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2821 return -ENOENT;
31b8006e
SW		 2822 }
2823
802c6d96
AE		 2824 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2825 dout("%s %p - already queued\n", __func__, con);
31b8006e 2826 con->ops->put(con);
802c6d96 2827 return -EBUSY;
31b8006e 2828 }
802c6d96
AE		 2829
2830 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2831 return 0;
2832}
2833
2834static void queue_con(struct ceph_connection *con)
2835{
2836 (void) queue_con_delay(con, 0);
31b8006e
SW		 2837}
2838
37ab77ac
ID		 2839static void cancel_con(struct ceph_connection *con)
2840{
2841 if (cancel_delayed_work(&con->work)) {
2842 dout("%s %p\n", __func__, con);
2843 con->ops->put(con);
2844 }
2845}
2846
7bb21d68
AE		 2847static bool con_sock_closed(struct ceph_connection *con)
2848{
c9ffc77a 2849 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2850 return false;
2851
2852#define CASE(x) \
2853 case CON_STATE_ ## x: \
2854 con->error_msg = "socket closed (con state " #x ")"; \
2855 break;
2856
2857 switch (con->state) {
2858 CASE(CLOSED);
2859 CASE(PREOPEN);
2860 CASE(CONNECTING);
2861 CASE(NEGOTIATING);
2862 CASE(OPEN);
2863 CASE(STANDBY);
2864 default:
b9a67899 2865 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2866 __func__, con, con->state);
2867 con->error_msg = "unrecognized con state";
2868 BUG();
2869 break;
2870 }
2871#undef CASE
2872
2873 return true;
2874}
2875
f20a39fd
AE		 2876static bool con_backoff(struct ceph_connection *con)
2877{
2878 int ret;
2879
2880 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2881 return false;
2882
2883 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2884 if (ret) {
2885 dout("%s: con %p FAILED to back off %lu\n", __func__,
2886 con, con->delay);
2887 BUG_ON(ret == -ENOENT);
2888 con_flag_set(con, CON_FLAG_BACKOFF);
2889 }
2890
2891 return true;
2892}
2893
93209264
AE		 2894/* Finish fault handling; con->mutex must *not* be held here */
2895
2896static void con_fault_finish(struct ceph_connection *con)
2897{
f6330cc1
ID		 2898 dout("%s %p\n", __func__, con);
2899
93209264
AE		 2900 /*
2901 * in case we faulted due to authentication, invalidate our
2902 * current tickets so that we can get new ones.
2903 */
f6330cc1
ID		 2904 if (con->auth_retry) {
2905 dout("auth_retry %d, invalidating\n", con->auth_retry);
2906 if (con->ops->invalidate_authorizer)
2907 con->ops->invalidate_authorizer(con);
2908 con->auth_retry = 0;
93209264
AE		 2909 }
2910
2911 if (con->ops->fault)
2912 con->ops->fault(con);
2913}
2914
31b8006e
SW		 2915/*
2916 * Do some work on a connection. Drop a connection ref when we're done.
2917 */
68931622 2918static void ceph_con_workfn(struct work_struct *work)
31b8006e
SW		 2919{
2920 struct ceph_connection *con = container_of(work, struct ceph_connection,
2921 work.work);
49659416 2922 bool fault;
31b8006e 2923
9dd4658d 2924 mutex_lock(&con->mutex);
49659416
AE		 2925 while (true) {
2926 int ret;
31b8006e 2927
49659416
AE		 2928 if ((fault = con_sock_closed(con))) {
2929 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2930 break;
2931 }
2932 if (con_backoff(con)) {
2933 dout("%s: con %p BACKOFF\n", __func__, con);
2934 break;
2935 }
2936 if (con->state == CON_STATE_STANDBY) {
2937 dout("%s: con %p STANDBY\n", __func__, con);
2938 break;
2939 }
2940 if (con->state == CON_STATE_CLOSED) {
2941 dout("%s: con %p CLOSED\n", __func__, con);
2942 BUG_ON(con->sock);
2943 break;
2944 }
2945 if (con->state == CON_STATE_PREOPEN) {
2946 dout("%s: con %p PREOPEN\n", __func__, con);
2947 BUG_ON(con->sock);
2948 }
0da5d703 2949
49659416
AE		 2950 ret = try_read(con);
2951 if (ret < 0) {
2952 if (ret == -EAGAIN)
2953 continue;
67c64eb7
ID		 2954 if (!con->error_msg)
2955 con->error_msg = "socket error on read";
49659416
AE		 2956 fault = true;
2957 break;
2958 }
2959
2960 ret = try_write(con);
2961 if (ret < 0) {
2962 if (ret == -EAGAIN)
2963 continue;
67c64eb7
ID		 2964 if (!con->error_msg)
2965 con->error_msg = "socket error on write";
49659416
AE		 2966 fault = true;
2967 }
2968
2969 break; /* If we make it to here, we're done */
3a140a0d 2970 }
b6e7b6a1
AE		 2971 if (fault)
2972 con_fault(con);
9dd4658d 2973 mutex_unlock(&con->mutex);
0da5d703 2974
b6e7b6a1
AE		 2975 if (fault)
2976 con_fault_finish(con);
2977
2978 con->ops->put(con);
31b8006e
SW		 2979}
2980
31b8006e
SW		 2981/*
2982 * Generic error/fault handler. A retry mechanism is used with
2983 * exponential backoff
2984 */
93209264 2985static void con_fault(struct ceph_connection *con)
31b8006e 2986{
31b8006e 2987 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2988 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2989
67c64eb7
ID		 2990 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2991 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2992 con->error_msg = NULL;
2993
122070a2 2994 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2995 con->state != CON_STATE_NEGOTIATING &&
2996 con->state != CON_STATE_OPEN);
ec302645 2997
31b8006e 2998 con_close_socket(con);
5e095e8b 2999
c9ffc77a 3000 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 3001 dout("fault on LOSSYTX channel, marking CLOSED\n");
3002 con->state = CON_STATE_CLOSED;
93209264 3003 return;
3b5ede07
SW		 3004 }
3005
5e095e8b 3006 if (con->in_msg) {
38941f80 3007 BUG_ON(con->in_msg->con != con);
5e095e8b
SW		 3008 ceph_msg_put(con->in_msg);
3009 con->in_msg = NULL;
3010 }
31b8006e 3011
e80a52d1
SW		 3012 /* Requeue anything that hasn't been acked */
3013 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 3014
e76661d0
SW		 3015 /* If there are no messages queued or keepalive pending, place
3016 * the connection in a STANDBY state */
3017 if (list_empty(&con->out_queue) &&
c9ffc77a 3018 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 3019 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 3020 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 3021 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 3022 } else {
3023 /* retry after a delay. */
8dacc7da 3024 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 3025 if (con->delay == 0)
3026 con->delay = BASE_DELAY_INTERVAL;
3027 else if (con->delay < MAX_DELAY_INTERVAL)
3028 con->delay *= 2;
c9ffc77a 3029 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 3030 queue_con(con);
31b8006e 3031 }
31b8006e
SW		 3032}
3033
3034
3035
3036/*
15d9882c 3037 * initialize a new messenger instance
31b8006e 3038 */
15d9882c 3039void ceph_messenger_init(struct ceph_messenger *msgr,
859bff51 3040 struct ceph_entity_addr *myaddr)
31b8006e 3041{
31b8006e
SW		 3042 spin_lock_init(&msgr->global_seq_lock);
3043
31b8006e
SW		 3044 if (myaddr)
3045 msgr->inst.addr = *myaddr;
3046
3047 /* select a random nonce */
ac8839d7 3048 msgr->inst.addr.type = 0;
103e2d3a 3049 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 3050 encode_my_addr(msgr);
31b8006e 3051
a2a32584 3052 atomic_set(&msgr->stopping, 0);
757856d2 3053 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
31b8006e 3054
15d9882c 3055 dout("%s %p\n", __func__, msgr);
31b8006e 3056}
15d9882c 3057EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 3058
757856d2
ID		 3059void ceph_messenger_fini(struct ceph_messenger *msgr)
3060{
3061 put_net(read_pnet(&msgr->net));
3062}
3063EXPORT_SYMBOL(ceph_messenger_fini);
3064
583d0fef
ID		 3065static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3066{
3067 if (msg->con)
3068 msg->con->ops->put(msg->con);
3069
3070 msg->con = con ? con->ops->get(con) : NULL;
3071 BUG_ON(msg->con != con);
3072}
3073
e00de341
SW		 3074static void clear_standby(struct ceph_connection *con)
3075{
3076 /* come back from STANDBY? */
8dacc7da 3077 if (con->state == CON_STATE_STANDBY) {
e00de341 3078 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 3079 con->state = CON_STATE_PREOPEN;
e00de341 3080 con->connect_seq++;
c9ffc77a
AE		 3081 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3082 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 3083 }
3084}
3085
31b8006e
SW		 3086/*
3087 * Queue up an outgoing message on the given connection.
3088 */
3089void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3090{
31b8006e 3091 /* set src+dst */
dbad185d 3092 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 3093 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 3094 msg->needs_out_seq = true;
3095
ec302645 3096 mutex_lock(&con->mutex);
92ce034b 3097
8dacc7da 3098 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 3099 dout("con_send %p closed, dropping %p\n", con, msg);
3100 ceph_msg_put(msg);
3101 mutex_unlock(&con->mutex);
3102 return;
3103 }
3104
583d0fef 3105 msg_con_set(msg, con);
92ce034b 3106
31b8006e
SW		 3107 BUG_ON(!list_empty(&msg->list_head));
3108 list_add_tail(&msg->list_head, &con->out_queue);
3109 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3110 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3111 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3112 le32_to_cpu(msg->hdr.front_len),
3113 le32_to_cpu(msg->hdr.middle_len),
3114 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 3115
3116 clear_standby(con);
ec302645 3117 mutex_unlock(&con->mutex);
31b8006e
SW		 3118
3119 /* if there wasn't anything waiting to send before, queue
3120 * new work */
c9ffc77a 3121 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3122 queue_con(con);
3123}
3d14c5d2 3124EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3125
3126/*
3127 * Revoke a message that was previously queued for send
3128 */
6740a845 3129void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3130{
6740a845
AE		 3131 struct ceph_connection *con = msg->con;
3132
583d0fef
ID		 3133 if (!con) {
3134 dout("%s msg %p null con\n", __func__, msg);
6740a845 3135 return; /* Message not in our possession */
583d0fef 3136 }
6740a845 3137
ec302645 3138 mutex_lock(&con->mutex);
31b8006e 3139 if (!list_empty(&msg->list_head)) {
38941f80 3140 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3141 list_del_init(&msg->list_head);
31b8006e 3142 msg->hdr.seq = 0;
38941f80 3143
31b8006e 3144 ceph_msg_put(msg);
ed98adad
SW		 3145 }
3146 if (con->out_msg == msg) {
67645d76
ID		 3147 BUG_ON(con->out_skip);
3148 /* footer */
3149 if (con->out_msg_done) {
3150 con->out_skip += con_out_kvec_skip(con);
3151 } else {
3152 BUG_ON(!msg->data_length);
89f08173 3153 con->out_skip += sizeof_footer(con);
31b8006e 3154 }
67645d76
ID		 3155 /* data, middle, front */
3156 if (msg->data_length)
3157 con->out_skip += msg->cursor.total_resid;
3158 if (msg->middle)
3159 con->out_skip += con_out_kvec_skip(con);
3160 con->out_skip += con_out_kvec_skip(con);
3161
3162 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3163 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
ed98adad 3164 msg->hdr.seq = 0;
67645d76 3165 con->out_msg = NULL;
92ce034b 3166 ceph_msg_put(msg);
31b8006e 3167 }
67645d76 3168
ec302645 3169 mutex_unlock(&con->mutex);
31b8006e
SW		 3170}
3171
350b1c32 3172/*
0d59ab81 3173 * Revoke a message that we may be reading data into
350b1c32 3174 */
8921d114 3175void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3176{
583d0fef 3177 struct ceph_connection *con = msg->con;
8921d114 3178
583d0fef 3179 if (!con) {
8921d114 3180 dout("%s msg %p null con\n", __func__, msg);
8921d114
AE		 3181 return; /* Message not in our possession */
3182 }
3183
350b1c32 3184 mutex_lock(&con->mutex);
8921d114 3185 if (con->in_msg == msg) {
95c96174
ED		 3186 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3187 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3188 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3189
3190 /* skip rest of message */
8921d114
AE		 3191 dout("%s %p msg %p revoked\n", __func__, con, msg);
3192 con->in_base_pos = con->in_base_pos -
350b1c32 3193 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3194 front_len -
3195 middle_len -
3196 data_len -
350b1c32 3197 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3198 ceph_msg_put(con->in_msg);
3199 con->in_msg = NULL;
3200 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3201 con->in_seq++;
350b1c32 3202 } else {
8921d114
AE		 3203 dout("%s %p in_msg %p msg %p no-op\n",
3204 __func__, con, con->in_msg, msg);
350b1c32
SW		 3205 }
3206 mutex_unlock(&con->mutex);
3207}
3208
31b8006e
SW		 3209/*
3210 * Queue a keepalive byte to ensure the tcp connection is alive.
3211 */
3212void ceph_con_keepalive(struct ceph_connection *con)
3213{
e00de341 3214 dout("con_keepalive %p\n", con);
00650931 3215 mutex_lock(&con->mutex);
e00de341 3216 clear_standby(con);
c5d8472f 3217 con_flag_set(con, CON_FLAG_KEEPALIVE_PENDING);
00650931 3218 mutex_unlock(&con->mutex);
c5d8472f
ID		 3219
3220 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3221 queue_con(con);
3222}
3d14c5d2 3223EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3224
8b9558aa
YZ		 3225bool ceph_con_keepalive_expired(struct ceph_connection *con,
3226 unsigned long interval)
3227{
3228 if (interval > 0 &&
3229 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
1134e091 3230 struct timespec now;
8b9558aa 3231 struct timespec ts;
1134e091 3232 ktime_get_real_ts(&now);
8b9558aa
YZ		 3233 jiffies_to_timespec(interval, &ts);
3234 ts = timespec_add(con->last_keepalive_ack, ts);
3235 return timespec_compare(&now, &ts) >= 0;
3236 }
3237 return false;
3238}
3239
6644ed7b 3240static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3241{
6644ed7b
AE		 3242 struct ceph_msg_data *data;
3243
3244 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3245 return NULL;
3246
81b36be4 3247 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
7c40b22f
DC		 3248 if (!data)
3249 return NULL;
3250
3251 data->type = type;
5240d9f9 3252 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3253
3254 return data;
3255}
3256
3257static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3258{
3259 if (!data)
3260 return;
3261
5240d9f9 3262 WARN_ON(!list_empty(&data->links));
e4339d28 3263 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3264 ceph_pagelist_release(data->pagelist);
81b36be4 3265 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3266}
3267
90af3602 3268void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3269 size_t length, size_t alignment)
02afca6c 3270{
6644ed7b
AE		 3271 struct ceph_msg_data *data;
3272
07aa1558
AE		 3273 BUG_ON(!pages);
3274 BUG_ON(!length);
6644ed7b
AE		 3275
3276 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3277 BUG_ON(!data);
3278 data->pages = pages;
3279 data->length = length;
3280 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3281
5240d9f9
AE		 3282 list_add_tail(&data->links, &msg->data);
3283 msg->data_length += length;
02afca6c 3284}
90af3602 3285EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3286
90af3602 3287void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3288 struct ceph_pagelist *pagelist)
3289{
6644ed7b
AE		 3290 struct ceph_msg_data *data;
3291
07aa1558
AE		 3292 BUG_ON(!pagelist);
3293 BUG_ON(!pagelist->length);
27fa8385 3294
6644ed7b
AE		 3295 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3296 BUG_ON(!data);
3297 data->pagelist = pagelist;
3298
5240d9f9
AE		 3299 list_add_tail(&data->links, &msg->data);
3300 msg->data_length += pagelist->length;
27fa8385 3301}
90af3602 3302EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3303
ea96571f 3304#ifdef CONFIG_BLOCK
90af3602 3305void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3306 size_t length)
27fa8385 3307{
6644ed7b
AE		 3308 struct ceph_msg_data *data;
3309
07aa1558 3310 BUG_ON(!bio);
27fa8385 3311
6644ed7b
AE		 3312 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3313 BUG_ON(!data);
3314 data->bio = bio;
c851c495 3315 data->bio_length = length;
6644ed7b 3316
5240d9f9
AE		 3317 list_add_tail(&data->links, &msg->data);
3318 msg->data_length += length;
27fa8385 3319}
90af3602 3320EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3321#endif /* CONFIG_BLOCK */
27fa8385 3322
31b8006e
SW		 3323/*
3324 * construct a new message with given type, size
3325 * the new msg has a ref count of 1.
3326 */
b61c2763
SW		 3327struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3328 bool can_fail)
31b8006e
SW		 3329{
3330 struct ceph_msg *m;
3331
e3d5d638 3332 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3333 if (m == NULL)
3334 goto out;
31b8006e
SW		 3335
3336 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3337 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3338 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3339
9516e45b
AE		 3340 INIT_LIST_HEAD(&m->list_head);
3341 kref_init(&m->kref);
5240d9f9 3342 INIT_LIST_HEAD(&m->data);
ca20892d 3343
31b8006e
SW		 3344 /* front */
3345 if (front_len) {
eeb0bed5 3346 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3347 if (m->front.iov_base == NULL) {
b61c2763 3348 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3349 front_len);
3350 goto out2;
3351 }
3352 } else {
3353 m->front.iov_base = NULL;
3354 }
f2be82b0 3355 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3356
bb257664 3357 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3358 return m;
3359
3360out2:
3361 ceph_msg_put(m);
3362out:
b61c2763
SW		 3363 if (!can_fail) {
3364 pr_err("msg_new can't create type %d front %d\n", type,
3365 front_len);
f0ed1b7c 3366 WARN_ON(1);
b61c2763
SW		 3367 } else {
3368 dout("msg_new can't create type %d front %d\n", type,
3369 front_len);
3370 }
a79832f2 3371 return NULL;
31b8006e 3372}
3d14c5d2 3373EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3374
31b8006e
SW		 3375/*
3376 * Allocate "middle" portion of a message, if it is needed and wasn't
3377 * allocated by alloc_msg. This allows us to read a small fixed-size
3378 * per-type header in the front and then gracefully fail (i.e.,
3379 * propagate the error to the caller based on info in the front) when
3380 * the middle is too large.
3381 */
2450418c 3382static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3383{
3384 int type = le16_to_cpu(msg->hdr.type);
3385 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3386
3387 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3388 ceph_msg_type_name(type), middle_len);
3389 BUG_ON(!middle_len);
3390 BUG_ON(msg->middle);
3391
b6c1d5b8 3392 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3393 if (!msg->middle)
3394 return -ENOMEM;
3395 return 0;
3396}
3397
2450418c 3398/*
1c20f2d2
AE		 3399 * Allocate a message for receiving an incoming message on a
3400 * connection, and save the result in con->in_msg. Uses the
3401 * connection's private alloc_msg op if available.
3402 *
4740a623
SW		 3403 * Returns 0 on success, or a negative error code.
3404 *
3405 * On success, if we set *skip = 1:
3406 * - the next message should be skipped and ignored.
3407 * - con->in_msg == NULL
3408 * or if we set *skip = 0:
3409 * - con->in_msg is non-null.
3410 * On error (ENOMEM, EAGAIN, ...),
3411 * - con->in_msg == NULL
2450418c 3412 */
4740a623 3413static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3414{
4740a623 3415 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3416 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3417 struct ceph_msg *msg;
4740a623 3418 int ret = 0;
2450418c 3419
1c20f2d2 3420 BUG_ON(con->in_msg != NULL);
53ded495 3421 BUG_ON(!con->ops->alloc_msg);
2450418c 3422
53ded495
AE		 3423 mutex_unlock(&con->mutex);
3424 msg = con->ops->alloc_msg(con, hdr, skip);
3425 mutex_lock(&con->mutex);
3426 if (con->state != CON_STATE_OPEN) {
3427 if (msg)
1d866d1c 3428 ceph_msg_put(msg);
53ded495
AE		 3429 return -EAGAIN;
3430 }
4137577a
AE		 3431 if (msg) {
3432 BUG_ON(*skip);
583d0fef 3433 msg_con_set(msg, con);
4137577a 3434 con->in_msg = msg;
4137577a
AE		 3435 } else {
3436 /*
3437 * Null message pointer means either we should skip
3438 * this message or we couldn't allocate memory. The
3439 * former is not an error.
3440 */
3441 if (*skip)
3442 return 0;
4137577a 3443
67c64eb7 3444 con->error_msg = "error allocating memory for incoming message";
53ded495 3445 return -ENOMEM;
2450418c 3446 }
1c20f2d2 3447 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3448
1c20f2d2
AE		 3449 if (middle_len && !con->in_msg->middle) {
3450 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3451 if (ret < 0) {
1c20f2d2
AE		 3452 ceph_msg_put(con->in_msg);
3453 con->in_msg = NULL;
2450418c
YS		 3454 }
3455 }
9d7f0f13 3456
4740a623 3457 return ret;
2450418c
YS		 3458}
3459
31b8006e
SW		 3460
3461/*
3462 * Free a generically kmalloc'd message.
3463 */
0215e44b 3464static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3465{
0215e44b 3466 dout("%s %p\n", __func__, m);
4965fc38 3467 kvfree(m->front.iov_base);
e3d5d638 3468 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3469}
3470
0215e44b 3471static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3472{
3473 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
10bcee14 3474 struct ceph_msg_data *data, *next;
31b8006e 3475
0215e44b 3476 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3477 WARN_ON(!list_empty(&m->list_head));
3478
583d0fef
ID		 3479 msg_con_set(m, NULL);
3480
c2e552e7
SW		 3481 /* drop middle, data, if any */
3482 if (m->middle) {
3483 ceph_buffer_put(m->middle);
3484 m->middle = NULL;
31b8006e 3485 }
5240d9f9 3486
10bcee14
GT		 3487 list_for_each_entry_safe(data, next, &m->data, links) {
3488 list_del_init(&data->links);
5240d9f9
AE		 3489 ceph_msg_data_destroy(data);
3490 }
a1930804 3491 m->data_length = 0;
58bb3b37 3492
c2e552e7
SW		 3493 if (m->pool)
3494 ceph_msgpool_put(m->pool, m);
3495 else
0215e44b
ID		 3496 ceph_msg_free(m);
3497}
3498
3499struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3500{
3501 dout("%s %p (was %d)\n", __func__, msg,
2c935bc5 3502 kref_read(&msg->kref));
0215e44b
ID		 3503 kref_get(&msg->kref);
3504 return msg;
3505}
3506EXPORT_SYMBOL(ceph_msg_get);
3507
3508void ceph_msg_put(struct ceph_msg *msg)
3509{
3510 dout("%s %p (was %d)\n", __func__, msg,
2c935bc5 3511 kref_read(&msg->kref));
0215e44b 3512 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3513}
0215e44b 3514EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3515
3516void ceph_msg_dump(struct ceph_msg *msg)
3517{
3cea4c30
ID		 3518 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3519 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3520 print_hex_dump(KERN_DEBUG, "header: ",
3521 DUMP_PREFIX_OFFSET, 16, 1,
3522 &msg->hdr, sizeof(msg->hdr), true);
3523 print_hex_dump(KERN_DEBUG, " front: ",
3524 DUMP_PREFIX_OFFSET, 16, 1,
3525 msg->front.iov_base, msg->front.iov_len, true);
3526 if (msg->middle)
3527 print_hex_dump(KERN_DEBUG, "middle: ",
3528 DUMP_PREFIX_OFFSET, 16, 1,
3529 msg->middle->vec.iov_base,
3530 msg->middle->vec.iov_len, true);
3531 print_hex_dump(KERN_DEBUG, "footer: ",
3532 DUMP_PREFIX_OFFSET, 16, 1,
3533 &msg->footer, sizeof(msg->footer), true);
3534}
3d14c5d2 3535EXPORT_SYMBOL(ceph_msg_dump);
ubuntu-bionic-kernel mirror