]> 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
Merge tag 'at91-ab-4.13-soc' of git://git.kernel.org/pub/scm/linux/kernel/git/abellon...
[mirror_ubuntu-bionic-kernel.git] / net / ceph / messenger.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
757856d2 9#include <linux/nsproxy.h>
633ee407 10#include <linux/sched/mm.h>
5a0e3ad6 11#include <linux/slab.h>
31b8006e
SW		 12#include <linux/socket.h>
13#include <linux/string.h>
3ebc21f7 14#ifdef CONFIG_BLOCK
68b4476b 15#include <linux/bio.h>
3ebc21f7 16#endif /* CONFIG_BLOCK */
ee3b56f2 17#include <linux/dns_resolver.h>
31b8006e
SW		 18#include <net/tcp.h>
19
2b3e0c90 20#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 21#include <linux/ceph/libceph.h>
22#include <linux/ceph/messenger.h>
23#include <linux/ceph/decode.h>
24#include <linux/ceph/pagelist.h>
bc3b2d7f 25#include <linux/export.h>
31b8006e
SW		 26
27/*
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
33 * the sender.
34 */
35
bc18f4b1
AE		 36/*
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
40 * unexpected state.
41 *
42 * --------
43 * | NEW* | transient initial state
44 * --------
45 * | con_sock_state_init()
46 * v
47 * ----------
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
50 * ^ \
51 * | \ con_sock_state_connecting()
52 * | ----------------------
53 * | \
54 * + con_sock_state_closed() \
fbb85a47
SW		 55 * |+--------------------------- \
56 * | \ \ \
57 * | ----------- \ \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
60 * | ^ \ |
61 * | | \ |
62 * | + con_sock_state_closing() \ |
63 * | / \ | |
64 * | / --------------- | |
65 * | / \ v v
bc18f4b1
AE		 66 * | / --------------
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
70 * | | v
71 * -------------
72 * | CONNECTED | TCP connection established
73 * -------------
74 *
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
76 */
ce2c8903
AE		 77
78#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
83
8dacc7da
SW		 84/*
85 * connection states
86 */
87#define CON_STATE_CLOSED 1 /* -> PREOPEN */
88#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
93
4a861692
SW		 94/*
95 * ceph_connection flag bits
96 */
97#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 103
c9ffc77a
AE		 104static bool con_flag_valid(unsigned long con_flag)
105{
106 switch (con_flag) {
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
112 return true;
113 default:
114 return false;
115 }
116}
117
118static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
119{
120 BUG_ON(!con_flag_valid(con_flag));
121
122 clear_bit(con_flag, &con->flags);
123}
124
125static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
126{
127 BUG_ON(!con_flag_valid(con_flag));
128
129 set_bit(con_flag, &con->flags);
130}
131
132static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
133{
134 BUG_ON(!con_flag_valid(con_flag));
135
136 return test_bit(con_flag, &con->flags);
137}
138
139static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
141{
142 BUG_ON(!con_flag_valid(con_flag));
143
144 return test_and_clear_bit(con_flag, &con->flags);
145}
146
147static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
149{
150 BUG_ON(!con_flag_valid(con_flag));
151
152 return test_and_set_bit(con_flag, &con->flags);
153}
154
e3d5d638
AE		 155/* Slab caches for frequently-allocated structures */
156
157static struct kmem_cache *ceph_msg_cache;
81b36be4 158static struct kmem_cache *ceph_msg_data_cache;
e3d5d638 159
31b8006e
SW		 160/* static tag bytes (protocol control messages) */
161static char tag_msg = CEPH_MSGR_TAG_MSG;
162static char tag_ack = CEPH_MSGR_TAG_ACK;
163static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
8b9558aa 164static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
31b8006e 165
a6a5349d
SW		 166#ifdef CONFIG_LOCKDEP
167static struct lock_class_key socket_class;
168#endif
169
84495f49
AE		 170/*
171 * When skipping (ignoring) a block of input we read it into a "skip
172 * buffer," which is this many bytes in size.
173 */
174#define SKIP_BUF_SIZE 1024
31b8006e
SW		 175
176static void queue_con(struct ceph_connection *con);
37ab77ac 177static void cancel_con(struct ceph_connection *con);
68931622 178static void ceph_con_workfn(struct work_struct *);
93209264 179static void con_fault(struct ceph_connection *con);
31b8006e 180
31b8006e 181/*
f64a9317
AE		 182 * Nicely render a sockaddr as a string. An array of formatted
183 * strings is used, to approximate reentrancy.
31b8006e 184 */
f64a9317
AE		 185#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
186#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
187#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
188#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
189
190static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
191static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 192
57666519 193static struct page *zero_page; /* used in certain error cases */
57666519 194
3d14c5d2 195const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 196{
197 int i;
198 char *s;
99f0f3b2
AE		 199 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
200 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 201
f64a9317 202 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 203 s = addr_str[i];
204
205 switch (ss->ss_family) {
206 case AF_INET:
bd406145
AE		 207 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
208 ntohs(in4->sin_port));
31b8006e
SW		 209 break;
210
211 case AF_INET6:
bd406145
AE		 212 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
213 ntohs(in6->sin6_port));
31b8006e
SW		 214 break;
215
216 default:
d3002b97
AE		 217 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
218 ss->ss_family);
31b8006e
SW		 219 }
220
221 return s;
222}
3d14c5d2 223EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 224
63f2d211
SW		 225static void encode_my_addr(struct ceph_messenger *msgr)
226{
227 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
228 ceph_encode_addr(&msgr->my_enc_addr);
229}
230
31b8006e
SW		 231/*
232 * work queue for all reading and writing to/from the socket.
233 */
e0f43c94 234static struct workqueue_struct *ceph_msgr_wq;
31b8006e 235
e3d5d638
AE		 236static int ceph_msgr_slab_init(void)
237{
238 BUG_ON(ceph_msg_cache);
5ee61e95 239 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
81b36be4
AE		 240 if (!ceph_msg_cache)
241 return -ENOMEM;
242
243 BUG_ON(ceph_msg_data_cache);
5ee61e95 244 ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
81b36be4
AE		 245 if (ceph_msg_data_cache)
246 return 0;
247
248 kmem_cache_destroy(ceph_msg_cache);
249 ceph_msg_cache = NULL;
250
251 return -ENOMEM;
e3d5d638
AE		 252}
253
254static void ceph_msgr_slab_exit(void)
255{
81b36be4
AE		 256 BUG_ON(!ceph_msg_data_cache);
257 kmem_cache_destroy(ceph_msg_data_cache);
258 ceph_msg_data_cache = NULL;
259
e3d5d638
AE		 260 BUG_ON(!ceph_msg_cache);
261 kmem_cache_destroy(ceph_msg_cache);
262 ceph_msg_cache = NULL;
263}
264
15417167 265static void _ceph_msgr_exit(void)
6173d1f0 266{
d3002b97 267 if (ceph_msgr_wq) {
6173d1f0 268 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 269 ceph_msgr_wq = NULL;
270 }
6173d1f0 271
6173d1f0 272 BUG_ON(zero_page == NULL);
09cbfeaf 273 put_page(zero_page);
6173d1f0 274 zero_page = NULL;
d920ff6f
BC		 275
276 ceph_msgr_slab_exit();
6173d1f0
AE		 277}
278
3d14c5d2 279int ceph_msgr_init(void)
31b8006e 280{
d920ff6f
BC		 281 if (ceph_msgr_slab_init())
282 return -ENOMEM;
283
57666519
AE		 284 BUG_ON(zero_page != NULL);
285 zero_page = ZERO_PAGE(0);
09cbfeaf 286 get_page(zero_page);
57666519 287
f9865f06
ID		 288 /*
289 * The number of active work items is limited by the number of
290 * connections, so leave @max_active at default.
291 */
292 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
6173d1f0
AE		 293 if (ceph_msgr_wq)
294 return 0;
57666519 295
6173d1f0
AE		 296 pr_err("msgr_init failed to create workqueue\n");
297 _ceph_msgr_exit();
57666519 298
6173d1f0 299 return -ENOMEM;
31b8006e 300}
3d14c5d2 301EXPORT_SYMBOL(ceph_msgr_init);
31b8006e
SW		 302
303void ceph_msgr_exit(void)
304{
57666519 305 BUG_ON(ceph_msgr_wq == NULL);
57666519 306
6173d1f0 307 _ceph_msgr_exit();
31b8006e 308}
3d14c5d2 309EXPORT_SYMBOL(ceph_msgr_exit);
31b8006e 310
cd84db6e 311void ceph_msgr_flush(void)
a922d38f
SW		 312{
313 flush_workqueue(ceph_msgr_wq);
314}
3d14c5d2 315EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 316
ce2c8903
AE		 317/* Connection socket state transition functions */
318
319static void con_sock_state_init(struct ceph_connection *con)
320{
321 int old_state;
322
323 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
324 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
325 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 326 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
327 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 328}
329
330static void con_sock_state_connecting(struct ceph_connection *con)
331{
332 int old_state;
333
334 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
335 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
336 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 337 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
338 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 339}
340
341static void con_sock_state_connected(struct ceph_connection *con)
342{
343 int old_state;
344
345 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
346 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
347 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 348 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
349 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 350}
351
352static void con_sock_state_closing(struct ceph_connection *con)
353{
354 int old_state;
355
356 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
357 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
358 old_state != CON_SOCK_STATE_CONNECTED &&
359 old_state != CON_SOCK_STATE_CLOSING))
360 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 361 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
362 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 363}
364
365static void con_sock_state_closed(struct ceph_connection *con)
366{
367 int old_state;
368
369 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
370 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 371 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 372 old_state != CON_SOCK_STATE_CONNECTING &&
373 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 374 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 375 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
376 CON_SOCK_STATE_CLOSED);
ce2c8903 377}
a922d38f 378
31b8006e
SW		 379/*
380 * socket callback functions
381 */
382
383/* data available on socket, or listen socket received a connect */
676d2369 384static void ceph_sock_data_ready(struct sock *sk)
31b8006e 385{
bd406145 386 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 387 if (atomic_read(&con->msgr->stopping)) {
388 return;
389 }
bd406145 390
31b8006e 391 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 392 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 393 con, con->state);
394 queue_con(con);
395 }
396}
397
398/* socket has buffer space for writing */
327800bd 399static void ceph_sock_write_space(struct sock *sk)
31b8006e 400{
d3002b97 401 struct ceph_connection *con = sk->sk_user_data;
31b8006e 402
182fac26
JS		 403 /* only queue to workqueue if there is data we want to write,
404 * and there is sufficient space in the socket buffer to accept
327800bd 405 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 406 * doesn't get called again until try_write() fills the socket
407 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
408 * and net/core/stream.c:sk_stream_write_space().
409 */
c9ffc77a 410 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
64dc6130 411 if (sk_stream_is_writeable(sk)) {
327800bd 412 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 413 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
414 queue_con(con);
415 }
31b8006e 416 } else {
327800bd 417 dout("%s %p nothing to write\n", __func__, con);
31b8006e 418 }
31b8006e
SW		 419}
420
421/* socket's state has changed */
327800bd 422static void ceph_sock_state_change(struct sock *sk)
31b8006e 423{
bd406145 424 struct ceph_connection *con = sk->sk_user_data;
31b8006e 425
327800bd 426 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 427 con, con->state, sk->sk_state);
428
31b8006e
SW		 429 switch (sk->sk_state) {
430 case TCP_CLOSE:
327800bd 431 dout("%s TCP_CLOSE\n", __func__);
31b8006e 432 case TCP_CLOSE_WAIT:
327800bd 433 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 434 con_sock_state_closing(con);
c9ffc77a 435 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 436 queue_con(con);
31b8006e
SW		 437 break;
438 case TCP_ESTABLISHED:
327800bd 439 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 440 con_sock_state_connected(con);
31b8006e
SW		 441 queue_con(con);
442 break;
d3002b97
AE		 443 default: /* Everything else is uninteresting */
444 break;
31b8006e
SW		 445 }
446}
447
448/*
449 * set up socket callbacks
450 */
451static void set_sock_callbacks(struct socket *sock,
452 struct ceph_connection *con)
453{
454 struct sock *sk = sock->sk;
bd406145 455 sk->sk_user_data = con;
327800bd
AE		 456 sk->sk_data_ready = ceph_sock_data_ready;
457 sk->sk_write_space = ceph_sock_write_space;
458 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 459}
460
461
462/*
463 * socket helpers
464 */
465
466/*
467 * initiate connection to a remote socket.
468 */
41617d0c 469static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 470{
f91d3471 471 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e 472 struct socket *sock;
633ee407 473 unsigned int noio_flag;
31b8006e
SW		 474 int ret;
475
476 BUG_ON(con->sock);
633ee407
ID		 477
478 /* sock_create_kern() allocates with GFP_KERNEL */
479 noio_flag = memalloc_noio_save();
757856d2 480 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
eeb1bd5c 481 SOCK_STREAM, IPPROTO_TCP, &sock);
633ee407 482 memalloc_noio_restore(noio_flag);
31b8006e 483 if (ret)
41617d0c 484 return ret;
6d7fdb0a 485 sock->sk->sk_allocation = GFP_NOFS;
31b8006e 486
a6a5349d
SW		 487#ifdef CONFIG_LOCKDEP
488 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
489#endif
490
31b8006e
SW		 491 set_sock_callbacks(sock, con);
492
3d14c5d2 493 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 494
89a86be0 495 con_sock_state_connecting(con);
f91d3471
SW		 496 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
497 O_NONBLOCK);
31b8006e
SW		 498 if (ret == -EINPROGRESS) {
499 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 500 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 501 sock->sk->sk_state);
a5bc3129 502 } else if (ret < 0) {
31b8006e 503 pr_err("connect %s error %d\n",
3d14c5d2 504 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 505 sock_release(sock);
41617d0c 506 return ret;
a5bc3129 507 }
89baaa57 508
859bff51 509 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
ba988f87
CH		 510 int optval = 1;
511
512 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
513 (char *)&optval, sizeof(optval));
514 if (ret)
515 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
516 ret);
517 }
518
a5bc3129 519 con->sock = sock;
41617d0c 520 return 0;
31b8006e
SW		 521}
522
523static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
524{
525 struct kvec iov = {buf, len};
526 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 527 int r;
31b8006e 528
100803a8
AV		 529 iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, &iov, 1, len);
530 r = sock_recvmsg(sock, &msg, msg.msg_flags);
98bdb0aa
SW		 531 if (r == -EAGAIN)
532 r = 0;
533 return r;
31b8006e
SW		 534}
535
afb3d90e
AE		 536static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
537 int page_offset, size_t length)
538{
100803a8
AV		 539 struct bio_vec bvec = {
540 .bv_page = page,
541 .bv_offset = page_offset,
542 .bv_len = length
543 };
544 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
545 int r;
afb3d90e
AE		 546
547 BUG_ON(page_offset + length > PAGE_SIZE);
100803a8
AV		 548 iov_iter_bvec(&msg.msg_iter, READ | ITER_BVEC, &bvec, 1, length);
549 r = sock_recvmsg(sock, &msg, msg.msg_flags);
550 if (r == -EAGAIN)
551 r = 0;
552 return r;
afb3d90e
AE		 553}
554
31b8006e
SW		 555/*
556 * write something. @more is true if caller will be sending more data
557 * shortly.
558 */
559static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
560 size_t kvlen, size_t len, int more)
561{
562 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
42961d23 563 int r;
31b8006e
SW		 564
565 if (more)
566 msg.msg_flags |= MSG_MORE;
567 else
568 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
569
42961d23
SW		 570 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
571 if (r == -EAGAIN)
572 r = 0;
573 return r;
31b8006e
SW		 574}
575
178eda29 576static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
e1dcb128 577 int offset, size_t size, bool more)
31739139
AE		 578{
579 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
580 int ret;
581
582 ret = kernel_sendpage(sock, page, offset, size, flags);
583 if (ret == -EAGAIN)
584 ret = 0;
585
586 return ret;
587}
588
178eda29
CC		 589static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
590 int offset, size_t size, bool more)
591{
61ff6e9b
AV		 592 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
593 struct bio_vec bvec;
178eda29 594 int ret;
178eda29
CC		 595
596 /* sendpage cannot properly handle pages with page_count == 0,
597 * we need to fallback to sendmsg if that's the case */
598 if (page_count(page) >= 1)
599 return __ceph_tcp_sendpage(sock, page, offset, size, more);
600
61ff6e9b
AV		 601 bvec.bv_page = page;
602 bvec.bv_offset = offset;
603 bvec.bv_len = size;
604
605 if (more)
606 msg.msg_flags |= MSG_MORE;
607 else
608 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
609
610 iov_iter_bvec(&msg.msg_iter, WRITE | ITER_BVEC, &bvec, 1, size);
611 ret = sock_sendmsg(sock, &msg);
612 if (ret == -EAGAIN)
613 ret = 0;
178eda29
CC		 614
615 return ret;
616}
31b8006e
SW		 617
618/*
619 * Shutdown/close the socket for the given connection.
620 */
621static int con_close_socket(struct ceph_connection *con)
622{
8007b8d6 623 int rc = 0;
31b8006e
SW		 624
625 dout("con_close_socket on %p sock %p\n", con, con->sock);
8007b8d6
SW		 626 if (con->sock) {
627 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
628 sock_release(con->sock);
629 con->sock = NULL;
630 }
456ea468
AE		 631
632 /*
4a861692 633 * Forcibly clear the SOCK_CLOSED flag. It gets set
456ea468
AE		 634 * independent of the connection mutex, and we could have
635 * received a socket close event before we had the chance to
636 * shut the socket down.
637 */
c9ffc77a 638 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
8007b8d6 639
ce2c8903 640 con_sock_state_closed(con);
31b8006e
SW		 641 return rc;
642}
643
644/*
645 * Reset a connection. Discard all incoming and outgoing messages
646 * and clear *_seq state.
647 */
648static void ceph_msg_remove(struct ceph_msg *msg)
649{
650 list_del_init(&msg->list_head);
38941f80 651
31b8006e
SW		 652 ceph_msg_put(msg);
653}
654static void ceph_msg_remove_list(struct list_head *head)
655{
656 while (!list_empty(head)) {
657 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
658 list_head);
659 ceph_msg_remove(msg);
660 }
661}
662
663static void reset_connection(struct ceph_connection *con)
664{
665 /* reset connection, out_queue, msg_ and connect_seq */
666 /* discard existing out_queue and msg_seq */
0fa6ebc6 667 dout("reset_connection %p\n", con);
31b8006e
SW		 668 ceph_msg_remove_list(&con->out_queue);
669 ceph_msg_remove_list(&con->out_sent);
670
cf3e5c40 671 if (con->in_msg) {
38941f80 672 BUG_ON(con->in_msg->con != con);
cf3e5c40
SW		 673 ceph_msg_put(con->in_msg);
674 con->in_msg = NULL;
675 }
676
31b8006e
SW		 677 con->connect_seq = 0;
678 con->out_seq = 0;
c86a2930 679 if (con->out_msg) {
583d0fef 680 BUG_ON(con->out_msg->con != con);
c86a2930
SW		 681 ceph_msg_put(con->out_msg);
682 con->out_msg = NULL;
683 }
31b8006e 684 con->in_seq = 0;
0e0d5e0c 685 con->in_seq_acked = 0;
67645d76
ID		 686
687 con->out_skip = 0;
31b8006e
SW		 688}
689
690/*
691 * mark a peer down. drop any open connections.
692 */
693void ceph_con_close(struct ceph_connection *con)
694{
8c50c817 695 mutex_lock(&con->mutex);
3d14c5d2
YS		 696 dout("con_close %p peer %s\n", con,
697 ceph_pr_addr(&con->peer_addr.in_addr));
8dacc7da 698 con->state = CON_STATE_CLOSED;
a5988c49 699
c9ffc77a
AE		 700 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
701 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
702 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
703 con_flag_clear(con, CON_FLAG_BACKOFF);
a5988c49 704
31b8006e 705 reset_connection(con);
6f2bc3ff 706 con->peer_global_seq = 0;
37ab77ac 707 cancel_con(con);
ee76e073 708 con_close_socket(con);
ec302645 709 mutex_unlock(&con->mutex);
31b8006e 710}
3d14c5d2 711EXPORT_SYMBOL(ceph_con_close);
31b8006e 712
31b8006e
SW		 713/*
714 * Reopen a closed connection, with a new peer address.
715 */
b7a9e5dd
SW		 716void ceph_con_open(struct ceph_connection *con,
717 __u8 entity_type, __u64 entity_num,
718 struct ceph_entity_addr *addr)
31b8006e 719{
5469155f 720 mutex_lock(&con->mutex);
3d14c5d2 721 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
8dacc7da 722
122070a2 723 WARN_ON(con->state != CON_STATE_CLOSED);
8dacc7da 724 con->state = CON_STATE_PREOPEN;
a5988c49 725
b7a9e5dd
SW		 726 con->peer_name.type = (__u8) entity_type;
727 con->peer_name.num = cpu_to_le64(entity_num);
728
31b8006e 729 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 730 con->delay = 0; /* reset backoff memory */
5469155f 731 mutex_unlock(&con->mutex);
31b8006e
SW		 732 queue_con(con);
733}
3d14c5d2 734EXPORT_SYMBOL(ceph_con_open);
31b8006e 735
87b315a5
SW		 736/*
737 * return true if this connection ever successfully opened
738 */
739bool ceph_con_opened(struct ceph_connection *con)
740{
741 return con->connect_seq > 0;
742}
743
31b8006e
SW		 744/*
745 * initialize a new connection.
746 */
1bfd89f4
AE		 747void ceph_con_init(struct ceph_connection *con, void *private,
748 const struct ceph_connection_operations *ops,
b7a9e5dd 749 struct ceph_messenger *msgr)
31b8006e
SW		 750{
751 dout("con_init %p\n", con);
752 memset(con, 0, sizeof(*con));
1bfd89f4
AE		 753 con->private = private;
754 con->ops = ops;
31b8006e 755 con->msgr = msgr;
ce2c8903
AE		 756
757 con_sock_state_init(con);
758
ec302645 759 mutex_init(&con->mutex);
31b8006e
SW		 760 INIT_LIST_HEAD(&con->out_queue);
761 INIT_LIST_HEAD(&con->out_sent);
68931622 762 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
a5988c49 763
8dacc7da 764 con->state = CON_STATE_CLOSED;
31b8006e 765}
3d14c5d2 766EXPORT_SYMBOL(ceph_con_init);
31b8006e
SW		 767
768
769/*
770 * We maintain a global counter to order connection attempts. Get
771 * a unique seq greater than @gt.
772 */
773static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
774{
775 u32 ret;
776
777 spin_lock(&msgr->global_seq_lock);
778 if (msgr->global_seq < gt)
779 msgr->global_seq = gt;
780 ret = ++msgr->global_seq;
781 spin_unlock(&msgr->global_seq_lock);
782 return ret;
783}
784
e2200423 785static void con_out_kvec_reset(struct ceph_connection *con)
859eb799 786{
67645d76
ID		 787 BUG_ON(con->out_skip);
788
859eb799
AE		 789 con->out_kvec_left = 0;
790 con->out_kvec_bytes = 0;
791 con->out_kvec_cur = &con->out_kvec[0];
792}
793
e2200423 794static void con_out_kvec_add(struct ceph_connection *con,
859eb799
AE		 795 size_t size, void *data)
796{
67645d76 797 int index = con->out_kvec_left;
859eb799 798
67645d76 799 BUG_ON(con->out_skip);
859eb799
AE		 800 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
801
802 con->out_kvec[index].iov_len = size;
803 con->out_kvec[index].iov_base = data;
804 con->out_kvec_left++;
805 con->out_kvec_bytes += size;
806}
31b8006e 807
67645d76
ID		 808/*
809 * Chop off a kvec from the end. Return residual number of bytes for
810 * that kvec, i.e. how many bytes would have been written if the kvec
811 * hadn't been nuked.
812 */
813static int con_out_kvec_skip(struct ceph_connection *con)
814{
815 int off = con->out_kvec_cur - con->out_kvec;
816 int skip = 0;
817
818 if (con->out_kvec_bytes > 0) {
819 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
820 BUG_ON(con->out_kvec_bytes < skip);
821 BUG_ON(!con->out_kvec_left);
822 con->out_kvec_bytes -= skip;
823 con->out_kvec_left--;
824 }
825
826 return skip;
827}
828
df6ad1f9 829#ifdef CONFIG_BLOCK
6aaa4511
AE		 830
831/*
832 * For a bio data item, a piece is whatever remains of the next
833 * entry in the current bio iovec, or the first entry in the next
834 * bio in the list.
835 */
8ae4f4f5 836static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 837 size_t length)
6aaa4511 838{
8ae4f4f5 839 struct ceph_msg_data *data = cursor->data;
6aaa4511
AE		 840 struct bio *bio;
841
842 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
843
844 bio = data->bio;
845 BUG_ON(!bio);
6aaa4511 846
ca8b3a69 847 cursor->resid = min(length, data->bio_length);
6aaa4511 848 cursor->bio = bio;
f38a5181
KO		 849 cursor->bvec_iter = bio->bi_iter;
850 cursor->last_piece =
851 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
6aaa4511
AE		 852}
853
8ae4f4f5 854static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 855 size_t *page_offset,
856 size_t *length)
857{
8ae4f4f5 858 struct ceph_msg_data *data = cursor->data;
6aaa4511 859 struct bio *bio;
f38a5181 860 struct bio_vec bio_vec;
6aaa4511
AE		 861
862 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
863
864 bio = cursor->bio;
865 BUG_ON(!bio);
866
f38a5181 867 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511 868
f38a5181 869 *page_offset = (size_t) bio_vec.bv_offset;
6aaa4511 870 BUG_ON(*page_offset >= PAGE_SIZE);
25aff7c5
AE		 871 if (cursor->last_piece) /* pagelist offset is always 0 */
872 *length = cursor->resid;
873 else
f38a5181 874 *length = (size_t) bio_vec.bv_len;
25aff7c5 875 BUG_ON(*length > cursor->resid);
5df521b1 876 BUG_ON(*page_offset + *length > PAGE_SIZE);
6aaa4511 877
f38a5181 878 return bio_vec.bv_page;
6aaa4511
AE		 879}
880
8ae4f4f5
AE		 881static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
882 size_t bytes)
6aaa4511 883{
6aaa4511 884 struct bio *bio;
f38a5181 885 struct bio_vec bio_vec;
6aaa4511 886
8ae4f4f5 887 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
6aaa4511
AE		 888
889 bio = cursor->bio;
890 BUG_ON(!bio);
891
f38a5181 892 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
6aaa4511
AE		 893
894 /* Advance the cursor offset */
895
25aff7c5
AE		 896 BUG_ON(cursor->resid < bytes);
897 cursor->resid -= bytes;
f38a5181
KO		 898
899 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
900
901 if (bytes < bio_vec.bv_len)
6aaa4511
AE		 902 return false; /* more bytes to process in this segment */
903
904 /* Move on to the next segment, and possibly the next bio */
905
f38a5181 906 if (!cursor->bvec_iter.bi_size) {
6aaa4511 907 bio = bio->bi_next;
0ec1d15e
ID		 908 cursor->bio = bio;
909 if (bio)
910 cursor->bvec_iter = bio->bi_iter;
911 else
912 memset(&cursor->bvec_iter, 0,
913 sizeof(cursor->bvec_iter));
6aaa4511 914 }
6aaa4511 915
25aff7c5
AE		 916 if (!cursor->last_piece) {
917 BUG_ON(!cursor->resid);
918 BUG_ON(!bio);
919 /* A short read is OK, so use <= rather than == */
f38a5181 920 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
6aaa4511 921 cursor->last_piece = true;
25aff7c5 922 }
6aaa4511
AE		 923
924 return true;
925}
ea96571f 926#endif /* CONFIG_BLOCK */
df6ad1f9 927
e766d7b5
AE		 928/*
929 * For a page array, a piece comes from the first page in the array
930 * that has not already been fully consumed.
931 */
8ae4f4f5 932static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 933 size_t length)
e766d7b5 934{
8ae4f4f5 935 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 936 int page_count;
937
938 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
939
940 BUG_ON(!data->pages);
941 BUG_ON(!data->length);
942
ca8b3a69 943 cursor->resid = min(length, data->length);
e766d7b5 944 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 945 cursor->page_offset = data->alignment & ~PAGE_MASK;
946 cursor->page_index = 0;
56fc5659
AE		 947 BUG_ON(page_count > (int)USHRT_MAX);
948 cursor->page_count = (unsigned short)page_count;
949 BUG_ON(length > SIZE_MAX - cursor->page_offset);
5f740d7e 950 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 951}
952
8ae4f4f5
AE		 953static struct page *
954ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
955 size_t *page_offset, size_t *length)
e766d7b5 956{
8ae4f4f5 957 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 958
959 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
960
961 BUG_ON(cursor->page_index >= cursor->page_count);
962 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 963
964 *page_offset = cursor->page_offset;
25aff7c5 965 if (cursor->last_piece)
e766d7b5 966 *length = cursor->resid;
25aff7c5 967 else
e766d7b5 968 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 969
970 return data->pages[cursor->page_index];
971}
972
8ae4f4f5 973static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 974 size_t bytes)
975{
8ae4f4f5 976 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 977
978 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 979
980 /* Advance the cursor page offset */
981
982 cursor->resid -= bytes;
5df521b1
AE		 983 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
984 if (!bytes || cursor->page_offset)
e766d7b5
AE		 985 return false; /* more bytes to process in the current page */
986
d90deda6
YZ		 987 if (!cursor->resid)
988 return false; /* no more data */
989
5df521b1 990 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 991
992 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 993 cursor->page_index++;
25aff7c5 994 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 995
996 return true;
997}
998
fe38a2b6 999/*
dd236fcb
AE		 1000 * For a pagelist, a piece is whatever remains to be consumed in the
1001 * first page in the list, or the front of the next page.
fe38a2b6 1002 */
8ae4f4f5
AE		 1003static void
1004ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 1005 size_t length)
fe38a2b6 1006{
8ae4f4f5 1007 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1008 struct ceph_pagelist *pagelist;
1009 struct page *page;
1010
dd236fcb 1011 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 1012
1013 pagelist = data->pagelist;
1014 BUG_ON(!pagelist);
25aff7c5
AE		 1015
1016 if (!length)
fe38a2b6
AE		 1017 return; /* pagelist can be assigned but empty */
1018
1019 BUG_ON(list_empty(&pagelist->head));
1020 page = list_first_entry(&pagelist->head, struct page, lru);
1021
ca8b3a69 1022 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 1023 cursor->page = page;
1024 cursor->offset = 0;
a51b272e 1025 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1026}
1027
8ae4f4f5
AE		 1028static struct page *
1029ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1030 size_t *page_offset, size_t *length)
fe38a2b6 1031{
8ae4f4f5 1032 struct ceph_msg_data *data = cursor->data;
fe38a2b6 1033 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 1034
1035 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1036
1037 pagelist = data->pagelist;
1038 BUG_ON(!pagelist);
1039
1040 BUG_ON(!cursor->page);
25aff7c5 1041 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 1042
5df521b1 1043 /* offset of first page in pagelist is always 0 */
fe38a2b6 1044 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 1045 if (cursor->last_piece)
25aff7c5
AE		 1046 *length = cursor->resid;
1047 else
1048 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1049
8ae4f4f5 1050 return cursor->page;
fe38a2b6
AE		 1051}
1052
8ae4f4f5 1053static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1054 size_t bytes)
fe38a2b6 1055{
8ae4f4f5 1056 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1057 struct ceph_pagelist *pagelist;
1058
1059 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1060
1061 pagelist = data->pagelist;
1062 BUG_ON(!pagelist);
25aff7c5
AE		 1063
1064 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1065 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1066
1067 /* Advance the cursor offset */
1068
25aff7c5 1069 cursor->resid -= bytes;
fe38a2b6 1070 cursor->offset += bytes;
5df521b1 1071 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1072 if (!bytes || cursor->offset & ~PAGE_MASK)
1073 return false; /* more bytes to process in the current page */
1074
d90deda6
YZ		 1075 if (!cursor->resid)
1076 return false; /* no more data */
1077
fe38a2b6
AE		 1078 /* Move on to the next page */
1079
1080 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
17ddc49b 1081 cursor->page = list_next_entry(cursor->page, lru);
25aff7c5 1082 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1083
1084 return true;
1085}
1086
dd236fcb
AE		 1087/*
1088 * Message data is handled (sent or received) in pieces, where each
1089 * piece resides on a single page. The network layer might not
1090 * consume an entire piece at once. A data item's cursor keeps
1091 * track of which piece is next to process and how much remains to
1092 * be processed in that piece. It also tracks whether the current
1093 * piece is the last one in the data item.
1094 */
ca8b3a69 1095static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1096{
ca8b3a69 1097 size_t length = cursor->total_resid;
8ae4f4f5 1098
8ae4f4f5 1099 switch (cursor->data->type) {
dd236fcb 1100 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1101 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1102 break;
e766d7b5 1103 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1104 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1105 break;
dd236fcb
AE		 1106#ifdef CONFIG_BLOCK
1107 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1108 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1109 break;
dd236fcb 1110#endif /* CONFIG_BLOCK */
6aaa4511 1111 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1112 default:
1113 /* BUG(); */
1114 break;
1115 }
8ae4f4f5 1116 cursor->need_crc = true;
dd236fcb
AE		 1117}
1118
ca8b3a69
AE		 1119static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1120{
1121 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1122 struct ceph_msg_data *data;
1123
1124 BUG_ON(!length);
1125 BUG_ON(length > msg->data_length);
1126 BUG_ON(list_empty(&msg->data));
1127
ca8b3a69
AE		 1128 cursor->data_head = &msg->data;
1129 cursor->total_resid = length;
1130 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1131 cursor->data = data;
1132
1133 __ceph_msg_data_cursor_init(cursor);
1134}
1135
dd236fcb
AE		 1136/*
1137 * Return the page containing the next piece to process for a given
1138 * data item, and supply the page offset and length of that piece.
1139 * Indicate whether this is the last piece in this data item.
1140 */
8ae4f4f5
AE		 1141static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1142 size_t *page_offset, size_t *length,
dd236fcb
AE		 1143 bool *last_piece)
1144{
1145 struct page *page;
1146
8ae4f4f5 1147 switch (cursor->data->type) {
dd236fcb 1148 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1149 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1150 break;
e766d7b5 1151 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1152 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1153 break;
dd236fcb
AE		 1154#ifdef CONFIG_BLOCK
1155 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1156 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1157 break;
dd236fcb 1158#endif /* CONFIG_BLOCK */
6aaa4511 1159 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1160 default:
1161 page = NULL;
1162 break;
1163 }
1164 BUG_ON(!page);
1165 BUG_ON(*page_offset + *length > PAGE_SIZE);
1166 BUG_ON(!*length);
1167 if (last_piece)
8ae4f4f5 1168 *last_piece = cursor->last_piece;
dd236fcb
AE		 1169
1170 return page;
1171}
1172
1173/*
1174 * Returns true if the result moves the cursor on to the next piece
1175 * of the data item.
1176 */
1759f7b0
ID		 1177static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1178 size_t bytes)
dd236fcb
AE		 1179{
1180 bool new_piece;
1181
25aff7c5 1182 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1183 switch (cursor->data->type) {
dd236fcb 1184 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1185 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1186 break;
e766d7b5 1187 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1188 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1189 break;
dd236fcb
AE		 1190#ifdef CONFIG_BLOCK
1191 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1192 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1193 break;
dd236fcb 1194#endif /* CONFIG_BLOCK */
6aaa4511 1195 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1196 default:
1197 BUG();
1198 break;
1199 }
ca8b3a69 1200 cursor->total_resid -= bytes;
dd236fcb 1201
ca8b3a69
AE		 1202 if (!cursor->resid && cursor->total_resid) {
1203 WARN_ON(!cursor->last_piece);
1204 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
17ddc49b 1205 cursor->data = list_next_entry(cursor->data, links);
ca8b3a69 1206 __ceph_msg_data_cursor_init(cursor);
a51b272e 1207 new_piece = true;
ca8b3a69 1208 }
a51b272e 1209 cursor->need_crc = new_piece;
dd236fcb
AE		 1210}
1211
dbc0d3ca
ID		 1212static size_t sizeof_footer(struct ceph_connection *con)
1213{
1214 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1215 sizeof(struct ceph_msg_footer) :
1216 sizeof(struct ceph_msg_footer_old);
1217}
1218
98fa5dd8 1219static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1220{
739c905b 1221 BUG_ON(!msg);
25aff7c5 1222 BUG_ON(!data_len);
739c905b 1223
4c59b4a2 1224 /* Initialize data cursor */
fe38a2b6 1225
8ae4f4f5 1226 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1227}
1228
31b8006e
SW		 1229/*
1230 * Prepare footer for currently outgoing message, and finish things
1231 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1232 */
859eb799 1233static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1234{
1235 struct ceph_msg *m = con->out_msg;
1236
fd154f3c
AE		 1237 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1238
31b8006e 1239 dout("prepare_write_message_footer %p\n", con);
89f08173 1240 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
33d07337
YZ		 1241 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1242 if (con->ops->sign_message)
79dbd1ba 1243 con->ops->sign_message(m);
33d07337
YZ		 1244 else
1245 m->footer.sig = 0;
33d07337
YZ		 1246 } else {
1247 m->old_footer.flags = m->footer.flags;
33d07337 1248 }
31b8006e 1249 con->out_more = m->more_to_follow;
c86a2930 1250 con->out_msg_done = true;
31b8006e
SW		 1251}
1252
1253/*
1254 * Prepare headers for the next outgoing message.
1255 */
1256static void prepare_write_message(struct ceph_connection *con)
1257{
1258 struct ceph_msg *m;
a9a0c51a 1259 u32 crc;
31b8006e 1260
e2200423 1261 con_out_kvec_reset(con);
c86a2930 1262 con->out_msg_done = false;
31b8006e
SW		 1263
1264 /* Sneak an ack in there first? If we can get it into the same
1265 * TCP packet that's a good thing. */
1266 if (con->in_seq > con->in_seq_acked) {
1267 con->in_seq_acked = con->in_seq;
e2200423 1268 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1269 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1270 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1271 &con->out_temp_ack);
31b8006e
SW		 1272 }
1273
38941f80 1274 BUG_ON(list_empty(&con->out_queue));
859eb799 1275 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1276 con->out_msg = m;
38941f80 1277 BUG_ON(m->con != con);
4cf9d544
SW		 1278
1279 /* put message on sent list */
1280 ceph_msg_get(m);
1281 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1282
e84346b7
SW		 1283 /*
1284 * only assign outgoing seq # if we haven't sent this message
1285 * yet. if it is requeued, resend with it's original seq.
1286 */
1287 if (m->needs_out_seq) {
1288 m->hdr.seq = cpu_to_le64(++con->out_seq);
1289 m->needs_out_seq = false;
1290 }
98fa5dd8 1291 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e 1292
98fa5dd8 1293 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1294 m, con->out_seq, le16_to_cpu(m->hdr.type),
1295 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1296 m->data_length);
31b8006e
SW		 1297 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1298
1299 /* tag + hdr + front + middle */
e2200423 1300 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
67645d76 1301 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
e2200423 1302 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1303
31b8006e 1304 if (m->middle)
e2200423 1305 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1306 m->middle->vec.iov_base);
31b8006e 1307
67645d76 1308 /* fill in hdr crc and finalize hdr */
a9a0c51a
AE		 1309 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1310 con->out_msg->hdr.crc = cpu_to_le32(crc);
67645d76 1311 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
a9a0c51a 1312
67645d76 1313 /* fill in front and middle crc, footer */
a9a0c51a
AE		 1314 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1315 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1316 if (m->middle) {
1317 crc = crc32c(0, m->middle->vec.iov_base,
1318 m->middle->vec.iov_len);
1319 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1320 } else
31b8006e 1321 con->out_msg->footer.middle_crc = 0;
739c905b 1322 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1323 le32_to_cpu(con->out_msg->footer.front_crc),
1324 le32_to_cpu(con->out_msg->footer.middle_crc));
67645d76 1325 con->out_msg->footer.flags = 0;
31b8006e
SW		 1326
1327 /* is there a data payload? */
739c905b 1328 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1329 if (m->data_length) {
1330 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1331 con->out_more = 1; /* data + footer will follow */
1332 } else {
31b8006e 1333 /* no, queue up footer too and be done */
859eb799 1334 prepare_write_message_footer(con);
78625051 1335 }
31b8006e 1336
c9ffc77a 1337 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1338}
1339
1340/*
1341 * Prepare an ack.
1342 */
1343static void prepare_write_ack(struct ceph_connection *con)
1344{
1345 dout("prepare_write_ack %p %llu -> %llu\n", con,
1346 con->in_seq_acked, con->in_seq);
1347 con->in_seq_acked = con->in_seq;
1348
e2200423 1349 con_out_kvec_reset(con);
859eb799 1350
e2200423 1351 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1352
31b8006e 1353 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1354 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1355 &con->out_temp_ack);
1356
31b8006e 1357 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1358 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1359}
1360
3a23083b
SW		 1361/*
1362 * Prepare to share the seq during handshake
1363 */
1364static void prepare_write_seq(struct ceph_connection *con)
1365{
1366 dout("prepare_write_seq %p %llu -> %llu\n", con,
1367 con->in_seq_acked, con->in_seq);
1368 con->in_seq_acked = con->in_seq;
1369
1370 con_out_kvec_reset(con);
1371
1372 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1373 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1374 &con->out_temp_ack);
1375
1376 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1377}
1378
31b8006e
SW		 1379/*
1380 * Prepare to write keepalive byte.
1381 */
1382static void prepare_write_keepalive(struct ceph_connection *con)
1383{
1384 dout("prepare_write_keepalive %p\n", con);
e2200423 1385 con_out_kvec_reset(con);
8b9558aa 1386 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1134e091 1387 struct timespec now;
7f61f545 1388
1134e091 1389 ktime_get_real_ts(&now);
8b9558aa 1390 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
7f61f545
ID		 1391 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1392 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1393 &con->out_temp_keepalive2);
8b9558aa
YZ		 1394 } else {
1395 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1396 }
c9ffc77a 1397 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1398}
1399
1400/*
1401 * Connection negotiation.
1402 */
1403
dac1e716
AE		 1404static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1405 int *auth_proto)
4e7a5dcd 1406{
a3530df3 1407 struct ceph_auth_handshake *auth;
b1c6b980
AE		 1408
1409 if (!con->ops->get_authorizer) {
1410 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1411 con->out_connect.authorizer_len = 0;
729796be 1412 return NULL;
b1c6b980
AE		 1413 }
1414
dac1e716 1415 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
a3530df3 1416 if (IS_ERR(auth))
729796be 1417 return auth;
0da5d703 1418
8f43fb53
AE		 1419 con->auth_reply_buf = auth->authorizer_reply_buf;
1420 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
729796be 1421 return auth;
4e7a5dcd
SW		 1422}
1423
31b8006e
SW		 1424/*
1425 * We connected to a peer and are saying hello.
1426 */
e825a66d 1427static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1428{
e2200423
AE		 1429 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1430 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1431 &con->msgr->my_enc_addr);
eed0ef2c 1432
eed0ef2c 1433 con->out_more = 0;
c9ffc77a 1434 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1435}
1436
e825a66d 1437static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1438{
95c96174 1439 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1440 int proto;
dac1e716 1441 int auth_proto;
729796be 1442 struct ceph_auth_handshake *auth;
31b8006e
SW		 1443
1444 switch (con->peer_name.type) {
1445 case CEPH_ENTITY_TYPE_MON:
1446 proto = CEPH_MONC_PROTOCOL;
1447 break;
1448 case CEPH_ENTITY_TYPE_OSD:
1449 proto = CEPH_OSDC_PROTOCOL;
1450 break;
1451 case CEPH_ENTITY_TYPE_MDS:
1452 proto = CEPH_MDSC_PROTOCOL;
1453 break;
1454 default:
1455 BUG();
1456 }
1457
1458 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1459 con->connect_seq, global_seq, proto);
4e7a5dcd 1460
859bff51
ID		 1461 con->out_connect.features =
1462 cpu_to_le64(from_msgr(con->msgr)->supported_features);
31b8006e
SW		 1463 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1464 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1465 con->out_connect.global_seq = cpu_to_le32(global_seq);
1466 con->out_connect.protocol_version = cpu_to_le32(proto);
1467 con->out_connect.flags = 0;
31b8006e 1468
dac1e716
AE		 1469 auth_proto = CEPH_AUTH_UNKNOWN;
1470 auth = get_connect_authorizer(con, &auth_proto);
729796be
AE		 1471 if (IS_ERR(auth))
1472 return PTR_ERR(auth);
3da54776 1473
dac1e716 1474 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
3da54776
AE		 1475 con->out_connect.authorizer_len = auth ?
1476 cpu_to_le32(auth->authorizer_buf_len) : 0;
1477
e2200423 1478 con_out_kvec_add(con, sizeof (con->out_connect),
3da54776
AE		 1479 &con->out_connect);
1480 if (auth && auth->authorizer_buf_len)
e2200423 1481 con_out_kvec_add(con, auth->authorizer_buf_len,
3da54776 1482 auth->authorizer_buf);
859eb799 1483
31b8006e 1484 con->out_more = 0;
c9ffc77a 1485 con_flag_set(con, CON_FLAG_WRITE_PENDING);
4e7a5dcd 1486
e10c758e 1487 return 0;
31b8006e
SW		 1488}
1489
31b8006e
SW		 1490/*
1491 * write as much of pending kvecs to the socket as we can.
1492 * 1 -> done
1493 * 0 -> socket full, but more to do
1494 * <0 -> error
1495 */
1496static int write_partial_kvec(struct ceph_connection *con)
1497{
1498 int ret;
1499
1500 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1501 while (con->out_kvec_bytes > 0) {
1502 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1503 con->out_kvec_left, con->out_kvec_bytes,
1504 con->out_more);
1505 if (ret <= 0)
1506 goto out;
1507 con->out_kvec_bytes -= ret;
1508 if (con->out_kvec_bytes == 0)
1509 break; /* done */
f42299e6
AE		 1510
1511 /* account for full iov entries consumed */
1512 while (ret >= con->out_kvec_cur->iov_len) {
1513 BUG_ON(!con->out_kvec_left);
1514 ret -= con->out_kvec_cur->iov_len;
1515 con->out_kvec_cur++;
1516 con->out_kvec_left--;
1517 }
1518 /* and for a partially-consumed entry */
1519 if (ret) {
1520 con->out_kvec_cur->iov_len -= ret;
1521 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1522 }
1523 }
1524 con->out_kvec_left = 0;
31b8006e
SW		 1525 ret = 1;
1526out:
1527 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1528 con->out_kvec_bytes, con->out_kvec_left, ret);
1529 return ret; /* done! */
1530}
1531
35b62808
AE		 1532static u32 ceph_crc32c_page(u32 crc, struct page *page,
1533 unsigned int page_offset,
1534 unsigned int length)
1535{
1536 char *kaddr;
1537
1538 kaddr = kmap(page);
1539 BUG_ON(kaddr == NULL);
1540 crc = crc32c(crc, kaddr + page_offset, length);
1541 kunmap(page);
1542
1543 return crc;
1544}
31b8006e
SW		 1545/*
1546 * Write as much message data payload as we can. If we finish, queue
1547 * up the footer.
1548 * 1 -> done, footer is now queued in out_kvec[].
1549 * 0 -> socket full, but more to do
1550 * <0 -> error
1551 */
34d2d200 1552static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1553{
1554 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1555 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 1556 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
f5db90bc 1557 u32 crc;
31b8006e 1558
859a35d5 1559 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1560
5240d9f9 1561 if (list_empty(&msg->data))
4c59b4a2
AE		 1562 return -EINVAL;
1563
5821bd8c
AE		 1564 /*
1565 * Iterate through each page that contains data to be
1566 * written, and send as much as possible for each.
1567 *
1568 * If we are calculating the data crc (the default), we will
1569 * need to map the page. If we have no pages, they have
1570 * been revoked, so use the zero page.
1571 */
f5db90bc 1572 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
643c68a4 1573 while (cursor->resid) {
8a166d05 1574 struct page *page;
e387d525
AE		 1575 size_t page_offset;
1576 size_t length;
8a166d05 1577 bool last_piece;
f5db90bc 1578 int ret;
68b4476b 1579
343128ce
SB		 1580 page = ceph_msg_data_next(cursor, &page_offset, &length,
1581 &last_piece);
e387d525 1582 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
c2cfa194 1583 length, !last_piece);
f5db90bc
AE		 1584 if (ret <= 0) {
1585 if (do_datacrc)
1586 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1587
f5db90bc
AE		 1588 return ret;
1589 }
143334ff
AE		 1590 if (do_datacrc && cursor->need_crc)
1591 crc = ceph_crc32c_page(crc, page, page_offset, length);
1759f7b0 1592 ceph_msg_data_advance(cursor, (size_t)ret);
31b8006e
SW		 1593 }
1594
34d2d200 1595 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1596
1597 /* prepare and queue up footer, too */
f5db90bc
AE		 1598 if (do_datacrc)
1599 msg->footer.data_crc = cpu_to_le32(crc);
1600 else
84ca8fc8 1601 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1602 con_out_kvec_reset(con);
859eb799 1603 prepare_write_message_footer(con);
f5db90bc
AE		 1604
1605 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1606}
1607
1608/*
1609 * write some zeros
1610 */
1611static int write_partial_skip(struct ceph_connection *con)
1612{
1613 int ret;
1614
67645d76 1615 dout("%s %p %d left\n", __func__, con, con->out_skip);
31b8006e 1616 while (con->out_skip > 0) {
09cbfeaf 1617 size_t size = min(con->out_skip, (int) PAGE_SIZE);
31b8006e 1618
e1dcb128 1619 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1620 if (ret <= 0)
1621 goto out;
1622 con->out_skip -= ret;
1623 }
1624 ret = 1;
1625out:
1626 return ret;
1627}
1628
1629/*
1630 * Prepare to read connection handshake, or an ack.
1631 */
eed0ef2c
SW		 1632static void prepare_read_banner(struct ceph_connection *con)
1633{
1634 dout("prepare_read_banner %p\n", con);
1635 con->in_base_pos = 0;
1636}
1637
31b8006e
SW		 1638static void prepare_read_connect(struct ceph_connection *con)
1639{
1640 dout("prepare_read_connect %p\n", con);
1641 con->in_base_pos = 0;
1642}
1643
1644static void prepare_read_ack(struct ceph_connection *con)
1645{
1646 dout("prepare_read_ack %p\n", con);
1647 con->in_base_pos = 0;
1648}
1649
3a23083b
SW		 1650static void prepare_read_seq(struct ceph_connection *con)
1651{
1652 dout("prepare_read_seq %p\n", con);
1653 con->in_base_pos = 0;
1654 con->in_tag = CEPH_MSGR_TAG_SEQ;
1655}
1656
31b8006e
SW		 1657static void prepare_read_tag(struct ceph_connection *con)
1658{
1659 dout("prepare_read_tag %p\n", con);
1660 con->in_base_pos = 0;
1661 con->in_tag = CEPH_MSGR_TAG_READY;
1662}
1663
8b9558aa
YZ		 1664static void prepare_read_keepalive_ack(struct ceph_connection *con)
1665{
1666 dout("prepare_read_keepalive_ack %p\n", con);
1667 con->in_base_pos = 0;
1668}
1669
31b8006e
SW		 1670/*
1671 * Prepare to read a message.
1672 */
1673static int prepare_read_message(struct ceph_connection *con)
1674{
1675 dout("prepare_read_message %p\n", con);
1676 BUG_ON(con->in_msg != NULL);
1677 con->in_base_pos = 0;
1678 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1679 return 0;
1680}
1681
1682
1683static int read_partial(struct ceph_connection *con,
fd51653f 1684 int end, int size, void *object)
31b8006e 1685{
e6cee71f
AE		 1686 while (con->in_base_pos < end) {
1687 int left = end - con->in_base_pos;
31b8006e
SW		 1688 int have = size - left;
1689 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1690 if (ret <= 0)
1691 return ret;
1692 con->in_base_pos += ret;
1693 }
1694 return 1;
1695}
1696
1697
1698/*
1699 * Read all or part of the connect-side handshake on a new connection
1700 */
eed0ef2c 1701static int read_partial_banner(struct ceph_connection *con)
31b8006e 1702{
fd51653f
AE		 1703 int size;
1704 int end;
1705 int ret;
31b8006e 1706
eed0ef2c 1707 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1708
1709 /* peer's banner */
fd51653f
AE		 1710 size = strlen(CEPH_BANNER);
1711 end = size;
1712 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1713 if (ret <= 0)
1714 goto out;
fd51653f
AE		 1715
1716 size = sizeof (con->actual_peer_addr);
1717 end += size;
1718 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1719 if (ret <= 0)
1720 goto out;
fd51653f
AE		 1721
1722 size = sizeof (con->peer_addr_for_me);
1723 end += size;
1724 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1725 if (ret <= 0)
1726 goto out;
fd51653f 1727
eed0ef2c
SW		 1728out:
1729 return ret;
1730}
1731
1732static int read_partial_connect(struct ceph_connection *con)
1733{
fd51653f
AE		 1734 int size;
1735 int end;
1736 int ret;
eed0ef2c
SW		 1737
1738 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1739
fd51653f
AE		 1740 size = sizeof (con->in_reply);
1741 end = size;
1742 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1743 if (ret <= 0)
1744 goto out;
fd51653f
AE		 1745
1746 size = le32_to_cpu(con->in_reply.authorizer_len);
1747 end += size;
1748 ret = read_partial(con, end, size, con->auth_reply_buf);
4e7a5dcd
SW		 1749 if (ret <= 0)
1750 goto out;
31b8006e 1751
4e7a5dcd
SW		 1752 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1753 con, (int)con->in_reply.tag,
1754 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1755 le32_to_cpu(con->in_reply.global_seq));
1756out:
1757 return ret;
eed0ef2c 1758
31b8006e
SW		 1759}
1760
1761/*
1762 * Verify the hello banner looks okay.
1763 */
1764static int verify_hello(struct ceph_connection *con)
1765{
1766 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1767 pr_err("connect to %s got bad banner\n",
3d14c5d2 1768 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1769 con->error_msg = "protocol error, bad banner";
1770 return -1;
1771 }
1772 return 0;
1773}
1774
1775static bool addr_is_blank(struct sockaddr_storage *ss)
1776{
c44bd69c
ID		 1777 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1778 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1779
31b8006e
SW		 1780 switch (ss->ss_family) {
1781 case AF_INET:
c44bd69c 1782 return addr->s_addr == htonl(INADDR_ANY);
31b8006e 1783 case AF_INET6:
c44bd69c
ID		 1784 return ipv6_addr_any(addr6);
1785 default:
1786 return true;
31b8006e 1787 }
31b8006e
SW		 1788}
1789
1790static int addr_port(struct sockaddr_storage *ss)
1791{
1792 switch (ss->ss_family) {
1793 case AF_INET:
f28bcfbe 1794 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1795 case AF_INET6:
f28bcfbe 1796 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1797 }
1798 return 0;
1799}
1800
1801static void addr_set_port(struct sockaddr_storage *ss, int p)
1802{
1803 switch (ss->ss_family) {
1804 case AF_INET:
1805 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1806 break;
31b8006e
SW		 1807 case AF_INET6:
1808 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1809 break;
31b8006e
SW		 1810 }
1811}
1812
ee3b56f2
NW		 1813/*
1814 * Unlike other *_pton function semantics, zero indicates success.
1815 */
1816static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1817 char delim, const char **ipend)
1818{
99f0f3b2
AE		 1819 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1820 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1821
1822 memset(ss, 0, sizeof(*ss));
1823
1824 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1825 ss->ss_family = AF_INET;
1826 return 0;
1827 }
1828
1829 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1830 ss->ss_family = AF_INET6;
1831 return 0;
1832 }
1833
1834 return -EINVAL;
1835}
1836
1837/*
1838 * Extract hostname string and resolve using kernel DNS facility.
1839 */
1840#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1841static int ceph_dns_resolve_name(const char *name, size_t namelen,
1842 struct sockaddr_storage *ss, char delim, const char **ipend)
1843{
1844 const char *end, *delim_p;
1845 char *colon_p, *ip_addr = NULL;
1846 int ip_len, ret;
1847
1848 /*
1849 * The end of the hostname occurs immediately preceding the delimiter or
1850 * the port marker (':') where the delimiter takes precedence.
1851 */
1852 delim_p = memchr(name, delim, namelen);
1853 colon_p = memchr(name, ':', namelen);
1854
1855 if (delim_p && colon_p)
1856 end = delim_p < colon_p ? delim_p : colon_p;
1857 else if (!delim_p && colon_p)
1858 end = colon_p;
1859 else {
1860 end = delim_p;
1861 if (!end) /* case: hostname:/ */
1862 end = name + namelen;
1863 }
1864
1865 if (end <= name)
1866 return -EINVAL;
1867
1868 /* do dns_resolve upcall */
1869 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1870 if (ip_len > 0)
1871 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1872 else
1873 ret = -ESRCH;
1874
1875 kfree(ip_addr);
1876
1877 *ipend = end;
1878
1879 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1880 ret, ret ? "failed" : ceph_pr_addr(ss));
1881
1882 return ret;
1883}
1884#else
1885static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1886 struct sockaddr_storage *ss, char delim, const char **ipend)
1887{
1888 return -EINVAL;
1889}
1890#endif
1891
1892/*
1893 * Parse a server name (IP or hostname). If a valid IP address is not found
1894 * then try to extract a hostname to resolve using userspace DNS upcall.
1895 */
1896static int ceph_parse_server_name(const char *name, size_t namelen,
1897 struct sockaddr_storage *ss, char delim, const char **ipend)
1898{
1899 int ret;
1900
1901 ret = ceph_pton(name, namelen, ss, delim, ipend);
1902 if (ret)
1903 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1904
1905 return ret;
1906}
1907
31b8006e
SW		 1908/*
1909 * Parse an ip[:port] list into an addr array. Use the default
1910 * monitor port if a port isn't specified.
1911 */
1912int ceph_parse_ips(const char *c, const char *end,
1913 struct ceph_entity_addr *addr,
1914 int max_count, int *count)
1915{
ee3b56f2 1916 int i, ret = -EINVAL;
31b8006e
SW		 1917 const char *p = c;
1918
1919 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1920 for (i = 0; i < max_count; i++) {
1921 const char *ipend;
1922 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1923 int port;
39139f64
SW		 1924 char delim = ',';
1925
1926 if (*p == '[') {
1927 delim = ']';
1928 p++;
1929 }
31b8006e 1930
ee3b56f2
NW		 1931 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1932 if (ret)
31b8006e 1933 goto bad;
ee3b56f2
NW		 1934 ret = -EINVAL;
1935
31b8006e
SW		 1936 p = ipend;
1937
39139f64
SW		 1938 if (delim == ']') {
1939 if (*p != ']') {
1940 dout("missing matching ']'\n");
1941 goto bad;
1942 }
1943 p++;
1944 }
1945
31b8006e
SW		 1946 /* port? */
1947 if (p < end && *p == ':') {
1948 port = 0;
1949 p++;
1950 while (p < end && *p >= '0' && *p <= '9') {
1951 port = (port * 10) + (*p - '0');
1952 p++;
1953 }
f48db1e9
ID		 1954 if (port == 0)
1955 port = CEPH_MON_PORT;
1956 else if (port > 65535)
31b8006e
SW		 1957 goto bad;
1958 } else {
1959 port = CEPH_MON_PORT;
1960 }
1961
1962 addr_set_port(ss, port);
1963
3d14c5d2 1964 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 1965
1966 if (p == end)
1967 break;
1968 if (*p != ',')
1969 goto bad;
1970 p++;
1971 }
1972
1973 if (p != end)
1974 goto bad;
1975
1976 if (count)
1977 *count = i + 1;
1978 return 0;
1979
1980bad:
39139f64 1981 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 1982 return ret;
31b8006e 1983}
3d14c5d2 1984EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 1985
eed0ef2c 1986static int process_banner(struct ceph_connection *con)
31b8006e 1987{
eed0ef2c 1988 dout("process_banner on %p\n", con);
31b8006e
SW		 1989
1990 if (verify_hello(con) < 0)
1991 return -1;
1992
63f2d211
SW		 1993 ceph_decode_addr(&con->actual_peer_addr);
1994 ceph_decode_addr(&con->peer_addr_for_me);
1995
31b8006e
SW		 1996 /*
1997 * Make sure the other end is who we wanted. note that the other
1998 * end may not yet know their ip address, so if it's 0.0.0.0, give
1999 * them the benefit of the doubt.
2000 */
103e2d3a
SW		 2001 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2002 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 2003 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2004 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 2005 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2006 ceph_pr_addr(&con->peer_addr.in_addr),
2007 (int)le32_to_cpu(con->peer_addr.nonce),
2008 ceph_pr_addr(&con->actual_peer_addr.in_addr),
2009 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 2010 con->error_msg = "wrong peer at address";
31b8006e
SW		 2011 return -1;
2012 }
2013
2014 /*
2015 * did we learn our address?
2016 */
2017 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2018 int port = addr_port(&con->msgr->inst.addr.in_addr);
2019
2020 memcpy(&con->msgr->inst.addr.in_addr,
2021 &con->peer_addr_for_me.in_addr,
2022 sizeof(con->peer_addr_for_me.in_addr));
2023 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 2024 encode_my_addr(con->msgr);
eed0ef2c 2025 dout("process_banner learned my addr is %s\n",
3d14c5d2 2026 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 2027 }
2028
eed0ef2c
SW		 2029 return 0;
2030}
2031
2032static int process_connect(struct ceph_connection *con)
2033{
859bff51
ID		 2034 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2035 u64 req_feat = from_msgr(con->msgr)->required_features;
2b3e0c90
ID		 2036 u64 server_feat = ceph_sanitize_features(
2037 le64_to_cpu(con->in_reply.features));
0da5d703 2038 int ret;
04a419f9 2039
eed0ef2c
SW		 2040 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2041
5c056fdc
ID		 2042 if (con->auth_reply_buf) {
2043 /*
2044 * Any connection that defines ->get_authorizer()
2045 * should also define ->verify_authorizer_reply().
2046 * See get_connect_authorizer().
2047 */
0dde5848 2048 ret = con->ops->verify_authorizer_reply(con);
5c056fdc
ID		 2049 if (ret < 0) {
2050 con->error_msg = "bad authorize reply";
2051 return ret;
2052 }
2053 }
2054
31b8006e 2055 switch (con->in_reply.tag) {
04a419f9
SW		 2056 case CEPH_MSGR_TAG_FEATURES:
2057 pr_err("%s%lld %s feature set mismatch,"
2058 " my %llx < server's %llx, missing %llx\n",
2059 ENTITY_NAME(con->peer_name),
3d14c5d2 2060 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2061 sup_feat, server_feat, server_feat & ~sup_feat);
2062 con->error_msg = "missing required protocol features";
0fa6ebc6 2063 reset_connection(con);
04a419f9
SW		 2064 return -1;
2065
31b8006e 2066 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2067 pr_err("%s%lld %s protocol version mismatch,"
2068 " my %d != server's %d\n",
2069 ENTITY_NAME(con->peer_name),
3d14c5d2 2070 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2071 le32_to_cpu(con->out_connect.protocol_version),
2072 le32_to_cpu(con->in_reply.protocol_version));
2073 con->error_msg = "protocol version mismatch";
0fa6ebc6 2074 reset_connection(con);
31b8006e
SW		 2075 return -1;
2076
4e7a5dcd
SW		 2077 case CEPH_MSGR_TAG_BADAUTHORIZER:
2078 con->auth_retry++;
2079 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2080 con->auth_retry);
2081 if (con->auth_retry == 2) {
2082 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2083 return -1;
2084 }
6d4221b5 2085 con_out_kvec_reset(con);
e825a66d 2086 ret = prepare_write_connect(con);
0da5d703
SW		 2087 if (ret < 0)
2088 return ret;
63733a0f 2089 prepare_read_connect(con);
4e7a5dcd 2090 break;
31b8006e
SW		 2091
2092 case CEPH_MSGR_TAG_RESETSESSION:
2093 /*
2094 * If we connected with a large connect_seq but the peer
2095 * has no record of a session with us (no connection, or
2096 * connect_seq == 0), they will send RESETSESION to indicate
2097 * that they must have reset their session, and may have
2098 * dropped messages.
2099 */
2100 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2101 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2102 pr_err("%s%lld %s connection reset\n",
2103 ENTITY_NAME(con->peer_name),
3d14c5d2 2104 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2105 reset_connection(con);
6d4221b5 2106 con_out_kvec_reset(con);
5a0f8fdd
AE		 2107 ret = prepare_write_connect(con);
2108 if (ret < 0)
2109 return ret;
31b8006e
SW		 2110 prepare_read_connect(con);
2111
2112 /* Tell ceph about it. */
ec302645 2113 mutex_unlock(&con->mutex);
31b8006e
SW		 2114 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2115 if (con->ops->peer_reset)
2116 con->ops->peer_reset(con);
ec302645 2117 mutex_lock(&con->mutex);
8dacc7da 2118 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2119 return -EAGAIN;
31b8006e
SW		 2120 break;
2121
2122 case CEPH_MSGR_TAG_RETRY_SESSION:
2123 /*
2124 * If we sent a smaller connect_seq than the peer has, try
2125 * again with a larger value.
2126 */
5bdca4e0 2127 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2128 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2129 le32_to_cpu(con->in_reply.connect_seq));
2130 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2131 con_out_kvec_reset(con);
5a0f8fdd
AE		 2132 ret = prepare_write_connect(con);
2133 if (ret < 0)
2134 return ret;
31b8006e
SW		 2135 prepare_read_connect(con);
2136 break;
2137
2138 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2139 /*
2140 * If we sent a smaller global_seq than the peer has, try
2141 * again with a larger value.
2142 */
eed0ef2c 2143 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2144 con->peer_global_seq,
5bdca4e0 2145 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2146 get_global_seq(con->msgr,
5bdca4e0 2147 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2148 con_out_kvec_reset(con);
5a0f8fdd
AE		 2149 ret = prepare_write_connect(con);
2150 if (ret < 0)
2151 return ret;
31b8006e
SW		 2152 prepare_read_connect(con);
2153 break;
2154
3a23083b 2155 case CEPH_MSGR_TAG_SEQ:
31b8006e 2156 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2157 if (req_feat & ~server_feat) {
2158 pr_err("%s%lld %s protocol feature mismatch,"
2159 " my required %llx > server's %llx, need %llx\n",
2160 ENTITY_NAME(con->peer_name),
3d14c5d2 2161 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2162 req_feat, server_feat, req_feat & ~server_feat);
2163 con->error_msg = "missing required protocol features";
0fa6ebc6 2164 reset_connection(con);
04a419f9
SW		 2165 return -1;
2166 }
8dacc7da 2167
122070a2 2168 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2169 con->state = CON_STATE_OPEN;
20e55c4c 2170 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2171 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2172 con->connect_seq++;
aba558e2 2173 con->peer_features = server_feat;
31b8006e
SW		 2174 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2175 con->peer_global_seq,
2176 le32_to_cpu(con->in_reply.connect_seq),
2177 con->connect_seq);
2178 WARN_ON(con->connect_seq !=
2179 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2180
2181 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2182 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2183
85effe18 2184 con->delay = 0; /* reset backoff memory */
92ac41d0 2185
3a23083b
SW		 2186 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2187 prepare_write_seq(con);
2188 prepare_read_seq(con);
2189 } else {
2190 prepare_read_tag(con);
2191 }
31b8006e
SW		 2192 break;
2193
2194 case CEPH_MSGR_TAG_WAIT:
2195 /*
2196 * If there is a connection race (we are opening
2197 * connections to each other), one of us may just have
2198 * to WAIT. This shouldn't happen if we are the
2199 * client.
2200 */
04177882
SW		 2201 con->error_msg = "protocol error, got WAIT as client";
2202 return -1;
31b8006e
SW		 2203
2204 default:
31b8006e
SW		 2205 con->error_msg = "protocol error, garbage tag during connect";
2206 return -1;
2207 }
2208 return 0;
2209}
2210
2211
2212/*
2213 * read (part of) an ack
2214 */
2215static int read_partial_ack(struct ceph_connection *con)
2216{
fd51653f
AE		 2217 int size = sizeof (con->in_temp_ack);
2218 int end = size;
31b8006e 2219
fd51653f 2220 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2221}
2222
31b8006e
SW		 2223/*
2224 * We can finally discard anything that's been acked.
2225 */
2226static void process_ack(struct ceph_connection *con)
2227{
2228 struct ceph_msg *m;
2229 u64 ack = le64_to_cpu(con->in_temp_ack);
2230 u64 seq;
0a2ad541
YZ		 2231 bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2232 struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
31b8006e 2233
0a2ad541
YZ		 2234 /*
2235 * In the reconnect case, con_fault() has requeued messages
2236 * in out_sent. We should cleanup old messages according to
2237 * the reconnect seq.
2238 */
2239 while (!list_empty(list)) {
2240 m = list_first_entry(list, struct ceph_msg, list_head);
2241 if (reconnect && m->needs_out_seq)
2242 break;
31b8006e
SW		 2243 seq = le64_to_cpu(m->hdr.seq);
2244 if (seq > ack)
2245 break;
2246 dout("got ack for seq %llu type %d at %p\n", seq,
2247 le16_to_cpu(m->hdr.type), m);
4cf9d544 2248 m->ack_stamp = jiffies;
31b8006e
SW		 2249 ceph_msg_remove(m);
2250 }
0a2ad541 2251
31b8006e
SW		 2252 prepare_read_tag(con);
2253}
2254
2255
2450418c 2256static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2257 struct kvec *section,
2258 unsigned int sec_len, u32 *crc)
2450418c 2259{
68b4476b 2260 int ret, left;
2450418c
YS		 2261
2262 BUG_ON(!section);
2263
2264 while (section->iov_len < sec_len) {
2265 BUG_ON(section->iov_base == NULL);
2266 left = sec_len - section->iov_len;
2267 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2268 section->iov_len, left);
2269 if (ret <= 0)
2270 return ret;
2271 section->iov_len += ret;
2450418c 2272 }
fe3ad593
AE		 2273 if (section->iov_len == sec_len)
2274 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2275
2450418c
YS		 2276 return 1;
2277}
31b8006e 2278
34d2d200
AE		 2279static int read_partial_msg_data(struct ceph_connection *con)
2280{
2281 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2282 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 2283 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
686be208
AE		 2284 struct page *page;
2285 size_t page_offset;
2286 size_t length;
f5db90bc 2287 u32 crc = 0;
34d2d200
AE		 2288 int ret;
2289
2290 BUG_ON(!msg);
5240d9f9 2291 if (list_empty(&msg->data))
4c59b4a2 2292 return -EIO;
34d2d200 2293
f5db90bc
AE		 2294 if (do_datacrc)
2295 crc = con->in_data_crc;
643c68a4 2296 while (cursor->resid) {
343128ce 2297 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
686be208 2298 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2299 if (ret <= 0) {
2300 if (do_datacrc)
2301 con->in_data_crc = crc;
2302
686be208 2303 return ret;
f5db90bc 2304 }
686be208
AE		 2305
2306 if (do_datacrc)
f5db90bc 2307 crc = ceph_crc32c_page(crc, page, page_offset, ret);
1759f7b0 2308 ceph_msg_data_advance(cursor, (size_t)ret);
34d2d200 2309 }
f5db90bc
AE		 2310 if (do_datacrc)
2311 con->in_data_crc = crc;
34d2d200
AE		 2312
2313 return 1; /* must return > 0 to indicate success */
2314}
2315
31b8006e
SW		 2316/*
2317 * read (part of) a message.
2318 */
686be208
AE		 2319static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2320
31b8006e
SW		 2321static int read_partial_message(struct ceph_connection *con)
2322{
2323 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2324 int size;
2325 int end;
31b8006e 2326 int ret;
95c96174 2327 unsigned int front_len, middle_len, data_len;
859bff51 2328 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
33d07337 2329 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
ae18756b 2330 u64 seq;
fe3ad593 2331 u32 crc;
31b8006e
SW		 2332
2333 dout("read_partial_message con %p msg %p\n", con, m);
2334
2335 /* header */
fd51653f
AE		 2336 size = sizeof (con->in_hdr);
2337 end = size;
2338 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2339 if (ret <= 0)
2340 return ret;
fe3ad593
AE		 2341
2342 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2343 if (cpu_to_le32(crc) != con->in_hdr.crc) {
67c64eb7 2344 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
fe3ad593
AE		 2345 crc, con->in_hdr.crc);
2346 return -EBADMSG;
2347 }
2348
31b8006e
SW		 2349 front_len = le32_to_cpu(con->in_hdr.front_len);
2350 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2351 return -EIO;
2352 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2353 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2354 return -EIO;
2355 data_len = le32_to_cpu(con->in_hdr.data_len);
2356 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2357 return -EIO;
2358
ae18756b
SW		 2359 /* verify seq# */
2360 seq = le64_to_cpu(con->in_hdr.seq);
2361 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2362 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2363 ENTITY_NAME(con->peer_name),
3d14c5d2 2364 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2365 seq, con->in_seq + 1);
2366 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2367 sizeof_footer(con);
ae18756b 2368 con->in_tag = CEPH_MSGR_TAG_READY;
e7a88e82 2369 return 1;
ae18756b
SW		 2370 } else if ((s64)seq - (s64)con->in_seq > 1) {
2371 pr_err("read_partial_message bad seq %lld expected %lld\n",
2372 seq, con->in_seq + 1);
2373 con->error_msg = "bad message sequence # for incoming message";
67c64eb7 2374 return -EBADE;
ae18756b
SW		 2375 }
2376
31b8006e
SW		 2377 /* allocate message? */
2378 if (!con->in_msg) {
4740a623
SW		 2379 int skip = 0;
2380
31b8006e 2381 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2382 front_len, data_len);
4740a623
SW		 2383 ret = ceph_con_in_msg_alloc(con, &skip);
2384 if (ret < 0)
2385 return ret;
f759ebb9
AE		 2386
2387 BUG_ON(!con->in_msg ^ skip);
2450418c 2388 if (skip) {
31b8006e 2389 /* skip this message */
a79832f2 2390 dout("alloc_msg said skip message\n");
31b8006e 2391 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2392 sizeof_footer(con);
31b8006e 2393 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2394 con->in_seq++;
e7a88e82 2395 return 1;
31b8006e 2396 }
38941f80 2397
4740a623 2398 BUG_ON(!con->in_msg);
38941f80 2399 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2400 m = con->in_msg;
2401 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2402 if (m->middle)
2403 m->middle->vec.iov_len = 0;
9d7f0f13 2404
78625051 2405 /* prepare for data payload, if any */
a4107026 2406
78625051 2407 if (data_len)
98fa5dd8 2408 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2409 }
2410
2411 /* front */
2450418c
YS		 2412 ret = read_partial_message_section(con, &m->front, front_len,
2413 &con->in_front_crc);
2414 if (ret <= 0)
2415 return ret;
31b8006e
SW		 2416
2417 /* middle */
2450418c 2418 if (m->middle) {
213c99ee
SW		 2419 ret = read_partial_message_section(con, &m->middle->vec,
2420 middle_len,
2450418c 2421 &con->in_middle_crc);
31b8006e
SW		 2422 if (ret <= 0)
2423 return ret;
31b8006e
SW		 2424 }
2425
2426 /* (page) data */
34d2d200
AE		 2427 if (data_len) {
2428 ret = read_partial_msg_data(con);
2429 if (ret <= 0)
2430 return ret;
31b8006e
SW		 2431 }
2432
31b8006e 2433 /* footer */
89f08173 2434 size = sizeof_footer(con);
fd51653f
AE		 2435 end += size;
2436 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2437 if (ret <= 0)
2438 return ret;
2439
33d07337
YZ		 2440 if (!need_sign) {
2441 m->footer.flags = m->old_footer.flags;
2442 m->footer.sig = 0;
2443 }
2444
31b8006e
SW		 2445 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2446 m, front_len, m->footer.front_crc, middle_len,
2447 m->footer.middle_crc, data_len, m->footer.data_crc);
2448
2449 /* crc ok? */
2450 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2451 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2452 m, con->in_front_crc, m->footer.front_crc);
2453 return -EBADMSG;
2454 }
2455 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2456 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2457 m, con->in_middle_crc, m->footer.middle_crc);
2458 return -EBADMSG;
2459 }
bca064d2 2460 if (do_datacrc &&
31b8006e
SW		 2461 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2462 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2463 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2464 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2465 return -EBADMSG;
2466 }
2467
33d07337 2468 if (need_sign && con->ops->check_message_signature &&
79dbd1ba 2469 con->ops->check_message_signature(m)) {
33d07337
YZ		 2470 pr_err("read_partial_message %p signature check failed\n", m);
2471 return -EBADMSG;
2472 }
2473
31b8006e
SW		 2474 return 1; /* done! */
2475}
2476
2477/*
2478 * Process message. This happens in the worker thread. The callback should
2479 * be careful not to do anything that waits on other incoming messages or it
2480 * may deadlock.
2481 */
2482static void process_message(struct ceph_connection *con)
2483{
583d0fef 2484 struct ceph_msg *msg = con->in_msg;
31b8006e 2485
38941f80 2486 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2487 con->in_msg = NULL;
2488
2489 /* if first message, set peer_name */
2490 if (con->peer_name.type == 0)
dbad185d 2491 con->peer_name = msg->hdr.src;
31b8006e 2492
31b8006e 2493 con->in_seq++;
ec302645 2494 mutex_unlock(&con->mutex);
31b8006e
SW		 2495
2496 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2497 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2498 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2499 le16_to_cpu(msg->hdr.type),
2500 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2501 le32_to_cpu(msg->hdr.front_len),
2502 le32_to_cpu(msg->hdr.data_len),
2503 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2504 con->ops->dispatch(con, msg);
ec302645
SW		 2505
2506 mutex_lock(&con->mutex);
31b8006e
SW		 2507}
2508
8b9558aa
YZ		 2509static int read_keepalive_ack(struct ceph_connection *con)
2510{
2511 struct ceph_timespec ceph_ts;
2512 size_t size = sizeof(ceph_ts);
2513 int ret = read_partial(con, size, size, &ceph_ts);
2514 if (ret <= 0)
2515 return ret;
2516 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2517 prepare_read_tag(con);
2518 return 1;
2519}
31b8006e
SW		 2520
2521/*
2522 * Write something to the socket. Called in a worker thread when the
2523 * socket appears to be writeable and we have something ready to send.
2524 */
2525static int try_write(struct ceph_connection *con)
2526{
31b8006e
SW		 2527 int ret = 1;
2528
d59315ca 2529 dout("try_write start %p state %lu\n", con, con->state);
31b8006e 2530
31b8006e
SW		 2531more:
2532 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2533
2534 /* open the socket first? */
8dacc7da
SW		 2535 if (con->state == CON_STATE_PREOPEN) {
2536 BUG_ON(con->sock);
2537 con->state = CON_STATE_CONNECTING;
a5988c49 2538
e2200423 2539 con_out_kvec_reset(con);
e825a66d 2540 prepare_write_banner(con);
eed0ef2c 2541 prepare_read_banner(con);
31b8006e 2542
cf3e5c40 2543 BUG_ON(con->in_msg);
31b8006e
SW		 2544 con->in_tag = CEPH_MSGR_TAG_READY;
2545 dout("try_write initiating connect on %p new state %lu\n",
2546 con, con->state);
41617d0c
AE		 2547 ret = ceph_tcp_connect(con);
2548 if (ret < 0) {
31b8006e 2549 con->error_msg = "connect error";
31b8006e
SW		 2550 goto out;
2551 }
2552 }
2553
2554more_kvec:
2555 /* kvec data queued? */
67645d76
ID		 2556 if (con->out_kvec_left) {
2557 ret = write_partial_kvec(con);
31b8006e 2558 if (ret <= 0)
42961d23 2559 goto out;
31b8006e 2560 }
67645d76
ID		 2561 if (con->out_skip) {
2562 ret = write_partial_skip(con);
31b8006e 2563 if (ret <= 0)
42961d23 2564 goto out;
31b8006e
SW		 2565 }
2566
2567 /* msg pages? */
2568 if (con->out_msg) {
c86a2930
SW		 2569 if (con->out_msg_done) {
2570 ceph_msg_put(con->out_msg);
2571 con->out_msg = NULL; /* we're done with this one */
2572 goto do_next;
2573 }
2574
34d2d200 2575 ret = write_partial_message_data(con);
31b8006e
SW		 2576 if (ret == 1)
2577 goto more_kvec; /* we need to send the footer, too! */
2578 if (ret == 0)
42961d23 2579 goto out;
31b8006e 2580 if (ret < 0) {
34d2d200 2581 dout("try_write write_partial_message_data err %d\n",
31b8006e 2582 ret);
42961d23 2583 goto out;
31b8006e
SW		 2584 }
2585 }
2586
c86a2930 2587do_next:
8dacc7da 2588 if (con->state == CON_STATE_OPEN) {
8b9558aa
YZ		 2589 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2590 prepare_write_keepalive(con);
2591 goto more;
2592 }
31b8006e
SW		 2593 /* is anything else pending? */
2594 if (!list_empty(&con->out_queue)) {
2595 prepare_write_message(con);
2596 goto more;
2597 }
2598 if (con->in_seq > con->in_seq_acked) {
2599 prepare_write_ack(con);
2600 goto more;
2601 }
31b8006e
SW		 2602 }
2603
2604 /* Nothing to do! */
c9ffc77a 2605 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2606 dout("try_write nothing else to write.\n");
31b8006e
SW		 2607 ret = 0;
2608out:
42961d23 2609 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2610 return ret;
2611}
2612
2613
2614
2615/*
2616 * Read what we can from the socket.
2617 */
2618static int try_read(struct ceph_connection *con)
2619{
31b8006e
SW		 2620 int ret = -1;
2621
8dacc7da
SW		 2622more:
2623 dout("try_read start on %p state %lu\n", con, con->state);
2624 if (con->state != CON_STATE_CONNECTING &&
2625 con->state != CON_STATE_NEGOTIATING &&
2626 con->state != CON_STATE_OPEN)
31b8006e
SW		 2627 return 0;
2628
8dacc7da 2629 BUG_ON(!con->sock);
ec302645 2630
31b8006e
SW		 2631 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2632 con->in_base_pos);
0da5d703 2633
8dacc7da 2634 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2635 dout("try_read connecting\n");
2636 ret = read_partial_banner(con);
2637 if (ret <= 0)
ab166d5a 2638 goto out;
7593af92
AE		 2639 ret = process_banner(con);
2640 if (ret < 0)
2641 goto out;
2642
8dacc7da 2643 con->state = CON_STATE_NEGOTIATING;
7593af92 2644
6d4221b5
JS		 2645 /*
2646 * Received banner is good, exchange connection info.
2647 * Do not reset out_kvec, as sending our banner raced
2648 * with receiving peer banner after connect completed.
2649 */
7593af92
AE		 2650 ret = prepare_write_connect(con);
2651 if (ret < 0)
2652 goto out;
2653 prepare_read_connect(con);
2654
2655 /* Send connection info before awaiting response */
0da5d703
SW		 2656 goto out;
2657 }
2658
8dacc7da 2659 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2660 dout("try_read negotiating\n");
31b8006e
SW		 2661 ret = read_partial_connect(con);
2662 if (ret <= 0)
31b8006e 2663 goto out;
98bdb0aa
SW		 2664 ret = process_connect(con);
2665 if (ret < 0)
2666 goto out;
31b8006e
SW		 2667 goto more;
2668 }
2669
122070a2 2670 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2671
31b8006e
SW		 2672 if (con->in_base_pos < 0) {
2673 /*
2674 * skipping + discarding content.
2675 *
2676 * FIXME: there must be a better way to do this!
2677 */
84495f49
AE		 2678 static char buf[SKIP_BUF_SIZE];
2679 int skip = min((int) sizeof (buf), -con->in_base_pos);
2680
31b8006e
SW		 2681 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2682 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2683 if (ret <= 0)
98bdb0aa 2684 goto out;
31b8006e
SW		 2685 con->in_base_pos += ret;
2686 if (con->in_base_pos)
2687 goto more;
2688 }
2689 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2690 /*
2691 * what's next?
2692 */
2693 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2694 if (ret <= 0)
98bdb0aa 2695 goto out;
31b8006e
SW		 2696 dout("try_read got tag %d\n", (int)con->in_tag);
2697 switch (con->in_tag) {
2698 case CEPH_MSGR_TAG_MSG:
2699 prepare_read_message(con);
2700 break;
2701 case CEPH_MSGR_TAG_ACK:
2702 prepare_read_ack(con);
2703 break;
8b9558aa
YZ		 2704 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2705 prepare_read_keepalive_ack(con);
2706 break;
31b8006e 2707 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2708 con_close_socket(con);
2709 con->state = CON_STATE_CLOSED;
98bdb0aa 2710 goto out;
31b8006e
SW		 2711 default:
2712 goto bad_tag;
2713 }
2714 }
2715 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2716 ret = read_partial_message(con);
2717 if (ret <= 0) {
2718 switch (ret) {
2719 case -EBADMSG:
a51983e4 2720 con->error_msg = "bad crc/signature";
67c64eb7
ID		 2721 /* fall through */
2722 case -EBADE:
31b8006e 2723 ret = -EIO;
98bdb0aa 2724 break;
31b8006e
SW		 2725 case -EIO:
2726 con->error_msg = "io error";
98bdb0aa 2727 break;
31b8006e 2728 }
98bdb0aa 2729 goto out;
31b8006e
SW		 2730 }
2731 if (con->in_tag == CEPH_MSGR_TAG_READY)
2732 goto more;
2733 process_message(con);
7b862e07
SW		 2734 if (con->state == CON_STATE_OPEN)
2735 prepare_read_tag(con);
31b8006e
SW		 2736 goto more;
2737 }
3a23083b
SW		 2738 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2739 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2740 /*
2741 * the final handshake seq exchange is semantically
2742 * equivalent to an ACK
2743 */
31b8006e
SW		 2744 ret = read_partial_ack(con);
2745 if (ret <= 0)
98bdb0aa 2746 goto out;
31b8006e
SW		 2747 process_ack(con);
2748 goto more;
2749 }
8b9558aa
YZ		 2750 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2751 ret = read_keepalive_ack(con);
2752 if (ret <= 0)
2753 goto out;
2754 goto more;
2755 }
31b8006e 2756
31b8006e 2757out:
98bdb0aa 2758 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2759 return ret;
2760
2761bad_tag:
2762 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2763 con->error_msg = "protocol error, garbage tag";
2764 ret = -1;
2765 goto out;
2766}
2767
2768
2769/*
802c6d96
AE		 2770 * Atomically queue work on a connection after the specified delay.
2771 * Bump @con reference to avoid races with connection teardown.
2772 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2773 */
802c6d96 2774static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2775{
31b8006e 2776 if (!con->ops->get(con)) {
802c6d96 2777 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2778 return -ENOENT;
31b8006e
SW		 2779 }
2780
802c6d96
AE		 2781 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2782 dout("%s %p - already queued\n", __func__, con);
31b8006e 2783 con->ops->put(con);
802c6d96 2784 return -EBUSY;
31b8006e 2785 }
802c6d96
AE		 2786
2787 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2788 return 0;
2789}
2790
2791static void queue_con(struct ceph_connection *con)
2792{
2793 (void) queue_con_delay(con, 0);
31b8006e
SW		 2794}
2795
37ab77ac
ID		 2796static void cancel_con(struct ceph_connection *con)
2797{
2798 if (cancel_delayed_work(&con->work)) {
2799 dout("%s %p\n", __func__, con);
2800 con->ops->put(con);
2801 }
2802}
2803
7bb21d68
AE		 2804static bool con_sock_closed(struct ceph_connection *con)
2805{
c9ffc77a 2806 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2807 return false;
2808
2809#define CASE(x) \
2810 case CON_STATE_ ## x: \
2811 con->error_msg = "socket closed (con state " #x ")"; \
2812 break;
2813
2814 switch (con->state) {
2815 CASE(CLOSED);
2816 CASE(PREOPEN);
2817 CASE(CONNECTING);
2818 CASE(NEGOTIATING);
2819 CASE(OPEN);
2820 CASE(STANDBY);
2821 default:
b9a67899 2822 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2823 __func__, con, con->state);
2824 con->error_msg = "unrecognized con state";
2825 BUG();
2826 break;
2827 }
2828#undef CASE
2829
2830 return true;
2831}
2832
f20a39fd
AE		 2833static bool con_backoff(struct ceph_connection *con)
2834{
2835 int ret;
2836
2837 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2838 return false;
2839
2840 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2841 if (ret) {
2842 dout("%s: con %p FAILED to back off %lu\n", __func__,
2843 con, con->delay);
2844 BUG_ON(ret == -ENOENT);
2845 con_flag_set(con, CON_FLAG_BACKOFF);
2846 }
2847
2848 return true;
2849}
2850
93209264
AE		 2851/* Finish fault handling; con->mutex must *not* be held here */
2852
2853static void con_fault_finish(struct ceph_connection *con)
2854{
f6330cc1
ID		 2855 dout("%s %p\n", __func__, con);
2856
93209264
AE		 2857 /*
2858 * in case we faulted due to authentication, invalidate our
2859 * current tickets so that we can get new ones.
2860 */
f6330cc1
ID		 2861 if (con->auth_retry) {
2862 dout("auth_retry %d, invalidating\n", con->auth_retry);
2863 if (con->ops->invalidate_authorizer)
2864 con->ops->invalidate_authorizer(con);
2865 con->auth_retry = 0;
93209264
AE		 2866 }
2867
2868 if (con->ops->fault)
2869 con->ops->fault(con);
2870}
2871
31b8006e
SW		 2872/*
2873 * Do some work on a connection. Drop a connection ref when we're done.
2874 */
68931622 2875static void ceph_con_workfn(struct work_struct *work)
31b8006e
SW		 2876{
2877 struct ceph_connection *con = container_of(work, struct ceph_connection,
2878 work.work);
49659416 2879 bool fault;
31b8006e 2880
9dd4658d 2881 mutex_lock(&con->mutex);
49659416
AE		 2882 while (true) {
2883 int ret;
31b8006e 2884
49659416
AE		 2885 if ((fault = con_sock_closed(con))) {
2886 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2887 break;
2888 }
2889 if (con_backoff(con)) {
2890 dout("%s: con %p BACKOFF\n", __func__, con);
2891 break;
2892 }
2893 if (con->state == CON_STATE_STANDBY) {
2894 dout("%s: con %p STANDBY\n", __func__, con);
2895 break;
2896 }
2897 if (con->state == CON_STATE_CLOSED) {
2898 dout("%s: con %p CLOSED\n", __func__, con);
2899 BUG_ON(con->sock);
2900 break;
2901 }
2902 if (con->state == CON_STATE_PREOPEN) {
2903 dout("%s: con %p PREOPEN\n", __func__, con);
2904 BUG_ON(con->sock);
2905 }
0da5d703 2906
49659416
AE		 2907 ret = try_read(con);
2908 if (ret < 0) {
2909 if (ret == -EAGAIN)
2910 continue;
67c64eb7
ID		 2911 if (!con->error_msg)
2912 con->error_msg = "socket error on read";
49659416
AE		 2913 fault = true;
2914 break;
2915 }
2916
2917 ret = try_write(con);
2918 if (ret < 0) {
2919 if (ret == -EAGAIN)
2920 continue;
67c64eb7
ID		 2921 if (!con->error_msg)
2922 con->error_msg = "socket error on write";
49659416
AE		 2923 fault = true;
2924 }
2925
2926 break; /* If we make it to here, we're done */
3a140a0d 2927 }
b6e7b6a1
AE		 2928 if (fault)
2929 con_fault(con);
9dd4658d 2930 mutex_unlock(&con->mutex);
0da5d703 2931
b6e7b6a1
AE		 2932 if (fault)
2933 con_fault_finish(con);
2934
2935 con->ops->put(con);
31b8006e
SW		 2936}
2937
31b8006e
SW		 2938/*
2939 * Generic error/fault handler. A retry mechanism is used with
2940 * exponential backoff
2941 */
93209264 2942static void con_fault(struct ceph_connection *con)
31b8006e 2943{
31b8006e 2944 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2945 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2946
67c64eb7
ID		 2947 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2948 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2949 con->error_msg = NULL;
2950
122070a2 2951 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2952 con->state != CON_STATE_NEGOTIATING &&
2953 con->state != CON_STATE_OPEN);
ec302645 2954
31b8006e 2955 con_close_socket(con);
5e095e8b 2956
c9ffc77a 2957 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2958 dout("fault on LOSSYTX channel, marking CLOSED\n");
2959 con->state = CON_STATE_CLOSED;
93209264 2960 return;
3b5ede07
SW		 2961 }
2962
5e095e8b 2963 if (con->in_msg) {
38941f80 2964 BUG_ON(con->in_msg->con != con);
5e095e8b
SW		 2965 ceph_msg_put(con->in_msg);
2966 con->in_msg = NULL;
2967 }
31b8006e 2968
e80a52d1
SW		 2969 /* Requeue anything that hasn't been acked */
2970 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 2971
e76661d0
SW		 2972 /* If there are no messages queued or keepalive pending, place
2973 * the connection in a STANDBY state */
2974 if (list_empty(&con->out_queue) &&
c9ffc77a 2975 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 2976 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 2977 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 2978 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 2979 } else {
2980 /* retry after a delay. */
8dacc7da 2981 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 2982 if (con->delay == 0)
2983 con->delay = BASE_DELAY_INTERVAL;
2984 else if (con->delay < MAX_DELAY_INTERVAL)
2985 con->delay *= 2;
c9ffc77a 2986 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 2987 queue_con(con);
31b8006e 2988 }
31b8006e
SW		 2989}
2990
2991
2992
2993/*
15d9882c 2994 * initialize a new messenger instance
31b8006e 2995 */
15d9882c 2996void ceph_messenger_init(struct ceph_messenger *msgr,
859bff51 2997 struct ceph_entity_addr *myaddr)
31b8006e 2998{
31b8006e
SW		 2999 spin_lock_init(&msgr->global_seq_lock);
3000
31b8006e
SW		 3001 if (myaddr)
3002 msgr->inst.addr = *myaddr;
3003
3004 /* select a random nonce */
ac8839d7 3005 msgr->inst.addr.type = 0;
103e2d3a 3006 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 3007 encode_my_addr(msgr);
31b8006e 3008
a2a32584 3009 atomic_set(&msgr->stopping, 0);
757856d2 3010 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
31b8006e 3011
15d9882c 3012 dout("%s %p\n", __func__, msgr);
31b8006e 3013}
15d9882c 3014EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 3015
757856d2
ID		 3016void ceph_messenger_fini(struct ceph_messenger *msgr)
3017{
3018 put_net(read_pnet(&msgr->net));
3019}
3020EXPORT_SYMBOL(ceph_messenger_fini);
3021
583d0fef
ID		 3022static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3023{
3024 if (msg->con)
3025 msg->con->ops->put(msg->con);
3026
3027 msg->con = con ? con->ops->get(con) : NULL;
3028 BUG_ON(msg->con != con);
3029}
3030
e00de341
SW		 3031static void clear_standby(struct ceph_connection *con)
3032{
3033 /* come back from STANDBY? */
8dacc7da 3034 if (con->state == CON_STATE_STANDBY) {
e00de341 3035 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 3036 con->state = CON_STATE_PREOPEN;
e00de341 3037 con->connect_seq++;
c9ffc77a
AE		 3038 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3039 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 3040 }
3041}
3042
31b8006e
SW		 3043/*
3044 * Queue up an outgoing message on the given connection.
3045 */
3046void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3047{
31b8006e 3048 /* set src+dst */
dbad185d 3049 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 3050 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 3051 msg->needs_out_seq = true;
3052
ec302645 3053 mutex_lock(&con->mutex);
92ce034b 3054
8dacc7da 3055 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 3056 dout("con_send %p closed, dropping %p\n", con, msg);
3057 ceph_msg_put(msg);
3058 mutex_unlock(&con->mutex);
3059 return;
3060 }
3061
583d0fef 3062 msg_con_set(msg, con);
92ce034b 3063
31b8006e
SW		 3064 BUG_ON(!list_empty(&msg->list_head));
3065 list_add_tail(&msg->list_head, &con->out_queue);
3066 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3067 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3068 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3069 le32_to_cpu(msg->hdr.front_len),
3070 le32_to_cpu(msg->hdr.middle_len),
3071 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 3072
3073 clear_standby(con);
ec302645 3074 mutex_unlock(&con->mutex);
31b8006e
SW		 3075
3076 /* if there wasn't anything waiting to send before, queue
3077 * new work */
c9ffc77a 3078 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3079 queue_con(con);
3080}
3d14c5d2 3081EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3082
3083/*
3084 * Revoke a message that was previously queued for send
3085 */
6740a845 3086void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3087{
6740a845
AE		 3088 struct ceph_connection *con = msg->con;
3089
583d0fef
ID		 3090 if (!con) {
3091 dout("%s msg %p null con\n", __func__, msg);
6740a845 3092 return; /* Message not in our possession */
583d0fef 3093 }
6740a845 3094
ec302645 3095 mutex_lock(&con->mutex);
31b8006e 3096 if (!list_empty(&msg->list_head)) {
38941f80 3097 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3098 list_del_init(&msg->list_head);
31b8006e 3099 msg->hdr.seq = 0;
38941f80 3100
31b8006e 3101 ceph_msg_put(msg);
ed98adad
SW		 3102 }
3103 if (con->out_msg == msg) {
67645d76
ID		 3104 BUG_ON(con->out_skip);
3105 /* footer */
3106 if (con->out_msg_done) {
3107 con->out_skip += con_out_kvec_skip(con);
3108 } else {
3109 BUG_ON(!msg->data_length);
89f08173 3110 con->out_skip += sizeof_footer(con);
31b8006e 3111 }
67645d76
ID		 3112 /* data, middle, front */
3113 if (msg->data_length)
3114 con->out_skip += msg->cursor.total_resid;
3115 if (msg->middle)
3116 con->out_skip += con_out_kvec_skip(con);
3117 con->out_skip += con_out_kvec_skip(con);
3118
3119 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3120 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
ed98adad 3121 msg->hdr.seq = 0;
67645d76 3122 con->out_msg = NULL;
92ce034b 3123 ceph_msg_put(msg);
31b8006e 3124 }
67645d76 3125
ec302645 3126 mutex_unlock(&con->mutex);
31b8006e
SW		 3127}
3128
350b1c32 3129/*
0d59ab81 3130 * Revoke a message that we may be reading data into
350b1c32 3131 */
8921d114 3132void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3133{
583d0fef 3134 struct ceph_connection *con = msg->con;
8921d114 3135
583d0fef 3136 if (!con) {
8921d114 3137 dout("%s msg %p null con\n", __func__, msg);
8921d114
AE		 3138 return; /* Message not in our possession */
3139 }
3140
350b1c32 3141 mutex_lock(&con->mutex);
8921d114 3142 if (con->in_msg == msg) {
95c96174
ED		 3143 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3144 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3145 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3146
3147 /* skip rest of message */
8921d114
AE		 3148 dout("%s %p msg %p revoked\n", __func__, con, msg);
3149 con->in_base_pos = con->in_base_pos -
350b1c32 3150 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3151 front_len -
3152 middle_len -
3153 data_len -
350b1c32 3154 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3155 ceph_msg_put(con->in_msg);
3156 con->in_msg = NULL;
3157 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3158 con->in_seq++;
350b1c32 3159 } else {
8921d114
AE		 3160 dout("%s %p in_msg %p msg %p no-op\n",
3161 __func__, con, con->in_msg, msg);
350b1c32
SW		 3162 }
3163 mutex_unlock(&con->mutex);
3164}
3165
31b8006e
SW		 3166/*
3167 * Queue a keepalive byte to ensure the tcp connection is alive.
3168 */
3169void ceph_con_keepalive(struct ceph_connection *con)
3170{
e00de341 3171 dout("con_keepalive %p\n", con);
00650931 3172 mutex_lock(&con->mutex);
e00de341 3173 clear_standby(con);
00650931 3174 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3175 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3176 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3177 queue_con(con);
3178}
3d14c5d2 3179EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3180
8b9558aa
YZ		 3181bool ceph_con_keepalive_expired(struct ceph_connection *con,
3182 unsigned long interval)
3183{
3184 if (interval > 0 &&
3185 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
1134e091 3186 struct timespec now;
8b9558aa 3187 struct timespec ts;
1134e091 3188 ktime_get_real_ts(&now);
8b9558aa
YZ		 3189 jiffies_to_timespec(interval, &ts);
3190 ts = timespec_add(con->last_keepalive_ack, ts);
3191 return timespec_compare(&now, &ts) >= 0;
3192 }
3193 return false;
3194}
3195
6644ed7b 3196static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3197{
6644ed7b
AE		 3198 struct ceph_msg_data *data;
3199
3200 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3201 return NULL;
3202
81b36be4 3203 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
6644ed7b
AE		 3204 if (data)
3205 data->type = type;
5240d9f9 3206 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3207
3208 return data;
3209}
3210
3211static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3212{
3213 if (!data)
3214 return;
3215
5240d9f9 3216 WARN_ON(!list_empty(&data->links));
e4339d28 3217 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3218 ceph_pagelist_release(data->pagelist);
81b36be4 3219 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3220}
3221
90af3602 3222void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3223 size_t length, size_t alignment)
02afca6c 3224{
6644ed7b
AE		 3225 struct ceph_msg_data *data;
3226
07aa1558
AE		 3227 BUG_ON(!pages);
3228 BUG_ON(!length);
6644ed7b
AE		 3229
3230 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3231 BUG_ON(!data);
3232 data->pages = pages;
3233 data->length = length;
3234 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3235
5240d9f9
AE		 3236 list_add_tail(&data->links, &msg->data);
3237 msg->data_length += length;
02afca6c 3238}
90af3602 3239EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3240
90af3602 3241void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3242 struct ceph_pagelist *pagelist)
3243{
6644ed7b
AE		 3244 struct ceph_msg_data *data;
3245
07aa1558
AE		 3246 BUG_ON(!pagelist);
3247 BUG_ON(!pagelist->length);
27fa8385 3248
6644ed7b
AE		 3249 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3250 BUG_ON(!data);
3251 data->pagelist = pagelist;
3252
5240d9f9
AE		 3253 list_add_tail(&data->links, &msg->data);
3254 msg->data_length += pagelist->length;
27fa8385 3255}
90af3602 3256EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3257
ea96571f 3258#ifdef CONFIG_BLOCK
90af3602 3259void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
a1930804 3260 size_t length)
27fa8385 3261{
6644ed7b
AE		 3262 struct ceph_msg_data *data;
3263
07aa1558 3264 BUG_ON(!bio);
27fa8385 3265
6644ed7b
AE		 3266 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3267 BUG_ON(!data);
3268 data->bio = bio;
c851c495 3269 data->bio_length = length;
6644ed7b 3270
5240d9f9
AE		 3271 list_add_tail(&data->links, &msg->data);
3272 msg->data_length += length;
27fa8385 3273}
90af3602 3274EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3275#endif /* CONFIG_BLOCK */
27fa8385 3276
31b8006e
SW		 3277/*
3278 * construct a new message with given type, size
3279 * the new msg has a ref count of 1.
3280 */
b61c2763
SW		 3281struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3282 bool can_fail)
31b8006e
SW		 3283{
3284 struct ceph_msg *m;
3285
e3d5d638 3286 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3287 if (m == NULL)
3288 goto out;
31b8006e
SW		 3289
3290 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3291 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3292 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3293
9516e45b
AE		 3294 INIT_LIST_HEAD(&m->list_head);
3295 kref_init(&m->kref);
5240d9f9 3296 INIT_LIST_HEAD(&m->data);
ca20892d 3297
31b8006e
SW		 3298 /* front */
3299 if (front_len) {
eeb0bed5 3300 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3301 if (m->front.iov_base == NULL) {
b61c2763 3302 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3303 front_len);
3304 goto out2;
3305 }
3306 } else {
3307 m->front.iov_base = NULL;
3308 }
f2be82b0 3309 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3310
bb257664 3311 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3312 return m;
3313
3314out2:
3315 ceph_msg_put(m);
3316out:
b61c2763
SW		 3317 if (!can_fail) {
3318 pr_err("msg_new can't create type %d front %d\n", type,
3319 front_len);
f0ed1b7c 3320 WARN_ON(1);
b61c2763
SW		 3321 } else {
3322 dout("msg_new can't create type %d front %d\n", type,
3323 front_len);
3324 }
a79832f2 3325 return NULL;
31b8006e 3326}
3d14c5d2 3327EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3328
31b8006e
SW		 3329/*
3330 * Allocate "middle" portion of a message, if it is needed and wasn't
3331 * allocated by alloc_msg. This allows us to read a small fixed-size
3332 * per-type header in the front and then gracefully fail (i.e.,
3333 * propagate the error to the caller based on info in the front) when
3334 * the middle is too large.
3335 */
2450418c 3336static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3337{
3338 int type = le16_to_cpu(msg->hdr.type);
3339 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3340
3341 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3342 ceph_msg_type_name(type), middle_len);
3343 BUG_ON(!middle_len);
3344 BUG_ON(msg->middle);
3345
b6c1d5b8 3346 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3347 if (!msg->middle)
3348 return -ENOMEM;
3349 return 0;
3350}
3351
2450418c 3352/*
1c20f2d2
AE		 3353 * Allocate a message for receiving an incoming message on a
3354 * connection, and save the result in con->in_msg. Uses the
3355 * connection's private alloc_msg op if available.
3356 *
4740a623
SW		 3357 * Returns 0 on success, or a negative error code.
3358 *
3359 * On success, if we set *skip = 1:
3360 * - the next message should be skipped and ignored.
3361 * - con->in_msg == NULL
3362 * or if we set *skip = 0:
3363 * - con->in_msg is non-null.
3364 * On error (ENOMEM, EAGAIN, ...),
3365 * - con->in_msg == NULL
2450418c 3366 */
4740a623 3367static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3368{
4740a623 3369 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3370 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3371 struct ceph_msg *msg;
4740a623 3372 int ret = 0;
2450418c 3373
1c20f2d2 3374 BUG_ON(con->in_msg != NULL);
53ded495 3375 BUG_ON(!con->ops->alloc_msg);
2450418c 3376
53ded495
AE		 3377 mutex_unlock(&con->mutex);
3378 msg = con->ops->alloc_msg(con, hdr, skip);
3379 mutex_lock(&con->mutex);
3380 if (con->state != CON_STATE_OPEN) {
3381 if (msg)
1d866d1c 3382 ceph_msg_put(msg);
53ded495
AE		 3383 return -EAGAIN;
3384 }
4137577a
AE		 3385 if (msg) {
3386 BUG_ON(*skip);
583d0fef 3387 msg_con_set(msg, con);
4137577a 3388 con->in_msg = msg;
4137577a
AE		 3389 } else {
3390 /*
3391 * Null message pointer means either we should skip
3392 * this message or we couldn't allocate memory. The
3393 * former is not an error.
3394 */
3395 if (*skip)
3396 return 0;
4137577a 3397
67c64eb7 3398 con->error_msg = "error allocating memory for incoming message";
53ded495 3399 return -ENOMEM;
2450418c 3400 }
1c20f2d2 3401 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3402
1c20f2d2
AE		 3403 if (middle_len && !con->in_msg->middle) {
3404 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3405 if (ret < 0) {
1c20f2d2
AE		 3406 ceph_msg_put(con->in_msg);
3407 con->in_msg = NULL;
2450418c
YS		 3408 }
3409 }
9d7f0f13 3410
4740a623 3411 return ret;
2450418c
YS		 3412}
3413
31b8006e
SW		 3414
3415/*
3416 * Free a generically kmalloc'd message.
3417 */
0215e44b 3418static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3419{
0215e44b 3420 dout("%s %p\n", __func__, m);
4965fc38 3421 kvfree(m->front.iov_base);
e3d5d638 3422 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3423}
3424
0215e44b 3425static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3426{
3427 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
10bcee14 3428 struct ceph_msg_data *data, *next;
31b8006e 3429
0215e44b 3430 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3431 WARN_ON(!list_empty(&m->list_head));
3432
583d0fef
ID		 3433 msg_con_set(m, NULL);
3434
c2e552e7
SW		 3435 /* drop middle, data, if any */
3436 if (m->middle) {
3437 ceph_buffer_put(m->middle);
3438 m->middle = NULL;
31b8006e 3439 }
5240d9f9 3440
10bcee14
GT		 3441 list_for_each_entry_safe(data, next, &m->data, links) {
3442 list_del_init(&data->links);
5240d9f9
AE		 3443 ceph_msg_data_destroy(data);
3444 }
a1930804 3445 m->data_length = 0;
58bb3b37 3446
c2e552e7
SW		 3447 if (m->pool)
3448 ceph_msgpool_put(m->pool, m);
3449 else
0215e44b
ID		 3450 ceph_msg_free(m);
3451}
3452
3453struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3454{
3455 dout("%s %p (was %d)\n", __func__, msg,
2c935bc5 3456 kref_read(&msg->kref));
0215e44b
ID		 3457 kref_get(&msg->kref);
3458 return msg;
3459}
3460EXPORT_SYMBOL(ceph_msg_get);
3461
3462void ceph_msg_put(struct ceph_msg *msg)
3463{
3464 dout("%s %p (was %d)\n", __func__, msg,
2c935bc5 3465 kref_read(&msg->kref));
0215e44b 3466 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3467}
0215e44b 3468EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3469
3470void ceph_msg_dump(struct ceph_msg *msg)
3471{
3cea4c30
ID		 3472 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3473 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3474 print_hex_dump(KERN_DEBUG, "header: ",
3475 DUMP_PREFIX_OFFSET, 16, 1,
3476 &msg->hdr, sizeof(msg->hdr), true);
3477 print_hex_dump(KERN_DEBUG, " front: ",
3478 DUMP_PREFIX_OFFSET, 16, 1,
3479 msg->front.iov_base, msg->front.iov_len, true);
3480 if (msg->middle)
3481 print_hex_dump(KERN_DEBUG, "middle: ",
3482 DUMP_PREFIX_OFFSET, 16, 1,
3483 msg->middle->vec.iov_base,
3484 msg->middle->vec.iov_len, true);
3485 print_hex_dump(KERN_DEBUG, "footer: ",
3486 DUMP_PREFIX_OFFSET, 16, 1,
3487 &msg->footer, sizeof(msg->footer), true);
3488}
3d14c5d2 3489EXPORT_SYMBOL(ceph_msg_dump);
ubuntu-bionic-kernel mirror