]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blame - net/ceph/messenger.c
projects / mirror_ubuntu-hirsute-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: factor out encrypt_authorizer()
[mirror_ubuntu-hirsute-kernel.git] / net / ceph / messenger.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
3d14c5d2 2#include <linux/ceph/ceph_debug.h>
31b8006e
SW		 3
4#include <linux/crc32c.h>
5#include <linux/ctype.h>
6#include <linux/highmem.h>
7#include <linux/inet.h>
8#include <linux/kthread.h>
9#include <linux/net.h>
757856d2 10#include <linux/nsproxy.h>
633ee407 11#include <linux/sched/mm.h>
5a0e3ad6 12#include <linux/slab.h>
31b8006e
SW		 13#include <linux/socket.h>
14#include <linux/string.h>
3ebc21f7 15#ifdef CONFIG_BLOCK
68b4476b 16#include <linux/bio.h>
3ebc21f7 17#endif /* CONFIG_BLOCK */
ee3b56f2 18#include <linux/dns_resolver.h>
31b8006e
SW		 19#include <net/tcp.h>
20
2b3e0c90 21#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 22#include <linux/ceph/libceph.h>
23#include <linux/ceph/messenger.h>
24#include <linux/ceph/decode.h>
25#include <linux/ceph/pagelist.h>
bc3b2d7f 26#include <linux/export.h>
31b8006e
SW		 27
28/*
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
34 * the sender.
35 */
36
bc18f4b1
AE		 37/*
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
41 * unexpected state.
42 *
43 * --------
44 * | NEW* | transient initial state
45 * --------
46 * | con_sock_state_init()
47 * v
48 * ----------
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
51 * ^ \
52 * | \ con_sock_state_connecting()
53 * | ----------------------
54 * | \
55 * + con_sock_state_closed() \
fbb85a47
SW		 56 * |+--------------------------- \
57 * | \ \ \
58 * | ----------- \ \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
61 * | ^ \ |
62 * | | \ |
63 * | + con_sock_state_closing() \ |
64 * | / \ | |
65 * | / --------------- | |
66 * | / \ v v
bc18f4b1
AE		 67 * | / --------------
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
71 * | | v
72 * -------------
73 * | CONNECTED | TCP connection established
74 * -------------
75 *
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77 */
ce2c8903
AE		 78
79#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
84
8dacc7da
SW		 85/*
86 * connection states
87 */
88#define CON_STATE_CLOSED 1 /* -> PREOPEN */
89#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
94
4a861692
SW		 95/*
96 * ceph_connection flag bits
97 */
98#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
8dacc7da 104
c9ffc77a
AE		 105static bool con_flag_valid(unsigned long con_flag)
106{
107 switch (con_flag) {
108 case CON_FLAG_LOSSYTX:
109 case CON_FLAG_KEEPALIVE_PENDING:
110 case CON_FLAG_WRITE_PENDING:
111 case CON_FLAG_SOCK_CLOSED:
112 case CON_FLAG_BACKOFF:
113 return true;
114 default:
115 return false;
116 }
117}
118
119static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120{
121 BUG_ON(!con_flag_valid(con_flag));
122
123 clear_bit(con_flag, &con->flags);
124}
125
126static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127{
128 BUG_ON(!con_flag_valid(con_flag));
129
130 set_bit(con_flag, &con->flags);
131}
132
133static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134{
135 BUG_ON(!con_flag_valid(con_flag));
136
137 return test_bit(con_flag, &con->flags);
138}
139
140static bool con_flag_test_and_clear(struct ceph_connection *con,
141 unsigned long con_flag)
142{
143 BUG_ON(!con_flag_valid(con_flag));
144
145 return test_and_clear_bit(con_flag, &con->flags);
146}
147
148static bool con_flag_test_and_set(struct ceph_connection *con,
149 unsigned long con_flag)
150{
151 BUG_ON(!con_flag_valid(con_flag));
152
153 return test_and_set_bit(con_flag, &con->flags);
154}
155
e3d5d638
AE		 156/* Slab caches for frequently-allocated structures */
157
158static struct kmem_cache *ceph_msg_cache;
81b36be4 159static struct kmem_cache *ceph_msg_data_cache;
e3d5d638 160
31b8006e
SW		 161/* static tag bytes (protocol control messages) */
162static char tag_msg = CEPH_MSGR_TAG_MSG;
163static char tag_ack = CEPH_MSGR_TAG_ACK;
164static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
8b9558aa 165static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
31b8006e 166
a6a5349d
SW		 167#ifdef CONFIG_LOCKDEP
168static struct lock_class_key socket_class;
169#endif
170
31b8006e 171static void queue_con(struct ceph_connection *con);
37ab77ac 172static void cancel_con(struct ceph_connection *con);
68931622 173static void ceph_con_workfn(struct work_struct *);
93209264 174static void con_fault(struct ceph_connection *con);
31b8006e 175
31b8006e 176/*
f64a9317
AE		 177 * Nicely render a sockaddr as a string. An array of formatted
178 * strings is used, to approximate reentrancy.
31b8006e 179 */
f64a9317
AE		 180#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
181#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
182#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
183#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
184
185static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
186static atomic_t addr_str_seq = ATOMIC_INIT(0);
31b8006e 187
57666519 188static struct page *zero_page; /* used in certain error cases */
57666519 189
3d14c5d2 190const char *ceph_pr_addr(const struct sockaddr_storage *ss)
31b8006e
SW		 191{
192 int i;
193 char *s;
99f0f3b2
AE		 194 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
195 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
31b8006e 196
f64a9317 197 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
31b8006e
SW		 198 s = addr_str[i];
199
200 switch (ss->ss_family) {
201 case AF_INET:
bd406145
AE		 202 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
203 ntohs(in4->sin_port));
31b8006e
SW		 204 break;
205
206 case AF_INET6:
bd406145
AE		 207 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
208 ntohs(in6->sin6_port));
31b8006e
SW		 209 break;
210
211 default:
d3002b97
AE		 212 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
213 ss->ss_family);
31b8006e
SW		 214 }
215
216 return s;
217}
3d14c5d2 218EXPORT_SYMBOL(ceph_pr_addr);
31b8006e 219
63f2d211
SW		 220static void encode_my_addr(struct ceph_messenger *msgr)
221{
222 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
223 ceph_encode_addr(&msgr->my_enc_addr);
224}
225
31b8006e
SW		 226/*
227 * work queue for all reading and writing to/from the socket.
228 */
e0f43c94 229static struct workqueue_struct *ceph_msgr_wq;
31b8006e 230
e3d5d638
AE		 231static int ceph_msgr_slab_init(void)
232{
233 BUG_ON(ceph_msg_cache);
5ee61e95 234 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
81b36be4
AE		 235 if (!ceph_msg_cache)
236 return -ENOMEM;
237
238 BUG_ON(ceph_msg_data_cache);
5ee61e95 239 ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
81b36be4
AE		 240 if (ceph_msg_data_cache)
241 return 0;
242
243 kmem_cache_destroy(ceph_msg_cache);
244 ceph_msg_cache = NULL;
245
246 return -ENOMEM;
e3d5d638
AE		 247}
248
249static void ceph_msgr_slab_exit(void)
250{
81b36be4
AE		 251 BUG_ON(!ceph_msg_data_cache);
252 kmem_cache_destroy(ceph_msg_data_cache);
253 ceph_msg_data_cache = NULL;
254
e3d5d638
AE		 255 BUG_ON(!ceph_msg_cache);
256 kmem_cache_destroy(ceph_msg_cache);
257 ceph_msg_cache = NULL;
258}
259
15417167 260static void _ceph_msgr_exit(void)
6173d1f0 261{
d3002b97 262 if (ceph_msgr_wq) {
6173d1f0 263 destroy_workqueue(ceph_msgr_wq);
d3002b97
AE		 264 ceph_msgr_wq = NULL;
265 }
6173d1f0 266
6173d1f0 267 BUG_ON(zero_page == NULL);
09cbfeaf 268 put_page(zero_page);
6173d1f0 269 zero_page = NULL;
d920ff6f
BC		 270
271 ceph_msgr_slab_exit();
6173d1f0
AE		 272}
273
57a35dfb 274int __init ceph_msgr_init(void)
31b8006e 275{
d920ff6f
BC		 276 if (ceph_msgr_slab_init())
277 return -ENOMEM;
278
57666519
AE		 279 BUG_ON(zero_page != NULL);
280 zero_page = ZERO_PAGE(0);
09cbfeaf 281 get_page(zero_page);
57666519 282
f9865f06
ID		 283 /*
284 * The number of active work items is limited by the number of
285 * connections, so leave @max_active at default.
286 */
287 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
6173d1f0
AE		 288 if (ceph_msgr_wq)
289 return 0;
57666519 290
6173d1f0
AE		 291 pr_err("msgr_init failed to create workqueue\n");
292 _ceph_msgr_exit();
57666519 293
6173d1f0 294 return -ENOMEM;
31b8006e
SW		 295}
296
297void ceph_msgr_exit(void)
298{
57666519 299 BUG_ON(ceph_msgr_wq == NULL);
57666519 300
6173d1f0 301 _ceph_msgr_exit();
31b8006e
SW		 302}
303
cd84db6e 304void ceph_msgr_flush(void)
a922d38f
SW		 305{
306 flush_workqueue(ceph_msgr_wq);
307}
3d14c5d2 308EXPORT_SYMBOL(ceph_msgr_flush);
a922d38f 309
ce2c8903
AE		 310/* Connection socket state transition functions */
311
312static void con_sock_state_init(struct ceph_connection *con)
313{
314 int old_state;
315
316 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
317 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
318 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 319 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
320 CON_SOCK_STATE_CLOSED);
ce2c8903
AE		 321}
322
323static void con_sock_state_connecting(struct ceph_connection *con)
324{
325 int old_state;
326
327 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
328 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
329 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 330 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
331 CON_SOCK_STATE_CONNECTING);
ce2c8903
AE		 332}
333
334static void con_sock_state_connected(struct ceph_connection *con)
335{
336 int old_state;
337
338 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
339 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
340 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 341 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
342 CON_SOCK_STATE_CONNECTED);
ce2c8903
AE		 343}
344
345static void con_sock_state_closing(struct ceph_connection *con)
346{
347 int old_state;
348
349 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
350 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
351 old_state != CON_SOCK_STATE_CONNECTED &&
352 old_state != CON_SOCK_STATE_CLOSING))
353 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 354 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
355 CON_SOCK_STATE_CLOSING);
ce2c8903
AE		 356}
357
358static void con_sock_state_closed(struct ceph_connection *con)
359{
360 int old_state;
361
362 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
363 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
fbb85a47 364 old_state != CON_SOCK_STATE_CLOSING &&
8007b8d6
SW		 365 old_state != CON_SOCK_STATE_CONNECTING &&
366 old_state != CON_SOCK_STATE_CLOSED))
ce2c8903 367 printk("%s: unexpected old state %d\n", __func__, old_state);
8007b8d6
SW		 368 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
369 CON_SOCK_STATE_CLOSED);
ce2c8903 370}
a922d38f 371
31b8006e
SW		 372/*
373 * socket callback functions
374 */
375
376/* data available on socket, or listen socket received a connect */
676d2369 377static void ceph_sock_data_ready(struct sock *sk)
31b8006e 378{
bd406145 379 struct ceph_connection *con = sk->sk_user_data;
a2a32584
GH		 380 if (atomic_read(&con->msgr->stopping)) {
381 return;
382 }
bd406145 383
31b8006e 384 if (sk->sk_state != TCP_CLOSE_WAIT) {
327800bd 385 dout("%s on %p state = %lu, queueing work\n", __func__,
31b8006e
SW		 386 con, con->state);
387 queue_con(con);
388 }
389}
390
391/* socket has buffer space for writing */
327800bd 392static void ceph_sock_write_space(struct sock *sk)
31b8006e 393{
d3002b97 394 struct ceph_connection *con = sk->sk_user_data;
31b8006e 395
182fac26
JS		 396 /* only queue to workqueue if there is data we want to write,
397 * and there is sufficient space in the socket buffer to accept
327800bd 398 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
182fac26
JS		 399 * doesn't get called again until try_write() fills the socket
400 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
401 * and net/core/stream.c:sk_stream_write_space().
402 */
c9ffc77a 403 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
64dc6130 404 if (sk_stream_is_writeable(sk)) {
327800bd 405 dout("%s %p queueing write work\n", __func__, con);
182fac26
JS		 406 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
407 queue_con(con);
408 }
31b8006e 409 } else {
327800bd 410 dout("%s %p nothing to write\n", __func__, con);
31b8006e 411 }
31b8006e
SW		 412}
413
414/* socket's state has changed */
327800bd 415static void ceph_sock_state_change(struct sock *sk)
31b8006e 416{
bd406145 417 struct ceph_connection *con = sk->sk_user_data;
31b8006e 418
327800bd 419 dout("%s %p state = %lu sk_state = %u\n", __func__,
31b8006e
SW		 420 con, con->state, sk->sk_state);
421
31b8006e
SW		 422 switch (sk->sk_state) {
423 case TCP_CLOSE:
327800bd 424 dout("%s TCP_CLOSE\n", __func__);
18370b36 425 /* fall through */
31b8006e 426 case TCP_CLOSE_WAIT:
327800bd 427 dout("%s TCP_CLOSE_WAIT\n", __func__);
ce2c8903 428 con_sock_state_closing(con);
c9ffc77a 429 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
d65c9e0b 430 queue_con(con);
31b8006e
SW		 431 break;
432 case TCP_ESTABLISHED:
327800bd 433 dout("%s TCP_ESTABLISHED\n", __func__);
ce2c8903 434 con_sock_state_connected(con);
31b8006e
SW		 435 queue_con(con);
436 break;
d3002b97
AE		 437 default: /* Everything else is uninteresting */
438 break;
31b8006e
SW		 439 }
440}
441
442/*
443 * set up socket callbacks
444 */
445static void set_sock_callbacks(struct socket *sock,
446 struct ceph_connection *con)
447{
448 struct sock *sk = sock->sk;
bd406145 449 sk->sk_user_data = con;
327800bd
AE		 450 sk->sk_data_ready = ceph_sock_data_ready;
451 sk->sk_write_space = ceph_sock_write_space;
452 sk->sk_state_change = ceph_sock_state_change;
31b8006e
SW		 453}
454
455
456/*
457 * socket helpers
458 */
459
460/*
461 * initiate connection to a remote socket.
462 */
41617d0c 463static int ceph_tcp_connect(struct ceph_connection *con)
31b8006e 464{
f91d3471 465 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
31b8006e 466 struct socket *sock;
633ee407 467 unsigned int noio_flag;
31b8006e
SW		 468 int ret;
469
470 BUG_ON(con->sock);
633ee407
ID		 471
472 /* sock_create_kern() allocates with GFP_KERNEL */
473 noio_flag = memalloc_noio_save();
757856d2 474 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
eeb1bd5c 475 SOCK_STREAM, IPPROTO_TCP, &sock);
633ee407 476 memalloc_noio_restore(noio_flag);
31b8006e 477 if (ret)
41617d0c 478 return ret;
6d7fdb0a 479 sock->sk->sk_allocation = GFP_NOFS;
31b8006e 480
a6a5349d
SW		 481#ifdef CONFIG_LOCKDEP
482 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
483#endif
484
31b8006e
SW		 485 set_sock_callbacks(sock, con);
486
3d14c5d2 487 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 488
89a86be0 489 con_sock_state_connecting(con);
f91d3471
SW		 490 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
491 O_NONBLOCK);
31b8006e
SW		 492 if (ret == -EINPROGRESS) {
493 dout("connect %s EINPROGRESS sk_state = %u\n",
3d14c5d2 494 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e 495 sock->sk->sk_state);
a5bc3129 496 } else if (ret < 0) {
31b8006e 497 pr_err("connect %s error %d\n",
3d14c5d2 498 ceph_pr_addr(&con->peer_addr.in_addr), ret);
31b8006e 499 sock_release(sock);
41617d0c 500 return ret;
a5bc3129 501 }
89baaa57 502
859bff51 503 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
ba988f87
CH		 504 int optval = 1;
505
506 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
507 (char *)&optval, sizeof(optval));
508 if (ret)
509 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
510 ret);
511 }
512
a5bc3129 513 con->sock = sock;
41617d0c 514 return 0;
31b8006e
SW		 515}
516
e5c93883
ID		 517/*
518 * If @buf is NULL, discard up to @len bytes.
519 */
31b8006e
SW		 520static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
521{
522 struct kvec iov = {buf, len};
523 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
98bdb0aa 524 int r;
31b8006e 525
e5c93883
ID		 526 if (!buf)
527 msg.msg_flags |= MSG_TRUNC;
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;
5359a17d 840 struct ceph_bio_iter *it = &cursor->bio_iter;
6aaa4511 841
5359a17d
ID		 842 cursor->resid = min_t(size_t, length, data->bio_length);
843 *it = data->bio_pos;
844 if (cursor->resid < it->iter.bi_size)
845 it->iter.bi_size = cursor->resid;
6aaa4511 846
5359a17d
ID		 847 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
848 cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
6aaa4511
AE		 849}
850
8ae4f4f5 851static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
6aaa4511
AE		 852 size_t *page_offset,
853 size_t *length)
854{
5359a17d
ID		 855 struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
856 cursor->bio_iter.iter);
6aaa4511 857
5359a17d
ID		 858 *page_offset = bv.bv_offset;
859 *length = bv.bv_len;
860 return bv.bv_page;
6aaa4511
AE		 861}
862
8ae4f4f5
AE		 863static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
864 size_t bytes)
6aaa4511 865{
5359a17d 866 struct ceph_bio_iter *it = &cursor->bio_iter;
6aaa4511 867
5359a17d
ID		 868 BUG_ON(bytes > cursor->resid);
869 BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
25aff7c5 870 cursor->resid -= bytes;
5359a17d 871 bio_advance_iter(it->bio, &it->iter, bytes);
f38a5181 872
5359a17d
ID		 873 if (!cursor->resid) {
874 BUG_ON(!cursor->last_piece);
875 return false; /* no more data */
876 }
f38a5181 877
5359a17d 878 if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done))
6aaa4511
AE		 879 return false; /* more bytes to process in this segment */
880
5359a17d
ID		 881 if (!it->iter.bi_size) {
882 it->bio = it->bio->bi_next;
883 it->iter = it->bio->bi_iter;
884 if (cursor->resid < it->iter.bi_size)
885 it->iter.bi_size = cursor->resid;
25aff7c5 886 }
6aaa4511 887
5359a17d
ID		 888 BUG_ON(cursor->last_piece);
889 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
890 cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
6aaa4511
AE		 891 return true;
892}
ea96571f 893#endif /* CONFIG_BLOCK */
df6ad1f9 894
b9e281c2
ID		 895static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
896 size_t length)
897{
898 struct ceph_msg_data *data = cursor->data;
899 struct bio_vec *bvecs = data->bvec_pos.bvecs;
900
901 cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
902 cursor->bvec_iter = data->bvec_pos.iter;
903 cursor->bvec_iter.bi_size = cursor->resid;
904
905 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
906 cursor->last_piece =
907 cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
908}
909
910static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
911 size_t *page_offset,
912 size_t *length)
913{
914 struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
915 cursor->bvec_iter);
916
917 *page_offset = bv.bv_offset;
918 *length = bv.bv_len;
919 return bv.bv_page;
920}
921
922static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
923 size_t bytes)
924{
925 struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
926
927 BUG_ON(bytes > cursor->resid);
928 BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
929 cursor->resid -= bytes;
930 bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
931
932 if (!cursor->resid) {
933 BUG_ON(!cursor->last_piece);
934 return false; /* no more data */
935 }
936
937 if (!bytes || cursor->bvec_iter.bi_bvec_done)
938 return false; /* more bytes to process in this segment */
939
940 BUG_ON(cursor->last_piece);
941 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
942 cursor->last_piece =
943 cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
944 return true;
945}
946
e766d7b5
AE		 947/*
948 * For a page array, a piece comes from the first page in the array
949 * that has not already been fully consumed.
950 */
8ae4f4f5 951static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 952 size_t length)
e766d7b5 953{
8ae4f4f5 954 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 955 int page_count;
956
957 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
958
959 BUG_ON(!data->pages);
960 BUG_ON(!data->length);
961
ca8b3a69 962 cursor->resid = min(length, data->length);
e766d7b5 963 page_count = calc_pages_for(data->alignment, (u64)data->length);
e766d7b5
AE		 964 cursor->page_offset = data->alignment & ~PAGE_MASK;
965 cursor->page_index = 0;
56fc5659
AE		 966 BUG_ON(page_count > (int)USHRT_MAX);
967 cursor->page_count = (unsigned short)page_count;
968 BUG_ON(length > SIZE_MAX - cursor->page_offset);
5f740d7e 969 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 970}
971
8ae4f4f5
AE		 972static struct page *
973ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
974 size_t *page_offset, size_t *length)
e766d7b5 975{
8ae4f4f5 976 struct ceph_msg_data *data = cursor->data;
e766d7b5
AE		 977
978 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
979
980 BUG_ON(cursor->page_index >= cursor->page_count);
981 BUG_ON(cursor->page_offset >= PAGE_SIZE);
e766d7b5
AE		 982
983 *page_offset = cursor->page_offset;
25aff7c5 984 if (cursor->last_piece)
e766d7b5 985 *length = cursor->resid;
25aff7c5 986 else
e766d7b5 987 *length = PAGE_SIZE - *page_offset;
e766d7b5
AE		 988
989 return data->pages[cursor->page_index];
990}
991
8ae4f4f5 992static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
e766d7b5
AE		 993 size_t bytes)
994{
8ae4f4f5 995 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
e766d7b5
AE		 996
997 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
e766d7b5
AE		 998
999 /* Advance the cursor page offset */
1000
1001 cursor->resid -= bytes;
5df521b1
AE		 1002 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
1003 if (!bytes || cursor->page_offset)
e766d7b5
AE		 1004 return false; /* more bytes to process in the current page */
1005
d90deda6
YZ		 1006 if (!cursor->resid)
1007 return false; /* no more data */
1008
5df521b1 1009 /* Move on to the next page; offset is already at 0 */
e766d7b5
AE		 1010
1011 BUG_ON(cursor->page_index >= cursor->page_count);
e766d7b5 1012 cursor->page_index++;
25aff7c5 1013 cursor->last_piece = cursor->resid <= PAGE_SIZE;
e766d7b5
AE		 1014
1015 return true;
1016}
1017
fe38a2b6 1018/*
dd236fcb
AE		 1019 * For a pagelist, a piece is whatever remains to be consumed in the
1020 * first page in the list, or the front of the next page.
fe38a2b6 1021 */
8ae4f4f5
AE		 1022static void
1023ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
25aff7c5 1024 size_t length)
fe38a2b6 1025{
8ae4f4f5 1026 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1027 struct ceph_pagelist *pagelist;
1028 struct page *page;
1029
dd236fcb 1030 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
fe38a2b6
AE		 1031
1032 pagelist = data->pagelist;
1033 BUG_ON(!pagelist);
25aff7c5
AE		 1034
1035 if (!length)
fe38a2b6
AE		 1036 return; /* pagelist can be assigned but empty */
1037
1038 BUG_ON(list_empty(&pagelist->head));
1039 page = list_first_entry(&pagelist->head, struct page, lru);
1040
ca8b3a69 1041 cursor->resid = min(length, pagelist->length);
fe38a2b6
AE		 1042 cursor->page = page;
1043 cursor->offset = 0;
a51b272e 1044 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1045}
1046
8ae4f4f5
AE		 1047static struct page *
1048ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1049 size_t *page_offset, size_t *length)
fe38a2b6 1050{
8ae4f4f5 1051 struct ceph_msg_data *data = cursor->data;
fe38a2b6 1052 struct ceph_pagelist *pagelist;
fe38a2b6
AE		 1053
1054 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1055
1056 pagelist = data->pagelist;
1057 BUG_ON(!pagelist);
1058
1059 BUG_ON(!cursor->page);
25aff7c5 1060 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6 1061
5df521b1 1062 /* offset of first page in pagelist is always 0 */
fe38a2b6 1063 *page_offset = cursor->offset & ~PAGE_MASK;
5df521b1 1064 if (cursor->last_piece)
25aff7c5
AE		 1065 *length = cursor->resid;
1066 else
1067 *length = PAGE_SIZE - *page_offset;
fe38a2b6 1068
8ae4f4f5 1069 return cursor->page;
fe38a2b6
AE		 1070}
1071
8ae4f4f5 1072static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
dd236fcb 1073 size_t bytes)
fe38a2b6 1074{
8ae4f4f5 1075 struct ceph_msg_data *data = cursor->data;
fe38a2b6
AE		 1076 struct ceph_pagelist *pagelist;
1077
1078 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1079
1080 pagelist = data->pagelist;
1081 BUG_ON(!pagelist);
25aff7c5
AE		 1082
1083 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
fe38a2b6
AE		 1084 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1085
1086 /* Advance the cursor offset */
1087
25aff7c5 1088 cursor->resid -= bytes;
fe38a2b6 1089 cursor->offset += bytes;
5df521b1 1090 /* offset of first page in pagelist is always 0 */
fe38a2b6
AE		 1091 if (!bytes || cursor->offset & ~PAGE_MASK)
1092 return false; /* more bytes to process in the current page */
1093
d90deda6
YZ		 1094 if (!cursor->resid)
1095 return false; /* no more data */
1096
fe38a2b6
AE		 1097 /* Move on to the next page */
1098
1099 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
17ddc49b 1100 cursor->page = list_next_entry(cursor->page, lru);
25aff7c5 1101 cursor->last_piece = cursor->resid <= PAGE_SIZE;
fe38a2b6
AE		 1102
1103 return true;
1104}
1105
dd236fcb
AE		 1106/*
1107 * Message data is handled (sent or received) in pieces, where each
1108 * piece resides on a single page. The network layer might not
1109 * consume an entire piece at once. A data item's cursor keeps
1110 * track of which piece is next to process and how much remains to
1111 * be processed in that piece. It also tracks whether the current
1112 * piece is the last one in the data item.
1113 */
ca8b3a69 1114static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
dd236fcb 1115{
ca8b3a69 1116 size_t length = cursor->total_resid;
8ae4f4f5 1117
8ae4f4f5 1118 switch (cursor->data->type) {
dd236fcb 1119 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1120 ceph_msg_data_pagelist_cursor_init(cursor, length);
dd236fcb 1121 break;
e766d7b5 1122 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1123 ceph_msg_data_pages_cursor_init(cursor, length);
e766d7b5 1124 break;
dd236fcb
AE		 1125#ifdef CONFIG_BLOCK
1126 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1127 ceph_msg_data_bio_cursor_init(cursor, length);
6aaa4511 1128 break;
dd236fcb 1129#endif /* CONFIG_BLOCK */
b9e281c2
ID		 1130 case CEPH_MSG_DATA_BVECS:
1131 ceph_msg_data_bvecs_cursor_init(cursor, length);
1132 break;
6aaa4511 1133 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1134 default:
1135 /* BUG(); */
1136 break;
1137 }
8ae4f4f5 1138 cursor->need_crc = true;
dd236fcb
AE		 1139}
1140
ca8b3a69
AE		 1141static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1142{
1143 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1144 struct ceph_msg_data *data;
1145
1146 BUG_ON(!length);
1147 BUG_ON(length > msg->data_length);
1148 BUG_ON(list_empty(&msg->data));
1149
ca8b3a69
AE		 1150 cursor->data_head = &msg->data;
1151 cursor->total_resid = length;
1152 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1153 cursor->data = data;
1154
1155 __ceph_msg_data_cursor_init(cursor);
1156}
1157
dd236fcb
AE		 1158/*
1159 * Return the page containing the next piece to process for a given
1160 * data item, and supply the page offset and length of that piece.
1161 * Indicate whether this is the last piece in this data item.
1162 */
8ae4f4f5
AE		 1163static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1164 size_t *page_offset, size_t *length,
dd236fcb
AE		 1165 bool *last_piece)
1166{
1167 struct page *page;
1168
8ae4f4f5 1169 switch (cursor->data->type) {
dd236fcb 1170 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1171 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
dd236fcb 1172 break;
e766d7b5 1173 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1174 page = ceph_msg_data_pages_next(cursor, page_offset, length);
e766d7b5 1175 break;
dd236fcb
AE		 1176#ifdef CONFIG_BLOCK
1177 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1178 page = ceph_msg_data_bio_next(cursor, page_offset, length);
6aaa4511 1179 break;
dd236fcb 1180#endif /* CONFIG_BLOCK */
b9e281c2
ID		 1181 case CEPH_MSG_DATA_BVECS:
1182 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1183 break;
6aaa4511 1184 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1185 default:
1186 page = NULL;
1187 break;
1188 }
5359a17d 1189
dd236fcb
AE		 1190 BUG_ON(!page);
1191 BUG_ON(*page_offset + *length > PAGE_SIZE);
1192 BUG_ON(!*length);
5359a17d 1193 BUG_ON(*length > cursor->resid);
dd236fcb 1194 if (last_piece)
8ae4f4f5 1195 *last_piece = cursor->last_piece;
dd236fcb
AE		 1196
1197 return page;
1198}
1199
1200/*
1201 * Returns true if the result moves the cursor on to the next piece
1202 * of the data item.
1203 */
1759f7b0
ID		 1204static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1205 size_t bytes)
dd236fcb
AE		 1206{
1207 bool new_piece;
1208
25aff7c5 1209 BUG_ON(bytes > cursor->resid);
8ae4f4f5 1210 switch (cursor->data->type) {
dd236fcb 1211 case CEPH_MSG_DATA_PAGELIST:
8ae4f4f5 1212 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
dd236fcb 1213 break;
e766d7b5 1214 case CEPH_MSG_DATA_PAGES:
8ae4f4f5 1215 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
e766d7b5 1216 break;
dd236fcb
AE		 1217#ifdef CONFIG_BLOCK
1218 case CEPH_MSG_DATA_BIO:
8ae4f4f5 1219 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
6aaa4511 1220 break;
dd236fcb 1221#endif /* CONFIG_BLOCK */
b9e281c2
ID		 1222 case CEPH_MSG_DATA_BVECS:
1223 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1224 break;
6aaa4511 1225 case CEPH_MSG_DATA_NONE:
dd236fcb
AE		 1226 default:
1227 BUG();
1228 break;
1229 }
ca8b3a69 1230 cursor->total_resid -= bytes;
dd236fcb 1231
ca8b3a69
AE		 1232 if (!cursor->resid && cursor->total_resid) {
1233 WARN_ON(!cursor->last_piece);
1234 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
17ddc49b 1235 cursor->data = list_next_entry(cursor->data, links);
ca8b3a69 1236 __ceph_msg_data_cursor_init(cursor);
a51b272e 1237 new_piece = true;
ca8b3a69 1238 }
a51b272e 1239 cursor->need_crc = new_piece;
dd236fcb
AE		 1240}
1241
dbc0d3ca
ID		 1242static size_t sizeof_footer(struct ceph_connection *con)
1243{
1244 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1245 sizeof(struct ceph_msg_footer) :
1246 sizeof(struct ceph_msg_footer_old);
1247}
1248
98fa5dd8 1249static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
739c905b 1250{
739c905b 1251 BUG_ON(!msg);
25aff7c5 1252 BUG_ON(!data_len);
739c905b 1253
4c59b4a2 1254 /* Initialize data cursor */
fe38a2b6 1255
8ae4f4f5 1256 ceph_msg_data_cursor_init(msg, (size_t)data_len);
739c905b
AE		 1257}
1258
31b8006e
SW		 1259/*
1260 * Prepare footer for currently outgoing message, and finish things
1261 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1262 */
859eb799 1263static void prepare_write_message_footer(struct ceph_connection *con)
31b8006e
SW		 1264{
1265 struct ceph_msg *m = con->out_msg;
1266
fd154f3c
AE		 1267 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1268
31b8006e 1269 dout("prepare_write_message_footer %p\n", con);
89f08173 1270 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
33d07337
YZ		 1271 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1272 if (con->ops->sign_message)
79dbd1ba 1273 con->ops->sign_message(m);
33d07337
YZ		 1274 else
1275 m->footer.sig = 0;
33d07337
YZ		 1276 } else {
1277 m->old_footer.flags = m->footer.flags;
33d07337 1278 }
31b8006e 1279 con->out_more = m->more_to_follow;
c86a2930 1280 con->out_msg_done = true;
31b8006e
SW		 1281}
1282
1283/*
1284 * Prepare headers for the next outgoing message.
1285 */
1286static void prepare_write_message(struct ceph_connection *con)
1287{
1288 struct ceph_msg *m;
a9a0c51a 1289 u32 crc;
31b8006e 1290
e2200423 1291 con_out_kvec_reset(con);
c86a2930 1292 con->out_msg_done = false;
31b8006e
SW		 1293
1294 /* Sneak an ack in there first? If we can get it into the same
1295 * TCP packet that's a good thing. */
1296 if (con->in_seq > con->in_seq_acked) {
1297 con->in_seq_acked = con->in_seq;
e2200423 1298 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
31b8006e 1299 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1300 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799 1301 &con->out_temp_ack);
31b8006e
SW		 1302 }
1303
38941f80 1304 BUG_ON(list_empty(&con->out_queue));
859eb799 1305 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
c86a2930 1306 con->out_msg = m;
38941f80 1307 BUG_ON(m->con != con);
4cf9d544
SW		 1308
1309 /* put message on sent list */
1310 ceph_msg_get(m);
1311 list_move_tail(&m->list_head, &con->out_sent);
31b8006e 1312
e84346b7
SW		 1313 /*
1314 * only assign outgoing seq # if we haven't sent this message
1315 * yet. if it is requeued, resend with it's original seq.
1316 */
1317 if (m->needs_out_seq) {
1318 m->hdr.seq = cpu_to_le64(++con->out_seq);
1319 m->needs_out_seq = false;
98ad5ebd 1320
4690faf0
ID		 1321 if (con->ops->reencode_message)
1322 con->ops->reencode_message(m);
1323 }
31b8006e 1324
98fa5dd8 1325 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
31b8006e
SW		 1326 m, con->out_seq, le16_to_cpu(m->hdr.type),
1327 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
98fa5dd8 1328 m->data_length);
98ad5ebd
ID		 1329 WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1330 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
31b8006e
SW		 1331
1332 /* tag + hdr + front + middle */
e2200423 1333 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
67645d76 1334 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
e2200423 1335 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
859eb799 1336
31b8006e 1337 if (m->middle)
e2200423 1338 con_out_kvec_add(con, m->middle->vec.iov_len,
859eb799 1339 m->middle->vec.iov_base);
31b8006e 1340
67645d76 1341 /* fill in hdr crc and finalize hdr */
a9a0c51a
AE		 1342 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1343 con->out_msg->hdr.crc = cpu_to_le32(crc);
67645d76 1344 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
a9a0c51a 1345
67645d76 1346 /* fill in front and middle crc, footer */
a9a0c51a
AE		 1347 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1348 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1349 if (m->middle) {
1350 crc = crc32c(0, m->middle->vec.iov_base,
1351 m->middle->vec.iov_len);
1352 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1353 } else
31b8006e 1354 con->out_msg->footer.middle_crc = 0;
739c905b 1355 dout("%s front_crc %u middle_crc %u\n", __func__,
31b8006e
SW		 1356 le32_to_cpu(con->out_msg->footer.front_crc),
1357 le32_to_cpu(con->out_msg->footer.middle_crc));
67645d76 1358 con->out_msg->footer.flags = 0;
31b8006e
SW		 1359
1360 /* is there a data payload? */
739c905b 1361 con->out_msg->footer.data_crc = 0;
98fa5dd8
AE		 1362 if (m->data_length) {
1363 prepare_message_data(con->out_msg, m->data_length);
78625051
AE		 1364 con->out_more = 1; /* data + footer will follow */
1365 } else {
31b8006e 1366 /* no, queue up footer too and be done */
859eb799 1367 prepare_write_message_footer(con);
78625051 1368 }
31b8006e 1369
c9ffc77a 1370 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1371}
1372
1373/*
1374 * Prepare an ack.
1375 */
1376static void prepare_write_ack(struct ceph_connection *con)
1377{
1378 dout("prepare_write_ack %p %llu -> %llu\n", con,
1379 con->in_seq_acked, con->in_seq);
1380 con->in_seq_acked = con->in_seq;
1381
e2200423 1382 con_out_kvec_reset(con);
859eb799 1383
e2200423 1384 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
859eb799 1385
31b8006e 1386 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
e2200423 1387 con_out_kvec_add(con, sizeof (con->out_temp_ack),
859eb799
AE		 1388 &con->out_temp_ack);
1389
31b8006e 1390 con->out_more = 1; /* more will follow.. eventually.. */
c9ffc77a 1391 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1392}
1393
3a23083b
SW		 1394/*
1395 * Prepare to share the seq during handshake
1396 */
1397static void prepare_write_seq(struct ceph_connection *con)
1398{
1399 dout("prepare_write_seq %p %llu -> %llu\n", con,
1400 con->in_seq_acked, con->in_seq);
1401 con->in_seq_acked = con->in_seq;
1402
1403 con_out_kvec_reset(con);
1404
1405 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1406 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1407 &con->out_temp_ack);
1408
1409 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1410}
1411
31b8006e
SW		 1412/*
1413 * Prepare to write keepalive byte.
1414 */
1415static void prepare_write_keepalive(struct ceph_connection *con)
1416{
1417 dout("prepare_write_keepalive %p\n", con);
e2200423 1418 con_out_kvec_reset(con);
8b9558aa 1419 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
473bd2d7 1420 struct timespec64 now;
7f61f545 1421
473bd2d7 1422 ktime_get_real_ts64(&now);
8b9558aa 1423 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
473bd2d7 1424 ceph_encode_timespec64(&con->out_temp_keepalive2, &now);
7f61f545
ID		 1425 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1426 &con->out_temp_keepalive2);
8b9558aa
YZ		 1427 } else {
1428 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1429 }
c9ffc77a 1430 con_flag_set(con, CON_FLAG_WRITE_PENDING);
31b8006e
SW		 1431}
1432
1433/*
1434 * Connection negotiation.
1435 */
1436
262614c4 1437static int get_connect_authorizer(struct ceph_connection *con)
4e7a5dcd 1438{
a3530df3 1439 struct ceph_auth_handshake *auth;
262614c4 1440 int auth_proto;
b1c6b980
AE		 1441
1442 if (!con->ops->get_authorizer) {
262614c4 1443 con->auth = NULL;
b1c6b980
AE		 1444 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1445 con->out_connect.authorizer_len = 0;
262614c4 1446 return 0;
b1c6b980
AE		 1447 }
1448
262614c4 1449 auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
a3530df3 1450 if (IS_ERR(auth))
262614c4 1451 return PTR_ERR(auth);
0da5d703 1452
262614c4
ID		 1453 con->auth = auth;
1454 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1455 con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1456 return 0;
4e7a5dcd
SW		 1457}
1458
31b8006e
SW		 1459/*
1460 * We connected to a peer and are saying hello.
1461 */
e825a66d 1462static void prepare_write_banner(struct ceph_connection *con)
31b8006e 1463{
e2200423
AE		 1464 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1465 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
e825a66d 1466 &con->msgr->my_enc_addr);
eed0ef2c 1467
eed0ef2c 1468 con->out_more = 0;
c9ffc77a 1469 con_flag_set(con, CON_FLAG_WRITE_PENDING);
eed0ef2c
SW		 1470}
1471
c0f56b48
ID		 1472static void __prepare_write_connect(struct ceph_connection *con)
1473{
1474 con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1475 if (con->auth)
1476 con_out_kvec_add(con, con->auth->authorizer_buf_len,
1477 con->auth->authorizer_buf);
1478
1479 con->out_more = 0;
1480 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1481}
1482
e825a66d 1483static int prepare_write_connect(struct ceph_connection *con)
eed0ef2c 1484{
95c96174 1485 unsigned int global_seq = get_global_seq(con->msgr, 0);
31b8006e 1486 int proto;
262614c4 1487 int ret;
31b8006e
SW		 1488
1489 switch (con->peer_name.type) {
1490 case CEPH_ENTITY_TYPE_MON:
1491 proto = CEPH_MONC_PROTOCOL;
1492 break;
1493 case CEPH_ENTITY_TYPE_OSD:
1494 proto = CEPH_OSDC_PROTOCOL;
1495 break;
1496 case CEPH_ENTITY_TYPE_MDS:
1497 proto = CEPH_MDSC_PROTOCOL;
1498 break;
1499 default:
1500 BUG();
1501 }
1502
1503 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1504 con->connect_seq, global_seq, proto);
4e7a5dcd 1505
859bff51
ID		 1506 con->out_connect.features =
1507 cpu_to_le64(from_msgr(con->msgr)->supported_features);
31b8006e
SW		 1508 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1509 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1510 con->out_connect.global_seq = cpu_to_le32(global_seq);
1511 con->out_connect.protocol_version = cpu_to_le32(proto);
1512 con->out_connect.flags = 0;
31b8006e 1513
262614c4
ID		 1514 ret = get_connect_authorizer(con);
1515 if (ret)
1516 return ret;
3da54776 1517
c0f56b48 1518 __prepare_write_connect(con);
e10c758e 1519 return 0;
31b8006e
SW		 1520}
1521
31b8006e
SW		 1522/*
1523 * write as much of pending kvecs to the socket as we can.
1524 * 1 -> done
1525 * 0 -> socket full, but more to do
1526 * <0 -> error
1527 */
1528static int write_partial_kvec(struct ceph_connection *con)
1529{
1530 int ret;
1531
1532 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1533 while (con->out_kvec_bytes > 0) {
1534 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1535 con->out_kvec_left, con->out_kvec_bytes,
1536 con->out_more);
1537 if (ret <= 0)
1538 goto out;
1539 con->out_kvec_bytes -= ret;
1540 if (con->out_kvec_bytes == 0)
1541 break; /* done */
f42299e6
AE		 1542
1543 /* account for full iov entries consumed */
1544 while (ret >= con->out_kvec_cur->iov_len) {
1545 BUG_ON(!con->out_kvec_left);
1546 ret -= con->out_kvec_cur->iov_len;
1547 con->out_kvec_cur++;
1548 con->out_kvec_left--;
1549 }
1550 /* and for a partially-consumed entry */
1551 if (ret) {
1552 con->out_kvec_cur->iov_len -= ret;
1553 con->out_kvec_cur->iov_base += ret;
31b8006e
SW		 1554 }
1555 }
1556 con->out_kvec_left = 0;
31b8006e
SW		 1557 ret = 1;
1558out:
1559 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1560 con->out_kvec_bytes, con->out_kvec_left, ret);
1561 return ret; /* done! */
1562}
1563
35b62808
AE		 1564static u32 ceph_crc32c_page(u32 crc, struct page *page,
1565 unsigned int page_offset,
1566 unsigned int length)
1567{
1568 char *kaddr;
1569
1570 kaddr = kmap(page);
1571 BUG_ON(kaddr == NULL);
1572 crc = crc32c(crc, kaddr + page_offset, length);
1573 kunmap(page);
1574
1575 return crc;
1576}
31b8006e
SW		 1577/*
1578 * Write as much message data payload as we can. If we finish, queue
1579 * up the footer.
1580 * 1 -> done, footer is now queued in out_kvec[].
1581 * 0 -> socket full, but more to do
1582 * <0 -> error
1583 */
34d2d200 1584static int write_partial_message_data(struct ceph_connection *con)
31b8006e
SW		 1585{
1586 struct ceph_msg *msg = con->out_msg;
8ae4f4f5 1587 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 1588 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
f5db90bc 1589 u32 crc;
31b8006e 1590
859a35d5 1591 dout("%s %p msg %p\n", __func__, con, msg);
31b8006e 1592
5240d9f9 1593 if (list_empty(&msg->data))
4c59b4a2
AE		 1594 return -EINVAL;
1595
5821bd8c
AE		 1596 /*
1597 * Iterate through each page that contains data to be
1598 * written, and send as much as possible for each.
1599 *
1600 * If we are calculating the data crc (the default), we will
1601 * need to map the page. If we have no pages, they have
1602 * been revoked, so use the zero page.
1603 */
f5db90bc 1604 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
45a267db 1605 while (cursor->total_resid) {
8a166d05 1606 struct page *page;
e387d525
AE		 1607 size_t page_offset;
1608 size_t length;
8a166d05 1609 bool last_piece;
f5db90bc 1610 int ret;
68b4476b 1611
45a267db
ID		 1612 if (!cursor->resid) {
1613 ceph_msg_data_advance(cursor, 0);
1614 continue;
1615 }
1616
343128ce
SB		 1617 page = ceph_msg_data_next(cursor, &page_offset, &length,
1618 &last_piece);
e387d525 1619 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
c2cfa194 1620 length, !last_piece);
f5db90bc
AE		 1621 if (ret <= 0) {
1622 if (do_datacrc)
1623 msg->footer.data_crc = cpu_to_le32(crc);
31b8006e 1624
f5db90bc
AE		 1625 return ret;
1626 }
143334ff
AE		 1627 if (do_datacrc && cursor->need_crc)
1628 crc = ceph_crc32c_page(crc, page, page_offset, length);
1759f7b0 1629 ceph_msg_data_advance(cursor, (size_t)ret);
31b8006e
SW		 1630 }
1631
34d2d200 1632 dout("%s %p msg %p done\n", __func__, con, msg);
31b8006e
SW		 1633
1634 /* prepare and queue up footer, too */
f5db90bc
AE		 1635 if (do_datacrc)
1636 msg->footer.data_crc = cpu_to_le32(crc);
1637 else
84ca8fc8 1638 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
e2200423 1639 con_out_kvec_reset(con);
859eb799 1640 prepare_write_message_footer(con);
f5db90bc
AE		 1641
1642 return 1; /* must return > 0 to indicate success */
31b8006e
SW		 1643}
1644
1645/*
1646 * write some zeros
1647 */
1648static int write_partial_skip(struct ceph_connection *con)
1649{
1650 int ret;
1651
67645d76 1652 dout("%s %p %d left\n", __func__, con, con->out_skip);
31b8006e 1653 while (con->out_skip > 0) {
09cbfeaf 1654 size_t size = min(con->out_skip, (int) PAGE_SIZE);
31b8006e 1655
e1dcb128 1656 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
31b8006e
SW		 1657 if (ret <= 0)
1658 goto out;
1659 con->out_skip -= ret;
1660 }
1661 ret = 1;
1662out:
1663 return ret;
1664}
1665
1666/*
1667 * Prepare to read connection handshake, or an ack.
1668 */
eed0ef2c
SW		 1669static void prepare_read_banner(struct ceph_connection *con)
1670{
1671 dout("prepare_read_banner %p\n", con);
1672 con->in_base_pos = 0;
1673}
1674
31b8006e
SW		 1675static void prepare_read_connect(struct ceph_connection *con)
1676{
1677 dout("prepare_read_connect %p\n", con);
1678 con->in_base_pos = 0;
1679}
1680
1681static void prepare_read_ack(struct ceph_connection *con)
1682{
1683 dout("prepare_read_ack %p\n", con);
1684 con->in_base_pos = 0;
1685}
1686
3a23083b
SW		 1687static void prepare_read_seq(struct ceph_connection *con)
1688{
1689 dout("prepare_read_seq %p\n", con);
1690 con->in_base_pos = 0;
1691 con->in_tag = CEPH_MSGR_TAG_SEQ;
1692}
1693
31b8006e
SW		 1694static void prepare_read_tag(struct ceph_connection *con)
1695{
1696 dout("prepare_read_tag %p\n", con);
1697 con->in_base_pos = 0;
1698 con->in_tag = CEPH_MSGR_TAG_READY;
1699}
1700
8b9558aa
YZ		 1701static void prepare_read_keepalive_ack(struct ceph_connection *con)
1702{
1703 dout("prepare_read_keepalive_ack %p\n", con);
1704 con->in_base_pos = 0;
1705}
1706
31b8006e
SW		 1707/*
1708 * Prepare to read a message.
1709 */
1710static int prepare_read_message(struct ceph_connection *con)
1711{
1712 dout("prepare_read_message %p\n", con);
1713 BUG_ON(con->in_msg != NULL);
1714 con->in_base_pos = 0;
1715 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1716 return 0;
1717}
1718
1719
1720static int read_partial(struct ceph_connection *con,
fd51653f 1721 int end, int size, void *object)
31b8006e 1722{
e6cee71f
AE		 1723 while (con->in_base_pos < end) {
1724 int left = end - con->in_base_pos;
31b8006e
SW		 1725 int have = size - left;
1726 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1727 if (ret <= 0)
1728 return ret;
1729 con->in_base_pos += ret;
1730 }
1731 return 1;
1732}
1733
1734
1735/*
1736 * Read all or part of the connect-side handshake on a new connection
1737 */
eed0ef2c 1738static int read_partial_banner(struct ceph_connection *con)
31b8006e 1739{
fd51653f
AE		 1740 int size;
1741 int end;
1742 int ret;
31b8006e 1743
eed0ef2c 1744 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 1745
1746 /* peer's banner */
fd51653f
AE		 1747 size = strlen(CEPH_BANNER);
1748 end = size;
1749 ret = read_partial(con, end, size, con->in_banner);
31b8006e
SW		 1750 if (ret <= 0)
1751 goto out;
fd51653f
AE		 1752
1753 size = sizeof (con->actual_peer_addr);
1754 end += size;
1755 ret = read_partial(con, end, size, &con->actual_peer_addr);
31b8006e
SW		 1756 if (ret <= 0)
1757 goto out;
fd51653f
AE		 1758
1759 size = sizeof (con->peer_addr_for_me);
1760 end += size;
1761 ret = read_partial(con, end, size, &con->peer_addr_for_me);
31b8006e
SW		 1762 if (ret <= 0)
1763 goto out;
fd51653f 1764
eed0ef2c
SW		 1765out:
1766 return ret;
1767}
1768
1769static int read_partial_connect(struct ceph_connection *con)
1770{
fd51653f
AE		 1771 int size;
1772 int end;
1773 int ret;
eed0ef2c
SW		 1774
1775 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1776
fd51653f
AE		 1777 size = sizeof (con->in_reply);
1778 end = size;
1779 ret = read_partial(con, end, size, &con->in_reply);
31b8006e
SW		 1780 if (ret <= 0)
1781 goto out;
fd51653f 1782
262614c4
ID		 1783 if (con->auth) {
1784 size = le32_to_cpu(con->in_reply.authorizer_len);
1785 end += size;
1786 ret = read_partial(con, end, size,
1787 con->auth->authorizer_reply_buf);
1788 if (ret <= 0)
1789 goto out;
1790 }
31b8006e 1791
4e7a5dcd
SW		 1792 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1793 con, (int)con->in_reply.tag,
1794 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 1795 le32_to_cpu(con->in_reply.global_seq));
1796out:
1797 return ret;
1798}
1799
1800/*
1801 * Verify the hello banner looks okay.
1802 */
1803static int verify_hello(struct ceph_connection *con)
1804{
1805 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 1806 pr_err("connect to %s got bad banner\n",
3d14c5d2 1807 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e
SW		 1808 con->error_msg = "protocol error, bad banner";
1809 return -1;
1810 }
1811 return 0;
1812}
1813
1814static bool addr_is_blank(struct sockaddr_storage *ss)
1815{
c44bd69c
ID		 1816 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1817 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1818
31b8006e
SW		 1819 switch (ss->ss_family) {
1820 case AF_INET:
c44bd69c 1821 return addr->s_addr == htonl(INADDR_ANY);
31b8006e 1822 case AF_INET6:
c44bd69c
ID		 1823 return ipv6_addr_any(addr6);
1824 default:
1825 return true;
31b8006e 1826 }
31b8006e
SW		 1827}
1828
1829static int addr_port(struct sockaddr_storage *ss)
1830{
1831 switch (ss->ss_family) {
1832 case AF_INET:
f28bcfbe 1833 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 1834 case AF_INET6:
f28bcfbe 1835 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 1836 }
1837 return 0;
1838}
1839
1840static void addr_set_port(struct sockaddr_storage *ss, int p)
1841{
1842 switch (ss->ss_family) {
1843 case AF_INET:
1844 ((struct sockaddr_in *)ss)->sin_port = htons(p);
a2a79609 1845 break;
31b8006e
SW		 1846 case AF_INET6:
1847 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
a2a79609 1848 break;
31b8006e
SW		 1849 }
1850}
1851
ee3b56f2
NW		 1852/*
1853 * Unlike other *_pton function semantics, zero indicates success.
1854 */
1855static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1856 char delim, const char **ipend)
1857{
99f0f3b2
AE		 1858 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1859 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
ee3b56f2
NW		 1860
1861 memset(ss, 0, sizeof(*ss));
1862
1863 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1864 ss->ss_family = AF_INET;
1865 return 0;
1866 }
1867
1868 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1869 ss->ss_family = AF_INET6;
1870 return 0;
1871 }
1872
1873 return -EINVAL;
1874}
1875
1876/*
1877 * Extract hostname string and resolve using kernel DNS facility.
1878 */
1879#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1880static int ceph_dns_resolve_name(const char *name, size_t namelen,
1881 struct sockaddr_storage *ss, char delim, const char **ipend)
1882{
1883 const char *end, *delim_p;
1884 char *colon_p, *ip_addr = NULL;
1885 int ip_len, ret;
1886
1887 /*
1888 * The end of the hostname occurs immediately preceding the delimiter or
1889 * the port marker (':') where the delimiter takes precedence.
1890 */
1891 delim_p = memchr(name, delim, namelen);
1892 colon_p = memchr(name, ':', namelen);
1893
1894 if (delim_p && colon_p)
1895 end = delim_p < colon_p ? delim_p : colon_p;
1896 else if (!delim_p && colon_p)
1897 end = colon_p;
1898 else {
1899 end = delim_p;
1900 if (!end) /* case: hostname:/ */
1901 end = name + namelen;
1902 }
1903
1904 if (end <= name)
1905 return -EINVAL;
1906
1907 /* do dns_resolve upcall */
1908 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1909 if (ip_len > 0)
1910 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1911 else
1912 ret = -ESRCH;
1913
1914 kfree(ip_addr);
1915
1916 *ipend = end;
1917
1918 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1919 ret, ret ? "failed" : ceph_pr_addr(ss));
1920
1921 return ret;
1922}
1923#else
1924static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1925 struct sockaddr_storage *ss, char delim, const char **ipend)
1926{
1927 return -EINVAL;
1928}
1929#endif
1930
1931/*
1932 * Parse a server name (IP or hostname). If a valid IP address is not found
1933 * then try to extract a hostname to resolve using userspace DNS upcall.
1934 */
1935static int ceph_parse_server_name(const char *name, size_t namelen,
1936 struct sockaddr_storage *ss, char delim, const char **ipend)
1937{
1938 int ret;
1939
1940 ret = ceph_pton(name, namelen, ss, delim, ipend);
1941 if (ret)
1942 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1943
1944 return ret;
1945}
1946
31b8006e
SW		 1947/*
1948 * Parse an ip[:port] list into an addr array. Use the default
1949 * monitor port if a port isn't specified.
1950 */
1951int ceph_parse_ips(const char *c, const char *end,
1952 struct ceph_entity_addr *addr,
1953 int max_count, int *count)
1954{
ee3b56f2 1955 int i, ret = -EINVAL;
31b8006e
SW		 1956 const char *p = c;
1957
1958 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1959 for (i = 0; i < max_count; i++) {
1960 const char *ipend;
1961 struct sockaddr_storage *ss = &addr[i].in_addr;
31b8006e 1962 int port;
39139f64
SW		 1963 char delim = ',';
1964
1965 if (*p == '[') {
1966 delim = ']';
1967 p++;
1968 }
31b8006e 1969
ee3b56f2
NW		 1970 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1971 if (ret)
31b8006e 1972 goto bad;
ee3b56f2
NW		 1973 ret = -EINVAL;
1974
31b8006e
SW		 1975 p = ipend;
1976
39139f64
SW		 1977 if (delim == ']') {
1978 if (*p != ']') {
1979 dout("missing matching ']'\n");
1980 goto bad;
1981 }
1982 p++;
1983 }
1984
31b8006e
SW		 1985 /* port? */
1986 if (p < end && *p == ':') {
1987 port = 0;
1988 p++;
1989 while (p < end && *p >= '0' && *p <= '9') {
1990 port = (port * 10) + (*p - '0');
1991 p++;
1992 }
f48db1e9
ID		 1993 if (port == 0)
1994 port = CEPH_MON_PORT;
1995 else if (port > 65535)
31b8006e
SW		 1996 goto bad;
1997 } else {
1998 port = CEPH_MON_PORT;
1999 }
2000
2001 addr_set_port(ss, port);
2002
3d14c5d2 2003 dout("parse_ips got %s\n", ceph_pr_addr(ss));
31b8006e
SW		 2004
2005 if (p == end)
2006 break;
2007 if (*p != ',')
2008 goto bad;
2009 p++;
2010 }
2011
2012 if (p != end)
2013 goto bad;
2014
2015 if (count)
2016 *count = i + 1;
2017 return 0;
2018
2019bad:
39139f64 2020 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
ee3b56f2 2021 return ret;
31b8006e 2022}
3d14c5d2 2023EXPORT_SYMBOL(ceph_parse_ips);
31b8006e 2024
eed0ef2c 2025static int process_banner(struct ceph_connection *con)
31b8006e 2026{
eed0ef2c 2027 dout("process_banner on %p\n", con);
31b8006e
SW		 2028
2029 if (verify_hello(con) < 0)
2030 return -1;
2031
63f2d211
SW		 2032 ceph_decode_addr(&con->actual_peer_addr);
2033 ceph_decode_addr(&con->peer_addr_for_me);
2034
31b8006e
SW		 2035 /*
2036 * Make sure the other end is who we wanted. note that the other
2037 * end may not yet know their ip address, so if it's 0.0.0.0, give
2038 * them the benefit of the doubt.
2039 */
103e2d3a
SW		 2040 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2041 sizeof(con->peer_addr)) != 0 &&
31b8006e
SW		 2042 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2043 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
b9a67899
JP		 2044 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2045 ceph_pr_addr(&con->peer_addr.in_addr),
2046 (int)le32_to_cpu(con->peer_addr.nonce),
2047 ceph_pr_addr(&con->actual_peer_addr.in_addr),
2048 (int)le32_to_cpu(con->actual_peer_addr.nonce));
58bb3b37 2049 con->error_msg = "wrong peer at address";
31b8006e
SW		 2050 return -1;
2051 }
2052
2053 /*
2054 * did we learn our address?
2055 */
2056 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2057 int port = addr_port(&con->msgr->inst.addr.in_addr);
2058
2059 memcpy(&con->msgr->inst.addr.in_addr,
2060 &con->peer_addr_for_me.in_addr,
2061 sizeof(con->peer_addr_for_me.in_addr));
2062 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 2063 encode_my_addr(con->msgr);
eed0ef2c 2064 dout("process_banner learned my addr is %s\n",
3d14c5d2 2065 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
31b8006e
SW		 2066 }
2067
eed0ef2c
SW		 2068 return 0;
2069}
2070
2071static int process_connect(struct ceph_connection *con)
2072{
859bff51
ID		 2073 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2074 u64 req_feat = from_msgr(con->msgr)->required_features;
dcbbd97c 2075 u64 server_feat = le64_to_cpu(con->in_reply.features);
0da5d703 2076 int ret;
04a419f9 2077
eed0ef2c
SW		 2078 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2079
262614c4 2080 if (con->auth) {
5c056fdc
ID		 2081 /*
2082 * Any connection that defines ->get_authorizer()
2083 * should also define ->verify_authorizer_reply().
2084 * See get_connect_authorizer().
2085 */
0dde5848 2086 ret = con->ops->verify_authorizer_reply(con);
5c056fdc
ID		 2087 if (ret < 0) {
2088 con->error_msg = "bad authorize reply";
2089 return ret;
2090 }
2091 }
2092
31b8006e 2093 switch (con->in_reply.tag) {
04a419f9
SW		 2094 case CEPH_MSGR_TAG_FEATURES:
2095 pr_err("%s%lld %s feature set mismatch,"
2096 " my %llx < server's %llx, missing %llx\n",
2097 ENTITY_NAME(con->peer_name),
3d14c5d2 2098 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2099 sup_feat, server_feat, server_feat & ~sup_feat);
2100 con->error_msg = "missing required protocol features";
0fa6ebc6 2101 reset_connection(con);
04a419f9
SW		 2102 return -1;
2103
31b8006e 2104 case CEPH_MSGR_TAG_BADPROTOVER:
31b8006e
SW		 2105 pr_err("%s%lld %s protocol version mismatch,"
2106 " my %d != server's %d\n",
2107 ENTITY_NAME(con->peer_name),
3d14c5d2 2108 ceph_pr_addr(&con->peer_addr.in_addr),
31b8006e
SW		 2109 le32_to_cpu(con->out_connect.protocol_version),
2110 le32_to_cpu(con->in_reply.protocol_version));
2111 con->error_msg = "protocol version mismatch";
0fa6ebc6 2112 reset_connection(con);
31b8006e
SW		 2113 return -1;
2114
4e7a5dcd
SW		 2115 case CEPH_MSGR_TAG_BADAUTHORIZER:
2116 con->auth_retry++;
2117 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2118 con->auth_retry);
2119 if (con->auth_retry == 2) {
2120 con->error_msg = "connect authorization failure";
4e7a5dcd
SW		 2121 return -1;
2122 }
6d4221b5 2123 con_out_kvec_reset(con);
e825a66d 2124 ret = prepare_write_connect(con);
0da5d703
SW		 2125 if (ret < 0)
2126 return ret;
63733a0f 2127 prepare_read_connect(con);
4e7a5dcd 2128 break;
31b8006e
SW		 2129
2130 case CEPH_MSGR_TAG_RESETSESSION:
2131 /*
2132 * If we connected with a large connect_seq but the peer
2133 * has no record of a session with us (no connection, or
2134 * connect_seq == 0), they will send RESETSESION to indicate
2135 * that they must have reset their session, and may have
2136 * dropped messages.
2137 */
2138 dout("process_connect got RESET peer seq %u\n",
5bdca4e0 2139 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 2140 pr_err("%s%lld %s connection reset\n",
2141 ENTITY_NAME(con->peer_name),
3d14c5d2 2142 ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2143 reset_connection(con);
6d4221b5 2144 con_out_kvec_reset(con);
5a0f8fdd
AE		 2145 ret = prepare_write_connect(con);
2146 if (ret < 0)
2147 return ret;
31b8006e
SW		 2148 prepare_read_connect(con);
2149
2150 /* Tell ceph about it. */
ec302645 2151 mutex_unlock(&con->mutex);
31b8006e
SW		 2152 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2153 if (con->ops->peer_reset)
2154 con->ops->peer_reset(con);
ec302645 2155 mutex_lock(&con->mutex);
8dacc7da 2156 if (con->state != CON_STATE_NEGOTIATING)
0da5d703 2157 return -EAGAIN;
31b8006e
SW		 2158 break;
2159
2160 case CEPH_MSGR_TAG_RETRY_SESSION:
2161 /*
2162 * If we sent a smaller connect_seq than the peer has, try
2163 * again with a larger value.
2164 */
5bdca4e0 2165 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
31b8006e 2166 le32_to_cpu(con->out_connect.connect_seq),
5bdca4e0
SW		 2167 le32_to_cpu(con->in_reply.connect_seq));
2168 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
6d4221b5 2169 con_out_kvec_reset(con);
5a0f8fdd
AE		 2170 ret = prepare_write_connect(con);
2171 if (ret < 0)
2172 return ret;
31b8006e
SW		 2173 prepare_read_connect(con);
2174 break;
2175
2176 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2177 /*
2178 * If we sent a smaller global_seq than the peer has, try
2179 * again with a larger value.
2180 */
eed0ef2c 2181 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e 2182 con->peer_global_seq,
5bdca4e0 2183 le32_to_cpu(con->in_reply.global_seq));
31b8006e 2184 get_global_seq(con->msgr,
5bdca4e0 2185 le32_to_cpu(con->in_reply.global_seq));
6d4221b5 2186 con_out_kvec_reset(con);
5a0f8fdd
AE		 2187 ret = prepare_write_connect(con);
2188 if (ret < 0)
2189 return ret;
31b8006e
SW		 2190 prepare_read_connect(con);
2191 break;
2192
3a23083b 2193 case CEPH_MSGR_TAG_SEQ:
31b8006e 2194 case CEPH_MSGR_TAG_READY:
04a419f9
SW		 2195 if (req_feat & ~server_feat) {
2196 pr_err("%s%lld %s protocol feature mismatch,"
2197 " my required %llx > server's %llx, need %llx\n",
2198 ENTITY_NAME(con->peer_name),
3d14c5d2 2199 ceph_pr_addr(&con->peer_addr.in_addr),
04a419f9
SW		 2200 req_feat, server_feat, req_feat & ~server_feat);
2201 con->error_msg = "missing required protocol features";
0fa6ebc6 2202 reset_connection(con);
04a419f9
SW		 2203 return -1;
2204 }
8dacc7da 2205
122070a2 2206 WARN_ON(con->state != CON_STATE_NEGOTIATING);
8dacc7da 2207 con->state = CON_STATE_OPEN;
20e55c4c 2208 con->auth_retry = 0; /* we authenticated; clear flag */
31b8006e
SW		 2209 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2210 con->connect_seq++;
aba558e2 2211 con->peer_features = server_feat;
31b8006e
SW		 2212 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2213 con->peer_global_seq,
2214 le32_to_cpu(con->in_reply.connect_seq),
2215 con->connect_seq);
2216 WARN_ON(con->connect_seq !=
2217 le32_to_cpu(con->in_reply.connect_seq));
92ac41d0
SW		 2218
2219 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
c9ffc77a 2220 con_flag_set(con, CON_FLAG_LOSSYTX);
92ac41d0 2221
85effe18 2222 con->delay = 0; /* reset backoff memory */
92ac41d0 2223
3a23083b
SW		 2224 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2225 prepare_write_seq(con);
2226 prepare_read_seq(con);
2227 } else {
2228 prepare_read_tag(con);
2229 }
31b8006e
SW		 2230 break;
2231
2232 case CEPH_MSGR_TAG_WAIT:
2233 /*
2234 * If there is a connection race (we are opening
2235 * connections to each other), one of us may just have
2236 * to WAIT. This shouldn't happen if we are the
2237 * client.
2238 */
04177882
SW		 2239 con->error_msg = "protocol error, got WAIT as client";
2240 return -1;
31b8006e
SW		 2241
2242 default:
31b8006e
SW		 2243 con->error_msg = "protocol error, garbage tag during connect";
2244 return -1;
2245 }
2246 return 0;
2247}
2248
2249
2250/*
2251 * read (part of) an ack
2252 */
2253static int read_partial_ack(struct ceph_connection *con)
2254{
fd51653f
AE		 2255 int size = sizeof (con->in_temp_ack);
2256 int end = size;
31b8006e 2257
fd51653f 2258 return read_partial(con, end, size, &con->in_temp_ack);
31b8006e
SW		 2259}
2260
31b8006e
SW		 2261/*
2262 * We can finally discard anything that's been acked.
2263 */
2264static void process_ack(struct ceph_connection *con)
2265{
2266 struct ceph_msg *m;
2267 u64 ack = le64_to_cpu(con->in_temp_ack);
2268 u64 seq;
0a2ad541
YZ		 2269 bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2270 struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
31b8006e 2271
0a2ad541
YZ		 2272 /*
2273 * In the reconnect case, con_fault() has requeued messages
2274 * in out_sent. We should cleanup old messages according to
2275 * the reconnect seq.
2276 */
2277 while (!list_empty(list)) {
2278 m = list_first_entry(list, struct ceph_msg, list_head);
2279 if (reconnect && m->needs_out_seq)
2280 break;
31b8006e
SW		 2281 seq = le64_to_cpu(m->hdr.seq);
2282 if (seq > ack)
2283 break;
2284 dout("got ack for seq %llu type %d at %p\n", seq,
2285 le16_to_cpu(m->hdr.type), m);
4cf9d544 2286 m->ack_stamp = jiffies;
31b8006e
SW		 2287 ceph_msg_remove(m);
2288 }
0a2ad541 2289
31b8006e
SW		 2290 prepare_read_tag(con);
2291}
2292
2293
2450418c 2294static int read_partial_message_section(struct ceph_connection *con,
213c99ee
SW		 2295 struct kvec *section,
2296 unsigned int sec_len, u32 *crc)
2450418c 2297{
68b4476b 2298 int ret, left;
2450418c
YS		 2299
2300 BUG_ON(!section);
2301
2302 while (section->iov_len < sec_len) {
2303 BUG_ON(section->iov_base == NULL);
2304 left = sec_len - section->iov_len;
2305 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2306 section->iov_len, left);
2307 if (ret <= 0)
2308 return ret;
2309 section->iov_len += ret;
2450418c 2310 }
fe3ad593
AE		 2311 if (section->iov_len == sec_len)
2312 *crc = crc32c(0, section->iov_base, section->iov_len);
31b8006e 2313
2450418c
YS		 2314 return 1;
2315}
31b8006e 2316
34d2d200
AE		 2317static int read_partial_msg_data(struct ceph_connection *con)
2318{
2319 struct ceph_msg *msg = con->in_msg;
8ae4f4f5 2320 struct ceph_msg_data_cursor *cursor = &msg->cursor;
859bff51 2321 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
686be208
AE		 2322 struct page *page;
2323 size_t page_offset;
2324 size_t length;
f5db90bc 2325 u32 crc = 0;
34d2d200
AE		 2326 int ret;
2327
2328 BUG_ON(!msg);
5240d9f9 2329 if (list_empty(&msg->data))
4c59b4a2 2330 return -EIO;
34d2d200 2331
f5db90bc
AE		 2332 if (do_datacrc)
2333 crc = con->in_data_crc;
45a267db
ID		 2334 while (cursor->total_resid) {
2335 if (!cursor->resid) {
2336 ceph_msg_data_advance(cursor, 0);
2337 continue;
2338 }
2339
343128ce 2340 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
686be208 2341 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
f5db90bc
AE		 2342 if (ret <= 0) {
2343 if (do_datacrc)
2344 con->in_data_crc = crc;
2345
686be208 2346 return ret;
f5db90bc 2347 }
686be208
AE		 2348
2349 if (do_datacrc)
f5db90bc 2350 crc = ceph_crc32c_page(crc, page, page_offset, ret);
1759f7b0 2351 ceph_msg_data_advance(cursor, (size_t)ret);
34d2d200 2352 }
f5db90bc
AE		 2353 if (do_datacrc)
2354 con->in_data_crc = crc;
34d2d200
AE		 2355
2356 return 1; /* must return > 0 to indicate success */
2357}
2358
31b8006e
SW		 2359/*
2360 * read (part of) a message.
2361 */
686be208
AE		 2362static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2363
31b8006e
SW		 2364static int read_partial_message(struct ceph_connection *con)
2365{
2366 struct ceph_msg *m = con->in_msg;
fd51653f
AE		 2367 int size;
2368 int end;
31b8006e 2369 int ret;
95c96174 2370 unsigned int front_len, middle_len, data_len;
859bff51 2371 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
33d07337 2372 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
ae18756b 2373 u64 seq;
fe3ad593 2374 u32 crc;
31b8006e
SW		 2375
2376 dout("read_partial_message con %p msg %p\n", con, m);
2377
2378 /* header */
fd51653f
AE		 2379 size = sizeof (con->in_hdr);
2380 end = size;
2381 ret = read_partial(con, end, size, &con->in_hdr);
57dac9d1
AE		 2382 if (ret <= 0)
2383 return ret;
fe3ad593
AE		 2384
2385 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2386 if (cpu_to_le32(crc) != con->in_hdr.crc) {
67c64eb7 2387 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
fe3ad593
AE		 2388 crc, con->in_hdr.crc);
2389 return -EBADMSG;
2390 }
2391
31b8006e
SW		 2392 front_len = le32_to_cpu(con->in_hdr.front_len);
2393 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2394 return -EIO;
2395 middle_len = le32_to_cpu(con->in_hdr.middle_len);
7b11ba37 2396 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
31b8006e
SW		 2397 return -EIO;
2398 data_len = le32_to_cpu(con->in_hdr.data_len);
2399 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2400 return -EIO;
2401
ae18756b
SW		 2402 /* verify seq# */
2403 seq = le64_to_cpu(con->in_hdr.seq);
2404 if ((s64)seq - (s64)con->in_seq < 1) {
df9f86fa 2405 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
ae18756b 2406 ENTITY_NAME(con->peer_name),
3d14c5d2 2407 ceph_pr_addr(&con->peer_addr.in_addr),
ae18756b
SW		 2408 seq, con->in_seq + 1);
2409 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2410 sizeof_footer(con);
ae18756b 2411 con->in_tag = CEPH_MSGR_TAG_READY;
e7a88e82 2412 return 1;
ae18756b
SW		 2413 } else if ((s64)seq - (s64)con->in_seq > 1) {
2414 pr_err("read_partial_message bad seq %lld expected %lld\n",
2415 seq, con->in_seq + 1);
2416 con->error_msg = "bad message sequence # for incoming message";
67c64eb7 2417 return -EBADE;
ae18756b
SW		 2418 }
2419
31b8006e
SW		 2420 /* allocate message? */
2421 if (!con->in_msg) {
4740a623
SW		 2422 int skip = 0;
2423
31b8006e 2424 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
6ebc8b32 2425 front_len, data_len);
4740a623
SW		 2426 ret = ceph_con_in_msg_alloc(con, &skip);
2427 if (ret < 0)
2428 return ret;
f759ebb9
AE		 2429
2430 BUG_ON(!con->in_msg ^ skip);
2450418c 2431 if (skip) {
31b8006e 2432 /* skip this message */
a79832f2 2433 dout("alloc_msg said skip message\n");
31b8006e 2434 con->in_base_pos = -front_len - middle_len - data_len -
dbc0d3ca 2435 sizeof_footer(con);
31b8006e 2436 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 2437 con->in_seq++;
e7a88e82 2438 return 1;
31b8006e 2439 }
38941f80 2440
4740a623 2441 BUG_ON(!con->in_msg);
38941f80 2442 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2443 m = con->in_msg;
2444 m->front.iov_len = 0; /* haven't read it yet */
2450418c
YS		 2445 if (m->middle)
2446 m->middle->vec.iov_len = 0;
9d7f0f13 2447
78625051 2448 /* prepare for data payload, if any */
a4107026 2449
78625051 2450 if (data_len)
98fa5dd8 2451 prepare_message_data(con->in_msg, data_len);
31b8006e
SW		 2452 }
2453
2454 /* front */
2450418c
YS		 2455 ret = read_partial_message_section(con, &m->front, front_len,
2456 &con->in_front_crc);
2457 if (ret <= 0)
2458 return ret;
31b8006e
SW		 2459
2460 /* middle */
2450418c 2461 if (m->middle) {
213c99ee
SW		 2462 ret = read_partial_message_section(con, &m->middle->vec,
2463 middle_len,
2450418c 2464 &con->in_middle_crc);
31b8006e
SW		 2465 if (ret <= 0)
2466 return ret;
31b8006e
SW		 2467 }
2468
2469 /* (page) data */
34d2d200
AE		 2470 if (data_len) {
2471 ret = read_partial_msg_data(con);
2472 if (ret <= 0)
2473 return ret;
31b8006e
SW		 2474 }
2475
31b8006e 2476 /* footer */
89f08173 2477 size = sizeof_footer(con);
fd51653f
AE		 2478 end += size;
2479 ret = read_partial(con, end, size, &m->footer);
57dac9d1
AE		 2480 if (ret <= 0)
2481 return ret;
2482
33d07337
YZ		 2483 if (!need_sign) {
2484 m->footer.flags = m->old_footer.flags;
2485 m->footer.sig = 0;
2486 }
2487
31b8006e
SW		 2488 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2489 m, front_len, m->footer.front_crc, middle_len,
2490 m->footer.middle_crc, data_len, m->footer.data_crc);
2491
2492 /* crc ok? */
2493 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2494 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2495 m, con->in_front_crc, m->footer.front_crc);
2496 return -EBADMSG;
2497 }
2498 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2499 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2500 m, con->in_middle_crc, m->footer.middle_crc);
2501 return -EBADMSG;
2502 }
bca064d2 2503 if (do_datacrc &&
31b8006e
SW		 2504 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2505 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2506 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2507 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2508 return -EBADMSG;
2509 }
2510
33d07337 2511 if (need_sign && con->ops->check_message_signature &&
79dbd1ba 2512 con->ops->check_message_signature(m)) {
33d07337
YZ		 2513 pr_err("read_partial_message %p signature check failed\n", m);
2514 return -EBADMSG;
2515 }
2516
31b8006e
SW		 2517 return 1; /* done! */
2518}
2519
2520/*
2521 * Process message. This happens in the worker thread. The callback should
2522 * be careful not to do anything that waits on other incoming messages or it
2523 * may deadlock.
2524 */
2525static void process_message(struct ceph_connection *con)
2526{
583d0fef 2527 struct ceph_msg *msg = con->in_msg;
31b8006e 2528
38941f80 2529 BUG_ON(con->in_msg->con != con);
31b8006e
SW		 2530 con->in_msg = NULL;
2531
2532 /* if first message, set peer_name */
2533 if (con->peer_name.type == 0)
dbad185d 2534 con->peer_name = msg->hdr.src;
31b8006e 2535
31b8006e 2536 con->in_seq++;
ec302645 2537 mutex_unlock(&con->mutex);
31b8006e
SW		 2538
2539 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2540 msg, le64_to_cpu(msg->hdr.seq),
dbad185d 2541 ENTITY_NAME(msg->hdr.src),
31b8006e
SW		 2542 le16_to_cpu(msg->hdr.type),
2543 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2544 le32_to_cpu(msg->hdr.front_len),
2545 le32_to_cpu(msg->hdr.data_len),
2546 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2547 con->ops->dispatch(con, msg);
ec302645
SW		 2548
2549 mutex_lock(&con->mutex);
31b8006e
SW		 2550}
2551
8b9558aa
YZ		 2552static int read_keepalive_ack(struct ceph_connection *con)
2553{
2554 struct ceph_timespec ceph_ts;
2555 size_t size = sizeof(ceph_ts);
2556 int ret = read_partial(con, size, size, &ceph_ts);
2557 if (ret <= 0)
2558 return ret;
473bd2d7 2559 ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
8b9558aa
YZ		 2560 prepare_read_tag(con);
2561 return 1;
2562}
31b8006e
SW		 2563
2564/*
2565 * Write something to the socket. Called in a worker thread when the
2566 * socket appears to be writeable and we have something ready to send.
2567 */
2568static int try_write(struct ceph_connection *con)
2569{
31b8006e
SW		 2570 int ret = 1;
2571
d59315ca 2572 dout("try_write start %p state %lu\n", con, con->state);
9c55ad1c
ID		 2573 if (con->state != CON_STATE_PREOPEN &&
2574 con->state != CON_STATE_CONNECTING &&
2575 con->state != CON_STATE_NEGOTIATING &&
2576 con->state != CON_STATE_OPEN)
2577 return 0;
31b8006e 2578
31b8006e 2579 /* open the socket first? */
8dacc7da
SW		 2580 if (con->state == CON_STATE_PREOPEN) {
2581 BUG_ON(con->sock);
2582 con->state = CON_STATE_CONNECTING;
a5988c49 2583
e2200423 2584 con_out_kvec_reset(con);
e825a66d 2585 prepare_write_banner(con);
eed0ef2c 2586 prepare_read_banner(con);
31b8006e 2587
cf3e5c40 2588 BUG_ON(con->in_msg);
31b8006e
SW		 2589 con->in_tag = CEPH_MSGR_TAG_READY;
2590 dout("try_write initiating connect on %p new state %lu\n",
2591 con, con->state);
41617d0c
AE		 2592 ret = ceph_tcp_connect(con);
2593 if (ret < 0) {
31b8006e 2594 con->error_msg = "connect error";
31b8006e
SW		 2595 goto out;
2596 }
2597 }
2598
d2935d6f
ID		 2599more:
2600 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
9c55ad1c
ID		 2601 BUG_ON(!con->sock);
2602
31b8006e 2603 /* kvec data queued? */
67645d76
ID		 2604 if (con->out_kvec_left) {
2605 ret = write_partial_kvec(con);
31b8006e 2606 if (ret <= 0)
42961d23 2607 goto out;
31b8006e 2608 }
67645d76
ID		 2609 if (con->out_skip) {
2610 ret = write_partial_skip(con);
31b8006e 2611 if (ret <= 0)
42961d23 2612 goto out;
31b8006e
SW		 2613 }
2614
2615 /* msg pages? */
2616 if (con->out_msg) {
c86a2930
SW		 2617 if (con->out_msg_done) {
2618 ceph_msg_put(con->out_msg);
2619 con->out_msg = NULL; /* we're done with this one */
2620 goto do_next;
2621 }
2622
34d2d200 2623 ret = write_partial_message_data(con);
31b8006e 2624 if (ret == 1)
d2935d6f 2625 goto more; /* we need to send the footer, too! */
31b8006e 2626 if (ret == 0)
42961d23 2627 goto out;
31b8006e 2628 if (ret < 0) {
34d2d200 2629 dout("try_write write_partial_message_data err %d\n",
31b8006e 2630 ret);
42961d23 2631 goto out;
31b8006e
SW		 2632 }
2633 }
2634
c86a2930 2635do_next:
8dacc7da 2636 if (con->state == CON_STATE_OPEN) {
8b9558aa
YZ		 2637 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2638 prepare_write_keepalive(con);
2639 goto more;
2640 }
31b8006e
SW		 2641 /* is anything else pending? */
2642 if (!list_empty(&con->out_queue)) {
2643 prepare_write_message(con);
2644 goto more;
2645 }
2646 if (con->in_seq > con->in_seq_acked) {
2647 prepare_write_ack(con);
2648 goto more;
2649 }
31b8006e
SW		 2650 }
2651
2652 /* Nothing to do! */
c9ffc77a 2653 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
31b8006e 2654 dout("try_write nothing else to write.\n");
31b8006e
SW		 2655 ret = 0;
2656out:
42961d23 2657 dout("try_write done on %p ret %d\n", con, ret);
31b8006e
SW		 2658 return ret;
2659}
2660
31b8006e
SW		 2661/*
2662 * Read what we can from the socket.
2663 */
2664static int try_read(struct ceph_connection *con)
2665{
31b8006e
SW		 2666 int ret = -1;
2667
8dacc7da
SW		 2668more:
2669 dout("try_read start on %p state %lu\n", con, con->state);
2670 if (con->state != CON_STATE_CONNECTING &&
2671 con->state != CON_STATE_NEGOTIATING &&
2672 con->state != CON_STATE_OPEN)
31b8006e
SW		 2673 return 0;
2674
8dacc7da 2675 BUG_ON(!con->sock);
ec302645 2676
31b8006e
SW		 2677 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2678 con->in_base_pos);
0da5d703 2679
8dacc7da 2680 if (con->state == CON_STATE_CONNECTING) {
7593af92
AE		 2681 dout("try_read connecting\n");
2682 ret = read_partial_banner(con);
2683 if (ret <= 0)
ab166d5a 2684 goto out;
7593af92
AE		 2685 ret = process_banner(con);
2686 if (ret < 0)
2687 goto out;
2688
8dacc7da 2689 con->state = CON_STATE_NEGOTIATING;
7593af92 2690
6d4221b5
JS		 2691 /*
2692 * Received banner is good, exchange connection info.
2693 * Do not reset out_kvec, as sending our banner raced
2694 * with receiving peer banner after connect completed.
2695 */
7593af92
AE		 2696 ret = prepare_write_connect(con);
2697 if (ret < 0)
2698 goto out;
2699 prepare_read_connect(con);
2700
2701 /* Send connection info before awaiting response */
0da5d703
SW		 2702 goto out;
2703 }
2704
8dacc7da 2705 if (con->state == CON_STATE_NEGOTIATING) {
7593af92 2706 dout("try_read negotiating\n");
31b8006e
SW		 2707 ret = read_partial_connect(con);
2708 if (ret <= 0)
31b8006e 2709 goto out;
98bdb0aa
SW		 2710 ret = process_connect(con);
2711 if (ret < 0)
2712 goto out;
31b8006e
SW		 2713 goto more;
2714 }
2715
122070a2 2716 WARN_ON(con->state != CON_STATE_OPEN);
8dacc7da 2717
31b8006e
SW		 2718 if (con->in_base_pos < 0) {
2719 /*
2720 * skipping + discarding content.
31b8006e 2721 */
e5c93883 2722 ret = ceph_tcp_recvmsg(con->sock, NULL, -con->in_base_pos);
31b8006e 2723 if (ret <= 0)
98bdb0aa 2724 goto out;
e5c93883 2725 dout("skipped %d / %d bytes\n", ret, -con->in_base_pos);
31b8006e
SW		 2726 con->in_base_pos += ret;
2727 if (con->in_base_pos)
2728 goto more;
2729 }
2730 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2731 /*
2732 * what's next?
2733 */
2734 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2735 if (ret <= 0)
98bdb0aa 2736 goto out;
31b8006e
SW		 2737 dout("try_read got tag %d\n", (int)con->in_tag);
2738 switch (con->in_tag) {
2739 case CEPH_MSGR_TAG_MSG:
2740 prepare_read_message(con);
2741 break;
2742 case CEPH_MSGR_TAG_ACK:
2743 prepare_read_ack(con);
2744 break;
8b9558aa
YZ		 2745 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2746 prepare_read_keepalive_ack(con);
2747 break;
31b8006e 2748 case CEPH_MSGR_TAG_CLOSE:
8dacc7da
SW		 2749 con_close_socket(con);
2750 con->state = CON_STATE_CLOSED;
98bdb0aa 2751 goto out;
31b8006e
SW		 2752 default:
2753 goto bad_tag;
2754 }
2755 }
2756 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2757 ret = read_partial_message(con);
2758 if (ret <= 0) {
2759 switch (ret) {
2760 case -EBADMSG:
a51983e4 2761 con->error_msg = "bad crc/signature";
67c64eb7
ID		 2762 /* fall through */
2763 case -EBADE:
31b8006e 2764 ret = -EIO;
98bdb0aa 2765 break;
31b8006e
SW		 2766 case -EIO:
2767 con->error_msg = "io error";
98bdb0aa 2768 break;
31b8006e 2769 }
98bdb0aa 2770 goto out;
31b8006e
SW		 2771 }
2772 if (con->in_tag == CEPH_MSGR_TAG_READY)
2773 goto more;
2774 process_message(con);
7b862e07
SW		 2775 if (con->state == CON_STATE_OPEN)
2776 prepare_read_tag(con);
31b8006e
SW		 2777 goto more;
2778 }
3a23083b
SW		 2779 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2780 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2781 /*
2782 * the final handshake seq exchange is semantically
2783 * equivalent to an ACK
2784 */
31b8006e
SW		 2785 ret = read_partial_ack(con);
2786 if (ret <= 0)
98bdb0aa 2787 goto out;
31b8006e
SW		 2788 process_ack(con);
2789 goto more;
2790 }
8b9558aa
YZ		 2791 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2792 ret = read_keepalive_ack(con);
2793 if (ret <= 0)
2794 goto out;
2795 goto more;
2796 }
31b8006e 2797
31b8006e 2798out:
98bdb0aa 2799 dout("try_read done on %p ret %d\n", con, ret);
31b8006e
SW		 2800 return ret;
2801
2802bad_tag:
2803 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2804 con->error_msg = "protocol error, garbage tag";
2805 ret = -1;
2806 goto out;
2807}
2808
2809
2810/*
802c6d96
AE		 2811 * Atomically queue work on a connection after the specified delay.
2812 * Bump @con reference to avoid races with connection teardown.
2813 * Returns 0 if work was queued, or an error code otherwise.
31b8006e 2814 */
802c6d96 2815static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
31b8006e 2816{
31b8006e 2817 if (!con->ops->get(con)) {
802c6d96 2818 dout("%s %p ref count 0\n", __func__, con);
802c6d96 2819 return -ENOENT;
31b8006e
SW		 2820 }
2821
802c6d96
AE		 2822 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2823 dout("%s %p - already queued\n", __func__, con);
31b8006e 2824 con->ops->put(con);
802c6d96 2825 return -EBUSY;
31b8006e 2826 }
802c6d96
AE		 2827
2828 dout("%s %p %lu\n", __func__, con, delay);
802c6d96
AE		 2829 return 0;
2830}
2831
2832static void queue_con(struct ceph_connection *con)
2833{
2834 (void) queue_con_delay(con, 0);
31b8006e
SW		 2835}
2836
37ab77ac
ID		 2837static void cancel_con(struct ceph_connection *con)
2838{
2839 if (cancel_delayed_work(&con->work)) {
2840 dout("%s %p\n", __func__, con);
2841 con->ops->put(con);
2842 }
2843}
2844
7bb21d68
AE		 2845static bool con_sock_closed(struct ceph_connection *con)
2846{
c9ffc77a 2847 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
7bb21d68
AE		 2848 return false;
2849
2850#define CASE(x) \
2851 case CON_STATE_ ## x: \
2852 con->error_msg = "socket closed (con state " #x ")"; \
2853 break;
2854
2855 switch (con->state) {
2856 CASE(CLOSED);
2857 CASE(PREOPEN);
2858 CASE(CONNECTING);
2859 CASE(NEGOTIATING);
2860 CASE(OPEN);
2861 CASE(STANDBY);
2862 default:
b9a67899 2863 pr_warn("%s con %p unrecognized state %lu\n",
7bb21d68
AE		 2864 __func__, con, con->state);
2865 con->error_msg = "unrecognized con state";
2866 BUG();
2867 break;
2868 }
2869#undef CASE
2870
2871 return true;
2872}
2873
f20a39fd
AE		 2874static bool con_backoff(struct ceph_connection *con)
2875{
2876 int ret;
2877
2878 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2879 return false;
2880
2881 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2882 if (ret) {
2883 dout("%s: con %p FAILED to back off %lu\n", __func__,
2884 con, con->delay);
2885 BUG_ON(ret == -ENOENT);
2886 con_flag_set(con, CON_FLAG_BACKOFF);
2887 }
2888
2889 return true;
2890}
2891
93209264
AE		 2892/* Finish fault handling; con->mutex must *not* be held here */
2893
2894static void con_fault_finish(struct ceph_connection *con)
2895{
f6330cc1
ID		 2896 dout("%s %p\n", __func__, con);
2897
93209264
AE		 2898 /*
2899 * in case we faulted due to authentication, invalidate our
2900 * current tickets so that we can get new ones.
2901 */
f6330cc1
ID		 2902 if (con->auth_retry) {
2903 dout("auth_retry %d, invalidating\n", con->auth_retry);
2904 if (con->ops->invalidate_authorizer)
2905 con->ops->invalidate_authorizer(con);
2906 con->auth_retry = 0;
93209264
AE		 2907 }
2908
2909 if (con->ops->fault)
2910 con->ops->fault(con);
2911}
2912
31b8006e
SW		 2913/*
2914 * Do some work on a connection. Drop a connection ref when we're done.
2915 */
68931622 2916static void ceph_con_workfn(struct work_struct *work)
31b8006e
SW		 2917{
2918 struct ceph_connection *con = container_of(work, struct ceph_connection,
2919 work.work);
49659416 2920 bool fault;
31b8006e 2921
9dd4658d 2922 mutex_lock(&con->mutex);
49659416
AE		 2923 while (true) {
2924 int ret;
31b8006e 2925
49659416
AE		 2926 if ((fault = con_sock_closed(con))) {
2927 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2928 break;
2929 }
2930 if (con_backoff(con)) {
2931 dout("%s: con %p BACKOFF\n", __func__, con);
2932 break;
2933 }
2934 if (con->state == CON_STATE_STANDBY) {
2935 dout("%s: con %p STANDBY\n", __func__, con);
2936 break;
2937 }
2938 if (con->state == CON_STATE_CLOSED) {
2939 dout("%s: con %p CLOSED\n", __func__, con);
2940 BUG_ON(con->sock);
2941 break;
2942 }
2943 if (con->state == CON_STATE_PREOPEN) {
2944 dout("%s: con %p PREOPEN\n", __func__, con);
2945 BUG_ON(con->sock);
2946 }
0da5d703 2947
49659416
AE		 2948 ret = try_read(con);
2949 if (ret < 0) {
2950 if (ret == -EAGAIN)
2951 continue;
67c64eb7
ID		 2952 if (!con->error_msg)
2953 con->error_msg = "socket error on read";
49659416
AE		 2954 fault = true;
2955 break;
2956 }
2957
2958 ret = try_write(con);
2959 if (ret < 0) {
2960 if (ret == -EAGAIN)
2961 continue;
67c64eb7
ID		 2962 if (!con->error_msg)
2963 con->error_msg = "socket error on write";
49659416
AE		 2964 fault = true;
2965 }
2966
2967 break; /* If we make it to here, we're done */
3a140a0d 2968 }
b6e7b6a1
AE		 2969 if (fault)
2970 con_fault(con);
9dd4658d 2971 mutex_unlock(&con->mutex);
0da5d703 2972
b6e7b6a1
AE		 2973 if (fault)
2974 con_fault_finish(con);
2975
2976 con->ops->put(con);
31b8006e
SW		 2977}
2978
31b8006e
SW		 2979/*
2980 * Generic error/fault handler. A retry mechanism is used with
2981 * exponential backoff
2982 */
93209264 2983static void con_fault(struct ceph_connection *con)
31b8006e 2984{
31b8006e 2985 dout("fault %p state %lu to peer %s\n",
3d14c5d2 2986 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
31b8006e 2987
67c64eb7
ID		 2988 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2989 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2990 con->error_msg = NULL;
2991
122070a2 2992 WARN_ON(con->state != CON_STATE_CONNECTING &&
8dacc7da
SW		 2993 con->state != CON_STATE_NEGOTIATING &&
2994 con->state != CON_STATE_OPEN);
ec302645 2995
31b8006e 2996 con_close_socket(con);
5e095e8b 2997
c9ffc77a 2998 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
8dacc7da
SW		 2999 dout("fault on LOSSYTX channel, marking CLOSED\n");
3000 con->state = CON_STATE_CLOSED;
93209264 3001 return;
3b5ede07
SW		 3002 }
3003
5e095e8b 3004 if (con->in_msg) {
38941f80 3005 BUG_ON(con->in_msg->con != con);
5e095e8b
SW		 3006 ceph_msg_put(con->in_msg);
3007 con->in_msg = NULL;
3008 }
31b8006e 3009
e80a52d1
SW		 3010 /* Requeue anything that hasn't been acked */
3011 list_splice_init(&con->out_sent, &con->out_queue);
9bd2e6f8 3012
e76661d0
SW		 3013 /* If there are no messages queued or keepalive pending, place
3014 * the connection in a STANDBY state */
3015 if (list_empty(&con->out_queue) &&
c9ffc77a 3016 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
e00de341 3017 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
c9ffc77a 3018 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
8dacc7da 3019 con->state = CON_STATE_STANDBY;
e80a52d1
SW		 3020 } else {
3021 /* retry after a delay. */
8dacc7da 3022 con->state = CON_STATE_PREOPEN;
e80a52d1
SW		 3023 if (con->delay == 0)
3024 con->delay = BASE_DELAY_INTERVAL;
3025 else if (con->delay < MAX_DELAY_INTERVAL)
3026 con->delay *= 2;
c9ffc77a 3027 con_flag_set(con, CON_FLAG_BACKOFF);
8618e30b 3028 queue_con(con);
31b8006e 3029 }
31b8006e
SW		 3030}
3031
3032
3033
3034/*
15d9882c 3035 * initialize a new messenger instance
31b8006e 3036 */
15d9882c 3037void ceph_messenger_init(struct ceph_messenger *msgr,
859bff51 3038 struct ceph_entity_addr *myaddr)
31b8006e 3039{
31b8006e
SW		 3040 spin_lock_init(&msgr->global_seq_lock);
3041
31b8006e
SW		 3042 if (myaddr)
3043 msgr->inst.addr = *myaddr;
3044
3045 /* select a random nonce */
ac8839d7 3046 msgr->inst.addr.type = 0;
103e2d3a 3047 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
63f2d211 3048 encode_my_addr(msgr);
31b8006e 3049
a2a32584 3050 atomic_set(&msgr->stopping, 0);
757856d2 3051 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
31b8006e 3052
15d9882c 3053 dout("%s %p\n", __func__, msgr);
31b8006e 3054}
15d9882c 3055EXPORT_SYMBOL(ceph_messenger_init);
31b8006e 3056
757856d2
ID		 3057void ceph_messenger_fini(struct ceph_messenger *msgr)
3058{
3059 put_net(read_pnet(&msgr->net));
3060}
3061EXPORT_SYMBOL(ceph_messenger_fini);
3062
583d0fef
ID		 3063static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3064{
3065 if (msg->con)
3066 msg->con->ops->put(msg->con);
3067
3068 msg->con = con ? con->ops->get(con) : NULL;
3069 BUG_ON(msg->con != con);
3070}
3071
e00de341
SW		 3072static void clear_standby(struct ceph_connection *con)
3073{
3074 /* come back from STANDBY? */
8dacc7da 3075 if (con->state == CON_STATE_STANDBY) {
e00de341 3076 dout("clear_standby %p and ++connect_seq\n", con);
8dacc7da 3077 con->state = CON_STATE_PREOPEN;
e00de341 3078 con->connect_seq++;
c9ffc77a
AE		 3079 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3080 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
e00de341
SW		 3081 }
3082}
3083
31b8006e
SW		 3084/*
3085 * Queue up an outgoing message on the given connection.
3086 */
3087void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3088{
31b8006e 3089 /* set src+dst */
dbad185d 3090 msg->hdr.src = con->msgr->inst.name;
3ca02ef9 3091 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
e84346b7
SW		 3092 msg->needs_out_seq = true;
3093
ec302645 3094 mutex_lock(&con->mutex);
92ce034b 3095
8dacc7da 3096 if (con->state == CON_STATE_CLOSED) {
a59b55a6
SW		 3097 dout("con_send %p closed, dropping %p\n", con, msg);
3098 ceph_msg_put(msg);
3099 mutex_unlock(&con->mutex);
3100 return;
3101 }
3102
583d0fef 3103 msg_con_set(msg, con);
92ce034b 3104
31b8006e
SW		 3105 BUG_ON(!list_empty(&msg->list_head));
3106 list_add_tail(&msg->list_head, &con->out_queue);
3107 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3108 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3109 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3110 le32_to_cpu(msg->hdr.front_len),
3111 le32_to_cpu(msg->hdr.middle_len),
3112 le32_to_cpu(msg->hdr.data_len));
00650931
SW		 3113
3114 clear_standby(con);
ec302645 3115 mutex_unlock(&con->mutex);
31b8006e
SW		 3116
3117 /* if there wasn't anything waiting to send before, queue
3118 * new work */
c9ffc77a 3119 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3120 queue_con(con);
3121}
3d14c5d2 3122EXPORT_SYMBOL(ceph_con_send);
31b8006e
SW		 3123
3124/*
3125 * Revoke a message that was previously queued for send
3126 */
6740a845 3127void ceph_msg_revoke(struct ceph_msg *msg)
31b8006e 3128{
6740a845
AE		 3129 struct ceph_connection *con = msg->con;
3130
583d0fef
ID		 3131 if (!con) {
3132 dout("%s msg %p null con\n", __func__, msg);
6740a845 3133 return; /* Message not in our possession */
583d0fef 3134 }
6740a845 3135
ec302645 3136 mutex_lock(&con->mutex);
31b8006e 3137 if (!list_empty(&msg->list_head)) {
38941f80 3138 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
31b8006e 3139 list_del_init(&msg->list_head);
31b8006e 3140 msg->hdr.seq = 0;
38941f80 3141
31b8006e 3142 ceph_msg_put(msg);
ed98adad
SW		 3143 }
3144 if (con->out_msg == msg) {
67645d76
ID		 3145 BUG_ON(con->out_skip);
3146 /* footer */
3147 if (con->out_msg_done) {
3148 con->out_skip += con_out_kvec_skip(con);
3149 } else {
3150 BUG_ON(!msg->data_length);
89f08173 3151 con->out_skip += sizeof_footer(con);
31b8006e 3152 }
67645d76
ID		 3153 /* data, middle, front */
3154 if (msg->data_length)
3155 con->out_skip += msg->cursor.total_resid;
3156 if (msg->middle)
3157 con->out_skip += con_out_kvec_skip(con);
3158 con->out_skip += con_out_kvec_skip(con);
3159
3160 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3161 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
ed98adad 3162 msg->hdr.seq = 0;
67645d76 3163 con->out_msg = NULL;
92ce034b 3164 ceph_msg_put(msg);
31b8006e 3165 }
67645d76 3166
ec302645 3167 mutex_unlock(&con->mutex);
31b8006e
SW		 3168}
3169
350b1c32 3170/*
0d59ab81 3171 * Revoke a message that we may be reading data into
350b1c32 3172 */
8921d114 3173void ceph_msg_revoke_incoming(struct ceph_msg *msg)
350b1c32 3174{
583d0fef 3175 struct ceph_connection *con = msg->con;
8921d114 3176
583d0fef 3177 if (!con) {
8921d114 3178 dout("%s msg %p null con\n", __func__, msg);
8921d114
AE		 3179 return; /* Message not in our possession */
3180 }
3181
350b1c32 3182 mutex_lock(&con->mutex);
8921d114 3183 if (con->in_msg == msg) {
95c96174
ED		 3184 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3185 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3186 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
350b1c32
SW		 3187
3188 /* skip rest of message */
8921d114
AE		 3189 dout("%s %p msg %p revoked\n", __func__, con, msg);
3190 con->in_base_pos = con->in_base_pos -
350b1c32 3191 sizeof(struct ceph_msg_header) -
0d59ab81
YS		 3192 front_len -
3193 middle_len -
3194 data_len -
350b1c32 3195 sizeof(struct ceph_msg_footer);
350b1c32
SW		 3196 ceph_msg_put(con->in_msg);
3197 con->in_msg = NULL;
3198 con->in_tag = CEPH_MSGR_TAG_READY;
684be25c 3199 con->in_seq++;
350b1c32 3200 } else {
8921d114
AE		 3201 dout("%s %p in_msg %p msg %p no-op\n",
3202 __func__, con, con->in_msg, msg);
350b1c32
SW		 3203 }
3204 mutex_unlock(&con->mutex);
3205}
3206
31b8006e
SW		 3207/*
3208 * Queue a keepalive byte to ensure the tcp connection is alive.
3209 */
3210void ceph_con_keepalive(struct ceph_connection *con)
3211{
e00de341 3212 dout("con_keepalive %p\n", con);
00650931 3213 mutex_lock(&con->mutex);
e00de341 3214 clear_standby(con);
00650931 3215 mutex_unlock(&con->mutex);
c9ffc77a
AE		 3216 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3217 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
31b8006e
SW		 3218 queue_con(con);
3219}
3d14c5d2 3220EXPORT_SYMBOL(ceph_con_keepalive);
31b8006e 3221
8b9558aa
YZ		 3222bool ceph_con_keepalive_expired(struct ceph_connection *con,
3223 unsigned long interval)
3224{
3225 if (interval > 0 &&
3226 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
473bd2d7
AB		 3227 struct timespec64 now;
3228 struct timespec64 ts;
3229 ktime_get_real_ts64(&now);
3230 jiffies_to_timespec64(interval, &ts);
3231 ts = timespec64_add(con->last_keepalive_ack, ts);
3232 return timespec64_compare(&now, &ts) >= 0;
8b9558aa
YZ		 3233 }
3234 return false;
3235}
3236
6644ed7b 3237static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
43794509 3238{
6644ed7b
AE		 3239 struct ceph_msg_data *data;
3240
3241 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3242 return NULL;
3243
81b36be4 3244 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
7c40b22f
DC		 3245 if (!data)
3246 return NULL;
3247
3248 data->type = type;
5240d9f9 3249 INIT_LIST_HEAD(&data->links);
6644ed7b
AE		 3250
3251 return data;
3252}
3253
3254static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3255{
3256 if (!data)
3257 return;
3258
5240d9f9 3259 WARN_ON(!list_empty(&data->links));
e4339d28 3260 if (data->type == CEPH_MSG_DATA_PAGELIST)
6644ed7b 3261 ceph_pagelist_release(data->pagelist);
81b36be4 3262 kmem_cache_free(ceph_msg_data_cache, data);
43794509
AE		 3263}
3264
90af3602 3265void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
f1baeb2b 3266 size_t length, size_t alignment)
02afca6c 3267{
6644ed7b
AE		 3268 struct ceph_msg_data *data;
3269
07aa1558
AE		 3270 BUG_ON(!pages);
3271 BUG_ON(!length);
6644ed7b
AE		 3272
3273 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3274 BUG_ON(!data);
3275 data->pages = pages;
3276 data->length = length;
3277 data->alignment = alignment & ~PAGE_MASK;
02afca6c 3278
5240d9f9
AE		 3279 list_add_tail(&data->links, &msg->data);
3280 msg->data_length += length;
02afca6c 3281}
90af3602 3282EXPORT_SYMBOL(ceph_msg_data_add_pages);
31b8006e 3283
90af3602 3284void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
27fa8385
AE		 3285 struct ceph_pagelist *pagelist)
3286{
6644ed7b
AE		 3287 struct ceph_msg_data *data;
3288
07aa1558
AE		 3289 BUG_ON(!pagelist);
3290 BUG_ON(!pagelist->length);
27fa8385 3291
6644ed7b
AE		 3292 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3293 BUG_ON(!data);
3294 data->pagelist = pagelist;
3295
5240d9f9
AE		 3296 list_add_tail(&data->links, &msg->data);
3297 msg->data_length += pagelist->length;
27fa8385 3298}
90af3602 3299EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
27fa8385 3300
ea96571f 3301#ifdef CONFIG_BLOCK
5359a17d
ID		 3302void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
3303 u32 length)
27fa8385 3304{
6644ed7b
AE		 3305 struct ceph_msg_data *data;
3306
6644ed7b
AE		 3307 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3308 BUG_ON(!data);
5359a17d 3309 data->bio_pos = *bio_pos;
c851c495 3310 data->bio_length = length;
6644ed7b 3311
5240d9f9
AE		 3312 list_add_tail(&data->links, &msg->data);
3313 msg->data_length += length;
27fa8385 3314}
90af3602 3315EXPORT_SYMBOL(ceph_msg_data_add_bio);
ea96571f 3316#endif /* CONFIG_BLOCK */
27fa8385 3317
b9e281c2
ID		 3318void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
3319 struct ceph_bvec_iter *bvec_pos)
3320{
3321 struct ceph_msg_data *data;
3322
3323 data = ceph_msg_data_create(CEPH_MSG_DATA_BVECS);
3324 BUG_ON(!data);
3325 data->bvec_pos = *bvec_pos;
3326
3327 list_add_tail(&data->links, &msg->data);
3328 msg->data_length += bvec_pos->iter.bi_size;
3329}
3330EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
3331
31b8006e
SW		 3332/*
3333 * construct a new message with given type, size
3334 * the new msg has a ref count of 1.
3335 */
b61c2763
SW		 3336struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3337 bool can_fail)
31b8006e
SW		 3338{
3339 struct ceph_msg *m;
3340
e3d5d638 3341 m = kmem_cache_zalloc(ceph_msg_cache, flags);
31b8006e
SW		 3342 if (m == NULL)
3343 goto out;
31b8006e
SW		 3344
3345 m->hdr.type = cpu_to_le16(type);
45c6ceb5 3346 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
31b8006e 3347 m->hdr.front_len = cpu_to_le32(front_len);
ca20892d 3348
9516e45b
AE		 3349 INIT_LIST_HEAD(&m->list_head);
3350 kref_init(&m->kref);
5240d9f9 3351 INIT_LIST_HEAD(&m->data);
ca20892d 3352
31b8006e
SW		 3353 /* front */
3354 if (front_len) {
eeb0bed5 3355 m->front.iov_base = ceph_kvmalloc(front_len, flags);
31b8006e 3356 if (m->front.iov_base == NULL) {
b61c2763 3357 dout("ceph_msg_new can't allocate %d bytes\n",
31b8006e
SW		 3358 front_len);
3359 goto out2;
3360 }
3361 } else {
3362 m->front.iov_base = NULL;
3363 }
f2be82b0 3364 m->front_alloc_len = m->front.iov_len = front_len;
31b8006e 3365
bb257664 3366 dout("ceph_msg_new %p front %d\n", m, front_len);
31b8006e
SW		 3367 return m;
3368
3369out2:
3370 ceph_msg_put(m);
3371out:
b61c2763
SW		 3372 if (!can_fail) {
3373 pr_err("msg_new can't create type %d front %d\n", type,
3374 front_len);
f0ed1b7c 3375 WARN_ON(1);
b61c2763
SW		 3376 } else {
3377 dout("msg_new can't create type %d front %d\n", type,
3378 front_len);
3379 }
a79832f2 3380 return NULL;
31b8006e 3381}
3d14c5d2 3382EXPORT_SYMBOL(ceph_msg_new);
31b8006e 3383
31b8006e
SW		 3384/*
3385 * Allocate "middle" portion of a message, if it is needed and wasn't
3386 * allocated by alloc_msg. This allows us to read a small fixed-size
3387 * per-type header in the front and then gracefully fail (i.e.,
3388 * propagate the error to the caller based on info in the front) when
3389 * the middle is too large.
3390 */
2450418c 3391static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
31b8006e
SW		 3392{
3393 int type = le16_to_cpu(msg->hdr.type);
3394 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3395
3396 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3397 ceph_msg_type_name(type), middle_len);
3398 BUG_ON(!middle_len);
3399 BUG_ON(msg->middle);
3400
b6c1d5b8 3401 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 3402 if (!msg->middle)
3403 return -ENOMEM;
3404 return 0;
3405}
3406
2450418c 3407/*
1c20f2d2
AE		 3408 * Allocate a message for receiving an incoming message on a
3409 * connection, and save the result in con->in_msg. Uses the
3410 * connection's private alloc_msg op if available.
3411 *
4740a623
SW		 3412 * Returns 0 on success, or a negative error code.
3413 *
3414 * On success, if we set *skip = 1:
3415 * - the next message should be skipped and ignored.
3416 * - con->in_msg == NULL
3417 * or if we set *skip = 0:
3418 * - con->in_msg is non-null.
3419 * On error (ENOMEM, EAGAIN, ...),
3420 * - con->in_msg == NULL
2450418c 3421 */
4740a623 3422static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2450418c 3423{
4740a623 3424 struct ceph_msg_header *hdr = &con->in_hdr;
2450418c 3425 int middle_len = le32_to_cpu(hdr->middle_len);
1d866d1c 3426 struct ceph_msg *msg;
4740a623 3427 int ret = 0;
2450418c 3428
1c20f2d2 3429 BUG_ON(con->in_msg != NULL);
53ded495 3430 BUG_ON(!con->ops->alloc_msg);
2450418c 3431
53ded495
AE		 3432 mutex_unlock(&con->mutex);
3433 msg = con->ops->alloc_msg(con, hdr, skip);
3434 mutex_lock(&con->mutex);
3435 if (con->state != CON_STATE_OPEN) {
3436 if (msg)
1d866d1c 3437 ceph_msg_put(msg);
53ded495
AE		 3438 return -EAGAIN;
3439 }
4137577a
AE		 3440 if (msg) {
3441 BUG_ON(*skip);
583d0fef 3442 msg_con_set(msg, con);
4137577a 3443 con->in_msg = msg;
4137577a
AE		 3444 } else {
3445 /*
3446 * Null message pointer means either we should skip
3447 * this message or we couldn't allocate memory. The
3448 * former is not an error.
3449 */
3450 if (*skip)
3451 return 0;
4137577a 3452
67c64eb7 3453 con->error_msg = "error allocating memory for incoming message";
53ded495 3454 return -ENOMEM;
2450418c 3455 }
1c20f2d2 3456 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2450418c 3457
1c20f2d2
AE		 3458 if (middle_len && !con->in_msg->middle) {
3459 ret = ceph_alloc_middle(con, con->in_msg);
2450418c 3460 if (ret < 0) {
1c20f2d2
AE		 3461 ceph_msg_put(con->in_msg);
3462 con->in_msg = NULL;
2450418c
YS		 3463 }
3464 }
9d7f0f13 3465
4740a623 3466 return ret;
2450418c
YS		 3467}
3468
31b8006e
SW		 3469
3470/*
3471 * Free a generically kmalloc'd message.
3472 */
0215e44b 3473static void ceph_msg_free(struct ceph_msg *m)
31b8006e 3474{
0215e44b 3475 dout("%s %p\n", __func__, m);
4965fc38 3476 kvfree(m->front.iov_base);
e3d5d638 3477 kmem_cache_free(ceph_msg_cache, m);
31b8006e
SW		 3478}
3479
0215e44b 3480static void ceph_msg_release(struct kref *kref)
c2e552e7
SW		 3481{
3482 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
10bcee14 3483 struct ceph_msg_data *data, *next;
31b8006e 3484
0215e44b 3485 dout("%s %p\n", __func__, m);
c2e552e7
SW		 3486 WARN_ON(!list_empty(&m->list_head));
3487
583d0fef
ID		 3488 msg_con_set(m, NULL);
3489
c2e552e7
SW		 3490 /* drop middle, data, if any */
3491 if (m->middle) {
3492 ceph_buffer_put(m->middle);
3493 m->middle = NULL;
31b8006e 3494 }
5240d9f9 3495
10bcee14
GT		 3496 list_for_each_entry_safe(data, next, &m->data, links) {
3497 list_del_init(&data->links);
5240d9f9
AE		 3498 ceph_msg_data_destroy(data);
3499 }
a1930804 3500 m->data_length = 0;
58bb3b37 3501
c2e552e7
SW		 3502 if (m->pool)
3503 ceph_msgpool_put(m->pool, m);
3504 else
0215e44b
ID		 3505 ceph_msg_free(m);
3506}
3507
3508struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3509{
3510 dout("%s %p (was %d)\n", __func__, msg,
2c935bc5 3511 kref_read(&msg->kref));
0215e44b
ID		 3512 kref_get(&msg->kref);
3513 return msg;
3514}
3515EXPORT_SYMBOL(ceph_msg_get);
3516
3517void ceph_msg_put(struct ceph_msg *msg)
3518{
3519 dout("%s %p (was %d)\n", __func__, msg,
2c935bc5 3520 kref_read(&msg->kref));
0215e44b 3521 kref_put(&msg->kref, ceph_msg_release);
31b8006e 3522}
0215e44b 3523EXPORT_SYMBOL(ceph_msg_put);
9ec7cab1
SW		 3524
3525void ceph_msg_dump(struct ceph_msg *msg)
3526{
3cea4c30
ID		 3527 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3528 msg->front_alloc_len, msg->data_length);
9ec7cab1
SW		 3529 print_hex_dump(KERN_DEBUG, "header: ",
3530 DUMP_PREFIX_OFFSET, 16, 1,
3531 &msg->hdr, sizeof(msg->hdr), true);
3532 print_hex_dump(KERN_DEBUG, " front: ",
3533 DUMP_PREFIX_OFFSET, 16, 1,
3534 msg->front.iov_base, msg->front.iov_len, true);
3535 if (msg->middle)
3536 print_hex_dump(KERN_DEBUG, "middle: ",
3537 DUMP_PREFIX_OFFSET, 16, 1,
3538 msg->middle->vec.iov_base,
3539 msg->middle->vec.iov_len, true);
3540 print_hex_dump(KERN_DEBUG, "footer: ",
3541 DUMP_PREFIX_OFFSET, 16, 1,
3542 &msg->footer, sizeof(msg->footer), true);
3543}
3d14c5d2 3544EXPORT_SYMBOL(ceph_msg_dump);
Ubuntu Hirsute Linux kernel mirror