blob: 30efc5c186222c64ea3d7d21b194f8ce9f4e0f47 [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weil31b80062009-10-06 11:31:13 -07002
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09009#include <linux/slab.h>
Sage Weil31b80062009-10-06 11:31:13 -070010#include <linux/socket.h>
11#include <linux/string.h>
Alex Elder3ebc21f2013-01-31 16:02:01 -060012#ifdef CONFIG_BLOCK
Yehuda Sadeh68b44762010-04-06 15:01:27 -070013#include <linux/bio.h>
Alex Elder3ebc21f2013-01-31 16:02:01 -060014#endif /* CONFIG_BLOCK */
Noah Watkinsee3b56f2011-09-23 11:48:42 -070015#include <linux/dns_resolver.h>
Sage Weil31b80062009-10-06 11:31:13 -070016#include <net/tcp.h>
17
Ilya Dryomov2b3e0c92013-12-24 21:19:24 +020018#include <linux/ceph/ceph_features.h>
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070019#include <linux/ceph/libceph.h>
20#include <linux/ceph/messenger.h>
21#include <linux/ceph/decode.h>
22#include <linux/ceph/pagelist.h>
Paul Gortmakerbc3b2d72011-07-15 11:47:34 -040023#include <linux/export.h>
Sage Weil31b80062009-10-06 11:31:13 -070024
Alex Elderfe38a2b2013-03-06 23:39:39 -060025#define list_entry_next(pos, member) \
26 list_entry(pos->member.next, typeof(*pos), member)
27
Sage Weil31b80062009-10-06 11:31:13 -070028/*
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
Alex Elderbc18f4b2012-06-20 21:53:53 -050037/*
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() \
Sage Weilfbb85a42012-06-27 12:31:02 -070056 * |+--------------------------- \
57 * | \ \ \
58 * | ----------- \ \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
61 * | ^ \ |
62 * | | \ |
63 * | + con_sock_state_closing() \ |
64 * | / \ | |
65 * | / --------------- | |
66 * | / \ v v
Alex Elderbc18f4b2012-06-20 21:53:53 -050067 * | / --------------
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 */
Alex Elderce2c8902012-05-22 22:15:49 -050078
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
Sage Weil8dacc7d2012-07-20 17:24:40 -070085/*
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
Sage Weil4a861692012-07-20 17:29:55 -070095/*
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 */
Sage Weil8dacc7d2012-07-20 17:24:40 -0700104
Alex Elderc9ffc772013-02-20 10:25:12 -0600105static 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
Alex Eldere3d5d632013-05-01 12:43:04 -0500156/* Slab caches for frequently-allocated structures */
157
158static struct kmem_cache *ceph_msg_cache;
Alex Elder81b36be2013-05-01 12:43:04 -0500159static struct kmem_cache *ceph_msg_data_cache;
Alex Eldere3d5d632013-05-01 12:43:04 -0500160
Sage Weil31b80062009-10-06 11:31:13 -0700161/* 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;
165
Sage Weila6a53492010-04-13 14:07:07 -0700166#ifdef CONFIG_LOCKDEP
167static struct lock_class_key socket_class;
168#endif
169
Alex Elder84495f42012-02-15 07:43:55 -0600170/*
171 * When skipping (ignoring) a block of input we read it into a "skip
172 * buffer," which is this many bytes in size.
173 */
174#define SKIP_BUF_SIZE 1024
Sage Weil31b80062009-10-06 11:31:13 -0700175
176static void queue_con(struct ceph_connection *con);
177static void con_work(struct work_struct *);
Alex Elder93209262013-02-19 12:25:57 -0600178static void con_fault(struct ceph_connection *con);
Sage Weil31b80062009-10-06 11:31:13 -0700179
Sage Weil31b80062009-10-06 11:31:13 -0700180/*
Alex Elderf64a9312012-01-23 15:49:27 -0600181 * Nicely render a sockaddr as a string. An array of formatted
182 * strings is used, to approximate reentrancy.
Sage Weil31b80062009-10-06 11:31:13 -0700183 */
Alex Elderf64a9312012-01-23 15:49:27 -0600184#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
185#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
186#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
187#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
188
189static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
190static atomic_t addr_str_seq = ATOMIC_INIT(0);
Sage Weil31b80062009-10-06 11:31:13 -0700191
Alex Elder57666512012-01-23 15:49:27 -0600192static struct page *zero_page; /* used in certain error cases */
Alex Elder57666512012-01-23 15:49:27 -0600193
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700194const char *ceph_pr_addr(const struct sockaddr_storage *ss)
Sage Weil31b80062009-10-06 11:31:13 -0700195{
196 int i;
197 char *s;
Alex Elder99f0f3b2012-01-23 15:49:27 -0600198 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
199 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
Sage Weil31b80062009-10-06 11:31:13 -0700200
Alex Elderf64a9312012-01-23 15:49:27 -0600201 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
Sage Weil31b80062009-10-06 11:31:13 -0700202 s = addr_str[i];
203
204 switch (ss->ss_family) {
205 case AF_INET:
Alex Elderbd406142012-01-23 15:49:27 -0600206 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
207 ntohs(in4->sin_port));
Sage Weil31b80062009-10-06 11:31:13 -0700208 break;
209
210 case AF_INET6:
Alex Elderbd406142012-01-23 15:49:27 -0600211 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
212 ntohs(in6->sin6_port));
Sage Weil31b80062009-10-06 11:31:13 -0700213 break;
214
215 default:
Alex Elderd3002b92012-02-14 14:05:33 -0600216 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
217 ss->ss_family);
Sage Weil31b80062009-10-06 11:31:13 -0700218 }
219
220 return s;
221}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700222EXPORT_SYMBOL(ceph_pr_addr);
Sage Weil31b80062009-10-06 11:31:13 -0700223
Sage Weil63f2d212009-11-03 15:17:56 -0800224static void encode_my_addr(struct ceph_messenger *msgr)
225{
226 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
227 ceph_encode_addr(&msgr->my_enc_addr);
228}
229
Sage Weil31b80062009-10-06 11:31:13 -0700230/*
231 * work queue for all reading and writing to/from the socket.
232 */
Alex Eldere0f43c92012-02-14 14:05:33 -0600233static struct workqueue_struct *ceph_msgr_wq;
Sage Weil31b80062009-10-06 11:31:13 -0700234
Alex Eldere3d5d632013-05-01 12:43:04 -0500235static int ceph_msgr_slab_init(void)
236{
237 BUG_ON(ceph_msg_cache);
238 ceph_msg_cache = kmem_cache_create("ceph_msg",
239 sizeof (struct ceph_msg),
240 __alignof__(struct ceph_msg), 0, NULL);
Alex Elder81b36be2013-05-01 12:43:04 -0500241
242 if (!ceph_msg_cache)
243 return -ENOMEM;
244
245 BUG_ON(ceph_msg_data_cache);
246 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
247 sizeof (struct ceph_msg_data),
248 __alignof__(struct ceph_msg_data),
249 0, NULL);
250 if (ceph_msg_data_cache)
251 return 0;
252
253 kmem_cache_destroy(ceph_msg_cache);
254 ceph_msg_cache = NULL;
255
256 return -ENOMEM;
Alex Eldere3d5d632013-05-01 12:43:04 -0500257}
258
259static void ceph_msgr_slab_exit(void)
260{
Alex Elder81b36be2013-05-01 12:43:04 -0500261 BUG_ON(!ceph_msg_data_cache);
262 kmem_cache_destroy(ceph_msg_data_cache);
263 ceph_msg_data_cache = NULL;
264
Alex Eldere3d5d632013-05-01 12:43:04 -0500265 BUG_ON(!ceph_msg_cache);
266 kmem_cache_destroy(ceph_msg_cache);
267 ceph_msg_cache = NULL;
268}
269
Alex Elder15417162013-02-19 12:25:56 -0600270static void _ceph_msgr_exit(void)
Alex Elder6173d1f2012-02-14 14:05:33 -0600271{
Alex Elderd3002b92012-02-14 14:05:33 -0600272 if (ceph_msgr_wq) {
Alex Elder6173d1f2012-02-14 14:05:33 -0600273 destroy_workqueue(ceph_msgr_wq);
Alex Elderd3002b92012-02-14 14:05:33 -0600274 ceph_msgr_wq = NULL;
275 }
Alex Elder6173d1f2012-02-14 14:05:33 -0600276
Alex Eldere3d5d632013-05-01 12:43:04 -0500277 ceph_msgr_slab_exit();
278
Alex Elder6173d1f2012-02-14 14:05:33 -0600279 BUG_ON(zero_page == NULL);
280 kunmap(zero_page);
281 page_cache_release(zero_page);
282 zero_page = NULL;
283}
284
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700285int ceph_msgr_init(void)
Sage Weil31b80062009-10-06 11:31:13 -0700286{
Alex Elder57666512012-01-23 15:49:27 -0600287 BUG_ON(zero_page != NULL);
288 zero_page = ZERO_PAGE(0);
289 page_cache_get(zero_page);
290
Alex Eldere3d5d632013-05-01 12:43:04 -0500291 if (ceph_msgr_slab_init())
292 return -ENOMEM;
293
Tejun Heo4d1829a2013-07-30 08:40:27 -0400294 ceph_msgr_wq = alloc_workqueue("ceph-msgr", 0, 0);
Alex Elder6173d1f2012-02-14 14:05:33 -0600295 if (ceph_msgr_wq)
296 return 0;
Alex Elder57666512012-01-23 15:49:27 -0600297
Alex Elder6173d1f2012-02-14 14:05:33 -0600298 pr_err("msgr_init failed to create workqueue\n");
299 _ceph_msgr_exit();
Alex Elder57666512012-01-23 15:49:27 -0600300
Alex Elder6173d1f2012-02-14 14:05:33 -0600301 return -ENOMEM;
Sage Weil31b80062009-10-06 11:31:13 -0700302}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700303EXPORT_SYMBOL(ceph_msgr_init);
Sage Weil31b80062009-10-06 11:31:13 -0700304
305void ceph_msgr_exit(void)
306{
Alex Elder57666512012-01-23 15:49:27 -0600307 BUG_ON(ceph_msgr_wq == NULL);
Alex Elder57666512012-01-23 15:49:27 -0600308
Alex Elder6173d1f2012-02-14 14:05:33 -0600309 _ceph_msgr_exit();
Sage Weil31b80062009-10-06 11:31:13 -0700310}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700311EXPORT_SYMBOL(ceph_msgr_exit);
Sage Weil31b80062009-10-06 11:31:13 -0700312
Yehuda Sadehcd84db62010-06-11 16:58:48 -0700313void ceph_msgr_flush(void)
Sage Weila922d382010-05-29 09:41:23 -0700314{
315 flush_workqueue(ceph_msgr_wq);
316}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700317EXPORT_SYMBOL(ceph_msgr_flush);
Sage Weila922d382010-05-29 09:41:23 -0700318
Alex Elderce2c8902012-05-22 22:15:49 -0500319/* Connection socket state transition functions */
320
321static void con_sock_state_init(struct ceph_connection *con)
322{
323 int old_state;
324
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
327 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700328 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
329 CON_SOCK_STATE_CLOSED);
Alex Elderce2c8902012-05-22 22:15:49 -0500330}
331
332static void con_sock_state_connecting(struct ceph_connection *con)
333{
334 int old_state;
335
336 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
337 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
338 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700339 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
340 CON_SOCK_STATE_CONNECTING);
Alex Elderce2c8902012-05-22 22:15:49 -0500341}
342
343static void con_sock_state_connected(struct ceph_connection *con)
344{
345 int old_state;
346
347 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
348 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
349 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700350 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
351 CON_SOCK_STATE_CONNECTED);
Alex Elderce2c8902012-05-22 22:15:49 -0500352}
353
354static void con_sock_state_closing(struct ceph_connection *con)
355{
356 int old_state;
357
358 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
359 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
360 old_state != CON_SOCK_STATE_CONNECTED &&
361 old_state != CON_SOCK_STATE_CLOSING))
362 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700363 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
364 CON_SOCK_STATE_CLOSING);
Alex Elderce2c8902012-05-22 22:15:49 -0500365}
366
367static void con_sock_state_closed(struct ceph_connection *con)
368{
369 int old_state;
370
371 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
372 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
Sage Weilfbb85a42012-06-27 12:31:02 -0700373 old_state != CON_SOCK_STATE_CLOSING &&
Sage Weil8007b8d2012-07-30 18:16:16 -0700374 old_state != CON_SOCK_STATE_CONNECTING &&
375 old_state != CON_SOCK_STATE_CLOSED))
Alex Elderce2c8902012-05-22 22:15:49 -0500376 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700377 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
378 CON_SOCK_STATE_CLOSED);
Alex Elderce2c8902012-05-22 22:15:49 -0500379}
Sage Weila922d382010-05-29 09:41:23 -0700380
Sage Weil31b80062009-10-06 11:31:13 -0700381/*
382 * socket callback functions
383 */
384
385/* data available on socket, or listen socket received a connect */
Alex Elder327800b2012-05-22 11:41:43 -0500386static void ceph_sock_data_ready(struct sock *sk, int count_unused)
Sage Weil31b80062009-10-06 11:31:13 -0700387{
Alex Elderbd406142012-01-23 15:49:27 -0600388 struct ceph_connection *con = sk->sk_user_data;
Guanjun Hea2a32582012-07-08 19:50:33 -0700389 if (atomic_read(&con->msgr->stopping)) {
390 return;
391 }
Alex Elderbd406142012-01-23 15:49:27 -0600392
Sage Weil31b80062009-10-06 11:31:13 -0700393 if (sk->sk_state != TCP_CLOSE_WAIT) {
Alex Elder327800b2012-05-22 11:41:43 -0500394 dout("%s on %p state = %lu, queueing work\n", __func__,
Sage Weil31b80062009-10-06 11:31:13 -0700395 con, con->state);
396 queue_con(con);
397 }
398}
399
400/* socket has buffer space for writing */
Alex Elder327800b2012-05-22 11:41:43 -0500401static void ceph_sock_write_space(struct sock *sk)
Sage Weil31b80062009-10-06 11:31:13 -0700402{
Alex Elderd3002b92012-02-14 14:05:33 -0600403 struct ceph_connection *con = sk->sk_user_data;
Sage Weil31b80062009-10-06 11:31:13 -0700404
Jim Schutt182fac22012-02-29 08:30:58 -0700405 /* only queue to workqueue if there is data we want to write,
406 * and there is sufficient space in the socket buffer to accept
Alex Elder327800b2012-05-22 11:41:43 -0500407 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
Jim Schutt182fac22012-02-29 08:30:58 -0700408 * doesn't get called again until try_write() fills the socket
409 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
410 * and net/core/stream.c:sk_stream_write_space().
411 */
Alex Elderc9ffc772013-02-20 10:25:12 -0600412 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
Eric Dumazet64dc6132013-07-22 20:26:31 -0700413 if (sk_stream_is_writeable(sk)) {
Alex Elder327800b2012-05-22 11:41:43 -0500414 dout("%s %p queueing write work\n", __func__, con);
Jim Schutt182fac22012-02-29 08:30:58 -0700415 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
416 queue_con(con);
417 }
Sage Weil31b80062009-10-06 11:31:13 -0700418 } else {
Alex Elder327800b2012-05-22 11:41:43 -0500419 dout("%s %p nothing to write\n", __func__, con);
Sage Weil31b80062009-10-06 11:31:13 -0700420 }
Sage Weil31b80062009-10-06 11:31:13 -0700421}
422
423/* socket's state has changed */
Alex Elder327800b2012-05-22 11:41:43 -0500424static void ceph_sock_state_change(struct sock *sk)
Sage Weil31b80062009-10-06 11:31:13 -0700425{
Alex Elderbd406142012-01-23 15:49:27 -0600426 struct ceph_connection *con = sk->sk_user_data;
Sage Weil31b80062009-10-06 11:31:13 -0700427
Alex Elder327800b2012-05-22 11:41:43 -0500428 dout("%s %p state = %lu sk_state = %u\n", __func__,
Sage Weil31b80062009-10-06 11:31:13 -0700429 con, con->state, sk->sk_state);
430
Sage Weil31b80062009-10-06 11:31:13 -0700431 switch (sk->sk_state) {
432 case TCP_CLOSE:
Alex Elder327800b2012-05-22 11:41:43 -0500433 dout("%s TCP_CLOSE\n", __func__);
Sage Weil31b80062009-10-06 11:31:13 -0700434 case TCP_CLOSE_WAIT:
Alex Elder327800b2012-05-22 11:41:43 -0500435 dout("%s TCP_CLOSE_WAIT\n", __func__);
Alex Elderce2c8902012-05-22 22:15:49 -0500436 con_sock_state_closing(con);
Alex Elderc9ffc772013-02-20 10:25:12 -0600437 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
Alex Elderd65c9e02012-06-20 21:53:53 -0500438 queue_con(con);
Sage Weil31b80062009-10-06 11:31:13 -0700439 break;
440 case TCP_ESTABLISHED:
Alex Elder327800b2012-05-22 11:41:43 -0500441 dout("%s TCP_ESTABLISHED\n", __func__);
Alex Elderce2c8902012-05-22 22:15:49 -0500442 con_sock_state_connected(con);
Sage Weil31b80062009-10-06 11:31:13 -0700443 queue_con(con);
444 break;
Alex Elderd3002b92012-02-14 14:05:33 -0600445 default: /* Everything else is uninteresting */
446 break;
Sage Weil31b80062009-10-06 11:31:13 -0700447 }
448}
449
450/*
451 * set up socket callbacks
452 */
453static void set_sock_callbacks(struct socket *sock,
454 struct ceph_connection *con)
455{
456 struct sock *sk = sock->sk;
Alex Elderbd406142012-01-23 15:49:27 -0600457 sk->sk_user_data = con;
Alex Elder327800b2012-05-22 11:41:43 -0500458 sk->sk_data_ready = ceph_sock_data_ready;
459 sk->sk_write_space = ceph_sock_write_space;
460 sk->sk_state_change = ceph_sock_state_change;
Sage Weil31b80062009-10-06 11:31:13 -0700461}
462
463
464/*
465 * socket helpers
466 */
467
468/*
469 * initiate connection to a remote socket.
470 */
Alex Elder41617d02012-02-14 14:05:33 -0600471static int ceph_tcp_connect(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -0700472{
Sage Weilf91d3472010-07-01 15:18:31 -0700473 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
Sage Weil31b80062009-10-06 11:31:13 -0700474 struct socket *sock;
475 int ret;
476
477 BUG_ON(con->sock);
Sage Weilf91d3472010-07-01 15:18:31 -0700478 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
479 IPPROTO_TCP, &sock);
Sage Weil31b80062009-10-06 11:31:13 -0700480 if (ret)
Alex Elder41617d02012-02-14 14:05:33 -0600481 return ret;
Sage Weil31b80062009-10-06 11:31:13 -0700482 sock->sk->sk_allocation = GFP_NOFS;
483
Sage Weila6a53492010-04-13 14:07:07 -0700484#ifdef CONFIG_LOCKDEP
485 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
486#endif
487
Sage Weil31b80062009-10-06 11:31:13 -0700488 set_sock_callbacks(sock, con);
489
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700490 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -0700491
Sage Weil89a86be2012-06-09 14:19:21 -0700492 con_sock_state_connecting(con);
Sage Weilf91d3472010-07-01 15:18:31 -0700493 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
494 O_NONBLOCK);
Sage Weil31b80062009-10-06 11:31:13 -0700495 if (ret == -EINPROGRESS) {
496 dout("connect %s EINPROGRESS sk_state = %u\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700497 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil31b80062009-10-06 11:31:13 -0700498 sock->sk->sk_state);
Alex Eldera5bc3129a2012-01-23 15:49:27 -0600499 } else if (ret < 0) {
Sage Weil31b80062009-10-06 11:31:13 -0700500 pr_err("connect %s error %d\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700501 ceph_pr_addr(&con->peer_addr.in_addr), ret);
Sage Weil31b80062009-10-06 11:31:13 -0700502 sock_release(sock);
Sage Weil31b80062009-10-06 11:31:13 -0700503 con->error_msg = "connect error";
Sage Weil31b80062009-10-06 11:31:13 -0700504
Alex Elder41617d02012-02-14 14:05:33 -0600505 return ret;
Alex Eldera5bc3129a2012-01-23 15:49:27 -0600506 }
507 con->sock = sock;
Alex Elder41617d02012-02-14 14:05:33 -0600508 return 0;
Sage Weil31b80062009-10-06 11:31:13 -0700509}
510
511static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
512{
513 struct kvec iov = {buf, len};
514 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
Sage Weil98bdb0a2011-01-25 08:17:48 -0800515 int r;
Sage Weil31b80062009-10-06 11:31:13 -0700516
Sage Weil98bdb0a2011-01-25 08:17:48 -0800517 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
518 if (r == -EAGAIN)
519 r = 0;
520 return r;
Sage Weil31b80062009-10-06 11:31:13 -0700521}
522
Alex Elderafb3d902013-03-08 20:58:59 -0600523static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
524 int page_offset, size_t length)
525{
526 void *kaddr;
527 int ret;
528
529 BUG_ON(page_offset + length > PAGE_SIZE);
530
531 kaddr = kmap(page);
532 BUG_ON(!kaddr);
533 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
534 kunmap(page);
535
536 return ret;
537}
538
Sage Weil31b80062009-10-06 11:31:13 -0700539/*
540 * write something. @more is true if caller will be sending more data
541 * shortly.
542 */
543static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
544 size_t kvlen, size_t len, int more)
545{
546 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
Sage Weil42961d22011-01-25 08:19:34 -0800547 int r;
Sage Weil31b80062009-10-06 11:31:13 -0700548
549 if (more)
550 msg.msg_flags |= MSG_MORE;
551 else
552 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
553
Sage Weil42961d22011-01-25 08:19:34 -0800554 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
555 if (r == -EAGAIN)
556 r = 0;
557 return r;
Sage Weil31b80062009-10-06 11:31:13 -0700558}
559
Alex Elder31739132012-03-07 11:40:08 -0600560static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
Alex Eldere1dcb122013-03-06 23:39:38 -0600561 int offset, size_t size, bool more)
Alex Elder31739132012-03-07 11:40:08 -0600562{
563 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
564 int ret;
565
566 ret = kernel_sendpage(sock, page, offset, size, flags);
567 if (ret == -EAGAIN)
568 ret = 0;
569
570 return ret;
571}
572
Sage Weil31b80062009-10-06 11:31:13 -0700573
574/*
575 * Shutdown/close the socket for the given connection.
576 */
577static int con_close_socket(struct ceph_connection *con)
578{
Sage Weil8007b8d2012-07-30 18:16:16 -0700579 int rc = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700580
581 dout("con_close_socket on %p sock %p\n", con, con->sock);
Sage Weil8007b8d2012-07-30 18:16:16 -0700582 if (con->sock) {
583 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
584 sock_release(con->sock);
585 con->sock = NULL;
586 }
Alex Elder456ea462012-06-20 21:53:53 -0500587
588 /*
Sage Weil4a861692012-07-20 17:29:55 -0700589 * Forcibly clear the SOCK_CLOSED flag. It gets set
Alex Elder456ea462012-06-20 21:53:53 -0500590 * independent of the connection mutex, and we could have
591 * received a socket close event before we had the chance to
592 * shut the socket down.
593 */
Alex Elderc9ffc772013-02-20 10:25:12 -0600594 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
Sage Weil8007b8d2012-07-30 18:16:16 -0700595
Alex Elderce2c8902012-05-22 22:15:49 -0500596 con_sock_state_closed(con);
Sage Weil31b80062009-10-06 11:31:13 -0700597 return rc;
598}
599
600/*
601 * Reset a connection. Discard all incoming and outgoing messages
602 * and clear *_seq state.
603 */
604static void ceph_msg_remove(struct ceph_msg *msg)
605{
606 list_del_init(&msg->list_head);
Alex Elder38941f82012-06-01 14:56:43 -0500607 BUG_ON(msg->con == NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -0700608 msg->con->ops->put(msg->con);
Alex Elder38941f82012-06-01 14:56:43 -0500609 msg->con = NULL;
610
Sage Weil31b80062009-10-06 11:31:13 -0700611 ceph_msg_put(msg);
612}
613static void ceph_msg_remove_list(struct list_head *head)
614{
615 while (!list_empty(head)) {
616 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
617 list_head);
618 ceph_msg_remove(msg);
619 }
620}
621
622static void reset_connection(struct ceph_connection *con)
623{
624 /* reset connection, out_queue, msg_ and connect_seq */
625 /* discard existing out_queue and msg_seq */
Sage Weil0fa6ebc2012-12-27 20:27:04 -0600626 dout("reset_connection %p\n", con);
Sage Weil31b80062009-10-06 11:31:13 -0700627 ceph_msg_remove_list(&con->out_queue);
628 ceph_msg_remove_list(&con->out_sent);
629
Sage Weilcf3e5c42009-12-11 09:48:05 -0800630 if (con->in_msg) {
Alex Elder38941f82012-06-01 14:56:43 -0500631 BUG_ON(con->in_msg->con != con);
632 con->in_msg->con = NULL;
Sage Weilcf3e5c42009-12-11 09:48:05 -0800633 ceph_msg_put(con->in_msg);
634 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -0700635 con->ops->put(con);
Sage Weilcf3e5c42009-12-11 09:48:05 -0800636 }
637
Sage Weil31b80062009-10-06 11:31:13 -0700638 con->connect_seq = 0;
639 con->out_seq = 0;
Sage Weilc86a2932009-12-14 14:04:30 -0800640 if (con->out_msg) {
641 ceph_msg_put(con->out_msg);
642 con->out_msg = NULL;
643 }
Sage Weil31b80062009-10-06 11:31:13 -0700644 con->in_seq = 0;
Sage Weil0e0d5e02010-04-02 16:07:19 -0700645 con->in_seq_acked = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700646}
647
648/*
649 * mark a peer down. drop any open connections.
650 */
651void ceph_con_close(struct ceph_connection *con)
652{
Sage Weil8c50c812012-07-30 16:24:37 -0700653 mutex_lock(&con->mutex);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700654 dout("con_close %p peer %s\n", con,
655 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil8dacc7d2012-07-20 17:24:40 -0700656 con->state = CON_STATE_CLOSED;
Alex Eldera5988c42012-05-29 11:04:58 -0500657
Alex Elderc9ffc772013-02-20 10:25:12 -0600658 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
659 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
660 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
661 con_flag_clear(con, CON_FLAG_BACKOFF);
Alex Eldera5988c42012-05-29 11:04:58 -0500662
Sage Weil31b80062009-10-06 11:31:13 -0700663 reset_connection(con);
Sage Weil6f2bc3f2010-04-02 16:16:34 -0700664 con->peer_global_seq = 0;
Sage Weil91e45ce32010-02-15 12:05:09 -0800665 cancel_delayed_work(&con->work);
Sage Weilee76e072012-07-20 16:45:49 -0700666 con_close_socket(con);
Sage Weilec302642009-12-22 10:43:42 -0800667 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700668}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700669EXPORT_SYMBOL(ceph_con_close);
Sage Weil31b80062009-10-06 11:31:13 -0700670
671/*
Sage Weil31b80062009-10-06 11:31:13 -0700672 * Reopen a closed connection, with a new peer address.
673 */
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700674void ceph_con_open(struct ceph_connection *con,
675 __u8 entity_type, __u64 entity_num,
676 struct ceph_entity_addr *addr)
Sage Weil31b80062009-10-06 11:31:13 -0700677{
Sage Weil54691552012-07-30 16:21:40 -0700678 mutex_lock(&con->mutex);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700679 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
Sage Weil8dacc7d2012-07-20 17:24:40 -0700680
Alex Elder122070a2012-12-26 10:43:57 -0600681 WARN_ON(con->state != CON_STATE_CLOSED);
Sage Weil8dacc7d2012-07-20 17:24:40 -0700682 con->state = CON_STATE_PREOPEN;
Alex Eldera5988c42012-05-29 11:04:58 -0500683
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700684 con->peer_name.type = (__u8) entity_type;
685 con->peer_name.num = cpu_to_le64(entity_num);
686
Sage Weil31b80062009-10-06 11:31:13 -0700687 memcpy(&con->peer_addr, addr, sizeof(*addr));
Sage Weil03c677e2009-11-20 15:14:15 -0800688 con->delay = 0; /* reset backoff memory */
Sage Weil54691552012-07-30 16:21:40 -0700689 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700690 queue_con(con);
691}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700692EXPORT_SYMBOL(ceph_con_open);
Sage Weil31b80062009-10-06 11:31:13 -0700693
694/*
Sage Weil87b315a2010-03-22 14:51:18 -0700695 * return true if this connection ever successfully opened
696 */
697bool ceph_con_opened(struct ceph_connection *con)
698{
699 return con->connect_seq > 0;
700}
701
702/*
Sage Weil31b80062009-10-06 11:31:13 -0700703 * initialize a new connection.
704 */
Alex Elder1bfd89f2012-05-26 23:26:43 -0500705void ceph_con_init(struct ceph_connection *con, void *private,
706 const struct ceph_connection_operations *ops,
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700707 struct ceph_messenger *msgr)
Sage Weil31b80062009-10-06 11:31:13 -0700708{
709 dout("con_init %p\n", con);
710 memset(con, 0, sizeof(*con));
Alex Elder1bfd89f2012-05-26 23:26:43 -0500711 con->private = private;
712 con->ops = ops;
Sage Weil31b80062009-10-06 11:31:13 -0700713 con->msgr = msgr;
Alex Elderce2c8902012-05-22 22:15:49 -0500714
715 con_sock_state_init(con);
716
Sage Weilec302642009-12-22 10:43:42 -0800717 mutex_init(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700718 INIT_LIST_HEAD(&con->out_queue);
719 INIT_LIST_HEAD(&con->out_sent);
720 INIT_DELAYED_WORK(&con->work, con_work);
Alex Eldera5988c42012-05-29 11:04:58 -0500721
Sage Weil8dacc7d2012-07-20 17:24:40 -0700722 con->state = CON_STATE_CLOSED;
Sage Weil31b80062009-10-06 11:31:13 -0700723}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700724EXPORT_SYMBOL(ceph_con_init);
Sage Weil31b80062009-10-06 11:31:13 -0700725
726
727/*
728 * We maintain a global counter to order connection attempts. Get
729 * a unique seq greater than @gt.
730 */
731static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
732{
733 u32 ret;
734
735 spin_lock(&msgr->global_seq_lock);
736 if (msgr->global_seq < gt)
737 msgr->global_seq = gt;
738 ret = ++msgr->global_seq;
739 spin_unlock(&msgr->global_seq_lock);
740 return ret;
741}
742
Alex Eldere2200422012-05-23 14:35:23 -0500743static void con_out_kvec_reset(struct ceph_connection *con)
Alex Elder859eb792012-02-14 14:05:33 -0600744{
745 con->out_kvec_left = 0;
746 con->out_kvec_bytes = 0;
747 con->out_kvec_cur = &con->out_kvec[0];
748}
749
Alex Eldere2200422012-05-23 14:35:23 -0500750static void con_out_kvec_add(struct ceph_connection *con,
Alex Elder859eb792012-02-14 14:05:33 -0600751 size_t size, void *data)
752{
753 int index;
754
755 index = con->out_kvec_left;
756 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
757
758 con->out_kvec[index].iov_len = size;
759 con->out_kvec[index].iov_base = data;
760 con->out_kvec_left++;
761 con->out_kvec_bytes += size;
762}
Sage Weil31b80062009-10-06 11:31:13 -0700763
Alex Elderdf6ad1f2012-06-11 14:57:13 -0500764#ifdef CONFIG_BLOCK
Alex Elder6aaa4512013-03-06 23:39:39 -0600765
766/*
767 * For a bio data item, a piece is whatever remains of the next
768 * entry in the current bio iovec, or the first entry in the next
769 * bio in the list.
770 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500771static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500772 size_t length)
Alex Elder6aaa4512013-03-06 23:39:39 -0600773{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500774 struct ceph_msg_data *data = cursor->data;
Alex Elder6aaa4512013-03-06 23:39:39 -0600775 struct bio *bio;
776
777 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
778
779 bio = data->bio;
780 BUG_ON(!bio);
Alex Elder6aaa4512013-03-06 23:39:39 -0600781
Alex Elderca8b3a62013-04-05 14:46:01 -0500782 cursor->resid = min(length, data->bio_length);
Alex Elder6aaa4512013-03-06 23:39:39 -0600783 cursor->bio = bio;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700784 cursor->bvec_iter = bio->bi_iter;
785 cursor->last_piece =
786 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
Alex Elder6aaa4512013-03-06 23:39:39 -0600787}
788
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500789static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
Alex Elder6aaa4512013-03-06 23:39:39 -0600790 size_t *page_offset,
791 size_t *length)
792{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500793 struct ceph_msg_data *data = cursor->data;
Alex Elder6aaa4512013-03-06 23:39:39 -0600794 struct bio *bio;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700795 struct bio_vec bio_vec;
Alex Elder6aaa4512013-03-06 23:39:39 -0600796
797 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
798
799 bio = cursor->bio;
800 BUG_ON(!bio);
801
Kent Overstreetf38a5182013-08-07 14:30:24 -0700802 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
Alex Elder6aaa4512013-03-06 23:39:39 -0600803
Kent Overstreetf38a5182013-08-07 14:30:24 -0700804 *page_offset = (size_t) bio_vec.bv_offset;
Alex Elder6aaa4512013-03-06 23:39:39 -0600805 BUG_ON(*page_offset >= PAGE_SIZE);
Alex Elder25aff7c2013-03-11 23:34:22 -0500806 if (cursor->last_piece) /* pagelist offset is always 0 */
807 *length = cursor->resid;
808 else
Kent Overstreetf38a5182013-08-07 14:30:24 -0700809 *length = (size_t) bio_vec.bv_len;
Alex Elder25aff7c2013-03-11 23:34:22 -0500810 BUG_ON(*length > cursor->resid);
Alex Elder5df521b2013-03-30 15:09:59 -0500811 BUG_ON(*page_offset + *length > PAGE_SIZE);
Alex Elder6aaa4512013-03-06 23:39:39 -0600812
Kent Overstreetf38a5182013-08-07 14:30:24 -0700813 return bio_vec.bv_page;
Alex Elder6aaa4512013-03-06 23:39:39 -0600814}
815
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500816static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
817 size_t bytes)
Alex Elder6aaa4512013-03-06 23:39:39 -0600818{
Alex Elder6aaa4512013-03-06 23:39:39 -0600819 struct bio *bio;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700820 struct bio_vec bio_vec;
Alex Elder6aaa4512013-03-06 23:39:39 -0600821
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500822 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
Alex Elder6aaa4512013-03-06 23:39:39 -0600823
824 bio = cursor->bio;
825 BUG_ON(!bio);
826
Kent Overstreetf38a5182013-08-07 14:30:24 -0700827 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
Alex Elder6aaa4512013-03-06 23:39:39 -0600828
829 /* Advance the cursor offset */
830
Alex Elder25aff7c2013-03-11 23:34:22 -0500831 BUG_ON(cursor->resid < bytes);
832 cursor->resid -= bytes;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700833
834 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
835
836 if (bytes < bio_vec.bv_len)
Alex Elder6aaa4512013-03-06 23:39:39 -0600837 return false; /* more bytes to process in this segment */
838
839 /* Move on to the next segment, and possibly the next bio */
840
Kent Overstreetf38a5182013-08-07 14:30:24 -0700841 if (!cursor->bvec_iter.bi_size) {
Alex Elder6aaa4512013-03-06 23:39:39 -0600842 bio = bio->bi_next;
Ilya Dryomov0ec1d152014-02-05 15:19:55 +0200843 cursor->bio = bio;
844 if (bio)
845 cursor->bvec_iter = bio->bi_iter;
846 else
847 memset(&cursor->bvec_iter, 0,
848 sizeof(cursor->bvec_iter));
Alex Elder6aaa4512013-03-06 23:39:39 -0600849 }
Alex Elder6aaa4512013-03-06 23:39:39 -0600850
Alex Elder25aff7c2013-03-11 23:34:22 -0500851 if (!cursor->last_piece) {
852 BUG_ON(!cursor->resid);
853 BUG_ON(!bio);
854 /* A short read is OK, so use <= rather than == */
Kent Overstreetf38a5182013-08-07 14:30:24 -0700855 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
Alex Elder6aaa4512013-03-06 23:39:39 -0600856 cursor->last_piece = true;
Alex Elder25aff7c2013-03-11 23:34:22 -0500857 }
Alex Elder6aaa4512013-03-06 23:39:39 -0600858
859 return true;
860}
Alex Elderea965712013-04-05 14:46:01 -0500861#endif /* CONFIG_BLOCK */
Alex Elderdf6ad1f2012-06-11 14:57:13 -0500862
Alex Elderfe38a2b2013-03-06 23:39:39 -0600863/*
Alex Eldere766d7b2013-03-07 15:38:28 -0600864 * For a page array, a piece comes from the first page in the array
865 * that has not already been fully consumed.
866 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500867static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500868 size_t length)
Alex Eldere766d7b2013-03-07 15:38:28 -0600869{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500870 struct ceph_msg_data *data = cursor->data;
Alex Eldere766d7b2013-03-07 15:38:28 -0600871 int page_count;
872
873 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
874
875 BUG_ON(!data->pages);
876 BUG_ON(!data->length);
877
Alex Elderca8b3a62013-04-05 14:46:01 -0500878 cursor->resid = min(length, data->length);
Alex Eldere766d7b2013-03-07 15:38:28 -0600879 page_count = calc_pages_for(data->alignment, (u64)data->length);
Alex Eldere766d7b2013-03-07 15:38:28 -0600880 cursor->page_offset = data->alignment & ~PAGE_MASK;
881 cursor->page_index = 0;
Alex Elder56fc5652013-03-30 23:46:55 -0500882 BUG_ON(page_count > (int)USHRT_MAX);
883 cursor->page_count = (unsigned short)page_count;
884 BUG_ON(length > SIZE_MAX - cursor->page_offset);
885 cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE;
Alex Eldere766d7b2013-03-07 15:38:28 -0600886}
887
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500888static struct page *
889ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
890 size_t *page_offset, size_t *length)
Alex Eldere766d7b2013-03-07 15:38:28 -0600891{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500892 struct ceph_msg_data *data = cursor->data;
Alex Eldere766d7b2013-03-07 15:38:28 -0600893
894 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
895
896 BUG_ON(cursor->page_index >= cursor->page_count);
897 BUG_ON(cursor->page_offset >= PAGE_SIZE);
Alex Eldere766d7b2013-03-07 15:38:28 -0600898
899 *page_offset = cursor->page_offset;
Alex Elder25aff7c2013-03-11 23:34:22 -0500900 if (cursor->last_piece)
Alex Eldere766d7b2013-03-07 15:38:28 -0600901 *length = cursor->resid;
Alex Elder25aff7c2013-03-11 23:34:22 -0500902 else
Alex Eldere766d7b2013-03-07 15:38:28 -0600903 *length = PAGE_SIZE - *page_offset;
Alex Eldere766d7b2013-03-07 15:38:28 -0600904
905 return data->pages[cursor->page_index];
906}
907
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500908static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
Alex Eldere766d7b2013-03-07 15:38:28 -0600909 size_t bytes)
910{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500911 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
Alex Eldere766d7b2013-03-07 15:38:28 -0600912
913 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
Alex Eldere766d7b2013-03-07 15:38:28 -0600914
915 /* Advance the cursor page offset */
916
917 cursor->resid -= bytes;
Alex Elder5df521b2013-03-30 15:09:59 -0500918 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
919 if (!bytes || cursor->page_offset)
Alex Eldere766d7b2013-03-07 15:38:28 -0600920 return false; /* more bytes to process in the current page */
921
Alex Elder5df521b2013-03-30 15:09:59 -0500922 /* Move on to the next page; offset is already at 0 */
Alex Eldere766d7b2013-03-07 15:38:28 -0600923
924 BUG_ON(cursor->page_index >= cursor->page_count);
Alex Eldere766d7b2013-03-07 15:38:28 -0600925 cursor->page_index++;
Alex Elder25aff7c2013-03-11 23:34:22 -0500926 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Eldere766d7b2013-03-07 15:38:28 -0600927
928 return true;
929}
930
931/*
Alex Elderdd236fc2013-03-06 23:39:39 -0600932 * For a pagelist, a piece is whatever remains to be consumed in the
933 * first page in the list, or the front of the next page.
Alex Elderfe38a2b2013-03-06 23:39:39 -0600934 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500935static void
936ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500937 size_t length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600938{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500939 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600940 struct ceph_pagelist *pagelist;
941 struct page *page;
942
Alex Elderdd236fc2013-03-06 23:39:39 -0600943 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600944
945 pagelist = data->pagelist;
946 BUG_ON(!pagelist);
Alex Elder25aff7c2013-03-11 23:34:22 -0500947
948 if (!length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600949 return; /* pagelist can be assigned but empty */
950
951 BUG_ON(list_empty(&pagelist->head));
952 page = list_first_entry(&pagelist->head, struct page, lru);
953
Alex Elderca8b3a62013-04-05 14:46:01 -0500954 cursor->resid = min(length, pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600955 cursor->page = page;
956 cursor->offset = 0;
Alex Eldera51b272e2013-04-19 15:34:49 -0500957 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600958}
959
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500960static struct page *
961ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
962 size_t *page_offset, size_t *length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600963{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500964 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600965 struct ceph_pagelist *pagelist;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600966
967 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
968
969 pagelist = data->pagelist;
970 BUG_ON(!pagelist);
971
972 BUG_ON(!cursor->page);
Alex Elder25aff7c2013-03-11 23:34:22 -0500973 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600974
Alex Elder5df521b2013-03-30 15:09:59 -0500975 /* offset of first page in pagelist is always 0 */
Alex Elderfe38a2b2013-03-06 23:39:39 -0600976 *page_offset = cursor->offset & ~PAGE_MASK;
Alex Elder5df521b2013-03-30 15:09:59 -0500977 if (cursor->last_piece)
Alex Elder25aff7c2013-03-11 23:34:22 -0500978 *length = cursor->resid;
979 else
980 *length = PAGE_SIZE - *page_offset;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600981
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500982 return cursor->page;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600983}
984
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500985static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
Alex Elderdd236fc2013-03-06 23:39:39 -0600986 size_t bytes)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600987{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500988 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600989 struct ceph_pagelist *pagelist;
990
991 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
992
993 pagelist = data->pagelist;
994 BUG_ON(!pagelist);
Alex Elder25aff7c2013-03-11 23:34:22 -0500995
996 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600997 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
998
999 /* Advance the cursor offset */
1000
Alex Elder25aff7c2013-03-11 23:34:22 -05001001 cursor->resid -= bytes;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001002 cursor->offset += bytes;
Alex Elder5df521b2013-03-30 15:09:59 -05001003 /* offset of first page in pagelist is always 0 */
Alex Elderfe38a2b2013-03-06 23:39:39 -06001004 if (!bytes || cursor->offset & ~PAGE_MASK)
1005 return false; /* more bytes to process in the current page */
1006
1007 /* Move on to the next page */
1008
1009 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1010 cursor->page = list_entry_next(cursor->page, lru);
Alex Elder25aff7c2013-03-11 23:34:22 -05001011 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001012
1013 return true;
1014}
1015
Alex Elderdd236fc2013-03-06 23:39:39 -06001016/*
1017 * Message data is handled (sent or received) in pieces, where each
1018 * piece resides on a single page. The network layer might not
1019 * consume an entire piece at once. A data item's cursor keeps
1020 * track of which piece is next to process and how much remains to
1021 * be processed in that piece. It also tracks whether the current
1022 * piece is the last one in the data item.
1023 */
Alex Elderca8b3a62013-04-05 14:46:01 -05001024static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
Alex Elderdd236fc2013-03-06 23:39:39 -06001025{
Alex Elderca8b3a62013-04-05 14:46:01 -05001026 size_t length = cursor->total_resid;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001027
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001028 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001029 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001030 ceph_msg_data_pagelist_cursor_init(cursor, length);
Alex Elderdd236fc2013-03-06 23:39:39 -06001031 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001032 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001033 ceph_msg_data_pages_cursor_init(cursor, length);
Alex Eldere766d7b2013-03-07 15:38:28 -06001034 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001035#ifdef CONFIG_BLOCK
1036 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001037 ceph_msg_data_bio_cursor_init(cursor, length);
Alex Elder6aaa4512013-03-06 23:39:39 -06001038 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001039#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001040 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001041 default:
1042 /* BUG(); */
1043 break;
1044 }
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001045 cursor->need_crc = true;
Alex Elderdd236fc2013-03-06 23:39:39 -06001046}
1047
Alex Elderca8b3a62013-04-05 14:46:01 -05001048static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1049{
1050 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1051 struct ceph_msg_data *data;
1052
1053 BUG_ON(!length);
1054 BUG_ON(length > msg->data_length);
1055 BUG_ON(list_empty(&msg->data));
1056
Alex Elderca8b3a62013-04-05 14:46:01 -05001057 cursor->data_head = &msg->data;
1058 cursor->total_resid = length;
1059 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1060 cursor->data = data;
1061
1062 __ceph_msg_data_cursor_init(cursor);
1063}
1064
Alex Elderdd236fc2013-03-06 23:39:39 -06001065/*
1066 * Return the page containing the next piece to process for a given
1067 * data item, and supply the page offset and length of that piece.
1068 * Indicate whether this is the last piece in this data item.
1069 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001070static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1071 size_t *page_offset, size_t *length,
Alex Elderdd236fc2013-03-06 23:39:39 -06001072 bool *last_piece)
1073{
1074 struct page *page;
1075
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001076 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001077 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001078 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
Alex Elderdd236fc2013-03-06 23:39:39 -06001079 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001080 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001081 page = ceph_msg_data_pages_next(cursor, page_offset, length);
Alex Eldere766d7b2013-03-07 15:38:28 -06001082 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001083#ifdef CONFIG_BLOCK
1084 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001085 page = ceph_msg_data_bio_next(cursor, page_offset, length);
Alex Elder6aaa4512013-03-06 23:39:39 -06001086 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001087#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001088 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001089 default:
1090 page = NULL;
1091 break;
1092 }
1093 BUG_ON(!page);
1094 BUG_ON(*page_offset + *length > PAGE_SIZE);
1095 BUG_ON(!*length);
1096 if (last_piece)
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001097 *last_piece = cursor->last_piece;
Alex Elderdd236fc2013-03-06 23:39:39 -06001098
1099 return page;
1100}
1101
1102/*
1103 * Returns true if the result moves the cursor on to the next piece
1104 * of the data item.
1105 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001106static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1107 size_t bytes)
Alex Elderdd236fc2013-03-06 23:39:39 -06001108{
1109 bool new_piece;
1110
Alex Elder25aff7c2013-03-11 23:34:22 -05001111 BUG_ON(bytes > cursor->resid);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001112 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001113 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001114 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
Alex Elderdd236fc2013-03-06 23:39:39 -06001115 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001116 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001117 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
Alex Eldere766d7b2013-03-07 15:38:28 -06001118 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001119#ifdef CONFIG_BLOCK
1120 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001121 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
Alex Elder6aaa4512013-03-06 23:39:39 -06001122 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001123#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001124 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001125 default:
1126 BUG();
1127 break;
1128 }
Alex Elderca8b3a62013-04-05 14:46:01 -05001129 cursor->total_resid -= bytes;
Alex Elderdd236fc2013-03-06 23:39:39 -06001130
Alex Elderca8b3a62013-04-05 14:46:01 -05001131 if (!cursor->resid && cursor->total_resid) {
1132 WARN_ON(!cursor->last_piece);
1133 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1134 cursor->data = list_entry_next(cursor->data, links);
1135 __ceph_msg_data_cursor_init(cursor);
Alex Eldera51b272e2013-04-19 15:34:49 -05001136 new_piece = true;
Alex Elderca8b3a62013-04-05 14:46:01 -05001137 }
Alex Eldera51b272e2013-04-19 15:34:49 -05001138 cursor->need_crc = new_piece;
Alex Elderca8b3a62013-04-05 14:46:01 -05001139
Alex Elderdd236fc2013-03-06 23:39:39 -06001140 return new_piece;
1141}
1142
Alex Elder98fa5dd2013-04-02 12:09:50 -05001143static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
Alex Elder739c9052012-06-11 14:57:13 -05001144{
Alex Elder739c9052012-06-11 14:57:13 -05001145 BUG_ON(!msg);
Alex Elder25aff7c2013-03-11 23:34:22 -05001146 BUG_ON(!data_len);
Alex Elder739c9052012-06-11 14:57:13 -05001147
Alex Elder4c59b4a2013-03-11 23:34:23 -05001148 /* Initialize data cursor */
Alex Elderfe38a2b2013-03-06 23:39:39 -06001149
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001150 ceph_msg_data_cursor_init(msg, (size_t)data_len);
Alex Elder739c9052012-06-11 14:57:13 -05001151}
1152
Sage Weil31b80062009-10-06 11:31:13 -07001153/*
1154 * Prepare footer for currently outgoing message, and finish things
1155 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1156 */
Alex Elder859eb792012-02-14 14:05:33 -06001157static void prepare_write_message_footer(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001158{
1159 struct ceph_msg *m = con->out_msg;
Alex Elder859eb792012-02-14 14:05:33 -06001160 int v = con->out_kvec_left;
Sage Weil31b80062009-10-06 11:31:13 -07001161
Alex Elderfd154f32012-06-11 14:57:13 -05001162 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1163
Sage Weil31b80062009-10-06 11:31:13 -07001164 dout("prepare_write_message_footer %p\n", con);
1165 con->out_kvec_is_msg = true;
1166 con->out_kvec[v].iov_base = &m->footer;
1167 con->out_kvec[v].iov_len = sizeof(m->footer);
1168 con->out_kvec_bytes += sizeof(m->footer);
1169 con->out_kvec_left++;
1170 con->out_more = m->more_to_follow;
Sage Weilc86a2932009-12-14 14:04:30 -08001171 con->out_msg_done = true;
Sage Weil31b80062009-10-06 11:31:13 -07001172}
1173
1174/*
1175 * Prepare headers for the next outgoing message.
1176 */
1177static void prepare_write_message(struct ceph_connection *con)
1178{
1179 struct ceph_msg *m;
Alex Eldera9a0c512012-02-15 07:43:54 -06001180 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07001181
Alex Eldere2200422012-05-23 14:35:23 -05001182 con_out_kvec_reset(con);
Sage Weil31b80062009-10-06 11:31:13 -07001183 con->out_kvec_is_msg = true;
Sage Weilc86a2932009-12-14 14:04:30 -08001184 con->out_msg_done = false;
Sage Weil31b80062009-10-06 11:31:13 -07001185
1186 /* Sneak an ack in there first? If we can get it into the same
1187 * TCP packet that's a good thing. */
1188 if (con->in_seq > con->in_seq_acked) {
1189 con->in_seq_acked = con->in_seq;
Alex Eldere2200422012-05-23 14:35:23 -05001190 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
Sage Weil31b80062009-10-06 11:31:13 -07001191 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
Alex Eldere2200422012-05-23 14:35:23 -05001192 con_out_kvec_add(con, sizeof (con->out_temp_ack),
Alex Elder859eb792012-02-14 14:05:33 -06001193 &con->out_temp_ack);
Sage Weil31b80062009-10-06 11:31:13 -07001194 }
1195
Alex Elder38941f82012-06-01 14:56:43 -05001196 BUG_ON(list_empty(&con->out_queue));
Alex Elder859eb792012-02-14 14:05:33 -06001197 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
Sage Weilc86a2932009-12-14 14:04:30 -08001198 con->out_msg = m;
Alex Elder38941f82012-06-01 14:56:43 -05001199 BUG_ON(m->con != con);
Sage Weil4cf9d542011-07-26 11:27:24 -07001200
1201 /* put message on sent list */
1202 ceph_msg_get(m);
1203 list_move_tail(&m->list_head, &con->out_sent);
Sage Weil31b80062009-10-06 11:31:13 -07001204
Sage Weile84346b2010-05-11 21:20:38 -07001205 /*
1206 * only assign outgoing seq # if we haven't sent this message
1207 * yet. if it is requeued, resend with it's original seq.
1208 */
1209 if (m->needs_out_seq) {
1210 m->hdr.seq = cpu_to_le64(++con->out_seq);
1211 m->needs_out_seq = false;
1212 }
Alex Elder98fa5dd2013-04-02 12:09:50 -05001213 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
Sage Weil31b80062009-10-06 11:31:13 -07001214
Alex Elder98fa5dd2013-04-02 12:09:50 -05001215 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
Sage Weil31b80062009-10-06 11:31:13 -07001216 m, con->out_seq, le16_to_cpu(m->hdr.type),
1217 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
Alex Elder98fa5dd2013-04-02 12:09:50 -05001218 m->data_length);
Sage Weil31b80062009-10-06 11:31:13 -07001219 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1220
1221 /* tag + hdr + front + middle */
Alex Eldere2200422012-05-23 14:35:23 -05001222 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1223 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1224 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
Alex Elder859eb792012-02-14 14:05:33 -06001225
Sage Weil31b80062009-10-06 11:31:13 -07001226 if (m->middle)
Alex Eldere2200422012-05-23 14:35:23 -05001227 con_out_kvec_add(con, m->middle->vec.iov_len,
Alex Elder859eb792012-02-14 14:05:33 -06001228 m->middle->vec.iov_base);
Sage Weil31b80062009-10-06 11:31:13 -07001229
1230 /* fill in crc (except data pages), footer */
Alex Eldera9a0c512012-02-15 07:43:54 -06001231 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1232 con->out_msg->hdr.crc = cpu_to_le32(crc);
Alex Elderfd154f32012-06-11 14:57:13 -05001233 con->out_msg->footer.flags = 0;
Alex Eldera9a0c512012-02-15 07:43:54 -06001234
1235 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1236 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1237 if (m->middle) {
1238 crc = crc32c(0, m->middle->vec.iov_base,
1239 m->middle->vec.iov_len);
1240 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1241 } else
Sage Weil31b80062009-10-06 11:31:13 -07001242 con->out_msg->footer.middle_crc = 0;
Alex Elder739c9052012-06-11 14:57:13 -05001243 dout("%s front_crc %u middle_crc %u\n", __func__,
Sage Weil31b80062009-10-06 11:31:13 -07001244 le32_to_cpu(con->out_msg->footer.front_crc),
1245 le32_to_cpu(con->out_msg->footer.middle_crc));
1246
1247 /* is there a data payload? */
Alex Elder739c9052012-06-11 14:57:13 -05001248 con->out_msg->footer.data_crc = 0;
Alex Elder98fa5dd2013-04-02 12:09:50 -05001249 if (m->data_length) {
1250 prepare_message_data(con->out_msg, m->data_length);
Alex Elder78625052013-03-06 23:39:39 -06001251 con->out_more = 1; /* data + footer will follow */
1252 } else {
Sage Weil31b80062009-10-06 11:31:13 -07001253 /* no, queue up footer too and be done */
Alex Elder859eb792012-02-14 14:05:33 -06001254 prepare_write_message_footer(con);
Alex Elder78625052013-03-06 23:39:39 -06001255 }
Sage Weil31b80062009-10-06 11:31:13 -07001256
Alex Elderc9ffc772013-02-20 10:25:12 -06001257 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001258}
1259
1260/*
1261 * Prepare an ack.
1262 */
1263static void prepare_write_ack(struct ceph_connection *con)
1264{
1265 dout("prepare_write_ack %p %llu -> %llu\n", con,
1266 con->in_seq_acked, con->in_seq);
1267 con->in_seq_acked = con->in_seq;
1268
Alex Eldere2200422012-05-23 14:35:23 -05001269 con_out_kvec_reset(con);
Alex Elder859eb792012-02-14 14:05:33 -06001270
Alex Eldere2200422012-05-23 14:35:23 -05001271 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
Alex Elder859eb792012-02-14 14:05:33 -06001272
Sage Weil31b80062009-10-06 11:31:13 -07001273 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
Alex Eldere2200422012-05-23 14:35:23 -05001274 con_out_kvec_add(con, sizeof (con->out_temp_ack),
Alex Elder859eb792012-02-14 14:05:33 -06001275 &con->out_temp_ack);
1276
Sage Weil31b80062009-10-06 11:31:13 -07001277 con->out_more = 1; /* more will follow.. eventually.. */
Alex Elderc9ffc772013-02-20 10:25:12 -06001278 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001279}
1280
1281/*
Sage Weil3a230832013-03-25 08:47:40 -07001282 * Prepare to share the seq during handshake
1283 */
1284static void prepare_write_seq(struct ceph_connection *con)
1285{
1286 dout("prepare_write_seq %p %llu -> %llu\n", con,
1287 con->in_seq_acked, con->in_seq);
1288 con->in_seq_acked = con->in_seq;
1289
1290 con_out_kvec_reset(con);
1291
1292 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1293 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1294 &con->out_temp_ack);
1295
1296 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1297}
1298
1299/*
Sage Weil31b80062009-10-06 11:31:13 -07001300 * Prepare to write keepalive byte.
1301 */
1302static void prepare_write_keepalive(struct ceph_connection *con)
1303{
1304 dout("prepare_write_keepalive %p\n", con);
Alex Eldere2200422012-05-23 14:35:23 -05001305 con_out_kvec_reset(con);
1306 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
Alex Elderc9ffc772013-02-20 10:25:12 -06001307 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001308}
1309
1310/*
1311 * Connection negotiation.
1312 */
1313
Alex Elderdac1e712012-05-16 15:16:39 -05001314static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1315 int *auth_proto)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001316{
Alex Eldera3530df2012-05-16 15:16:39 -05001317 struct ceph_auth_handshake *auth;
Alex Elderb1c6b982012-05-16 15:16:38 -05001318
1319 if (!con->ops->get_authorizer) {
1320 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1321 con->out_connect.authorizer_len = 0;
Alex Elder729796b2012-05-16 15:16:39 -05001322 return NULL;
Alex Elderb1c6b982012-05-16 15:16:38 -05001323 }
1324
1325 /* Can't hold the mutex while getting authorizer */
Sage Weilec302642009-12-22 10:43:42 -08001326 mutex_unlock(&con->mutex);
Alex Elderdac1e712012-05-16 15:16:39 -05001327 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
Sage Weilec302642009-12-22 10:43:42 -08001328 mutex_lock(&con->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001329
Alex Eldera3530df2012-05-16 15:16:39 -05001330 if (IS_ERR(auth))
Alex Elder729796b2012-05-16 15:16:39 -05001331 return auth;
Sage Weil8dacc7d2012-07-20 17:24:40 -07001332 if (con->state != CON_STATE_NEGOTIATING)
Alex Elder729796b2012-05-16 15:16:39 -05001333 return ERR_PTR(-EAGAIN);
Sage Weil0da5d702011-05-19 11:21:05 -07001334
Alex Elder8f43fb52012-05-16 15:16:39 -05001335 con->auth_reply_buf = auth->authorizer_reply_buf;
1336 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
Alex Elder729796b2012-05-16 15:16:39 -05001337 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001338}
1339
Sage Weil31b80062009-10-06 11:31:13 -07001340/*
1341 * We connected to a peer and are saying hello.
1342 */
Alex Eldere825a662012-05-16 15:16:38 -05001343static void prepare_write_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001344{
Alex Eldere2200422012-05-23 14:35:23 -05001345 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1346 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
Alex Eldere825a662012-05-16 15:16:38 -05001347 &con->msgr->my_enc_addr);
Sage Weileed0ef22009-11-10 14:34:36 -08001348
Sage Weileed0ef22009-11-10 14:34:36 -08001349 con->out_more = 0;
Alex Elderc9ffc772013-02-20 10:25:12 -06001350 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weileed0ef22009-11-10 14:34:36 -08001351}
1352
Alex Eldere825a662012-05-16 15:16:38 -05001353static int prepare_write_connect(struct ceph_connection *con)
Sage Weileed0ef22009-11-10 14:34:36 -08001354{
Eric Dumazet95c96172012-04-15 05:58:06 +00001355 unsigned int global_seq = get_global_seq(con->msgr, 0);
Sage Weil31b80062009-10-06 11:31:13 -07001356 int proto;
Alex Elderdac1e712012-05-16 15:16:39 -05001357 int auth_proto;
Alex Elder729796b2012-05-16 15:16:39 -05001358 struct ceph_auth_handshake *auth;
Sage Weil31b80062009-10-06 11:31:13 -07001359
1360 switch (con->peer_name.type) {
1361 case CEPH_ENTITY_TYPE_MON:
1362 proto = CEPH_MONC_PROTOCOL;
1363 break;
1364 case CEPH_ENTITY_TYPE_OSD:
1365 proto = CEPH_OSDC_PROTOCOL;
1366 break;
1367 case CEPH_ENTITY_TYPE_MDS:
1368 proto = CEPH_MDSC_PROTOCOL;
1369 break;
1370 default:
1371 BUG();
1372 }
1373
1374 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1375 con->connect_seq, global_seq, proto);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001376
Alex Eldere825a662012-05-16 15:16:38 -05001377 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
Sage Weil31b80062009-10-06 11:31:13 -07001378 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1379 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1380 con->out_connect.global_seq = cpu_to_le32(global_seq);
1381 con->out_connect.protocol_version = cpu_to_le32(proto);
1382 con->out_connect.flags = 0;
Sage Weil31b80062009-10-06 11:31:13 -07001383
Alex Elderdac1e712012-05-16 15:16:39 -05001384 auth_proto = CEPH_AUTH_UNKNOWN;
1385 auth = get_connect_authorizer(con, &auth_proto);
Alex Elder729796b2012-05-16 15:16:39 -05001386 if (IS_ERR(auth))
1387 return PTR_ERR(auth);
Alex Elder3da54772012-05-16 15:16:39 -05001388
Alex Elderdac1e712012-05-16 15:16:39 -05001389 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
Alex Elder3da54772012-05-16 15:16:39 -05001390 con->out_connect.authorizer_len = auth ?
1391 cpu_to_le32(auth->authorizer_buf_len) : 0;
1392
Alex Eldere2200422012-05-23 14:35:23 -05001393 con_out_kvec_add(con, sizeof (con->out_connect),
Alex Elder3da54772012-05-16 15:16:39 -05001394 &con->out_connect);
1395 if (auth && auth->authorizer_buf_len)
Alex Eldere2200422012-05-23 14:35:23 -05001396 con_out_kvec_add(con, auth->authorizer_buf_len,
Alex Elder3da54772012-05-16 15:16:39 -05001397 auth->authorizer_buf);
Alex Elder859eb792012-02-14 14:05:33 -06001398
Sage Weil31b80062009-10-06 11:31:13 -07001399 con->out_more = 0;
Alex Elderc9ffc772013-02-20 10:25:12 -06001400 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001401
Alex Eldere10c7582012-05-16 15:16:38 -05001402 return 0;
Sage Weil31b80062009-10-06 11:31:13 -07001403}
1404
Sage Weil31b80062009-10-06 11:31:13 -07001405/*
1406 * write as much of pending kvecs to the socket as we can.
1407 * 1 -> done
1408 * 0 -> socket full, but more to do
1409 * <0 -> error
1410 */
1411static int write_partial_kvec(struct ceph_connection *con)
1412{
1413 int ret;
1414
1415 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1416 while (con->out_kvec_bytes > 0) {
1417 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1418 con->out_kvec_left, con->out_kvec_bytes,
1419 con->out_more);
1420 if (ret <= 0)
1421 goto out;
1422 con->out_kvec_bytes -= ret;
1423 if (con->out_kvec_bytes == 0)
1424 break; /* done */
Alex Elderf42299e2012-02-15 07:43:54 -06001425
1426 /* account for full iov entries consumed */
1427 while (ret >= con->out_kvec_cur->iov_len) {
1428 BUG_ON(!con->out_kvec_left);
1429 ret -= con->out_kvec_cur->iov_len;
1430 con->out_kvec_cur++;
1431 con->out_kvec_left--;
1432 }
1433 /* and for a partially-consumed entry */
1434 if (ret) {
1435 con->out_kvec_cur->iov_len -= ret;
1436 con->out_kvec_cur->iov_base += ret;
Sage Weil31b80062009-10-06 11:31:13 -07001437 }
1438 }
1439 con->out_kvec_left = 0;
1440 con->out_kvec_is_msg = false;
1441 ret = 1;
1442out:
1443 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1444 con->out_kvec_bytes, con->out_kvec_left, ret);
1445 return ret; /* done! */
1446}
1447
Alex Elder35b62802013-03-08 20:59:00 -06001448static u32 ceph_crc32c_page(u32 crc, struct page *page,
1449 unsigned int page_offset,
1450 unsigned int length)
1451{
1452 char *kaddr;
1453
1454 kaddr = kmap(page);
1455 BUG_ON(kaddr == NULL);
1456 crc = crc32c(crc, kaddr + page_offset, length);
1457 kunmap(page);
1458
1459 return crc;
1460}
Sage Weil31b80062009-10-06 11:31:13 -07001461/*
1462 * Write as much message data payload as we can. If we finish, queue
1463 * up the footer.
1464 * 1 -> done, footer is now queued in out_kvec[].
1465 * 0 -> socket full, but more to do
1466 * <0 -> error
1467 */
Alex Elder34d2d202013-03-08 20:58:59 -06001468static int write_partial_message_data(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001469{
1470 struct ceph_msg *msg = con->out_msg;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001471 struct ceph_msg_data_cursor *cursor = &msg->cursor;
Alex Elder37675b02012-03-07 11:40:08 -06001472 bool do_datacrc = !con->msgr->nocrc;
Alex Elderf5db90b2013-03-11 23:34:23 -05001473 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07001474
Alex Elder859a35d2013-03-11 23:34:23 -05001475 dout("%s %p msg %p\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07001476
Alex Elder5240d9f2013-03-14 14:09:06 -05001477 if (list_empty(&msg->data))
Alex Elder4c59b4a2013-03-11 23:34:23 -05001478 return -EINVAL;
1479
Alex Elder5821bd82012-06-11 14:57:13 -05001480 /*
1481 * Iterate through each page that contains data to be
1482 * written, and send as much as possible for each.
1483 *
1484 * If we are calculating the data crc (the default), we will
1485 * need to map the page. If we have no pages, they have
1486 * been revoked, so use the zero page.
1487 */
Alex Elderf5db90b2013-03-11 23:34:23 -05001488 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
Alex Elder643c68a2013-03-11 23:34:23 -05001489 while (cursor->resid) {
Alex Elder8a166d02013-03-08 13:35:36 -06001490 struct page *page;
Alex Eldere387d522013-03-06 23:39:38 -06001491 size_t page_offset;
1492 size_t length;
Alex Elder8a166d02013-03-08 13:35:36 -06001493 bool last_piece;
Alex Elder8ea299b2013-03-11 23:34:23 -05001494 bool need_crc;
Alex Elderf5db90b2013-03-11 23:34:23 -05001495 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07001496
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001497 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
Alex Elder4c59b4a2013-03-11 23:34:23 -05001498 &last_piece);
Alex Eldere387d522013-03-06 23:39:38 -06001499 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
Alex Elderfe38a2b2013-03-06 23:39:39 -06001500 length, last_piece);
Alex Elderf5db90b2013-03-11 23:34:23 -05001501 if (ret <= 0) {
1502 if (do_datacrc)
1503 msg->footer.data_crc = cpu_to_le32(crc);
Sage Weil31b80062009-10-06 11:31:13 -07001504
Alex Elderf5db90b2013-03-11 23:34:23 -05001505 return ret;
1506 }
Alex Elder143334f2013-03-29 11:44:10 -05001507 if (do_datacrc && cursor->need_crc)
1508 crc = ceph_crc32c_page(crc, page, page_offset, length);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001509 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
Sage Weil31b80062009-10-06 11:31:13 -07001510 }
1511
Alex Elder34d2d202013-03-08 20:58:59 -06001512 dout("%s %p msg %p done\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07001513
1514 /* prepare and queue up footer, too */
Alex Elderf5db90b2013-03-11 23:34:23 -05001515 if (do_datacrc)
1516 msg->footer.data_crc = cpu_to_le32(crc);
1517 else
Alex Elder84ca8fc2012-06-11 14:57:13 -05001518 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
Alex Eldere2200422012-05-23 14:35:23 -05001519 con_out_kvec_reset(con);
Alex Elder859eb792012-02-14 14:05:33 -06001520 prepare_write_message_footer(con);
Alex Elderf5db90b2013-03-11 23:34:23 -05001521
1522 return 1; /* must return > 0 to indicate success */
Sage Weil31b80062009-10-06 11:31:13 -07001523}
1524
1525/*
1526 * write some zeros
1527 */
1528static int write_partial_skip(struct ceph_connection *con)
1529{
1530 int ret;
1531
1532 while (con->out_skip > 0) {
Alex Elder31739132012-03-07 11:40:08 -06001533 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
Sage Weil31b80062009-10-06 11:31:13 -07001534
Alex Eldere1dcb122013-03-06 23:39:38 -06001535 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
Sage Weil31b80062009-10-06 11:31:13 -07001536 if (ret <= 0)
1537 goto out;
1538 con->out_skip -= ret;
1539 }
1540 ret = 1;
1541out:
1542 return ret;
1543}
1544
1545/*
1546 * Prepare to read connection handshake, or an ack.
1547 */
Sage Weileed0ef22009-11-10 14:34:36 -08001548static void prepare_read_banner(struct ceph_connection *con)
1549{
1550 dout("prepare_read_banner %p\n", con);
1551 con->in_base_pos = 0;
1552}
1553
Sage Weil31b80062009-10-06 11:31:13 -07001554static void prepare_read_connect(struct ceph_connection *con)
1555{
1556 dout("prepare_read_connect %p\n", con);
1557 con->in_base_pos = 0;
1558}
1559
1560static void prepare_read_ack(struct ceph_connection *con)
1561{
1562 dout("prepare_read_ack %p\n", con);
1563 con->in_base_pos = 0;
1564}
1565
Sage Weil3a230832013-03-25 08:47:40 -07001566static void prepare_read_seq(struct ceph_connection *con)
1567{
1568 dout("prepare_read_seq %p\n", con);
1569 con->in_base_pos = 0;
1570 con->in_tag = CEPH_MSGR_TAG_SEQ;
1571}
1572
Sage Weil31b80062009-10-06 11:31:13 -07001573static void prepare_read_tag(struct ceph_connection *con)
1574{
1575 dout("prepare_read_tag %p\n", con);
1576 con->in_base_pos = 0;
1577 con->in_tag = CEPH_MSGR_TAG_READY;
1578}
1579
1580/*
1581 * Prepare to read a message.
1582 */
1583static int prepare_read_message(struct ceph_connection *con)
1584{
1585 dout("prepare_read_message %p\n", con);
1586 BUG_ON(con->in_msg != NULL);
1587 con->in_base_pos = 0;
1588 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1589 return 0;
1590}
1591
1592
1593static int read_partial(struct ceph_connection *con,
Alex Elderfd516532012-05-10 10:29:50 -05001594 int end, int size, void *object)
Sage Weil31b80062009-10-06 11:31:13 -07001595{
Alex Eldere6cee712012-05-10 10:29:50 -05001596 while (con->in_base_pos < end) {
1597 int left = end - con->in_base_pos;
Sage Weil31b80062009-10-06 11:31:13 -07001598 int have = size - left;
1599 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1600 if (ret <= 0)
1601 return ret;
1602 con->in_base_pos += ret;
1603 }
1604 return 1;
1605}
1606
1607
1608/*
1609 * Read all or part of the connect-side handshake on a new connection
1610 */
Sage Weileed0ef22009-11-10 14:34:36 -08001611static int read_partial_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001612{
Alex Elderfd516532012-05-10 10:29:50 -05001613 int size;
1614 int end;
1615 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07001616
Sage Weileed0ef22009-11-10 14:34:36 -08001617 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
Sage Weil31b80062009-10-06 11:31:13 -07001618
1619 /* peer's banner */
Alex Elderfd516532012-05-10 10:29:50 -05001620 size = strlen(CEPH_BANNER);
1621 end = size;
1622 ret = read_partial(con, end, size, con->in_banner);
Sage Weil31b80062009-10-06 11:31:13 -07001623 if (ret <= 0)
1624 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001625
1626 size = sizeof (con->actual_peer_addr);
1627 end += size;
1628 ret = read_partial(con, end, size, &con->actual_peer_addr);
Sage Weil31b80062009-10-06 11:31:13 -07001629 if (ret <= 0)
1630 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001631
1632 size = sizeof (con->peer_addr_for_me);
1633 end += size;
1634 ret = read_partial(con, end, size, &con->peer_addr_for_me);
Sage Weil31b80062009-10-06 11:31:13 -07001635 if (ret <= 0)
1636 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001637
Sage Weileed0ef22009-11-10 14:34:36 -08001638out:
1639 return ret;
1640}
1641
1642static int read_partial_connect(struct ceph_connection *con)
1643{
Alex Elderfd516532012-05-10 10:29:50 -05001644 int size;
1645 int end;
1646 int ret;
Sage Weileed0ef22009-11-10 14:34:36 -08001647
1648 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1649
Alex Elderfd516532012-05-10 10:29:50 -05001650 size = sizeof (con->in_reply);
1651 end = size;
1652 ret = read_partial(con, end, size, &con->in_reply);
Sage Weil31b80062009-10-06 11:31:13 -07001653 if (ret <= 0)
1654 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001655
1656 size = le32_to_cpu(con->in_reply.authorizer_len);
1657 end += size;
1658 ret = read_partial(con, end, size, con->auth_reply_buf);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001659 if (ret <= 0)
1660 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07001661
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001662 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1663 con, (int)con->in_reply.tag,
1664 le32_to_cpu(con->in_reply.connect_seq),
Sage Weil31b80062009-10-06 11:31:13 -07001665 le32_to_cpu(con->in_reply.global_seq));
1666out:
1667 return ret;
Sage Weileed0ef22009-11-10 14:34:36 -08001668
Sage Weil31b80062009-10-06 11:31:13 -07001669}
1670
1671/*
1672 * Verify the hello banner looks okay.
1673 */
1674static int verify_hello(struct ceph_connection *con)
1675{
1676 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
Sage Weil13e38c82009-10-09 16:36:34 -07001677 pr_err("connect to %s got bad banner\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001678 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07001679 con->error_msg = "protocol error, bad banner";
1680 return -1;
1681 }
1682 return 0;
1683}
1684
1685static bool addr_is_blank(struct sockaddr_storage *ss)
1686{
1687 switch (ss->ss_family) {
1688 case AF_INET:
1689 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1690 case AF_INET6:
1691 return
1692 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1693 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1694 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1695 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1696 }
1697 return false;
1698}
1699
1700static int addr_port(struct sockaddr_storage *ss)
1701{
1702 switch (ss->ss_family) {
1703 case AF_INET:
Sage Weilf28bcfb2009-11-04 11:46:35 -08001704 return ntohs(((struct sockaddr_in *)ss)->sin_port);
Sage Weil31b80062009-10-06 11:31:13 -07001705 case AF_INET6:
Sage Weilf28bcfb2009-11-04 11:46:35 -08001706 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
Sage Weil31b80062009-10-06 11:31:13 -07001707 }
1708 return 0;
1709}
1710
1711static void addr_set_port(struct sockaddr_storage *ss, int p)
1712{
1713 switch (ss->ss_family) {
1714 case AF_INET:
1715 ((struct sockaddr_in *)ss)->sin_port = htons(p);
Sage Weila2a79602011-05-12 15:34:24 -07001716 break;
Sage Weil31b80062009-10-06 11:31:13 -07001717 case AF_INET6:
1718 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
Sage Weila2a79602011-05-12 15:34:24 -07001719 break;
Sage Weil31b80062009-10-06 11:31:13 -07001720 }
1721}
1722
1723/*
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001724 * Unlike other *_pton function semantics, zero indicates success.
1725 */
1726static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1727 char delim, const char **ipend)
1728{
Alex Elder99f0f3b2012-01-23 15:49:27 -06001729 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1730 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001731
1732 memset(ss, 0, sizeof(*ss));
1733
1734 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1735 ss->ss_family = AF_INET;
1736 return 0;
1737 }
1738
1739 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1740 ss->ss_family = AF_INET6;
1741 return 0;
1742 }
1743
1744 return -EINVAL;
1745}
1746
1747/*
1748 * Extract hostname string and resolve using kernel DNS facility.
1749 */
1750#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1751static int ceph_dns_resolve_name(const char *name, size_t namelen,
1752 struct sockaddr_storage *ss, char delim, const char **ipend)
1753{
1754 const char *end, *delim_p;
1755 char *colon_p, *ip_addr = NULL;
1756 int ip_len, ret;
1757
1758 /*
1759 * The end of the hostname occurs immediately preceding the delimiter or
1760 * the port marker (':') where the delimiter takes precedence.
1761 */
1762 delim_p = memchr(name, delim, namelen);
1763 colon_p = memchr(name, ':', namelen);
1764
1765 if (delim_p && colon_p)
1766 end = delim_p < colon_p ? delim_p : colon_p;
1767 else if (!delim_p && colon_p)
1768 end = colon_p;
1769 else {
1770 end = delim_p;
1771 if (!end) /* case: hostname:/ */
1772 end = name + namelen;
1773 }
1774
1775 if (end <= name)
1776 return -EINVAL;
1777
1778 /* do dns_resolve upcall */
1779 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1780 if (ip_len > 0)
1781 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1782 else
1783 ret = -ESRCH;
1784
1785 kfree(ip_addr);
1786
1787 *ipend = end;
1788
1789 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1790 ret, ret ? "failed" : ceph_pr_addr(ss));
1791
1792 return ret;
1793}
1794#else
1795static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1796 struct sockaddr_storage *ss, char delim, const char **ipend)
1797{
1798 return -EINVAL;
1799}
1800#endif
1801
1802/*
1803 * Parse a server name (IP or hostname). If a valid IP address is not found
1804 * then try to extract a hostname to resolve using userspace DNS upcall.
1805 */
1806static int ceph_parse_server_name(const char *name, size_t namelen,
1807 struct sockaddr_storage *ss, char delim, const char **ipend)
1808{
1809 int ret;
1810
1811 ret = ceph_pton(name, namelen, ss, delim, ipend);
1812 if (ret)
1813 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1814
1815 return ret;
1816}
1817
1818/*
Sage Weil31b80062009-10-06 11:31:13 -07001819 * Parse an ip[:port] list into an addr array. Use the default
1820 * monitor port if a port isn't specified.
1821 */
1822int ceph_parse_ips(const char *c, const char *end,
1823 struct ceph_entity_addr *addr,
1824 int max_count, int *count)
1825{
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001826 int i, ret = -EINVAL;
Sage Weil31b80062009-10-06 11:31:13 -07001827 const char *p = c;
1828
1829 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1830 for (i = 0; i < max_count; i++) {
1831 const char *ipend;
1832 struct sockaddr_storage *ss = &addr[i].in_addr;
Sage Weil31b80062009-10-06 11:31:13 -07001833 int port;
Sage Weil39139f62010-07-08 09:54:52 -07001834 char delim = ',';
1835
1836 if (*p == '[') {
1837 delim = ']';
1838 p++;
1839 }
Sage Weil31b80062009-10-06 11:31:13 -07001840
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001841 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1842 if (ret)
Sage Weil31b80062009-10-06 11:31:13 -07001843 goto bad;
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001844 ret = -EINVAL;
1845
Sage Weil31b80062009-10-06 11:31:13 -07001846 p = ipend;
1847
Sage Weil39139f62010-07-08 09:54:52 -07001848 if (delim == ']') {
1849 if (*p != ']') {
1850 dout("missing matching ']'\n");
1851 goto bad;
1852 }
1853 p++;
1854 }
1855
Sage Weil31b80062009-10-06 11:31:13 -07001856 /* port? */
1857 if (p < end && *p == ':') {
1858 port = 0;
1859 p++;
1860 while (p < end && *p >= '0' && *p <= '9') {
1861 port = (port * 10) + (*p - '0');
1862 p++;
1863 }
Ilya Dryomovf48db1e2013-12-30 19:21:29 +02001864 if (port == 0)
1865 port = CEPH_MON_PORT;
1866 else if (port > 65535)
Sage Weil31b80062009-10-06 11:31:13 -07001867 goto bad;
1868 } else {
1869 port = CEPH_MON_PORT;
1870 }
1871
1872 addr_set_port(ss, port);
1873
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001874 dout("parse_ips got %s\n", ceph_pr_addr(ss));
Sage Weil31b80062009-10-06 11:31:13 -07001875
1876 if (p == end)
1877 break;
1878 if (*p != ',')
1879 goto bad;
1880 p++;
1881 }
1882
1883 if (p != end)
1884 goto bad;
1885
1886 if (count)
1887 *count = i + 1;
1888 return 0;
1889
1890bad:
Sage Weil39139f62010-07-08 09:54:52 -07001891 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001892 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07001893}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001894EXPORT_SYMBOL(ceph_parse_ips);
Sage Weil31b80062009-10-06 11:31:13 -07001895
Sage Weileed0ef22009-11-10 14:34:36 -08001896static int process_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001897{
Sage Weileed0ef22009-11-10 14:34:36 -08001898 dout("process_banner on %p\n", con);
Sage Weil31b80062009-10-06 11:31:13 -07001899
1900 if (verify_hello(con) < 0)
1901 return -1;
1902
Sage Weil63f2d212009-11-03 15:17:56 -08001903 ceph_decode_addr(&con->actual_peer_addr);
1904 ceph_decode_addr(&con->peer_addr_for_me);
1905
Sage Weil31b80062009-10-06 11:31:13 -07001906 /*
1907 * Make sure the other end is who we wanted. note that the other
1908 * end may not yet know their ip address, so if it's 0.0.0.0, give
1909 * them the benefit of the doubt.
1910 */
Sage Weil103e2d32010-01-07 16:12:36 -08001911 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1912 sizeof(con->peer_addr)) != 0 &&
Sage Weil31b80062009-10-06 11:31:13 -07001913 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1914 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001915 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001916 ceph_pr_addr(&con->peer_addr.in_addr),
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001917 (int)le32_to_cpu(con->peer_addr.nonce),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001918 ceph_pr_addr(&con->actual_peer_addr.in_addr),
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001919 (int)le32_to_cpu(con->actual_peer_addr.nonce));
Sage Weil58bb3b32009-12-23 12:12:31 -08001920 con->error_msg = "wrong peer at address";
Sage Weil31b80062009-10-06 11:31:13 -07001921 return -1;
1922 }
1923
1924 /*
1925 * did we learn our address?
1926 */
1927 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1928 int port = addr_port(&con->msgr->inst.addr.in_addr);
1929
1930 memcpy(&con->msgr->inst.addr.in_addr,
1931 &con->peer_addr_for_me.in_addr,
1932 sizeof(con->peer_addr_for_me.in_addr));
1933 addr_set_port(&con->msgr->inst.addr.in_addr, port);
Sage Weil63f2d212009-11-03 15:17:56 -08001934 encode_my_addr(con->msgr);
Sage Weileed0ef22009-11-10 14:34:36 -08001935 dout("process_banner learned my addr is %s\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001936 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07001937 }
1938
Sage Weileed0ef22009-11-10 14:34:36 -08001939 return 0;
1940}
1941
1942static int process_connect(struct ceph_connection *con)
1943{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001944 u64 sup_feat = con->msgr->supported_features;
1945 u64 req_feat = con->msgr->required_features;
Ilya Dryomov2b3e0c92013-12-24 21:19:24 +02001946 u64 server_feat = ceph_sanitize_features(
1947 le64_to_cpu(con->in_reply.features));
Sage Weil0da5d702011-05-19 11:21:05 -07001948 int ret;
Sage Weil04a419f2009-12-23 09:30:21 -08001949
Sage Weileed0ef22009-11-10 14:34:36 -08001950 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1951
Sage Weil31b80062009-10-06 11:31:13 -07001952 switch (con->in_reply.tag) {
Sage Weil04a419f2009-12-23 09:30:21 -08001953 case CEPH_MSGR_TAG_FEATURES:
1954 pr_err("%s%lld %s feature set mismatch,"
1955 " my %llx < server's %llx, missing %llx\n",
1956 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001957 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil04a419f2009-12-23 09:30:21 -08001958 sup_feat, server_feat, server_feat & ~sup_feat);
1959 con->error_msg = "missing required protocol features";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06001960 reset_connection(con);
Sage Weil04a419f2009-12-23 09:30:21 -08001961 return -1;
1962
Sage Weil31b80062009-10-06 11:31:13 -07001963 case CEPH_MSGR_TAG_BADPROTOVER:
Sage Weil31b80062009-10-06 11:31:13 -07001964 pr_err("%s%lld %s protocol version mismatch,"
1965 " my %d != server's %d\n",
1966 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001967 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil31b80062009-10-06 11:31:13 -07001968 le32_to_cpu(con->out_connect.protocol_version),
1969 le32_to_cpu(con->in_reply.protocol_version));
1970 con->error_msg = "protocol version mismatch";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06001971 reset_connection(con);
Sage Weil31b80062009-10-06 11:31:13 -07001972 return -1;
1973
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001974 case CEPH_MSGR_TAG_BADAUTHORIZER:
1975 con->auth_retry++;
1976 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1977 con->auth_retry);
1978 if (con->auth_retry == 2) {
1979 con->error_msg = "connect authorization failure";
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001980 return -1;
1981 }
Jim Schutt6d4221b2012-08-10 10:37:38 -07001982 con_out_kvec_reset(con);
Alex Eldere825a662012-05-16 15:16:38 -05001983 ret = prepare_write_connect(con);
Sage Weil0da5d702011-05-19 11:21:05 -07001984 if (ret < 0)
1985 return ret;
Sage Weil63733a02010-03-15 15:47:22 -07001986 prepare_read_connect(con);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001987 break;
Sage Weil31b80062009-10-06 11:31:13 -07001988
1989 case CEPH_MSGR_TAG_RESETSESSION:
1990 /*
1991 * If we connected with a large connect_seq but the peer
1992 * has no record of a session with us (no connection, or
1993 * connect_seq == 0), they will send RESETSESION to indicate
1994 * that they must have reset their session, and may have
1995 * dropped messages.
1996 */
1997 dout("process_connect got RESET peer seq %u\n",
Sage Weil5bdca4e2012-07-10 11:53:34 -07001998 le32_to_cpu(con->in_reply.connect_seq));
Sage Weil31b80062009-10-06 11:31:13 -07001999 pr_err("%s%lld %s connection reset\n",
2000 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002001 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07002002 reset_connection(con);
Jim Schutt6d4221b2012-08-10 10:37:38 -07002003 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002004 ret = prepare_write_connect(con);
2005 if (ret < 0)
2006 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002007 prepare_read_connect(con);
2008
2009 /* Tell ceph about it. */
Sage Weilec302642009-12-22 10:43:42 -08002010 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002011 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2012 if (con->ops->peer_reset)
2013 con->ops->peer_reset(con);
Sage Weilec302642009-12-22 10:43:42 -08002014 mutex_lock(&con->mutex);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002015 if (con->state != CON_STATE_NEGOTIATING)
Sage Weil0da5d702011-05-19 11:21:05 -07002016 return -EAGAIN;
Sage Weil31b80062009-10-06 11:31:13 -07002017 break;
2018
2019 case CEPH_MSGR_TAG_RETRY_SESSION:
2020 /*
2021 * If we sent a smaller connect_seq than the peer has, try
2022 * again with a larger value.
2023 */
Sage Weil5bdca4e2012-07-10 11:53:34 -07002024 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
Sage Weil31b80062009-10-06 11:31:13 -07002025 le32_to_cpu(con->out_connect.connect_seq),
Sage Weil5bdca4e2012-07-10 11:53:34 -07002026 le32_to_cpu(con->in_reply.connect_seq));
2027 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
Jim Schutt6d4221b2012-08-10 10:37:38 -07002028 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002029 ret = prepare_write_connect(con);
2030 if (ret < 0)
2031 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002032 prepare_read_connect(con);
2033 break;
2034
2035 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2036 /*
2037 * If we sent a smaller global_seq than the peer has, try
2038 * again with a larger value.
2039 */
Sage Weileed0ef22009-11-10 14:34:36 -08002040 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
Sage Weil31b80062009-10-06 11:31:13 -07002041 con->peer_global_seq,
Sage Weil5bdca4e2012-07-10 11:53:34 -07002042 le32_to_cpu(con->in_reply.global_seq));
Sage Weil31b80062009-10-06 11:31:13 -07002043 get_global_seq(con->msgr,
Sage Weil5bdca4e2012-07-10 11:53:34 -07002044 le32_to_cpu(con->in_reply.global_seq));
Jim Schutt6d4221b2012-08-10 10:37:38 -07002045 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002046 ret = prepare_write_connect(con);
2047 if (ret < 0)
2048 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002049 prepare_read_connect(con);
2050 break;
2051
Sage Weil3a230832013-03-25 08:47:40 -07002052 case CEPH_MSGR_TAG_SEQ:
Sage Weil31b80062009-10-06 11:31:13 -07002053 case CEPH_MSGR_TAG_READY:
Sage Weil04a419f2009-12-23 09:30:21 -08002054 if (req_feat & ~server_feat) {
2055 pr_err("%s%lld %s protocol feature mismatch,"
2056 " my required %llx > server's %llx, need %llx\n",
2057 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002058 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil04a419f2009-12-23 09:30:21 -08002059 req_feat, server_feat, req_feat & ~server_feat);
2060 con->error_msg = "missing required protocol features";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06002061 reset_connection(con);
Sage Weil04a419f2009-12-23 09:30:21 -08002062 return -1;
2063 }
Sage Weil8dacc7d2012-07-20 17:24:40 -07002064
Alex Elder122070a2012-12-26 10:43:57 -06002065 WARN_ON(con->state != CON_STATE_NEGOTIATING);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002066 con->state = CON_STATE_OPEN;
Sage Weil20e55c42013-03-25 09:30:13 -07002067 con->auth_retry = 0; /* we authenticated; clear flag */
Sage Weil31b80062009-10-06 11:31:13 -07002068 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2069 con->connect_seq++;
Sage Weilaba558e2010-05-12 15:23:30 -07002070 con->peer_features = server_feat;
Sage Weil31b80062009-10-06 11:31:13 -07002071 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2072 con->peer_global_seq,
2073 le32_to_cpu(con->in_reply.connect_seq),
2074 con->connect_seq);
2075 WARN_ON(con->connect_seq !=
2076 le32_to_cpu(con->in_reply.connect_seq));
Sage Weil92ac41d2009-12-14 14:56:56 -08002077
2078 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
Alex Elderc9ffc772013-02-20 10:25:12 -06002079 con_flag_set(con, CON_FLAG_LOSSYTX);
Sage Weil92ac41d2009-12-14 14:56:56 -08002080
Sage Weil85effe12012-07-30 16:22:05 -07002081 con->delay = 0; /* reset backoff memory */
Sage Weil31b80062009-10-06 11:31:13 -07002082
Sage Weil3a230832013-03-25 08:47:40 -07002083 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2084 prepare_write_seq(con);
2085 prepare_read_seq(con);
2086 } else {
2087 prepare_read_tag(con);
2088 }
Sage Weil31b80062009-10-06 11:31:13 -07002089 break;
2090
2091 case CEPH_MSGR_TAG_WAIT:
2092 /*
2093 * If there is a connection race (we are opening
2094 * connections to each other), one of us may just have
2095 * to WAIT. This shouldn't happen if we are the
2096 * client.
2097 */
Sage Weil04177882011-05-12 15:33:17 -07002098 pr_err("process_connect got WAIT as client\n");
2099 con->error_msg = "protocol error, got WAIT as client";
2100 return -1;
Sage Weil31b80062009-10-06 11:31:13 -07002101
2102 default:
2103 pr_err("connect protocol error, will retry\n");
2104 con->error_msg = "protocol error, garbage tag during connect";
2105 return -1;
2106 }
2107 return 0;
2108}
2109
2110
2111/*
2112 * read (part of) an ack
2113 */
2114static int read_partial_ack(struct ceph_connection *con)
2115{
Alex Elderfd516532012-05-10 10:29:50 -05002116 int size = sizeof (con->in_temp_ack);
2117 int end = size;
Sage Weil31b80062009-10-06 11:31:13 -07002118
Alex Elderfd516532012-05-10 10:29:50 -05002119 return read_partial(con, end, size, &con->in_temp_ack);
Sage Weil31b80062009-10-06 11:31:13 -07002120}
2121
Sage Weil31b80062009-10-06 11:31:13 -07002122/*
2123 * We can finally discard anything that's been acked.
2124 */
2125static void process_ack(struct ceph_connection *con)
2126{
2127 struct ceph_msg *m;
2128 u64 ack = le64_to_cpu(con->in_temp_ack);
2129 u64 seq;
2130
Sage Weil31b80062009-10-06 11:31:13 -07002131 while (!list_empty(&con->out_sent)) {
2132 m = list_first_entry(&con->out_sent, struct ceph_msg,
2133 list_head);
2134 seq = le64_to_cpu(m->hdr.seq);
2135 if (seq > ack)
2136 break;
2137 dout("got ack for seq %llu type %d at %p\n", seq,
2138 le16_to_cpu(m->hdr.type), m);
Sage Weil4cf9d542011-07-26 11:27:24 -07002139 m->ack_stamp = jiffies;
Sage Weil31b80062009-10-06 11:31:13 -07002140 ceph_msg_remove(m);
2141 }
Sage Weil31b80062009-10-06 11:31:13 -07002142 prepare_read_tag(con);
2143}
2144
2145
Yehuda Sadeh24504182010-01-08 13:58:34 -08002146static int read_partial_message_section(struct ceph_connection *con,
Sage Weil213c99e2010-08-03 10:25:11 -07002147 struct kvec *section,
2148 unsigned int sec_len, u32 *crc)
Yehuda Sadeh24504182010-01-08 13:58:34 -08002149{
Yehuda Sadeh68b44762010-04-06 15:01:27 -07002150 int ret, left;
Sage Weil31b80062009-10-06 11:31:13 -07002151
Yehuda Sadeh24504182010-01-08 13:58:34 -08002152 BUG_ON(!section);
Sage Weil31b80062009-10-06 11:31:13 -07002153
Yehuda Sadeh24504182010-01-08 13:58:34 -08002154 while (section->iov_len < sec_len) {
2155 BUG_ON(section->iov_base == NULL);
2156 left = sec_len - section->iov_len;
2157 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2158 section->iov_len, left);
2159 if (ret <= 0)
2160 return ret;
2161 section->iov_len += ret;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002162 }
Alex Elderfe3ad592012-02-15 07:43:54 -06002163 if (section->iov_len == sec_len)
2164 *crc = crc32c(0, section->iov_base, section->iov_len);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002165
2166 return 1;
2167}
2168
Alex Elder34d2d202013-03-08 20:58:59 -06002169static int read_partial_msg_data(struct ceph_connection *con)
2170{
2171 struct ceph_msg *msg = con->in_msg;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002172 struct ceph_msg_data_cursor *cursor = &msg->cursor;
Alex Elder34d2d202013-03-08 20:58:59 -06002173 const bool do_datacrc = !con->msgr->nocrc;
Alex Elder686be202013-03-11 23:34:23 -05002174 struct page *page;
2175 size_t page_offset;
2176 size_t length;
Alex Elderf5db90b2013-03-11 23:34:23 -05002177 u32 crc = 0;
Alex Elder34d2d202013-03-08 20:58:59 -06002178 int ret;
2179
2180 BUG_ON(!msg);
Alex Elder5240d9f2013-03-14 14:09:06 -05002181 if (list_empty(&msg->data))
Alex Elder4c59b4a2013-03-11 23:34:23 -05002182 return -EIO;
Alex Elder34d2d202013-03-08 20:58:59 -06002183
Alex Elderf5db90b2013-03-11 23:34:23 -05002184 if (do_datacrc)
2185 crc = con->in_data_crc;
Alex Elder643c68a2013-03-11 23:34:23 -05002186 while (cursor->resid) {
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002187 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
Alex Elder4c59b4a2013-03-11 23:34:23 -05002188 NULL);
Alex Elder686be202013-03-11 23:34:23 -05002189 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
Alex Elderf5db90b2013-03-11 23:34:23 -05002190 if (ret <= 0) {
2191 if (do_datacrc)
2192 con->in_data_crc = crc;
2193
Alex Elder686be202013-03-11 23:34:23 -05002194 return ret;
Alex Elderf5db90b2013-03-11 23:34:23 -05002195 }
Alex Elder686be202013-03-11 23:34:23 -05002196
2197 if (do_datacrc)
Alex Elderf5db90b2013-03-11 23:34:23 -05002198 crc = ceph_crc32c_page(crc, page, page_offset, ret);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002199 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
Alex Elder34d2d202013-03-08 20:58:59 -06002200 }
Alex Elderf5db90b2013-03-11 23:34:23 -05002201 if (do_datacrc)
2202 con->in_data_crc = crc;
Alex Elder34d2d202013-03-08 20:58:59 -06002203
2204 return 1; /* must return > 0 to indicate success */
2205}
2206
Sage Weil31b80062009-10-06 11:31:13 -07002207/*
2208 * read (part of) a message.
2209 */
Alex Elder686be202013-03-11 23:34:23 -05002210static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2211
Sage Weil31b80062009-10-06 11:31:13 -07002212static int read_partial_message(struct ceph_connection *con)
2213{
2214 struct ceph_msg *m = con->in_msg;
Alex Elderfd516532012-05-10 10:29:50 -05002215 int size;
2216 int end;
Sage Weil31b80062009-10-06 11:31:13 -07002217 int ret;
Eric Dumazet95c96172012-04-15 05:58:06 +00002218 unsigned int front_len, middle_len, data_len;
Alex Elder37675b02012-03-07 11:40:08 -06002219 bool do_datacrc = !con->msgr->nocrc;
Sage Weilae187562010-04-22 07:47:01 -07002220 u64 seq;
Alex Elderfe3ad592012-02-15 07:43:54 -06002221 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07002222
2223 dout("read_partial_message con %p msg %p\n", con, m);
2224
2225 /* header */
Alex Elderfd516532012-05-10 10:29:50 -05002226 size = sizeof (con->in_hdr);
2227 end = size;
2228 ret = read_partial(con, end, size, &con->in_hdr);
Alex Elder57dac9d2012-05-10 10:29:50 -05002229 if (ret <= 0)
2230 return ret;
Alex Elderfe3ad592012-02-15 07:43:54 -06002231
2232 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2233 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2234 pr_err("read_partial_message bad hdr "
2235 " crc %u != expected %u\n",
2236 crc, con->in_hdr.crc);
2237 return -EBADMSG;
2238 }
2239
Sage Weil31b80062009-10-06 11:31:13 -07002240 front_len = le32_to_cpu(con->in_hdr.front_len);
2241 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2242 return -EIO;
2243 middle_len = le32_to_cpu(con->in_hdr.middle_len);
Alex Elder7b11ba32013-03-08 18:51:03 -06002244 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
Sage Weil31b80062009-10-06 11:31:13 -07002245 return -EIO;
2246 data_len = le32_to_cpu(con->in_hdr.data_len);
2247 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2248 return -EIO;
2249
Sage Weilae187562010-04-22 07:47:01 -07002250 /* verify seq# */
2251 seq = le64_to_cpu(con->in_hdr.seq);
2252 if ((s64)seq - (s64)con->in_seq < 1) {
Sage Weildf9f86f2010-11-01 15:49:23 -07002253 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
Sage Weilae187562010-04-22 07:47:01 -07002254 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002255 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weilae187562010-04-22 07:47:01 -07002256 seq, con->in_seq + 1);
2257 con->in_base_pos = -front_len - middle_len - data_len -
2258 sizeof(m->footer);
2259 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weilae187562010-04-22 07:47:01 -07002260 return 0;
2261 } else if ((s64)seq - (s64)con->in_seq > 1) {
2262 pr_err("read_partial_message bad seq %lld expected %lld\n",
2263 seq, con->in_seq + 1);
2264 con->error_msg = "bad message sequence # for incoming message";
2265 return -EBADMSG;
2266 }
2267
Sage Weil31b80062009-10-06 11:31:13 -07002268 /* allocate message? */
2269 if (!con->in_msg) {
Sage Weil4740a622012-07-30 18:19:30 -07002270 int skip = 0;
2271
Sage Weil31b80062009-10-06 11:31:13 -07002272 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
Alex Elder6ebc8b32013-03-08 18:51:03 -06002273 front_len, data_len);
Sage Weil4740a622012-07-30 18:19:30 -07002274 ret = ceph_con_in_msg_alloc(con, &skip);
2275 if (ret < 0)
2276 return ret;
Alex Elderf759ebb2013-04-05 14:46:01 -05002277
2278 BUG_ON(!con->in_msg ^ skip);
2279 if (con->in_msg && data_len > con->in_msg->data_length) {
2280 pr_warning("%s skipping long message (%u > %zd)\n",
2281 __func__, data_len, con->in_msg->data_length);
2282 ceph_msg_put(con->in_msg);
2283 con->in_msg = NULL;
2284 skip = 1;
2285 }
Yehuda Sadeh24504182010-01-08 13:58:34 -08002286 if (skip) {
Sage Weil31b80062009-10-06 11:31:13 -07002287 /* skip this message */
Sage Weila79832f2010-04-01 16:06:19 -07002288 dout("alloc_msg said skip message\n");
Sage Weil31b80062009-10-06 11:31:13 -07002289 con->in_base_pos = -front_len - middle_len - data_len -
2290 sizeof(m->footer);
2291 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weil684be252010-04-21 20:45:59 -07002292 con->in_seq++;
Sage Weil31b80062009-10-06 11:31:13 -07002293 return 0;
2294 }
Alex Elder38941f82012-06-01 14:56:43 -05002295
Sage Weil4740a622012-07-30 18:19:30 -07002296 BUG_ON(!con->in_msg);
Alex Elder38941f82012-06-01 14:56:43 -05002297 BUG_ON(con->in_msg->con != con);
Sage Weil31b80062009-10-06 11:31:13 -07002298 m = con->in_msg;
2299 m->front.iov_len = 0; /* haven't read it yet */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002300 if (m->middle)
2301 m->middle->vec.iov_len = 0;
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08002302
Alex Elder78625052013-03-06 23:39:39 -06002303 /* prepare for data payload, if any */
Sage Weila4107022012-07-30 16:20:25 -07002304
Alex Elder78625052013-03-06 23:39:39 -06002305 if (data_len)
Alex Elder98fa5dd2013-04-02 12:09:50 -05002306 prepare_message_data(con->in_msg, data_len);
Sage Weil31b80062009-10-06 11:31:13 -07002307 }
2308
2309 /* front */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002310 ret = read_partial_message_section(con, &m->front, front_len,
2311 &con->in_front_crc);
2312 if (ret <= 0)
2313 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002314
2315 /* middle */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002316 if (m->middle) {
Sage Weil213c99e2010-08-03 10:25:11 -07002317 ret = read_partial_message_section(con, &m->middle->vec,
2318 middle_len,
Yehuda Sadeh24504182010-01-08 13:58:34 -08002319 &con->in_middle_crc);
Sage Weil31b80062009-10-06 11:31:13 -07002320 if (ret <= 0)
2321 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002322 }
2323
2324 /* (page) data */
Alex Elder34d2d202013-03-08 20:58:59 -06002325 if (data_len) {
2326 ret = read_partial_msg_data(con);
2327 if (ret <= 0)
2328 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002329 }
2330
Sage Weil31b80062009-10-06 11:31:13 -07002331 /* footer */
Alex Elderfd516532012-05-10 10:29:50 -05002332 size = sizeof (m->footer);
2333 end += size;
2334 ret = read_partial(con, end, size, &m->footer);
Alex Elder57dac9d2012-05-10 10:29:50 -05002335 if (ret <= 0)
2336 return ret;
2337
Sage Weil31b80062009-10-06 11:31:13 -07002338 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2339 m, front_len, m->footer.front_crc, middle_len,
2340 m->footer.middle_crc, data_len, m->footer.data_crc);
2341
2342 /* crc ok? */
2343 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2344 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2345 m, con->in_front_crc, m->footer.front_crc);
2346 return -EBADMSG;
2347 }
2348 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2349 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2350 m, con->in_middle_crc, m->footer.middle_crc);
2351 return -EBADMSG;
2352 }
Alex Elderbca064d2012-02-15 07:43:54 -06002353 if (do_datacrc &&
Sage Weil31b80062009-10-06 11:31:13 -07002354 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2355 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2356 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2357 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2358 return -EBADMSG;
2359 }
2360
2361 return 1; /* done! */
2362}
2363
2364/*
2365 * Process message. This happens in the worker thread. The callback should
2366 * be careful not to do anything that waits on other incoming messages or it
2367 * may deadlock.
2368 */
2369static void process_message(struct ceph_connection *con)
2370{
Sage Weil5e095e82009-12-14 14:30:34 -08002371 struct ceph_msg *msg;
Sage Weil31b80062009-10-06 11:31:13 -07002372
Alex Elder38941f82012-06-01 14:56:43 -05002373 BUG_ON(con->in_msg->con != con);
2374 con->in_msg->con = NULL;
Sage Weil5e095e82009-12-14 14:30:34 -08002375 msg = con->in_msg;
Sage Weil31b80062009-10-06 11:31:13 -07002376 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -07002377 con->ops->put(con);
Sage Weil31b80062009-10-06 11:31:13 -07002378
2379 /* if first message, set peer_name */
2380 if (con->peer_name.type == 0)
Sage Weildbad1852010-03-25 15:45:38 -07002381 con->peer_name = msg->hdr.src;
Sage Weil31b80062009-10-06 11:31:13 -07002382
Sage Weil31b80062009-10-06 11:31:13 -07002383 con->in_seq++;
Sage Weilec302642009-12-22 10:43:42 -08002384 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002385
2386 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2387 msg, le64_to_cpu(msg->hdr.seq),
Sage Weildbad1852010-03-25 15:45:38 -07002388 ENTITY_NAME(msg->hdr.src),
Sage Weil31b80062009-10-06 11:31:13 -07002389 le16_to_cpu(msg->hdr.type),
2390 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2391 le32_to_cpu(msg->hdr.front_len),
2392 le32_to_cpu(msg->hdr.data_len),
2393 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2394 con->ops->dispatch(con, msg);
Sage Weilec302642009-12-22 10:43:42 -08002395
2396 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002397}
2398
2399
2400/*
2401 * Write something to the socket. Called in a worker thread when the
2402 * socket appears to be writeable and we have something ready to send.
2403 */
2404static int try_write(struct ceph_connection *con)
2405{
Sage Weil31b80062009-10-06 11:31:13 -07002406 int ret = 1;
2407
Sage Weild59315c2012-06-21 12:49:23 -07002408 dout("try_write start %p state %lu\n", con, con->state);
Sage Weil31b80062009-10-06 11:31:13 -07002409
Sage Weil31b80062009-10-06 11:31:13 -07002410more:
2411 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2412
2413 /* open the socket first? */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002414 if (con->state == CON_STATE_PREOPEN) {
2415 BUG_ON(con->sock);
2416 con->state = CON_STATE_CONNECTING;
Alex Eldera5988c42012-05-29 11:04:58 -05002417
Alex Eldere2200422012-05-23 14:35:23 -05002418 con_out_kvec_reset(con);
Alex Eldere825a662012-05-16 15:16:38 -05002419 prepare_write_banner(con);
Sage Weileed0ef22009-11-10 14:34:36 -08002420 prepare_read_banner(con);
Sage Weil31b80062009-10-06 11:31:13 -07002421
Sage Weilcf3e5c42009-12-11 09:48:05 -08002422 BUG_ON(con->in_msg);
Sage Weil31b80062009-10-06 11:31:13 -07002423 con->in_tag = CEPH_MSGR_TAG_READY;
2424 dout("try_write initiating connect on %p new state %lu\n",
2425 con, con->state);
Alex Elder41617d02012-02-14 14:05:33 -06002426 ret = ceph_tcp_connect(con);
2427 if (ret < 0) {
Sage Weil31b80062009-10-06 11:31:13 -07002428 con->error_msg = "connect error";
Sage Weil31b80062009-10-06 11:31:13 -07002429 goto out;
2430 }
2431 }
2432
2433more_kvec:
2434 /* kvec data queued? */
2435 if (con->out_skip) {
2436 ret = write_partial_skip(con);
2437 if (ret <= 0)
Sage Weil42961d22011-01-25 08:19:34 -08002438 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002439 }
2440 if (con->out_kvec_left) {
2441 ret = write_partial_kvec(con);
2442 if (ret <= 0)
Sage Weil42961d22011-01-25 08:19:34 -08002443 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002444 }
2445
2446 /* msg pages? */
2447 if (con->out_msg) {
Sage Weilc86a2932009-12-14 14:04:30 -08002448 if (con->out_msg_done) {
2449 ceph_msg_put(con->out_msg);
2450 con->out_msg = NULL; /* we're done with this one */
2451 goto do_next;
2452 }
2453
Alex Elder34d2d202013-03-08 20:58:59 -06002454 ret = write_partial_message_data(con);
Sage Weil31b80062009-10-06 11:31:13 -07002455 if (ret == 1)
2456 goto more_kvec; /* we need to send the footer, too! */
2457 if (ret == 0)
Sage Weil42961d22011-01-25 08:19:34 -08002458 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002459 if (ret < 0) {
Alex Elder34d2d202013-03-08 20:58:59 -06002460 dout("try_write write_partial_message_data err %d\n",
Sage Weil31b80062009-10-06 11:31:13 -07002461 ret);
Sage Weil42961d22011-01-25 08:19:34 -08002462 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002463 }
2464 }
2465
Sage Weilc86a2932009-12-14 14:04:30 -08002466do_next:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002467 if (con->state == CON_STATE_OPEN) {
Sage Weil31b80062009-10-06 11:31:13 -07002468 /* is anything else pending? */
2469 if (!list_empty(&con->out_queue)) {
2470 prepare_write_message(con);
2471 goto more;
2472 }
2473 if (con->in_seq > con->in_seq_acked) {
2474 prepare_write_ack(con);
2475 goto more;
2476 }
Alex Elderc9ffc772013-02-20 10:25:12 -06002477 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
Sage Weil31b80062009-10-06 11:31:13 -07002478 prepare_write_keepalive(con);
2479 goto more;
2480 }
2481 }
2482
2483 /* Nothing to do! */
Alex Elderc9ffc772013-02-20 10:25:12 -06002484 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07002485 dout("try_write nothing else to write.\n");
Sage Weil31b80062009-10-06 11:31:13 -07002486 ret = 0;
2487out:
Sage Weil42961d22011-01-25 08:19:34 -08002488 dout("try_write done on %p ret %d\n", con, ret);
Sage Weil31b80062009-10-06 11:31:13 -07002489 return ret;
2490}
2491
2492
2493
2494/*
2495 * Read what we can from the socket.
2496 */
2497static int try_read(struct ceph_connection *con)
2498{
Sage Weil31b80062009-10-06 11:31:13 -07002499 int ret = -1;
2500
Sage Weil31b80062009-10-06 11:31:13 -07002501more:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002502 dout("try_read start on %p state %lu\n", con, con->state);
2503 if (con->state != CON_STATE_CONNECTING &&
2504 con->state != CON_STATE_NEGOTIATING &&
2505 con->state != CON_STATE_OPEN)
2506 return 0;
2507
2508 BUG_ON(!con->sock);
2509
Sage Weil31b80062009-10-06 11:31:13 -07002510 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2511 con->in_base_pos);
Sage Weil0da5d702011-05-19 11:21:05 -07002512
Sage Weil8dacc7d2012-07-20 17:24:40 -07002513 if (con->state == CON_STATE_CONNECTING) {
Alex Elder7593af92012-05-24 11:55:03 -05002514 dout("try_read connecting\n");
2515 ret = read_partial_banner(con);
2516 if (ret <= 0)
Alex Elderab166d52012-05-31 11:37:29 -05002517 goto out;
Alex Elder7593af92012-05-24 11:55:03 -05002518 ret = process_banner(con);
2519 if (ret < 0)
2520 goto out;
2521
Sage Weil8dacc7d2012-07-20 17:24:40 -07002522 con->state = CON_STATE_NEGOTIATING;
Alex Elder7593af92012-05-24 11:55:03 -05002523
Jim Schutt6d4221b2012-08-10 10:37:38 -07002524 /*
2525 * Received banner is good, exchange connection info.
2526 * Do not reset out_kvec, as sending our banner raced
2527 * with receiving peer banner after connect completed.
2528 */
Alex Elder7593af92012-05-24 11:55:03 -05002529 ret = prepare_write_connect(con);
2530 if (ret < 0)
2531 goto out;
2532 prepare_read_connect(con);
2533
2534 /* Send connection info before awaiting response */
Sage Weil0da5d702011-05-19 11:21:05 -07002535 goto out;
2536 }
2537
Sage Weil8dacc7d2012-07-20 17:24:40 -07002538 if (con->state == CON_STATE_NEGOTIATING) {
Alex Elder7593af92012-05-24 11:55:03 -05002539 dout("try_read negotiating\n");
Sage Weil31b80062009-10-06 11:31:13 -07002540 ret = read_partial_connect(con);
2541 if (ret <= 0)
Sage Weil31b80062009-10-06 11:31:13 -07002542 goto out;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002543 ret = process_connect(con);
2544 if (ret < 0)
2545 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002546 goto more;
2547 }
2548
Alex Elder122070a2012-12-26 10:43:57 -06002549 WARN_ON(con->state != CON_STATE_OPEN);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002550
Sage Weil31b80062009-10-06 11:31:13 -07002551 if (con->in_base_pos < 0) {
2552 /*
2553 * skipping + discarding content.
2554 *
2555 * FIXME: there must be a better way to do this!
2556 */
Alex Elder84495f42012-02-15 07:43:55 -06002557 static char buf[SKIP_BUF_SIZE];
2558 int skip = min((int) sizeof (buf), -con->in_base_pos);
2559
Sage Weil31b80062009-10-06 11:31:13 -07002560 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2561 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2562 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002563 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002564 con->in_base_pos += ret;
2565 if (con->in_base_pos)
2566 goto more;
2567 }
2568 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2569 /*
2570 * what's next?
2571 */
2572 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2573 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002574 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002575 dout("try_read got tag %d\n", (int)con->in_tag);
2576 switch (con->in_tag) {
2577 case CEPH_MSGR_TAG_MSG:
2578 prepare_read_message(con);
2579 break;
2580 case CEPH_MSGR_TAG_ACK:
2581 prepare_read_ack(con);
2582 break;
2583 case CEPH_MSGR_TAG_CLOSE:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002584 con_close_socket(con);
2585 con->state = CON_STATE_CLOSED;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002586 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002587 default:
2588 goto bad_tag;
2589 }
2590 }
2591 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2592 ret = read_partial_message(con);
2593 if (ret <= 0) {
2594 switch (ret) {
2595 case -EBADMSG:
2596 con->error_msg = "bad crc";
2597 ret = -EIO;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002598 break;
Sage Weil31b80062009-10-06 11:31:13 -07002599 case -EIO:
2600 con->error_msg = "io error";
Sage Weil98bdb0a2011-01-25 08:17:48 -08002601 break;
Sage Weil31b80062009-10-06 11:31:13 -07002602 }
Sage Weil98bdb0a2011-01-25 08:17:48 -08002603 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002604 }
2605 if (con->in_tag == CEPH_MSGR_TAG_READY)
2606 goto more;
2607 process_message(con);
Sage Weil7b862e02012-07-30 18:16:56 -07002608 if (con->state == CON_STATE_OPEN)
2609 prepare_read_tag(con);
Sage Weil31b80062009-10-06 11:31:13 -07002610 goto more;
2611 }
Sage Weil3a230832013-03-25 08:47:40 -07002612 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2613 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2614 /*
2615 * the final handshake seq exchange is semantically
2616 * equivalent to an ACK
2617 */
Sage Weil31b80062009-10-06 11:31:13 -07002618 ret = read_partial_ack(con);
2619 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002620 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002621 process_ack(con);
2622 goto more;
2623 }
2624
Sage Weil31b80062009-10-06 11:31:13 -07002625out:
Sage Weil98bdb0a2011-01-25 08:17:48 -08002626 dout("try_read done on %p ret %d\n", con, ret);
Sage Weil31b80062009-10-06 11:31:13 -07002627 return ret;
2628
2629bad_tag:
2630 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2631 con->error_msg = "protocol error, garbage tag";
2632 ret = -1;
2633 goto out;
2634}
2635
2636
2637/*
Alex Elder802c6d92012-10-08 20:37:30 -07002638 * Atomically queue work on a connection after the specified delay.
2639 * Bump @con reference to avoid races with connection teardown.
2640 * Returns 0 if work was queued, or an error code otherwise.
Sage Weil31b80062009-10-06 11:31:13 -07002641 */
Alex Elder802c6d92012-10-08 20:37:30 -07002642static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
Sage Weil31b80062009-10-06 11:31:13 -07002643{
Sage Weil31b80062009-10-06 11:31:13 -07002644 if (!con->ops->get(con)) {
Alex Elder802c6d92012-10-08 20:37:30 -07002645 dout("%s %p ref count 0\n", __func__, con);
2646
2647 return -ENOENT;
Sage Weil31b80062009-10-06 11:31:13 -07002648 }
2649
Alex Elder802c6d92012-10-08 20:37:30 -07002650 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2651 dout("%s %p - already queued\n", __func__, con);
Sage Weil31b80062009-10-06 11:31:13 -07002652 con->ops->put(con);
Alex Elder802c6d92012-10-08 20:37:30 -07002653
2654 return -EBUSY;
Sage Weil31b80062009-10-06 11:31:13 -07002655 }
Alex Elder802c6d92012-10-08 20:37:30 -07002656
2657 dout("%s %p %lu\n", __func__, con, delay);
2658
2659 return 0;
2660}
2661
2662static void queue_con(struct ceph_connection *con)
2663{
2664 (void) queue_con_delay(con, 0);
Sage Weil31b80062009-10-06 11:31:13 -07002665}
2666
Alex Elder7bb21d682012-12-07 19:50:07 -06002667static bool con_sock_closed(struct ceph_connection *con)
2668{
Alex Elderc9ffc772013-02-20 10:25:12 -06002669 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
Alex Elder7bb21d682012-12-07 19:50:07 -06002670 return false;
2671
2672#define CASE(x) \
2673 case CON_STATE_ ## x: \
2674 con->error_msg = "socket closed (con state " #x ")"; \
2675 break;
2676
2677 switch (con->state) {
2678 CASE(CLOSED);
2679 CASE(PREOPEN);
2680 CASE(CONNECTING);
2681 CASE(NEGOTIATING);
2682 CASE(OPEN);
2683 CASE(STANDBY);
2684 default:
2685 pr_warning("%s con %p unrecognized state %lu\n",
2686 __func__, con, con->state);
2687 con->error_msg = "unrecognized con state";
2688 BUG();
2689 break;
2690 }
2691#undef CASE
2692
2693 return true;
2694}
2695
Alex Elderf20a39f2013-02-19 12:25:57 -06002696static bool con_backoff(struct ceph_connection *con)
2697{
2698 int ret;
2699
2700 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2701 return false;
2702
2703 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2704 if (ret) {
2705 dout("%s: con %p FAILED to back off %lu\n", __func__,
2706 con, con->delay);
2707 BUG_ON(ret == -ENOENT);
2708 con_flag_set(con, CON_FLAG_BACKOFF);
2709 }
2710
2711 return true;
2712}
2713
Alex Elder93209262013-02-19 12:25:57 -06002714/* Finish fault handling; con->mutex must *not* be held here */
2715
2716static void con_fault_finish(struct ceph_connection *con)
2717{
2718 /*
2719 * in case we faulted due to authentication, invalidate our
2720 * current tickets so that we can get new ones.
2721 */
2722 if (con->auth_retry && con->ops->invalidate_authorizer) {
2723 dout("calling invalidate_authorizer()\n");
2724 con->ops->invalidate_authorizer(con);
2725 }
2726
2727 if (con->ops->fault)
2728 con->ops->fault(con);
2729}
2730
Sage Weil31b80062009-10-06 11:31:13 -07002731/*
2732 * Do some work on a connection. Drop a connection ref when we're done.
2733 */
2734static void con_work(struct work_struct *work)
2735{
2736 struct ceph_connection *con = container_of(work, struct ceph_connection,
2737 work.work);
Alex Elder49659412013-02-19 12:25:57 -06002738 bool fault;
Sage Weil31b80062009-10-06 11:31:13 -07002739
Sage Weil9dd46582010-04-28 13:51:50 -07002740 mutex_lock(&con->mutex);
Alex Elder49659412013-02-19 12:25:57 -06002741 while (true) {
2742 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07002743
Alex Elder49659412013-02-19 12:25:57 -06002744 if ((fault = con_sock_closed(con))) {
2745 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2746 break;
2747 }
2748 if (con_backoff(con)) {
2749 dout("%s: con %p BACKOFF\n", __func__, con);
2750 break;
2751 }
2752 if (con->state == CON_STATE_STANDBY) {
2753 dout("%s: con %p STANDBY\n", __func__, con);
2754 break;
2755 }
2756 if (con->state == CON_STATE_CLOSED) {
2757 dout("%s: con %p CLOSED\n", __func__, con);
2758 BUG_ON(con->sock);
2759 break;
2760 }
2761 if (con->state == CON_STATE_PREOPEN) {
2762 dout("%s: con %p PREOPEN\n", __func__, con);
2763 BUG_ON(con->sock);
2764 }
Sage Weil0da5d702011-05-19 11:21:05 -07002765
Alex Elder49659412013-02-19 12:25:57 -06002766 ret = try_read(con);
2767 if (ret < 0) {
2768 if (ret == -EAGAIN)
2769 continue;
2770 con->error_msg = "socket error on read";
2771 fault = true;
2772 break;
2773 }
2774
2775 ret = try_write(con);
2776 if (ret < 0) {
2777 if (ret == -EAGAIN)
2778 continue;
2779 con->error_msg = "socket error on write";
2780 fault = true;
2781 }
2782
2783 break; /* If we make it to here, we're done */
Sage Weil3a140a02012-07-30 16:24:21 -07002784 }
Alex Elderb6e7b6a2013-02-19 12:25:57 -06002785 if (fault)
2786 con_fault(con);
Sage Weil9dd46582010-04-28 13:51:50 -07002787 mutex_unlock(&con->mutex);
Sage Weil0da5d702011-05-19 11:21:05 -07002788
Alex Elderb6e7b6a2013-02-19 12:25:57 -06002789 if (fault)
2790 con_fault_finish(con);
2791
2792 con->ops->put(con);
Sage Weil31b80062009-10-06 11:31:13 -07002793}
2794
Sage Weil31b80062009-10-06 11:31:13 -07002795/*
2796 * Generic error/fault handler. A retry mechanism is used with
2797 * exponential backoff
2798 */
Alex Elder93209262013-02-19 12:25:57 -06002799static void con_fault(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07002800{
Alex Elder28362982012-12-14 16:47:41 -06002801 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002802 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
Sage Weil31b80062009-10-06 11:31:13 -07002803 dout("fault %p state %lu to peer %s\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002804 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07002805
Alex Elder122070a2012-12-26 10:43:57 -06002806 WARN_ON(con->state != CON_STATE_CONNECTING &&
Sage Weil8dacc7d2012-07-20 17:24:40 -07002807 con->state != CON_STATE_NEGOTIATING &&
2808 con->state != CON_STATE_OPEN);
Sage Weilec302642009-12-22 10:43:42 -08002809
Sage Weil31b80062009-10-06 11:31:13 -07002810 con_close_socket(con);
Sage Weil5e095e82009-12-14 14:30:34 -08002811
Alex Elderc9ffc772013-02-20 10:25:12 -06002812 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
Sage Weil8dacc7d2012-07-20 17:24:40 -07002813 dout("fault on LOSSYTX channel, marking CLOSED\n");
2814 con->state = CON_STATE_CLOSED;
Alex Elder93209262013-02-19 12:25:57 -06002815 return;
Sage Weil3b5ede02012-07-20 15:22:53 -07002816 }
2817
Sage Weil5e095e82009-12-14 14:30:34 -08002818 if (con->in_msg) {
Alex Elder38941f82012-06-01 14:56:43 -05002819 BUG_ON(con->in_msg->con != con);
2820 con->in_msg->con = NULL;
Sage Weil5e095e82009-12-14 14:30:34 -08002821 ceph_msg_put(con->in_msg);
2822 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -07002823 con->ops->put(con);
Sage Weil5e095e82009-12-14 14:30:34 -08002824 }
Sage Weil31b80062009-10-06 11:31:13 -07002825
Sage Weile80a52d2010-02-25 12:40:45 -08002826 /* Requeue anything that hasn't been acked */
2827 list_splice_init(&con->out_sent, &con->out_queue);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002828
Sage Weile76661d2011-03-03 10:10:15 -08002829 /* If there are no messages queued or keepalive pending, place
2830 * the connection in a STANDBY state */
2831 if (list_empty(&con->out_queue) &&
Alex Elderc9ffc772013-02-20 10:25:12 -06002832 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
Sage Weile00de342011-03-04 12:25:05 -08002833 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
Alex Elderc9ffc772013-02-20 10:25:12 -06002834 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002835 con->state = CON_STATE_STANDBY;
Sage Weile80a52d2010-02-25 12:40:45 -08002836 } else {
2837 /* retry after a delay. */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002838 con->state = CON_STATE_PREOPEN;
Sage Weile80a52d2010-02-25 12:40:45 -08002839 if (con->delay == 0)
2840 con->delay = BASE_DELAY_INTERVAL;
2841 else if (con->delay < MAX_DELAY_INTERVAL)
2842 con->delay *= 2;
Alex Elderc9ffc772013-02-20 10:25:12 -06002843 con_flag_set(con, CON_FLAG_BACKOFF);
Alex Elder8618e302012-10-08 20:37:30 -07002844 queue_con(con);
Sage Weil31b80062009-10-06 11:31:13 -07002845 }
Sage Weil31b80062009-10-06 11:31:13 -07002846}
2847
2848
2849
2850/*
Alex Elder15d98822012-05-26 23:26:43 -05002851 * initialize a new messenger instance
Sage Weil31b80062009-10-06 11:31:13 -07002852 */
Alex Elder15d98822012-05-26 23:26:43 -05002853void ceph_messenger_init(struct ceph_messenger *msgr,
2854 struct ceph_entity_addr *myaddr,
Ilya Dryomov12b46292013-12-24 21:19:23 +02002855 u64 supported_features,
2856 u64 required_features,
Alex Elder15d98822012-05-26 23:26:43 -05002857 bool nocrc)
Sage Weil31b80062009-10-06 11:31:13 -07002858{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002859 msgr->supported_features = supported_features;
2860 msgr->required_features = required_features;
2861
Sage Weil31b80062009-10-06 11:31:13 -07002862 spin_lock_init(&msgr->global_seq_lock);
2863
Sage Weil31b80062009-10-06 11:31:13 -07002864 if (myaddr)
2865 msgr->inst.addr = *myaddr;
2866
2867 /* select a random nonce */
Sage Weilac8839d2010-01-27 14:28:10 -08002868 msgr->inst.addr.type = 0;
Sage Weil103e2d32010-01-07 16:12:36 -08002869 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
Sage Weil63f2d212009-11-03 15:17:56 -08002870 encode_my_addr(msgr);
Alex Elder15d98822012-05-26 23:26:43 -05002871 msgr->nocrc = nocrc;
Sage Weil31b80062009-10-06 11:31:13 -07002872
Guanjun Hea2a32582012-07-08 19:50:33 -07002873 atomic_set(&msgr->stopping, 0);
Sage Weil31b80062009-10-06 11:31:13 -07002874
Alex Elder15d98822012-05-26 23:26:43 -05002875 dout("%s %p\n", __func__, msgr);
Sage Weil31b80062009-10-06 11:31:13 -07002876}
Alex Elder15d98822012-05-26 23:26:43 -05002877EXPORT_SYMBOL(ceph_messenger_init);
Sage Weil31b80062009-10-06 11:31:13 -07002878
Sage Weile00de342011-03-04 12:25:05 -08002879static void clear_standby(struct ceph_connection *con)
2880{
2881 /* come back from STANDBY? */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002882 if (con->state == CON_STATE_STANDBY) {
Sage Weile00de342011-03-04 12:25:05 -08002883 dout("clear_standby %p and ++connect_seq\n", con);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002884 con->state = CON_STATE_PREOPEN;
Sage Weile00de342011-03-04 12:25:05 -08002885 con->connect_seq++;
Alex Elderc9ffc772013-02-20 10:25:12 -06002886 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2887 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
Sage Weile00de342011-03-04 12:25:05 -08002888 }
2889}
2890
Sage Weil31b80062009-10-06 11:31:13 -07002891/*
2892 * Queue up an outgoing message on the given connection.
2893 */
2894void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2895{
Sage Weila59b55a2012-07-20 15:34:04 -07002896 /* set src+dst */
2897 msg->hdr.src = con->msgr->inst.name;
2898 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2899 msg->needs_out_seq = true;
2900
2901 mutex_lock(&con->mutex);
2902
Sage Weil8dacc7d2012-07-20 17:24:40 -07002903 if (con->state == CON_STATE_CLOSED) {
Sage Weil31b80062009-10-06 11:31:13 -07002904 dout("con_send %p closed, dropping %p\n", con, msg);
2905 ceph_msg_put(msg);
Sage Weila59b55a2012-07-20 15:34:04 -07002906 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002907 return;
2908 }
2909
Alex Elder38941f82012-06-01 14:56:43 -05002910 BUG_ON(msg->con != NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -07002911 msg->con = con->ops->get(con);
Alex Elder92ce0342012-06-04 14:43:33 -05002912 BUG_ON(msg->con == NULL);
Sage Weil31b80062009-10-06 11:31:13 -07002913
Sage Weil31b80062009-10-06 11:31:13 -07002914 BUG_ON(!list_empty(&msg->list_head));
2915 list_add_tail(&msg->list_head, &con->out_queue);
2916 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2917 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2918 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2919 le32_to_cpu(msg->hdr.front_len),
2920 le32_to_cpu(msg->hdr.middle_len),
2921 le32_to_cpu(msg->hdr.data_len));
Sage Weil00650932012-07-20 15:33:04 -07002922
2923 clear_standby(con);
Sage Weilec302642009-12-22 10:43:42 -08002924 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002925
2926 /* if there wasn't anything waiting to send before, queue
2927 * new work */
Alex Elderc9ffc772013-02-20 10:25:12 -06002928 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
Sage Weil31b80062009-10-06 11:31:13 -07002929 queue_con(con);
2930}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002931EXPORT_SYMBOL(ceph_con_send);
Sage Weil31b80062009-10-06 11:31:13 -07002932
2933/*
2934 * Revoke a message that was previously queued for send
2935 */
Alex Elder6740a842012-06-01 14:56:43 -05002936void ceph_msg_revoke(struct ceph_msg *msg)
Sage Weil31b80062009-10-06 11:31:13 -07002937{
Alex Elder6740a842012-06-01 14:56:43 -05002938 struct ceph_connection *con = msg->con;
2939
2940 if (!con)
2941 return; /* Message not in our possession */
2942
Sage Weilec302642009-12-22 10:43:42 -08002943 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002944 if (!list_empty(&msg->list_head)) {
Alex Elder38941f82012-06-01 14:56:43 -05002945 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07002946 list_del_init(&msg->list_head);
Alex Elder38941f82012-06-01 14:56:43 -05002947 BUG_ON(msg->con == NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -07002948 msg->con->ops->put(msg->con);
Alex Elder38941f82012-06-01 14:56:43 -05002949 msg->con = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002950 msg->hdr.seq = 0;
Alex Elder38941f82012-06-01 14:56:43 -05002951
Sage Weil31b80062009-10-06 11:31:13 -07002952 ceph_msg_put(msg);
Sage Weiled98ada2010-07-05 12:15:14 -07002953 }
2954 if (con->out_msg == msg) {
Alex Elder38941f82012-06-01 14:56:43 -05002955 dout("%s %p msg %p - was sending\n", __func__, con, msg);
Sage Weiled98ada2010-07-05 12:15:14 -07002956 con->out_msg = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002957 if (con->out_kvec_is_msg) {
2958 con->out_skip = con->out_kvec_bytes;
2959 con->out_kvec_is_msg = false;
2960 }
Sage Weiled98ada2010-07-05 12:15:14 -07002961 msg->hdr.seq = 0;
Alex Elder92ce0342012-06-04 14:43:33 -05002962
2963 ceph_msg_put(msg);
Sage Weil31b80062009-10-06 11:31:13 -07002964 }
Sage Weilec302642009-12-22 10:43:42 -08002965 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002966}
2967
2968/*
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002969 * Revoke a message that we may be reading data into
Sage Weil350b1c32009-12-22 10:45:45 -08002970 */
Alex Elder8921d112012-06-01 14:56:43 -05002971void ceph_msg_revoke_incoming(struct ceph_msg *msg)
Sage Weil350b1c32009-12-22 10:45:45 -08002972{
Alex Elder8921d112012-06-01 14:56:43 -05002973 struct ceph_connection *con;
2974
2975 BUG_ON(msg == NULL);
2976 if (!msg->con) {
2977 dout("%s msg %p null con\n", __func__, msg);
2978
2979 return; /* Message not in our possession */
2980 }
2981
2982 con = msg->con;
Sage Weil350b1c32009-12-22 10:45:45 -08002983 mutex_lock(&con->mutex);
Alex Elder8921d112012-06-01 14:56:43 -05002984 if (con->in_msg == msg) {
Eric Dumazet95c96172012-04-15 05:58:06 +00002985 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2986 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2987 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
Sage Weil350b1c32009-12-22 10:45:45 -08002988
2989 /* skip rest of message */
Alex Elder8921d112012-06-01 14:56:43 -05002990 dout("%s %p msg %p revoked\n", __func__, con, msg);
2991 con->in_base_pos = con->in_base_pos -
Sage Weil350b1c32009-12-22 10:45:45 -08002992 sizeof(struct ceph_msg_header) -
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002993 front_len -
2994 middle_len -
2995 data_len -
Sage Weil350b1c32009-12-22 10:45:45 -08002996 sizeof(struct ceph_msg_footer);
Sage Weil350b1c32009-12-22 10:45:45 -08002997 ceph_msg_put(con->in_msg);
2998 con->in_msg = NULL;
2999 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weil684be252010-04-21 20:45:59 -07003000 con->in_seq++;
Sage Weil350b1c32009-12-22 10:45:45 -08003001 } else {
Alex Elder8921d112012-06-01 14:56:43 -05003002 dout("%s %p in_msg %p msg %p no-op\n",
3003 __func__, con, con->in_msg, msg);
Sage Weil350b1c32009-12-22 10:45:45 -08003004 }
3005 mutex_unlock(&con->mutex);
3006}
3007
3008/*
Sage Weil31b80062009-10-06 11:31:13 -07003009 * Queue a keepalive byte to ensure the tcp connection is alive.
3010 */
3011void ceph_con_keepalive(struct ceph_connection *con)
3012{
Sage Weile00de342011-03-04 12:25:05 -08003013 dout("con_keepalive %p\n", con);
Sage Weil00650932012-07-20 15:33:04 -07003014 mutex_lock(&con->mutex);
Sage Weile00de342011-03-04 12:25:05 -08003015 clear_standby(con);
Sage Weil00650932012-07-20 15:33:04 -07003016 mutex_unlock(&con->mutex);
Alex Elderc9ffc772013-02-20 10:25:12 -06003017 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3018 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
Sage Weil31b80062009-10-06 11:31:13 -07003019 queue_con(con);
3020}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003021EXPORT_SYMBOL(ceph_con_keepalive);
Sage Weil31b80062009-10-06 11:31:13 -07003022
Alex Elder6644ed72013-03-11 23:34:24 -05003023static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
Alex Elder43794502013-03-01 18:00:16 -06003024{
Alex Elder6644ed72013-03-11 23:34:24 -05003025 struct ceph_msg_data *data;
3026
3027 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3028 return NULL;
3029
Alex Elder81b36be2013-05-01 12:43:04 -05003030 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
Alex Elder6644ed72013-03-11 23:34:24 -05003031 if (data)
3032 data->type = type;
Alex Elder5240d9f2013-03-14 14:09:06 -05003033 INIT_LIST_HEAD(&data->links);
Alex Elder6644ed72013-03-11 23:34:24 -05003034
3035 return data;
3036}
3037
3038static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3039{
3040 if (!data)
3041 return;
3042
Alex Elder5240d9f2013-03-14 14:09:06 -05003043 WARN_ON(!list_empty(&data->links));
Alex Elder6644ed72013-03-11 23:34:24 -05003044 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3045 ceph_pagelist_release(data->pagelist);
3046 kfree(data->pagelist);
3047 }
Alex Elder81b36be2013-05-01 12:43:04 -05003048 kmem_cache_free(ceph_msg_data_cache, data);
Alex Elder43794502013-03-01 18:00:16 -06003049}
3050
Alex Elder90af3602013-04-05 14:46:01 -05003051void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
Alex Elderf1baeb22013-03-07 15:38:26 -06003052 size_t length, size_t alignment)
Alex Elder02afca62013-02-14 12:16:43 -06003053{
Alex Elder6644ed72013-03-11 23:34:24 -05003054 struct ceph_msg_data *data;
3055
Alex Elder07aa1552013-03-04 18:29:06 -06003056 BUG_ON(!pages);
3057 BUG_ON(!length);
Alex Elder02afca62013-02-14 12:16:43 -06003058
Alex Elder6644ed72013-03-11 23:34:24 -05003059 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3060 BUG_ON(!data);
3061 data->pages = pages;
3062 data->length = length;
3063 data->alignment = alignment & ~PAGE_MASK;
3064
Alex Elder5240d9f2013-03-14 14:09:06 -05003065 list_add_tail(&data->links, &msg->data);
3066 msg->data_length += length;
Alex Elder02afca62013-02-14 12:16:43 -06003067}
Alex Elder90af3602013-04-05 14:46:01 -05003068EXPORT_SYMBOL(ceph_msg_data_add_pages);
Sage Weil31b80062009-10-06 11:31:13 -07003069
Alex Elder90af3602013-04-05 14:46:01 -05003070void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
Alex Elder27fa8382013-02-14 12:16:43 -06003071 struct ceph_pagelist *pagelist)
3072{
Alex Elder6644ed72013-03-11 23:34:24 -05003073 struct ceph_msg_data *data;
3074
Alex Elder07aa1552013-03-04 18:29:06 -06003075 BUG_ON(!pagelist);
3076 BUG_ON(!pagelist->length);
Alex Elder27fa8382013-02-14 12:16:43 -06003077
Alex Elder6644ed72013-03-11 23:34:24 -05003078 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3079 BUG_ON(!data);
3080 data->pagelist = pagelist;
3081
Alex Elder5240d9f2013-03-14 14:09:06 -05003082 list_add_tail(&data->links, &msg->data);
3083 msg->data_length += pagelist->length;
Alex Elder27fa8382013-02-14 12:16:43 -06003084}
Alex Elder90af3602013-04-05 14:46:01 -05003085EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
Alex Elder27fa8382013-02-14 12:16:43 -06003086
Alex Elderea965712013-04-05 14:46:01 -05003087#ifdef CONFIG_BLOCK
Alex Elder90af3602013-04-05 14:46:01 -05003088void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
Alex Eldera1930802013-03-14 14:09:06 -05003089 size_t length)
Alex Elder27fa8382013-02-14 12:16:43 -06003090{
Alex Elder6644ed72013-03-11 23:34:24 -05003091 struct ceph_msg_data *data;
Alex Elder27fa8382013-02-14 12:16:43 -06003092
Alex Elder6644ed72013-03-11 23:34:24 -05003093 BUG_ON(!bio);
Alex Elder6644ed72013-03-11 23:34:24 -05003094
3095 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3096 BUG_ON(!data);
3097 data->bio = bio;
Alex Elderc851c492013-04-05 14:46:01 -05003098 data->bio_length = length;
Alex Elder6644ed72013-03-11 23:34:24 -05003099
Alex Elder5240d9f2013-03-14 14:09:06 -05003100 list_add_tail(&data->links, &msg->data);
3101 msg->data_length += length;
Alex Elder27fa8382013-02-14 12:16:43 -06003102}
Alex Elder90af3602013-04-05 14:46:01 -05003103EXPORT_SYMBOL(ceph_msg_data_add_bio);
Alex Elderea965712013-04-05 14:46:01 -05003104#endif /* CONFIG_BLOCK */
Alex Elder27fa8382013-02-14 12:16:43 -06003105
Sage Weil31b80062009-10-06 11:31:13 -07003106/*
3107 * construct a new message with given type, size
3108 * the new msg has a ref count of 1.
3109 */
Sage Weilb61c2762011-08-09 15:03:46 -07003110struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3111 bool can_fail)
Sage Weil31b80062009-10-06 11:31:13 -07003112{
3113 struct ceph_msg *m;
3114
Alex Eldere3d5d632013-05-01 12:43:04 -05003115 m = kmem_cache_zalloc(ceph_msg_cache, flags);
Sage Weil31b80062009-10-06 11:31:13 -07003116 if (m == NULL)
3117 goto out;
Alex Elder38941f82012-06-01 14:56:43 -05003118
Sage Weil31b80062009-10-06 11:31:13 -07003119 m->hdr.type = cpu_to_le16(type);
Sage Weil45c6ceb2010-05-11 15:01:51 -07003120 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
Sage Weil31b80062009-10-06 11:31:13 -07003121 m->hdr.front_len = cpu_to_le32(front_len);
Sage Weil31b80062009-10-06 11:31:13 -07003122
Alex Elder9516e452013-03-01 18:00:16 -06003123 INIT_LIST_HEAD(&m->list_head);
3124 kref_init(&m->kref);
Alex Elder5240d9f2013-03-14 14:09:06 -05003125 INIT_LIST_HEAD(&m->data);
Henry C Changca208922011-05-03 02:29:56 +00003126
Sage Weil31b80062009-10-06 11:31:13 -07003127 /* front */
3128 if (front_len) {
Ilya Dryomoveeb0bed2014-01-09 20:08:21 +02003129 m->front.iov_base = ceph_kvmalloc(front_len, flags);
Sage Weil31b80062009-10-06 11:31:13 -07003130 if (m->front.iov_base == NULL) {
Sage Weilb61c2762011-08-09 15:03:46 -07003131 dout("ceph_msg_new can't allocate %d bytes\n",
Sage Weil31b80062009-10-06 11:31:13 -07003132 front_len);
3133 goto out2;
3134 }
3135 } else {
3136 m->front.iov_base = NULL;
3137 }
Ilya Dryomovf2be82b2014-01-09 20:08:21 +02003138 m->front_alloc_len = m->front.iov_len = front_len;
Sage Weil31b80062009-10-06 11:31:13 -07003139
Sage Weilbb257662010-04-01 16:07:23 -07003140 dout("ceph_msg_new %p front %d\n", m, front_len);
Sage Weil31b80062009-10-06 11:31:13 -07003141 return m;
3142
3143out2:
3144 ceph_msg_put(m);
3145out:
Sage Weilb61c2762011-08-09 15:03:46 -07003146 if (!can_fail) {
3147 pr_err("msg_new can't create type %d front %d\n", type,
3148 front_len);
Sage Weilf0ed1b72011-08-09 15:05:07 -07003149 WARN_ON(1);
Sage Weilb61c2762011-08-09 15:03:46 -07003150 } else {
3151 dout("msg_new can't create type %d front %d\n", type,
3152 front_len);
3153 }
Sage Weila79832f2010-04-01 16:06:19 -07003154 return NULL;
Sage Weil31b80062009-10-06 11:31:13 -07003155}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003156EXPORT_SYMBOL(ceph_msg_new);
Sage Weil31b80062009-10-06 11:31:13 -07003157
3158/*
Sage Weil31b80062009-10-06 11:31:13 -07003159 * Allocate "middle" portion of a message, if it is needed and wasn't
3160 * allocated by alloc_msg. This allows us to read a small fixed-size
3161 * per-type header in the front and then gracefully fail (i.e.,
3162 * propagate the error to the caller based on info in the front) when
3163 * the middle is too large.
3164 */
Yehuda Sadeh24504182010-01-08 13:58:34 -08003165static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
Sage Weil31b80062009-10-06 11:31:13 -07003166{
3167 int type = le16_to_cpu(msg->hdr.type);
3168 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3169
3170 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3171 ceph_msg_type_name(type), middle_len);
3172 BUG_ON(!middle_len);
3173 BUG_ON(msg->middle);
3174
Sage Weilb6c1d5b2009-12-07 12:17:17 -08003175 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
Sage Weil31b80062009-10-06 11:31:13 -07003176 if (!msg->middle)
3177 return -ENOMEM;
3178 return 0;
3179}
3180
Yehuda Sadeh24504182010-01-08 13:58:34 -08003181/*
Alex Elder1c20f2d2012-06-04 14:43:32 -05003182 * Allocate a message for receiving an incoming message on a
3183 * connection, and save the result in con->in_msg. Uses the
3184 * connection's private alloc_msg op if available.
3185 *
Sage Weil4740a622012-07-30 18:19:30 -07003186 * Returns 0 on success, or a negative error code.
3187 *
3188 * On success, if we set *skip = 1:
3189 * - the next message should be skipped and ignored.
3190 * - con->in_msg == NULL
3191 * or if we set *skip = 0:
3192 * - con->in_msg is non-null.
3193 * On error (ENOMEM, EAGAIN, ...),
3194 * - con->in_msg == NULL
Yehuda Sadeh24504182010-01-08 13:58:34 -08003195 */
Sage Weil4740a622012-07-30 18:19:30 -07003196static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
Yehuda Sadeh24504182010-01-08 13:58:34 -08003197{
Sage Weil4740a622012-07-30 18:19:30 -07003198 struct ceph_msg_header *hdr = &con->in_hdr;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003199 int middle_len = le32_to_cpu(hdr->middle_len);
Alex Elder1d866d12013-03-01 18:00:14 -06003200 struct ceph_msg *msg;
Sage Weil4740a622012-07-30 18:19:30 -07003201 int ret = 0;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003202
Alex Elder1c20f2d2012-06-04 14:43:32 -05003203 BUG_ON(con->in_msg != NULL);
Alex Elder53ded492013-03-01 18:00:14 -06003204 BUG_ON(!con->ops->alloc_msg);
Yehuda Sadeh24504182010-01-08 13:58:34 -08003205
Alex Elder53ded492013-03-01 18:00:14 -06003206 mutex_unlock(&con->mutex);
3207 msg = con->ops->alloc_msg(con, hdr, skip);
3208 mutex_lock(&con->mutex);
3209 if (con->state != CON_STATE_OPEN) {
3210 if (msg)
Alex Elder1d866d12013-03-01 18:00:14 -06003211 ceph_msg_put(msg);
Alex Elder53ded492013-03-01 18:00:14 -06003212 return -EAGAIN;
3213 }
Alex Elder41375772013-03-05 09:25:10 -06003214 if (msg) {
3215 BUG_ON(*skip);
3216 con->in_msg = msg;
Sage Weil36eb71a2012-06-21 12:47:08 -07003217 con->in_msg->con = con->ops->get(con);
Alex Elder92ce0342012-06-04 14:43:33 -05003218 BUG_ON(con->in_msg->con == NULL);
Alex Elder41375772013-03-05 09:25:10 -06003219 } else {
3220 /*
3221 * Null message pointer means either we should skip
3222 * this message or we couldn't allocate memory. The
3223 * former is not an error.
3224 */
3225 if (*skip)
3226 return 0;
3227 con->error_msg = "error allocating memory for incoming message";
3228
Alex Elder53ded492013-03-01 18:00:14 -06003229 return -ENOMEM;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003230 }
Alex Elder1c20f2d2012-06-04 14:43:32 -05003231 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
Yehuda Sadeh24504182010-01-08 13:58:34 -08003232
Alex Elder1c20f2d2012-06-04 14:43:32 -05003233 if (middle_len && !con->in_msg->middle) {
3234 ret = ceph_alloc_middle(con, con->in_msg);
Yehuda Sadeh24504182010-01-08 13:58:34 -08003235 if (ret < 0) {
Alex Elder1c20f2d2012-06-04 14:43:32 -05003236 ceph_msg_put(con->in_msg);
3237 con->in_msg = NULL;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003238 }
3239 }
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08003240
Sage Weil4740a622012-07-30 18:19:30 -07003241 return ret;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003242}
3243
Sage Weil31b80062009-10-06 11:31:13 -07003244
3245/*
3246 * Free a generically kmalloc'd message.
3247 */
3248void ceph_msg_kfree(struct ceph_msg *m)
3249{
3250 dout("msg_kfree %p\n", m);
Ilya Dryomoveeb0bed2014-01-09 20:08:21 +02003251 ceph_kvfree(m->front.iov_base);
Alex Eldere3d5d632013-05-01 12:43:04 -05003252 kmem_cache_free(ceph_msg_cache, m);
Sage Weil31b80062009-10-06 11:31:13 -07003253}
3254
3255/*
3256 * Drop a msg ref. Destroy as needed.
3257 */
Sage Weilc2e552e2009-12-07 15:55:05 -08003258void ceph_msg_last_put(struct kref *kref)
Sage Weil31b80062009-10-06 11:31:13 -07003259{
Sage Weilc2e552e2009-12-07 15:55:05 -08003260 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
Alex Elder5240d9f2013-03-14 14:09:06 -05003261 LIST_HEAD(data);
3262 struct list_head *links;
3263 struct list_head *next;
Sage Weil31b80062009-10-06 11:31:13 -07003264
Sage Weilc2e552e2009-12-07 15:55:05 -08003265 dout("ceph_msg_put last one on %p\n", m);
3266 WARN_ON(!list_empty(&m->list_head));
Sage Weil31b80062009-10-06 11:31:13 -07003267
Sage Weilc2e552e2009-12-07 15:55:05 -08003268 /* drop middle, data, if any */
3269 if (m->middle) {
3270 ceph_buffer_put(m->middle);
3271 m->middle = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07003272 }
Alex Elder5240d9f2013-03-14 14:09:06 -05003273
3274 list_splice_init(&m->data, &data);
3275 list_for_each_safe(links, next, &data) {
3276 struct ceph_msg_data *data;
3277
3278 data = list_entry(links, struct ceph_msg_data, links);
3279 list_del_init(links);
3280 ceph_msg_data_destroy(data);
3281 }
Alex Eldera1930802013-03-14 14:09:06 -05003282 m->data_length = 0;
Sage Weil58bb3b32009-12-23 12:12:31 -08003283
Sage Weilc2e552e2009-12-07 15:55:05 -08003284 if (m->pool)
3285 ceph_msgpool_put(m->pool, m);
3286 else
3287 ceph_msg_kfree(m);
Sage Weil31b80062009-10-06 11:31:13 -07003288}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003289EXPORT_SYMBOL(ceph_msg_last_put);
Sage Weil9ec7cab2009-12-14 15:13:47 -08003290
3291void ceph_msg_dump(struct ceph_msg *msg)
3292{
Ilya Dryomov3cea4c32014-01-09 20:08:21 +02003293 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3294 msg->front_alloc_len, msg->data_length);
Sage Weil9ec7cab2009-12-14 15:13:47 -08003295 print_hex_dump(KERN_DEBUG, "header: ",
3296 DUMP_PREFIX_OFFSET, 16, 1,
3297 &msg->hdr, sizeof(msg->hdr), true);
3298 print_hex_dump(KERN_DEBUG, " front: ",
3299 DUMP_PREFIX_OFFSET, 16, 1,
3300 msg->front.iov_base, msg->front.iov_len, true);
3301 if (msg->middle)
3302 print_hex_dump(KERN_DEBUG, "middle: ",
3303 DUMP_PREFIX_OFFSET, 16, 1,
3304 msg->middle->vec.iov_base,
3305 msg->middle->vec.iov_len, true);
3306 print_hex_dump(KERN_DEBUG, "footer: ",
3307 DUMP_PREFIX_OFFSET, 16, 1,
3308 &msg->footer, sizeof(msg->footer), true);
3309}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003310EXPORT_SYMBOL(ceph_msg_dump);