blob: 1948d592aa54c7a1831df546702904898cd68da4 [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 */
David S. Miller676d2362014-04-11 16:15:36 -0400386static void ceph_sock_data_ready(struct sock *sk)
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 Eldera5bc3122012-01-23 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 Eldera5bc3122012-01-23 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
Chunwei Chen178eda22014-04-23 12:35:09 +0800560static 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
Chunwei Chen178eda22014-04-23 12:35:09 +0800573static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
574 int offset, size_t size, bool more)
575{
576 int ret;
577 struct kvec iov;
578
579 /* sendpage cannot properly handle pages with page_count == 0,
580 * we need to fallback to sendmsg if that's the case */
581 if (page_count(page) >= 1)
582 return __ceph_tcp_sendpage(sock, page, offset, size, more);
583
584 iov.iov_base = kmap(page) + offset;
585 iov.iov_len = size;
586 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
587 kunmap(page);
588
589 return ret;
590}
Sage Weil31b80062009-10-06 11:31:13 -0700591
592/*
593 * Shutdown/close the socket for the given connection.
594 */
595static int con_close_socket(struct ceph_connection *con)
596{
Sage Weil8007b8d2012-07-30 18:16:16 -0700597 int rc = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700598
599 dout("con_close_socket on %p sock %p\n", con, con->sock);
Sage Weil8007b8d2012-07-30 18:16:16 -0700600 if (con->sock) {
601 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
602 sock_release(con->sock);
603 con->sock = NULL;
604 }
Alex Elder456ea462012-06-20 21:53:53 -0500605
606 /*
Sage Weil4a861692012-07-20 17:29:55 -0700607 * Forcibly clear the SOCK_CLOSED flag. It gets set
Alex Elder456ea462012-06-20 21:53:53 -0500608 * independent of the connection mutex, and we could have
609 * received a socket close event before we had the chance to
610 * shut the socket down.
611 */
Alex Elderc9ffc772013-02-20 10:25:12 -0600612 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
Sage Weil8007b8d2012-07-30 18:16:16 -0700613
Alex Elderce2c8902012-05-22 22:15:49 -0500614 con_sock_state_closed(con);
Sage Weil31b80062009-10-06 11:31:13 -0700615 return rc;
616}
617
618/*
619 * Reset a connection. Discard all incoming and outgoing messages
620 * and clear *_seq state.
621 */
622static void ceph_msg_remove(struct ceph_msg *msg)
623{
624 list_del_init(&msg->list_head);
Alex Elder38941f82012-06-01 14:56:43 -0500625 BUG_ON(msg->con == NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -0700626 msg->con->ops->put(msg->con);
Alex Elder38941f82012-06-01 14:56:43 -0500627 msg->con = NULL;
628
Sage Weil31b80062009-10-06 11:31:13 -0700629 ceph_msg_put(msg);
630}
631static void ceph_msg_remove_list(struct list_head *head)
632{
633 while (!list_empty(head)) {
634 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
635 list_head);
636 ceph_msg_remove(msg);
637 }
638}
639
640static void reset_connection(struct ceph_connection *con)
641{
642 /* reset connection, out_queue, msg_ and connect_seq */
643 /* discard existing out_queue and msg_seq */
Sage Weil0fa6ebc2012-12-27 20:27:04 -0600644 dout("reset_connection %p\n", con);
Sage Weil31b80062009-10-06 11:31:13 -0700645 ceph_msg_remove_list(&con->out_queue);
646 ceph_msg_remove_list(&con->out_sent);
647
Sage Weilcf3e5c42009-12-11 09:48:05 -0800648 if (con->in_msg) {
Alex Elder38941f82012-06-01 14:56:43 -0500649 BUG_ON(con->in_msg->con != con);
650 con->in_msg->con = NULL;
Sage Weilcf3e5c42009-12-11 09:48:05 -0800651 ceph_msg_put(con->in_msg);
652 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -0700653 con->ops->put(con);
Sage Weilcf3e5c42009-12-11 09:48:05 -0800654 }
655
Sage Weil31b80062009-10-06 11:31:13 -0700656 con->connect_seq = 0;
657 con->out_seq = 0;
Sage Weilc86a2932009-12-14 14:04:30 -0800658 if (con->out_msg) {
659 ceph_msg_put(con->out_msg);
660 con->out_msg = NULL;
661 }
Sage Weil31b80062009-10-06 11:31:13 -0700662 con->in_seq = 0;
Sage Weil0e0d5e02010-04-02 16:07:19 -0700663 con->in_seq_acked = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700664}
665
666/*
667 * mark a peer down. drop any open connections.
668 */
669void ceph_con_close(struct ceph_connection *con)
670{
Sage Weil8c50c812012-07-30 16:24:37 -0700671 mutex_lock(&con->mutex);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700672 dout("con_close %p peer %s\n", con,
673 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil8dacc7d2012-07-20 17:24:40 -0700674 con->state = CON_STATE_CLOSED;
Alex Eldera5988c42012-05-29 11:04:58 -0500675
Alex Elderc9ffc772013-02-20 10:25:12 -0600676 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
677 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
678 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
679 con_flag_clear(con, CON_FLAG_BACKOFF);
Alex Eldera5988c42012-05-29 11:04:58 -0500680
Sage Weil31b80062009-10-06 11:31:13 -0700681 reset_connection(con);
Sage Weil6f2bc3f2010-04-02 16:16:34 -0700682 con->peer_global_seq = 0;
Sage Weil91e45ce32010-02-15 12:05:09 -0800683 cancel_delayed_work(&con->work);
Sage Weilee76e072012-07-20 16:45:49 -0700684 con_close_socket(con);
Sage Weilec302642009-12-22 10:43:42 -0800685 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700686}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700687EXPORT_SYMBOL(ceph_con_close);
Sage Weil31b80062009-10-06 11:31:13 -0700688
689/*
Sage Weil31b80062009-10-06 11:31:13 -0700690 * Reopen a closed connection, with a new peer address.
691 */
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700692void ceph_con_open(struct ceph_connection *con,
693 __u8 entity_type, __u64 entity_num,
694 struct ceph_entity_addr *addr)
Sage Weil31b80062009-10-06 11:31:13 -0700695{
Sage Weil54691552012-07-30 16:21:40 -0700696 mutex_lock(&con->mutex);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700697 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
Sage Weil8dacc7d2012-07-20 17:24:40 -0700698
Alex Elder122070a2012-12-26 10:43:57 -0600699 WARN_ON(con->state != CON_STATE_CLOSED);
Sage Weil8dacc7d2012-07-20 17:24:40 -0700700 con->state = CON_STATE_PREOPEN;
Alex Eldera5988c42012-05-29 11:04:58 -0500701
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700702 con->peer_name.type = (__u8) entity_type;
703 con->peer_name.num = cpu_to_le64(entity_num);
704
Sage Weil31b80062009-10-06 11:31:13 -0700705 memcpy(&con->peer_addr, addr, sizeof(*addr));
Sage Weil03c677e2009-11-20 15:14:15 -0800706 con->delay = 0; /* reset backoff memory */
Sage Weil54691552012-07-30 16:21:40 -0700707 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700708 queue_con(con);
709}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700710EXPORT_SYMBOL(ceph_con_open);
Sage Weil31b80062009-10-06 11:31:13 -0700711
712/*
Sage Weil87b315a2010-03-22 14:51:18 -0700713 * return true if this connection ever successfully opened
714 */
715bool ceph_con_opened(struct ceph_connection *con)
716{
717 return con->connect_seq > 0;
718}
719
720/*
Sage Weil31b80062009-10-06 11:31:13 -0700721 * initialize a new connection.
722 */
Alex Elder1bfd89f2012-05-26 23:26:43 -0500723void ceph_con_init(struct ceph_connection *con, void *private,
724 const struct ceph_connection_operations *ops,
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700725 struct ceph_messenger *msgr)
Sage Weil31b80062009-10-06 11:31:13 -0700726{
727 dout("con_init %p\n", con);
728 memset(con, 0, sizeof(*con));
Alex Elder1bfd89f2012-05-26 23:26:43 -0500729 con->private = private;
730 con->ops = ops;
Sage Weil31b80062009-10-06 11:31:13 -0700731 con->msgr = msgr;
Alex Elderce2c8902012-05-22 22:15:49 -0500732
733 con_sock_state_init(con);
734
Sage Weilec302642009-12-22 10:43:42 -0800735 mutex_init(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700736 INIT_LIST_HEAD(&con->out_queue);
737 INIT_LIST_HEAD(&con->out_sent);
738 INIT_DELAYED_WORK(&con->work, con_work);
Alex Eldera5988c42012-05-29 11:04:58 -0500739
Sage Weil8dacc7d2012-07-20 17:24:40 -0700740 con->state = CON_STATE_CLOSED;
Sage Weil31b80062009-10-06 11:31:13 -0700741}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700742EXPORT_SYMBOL(ceph_con_init);
Sage Weil31b80062009-10-06 11:31:13 -0700743
744
745/*
746 * We maintain a global counter to order connection attempts. Get
747 * a unique seq greater than @gt.
748 */
749static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
750{
751 u32 ret;
752
753 spin_lock(&msgr->global_seq_lock);
754 if (msgr->global_seq < gt)
755 msgr->global_seq = gt;
756 ret = ++msgr->global_seq;
757 spin_unlock(&msgr->global_seq_lock);
758 return ret;
759}
760
Alex Eldere2200422012-05-23 14:35:23 -0500761static void con_out_kvec_reset(struct ceph_connection *con)
Alex Elder859eb792012-02-14 14:05:33 -0600762{
763 con->out_kvec_left = 0;
764 con->out_kvec_bytes = 0;
765 con->out_kvec_cur = &con->out_kvec[0];
766}
767
Alex Eldere2200422012-05-23 14:35:23 -0500768static void con_out_kvec_add(struct ceph_connection *con,
Alex Elder859eb792012-02-14 14:05:33 -0600769 size_t size, void *data)
770{
771 int index;
772
773 index = con->out_kvec_left;
774 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
775
776 con->out_kvec[index].iov_len = size;
777 con->out_kvec[index].iov_base = data;
778 con->out_kvec_left++;
779 con->out_kvec_bytes += size;
780}
Sage Weil31b80062009-10-06 11:31:13 -0700781
Alex Elderdf6ad1f2012-06-11 14:57:13 -0500782#ifdef CONFIG_BLOCK
Alex Elder6aaa4512013-03-06 23:39:39 -0600783
784/*
785 * For a bio data item, a piece is whatever remains of the next
786 * entry in the current bio iovec, or the first entry in the next
787 * bio in the list.
788 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500789static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500790 size_t length)
Alex Elder6aaa4512013-03-06 23:39:39 -0600791{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500792 struct ceph_msg_data *data = cursor->data;
Alex Elder6aaa4512013-03-06 23:39:39 -0600793 struct bio *bio;
794
795 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
796
797 bio = data->bio;
798 BUG_ON(!bio);
Alex Elder6aaa4512013-03-06 23:39:39 -0600799
Alex Elderca8b3a62013-04-05 14:46:01 -0500800 cursor->resid = min(length, data->bio_length);
Alex Elder6aaa4512013-03-06 23:39:39 -0600801 cursor->bio = bio;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700802 cursor->bvec_iter = bio->bi_iter;
803 cursor->last_piece =
804 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
Alex Elder6aaa4512013-03-06 23:39:39 -0600805}
806
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500807static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
Alex Elder6aaa4512013-03-06 23:39:39 -0600808 size_t *page_offset,
809 size_t *length)
810{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500811 struct ceph_msg_data *data = cursor->data;
Alex Elder6aaa4512013-03-06 23:39:39 -0600812 struct bio *bio;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700813 struct bio_vec bio_vec;
Alex Elder6aaa4512013-03-06 23:39:39 -0600814
815 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
816
817 bio = cursor->bio;
818 BUG_ON(!bio);
819
Kent Overstreetf38a5182013-08-07 14:30:24 -0700820 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
Alex Elder6aaa4512013-03-06 23:39:39 -0600821
Kent Overstreetf38a5182013-08-07 14:30:24 -0700822 *page_offset = (size_t) bio_vec.bv_offset;
Alex Elder6aaa4512013-03-06 23:39:39 -0600823 BUG_ON(*page_offset >= PAGE_SIZE);
Alex Elder25aff7c2013-03-11 23:34:22 -0500824 if (cursor->last_piece) /* pagelist offset is always 0 */
825 *length = cursor->resid;
826 else
Kent Overstreetf38a5182013-08-07 14:30:24 -0700827 *length = (size_t) bio_vec.bv_len;
Alex Elder25aff7c2013-03-11 23:34:22 -0500828 BUG_ON(*length > cursor->resid);
Alex Elder5df521b2013-03-30 15:09:59 -0500829 BUG_ON(*page_offset + *length > PAGE_SIZE);
Alex Elder6aaa4512013-03-06 23:39:39 -0600830
Kent Overstreetf38a5182013-08-07 14:30:24 -0700831 return bio_vec.bv_page;
Alex Elder6aaa4512013-03-06 23:39:39 -0600832}
833
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500834static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
835 size_t bytes)
Alex Elder6aaa4512013-03-06 23:39:39 -0600836{
Alex Elder6aaa4512013-03-06 23:39:39 -0600837 struct bio *bio;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700838 struct bio_vec bio_vec;
Alex Elder6aaa4512013-03-06 23:39:39 -0600839
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500840 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
Alex Elder6aaa4512013-03-06 23:39:39 -0600841
842 bio = cursor->bio;
843 BUG_ON(!bio);
844
Kent Overstreetf38a5182013-08-07 14:30:24 -0700845 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
Alex Elder6aaa4512013-03-06 23:39:39 -0600846
847 /* Advance the cursor offset */
848
Alex Elder25aff7c2013-03-11 23:34:22 -0500849 BUG_ON(cursor->resid < bytes);
850 cursor->resid -= bytes;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700851
852 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
853
854 if (bytes < bio_vec.bv_len)
Alex Elder6aaa4512013-03-06 23:39:39 -0600855 return false; /* more bytes to process in this segment */
856
857 /* Move on to the next segment, and possibly the next bio */
858
Kent Overstreetf38a5182013-08-07 14:30:24 -0700859 if (!cursor->bvec_iter.bi_size) {
Alex Elder6aaa4512013-03-06 23:39:39 -0600860 bio = bio->bi_next;
Ilya Dryomov0ec1d152014-02-05 15:19:55 +0200861 cursor->bio = bio;
862 if (bio)
863 cursor->bvec_iter = bio->bi_iter;
864 else
865 memset(&cursor->bvec_iter, 0,
866 sizeof(cursor->bvec_iter));
Alex Elder6aaa4512013-03-06 23:39:39 -0600867 }
Alex Elder6aaa4512013-03-06 23:39:39 -0600868
Alex Elder25aff7c2013-03-11 23:34:22 -0500869 if (!cursor->last_piece) {
870 BUG_ON(!cursor->resid);
871 BUG_ON(!bio);
872 /* A short read is OK, so use <= rather than == */
Kent Overstreetf38a5182013-08-07 14:30:24 -0700873 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
Alex Elder6aaa4512013-03-06 23:39:39 -0600874 cursor->last_piece = true;
Alex Elder25aff7c2013-03-11 23:34:22 -0500875 }
Alex Elder6aaa4512013-03-06 23:39:39 -0600876
877 return true;
878}
Alex Elderea965712013-04-05 14:46:01 -0500879#endif /* CONFIG_BLOCK */
Alex Elderdf6ad1f2012-06-11 14:57:13 -0500880
Alex Elderfe38a2b2013-03-06 23:39:39 -0600881/*
Alex Eldere766d7b2013-03-07 15:38:28 -0600882 * For a page array, a piece comes from the first page in the array
883 * that has not already been fully consumed.
884 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500885static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500886 size_t length)
Alex Eldere766d7b2013-03-07 15:38:28 -0600887{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500888 struct ceph_msg_data *data = cursor->data;
Alex Eldere766d7b2013-03-07 15:38:28 -0600889 int page_count;
890
891 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
892
893 BUG_ON(!data->pages);
894 BUG_ON(!data->length);
895
Alex Elderca8b3a62013-04-05 14:46:01 -0500896 cursor->resid = min(length, data->length);
Alex Eldere766d7b2013-03-07 15:38:28 -0600897 page_count = calc_pages_for(data->alignment, (u64)data->length);
Alex Eldere766d7b2013-03-07 15:38:28 -0600898 cursor->page_offset = data->alignment & ~PAGE_MASK;
899 cursor->page_index = 0;
Alex Elder56fc5652013-03-30 23:46:55 -0500900 BUG_ON(page_count > (int)USHRT_MAX);
901 cursor->page_count = (unsigned short)page_count;
902 BUG_ON(length > SIZE_MAX - cursor->page_offset);
903 cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE;
Alex Eldere766d7b2013-03-07 15:38:28 -0600904}
905
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500906static struct page *
907ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
908 size_t *page_offset, size_t *length)
Alex Eldere766d7b2013-03-07 15:38:28 -0600909{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500910 struct ceph_msg_data *data = cursor->data;
Alex Eldere766d7b2013-03-07 15:38:28 -0600911
912 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
913
914 BUG_ON(cursor->page_index >= cursor->page_count);
915 BUG_ON(cursor->page_offset >= PAGE_SIZE);
Alex Eldere766d7b2013-03-07 15:38:28 -0600916
917 *page_offset = cursor->page_offset;
Alex Elder25aff7c2013-03-11 23:34:22 -0500918 if (cursor->last_piece)
Alex Eldere766d7b2013-03-07 15:38:28 -0600919 *length = cursor->resid;
Alex Elder25aff7c2013-03-11 23:34:22 -0500920 else
Alex Eldere766d7b2013-03-07 15:38:28 -0600921 *length = PAGE_SIZE - *page_offset;
Alex Eldere766d7b2013-03-07 15:38:28 -0600922
923 return data->pages[cursor->page_index];
924}
925
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500926static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
Alex Eldere766d7b2013-03-07 15:38:28 -0600927 size_t bytes)
928{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500929 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
Alex Eldere766d7b2013-03-07 15:38:28 -0600930
931 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
Alex Eldere766d7b2013-03-07 15:38:28 -0600932
933 /* Advance the cursor page offset */
934
935 cursor->resid -= bytes;
Alex Elder5df521b2013-03-30 15:09:59 -0500936 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
937 if (!bytes || cursor->page_offset)
Alex Eldere766d7b2013-03-07 15:38:28 -0600938 return false; /* more bytes to process in the current page */
939
Yan, Zhengd90deda2014-03-23 06:50:39 +0800940 if (!cursor->resid)
941 return false; /* no more data */
942
Alex Elder5df521b2013-03-30 15:09:59 -0500943 /* Move on to the next page; offset is already at 0 */
Alex Eldere766d7b2013-03-07 15:38:28 -0600944
945 BUG_ON(cursor->page_index >= cursor->page_count);
Alex Eldere766d7b2013-03-07 15:38:28 -0600946 cursor->page_index++;
Alex Elder25aff7c2013-03-11 23:34:22 -0500947 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Eldere766d7b2013-03-07 15:38:28 -0600948
949 return true;
950}
951
952/*
Alex Elderdd236fc2013-03-06 23:39:39 -0600953 * For a pagelist, a piece is whatever remains to be consumed in the
954 * first page in the list, or the front of the next page.
Alex Elderfe38a2b2013-03-06 23:39:39 -0600955 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500956static void
957ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500958 size_t length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600959{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500960 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600961 struct ceph_pagelist *pagelist;
962 struct page *page;
963
Alex Elderdd236fc2013-03-06 23:39:39 -0600964 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600965
966 pagelist = data->pagelist;
967 BUG_ON(!pagelist);
Alex Elder25aff7c2013-03-11 23:34:22 -0500968
969 if (!length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600970 return; /* pagelist can be assigned but empty */
971
972 BUG_ON(list_empty(&pagelist->head));
973 page = list_first_entry(&pagelist->head, struct page, lru);
974
Alex Elderca8b3a62013-04-05 14:46:01 -0500975 cursor->resid = min(length, pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600976 cursor->page = page;
977 cursor->offset = 0;
Alex Eldera51b272e2013-04-19 15:34:49 -0500978 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600979}
980
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500981static struct page *
982ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
983 size_t *page_offset, size_t *length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600984{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500985 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600986 struct ceph_pagelist *pagelist;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600987
988 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
989
990 pagelist = data->pagelist;
991 BUG_ON(!pagelist);
992
993 BUG_ON(!cursor->page);
Alex Elder25aff7c2013-03-11 23:34:22 -0500994 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600995
Alex Elder5df521b2013-03-30 15:09:59 -0500996 /* offset of first page in pagelist is always 0 */
Alex Elderfe38a2b2013-03-06 23:39:39 -0600997 *page_offset = cursor->offset & ~PAGE_MASK;
Alex Elder5df521b2013-03-30 15:09:59 -0500998 if (cursor->last_piece)
Alex Elder25aff7c2013-03-11 23:34:22 -0500999 *length = cursor->resid;
1000 else
1001 *length = PAGE_SIZE - *page_offset;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001002
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001003 return cursor->page;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001004}
1005
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001006static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
Alex Elderdd236fc2013-03-06 23:39:39 -06001007 size_t bytes)
Alex Elderfe38a2b2013-03-06 23:39:39 -06001008{
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001009 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001010 struct ceph_pagelist *pagelist;
1011
1012 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1013
1014 pagelist = data->pagelist;
1015 BUG_ON(!pagelist);
Alex Elder25aff7c2013-03-11 23:34:22 -05001016
1017 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -06001018 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1019
1020 /* Advance the cursor offset */
1021
Alex Elder25aff7c2013-03-11 23:34:22 -05001022 cursor->resid -= bytes;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001023 cursor->offset += bytes;
Alex Elder5df521b2013-03-30 15:09:59 -05001024 /* offset of first page in pagelist is always 0 */
Alex Elderfe38a2b2013-03-06 23:39:39 -06001025 if (!bytes || cursor->offset & ~PAGE_MASK)
1026 return false; /* more bytes to process in the current page */
1027
Yan, Zhengd90deda2014-03-23 06:50:39 +08001028 if (!cursor->resid)
1029 return false; /* no more data */
1030
Alex Elderfe38a2b2013-03-06 23:39:39 -06001031 /* Move on to the next page */
1032
1033 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1034 cursor->page = list_entry_next(cursor->page, lru);
Alex Elder25aff7c2013-03-11 23:34:22 -05001035 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001036
1037 return true;
1038}
1039
Alex Elderdd236fc2013-03-06 23:39:39 -06001040/*
1041 * Message data is handled (sent or received) in pieces, where each
1042 * piece resides on a single page. The network layer might not
1043 * consume an entire piece at once. A data item's cursor keeps
1044 * track of which piece is next to process and how much remains to
1045 * be processed in that piece. It also tracks whether the current
1046 * piece is the last one in the data item.
1047 */
Alex Elderca8b3a62013-04-05 14:46:01 -05001048static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
Alex Elderdd236fc2013-03-06 23:39:39 -06001049{
Alex Elderca8b3a62013-04-05 14:46:01 -05001050 size_t length = cursor->total_resid;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001051
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001052 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001053 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001054 ceph_msg_data_pagelist_cursor_init(cursor, length);
Alex Elderdd236fc2013-03-06 23:39:39 -06001055 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001056 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001057 ceph_msg_data_pages_cursor_init(cursor, length);
Alex Eldere766d7b2013-03-07 15:38:28 -06001058 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001059#ifdef CONFIG_BLOCK
1060 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001061 ceph_msg_data_bio_cursor_init(cursor, length);
Alex Elder6aaa4512013-03-06 23:39:39 -06001062 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001063#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001064 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001065 default:
1066 /* BUG(); */
1067 break;
1068 }
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001069 cursor->need_crc = true;
Alex Elderdd236fc2013-03-06 23:39:39 -06001070}
1071
Alex Elderca8b3a62013-04-05 14:46:01 -05001072static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1073{
1074 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1075 struct ceph_msg_data *data;
1076
1077 BUG_ON(!length);
1078 BUG_ON(length > msg->data_length);
1079 BUG_ON(list_empty(&msg->data));
1080
Alex Elderca8b3a62013-04-05 14:46:01 -05001081 cursor->data_head = &msg->data;
1082 cursor->total_resid = length;
1083 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1084 cursor->data = data;
1085
1086 __ceph_msg_data_cursor_init(cursor);
1087}
1088
Alex Elderdd236fc2013-03-06 23:39:39 -06001089/*
1090 * Return the page containing the next piece to process for a given
1091 * data item, and supply the page offset and length of that piece.
1092 * Indicate whether this is the last piece in this data item.
1093 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001094static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1095 size_t *page_offset, size_t *length,
Alex Elderdd236fc2013-03-06 23:39:39 -06001096 bool *last_piece)
1097{
1098 struct page *page;
1099
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001100 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001101 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001102 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
Alex Elderdd236fc2013-03-06 23:39:39 -06001103 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001104 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001105 page = ceph_msg_data_pages_next(cursor, page_offset, length);
Alex Eldere766d7b2013-03-07 15:38:28 -06001106 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001107#ifdef CONFIG_BLOCK
1108 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001109 page = ceph_msg_data_bio_next(cursor, page_offset, length);
Alex Elder6aaa4512013-03-06 23:39:39 -06001110 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001111#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001112 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001113 default:
1114 page = NULL;
1115 break;
1116 }
1117 BUG_ON(!page);
1118 BUG_ON(*page_offset + *length > PAGE_SIZE);
1119 BUG_ON(!*length);
1120 if (last_piece)
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001121 *last_piece = cursor->last_piece;
Alex Elderdd236fc2013-03-06 23:39:39 -06001122
1123 return page;
1124}
1125
1126/*
1127 * Returns true if the result moves the cursor on to the next piece
1128 * of the data item.
1129 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001130static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1131 size_t bytes)
Alex Elderdd236fc2013-03-06 23:39:39 -06001132{
1133 bool new_piece;
1134
Alex Elder25aff7c2013-03-11 23:34:22 -05001135 BUG_ON(bytes > cursor->resid);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001136 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001137 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001138 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
Alex Elderdd236fc2013-03-06 23:39:39 -06001139 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001140 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001141 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
Alex Eldere766d7b2013-03-07 15:38:28 -06001142 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001143#ifdef CONFIG_BLOCK
1144 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001145 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
Alex Elder6aaa4512013-03-06 23:39:39 -06001146 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001147#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001148 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001149 default:
1150 BUG();
1151 break;
1152 }
Alex Elderca8b3a62013-04-05 14:46:01 -05001153 cursor->total_resid -= bytes;
Alex Elderdd236fc2013-03-06 23:39:39 -06001154
Alex Elderca8b3a62013-04-05 14:46:01 -05001155 if (!cursor->resid && cursor->total_resid) {
1156 WARN_ON(!cursor->last_piece);
1157 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1158 cursor->data = list_entry_next(cursor->data, links);
1159 __ceph_msg_data_cursor_init(cursor);
Alex Eldera51b272e2013-04-19 15:34:49 -05001160 new_piece = true;
Alex Elderca8b3a62013-04-05 14:46:01 -05001161 }
Alex Eldera51b272e2013-04-19 15:34:49 -05001162 cursor->need_crc = new_piece;
Alex Elderca8b3a62013-04-05 14:46:01 -05001163
Alex Elderdd236fc2013-03-06 23:39:39 -06001164 return new_piece;
1165}
1166
Alex Elder98fa5dd2013-04-02 12:09:50 -05001167static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
Alex Elder739c9052012-06-11 14:57:13 -05001168{
Alex Elder739c9052012-06-11 14:57:13 -05001169 BUG_ON(!msg);
Alex Elder25aff7c2013-03-11 23:34:22 -05001170 BUG_ON(!data_len);
Alex Elder739c9052012-06-11 14:57:13 -05001171
Alex Elder4c59b4a2013-03-11 23:34:23 -05001172 /* Initialize data cursor */
Alex Elderfe38a2b2013-03-06 23:39:39 -06001173
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001174 ceph_msg_data_cursor_init(msg, (size_t)data_len);
Alex Elder739c9052012-06-11 14:57:13 -05001175}
1176
Sage Weil31b80062009-10-06 11:31:13 -07001177/*
1178 * Prepare footer for currently outgoing message, and finish things
1179 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1180 */
Alex Elder859eb792012-02-14 14:05:33 -06001181static void prepare_write_message_footer(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001182{
1183 struct ceph_msg *m = con->out_msg;
Alex Elder859eb792012-02-14 14:05:33 -06001184 int v = con->out_kvec_left;
Sage Weil31b80062009-10-06 11:31:13 -07001185
Alex Elderfd154f32012-06-11 14:57:13 -05001186 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1187
Sage Weil31b80062009-10-06 11:31:13 -07001188 dout("prepare_write_message_footer %p\n", con);
1189 con->out_kvec_is_msg = true;
1190 con->out_kvec[v].iov_base = &m->footer;
1191 con->out_kvec[v].iov_len = sizeof(m->footer);
1192 con->out_kvec_bytes += sizeof(m->footer);
1193 con->out_kvec_left++;
1194 con->out_more = m->more_to_follow;
Sage Weilc86a2932009-12-14 14:04:30 -08001195 con->out_msg_done = true;
Sage Weil31b80062009-10-06 11:31:13 -07001196}
1197
1198/*
1199 * Prepare headers for the next outgoing message.
1200 */
1201static void prepare_write_message(struct ceph_connection *con)
1202{
1203 struct ceph_msg *m;
Alex Eldera9a0c512012-02-15 07:43:54 -06001204 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07001205
Alex Eldere2200422012-05-23 14:35:23 -05001206 con_out_kvec_reset(con);
Sage Weil31b80062009-10-06 11:31:13 -07001207 con->out_kvec_is_msg = true;
Sage Weilc86a2932009-12-14 14:04:30 -08001208 con->out_msg_done = false;
Sage Weil31b80062009-10-06 11:31:13 -07001209
1210 /* Sneak an ack in there first? If we can get it into the same
1211 * TCP packet that's a good thing. */
1212 if (con->in_seq > con->in_seq_acked) {
1213 con->in_seq_acked = con->in_seq;
Alex Eldere2200422012-05-23 14:35:23 -05001214 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
Sage Weil31b80062009-10-06 11:31:13 -07001215 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
Alex Eldere2200422012-05-23 14:35:23 -05001216 con_out_kvec_add(con, sizeof (con->out_temp_ack),
Alex Elder859eb792012-02-14 14:05:33 -06001217 &con->out_temp_ack);
Sage Weil31b80062009-10-06 11:31:13 -07001218 }
1219
Alex Elder38941f82012-06-01 14:56:43 -05001220 BUG_ON(list_empty(&con->out_queue));
Alex Elder859eb792012-02-14 14:05:33 -06001221 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
Sage Weilc86a2932009-12-14 14:04:30 -08001222 con->out_msg = m;
Alex Elder38941f82012-06-01 14:56:43 -05001223 BUG_ON(m->con != con);
Sage Weil4cf9d542011-07-26 11:27:24 -07001224
1225 /* put message on sent list */
1226 ceph_msg_get(m);
1227 list_move_tail(&m->list_head, &con->out_sent);
Sage Weil31b80062009-10-06 11:31:13 -07001228
Sage Weile84346b2010-05-11 21:20:38 -07001229 /*
1230 * only assign outgoing seq # if we haven't sent this message
1231 * yet. if it is requeued, resend with it's original seq.
1232 */
1233 if (m->needs_out_seq) {
1234 m->hdr.seq = cpu_to_le64(++con->out_seq);
1235 m->needs_out_seq = false;
1236 }
Alex Elder98fa5dd2013-04-02 12:09:50 -05001237 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
Sage Weil31b80062009-10-06 11:31:13 -07001238
Alex Elder98fa5dd2013-04-02 12:09:50 -05001239 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
Sage Weil31b80062009-10-06 11:31:13 -07001240 m, con->out_seq, le16_to_cpu(m->hdr.type),
1241 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
Alex Elder98fa5dd2013-04-02 12:09:50 -05001242 m->data_length);
Sage Weil31b80062009-10-06 11:31:13 -07001243 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1244
1245 /* tag + hdr + front + middle */
Alex Eldere2200422012-05-23 14:35:23 -05001246 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1247 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1248 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
Alex Elder859eb792012-02-14 14:05:33 -06001249
Sage Weil31b80062009-10-06 11:31:13 -07001250 if (m->middle)
Alex Eldere2200422012-05-23 14:35:23 -05001251 con_out_kvec_add(con, m->middle->vec.iov_len,
Alex Elder859eb792012-02-14 14:05:33 -06001252 m->middle->vec.iov_base);
Sage Weil31b80062009-10-06 11:31:13 -07001253
1254 /* fill in crc (except data pages), footer */
Alex Eldera9a0c512012-02-15 07:43:54 -06001255 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1256 con->out_msg->hdr.crc = cpu_to_le32(crc);
Alex Elderfd154f32012-06-11 14:57:13 -05001257 con->out_msg->footer.flags = 0;
Alex Eldera9a0c512012-02-15 07:43:54 -06001258
1259 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1260 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1261 if (m->middle) {
1262 crc = crc32c(0, m->middle->vec.iov_base,
1263 m->middle->vec.iov_len);
1264 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1265 } else
Sage Weil31b80062009-10-06 11:31:13 -07001266 con->out_msg->footer.middle_crc = 0;
Alex Elder739c9052012-06-11 14:57:13 -05001267 dout("%s front_crc %u middle_crc %u\n", __func__,
Sage Weil31b80062009-10-06 11:31:13 -07001268 le32_to_cpu(con->out_msg->footer.front_crc),
1269 le32_to_cpu(con->out_msg->footer.middle_crc));
1270
1271 /* is there a data payload? */
Alex Elder739c9052012-06-11 14:57:13 -05001272 con->out_msg->footer.data_crc = 0;
Alex Elder98fa5dd2013-04-02 12:09:50 -05001273 if (m->data_length) {
1274 prepare_message_data(con->out_msg, m->data_length);
Alex Elder78625052013-03-06 23:39:39 -06001275 con->out_more = 1; /* data + footer will follow */
1276 } else {
Sage Weil31b80062009-10-06 11:31:13 -07001277 /* no, queue up footer too and be done */
Alex Elder859eb792012-02-14 14:05:33 -06001278 prepare_write_message_footer(con);
Alex Elder78625052013-03-06 23:39:39 -06001279 }
Sage Weil31b80062009-10-06 11:31:13 -07001280
Alex Elderc9ffc772013-02-20 10:25:12 -06001281 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001282}
1283
1284/*
1285 * Prepare an ack.
1286 */
1287static void prepare_write_ack(struct ceph_connection *con)
1288{
1289 dout("prepare_write_ack %p %llu -> %llu\n", con,
1290 con->in_seq_acked, con->in_seq);
1291 con->in_seq_acked = con->in_seq;
1292
Alex Eldere2200422012-05-23 14:35:23 -05001293 con_out_kvec_reset(con);
Alex Elder859eb792012-02-14 14:05:33 -06001294
Alex Eldere2200422012-05-23 14:35:23 -05001295 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
Alex Elder859eb792012-02-14 14:05:33 -06001296
Sage Weil31b80062009-10-06 11:31:13 -07001297 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
Alex Eldere2200422012-05-23 14:35:23 -05001298 con_out_kvec_add(con, sizeof (con->out_temp_ack),
Alex Elder859eb792012-02-14 14:05:33 -06001299 &con->out_temp_ack);
1300
Sage Weil31b80062009-10-06 11:31:13 -07001301 con->out_more = 1; /* more will follow.. eventually.. */
Alex Elderc9ffc772013-02-20 10:25:12 -06001302 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001303}
1304
1305/*
Sage Weil3a230832013-03-25 08:47:40 -07001306 * Prepare to share the seq during handshake
1307 */
1308static void prepare_write_seq(struct ceph_connection *con)
1309{
1310 dout("prepare_write_seq %p %llu -> %llu\n", con,
1311 con->in_seq_acked, con->in_seq);
1312 con->in_seq_acked = con->in_seq;
1313
1314 con_out_kvec_reset(con);
1315
1316 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1317 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1318 &con->out_temp_ack);
1319
1320 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1321}
1322
1323/*
Sage Weil31b80062009-10-06 11:31:13 -07001324 * Prepare to write keepalive byte.
1325 */
1326static void prepare_write_keepalive(struct ceph_connection *con)
1327{
1328 dout("prepare_write_keepalive %p\n", con);
Alex Eldere2200422012-05-23 14:35:23 -05001329 con_out_kvec_reset(con);
1330 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
Alex Elderc9ffc772013-02-20 10:25:12 -06001331 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001332}
1333
1334/*
1335 * Connection negotiation.
1336 */
1337
Alex Elderdac1e712012-05-16 15:16:39 -05001338static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1339 int *auth_proto)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001340{
Alex Eldera3530df2012-05-16 15:16:39 -05001341 struct ceph_auth_handshake *auth;
Alex Elderb1c6b982012-05-16 15:16:38 -05001342
1343 if (!con->ops->get_authorizer) {
1344 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1345 con->out_connect.authorizer_len = 0;
Alex Elder729796b2012-05-16 15:16:39 -05001346 return NULL;
Alex Elderb1c6b982012-05-16 15:16:38 -05001347 }
1348
1349 /* Can't hold the mutex while getting authorizer */
Sage Weilec302642009-12-22 10:43:42 -08001350 mutex_unlock(&con->mutex);
Alex Elderdac1e712012-05-16 15:16:39 -05001351 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
Sage Weilec302642009-12-22 10:43:42 -08001352 mutex_lock(&con->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001353
Alex Eldera3530df2012-05-16 15:16:39 -05001354 if (IS_ERR(auth))
Alex Elder729796b2012-05-16 15:16:39 -05001355 return auth;
Sage Weil8dacc7d2012-07-20 17:24:40 -07001356 if (con->state != CON_STATE_NEGOTIATING)
Alex Elder729796b2012-05-16 15:16:39 -05001357 return ERR_PTR(-EAGAIN);
Sage Weil0da5d702011-05-19 11:21:05 -07001358
Alex Elder8f43fb52012-05-16 15:16:39 -05001359 con->auth_reply_buf = auth->authorizer_reply_buf;
1360 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
Alex Elder729796b2012-05-16 15:16:39 -05001361 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001362}
1363
Sage Weil31b80062009-10-06 11:31:13 -07001364/*
1365 * We connected to a peer and are saying hello.
1366 */
Alex Eldere825a662012-05-16 15:16:38 -05001367static void prepare_write_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001368{
Alex Eldere2200422012-05-23 14:35:23 -05001369 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1370 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
Alex Eldere825a662012-05-16 15:16:38 -05001371 &con->msgr->my_enc_addr);
Sage Weileed0ef22009-11-10 14:34:36 -08001372
Sage Weileed0ef22009-11-10 14:34:36 -08001373 con->out_more = 0;
Alex Elderc9ffc772013-02-20 10:25:12 -06001374 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weileed0ef22009-11-10 14:34:36 -08001375}
1376
Alex Eldere825a662012-05-16 15:16:38 -05001377static int prepare_write_connect(struct ceph_connection *con)
Sage Weileed0ef22009-11-10 14:34:36 -08001378{
Eric Dumazet95c96172012-04-15 05:58:06 +00001379 unsigned int global_seq = get_global_seq(con->msgr, 0);
Sage Weil31b80062009-10-06 11:31:13 -07001380 int proto;
Alex Elderdac1e712012-05-16 15:16:39 -05001381 int auth_proto;
Alex Elder729796b2012-05-16 15:16:39 -05001382 struct ceph_auth_handshake *auth;
Sage Weil31b80062009-10-06 11:31:13 -07001383
1384 switch (con->peer_name.type) {
1385 case CEPH_ENTITY_TYPE_MON:
1386 proto = CEPH_MONC_PROTOCOL;
1387 break;
1388 case CEPH_ENTITY_TYPE_OSD:
1389 proto = CEPH_OSDC_PROTOCOL;
1390 break;
1391 case CEPH_ENTITY_TYPE_MDS:
1392 proto = CEPH_MDSC_PROTOCOL;
1393 break;
1394 default:
1395 BUG();
1396 }
1397
1398 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1399 con->connect_seq, global_seq, proto);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001400
Alex Eldere825a662012-05-16 15:16:38 -05001401 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
Sage Weil31b80062009-10-06 11:31:13 -07001402 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1403 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1404 con->out_connect.global_seq = cpu_to_le32(global_seq);
1405 con->out_connect.protocol_version = cpu_to_le32(proto);
1406 con->out_connect.flags = 0;
Sage Weil31b80062009-10-06 11:31:13 -07001407
Alex Elderdac1e712012-05-16 15:16:39 -05001408 auth_proto = CEPH_AUTH_UNKNOWN;
1409 auth = get_connect_authorizer(con, &auth_proto);
Alex Elder729796b2012-05-16 15:16:39 -05001410 if (IS_ERR(auth))
1411 return PTR_ERR(auth);
Alex Elder3da54772012-05-16 15:16:39 -05001412
Alex Elderdac1e712012-05-16 15:16:39 -05001413 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
Alex Elder3da54772012-05-16 15:16:39 -05001414 con->out_connect.authorizer_len = auth ?
1415 cpu_to_le32(auth->authorizer_buf_len) : 0;
1416
Alex Eldere2200422012-05-23 14:35:23 -05001417 con_out_kvec_add(con, sizeof (con->out_connect),
Alex Elder3da54772012-05-16 15:16:39 -05001418 &con->out_connect);
1419 if (auth && auth->authorizer_buf_len)
Alex Eldere2200422012-05-23 14:35:23 -05001420 con_out_kvec_add(con, auth->authorizer_buf_len,
Alex Elder3da54772012-05-16 15:16:39 -05001421 auth->authorizer_buf);
Alex Elder859eb792012-02-14 14:05:33 -06001422
Sage Weil31b80062009-10-06 11:31:13 -07001423 con->out_more = 0;
Alex Elderc9ffc772013-02-20 10:25:12 -06001424 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001425
Alex Eldere10c7582012-05-16 15:16:38 -05001426 return 0;
Sage Weil31b80062009-10-06 11:31:13 -07001427}
1428
Sage Weil31b80062009-10-06 11:31:13 -07001429/*
1430 * write as much of pending kvecs to the socket as we can.
1431 * 1 -> done
1432 * 0 -> socket full, but more to do
1433 * <0 -> error
1434 */
1435static int write_partial_kvec(struct ceph_connection *con)
1436{
1437 int ret;
1438
1439 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1440 while (con->out_kvec_bytes > 0) {
1441 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1442 con->out_kvec_left, con->out_kvec_bytes,
1443 con->out_more);
1444 if (ret <= 0)
1445 goto out;
1446 con->out_kvec_bytes -= ret;
1447 if (con->out_kvec_bytes == 0)
1448 break; /* done */
Alex Elderf42299e2012-02-15 07:43:54 -06001449
1450 /* account for full iov entries consumed */
1451 while (ret >= con->out_kvec_cur->iov_len) {
1452 BUG_ON(!con->out_kvec_left);
1453 ret -= con->out_kvec_cur->iov_len;
1454 con->out_kvec_cur++;
1455 con->out_kvec_left--;
1456 }
1457 /* and for a partially-consumed entry */
1458 if (ret) {
1459 con->out_kvec_cur->iov_len -= ret;
1460 con->out_kvec_cur->iov_base += ret;
Sage Weil31b80062009-10-06 11:31:13 -07001461 }
1462 }
1463 con->out_kvec_left = 0;
1464 con->out_kvec_is_msg = false;
1465 ret = 1;
1466out:
1467 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1468 con->out_kvec_bytes, con->out_kvec_left, ret);
1469 return ret; /* done! */
1470}
1471
Alex Elder35b62802013-03-08 20:59:00 -06001472static u32 ceph_crc32c_page(u32 crc, struct page *page,
1473 unsigned int page_offset,
1474 unsigned int length)
1475{
1476 char *kaddr;
1477
1478 kaddr = kmap(page);
1479 BUG_ON(kaddr == NULL);
1480 crc = crc32c(crc, kaddr + page_offset, length);
1481 kunmap(page);
1482
1483 return crc;
1484}
Sage Weil31b80062009-10-06 11:31:13 -07001485/*
1486 * Write as much message data payload as we can. If we finish, queue
1487 * up the footer.
1488 * 1 -> done, footer is now queued in out_kvec[].
1489 * 0 -> socket full, but more to do
1490 * <0 -> error
1491 */
Alex Elder34d2d202013-03-08 20:58:59 -06001492static int write_partial_message_data(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001493{
1494 struct ceph_msg *msg = con->out_msg;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001495 struct ceph_msg_data_cursor *cursor = &msg->cursor;
Alex Elder37675b02012-03-07 11:40:08 -06001496 bool do_datacrc = !con->msgr->nocrc;
Alex Elderf5db90b2013-03-11 23:34:23 -05001497 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07001498
Alex Elder859a35d2013-03-11 23:34:23 -05001499 dout("%s %p msg %p\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07001500
Alex Elder5240d9f2013-03-14 14:09:06 -05001501 if (list_empty(&msg->data))
Alex Elder4c59b4a2013-03-11 23:34:23 -05001502 return -EINVAL;
1503
Alex Elder5821bd82012-06-11 14:57:13 -05001504 /*
1505 * Iterate through each page that contains data to be
1506 * written, and send as much as possible for each.
1507 *
1508 * If we are calculating the data crc (the default), we will
1509 * need to map the page. If we have no pages, they have
1510 * been revoked, so use the zero page.
1511 */
Alex Elderf5db90b2013-03-11 23:34:23 -05001512 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
Alex Elder643c68a2013-03-11 23:34:23 -05001513 while (cursor->resid) {
Alex Elder8a166d02013-03-08 13:35:36 -06001514 struct page *page;
Alex Eldere387d522013-03-06 23:39:38 -06001515 size_t page_offset;
1516 size_t length;
Alex Elder8a166d02013-03-08 13:35:36 -06001517 bool last_piece;
Alex Elder8ea299b2013-03-11 23:34:23 -05001518 bool need_crc;
Alex Elderf5db90b2013-03-11 23:34:23 -05001519 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07001520
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001521 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
Alex Elder4c59b4a2013-03-11 23:34:23 -05001522 &last_piece);
Alex Eldere387d522013-03-06 23:39:38 -06001523 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
Alex Elderfe38a2b2013-03-06 23:39:39 -06001524 length, last_piece);
Alex Elderf5db90b2013-03-11 23:34:23 -05001525 if (ret <= 0) {
1526 if (do_datacrc)
1527 msg->footer.data_crc = cpu_to_le32(crc);
Sage Weil31b80062009-10-06 11:31:13 -07001528
Alex Elderf5db90b2013-03-11 23:34:23 -05001529 return ret;
1530 }
Alex Elder143334f2013-03-29 11:44:10 -05001531 if (do_datacrc && cursor->need_crc)
1532 crc = ceph_crc32c_page(crc, page, page_offset, length);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001533 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
Sage Weil31b80062009-10-06 11:31:13 -07001534 }
1535
Alex Elder34d2d202013-03-08 20:58:59 -06001536 dout("%s %p msg %p done\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07001537
1538 /* prepare and queue up footer, too */
Alex Elderf5db90b2013-03-11 23:34:23 -05001539 if (do_datacrc)
1540 msg->footer.data_crc = cpu_to_le32(crc);
1541 else
Alex Elder84ca8fc2012-06-11 14:57:13 -05001542 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
Alex Eldere2200422012-05-23 14:35:23 -05001543 con_out_kvec_reset(con);
Alex Elder859eb792012-02-14 14:05:33 -06001544 prepare_write_message_footer(con);
Alex Elderf5db90b2013-03-11 23:34:23 -05001545
1546 return 1; /* must return > 0 to indicate success */
Sage Weil31b80062009-10-06 11:31:13 -07001547}
1548
1549/*
1550 * write some zeros
1551 */
1552static int write_partial_skip(struct ceph_connection *con)
1553{
1554 int ret;
1555
1556 while (con->out_skip > 0) {
Alex Elder31739132012-03-07 11:40:08 -06001557 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
Sage Weil31b80062009-10-06 11:31:13 -07001558
Alex Eldere1dcb122013-03-06 23:39:38 -06001559 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
Sage Weil31b80062009-10-06 11:31:13 -07001560 if (ret <= 0)
1561 goto out;
1562 con->out_skip -= ret;
1563 }
1564 ret = 1;
1565out:
1566 return ret;
1567}
1568
1569/*
1570 * Prepare to read connection handshake, or an ack.
1571 */
Sage Weileed0ef22009-11-10 14:34:36 -08001572static void prepare_read_banner(struct ceph_connection *con)
1573{
1574 dout("prepare_read_banner %p\n", con);
1575 con->in_base_pos = 0;
1576}
1577
Sage Weil31b80062009-10-06 11:31:13 -07001578static void prepare_read_connect(struct ceph_connection *con)
1579{
1580 dout("prepare_read_connect %p\n", con);
1581 con->in_base_pos = 0;
1582}
1583
1584static void prepare_read_ack(struct ceph_connection *con)
1585{
1586 dout("prepare_read_ack %p\n", con);
1587 con->in_base_pos = 0;
1588}
1589
Sage Weil3a230832013-03-25 08:47:40 -07001590static void prepare_read_seq(struct ceph_connection *con)
1591{
1592 dout("prepare_read_seq %p\n", con);
1593 con->in_base_pos = 0;
1594 con->in_tag = CEPH_MSGR_TAG_SEQ;
1595}
1596
Sage Weil31b80062009-10-06 11:31:13 -07001597static void prepare_read_tag(struct ceph_connection *con)
1598{
1599 dout("prepare_read_tag %p\n", con);
1600 con->in_base_pos = 0;
1601 con->in_tag = CEPH_MSGR_TAG_READY;
1602}
1603
1604/*
1605 * Prepare to read a message.
1606 */
1607static int prepare_read_message(struct ceph_connection *con)
1608{
1609 dout("prepare_read_message %p\n", con);
1610 BUG_ON(con->in_msg != NULL);
1611 con->in_base_pos = 0;
1612 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1613 return 0;
1614}
1615
1616
1617static int read_partial(struct ceph_connection *con,
Alex Elderfd516532012-05-10 10:29:50 -05001618 int end, int size, void *object)
Sage Weil31b80062009-10-06 11:31:13 -07001619{
Alex Eldere6cee712012-05-10 10:29:50 -05001620 while (con->in_base_pos < end) {
1621 int left = end - con->in_base_pos;
Sage Weil31b80062009-10-06 11:31:13 -07001622 int have = size - left;
1623 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1624 if (ret <= 0)
1625 return ret;
1626 con->in_base_pos += ret;
1627 }
1628 return 1;
1629}
1630
1631
1632/*
1633 * Read all or part of the connect-side handshake on a new connection
1634 */
Sage Weileed0ef22009-11-10 14:34:36 -08001635static int read_partial_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001636{
Alex Elderfd516532012-05-10 10:29:50 -05001637 int size;
1638 int end;
1639 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07001640
Sage Weileed0ef22009-11-10 14:34:36 -08001641 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
Sage Weil31b80062009-10-06 11:31:13 -07001642
1643 /* peer's banner */
Alex Elderfd516532012-05-10 10:29:50 -05001644 size = strlen(CEPH_BANNER);
1645 end = size;
1646 ret = read_partial(con, end, size, con->in_banner);
Sage Weil31b80062009-10-06 11:31:13 -07001647 if (ret <= 0)
1648 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001649
1650 size = sizeof (con->actual_peer_addr);
1651 end += size;
1652 ret = read_partial(con, end, size, &con->actual_peer_addr);
Sage Weil31b80062009-10-06 11:31:13 -07001653 if (ret <= 0)
1654 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001655
1656 size = sizeof (con->peer_addr_for_me);
1657 end += size;
1658 ret = read_partial(con, end, size, &con->peer_addr_for_me);
Sage Weil31b80062009-10-06 11:31:13 -07001659 if (ret <= 0)
1660 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001661
Sage Weileed0ef22009-11-10 14:34:36 -08001662out:
1663 return ret;
1664}
1665
1666static int read_partial_connect(struct ceph_connection *con)
1667{
Alex Elderfd516532012-05-10 10:29:50 -05001668 int size;
1669 int end;
1670 int ret;
Sage Weileed0ef22009-11-10 14:34:36 -08001671
1672 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1673
Alex Elderfd516532012-05-10 10:29:50 -05001674 size = sizeof (con->in_reply);
1675 end = size;
1676 ret = read_partial(con, end, size, &con->in_reply);
Sage Weil31b80062009-10-06 11:31:13 -07001677 if (ret <= 0)
1678 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001679
1680 size = le32_to_cpu(con->in_reply.authorizer_len);
1681 end += size;
1682 ret = read_partial(con, end, size, con->auth_reply_buf);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001683 if (ret <= 0)
1684 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07001685
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001686 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1687 con, (int)con->in_reply.tag,
1688 le32_to_cpu(con->in_reply.connect_seq),
Sage Weil31b80062009-10-06 11:31:13 -07001689 le32_to_cpu(con->in_reply.global_seq));
1690out:
1691 return ret;
Sage Weileed0ef22009-11-10 14:34:36 -08001692
Sage Weil31b80062009-10-06 11:31:13 -07001693}
1694
1695/*
1696 * Verify the hello banner looks okay.
1697 */
1698static int verify_hello(struct ceph_connection *con)
1699{
1700 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
Sage Weil13e38c82009-10-09 16:36:34 -07001701 pr_err("connect to %s got bad banner\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001702 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07001703 con->error_msg = "protocol error, bad banner";
1704 return -1;
1705 }
1706 return 0;
1707}
1708
1709static bool addr_is_blank(struct sockaddr_storage *ss)
1710{
1711 switch (ss->ss_family) {
1712 case AF_INET:
1713 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1714 case AF_INET6:
1715 return
1716 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1717 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1718 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1719 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1720 }
1721 return false;
1722}
1723
1724static int addr_port(struct sockaddr_storage *ss)
1725{
1726 switch (ss->ss_family) {
1727 case AF_INET:
Sage Weilf28bcfb2009-11-04 11:46:35 -08001728 return ntohs(((struct sockaddr_in *)ss)->sin_port);
Sage Weil31b80062009-10-06 11:31:13 -07001729 case AF_INET6:
Sage Weilf28bcfb2009-11-04 11:46:35 -08001730 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
Sage Weil31b80062009-10-06 11:31:13 -07001731 }
1732 return 0;
1733}
1734
1735static void addr_set_port(struct sockaddr_storage *ss, int p)
1736{
1737 switch (ss->ss_family) {
1738 case AF_INET:
1739 ((struct sockaddr_in *)ss)->sin_port = htons(p);
Sage Weila2a79602011-05-12 15:34:24 -07001740 break;
Sage Weil31b80062009-10-06 11:31:13 -07001741 case AF_INET6:
1742 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
Sage Weila2a79602011-05-12 15:34:24 -07001743 break;
Sage Weil31b80062009-10-06 11:31:13 -07001744 }
1745}
1746
1747/*
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001748 * Unlike other *_pton function semantics, zero indicates success.
1749 */
1750static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1751 char delim, const char **ipend)
1752{
Alex Elder99f0f3b2012-01-23 15:49:27 -06001753 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1754 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001755
1756 memset(ss, 0, sizeof(*ss));
1757
1758 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1759 ss->ss_family = AF_INET;
1760 return 0;
1761 }
1762
1763 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1764 ss->ss_family = AF_INET6;
1765 return 0;
1766 }
1767
1768 return -EINVAL;
1769}
1770
1771/*
1772 * Extract hostname string and resolve using kernel DNS facility.
1773 */
1774#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1775static int ceph_dns_resolve_name(const char *name, size_t namelen,
1776 struct sockaddr_storage *ss, char delim, const char **ipend)
1777{
1778 const char *end, *delim_p;
1779 char *colon_p, *ip_addr = NULL;
1780 int ip_len, ret;
1781
1782 /*
1783 * The end of the hostname occurs immediately preceding the delimiter or
1784 * the port marker (':') where the delimiter takes precedence.
1785 */
1786 delim_p = memchr(name, delim, namelen);
1787 colon_p = memchr(name, ':', namelen);
1788
1789 if (delim_p && colon_p)
1790 end = delim_p < colon_p ? delim_p : colon_p;
1791 else if (!delim_p && colon_p)
1792 end = colon_p;
1793 else {
1794 end = delim_p;
1795 if (!end) /* case: hostname:/ */
1796 end = name + namelen;
1797 }
1798
1799 if (end <= name)
1800 return -EINVAL;
1801
1802 /* do dns_resolve upcall */
1803 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1804 if (ip_len > 0)
1805 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1806 else
1807 ret = -ESRCH;
1808
1809 kfree(ip_addr);
1810
1811 *ipend = end;
1812
1813 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1814 ret, ret ? "failed" : ceph_pr_addr(ss));
1815
1816 return ret;
1817}
1818#else
1819static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1820 struct sockaddr_storage *ss, char delim, const char **ipend)
1821{
1822 return -EINVAL;
1823}
1824#endif
1825
1826/*
1827 * Parse a server name (IP or hostname). If a valid IP address is not found
1828 * then try to extract a hostname to resolve using userspace DNS upcall.
1829 */
1830static int ceph_parse_server_name(const char *name, size_t namelen,
1831 struct sockaddr_storage *ss, char delim, const char **ipend)
1832{
1833 int ret;
1834
1835 ret = ceph_pton(name, namelen, ss, delim, ipend);
1836 if (ret)
1837 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1838
1839 return ret;
1840}
1841
1842/*
Sage Weil31b80062009-10-06 11:31:13 -07001843 * Parse an ip[:port] list into an addr array. Use the default
1844 * monitor port if a port isn't specified.
1845 */
1846int ceph_parse_ips(const char *c, const char *end,
1847 struct ceph_entity_addr *addr,
1848 int max_count, int *count)
1849{
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001850 int i, ret = -EINVAL;
Sage Weil31b80062009-10-06 11:31:13 -07001851 const char *p = c;
1852
1853 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1854 for (i = 0; i < max_count; i++) {
1855 const char *ipend;
1856 struct sockaddr_storage *ss = &addr[i].in_addr;
Sage Weil31b80062009-10-06 11:31:13 -07001857 int port;
Sage Weil39139f62010-07-08 09:54:52 -07001858 char delim = ',';
1859
1860 if (*p == '[') {
1861 delim = ']';
1862 p++;
1863 }
Sage Weil31b80062009-10-06 11:31:13 -07001864
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001865 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1866 if (ret)
Sage Weil31b80062009-10-06 11:31:13 -07001867 goto bad;
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001868 ret = -EINVAL;
1869
Sage Weil31b80062009-10-06 11:31:13 -07001870 p = ipend;
1871
Sage Weil39139f62010-07-08 09:54:52 -07001872 if (delim == ']') {
1873 if (*p != ']') {
1874 dout("missing matching ']'\n");
1875 goto bad;
1876 }
1877 p++;
1878 }
1879
Sage Weil31b80062009-10-06 11:31:13 -07001880 /* port? */
1881 if (p < end && *p == ':') {
1882 port = 0;
1883 p++;
1884 while (p < end && *p >= '0' && *p <= '9') {
1885 port = (port * 10) + (*p - '0');
1886 p++;
1887 }
Ilya Dryomovf48db1e2013-12-30 19:21:29 +02001888 if (port == 0)
1889 port = CEPH_MON_PORT;
1890 else if (port > 65535)
Sage Weil31b80062009-10-06 11:31:13 -07001891 goto bad;
1892 } else {
1893 port = CEPH_MON_PORT;
1894 }
1895
1896 addr_set_port(ss, port);
1897
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001898 dout("parse_ips got %s\n", ceph_pr_addr(ss));
Sage Weil31b80062009-10-06 11:31:13 -07001899
1900 if (p == end)
1901 break;
1902 if (*p != ',')
1903 goto bad;
1904 p++;
1905 }
1906
1907 if (p != end)
1908 goto bad;
1909
1910 if (count)
1911 *count = i + 1;
1912 return 0;
1913
1914bad:
Sage Weil39139f62010-07-08 09:54:52 -07001915 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001916 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07001917}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001918EXPORT_SYMBOL(ceph_parse_ips);
Sage Weil31b80062009-10-06 11:31:13 -07001919
Sage Weileed0ef22009-11-10 14:34:36 -08001920static int process_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001921{
Sage Weileed0ef22009-11-10 14:34:36 -08001922 dout("process_banner on %p\n", con);
Sage Weil31b80062009-10-06 11:31:13 -07001923
1924 if (verify_hello(con) < 0)
1925 return -1;
1926
Sage Weil63f2d212009-11-03 15:17:56 -08001927 ceph_decode_addr(&con->actual_peer_addr);
1928 ceph_decode_addr(&con->peer_addr_for_me);
1929
Sage Weil31b80062009-10-06 11:31:13 -07001930 /*
1931 * Make sure the other end is who we wanted. note that the other
1932 * end may not yet know their ip address, so if it's 0.0.0.0, give
1933 * them the benefit of the doubt.
1934 */
Sage Weil103e2d32010-01-07 16:12:36 -08001935 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1936 sizeof(con->peer_addr)) != 0 &&
Sage Weil31b80062009-10-06 11:31:13 -07001937 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1938 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001939 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001940 ceph_pr_addr(&con->peer_addr.in_addr),
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001941 (int)le32_to_cpu(con->peer_addr.nonce),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001942 ceph_pr_addr(&con->actual_peer_addr.in_addr),
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001943 (int)le32_to_cpu(con->actual_peer_addr.nonce));
Sage Weil58bb3b32009-12-23 12:12:31 -08001944 con->error_msg = "wrong peer at address";
Sage Weil31b80062009-10-06 11:31:13 -07001945 return -1;
1946 }
1947
1948 /*
1949 * did we learn our address?
1950 */
1951 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1952 int port = addr_port(&con->msgr->inst.addr.in_addr);
1953
1954 memcpy(&con->msgr->inst.addr.in_addr,
1955 &con->peer_addr_for_me.in_addr,
1956 sizeof(con->peer_addr_for_me.in_addr));
1957 addr_set_port(&con->msgr->inst.addr.in_addr, port);
Sage Weil63f2d212009-11-03 15:17:56 -08001958 encode_my_addr(con->msgr);
Sage Weileed0ef22009-11-10 14:34:36 -08001959 dout("process_banner learned my addr is %s\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001960 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07001961 }
1962
Sage Weileed0ef22009-11-10 14:34:36 -08001963 return 0;
1964}
1965
1966static int process_connect(struct ceph_connection *con)
1967{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001968 u64 sup_feat = con->msgr->supported_features;
1969 u64 req_feat = con->msgr->required_features;
Ilya Dryomov2b3e0c92013-12-24 21:19:24 +02001970 u64 server_feat = ceph_sanitize_features(
1971 le64_to_cpu(con->in_reply.features));
Sage Weil0da5d702011-05-19 11:21:05 -07001972 int ret;
Sage Weil04a419f2009-12-23 09:30:21 -08001973
Sage Weileed0ef22009-11-10 14:34:36 -08001974 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1975
Sage Weil31b80062009-10-06 11:31:13 -07001976 switch (con->in_reply.tag) {
Sage Weil04a419f2009-12-23 09:30:21 -08001977 case CEPH_MSGR_TAG_FEATURES:
1978 pr_err("%s%lld %s feature set mismatch,"
1979 " my %llx < server's %llx, missing %llx\n",
1980 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001981 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil04a419f2009-12-23 09:30:21 -08001982 sup_feat, server_feat, server_feat & ~sup_feat);
1983 con->error_msg = "missing required protocol features";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06001984 reset_connection(con);
Sage Weil04a419f2009-12-23 09:30:21 -08001985 return -1;
1986
Sage Weil31b80062009-10-06 11:31:13 -07001987 case CEPH_MSGR_TAG_BADPROTOVER:
Sage Weil31b80062009-10-06 11:31:13 -07001988 pr_err("%s%lld %s protocol version mismatch,"
1989 " my %d != server's %d\n",
1990 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001991 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil31b80062009-10-06 11:31:13 -07001992 le32_to_cpu(con->out_connect.protocol_version),
1993 le32_to_cpu(con->in_reply.protocol_version));
1994 con->error_msg = "protocol version mismatch";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06001995 reset_connection(con);
Sage Weil31b80062009-10-06 11:31:13 -07001996 return -1;
1997
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001998 case CEPH_MSGR_TAG_BADAUTHORIZER:
1999 con->auth_retry++;
2000 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2001 con->auth_retry);
2002 if (con->auth_retry == 2) {
2003 con->error_msg = "connect authorization failure";
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002004 return -1;
2005 }
Jim Schutt6d4221b2012-08-10 10:37:38 -07002006 con_out_kvec_reset(con);
Alex Eldere825a662012-05-16 15:16:38 -05002007 ret = prepare_write_connect(con);
Sage Weil0da5d702011-05-19 11:21:05 -07002008 if (ret < 0)
2009 return ret;
Sage Weil63733a02010-03-15 15:47:22 -07002010 prepare_read_connect(con);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002011 break;
Sage Weil31b80062009-10-06 11:31:13 -07002012
2013 case CEPH_MSGR_TAG_RESETSESSION:
2014 /*
2015 * If we connected with a large connect_seq but the peer
2016 * has no record of a session with us (no connection, or
2017 * connect_seq == 0), they will send RESETSESION to indicate
2018 * that they must have reset their session, and may have
2019 * dropped messages.
2020 */
2021 dout("process_connect got RESET peer seq %u\n",
Sage Weil5bdca4e2012-07-10 11:53:34 -07002022 le32_to_cpu(con->in_reply.connect_seq));
Sage Weil31b80062009-10-06 11:31:13 -07002023 pr_err("%s%lld %s connection reset\n",
2024 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002025 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07002026 reset_connection(con);
Jim Schutt6d4221b2012-08-10 10:37:38 -07002027 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002028 ret = prepare_write_connect(con);
2029 if (ret < 0)
2030 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002031 prepare_read_connect(con);
2032
2033 /* Tell ceph about it. */
Sage Weilec302642009-12-22 10:43:42 -08002034 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002035 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2036 if (con->ops->peer_reset)
2037 con->ops->peer_reset(con);
Sage Weilec302642009-12-22 10:43:42 -08002038 mutex_lock(&con->mutex);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002039 if (con->state != CON_STATE_NEGOTIATING)
Sage Weil0da5d702011-05-19 11:21:05 -07002040 return -EAGAIN;
Sage Weil31b80062009-10-06 11:31:13 -07002041 break;
2042
2043 case CEPH_MSGR_TAG_RETRY_SESSION:
2044 /*
2045 * If we sent a smaller connect_seq than the peer has, try
2046 * again with a larger value.
2047 */
Sage Weil5bdca4e2012-07-10 11:53:34 -07002048 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
Sage Weil31b80062009-10-06 11:31:13 -07002049 le32_to_cpu(con->out_connect.connect_seq),
Sage Weil5bdca4e2012-07-10 11:53:34 -07002050 le32_to_cpu(con->in_reply.connect_seq));
2051 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
Jim Schutt6d4221b2012-08-10 10:37:38 -07002052 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002053 ret = prepare_write_connect(con);
2054 if (ret < 0)
2055 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002056 prepare_read_connect(con);
2057 break;
2058
2059 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2060 /*
2061 * If we sent a smaller global_seq than the peer has, try
2062 * again with a larger value.
2063 */
Sage Weileed0ef22009-11-10 14:34:36 -08002064 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
Sage Weil31b80062009-10-06 11:31:13 -07002065 con->peer_global_seq,
Sage Weil5bdca4e2012-07-10 11:53:34 -07002066 le32_to_cpu(con->in_reply.global_seq));
Sage Weil31b80062009-10-06 11:31:13 -07002067 get_global_seq(con->msgr,
Sage Weil5bdca4e2012-07-10 11:53:34 -07002068 le32_to_cpu(con->in_reply.global_seq));
Jim Schutt6d4221b2012-08-10 10:37:38 -07002069 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002070 ret = prepare_write_connect(con);
2071 if (ret < 0)
2072 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002073 prepare_read_connect(con);
2074 break;
2075
Sage Weil3a230832013-03-25 08:47:40 -07002076 case CEPH_MSGR_TAG_SEQ:
Sage Weil31b80062009-10-06 11:31:13 -07002077 case CEPH_MSGR_TAG_READY:
Sage Weil04a419f2009-12-23 09:30:21 -08002078 if (req_feat & ~server_feat) {
2079 pr_err("%s%lld %s protocol feature mismatch,"
2080 " my required %llx > server's %llx, need %llx\n",
2081 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002082 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil04a419f2009-12-23 09:30:21 -08002083 req_feat, server_feat, req_feat & ~server_feat);
2084 con->error_msg = "missing required protocol features";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06002085 reset_connection(con);
Sage Weil04a419f2009-12-23 09:30:21 -08002086 return -1;
2087 }
Sage Weil8dacc7d2012-07-20 17:24:40 -07002088
Alex Elder122070a2012-12-26 10:43:57 -06002089 WARN_ON(con->state != CON_STATE_NEGOTIATING);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002090 con->state = CON_STATE_OPEN;
Sage Weil20e55c42013-03-25 09:30:13 -07002091 con->auth_retry = 0; /* we authenticated; clear flag */
Sage Weil31b80062009-10-06 11:31:13 -07002092 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2093 con->connect_seq++;
Sage Weilaba558e2010-05-12 15:23:30 -07002094 con->peer_features = server_feat;
Sage Weil31b80062009-10-06 11:31:13 -07002095 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2096 con->peer_global_seq,
2097 le32_to_cpu(con->in_reply.connect_seq),
2098 con->connect_seq);
2099 WARN_ON(con->connect_seq !=
2100 le32_to_cpu(con->in_reply.connect_seq));
Sage Weil92ac41d2009-12-14 14:56:56 -08002101
2102 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
Alex Elderc9ffc772013-02-20 10:25:12 -06002103 con_flag_set(con, CON_FLAG_LOSSYTX);
Sage Weil92ac41d2009-12-14 14:56:56 -08002104
Sage Weil85effe12012-07-30 16:22:05 -07002105 con->delay = 0; /* reset backoff memory */
Sage Weil31b80062009-10-06 11:31:13 -07002106
Sage Weil3a230832013-03-25 08:47:40 -07002107 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2108 prepare_write_seq(con);
2109 prepare_read_seq(con);
2110 } else {
2111 prepare_read_tag(con);
2112 }
Sage Weil31b80062009-10-06 11:31:13 -07002113 break;
2114
2115 case CEPH_MSGR_TAG_WAIT:
2116 /*
2117 * If there is a connection race (we are opening
2118 * connections to each other), one of us may just have
2119 * to WAIT. This shouldn't happen if we are the
2120 * client.
2121 */
Sage Weil04177882011-05-12 15:33:17 -07002122 pr_err("process_connect got WAIT as client\n");
2123 con->error_msg = "protocol error, got WAIT as client";
2124 return -1;
Sage Weil31b80062009-10-06 11:31:13 -07002125
2126 default:
2127 pr_err("connect protocol error, will retry\n");
2128 con->error_msg = "protocol error, garbage tag during connect";
2129 return -1;
2130 }
2131 return 0;
2132}
2133
2134
2135/*
2136 * read (part of) an ack
2137 */
2138static int read_partial_ack(struct ceph_connection *con)
2139{
Alex Elderfd516532012-05-10 10:29:50 -05002140 int size = sizeof (con->in_temp_ack);
2141 int end = size;
Sage Weil31b80062009-10-06 11:31:13 -07002142
Alex Elderfd516532012-05-10 10:29:50 -05002143 return read_partial(con, end, size, &con->in_temp_ack);
Sage Weil31b80062009-10-06 11:31:13 -07002144}
2145
Sage Weil31b80062009-10-06 11:31:13 -07002146/*
2147 * We can finally discard anything that's been acked.
2148 */
2149static void process_ack(struct ceph_connection *con)
2150{
2151 struct ceph_msg *m;
2152 u64 ack = le64_to_cpu(con->in_temp_ack);
2153 u64 seq;
2154
Sage Weil31b80062009-10-06 11:31:13 -07002155 while (!list_empty(&con->out_sent)) {
2156 m = list_first_entry(&con->out_sent, struct ceph_msg,
2157 list_head);
2158 seq = le64_to_cpu(m->hdr.seq);
2159 if (seq > ack)
2160 break;
2161 dout("got ack for seq %llu type %d at %p\n", seq,
2162 le16_to_cpu(m->hdr.type), m);
Sage Weil4cf9d542011-07-26 11:27:24 -07002163 m->ack_stamp = jiffies;
Sage Weil31b80062009-10-06 11:31:13 -07002164 ceph_msg_remove(m);
2165 }
Sage Weil31b80062009-10-06 11:31:13 -07002166 prepare_read_tag(con);
2167}
2168
2169
Yehuda Sadeh24504182010-01-08 13:58:34 -08002170static int read_partial_message_section(struct ceph_connection *con,
Sage Weil213c99e2010-08-03 10:25:11 -07002171 struct kvec *section,
2172 unsigned int sec_len, u32 *crc)
Yehuda Sadeh24504182010-01-08 13:58:34 -08002173{
Yehuda Sadeh68b44762010-04-06 15:01:27 -07002174 int ret, left;
Sage Weil31b80062009-10-06 11:31:13 -07002175
Yehuda Sadeh24504182010-01-08 13:58:34 -08002176 BUG_ON(!section);
Sage Weil31b80062009-10-06 11:31:13 -07002177
Yehuda Sadeh24504182010-01-08 13:58:34 -08002178 while (section->iov_len < sec_len) {
2179 BUG_ON(section->iov_base == NULL);
2180 left = sec_len - section->iov_len;
2181 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2182 section->iov_len, left);
2183 if (ret <= 0)
2184 return ret;
2185 section->iov_len += ret;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002186 }
Alex Elderfe3ad592012-02-15 07:43:54 -06002187 if (section->iov_len == sec_len)
2188 *crc = crc32c(0, section->iov_base, section->iov_len);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002189
2190 return 1;
2191}
2192
Alex Elder34d2d202013-03-08 20:58:59 -06002193static int read_partial_msg_data(struct ceph_connection *con)
2194{
2195 struct ceph_msg *msg = con->in_msg;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002196 struct ceph_msg_data_cursor *cursor = &msg->cursor;
Alex Elder34d2d202013-03-08 20:58:59 -06002197 const bool do_datacrc = !con->msgr->nocrc;
Alex Elder686be202013-03-11 23:34:23 -05002198 struct page *page;
2199 size_t page_offset;
2200 size_t length;
Alex Elderf5db90b2013-03-11 23:34:23 -05002201 u32 crc = 0;
Alex Elder34d2d202013-03-08 20:58:59 -06002202 int ret;
2203
2204 BUG_ON(!msg);
Alex Elder5240d9f2013-03-14 14:09:06 -05002205 if (list_empty(&msg->data))
Alex Elder4c59b4a2013-03-11 23:34:23 -05002206 return -EIO;
Alex Elder34d2d202013-03-08 20:58:59 -06002207
Alex Elderf5db90b2013-03-11 23:34:23 -05002208 if (do_datacrc)
2209 crc = con->in_data_crc;
Alex Elder643c68a2013-03-11 23:34:23 -05002210 while (cursor->resid) {
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002211 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
Alex Elder4c59b4a2013-03-11 23:34:23 -05002212 NULL);
Alex Elder686be202013-03-11 23:34:23 -05002213 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
Alex Elderf5db90b2013-03-11 23:34:23 -05002214 if (ret <= 0) {
2215 if (do_datacrc)
2216 con->in_data_crc = crc;
2217
Alex Elder686be202013-03-11 23:34:23 -05002218 return ret;
Alex Elderf5db90b2013-03-11 23:34:23 -05002219 }
Alex Elder686be202013-03-11 23:34:23 -05002220
2221 if (do_datacrc)
Alex Elderf5db90b2013-03-11 23:34:23 -05002222 crc = ceph_crc32c_page(crc, page, page_offset, ret);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002223 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
Alex Elder34d2d202013-03-08 20:58:59 -06002224 }
Alex Elderf5db90b2013-03-11 23:34:23 -05002225 if (do_datacrc)
2226 con->in_data_crc = crc;
Alex Elder34d2d202013-03-08 20:58:59 -06002227
2228 return 1; /* must return > 0 to indicate success */
2229}
2230
Sage Weil31b80062009-10-06 11:31:13 -07002231/*
2232 * read (part of) a message.
2233 */
Alex Elder686be202013-03-11 23:34:23 -05002234static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2235
Sage Weil31b80062009-10-06 11:31:13 -07002236static int read_partial_message(struct ceph_connection *con)
2237{
2238 struct ceph_msg *m = con->in_msg;
Alex Elderfd516532012-05-10 10:29:50 -05002239 int size;
2240 int end;
Sage Weil31b80062009-10-06 11:31:13 -07002241 int ret;
Eric Dumazet95c96172012-04-15 05:58:06 +00002242 unsigned int front_len, middle_len, data_len;
Alex Elder37675b02012-03-07 11:40:08 -06002243 bool do_datacrc = !con->msgr->nocrc;
Sage Weilae187562010-04-22 07:47:01 -07002244 u64 seq;
Alex Elderfe3ad592012-02-15 07:43:54 -06002245 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07002246
2247 dout("read_partial_message con %p msg %p\n", con, m);
2248
2249 /* header */
Alex Elderfd516532012-05-10 10:29:50 -05002250 size = sizeof (con->in_hdr);
2251 end = size;
2252 ret = read_partial(con, end, size, &con->in_hdr);
Alex Elder57dac9d2012-05-10 10:29:50 -05002253 if (ret <= 0)
2254 return ret;
Alex Elderfe3ad592012-02-15 07:43:54 -06002255
2256 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2257 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2258 pr_err("read_partial_message bad hdr "
2259 " crc %u != expected %u\n",
2260 crc, con->in_hdr.crc);
2261 return -EBADMSG;
2262 }
2263
Sage Weil31b80062009-10-06 11:31:13 -07002264 front_len = le32_to_cpu(con->in_hdr.front_len);
2265 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2266 return -EIO;
2267 middle_len = le32_to_cpu(con->in_hdr.middle_len);
Alex Elder7b11ba32013-03-08 18:51:03 -06002268 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
Sage Weil31b80062009-10-06 11:31:13 -07002269 return -EIO;
2270 data_len = le32_to_cpu(con->in_hdr.data_len);
2271 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2272 return -EIO;
2273
Sage Weilae187562010-04-22 07:47:01 -07002274 /* verify seq# */
2275 seq = le64_to_cpu(con->in_hdr.seq);
2276 if ((s64)seq - (s64)con->in_seq < 1) {
Sage Weildf9f86f2010-11-01 15:49:23 -07002277 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
Sage Weilae187562010-04-22 07:47:01 -07002278 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002279 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weilae187562010-04-22 07:47:01 -07002280 seq, con->in_seq + 1);
2281 con->in_base_pos = -front_len - middle_len - data_len -
2282 sizeof(m->footer);
2283 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weilae187562010-04-22 07:47:01 -07002284 return 0;
2285 } else if ((s64)seq - (s64)con->in_seq > 1) {
2286 pr_err("read_partial_message bad seq %lld expected %lld\n",
2287 seq, con->in_seq + 1);
2288 con->error_msg = "bad message sequence # for incoming message";
2289 return -EBADMSG;
2290 }
2291
Sage Weil31b80062009-10-06 11:31:13 -07002292 /* allocate message? */
2293 if (!con->in_msg) {
Sage Weil4740a622012-07-30 18:19:30 -07002294 int skip = 0;
2295
Sage Weil31b80062009-10-06 11:31:13 -07002296 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
Alex Elder6ebc8b32013-03-08 18:51:03 -06002297 front_len, data_len);
Sage Weil4740a622012-07-30 18:19:30 -07002298 ret = ceph_con_in_msg_alloc(con, &skip);
2299 if (ret < 0)
2300 return ret;
Alex Elderf759ebb2013-04-05 14:46:01 -05002301
2302 BUG_ON(!con->in_msg ^ skip);
2303 if (con->in_msg && data_len > con->in_msg->data_length) {
2304 pr_warning("%s skipping long message (%u > %zd)\n",
2305 __func__, data_len, con->in_msg->data_length);
2306 ceph_msg_put(con->in_msg);
2307 con->in_msg = NULL;
2308 skip = 1;
2309 }
Yehuda Sadeh24504182010-01-08 13:58:34 -08002310 if (skip) {
Sage Weil31b80062009-10-06 11:31:13 -07002311 /* skip this message */
Sage Weila79832f2010-04-01 16:06:19 -07002312 dout("alloc_msg said skip message\n");
Sage Weil31b80062009-10-06 11:31:13 -07002313 con->in_base_pos = -front_len - middle_len - data_len -
2314 sizeof(m->footer);
2315 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weil684be252010-04-21 20:45:59 -07002316 con->in_seq++;
Sage Weil31b80062009-10-06 11:31:13 -07002317 return 0;
2318 }
Alex Elder38941f82012-06-01 14:56:43 -05002319
Sage Weil4740a622012-07-30 18:19:30 -07002320 BUG_ON(!con->in_msg);
Alex Elder38941f82012-06-01 14:56:43 -05002321 BUG_ON(con->in_msg->con != con);
Sage Weil31b80062009-10-06 11:31:13 -07002322 m = con->in_msg;
2323 m->front.iov_len = 0; /* haven't read it yet */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002324 if (m->middle)
2325 m->middle->vec.iov_len = 0;
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08002326
Alex Elder78625052013-03-06 23:39:39 -06002327 /* prepare for data payload, if any */
Sage Weila4107022012-07-30 16:20:25 -07002328
Alex Elder78625052013-03-06 23:39:39 -06002329 if (data_len)
Alex Elder98fa5dd2013-04-02 12:09:50 -05002330 prepare_message_data(con->in_msg, data_len);
Sage Weil31b80062009-10-06 11:31:13 -07002331 }
2332
2333 /* front */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002334 ret = read_partial_message_section(con, &m->front, front_len,
2335 &con->in_front_crc);
2336 if (ret <= 0)
2337 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002338
2339 /* middle */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002340 if (m->middle) {
Sage Weil213c99e2010-08-03 10:25:11 -07002341 ret = read_partial_message_section(con, &m->middle->vec,
2342 middle_len,
Yehuda Sadeh24504182010-01-08 13:58:34 -08002343 &con->in_middle_crc);
Sage Weil31b80062009-10-06 11:31:13 -07002344 if (ret <= 0)
2345 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002346 }
2347
2348 /* (page) data */
Alex Elder34d2d202013-03-08 20:58:59 -06002349 if (data_len) {
2350 ret = read_partial_msg_data(con);
2351 if (ret <= 0)
2352 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002353 }
2354
Sage Weil31b80062009-10-06 11:31:13 -07002355 /* footer */
Alex Elderfd516532012-05-10 10:29:50 -05002356 size = sizeof (m->footer);
2357 end += size;
2358 ret = read_partial(con, end, size, &m->footer);
Alex Elder57dac9d2012-05-10 10:29:50 -05002359 if (ret <= 0)
2360 return ret;
2361
Sage Weil31b80062009-10-06 11:31:13 -07002362 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2363 m, front_len, m->footer.front_crc, middle_len,
2364 m->footer.middle_crc, data_len, m->footer.data_crc);
2365
2366 /* crc ok? */
2367 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2368 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2369 m, con->in_front_crc, m->footer.front_crc);
2370 return -EBADMSG;
2371 }
2372 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2373 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2374 m, con->in_middle_crc, m->footer.middle_crc);
2375 return -EBADMSG;
2376 }
Alex Elderbca064d2012-02-15 07:43:54 -06002377 if (do_datacrc &&
Sage Weil31b80062009-10-06 11:31:13 -07002378 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2379 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2380 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2381 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2382 return -EBADMSG;
2383 }
2384
2385 return 1; /* done! */
2386}
2387
2388/*
2389 * Process message. This happens in the worker thread. The callback should
2390 * be careful not to do anything that waits on other incoming messages or it
2391 * may deadlock.
2392 */
2393static void process_message(struct ceph_connection *con)
2394{
Sage Weil5e095e82009-12-14 14:30:34 -08002395 struct ceph_msg *msg;
Sage Weil31b80062009-10-06 11:31:13 -07002396
Alex Elder38941f82012-06-01 14:56:43 -05002397 BUG_ON(con->in_msg->con != con);
2398 con->in_msg->con = NULL;
Sage Weil5e095e82009-12-14 14:30:34 -08002399 msg = con->in_msg;
Sage Weil31b80062009-10-06 11:31:13 -07002400 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -07002401 con->ops->put(con);
Sage Weil31b80062009-10-06 11:31:13 -07002402
2403 /* if first message, set peer_name */
2404 if (con->peer_name.type == 0)
Sage Weildbad1852010-03-25 15:45:38 -07002405 con->peer_name = msg->hdr.src;
Sage Weil31b80062009-10-06 11:31:13 -07002406
Sage Weil31b80062009-10-06 11:31:13 -07002407 con->in_seq++;
Sage Weilec302642009-12-22 10:43:42 -08002408 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002409
2410 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2411 msg, le64_to_cpu(msg->hdr.seq),
Sage Weildbad1852010-03-25 15:45:38 -07002412 ENTITY_NAME(msg->hdr.src),
Sage Weil31b80062009-10-06 11:31:13 -07002413 le16_to_cpu(msg->hdr.type),
2414 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2415 le32_to_cpu(msg->hdr.front_len),
2416 le32_to_cpu(msg->hdr.data_len),
2417 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2418 con->ops->dispatch(con, msg);
Sage Weilec302642009-12-22 10:43:42 -08002419
2420 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002421}
2422
2423
2424/*
2425 * Write something to the socket. Called in a worker thread when the
2426 * socket appears to be writeable and we have something ready to send.
2427 */
2428static int try_write(struct ceph_connection *con)
2429{
Sage Weil31b80062009-10-06 11:31:13 -07002430 int ret = 1;
2431
Sage Weild59315c2012-06-21 12:49:23 -07002432 dout("try_write start %p state %lu\n", con, con->state);
Sage Weil31b80062009-10-06 11:31:13 -07002433
Sage Weil31b80062009-10-06 11:31:13 -07002434more:
2435 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2436
2437 /* open the socket first? */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002438 if (con->state == CON_STATE_PREOPEN) {
2439 BUG_ON(con->sock);
2440 con->state = CON_STATE_CONNECTING;
Alex Eldera5988c42012-05-29 11:04:58 -05002441
Alex Eldere2200422012-05-23 14:35:23 -05002442 con_out_kvec_reset(con);
Alex Eldere825a662012-05-16 15:16:38 -05002443 prepare_write_banner(con);
Sage Weileed0ef22009-11-10 14:34:36 -08002444 prepare_read_banner(con);
Sage Weil31b80062009-10-06 11:31:13 -07002445
Sage Weilcf3e5c42009-12-11 09:48:05 -08002446 BUG_ON(con->in_msg);
Sage Weil31b80062009-10-06 11:31:13 -07002447 con->in_tag = CEPH_MSGR_TAG_READY;
2448 dout("try_write initiating connect on %p new state %lu\n",
2449 con, con->state);
Alex Elder41617d02012-02-14 14:05:33 -06002450 ret = ceph_tcp_connect(con);
2451 if (ret < 0) {
Sage Weil31b80062009-10-06 11:31:13 -07002452 con->error_msg = "connect error";
Sage Weil31b80062009-10-06 11:31:13 -07002453 goto out;
2454 }
2455 }
2456
2457more_kvec:
2458 /* kvec data queued? */
2459 if (con->out_skip) {
2460 ret = write_partial_skip(con);
2461 if (ret <= 0)
Sage Weil42961d22011-01-25 08:19:34 -08002462 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002463 }
2464 if (con->out_kvec_left) {
2465 ret = write_partial_kvec(con);
2466 if (ret <= 0)
Sage Weil42961d22011-01-25 08:19:34 -08002467 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002468 }
2469
2470 /* msg pages? */
2471 if (con->out_msg) {
Sage Weilc86a2932009-12-14 14:04:30 -08002472 if (con->out_msg_done) {
2473 ceph_msg_put(con->out_msg);
2474 con->out_msg = NULL; /* we're done with this one */
2475 goto do_next;
2476 }
2477
Alex Elder34d2d202013-03-08 20:58:59 -06002478 ret = write_partial_message_data(con);
Sage Weil31b80062009-10-06 11:31:13 -07002479 if (ret == 1)
2480 goto more_kvec; /* we need to send the footer, too! */
2481 if (ret == 0)
Sage Weil42961d22011-01-25 08:19:34 -08002482 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002483 if (ret < 0) {
Alex Elder34d2d202013-03-08 20:58:59 -06002484 dout("try_write write_partial_message_data err %d\n",
Sage Weil31b80062009-10-06 11:31:13 -07002485 ret);
Sage Weil42961d22011-01-25 08:19:34 -08002486 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002487 }
2488 }
2489
Sage Weilc86a2932009-12-14 14:04:30 -08002490do_next:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002491 if (con->state == CON_STATE_OPEN) {
Sage Weil31b80062009-10-06 11:31:13 -07002492 /* is anything else pending? */
2493 if (!list_empty(&con->out_queue)) {
2494 prepare_write_message(con);
2495 goto more;
2496 }
2497 if (con->in_seq > con->in_seq_acked) {
2498 prepare_write_ack(con);
2499 goto more;
2500 }
Alex Elderc9ffc772013-02-20 10:25:12 -06002501 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
Sage Weil31b80062009-10-06 11:31:13 -07002502 prepare_write_keepalive(con);
2503 goto more;
2504 }
2505 }
2506
2507 /* Nothing to do! */
Alex Elderc9ffc772013-02-20 10:25:12 -06002508 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07002509 dout("try_write nothing else to write.\n");
Sage Weil31b80062009-10-06 11:31:13 -07002510 ret = 0;
2511out:
Sage Weil42961d22011-01-25 08:19:34 -08002512 dout("try_write done on %p ret %d\n", con, ret);
Sage Weil31b80062009-10-06 11:31:13 -07002513 return ret;
2514}
2515
2516
2517
2518/*
2519 * Read what we can from the socket.
2520 */
2521static int try_read(struct ceph_connection *con)
2522{
Sage Weil31b80062009-10-06 11:31:13 -07002523 int ret = -1;
2524
Sage Weil31b80062009-10-06 11:31:13 -07002525more:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002526 dout("try_read start on %p state %lu\n", con, con->state);
2527 if (con->state != CON_STATE_CONNECTING &&
2528 con->state != CON_STATE_NEGOTIATING &&
2529 con->state != CON_STATE_OPEN)
2530 return 0;
2531
2532 BUG_ON(!con->sock);
2533
Sage Weil31b80062009-10-06 11:31:13 -07002534 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2535 con->in_base_pos);
Sage Weil0da5d702011-05-19 11:21:05 -07002536
Sage Weil8dacc7d2012-07-20 17:24:40 -07002537 if (con->state == CON_STATE_CONNECTING) {
Alex Elder7593af92012-05-24 11:55:03 -05002538 dout("try_read connecting\n");
2539 ret = read_partial_banner(con);
2540 if (ret <= 0)
Alex Elderab166d52012-05-31 11:37:29 -05002541 goto out;
Alex Elder7593af92012-05-24 11:55:03 -05002542 ret = process_banner(con);
2543 if (ret < 0)
2544 goto out;
2545
Sage Weil8dacc7d2012-07-20 17:24:40 -07002546 con->state = CON_STATE_NEGOTIATING;
Alex Elder7593af92012-05-24 11:55:03 -05002547
Jim Schutt6d4221b2012-08-10 10:37:38 -07002548 /*
2549 * Received banner is good, exchange connection info.
2550 * Do not reset out_kvec, as sending our banner raced
2551 * with receiving peer banner after connect completed.
2552 */
Alex Elder7593af92012-05-24 11:55:03 -05002553 ret = prepare_write_connect(con);
2554 if (ret < 0)
2555 goto out;
2556 prepare_read_connect(con);
2557
2558 /* Send connection info before awaiting response */
Sage Weil0da5d702011-05-19 11:21:05 -07002559 goto out;
2560 }
2561
Sage Weil8dacc7d2012-07-20 17:24:40 -07002562 if (con->state == CON_STATE_NEGOTIATING) {
Alex Elder7593af92012-05-24 11:55:03 -05002563 dout("try_read negotiating\n");
Sage Weil31b80062009-10-06 11:31:13 -07002564 ret = read_partial_connect(con);
2565 if (ret <= 0)
Sage Weil31b80062009-10-06 11:31:13 -07002566 goto out;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002567 ret = process_connect(con);
2568 if (ret < 0)
2569 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002570 goto more;
2571 }
2572
Alex Elder122070a2012-12-26 10:43:57 -06002573 WARN_ON(con->state != CON_STATE_OPEN);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002574
Sage Weil31b80062009-10-06 11:31:13 -07002575 if (con->in_base_pos < 0) {
2576 /*
2577 * skipping + discarding content.
2578 *
2579 * FIXME: there must be a better way to do this!
2580 */
Alex Elder84495f42012-02-15 07:43:55 -06002581 static char buf[SKIP_BUF_SIZE];
2582 int skip = min((int) sizeof (buf), -con->in_base_pos);
2583
Sage Weil31b80062009-10-06 11:31:13 -07002584 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2585 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2586 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002587 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002588 con->in_base_pos += ret;
2589 if (con->in_base_pos)
2590 goto more;
2591 }
2592 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2593 /*
2594 * what's next?
2595 */
2596 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2597 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002598 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002599 dout("try_read got tag %d\n", (int)con->in_tag);
2600 switch (con->in_tag) {
2601 case CEPH_MSGR_TAG_MSG:
2602 prepare_read_message(con);
2603 break;
2604 case CEPH_MSGR_TAG_ACK:
2605 prepare_read_ack(con);
2606 break;
2607 case CEPH_MSGR_TAG_CLOSE:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002608 con_close_socket(con);
2609 con->state = CON_STATE_CLOSED;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002610 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002611 default:
2612 goto bad_tag;
2613 }
2614 }
2615 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2616 ret = read_partial_message(con);
2617 if (ret <= 0) {
2618 switch (ret) {
2619 case -EBADMSG:
2620 con->error_msg = "bad crc";
2621 ret = -EIO;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002622 break;
Sage Weil31b80062009-10-06 11:31:13 -07002623 case -EIO:
2624 con->error_msg = "io error";
Sage Weil98bdb0a2011-01-25 08:17:48 -08002625 break;
Sage Weil31b80062009-10-06 11:31:13 -07002626 }
Sage Weil98bdb0a2011-01-25 08:17:48 -08002627 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002628 }
2629 if (con->in_tag == CEPH_MSGR_TAG_READY)
2630 goto more;
2631 process_message(con);
Sage Weil7b862e02012-07-30 18:16:56 -07002632 if (con->state == CON_STATE_OPEN)
2633 prepare_read_tag(con);
Sage Weil31b80062009-10-06 11:31:13 -07002634 goto more;
2635 }
Sage Weil3a230832013-03-25 08:47:40 -07002636 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2637 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2638 /*
2639 * the final handshake seq exchange is semantically
2640 * equivalent to an ACK
2641 */
Sage Weil31b80062009-10-06 11:31:13 -07002642 ret = read_partial_ack(con);
2643 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002644 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002645 process_ack(con);
2646 goto more;
2647 }
2648
Sage Weil31b80062009-10-06 11:31:13 -07002649out:
Sage Weil98bdb0a2011-01-25 08:17:48 -08002650 dout("try_read done on %p ret %d\n", con, ret);
Sage Weil31b80062009-10-06 11:31:13 -07002651 return ret;
2652
2653bad_tag:
2654 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2655 con->error_msg = "protocol error, garbage tag";
2656 ret = -1;
2657 goto out;
2658}
2659
2660
2661/*
Alex Elder802c6d92012-10-08 20:37:30 -07002662 * Atomically queue work on a connection after the specified delay.
2663 * Bump @con reference to avoid races with connection teardown.
2664 * Returns 0 if work was queued, or an error code otherwise.
Sage Weil31b80062009-10-06 11:31:13 -07002665 */
Alex Elder802c6d92012-10-08 20:37:30 -07002666static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
Sage Weil31b80062009-10-06 11:31:13 -07002667{
Sage Weil31b80062009-10-06 11:31:13 -07002668 if (!con->ops->get(con)) {
Alex Elder802c6d92012-10-08 20:37:30 -07002669 dout("%s %p ref count 0\n", __func__, con);
2670
2671 return -ENOENT;
Sage Weil31b80062009-10-06 11:31:13 -07002672 }
2673
Alex Elder802c6d92012-10-08 20:37:30 -07002674 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2675 dout("%s %p - already queued\n", __func__, con);
Sage Weil31b80062009-10-06 11:31:13 -07002676 con->ops->put(con);
Alex Elder802c6d92012-10-08 20:37:30 -07002677
2678 return -EBUSY;
Sage Weil31b80062009-10-06 11:31:13 -07002679 }
Alex Elder802c6d92012-10-08 20:37:30 -07002680
2681 dout("%s %p %lu\n", __func__, con, delay);
2682
2683 return 0;
2684}
2685
2686static void queue_con(struct ceph_connection *con)
2687{
2688 (void) queue_con_delay(con, 0);
Sage Weil31b80062009-10-06 11:31:13 -07002689}
2690
Alex Elder7bb21d682012-12-07 19:50:07 -06002691static bool con_sock_closed(struct ceph_connection *con)
2692{
Alex Elderc9ffc772013-02-20 10:25:12 -06002693 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
Alex Elder7bb21d682012-12-07 19:50:07 -06002694 return false;
2695
2696#define CASE(x) \
2697 case CON_STATE_ ## x: \
2698 con->error_msg = "socket closed (con state " #x ")"; \
2699 break;
2700
2701 switch (con->state) {
2702 CASE(CLOSED);
2703 CASE(PREOPEN);
2704 CASE(CONNECTING);
2705 CASE(NEGOTIATING);
2706 CASE(OPEN);
2707 CASE(STANDBY);
2708 default:
2709 pr_warning("%s con %p unrecognized state %lu\n",
2710 __func__, con, con->state);
2711 con->error_msg = "unrecognized con state";
2712 BUG();
2713 break;
2714 }
2715#undef CASE
2716
2717 return true;
2718}
2719
Alex Elderf20a39f2013-02-19 12:25:57 -06002720static bool con_backoff(struct ceph_connection *con)
2721{
2722 int ret;
2723
2724 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2725 return false;
2726
2727 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2728 if (ret) {
2729 dout("%s: con %p FAILED to back off %lu\n", __func__,
2730 con, con->delay);
2731 BUG_ON(ret == -ENOENT);
2732 con_flag_set(con, CON_FLAG_BACKOFF);
2733 }
2734
2735 return true;
2736}
2737
Alex Elder93209262013-02-19 12:25:57 -06002738/* Finish fault handling; con->mutex must *not* be held here */
2739
2740static void con_fault_finish(struct ceph_connection *con)
2741{
2742 /*
2743 * in case we faulted due to authentication, invalidate our
2744 * current tickets so that we can get new ones.
2745 */
2746 if (con->auth_retry && con->ops->invalidate_authorizer) {
2747 dout("calling invalidate_authorizer()\n");
2748 con->ops->invalidate_authorizer(con);
2749 }
2750
2751 if (con->ops->fault)
2752 con->ops->fault(con);
2753}
2754
Sage Weil31b80062009-10-06 11:31:13 -07002755/*
2756 * Do some work on a connection. Drop a connection ref when we're done.
2757 */
2758static void con_work(struct work_struct *work)
2759{
2760 struct ceph_connection *con = container_of(work, struct ceph_connection,
2761 work.work);
Alex Elder49659412013-02-19 12:25:57 -06002762 bool fault;
Sage Weil31b80062009-10-06 11:31:13 -07002763
Sage Weil9dd46582010-04-28 13:51:50 -07002764 mutex_lock(&con->mutex);
Alex Elder49659412013-02-19 12:25:57 -06002765 while (true) {
2766 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07002767
Alex Elder49659412013-02-19 12:25:57 -06002768 if ((fault = con_sock_closed(con))) {
2769 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2770 break;
2771 }
2772 if (con_backoff(con)) {
2773 dout("%s: con %p BACKOFF\n", __func__, con);
2774 break;
2775 }
2776 if (con->state == CON_STATE_STANDBY) {
2777 dout("%s: con %p STANDBY\n", __func__, con);
2778 break;
2779 }
2780 if (con->state == CON_STATE_CLOSED) {
2781 dout("%s: con %p CLOSED\n", __func__, con);
2782 BUG_ON(con->sock);
2783 break;
2784 }
2785 if (con->state == CON_STATE_PREOPEN) {
2786 dout("%s: con %p PREOPEN\n", __func__, con);
2787 BUG_ON(con->sock);
2788 }
Sage Weil0da5d702011-05-19 11:21:05 -07002789
Alex Elder49659412013-02-19 12:25:57 -06002790 ret = try_read(con);
2791 if (ret < 0) {
2792 if (ret == -EAGAIN)
2793 continue;
2794 con->error_msg = "socket error on read";
2795 fault = true;
2796 break;
2797 }
2798
2799 ret = try_write(con);
2800 if (ret < 0) {
2801 if (ret == -EAGAIN)
2802 continue;
2803 con->error_msg = "socket error on write";
2804 fault = true;
2805 }
2806
2807 break; /* If we make it to here, we're done */
Sage Weil3a140a02012-07-30 16:24:21 -07002808 }
Alex Elderb6e7b6a2013-02-19 12:25:57 -06002809 if (fault)
2810 con_fault(con);
Sage Weil9dd46582010-04-28 13:51:50 -07002811 mutex_unlock(&con->mutex);
Sage Weil0da5d702011-05-19 11:21:05 -07002812
Alex Elderb6e7b6a2013-02-19 12:25:57 -06002813 if (fault)
2814 con_fault_finish(con);
2815
2816 con->ops->put(con);
Sage Weil31b80062009-10-06 11:31:13 -07002817}
2818
Sage Weil31b80062009-10-06 11:31:13 -07002819/*
2820 * Generic error/fault handler. A retry mechanism is used with
2821 * exponential backoff
2822 */
Alex Elder93209262013-02-19 12:25:57 -06002823static void con_fault(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07002824{
Alex Elder28362982012-12-14 16:47:41 -06002825 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002826 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
Sage Weil31b80062009-10-06 11:31:13 -07002827 dout("fault %p state %lu to peer %s\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002828 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07002829
Alex Elder122070a2012-12-26 10:43:57 -06002830 WARN_ON(con->state != CON_STATE_CONNECTING &&
Sage Weil8dacc7d2012-07-20 17:24:40 -07002831 con->state != CON_STATE_NEGOTIATING &&
2832 con->state != CON_STATE_OPEN);
Sage Weilec302642009-12-22 10:43:42 -08002833
Sage Weil31b80062009-10-06 11:31:13 -07002834 con_close_socket(con);
Sage Weil5e095e82009-12-14 14:30:34 -08002835
Alex Elderc9ffc772013-02-20 10:25:12 -06002836 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
Sage Weil8dacc7d2012-07-20 17:24:40 -07002837 dout("fault on LOSSYTX channel, marking CLOSED\n");
2838 con->state = CON_STATE_CLOSED;
Alex Elder93209262013-02-19 12:25:57 -06002839 return;
Sage Weil3b5ede02012-07-20 15:22:53 -07002840 }
2841
Sage Weil5e095e82009-12-14 14:30:34 -08002842 if (con->in_msg) {
Alex Elder38941f82012-06-01 14:56:43 -05002843 BUG_ON(con->in_msg->con != con);
2844 con->in_msg->con = NULL;
Sage Weil5e095e82009-12-14 14:30:34 -08002845 ceph_msg_put(con->in_msg);
2846 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -07002847 con->ops->put(con);
Sage Weil5e095e82009-12-14 14:30:34 -08002848 }
Sage Weil31b80062009-10-06 11:31:13 -07002849
Sage Weile80a52d2010-02-25 12:40:45 -08002850 /* Requeue anything that hasn't been acked */
2851 list_splice_init(&con->out_sent, &con->out_queue);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002852
Sage Weile76661d2011-03-03 10:10:15 -08002853 /* If there are no messages queued or keepalive pending, place
2854 * the connection in a STANDBY state */
2855 if (list_empty(&con->out_queue) &&
Alex Elderc9ffc772013-02-20 10:25:12 -06002856 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
Sage Weile00de342011-03-04 12:25:05 -08002857 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
Alex Elderc9ffc772013-02-20 10:25:12 -06002858 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002859 con->state = CON_STATE_STANDBY;
Sage Weile80a52d2010-02-25 12:40:45 -08002860 } else {
2861 /* retry after a delay. */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002862 con->state = CON_STATE_PREOPEN;
Sage Weile80a52d2010-02-25 12:40:45 -08002863 if (con->delay == 0)
2864 con->delay = BASE_DELAY_INTERVAL;
2865 else if (con->delay < MAX_DELAY_INTERVAL)
2866 con->delay *= 2;
Alex Elderc9ffc772013-02-20 10:25:12 -06002867 con_flag_set(con, CON_FLAG_BACKOFF);
Alex Elder8618e302012-10-08 20:37:30 -07002868 queue_con(con);
Sage Weil31b80062009-10-06 11:31:13 -07002869 }
Sage Weil31b80062009-10-06 11:31:13 -07002870}
2871
2872
2873
2874/*
Alex Elder15d98822012-05-26 23:26:43 -05002875 * initialize a new messenger instance
Sage Weil31b80062009-10-06 11:31:13 -07002876 */
Alex Elder15d98822012-05-26 23:26:43 -05002877void ceph_messenger_init(struct ceph_messenger *msgr,
2878 struct ceph_entity_addr *myaddr,
Ilya Dryomov12b46292013-12-24 21:19:23 +02002879 u64 supported_features,
2880 u64 required_features,
Alex Elder15d98822012-05-26 23:26:43 -05002881 bool nocrc)
Sage Weil31b80062009-10-06 11:31:13 -07002882{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002883 msgr->supported_features = supported_features;
2884 msgr->required_features = required_features;
2885
Sage Weil31b80062009-10-06 11:31:13 -07002886 spin_lock_init(&msgr->global_seq_lock);
2887
Sage Weil31b80062009-10-06 11:31:13 -07002888 if (myaddr)
2889 msgr->inst.addr = *myaddr;
2890
2891 /* select a random nonce */
Sage Weilac8839d2010-01-27 14:28:10 -08002892 msgr->inst.addr.type = 0;
Sage Weil103e2d32010-01-07 16:12:36 -08002893 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
Sage Weil63f2d212009-11-03 15:17:56 -08002894 encode_my_addr(msgr);
Alex Elder15d98822012-05-26 23:26:43 -05002895 msgr->nocrc = nocrc;
Sage Weil31b80062009-10-06 11:31:13 -07002896
Guanjun Hea2a32582012-07-08 19:50:33 -07002897 atomic_set(&msgr->stopping, 0);
Sage Weil31b80062009-10-06 11:31:13 -07002898
Alex Elder15d98822012-05-26 23:26:43 -05002899 dout("%s %p\n", __func__, msgr);
Sage Weil31b80062009-10-06 11:31:13 -07002900}
Alex Elder15d98822012-05-26 23:26:43 -05002901EXPORT_SYMBOL(ceph_messenger_init);
Sage Weil31b80062009-10-06 11:31:13 -07002902
Sage Weile00de342011-03-04 12:25:05 -08002903static void clear_standby(struct ceph_connection *con)
2904{
2905 /* come back from STANDBY? */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002906 if (con->state == CON_STATE_STANDBY) {
Sage Weile00de342011-03-04 12:25:05 -08002907 dout("clear_standby %p and ++connect_seq\n", con);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002908 con->state = CON_STATE_PREOPEN;
Sage Weile00de342011-03-04 12:25:05 -08002909 con->connect_seq++;
Alex Elderc9ffc772013-02-20 10:25:12 -06002910 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2911 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
Sage Weile00de342011-03-04 12:25:05 -08002912 }
2913}
2914
Sage Weil31b80062009-10-06 11:31:13 -07002915/*
2916 * Queue up an outgoing message on the given connection.
2917 */
2918void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2919{
Sage Weila59b55a2012-07-20 15:34:04 -07002920 /* set src+dst */
2921 msg->hdr.src = con->msgr->inst.name;
2922 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2923 msg->needs_out_seq = true;
2924
2925 mutex_lock(&con->mutex);
2926
Sage Weil8dacc7d2012-07-20 17:24:40 -07002927 if (con->state == CON_STATE_CLOSED) {
Sage Weil31b80062009-10-06 11:31:13 -07002928 dout("con_send %p closed, dropping %p\n", con, msg);
2929 ceph_msg_put(msg);
Sage Weila59b55a2012-07-20 15:34:04 -07002930 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002931 return;
2932 }
2933
Alex Elder38941f82012-06-01 14:56:43 -05002934 BUG_ON(msg->con != NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -07002935 msg->con = con->ops->get(con);
Alex Elder92ce0342012-06-04 14:43:33 -05002936 BUG_ON(msg->con == NULL);
Sage Weil31b80062009-10-06 11:31:13 -07002937
Sage Weil31b80062009-10-06 11:31:13 -07002938 BUG_ON(!list_empty(&msg->list_head));
2939 list_add_tail(&msg->list_head, &con->out_queue);
2940 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2941 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2942 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2943 le32_to_cpu(msg->hdr.front_len),
2944 le32_to_cpu(msg->hdr.middle_len),
2945 le32_to_cpu(msg->hdr.data_len));
Sage Weil00650932012-07-20 15:33:04 -07002946
2947 clear_standby(con);
Sage Weilec302642009-12-22 10:43:42 -08002948 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002949
2950 /* if there wasn't anything waiting to send before, queue
2951 * new work */
Alex Elderc9ffc772013-02-20 10:25:12 -06002952 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
Sage Weil31b80062009-10-06 11:31:13 -07002953 queue_con(con);
2954}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002955EXPORT_SYMBOL(ceph_con_send);
Sage Weil31b80062009-10-06 11:31:13 -07002956
2957/*
2958 * Revoke a message that was previously queued for send
2959 */
Alex Elder6740a842012-06-01 14:56:43 -05002960void ceph_msg_revoke(struct ceph_msg *msg)
Sage Weil31b80062009-10-06 11:31:13 -07002961{
Alex Elder6740a842012-06-01 14:56:43 -05002962 struct ceph_connection *con = msg->con;
2963
2964 if (!con)
2965 return; /* Message not in our possession */
2966
Sage Weilec302642009-12-22 10:43:42 -08002967 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002968 if (!list_empty(&msg->list_head)) {
Alex Elder38941f82012-06-01 14:56:43 -05002969 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07002970 list_del_init(&msg->list_head);
Alex Elder38941f82012-06-01 14:56:43 -05002971 BUG_ON(msg->con == NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -07002972 msg->con->ops->put(msg->con);
Alex Elder38941f82012-06-01 14:56:43 -05002973 msg->con = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002974 msg->hdr.seq = 0;
Alex Elder38941f82012-06-01 14:56:43 -05002975
Sage Weil31b80062009-10-06 11:31:13 -07002976 ceph_msg_put(msg);
Sage Weiled98ada2010-07-05 12:15:14 -07002977 }
2978 if (con->out_msg == msg) {
Alex Elder38941f82012-06-01 14:56:43 -05002979 dout("%s %p msg %p - was sending\n", __func__, con, msg);
Sage Weiled98ada2010-07-05 12:15:14 -07002980 con->out_msg = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002981 if (con->out_kvec_is_msg) {
2982 con->out_skip = con->out_kvec_bytes;
2983 con->out_kvec_is_msg = false;
2984 }
Sage Weiled98ada2010-07-05 12:15:14 -07002985 msg->hdr.seq = 0;
Alex Elder92ce0342012-06-04 14:43:33 -05002986
2987 ceph_msg_put(msg);
Sage Weil31b80062009-10-06 11:31:13 -07002988 }
Sage Weilec302642009-12-22 10:43:42 -08002989 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002990}
2991
2992/*
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002993 * Revoke a message that we may be reading data into
Sage Weil350b1c32009-12-22 10:45:45 -08002994 */
Alex Elder8921d112012-06-01 14:56:43 -05002995void ceph_msg_revoke_incoming(struct ceph_msg *msg)
Sage Weil350b1c32009-12-22 10:45:45 -08002996{
Alex Elder8921d112012-06-01 14:56:43 -05002997 struct ceph_connection *con;
2998
2999 BUG_ON(msg == NULL);
3000 if (!msg->con) {
3001 dout("%s msg %p null con\n", __func__, msg);
3002
3003 return; /* Message not in our possession */
3004 }
3005
3006 con = msg->con;
Sage Weil350b1c32009-12-22 10:45:45 -08003007 mutex_lock(&con->mutex);
Alex Elder8921d112012-06-01 14:56:43 -05003008 if (con->in_msg == msg) {
Eric Dumazet95c96172012-04-15 05:58:06 +00003009 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3010 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3011 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
Sage Weil350b1c32009-12-22 10:45:45 -08003012
3013 /* skip rest of message */
Alex Elder8921d112012-06-01 14:56:43 -05003014 dout("%s %p msg %p revoked\n", __func__, con, msg);
3015 con->in_base_pos = con->in_base_pos -
Sage Weil350b1c32009-12-22 10:45:45 -08003016 sizeof(struct ceph_msg_header) -
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08003017 front_len -
3018 middle_len -
3019 data_len -
Sage Weil350b1c32009-12-22 10:45:45 -08003020 sizeof(struct ceph_msg_footer);
Sage Weil350b1c32009-12-22 10:45:45 -08003021 ceph_msg_put(con->in_msg);
3022 con->in_msg = NULL;
3023 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weil684be252010-04-21 20:45:59 -07003024 con->in_seq++;
Sage Weil350b1c32009-12-22 10:45:45 -08003025 } else {
Alex Elder8921d112012-06-01 14:56:43 -05003026 dout("%s %p in_msg %p msg %p no-op\n",
3027 __func__, con, con->in_msg, msg);
Sage Weil350b1c32009-12-22 10:45:45 -08003028 }
3029 mutex_unlock(&con->mutex);
3030}
3031
3032/*
Sage Weil31b80062009-10-06 11:31:13 -07003033 * Queue a keepalive byte to ensure the tcp connection is alive.
3034 */
3035void ceph_con_keepalive(struct ceph_connection *con)
3036{
Sage Weile00de342011-03-04 12:25:05 -08003037 dout("con_keepalive %p\n", con);
Sage Weil00650932012-07-20 15:33:04 -07003038 mutex_lock(&con->mutex);
Sage Weile00de342011-03-04 12:25:05 -08003039 clear_standby(con);
Sage Weil00650932012-07-20 15:33:04 -07003040 mutex_unlock(&con->mutex);
Alex Elderc9ffc772013-02-20 10:25:12 -06003041 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3042 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
Sage Weil31b80062009-10-06 11:31:13 -07003043 queue_con(con);
3044}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003045EXPORT_SYMBOL(ceph_con_keepalive);
Sage Weil31b80062009-10-06 11:31:13 -07003046
Alex Elder6644ed72013-03-11 23:34:24 -05003047static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
Alex Elder43794502013-03-01 18:00:16 -06003048{
Alex Elder6644ed72013-03-11 23:34:24 -05003049 struct ceph_msg_data *data;
3050
3051 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3052 return NULL;
3053
Alex Elder81b36be2013-05-01 12:43:04 -05003054 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
Alex Elder6644ed72013-03-11 23:34:24 -05003055 if (data)
3056 data->type = type;
Alex Elder5240d9f2013-03-14 14:09:06 -05003057 INIT_LIST_HEAD(&data->links);
Alex Elder6644ed72013-03-11 23:34:24 -05003058
3059 return data;
3060}
3061
3062static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3063{
3064 if (!data)
3065 return;
3066
Alex Elder5240d9f2013-03-14 14:09:06 -05003067 WARN_ON(!list_empty(&data->links));
Alex Elder6644ed72013-03-11 23:34:24 -05003068 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3069 ceph_pagelist_release(data->pagelist);
3070 kfree(data->pagelist);
3071 }
Alex Elder81b36be2013-05-01 12:43:04 -05003072 kmem_cache_free(ceph_msg_data_cache, data);
Alex Elder43794502013-03-01 18:00:16 -06003073}
3074
Alex Elder90af3602013-04-05 14:46:01 -05003075void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
Alex Elderf1baeb22013-03-07 15:38:26 -06003076 size_t length, size_t alignment)
Alex Elder02afca62013-02-14 12:16:43 -06003077{
Alex Elder6644ed72013-03-11 23:34:24 -05003078 struct ceph_msg_data *data;
3079
Alex Elder07aa1552013-03-04 18:29:06 -06003080 BUG_ON(!pages);
3081 BUG_ON(!length);
Alex Elder02afca62013-02-14 12:16:43 -06003082
Alex Elder6644ed72013-03-11 23:34:24 -05003083 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3084 BUG_ON(!data);
3085 data->pages = pages;
3086 data->length = length;
3087 data->alignment = alignment & ~PAGE_MASK;
3088
Alex Elder5240d9f2013-03-14 14:09:06 -05003089 list_add_tail(&data->links, &msg->data);
3090 msg->data_length += length;
Alex Elder02afca62013-02-14 12:16:43 -06003091}
Alex Elder90af3602013-04-05 14:46:01 -05003092EXPORT_SYMBOL(ceph_msg_data_add_pages);
Sage Weil31b80062009-10-06 11:31:13 -07003093
Alex Elder90af3602013-04-05 14:46:01 -05003094void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
Alex Elder27fa8382013-02-14 12:16:43 -06003095 struct ceph_pagelist *pagelist)
3096{
Alex Elder6644ed72013-03-11 23:34:24 -05003097 struct ceph_msg_data *data;
3098
Alex Elder07aa1552013-03-04 18:29:06 -06003099 BUG_ON(!pagelist);
3100 BUG_ON(!pagelist->length);
Alex Elder27fa8382013-02-14 12:16:43 -06003101
Alex Elder6644ed72013-03-11 23:34:24 -05003102 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3103 BUG_ON(!data);
3104 data->pagelist = pagelist;
3105
Alex Elder5240d9f2013-03-14 14:09:06 -05003106 list_add_tail(&data->links, &msg->data);
3107 msg->data_length += pagelist->length;
Alex Elder27fa8382013-02-14 12:16:43 -06003108}
Alex Elder90af3602013-04-05 14:46:01 -05003109EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
Alex Elder27fa8382013-02-14 12:16:43 -06003110
Alex Elderea965712013-04-05 14:46:01 -05003111#ifdef CONFIG_BLOCK
Alex Elder90af3602013-04-05 14:46:01 -05003112void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
Alex Eldera1930802013-03-14 14:09:06 -05003113 size_t length)
Alex Elder27fa8382013-02-14 12:16:43 -06003114{
Alex Elder6644ed72013-03-11 23:34:24 -05003115 struct ceph_msg_data *data;
Alex Elder27fa8382013-02-14 12:16:43 -06003116
Alex Elder6644ed72013-03-11 23:34:24 -05003117 BUG_ON(!bio);
Alex Elder6644ed72013-03-11 23:34:24 -05003118
3119 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3120 BUG_ON(!data);
3121 data->bio = bio;
Alex Elderc851c492013-04-05 14:46:01 -05003122 data->bio_length = length;
Alex Elder6644ed72013-03-11 23:34:24 -05003123
Alex Elder5240d9f2013-03-14 14:09:06 -05003124 list_add_tail(&data->links, &msg->data);
3125 msg->data_length += length;
Alex Elder27fa8382013-02-14 12:16:43 -06003126}
Alex Elder90af3602013-04-05 14:46:01 -05003127EXPORT_SYMBOL(ceph_msg_data_add_bio);
Alex Elderea965712013-04-05 14:46:01 -05003128#endif /* CONFIG_BLOCK */
Alex Elder27fa8382013-02-14 12:16:43 -06003129
Sage Weil31b80062009-10-06 11:31:13 -07003130/*
3131 * construct a new message with given type, size
3132 * the new msg has a ref count of 1.
3133 */
Sage Weilb61c2762011-08-09 15:03:46 -07003134struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3135 bool can_fail)
Sage Weil31b80062009-10-06 11:31:13 -07003136{
3137 struct ceph_msg *m;
3138
Alex Eldere3d5d632013-05-01 12:43:04 -05003139 m = kmem_cache_zalloc(ceph_msg_cache, flags);
Sage Weil31b80062009-10-06 11:31:13 -07003140 if (m == NULL)
3141 goto out;
Alex Elder38941f82012-06-01 14:56:43 -05003142
Sage Weil31b80062009-10-06 11:31:13 -07003143 m->hdr.type = cpu_to_le16(type);
Sage Weil45c6ceb2010-05-11 15:01:51 -07003144 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
Sage Weil31b80062009-10-06 11:31:13 -07003145 m->hdr.front_len = cpu_to_le32(front_len);
Sage Weil31b80062009-10-06 11:31:13 -07003146
Alex Elder9516e452013-03-01 18:00:16 -06003147 INIT_LIST_HEAD(&m->list_head);
3148 kref_init(&m->kref);
Alex Elder5240d9f2013-03-14 14:09:06 -05003149 INIT_LIST_HEAD(&m->data);
Henry C Changca208922011-05-03 02:29:56 +00003150
Sage Weil31b80062009-10-06 11:31:13 -07003151 /* front */
3152 if (front_len) {
Ilya Dryomoveeb0bed2014-01-09 20:08:21 +02003153 m->front.iov_base = ceph_kvmalloc(front_len, flags);
Sage Weil31b80062009-10-06 11:31:13 -07003154 if (m->front.iov_base == NULL) {
Sage Weilb61c2762011-08-09 15:03:46 -07003155 dout("ceph_msg_new can't allocate %d bytes\n",
Sage Weil31b80062009-10-06 11:31:13 -07003156 front_len);
3157 goto out2;
3158 }
3159 } else {
3160 m->front.iov_base = NULL;
3161 }
Ilya Dryomovf2be82b2014-01-09 20:08:21 +02003162 m->front_alloc_len = m->front.iov_len = front_len;
Sage Weil31b80062009-10-06 11:31:13 -07003163
Sage Weilbb257662010-04-01 16:07:23 -07003164 dout("ceph_msg_new %p front %d\n", m, front_len);
Sage Weil31b80062009-10-06 11:31:13 -07003165 return m;
3166
3167out2:
3168 ceph_msg_put(m);
3169out:
Sage Weilb61c2762011-08-09 15:03:46 -07003170 if (!can_fail) {
3171 pr_err("msg_new can't create type %d front %d\n", type,
3172 front_len);
Sage Weilf0ed1b72011-08-09 15:05:07 -07003173 WARN_ON(1);
Sage Weilb61c2762011-08-09 15:03:46 -07003174 } else {
3175 dout("msg_new can't create type %d front %d\n", type,
3176 front_len);
3177 }
Sage Weila79832f2010-04-01 16:06:19 -07003178 return NULL;
Sage Weil31b80062009-10-06 11:31:13 -07003179}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003180EXPORT_SYMBOL(ceph_msg_new);
Sage Weil31b80062009-10-06 11:31:13 -07003181
3182/*
Sage Weil31b80062009-10-06 11:31:13 -07003183 * Allocate "middle" portion of a message, if it is needed and wasn't
3184 * allocated by alloc_msg. This allows us to read a small fixed-size
3185 * per-type header in the front and then gracefully fail (i.e.,
3186 * propagate the error to the caller based on info in the front) when
3187 * the middle is too large.
3188 */
Yehuda Sadeh24504182010-01-08 13:58:34 -08003189static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
Sage Weil31b80062009-10-06 11:31:13 -07003190{
3191 int type = le16_to_cpu(msg->hdr.type);
3192 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3193
3194 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3195 ceph_msg_type_name(type), middle_len);
3196 BUG_ON(!middle_len);
3197 BUG_ON(msg->middle);
3198
Sage Weilb6c1d5b2009-12-07 12:17:17 -08003199 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
Sage Weil31b80062009-10-06 11:31:13 -07003200 if (!msg->middle)
3201 return -ENOMEM;
3202 return 0;
3203}
3204
Yehuda Sadeh24504182010-01-08 13:58:34 -08003205/*
Alex Elder1c20f2d2012-06-04 14:43:32 -05003206 * Allocate a message for receiving an incoming message on a
3207 * connection, and save the result in con->in_msg. Uses the
3208 * connection's private alloc_msg op if available.
3209 *
Sage Weil4740a622012-07-30 18:19:30 -07003210 * Returns 0 on success, or a negative error code.
3211 *
3212 * On success, if we set *skip = 1:
3213 * - the next message should be skipped and ignored.
3214 * - con->in_msg == NULL
3215 * or if we set *skip = 0:
3216 * - con->in_msg is non-null.
3217 * On error (ENOMEM, EAGAIN, ...),
3218 * - con->in_msg == NULL
Yehuda Sadeh24504182010-01-08 13:58:34 -08003219 */
Sage Weil4740a622012-07-30 18:19:30 -07003220static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
Yehuda Sadeh24504182010-01-08 13:58:34 -08003221{
Sage Weil4740a622012-07-30 18:19:30 -07003222 struct ceph_msg_header *hdr = &con->in_hdr;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003223 int middle_len = le32_to_cpu(hdr->middle_len);
Alex Elder1d866d12013-03-01 18:00:14 -06003224 struct ceph_msg *msg;
Sage Weil4740a622012-07-30 18:19:30 -07003225 int ret = 0;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003226
Alex Elder1c20f2d2012-06-04 14:43:32 -05003227 BUG_ON(con->in_msg != NULL);
Alex Elder53ded492013-03-01 18:00:14 -06003228 BUG_ON(!con->ops->alloc_msg);
Yehuda Sadeh24504182010-01-08 13:58:34 -08003229
Alex Elder53ded492013-03-01 18:00:14 -06003230 mutex_unlock(&con->mutex);
3231 msg = con->ops->alloc_msg(con, hdr, skip);
3232 mutex_lock(&con->mutex);
3233 if (con->state != CON_STATE_OPEN) {
3234 if (msg)
Alex Elder1d866d12013-03-01 18:00:14 -06003235 ceph_msg_put(msg);
Alex Elder53ded492013-03-01 18:00:14 -06003236 return -EAGAIN;
3237 }
Alex Elder41375772013-03-05 09:25:10 -06003238 if (msg) {
3239 BUG_ON(*skip);
3240 con->in_msg = msg;
Sage Weil36eb71a2012-06-21 12:47:08 -07003241 con->in_msg->con = con->ops->get(con);
Alex Elder92ce0342012-06-04 14:43:33 -05003242 BUG_ON(con->in_msg->con == NULL);
Alex Elder41375772013-03-05 09:25:10 -06003243 } else {
3244 /*
3245 * Null message pointer means either we should skip
3246 * this message or we couldn't allocate memory. The
3247 * former is not an error.
3248 */
3249 if (*skip)
3250 return 0;
3251 con->error_msg = "error allocating memory for incoming message";
3252
Alex Elder53ded492013-03-01 18:00:14 -06003253 return -ENOMEM;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003254 }
Alex Elder1c20f2d2012-06-04 14:43:32 -05003255 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
Yehuda Sadeh24504182010-01-08 13:58:34 -08003256
Alex Elder1c20f2d2012-06-04 14:43:32 -05003257 if (middle_len && !con->in_msg->middle) {
3258 ret = ceph_alloc_middle(con, con->in_msg);
Yehuda Sadeh24504182010-01-08 13:58:34 -08003259 if (ret < 0) {
Alex Elder1c20f2d2012-06-04 14:43:32 -05003260 ceph_msg_put(con->in_msg);
3261 con->in_msg = NULL;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003262 }
3263 }
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08003264
Sage Weil4740a622012-07-30 18:19:30 -07003265 return ret;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003266}
3267
Sage Weil31b80062009-10-06 11:31:13 -07003268
3269/*
3270 * Free a generically kmalloc'd message.
3271 */
3272void ceph_msg_kfree(struct ceph_msg *m)
3273{
3274 dout("msg_kfree %p\n", m);
Ilya Dryomoveeb0bed2014-01-09 20:08:21 +02003275 ceph_kvfree(m->front.iov_base);
Alex Eldere3d5d632013-05-01 12:43:04 -05003276 kmem_cache_free(ceph_msg_cache, m);
Sage Weil31b80062009-10-06 11:31:13 -07003277}
3278
3279/*
3280 * Drop a msg ref. Destroy as needed.
3281 */
Sage Weilc2e552e2009-12-07 15:55:05 -08003282void ceph_msg_last_put(struct kref *kref)
Sage Weil31b80062009-10-06 11:31:13 -07003283{
Sage Weilc2e552e2009-12-07 15:55:05 -08003284 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
Alex Elder5240d9f2013-03-14 14:09:06 -05003285 LIST_HEAD(data);
3286 struct list_head *links;
3287 struct list_head *next;
Sage Weil31b80062009-10-06 11:31:13 -07003288
Sage Weilc2e552e2009-12-07 15:55:05 -08003289 dout("ceph_msg_put last one on %p\n", m);
3290 WARN_ON(!list_empty(&m->list_head));
Sage Weil31b80062009-10-06 11:31:13 -07003291
Sage Weilc2e552e2009-12-07 15:55:05 -08003292 /* drop middle, data, if any */
3293 if (m->middle) {
3294 ceph_buffer_put(m->middle);
3295 m->middle = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07003296 }
Alex Elder5240d9f2013-03-14 14:09:06 -05003297
3298 list_splice_init(&m->data, &data);
3299 list_for_each_safe(links, next, &data) {
3300 struct ceph_msg_data *data;
3301
3302 data = list_entry(links, struct ceph_msg_data, links);
3303 list_del_init(links);
3304 ceph_msg_data_destroy(data);
3305 }
Alex Eldera1930802013-03-14 14:09:06 -05003306 m->data_length = 0;
Sage Weil58bb3b32009-12-23 12:12:31 -08003307
Sage Weilc2e552e2009-12-07 15:55:05 -08003308 if (m->pool)
3309 ceph_msgpool_put(m->pool, m);
3310 else
3311 ceph_msg_kfree(m);
Sage Weil31b80062009-10-06 11:31:13 -07003312}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003313EXPORT_SYMBOL(ceph_msg_last_put);
Sage Weil9ec7cab2009-12-14 15:13:47 -08003314
3315void ceph_msg_dump(struct ceph_msg *msg)
3316{
Ilya Dryomov3cea4c32014-01-09 20:08:21 +02003317 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3318 msg->front_alloc_len, msg->data_length);
Sage Weil9ec7cab2009-12-14 15:13:47 -08003319 print_hex_dump(KERN_DEBUG, "header: ",
3320 DUMP_PREFIX_OFFSET, 16, 1,
3321 &msg->hdr, sizeof(msg->hdr), true);
3322 print_hex_dump(KERN_DEBUG, " front: ",
3323 DUMP_PREFIX_OFFSET, 16, 1,
3324 msg->front.iov_base, msg->front.iov_len, true);
3325 if (msg->middle)
3326 print_hex_dump(KERN_DEBUG, "middle: ",
3327 DUMP_PREFIX_OFFSET, 16, 1,
3328 msg->middle->vec.iov_base,
3329 msg->middle->vec.iov_len, true);
3330 print_hex_dump(KERN_DEBUG, "footer: ",
3331 DUMP_PREFIX_OFFSET, 16, 1,
3332 &msg->footer, sizeof(msg->footer), true);
3333}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003334EXPORT_SYMBOL(ceph_msg_dump);