blob: 4a5df7b1cc9ff5652185e0248cf5a4df4006effd [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
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070018#include <linux/ceph/libceph.h>
19#include <linux/ceph/messenger.h>
20#include <linux/ceph/decode.h>
21#include <linux/ceph/pagelist.h>
Paul Gortmakerbc3b2d72011-07-15 11:47:34 -040022#include <linux/export.h>
Sage Weil31b80062009-10-06 11:31:13 -070023
Alex Elderfe38a2b2013-03-06 23:39:39 -060024#define list_entry_next(pos, member) \
25 list_entry(pos->member.next, typeof(*pos), member)
26
Sage Weil31b80062009-10-06 11:31:13 -070027/*
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
33 * the sender.
34 */
35
Alex Elderbc18f4b2012-06-20 21:53:53 -050036/*
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
40 * unexpected state.
41 *
42 * --------
43 * | NEW* | transient initial state
44 * --------
45 * | con_sock_state_init()
46 * v
47 * ----------
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
50 * ^ \
51 * | \ con_sock_state_connecting()
52 * | ----------------------
53 * | \
54 * + con_sock_state_closed() \
Sage Weilfbb85a42012-06-27 12:31:02 -070055 * |+--------------------------- \
56 * | \ \ \
57 * | ----------- \ \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
60 * | ^ \ |
61 * | | \ |
62 * | + con_sock_state_closing() \ |
63 * | / \ | |
64 * | / --------------- | |
65 * | / \ v v
Alex Elderbc18f4b2012-06-20 21:53:53 -050066 * | / --------------
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
70 * | | v
71 * -------------
72 * | CONNECTED | TCP connection established
73 * -------------
74 *
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
76 */
Alex Elderce2c8902012-05-22 22:15:49 -050077
78#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
83
Sage Weil8dacc7d2012-07-20 17:24:40 -070084/*
85 * connection states
86 */
87#define CON_STATE_CLOSED 1 /* -> PREOPEN */
88#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
93
Sage Weil4a861692012-07-20 17:29:55 -070094/*
95 * ceph_connection flag bits
96 */
97#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
Sage Weil8dacc7d2012-07-20 17:24:40 -0700103
Alex Elderc9ffc772013-02-20 10:25:12 -0600104static bool con_flag_valid(unsigned long con_flag)
105{
106 switch (con_flag) {
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
112 return true;
113 default:
114 return false;
115 }
116}
117
118static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
119{
120 BUG_ON(!con_flag_valid(con_flag));
121
122 clear_bit(con_flag, &con->flags);
123}
124
125static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
126{
127 BUG_ON(!con_flag_valid(con_flag));
128
129 set_bit(con_flag, &con->flags);
130}
131
132static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
133{
134 BUG_ON(!con_flag_valid(con_flag));
135
136 return test_bit(con_flag, &con->flags);
137}
138
139static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
141{
142 BUG_ON(!con_flag_valid(con_flag));
143
144 return test_and_clear_bit(con_flag, &con->flags);
145}
146
147static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
149{
150 BUG_ON(!con_flag_valid(con_flag));
151
152 return test_and_set_bit(con_flag, &con->flags);
153}
154
Alex Eldere3d5d632013-05-01 12:43:04 -0500155/* Slab caches for frequently-allocated structures */
156
157static struct kmem_cache *ceph_msg_cache;
Alex Elder81b36be2013-05-01 12:43:04 -0500158static struct kmem_cache *ceph_msg_data_cache;
Alex Eldere3d5d632013-05-01 12:43:04 -0500159
Sage Weil31b80062009-10-06 11:31:13 -0700160/* static tag bytes (protocol control messages) */
161static char tag_msg = CEPH_MSGR_TAG_MSG;
162static char tag_ack = CEPH_MSGR_TAG_ACK;
163static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
164
Sage Weila6a53492010-04-13 14:07:07 -0700165#ifdef CONFIG_LOCKDEP
166static struct lock_class_key socket_class;
167#endif
168
Alex Elder84495f42012-02-15 07:43:55 -0600169/*
170 * When skipping (ignoring) a block of input we read it into a "skip
171 * buffer," which is this many bytes in size.
172 */
173#define SKIP_BUF_SIZE 1024
Sage Weil31b80062009-10-06 11:31:13 -0700174
175static void queue_con(struct ceph_connection *con);
176static void con_work(struct work_struct *);
Alex Elder93209262013-02-19 12:25:57 -0600177static void con_fault(struct ceph_connection *con);
Sage Weil31b80062009-10-06 11:31:13 -0700178
Sage Weil31b80062009-10-06 11:31:13 -0700179/*
Alex Elderf64a9312012-01-23 15:49:27 -0600180 * Nicely render a sockaddr as a string. An array of formatted
181 * strings is used, to approximate reentrancy.
Sage Weil31b80062009-10-06 11:31:13 -0700182 */
Alex Elderf64a9312012-01-23 15:49:27 -0600183#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
184#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
185#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
186#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
187
188static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
189static atomic_t addr_str_seq = ATOMIC_INIT(0);
Sage Weil31b80062009-10-06 11:31:13 -0700190
Alex Elder57666512012-01-23 15:49:27 -0600191static struct page *zero_page; /* used in certain error cases */
Alex Elder57666512012-01-23 15:49:27 -0600192
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700193const char *ceph_pr_addr(const struct sockaddr_storage *ss)
Sage Weil31b80062009-10-06 11:31:13 -0700194{
195 int i;
196 char *s;
Alex Elder99f0f3b2012-01-23 15:49:27 -0600197 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
198 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
Sage Weil31b80062009-10-06 11:31:13 -0700199
Alex Elderf64a9312012-01-23 15:49:27 -0600200 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
Sage Weil31b80062009-10-06 11:31:13 -0700201 s = addr_str[i];
202
203 switch (ss->ss_family) {
204 case AF_INET:
Alex Elderbd406142012-01-23 15:49:27 -0600205 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
206 ntohs(in4->sin_port));
Sage Weil31b80062009-10-06 11:31:13 -0700207 break;
208
209 case AF_INET6:
Alex Elderbd406142012-01-23 15:49:27 -0600210 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
211 ntohs(in6->sin6_port));
Sage Weil31b80062009-10-06 11:31:13 -0700212 break;
213
214 default:
Alex Elderd3002b92012-02-14 14:05:33 -0600215 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
216 ss->ss_family);
Sage Weil31b80062009-10-06 11:31:13 -0700217 }
218
219 return s;
220}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700221EXPORT_SYMBOL(ceph_pr_addr);
Sage Weil31b80062009-10-06 11:31:13 -0700222
Sage Weil63f2d212009-11-03 15:17:56 -0800223static void encode_my_addr(struct ceph_messenger *msgr)
224{
225 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
226 ceph_encode_addr(&msgr->my_enc_addr);
227}
228
Sage Weil31b80062009-10-06 11:31:13 -0700229/*
230 * work queue for all reading and writing to/from the socket.
231 */
Alex Eldere0f43c92012-02-14 14:05:33 -0600232static struct workqueue_struct *ceph_msgr_wq;
Sage Weil31b80062009-10-06 11:31:13 -0700233
Alex Eldere3d5d632013-05-01 12:43:04 -0500234static int ceph_msgr_slab_init(void)
235{
236 BUG_ON(ceph_msg_cache);
237 ceph_msg_cache = kmem_cache_create("ceph_msg",
238 sizeof (struct ceph_msg),
239 __alignof__(struct ceph_msg), 0, NULL);
Alex Elder81b36be2013-05-01 12:43:04 -0500240
241 if (!ceph_msg_cache)
242 return -ENOMEM;
243
244 BUG_ON(ceph_msg_data_cache);
245 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
246 sizeof (struct ceph_msg_data),
247 __alignof__(struct ceph_msg_data),
248 0, NULL);
249 if (ceph_msg_data_cache)
250 return 0;
251
252 kmem_cache_destroy(ceph_msg_cache);
253 ceph_msg_cache = NULL;
254
255 return -ENOMEM;
Alex Eldere3d5d632013-05-01 12:43:04 -0500256}
257
258static void ceph_msgr_slab_exit(void)
259{
Alex Elder81b36be2013-05-01 12:43:04 -0500260 BUG_ON(!ceph_msg_data_cache);
261 kmem_cache_destroy(ceph_msg_data_cache);
262 ceph_msg_data_cache = NULL;
263
Alex Eldere3d5d632013-05-01 12:43:04 -0500264 BUG_ON(!ceph_msg_cache);
265 kmem_cache_destroy(ceph_msg_cache);
266 ceph_msg_cache = NULL;
267}
268
Alex Elder15417162013-02-19 12:25:56 -0600269static void _ceph_msgr_exit(void)
Alex Elder6173d1f2012-02-14 14:05:33 -0600270{
Alex Elderd3002b92012-02-14 14:05:33 -0600271 if (ceph_msgr_wq) {
Alex Elder6173d1f2012-02-14 14:05:33 -0600272 destroy_workqueue(ceph_msgr_wq);
Alex Elderd3002b92012-02-14 14:05:33 -0600273 ceph_msgr_wq = NULL;
274 }
Alex Elder6173d1f2012-02-14 14:05:33 -0600275
Alex Eldere3d5d632013-05-01 12:43:04 -0500276 ceph_msgr_slab_exit();
277
Alex Elder6173d1f2012-02-14 14:05:33 -0600278 BUG_ON(zero_page == NULL);
279 kunmap(zero_page);
280 page_cache_release(zero_page);
281 zero_page = NULL;
282}
283
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700284int ceph_msgr_init(void)
Sage Weil31b80062009-10-06 11:31:13 -0700285{
Alex Elder57666512012-01-23 15:49:27 -0600286 BUG_ON(zero_page != NULL);
287 zero_page = ZERO_PAGE(0);
288 page_cache_get(zero_page);
289
Alex Eldere3d5d632013-05-01 12:43:04 -0500290 if (ceph_msgr_slab_init())
291 return -ENOMEM;
292
Tejun Heo4d1829a2013-07-30 08:40:27 -0400293 ceph_msgr_wq = alloc_workqueue("ceph-msgr", 0, 0);
Alex Elder6173d1f2012-02-14 14:05:33 -0600294 if (ceph_msgr_wq)
295 return 0;
Alex Elder57666512012-01-23 15:49:27 -0600296
Alex Elder6173d1f2012-02-14 14:05:33 -0600297 pr_err("msgr_init failed to create workqueue\n");
298 _ceph_msgr_exit();
Alex Elder57666512012-01-23 15:49:27 -0600299
Alex Elder6173d1f2012-02-14 14:05:33 -0600300 return -ENOMEM;
Sage Weil31b80062009-10-06 11:31:13 -0700301}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700302EXPORT_SYMBOL(ceph_msgr_init);
Sage Weil31b80062009-10-06 11:31:13 -0700303
304void ceph_msgr_exit(void)
305{
Alex Elder57666512012-01-23 15:49:27 -0600306 BUG_ON(ceph_msgr_wq == NULL);
Alex Elder57666512012-01-23 15:49:27 -0600307
Alex Elder6173d1f2012-02-14 14:05:33 -0600308 _ceph_msgr_exit();
Sage Weil31b80062009-10-06 11:31:13 -0700309}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700310EXPORT_SYMBOL(ceph_msgr_exit);
Sage Weil31b80062009-10-06 11:31:13 -0700311
Yehuda Sadehcd84db62010-06-11 16:58:48 -0700312void ceph_msgr_flush(void)
Sage Weila922d382010-05-29 09:41:23 -0700313{
314 flush_workqueue(ceph_msgr_wq);
315}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700316EXPORT_SYMBOL(ceph_msgr_flush);
Sage Weila922d382010-05-29 09:41:23 -0700317
Alex Elderce2c8902012-05-22 22:15:49 -0500318/* Connection socket state transition functions */
319
320static void con_sock_state_init(struct ceph_connection *con)
321{
322 int old_state;
323
324 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
325 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
326 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700327 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
328 CON_SOCK_STATE_CLOSED);
Alex Elderce2c8902012-05-22 22:15:49 -0500329}
330
331static void con_sock_state_connecting(struct ceph_connection *con)
332{
333 int old_state;
334
335 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
336 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
337 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700338 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
339 CON_SOCK_STATE_CONNECTING);
Alex Elderce2c8902012-05-22 22:15:49 -0500340}
341
342static void con_sock_state_connected(struct ceph_connection *con)
343{
344 int old_state;
345
346 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
347 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
348 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700349 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
350 CON_SOCK_STATE_CONNECTED);
Alex Elderce2c8902012-05-22 22:15:49 -0500351}
352
353static void con_sock_state_closing(struct ceph_connection *con)
354{
355 int old_state;
356
357 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
358 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
359 old_state != CON_SOCK_STATE_CONNECTED &&
360 old_state != CON_SOCK_STATE_CLOSING))
361 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700362 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
363 CON_SOCK_STATE_CLOSING);
Alex Elderce2c8902012-05-22 22:15:49 -0500364}
365
366static void con_sock_state_closed(struct ceph_connection *con)
367{
368 int old_state;
369
370 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
371 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
Sage Weilfbb85a42012-06-27 12:31:02 -0700372 old_state != CON_SOCK_STATE_CLOSING &&
Sage Weil8007b8d2012-07-30 18:16:16 -0700373 old_state != CON_SOCK_STATE_CONNECTING &&
374 old_state != CON_SOCK_STATE_CLOSED))
Alex Elderce2c8902012-05-22 22:15:49 -0500375 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700376 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
377 CON_SOCK_STATE_CLOSED);
Alex Elderce2c8902012-05-22 22:15:49 -0500378}
Sage Weila922d382010-05-29 09:41:23 -0700379
Sage Weil31b80062009-10-06 11:31:13 -0700380/*
381 * socket callback functions
382 */
383
384/* data available on socket, or listen socket received a connect */
Alex Elder327800b2012-05-22 11:41:43 -0500385static void ceph_sock_data_ready(struct sock *sk, int count_unused)
Sage Weil31b80062009-10-06 11:31:13 -0700386{
Alex Elderbd406142012-01-23 15:49:27 -0600387 struct ceph_connection *con = sk->sk_user_data;
Guanjun Hea2a32582012-07-08 19:50:33 -0700388 if (atomic_read(&con->msgr->stopping)) {
389 return;
390 }
Alex Elderbd406142012-01-23 15:49:27 -0600391
Sage Weil31b80062009-10-06 11:31:13 -0700392 if (sk->sk_state != TCP_CLOSE_WAIT) {
Alex Elder327800b2012-05-22 11:41:43 -0500393 dout("%s on %p state = %lu, queueing work\n", __func__,
Sage Weil31b80062009-10-06 11:31:13 -0700394 con, con->state);
395 queue_con(con);
396 }
397}
398
399/* socket has buffer space for writing */
Alex Elder327800b2012-05-22 11:41:43 -0500400static void ceph_sock_write_space(struct sock *sk)
Sage Weil31b80062009-10-06 11:31:13 -0700401{
Alex Elderd3002b92012-02-14 14:05:33 -0600402 struct ceph_connection *con = sk->sk_user_data;
Sage Weil31b80062009-10-06 11:31:13 -0700403
Jim Schutt182fac22012-02-29 08:30:58 -0700404 /* only queue to workqueue if there is data we want to write,
405 * and there is sufficient space in the socket buffer to accept
Alex Elder327800b2012-05-22 11:41:43 -0500406 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
Jim Schutt182fac22012-02-29 08:30:58 -0700407 * doesn't get called again until try_write() fills the socket
408 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
409 * and net/core/stream.c:sk_stream_write_space().
410 */
Alex Elderc9ffc772013-02-20 10:25:12 -0600411 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
Eric Dumazet64dc6132013-07-22 20:26:31 -0700412 if (sk_stream_is_writeable(sk)) {
Alex Elder327800b2012-05-22 11:41:43 -0500413 dout("%s %p queueing write work\n", __func__, con);
Jim Schutt182fac22012-02-29 08:30:58 -0700414 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
415 queue_con(con);
416 }
Sage Weil31b80062009-10-06 11:31:13 -0700417 } else {
Alex Elder327800b2012-05-22 11:41:43 -0500418 dout("%s %p nothing to write\n", __func__, con);
Sage Weil31b80062009-10-06 11:31:13 -0700419 }
Sage Weil31b80062009-10-06 11:31:13 -0700420}
421
422/* socket's state has changed */
Alex Elder327800b2012-05-22 11:41:43 -0500423static void ceph_sock_state_change(struct sock *sk)
Sage Weil31b80062009-10-06 11:31:13 -0700424{
Alex Elderbd406142012-01-23 15:49:27 -0600425 struct ceph_connection *con = sk->sk_user_data;
Sage Weil31b80062009-10-06 11:31:13 -0700426
Alex Elder327800b2012-05-22 11:41:43 -0500427 dout("%s %p state = %lu sk_state = %u\n", __func__,
Sage Weil31b80062009-10-06 11:31:13 -0700428 con, con->state, sk->sk_state);
429
Sage Weil31b80062009-10-06 11:31:13 -0700430 switch (sk->sk_state) {
431 case TCP_CLOSE:
Alex Elder327800b2012-05-22 11:41:43 -0500432 dout("%s TCP_CLOSE\n", __func__);
Sage Weil31b80062009-10-06 11:31:13 -0700433 case TCP_CLOSE_WAIT:
Alex Elder327800b2012-05-22 11:41:43 -0500434 dout("%s TCP_CLOSE_WAIT\n", __func__);
Alex Elderce2c8902012-05-22 22:15:49 -0500435 con_sock_state_closing(con);
Alex Elderc9ffc772013-02-20 10:25:12 -0600436 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
Alex Elderd65c9e02012-06-20 21:53:53 -0500437 queue_con(con);
Sage Weil31b80062009-10-06 11:31:13 -0700438 break;
439 case TCP_ESTABLISHED:
Alex Elder327800b2012-05-22 11:41:43 -0500440 dout("%s TCP_ESTABLISHED\n", __func__);
Alex Elderce2c8902012-05-22 22:15:49 -0500441 con_sock_state_connected(con);
Sage Weil31b80062009-10-06 11:31:13 -0700442 queue_con(con);
443 break;
Alex Elderd3002b92012-02-14 14:05:33 -0600444 default: /* Everything else is uninteresting */
445 break;
Sage Weil31b80062009-10-06 11:31:13 -0700446 }
447}
448
449/*
450 * set up socket callbacks
451 */
452static void set_sock_callbacks(struct socket *sock,
453 struct ceph_connection *con)
454{
455 struct sock *sk = sock->sk;
Alex Elderbd406142012-01-23 15:49:27 -0600456 sk->sk_user_data = con;
Alex Elder327800b2012-05-22 11:41:43 -0500457 sk->sk_data_ready = ceph_sock_data_ready;
458 sk->sk_write_space = ceph_sock_write_space;
459 sk->sk_state_change = ceph_sock_state_change;
Sage Weil31b80062009-10-06 11:31:13 -0700460}
461
462
463/*
464 * socket helpers
465 */
466
467/*
468 * initiate connection to a remote socket.
469 */
Alex Elder41617d02012-02-14 14:05:33 -0600470static int ceph_tcp_connect(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -0700471{
Sage Weilf91d3472010-07-01 15:18:31 -0700472 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
Sage Weil31b80062009-10-06 11:31:13 -0700473 struct socket *sock;
474 int ret;
475
476 BUG_ON(con->sock);
Sage Weilf91d3472010-07-01 15:18:31 -0700477 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
478 IPPROTO_TCP, &sock);
Sage Weil31b80062009-10-06 11:31:13 -0700479 if (ret)
Alex Elder41617d02012-02-14 14:05:33 -0600480 return ret;
Sage Weil31b80062009-10-06 11:31:13 -0700481 sock->sk->sk_allocation = GFP_NOFS;
482
Sage Weila6a53492010-04-13 14:07:07 -0700483#ifdef CONFIG_LOCKDEP
484 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
485#endif
486
Sage Weil31b80062009-10-06 11:31:13 -0700487 set_sock_callbacks(sock, con);
488
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700489 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -0700490
Sage Weil89a86be2012-06-09 14:19:21 -0700491 con_sock_state_connecting(con);
Sage Weilf91d3472010-07-01 15:18:31 -0700492 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
493 O_NONBLOCK);
Sage Weil31b80062009-10-06 11:31:13 -0700494 if (ret == -EINPROGRESS) {
495 dout("connect %s EINPROGRESS sk_state = %u\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700496 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil31b80062009-10-06 11:31:13 -0700497 sock->sk->sk_state);
Alex Eldera5bc3129a2012-01-23 15:49:27 -0600498 } else if (ret < 0) {
Sage Weil31b80062009-10-06 11:31:13 -0700499 pr_err("connect %s error %d\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700500 ceph_pr_addr(&con->peer_addr.in_addr), ret);
Sage Weil31b80062009-10-06 11:31:13 -0700501 sock_release(sock);
Sage Weil31b80062009-10-06 11:31:13 -0700502 con->error_msg = "connect error";
Sage Weil31b80062009-10-06 11:31:13 -0700503
Alex Elder41617d02012-02-14 14:05:33 -0600504 return ret;
Alex Eldera5bc3129a2012-01-23 15:49:27 -0600505 }
506 con->sock = sock;
Alex Elder41617d02012-02-14 14:05:33 -0600507 return 0;
Sage Weil31b80062009-10-06 11:31:13 -0700508}
509
510static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
511{
512 struct kvec iov = {buf, len};
513 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
Sage Weil98bdb0a2011-01-25 08:17:48 -0800514 int r;
Sage Weil31b80062009-10-06 11:31:13 -0700515
Sage Weil98bdb0a2011-01-25 08:17:48 -0800516 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
517 if (r == -EAGAIN)
518 r = 0;
519 return r;
Sage Weil31b80062009-10-06 11:31:13 -0700520}
521
Alex Elderafb3d902013-03-08 20:58:59 -0600522static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
523 int page_offset, size_t length)
524{
525 void *kaddr;
526 int ret;
527
528 BUG_ON(page_offset + length > PAGE_SIZE);
529
530 kaddr = kmap(page);
531 BUG_ON(!kaddr);
532 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
533 kunmap(page);
534
535 return ret;
536}
537
Sage Weil31b80062009-10-06 11:31:13 -0700538/*
539 * write something. @more is true if caller will be sending more data
540 * shortly.
541 */
542static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
543 size_t kvlen, size_t len, int more)
544{
545 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
Sage Weil42961d22011-01-25 08:19:34 -0800546 int r;
Sage Weil31b80062009-10-06 11:31:13 -0700547
548 if (more)
549 msg.msg_flags |= MSG_MORE;
550 else
551 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
552
Sage Weil42961d22011-01-25 08:19:34 -0800553 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
554 if (r == -EAGAIN)
555 r = 0;
556 return r;
Sage Weil31b80062009-10-06 11:31:13 -0700557}
558
Alex Elder31739132012-03-07 11:40:08 -0600559static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
Alex Eldere1dcb122013-03-06 23:39:38 -0600560 int offset, size_t size, bool more)
Alex Elder31739132012-03-07 11:40:08 -0600561{
562 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
563 int ret;
564
565 ret = kernel_sendpage(sock, page, offset, size, flags);
566 if (ret == -EAGAIN)
567 ret = 0;
568
569 return ret;
570}
571
Sage Weil31b80062009-10-06 11:31:13 -0700572
573/*
574 * Shutdown/close the socket for the given connection.
575 */
576static int con_close_socket(struct ceph_connection *con)
577{
Sage Weil8007b8d2012-07-30 18:16:16 -0700578 int rc = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700579
580 dout("con_close_socket on %p sock %p\n", con, con->sock);
Sage Weil8007b8d2012-07-30 18:16:16 -0700581 if (con->sock) {
582 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
583 sock_release(con->sock);
584 con->sock = NULL;
585 }
Alex Elder456ea462012-06-20 21:53:53 -0500586
587 /*
Sage Weil4a861692012-07-20 17:29:55 -0700588 * Forcibly clear the SOCK_CLOSED flag. It gets set
Alex Elder456ea462012-06-20 21:53:53 -0500589 * independent of the connection mutex, and we could have
590 * received a socket close event before we had the chance to
591 * shut the socket down.
592 */
Alex Elderc9ffc772013-02-20 10:25:12 -0600593 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
Sage Weil8007b8d2012-07-30 18:16:16 -0700594
Alex Elderce2c8902012-05-22 22:15:49 -0500595 con_sock_state_closed(con);
Sage Weil31b80062009-10-06 11:31:13 -0700596 return rc;
597}
598
599/*
600 * Reset a connection. Discard all incoming and outgoing messages
601 * and clear *_seq state.
602 */
603static void ceph_msg_remove(struct ceph_msg *msg)
604{
605 list_del_init(&msg->list_head);
Alex Elder38941f82012-06-01 14:56:43 -0500606 BUG_ON(msg->con == NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -0700607 msg->con->ops->put(msg->con);
Alex Elder38941f82012-06-01 14:56:43 -0500608 msg->con = NULL;
609
Sage Weil31b80062009-10-06 11:31:13 -0700610 ceph_msg_put(msg);
611}
612static void ceph_msg_remove_list(struct list_head *head)
613{
614 while (!list_empty(head)) {
615 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
616 list_head);
617 ceph_msg_remove(msg);
618 }
619}
620
621static void reset_connection(struct ceph_connection *con)
622{
623 /* reset connection, out_queue, msg_ and connect_seq */
624 /* discard existing out_queue and msg_seq */
Sage Weil0fa6ebc2012-12-27 20:27:04 -0600625 dout("reset_connection %p\n", con);
Sage Weil31b80062009-10-06 11:31:13 -0700626 ceph_msg_remove_list(&con->out_queue);
627 ceph_msg_remove_list(&con->out_sent);
628
Sage Weilcf3e5c42009-12-11 09:48:05 -0800629 if (con->in_msg) {
Alex Elder38941f82012-06-01 14:56:43 -0500630 BUG_ON(con->in_msg->con != con);
631 con->in_msg->con = NULL;
Sage Weilcf3e5c42009-12-11 09:48:05 -0800632 ceph_msg_put(con->in_msg);
633 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -0700634 con->ops->put(con);
Sage Weilcf3e5c42009-12-11 09:48:05 -0800635 }
636
Sage Weil31b80062009-10-06 11:31:13 -0700637 con->connect_seq = 0;
638 con->out_seq = 0;
Sage Weilc86a2932009-12-14 14:04:30 -0800639 if (con->out_msg) {
640 ceph_msg_put(con->out_msg);
641 con->out_msg = NULL;
642 }
Sage Weil31b80062009-10-06 11:31:13 -0700643 con->in_seq = 0;
Sage Weil0e0d5e02010-04-02 16:07:19 -0700644 con->in_seq_acked = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700645}
646
647/*
648 * mark a peer down. drop any open connections.
649 */
650void ceph_con_close(struct ceph_connection *con)
651{
Sage Weil8c50c812012-07-30 16:24:37 -0700652 mutex_lock(&con->mutex);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700653 dout("con_close %p peer %s\n", con,
654 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil8dacc7d2012-07-20 17:24:40 -0700655 con->state = CON_STATE_CLOSED;
Alex Eldera5988c42012-05-29 11:04:58 -0500656
Alex Elderc9ffc772013-02-20 10:25:12 -0600657 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
658 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
659 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
660 con_flag_clear(con, CON_FLAG_BACKOFF);
Alex Eldera5988c42012-05-29 11:04:58 -0500661
Sage Weil31b80062009-10-06 11:31:13 -0700662 reset_connection(con);
Sage Weil6f2bc3f2010-04-02 16:16:34 -0700663 con->peer_global_seq = 0;
Sage Weil91e45ce32010-02-15 12:05:09 -0800664 cancel_delayed_work(&con->work);
Sage Weilee76e072012-07-20 16:45:49 -0700665 con_close_socket(con);
Sage Weilec302642009-12-22 10:43:42 -0800666 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700667}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700668EXPORT_SYMBOL(ceph_con_close);
Sage Weil31b80062009-10-06 11:31:13 -0700669
670/*
Sage Weil31b80062009-10-06 11:31:13 -0700671 * Reopen a closed connection, with a new peer address.
672 */
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700673void ceph_con_open(struct ceph_connection *con,
674 __u8 entity_type, __u64 entity_num,
675 struct ceph_entity_addr *addr)
Sage Weil31b80062009-10-06 11:31:13 -0700676{
Sage Weil54691552012-07-30 16:21:40 -0700677 mutex_lock(&con->mutex);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700678 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
Sage Weil8dacc7d2012-07-20 17:24:40 -0700679
Alex Elder122070a2012-12-26 10:43:57 -0600680 WARN_ON(con->state != CON_STATE_CLOSED);
Sage Weil8dacc7d2012-07-20 17:24:40 -0700681 con->state = CON_STATE_PREOPEN;
Alex Eldera5988c42012-05-29 11:04:58 -0500682
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700683 con->peer_name.type = (__u8) entity_type;
684 con->peer_name.num = cpu_to_le64(entity_num);
685
Sage Weil31b80062009-10-06 11:31:13 -0700686 memcpy(&con->peer_addr, addr, sizeof(*addr));
Sage Weil03c677e2009-11-20 15:14:15 -0800687 con->delay = 0; /* reset backoff memory */
Sage Weil54691552012-07-30 16:21:40 -0700688 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700689 queue_con(con);
690}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700691EXPORT_SYMBOL(ceph_con_open);
Sage Weil31b80062009-10-06 11:31:13 -0700692
693/*
Sage Weil87b315a2010-03-22 14:51:18 -0700694 * return true if this connection ever successfully opened
695 */
696bool ceph_con_opened(struct ceph_connection *con)
697{
698 return con->connect_seq > 0;
699}
700
701/*
Sage Weil31b80062009-10-06 11:31:13 -0700702 * initialize a new connection.
703 */
Alex Elder1bfd89f2012-05-26 23:26:43 -0500704void ceph_con_init(struct ceph_connection *con, void *private,
705 const struct ceph_connection_operations *ops,
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700706 struct ceph_messenger *msgr)
Sage Weil31b80062009-10-06 11:31:13 -0700707{
708 dout("con_init %p\n", con);
709 memset(con, 0, sizeof(*con));
Alex Elder1bfd89f2012-05-26 23:26:43 -0500710 con->private = private;
711 con->ops = ops;
Sage Weil31b80062009-10-06 11:31:13 -0700712 con->msgr = msgr;
Alex Elderce2c8902012-05-22 22:15:49 -0500713
714 con_sock_state_init(con);
715
Sage Weilec302642009-12-22 10:43:42 -0800716 mutex_init(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700717 INIT_LIST_HEAD(&con->out_queue);
718 INIT_LIST_HEAD(&con->out_sent);
719 INIT_DELAYED_WORK(&con->work, con_work);
Alex Eldera5988c42012-05-29 11:04:58 -0500720
Sage Weil8dacc7d2012-07-20 17:24:40 -0700721 con->state = CON_STATE_CLOSED;
Sage Weil31b80062009-10-06 11:31:13 -0700722}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700723EXPORT_SYMBOL(ceph_con_init);
Sage Weil31b80062009-10-06 11:31:13 -0700724
725
726/*
727 * We maintain a global counter to order connection attempts. Get
728 * a unique seq greater than @gt.
729 */
730static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
731{
732 u32 ret;
733
734 spin_lock(&msgr->global_seq_lock);
735 if (msgr->global_seq < gt)
736 msgr->global_seq = gt;
737 ret = ++msgr->global_seq;
738 spin_unlock(&msgr->global_seq_lock);
739 return ret;
740}
741
Alex Eldere2200422012-05-23 14:35:23 -0500742static void con_out_kvec_reset(struct ceph_connection *con)
Alex Elder859eb792012-02-14 14:05:33 -0600743{
744 con->out_kvec_left = 0;
745 con->out_kvec_bytes = 0;
746 con->out_kvec_cur = &con->out_kvec[0];
747}
748
Alex Eldere2200422012-05-23 14:35:23 -0500749static void con_out_kvec_add(struct ceph_connection *con,
Alex Elder859eb792012-02-14 14:05:33 -0600750 size_t size, void *data)
751{
752 int index;
753
754 index = con->out_kvec_left;
755 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
756
757 con->out_kvec[index].iov_len = size;
758 con->out_kvec[index].iov_base = data;
759 con->out_kvec_left++;
760 con->out_kvec_bytes += size;
761}
Sage Weil31b80062009-10-06 11:31:13 -0700762
Alex Elderdf6ad1f2012-06-11 14:57:13 -0500763#ifdef CONFIG_BLOCK
Alex Elder6aaa4512013-03-06 23:39:39 -0600764
765/*
766 * For a bio data item, a piece is whatever remains of the next
767 * entry in the current bio iovec, or the first entry in the next
768 * bio in the list.
769 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500770static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500771 size_t length)
Alex Elder6aaa4512013-03-06 23:39:39 -0600772{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500773 struct ceph_msg_data *data = cursor->data;
Alex Elder6aaa4512013-03-06 23:39:39 -0600774 struct bio *bio;
775
776 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
777
778 bio = data->bio;
779 BUG_ON(!bio);
780 BUG_ON(!bio->bi_vcnt);
Alex Elder6aaa4512013-03-06 23:39:39 -0600781
Alex Elderca8b3a62013-04-05 14:46:01 -0500782 cursor->resid = min(length, data->bio_length);
Alex Elder6aaa4512013-03-06 23:39:39 -0600783 cursor->bio = bio;
784 cursor->vector_index = 0;
785 cursor->vector_offset = 0;
Alex Elder25aff7c2013-03-11 23:34:22 -0500786 cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
Alex Elder6aaa4512013-03-06 23:39:39 -0600787}
788
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500789static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
Alex Elder6aaa4512013-03-06 23:39:39 -0600790 size_t *page_offset,
791 size_t *length)
792{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500793 struct ceph_msg_data *data = cursor->data;
Alex Elder6aaa4512013-03-06 23:39:39 -0600794 struct bio *bio;
795 struct bio_vec *bio_vec;
796 unsigned int index;
797
798 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
799
800 bio = cursor->bio;
801 BUG_ON(!bio);
802
803 index = cursor->vector_index;
804 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
805
806 bio_vec = &bio->bi_io_vec[index];
807 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
808 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
809 BUG_ON(*page_offset >= PAGE_SIZE);
Alex Elder25aff7c2013-03-11 23:34:22 -0500810 if (cursor->last_piece) /* pagelist offset is always 0 */
811 *length = cursor->resid;
812 else
813 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
Alex Elder25aff7c2013-03-11 23:34:22 -0500814 BUG_ON(*length > cursor->resid);
Alex Elder5df521b2013-03-30 15:09:59 -0500815 BUG_ON(*page_offset + *length > PAGE_SIZE);
Alex Elder6aaa4512013-03-06 23:39:39 -0600816
817 return bio_vec->bv_page;
818}
819
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500820static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
821 size_t bytes)
Alex Elder6aaa4512013-03-06 23:39:39 -0600822{
Alex Elder6aaa4512013-03-06 23:39:39 -0600823 struct bio *bio;
824 struct bio_vec *bio_vec;
825 unsigned int index;
826
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500827 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
Alex Elder6aaa4512013-03-06 23:39:39 -0600828
829 bio = cursor->bio;
830 BUG_ON(!bio);
831
832 index = cursor->vector_index;
833 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
834 bio_vec = &bio->bi_io_vec[index];
Alex Elder6aaa4512013-03-06 23:39:39 -0600835
836 /* Advance the cursor offset */
837
Alex Elder25aff7c2013-03-11 23:34:22 -0500838 BUG_ON(cursor->resid < bytes);
839 cursor->resid -= bytes;
Alex Elder6aaa4512013-03-06 23:39:39 -0600840 cursor->vector_offset += bytes;
841 if (cursor->vector_offset < bio_vec->bv_len)
842 return false; /* more bytes to process in this segment */
Alex Elder25aff7c2013-03-11 23:34:22 -0500843 BUG_ON(cursor->vector_offset != bio_vec->bv_len);
Alex Elder6aaa4512013-03-06 23:39:39 -0600844
845 /* Move on to the next segment, and possibly the next bio */
846
Alex Elder25aff7c2013-03-11 23:34:22 -0500847 if (++index == (unsigned int) bio->bi_vcnt) {
Alex Elder6aaa4512013-03-06 23:39:39 -0600848 bio = bio->bi_next;
Alex Elder25aff7c2013-03-11 23:34:22 -0500849 index = 0;
Alex Elder6aaa4512013-03-06 23:39:39 -0600850 }
Alex Elder25aff7c2013-03-11 23:34:22 -0500851 cursor->bio = bio;
852 cursor->vector_index = index;
Alex Elder6aaa4512013-03-06 23:39:39 -0600853 cursor->vector_offset = 0;
854
Alex Elder25aff7c2013-03-11 23:34:22 -0500855 if (!cursor->last_piece) {
856 BUG_ON(!cursor->resid);
857 BUG_ON(!bio);
858 /* A short read is OK, so use <= rather than == */
859 if (cursor->resid <= bio->bi_io_vec[index].bv_len)
Alex Elder6aaa4512013-03-06 23:39:39 -0600860 cursor->last_piece = true;
Alex Elder25aff7c2013-03-11 23:34:22 -0500861 }
Alex Elder6aaa4512013-03-06 23:39:39 -0600862
863 return true;
864}
Alex Elderea965712013-04-05 14:46:01 -0500865#endif /* CONFIG_BLOCK */
Alex Elderdf6ad1f2012-06-11 14:57:13 -0500866
Alex Elderfe38a2b2013-03-06 23:39:39 -0600867/*
Alex Eldere766d7b2013-03-07 15:38:28 -0600868 * For a page array, a piece comes from the first page in the array
869 * that has not already been fully consumed.
870 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500871static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500872 size_t length)
Alex Eldere766d7b2013-03-07 15:38:28 -0600873{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500874 struct ceph_msg_data *data = cursor->data;
Alex Eldere766d7b2013-03-07 15:38:28 -0600875 int page_count;
876
877 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
878
879 BUG_ON(!data->pages);
880 BUG_ON(!data->length);
881
Alex Elderca8b3a62013-04-05 14:46:01 -0500882 cursor->resid = min(length, data->length);
Alex Eldere766d7b2013-03-07 15:38:28 -0600883 page_count = calc_pages_for(data->alignment, (u64)data->length);
Alex Eldere766d7b2013-03-07 15:38:28 -0600884 cursor->page_offset = data->alignment & ~PAGE_MASK;
885 cursor->page_index = 0;
Alex Elder56fc5652013-03-30 23:46:55 -0500886 BUG_ON(page_count > (int)USHRT_MAX);
887 cursor->page_count = (unsigned short)page_count;
888 BUG_ON(length > SIZE_MAX - cursor->page_offset);
889 cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE;
Alex Eldere766d7b2013-03-07 15:38:28 -0600890}
891
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500892static struct page *
893ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
894 size_t *page_offset, size_t *length)
Alex Eldere766d7b2013-03-07 15:38:28 -0600895{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500896 struct ceph_msg_data *data = cursor->data;
Alex Eldere766d7b2013-03-07 15:38:28 -0600897
898 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
899
900 BUG_ON(cursor->page_index >= cursor->page_count);
901 BUG_ON(cursor->page_offset >= PAGE_SIZE);
Alex Eldere766d7b2013-03-07 15:38:28 -0600902
903 *page_offset = cursor->page_offset;
Alex Elder25aff7c2013-03-11 23:34:22 -0500904 if (cursor->last_piece)
Alex Eldere766d7b2013-03-07 15:38:28 -0600905 *length = cursor->resid;
Alex Elder25aff7c2013-03-11 23:34:22 -0500906 else
Alex Eldere766d7b2013-03-07 15:38:28 -0600907 *length = PAGE_SIZE - *page_offset;
Alex Eldere766d7b2013-03-07 15:38:28 -0600908
909 return data->pages[cursor->page_index];
910}
911
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500912static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
Alex Eldere766d7b2013-03-07 15:38:28 -0600913 size_t bytes)
914{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500915 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
Alex Eldere766d7b2013-03-07 15:38:28 -0600916
917 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
Alex Eldere766d7b2013-03-07 15:38:28 -0600918
919 /* Advance the cursor page offset */
920
921 cursor->resid -= bytes;
Alex Elder5df521b2013-03-30 15:09:59 -0500922 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
923 if (!bytes || cursor->page_offset)
Alex Eldere766d7b2013-03-07 15:38:28 -0600924 return false; /* more bytes to process in the current page */
925
Alex Elder5df521b2013-03-30 15:09:59 -0500926 /* Move on to the next page; offset is already at 0 */
Alex Eldere766d7b2013-03-07 15:38:28 -0600927
928 BUG_ON(cursor->page_index >= cursor->page_count);
Alex Eldere766d7b2013-03-07 15:38:28 -0600929 cursor->page_index++;
Alex Elder25aff7c2013-03-11 23:34:22 -0500930 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Eldere766d7b2013-03-07 15:38:28 -0600931
932 return true;
933}
934
935/*
Alex Elderdd236fc2013-03-06 23:39:39 -0600936 * For a pagelist, a piece is whatever remains to be consumed in the
937 * first page in the list, or the front of the next page.
Alex Elderfe38a2b2013-03-06 23:39:39 -0600938 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500939static void
940ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500941 size_t length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600942{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500943 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600944 struct ceph_pagelist *pagelist;
945 struct page *page;
946
Alex Elderdd236fc2013-03-06 23:39:39 -0600947 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600948
949 pagelist = data->pagelist;
950 BUG_ON(!pagelist);
Alex Elder25aff7c2013-03-11 23:34:22 -0500951
952 if (!length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600953 return; /* pagelist can be assigned but empty */
954
955 BUG_ON(list_empty(&pagelist->head));
956 page = list_first_entry(&pagelist->head, struct page, lru);
957
Alex Elderca8b3a62013-04-05 14:46:01 -0500958 cursor->resid = min(length, pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600959 cursor->page = page;
960 cursor->offset = 0;
Alex Eldera51b272e2013-04-19 15:34:49 -0500961 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600962}
963
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500964static struct page *
965ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
966 size_t *page_offset, size_t *length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600967{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500968 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600969 struct ceph_pagelist *pagelist;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600970
971 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
972
973 pagelist = data->pagelist;
974 BUG_ON(!pagelist);
975
976 BUG_ON(!cursor->page);
Alex Elder25aff7c2013-03-11 23:34:22 -0500977 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600978
Alex Elder5df521b2013-03-30 15:09:59 -0500979 /* offset of first page in pagelist is always 0 */
Alex Elderfe38a2b2013-03-06 23:39:39 -0600980 *page_offset = cursor->offset & ~PAGE_MASK;
Alex Elder5df521b2013-03-30 15:09:59 -0500981 if (cursor->last_piece)
Alex Elder25aff7c2013-03-11 23:34:22 -0500982 *length = cursor->resid;
983 else
984 *length = PAGE_SIZE - *page_offset;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600985
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500986 return cursor->page;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600987}
988
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500989static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
Alex Elderdd236fc2013-03-06 23:39:39 -0600990 size_t bytes)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600991{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500992 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600993 struct ceph_pagelist *pagelist;
994
995 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
996
997 pagelist = data->pagelist;
998 BUG_ON(!pagelist);
Alex Elder25aff7c2013-03-11 23:34:22 -0500999
1000 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -06001001 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1002
1003 /* Advance the cursor offset */
1004
Alex Elder25aff7c2013-03-11 23:34:22 -05001005 cursor->resid -= bytes;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001006 cursor->offset += bytes;
Alex Elder5df521b2013-03-30 15:09:59 -05001007 /* offset of first page in pagelist is always 0 */
Alex Elderfe38a2b2013-03-06 23:39:39 -06001008 if (!bytes || cursor->offset & ~PAGE_MASK)
1009 return false; /* more bytes to process in the current page */
1010
1011 /* Move on to the next page */
1012
1013 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1014 cursor->page = list_entry_next(cursor->page, lru);
Alex Elder25aff7c2013-03-11 23:34:22 -05001015 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001016
1017 return true;
1018}
1019
Alex Elderdd236fc2013-03-06 23:39:39 -06001020/*
1021 * Message data is handled (sent or received) in pieces, where each
1022 * piece resides on a single page. The network layer might not
1023 * consume an entire piece at once. A data item's cursor keeps
1024 * track of which piece is next to process and how much remains to
1025 * be processed in that piece. It also tracks whether the current
1026 * piece is the last one in the data item.
1027 */
Alex Elderca8b3a62013-04-05 14:46:01 -05001028static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
Alex Elderdd236fc2013-03-06 23:39:39 -06001029{
Alex Elderca8b3a62013-04-05 14:46:01 -05001030 size_t length = cursor->total_resid;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001031
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001032 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001033 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001034 ceph_msg_data_pagelist_cursor_init(cursor, length);
Alex Elderdd236fc2013-03-06 23:39:39 -06001035 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001036 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001037 ceph_msg_data_pages_cursor_init(cursor, length);
Alex Eldere766d7b2013-03-07 15:38:28 -06001038 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001039#ifdef CONFIG_BLOCK
1040 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001041 ceph_msg_data_bio_cursor_init(cursor, length);
Alex Elder6aaa4512013-03-06 23:39:39 -06001042 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001043#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001044 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001045 default:
1046 /* BUG(); */
1047 break;
1048 }
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001049 cursor->need_crc = true;
Alex Elderdd236fc2013-03-06 23:39:39 -06001050}
1051
Alex Elderca8b3a62013-04-05 14:46:01 -05001052static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1053{
1054 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1055 struct ceph_msg_data *data;
1056
1057 BUG_ON(!length);
1058 BUG_ON(length > msg->data_length);
1059 BUG_ON(list_empty(&msg->data));
1060
Alex Elderca8b3a62013-04-05 14:46:01 -05001061 cursor->data_head = &msg->data;
1062 cursor->total_resid = length;
1063 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1064 cursor->data = data;
1065
1066 __ceph_msg_data_cursor_init(cursor);
1067}
1068
Alex Elderdd236fc2013-03-06 23:39:39 -06001069/*
1070 * Return the page containing the next piece to process for a given
1071 * data item, and supply the page offset and length of that piece.
1072 * Indicate whether this is the last piece in this data item.
1073 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001074static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1075 size_t *page_offset, size_t *length,
Alex Elderdd236fc2013-03-06 23:39:39 -06001076 bool *last_piece)
1077{
1078 struct page *page;
1079
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001080 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001081 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001082 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
Alex Elderdd236fc2013-03-06 23:39:39 -06001083 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001084 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001085 page = ceph_msg_data_pages_next(cursor, page_offset, length);
Alex Eldere766d7b2013-03-07 15:38:28 -06001086 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001087#ifdef CONFIG_BLOCK
1088 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001089 page = ceph_msg_data_bio_next(cursor, page_offset, length);
Alex Elder6aaa4512013-03-06 23:39:39 -06001090 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001091#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001092 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001093 default:
1094 page = NULL;
1095 break;
1096 }
1097 BUG_ON(!page);
1098 BUG_ON(*page_offset + *length > PAGE_SIZE);
1099 BUG_ON(!*length);
1100 if (last_piece)
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001101 *last_piece = cursor->last_piece;
Alex Elderdd236fc2013-03-06 23:39:39 -06001102
1103 return page;
1104}
1105
1106/*
1107 * Returns true if the result moves the cursor on to the next piece
1108 * of the data item.
1109 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001110static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1111 size_t bytes)
Alex Elderdd236fc2013-03-06 23:39:39 -06001112{
1113 bool new_piece;
1114
Alex Elder25aff7c2013-03-11 23:34:22 -05001115 BUG_ON(bytes > cursor->resid);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001116 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001117 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001118 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
Alex Elderdd236fc2013-03-06 23:39:39 -06001119 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001120 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001121 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
Alex Eldere766d7b2013-03-07 15:38:28 -06001122 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001123#ifdef CONFIG_BLOCK
1124 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001125 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
Alex Elder6aaa4512013-03-06 23:39:39 -06001126 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001127#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001128 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001129 default:
1130 BUG();
1131 break;
1132 }
Alex Elderca8b3a62013-04-05 14:46:01 -05001133 cursor->total_resid -= bytes;
Alex Elderdd236fc2013-03-06 23:39:39 -06001134
Alex Elderca8b3a62013-04-05 14:46:01 -05001135 if (!cursor->resid && cursor->total_resid) {
1136 WARN_ON(!cursor->last_piece);
1137 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1138 cursor->data = list_entry_next(cursor->data, links);
1139 __ceph_msg_data_cursor_init(cursor);
Alex Eldera51b272e2013-04-19 15:34:49 -05001140 new_piece = true;
Alex Elderca8b3a62013-04-05 14:46:01 -05001141 }
Alex Eldera51b272e2013-04-19 15:34:49 -05001142 cursor->need_crc = new_piece;
Alex Elderca8b3a62013-04-05 14:46:01 -05001143
Alex Elderdd236fc2013-03-06 23:39:39 -06001144 return new_piece;
1145}
1146
Alex Elder98fa5dd2013-04-02 12:09:50 -05001147static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
Alex Elder739c9052012-06-11 14:57:13 -05001148{
Alex Elder739c9052012-06-11 14:57:13 -05001149 BUG_ON(!msg);
Alex Elder25aff7c2013-03-11 23:34:22 -05001150 BUG_ON(!data_len);
Alex Elder739c9052012-06-11 14:57:13 -05001151
Alex Elder4c59b4a2013-03-11 23:34:23 -05001152 /* Initialize data cursor */
Alex Elderfe38a2b2013-03-06 23:39:39 -06001153
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001154 ceph_msg_data_cursor_init(msg, (size_t)data_len);
Alex Elder739c9052012-06-11 14:57:13 -05001155}
1156
Sage Weil31b80062009-10-06 11:31:13 -07001157/*
1158 * Prepare footer for currently outgoing message, and finish things
1159 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1160 */
Alex Elder859eb792012-02-14 14:05:33 -06001161static void prepare_write_message_footer(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001162{
1163 struct ceph_msg *m = con->out_msg;
Alex Elder859eb792012-02-14 14:05:33 -06001164 int v = con->out_kvec_left;
Sage Weil31b80062009-10-06 11:31:13 -07001165
Alex Elderfd154f32012-06-11 14:57:13 -05001166 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1167
Sage Weil31b80062009-10-06 11:31:13 -07001168 dout("prepare_write_message_footer %p\n", con);
1169 con->out_kvec_is_msg = true;
1170 con->out_kvec[v].iov_base = &m->footer;
1171 con->out_kvec[v].iov_len = sizeof(m->footer);
1172 con->out_kvec_bytes += sizeof(m->footer);
1173 con->out_kvec_left++;
1174 con->out_more = m->more_to_follow;
Sage Weilc86a2932009-12-14 14:04:30 -08001175 con->out_msg_done = true;
Sage Weil31b80062009-10-06 11:31:13 -07001176}
1177
1178/*
1179 * Prepare headers for the next outgoing message.
1180 */
1181static void prepare_write_message(struct ceph_connection *con)
1182{
1183 struct ceph_msg *m;
Alex Eldera9a0c512012-02-15 07:43:54 -06001184 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07001185
Alex Eldere2200422012-05-23 14:35:23 -05001186 con_out_kvec_reset(con);
Sage Weil31b80062009-10-06 11:31:13 -07001187 con->out_kvec_is_msg = true;
Sage Weilc86a2932009-12-14 14:04:30 -08001188 con->out_msg_done = false;
Sage Weil31b80062009-10-06 11:31:13 -07001189
1190 /* Sneak an ack in there first? If we can get it into the same
1191 * TCP packet that's a good thing. */
1192 if (con->in_seq > con->in_seq_acked) {
1193 con->in_seq_acked = con->in_seq;
Alex Eldere2200422012-05-23 14:35:23 -05001194 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
Sage Weil31b80062009-10-06 11:31:13 -07001195 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
Alex Eldere2200422012-05-23 14:35:23 -05001196 con_out_kvec_add(con, sizeof (con->out_temp_ack),
Alex Elder859eb792012-02-14 14:05:33 -06001197 &con->out_temp_ack);
Sage Weil31b80062009-10-06 11:31:13 -07001198 }
1199
Alex Elder38941f82012-06-01 14:56:43 -05001200 BUG_ON(list_empty(&con->out_queue));
Alex Elder859eb792012-02-14 14:05:33 -06001201 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
Sage Weilc86a2932009-12-14 14:04:30 -08001202 con->out_msg = m;
Alex Elder38941f82012-06-01 14:56:43 -05001203 BUG_ON(m->con != con);
Sage Weil4cf9d542011-07-26 11:27:24 -07001204
1205 /* put message on sent list */
1206 ceph_msg_get(m);
1207 list_move_tail(&m->list_head, &con->out_sent);
Sage Weil31b80062009-10-06 11:31:13 -07001208
Sage Weile84346b2010-05-11 21:20:38 -07001209 /*
1210 * only assign outgoing seq # if we haven't sent this message
1211 * yet. if it is requeued, resend with it's original seq.
1212 */
1213 if (m->needs_out_seq) {
1214 m->hdr.seq = cpu_to_le64(++con->out_seq);
1215 m->needs_out_seq = false;
1216 }
Alex Elder98fa5dd2013-04-02 12:09:50 -05001217 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
Sage Weil31b80062009-10-06 11:31:13 -07001218
Alex Elder98fa5dd2013-04-02 12:09:50 -05001219 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
Sage Weil31b80062009-10-06 11:31:13 -07001220 m, con->out_seq, le16_to_cpu(m->hdr.type),
1221 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
Alex Elder98fa5dd2013-04-02 12:09:50 -05001222 m->data_length);
Sage Weil31b80062009-10-06 11:31:13 -07001223 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1224
1225 /* tag + hdr + front + middle */
Alex Eldere2200422012-05-23 14:35:23 -05001226 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1227 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1228 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
Alex Elder859eb792012-02-14 14:05:33 -06001229
Sage Weil31b80062009-10-06 11:31:13 -07001230 if (m->middle)
Alex Eldere2200422012-05-23 14:35:23 -05001231 con_out_kvec_add(con, m->middle->vec.iov_len,
Alex Elder859eb792012-02-14 14:05:33 -06001232 m->middle->vec.iov_base);
Sage Weil31b80062009-10-06 11:31:13 -07001233
1234 /* fill in crc (except data pages), footer */
Alex Eldera9a0c512012-02-15 07:43:54 -06001235 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1236 con->out_msg->hdr.crc = cpu_to_le32(crc);
Alex Elderfd154f32012-06-11 14:57:13 -05001237 con->out_msg->footer.flags = 0;
Alex Eldera9a0c512012-02-15 07:43:54 -06001238
1239 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1240 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1241 if (m->middle) {
1242 crc = crc32c(0, m->middle->vec.iov_base,
1243 m->middle->vec.iov_len);
1244 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1245 } else
Sage Weil31b80062009-10-06 11:31:13 -07001246 con->out_msg->footer.middle_crc = 0;
Alex Elder739c9052012-06-11 14:57:13 -05001247 dout("%s front_crc %u middle_crc %u\n", __func__,
Sage Weil31b80062009-10-06 11:31:13 -07001248 le32_to_cpu(con->out_msg->footer.front_crc),
1249 le32_to_cpu(con->out_msg->footer.middle_crc));
1250
1251 /* is there a data payload? */
Alex Elder739c9052012-06-11 14:57:13 -05001252 con->out_msg->footer.data_crc = 0;
Alex Elder98fa5dd2013-04-02 12:09:50 -05001253 if (m->data_length) {
1254 prepare_message_data(con->out_msg, m->data_length);
Alex Elder78625052013-03-06 23:39:39 -06001255 con->out_more = 1; /* data + footer will follow */
1256 } else {
Sage Weil31b80062009-10-06 11:31:13 -07001257 /* no, queue up footer too and be done */
Alex Elder859eb792012-02-14 14:05:33 -06001258 prepare_write_message_footer(con);
Alex Elder78625052013-03-06 23:39:39 -06001259 }
Sage Weil31b80062009-10-06 11:31:13 -07001260
Alex Elderc9ffc772013-02-20 10:25:12 -06001261 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001262}
1263
1264/*
1265 * Prepare an ack.
1266 */
1267static void prepare_write_ack(struct ceph_connection *con)
1268{
1269 dout("prepare_write_ack %p %llu -> %llu\n", con,
1270 con->in_seq_acked, con->in_seq);
1271 con->in_seq_acked = con->in_seq;
1272
Alex Eldere2200422012-05-23 14:35:23 -05001273 con_out_kvec_reset(con);
Alex Elder859eb792012-02-14 14:05:33 -06001274
Alex Eldere2200422012-05-23 14:35:23 -05001275 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
Alex Elder859eb792012-02-14 14:05:33 -06001276
Sage Weil31b80062009-10-06 11:31:13 -07001277 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
Alex Eldere2200422012-05-23 14:35:23 -05001278 con_out_kvec_add(con, sizeof (con->out_temp_ack),
Alex Elder859eb792012-02-14 14:05:33 -06001279 &con->out_temp_ack);
1280
Sage Weil31b80062009-10-06 11:31:13 -07001281 con->out_more = 1; /* more will follow.. eventually.. */
Alex Elderc9ffc772013-02-20 10:25:12 -06001282 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001283}
1284
1285/*
Sage Weil3a230832013-03-25 08:47:40 -07001286 * Prepare to share the seq during handshake
1287 */
1288static void prepare_write_seq(struct ceph_connection *con)
1289{
1290 dout("prepare_write_seq %p %llu -> %llu\n", con,
1291 con->in_seq_acked, con->in_seq);
1292 con->in_seq_acked = con->in_seq;
1293
1294 con_out_kvec_reset(con);
1295
1296 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1297 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1298 &con->out_temp_ack);
1299
1300 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1301}
1302
1303/*
Sage Weil31b80062009-10-06 11:31:13 -07001304 * Prepare to write keepalive byte.
1305 */
1306static void prepare_write_keepalive(struct ceph_connection *con)
1307{
1308 dout("prepare_write_keepalive %p\n", con);
Alex Eldere2200422012-05-23 14:35:23 -05001309 con_out_kvec_reset(con);
1310 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
Alex Elderc9ffc772013-02-20 10:25:12 -06001311 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001312}
1313
1314/*
1315 * Connection negotiation.
1316 */
1317
Alex Elderdac1e712012-05-16 15:16:39 -05001318static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1319 int *auth_proto)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001320{
Alex Eldera3530df2012-05-16 15:16:39 -05001321 struct ceph_auth_handshake *auth;
Alex Elderb1c6b982012-05-16 15:16:38 -05001322
1323 if (!con->ops->get_authorizer) {
1324 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1325 con->out_connect.authorizer_len = 0;
Alex Elder729796b2012-05-16 15:16:39 -05001326 return NULL;
Alex Elderb1c6b982012-05-16 15:16:38 -05001327 }
1328
1329 /* Can't hold the mutex while getting authorizer */
Sage Weilec302642009-12-22 10:43:42 -08001330 mutex_unlock(&con->mutex);
Alex Elderdac1e712012-05-16 15:16:39 -05001331 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
Sage Weilec302642009-12-22 10:43:42 -08001332 mutex_lock(&con->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001333
Alex Eldera3530df2012-05-16 15:16:39 -05001334 if (IS_ERR(auth))
Alex Elder729796b2012-05-16 15:16:39 -05001335 return auth;
Sage Weil8dacc7d2012-07-20 17:24:40 -07001336 if (con->state != CON_STATE_NEGOTIATING)
Alex Elder729796b2012-05-16 15:16:39 -05001337 return ERR_PTR(-EAGAIN);
Sage Weil0da5d702011-05-19 11:21:05 -07001338
Alex Elder8f43fb52012-05-16 15:16:39 -05001339 con->auth_reply_buf = auth->authorizer_reply_buf;
1340 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
Alex Elder729796b2012-05-16 15:16:39 -05001341 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001342}
1343
Sage Weil31b80062009-10-06 11:31:13 -07001344/*
1345 * We connected to a peer and are saying hello.
1346 */
Alex Eldere825a662012-05-16 15:16:38 -05001347static void prepare_write_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001348{
Alex Eldere2200422012-05-23 14:35:23 -05001349 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1350 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
Alex Eldere825a662012-05-16 15:16:38 -05001351 &con->msgr->my_enc_addr);
Sage Weileed0ef22009-11-10 14:34:36 -08001352
Sage Weileed0ef22009-11-10 14:34:36 -08001353 con->out_more = 0;
Alex Elderc9ffc772013-02-20 10:25:12 -06001354 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weileed0ef22009-11-10 14:34:36 -08001355}
1356
Alex Eldere825a662012-05-16 15:16:38 -05001357static int prepare_write_connect(struct ceph_connection *con)
Sage Weileed0ef22009-11-10 14:34:36 -08001358{
Eric Dumazet95c96172012-04-15 05:58:06 +00001359 unsigned int global_seq = get_global_seq(con->msgr, 0);
Sage Weil31b80062009-10-06 11:31:13 -07001360 int proto;
Alex Elderdac1e712012-05-16 15:16:39 -05001361 int auth_proto;
Alex Elder729796b2012-05-16 15:16:39 -05001362 struct ceph_auth_handshake *auth;
Sage Weil31b80062009-10-06 11:31:13 -07001363
1364 switch (con->peer_name.type) {
1365 case CEPH_ENTITY_TYPE_MON:
1366 proto = CEPH_MONC_PROTOCOL;
1367 break;
1368 case CEPH_ENTITY_TYPE_OSD:
1369 proto = CEPH_OSDC_PROTOCOL;
1370 break;
1371 case CEPH_ENTITY_TYPE_MDS:
1372 proto = CEPH_MDSC_PROTOCOL;
1373 break;
1374 default:
1375 BUG();
1376 }
1377
1378 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1379 con->connect_seq, global_seq, proto);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001380
Alex Eldere825a662012-05-16 15:16:38 -05001381 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
Sage Weil31b80062009-10-06 11:31:13 -07001382 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1383 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1384 con->out_connect.global_seq = cpu_to_le32(global_seq);
1385 con->out_connect.protocol_version = cpu_to_le32(proto);
1386 con->out_connect.flags = 0;
Sage Weil31b80062009-10-06 11:31:13 -07001387
Alex Elderdac1e712012-05-16 15:16:39 -05001388 auth_proto = CEPH_AUTH_UNKNOWN;
1389 auth = get_connect_authorizer(con, &auth_proto);
Alex Elder729796b2012-05-16 15:16:39 -05001390 if (IS_ERR(auth))
1391 return PTR_ERR(auth);
Alex Elder3da54772012-05-16 15:16:39 -05001392
Alex Elderdac1e712012-05-16 15:16:39 -05001393 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
Alex Elder3da54772012-05-16 15:16:39 -05001394 con->out_connect.authorizer_len = auth ?
1395 cpu_to_le32(auth->authorizer_buf_len) : 0;
1396
Alex Eldere2200422012-05-23 14:35:23 -05001397 con_out_kvec_add(con, sizeof (con->out_connect),
Alex Elder3da54772012-05-16 15:16:39 -05001398 &con->out_connect);
1399 if (auth && auth->authorizer_buf_len)
Alex Eldere2200422012-05-23 14:35:23 -05001400 con_out_kvec_add(con, auth->authorizer_buf_len,
Alex Elder3da54772012-05-16 15:16:39 -05001401 auth->authorizer_buf);
Alex Elder859eb792012-02-14 14:05:33 -06001402
Sage Weil31b80062009-10-06 11:31:13 -07001403 con->out_more = 0;
Alex Elderc9ffc772013-02-20 10:25:12 -06001404 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001405
Alex Eldere10c7582012-05-16 15:16:38 -05001406 return 0;
Sage Weil31b80062009-10-06 11:31:13 -07001407}
1408
Sage Weil31b80062009-10-06 11:31:13 -07001409/*
1410 * write as much of pending kvecs to the socket as we can.
1411 * 1 -> done
1412 * 0 -> socket full, but more to do
1413 * <0 -> error
1414 */
1415static int write_partial_kvec(struct ceph_connection *con)
1416{
1417 int ret;
1418
1419 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1420 while (con->out_kvec_bytes > 0) {
1421 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1422 con->out_kvec_left, con->out_kvec_bytes,
1423 con->out_more);
1424 if (ret <= 0)
1425 goto out;
1426 con->out_kvec_bytes -= ret;
1427 if (con->out_kvec_bytes == 0)
1428 break; /* done */
Alex Elderf42299e2012-02-15 07:43:54 -06001429
1430 /* account for full iov entries consumed */
1431 while (ret >= con->out_kvec_cur->iov_len) {
1432 BUG_ON(!con->out_kvec_left);
1433 ret -= con->out_kvec_cur->iov_len;
1434 con->out_kvec_cur++;
1435 con->out_kvec_left--;
1436 }
1437 /* and for a partially-consumed entry */
1438 if (ret) {
1439 con->out_kvec_cur->iov_len -= ret;
1440 con->out_kvec_cur->iov_base += ret;
Sage Weil31b80062009-10-06 11:31:13 -07001441 }
1442 }
1443 con->out_kvec_left = 0;
1444 con->out_kvec_is_msg = false;
1445 ret = 1;
1446out:
1447 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1448 con->out_kvec_bytes, con->out_kvec_left, ret);
1449 return ret; /* done! */
1450}
1451
Alex Elder35b62802013-03-08 20:59:00 -06001452static u32 ceph_crc32c_page(u32 crc, struct page *page,
1453 unsigned int page_offset,
1454 unsigned int length)
1455{
1456 char *kaddr;
1457
1458 kaddr = kmap(page);
1459 BUG_ON(kaddr == NULL);
1460 crc = crc32c(crc, kaddr + page_offset, length);
1461 kunmap(page);
1462
1463 return crc;
1464}
Sage Weil31b80062009-10-06 11:31:13 -07001465/*
1466 * Write as much message data payload as we can. If we finish, queue
1467 * up the footer.
1468 * 1 -> done, footer is now queued in out_kvec[].
1469 * 0 -> socket full, but more to do
1470 * <0 -> error
1471 */
Alex Elder34d2d202013-03-08 20:58:59 -06001472static int write_partial_message_data(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001473{
1474 struct ceph_msg *msg = con->out_msg;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001475 struct ceph_msg_data_cursor *cursor = &msg->cursor;
Alex Elder37675b02012-03-07 11:40:08 -06001476 bool do_datacrc = !con->msgr->nocrc;
Alex Elderf5db90b2013-03-11 23:34:23 -05001477 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07001478
Alex Elder859a35d2013-03-11 23:34:23 -05001479 dout("%s %p msg %p\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07001480
Alex Elder5240d9f2013-03-14 14:09:06 -05001481 if (list_empty(&msg->data))
Alex Elder4c59b4a2013-03-11 23:34:23 -05001482 return -EINVAL;
1483
Alex Elder5821bd82012-06-11 14:57:13 -05001484 /*
1485 * Iterate through each page that contains data to be
1486 * written, and send as much as possible for each.
1487 *
1488 * If we are calculating the data crc (the default), we will
1489 * need to map the page. If we have no pages, they have
1490 * been revoked, so use the zero page.
1491 */
Alex Elderf5db90b2013-03-11 23:34:23 -05001492 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
Alex Elder643c68a2013-03-11 23:34:23 -05001493 while (cursor->resid) {
Alex Elder8a166d02013-03-08 13:35:36 -06001494 struct page *page;
Alex Eldere387d522013-03-06 23:39:38 -06001495 size_t page_offset;
1496 size_t length;
Alex Elder8a166d02013-03-08 13:35:36 -06001497 bool last_piece;
Alex Elder8ea299b2013-03-11 23:34:23 -05001498 bool need_crc;
Alex Elderf5db90b2013-03-11 23:34:23 -05001499 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07001500
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001501 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
Alex Elder4c59b4a2013-03-11 23:34:23 -05001502 &last_piece);
Alex Eldere387d522013-03-06 23:39:38 -06001503 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
Alex Elderfe38a2b2013-03-06 23:39:39 -06001504 length, last_piece);
Alex Elderf5db90b2013-03-11 23:34:23 -05001505 if (ret <= 0) {
1506 if (do_datacrc)
1507 msg->footer.data_crc = cpu_to_le32(crc);
Sage Weil31b80062009-10-06 11:31:13 -07001508
Alex Elderf5db90b2013-03-11 23:34:23 -05001509 return ret;
1510 }
Alex Elder143334f2013-03-29 11:44:10 -05001511 if (do_datacrc && cursor->need_crc)
1512 crc = ceph_crc32c_page(crc, page, page_offset, length);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001513 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
Sage Weil31b80062009-10-06 11:31:13 -07001514 }
1515
Alex Elder34d2d202013-03-08 20:58:59 -06001516 dout("%s %p msg %p done\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07001517
1518 /* prepare and queue up footer, too */
Alex Elderf5db90b2013-03-11 23:34:23 -05001519 if (do_datacrc)
1520 msg->footer.data_crc = cpu_to_le32(crc);
1521 else
Alex Elder84ca8fc2012-06-11 14:57:13 -05001522 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
Alex Eldere2200422012-05-23 14:35:23 -05001523 con_out_kvec_reset(con);
Alex Elder859eb792012-02-14 14:05:33 -06001524 prepare_write_message_footer(con);
Alex Elderf5db90b2013-03-11 23:34:23 -05001525
1526 return 1; /* must return > 0 to indicate success */
Sage Weil31b80062009-10-06 11:31:13 -07001527}
1528
1529/*
1530 * write some zeros
1531 */
1532static int write_partial_skip(struct ceph_connection *con)
1533{
1534 int ret;
1535
1536 while (con->out_skip > 0) {
Alex Elder31739132012-03-07 11:40:08 -06001537 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
Sage Weil31b80062009-10-06 11:31:13 -07001538
Alex Eldere1dcb122013-03-06 23:39:38 -06001539 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
Sage Weil31b80062009-10-06 11:31:13 -07001540 if (ret <= 0)
1541 goto out;
1542 con->out_skip -= ret;
1543 }
1544 ret = 1;
1545out:
1546 return ret;
1547}
1548
1549/*
1550 * Prepare to read connection handshake, or an ack.
1551 */
Sage Weileed0ef22009-11-10 14:34:36 -08001552static void prepare_read_banner(struct ceph_connection *con)
1553{
1554 dout("prepare_read_banner %p\n", con);
1555 con->in_base_pos = 0;
1556}
1557
Sage Weil31b80062009-10-06 11:31:13 -07001558static void prepare_read_connect(struct ceph_connection *con)
1559{
1560 dout("prepare_read_connect %p\n", con);
1561 con->in_base_pos = 0;
1562}
1563
1564static void prepare_read_ack(struct ceph_connection *con)
1565{
1566 dout("prepare_read_ack %p\n", con);
1567 con->in_base_pos = 0;
1568}
1569
Sage Weil3a230832013-03-25 08:47:40 -07001570static void prepare_read_seq(struct ceph_connection *con)
1571{
1572 dout("prepare_read_seq %p\n", con);
1573 con->in_base_pos = 0;
1574 con->in_tag = CEPH_MSGR_TAG_SEQ;
1575}
1576
Sage Weil31b80062009-10-06 11:31:13 -07001577static void prepare_read_tag(struct ceph_connection *con)
1578{
1579 dout("prepare_read_tag %p\n", con);
1580 con->in_base_pos = 0;
1581 con->in_tag = CEPH_MSGR_TAG_READY;
1582}
1583
1584/*
1585 * Prepare to read a message.
1586 */
1587static int prepare_read_message(struct ceph_connection *con)
1588{
1589 dout("prepare_read_message %p\n", con);
1590 BUG_ON(con->in_msg != NULL);
1591 con->in_base_pos = 0;
1592 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1593 return 0;
1594}
1595
1596
1597static int read_partial(struct ceph_connection *con,
Alex Elderfd516532012-05-10 10:29:50 -05001598 int end, int size, void *object)
Sage Weil31b80062009-10-06 11:31:13 -07001599{
Alex Eldere6cee712012-05-10 10:29:50 -05001600 while (con->in_base_pos < end) {
1601 int left = end - con->in_base_pos;
Sage Weil31b80062009-10-06 11:31:13 -07001602 int have = size - left;
1603 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1604 if (ret <= 0)
1605 return ret;
1606 con->in_base_pos += ret;
1607 }
1608 return 1;
1609}
1610
1611
1612/*
1613 * Read all or part of the connect-side handshake on a new connection
1614 */
Sage Weileed0ef22009-11-10 14:34:36 -08001615static int read_partial_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001616{
Alex Elderfd516532012-05-10 10:29:50 -05001617 int size;
1618 int end;
1619 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07001620
Sage Weileed0ef22009-11-10 14:34:36 -08001621 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
Sage Weil31b80062009-10-06 11:31:13 -07001622
1623 /* peer's banner */
Alex Elderfd516532012-05-10 10:29:50 -05001624 size = strlen(CEPH_BANNER);
1625 end = size;
1626 ret = read_partial(con, end, size, con->in_banner);
Sage Weil31b80062009-10-06 11:31:13 -07001627 if (ret <= 0)
1628 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001629
1630 size = sizeof (con->actual_peer_addr);
1631 end += size;
1632 ret = read_partial(con, end, size, &con->actual_peer_addr);
Sage Weil31b80062009-10-06 11:31:13 -07001633 if (ret <= 0)
1634 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001635
1636 size = sizeof (con->peer_addr_for_me);
1637 end += size;
1638 ret = read_partial(con, end, size, &con->peer_addr_for_me);
Sage Weil31b80062009-10-06 11:31:13 -07001639 if (ret <= 0)
1640 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001641
Sage Weileed0ef22009-11-10 14:34:36 -08001642out:
1643 return ret;
1644}
1645
1646static int read_partial_connect(struct ceph_connection *con)
1647{
Alex Elderfd516532012-05-10 10:29:50 -05001648 int size;
1649 int end;
1650 int ret;
Sage Weileed0ef22009-11-10 14:34:36 -08001651
1652 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1653
Alex Elderfd516532012-05-10 10:29:50 -05001654 size = sizeof (con->in_reply);
1655 end = size;
1656 ret = read_partial(con, end, size, &con->in_reply);
Sage Weil31b80062009-10-06 11:31:13 -07001657 if (ret <= 0)
1658 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001659
1660 size = le32_to_cpu(con->in_reply.authorizer_len);
1661 end += size;
1662 ret = read_partial(con, end, size, con->auth_reply_buf);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001663 if (ret <= 0)
1664 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07001665
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001666 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1667 con, (int)con->in_reply.tag,
1668 le32_to_cpu(con->in_reply.connect_seq),
Sage Weil31b80062009-10-06 11:31:13 -07001669 le32_to_cpu(con->in_reply.global_seq));
1670out:
1671 return ret;
Sage Weileed0ef22009-11-10 14:34:36 -08001672
Sage Weil31b80062009-10-06 11:31:13 -07001673}
1674
1675/*
1676 * Verify the hello banner looks okay.
1677 */
1678static int verify_hello(struct ceph_connection *con)
1679{
1680 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
Sage Weil13e38c82009-10-09 16:36:34 -07001681 pr_err("connect to %s got bad banner\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001682 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07001683 con->error_msg = "protocol error, bad banner";
1684 return -1;
1685 }
1686 return 0;
1687}
1688
1689static bool addr_is_blank(struct sockaddr_storage *ss)
1690{
1691 switch (ss->ss_family) {
1692 case AF_INET:
1693 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1694 case AF_INET6:
1695 return
1696 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1697 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1698 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1699 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1700 }
1701 return false;
1702}
1703
1704static int addr_port(struct sockaddr_storage *ss)
1705{
1706 switch (ss->ss_family) {
1707 case AF_INET:
Sage Weilf28bcfb2009-11-04 11:46:35 -08001708 return ntohs(((struct sockaddr_in *)ss)->sin_port);
Sage Weil31b80062009-10-06 11:31:13 -07001709 case AF_INET6:
Sage Weilf28bcfb2009-11-04 11:46:35 -08001710 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
Sage Weil31b80062009-10-06 11:31:13 -07001711 }
1712 return 0;
1713}
1714
1715static void addr_set_port(struct sockaddr_storage *ss, int p)
1716{
1717 switch (ss->ss_family) {
1718 case AF_INET:
1719 ((struct sockaddr_in *)ss)->sin_port = htons(p);
Sage Weila2a79602011-05-12 15:34:24 -07001720 break;
Sage Weil31b80062009-10-06 11:31:13 -07001721 case AF_INET6:
1722 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
Sage Weila2a79602011-05-12 15:34:24 -07001723 break;
Sage Weil31b80062009-10-06 11:31:13 -07001724 }
1725}
1726
1727/*
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001728 * Unlike other *_pton function semantics, zero indicates success.
1729 */
1730static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1731 char delim, const char **ipend)
1732{
Alex Elder99f0f3b2012-01-23 15:49:27 -06001733 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1734 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001735
1736 memset(ss, 0, sizeof(*ss));
1737
1738 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1739 ss->ss_family = AF_INET;
1740 return 0;
1741 }
1742
1743 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1744 ss->ss_family = AF_INET6;
1745 return 0;
1746 }
1747
1748 return -EINVAL;
1749}
1750
1751/*
1752 * Extract hostname string and resolve using kernel DNS facility.
1753 */
1754#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1755static int ceph_dns_resolve_name(const char *name, size_t namelen,
1756 struct sockaddr_storage *ss, char delim, const char **ipend)
1757{
1758 const char *end, *delim_p;
1759 char *colon_p, *ip_addr = NULL;
1760 int ip_len, ret;
1761
1762 /*
1763 * The end of the hostname occurs immediately preceding the delimiter or
1764 * the port marker (':') where the delimiter takes precedence.
1765 */
1766 delim_p = memchr(name, delim, namelen);
1767 colon_p = memchr(name, ':', namelen);
1768
1769 if (delim_p && colon_p)
1770 end = delim_p < colon_p ? delim_p : colon_p;
1771 else if (!delim_p && colon_p)
1772 end = colon_p;
1773 else {
1774 end = delim_p;
1775 if (!end) /* case: hostname:/ */
1776 end = name + namelen;
1777 }
1778
1779 if (end <= name)
1780 return -EINVAL;
1781
1782 /* do dns_resolve upcall */
1783 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1784 if (ip_len > 0)
1785 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1786 else
1787 ret = -ESRCH;
1788
1789 kfree(ip_addr);
1790
1791 *ipend = end;
1792
1793 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1794 ret, ret ? "failed" : ceph_pr_addr(ss));
1795
1796 return ret;
1797}
1798#else
1799static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1800 struct sockaddr_storage *ss, char delim, const char **ipend)
1801{
1802 return -EINVAL;
1803}
1804#endif
1805
1806/*
1807 * Parse a server name (IP or hostname). If a valid IP address is not found
1808 * then try to extract a hostname to resolve using userspace DNS upcall.
1809 */
1810static int ceph_parse_server_name(const char *name, size_t namelen,
1811 struct sockaddr_storage *ss, char delim, const char **ipend)
1812{
1813 int ret;
1814
1815 ret = ceph_pton(name, namelen, ss, delim, ipend);
1816 if (ret)
1817 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1818
1819 return ret;
1820}
1821
1822/*
Sage Weil31b80062009-10-06 11:31:13 -07001823 * Parse an ip[:port] list into an addr array. Use the default
1824 * monitor port if a port isn't specified.
1825 */
1826int ceph_parse_ips(const char *c, const char *end,
1827 struct ceph_entity_addr *addr,
1828 int max_count, int *count)
1829{
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001830 int i, ret = -EINVAL;
Sage Weil31b80062009-10-06 11:31:13 -07001831 const char *p = c;
1832
1833 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1834 for (i = 0; i < max_count; i++) {
1835 const char *ipend;
1836 struct sockaddr_storage *ss = &addr[i].in_addr;
Sage Weil31b80062009-10-06 11:31:13 -07001837 int port;
Sage Weil39139f62010-07-08 09:54:52 -07001838 char delim = ',';
1839
1840 if (*p == '[') {
1841 delim = ']';
1842 p++;
1843 }
Sage Weil31b80062009-10-06 11:31:13 -07001844
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001845 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1846 if (ret)
Sage Weil31b80062009-10-06 11:31:13 -07001847 goto bad;
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001848 ret = -EINVAL;
1849
Sage Weil31b80062009-10-06 11:31:13 -07001850 p = ipend;
1851
Sage Weil39139f62010-07-08 09:54:52 -07001852 if (delim == ']') {
1853 if (*p != ']') {
1854 dout("missing matching ']'\n");
1855 goto bad;
1856 }
1857 p++;
1858 }
1859
Sage Weil31b80062009-10-06 11:31:13 -07001860 /* port? */
1861 if (p < end && *p == ':') {
1862 port = 0;
1863 p++;
1864 while (p < end && *p >= '0' && *p <= '9') {
1865 port = (port * 10) + (*p - '0');
1866 p++;
1867 }
1868 if (port > 65535 || port == 0)
1869 goto bad;
1870 } else {
1871 port = CEPH_MON_PORT;
1872 }
1873
1874 addr_set_port(ss, port);
1875
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001876 dout("parse_ips got %s\n", ceph_pr_addr(ss));
Sage Weil31b80062009-10-06 11:31:13 -07001877
1878 if (p == end)
1879 break;
1880 if (*p != ',')
1881 goto bad;
1882 p++;
1883 }
1884
1885 if (p != end)
1886 goto bad;
1887
1888 if (count)
1889 *count = i + 1;
1890 return 0;
1891
1892bad:
Sage Weil39139f62010-07-08 09:54:52 -07001893 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001894 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07001895}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001896EXPORT_SYMBOL(ceph_parse_ips);
Sage Weil31b80062009-10-06 11:31:13 -07001897
Sage Weileed0ef22009-11-10 14:34:36 -08001898static int process_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001899{
Sage Weileed0ef22009-11-10 14:34:36 -08001900 dout("process_banner on %p\n", con);
Sage Weil31b80062009-10-06 11:31:13 -07001901
1902 if (verify_hello(con) < 0)
1903 return -1;
1904
Sage Weil63f2d212009-11-03 15:17:56 -08001905 ceph_decode_addr(&con->actual_peer_addr);
1906 ceph_decode_addr(&con->peer_addr_for_me);
1907
Sage Weil31b80062009-10-06 11:31:13 -07001908 /*
1909 * Make sure the other end is who we wanted. note that the other
1910 * end may not yet know their ip address, so if it's 0.0.0.0, give
1911 * them the benefit of the doubt.
1912 */
Sage Weil103e2d32010-01-07 16:12:36 -08001913 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1914 sizeof(con->peer_addr)) != 0 &&
Sage Weil31b80062009-10-06 11:31:13 -07001915 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1916 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001917 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001918 ceph_pr_addr(&con->peer_addr.in_addr),
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001919 (int)le32_to_cpu(con->peer_addr.nonce),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001920 ceph_pr_addr(&con->actual_peer_addr.in_addr),
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001921 (int)le32_to_cpu(con->actual_peer_addr.nonce));
Sage Weil58bb3b32009-12-23 12:12:31 -08001922 con->error_msg = "wrong peer at address";
Sage Weil31b80062009-10-06 11:31:13 -07001923 return -1;
1924 }
1925
1926 /*
1927 * did we learn our address?
1928 */
1929 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1930 int port = addr_port(&con->msgr->inst.addr.in_addr);
1931
1932 memcpy(&con->msgr->inst.addr.in_addr,
1933 &con->peer_addr_for_me.in_addr,
1934 sizeof(con->peer_addr_for_me.in_addr));
1935 addr_set_port(&con->msgr->inst.addr.in_addr, port);
Sage Weil63f2d212009-11-03 15:17:56 -08001936 encode_my_addr(con->msgr);
Sage Weileed0ef22009-11-10 14:34:36 -08001937 dout("process_banner learned my addr is %s\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001938 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07001939 }
1940
Sage Weileed0ef22009-11-10 14:34:36 -08001941 return 0;
1942}
1943
1944static int process_connect(struct ceph_connection *con)
1945{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001946 u64 sup_feat = con->msgr->supported_features;
1947 u64 req_feat = con->msgr->required_features;
Sage Weil04a419f2009-12-23 09:30:21 -08001948 u64 server_feat = le64_to_cpu(con->in_reply.features);
Sage Weil0da5d702011-05-19 11:21:05 -07001949 int ret;
Sage Weil04a419f2009-12-23 09:30:21 -08001950
Sage Weileed0ef22009-11-10 14:34:36 -08001951 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1952
Sage Weil31b80062009-10-06 11:31:13 -07001953 switch (con->in_reply.tag) {
Sage Weil04a419f2009-12-23 09:30:21 -08001954 case CEPH_MSGR_TAG_FEATURES:
1955 pr_err("%s%lld %s feature set mismatch,"
1956 " my %llx < server's %llx, missing %llx\n",
1957 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001958 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil04a419f2009-12-23 09:30:21 -08001959 sup_feat, server_feat, server_feat & ~sup_feat);
1960 con->error_msg = "missing required protocol features";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06001961 reset_connection(con);
Sage Weil04a419f2009-12-23 09:30:21 -08001962 return -1;
1963
Sage Weil31b80062009-10-06 11:31:13 -07001964 case CEPH_MSGR_TAG_BADPROTOVER:
Sage Weil31b80062009-10-06 11:31:13 -07001965 pr_err("%s%lld %s protocol version mismatch,"
1966 " my %d != server's %d\n",
1967 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001968 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil31b80062009-10-06 11:31:13 -07001969 le32_to_cpu(con->out_connect.protocol_version),
1970 le32_to_cpu(con->in_reply.protocol_version));
1971 con->error_msg = "protocol version mismatch";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06001972 reset_connection(con);
Sage Weil31b80062009-10-06 11:31:13 -07001973 return -1;
1974
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001975 case CEPH_MSGR_TAG_BADAUTHORIZER:
1976 con->auth_retry++;
1977 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1978 con->auth_retry);
1979 if (con->auth_retry == 2) {
1980 con->error_msg = "connect authorization failure";
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001981 return -1;
1982 }
Jim Schutt6d4221b2012-08-10 10:37:38 -07001983 con_out_kvec_reset(con);
Alex Eldere825a662012-05-16 15:16:38 -05001984 ret = prepare_write_connect(con);
Sage Weil0da5d702011-05-19 11:21:05 -07001985 if (ret < 0)
1986 return ret;
Sage Weil63733a02010-03-15 15:47:22 -07001987 prepare_read_connect(con);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001988 break;
Sage Weil31b80062009-10-06 11:31:13 -07001989
1990 case CEPH_MSGR_TAG_RESETSESSION:
1991 /*
1992 * If we connected with a large connect_seq but the peer
1993 * has no record of a session with us (no connection, or
1994 * connect_seq == 0), they will send RESETSESION to indicate
1995 * that they must have reset their session, and may have
1996 * dropped messages.
1997 */
1998 dout("process_connect got RESET peer seq %u\n",
Sage Weil5bdca4e2012-07-10 11:53:34 -07001999 le32_to_cpu(con->in_reply.connect_seq));
Sage Weil31b80062009-10-06 11:31:13 -07002000 pr_err("%s%lld %s connection reset\n",
2001 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002002 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07002003 reset_connection(con);
Jim Schutt6d4221b2012-08-10 10:37:38 -07002004 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002005 ret = prepare_write_connect(con);
2006 if (ret < 0)
2007 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002008 prepare_read_connect(con);
2009
2010 /* Tell ceph about it. */
Sage Weilec302642009-12-22 10:43:42 -08002011 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002012 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2013 if (con->ops->peer_reset)
2014 con->ops->peer_reset(con);
Sage Weilec302642009-12-22 10:43:42 -08002015 mutex_lock(&con->mutex);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002016 if (con->state != CON_STATE_NEGOTIATING)
Sage Weil0da5d702011-05-19 11:21:05 -07002017 return -EAGAIN;
Sage Weil31b80062009-10-06 11:31:13 -07002018 break;
2019
2020 case CEPH_MSGR_TAG_RETRY_SESSION:
2021 /*
2022 * If we sent a smaller connect_seq than the peer has, try
2023 * again with a larger value.
2024 */
Sage Weil5bdca4e2012-07-10 11:53:34 -07002025 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
Sage Weil31b80062009-10-06 11:31:13 -07002026 le32_to_cpu(con->out_connect.connect_seq),
Sage Weil5bdca4e2012-07-10 11:53:34 -07002027 le32_to_cpu(con->in_reply.connect_seq));
2028 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
Jim Schutt6d4221b2012-08-10 10:37:38 -07002029 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002030 ret = prepare_write_connect(con);
2031 if (ret < 0)
2032 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002033 prepare_read_connect(con);
2034 break;
2035
2036 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2037 /*
2038 * If we sent a smaller global_seq than the peer has, try
2039 * again with a larger value.
2040 */
Sage Weileed0ef22009-11-10 14:34:36 -08002041 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
Sage Weil31b80062009-10-06 11:31:13 -07002042 con->peer_global_seq,
Sage Weil5bdca4e2012-07-10 11:53:34 -07002043 le32_to_cpu(con->in_reply.global_seq));
Sage Weil31b80062009-10-06 11:31:13 -07002044 get_global_seq(con->msgr,
Sage Weil5bdca4e2012-07-10 11:53:34 -07002045 le32_to_cpu(con->in_reply.global_seq));
Jim Schutt6d4221b2012-08-10 10:37:38 -07002046 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002047 ret = prepare_write_connect(con);
2048 if (ret < 0)
2049 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002050 prepare_read_connect(con);
2051 break;
2052
Sage Weil3a230832013-03-25 08:47:40 -07002053 case CEPH_MSGR_TAG_SEQ:
Sage Weil31b80062009-10-06 11:31:13 -07002054 case CEPH_MSGR_TAG_READY:
Sage Weil04a419f2009-12-23 09:30:21 -08002055 if (req_feat & ~server_feat) {
2056 pr_err("%s%lld %s protocol feature mismatch,"
2057 " my required %llx > server's %llx, need %llx\n",
2058 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002059 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil04a419f2009-12-23 09:30:21 -08002060 req_feat, server_feat, req_feat & ~server_feat);
2061 con->error_msg = "missing required protocol features";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06002062 reset_connection(con);
Sage Weil04a419f2009-12-23 09:30:21 -08002063 return -1;
2064 }
Sage Weil8dacc7d2012-07-20 17:24:40 -07002065
Alex Elder122070a2012-12-26 10:43:57 -06002066 WARN_ON(con->state != CON_STATE_NEGOTIATING);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002067 con->state = CON_STATE_OPEN;
Sage Weil20e55c42013-03-25 09:30:13 -07002068 con->auth_retry = 0; /* we authenticated; clear flag */
Sage Weil31b80062009-10-06 11:31:13 -07002069 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2070 con->connect_seq++;
Sage Weilaba558e2010-05-12 15:23:30 -07002071 con->peer_features = server_feat;
Sage Weil31b80062009-10-06 11:31:13 -07002072 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2073 con->peer_global_seq,
2074 le32_to_cpu(con->in_reply.connect_seq),
2075 con->connect_seq);
2076 WARN_ON(con->connect_seq !=
2077 le32_to_cpu(con->in_reply.connect_seq));
Sage Weil92ac41d2009-12-14 14:56:56 -08002078
2079 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
Alex Elderc9ffc772013-02-20 10:25:12 -06002080 con_flag_set(con, CON_FLAG_LOSSYTX);
Sage Weil92ac41d2009-12-14 14:56:56 -08002081
Sage Weil85effe12012-07-30 16:22:05 -07002082 con->delay = 0; /* reset backoff memory */
Sage Weil31b80062009-10-06 11:31:13 -07002083
Sage Weil3a230832013-03-25 08:47:40 -07002084 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2085 prepare_write_seq(con);
2086 prepare_read_seq(con);
2087 } else {
2088 prepare_read_tag(con);
2089 }
Sage Weil31b80062009-10-06 11:31:13 -07002090 break;
2091
2092 case CEPH_MSGR_TAG_WAIT:
2093 /*
2094 * If there is a connection race (we are opening
2095 * connections to each other), one of us may just have
2096 * to WAIT. This shouldn't happen if we are the
2097 * client.
2098 */
Sage Weil04177882011-05-12 15:33:17 -07002099 pr_err("process_connect got WAIT as client\n");
2100 con->error_msg = "protocol error, got WAIT as client";
2101 return -1;
Sage Weil31b80062009-10-06 11:31:13 -07002102
2103 default:
2104 pr_err("connect protocol error, will retry\n");
2105 con->error_msg = "protocol error, garbage tag during connect";
2106 return -1;
2107 }
2108 return 0;
2109}
2110
2111
2112/*
2113 * read (part of) an ack
2114 */
2115static int read_partial_ack(struct ceph_connection *con)
2116{
Alex Elderfd516532012-05-10 10:29:50 -05002117 int size = sizeof (con->in_temp_ack);
2118 int end = size;
Sage Weil31b80062009-10-06 11:31:13 -07002119
Alex Elderfd516532012-05-10 10:29:50 -05002120 return read_partial(con, end, size, &con->in_temp_ack);
Sage Weil31b80062009-10-06 11:31:13 -07002121}
2122
Sage Weil31b80062009-10-06 11:31:13 -07002123/*
2124 * We can finally discard anything that's been acked.
2125 */
2126static void process_ack(struct ceph_connection *con)
2127{
2128 struct ceph_msg *m;
2129 u64 ack = le64_to_cpu(con->in_temp_ack);
2130 u64 seq;
2131
Sage Weil31b80062009-10-06 11:31:13 -07002132 while (!list_empty(&con->out_sent)) {
2133 m = list_first_entry(&con->out_sent, struct ceph_msg,
2134 list_head);
2135 seq = le64_to_cpu(m->hdr.seq);
2136 if (seq > ack)
2137 break;
2138 dout("got ack for seq %llu type %d at %p\n", seq,
2139 le16_to_cpu(m->hdr.type), m);
Sage Weil4cf9d542011-07-26 11:27:24 -07002140 m->ack_stamp = jiffies;
Sage Weil31b80062009-10-06 11:31:13 -07002141 ceph_msg_remove(m);
2142 }
Sage Weil31b80062009-10-06 11:31:13 -07002143 prepare_read_tag(con);
2144}
2145
2146
Yehuda Sadeh24504182010-01-08 13:58:34 -08002147static int read_partial_message_section(struct ceph_connection *con,
Sage Weil213c99e2010-08-03 10:25:11 -07002148 struct kvec *section,
2149 unsigned int sec_len, u32 *crc)
Yehuda Sadeh24504182010-01-08 13:58:34 -08002150{
Yehuda Sadeh68b44762010-04-06 15:01:27 -07002151 int ret, left;
Sage Weil31b80062009-10-06 11:31:13 -07002152
Yehuda Sadeh24504182010-01-08 13:58:34 -08002153 BUG_ON(!section);
Sage Weil31b80062009-10-06 11:31:13 -07002154
Yehuda Sadeh24504182010-01-08 13:58:34 -08002155 while (section->iov_len < sec_len) {
2156 BUG_ON(section->iov_base == NULL);
2157 left = sec_len - section->iov_len;
2158 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2159 section->iov_len, left);
2160 if (ret <= 0)
2161 return ret;
2162 section->iov_len += ret;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002163 }
Alex Elderfe3ad592012-02-15 07:43:54 -06002164 if (section->iov_len == sec_len)
2165 *crc = crc32c(0, section->iov_base, section->iov_len);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002166
2167 return 1;
2168}
2169
Alex Elder34d2d202013-03-08 20:58:59 -06002170static int read_partial_msg_data(struct ceph_connection *con)
2171{
2172 struct ceph_msg *msg = con->in_msg;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002173 struct ceph_msg_data_cursor *cursor = &msg->cursor;
Alex Elder34d2d202013-03-08 20:58:59 -06002174 const bool do_datacrc = !con->msgr->nocrc;
Alex Elder686be202013-03-11 23:34:23 -05002175 struct page *page;
2176 size_t page_offset;
2177 size_t length;
Alex Elderf5db90b2013-03-11 23:34:23 -05002178 u32 crc = 0;
Alex Elder34d2d202013-03-08 20:58:59 -06002179 int ret;
2180
2181 BUG_ON(!msg);
Alex Elder5240d9f2013-03-14 14:09:06 -05002182 if (list_empty(&msg->data))
Alex Elder4c59b4a2013-03-11 23:34:23 -05002183 return -EIO;
Alex Elder34d2d202013-03-08 20:58:59 -06002184
Alex Elderf5db90b2013-03-11 23:34:23 -05002185 if (do_datacrc)
2186 crc = con->in_data_crc;
Alex Elder643c68a2013-03-11 23:34:23 -05002187 while (cursor->resid) {
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002188 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
Alex Elder4c59b4a2013-03-11 23:34:23 -05002189 NULL);
Alex Elder686be202013-03-11 23:34:23 -05002190 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
Alex Elderf5db90b2013-03-11 23:34:23 -05002191 if (ret <= 0) {
2192 if (do_datacrc)
2193 con->in_data_crc = crc;
2194
Alex Elder686be202013-03-11 23:34:23 -05002195 return ret;
Alex Elderf5db90b2013-03-11 23:34:23 -05002196 }
Alex Elder686be202013-03-11 23:34:23 -05002197
2198 if (do_datacrc)
Alex Elderf5db90b2013-03-11 23:34:23 -05002199 crc = ceph_crc32c_page(crc, page, page_offset, ret);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002200 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
Alex Elder34d2d202013-03-08 20:58:59 -06002201 }
Alex Elderf5db90b2013-03-11 23:34:23 -05002202 if (do_datacrc)
2203 con->in_data_crc = crc;
Alex Elder34d2d202013-03-08 20:58:59 -06002204
2205 return 1; /* must return > 0 to indicate success */
2206}
2207
Sage Weil31b80062009-10-06 11:31:13 -07002208/*
2209 * read (part of) a message.
2210 */
Alex Elder686be202013-03-11 23:34:23 -05002211static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2212
Sage Weil31b80062009-10-06 11:31:13 -07002213static int read_partial_message(struct ceph_connection *con)
2214{
2215 struct ceph_msg *m = con->in_msg;
Alex Elderfd516532012-05-10 10:29:50 -05002216 int size;
2217 int end;
Sage Weil31b80062009-10-06 11:31:13 -07002218 int ret;
Eric Dumazet95c96172012-04-15 05:58:06 +00002219 unsigned int front_len, middle_len, data_len;
Alex Elder37675b02012-03-07 11:40:08 -06002220 bool do_datacrc = !con->msgr->nocrc;
Sage Weilae187562010-04-22 07:47:01 -07002221 u64 seq;
Alex Elderfe3ad592012-02-15 07:43:54 -06002222 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07002223
2224 dout("read_partial_message con %p msg %p\n", con, m);
2225
2226 /* header */
Alex Elderfd516532012-05-10 10:29:50 -05002227 size = sizeof (con->in_hdr);
2228 end = size;
2229 ret = read_partial(con, end, size, &con->in_hdr);
Alex Elder57dac9d2012-05-10 10:29:50 -05002230 if (ret <= 0)
2231 return ret;
Alex Elderfe3ad592012-02-15 07:43:54 -06002232
2233 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2234 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2235 pr_err("read_partial_message bad hdr "
2236 " crc %u != expected %u\n",
2237 crc, con->in_hdr.crc);
2238 return -EBADMSG;
2239 }
2240
Sage Weil31b80062009-10-06 11:31:13 -07002241 front_len = le32_to_cpu(con->in_hdr.front_len);
2242 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2243 return -EIO;
2244 middle_len = le32_to_cpu(con->in_hdr.middle_len);
Alex Elder7b11ba32013-03-08 18:51:03 -06002245 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
Sage Weil31b80062009-10-06 11:31:13 -07002246 return -EIO;
2247 data_len = le32_to_cpu(con->in_hdr.data_len);
2248 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2249 return -EIO;
2250
Sage Weilae187562010-04-22 07:47:01 -07002251 /* verify seq# */
2252 seq = le64_to_cpu(con->in_hdr.seq);
2253 if ((s64)seq - (s64)con->in_seq < 1) {
Sage Weildf9f86f2010-11-01 15:49:23 -07002254 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
Sage Weilae187562010-04-22 07:47:01 -07002255 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002256 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weilae187562010-04-22 07:47:01 -07002257 seq, con->in_seq + 1);
2258 con->in_base_pos = -front_len - middle_len - data_len -
2259 sizeof(m->footer);
2260 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weilae187562010-04-22 07:47:01 -07002261 return 0;
2262 } else if ((s64)seq - (s64)con->in_seq > 1) {
2263 pr_err("read_partial_message bad seq %lld expected %lld\n",
2264 seq, con->in_seq + 1);
2265 con->error_msg = "bad message sequence # for incoming message";
2266 return -EBADMSG;
2267 }
2268
Sage Weil31b80062009-10-06 11:31:13 -07002269 /* allocate message? */
2270 if (!con->in_msg) {
Sage Weil4740a622012-07-30 18:19:30 -07002271 int skip = 0;
2272
Sage Weil31b80062009-10-06 11:31:13 -07002273 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
Alex Elder6ebc8b32013-03-08 18:51:03 -06002274 front_len, data_len);
Sage Weil4740a622012-07-30 18:19:30 -07002275 ret = ceph_con_in_msg_alloc(con, &skip);
2276 if (ret < 0)
2277 return ret;
Alex Elderf759ebb2013-04-05 14:46:01 -05002278
2279 BUG_ON(!con->in_msg ^ skip);
2280 if (con->in_msg && data_len > con->in_msg->data_length) {
2281 pr_warning("%s skipping long message (%u > %zd)\n",
2282 __func__, data_len, con->in_msg->data_length);
2283 ceph_msg_put(con->in_msg);
2284 con->in_msg = NULL;
2285 skip = 1;
2286 }
Yehuda Sadeh24504182010-01-08 13:58:34 -08002287 if (skip) {
Sage Weil31b80062009-10-06 11:31:13 -07002288 /* skip this message */
Sage Weila79832f2010-04-01 16:06:19 -07002289 dout("alloc_msg said skip message\n");
Sage Weil31b80062009-10-06 11:31:13 -07002290 con->in_base_pos = -front_len - middle_len - data_len -
2291 sizeof(m->footer);
2292 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weil684be252010-04-21 20:45:59 -07002293 con->in_seq++;
Sage Weil31b80062009-10-06 11:31:13 -07002294 return 0;
2295 }
Alex Elder38941f82012-06-01 14:56:43 -05002296
Sage Weil4740a622012-07-30 18:19:30 -07002297 BUG_ON(!con->in_msg);
Alex Elder38941f82012-06-01 14:56:43 -05002298 BUG_ON(con->in_msg->con != con);
Sage Weil31b80062009-10-06 11:31:13 -07002299 m = con->in_msg;
2300 m->front.iov_len = 0; /* haven't read it yet */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002301 if (m->middle)
2302 m->middle->vec.iov_len = 0;
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08002303
Alex Elder78625052013-03-06 23:39:39 -06002304 /* prepare for data payload, if any */
Sage Weila4107022012-07-30 16:20:25 -07002305
Alex Elder78625052013-03-06 23:39:39 -06002306 if (data_len)
Alex Elder98fa5dd2013-04-02 12:09:50 -05002307 prepare_message_data(con->in_msg, data_len);
Sage Weil31b80062009-10-06 11:31:13 -07002308 }
2309
2310 /* front */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002311 ret = read_partial_message_section(con, &m->front, front_len,
2312 &con->in_front_crc);
2313 if (ret <= 0)
2314 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002315
2316 /* middle */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002317 if (m->middle) {
Sage Weil213c99e2010-08-03 10:25:11 -07002318 ret = read_partial_message_section(con, &m->middle->vec,
2319 middle_len,
Yehuda Sadeh24504182010-01-08 13:58:34 -08002320 &con->in_middle_crc);
Sage Weil31b80062009-10-06 11:31:13 -07002321 if (ret <= 0)
2322 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002323 }
2324
2325 /* (page) data */
Alex Elder34d2d202013-03-08 20:58:59 -06002326 if (data_len) {
2327 ret = read_partial_msg_data(con);
2328 if (ret <= 0)
2329 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002330 }
2331
Sage Weil31b80062009-10-06 11:31:13 -07002332 /* footer */
Alex Elderfd516532012-05-10 10:29:50 -05002333 size = sizeof (m->footer);
2334 end += size;
2335 ret = read_partial(con, end, size, &m->footer);
Alex Elder57dac9d2012-05-10 10:29:50 -05002336 if (ret <= 0)
2337 return ret;
2338
Sage Weil31b80062009-10-06 11:31:13 -07002339 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2340 m, front_len, m->footer.front_crc, middle_len,
2341 m->footer.middle_crc, data_len, m->footer.data_crc);
2342
2343 /* crc ok? */
2344 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2345 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2346 m, con->in_front_crc, m->footer.front_crc);
2347 return -EBADMSG;
2348 }
2349 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2350 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2351 m, con->in_middle_crc, m->footer.middle_crc);
2352 return -EBADMSG;
2353 }
Alex Elderbca064d2012-02-15 07:43:54 -06002354 if (do_datacrc &&
Sage Weil31b80062009-10-06 11:31:13 -07002355 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2356 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2357 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2358 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2359 return -EBADMSG;
2360 }
2361
2362 return 1; /* done! */
2363}
2364
2365/*
2366 * Process message. This happens in the worker thread. The callback should
2367 * be careful not to do anything that waits on other incoming messages or it
2368 * may deadlock.
2369 */
2370static void process_message(struct ceph_connection *con)
2371{
Sage Weil5e095e82009-12-14 14:30:34 -08002372 struct ceph_msg *msg;
Sage Weil31b80062009-10-06 11:31:13 -07002373
Alex Elder38941f82012-06-01 14:56:43 -05002374 BUG_ON(con->in_msg->con != con);
2375 con->in_msg->con = NULL;
Sage Weil5e095e82009-12-14 14:30:34 -08002376 msg = con->in_msg;
Sage Weil31b80062009-10-06 11:31:13 -07002377 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -07002378 con->ops->put(con);
Sage Weil31b80062009-10-06 11:31:13 -07002379
2380 /* if first message, set peer_name */
2381 if (con->peer_name.type == 0)
Sage Weildbad1852010-03-25 15:45:38 -07002382 con->peer_name = msg->hdr.src;
Sage Weil31b80062009-10-06 11:31:13 -07002383
Sage Weil31b80062009-10-06 11:31:13 -07002384 con->in_seq++;
Sage Weilec302642009-12-22 10:43:42 -08002385 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002386
2387 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2388 msg, le64_to_cpu(msg->hdr.seq),
Sage Weildbad1852010-03-25 15:45:38 -07002389 ENTITY_NAME(msg->hdr.src),
Sage Weil31b80062009-10-06 11:31:13 -07002390 le16_to_cpu(msg->hdr.type),
2391 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2392 le32_to_cpu(msg->hdr.front_len),
2393 le32_to_cpu(msg->hdr.data_len),
2394 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2395 con->ops->dispatch(con, msg);
Sage Weilec302642009-12-22 10:43:42 -08002396
2397 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002398}
2399
2400
2401/*
2402 * Write something to the socket. Called in a worker thread when the
2403 * socket appears to be writeable and we have something ready to send.
2404 */
2405static int try_write(struct ceph_connection *con)
2406{
Sage Weil31b80062009-10-06 11:31:13 -07002407 int ret = 1;
2408
Sage Weild59315c2012-06-21 12:49:23 -07002409 dout("try_write start %p state %lu\n", con, con->state);
Sage Weil31b80062009-10-06 11:31:13 -07002410
Sage Weil31b80062009-10-06 11:31:13 -07002411more:
2412 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2413
2414 /* open the socket first? */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002415 if (con->state == CON_STATE_PREOPEN) {
2416 BUG_ON(con->sock);
2417 con->state = CON_STATE_CONNECTING;
Alex Eldera5988c42012-05-29 11:04:58 -05002418
Alex Eldere2200422012-05-23 14:35:23 -05002419 con_out_kvec_reset(con);
Alex Eldere825a662012-05-16 15:16:38 -05002420 prepare_write_banner(con);
Sage Weileed0ef22009-11-10 14:34:36 -08002421 prepare_read_banner(con);
Sage Weil31b80062009-10-06 11:31:13 -07002422
Sage Weilcf3e5c42009-12-11 09:48:05 -08002423 BUG_ON(con->in_msg);
Sage Weil31b80062009-10-06 11:31:13 -07002424 con->in_tag = CEPH_MSGR_TAG_READY;
2425 dout("try_write initiating connect on %p new state %lu\n",
2426 con, con->state);
Alex Elder41617d02012-02-14 14:05:33 -06002427 ret = ceph_tcp_connect(con);
2428 if (ret < 0) {
Sage Weil31b80062009-10-06 11:31:13 -07002429 con->error_msg = "connect error";
Sage Weil31b80062009-10-06 11:31:13 -07002430 goto out;
2431 }
2432 }
2433
2434more_kvec:
2435 /* kvec data queued? */
2436 if (con->out_skip) {
2437 ret = write_partial_skip(con);
2438 if (ret <= 0)
Sage Weil42961d22011-01-25 08:19:34 -08002439 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002440 }
2441 if (con->out_kvec_left) {
2442 ret = write_partial_kvec(con);
2443 if (ret <= 0)
Sage Weil42961d22011-01-25 08:19:34 -08002444 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002445 }
2446
2447 /* msg pages? */
2448 if (con->out_msg) {
Sage Weilc86a2932009-12-14 14:04:30 -08002449 if (con->out_msg_done) {
2450 ceph_msg_put(con->out_msg);
2451 con->out_msg = NULL; /* we're done with this one */
2452 goto do_next;
2453 }
2454
Alex Elder34d2d202013-03-08 20:58:59 -06002455 ret = write_partial_message_data(con);
Sage Weil31b80062009-10-06 11:31:13 -07002456 if (ret == 1)
2457 goto more_kvec; /* we need to send the footer, too! */
2458 if (ret == 0)
Sage Weil42961d22011-01-25 08:19:34 -08002459 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002460 if (ret < 0) {
Alex Elder34d2d202013-03-08 20:58:59 -06002461 dout("try_write write_partial_message_data err %d\n",
Sage Weil31b80062009-10-06 11:31:13 -07002462 ret);
Sage Weil42961d22011-01-25 08:19:34 -08002463 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002464 }
2465 }
2466
Sage Weilc86a2932009-12-14 14:04:30 -08002467do_next:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002468 if (con->state == CON_STATE_OPEN) {
Sage Weil31b80062009-10-06 11:31:13 -07002469 /* is anything else pending? */
2470 if (!list_empty(&con->out_queue)) {
2471 prepare_write_message(con);
2472 goto more;
2473 }
2474 if (con->in_seq > con->in_seq_acked) {
2475 prepare_write_ack(con);
2476 goto more;
2477 }
Alex Elderc9ffc772013-02-20 10:25:12 -06002478 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
Sage Weil31b80062009-10-06 11:31:13 -07002479 prepare_write_keepalive(con);
2480 goto more;
2481 }
2482 }
2483
2484 /* Nothing to do! */
Alex Elderc9ffc772013-02-20 10:25:12 -06002485 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07002486 dout("try_write nothing else to write.\n");
Sage Weil31b80062009-10-06 11:31:13 -07002487 ret = 0;
2488out:
Sage Weil42961d22011-01-25 08:19:34 -08002489 dout("try_write done on %p ret %d\n", con, ret);
Sage Weil31b80062009-10-06 11:31:13 -07002490 return ret;
2491}
2492
2493
2494
2495/*
2496 * Read what we can from the socket.
2497 */
2498static int try_read(struct ceph_connection *con)
2499{
Sage Weil31b80062009-10-06 11:31:13 -07002500 int ret = -1;
2501
Sage Weil31b80062009-10-06 11:31:13 -07002502more:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002503 dout("try_read start on %p state %lu\n", con, con->state);
2504 if (con->state != CON_STATE_CONNECTING &&
2505 con->state != CON_STATE_NEGOTIATING &&
2506 con->state != CON_STATE_OPEN)
2507 return 0;
2508
2509 BUG_ON(!con->sock);
2510
Sage Weil31b80062009-10-06 11:31:13 -07002511 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2512 con->in_base_pos);
Sage Weil0da5d702011-05-19 11:21:05 -07002513
Sage Weil8dacc7d2012-07-20 17:24:40 -07002514 if (con->state == CON_STATE_CONNECTING) {
Alex Elder7593af92012-05-24 11:55:03 -05002515 dout("try_read connecting\n");
2516 ret = read_partial_banner(con);
2517 if (ret <= 0)
Alex Elderab166d52012-05-31 11:37:29 -05002518 goto out;
Alex Elder7593af92012-05-24 11:55:03 -05002519 ret = process_banner(con);
2520 if (ret < 0)
2521 goto out;
2522
Sage Weil8dacc7d2012-07-20 17:24:40 -07002523 con->state = CON_STATE_NEGOTIATING;
Alex Elder7593af92012-05-24 11:55:03 -05002524
Jim Schutt6d4221b2012-08-10 10:37:38 -07002525 /*
2526 * Received banner is good, exchange connection info.
2527 * Do not reset out_kvec, as sending our banner raced
2528 * with receiving peer banner after connect completed.
2529 */
Alex Elder7593af92012-05-24 11:55:03 -05002530 ret = prepare_write_connect(con);
2531 if (ret < 0)
2532 goto out;
2533 prepare_read_connect(con);
2534
2535 /* Send connection info before awaiting response */
Sage Weil0da5d702011-05-19 11:21:05 -07002536 goto out;
2537 }
2538
Sage Weil8dacc7d2012-07-20 17:24:40 -07002539 if (con->state == CON_STATE_NEGOTIATING) {
Alex Elder7593af92012-05-24 11:55:03 -05002540 dout("try_read negotiating\n");
Sage Weil31b80062009-10-06 11:31:13 -07002541 ret = read_partial_connect(con);
2542 if (ret <= 0)
Sage Weil31b80062009-10-06 11:31:13 -07002543 goto out;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002544 ret = process_connect(con);
2545 if (ret < 0)
2546 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002547 goto more;
2548 }
2549
Alex Elder122070a2012-12-26 10:43:57 -06002550 WARN_ON(con->state != CON_STATE_OPEN);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002551
Sage Weil31b80062009-10-06 11:31:13 -07002552 if (con->in_base_pos < 0) {
2553 /*
2554 * skipping + discarding content.
2555 *
2556 * FIXME: there must be a better way to do this!
2557 */
Alex Elder84495f42012-02-15 07:43:55 -06002558 static char buf[SKIP_BUF_SIZE];
2559 int skip = min((int) sizeof (buf), -con->in_base_pos);
2560
Sage Weil31b80062009-10-06 11:31:13 -07002561 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2562 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2563 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002564 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002565 con->in_base_pos += ret;
2566 if (con->in_base_pos)
2567 goto more;
2568 }
2569 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2570 /*
2571 * what's next?
2572 */
2573 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2574 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002575 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002576 dout("try_read got tag %d\n", (int)con->in_tag);
2577 switch (con->in_tag) {
2578 case CEPH_MSGR_TAG_MSG:
2579 prepare_read_message(con);
2580 break;
2581 case CEPH_MSGR_TAG_ACK:
2582 prepare_read_ack(con);
2583 break;
2584 case CEPH_MSGR_TAG_CLOSE:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002585 con_close_socket(con);
2586 con->state = CON_STATE_CLOSED;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002587 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002588 default:
2589 goto bad_tag;
2590 }
2591 }
2592 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2593 ret = read_partial_message(con);
2594 if (ret <= 0) {
2595 switch (ret) {
2596 case -EBADMSG:
2597 con->error_msg = "bad crc";
2598 ret = -EIO;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002599 break;
Sage Weil31b80062009-10-06 11:31:13 -07002600 case -EIO:
2601 con->error_msg = "io error";
Sage Weil98bdb0a2011-01-25 08:17:48 -08002602 break;
Sage Weil31b80062009-10-06 11:31:13 -07002603 }
Sage Weil98bdb0a2011-01-25 08:17:48 -08002604 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002605 }
2606 if (con->in_tag == CEPH_MSGR_TAG_READY)
2607 goto more;
2608 process_message(con);
Sage Weil7b862e02012-07-30 18:16:56 -07002609 if (con->state == CON_STATE_OPEN)
2610 prepare_read_tag(con);
Sage Weil31b80062009-10-06 11:31:13 -07002611 goto more;
2612 }
Sage Weil3a230832013-03-25 08:47:40 -07002613 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2614 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2615 /*
2616 * the final handshake seq exchange is semantically
2617 * equivalent to an ACK
2618 */
Sage Weil31b80062009-10-06 11:31:13 -07002619 ret = read_partial_ack(con);
2620 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002621 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002622 process_ack(con);
2623 goto more;
2624 }
2625
Sage Weil31b80062009-10-06 11:31:13 -07002626out:
Sage Weil98bdb0a2011-01-25 08:17:48 -08002627 dout("try_read done on %p ret %d\n", con, ret);
Sage Weil31b80062009-10-06 11:31:13 -07002628 return ret;
2629
2630bad_tag:
2631 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2632 con->error_msg = "protocol error, garbage tag";
2633 ret = -1;
2634 goto out;
2635}
2636
2637
2638/*
Alex Elder802c6d92012-10-08 20:37:30 -07002639 * Atomically queue work on a connection after the specified delay.
2640 * Bump @con reference to avoid races with connection teardown.
2641 * Returns 0 if work was queued, or an error code otherwise.
Sage Weil31b80062009-10-06 11:31:13 -07002642 */
Alex Elder802c6d92012-10-08 20:37:30 -07002643static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
Sage Weil31b80062009-10-06 11:31:13 -07002644{
Sage Weil31b80062009-10-06 11:31:13 -07002645 if (!con->ops->get(con)) {
Alex Elder802c6d92012-10-08 20:37:30 -07002646 dout("%s %p ref count 0\n", __func__, con);
2647
2648 return -ENOENT;
Sage Weil31b80062009-10-06 11:31:13 -07002649 }
2650
Alex Elder802c6d92012-10-08 20:37:30 -07002651 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2652 dout("%s %p - already queued\n", __func__, con);
Sage Weil31b80062009-10-06 11:31:13 -07002653 con->ops->put(con);
Alex Elder802c6d92012-10-08 20:37:30 -07002654
2655 return -EBUSY;
Sage Weil31b80062009-10-06 11:31:13 -07002656 }
Alex Elder802c6d92012-10-08 20:37:30 -07002657
2658 dout("%s %p %lu\n", __func__, con, delay);
2659
2660 return 0;
2661}
2662
2663static void queue_con(struct ceph_connection *con)
2664{
2665 (void) queue_con_delay(con, 0);
Sage Weil31b80062009-10-06 11:31:13 -07002666}
2667
Alex Elder7bb21d682012-12-07 19:50:07 -06002668static bool con_sock_closed(struct ceph_connection *con)
2669{
Alex Elderc9ffc772013-02-20 10:25:12 -06002670 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
Alex Elder7bb21d682012-12-07 19:50:07 -06002671 return false;
2672
2673#define CASE(x) \
2674 case CON_STATE_ ## x: \
2675 con->error_msg = "socket closed (con state " #x ")"; \
2676 break;
2677
2678 switch (con->state) {
2679 CASE(CLOSED);
2680 CASE(PREOPEN);
2681 CASE(CONNECTING);
2682 CASE(NEGOTIATING);
2683 CASE(OPEN);
2684 CASE(STANDBY);
2685 default:
2686 pr_warning("%s con %p unrecognized state %lu\n",
2687 __func__, con, con->state);
2688 con->error_msg = "unrecognized con state";
2689 BUG();
2690 break;
2691 }
2692#undef CASE
2693
2694 return true;
2695}
2696
Alex Elderf20a39f2013-02-19 12:25:57 -06002697static bool con_backoff(struct ceph_connection *con)
2698{
2699 int ret;
2700
2701 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2702 return false;
2703
2704 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2705 if (ret) {
2706 dout("%s: con %p FAILED to back off %lu\n", __func__,
2707 con, con->delay);
2708 BUG_ON(ret == -ENOENT);
2709 con_flag_set(con, CON_FLAG_BACKOFF);
2710 }
2711
2712 return true;
2713}
2714
Alex Elder93209262013-02-19 12:25:57 -06002715/* Finish fault handling; con->mutex must *not* be held here */
2716
2717static void con_fault_finish(struct ceph_connection *con)
2718{
2719 /*
2720 * in case we faulted due to authentication, invalidate our
2721 * current tickets so that we can get new ones.
2722 */
2723 if (con->auth_retry && con->ops->invalidate_authorizer) {
2724 dout("calling invalidate_authorizer()\n");
2725 con->ops->invalidate_authorizer(con);
2726 }
2727
2728 if (con->ops->fault)
2729 con->ops->fault(con);
2730}
2731
Sage Weil31b80062009-10-06 11:31:13 -07002732/*
2733 * Do some work on a connection. Drop a connection ref when we're done.
2734 */
2735static void con_work(struct work_struct *work)
2736{
2737 struct ceph_connection *con = container_of(work, struct ceph_connection,
2738 work.work);
Alex Elder49659412013-02-19 12:25:57 -06002739 bool fault;
Sage Weil31b80062009-10-06 11:31:13 -07002740
Sage Weil9dd46582010-04-28 13:51:50 -07002741 mutex_lock(&con->mutex);
Alex Elder49659412013-02-19 12:25:57 -06002742 while (true) {
2743 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07002744
Alex Elder49659412013-02-19 12:25:57 -06002745 if ((fault = con_sock_closed(con))) {
2746 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2747 break;
2748 }
2749 if (con_backoff(con)) {
2750 dout("%s: con %p BACKOFF\n", __func__, con);
2751 break;
2752 }
2753 if (con->state == CON_STATE_STANDBY) {
2754 dout("%s: con %p STANDBY\n", __func__, con);
2755 break;
2756 }
2757 if (con->state == CON_STATE_CLOSED) {
2758 dout("%s: con %p CLOSED\n", __func__, con);
2759 BUG_ON(con->sock);
2760 break;
2761 }
2762 if (con->state == CON_STATE_PREOPEN) {
2763 dout("%s: con %p PREOPEN\n", __func__, con);
2764 BUG_ON(con->sock);
2765 }
Sage Weil0da5d702011-05-19 11:21:05 -07002766
Alex Elder49659412013-02-19 12:25:57 -06002767 ret = try_read(con);
2768 if (ret < 0) {
2769 if (ret == -EAGAIN)
2770 continue;
2771 con->error_msg = "socket error on read";
2772 fault = true;
2773 break;
2774 }
2775
2776 ret = try_write(con);
2777 if (ret < 0) {
2778 if (ret == -EAGAIN)
2779 continue;
2780 con->error_msg = "socket error on write";
2781 fault = true;
2782 }
2783
2784 break; /* If we make it to here, we're done */
Sage Weil3a140a02012-07-30 16:24:21 -07002785 }
Alex Elderb6e7b6a2013-02-19 12:25:57 -06002786 if (fault)
2787 con_fault(con);
Sage Weil9dd46582010-04-28 13:51:50 -07002788 mutex_unlock(&con->mutex);
Sage Weil0da5d702011-05-19 11:21:05 -07002789
Alex Elderb6e7b6a2013-02-19 12:25:57 -06002790 if (fault)
2791 con_fault_finish(con);
2792
2793 con->ops->put(con);
Sage Weil31b80062009-10-06 11:31:13 -07002794}
2795
Sage Weil31b80062009-10-06 11:31:13 -07002796/*
2797 * Generic error/fault handler. A retry mechanism is used with
2798 * exponential backoff
2799 */
Alex Elder93209262013-02-19 12:25:57 -06002800static void con_fault(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07002801{
Alex Elder28362982012-12-14 16:47:41 -06002802 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002803 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
Sage Weil31b80062009-10-06 11:31:13 -07002804 dout("fault %p state %lu to peer %s\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002805 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07002806
Alex Elder122070a2012-12-26 10:43:57 -06002807 WARN_ON(con->state != CON_STATE_CONNECTING &&
Sage Weil8dacc7d2012-07-20 17:24:40 -07002808 con->state != CON_STATE_NEGOTIATING &&
2809 con->state != CON_STATE_OPEN);
Sage Weilec302642009-12-22 10:43:42 -08002810
Sage Weil31b80062009-10-06 11:31:13 -07002811 con_close_socket(con);
Sage Weil5e095e82009-12-14 14:30:34 -08002812
Alex Elderc9ffc772013-02-20 10:25:12 -06002813 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
Sage Weil8dacc7d2012-07-20 17:24:40 -07002814 dout("fault on LOSSYTX channel, marking CLOSED\n");
2815 con->state = CON_STATE_CLOSED;
Alex Elder93209262013-02-19 12:25:57 -06002816 return;
Sage Weil3b5ede02012-07-20 15:22:53 -07002817 }
2818
Sage Weil5e095e82009-12-14 14:30:34 -08002819 if (con->in_msg) {
Alex Elder38941f82012-06-01 14:56:43 -05002820 BUG_ON(con->in_msg->con != con);
2821 con->in_msg->con = NULL;
Sage Weil5e095e82009-12-14 14:30:34 -08002822 ceph_msg_put(con->in_msg);
2823 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -07002824 con->ops->put(con);
Sage Weil5e095e82009-12-14 14:30:34 -08002825 }
Sage Weil31b80062009-10-06 11:31:13 -07002826
Sage Weile80a52d2010-02-25 12:40:45 -08002827 /* Requeue anything that hasn't been acked */
2828 list_splice_init(&con->out_sent, &con->out_queue);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002829
Sage Weile76661d2011-03-03 10:10:15 -08002830 /* If there are no messages queued or keepalive pending, place
2831 * the connection in a STANDBY state */
2832 if (list_empty(&con->out_queue) &&
Alex Elderc9ffc772013-02-20 10:25:12 -06002833 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
Sage Weile00de342011-03-04 12:25:05 -08002834 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
Alex Elderc9ffc772013-02-20 10:25:12 -06002835 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002836 con->state = CON_STATE_STANDBY;
Sage Weile80a52d2010-02-25 12:40:45 -08002837 } else {
2838 /* retry after a delay. */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002839 con->state = CON_STATE_PREOPEN;
Sage Weile80a52d2010-02-25 12:40:45 -08002840 if (con->delay == 0)
2841 con->delay = BASE_DELAY_INTERVAL;
2842 else if (con->delay < MAX_DELAY_INTERVAL)
2843 con->delay *= 2;
Alex Elderc9ffc772013-02-20 10:25:12 -06002844 con_flag_set(con, CON_FLAG_BACKOFF);
Alex Elder8618e302012-10-08 20:37:30 -07002845 queue_con(con);
Sage Weil31b80062009-10-06 11:31:13 -07002846 }
Sage Weil31b80062009-10-06 11:31:13 -07002847}
2848
2849
2850
2851/*
Alex Elder15d98822012-05-26 23:26:43 -05002852 * initialize a new messenger instance
Sage Weil31b80062009-10-06 11:31:13 -07002853 */
Alex Elder15d98822012-05-26 23:26:43 -05002854void ceph_messenger_init(struct ceph_messenger *msgr,
2855 struct ceph_entity_addr *myaddr,
2856 u32 supported_features,
2857 u32 required_features,
2858 bool nocrc)
Sage Weil31b80062009-10-06 11:31:13 -07002859{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002860 msgr->supported_features = supported_features;
2861 msgr->required_features = required_features;
2862
Sage Weil31b80062009-10-06 11:31:13 -07002863 spin_lock_init(&msgr->global_seq_lock);
2864
Sage Weil31b80062009-10-06 11:31:13 -07002865 if (myaddr)
2866 msgr->inst.addr = *myaddr;
2867
2868 /* select a random nonce */
Sage Weilac8839d2010-01-27 14:28:10 -08002869 msgr->inst.addr.type = 0;
Sage Weil103e2d32010-01-07 16:12:36 -08002870 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
Sage Weil63f2d212009-11-03 15:17:56 -08002871 encode_my_addr(msgr);
Alex Elder15d98822012-05-26 23:26:43 -05002872 msgr->nocrc = nocrc;
Sage Weil31b80062009-10-06 11:31:13 -07002873
Guanjun Hea2a32582012-07-08 19:50:33 -07002874 atomic_set(&msgr->stopping, 0);
Sage Weil31b80062009-10-06 11:31:13 -07002875
Alex Elder15d98822012-05-26 23:26:43 -05002876 dout("%s %p\n", __func__, msgr);
Sage Weil31b80062009-10-06 11:31:13 -07002877}
Alex Elder15d98822012-05-26 23:26:43 -05002878EXPORT_SYMBOL(ceph_messenger_init);
Sage Weil31b80062009-10-06 11:31:13 -07002879
Sage Weile00de342011-03-04 12:25:05 -08002880static void clear_standby(struct ceph_connection *con)
2881{
2882 /* come back from STANDBY? */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002883 if (con->state == CON_STATE_STANDBY) {
Sage Weile00de342011-03-04 12:25:05 -08002884 dout("clear_standby %p and ++connect_seq\n", con);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002885 con->state = CON_STATE_PREOPEN;
Sage Weile00de342011-03-04 12:25:05 -08002886 con->connect_seq++;
Alex Elderc9ffc772013-02-20 10:25:12 -06002887 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2888 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
Sage Weile00de342011-03-04 12:25:05 -08002889 }
2890}
2891
Sage Weil31b80062009-10-06 11:31:13 -07002892/*
2893 * Queue up an outgoing message on the given connection.
2894 */
2895void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2896{
Sage Weila59b55a2012-07-20 15:34:04 -07002897 /* set src+dst */
2898 msg->hdr.src = con->msgr->inst.name;
2899 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2900 msg->needs_out_seq = true;
2901
2902 mutex_lock(&con->mutex);
2903
Sage Weil8dacc7d2012-07-20 17:24:40 -07002904 if (con->state == CON_STATE_CLOSED) {
Sage Weil31b80062009-10-06 11:31:13 -07002905 dout("con_send %p closed, dropping %p\n", con, msg);
2906 ceph_msg_put(msg);
Sage Weila59b55a2012-07-20 15:34:04 -07002907 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002908 return;
2909 }
2910
Alex Elder38941f82012-06-01 14:56:43 -05002911 BUG_ON(msg->con != NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -07002912 msg->con = con->ops->get(con);
Alex Elder92ce0342012-06-04 14:43:33 -05002913 BUG_ON(msg->con == NULL);
Sage Weil31b80062009-10-06 11:31:13 -07002914
Sage Weil31b80062009-10-06 11:31:13 -07002915 BUG_ON(!list_empty(&msg->list_head));
2916 list_add_tail(&msg->list_head, &con->out_queue);
2917 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2918 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2919 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2920 le32_to_cpu(msg->hdr.front_len),
2921 le32_to_cpu(msg->hdr.middle_len),
2922 le32_to_cpu(msg->hdr.data_len));
Sage Weil00650932012-07-20 15:33:04 -07002923
2924 clear_standby(con);
Sage Weilec302642009-12-22 10:43:42 -08002925 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002926
2927 /* if there wasn't anything waiting to send before, queue
2928 * new work */
Alex Elderc9ffc772013-02-20 10:25:12 -06002929 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
Sage Weil31b80062009-10-06 11:31:13 -07002930 queue_con(con);
2931}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002932EXPORT_SYMBOL(ceph_con_send);
Sage Weil31b80062009-10-06 11:31:13 -07002933
2934/*
2935 * Revoke a message that was previously queued for send
2936 */
Alex Elder6740a842012-06-01 14:56:43 -05002937void ceph_msg_revoke(struct ceph_msg *msg)
Sage Weil31b80062009-10-06 11:31:13 -07002938{
Alex Elder6740a842012-06-01 14:56:43 -05002939 struct ceph_connection *con = msg->con;
2940
2941 if (!con)
2942 return; /* Message not in our possession */
2943
Sage Weilec302642009-12-22 10:43:42 -08002944 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002945 if (!list_empty(&msg->list_head)) {
Alex Elder38941f82012-06-01 14:56:43 -05002946 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07002947 list_del_init(&msg->list_head);
Alex Elder38941f82012-06-01 14:56:43 -05002948 BUG_ON(msg->con == NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -07002949 msg->con->ops->put(msg->con);
Alex Elder38941f82012-06-01 14:56:43 -05002950 msg->con = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002951 msg->hdr.seq = 0;
Alex Elder38941f82012-06-01 14:56:43 -05002952
Sage Weil31b80062009-10-06 11:31:13 -07002953 ceph_msg_put(msg);
Sage Weiled98ada2010-07-05 12:15:14 -07002954 }
2955 if (con->out_msg == msg) {
Alex Elder38941f82012-06-01 14:56:43 -05002956 dout("%s %p msg %p - was sending\n", __func__, con, msg);
Sage Weiled98ada2010-07-05 12:15:14 -07002957 con->out_msg = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002958 if (con->out_kvec_is_msg) {
2959 con->out_skip = con->out_kvec_bytes;
2960 con->out_kvec_is_msg = false;
2961 }
Sage Weiled98ada2010-07-05 12:15:14 -07002962 msg->hdr.seq = 0;
Alex Elder92ce0342012-06-04 14:43:33 -05002963
2964 ceph_msg_put(msg);
Sage Weil31b80062009-10-06 11:31:13 -07002965 }
Sage Weilec302642009-12-22 10:43:42 -08002966 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002967}
2968
2969/*
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002970 * Revoke a message that we may be reading data into
Sage Weil350b1c32009-12-22 10:45:45 -08002971 */
Alex Elder8921d112012-06-01 14:56:43 -05002972void ceph_msg_revoke_incoming(struct ceph_msg *msg)
Sage Weil350b1c32009-12-22 10:45:45 -08002973{
Alex Elder8921d112012-06-01 14:56:43 -05002974 struct ceph_connection *con;
2975
2976 BUG_ON(msg == NULL);
2977 if (!msg->con) {
2978 dout("%s msg %p null con\n", __func__, msg);
2979
2980 return; /* Message not in our possession */
2981 }
2982
2983 con = msg->con;
Sage Weil350b1c32009-12-22 10:45:45 -08002984 mutex_lock(&con->mutex);
Alex Elder8921d112012-06-01 14:56:43 -05002985 if (con->in_msg == msg) {
Eric Dumazet95c96172012-04-15 05:58:06 +00002986 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2987 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2988 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
Sage Weil350b1c32009-12-22 10:45:45 -08002989
2990 /* skip rest of message */
Alex Elder8921d112012-06-01 14:56:43 -05002991 dout("%s %p msg %p revoked\n", __func__, con, msg);
2992 con->in_base_pos = con->in_base_pos -
Sage Weil350b1c32009-12-22 10:45:45 -08002993 sizeof(struct ceph_msg_header) -
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002994 front_len -
2995 middle_len -
2996 data_len -
Sage Weil350b1c32009-12-22 10:45:45 -08002997 sizeof(struct ceph_msg_footer);
Sage Weil350b1c32009-12-22 10:45:45 -08002998 ceph_msg_put(con->in_msg);
2999 con->in_msg = NULL;
3000 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weil684be252010-04-21 20:45:59 -07003001 con->in_seq++;
Sage Weil350b1c32009-12-22 10:45:45 -08003002 } else {
Alex Elder8921d112012-06-01 14:56:43 -05003003 dout("%s %p in_msg %p msg %p no-op\n",
3004 __func__, con, con->in_msg, msg);
Sage Weil350b1c32009-12-22 10:45:45 -08003005 }
3006 mutex_unlock(&con->mutex);
3007}
3008
3009/*
Sage Weil31b80062009-10-06 11:31:13 -07003010 * Queue a keepalive byte to ensure the tcp connection is alive.
3011 */
3012void ceph_con_keepalive(struct ceph_connection *con)
3013{
Sage Weile00de342011-03-04 12:25:05 -08003014 dout("con_keepalive %p\n", con);
Sage Weil00650932012-07-20 15:33:04 -07003015 mutex_lock(&con->mutex);
Sage Weile00de342011-03-04 12:25:05 -08003016 clear_standby(con);
Sage Weil00650932012-07-20 15:33:04 -07003017 mutex_unlock(&con->mutex);
Alex Elderc9ffc772013-02-20 10:25:12 -06003018 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3019 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
Sage Weil31b80062009-10-06 11:31:13 -07003020 queue_con(con);
3021}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003022EXPORT_SYMBOL(ceph_con_keepalive);
Sage Weil31b80062009-10-06 11:31:13 -07003023
Alex Elder6644ed72013-03-11 23:34:24 -05003024static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
Alex Elder43794502013-03-01 18:00:16 -06003025{
Alex Elder6644ed72013-03-11 23:34:24 -05003026 struct ceph_msg_data *data;
3027
3028 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3029 return NULL;
3030
Alex Elder81b36be2013-05-01 12:43:04 -05003031 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
Alex Elder6644ed72013-03-11 23:34:24 -05003032 if (data)
3033 data->type = type;
Alex Elder5240d9f2013-03-14 14:09:06 -05003034 INIT_LIST_HEAD(&data->links);
Alex Elder6644ed72013-03-11 23:34:24 -05003035
3036 return data;
3037}
3038
3039static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3040{
3041 if (!data)
3042 return;
3043
Alex Elder5240d9f2013-03-14 14:09:06 -05003044 WARN_ON(!list_empty(&data->links));
Alex Elder6644ed72013-03-11 23:34:24 -05003045 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3046 ceph_pagelist_release(data->pagelist);
3047 kfree(data->pagelist);
3048 }
Alex Elder81b36be2013-05-01 12:43:04 -05003049 kmem_cache_free(ceph_msg_data_cache, data);
Alex Elder43794502013-03-01 18:00:16 -06003050}
3051
Alex Elder90af3602013-04-05 14:46:01 -05003052void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
Alex Elderf1baeb22013-03-07 15:38:26 -06003053 size_t length, size_t alignment)
Alex Elder02afca62013-02-14 12:16:43 -06003054{
Alex Elder6644ed72013-03-11 23:34:24 -05003055 struct ceph_msg_data *data;
3056
Alex Elder07aa1552013-03-04 18:29:06 -06003057 BUG_ON(!pages);
3058 BUG_ON(!length);
Alex Elder02afca62013-02-14 12:16:43 -06003059
Alex Elder6644ed72013-03-11 23:34:24 -05003060 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3061 BUG_ON(!data);
3062 data->pages = pages;
3063 data->length = length;
3064 data->alignment = alignment & ~PAGE_MASK;
3065
Alex Elder5240d9f2013-03-14 14:09:06 -05003066 list_add_tail(&data->links, &msg->data);
3067 msg->data_length += length;
Alex Elder02afca62013-02-14 12:16:43 -06003068}
Alex Elder90af3602013-04-05 14:46:01 -05003069EXPORT_SYMBOL(ceph_msg_data_add_pages);
Sage Weil31b80062009-10-06 11:31:13 -07003070
Alex Elder90af3602013-04-05 14:46:01 -05003071void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
Alex Elder27fa8382013-02-14 12:16:43 -06003072 struct ceph_pagelist *pagelist)
3073{
Alex Elder6644ed72013-03-11 23:34:24 -05003074 struct ceph_msg_data *data;
3075
Alex Elder07aa1552013-03-04 18:29:06 -06003076 BUG_ON(!pagelist);
3077 BUG_ON(!pagelist->length);
Alex Elder27fa8382013-02-14 12:16:43 -06003078
Alex Elder6644ed72013-03-11 23:34:24 -05003079 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3080 BUG_ON(!data);
3081 data->pagelist = pagelist;
3082
Alex Elder5240d9f2013-03-14 14:09:06 -05003083 list_add_tail(&data->links, &msg->data);
3084 msg->data_length += pagelist->length;
Alex Elder27fa8382013-02-14 12:16:43 -06003085}
Alex Elder90af3602013-04-05 14:46:01 -05003086EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
Alex Elder27fa8382013-02-14 12:16:43 -06003087
Alex Elderea965712013-04-05 14:46:01 -05003088#ifdef CONFIG_BLOCK
Alex Elder90af3602013-04-05 14:46:01 -05003089void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
Alex Eldera1930802013-03-14 14:09:06 -05003090 size_t length)
Alex Elder27fa8382013-02-14 12:16:43 -06003091{
Alex Elder6644ed72013-03-11 23:34:24 -05003092 struct ceph_msg_data *data;
Alex Elder27fa8382013-02-14 12:16:43 -06003093
Alex Elder6644ed72013-03-11 23:34:24 -05003094 BUG_ON(!bio);
Alex Elder6644ed72013-03-11 23:34:24 -05003095
3096 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3097 BUG_ON(!data);
3098 data->bio = bio;
Alex Elderc851c492013-04-05 14:46:01 -05003099 data->bio_length = length;
Alex Elder6644ed72013-03-11 23:34:24 -05003100
Alex Elder5240d9f2013-03-14 14:09:06 -05003101 list_add_tail(&data->links, &msg->data);
3102 msg->data_length += length;
Alex Elder27fa8382013-02-14 12:16:43 -06003103}
Alex Elder90af3602013-04-05 14:46:01 -05003104EXPORT_SYMBOL(ceph_msg_data_add_bio);
Alex Elderea965712013-04-05 14:46:01 -05003105#endif /* CONFIG_BLOCK */
Alex Elder27fa8382013-02-14 12:16:43 -06003106
Sage Weil31b80062009-10-06 11:31:13 -07003107/*
3108 * construct a new message with given type, size
3109 * the new msg has a ref count of 1.
3110 */
Sage Weilb61c2762011-08-09 15:03:46 -07003111struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3112 bool can_fail)
Sage Weil31b80062009-10-06 11:31:13 -07003113{
3114 struct ceph_msg *m;
3115
Alex Eldere3d5d632013-05-01 12:43:04 -05003116 m = kmem_cache_zalloc(ceph_msg_cache, flags);
Sage Weil31b80062009-10-06 11:31:13 -07003117 if (m == NULL)
3118 goto out;
Alex Elder38941f82012-06-01 14:56:43 -05003119
Sage Weil31b80062009-10-06 11:31:13 -07003120 m->hdr.type = cpu_to_le16(type);
Sage Weil45c6ceb2010-05-11 15:01:51 -07003121 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
Sage Weil31b80062009-10-06 11:31:13 -07003122 m->hdr.front_len = cpu_to_le32(front_len);
Sage Weil31b80062009-10-06 11:31:13 -07003123
Alex Elder9516e452013-03-01 18:00:16 -06003124 INIT_LIST_HEAD(&m->list_head);
3125 kref_init(&m->kref);
Alex Elder5240d9f2013-03-14 14:09:06 -05003126 INIT_LIST_HEAD(&m->data);
Henry C Changca208922011-05-03 02:29:56 +00003127
Sage Weil31b80062009-10-06 11:31:13 -07003128 /* front */
Alex Elder9516e452013-03-01 18:00:16 -06003129 m->front_max = front_len;
Sage Weil31b80062009-10-06 11:31:13 -07003130 if (front_len) {
3131 if (front_len > PAGE_CACHE_SIZE) {
Yehuda Sadeh34d23762010-04-06 14:33:58 -07003132 m->front.iov_base = __vmalloc(front_len, flags,
Sage Weil31b80062009-10-06 11:31:13 -07003133 PAGE_KERNEL);
3134 m->front_is_vmalloc = true;
3135 } else {
Yehuda Sadeh34d23762010-04-06 14:33:58 -07003136 m->front.iov_base = kmalloc(front_len, flags);
Sage Weil31b80062009-10-06 11:31:13 -07003137 }
3138 if (m->front.iov_base == NULL) {
Sage Weilb61c2762011-08-09 15:03:46 -07003139 dout("ceph_msg_new can't allocate %d bytes\n",
Sage Weil31b80062009-10-06 11:31:13 -07003140 front_len);
3141 goto out2;
3142 }
3143 } else {
3144 m->front.iov_base = NULL;
3145 }
3146 m->front.iov_len = front_len;
3147
Sage Weilbb257662010-04-01 16:07:23 -07003148 dout("ceph_msg_new %p front %d\n", m, front_len);
Sage Weil31b80062009-10-06 11:31:13 -07003149 return m;
3150
3151out2:
3152 ceph_msg_put(m);
3153out:
Sage Weilb61c2762011-08-09 15:03:46 -07003154 if (!can_fail) {
3155 pr_err("msg_new can't create type %d front %d\n", type,
3156 front_len);
Sage Weilf0ed1b72011-08-09 15:05:07 -07003157 WARN_ON(1);
Sage Weilb61c2762011-08-09 15:03:46 -07003158 } else {
3159 dout("msg_new can't create type %d front %d\n", type,
3160 front_len);
3161 }
Sage Weila79832f2010-04-01 16:06:19 -07003162 return NULL;
Sage Weil31b80062009-10-06 11:31:13 -07003163}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003164EXPORT_SYMBOL(ceph_msg_new);
Sage Weil31b80062009-10-06 11:31:13 -07003165
3166/*
Sage Weil31b80062009-10-06 11:31:13 -07003167 * Allocate "middle" portion of a message, if it is needed and wasn't
3168 * allocated by alloc_msg. This allows us to read a small fixed-size
3169 * per-type header in the front and then gracefully fail (i.e.,
3170 * propagate the error to the caller based on info in the front) when
3171 * the middle is too large.
3172 */
Yehuda Sadeh24504182010-01-08 13:58:34 -08003173static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
Sage Weil31b80062009-10-06 11:31:13 -07003174{
3175 int type = le16_to_cpu(msg->hdr.type);
3176 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3177
3178 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3179 ceph_msg_type_name(type), middle_len);
3180 BUG_ON(!middle_len);
3181 BUG_ON(msg->middle);
3182
Sage Weilb6c1d5b2009-12-07 12:17:17 -08003183 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
Sage Weil31b80062009-10-06 11:31:13 -07003184 if (!msg->middle)
3185 return -ENOMEM;
3186 return 0;
3187}
3188
Yehuda Sadeh24504182010-01-08 13:58:34 -08003189/*
Alex Elder1c20f2d2012-06-04 14:43:32 -05003190 * Allocate a message for receiving an incoming message on a
3191 * connection, and save the result in con->in_msg. Uses the
3192 * connection's private alloc_msg op if available.
3193 *
Sage Weil4740a622012-07-30 18:19:30 -07003194 * Returns 0 on success, or a negative error code.
3195 *
3196 * On success, if we set *skip = 1:
3197 * - the next message should be skipped and ignored.
3198 * - con->in_msg == NULL
3199 * or if we set *skip = 0:
3200 * - con->in_msg is non-null.
3201 * On error (ENOMEM, EAGAIN, ...),
3202 * - con->in_msg == NULL
Yehuda Sadeh24504182010-01-08 13:58:34 -08003203 */
Sage Weil4740a622012-07-30 18:19:30 -07003204static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
Yehuda Sadeh24504182010-01-08 13:58:34 -08003205{
Sage Weil4740a622012-07-30 18:19:30 -07003206 struct ceph_msg_header *hdr = &con->in_hdr;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003207 int middle_len = le32_to_cpu(hdr->middle_len);
Alex Elder1d866d12013-03-01 18:00:14 -06003208 struct ceph_msg *msg;
Sage Weil4740a622012-07-30 18:19:30 -07003209 int ret = 0;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003210
Alex Elder1c20f2d2012-06-04 14:43:32 -05003211 BUG_ON(con->in_msg != NULL);
Alex Elder53ded492013-03-01 18:00:14 -06003212 BUG_ON(!con->ops->alloc_msg);
Yehuda Sadeh24504182010-01-08 13:58:34 -08003213
Alex Elder53ded492013-03-01 18:00:14 -06003214 mutex_unlock(&con->mutex);
3215 msg = con->ops->alloc_msg(con, hdr, skip);
3216 mutex_lock(&con->mutex);
3217 if (con->state != CON_STATE_OPEN) {
3218 if (msg)
Alex Elder1d866d12013-03-01 18:00:14 -06003219 ceph_msg_put(msg);
Alex Elder53ded492013-03-01 18:00:14 -06003220 return -EAGAIN;
3221 }
Alex Elder41375772013-03-05 09:25:10 -06003222 if (msg) {
3223 BUG_ON(*skip);
3224 con->in_msg = msg;
Sage Weil36eb71a2012-06-21 12:47:08 -07003225 con->in_msg->con = con->ops->get(con);
Alex Elder92ce0342012-06-04 14:43:33 -05003226 BUG_ON(con->in_msg->con == NULL);
Alex Elder41375772013-03-05 09:25:10 -06003227 } else {
3228 /*
3229 * Null message pointer means either we should skip
3230 * this message or we couldn't allocate memory. The
3231 * former is not an error.
3232 */
3233 if (*skip)
3234 return 0;
3235 con->error_msg = "error allocating memory for incoming message";
3236
Alex Elder53ded492013-03-01 18:00:14 -06003237 return -ENOMEM;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003238 }
Alex Elder1c20f2d2012-06-04 14:43:32 -05003239 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
Yehuda Sadeh24504182010-01-08 13:58:34 -08003240
Alex Elder1c20f2d2012-06-04 14:43:32 -05003241 if (middle_len && !con->in_msg->middle) {
3242 ret = ceph_alloc_middle(con, con->in_msg);
Yehuda Sadeh24504182010-01-08 13:58:34 -08003243 if (ret < 0) {
Alex Elder1c20f2d2012-06-04 14:43:32 -05003244 ceph_msg_put(con->in_msg);
3245 con->in_msg = NULL;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003246 }
3247 }
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08003248
Sage Weil4740a622012-07-30 18:19:30 -07003249 return ret;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003250}
3251
Sage Weil31b80062009-10-06 11:31:13 -07003252
3253/*
3254 * Free a generically kmalloc'd message.
3255 */
3256void ceph_msg_kfree(struct ceph_msg *m)
3257{
3258 dout("msg_kfree %p\n", m);
3259 if (m->front_is_vmalloc)
3260 vfree(m->front.iov_base);
3261 else
3262 kfree(m->front.iov_base);
Alex Eldere3d5d632013-05-01 12:43:04 -05003263 kmem_cache_free(ceph_msg_cache, m);
Sage Weil31b80062009-10-06 11:31:13 -07003264}
3265
3266/*
3267 * Drop a msg ref. Destroy as needed.
3268 */
Sage Weilc2e552e2009-12-07 15:55:05 -08003269void ceph_msg_last_put(struct kref *kref)
Sage Weil31b80062009-10-06 11:31:13 -07003270{
Sage Weilc2e552e2009-12-07 15:55:05 -08003271 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
Alex Elder5240d9f2013-03-14 14:09:06 -05003272 LIST_HEAD(data);
3273 struct list_head *links;
3274 struct list_head *next;
Sage Weil31b80062009-10-06 11:31:13 -07003275
Sage Weilc2e552e2009-12-07 15:55:05 -08003276 dout("ceph_msg_put last one on %p\n", m);
3277 WARN_ON(!list_empty(&m->list_head));
Sage Weil31b80062009-10-06 11:31:13 -07003278
Sage Weilc2e552e2009-12-07 15:55:05 -08003279 /* drop middle, data, if any */
3280 if (m->middle) {
3281 ceph_buffer_put(m->middle);
3282 m->middle = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07003283 }
Alex Elder5240d9f2013-03-14 14:09:06 -05003284
3285 list_splice_init(&m->data, &data);
3286 list_for_each_safe(links, next, &data) {
3287 struct ceph_msg_data *data;
3288
3289 data = list_entry(links, struct ceph_msg_data, links);
3290 list_del_init(links);
3291 ceph_msg_data_destroy(data);
3292 }
Alex Eldera1930802013-03-14 14:09:06 -05003293 m->data_length = 0;
Sage Weil58bb3b32009-12-23 12:12:31 -08003294
Sage Weilc2e552e2009-12-07 15:55:05 -08003295 if (m->pool)
3296 ceph_msgpool_put(m->pool, m);
3297 else
3298 ceph_msg_kfree(m);
Sage Weil31b80062009-10-06 11:31:13 -07003299}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003300EXPORT_SYMBOL(ceph_msg_last_put);
Sage Weil9ec7cab2009-12-14 15:13:47 -08003301
3302void ceph_msg_dump(struct ceph_msg *msg)
3303{
Alex Elder4a73ef22013-03-07 15:38:26 -06003304 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
Alex Elder5240d9f2013-03-14 14:09:06 -05003305 msg->front_max, msg->data_length);
Sage Weil9ec7cab2009-12-14 15:13:47 -08003306 print_hex_dump(KERN_DEBUG, "header: ",
3307 DUMP_PREFIX_OFFSET, 16, 1,
3308 &msg->hdr, sizeof(msg->hdr), true);
3309 print_hex_dump(KERN_DEBUG, " front: ",
3310 DUMP_PREFIX_OFFSET, 16, 1,
3311 msg->front.iov_base, msg->front.iov_len, true);
3312 if (msg->middle)
3313 print_hex_dump(KERN_DEBUG, "middle: ",
3314 DUMP_PREFIX_OFFSET, 16, 1,
3315 msg->middle->vec.iov_base,
3316 msg->middle->vec.iov_len, true);
3317 print_hex_dump(KERN_DEBUG, "footer: ",
3318 DUMP_PREFIX_OFFSET, 16, 1,
3319 &msg->footer, sizeof(msg->footer), true);
3320}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003321EXPORT_SYMBOL(ceph_msg_dump);