blob: e3be1d22a2477dd2d9e271a4594d70ebdb1148e7 [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>
Ilya Dryomov757856d2015-06-25 17:47:45 +03009#include <linux/nsproxy.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090010#include <linux/slab.h>
Sage Weil31b80062009-10-06 11:31:13 -070011#include <linux/socket.h>
12#include <linux/string.h>
Alex Elder3ebc21f2013-01-31 16:02:01 -060013#ifdef CONFIG_BLOCK
Yehuda Sadeh68b44762010-04-06 15:01:27 -070014#include <linux/bio.h>
Alex Elder3ebc21f2013-01-31 16:02:01 -060015#endif /* CONFIG_BLOCK */
Noah Watkinsee3b56f2011-09-23 11:48:42 -070016#include <linux/dns_resolver.h>
Sage Weil31b80062009-10-06 11:31:13 -070017#include <net/tcp.h>
18
Ilya Dryomov2b3e0c92013-12-24 21:19:24 +020019#include <linux/ceph/ceph_features.h>
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070020#include <linux/ceph/libceph.h>
21#include <linux/ceph/messenger.h>
22#include <linux/ceph/decode.h>
23#include <linux/ceph/pagelist.h>
Paul Gortmakerbc3b2d72011-07-15 11:47:34 -040024#include <linux/export.h>
Sage Weil31b80062009-10-06 11:31:13 -070025
Alex Elderfe38a2b2013-03-06 23:39:39 -060026#define list_entry_next(pos, member) \
27 list_entry(pos->member.next, typeof(*pos), member)
28
Sage Weil31b80062009-10-06 11:31:13 -070029/*
30 * Ceph uses the messenger to exchange ceph_msg messages with other
31 * hosts in the system. The messenger provides ordered and reliable
32 * delivery. We tolerate TCP disconnects by reconnecting (with
33 * exponential backoff) in the case of a fault (disconnection, bad
34 * crc, protocol error). Acks allow sent messages to be discarded by
35 * the sender.
36 */
37
Alex Elderbc18f4b2012-06-20 21:53:53 -050038/*
39 * We track the state of the socket on a given connection using
40 * values defined below. The transition to a new socket state is
41 * handled by a function which verifies we aren't coming from an
42 * unexpected state.
43 *
44 * --------
45 * | NEW* | transient initial state
46 * --------
47 * | con_sock_state_init()
48 * v
49 * ----------
50 * | CLOSED | initialized, but no socket (and no
51 * ---------- TCP connection)
52 * ^ \
53 * | \ con_sock_state_connecting()
54 * | ----------------------
55 * | \
56 * + con_sock_state_closed() \
Sage Weilfbb85a42012-06-27 12:31:02 -070057 * |+--------------------------- \
58 * | \ \ \
59 * | ----------- \ \
60 * | | CLOSING | socket event; \ \
61 * | ----------- await close \ \
62 * | ^ \ |
63 * | | \ |
64 * | + con_sock_state_closing() \ |
65 * | / \ | |
66 * | / --------------- | |
67 * | / \ v v
Alex Elderbc18f4b2012-06-20 21:53:53 -050068 * | / --------------
69 * | / -----------------| CONNECTING | socket created, TCP
70 * | | / -------------- connect initiated
71 * | | | con_sock_state_connected()
72 * | | v
73 * -------------
74 * | CONNECTED | TCP connection established
75 * -------------
76 *
77 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
78 */
Alex Elderce2c8902012-05-22 22:15:49 -050079
80#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
81#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
82#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
83#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
84#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
85
Sage Weil8dacc7d2012-07-20 17:24:40 -070086/*
87 * connection states
88 */
89#define CON_STATE_CLOSED 1 /* -> PREOPEN */
90#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
91#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
92#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
93#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
94#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
95
Sage Weil4a861692012-07-20 17:29:55 -070096/*
97 * ceph_connection flag bits
98 */
99#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
100 * messages on errors */
101#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
102#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
103#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
104#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
Sage Weil8dacc7d2012-07-20 17:24:40 -0700105
Alex Elderc9ffc772013-02-20 10:25:12 -0600106static bool con_flag_valid(unsigned long con_flag)
107{
108 switch (con_flag) {
109 case CON_FLAG_LOSSYTX:
110 case CON_FLAG_KEEPALIVE_PENDING:
111 case CON_FLAG_WRITE_PENDING:
112 case CON_FLAG_SOCK_CLOSED:
113 case CON_FLAG_BACKOFF:
114 return true;
115 default:
116 return false;
117 }
118}
119
120static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
121{
122 BUG_ON(!con_flag_valid(con_flag));
123
124 clear_bit(con_flag, &con->flags);
125}
126
127static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
128{
129 BUG_ON(!con_flag_valid(con_flag));
130
131 set_bit(con_flag, &con->flags);
132}
133
134static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
135{
136 BUG_ON(!con_flag_valid(con_flag));
137
138 return test_bit(con_flag, &con->flags);
139}
140
141static bool con_flag_test_and_clear(struct ceph_connection *con,
142 unsigned long con_flag)
143{
144 BUG_ON(!con_flag_valid(con_flag));
145
146 return test_and_clear_bit(con_flag, &con->flags);
147}
148
149static bool con_flag_test_and_set(struct ceph_connection *con,
150 unsigned long con_flag)
151{
152 BUG_ON(!con_flag_valid(con_flag));
153
154 return test_and_set_bit(con_flag, &con->flags);
155}
156
Alex Eldere3d5d632013-05-01 12:43:04 -0500157/* Slab caches for frequently-allocated structures */
158
159static struct kmem_cache *ceph_msg_cache;
Alex Elder81b36be2013-05-01 12:43:04 -0500160static struct kmem_cache *ceph_msg_data_cache;
Alex Eldere3d5d632013-05-01 12:43:04 -0500161
Sage Weil31b80062009-10-06 11:31:13 -0700162/* static tag bytes (protocol control messages) */
163static char tag_msg = CEPH_MSGR_TAG_MSG;
164static char tag_ack = CEPH_MSGR_TAG_ACK;
165static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
166
Sage Weila6a53492010-04-13 14:07:07 -0700167#ifdef CONFIG_LOCKDEP
168static struct lock_class_key socket_class;
169#endif
170
Alex Elder84495f42012-02-15 07:43:55 -0600171/*
172 * When skipping (ignoring) a block of input we read it into a "skip
173 * buffer," which is this many bytes in size.
174 */
175#define SKIP_BUF_SIZE 1024
Sage Weil31b80062009-10-06 11:31:13 -0700176
177static void queue_con(struct ceph_connection *con);
Ilya Dryomov37ab77a2014-06-24 16:21:45 +0400178static void cancel_con(struct ceph_connection *con);
Sage Weil31b80062009-10-06 11:31:13 -0700179static void con_work(struct work_struct *);
Alex Elder93209262013-02-19 12:25:57 -0600180static void con_fault(struct ceph_connection *con);
Sage Weil31b80062009-10-06 11:31:13 -0700181
Sage Weil31b80062009-10-06 11:31:13 -0700182/*
Alex Elderf64a9312012-01-23 15:49:27 -0600183 * Nicely render a sockaddr as a string. An array of formatted
184 * strings is used, to approximate reentrancy.
Sage Weil31b80062009-10-06 11:31:13 -0700185 */
Alex Elderf64a9312012-01-23 15:49:27 -0600186#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
187#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
188#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
189#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
190
191static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
192static atomic_t addr_str_seq = ATOMIC_INIT(0);
Sage Weil31b80062009-10-06 11:31:13 -0700193
Alex Elder57666512012-01-23 15:49:27 -0600194static struct page *zero_page; /* used in certain error cases */
Alex Elder57666512012-01-23 15:49:27 -0600195
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700196const char *ceph_pr_addr(const struct sockaddr_storage *ss)
Sage Weil31b80062009-10-06 11:31:13 -0700197{
198 int i;
199 char *s;
Alex Elder99f0f3b2012-01-23 15:49:27 -0600200 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
201 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
Sage Weil31b80062009-10-06 11:31:13 -0700202
Alex Elderf64a9312012-01-23 15:49:27 -0600203 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
Sage Weil31b80062009-10-06 11:31:13 -0700204 s = addr_str[i];
205
206 switch (ss->ss_family) {
207 case AF_INET:
Alex Elderbd406142012-01-23 15:49:27 -0600208 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
209 ntohs(in4->sin_port));
Sage Weil31b80062009-10-06 11:31:13 -0700210 break;
211
212 case AF_INET6:
Alex Elderbd406142012-01-23 15:49:27 -0600213 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
214 ntohs(in6->sin6_port));
Sage Weil31b80062009-10-06 11:31:13 -0700215 break;
216
217 default:
Alex Elderd3002b92012-02-14 14:05:33 -0600218 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
219 ss->ss_family);
Sage Weil31b80062009-10-06 11:31:13 -0700220 }
221
222 return s;
223}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700224EXPORT_SYMBOL(ceph_pr_addr);
Sage Weil31b80062009-10-06 11:31:13 -0700225
Sage Weil63f2d212009-11-03 15:17:56 -0800226static void encode_my_addr(struct ceph_messenger *msgr)
227{
228 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
229 ceph_encode_addr(&msgr->my_enc_addr);
230}
231
Sage Weil31b80062009-10-06 11:31:13 -0700232/*
233 * work queue for all reading and writing to/from the socket.
234 */
Alex Eldere0f43c92012-02-14 14:05:33 -0600235static struct workqueue_struct *ceph_msgr_wq;
Sage Weil31b80062009-10-06 11:31:13 -0700236
Alex Eldere3d5d632013-05-01 12:43:04 -0500237static int ceph_msgr_slab_init(void)
238{
239 BUG_ON(ceph_msg_cache);
240 ceph_msg_cache = kmem_cache_create("ceph_msg",
241 sizeof (struct ceph_msg),
242 __alignof__(struct ceph_msg), 0, NULL);
Alex Elder81b36be2013-05-01 12:43:04 -0500243
244 if (!ceph_msg_cache)
245 return -ENOMEM;
246
247 BUG_ON(ceph_msg_data_cache);
248 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
249 sizeof (struct ceph_msg_data),
250 __alignof__(struct ceph_msg_data),
251 0, NULL);
252 if (ceph_msg_data_cache)
253 return 0;
254
255 kmem_cache_destroy(ceph_msg_cache);
256 ceph_msg_cache = NULL;
257
258 return -ENOMEM;
Alex Eldere3d5d632013-05-01 12:43:04 -0500259}
260
261static void ceph_msgr_slab_exit(void)
262{
Alex Elder81b36be2013-05-01 12:43:04 -0500263 BUG_ON(!ceph_msg_data_cache);
264 kmem_cache_destroy(ceph_msg_data_cache);
265 ceph_msg_data_cache = NULL;
266
Alex Eldere3d5d632013-05-01 12:43:04 -0500267 BUG_ON(!ceph_msg_cache);
268 kmem_cache_destroy(ceph_msg_cache);
269 ceph_msg_cache = NULL;
270}
271
Alex Elder15417162013-02-19 12:25:56 -0600272static void _ceph_msgr_exit(void)
Alex Elder6173d1f2012-02-14 14:05:33 -0600273{
Alex Elderd3002b92012-02-14 14:05:33 -0600274 if (ceph_msgr_wq) {
Alex Elder6173d1f2012-02-14 14:05:33 -0600275 destroy_workqueue(ceph_msgr_wq);
Alex Elderd3002b92012-02-14 14:05:33 -0600276 ceph_msgr_wq = NULL;
277 }
Alex Elder6173d1f2012-02-14 14:05:33 -0600278
Alex Eldere3d5d632013-05-01 12:43:04 -0500279 ceph_msgr_slab_exit();
280
Alex Elder6173d1f2012-02-14 14:05:33 -0600281 BUG_ON(zero_page == NULL);
Alex Elder6173d1f2012-02-14 14:05:33 -0600282 page_cache_release(zero_page);
283 zero_page = NULL;
284}
285
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700286int ceph_msgr_init(void)
Sage Weil31b80062009-10-06 11:31:13 -0700287{
Alex Elder57666512012-01-23 15:49:27 -0600288 BUG_ON(zero_page != NULL);
289 zero_page = ZERO_PAGE(0);
290 page_cache_get(zero_page);
291
Alex Eldere3d5d632013-05-01 12:43:04 -0500292 if (ceph_msgr_slab_init())
293 return -ENOMEM;
294
Ilya Dryomovf9865f02014-10-10 16:39:05 +0400295 /*
296 * The number of active work items is limited by the number of
297 * connections, so leave @max_active at default.
298 */
299 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
Alex Elder6173d1f2012-02-14 14:05:33 -0600300 if (ceph_msgr_wq)
301 return 0;
Alex Elder57666512012-01-23 15:49:27 -0600302
Alex Elder6173d1f2012-02-14 14:05:33 -0600303 pr_err("msgr_init failed to create workqueue\n");
304 _ceph_msgr_exit();
Alex Elder57666512012-01-23 15:49:27 -0600305
Alex Elder6173d1f2012-02-14 14:05:33 -0600306 return -ENOMEM;
Sage Weil31b80062009-10-06 11:31:13 -0700307}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700308EXPORT_SYMBOL(ceph_msgr_init);
Sage Weil31b80062009-10-06 11:31:13 -0700309
310void ceph_msgr_exit(void)
311{
Alex Elder57666512012-01-23 15:49:27 -0600312 BUG_ON(ceph_msgr_wq == NULL);
Alex Elder57666512012-01-23 15:49:27 -0600313
Alex Elder6173d1f2012-02-14 14:05:33 -0600314 _ceph_msgr_exit();
Sage Weil31b80062009-10-06 11:31:13 -0700315}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700316EXPORT_SYMBOL(ceph_msgr_exit);
Sage Weil31b80062009-10-06 11:31:13 -0700317
Yehuda Sadehcd84db62010-06-11 16:58:48 -0700318void ceph_msgr_flush(void)
Sage Weila922d382010-05-29 09:41:23 -0700319{
320 flush_workqueue(ceph_msgr_wq);
321}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700322EXPORT_SYMBOL(ceph_msgr_flush);
Sage Weila922d382010-05-29 09:41:23 -0700323
Alex Elderce2c8902012-05-22 22:15:49 -0500324/* Connection socket state transition functions */
325
326static void con_sock_state_init(struct ceph_connection *con)
327{
328 int old_state;
329
330 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
331 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
332 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700333 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
334 CON_SOCK_STATE_CLOSED);
Alex Elderce2c8902012-05-22 22:15:49 -0500335}
336
337static void con_sock_state_connecting(struct ceph_connection *con)
338{
339 int old_state;
340
341 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
342 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
343 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700344 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
345 CON_SOCK_STATE_CONNECTING);
Alex Elderce2c8902012-05-22 22:15:49 -0500346}
347
348static void con_sock_state_connected(struct ceph_connection *con)
349{
350 int old_state;
351
352 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
353 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
354 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700355 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
356 CON_SOCK_STATE_CONNECTED);
Alex Elderce2c8902012-05-22 22:15:49 -0500357}
358
359static void con_sock_state_closing(struct ceph_connection *con)
360{
361 int old_state;
362
363 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
364 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
365 old_state != CON_SOCK_STATE_CONNECTED &&
366 old_state != CON_SOCK_STATE_CLOSING))
367 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700368 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
369 CON_SOCK_STATE_CLOSING);
Alex Elderce2c8902012-05-22 22:15:49 -0500370}
371
372static void con_sock_state_closed(struct ceph_connection *con)
373{
374 int old_state;
375
376 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
377 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
Sage Weilfbb85a42012-06-27 12:31:02 -0700378 old_state != CON_SOCK_STATE_CLOSING &&
Sage Weil8007b8d2012-07-30 18:16:16 -0700379 old_state != CON_SOCK_STATE_CONNECTING &&
380 old_state != CON_SOCK_STATE_CLOSED))
Alex Elderce2c8902012-05-22 22:15:49 -0500381 printk("%s: unexpected old state %d\n", __func__, old_state);
Sage Weil8007b8d2012-07-30 18:16:16 -0700382 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
383 CON_SOCK_STATE_CLOSED);
Alex Elderce2c8902012-05-22 22:15:49 -0500384}
Sage Weila922d382010-05-29 09:41:23 -0700385
Sage Weil31b80062009-10-06 11:31:13 -0700386/*
387 * socket callback functions
388 */
389
390/* data available on socket, or listen socket received a connect */
David S. Miller676d2362014-04-11 16:15:36 -0400391static void ceph_sock_data_ready(struct sock *sk)
Sage Weil31b80062009-10-06 11:31:13 -0700392{
Alex Elderbd406142012-01-23 15:49:27 -0600393 struct ceph_connection *con = sk->sk_user_data;
Guanjun Hea2a32582012-07-08 19:50:33 -0700394 if (atomic_read(&con->msgr->stopping)) {
395 return;
396 }
Alex Elderbd406142012-01-23 15:49:27 -0600397
Sage Weil31b80062009-10-06 11:31:13 -0700398 if (sk->sk_state != TCP_CLOSE_WAIT) {
Alex Elder327800b2012-05-22 11:41:43 -0500399 dout("%s on %p state = %lu, queueing work\n", __func__,
Sage Weil31b80062009-10-06 11:31:13 -0700400 con, con->state);
401 queue_con(con);
402 }
403}
404
405/* socket has buffer space for writing */
Alex Elder327800b2012-05-22 11:41:43 -0500406static void ceph_sock_write_space(struct sock *sk)
Sage Weil31b80062009-10-06 11:31:13 -0700407{
Alex Elderd3002b92012-02-14 14:05:33 -0600408 struct ceph_connection *con = sk->sk_user_data;
Sage Weil31b80062009-10-06 11:31:13 -0700409
Jim Schutt182fac22012-02-29 08:30:58 -0700410 /* only queue to workqueue if there is data we want to write,
411 * and there is sufficient space in the socket buffer to accept
Alex Elder327800b2012-05-22 11:41:43 -0500412 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
Jim Schutt182fac22012-02-29 08:30:58 -0700413 * doesn't get called again until try_write() fills the socket
414 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
415 * and net/core/stream.c:sk_stream_write_space().
416 */
Alex Elderc9ffc772013-02-20 10:25:12 -0600417 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
Eric Dumazet64dc6132013-07-22 20:26:31 -0700418 if (sk_stream_is_writeable(sk)) {
Alex Elder327800b2012-05-22 11:41:43 -0500419 dout("%s %p queueing write work\n", __func__, con);
Jim Schutt182fac22012-02-29 08:30:58 -0700420 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
421 queue_con(con);
422 }
Sage Weil31b80062009-10-06 11:31:13 -0700423 } else {
Alex Elder327800b2012-05-22 11:41:43 -0500424 dout("%s %p nothing to write\n", __func__, con);
Sage Weil31b80062009-10-06 11:31:13 -0700425 }
Sage Weil31b80062009-10-06 11:31:13 -0700426}
427
428/* socket's state has changed */
Alex Elder327800b2012-05-22 11:41:43 -0500429static void ceph_sock_state_change(struct sock *sk)
Sage Weil31b80062009-10-06 11:31:13 -0700430{
Alex Elderbd406142012-01-23 15:49:27 -0600431 struct ceph_connection *con = sk->sk_user_data;
Sage Weil31b80062009-10-06 11:31:13 -0700432
Alex Elder327800b2012-05-22 11:41:43 -0500433 dout("%s %p state = %lu sk_state = %u\n", __func__,
Sage Weil31b80062009-10-06 11:31:13 -0700434 con, con->state, sk->sk_state);
435
Sage Weil31b80062009-10-06 11:31:13 -0700436 switch (sk->sk_state) {
437 case TCP_CLOSE:
Alex Elder327800b2012-05-22 11:41:43 -0500438 dout("%s TCP_CLOSE\n", __func__);
Sage Weil31b80062009-10-06 11:31:13 -0700439 case TCP_CLOSE_WAIT:
Alex Elder327800b2012-05-22 11:41:43 -0500440 dout("%s TCP_CLOSE_WAIT\n", __func__);
Alex Elderce2c8902012-05-22 22:15:49 -0500441 con_sock_state_closing(con);
Alex Elderc9ffc772013-02-20 10:25:12 -0600442 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
Alex Elderd65c9e02012-06-20 21:53:53 -0500443 queue_con(con);
Sage Weil31b80062009-10-06 11:31:13 -0700444 break;
445 case TCP_ESTABLISHED:
Alex Elder327800b2012-05-22 11:41:43 -0500446 dout("%s TCP_ESTABLISHED\n", __func__);
Alex Elderce2c8902012-05-22 22:15:49 -0500447 con_sock_state_connected(con);
Sage Weil31b80062009-10-06 11:31:13 -0700448 queue_con(con);
449 break;
Alex Elderd3002b92012-02-14 14:05:33 -0600450 default: /* Everything else is uninteresting */
451 break;
Sage Weil31b80062009-10-06 11:31:13 -0700452 }
453}
454
455/*
456 * set up socket callbacks
457 */
458static void set_sock_callbacks(struct socket *sock,
459 struct ceph_connection *con)
460{
461 struct sock *sk = sock->sk;
Alex Elderbd406142012-01-23 15:49:27 -0600462 sk->sk_user_data = con;
Alex Elder327800b2012-05-22 11:41:43 -0500463 sk->sk_data_ready = ceph_sock_data_ready;
464 sk->sk_write_space = ceph_sock_write_space;
465 sk->sk_state_change = ceph_sock_state_change;
Sage Weil31b80062009-10-06 11:31:13 -0700466}
467
468
469/*
470 * socket helpers
471 */
472
473/*
474 * initiate connection to a remote socket.
475 */
Alex Elder41617d02012-02-14 14:05:33 -0600476static int ceph_tcp_connect(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -0700477{
Sage Weilf91d3472010-07-01 15:18:31 -0700478 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
Sage Weil31b80062009-10-06 11:31:13 -0700479 struct socket *sock;
480 int ret;
481
482 BUG_ON(con->sock);
Ilya Dryomov757856d2015-06-25 17:47:45 +0300483 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
Eric W. Biedermaneeb1bd52015-05-08 21:08:05 -0500484 SOCK_STREAM, IPPROTO_TCP, &sock);
Sage Weil31b80062009-10-06 11:31:13 -0700485 if (ret)
Alex Elder41617d02012-02-14 14:05:33 -0600486 return ret;
Ilya Dryomov6d7fdb02015-04-02 14:40:58 +0300487 sock->sk->sk_allocation = GFP_NOFS;
Sage Weil31b80062009-10-06 11:31:13 -0700488
Sage Weila6a53492010-04-13 14:07:07 -0700489#ifdef CONFIG_LOCKDEP
490 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
491#endif
492
Sage Weil31b80062009-10-06 11:31:13 -0700493 set_sock_callbacks(sock, con);
494
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700495 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -0700496
Sage Weil89a86be2012-06-09 14:19:21 -0700497 con_sock_state_connecting(con);
Sage Weilf91d3472010-07-01 15:18:31 -0700498 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
499 O_NONBLOCK);
Sage Weil31b80062009-10-06 11:31:13 -0700500 if (ret == -EINPROGRESS) {
501 dout("connect %s EINPROGRESS sk_state = %u\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700502 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil31b80062009-10-06 11:31:13 -0700503 sock->sk->sk_state);
Alex Eldera5bc3129a2012-01-23 15:49:27 -0600504 } else if (ret < 0) {
Sage Weil31b80062009-10-06 11:31:13 -0700505 pr_err("connect %s error %d\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700506 ceph_pr_addr(&con->peer_addr.in_addr), ret);
Sage Weil31b80062009-10-06 11:31:13 -0700507 sock_release(sock);
Alex Elder41617d02012-02-14 14:05:33 -0600508 return ret;
Alex Eldera5bc3129a2012-01-23 15:49:27 -0600509 }
Mike Christie89baaa52014-10-16 01:50:19 -0500510
Chaitanya Huilgolba988f82015-01-23 16:41:25 +0530511 if (con->msgr->tcp_nodelay) {
512 int optval = 1;
513
514 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
515 (char *)&optval, sizeof(optval));
516 if (ret)
517 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
518 ret);
519 }
520
Alex Eldera5bc3129a2012-01-23 15:49:27 -0600521 con->sock = sock;
Alex Elder41617d02012-02-14 14:05:33 -0600522 return 0;
Sage Weil31b80062009-10-06 11:31:13 -0700523}
524
525static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
526{
527 struct kvec iov = {buf, len};
528 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
Sage Weil98bdb0a2011-01-25 08:17:48 -0800529 int r;
Sage Weil31b80062009-10-06 11:31:13 -0700530
Sage Weil98bdb0a2011-01-25 08:17:48 -0800531 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
532 if (r == -EAGAIN)
533 r = 0;
534 return r;
Sage Weil31b80062009-10-06 11:31:13 -0700535}
536
Alex Elderafb3d902013-03-08 20:58:59 -0600537static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
538 int page_offset, size_t length)
539{
540 void *kaddr;
541 int ret;
542
543 BUG_ON(page_offset + length > PAGE_SIZE);
544
545 kaddr = kmap(page);
546 BUG_ON(!kaddr);
547 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
548 kunmap(page);
549
550 return ret;
551}
552
Sage Weil31b80062009-10-06 11:31:13 -0700553/*
554 * write something. @more is true if caller will be sending more data
555 * shortly.
556 */
557static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
558 size_t kvlen, size_t len, int more)
559{
560 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
Sage Weil42961d22011-01-25 08:19:34 -0800561 int r;
Sage Weil31b80062009-10-06 11:31:13 -0700562
563 if (more)
564 msg.msg_flags |= MSG_MORE;
565 else
566 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
567
Sage Weil42961d22011-01-25 08:19:34 -0800568 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
569 if (r == -EAGAIN)
570 r = 0;
571 return r;
Sage Weil31b80062009-10-06 11:31:13 -0700572}
573
Chunwei Chen178eda22014-04-23 12:35:09 +0800574static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
Alex Eldere1dcb122013-03-06 23:39:38 -0600575 int offset, size_t size, bool more)
Alex Elder31739132012-03-07 11:40:08 -0600576{
577 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
578 int ret;
579
580 ret = kernel_sendpage(sock, page, offset, size, flags);
581 if (ret == -EAGAIN)
582 ret = 0;
583
584 return ret;
585}
586
Chunwei Chen178eda22014-04-23 12:35:09 +0800587static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
588 int offset, size_t size, bool more)
589{
590 int ret;
591 struct kvec iov;
592
593 /* sendpage cannot properly handle pages with page_count == 0,
594 * we need to fallback to sendmsg if that's the case */
595 if (page_count(page) >= 1)
596 return __ceph_tcp_sendpage(sock, page, offset, size, more);
597
598 iov.iov_base = kmap(page) + offset;
599 iov.iov_len = size;
600 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
601 kunmap(page);
602
603 return ret;
604}
Sage Weil31b80062009-10-06 11:31:13 -0700605
606/*
607 * Shutdown/close the socket for the given connection.
608 */
609static int con_close_socket(struct ceph_connection *con)
610{
Sage Weil8007b8d2012-07-30 18:16:16 -0700611 int rc = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700612
613 dout("con_close_socket on %p sock %p\n", con, con->sock);
Sage Weil8007b8d2012-07-30 18:16:16 -0700614 if (con->sock) {
615 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
616 sock_release(con->sock);
617 con->sock = NULL;
618 }
Alex Elder456ea462012-06-20 21:53:53 -0500619
620 /*
Sage Weil4a861692012-07-20 17:29:55 -0700621 * Forcibly clear the SOCK_CLOSED flag. It gets set
Alex Elder456ea462012-06-20 21:53:53 -0500622 * independent of the connection mutex, and we could have
623 * received a socket close event before we had the chance to
624 * shut the socket down.
625 */
Alex Elderc9ffc772013-02-20 10:25:12 -0600626 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
Sage Weil8007b8d2012-07-30 18:16:16 -0700627
Alex Elderce2c8902012-05-22 22:15:49 -0500628 con_sock_state_closed(con);
Sage Weil31b80062009-10-06 11:31:13 -0700629 return rc;
630}
631
632/*
633 * Reset a connection. Discard all incoming and outgoing messages
634 * and clear *_seq state.
635 */
636static void ceph_msg_remove(struct ceph_msg *msg)
637{
638 list_del_init(&msg->list_head);
Alex Elder38941f82012-06-01 14:56:43 -0500639 BUG_ON(msg->con == NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -0700640 msg->con->ops->put(msg->con);
Alex Elder38941f82012-06-01 14:56:43 -0500641 msg->con = NULL;
642
Sage Weil31b80062009-10-06 11:31:13 -0700643 ceph_msg_put(msg);
644}
645static void ceph_msg_remove_list(struct list_head *head)
646{
647 while (!list_empty(head)) {
648 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
649 list_head);
650 ceph_msg_remove(msg);
651 }
652}
653
654static void reset_connection(struct ceph_connection *con)
655{
656 /* reset connection, out_queue, msg_ and connect_seq */
657 /* discard existing out_queue and msg_seq */
Sage Weil0fa6ebc2012-12-27 20:27:04 -0600658 dout("reset_connection %p\n", con);
Sage Weil31b80062009-10-06 11:31:13 -0700659 ceph_msg_remove_list(&con->out_queue);
660 ceph_msg_remove_list(&con->out_sent);
661
Sage Weilcf3e5c42009-12-11 09:48:05 -0800662 if (con->in_msg) {
Alex Elder38941f82012-06-01 14:56:43 -0500663 BUG_ON(con->in_msg->con != con);
664 con->in_msg->con = NULL;
Sage Weilcf3e5c42009-12-11 09:48:05 -0800665 ceph_msg_put(con->in_msg);
666 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -0700667 con->ops->put(con);
Sage Weilcf3e5c42009-12-11 09:48:05 -0800668 }
669
Sage Weil31b80062009-10-06 11:31:13 -0700670 con->connect_seq = 0;
671 con->out_seq = 0;
Sage Weilc86a2932009-12-14 14:04:30 -0800672 if (con->out_msg) {
673 ceph_msg_put(con->out_msg);
674 con->out_msg = NULL;
675 }
Sage Weil31b80062009-10-06 11:31:13 -0700676 con->in_seq = 0;
Sage Weil0e0d5e02010-04-02 16:07:19 -0700677 con->in_seq_acked = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700678}
679
680/*
681 * mark a peer down. drop any open connections.
682 */
683void ceph_con_close(struct ceph_connection *con)
684{
Sage Weil8c50c812012-07-30 16:24:37 -0700685 mutex_lock(&con->mutex);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700686 dout("con_close %p peer %s\n", con,
687 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil8dacc7d2012-07-20 17:24:40 -0700688 con->state = CON_STATE_CLOSED;
Alex Eldera5988c42012-05-29 11:04:58 -0500689
Alex Elderc9ffc772013-02-20 10:25:12 -0600690 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
691 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
692 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
693 con_flag_clear(con, CON_FLAG_BACKOFF);
Alex Eldera5988c42012-05-29 11:04:58 -0500694
Sage Weil31b80062009-10-06 11:31:13 -0700695 reset_connection(con);
Sage Weil6f2bc3f2010-04-02 16:16:34 -0700696 con->peer_global_seq = 0;
Ilya Dryomov37ab77a2014-06-24 16:21:45 +0400697 cancel_con(con);
Sage Weilee76e072012-07-20 16:45:49 -0700698 con_close_socket(con);
Sage Weilec302642009-12-22 10:43:42 -0800699 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700700}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700701EXPORT_SYMBOL(ceph_con_close);
Sage Weil31b80062009-10-06 11:31:13 -0700702
703/*
Sage Weil31b80062009-10-06 11:31:13 -0700704 * Reopen a closed connection, with a new peer address.
705 */
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700706void ceph_con_open(struct ceph_connection *con,
707 __u8 entity_type, __u64 entity_num,
708 struct ceph_entity_addr *addr)
Sage Weil31b80062009-10-06 11:31:13 -0700709{
Sage Weil54691552012-07-30 16:21:40 -0700710 mutex_lock(&con->mutex);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700711 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
Sage Weil8dacc7d2012-07-20 17:24:40 -0700712
Alex Elder122070a2012-12-26 10:43:57 -0600713 WARN_ON(con->state != CON_STATE_CLOSED);
Sage Weil8dacc7d2012-07-20 17:24:40 -0700714 con->state = CON_STATE_PREOPEN;
Alex Eldera5988c42012-05-29 11:04:58 -0500715
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700716 con->peer_name.type = (__u8) entity_type;
717 con->peer_name.num = cpu_to_le64(entity_num);
718
Sage Weil31b80062009-10-06 11:31:13 -0700719 memcpy(&con->peer_addr, addr, sizeof(*addr));
Sage Weil03c677e2009-11-20 15:14:15 -0800720 con->delay = 0; /* reset backoff memory */
Sage Weil54691552012-07-30 16:21:40 -0700721 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700722 queue_con(con);
723}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700724EXPORT_SYMBOL(ceph_con_open);
Sage Weil31b80062009-10-06 11:31:13 -0700725
726/*
Sage Weil87b315a2010-03-22 14:51:18 -0700727 * return true if this connection ever successfully opened
728 */
729bool ceph_con_opened(struct ceph_connection *con)
730{
731 return con->connect_seq > 0;
732}
733
734/*
Sage Weil31b80062009-10-06 11:31:13 -0700735 * initialize a new connection.
736 */
Alex Elder1bfd89f2012-05-26 23:26:43 -0500737void ceph_con_init(struct ceph_connection *con, void *private,
738 const struct ceph_connection_operations *ops,
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700739 struct ceph_messenger *msgr)
Sage Weil31b80062009-10-06 11:31:13 -0700740{
741 dout("con_init %p\n", con);
742 memset(con, 0, sizeof(*con));
Alex Elder1bfd89f2012-05-26 23:26:43 -0500743 con->private = private;
744 con->ops = ops;
Sage Weil31b80062009-10-06 11:31:13 -0700745 con->msgr = msgr;
Alex Elderce2c8902012-05-22 22:15:49 -0500746
747 con_sock_state_init(con);
748
Sage Weilec302642009-12-22 10:43:42 -0800749 mutex_init(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700750 INIT_LIST_HEAD(&con->out_queue);
751 INIT_LIST_HEAD(&con->out_sent);
752 INIT_DELAYED_WORK(&con->work, con_work);
Alex Eldera5988c42012-05-29 11:04:58 -0500753
Sage Weil8dacc7d2012-07-20 17:24:40 -0700754 con->state = CON_STATE_CLOSED;
Sage Weil31b80062009-10-06 11:31:13 -0700755}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700756EXPORT_SYMBOL(ceph_con_init);
Sage Weil31b80062009-10-06 11:31:13 -0700757
758
759/*
760 * We maintain a global counter to order connection attempts. Get
761 * a unique seq greater than @gt.
762 */
763static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
764{
765 u32 ret;
766
767 spin_lock(&msgr->global_seq_lock);
768 if (msgr->global_seq < gt)
769 msgr->global_seq = gt;
770 ret = ++msgr->global_seq;
771 spin_unlock(&msgr->global_seq_lock);
772 return ret;
773}
774
Alex Eldere2200422012-05-23 14:35:23 -0500775static void con_out_kvec_reset(struct ceph_connection *con)
Alex Elder859eb792012-02-14 14:05:33 -0600776{
777 con->out_kvec_left = 0;
778 con->out_kvec_bytes = 0;
779 con->out_kvec_cur = &con->out_kvec[0];
780}
781
Alex Eldere2200422012-05-23 14:35:23 -0500782static void con_out_kvec_add(struct ceph_connection *con,
Alex Elder859eb792012-02-14 14:05:33 -0600783 size_t size, void *data)
784{
785 int index;
786
787 index = con->out_kvec_left;
788 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
789
790 con->out_kvec[index].iov_len = size;
791 con->out_kvec[index].iov_base = data;
792 con->out_kvec_left++;
793 con->out_kvec_bytes += size;
794}
Sage Weil31b80062009-10-06 11:31:13 -0700795
Alex Elderdf6ad1f2012-06-11 14:57:13 -0500796#ifdef CONFIG_BLOCK
Alex Elder6aaa4512013-03-06 23:39:39 -0600797
798/*
799 * For a bio data item, a piece is whatever remains of the next
800 * entry in the current bio iovec, or the first entry in the next
801 * bio in the list.
802 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500803static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500804 size_t length)
Alex Elder6aaa4512013-03-06 23:39:39 -0600805{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500806 struct ceph_msg_data *data = cursor->data;
Alex Elder6aaa4512013-03-06 23:39:39 -0600807 struct bio *bio;
808
809 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
810
811 bio = data->bio;
812 BUG_ON(!bio);
Alex Elder6aaa4512013-03-06 23:39:39 -0600813
Alex Elderca8b3a62013-04-05 14:46:01 -0500814 cursor->resid = min(length, data->bio_length);
Alex Elder6aaa4512013-03-06 23:39:39 -0600815 cursor->bio = bio;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700816 cursor->bvec_iter = bio->bi_iter;
817 cursor->last_piece =
818 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
Alex Elder6aaa4512013-03-06 23:39:39 -0600819}
820
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500821static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
Alex Elder6aaa4512013-03-06 23:39:39 -0600822 size_t *page_offset,
823 size_t *length)
824{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500825 struct ceph_msg_data *data = cursor->data;
Alex Elder6aaa4512013-03-06 23:39:39 -0600826 struct bio *bio;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700827 struct bio_vec bio_vec;
Alex Elder6aaa4512013-03-06 23:39:39 -0600828
829 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
830
831 bio = cursor->bio;
832 BUG_ON(!bio);
833
Kent Overstreetf38a5182013-08-07 14:30:24 -0700834 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
Alex Elder6aaa4512013-03-06 23:39:39 -0600835
Kent Overstreetf38a5182013-08-07 14:30:24 -0700836 *page_offset = (size_t) bio_vec.bv_offset;
Alex Elder6aaa4512013-03-06 23:39:39 -0600837 BUG_ON(*page_offset >= PAGE_SIZE);
Alex Elder25aff7c2013-03-11 23:34:22 -0500838 if (cursor->last_piece) /* pagelist offset is always 0 */
839 *length = cursor->resid;
840 else
Kent Overstreetf38a5182013-08-07 14:30:24 -0700841 *length = (size_t) bio_vec.bv_len;
Alex Elder25aff7c2013-03-11 23:34:22 -0500842 BUG_ON(*length > cursor->resid);
Alex Elder5df521b2013-03-30 15:09:59 -0500843 BUG_ON(*page_offset + *length > PAGE_SIZE);
Alex Elder6aaa4512013-03-06 23:39:39 -0600844
Kent Overstreetf38a5182013-08-07 14:30:24 -0700845 return bio_vec.bv_page;
Alex Elder6aaa4512013-03-06 23:39:39 -0600846}
847
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500848static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
849 size_t bytes)
Alex Elder6aaa4512013-03-06 23:39:39 -0600850{
Alex Elder6aaa4512013-03-06 23:39:39 -0600851 struct bio *bio;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700852 struct bio_vec bio_vec;
Alex Elder6aaa4512013-03-06 23:39:39 -0600853
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500854 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
Alex Elder6aaa4512013-03-06 23:39:39 -0600855
856 bio = cursor->bio;
857 BUG_ON(!bio);
858
Kent Overstreetf38a5182013-08-07 14:30:24 -0700859 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
Alex Elder6aaa4512013-03-06 23:39:39 -0600860
861 /* Advance the cursor offset */
862
Alex Elder25aff7c2013-03-11 23:34:22 -0500863 BUG_ON(cursor->resid < bytes);
864 cursor->resid -= bytes;
Kent Overstreetf38a5182013-08-07 14:30:24 -0700865
866 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
867
868 if (bytes < bio_vec.bv_len)
Alex Elder6aaa4512013-03-06 23:39:39 -0600869 return false; /* more bytes to process in this segment */
870
871 /* Move on to the next segment, and possibly the next bio */
872
Kent Overstreetf38a5182013-08-07 14:30:24 -0700873 if (!cursor->bvec_iter.bi_size) {
Alex Elder6aaa4512013-03-06 23:39:39 -0600874 bio = bio->bi_next;
Ilya Dryomov0ec1d152014-02-05 15:19:55 +0200875 cursor->bio = bio;
876 if (bio)
877 cursor->bvec_iter = bio->bi_iter;
878 else
879 memset(&cursor->bvec_iter, 0,
880 sizeof(cursor->bvec_iter));
Alex Elder6aaa4512013-03-06 23:39:39 -0600881 }
Alex Elder6aaa4512013-03-06 23:39:39 -0600882
Alex Elder25aff7c2013-03-11 23:34:22 -0500883 if (!cursor->last_piece) {
884 BUG_ON(!cursor->resid);
885 BUG_ON(!bio);
886 /* A short read is OK, so use <= rather than == */
Kent Overstreetf38a5182013-08-07 14:30:24 -0700887 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
Alex Elder6aaa4512013-03-06 23:39:39 -0600888 cursor->last_piece = true;
Alex Elder25aff7c2013-03-11 23:34:22 -0500889 }
Alex Elder6aaa4512013-03-06 23:39:39 -0600890
891 return true;
892}
Alex Elderea965712013-04-05 14:46:01 -0500893#endif /* CONFIG_BLOCK */
Alex Elderdf6ad1f2012-06-11 14:57:13 -0500894
Alex Elderfe38a2b2013-03-06 23:39:39 -0600895/*
Alex Eldere766d7b2013-03-07 15:38:28 -0600896 * For a page array, a piece comes from the first page in the array
897 * that has not already been fully consumed.
898 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500899static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500900 size_t length)
Alex Eldere766d7b2013-03-07 15:38:28 -0600901{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500902 struct ceph_msg_data *data = cursor->data;
Alex Eldere766d7b2013-03-07 15:38:28 -0600903 int page_count;
904
905 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
906
907 BUG_ON(!data->pages);
908 BUG_ON(!data->length);
909
Alex Elderca8b3a62013-04-05 14:46:01 -0500910 cursor->resid = min(length, data->length);
Alex Eldere766d7b2013-03-07 15:38:28 -0600911 page_count = calc_pages_for(data->alignment, (u64)data->length);
Alex Eldere766d7b2013-03-07 15:38:28 -0600912 cursor->page_offset = data->alignment & ~PAGE_MASK;
913 cursor->page_index = 0;
Alex Elder56fc5652013-03-30 23:46:55 -0500914 BUG_ON(page_count > (int)USHRT_MAX);
915 cursor->page_count = (unsigned short)page_count;
916 BUG_ON(length > SIZE_MAX - cursor->page_offset);
Ilya Dryomov5f740d72014-08-08 12:43:39 +0400917 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
Alex Eldere766d7b2013-03-07 15:38:28 -0600918}
919
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500920static struct page *
921ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
922 size_t *page_offset, size_t *length)
Alex Eldere766d7b2013-03-07 15:38:28 -0600923{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500924 struct ceph_msg_data *data = cursor->data;
Alex Eldere766d7b2013-03-07 15:38:28 -0600925
926 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
927
928 BUG_ON(cursor->page_index >= cursor->page_count);
929 BUG_ON(cursor->page_offset >= PAGE_SIZE);
Alex Eldere766d7b2013-03-07 15:38:28 -0600930
931 *page_offset = cursor->page_offset;
Alex Elder25aff7c2013-03-11 23:34:22 -0500932 if (cursor->last_piece)
Alex Eldere766d7b2013-03-07 15:38:28 -0600933 *length = cursor->resid;
Alex Elder25aff7c2013-03-11 23:34:22 -0500934 else
Alex Eldere766d7b2013-03-07 15:38:28 -0600935 *length = PAGE_SIZE - *page_offset;
Alex Eldere766d7b2013-03-07 15:38:28 -0600936
937 return data->pages[cursor->page_index];
938}
939
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500940static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
Alex Eldere766d7b2013-03-07 15:38:28 -0600941 size_t bytes)
942{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500943 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
Alex Eldere766d7b2013-03-07 15:38:28 -0600944
945 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
Alex Eldere766d7b2013-03-07 15:38:28 -0600946
947 /* Advance the cursor page offset */
948
949 cursor->resid -= bytes;
Alex Elder5df521b2013-03-30 15:09:59 -0500950 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
951 if (!bytes || cursor->page_offset)
Alex Eldere766d7b2013-03-07 15:38:28 -0600952 return false; /* more bytes to process in the current page */
953
Yan, Zhengd90deda2014-03-23 06:50:39 +0800954 if (!cursor->resid)
955 return false; /* no more data */
956
Alex Elder5df521b2013-03-30 15:09:59 -0500957 /* Move on to the next page; offset is already at 0 */
Alex Eldere766d7b2013-03-07 15:38:28 -0600958
959 BUG_ON(cursor->page_index >= cursor->page_count);
Alex Eldere766d7b2013-03-07 15:38:28 -0600960 cursor->page_index++;
Alex Elder25aff7c2013-03-11 23:34:22 -0500961 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Eldere766d7b2013-03-07 15:38:28 -0600962
963 return true;
964}
965
966/*
Alex Elderdd236fc2013-03-06 23:39:39 -0600967 * For a pagelist, a piece is whatever remains to be consumed in the
968 * first page in the list, or the front of the next page.
Alex Elderfe38a2b2013-03-06 23:39:39 -0600969 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500970static void
971ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
Alex Elder25aff7c2013-03-11 23:34:22 -0500972 size_t length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600973{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500974 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600975 struct ceph_pagelist *pagelist;
976 struct page *page;
977
Alex Elderdd236fc2013-03-06 23:39:39 -0600978 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600979
980 pagelist = data->pagelist;
981 BUG_ON(!pagelist);
Alex Elder25aff7c2013-03-11 23:34:22 -0500982
983 if (!length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600984 return; /* pagelist can be assigned but empty */
985
986 BUG_ON(list_empty(&pagelist->head));
987 page = list_first_entry(&pagelist->head, struct page, lru);
988
Alex Elderca8b3a62013-04-05 14:46:01 -0500989 cursor->resid = min(length, pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -0600990 cursor->page = page;
991 cursor->offset = 0;
Alex Eldera51b272e2013-04-19 15:34:49 -0500992 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Elderfe38a2b2013-03-06 23:39:39 -0600993}
994
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500995static struct page *
996ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
997 size_t *page_offset, size_t *length)
Alex Elderfe38a2b2013-03-06 23:39:39 -0600998{
Alex Elder8ae4f4f2013-03-14 14:09:06 -0500999 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001000 struct ceph_pagelist *pagelist;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001001
1002 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1003
1004 pagelist = data->pagelist;
1005 BUG_ON(!pagelist);
1006
1007 BUG_ON(!cursor->page);
Alex Elder25aff7c2013-03-11 23:34:22 -05001008 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -06001009
Alex Elder5df521b2013-03-30 15:09:59 -05001010 /* offset of first page in pagelist is always 0 */
Alex Elderfe38a2b2013-03-06 23:39:39 -06001011 *page_offset = cursor->offset & ~PAGE_MASK;
Alex Elder5df521b2013-03-30 15:09:59 -05001012 if (cursor->last_piece)
Alex Elder25aff7c2013-03-11 23:34:22 -05001013 *length = cursor->resid;
1014 else
1015 *length = PAGE_SIZE - *page_offset;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001016
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001017 return cursor->page;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001018}
1019
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001020static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
Alex Elderdd236fc2013-03-06 23:39:39 -06001021 size_t bytes)
Alex Elderfe38a2b2013-03-06 23:39:39 -06001022{
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001023 struct ceph_msg_data *data = cursor->data;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001024 struct ceph_pagelist *pagelist;
1025
1026 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1027
1028 pagelist = data->pagelist;
1029 BUG_ON(!pagelist);
Alex Elder25aff7c2013-03-11 23:34:22 -05001030
1031 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
Alex Elderfe38a2b2013-03-06 23:39:39 -06001032 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1033
1034 /* Advance the cursor offset */
1035
Alex Elder25aff7c2013-03-11 23:34:22 -05001036 cursor->resid -= bytes;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001037 cursor->offset += bytes;
Alex Elder5df521b2013-03-30 15:09:59 -05001038 /* offset of first page in pagelist is always 0 */
Alex Elderfe38a2b2013-03-06 23:39:39 -06001039 if (!bytes || cursor->offset & ~PAGE_MASK)
1040 return false; /* more bytes to process in the current page */
1041
Yan, Zhengd90deda2014-03-23 06:50:39 +08001042 if (!cursor->resid)
1043 return false; /* no more data */
1044
Alex Elderfe38a2b2013-03-06 23:39:39 -06001045 /* Move on to the next page */
1046
1047 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1048 cursor->page = list_entry_next(cursor->page, lru);
Alex Elder25aff7c2013-03-11 23:34:22 -05001049 cursor->last_piece = cursor->resid <= PAGE_SIZE;
Alex Elderfe38a2b2013-03-06 23:39:39 -06001050
1051 return true;
1052}
1053
Alex Elderdd236fc2013-03-06 23:39:39 -06001054/*
1055 * Message data is handled (sent or received) in pieces, where each
1056 * piece resides on a single page. The network layer might not
1057 * consume an entire piece at once. A data item's cursor keeps
1058 * track of which piece is next to process and how much remains to
1059 * be processed in that piece. It also tracks whether the current
1060 * piece is the last one in the data item.
1061 */
Alex Elderca8b3a62013-04-05 14:46:01 -05001062static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
Alex Elderdd236fc2013-03-06 23:39:39 -06001063{
Alex Elderca8b3a62013-04-05 14:46:01 -05001064 size_t length = cursor->total_resid;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001065
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001066 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001067 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001068 ceph_msg_data_pagelist_cursor_init(cursor, length);
Alex Elderdd236fc2013-03-06 23:39:39 -06001069 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001070 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001071 ceph_msg_data_pages_cursor_init(cursor, length);
Alex Eldere766d7b2013-03-07 15:38:28 -06001072 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001073#ifdef CONFIG_BLOCK
1074 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001075 ceph_msg_data_bio_cursor_init(cursor, length);
Alex Elder6aaa4512013-03-06 23:39:39 -06001076 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001077#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001078 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001079 default:
1080 /* BUG(); */
1081 break;
1082 }
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001083 cursor->need_crc = true;
Alex Elderdd236fc2013-03-06 23:39:39 -06001084}
1085
Alex Elderca8b3a62013-04-05 14:46:01 -05001086static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1087{
1088 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1089 struct ceph_msg_data *data;
1090
1091 BUG_ON(!length);
1092 BUG_ON(length > msg->data_length);
1093 BUG_ON(list_empty(&msg->data));
1094
Alex Elderca8b3a62013-04-05 14:46:01 -05001095 cursor->data_head = &msg->data;
1096 cursor->total_resid = length;
1097 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1098 cursor->data = data;
1099
1100 __ceph_msg_data_cursor_init(cursor);
1101}
1102
Alex Elderdd236fc2013-03-06 23:39:39 -06001103/*
1104 * Return the page containing the next piece to process for a given
1105 * data item, and supply the page offset and length of that piece.
1106 * Indicate whether this is the last piece in this data item.
1107 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001108static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1109 size_t *page_offset, size_t *length,
Alex Elderdd236fc2013-03-06 23:39:39 -06001110 bool *last_piece)
1111{
1112 struct page *page;
1113
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001114 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001115 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001116 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
Alex Elderdd236fc2013-03-06 23:39:39 -06001117 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001118 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001119 page = ceph_msg_data_pages_next(cursor, page_offset, length);
Alex Eldere766d7b2013-03-07 15:38:28 -06001120 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001121#ifdef CONFIG_BLOCK
1122 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001123 page = ceph_msg_data_bio_next(cursor, page_offset, length);
Alex Elder6aaa4512013-03-06 23:39:39 -06001124 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001125#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001126 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001127 default:
1128 page = NULL;
1129 break;
1130 }
1131 BUG_ON(!page);
1132 BUG_ON(*page_offset + *length > PAGE_SIZE);
1133 BUG_ON(!*length);
1134 if (last_piece)
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001135 *last_piece = cursor->last_piece;
Alex Elderdd236fc2013-03-06 23:39:39 -06001136
1137 return page;
1138}
1139
1140/*
1141 * Returns true if the result moves the cursor on to the next piece
1142 * of the data item.
1143 */
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001144static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1145 size_t bytes)
Alex Elderdd236fc2013-03-06 23:39:39 -06001146{
1147 bool new_piece;
1148
Alex Elder25aff7c2013-03-11 23:34:22 -05001149 BUG_ON(bytes > cursor->resid);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001150 switch (cursor->data->type) {
Alex Elderdd236fc2013-03-06 23:39:39 -06001151 case CEPH_MSG_DATA_PAGELIST:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001152 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
Alex Elderdd236fc2013-03-06 23:39:39 -06001153 break;
Alex Eldere766d7b2013-03-07 15:38:28 -06001154 case CEPH_MSG_DATA_PAGES:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001155 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
Alex Eldere766d7b2013-03-07 15:38:28 -06001156 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001157#ifdef CONFIG_BLOCK
1158 case CEPH_MSG_DATA_BIO:
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001159 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
Alex Elder6aaa4512013-03-06 23:39:39 -06001160 break;
Alex Elderdd236fc2013-03-06 23:39:39 -06001161#endif /* CONFIG_BLOCK */
Alex Elder6aaa4512013-03-06 23:39:39 -06001162 case CEPH_MSG_DATA_NONE:
Alex Elderdd236fc2013-03-06 23:39:39 -06001163 default:
1164 BUG();
1165 break;
1166 }
Alex Elderca8b3a62013-04-05 14:46:01 -05001167 cursor->total_resid -= bytes;
Alex Elderdd236fc2013-03-06 23:39:39 -06001168
Alex Elderca8b3a62013-04-05 14:46:01 -05001169 if (!cursor->resid && cursor->total_resid) {
1170 WARN_ON(!cursor->last_piece);
1171 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1172 cursor->data = list_entry_next(cursor->data, links);
1173 __ceph_msg_data_cursor_init(cursor);
Alex Eldera51b272e2013-04-19 15:34:49 -05001174 new_piece = true;
Alex Elderca8b3a62013-04-05 14:46:01 -05001175 }
Alex Eldera51b272e2013-04-19 15:34:49 -05001176 cursor->need_crc = new_piece;
Alex Elderca8b3a62013-04-05 14:46:01 -05001177
Alex Elderdd236fc2013-03-06 23:39:39 -06001178 return new_piece;
1179}
1180
Alex Elder98fa5dd2013-04-02 12:09:50 -05001181static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
Alex Elder739c9052012-06-11 14:57:13 -05001182{
Alex Elder739c9052012-06-11 14:57:13 -05001183 BUG_ON(!msg);
Alex Elder25aff7c2013-03-11 23:34:22 -05001184 BUG_ON(!data_len);
Alex Elder739c9052012-06-11 14:57:13 -05001185
Alex Elder4c59b4a2013-03-11 23:34:23 -05001186 /* Initialize data cursor */
Alex Elderfe38a2b2013-03-06 23:39:39 -06001187
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001188 ceph_msg_data_cursor_init(msg, (size_t)data_len);
Alex Elder739c9052012-06-11 14:57:13 -05001189}
1190
Sage Weil31b80062009-10-06 11:31:13 -07001191/*
1192 * Prepare footer for currently outgoing message, and finish things
1193 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1194 */
Alex Elder859eb792012-02-14 14:05:33 -06001195static void prepare_write_message_footer(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001196{
1197 struct ceph_msg *m = con->out_msg;
Alex Elder859eb792012-02-14 14:05:33 -06001198 int v = con->out_kvec_left;
Sage Weil31b80062009-10-06 11:31:13 -07001199
Alex Elderfd154f32012-06-11 14:57:13 -05001200 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1201
Sage Weil31b80062009-10-06 11:31:13 -07001202 dout("prepare_write_message_footer %p\n", con);
1203 con->out_kvec_is_msg = true;
1204 con->out_kvec[v].iov_base = &m->footer;
Yan, Zheng33d07332014-11-04 16:33:37 +08001205 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1206 if (con->ops->sign_message)
1207 con->ops->sign_message(con, m);
1208 else
1209 m->footer.sig = 0;
1210 con->out_kvec[v].iov_len = sizeof(m->footer);
1211 con->out_kvec_bytes += sizeof(m->footer);
1212 } else {
1213 m->old_footer.flags = m->footer.flags;
1214 con->out_kvec[v].iov_len = sizeof(m->old_footer);
1215 con->out_kvec_bytes += sizeof(m->old_footer);
1216 }
Sage Weil31b80062009-10-06 11:31:13 -07001217 con->out_kvec_left++;
1218 con->out_more = m->more_to_follow;
Sage Weilc86a2932009-12-14 14:04:30 -08001219 con->out_msg_done = true;
Sage Weil31b80062009-10-06 11:31:13 -07001220}
1221
1222/*
1223 * Prepare headers for the next outgoing message.
1224 */
1225static void prepare_write_message(struct ceph_connection *con)
1226{
1227 struct ceph_msg *m;
Alex Eldera9a0c512012-02-15 07:43:54 -06001228 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07001229
Alex Eldere2200422012-05-23 14:35:23 -05001230 con_out_kvec_reset(con);
Sage Weil31b80062009-10-06 11:31:13 -07001231 con->out_kvec_is_msg = true;
Sage Weilc86a2932009-12-14 14:04:30 -08001232 con->out_msg_done = false;
Sage Weil31b80062009-10-06 11:31:13 -07001233
1234 /* Sneak an ack in there first? If we can get it into the same
1235 * TCP packet that's a good thing. */
1236 if (con->in_seq > con->in_seq_acked) {
1237 con->in_seq_acked = con->in_seq;
Alex Eldere2200422012-05-23 14:35:23 -05001238 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
Sage Weil31b80062009-10-06 11:31:13 -07001239 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
Alex Eldere2200422012-05-23 14:35:23 -05001240 con_out_kvec_add(con, sizeof (con->out_temp_ack),
Alex Elder859eb792012-02-14 14:05:33 -06001241 &con->out_temp_ack);
Sage Weil31b80062009-10-06 11:31:13 -07001242 }
1243
Alex Elder38941f82012-06-01 14:56:43 -05001244 BUG_ON(list_empty(&con->out_queue));
Alex Elder859eb792012-02-14 14:05:33 -06001245 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
Sage Weilc86a2932009-12-14 14:04:30 -08001246 con->out_msg = m;
Alex Elder38941f82012-06-01 14:56:43 -05001247 BUG_ON(m->con != con);
Sage Weil4cf9d542011-07-26 11:27:24 -07001248
1249 /* put message on sent list */
1250 ceph_msg_get(m);
1251 list_move_tail(&m->list_head, &con->out_sent);
Sage Weil31b80062009-10-06 11:31:13 -07001252
Sage Weile84346b2010-05-11 21:20:38 -07001253 /*
1254 * only assign outgoing seq # if we haven't sent this message
1255 * yet. if it is requeued, resend with it's original seq.
1256 */
1257 if (m->needs_out_seq) {
1258 m->hdr.seq = cpu_to_le64(++con->out_seq);
1259 m->needs_out_seq = false;
1260 }
Alex Elder98fa5dd2013-04-02 12:09:50 -05001261 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
Sage Weil31b80062009-10-06 11:31:13 -07001262
Alex Elder98fa5dd2013-04-02 12:09:50 -05001263 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
Sage Weil31b80062009-10-06 11:31:13 -07001264 m, con->out_seq, le16_to_cpu(m->hdr.type),
1265 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
Alex Elder98fa5dd2013-04-02 12:09:50 -05001266 m->data_length);
Sage Weil31b80062009-10-06 11:31:13 -07001267 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1268
1269 /* tag + hdr + front + middle */
Alex Eldere2200422012-05-23 14:35:23 -05001270 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1271 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1272 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
Alex Elder859eb792012-02-14 14:05:33 -06001273
Sage Weil31b80062009-10-06 11:31:13 -07001274 if (m->middle)
Alex Eldere2200422012-05-23 14:35:23 -05001275 con_out_kvec_add(con, m->middle->vec.iov_len,
Alex Elder859eb792012-02-14 14:05:33 -06001276 m->middle->vec.iov_base);
Sage Weil31b80062009-10-06 11:31:13 -07001277
1278 /* fill in crc (except data pages), footer */
Alex Eldera9a0c512012-02-15 07:43:54 -06001279 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1280 con->out_msg->hdr.crc = cpu_to_le32(crc);
Alex Elderfd154f32012-06-11 14:57:13 -05001281 con->out_msg->footer.flags = 0;
Alex Eldera9a0c512012-02-15 07:43:54 -06001282
1283 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1284 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1285 if (m->middle) {
1286 crc = crc32c(0, m->middle->vec.iov_base,
1287 m->middle->vec.iov_len);
1288 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1289 } else
Sage Weil31b80062009-10-06 11:31:13 -07001290 con->out_msg->footer.middle_crc = 0;
Alex Elder739c9052012-06-11 14:57:13 -05001291 dout("%s front_crc %u middle_crc %u\n", __func__,
Sage Weil31b80062009-10-06 11:31:13 -07001292 le32_to_cpu(con->out_msg->footer.front_crc),
1293 le32_to_cpu(con->out_msg->footer.middle_crc));
1294
1295 /* is there a data payload? */
Alex Elder739c9052012-06-11 14:57:13 -05001296 con->out_msg->footer.data_crc = 0;
Alex Elder98fa5dd2013-04-02 12:09:50 -05001297 if (m->data_length) {
1298 prepare_message_data(con->out_msg, m->data_length);
Alex Elder78625052013-03-06 23:39:39 -06001299 con->out_more = 1; /* data + footer will follow */
1300 } else {
Sage Weil31b80062009-10-06 11:31:13 -07001301 /* no, queue up footer too and be done */
Alex Elder859eb792012-02-14 14:05:33 -06001302 prepare_write_message_footer(con);
Alex Elder78625052013-03-06 23:39:39 -06001303 }
Sage Weil31b80062009-10-06 11:31:13 -07001304
Alex Elderc9ffc772013-02-20 10:25:12 -06001305 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001306}
1307
1308/*
1309 * Prepare an ack.
1310 */
1311static void prepare_write_ack(struct ceph_connection *con)
1312{
1313 dout("prepare_write_ack %p %llu -> %llu\n", con,
1314 con->in_seq_acked, con->in_seq);
1315 con->in_seq_acked = con->in_seq;
1316
Alex Eldere2200422012-05-23 14:35:23 -05001317 con_out_kvec_reset(con);
Alex Elder859eb792012-02-14 14:05:33 -06001318
Alex Eldere2200422012-05-23 14:35:23 -05001319 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
Alex Elder859eb792012-02-14 14:05:33 -06001320
Sage Weil31b80062009-10-06 11:31:13 -07001321 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
Alex Eldere2200422012-05-23 14:35:23 -05001322 con_out_kvec_add(con, sizeof (con->out_temp_ack),
Alex Elder859eb792012-02-14 14:05:33 -06001323 &con->out_temp_ack);
1324
Sage Weil31b80062009-10-06 11:31:13 -07001325 con->out_more = 1; /* more will follow.. eventually.. */
Alex Elderc9ffc772013-02-20 10:25:12 -06001326 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001327}
1328
1329/*
Sage Weil3a230832013-03-25 08:47:40 -07001330 * Prepare to share the seq during handshake
1331 */
1332static void prepare_write_seq(struct ceph_connection *con)
1333{
1334 dout("prepare_write_seq %p %llu -> %llu\n", con,
1335 con->in_seq_acked, con->in_seq);
1336 con->in_seq_acked = con->in_seq;
1337
1338 con_out_kvec_reset(con);
1339
1340 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1341 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1342 &con->out_temp_ack);
1343
1344 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1345}
1346
1347/*
Sage Weil31b80062009-10-06 11:31:13 -07001348 * Prepare to write keepalive byte.
1349 */
1350static void prepare_write_keepalive(struct ceph_connection *con)
1351{
1352 dout("prepare_write_keepalive %p\n", con);
Alex Eldere2200422012-05-23 14:35:23 -05001353 con_out_kvec_reset(con);
1354 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
Alex Elderc9ffc772013-02-20 10:25:12 -06001355 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07001356}
1357
1358/*
1359 * Connection negotiation.
1360 */
1361
Alex Elderdac1e712012-05-16 15:16:39 -05001362static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1363 int *auth_proto)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001364{
Alex Eldera3530df2012-05-16 15:16:39 -05001365 struct ceph_auth_handshake *auth;
Alex Elderb1c6b982012-05-16 15:16:38 -05001366
1367 if (!con->ops->get_authorizer) {
1368 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1369 con->out_connect.authorizer_len = 0;
Alex Elder729796b2012-05-16 15:16:39 -05001370 return NULL;
Alex Elderb1c6b982012-05-16 15:16:38 -05001371 }
1372
1373 /* Can't hold the mutex while getting authorizer */
Sage Weilec302642009-12-22 10:43:42 -08001374 mutex_unlock(&con->mutex);
Alex Elderdac1e712012-05-16 15:16:39 -05001375 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
Sage Weilec302642009-12-22 10:43:42 -08001376 mutex_lock(&con->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001377
Alex Eldera3530df2012-05-16 15:16:39 -05001378 if (IS_ERR(auth))
Alex Elder729796b2012-05-16 15:16:39 -05001379 return auth;
Sage Weil8dacc7d2012-07-20 17:24:40 -07001380 if (con->state != CON_STATE_NEGOTIATING)
Alex Elder729796b2012-05-16 15:16:39 -05001381 return ERR_PTR(-EAGAIN);
Sage Weil0da5d702011-05-19 11:21:05 -07001382
Alex Elder8f43fb52012-05-16 15:16:39 -05001383 con->auth_reply_buf = auth->authorizer_reply_buf;
1384 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
Alex Elder729796b2012-05-16 15:16:39 -05001385 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001386}
1387
Sage Weil31b80062009-10-06 11:31:13 -07001388/*
1389 * We connected to a peer and are saying hello.
1390 */
Alex Eldere825a662012-05-16 15:16:38 -05001391static void prepare_write_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001392{
Alex Eldere2200422012-05-23 14:35:23 -05001393 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1394 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
Alex Eldere825a662012-05-16 15:16:38 -05001395 &con->msgr->my_enc_addr);
Sage Weileed0ef22009-11-10 14:34:36 -08001396
Sage Weileed0ef22009-11-10 14:34:36 -08001397 con->out_more = 0;
Alex Elderc9ffc772013-02-20 10:25:12 -06001398 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weileed0ef22009-11-10 14:34:36 -08001399}
1400
Alex Eldere825a662012-05-16 15:16:38 -05001401static int prepare_write_connect(struct ceph_connection *con)
Sage Weileed0ef22009-11-10 14:34:36 -08001402{
Eric Dumazet95c96172012-04-15 05:58:06 +00001403 unsigned int global_seq = get_global_seq(con->msgr, 0);
Sage Weil31b80062009-10-06 11:31:13 -07001404 int proto;
Alex Elderdac1e712012-05-16 15:16:39 -05001405 int auth_proto;
Alex Elder729796b2012-05-16 15:16:39 -05001406 struct ceph_auth_handshake *auth;
Sage Weil31b80062009-10-06 11:31:13 -07001407
1408 switch (con->peer_name.type) {
1409 case CEPH_ENTITY_TYPE_MON:
1410 proto = CEPH_MONC_PROTOCOL;
1411 break;
1412 case CEPH_ENTITY_TYPE_OSD:
1413 proto = CEPH_OSDC_PROTOCOL;
1414 break;
1415 case CEPH_ENTITY_TYPE_MDS:
1416 proto = CEPH_MDSC_PROTOCOL;
1417 break;
1418 default:
1419 BUG();
1420 }
1421
1422 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1423 con->connect_seq, global_seq, proto);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001424
Alex Eldere825a662012-05-16 15:16:38 -05001425 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
Sage Weil31b80062009-10-06 11:31:13 -07001426 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1427 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1428 con->out_connect.global_seq = cpu_to_le32(global_seq);
1429 con->out_connect.protocol_version = cpu_to_le32(proto);
1430 con->out_connect.flags = 0;
Sage Weil31b80062009-10-06 11:31:13 -07001431
Alex Elderdac1e712012-05-16 15:16:39 -05001432 auth_proto = CEPH_AUTH_UNKNOWN;
1433 auth = get_connect_authorizer(con, &auth_proto);
Alex Elder729796b2012-05-16 15:16:39 -05001434 if (IS_ERR(auth))
1435 return PTR_ERR(auth);
Alex Elder3da54772012-05-16 15:16:39 -05001436
Alex Elderdac1e712012-05-16 15:16:39 -05001437 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
Alex Elder3da54772012-05-16 15:16:39 -05001438 con->out_connect.authorizer_len = auth ?
1439 cpu_to_le32(auth->authorizer_buf_len) : 0;
1440
Alex Eldere2200422012-05-23 14:35:23 -05001441 con_out_kvec_add(con, sizeof (con->out_connect),
Alex Elder3da54772012-05-16 15:16:39 -05001442 &con->out_connect);
1443 if (auth && auth->authorizer_buf_len)
Alex Eldere2200422012-05-23 14:35:23 -05001444 con_out_kvec_add(con, auth->authorizer_buf_len,
Alex Elder3da54772012-05-16 15:16:39 -05001445 auth->authorizer_buf);
Alex Elder859eb792012-02-14 14:05:33 -06001446
Sage Weil31b80062009-10-06 11:31:13 -07001447 con->out_more = 0;
Alex Elderc9ffc772013-02-20 10:25:12 -06001448 con_flag_set(con, CON_FLAG_WRITE_PENDING);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001449
Alex Eldere10c7582012-05-16 15:16:38 -05001450 return 0;
Sage Weil31b80062009-10-06 11:31:13 -07001451}
1452
Sage Weil31b80062009-10-06 11:31:13 -07001453/*
1454 * write as much of pending kvecs to the socket as we can.
1455 * 1 -> done
1456 * 0 -> socket full, but more to do
1457 * <0 -> error
1458 */
1459static int write_partial_kvec(struct ceph_connection *con)
1460{
1461 int ret;
1462
1463 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1464 while (con->out_kvec_bytes > 0) {
1465 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1466 con->out_kvec_left, con->out_kvec_bytes,
1467 con->out_more);
1468 if (ret <= 0)
1469 goto out;
1470 con->out_kvec_bytes -= ret;
1471 if (con->out_kvec_bytes == 0)
1472 break; /* done */
Alex Elderf42299e2012-02-15 07:43:54 -06001473
1474 /* account for full iov entries consumed */
1475 while (ret >= con->out_kvec_cur->iov_len) {
1476 BUG_ON(!con->out_kvec_left);
1477 ret -= con->out_kvec_cur->iov_len;
1478 con->out_kvec_cur++;
1479 con->out_kvec_left--;
1480 }
1481 /* and for a partially-consumed entry */
1482 if (ret) {
1483 con->out_kvec_cur->iov_len -= ret;
1484 con->out_kvec_cur->iov_base += ret;
Sage Weil31b80062009-10-06 11:31:13 -07001485 }
1486 }
1487 con->out_kvec_left = 0;
1488 con->out_kvec_is_msg = false;
1489 ret = 1;
1490out:
1491 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1492 con->out_kvec_bytes, con->out_kvec_left, ret);
1493 return ret; /* done! */
1494}
1495
Alex Elder35b62802013-03-08 20:59:00 -06001496static u32 ceph_crc32c_page(u32 crc, struct page *page,
1497 unsigned int page_offset,
1498 unsigned int length)
1499{
1500 char *kaddr;
1501
1502 kaddr = kmap(page);
1503 BUG_ON(kaddr == NULL);
1504 crc = crc32c(crc, kaddr + page_offset, length);
1505 kunmap(page);
1506
1507 return crc;
1508}
Sage Weil31b80062009-10-06 11:31:13 -07001509/*
1510 * Write as much message data payload as we can. If we finish, queue
1511 * up the footer.
1512 * 1 -> done, footer is now queued in out_kvec[].
1513 * 0 -> socket full, but more to do
1514 * <0 -> error
1515 */
Alex Elder34d2d202013-03-08 20:58:59 -06001516static int write_partial_message_data(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001517{
1518 struct ceph_msg *msg = con->out_msg;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001519 struct ceph_msg_data_cursor *cursor = &msg->cursor;
Alex Elder37675b02012-03-07 11:40:08 -06001520 bool do_datacrc = !con->msgr->nocrc;
Alex Elderf5db90b2013-03-11 23:34:23 -05001521 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07001522
Alex Elder859a35d2013-03-11 23:34:23 -05001523 dout("%s %p msg %p\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07001524
Alex Elder5240d9f2013-03-14 14:09:06 -05001525 if (list_empty(&msg->data))
Alex Elder4c59b4a2013-03-11 23:34:23 -05001526 return -EINVAL;
1527
Alex Elder5821bd82012-06-11 14:57:13 -05001528 /*
1529 * Iterate through each page that contains data to be
1530 * written, and send as much as possible for each.
1531 *
1532 * If we are calculating the data crc (the default), we will
1533 * need to map the page. If we have no pages, they have
1534 * been revoked, so use the zero page.
1535 */
Alex Elderf5db90b2013-03-11 23:34:23 -05001536 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
Alex Elder643c68a2013-03-11 23:34:23 -05001537 while (cursor->resid) {
Alex Elder8a166d02013-03-08 13:35:36 -06001538 struct page *page;
Alex Eldere387d522013-03-06 23:39:38 -06001539 size_t page_offset;
1540 size_t length;
Alex Elder8a166d02013-03-08 13:35:36 -06001541 bool last_piece;
Alex Elder8ea299b2013-03-11 23:34:23 -05001542 bool need_crc;
Alex Elderf5db90b2013-03-11 23:34:23 -05001543 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07001544
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001545 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
Alex Elder4c59b4a2013-03-11 23:34:23 -05001546 &last_piece);
Alex Eldere387d522013-03-06 23:39:38 -06001547 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
Benoît Canetc2cfa192015-06-25 20:32:34 +03001548 length, !last_piece);
Alex Elderf5db90b2013-03-11 23:34:23 -05001549 if (ret <= 0) {
1550 if (do_datacrc)
1551 msg->footer.data_crc = cpu_to_le32(crc);
Sage Weil31b80062009-10-06 11:31:13 -07001552
Alex Elderf5db90b2013-03-11 23:34:23 -05001553 return ret;
1554 }
Alex Elder143334f2013-03-29 11:44:10 -05001555 if (do_datacrc && cursor->need_crc)
1556 crc = ceph_crc32c_page(crc, page, page_offset, length);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05001557 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
Sage Weil31b80062009-10-06 11:31:13 -07001558 }
1559
Alex Elder34d2d202013-03-08 20:58:59 -06001560 dout("%s %p msg %p done\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07001561
1562 /* prepare and queue up footer, too */
Alex Elderf5db90b2013-03-11 23:34:23 -05001563 if (do_datacrc)
1564 msg->footer.data_crc = cpu_to_le32(crc);
1565 else
Alex Elder84ca8fc2012-06-11 14:57:13 -05001566 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
Alex Eldere2200422012-05-23 14:35:23 -05001567 con_out_kvec_reset(con);
Alex Elder859eb792012-02-14 14:05:33 -06001568 prepare_write_message_footer(con);
Alex Elderf5db90b2013-03-11 23:34:23 -05001569
1570 return 1; /* must return > 0 to indicate success */
Sage Weil31b80062009-10-06 11:31:13 -07001571}
1572
1573/*
1574 * write some zeros
1575 */
1576static int write_partial_skip(struct ceph_connection *con)
1577{
1578 int ret;
1579
1580 while (con->out_skip > 0) {
Alex Elder31739132012-03-07 11:40:08 -06001581 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
Sage Weil31b80062009-10-06 11:31:13 -07001582
Alex Eldere1dcb122013-03-06 23:39:38 -06001583 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
Sage Weil31b80062009-10-06 11:31:13 -07001584 if (ret <= 0)
1585 goto out;
1586 con->out_skip -= ret;
1587 }
1588 ret = 1;
1589out:
1590 return ret;
1591}
1592
1593/*
1594 * Prepare to read connection handshake, or an ack.
1595 */
Sage Weileed0ef22009-11-10 14:34:36 -08001596static void prepare_read_banner(struct ceph_connection *con)
1597{
1598 dout("prepare_read_banner %p\n", con);
1599 con->in_base_pos = 0;
1600}
1601
Sage Weil31b80062009-10-06 11:31:13 -07001602static void prepare_read_connect(struct ceph_connection *con)
1603{
1604 dout("prepare_read_connect %p\n", con);
1605 con->in_base_pos = 0;
1606}
1607
1608static void prepare_read_ack(struct ceph_connection *con)
1609{
1610 dout("prepare_read_ack %p\n", con);
1611 con->in_base_pos = 0;
1612}
1613
Sage Weil3a230832013-03-25 08:47:40 -07001614static void prepare_read_seq(struct ceph_connection *con)
1615{
1616 dout("prepare_read_seq %p\n", con);
1617 con->in_base_pos = 0;
1618 con->in_tag = CEPH_MSGR_TAG_SEQ;
1619}
1620
Sage Weil31b80062009-10-06 11:31:13 -07001621static void prepare_read_tag(struct ceph_connection *con)
1622{
1623 dout("prepare_read_tag %p\n", con);
1624 con->in_base_pos = 0;
1625 con->in_tag = CEPH_MSGR_TAG_READY;
1626}
1627
1628/*
1629 * Prepare to read a message.
1630 */
1631static int prepare_read_message(struct ceph_connection *con)
1632{
1633 dout("prepare_read_message %p\n", con);
1634 BUG_ON(con->in_msg != NULL);
1635 con->in_base_pos = 0;
1636 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1637 return 0;
1638}
1639
1640
1641static int read_partial(struct ceph_connection *con,
Alex Elderfd516532012-05-10 10:29:50 -05001642 int end, int size, void *object)
Sage Weil31b80062009-10-06 11:31:13 -07001643{
Alex Eldere6cee712012-05-10 10:29:50 -05001644 while (con->in_base_pos < end) {
1645 int left = end - con->in_base_pos;
Sage Weil31b80062009-10-06 11:31:13 -07001646 int have = size - left;
1647 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1648 if (ret <= 0)
1649 return ret;
1650 con->in_base_pos += ret;
1651 }
1652 return 1;
1653}
1654
1655
1656/*
1657 * Read all or part of the connect-side handshake on a new connection
1658 */
Sage Weileed0ef22009-11-10 14:34:36 -08001659static int read_partial_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001660{
Alex Elderfd516532012-05-10 10:29:50 -05001661 int size;
1662 int end;
1663 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07001664
Sage Weileed0ef22009-11-10 14:34:36 -08001665 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
Sage Weil31b80062009-10-06 11:31:13 -07001666
1667 /* peer's banner */
Alex Elderfd516532012-05-10 10:29:50 -05001668 size = strlen(CEPH_BANNER);
1669 end = size;
1670 ret = read_partial(con, end, size, con->in_banner);
Sage Weil31b80062009-10-06 11:31:13 -07001671 if (ret <= 0)
1672 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001673
1674 size = sizeof (con->actual_peer_addr);
1675 end += size;
1676 ret = read_partial(con, end, size, &con->actual_peer_addr);
Sage Weil31b80062009-10-06 11:31:13 -07001677 if (ret <= 0)
1678 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001679
1680 size = sizeof (con->peer_addr_for_me);
1681 end += size;
1682 ret = read_partial(con, end, size, &con->peer_addr_for_me);
Sage Weil31b80062009-10-06 11:31:13 -07001683 if (ret <= 0)
1684 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001685
Sage Weileed0ef22009-11-10 14:34:36 -08001686out:
1687 return ret;
1688}
1689
1690static int read_partial_connect(struct ceph_connection *con)
1691{
Alex Elderfd516532012-05-10 10:29:50 -05001692 int size;
1693 int end;
1694 int ret;
Sage Weileed0ef22009-11-10 14:34:36 -08001695
1696 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1697
Alex Elderfd516532012-05-10 10:29:50 -05001698 size = sizeof (con->in_reply);
1699 end = size;
1700 ret = read_partial(con, end, size, &con->in_reply);
Sage Weil31b80062009-10-06 11:31:13 -07001701 if (ret <= 0)
1702 goto out;
Alex Elderfd516532012-05-10 10:29:50 -05001703
1704 size = le32_to_cpu(con->in_reply.authorizer_len);
1705 end += size;
1706 ret = read_partial(con, end, size, con->auth_reply_buf);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001707 if (ret <= 0)
1708 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07001709
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001710 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1711 con, (int)con->in_reply.tag,
1712 le32_to_cpu(con->in_reply.connect_seq),
Sage Weil31b80062009-10-06 11:31:13 -07001713 le32_to_cpu(con->in_reply.global_seq));
1714out:
1715 return ret;
Sage Weileed0ef22009-11-10 14:34:36 -08001716
Sage Weil31b80062009-10-06 11:31:13 -07001717}
1718
1719/*
1720 * Verify the hello banner looks okay.
1721 */
1722static int verify_hello(struct ceph_connection *con)
1723{
1724 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
Sage Weil13e38c82009-10-09 16:36:34 -07001725 pr_err("connect to %s got bad banner\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001726 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07001727 con->error_msg = "protocol error, bad banner";
1728 return -1;
1729 }
1730 return 0;
1731}
1732
1733static bool addr_is_blank(struct sockaddr_storage *ss)
1734{
Ilya Dryomovc44bd692015-07-09 13:57:52 +03001735 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1736 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1737
Sage Weil31b80062009-10-06 11:31:13 -07001738 switch (ss->ss_family) {
1739 case AF_INET:
Ilya Dryomovc44bd692015-07-09 13:57:52 +03001740 return addr->s_addr == htonl(INADDR_ANY);
Sage Weil31b80062009-10-06 11:31:13 -07001741 case AF_INET6:
Ilya Dryomovc44bd692015-07-09 13:57:52 +03001742 return ipv6_addr_any(addr6);
1743 default:
1744 return true;
Sage Weil31b80062009-10-06 11:31:13 -07001745 }
Sage Weil31b80062009-10-06 11:31:13 -07001746}
1747
1748static int addr_port(struct sockaddr_storage *ss)
1749{
1750 switch (ss->ss_family) {
1751 case AF_INET:
Sage Weilf28bcfb2009-11-04 11:46:35 -08001752 return ntohs(((struct sockaddr_in *)ss)->sin_port);
Sage Weil31b80062009-10-06 11:31:13 -07001753 case AF_INET6:
Sage Weilf28bcfb2009-11-04 11:46:35 -08001754 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
Sage Weil31b80062009-10-06 11:31:13 -07001755 }
1756 return 0;
1757}
1758
1759static void addr_set_port(struct sockaddr_storage *ss, int p)
1760{
1761 switch (ss->ss_family) {
1762 case AF_INET:
1763 ((struct sockaddr_in *)ss)->sin_port = htons(p);
Sage Weila2a79602011-05-12 15:34:24 -07001764 break;
Sage Weil31b80062009-10-06 11:31:13 -07001765 case AF_INET6:
1766 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
Sage Weila2a79602011-05-12 15:34:24 -07001767 break;
Sage Weil31b80062009-10-06 11:31:13 -07001768 }
1769}
1770
1771/*
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001772 * Unlike other *_pton function semantics, zero indicates success.
1773 */
1774static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1775 char delim, const char **ipend)
1776{
Alex Elder99f0f3b2012-01-23 15:49:27 -06001777 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1778 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001779
1780 memset(ss, 0, sizeof(*ss));
1781
1782 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1783 ss->ss_family = AF_INET;
1784 return 0;
1785 }
1786
1787 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1788 ss->ss_family = AF_INET6;
1789 return 0;
1790 }
1791
1792 return -EINVAL;
1793}
1794
1795/*
1796 * Extract hostname string and resolve using kernel DNS facility.
1797 */
1798#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1799static int ceph_dns_resolve_name(const char *name, size_t namelen,
1800 struct sockaddr_storage *ss, char delim, const char **ipend)
1801{
1802 const char *end, *delim_p;
1803 char *colon_p, *ip_addr = NULL;
1804 int ip_len, ret;
1805
1806 /*
1807 * The end of the hostname occurs immediately preceding the delimiter or
1808 * the port marker (':') where the delimiter takes precedence.
1809 */
1810 delim_p = memchr(name, delim, namelen);
1811 colon_p = memchr(name, ':', namelen);
1812
1813 if (delim_p && colon_p)
1814 end = delim_p < colon_p ? delim_p : colon_p;
1815 else if (!delim_p && colon_p)
1816 end = colon_p;
1817 else {
1818 end = delim_p;
1819 if (!end) /* case: hostname:/ */
1820 end = name + namelen;
1821 }
1822
1823 if (end <= name)
1824 return -EINVAL;
1825
1826 /* do dns_resolve upcall */
1827 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1828 if (ip_len > 0)
1829 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1830 else
1831 ret = -ESRCH;
1832
1833 kfree(ip_addr);
1834
1835 *ipend = end;
1836
1837 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1838 ret, ret ? "failed" : ceph_pr_addr(ss));
1839
1840 return ret;
1841}
1842#else
1843static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1844 struct sockaddr_storage *ss, char delim, const char **ipend)
1845{
1846 return -EINVAL;
1847}
1848#endif
1849
1850/*
1851 * Parse a server name (IP or hostname). If a valid IP address is not found
1852 * then try to extract a hostname to resolve using userspace DNS upcall.
1853 */
1854static int ceph_parse_server_name(const char *name, size_t namelen,
1855 struct sockaddr_storage *ss, char delim, const char **ipend)
1856{
1857 int ret;
1858
1859 ret = ceph_pton(name, namelen, ss, delim, ipend);
1860 if (ret)
1861 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1862
1863 return ret;
1864}
1865
1866/*
Sage Weil31b80062009-10-06 11:31:13 -07001867 * Parse an ip[:port] list into an addr array. Use the default
1868 * monitor port if a port isn't specified.
1869 */
1870int ceph_parse_ips(const char *c, const char *end,
1871 struct ceph_entity_addr *addr,
1872 int max_count, int *count)
1873{
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001874 int i, ret = -EINVAL;
Sage Weil31b80062009-10-06 11:31:13 -07001875 const char *p = c;
1876
1877 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1878 for (i = 0; i < max_count; i++) {
1879 const char *ipend;
1880 struct sockaddr_storage *ss = &addr[i].in_addr;
Sage Weil31b80062009-10-06 11:31:13 -07001881 int port;
Sage Weil39139f62010-07-08 09:54:52 -07001882 char delim = ',';
1883
1884 if (*p == '[') {
1885 delim = ']';
1886 p++;
1887 }
Sage Weil31b80062009-10-06 11:31:13 -07001888
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001889 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1890 if (ret)
Sage Weil31b80062009-10-06 11:31:13 -07001891 goto bad;
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001892 ret = -EINVAL;
1893
Sage Weil31b80062009-10-06 11:31:13 -07001894 p = ipend;
1895
Sage Weil39139f62010-07-08 09:54:52 -07001896 if (delim == ']') {
1897 if (*p != ']') {
1898 dout("missing matching ']'\n");
1899 goto bad;
1900 }
1901 p++;
1902 }
1903
Sage Weil31b80062009-10-06 11:31:13 -07001904 /* port? */
1905 if (p < end && *p == ':') {
1906 port = 0;
1907 p++;
1908 while (p < end && *p >= '0' && *p <= '9') {
1909 port = (port * 10) + (*p - '0');
1910 p++;
1911 }
Ilya Dryomovf48db1e2013-12-30 19:21:29 +02001912 if (port == 0)
1913 port = CEPH_MON_PORT;
1914 else if (port > 65535)
Sage Weil31b80062009-10-06 11:31:13 -07001915 goto bad;
1916 } else {
1917 port = CEPH_MON_PORT;
1918 }
1919
1920 addr_set_port(ss, port);
1921
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001922 dout("parse_ips got %s\n", ceph_pr_addr(ss));
Sage Weil31b80062009-10-06 11:31:13 -07001923
1924 if (p == end)
1925 break;
1926 if (*p != ',')
1927 goto bad;
1928 p++;
1929 }
1930
1931 if (p != end)
1932 goto bad;
1933
1934 if (count)
1935 *count = i + 1;
1936 return 0;
1937
1938bad:
Sage Weil39139f62010-07-08 09:54:52 -07001939 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
Noah Watkinsee3b56f2011-09-23 11:48:42 -07001940 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07001941}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001942EXPORT_SYMBOL(ceph_parse_ips);
Sage Weil31b80062009-10-06 11:31:13 -07001943
Sage Weileed0ef22009-11-10 14:34:36 -08001944static int process_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001945{
Sage Weileed0ef22009-11-10 14:34:36 -08001946 dout("process_banner on %p\n", con);
Sage Weil31b80062009-10-06 11:31:13 -07001947
1948 if (verify_hello(con) < 0)
1949 return -1;
1950
Sage Weil63f2d212009-11-03 15:17:56 -08001951 ceph_decode_addr(&con->actual_peer_addr);
1952 ceph_decode_addr(&con->peer_addr_for_me);
1953
Sage Weil31b80062009-10-06 11:31:13 -07001954 /*
1955 * Make sure the other end is who we wanted. note that the other
1956 * end may not yet know their ip address, so if it's 0.0.0.0, give
1957 * them the benefit of the doubt.
1958 */
Sage Weil103e2d32010-01-07 16:12:36 -08001959 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1960 sizeof(con->peer_addr)) != 0 &&
Sage Weil31b80062009-10-06 11:31:13 -07001961 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1962 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
Joe Perchesb9a67892014-09-09 21:17:29 -07001963 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1964 ceph_pr_addr(&con->peer_addr.in_addr),
1965 (int)le32_to_cpu(con->peer_addr.nonce),
1966 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1967 (int)le32_to_cpu(con->actual_peer_addr.nonce));
Sage Weil58bb3b32009-12-23 12:12:31 -08001968 con->error_msg = "wrong peer at address";
Sage Weil31b80062009-10-06 11:31:13 -07001969 return -1;
1970 }
1971
1972 /*
1973 * did we learn our address?
1974 */
1975 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1976 int port = addr_port(&con->msgr->inst.addr.in_addr);
1977
1978 memcpy(&con->msgr->inst.addr.in_addr,
1979 &con->peer_addr_for_me.in_addr,
1980 sizeof(con->peer_addr_for_me.in_addr));
1981 addr_set_port(&con->msgr->inst.addr.in_addr, port);
Sage Weil63f2d212009-11-03 15:17:56 -08001982 encode_my_addr(con->msgr);
Sage Weileed0ef22009-11-10 14:34:36 -08001983 dout("process_banner learned my addr is %s\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001984 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07001985 }
1986
Sage Weileed0ef22009-11-10 14:34:36 -08001987 return 0;
1988}
1989
1990static int process_connect(struct ceph_connection *con)
1991{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001992 u64 sup_feat = con->msgr->supported_features;
1993 u64 req_feat = con->msgr->required_features;
Ilya Dryomov2b3e0c92013-12-24 21:19:24 +02001994 u64 server_feat = ceph_sanitize_features(
1995 le64_to_cpu(con->in_reply.features));
Sage Weil0da5d702011-05-19 11:21:05 -07001996 int ret;
Sage Weil04a419f2009-12-23 09:30:21 -08001997
Sage Weileed0ef22009-11-10 14:34:36 -08001998 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1999
Sage Weil31b80062009-10-06 11:31:13 -07002000 switch (con->in_reply.tag) {
Sage Weil04a419f2009-12-23 09:30:21 -08002001 case CEPH_MSGR_TAG_FEATURES:
2002 pr_err("%s%lld %s feature set mismatch,"
2003 " my %llx < server's %llx, missing %llx\n",
2004 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002005 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil04a419f2009-12-23 09:30:21 -08002006 sup_feat, server_feat, server_feat & ~sup_feat);
2007 con->error_msg = "missing required protocol features";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06002008 reset_connection(con);
Sage Weil04a419f2009-12-23 09:30:21 -08002009 return -1;
2010
Sage Weil31b80062009-10-06 11:31:13 -07002011 case CEPH_MSGR_TAG_BADPROTOVER:
Sage Weil31b80062009-10-06 11:31:13 -07002012 pr_err("%s%lld %s protocol version mismatch,"
2013 " my %d != server's %d\n",
2014 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002015 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil31b80062009-10-06 11:31:13 -07002016 le32_to_cpu(con->out_connect.protocol_version),
2017 le32_to_cpu(con->in_reply.protocol_version));
2018 con->error_msg = "protocol version mismatch";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06002019 reset_connection(con);
Sage Weil31b80062009-10-06 11:31:13 -07002020 return -1;
2021
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002022 case CEPH_MSGR_TAG_BADAUTHORIZER:
2023 con->auth_retry++;
2024 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2025 con->auth_retry);
2026 if (con->auth_retry == 2) {
2027 con->error_msg = "connect authorization failure";
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002028 return -1;
2029 }
Jim Schutt6d4221b2012-08-10 10:37:38 -07002030 con_out_kvec_reset(con);
Alex Eldere825a662012-05-16 15:16:38 -05002031 ret = prepare_write_connect(con);
Sage Weil0da5d702011-05-19 11:21:05 -07002032 if (ret < 0)
2033 return ret;
Sage Weil63733a02010-03-15 15:47:22 -07002034 prepare_read_connect(con);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002035 break;
Sage Weil31b80062009-10-06 11:31:13 -07002036
2037 case CEPH_MSGR_TAG_RESETSESSION:
2038 /*
2039 * If we connected with a large connect_seq but the peer
2040 * has no record of a session with us (no connection, or
2041 * connect_seq == 0), they will send RESETSESION to indicate
2042 * that they must have reset their session, and may have
2043 * dropped messages.
2044 */
2045 dout("process_connect got RESET peer seq %u\n",
Sage Weil5bdca4e2012-07-10 11:53:34 -07002046 le32_to_cpu(con->in_reply.connect_seq));
Sage Weil31b80062009-10-06 11:31:13 -07002047 pr_err("%s%lld %s connection reset\n",
2048 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002049 ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07002050 reset_connection(con);
Jim Schutt6d4221b2012-08-10 10:37:38 -07002051 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002052 ret = prepare_write_connect(con);
2053 if (ret < 0)
2054 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002055 prepare_read_connect(con);
2056
2057 /* Tell ceph about it. */
Sage Weilec302642009-12-22 10:43:42 -08002058 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002059 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2060 if (con->ops->peer_reset)
2061 con->ops->peer_reset(con);
Sage Weilec302642009-12-22 10:43:42 -08002062 mutex_lock(&con->mutex);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002063 if (con->state != CON_STATE_NEGOTIATING)
Sage Weil0da5d702011-05-19 11:21:05 -07002064 return -EAGAIN;
Sage Weil31b80062009-10-06 11:31:13 -07002065 break;
2066
2067 case CEPH_MSGR_TAG_RETRY_SESSION:
2068 /*
2069 * If we sent a smaller connect_seq than the peer has, try
2070 * again with a larger value.
2071 */
Sage Weil5bdca4e2012-07-10 11:53:34 -07002072 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
Sage Weil31b80062009-10-06 11:31:13 -07002073 le32_to_cpu(con->out_connect.connect_seq),
Sage Weil5bdca4e2012-07-10 11:53:34 -07002074 le32_to_cpu(con->in_reply.connect_seq));
2075 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
Jim Schutt6d4221b2012-08-10 10:37:38 -07002076 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002077 ret = prepare_write_connect(con);
2078 if (ret < 0)
2079 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002080 prepare_read_connect(con);
2081 break;
2082
2083 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2084 /*
2085 * If we sent a smaller global_seq than the peer has, try
2086 * again with a larger value.
2087 */
Sage Weileed0ef22009-11-10 14:34:36 -08002088 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
Sage Weil31b80062009-10-06 11:31:13 -07002089 con->peer_global_seq,
Sage Weil5bdca4e2012-07-10 11:53:34 -07002090 le32_to_cpu(con->in_reply.global_seq));
Sage Weil31b80062009-10-06 11:31:13 -07002091 get_global_seq(con->msgr,
Sage Weil5bdca4e2012-07-10 11:53:34 -07002092 le32_to_cpu(con->in_reply.global_seq));
Jim Schutt6d4221b2012-08-10 10:37:38 -07002093 con_out_kvec_reset(con);
Alex Elder5a0f8fd2012-05-16 21:51:59 -05002094 ret = prepare_write_connect(con);
2095 if (ret < 0)
2096 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002097 prepare_read_connect(con);
2098 break;
2099
Sage Weil3a230832013-03-25 08:47:40 -07002100 case CEPH_MSGR_TAG_SEQ:
Sage Weil31b80062009-10-06 11:31:13 -07002101 case CEPH_MSGR_TAG_READY:
Sage Weil04a419f2009-12-23 09:30:21 -08002102 if (req_feat & ~server_feat) {
2103 pr_err("%s%lld %s protocol feature mismatch,"
2104 " my required %llx > server's %llx, need %llx\n",
2105 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002106 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weil04a419f2009-12-23 09:30:21 -08002107 req_feat, server_feat, req_feat & ~server_feat);
2108 con->error_msg = "missing required protocol features";
Sage Weil0fa6ebc2012-12-27 20:27:04 -06002109 reset_connection(con);
Sage Weil04a419f2009-12-23 09:30:21 -08002110 return -1;
2111 }
Sage Weil8dacc7d2012-07-20 17:24:40 -07002112
Alex Elder122070a2012-12-26 10:43:57 -06002113 WARN_ON(con->state != CON_STATE_NEGOTIATING);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002114 con->state = CON_STATE_OPEN;
Sage Weil20e55c42013-03-25 09:30:13 -07002115 con->auth_retry = 0; /* we authenticated; clear flag */
Sage Weil31b80062009-10-06 11:31:13 -07002116 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2117 con->connect_seq++;
Sage Weilaba558e2010-05-12 15:23:30 -07002118 con->peer_features = server_feat;
Sage Weil31b80062009-10-06 11:31:13 -07002119 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2120 con->peer_global_seq,
2121 le32_to_cpu(con->in_reply.connect_seq),
2122 con->connect_seq);
2123 WARN_ON(con->connect_seq !=
2124 le32_to_cpu(con->in_reply.connect_seq));
Sage Weil92ac41d2009-12-14 14:56:56 -08002125
2126 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
Alex Elderc9ffc772013-02-20 10:25:12 -06002127 con_flag_set(con, CON_FLAG_LOSSYTX);
Sage Weil92ac41d2009-12-14 14:56:56 -08002128
Sage Weil85effe12012-07-30 16:22:05 -07002129 con->delay = 0; /* reset backoff memory */
Sage Weil31b80062009-10-06 11:31:13 -07002130
Sage Weil3a230832013-03-25 08:47:40 -07002131 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2132 prepare_write_seq(con);
2133 prepare_read_seq(con);
2134 } else {
2135 prepare_read_tag(con);
2136 }
Sage Weil31b80062009-10-06 11:31:13 -07002137 break;
2138
2139 case CEPH_MSGR_TAG_WAIT:
2140 /*
2141 * If there is a connection race (we are opening
2142 * connections to each other), one of us may just have
2143 * to WAIT. This shouldn't happen if we are the
2144 * client.
2145 */
Sage Weil04177882011-05-12 15:33:17 -07002146 con->error_msg = "protocol error, got WAIT as client";
2147 return -1;
Sage Weil31b80062009-10-06 11:31:13 -07002148
2149 default:
Sage Weil31b80062009-10-06 11:31:13 -07002150 con->error_msg = "protocol error, garbage tag during connect";
2151 return -1;
2152 }
2153 return 0;
2154}
2155
2156
2157/*
2158 * read (part of) an ack
2159 */
2160static int read_partial_ack(struct ceph_connection *con)
2161{
Alex Elderfd516532012-05-10 10:29:50 -05002162 int size = sizeof (con->in_temp_ack);
2163 int end = size;
Sage Weil31b80062009-10-06 11:31:13 -07002164
Alex Elderfd516532012-05-10 10:29:50 -05002165 return read_partial(con, end, size, &con->in_temp_ack);
Sage Weil31b80062009-10-06 11:31:13 -07002166}
2167
Sage Weil31b80062009-10-06 11:31:13 -07002168/*
2169 * We can finally discard anything that's been acked.
2170 */
2171static void process_ack(struct ceph_connection *con)
2172{
2173 struct ceph_msg *m;
2174 u64 ack = le64_to_cpu(con->in_temp_ack);
2175 u64 seq;
2176
Sage Weil31b80062009-10-06 11:31:13 -07002177 while (!list_empty(&con->out_sent)) {
2178 m = list_first_entry(&con->out_sent, struct ceph_msg,
2179 list_head);
2180 seq = le64_to_cpu(m->hdr.seq);
2181 if (seq > ack)
2182 break;
2183 dout("got ack for seq %llu type %d at %p\n", seq,
2184 le16_to_cpu(m->hdr.type), m);
Sage Weil4cf9d542011-07-26 11:27:24 -07002185 m->ack_stamp = jiffies;
Sage Weil31b80062009-10-06 11:31:13 -07002186 ceph_msg_remove(m);
2187 }
Sage Weil31b80062009-10-06 11:31:13 -07002188 prepare_read_tag(con);
2189}
2190
2191
Yehuda Sadeh24504182010-01-08 13:58:34 -08002192static int read_partial_message_section(struct ceph_connection *con,
Sage Weil213c99e2010-08-03 10:25:11 -07002193 struct kvec *section,
2194 unsigned int sec_len, u32 *crc)
Yehuda Sadeh24504182010-01-08 13:58:34 -08002195{
Yehuda Sadeh68b44762010-04-06 15:01:27 -07002196 int ret, left;
Sage Weil31b80062009-10-06 11:31:13 -07002197
Yehuda Sadeh24504182010-01-08 13:58:34 -08002198 BUG_ON(!section);
Sage Weil31b80062009-10-06 11:31:13 -07002199
Yehuda Sadeh24504182010-01-08 13:58:34 -08002200 while (section->iov_len < sec_len) {
2201 BUG_ON(section->iov_base == NULL);
2202 left = sec_len - section->iov_len;
2203 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2204 section->iov_len, left);
2205 if (ret <= 0)
2206 return ret;
2207 section->iov_len += ret;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002208 }
Alex Elderfe3ad592012-02-15 07:43:54 -06002209 if (section->iov_len == sec_len)
2210 *crc = crc32c(0, section->iov_base, section->iov_len);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002211
2212 return 1;
2213}
2214
Alex Elder34d2d202013-03-08 20:58:59 -06002215static int read_partial_msg_data(struct ceph_connection *con)
2216{
2217 struct ceph_msg *msg = con->in_msg;
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002218 struct ceph_msg_data_cursor *cursor = &msg->cursor;
Alex Elder34d2d202013-03-08 20:58:59 -06002219 const bool do_datacrc = !con->msgr->nocrc;
Alex Elder686be202013-03-11 23:34:23 -05002220 struct page *page;
2221 size_t page_offset;
2222 size_t length;
Alex Elderf5db90b2013-03-11 23:34:23 -05002223 u32 crc = 0;
Alex Elder34d2d202013-03-08 20:58:59 -06002224 int ret;
2225
2226 BUG_ON(!msg);
Alex Elder5240d9f2013-03-14 14:09:06 -05002227 if (list_empty(&msg->data))
Alex Elder4c59b4a2013-03-11 23:34:23 -05002228 return -EIO;
Alex Elder34d2d202013-03-08 20:58:59 -06002229
Alex Elderf5db90b2013-03-11 23:34:23 -05002230 if (do_datacrc)
2231 crc = con->in_data_crc;
Alex Elder643c68a2013-03-11 23:34:23 -05002232 while (cursor->resid) {
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002233 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
Alex Elder4c59b4a2013-03-11 23:34:23 -05002234 NULL);
Alex Elder686be202013-03-11 23:34:23 -05002235 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
Alex Elderf5db90b2013-03-11 23:34:23 -05002236 if (ret <= 0) {
2237 if (do_datacrc)
2238 con->in_data_crc = crc;
2239
Alex Elder686be202013-03-11 23:34:23 -05002240 return ret;
Alex Elderf5db90b2013-03-11 23:34:23 -05002241 }
Alex Elder686be202013-03-11 23:34:23 -05002242
2243 if (do_datacrc)
Alex Elderf5db90b2013-03-11 23:34:23 -05002244 crc = ceph_crc32c_page(crc, page, page_offset, ret);
Alex Elder8ae4f4f2013-03-14 14:09:06 -05002245 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
Alex Elder34d2d202013-03-08 20:58:59 -06002246 }
Alex Elderf5db90b2013-03-11 23:34:23 -05002247 if (do_datacrc)
2248 con->in_data_crc = crc;
Alex Elder34d2d202013-03-08 20:58:59 -06002249
2250 return 1; /* must return > 0 to indicate success */
2251}
2252
Sage Weil31b80062009-10-06 11:31:13 -07002253/*
2254 * read (part of) a message.
2255 */
Alex Elder686be202013-03-11 23:34:23 -05002256static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2257
Sage Weil31b80062009-10-06 11:31:13 -07002258static int read_partial_message(struct ceph_connection *con)
2259{
2260 struct ceph_msg *m = con->in_msg;
Alex Elderfd516532012-05-10 10:29:50 -05002261 int size;
2262 int end;
Sage Weil31b80062009-10-06 11:31:13 -07002263 int ret;
Eric Dumazet95c96172012-04-15 05:58:06 +00002264 unsigned int front_len, middle_len, data_len;
Alex Elder37675b02012-03-07 11:40:08 -06002265 bool do_datacrc = !con->msgr->nocrc;
Yan, Zheng33d07332014-11-04 16:33:37 +08002266 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
Sage Weilae187562010-04-22 07:47:01 -07002267 u64 seq;
Alex Elderfe3ad592012-02-15 07:43:54 -06002268 u32 crc;
Sage Weil31b80062009-10-06 11:31:13 -07002269
2270 dout("read_partial_message con %p msg %p\n", con, m);
2271
2272 /* header */
Alex Elderfd516532012-05-10 10:29:50 -05002273 size = sizeof (con->in_hdr);
2274 end = size;
2275 ret = read_partial(con, end, size, &con->in_hdr);
Alex Elder57dac9d2012-05-10 10:29:50 -05002276 if (ret <= 0)
2277 return ret;
Alex Elderfe3ad592012-02-15 07:43:54 -06002278
2279 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2280 if (cpu_to_le32(crc) != con->in_hdr.crc) {
Ilya Dryomov67c64eb2015-03-23 14:52:40 +03002281 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
Alex Elderfe3ad592012-02-15 07:43:54 -06002282 crc, con->in_hdr.crc);
2283 return -EBADMSG;
2284 }
2285
Sage Weil31b80062009-10-06 11:31:13 -07002286 front_len = le32_to_cpu(con->in_hdr.front_len);
2287 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2288 return -EIO;
2289 middle_len = le32_to_cpu(con->in_hdr.middle_len);
Alex Elder7b11ba32013-03-08 18:51:03 -06002290 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
Sage Weil31b80062009-10-06 11:31:13 -07002291 return -EIO;
2292 data_len = le32_to_cpu(con->in_hdr.data_len);
2293 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2294 return -EIO;
2295
Sage Weilae187562010-04-22 07:47:01 -07002296 /* verify seq# */
2297 seq = le64_to_cpu(con->in_hdr.seq);
2298 if ((s64)seq - (s64)con->in_seq < 1) {
Sage Weildf9f86f2010-11-01 15:49:23 -07002299 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
Sage Weilae187562010-04-22 07:47:01 -07002300 ENTITY_NAME(con->peer_name),
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002301 ceph_pr_addr(&con->peer_addr.in_addr),
Sage Weilae187562010-04-22 07:47:01 -07002302 seq, con->in_seq + 1);
2303 con->in_base_pos = -front_len - middle_len - data_len -
2304 sizeof(m->footer);
2305 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weilae187562010-04-22 07:47:01 -07002306 return 0;
2307 } else if ((s64)seq - (s64)con->in_seq > 1) {
2308 pr_err("read_partial_message bad seq %lld expected %lld\n",
2309 seq, con->in_seq + 1);
2310 con->error_msg = "bad message sequence # for incoming message";
Ilya Dryomov67c64eb2015-03-23 14:52:40 +03002311 return -EBADE;
Sage Weilae187562010-04-22 07:47:01 -07002312 }
2313
Sage Weil31b80062009-10-06 11:31:13 -07002314 /* allocate message? */
2315 if (!con->in_msg) {
Sage Weil4740a622012-07-30 18:19:30 -07002316 int skip = 0;
2317
Sage Weil31b80062009-10-06 11:31:13 -07002318 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
Alex Elder6ebc8b32013-03-08 18:51:03 -06002319 front_len, data_len);
Sage Weil4740a622012-07-30 18:19:30 -07002320 ret = ceph_con_in_msg_alloc(con, &skip);
2321 if (ret < 0)
2322 return ret;
Alex Elderf759ebb2013-04-05 14:46:01 -05002323
2324 BUG_ON(!con->in_msg ^ skip);
2325 if (con->in_msg && data_len > con->in_msg->data_length) {
Joe Perchesb9a67892014-09-09 21:17:29 -07002326 pr_warn("%s skipping long message (%u > %zd)\n",
Alex Elderf759ebb2013-04-05 14:46:01 -05002327 __func__, data_len, con->in_msg->data_length);
2328 ceph_msg_put(con->in_msg);
2329 con->in_msg = NULL;
2330 skip = 1;
2331 }
Yehuda Sadeh24504182010-01-08 13:58:34 -08002332 if (skip) {
Sage Weil31b80062009-10-06 11:31:13 -07002333 /* skip this message */
Sage Weila79832f2010-04-01 16:06:19 -07002334 dout("alloc_msg said skip message\n");
Sage Weil31b80062009-10-06 11:31:13 -07002335 con->in_base_pos = -front_len - middle_len - data_len -
2336 sizeof(m->footer);
2337 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weil684be252010-04-21 20:45:59 -07002338 con->in_seq++;
Sage Weil31b80062009-10-06 11:31:13 -07002339 return 0;
2340 }
Alex Elder38941f82012-06-01 14:56:43 -05002341
Sage Weil4740a622012-07-30 18:19:30 -07002342 BUG_ON(!con->in_msg);
Alex Elder38941f82012-06-01 14:56:43 -05002343 BUG_ON(con->in_msg->con != con);
Sage Weil31b80062009-10-06 11:31:13 -07002344 m = con->in_msg;
2345 m->front.iov_len = 0; /* haven't read it yet */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002346 if (m->middle)
2347 m->middle->vec.iov_len = 0;
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08002348
Alex Elder78625052013-03-06 23:39:39 -06002349 /* prepare for data payload, if any */
Sage Weila4107022012-07-30 16:20:25 -07002350
Alex Elder78625052013-03-06 23:39:39 -06002351 if (data_len)
Alex Elder98fa5dd2013-04-02 12:09:50 -05002352 prepare_message_data(con->in_msg, data_len);
Sage Weil31b80062009-10-06 11:31:13 -07002353 }
2354
2355 /* front */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002356 ret = read_partial_message_section(con, &m->front, front_len,
2357 &con->in_front_crc);
2358 if (ret <= 0)
2359 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002360
2361 /* middle */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002362 if (m->middle) {
Sage Weil213c99e2010-08-03 10:25:11 -07002363 ret = read_partial_message_section(con, &m->middle->vec,
2364 middle_len,
Yehuda Sadeh24504182010-01-08 13:58:34 -08002365 &con->in_middle_crc);
Sage Weil31b80062009-10-06 11:31:13 -07002366 if (ret <= 0)
2367 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002368 }
2369
2370 /* (page) data */
Alex Elder34d2d202013-03-08 20:58:59 -06002371 if (data_len) {
2372 ret = read_partial_msg_data(con);
2373 if (ret <= 0)
2374 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07002375 }
2376
Sage Weil31b80062009-10-06 11:31:13 -07002377 /* footer */
Yan, Zheng33d07332014-11-04 16:33:37 +08002378 if (need_sign)
2379 size = sizeof(m->footer);
2380 else
2381 size = sizeof(m->old_footer);
2382
Alex Elderfd516532012-05-10 10:29:50 -05002383 end += size;
2384 ret = read_partial(con, end, size, &m->footer);
Alex Elder57dac9d2012-05-10 10:29:50 -05002385 if (ret <= 0)
2386 return ret;
2387
Yan, Zheng33d07332014-11-04 16:33:37 +08002388 if (!need_sign) {
2389 m->footer.flags = m->old_footer.flags;
2390 m->footer.sig = 0;
2391 }
2392
Sage Weil31b80062009-10-06 11:31:13 -07002393 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2394 m, front_len, m->footer.front_crc, middle_len,
2395 m->footer.middle_crc, data_len, m->footer.data_crc);
2396
2397 /* crc ok? */
2398 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2399 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2400 m, con->in_front_crc, m->footer.front_crc);
2401 return -EBADMSG;
2402 }
2403 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2404 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2405 m, con->in_middle_crc, m->footer.middle_crc);
2406 return -EBADMSG;
2407 }
Alex Elderbca064d2012-02-15 07:43:54 -06002408 if (do_datacrc &&
Sage Weil31b80062009-10-06 11:31:13 -07002409 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2410 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2411 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2412 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2413 return -EBADMSG;
2414 }
2415
Yan, Zheng33d07332014-11-04 16:33:37 +08002416 if (need_sign && con->ops->check_message_signature &&
2417 con->ops->check_message_signature(con, m)) {
2418 pr_err("read_partial_message %p signature check failed\n", m);
2419 return -EBADMSG;
2420 }
2421
Sage Weil31b80062009-10-06 11:31:13 -07002422 return 1; /* done! */
2423}
2424
2425/*
2426 * Process message. This happens in the worker thread. The callback should
2427 * be careful not to do anything that waits on other incoming messages or it
2428 * may deadlock.
2429 */
2430static void process_message(struct ceph_connection *con)
2431{
Sage Weil5e095e82009-12-14 14:30:34 -08002432 struct ceph_msg *msg;
Sage Weil31b80062009-10-06 11:31:13 -07002433
Alex Elder38941f82012-06-01 14:56:43 -05002434 BUG_ON(con->in_msg->con != con);
2435 con->in_msg->con = NULL;
Sage Weil5e095e82009-12-14 14:30:34 -08002436 msg = con->in_msg;
Sage Weil31b80062009-10-06 11:31:13 -07002437 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -07002438 con->ops->put(con);
Sage Weil31b80062009-10-06 11:31:13 -07002439
2440 /* if first message, set peer_name */
2441 if (con->peer_name.type == 0)
Sage Weildbad1852010-03-25 15:45:38 -07002442 con->peer_name = msg->hdr.src;
Sage Weil31b80062009-10-06 11:31:13 -07002443
Sage Weil31b80062009-10-06 11:31:13 -07002444 con->in_seq++;
Sage Weilec302642009-12-22 10:43:42 -08002445 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002446
2447 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2448 msg, le64_to_cpu(msg->hdr.seq),
Sage Weildbad1852010-03-25 15:45:38 -07002449 ENTITY_NAME(msg->hdr.src),
Sage Weil31b80062009-10-06 11:31:13 -07002450 le16_to_cpu(msg->hdr.type),
2451 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2452 le32_to_cpu(msg->hdr.front_len),
2453 le32_to_cpu(msg->hdr.data_len),
2454 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2455 con->ops->dispatch(con, msg);
Sage Weilec302642009-12-22 10:43:42 -08002456
2457 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002458}
2459
2460
2461/*
2462 * Write something to the socket. Called in a worker thread when the
2463 * socket appears to be writeable and we have something ready to send.
2464 */
2465static int try_write(struct ceph_connection *con)
2466{
Sage Weil31b80062009-10-06 11:31:13 -07002467 int ret = 1;
2468
Sage Weild59315c2012-06-21 12:49:23 -07002469 dout("try_write start %p state %lu\n", con, con->state);
Sage Weil31b80062009-10-06 11:31:13 -07002470
Sage Weil31b80062009-10-06 11:31:13 -07002471more:
2472 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2473
2474 /* open the socket first? */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002475 if (con->state == CON_STATE_PREOPEN) {
2476 BUG_ON(con->sock);
2477 con->state = CON_STATE_CONNECTING;
Alex Eldera5988c42012-05-29 11:04:58 -05002478
Alex Eldere2200422012-05-23 14:35:23 -05002479 con_out_kvec_reset(con);
Alex Eldere825a662012-05-16 15:16:38 -05002480 prepare_write_banner(con);
Sage Weileed0ef22009-11-10 14:34:36 -08002481 prepare_read_banner(con);
Sage Weil31b80062009-10-06 11:31:13 -07002482
Sage Weilcf3e5c42009-12-11 09:48:05 -08002483 BUG_ON(con->in_msg);
Sage Weil31b80062009-10-06 11:31:13 -07002484 con->in_tag = CEPH_MSGR_TAG_READY;
2485 dout("try_write initiating connect on %p new state %lu\n",
2486 con, con->state);
Alex Elder41617d02012-02-14 14:05:33 -06002487 ret = ceph_tcp_connect(con);
2488 if (ret < 0) {
Sage Weil31b80062009-10-06 11:31:13 -07002489 con->error_msg = "connect error";
Sage Weil31b80062009-10-06 11:31:13 -07002490 goto out;
2491 }
2492 }
2493
2494more_kvec:
2495 /* kvec data queued? */
2496 if (con->out_skip) {
2497 ret = write_partial_skip(con);
2498 if (ret <= 0)
Sage Weil42961d22011-01-25 08:19:34 -08002499 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002500 }
2501 if (con->out_kvec_left) {
2502 ret = write_partial_kvec(con);
2503 if (ret <= 0)
Sage Weil42961d22011-01-25 08:19:34 -08002504 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002505 }
2506
2507 /* msg pages? */
2508 if (con->out_msg) {
Sage Weilc86a2932009-12-14 14:04:30 -08002509 if (con->out_msg_done) {
2510 ceph_msg_put(con->out_msg);
2511 con->out_msg = NULL; /* we're done with this one */
2512 goto do_next;
2513 }
2514
Alex Elder34d2d202013-03-08 20:58:59 -06002515 ret = write_partial_message_data(con);
Sage Weil31b80062009-10-06 11:31:13 -07002516 if (ret == 1)
2517 goto more_kvec; /* we need to send the footer, too! */
2518 if (ret == 0)
Sage Weil42961d22011-01-25 08:19:34 -08002519 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002520 if (ret < 0) {
Alex Elder34d2d202013-03-08 20:58:59 -06002521 dout("try_write write_partial_message_data err %d\n",
Sage Weil31b80062009-10-06 11:31:13 -07002522 ret);
Sage Weil42961d22011-01-25 08:19:34 -08002523 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002524 }
2525 }
2526
Sage Weilc86a2932009-12-14 14:04:30 -08002527do_next:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002528 if (con->state == CON_STATE_OPEN) {
Sage Weil31b80062009-10-06 11:31:13 -07002529 /* is anything else pending? */
2530 if (!list_empty(&con->out_queue)) {
2531 prepare_write_message(con);
2532 goto more;
2533 }
2534 if (con->in_seq > con->in_seq_acked) {
2535 prepare_write_ack(con);
2536 goto more;
2537 }
Alex Elderc9ffc772013-02-20 10:25:12 -06002538 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
Sage Weil31b80062009-10-06 11:31:13 -07002539 prepare_write_keepalive(con);
2540 goto more;
2541 }
2542 }
2543
2544 /* Nothing to do! */
Alex Elderc9ffc772013-02-20 10:25:12 -06002545 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
Sage Weil31b80062009-10-06 11:31:13 -07002546 dout("try_write nothing else to write.\n");
Sage Weil31b80062009-10-06 11:31:13 -07002547 ret = 0;
2548out:
Sage Weil42961d22011-01-25 08:19:34 -08002549 dout("try_write done on %p ret %d\n", con, ret);
Sage Weil31b80062009-10-06 11:31:13 -07002550 return ret;
2551}
2552
2553
2554
2555/*
2556 * Read what we can from the socket.
2557 */
2558static int try_read(struct ceph_connection *con)
2559{
Sage Weil31b80062009-10-06 11:31:13 -07002560 int ret = -1;
2561
Sage Weil31b80062009-10-06 11:31:13 -07002562more:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002563 dout("try_read start on %p state %lu\n", con, con->state);
2564 if (con->state != CON_STATE_CONNECTING &&
2565 con->state != CON_STATE_NEGOTIATING &&
2566 con->state != CON_STATE_OPEN)
2567 return 0;
2568
2569 BUG_ON(!con->sock);
2570
Sage Weil31b80062009-10-06 11:31:13 -07002571 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2572 con->in_base_pos);
Sage Weil0da5d702011-05-19 11:21:05 -07002573
Sage Weil8dacc7d2012-07-20 17:24:40 -07002574 if (con->state == CON_STATE_CONNECTING) {
Alex Elder7593af92012-05-24 11:55:03 -05002575 dout("try_read connecting\n");
2576 ret = read_partial_banner(con);
2577 if (ret <= 0)
Alex Elderab166d52012-05-31 11:37:29 -05002578 goto out;
Alex Elder7593af92012-05-24 11:55:03 -05002579 ret = process_banner(con);
2580 if (ret < 0)
2581 goto out;
2582
Sage Weil8dacc7d2012-07-20 17:24:40 -07002583 con->state = CON_STATE_NEGOTIATING;
Alex Elder7593af92012-05-24 11:55:03 -05002584
Jim Schutt6d4221b2012-08-10 10:37:38 -07002585 /*
2586 * Received banner is good, exchange connection info.
2587 * Do not reset out_kvec, as sending our banner raced
2588 * with receiving peer banner after connect completed.
2589 */
Alex Elder7593af92012-05-24 11:55:03 -05002590 ret = prepare_write_connect(con);
2591 if (ret < 0)
2592 goto out;
2593 prepare_read_connect(con);
2594
2595 /* Send connection info before awaiting response */
Sage Weil0da5d702011-05-19 11:21:05 -07002596 goto out;
2597 }
2598
Sage Weil8dacc7d2012-07-20 17:24:40 -07002599 if (con->state == CON_STATE_NEGOTIATING) {
Alex Elder7593af92012-05-24 11:55:03 -05002600 dout("try_read negotiating\n");
Sage Weil31b80062009-10-06 11:31:13 -07002601 ret = read_partial_connect(con);
2602 if (ret <= 0)
Sage Weil31b80062009-10-06 11:31:13 -07002603 goto out;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002604 ret = process_connect(con);
2605 if (ret < 0)
2606 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002607 goto more;
2608 }
2609
Alex Elder122070a2012-12-26 10:43:57 -06002610 WARN_ON(con->state != CON_STATE_OPEN);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002611
Sage Weil31b80062009-10-06 11:31:13 -07002612 if (con->in_base_pos < 0) {
2613 /*
2614 * skipping + discarding content.
2615 *
2616 * FIXME: there must be a better way to do this!
2617 */
Alex Elder84495f42012-02-15 07:43:55 -06002618 static char buf[SKIP_BUF_SIZE];
2619 int skip = min((int) sizeof (buf), -con->in_base_pos);
2620
Sage Weil31b80062009-10-06 11:31:13 -07002621 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2622 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2623 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002624 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002625 con->in_base_pos += ret;
2626 if (con->in_base_pos)
2627 goto more;
2628 }
2629 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2630 /*
2631 * what's next?
2632 */
2633 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2634 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002635 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002636 dout("try_read got tag %d\n", (int)con->in_tag);
2637 switch (con->in_tag) {
2638 case CEPH_MSGR_TAG_MSG:
2639 prepare_read_message(con);
2640 break;
2641 case CEPH_MSGR_TAG_ACK:
2642 prepare_read_ack(con);
2643 break;
2644 case CEPH_MSGR_TAG_CLOSE:
Sage Weil8dacc7d2012-07-20 17:24:40 -07002645 con_close_socket(con);
2646 con->state = CON_STATE_CLOSED;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002647 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002648 default:
2649 goto bad_tag;
2650 }
2651 }
2652 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2653 ret = read_partial_message(con);
2654 if (ret <= 0) {
2655 switch (ret) {
2656 case -EBADMSG:
2657 con->error_msg = "bad crc";
Ilya Dryomov67c64eb2015-03-23 14:52:40 +03002658 /* fall through */
2659 case -EBADE:
Sage Weil31b80062009-10-06 11:31:13 -07002660 ret = -EIO;
Sage Weil98bdb0a2011-01-25 08:17:48 -08002661 break;
Sage Weil31b80062009-10-06 11:31:13 -07002662 case -EIO:
2663 con->error_msg = "io error";
Sage Weil98bdb0a2011-01-25 08:17:48 -08002664 break;
Sage Weil31b80062009-10-06 11:31:13 -07002665 }
Sage Weil98bdb0a2011-01-25 08:17:48 -08002666 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002667 }
2668 if (con->in_tag == CEPH_MSGR_TAG_READY)
2669 goto more;
2670 process_message(con);
Sage Weil7b862e02012-07-30 18:16:56 -07002671 if (con->state == CON_STATE_OPEN)
2672 prepare_read_tag(con);
Sage Weil31b80062009-10-06 11:31:13 -07002673 goto more;
2674 }
Sage Weil3a230832013-03-25 08:47:40 -07002675 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2676 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2677 /*
2678 * the final handshake seq exchange is semantically
2679 * equivalent to an ACK
2680 */
Sage Weil31b80062009-10-06 11:31:13 -07002681 ret = read_partial_ack(con);
2682 if (ret <= 0)
Sage Weil98bdb0a2011-01-25 08:17:48 -08002683 goto out;
Sage Weil31b80062009-10-06 11:31:13 -07002684 process_ack(con);
2685 goto more;
2686 }
2687
Sage Weil31b80062009-10-06 11:31:13 -07002688out:
Sage Weil98bdb0a2011-01-25 08:17:48 -08002689 dout("try_read done on %p ret %d\n", con, ret);
Sage Weil31b80062009-10-06 11:31:13 -07002690 return ret;
2691
2692bad_tag:
2693 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2694 con->error_msg = "protocol error, garbage tag";
2695 ret = -1;
2696 goto out;
2697}
2698
2699
2700/*
Alex Elder802c6d92012-10-08 20:37:30 -07002701 * Atomically queue work on a connection after the specified delay.
2702 * Bump @con reference to avoid races with connection teardown.
2703 * Returns 0 if work was queued, or an error code otherwise.
Sage Weil31b80062009-10-06 11:31:13 -07002704 */
Alex Elder802c6d92012-10-08 20:37:30 -07002705static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
Sage Weil31b80062009-10-06 11:31:13 -07002706{
Sage Weil31b80062009-10-06 11:31:13 -07002707 if (!con->ops->get(con)) {
Alex Elder802c6d92012-10-08 20:37:30 -07002708 dout("%s %p ref count 0\n", __func__, con);
Alex Elder802c6d92012-10-08 20:37:30 -07002709 return -ENOENT;
Sage Weil31b80062009-10-06 11:31:13 -07002710 }
2711
Alex Elder802c6d92012-10-08 20:37:30 -07002712 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2713 dout("%s %p - already queued\n", __func__, con);
Sage Weil31b80062009-10-06 11:31:13 -07002714 con->ops->put(con);
Alex Elder802c6d92012-10-08 20:37:30 -07002715 return -EBUSY;
Sage Weil31b80062009-10-06 11:31:13 -07002716 }
Alex Elder802c6d92012-10-08 20:37:30 -07002717
2718 dout("%s %p %lu\n", __func__, con, delay);
Alex Elder802c6d92012-10-08 20:37:30 -07002719 return 0;
2720}
2721
2722static void queue_con(struct ceph_connection *con)
2723{
2724 (void) queue_con_delay(con, 0);
Sage Weil31b80062009-10-06 11:31:13 -07002725}
2726
Ilya Dryomov37ab77a2014-06-24 16:21:45 +04002727static void cancel_con(struct ceph_connection *con)
2728{
2729 if (cancel_delayed_work(&con->work)) {
2730 dout("%s %p\n", __func__, con);
2731 con->ops->put(con);
2732 }
2733}
2734
Alex Elder7bb21d682012-12-07 19:50:07 -06002735static bool con_sock_closed(struct ceph_connection *con)
2736{
Alex Elderc9ffc772013-02-20 10:25:12 -06002737 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
Alex Elder7bb21d682012-12-07 19:50:07 -06002738 return false;
2739
2740#define CASE(x) \
2741 case CON_STATE_ ## x: \
2742 con->error_msg = "socket closed (con state " #x ")"; \
2743 break;
2744
2745 switch (con->state) {
2746 CASE(CLOSED);
2747 CASE(PREOPEN);
2748 CASE(CONNECTING);
2749 CASE(NEGOTIATING);
2750 CASE(OPEN);
2751 CASE(STANDBY);
2752 default:
Joe Perchesb9a67892014-09-09 21:17:29 -07002753 pr_warn("%s con %p unrecognized state %lu\n",
Alex Elder7bb21d682012-12-07 19:50:07 -06002754 __func__, con, con->state);
2755 con->error_msg = "unrecognized con state";
2756 BUG();
2757 break;
2758 }
2759#undef CASE
2760
2761 return true;
2762}
2763
Alex Elderf20a39f2013-02-19 12:25:57 -06002764static bool con_backoff(struct ceph_connection *con)
2765{
2766 int ret;
2767
2768 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2769 return false;
2770
2771 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2772 if (ret) {
2773 dout("%s: con %p FAILED to back off %lu\n", __func__,
2774 con, con->delay);
2775 BUG_ON(ret == -ENOENT);
2776 con_flag_set(con, CON_FLAG_BACKOFF);
2777 }
2778
2779 return true;
2780}
2781
Alex Elder93209262013-02-19 12:25:57 -06002782/* Finish fault handling; con->mutex must *not* be held here */
2783
2784static void con_fault_finish(struct ceph_connection *con)
2785{
2786 /*
2787 * in case we faulted due to authentication, invalidate our
2788 * current tickets so that we can get new ones.
2789 */
2790 if (con->auth_retry && con->ops->invalidate_authorizer) {
2791 dout("calling invalidate_authorizer()\n");
2792 con->ops->invalidate_authorizer(con);
2793 }
2794
2795 if (con->ops->fault)
2796 con->ops->fault(con);
2797}
2798
Sage Weil31b80062009-10-06 11:31:13 -07002799/*
2800 * Do some work on a connection. Drop a connection ref when we're done.
2801 */
2802static void con_work(struct work_struct *work)
2803{
2804 struct ceph_connection *con = container_of(work, struct ceph_connection,
2805 work.work);
Alex Elder49659412013-02-19 12:25:57 -06002806 bool fault;
Sage Weil31b80062009-10-06 11:31:13 -07002807
Sage Weil9dd46582010-04-28 13:51:50 -07002808 mutex_lock(&con->mutex);
Alex Elder49659412013-02-19 12:25:57 -06002809 while (true) {
2810 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07002811
Alex Elder49659412013-02-19 12:25:57 -06002812 if ((fault = con_sock_closed(con))) {
2813 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2814 break;
2815 }
2816 if (con_backoff(con)) {
2817 dout("%s: con %p BACKOFF\n", __func__, con);
2818 break;
2819 }
2820 if (con->state == CON_STATE_STANDBY) {
2821 dout("%s: con %p STANDBY\n", __func__, con);
2822 break;
2823 }
2824 if (con->state == CON_STATE_CLOSED) {
2825 dout("%s: con %p CLOSED\n", __func__, con);
2826 BUG_ON(con->sock);
2827 break;
2828 }
2829 if (con->state == CON_STATE_PREOPEN) {
2830 dout("%s: con %p PREOPEN\n", __func__, con);
2831 BUG_ON(con->sock);
2832 }
Sage Weil0da5d702011-05-19 11:21:05 -07002833
Alex Elder49659412013-02-19 12:25:57 -06002834 ret = try_read(con);
2835 if (ret < 0) {
2836 if (ret == -EAGAIN)
2837 continue;
Ilya Dryomov67c64eb2015-03-23 14:52:40 +03002838 if (!con->error_msg)
2839 con->error_msg = "socket error on read";
Alex Elder49659412013-02-19 12:25:57 -06002840 fault = true;
2841 break;
2842 }
2843
2844 ret = try_write(con);
2845 if (ret < 0) {
2846 if (ret == -EAGAIN)
2847 continue;
Ilya Dryomov67c64eb2015-03-23 14:52:40 +03002848 if (!con->error_msg)
2849 con->error_msg = "socket error on write";
Alex Elder49659412013-02-19 12:25:57 -06002850 fault = true;
2851 }
2852
2853 break; /* If we make it to here, we're done */
Sage Weil3a140a02012-07-30 16:24:21 -07002854 }
Alex Elderb6e7b6a2013-02-19 12:25:57 -06002855 if (fault)
2856 con_fault(con);
Sage Weil9dd46582010-04-28 13:51:50 -07002857 mutex_unlock(&con->mutex);
Sage Weil0da5d702011-05-19 11:21:05 -07002858
Alex Elderb6e7b6a2013-02-19 12:25:57 -06002859 if (fault)
2860 con_fault_finish(con);
2861
2862 con->ops->put(con);
Sage Weil31b80062009-10-06 11:31:13 -07002863}
2864
Sage Weil31b80062009-10-06 11:31:13 -07002865/*
2866 * Generic error/fault handler. A retry mechanism is used with
2867 * exponential backoff
2868 */
Alex Elder93209262013-02-19 12:25:57 -06002869static void con_fault(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07002870{
Sage Weil31b80062009-10-06 11:31:13 -07002871 dout("fault %p state %lu to peer %s\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002872 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
Sage Weil31b80062009-10-06 11:31:13 -07002873
Ilya Dryomov67c64eb2015-03-23 14:52:40 +03002874 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2875 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2876 con->error_msg = NULL;
2877
Alex Elder122070a2012-12-26 10:43:57 -06002878 WARN_ON(con->state != CON_STATE_CONNECTING &&
Sage Weil8dacc7d2012-07-20 17:24:40 -07002879 con->state != CON_STATE_NEGOTIATING &&
2880 con->state != CON_STATE_OPEN);
Sage Weilec302642009-12-22 10:43:42 -08002881
Sage Weil31b80062009-10-06 11:31:13 -07002882 con_close_socket(con);
Sage Weil5e095e82009-12-14 14:30:34 -08002883
Alex Elderc9ffc772013-02-20 10:25:12 -06002884 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
Sage Weil8dacc7d2012-07-20 17:24:40 -07002885 dout("fault on LOSSYTX channel, marking CLOSED\n");
2886 con->state = CON_STATE_CLOSED;
Alex Elder93209262013-02-19 12:25:57 -06002887 return;
Sage Weil3b5ede02012-07-20 15:22:53 -07002888 }
2889
Sage Weil5e095e82009-12-14 14:30:34 -08002890 if (con->in_msg) {
Alex Elder38941f82012-06-01 14:56:43 -05002891 BUG_ON(con->in_msg->con != con);
2892 con->in_msg->con = NULL;
Sage Weil5e095e82009-12-14 14:30:34 -08002893 ceph_msg_put(con->in_msg);
2894 con->in_msg = NULL;
Sage Weil36eb71a2012-06-21 12:47:08 -07002895 con->ops->put(con);
Sage Weil5e095e82009-12-14 14:30:34 -08002896 }
Sage Weil31b80062009-10-06 11:31:13 -07002897
Sage Weile80a52d2010-02-25 12:40:45 -08002898 /* Requeue anything that hasn't been acked */
2899 list_splice_init(&con->out_sent, &con->out_queue);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002900
Sage Weile76661d2011-03-03 10:10:15 -08002901 /* If there are no messages queued or keepalive pending, place
2902 * the connection in a STANDBY state */
2903 if (list_empty(&con->out_queue) &&
Alex Elderc9ffc772013-02-20 10:25:12 -06002904 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
Sage Weile00de342011-03-04 12:25:05 -08002905 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
Alex Elderc9ffc772013-02-20 10:25:12 -06002906 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002907 con->state = CON_STATE_STANDBY;
Sage Weile80a52d2010-02-25 12:40:45 -08002908 } else {
2909 /* retry after a delay. */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002910 con->state = CON_STATE_PREOPEN;
Sage Weile80a52d2010-02-25 12:40:45 -08002911 if (con->delay == 0)
2912 con->delay = BASE_DELAY_INTERVAL;
2913 else if (con->delay < MAX_DELAY_INTERVAL)
2914 con->delay *= 2;
Alex Elderc9ffc772013-02-20 10:25:12 -06002915 con_flag_set(con, CON_FLAG_BACKOFF);
Alex Elder8618e302012-10-08 20:37:30 -07002916 queue_con(con);
Sage Weil31b80062009-10-06 11:31:13 -07002917 }
Sage Weil31b80062009-10-06 11:31:13 -07002918}
2919
2920
2921
2922/*
Alex Elder15d98822012-05-26 23:26:43 -05002923 * initialize a new messenger instance
Sage Weil31b80062009-10-06 11:31:13 -07002924 */
Alex Elder15d98822012-05-26 23:26:43 -05002925void ceph_messenger_init(struct ceph_messenger *msgr,
2926 struct ceph_entity_addr *myaddr,
Ilya Dryomov12b46292013-12-24 21:19:23 +02002927 u64 supported_features,
2928 u64 required_features,
Chaitanya Huilgolba988f82015-01-23 16:41:25 +05302929 bool nocrc,
2930 bool tcp_nodelay)
Sage Weil31b80062009-10-06 11:31:13 -07002931{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002932 msgr->supported_features = supported_features;
2933 msgr->required_features = required_features;
2934
Sage Weil31b80062009-10-06 11:31:13 -07002935 spin_lock_init(&msgr->global_seq_lock);
2936
Sage Weil31b80062009-10-06 11:31:13 -07002937 if (myaddr)
2938 msgr->inst.addr = *myaddr;
2939
2940 /* select a random nonce */
Sage Weilac8839d2010-01-27 14:28:10 -08002941 msgr->inst.addr.type = 0;
Sage Weil103e2d32010-01-07 16:12:36 -08002942 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
Sage Weil63f2d212009-11-03 15:17:56 -08002943 encode_my_addr(msgr);
Alex Elder15d98822012-05-26 23:26:43 -05002944 msgr->nocrc = nocrc;
Chaitanya Huilgolba988f82015-01-23 16:41:25 +05302945 msgr->tcp_nodelay = tcp_nodelay;
Sage Weil31b80062009-10-06 11:31:13 -07002946
Guanjun Hea2a32582012-07-08 19:50:33 -07002947 atomic_set(&msgr->stopping, 0);
Ilya Dryomov757856d2015-06-25 17:47:45 +03002948 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
Sage Weil31b80062009-10-06 11:31:13 -07002949
Alex Elder15d98822012-05-26 23:26:43 -05002950 dout("%s %p\n", __func__, msgr);
Sage Weil31b80062009-10-06 11:31:13 -07002951}
Alex Elder15d98822012-05-26 23:26:43 -05002952EXPORT_SYMBOL(ceph_messenger_init);
Sage Weil31b80062009-10-06 11:31:13 -07002953
Ilya Dryomov757856d2015-06-25 17:47:45 +03002954void ceph_messenger_fini(struct ceph_messenger *msgr)
2955{
2956 put_net(read_pnet(&msgr->net));
2957}
2958EXPORT_SYMBOL(ceph_messenger_fini);
2959
Sage Weile00de342011-03-04 12:25:05 -08002960static void clear_standby(struct ceph_connection *con)
2961{
2962 /* come back from STANDBY? */
Sage Weil8dacc7d2012-07-20 17:24:40 -07002963 if (con->state == CON_STATE_STANDBY) {
Sage Weile00de342011-03-04 12:25:05 -08002964 dout("clear_standby %p and ++connect_seq\n", con);
Sage Weil8dacc7d2012-07-20 17:24:40 -07002965 con->state = CON_STATE_PREOPEN;
Sage Weile00de342011-03-04 12:25:05 -08002966 con->connect_seq++;
Alex Elderc9ffc772013-02-20 10:25:12 -06002967 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2968 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
Sage Weile00de342011-03-04 12:25:05 -08002969 }
2970}
2971
Sage Weil31b80062009-10-06 11:31:13 -07002972/*
2973 * Queue up an outgoing message on the given connection.
2974 */
2975void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2976{
Sage Weila59b55a2012-07-20 15:34:04 -07002977 /* set src+dst */
2978 msg->hdr.src = con->msgr->inst.name;
2979 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2980 msg->needs_out_seq = true;
2981
2982 mutex_lock(&con->mutex);
2983
Sage Weil8dacc7d2012-07-20 17:24:40 -07002984 if (con->state == CON_STATE_CLOSED) {
Sage Weil31b80062009-10-06 11:31:13 -07002985 dout("con_send %p closed, dropping %p\n", con, msg);
2986 ceph_msg_put(msg);
Sage Weila59b55a2012-07-20 15:34:04 -07002987 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002988 return;
2989 }
2990
Alex Elder38941f82012-06-01 14:56:43 -05002991 BUG_ON(msg->con != NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -07002992 msg->con = con->ops->get(con);
Alex Elder92ce0342012-06-04 14:43:33 -05002993 BUG_ON(msg->con == NULL);
Sage Weil31b80062009-10-06 11:31:13 -07002994
Sage Weil31b80062009-10-06 11:31:13 -07002995 BUG_ON(!list_empty(&msg->list_head));
2996 list_add_tail(&msg->list_head, &con->out_queue);
2997 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2998 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2999 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3000 le32_to_cpu(msg->hdr.front_len),
3001 le32_to_cpu(msg->hdr.middle_len),
3002 le32_to_cpu(msg->hdr.data_len));
Sage Weil00650932012-07-20 15:33:04 -07003003
3004 clear_standby(con);
Sage Weilec302642009-12-22 10:43:42 -08003005 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07003006
3007 /* if there wasn't anything waiting to send before, queue
3008 * new work */
Alex Elderc9ffc772013-02-20 10:25:12 -06003009 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
Sage Weil31b80062009-10-06 11:31:13 -07003010 queue_con(con);
3011}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003012EXPORT_SYMBOL(ceph_con_send);
Sage Weil31b80062009-10-06 11:31:13 -07003013
3014/*
3015 * Revoke a message that was previously queued for send
3016 */
Alex Elder6740a842012-06-01 14:56:43 -05003017void ceph_msg_revoke(struct ceph_msg *msg)
Sage Weil31b80062009-10-06 11:31:13 -07003018{
Alex Elder6740a842012-06-01 14:56:43 -05003019 struct ceph_connection *con = msg->con;
3020
3021 if (!con)
3022 return; /* Message not in our possession */
3023
Sage Weilec302642009-12-22 10:43:42 -08003024 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07003025 if (!list_empty(&msg->list_head)) {
Alex Elder38941f82012-06-01 14:56:43 -05003026 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07003027 list_del_init(&msg->list_head);
Alex Elder38941f82012-06-01 14:56:43 -05003028 BUG_ON(msg->con == NULL);
Sage Weil36eb71a2012-06-21 12:47:08 -07003029 msg->con->ops->put(msg->con);
Alex Elder38941f82012-06-01 14:56:43 -05003030 msg->con = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07003031 msg->hdr.seq = 0;
Alex Elder38941f82012-06-01 14:56:43 -05003032
Sage Weil31b80062009-10-06 11:31:13 -07003033 ceph_msg_put(msg);
Sage Weiled98ada2010-07-05 12:15:14 -07003034 }
3035 if (con->out_msg == msg) {
Alex Elder38941f82012-06-01 14:56:43 -05003036 dout("%s %p msg %p - was sending\n", __func__, con, msg);
Sage Weiled98ada2010-07-05 12:15:14 -07003037 con->out_msg = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07003038 if (con->out_kvec_is_msg) {
3039 con->out_skip = con->out_kvec_bytes;
3040 con->out_kvec_is_msg = false;
3041 }
Sage Weiled98ada2010-07-05 12:15:14 -07003042 msg->hdr.seq = 0;
Alex Elder92ce0342012-06-04 14:43:33 -05003043
3044 ceph_msg_put(msg);
Sage Weil31b80062009-10-06 11:31:13 -07003045 }
Sage Weilec302642009-12-22 10:43:42 -08003046 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07003047}
3048
3049/*
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08003050 * Revoke a message that we may be reading data into
Sage Weil350b1c32009-12-22 10:45:45 -08003051 */
Alex Elder8921d112012-06-01 14:56:43 -05003052void ceph_msg_revoke_incoming(struct ceph_msg *msg)
Sage Weil350b1c32009-12-22 10:45:45 -08003053{
Alex Elder8921d112012-06-01 14:56:43 -05003054 struct ceph_connection *con;
3055
3056 BUG_ON(msg == NULL);
3057 if (!msg->con) {
3058 dout("%s msg %p null con\n", __func__, msg);
3059
3060 return; /* Message not in our possession */
3061 }
3062
3063 con = msg->con;
Sage Weil350b1c32009-12-22 10:45:45 -08003064 mutex_lock(&con->mutex);
Alex Elder8921d112012-06-01 14:56:43 -05003065 if (con->in_msg == msg) {
Eric Dumazet95c96172012-04-15 05:58:06 +00003066 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3067 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3068 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
Sage Weil350b1c32009-12-22 10:45:45 -08003069
3070 /* skip rest of message */
Alex Elder8921d112012-06-01 14:56:43 -05003071 dout("%s %p msg %p revoked\n", __func__, con, msg);
3072 con->in_base_pos = con->in_base_pos -
Sage Weil350b1c32009-12-22 10:45:45 -08003073 sizeof(struct ceph_msg_header) -
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08003074 front_len -
3075 middle_len -
3076 data_len -
Sage Weil350b1c32009-12-22 10:45:45 -08003077 sizeof(struct ceph_msg_footer);
Sage Weil350b1c32009-12-22 10:45:45 -08003078 ceph_msg_put(con->in_msg);
3079 con->in_msg = NULL;
3080 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weil684be252010-04-21 20:45:59 -07003081 con->in_seq++;
Sage Weil350b1c32009-12-22 10:45:45 -08003082 } else {
Alex Elder8921d112012-06-01 14:56:43 -05003083 dout("%s %p in_msg %p msg %p no-op\n",
3084 __func__, con, con->in_msg, msg);
Sage Weil350b1c32009-12-22 10:45:45 -08003085 }
3086 mutex_unlock(&con->mutex);
3087}
3088
3089/*
Sage Weil31b80062009-10-06 11:31:13 -07003090 * Queue a keepalive byte to ensure the tcp connection is alive.
3091 */
3092void ceph_con_keepalive(struct ceph_connection *con)
3093{
Sage Weile00de342011-03-04 12:25:05 -08003094 dout("con_keepalive %p\n", con);
Sage Weil00650932012-07-20 15:33:04 -07003095 mutex_lock(&con->mutex);
Sage Weile00de342011-03-04 12:25:05 -08003096 clear_standby(con);
Sage Weil00650932012-07-20 15:33:04 -07003097 mutex_unlock(&con->mutex);
Alex Elderc9ffc772013-02-20 10:25:12 -06003098 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3099 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
Sage Weil31b80062009-10-06 11:31:13 -07003100 queue_con(con);
3101}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003102EXPORT_SYMBOL(ceph_con_keepalive);
Sage Weil31b80062009-10-06 11:31:13 -07003103
Alex Elder6644ed72013-03-11 23:34:24 -05003104static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
Alex Elder43794502013-03-01 18:00:16 -06003105{
Alex Elder6644ed72013-03-11 23:34:24 -05003106 struct ceph_msg_data *data;
3107
3108 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3109 return NULL;
3110
Alex Elder81b36be2013-05-01 12:43:04 -05003111 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
Alex Elder6644ed72013-03-11 23:34:24 -05003112 if (data)
3113 data->type = type;
Alex Elder5240d9f2013-03-14 14:09:06 -05003114 INIT_LIST_HEAD(&data->links);
Alex Elder6644ed72013-03-11 23:34:24 -05003115
3116 return data;
3117}
3118
3119static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3120{
3121 if (!data)
3122 return;
3123
Alex Elder5240d9f2013-03-14 14:09:06 -05003124 WARN_ON(!list_empty(&data->links));
Yan, Zhenge4339d282014-09-16 17:50:45 +08003125 if (data->type == CEPH_MSG_DATA_PAGELIST)
Alex Elder6644ed72013-03-11 23:34:24 -05003126 ceph_pagelist_release(data->pagelist);
Alex Elder81b36be2013-05-01 12:43:04 -05003127 kmem_cache_free(ceph_msg_data_cache, data);
Alex Elder43794502013-03-01 18:00:16 -06003128}
3129
Alex Elder90af3602013-04-05 14:46:01 -05003130void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
Alex Elderf1baeb22013-03-07 15:38:26 -06003131 size_t length, size_t alignment)
Alex Elder02afca62013-02-14 12:16:43 -06003132{
Alex Elder6644ed72013-03-11 23:34:24 -05003133 struct ceph_msg_data *data;
3134
Alex Elder07aa1552013-03-04 18:29:06 -06003135 BUG_ON(!pages);
3136 BUG_ON(!length);
Alex Elder02afca62013-02-14 12:16:43 -06003137
Alex Elder6644ed72013-03-11 23:34:24 -05003138 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3139 BUG_ON(!data);
3140 data->pages = pages;
3141 data->length = length;
3142 data->alignment = alignment & ~PAGE_MASK;
3143
Alex Elder5240d9f2013-03-14 14:09:06 -05003144 list_add_tail(&data->links, &msg->data);
3145 msg->data_length += length;
Alex Elder02afca62013-02-14 12:16:43 -06003146}
Alex Elder90af3602013-04-05 14:46:01 -05003147EXPORT_SYMBOL(ceph_msg_data_add_pages);
Sage Weil31b80062009-10-06 11:31:13 -07003148
Alex Elder90af3602013-04-05 14:46:01 -05003149void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
Alex Elder27fa8382013-02-14 12:16:43 -06003150 struct ceph_pagelist *pagelist)
3151{
Alex Elder6644ed72013-03-11 23:34:24 -05003152 struct ceph_msg_data *data;
3153
Alex Elder07aa1552013-03-04 18:29:06 -06003154 BUG_ON(!pagelist);
3155 BUG_ON(!pagelist->length);
Alex Elder27fa8382013-02-14 12:16:43 -06003156
Alex Elder6644ed72013-03-11 23:34:24 -05003157 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3158 BUG_ON(!data);
3159 data->pagelist = pagelist;
3160
Alex Elder5240d9f2013-03-14 14:09:06 -05003161 list_add_tail(&data->links, &msg->data);
3162 msg->data_length += pagelist->length;
Alex Elder27fa8382013-02-14 12:16:43 -06003163}
Alex Elder90af3602013-04-05 14:46:01 -05003164EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
Alex Elder27fa8382013-02-14 12:16:43 -06003165
Alex Elderea965712013-04-05 14:46:01 -05003166#ifdef CONFIG_BLOCK
Alex Elder90af3602013-04-05 14:46:01 -05003167void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
Alex Eldera1930802013-03-14 14:09:06 -05003168 size_t length)
Alex Elder27fa8382013-02-14 12:16:43 -06003169{
Alex Elder6644ed72013-03-11 23:34:24 -05003170 struct ceph_msg_data *data;
Alex Elder27fa8382013-02-14 12:16:43 -06003171
Alex Elder6644ed72013-03-11 23:34:24 -05003172 BUG_ON(!bio);
Alex Elder6644ed72013-03-11 23:34:24 -05003173
3174 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3175 BUG_ON(!data);
3176 data->bio = bio;
Alex Elderc851c492013-04-05 14:46:01 -05003177 data->bio_length = length;
Alex Elder6644ed72013-03-11 23:34:24 -05003178
Alex Elder5240d9f2013-03-14 14:09:06 -05003179 list_add_tail(&data->links, &msg->data);
3180 msg->data_length += length;
Alex Elder27fa8382013-02-14 12:16:43 -06003181}
Alex Elder90af3602013-04-05 14:46:01 -05003182EXPORT_SYMBOL(ceph_msg_data_add_bio);
Alex Elderea965712013-04-05 14:46:01 -05003183#endif /* CONFIG_BLOCK */
Alex Elder27fa8382013-02-14 12:16:43 -06003184
Sage Weil31b80062009-10-06 11:31:13 -07003185/*
3186 * construct a new message with given type, size
3187 * the new msg has a ref count of 1.
3188 */
Sage Weilb61c2762011-08-09 15:03:46 -07003189struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3190 bool can_fail)
Sage Weil31b80062009-10-06 11:31:13 -07003191{
3192 struct ceph_msg *m;
3193
Alex Eldere3d5d632013-05-01 12:43:04 -05003194 m = kmem_cache_zalloc(ceph_msg_cache, flags);
Sage Weil31b80062009-10-06 11:31:13 -07003195 if (m == NULL)
3196 goto out;
Alex Elder38941f82012-06-01 14:56:43 -05003197
Sage Weil31b80062009-10-06 11:31:13 -07003198 m->hdr.type = cpu_to_le16(type);
Sage Weil45c6ceb2010-05-11 15:01:51 -07003199 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
Sage Weil31b80062009-10-06 11:31:13 -07003200 m->hdr.front_len = cpu_to_le32(front_len);
Sage Weil31b80062009-10-06 11:31:13 -07003201
Alex Elder9516e452013-03-01 18:00:16 -06003202 INIT_LIST_HEAD(&m->list_head);
3203 kref_init(&m->kref);
Alex Elder5240d9f2013-03-14 14:09:06 -05003204 INIT_LIST_HEAD(&m->data);
Henry C Changca208922011-05-03 02:29:56 +00003205
Sage Weil31b80062009-10-06 11:31:13 -07003206 /* front */
3207 if (front_len) {
Ilya Dryomoveeb0bed2014-01-09 20:08:21 +02003208 m->front.iov_base = ceph_kvmalloc(front_len, flags);
Sage Weil31b80062009-10-06 11:31:13 -07003209 if (m->front.iov_base == NULL) {
Sage Weilb61c2762011-08-09 15:03:46 -07003210 dout("ceph_msg_new can't allocate %d bytes\n",
Sage Weil31b80062009-10-06 11:31:13 -07003211 front_len);
3212 goto out2;
3213 }
3214 } else {
3215 m->front.iov_base = NULL;
3216 }
Ilya Dryomovf2be82b2014-01-09 20:08:21 +02003217 m->front_alloc_len = m->front.iov_len = front_len;
Sage Weil31b80062009-10-06 11:31:13 -07003218
Sage Weilbb257662010-04-01 16:07:23 -07003219 dout("ceph_msg_new %p front %d\n", m, front_len);
Sage Weil31b80062009-10-06 11:31:13 -07003220 return m;
3221
3222out2:
3223 ceph_msg_put(m);
3224out:
Sage Weilb61c2762011-08-09 15:03:46 -07003225 if (!can_fail) {
3226 pr_err("msg_new can't create type %d front %d\n", type,
3227 front_len);
Sage Weilf0ed1b72011-08-09 15:05:07 -07003228 WARN_ON(1);
Sage Weilb61c2762011-08-09 15:03:46 -07003229 } else {
3230 dout("msg_new can't create type %d front %d\n", type,
3231 front_len);
3232 }
Sage Weila79832f2010-04-01 16:06:19 -07003233 return NULL;
Sage Weil31b80062009-10-06 11:31:13 -07003234}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003235EXPORT_SYMBOL(ceph_msg_new);
Sage Weil31b80062009-10-06 11:31:13 -07003236
3237/*
Sage Weil31b80062009-10-06 11:31:13 -07003238 * Allocate "middle" portion of a message, if it is needed and wasn't
3239 * allocated by alloc_msg. This allows us to read a small fixed-size
3240 * per-type header in the front and then gracefully fail (i.e.,
3241 * propagate the error to the caller based on info in the front) when
3242 * the middle is too large.
3243 */
Yehuda Sadeh24504182010-01-08 13:58:34 -08003244static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
Sage Weil31b80062009-10-06 11:31:13 -07003245{
3246 int type = le16_to_cpu(msg->hdr.type);
3247 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3248
3249 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3250 ceph_msg_type_name(type), middle_len);
3251 BUG_ON(!middle_len);
3252 BUG_ON(msg->middle);
3253
Sage Weilb6c1d5b2009-12-07 12:17:17 -08003254 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
Sage Weil31b80062009-10-06 11:31:13 -07003255 if (!msg->middle)
3256 return -ENOMEM;
3257 return 0;
3258}
3259
Yehuda Sadeh24504182010-01-08 13:58:34 -08003260/*
Alex Elder1c20f2d2012-06-04 14:43:32 -05003261 * Allocate a message for receiving an incoming message on a
3262 * connection, and save the result in con->in_msg. Uses the
3263 * connection's private alloc_msg op if available.
3264 *
Sage Weil4740a622012-07-30 18:19:30 -07003265 * Returns 0 on success, or a negative error code.
3266 *
3267 * On success, if we set *skip = 1:
3268 * - the next message should be skipped and ignored.
3269 * - con->in_msg == NULL
3270 * or if we set *skip = 0:
3271 * - con->in_msg is non-null.
3272 * On error (ENOMEM, EAGAIN, ...),
3273 * - con->in_msg == NULL
Yehuda Sadeh24504182010-01-08 13:58:34 -08003274 */
Sage Weil4740a622012-07-30 18:19:30 -07003275static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
Yehuda Sadeh24504182010-01-08 13:58:34 -08003276{
Sage Weil4740a622012-07-30 18:19:30 -07003277 struct ceph_msg_header *hdr = &con->in_hdr;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003278 int middle_len = le32_to_cpu(hdr->middle_len);
Alex Elder1d866d12013-03-01 18:00:14 -06003279 struct ceph_msg *msg;
Sage Weil4740a622012-07-30 18:19:30 -07003280 int ret = 0;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003281
Alex Elder1c20f2d2012-06-04 14:43:32 -05003282 BUG_ON(con->in_msg != NULL);
Alex Elder53ded492013-03-01 18:00:14 -06003283 BUG_ON(!con->ops->alloc_msg);
Yehuda Sadeh24504182010-01-08 13:58:34 -08003284
Alex Elder53ded492013-03-01 18:00:14 -06003285 mutex_unlock(&con->mutex);
3286 msg = con->ops->alloc_msg(con, hdr, skip);
3287 mutex_lock(&con->mutex);
3288 if (con->state != CON_STATE_OPEN) {
3289 if (msg)
Alex Elder1d866d12013-03-01 18:00:14 -06003290 ceph_msg_put(msg);
Alex Elder53ded492013-03-01 18:00:14 -06003291 return -EAGAIN;
3292 }
Alex Elder41375772013-03-05 09:25:10 -06003293 if (msg) {
3294 BUG_ON(*skip);
3295 con->in_msg = msg;
Sage Weil36eb71a2012-06-21 12:47:08 -07003296 con->in_msg->con = con->ops->get(con);
Alex Elder92ce0342012-06-04 14:43:33 -05003297 BUG_ON(con->in_msg->con == NULL);
Alex Elder41375772013-03-05 09:25:10 -06003298 } else {
3299 /*
3300 * Null message pointer means either we should skip
3301 * this message or we couldn't allocate memory. The
3302 * former is not an error.
3303 */
3304 if (*skip)
3305 return 0;
Alex Elder41375772013-03-05 09:25:10 -06003306
Ilya Dryomov67c64eb2015-03-23 14:52:40 +03003307 con->error_msg = "error allocating memory for incoming message";
Alex Elder53ded492013-03-01 18:00:14 -06003308 return -ENOMEM;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003309 }
Alex Elder1c20f2d2012-06-04 14:43:32 -05003310 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
Yehuda Sadeh24504182010-01-08 13:58:34 -08003311
Alex Elder1c20f2d2012-06-04 14:43:32 -05003312 if (middle_len && !con->in_msg->middle) {
3313 ret = ceph_alloc_middle(con, con->in_msg);
Yehuda Sadeh24504182010-01-08 13:58:34 -08003314 if (ret < 0) {
Alex Elder1c20f2d2012-06-04 14:43:32 -05003315 ceph_msg_put(con->in_msg);
3316 con->in_msg = NULL;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003317 }
3318 }
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08003319
Sage Weil4740a622012-07-30 18:19:30 -07003320 return ret;
Yehuda Sadeh24504182010-01-08 13:58:34 -08003321}
3322
Sage Weil31b80062009-10-06 11:31:13 -07003323
3324/*
3325 * Free a generically kmalloc'd message.
3326 */
Ilya Dryomov0215e442014-06-20 14:14:41 +04003327static void ceph_msg_free(struct ceph_msg *m)
Sage Weil31b80062009-10-06 11:31:13 -07003328{
Ilya Dryomov0215e442014-06-20 14:14:41 +04003329 dout("%s %p\n", __func__, m);
Ilya Dryomov4965fc32014-10-23 16:32:57 +04003330 kvfree(m->front.iov_base);
Alex Eldere3d5d632013-05-01 12:43:04 -05003331 kmem_cache_free(ceph_msg_cache, m);
Sage Weil31b80062009-10-06 11:31:13 -07003332}
3333
Ilya Dryomov0215e442014-06-20 14:14:41 +04003334static void ceph_msg_release(struct kref *kref)
Sage Weil31b80062009-10-06 11:31:13 -07003335{
Sage Weilc2e552e2009-12-07 15:55:05 -08003336 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
Alex Elder5240d9f2013-03-14 14:09:06 -05003337 LIST_HEAD(data);
3338 struct list_head *links;
3339 struct list_head *next;
Sage Weil31b80062009-10-06 11:31:13 -07003340
Ilya Dryomov0215e442014-06-20 14:14:41 +04003341 dout("%s %p\n", __func__, m);
Sage Weilc2e552e2009-12-07 15:55:05 -08003342 WARN_ON(!list_empty(&m->list_head));
Sage Weil31b80062009-10-06 11:31:13 -07003343
Sage Weilc2e552e2009-12-07 15:55:05 -08003344 /* drop middle, data, if any */
3345 if (m->middle) {
3346 ceph_buffer_put(m->middle);
3347 m->middle = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07003348 }
Alex Elder5240d9f2013-03-14 14:09:06 -05003349
3350 list_splice_init(&m->data, &data);
3351 list_for_each_safe(links, next, &data) {
3352 struct ceph_msg_data *data;
3353
3354 data = list_entry(links, struct ceph_msg_data, links);
3355 list_del_init(links);
3356 ceph_msg_data_destroy(data);
3357 }
Alex Eldera1930802013-03-14 14:09:06 -05003358 m->data_length = 0;
Sage Weil58bb3b32009-12-23 12:12:31 -08003359
Sage Weilc2e552e2009-12-07 15:55:05 -08003360 if (m->pool)
3361 ceph_msgpool_put(m->pool, m);
3362 else
Ilya Dryomov0215e442014-06-20 14:14:41 +04003363 ceph_msg_free(m);
Sage Weil31b80062009-10-06 11:31:13 -07003364}
Ilya Dryomov0215e442014-06-20 14:14:41 +04003365
3366struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3367{
3368 dout("%s %p (was %d)\n", __func__, msg,
3369 atomic_read(&msg->kref.refcount));
3370 kref_get(&msg->kref);
3371 return msg;
3372}
3373EXPORT_SYMBOL(ceph_msg_get);
3374
3375void ceph_msg_put(struct ceph_msg *msg)
3376{
3377 dout("%s %p (was %d)\n", __func__, msg,
3378 atomic_read(&msg->kref.refcount));
3379 kref_put(&msg->kref, ceph_msg_release);
3380}
3381EXPORT_SYMBOL(ceph_msg_put);
Sage Weil9ec7cab2009-12-14 15:13:47 -08003382
3383void ceph_msg_dump(struct ceph_msg *msg)
3384{
Ilya Dryomov3cea4c32014-01-09 20:08:21 +02003385 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3386 msg->front_alloc_len, msg->data_length);
Sage Weil9ec7cab2009-12-14 15:13:47 -08003387 print_hex_dump(KERN_DEBUG, "header: ",
3388 DUMP_PREFIX_OFFSET, 16, 1,
3389 &msg->hdr, sizeof(msg->hdr), true);
3390 print_hex_dump(KERN_DEBUG, " front: ",
3391 DUMP_PREFIX_OFFSET, 16, 1,
3392 msg->front.iov_base, msg->front.iov_len, true);
3393 if (msg->middle)
3394 print_hex_dump(KERN_DEBUG, "middle: ",
3395 DUMP_PREFIX_OFFSET, 16, 1,
3396 msg->middle->vec.iov_base,
3397 msg->middle->vec.iov_len, true);
3398 print_hex_dump(KERN_DEBUG, "footer: ",
3399 DUMP_PREFIX_OFFSET, 16, 1,
3400 &msg->footer, sizeof(msg->footer), true);
3401}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003402EXPORT_SYMBOL(ceph_msg_dump);