Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
31b8006e SW |
2 | |
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> | |
5a0e3ad6 | 9 | #include <linux/slab.h> |
31b8006e SW |
10 | #include <linux/socket.h> |
11 | #include <linux/string.h> | |
3ebc21f7 | 12 | #ifdef CONFIG_BLOCK |
68b4476b | 13 | #include <linux/bio.h> |
3ebc21f7 | 14 | #endif /* CONFIG_BLOCK */ |
ee3b56f2 | 15 | #include <linux/dns_resolver.h> |
31b8006e SW |
16 | #include <net/tcp.h> |
17 | ||
3d14c5d2 YS |
18 | #include <linux/ceph/libceph.h> |
19 | #include <linux/ceph/messenger.h> | |
20 | #include <linux/ceph/decode.h> | |
21 | #include <linux/ceph/pagelist.h> | |
bc3b2d7f | 22 | #include <linux/export.h> |
31b8006e SW |
23 | |
24 | /* | |
25 | * Ceph uses the messenger to exchange ceph_msg messages with other | |
26 | * hosts in the system. The messenger provides ordered and reliable | |
27 | * delivery. We tolerate TCP disconnects by reconnecting (with | |
28 | * exponential backoff) in the case of a fault (disconnection, bad | |
29 | * crc, protocol error). Acks allow sent messages to be discarded by | |
30 | * the sender. | |
31 | */ | |
32 | ||
bc18f4b1 AE |
33 | /* |
34 | * We track the state of the socket on a given connection using | |
35 | * values defined below. The transition to a new socket state is | |
36 | * handled by a function which verifies we aren't coming from an | |
37 | * unexpected state. | |
38 | * | |
39 | * -------- | |
40 | * | NEW* | transient initial state | |
41 | * -------- | |
42 | * | con_sock_state_init() | |
43 | * v | |
44 | * ---------- | |
45 | * | CLOSED | initialized, but no socket (and no | |
46 | * ---------- TCP connection) | |
47 | * ^ \ | |
48 | * | \ con_sock_state_connecting() | |
49 | * | ---------------------- | |
50 | * | \ | |
51 | * + con_sock_state_closed() \ | |
fbb85a47 SW |
52 | * |+--------------------------- \ |
53 | * | \ \ \ | |
54 | * | ----------- \ \ | |
55 | * | | CLOSING | socket event; \ \ | |
56 | * | ----------- await close \ \ | |
57 | * | ^ \ | | |
58 | * | | \ | | |
59 | * | + con_sock_state_closing() \ | | |
60 | * | / \ | | | |
61 | * | / --------------- | | | |
62 | * | / \ v v | |
bc18f4b1 AE |
63 | * | / -------------- |
64 | * | / -----------------| CONNECTING | socket created, TCP | |
65 | * | | / -------------- connect initiated | |
66 | * | | | con_sock_state_connected() | |
67 | * | | v | |
68 | * ------------- | |
69 | * | CONNECTED | TCP connection established | |
70 | * ------------- | |
71 | * | |
72 | * State values for ceph_connection->sock_state; NEW is assumed to be 0. | |
73 | */ | |
ce2c8903 AE |
74 | |
75 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ | |
76 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ | |
77 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ | |
78 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ | |
79 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ | |
80 | ||
8dacc7da SW |
81 | /* |
82 | * connection states | |
83 | */ | |
84 | #define CON_STATE_CLOSED 1 /* -> PREOPEN */ | |
85 | #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */ | |
86 | #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */ | |
87 | #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */ | |
88 | #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */ | |
89 | #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */ | |
90 | ||
4a861692 SW |
91 | /* |
92 | * ceph_connection flag bits | |
93 | */ | |
94 | #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop | |
95 | * messages on errors */ | |
96 | #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */ | |
97 | #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */ | |
98 | #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */ | |
99 | #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */ | |
8dacc7da | 100 | |
c9ffc77a AE |
101 | static bool con_flag_valid(unsigned long con_flag) |
102 | { | |
103 | switch (con_flag) { | |
104 | case CON_FLAG_LOSSYTX: | |
105 | case CON_FLAG_KEEPALIVE_PENDING: | |
106 | case CON_FLAG_WRITE_PENDING: | |
107 | case CON_FLAG_SOCK_CLOSED: | |
108 | case CON_FLAG_BACKOFF: | |
109 | return true; | |
110 | default: | |
111 | return false; | |
112 | } | |
113 | } | |
114 | ||
115 | static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag) | |
116 | { | |
117 | BUG_ON(!con_flag_valid(con_flag)); | |
118 | ||
119 | clear_bit(con_flag, &con->flags); | |
120 | } | |
121 | ||
122 | static void con_flag_set(struct ceph_connection *con, unsigned long con_flag) | |
123 | { | |
124 | BUG_ON(!con_flag_valid(con_flag)); | |
125 | ||
126 | set_bit(con_flag, &con->flags); | |
127 | } | |
128 | ||
129 | static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag) | |
130 | { | |
131 | BUG_ON(!con_flag_valid(con_flag)); | |
132 | ||
133 | return test_bit(con_flag, &con->flags); | |
134 | } | |
135 | ||
136 | static bool con_flag_test_and_clear(struct ceph_connection *con, | |
137 | unsigned long con_flag) | |
138 | { | |
139 | BUG_ON(!con_flag_valid(con_flag)); | |
140 | ||
141 | return test_and_clear_bit(con_flag, &con->flags); | |
142 | } | |
143 | ||
144 | static bool con_flag_test_and_set(struct ceph_connection *con, | |
145 | unsigned long con_flag) | |
146 | { | |
147 | BUG_ON(!con_flag_valid(con_flag)); | |
148 | ||
149 | return test_and_set_bit(con_flag, &con->flags); | |
150 | } | |
151 | ||
31b8006e SW |
152 | /* static tag bytes (protocol control messages) */ |
153 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
154 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
155 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
156 | ||
a6a5349d SW |
157 | #ifdef CONFIG_LOCKDEP |
158 | static struct lock_class_key socket_class; | |
159 | #endif | |
160 | ||
84495f49 AE |
161 | /* |
162 | * When skipping (ignoring) a block of input we read it into a "skip | |
163 | * buffer," which is this many bytes in size. | |
164 | */ | |
165 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
166 | |
167 | static void queue_con(struct ceph_connection *con); | |
168 | static void con_work(struct work_struct *); | |
93209264 | 169 | static void con_fault(struct ceph_connection *con); |
31b8006e | 170 | |
31b8006e | 171 | /* |
f64a9317 AE |
172 | * Nicely render a sockaddr as a string. An array of formatted |
173 | * strings is used, to approximate reentrancy. | |
31b8006e | 174 | */ |
f64a9317 AE |
175 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
176 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
177 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
178 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
179 | ||
180 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
181 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 182 | |
57666519 | 183 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 184 | |
3d14c5d2 | 185 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
186 | { |
187 | int i; | |
188 | char *s; | |
99f0f3b2 AE |
189 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
190 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 191 | |
f64a9317 | 192 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
193 | s = addr_str[i]; |
194 | ||
195 | switch (ss->ss_family) { | |
196 | case AF_INET: | |
bd406145 AE |
197 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
198 | ntohs(in4->sin_port)); | |
31b8006e SW |
199 | break; |
200 | ||
201 | case AF_INET6: | |
bd406145 AE |
202 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
203 | ntohs(in6->sin6_port)); | |
31b8006e SW |
204 | break; |
205 | ||
206 | default: | |
d3002b97 AE |
207 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
208 | ss->ss_family); | |
31b8006e SW |
209 | } |
210 | ||
211 | return s; | |
212 | } | |
3d14c5d2 | 213 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 214 | |
63f2d211 SW |
215 | static void encode_my_addr(struct ceph_messenger *msgr) |
216 | { | |
217 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
218 | ceph_encode_addr(&msgr->my_enc_addr); | |
219 | } | |
220 | ||
31b8006e SW |
221 | /* |
222 | * work queue for all reading and writing to/from the socket. | |
223 | */ | |
e0f43c94 | 224 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 225 | |
15417167 | 226 | static void _ceph_msgr_exit(void) |
6173d1f0 | 227 | { |
d3002b97 | 228 | if (ceph_msgr_wq) { |
6173d1f0 | 229 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
230 | ceph_msgr_wq = NULL; |
231 | } | |
6173d1f0 | 232 | |
6173d1f0 AE |
233 | BUG_ON(zero_page == NULL); |
234 | kunmap(zero_page); | |
235 | page_cache_release(zero_page); | |
236 | zero_page = NULL; | |
237 | } | |
238 | ||
3d14c5d2 | 239 | int ceph_msgr_init(void) |
31b8006e | 240 | { |
57666519 AE |
241 | BUG_ON(zero_page != NULL); |
242 | zero_page = ZERO_PAGE(0); | |
243 | page_cache_get(zero_page); | |
244 | ||
f363e45f | 245 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0); |
6173d1f0 AE |
246 | if (ceph_msgr_wq) |
247 | return 0; | |
57666519 | 248 | |
6173d1f0 AE |
249 | pr_err("msgr_init failed to create workqueue\n"); |
250 | _ceph_msgr_exit(); | |
57666519 | 251 | |
6173d1f0 | 252 | return -ENOMEM; |
31b8006e | 253 | } |
3d14c5d2 | 254 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
255 | |
256 | void ceph_msgr_exit(void) | |
257 | { | |
57666519 | 258 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 259 | |
6173d1f0 | 260 | _ceph_msgr_exit(); |
31b8006e | 261 | } |
3d14c5d2 | 262 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 263 | |
cd84db6e | 264 | void ceph_msgr_flush(void) |
a922d38f SW |
265 | { |
266 | flush_workqueue(ceph_msgr_wq); | |
267 | } | |
3d14c5d2 | 268 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 269 | |
ce2c8903 AE |
270 | /* Connection socket state transition functions */ |
271 | ||
272 | static void con_sock_state_init(struct ceph_connection *con) | |
273 | { | |
274 | int old_state; | |
275 | ||
276 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
277 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
278 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
279 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
280 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 AE |
281 | } |
282 | ||
283 | static void con_sock_state_connecting(struct ceph_connection *con) | |
284 | { | |
285 | int old_state; | |
286 | ||
287 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
288 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
289 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
290 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
291 | CON_SOCK_STATE_CONNECTING); | |
ce2c8903 AE |
292 | } |
293 | ||
294 | static void con_sock_state_connected(struct ceph_connection *con) | |
295 | { | |
296 | int old_state; | |
297 | ||
298 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
299 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
300 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
301 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
302 | CON_SOCK_STATE_CONNECTED); | |
ce2c8903 AE |
303 | } |
304 | ||
305 | static void con_sock_state_closing(struct ceph_connection *con) | |
306 | { | |
307 | int old_state; | |
308 | ||
309 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
310 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
311 | old_state != CON_SOCK_STATE_CONNECTED && | |
312 | old_state != CON_SOCK_STATE_CLOSING)) | |
313 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
314 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
315 | CON_SOCK_STATE_CLOSING); | |
ce2c8903 AE |
316 | } |
317 | ||
318 | static void con_sock_state_closed(struct ceph_connection *con) | |
319 | { | |
320 | int old_state; | |
321 | ||
322 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
323 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 | 324 | old_state != CON_SOCK_STATE_CLOSING && |
8007b8d6 SW |
325 | old_state != CON_SOCK_STATE_CONNECTING && |
326 | old_state != CON_SOCK_STATE_CLOSED)) | |
ce2c8903 | 327 | printk("%s: unexpected old state %d\n", __func__, old_state); |
8007b8d6 SW |
328 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
329 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 | 330 | } |
a922d38f | 331 | |
31b8006e SW |
332 | /* |
333 | * socket callback functions | |
334 | */ | |
335 | ||
336 | /* data available on socket, or listen socket received a connect */ | |
327800bd | 337 | static void ceph_sock_data_ready(struct sock *sk, int count_unused) |
31b8006e | 338 | { |
bd406145 | 339 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
340 | if (atomic_read(&con->msgr->stopping)) { |
341 | return; | |
342 | } | |
bd406145 | 343 | |
31b8006e | 344 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 345 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
346 | con, con->state); |
347 | queue_con(con); | |
348 | } | |
349 | } | |
350 | ||
351 | /* socket has buffer space for writing */ | |
327800bd | 352 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 353 | { |
d3002b97 | 354 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 355 | |
182fac26 JS |
356 | /* only queue to workqueue if there is data we want to write, |
357 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 358 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
359 | * doesn't get called again until try_write() fills the socket |
360 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
361 | * and net/core/stream.c:sk_stream_write_space(). | |
362 | */ | |
c9ffc77a | 363 | if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) { |
182fac26 | 364 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
327800bd | 365 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
366 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
367 | queue_con(con); | |
368 | } | |
31b8006e | 369 | } else { |
327800bd | 370 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 371 | } |
31b8006e SW |
372 | } |
373 | ||
374 | /* socket's state has changed */ | |
327800bd | 375 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 376 | { |
bd406145 | 377 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 378 | |
327800bd | 379 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
380 | con, con->state, sk->sk_state); |
381 | ||
31b8006e SW |
382 | switch (sk->sk_state) { |
383 | case TCP_CLOSE: | |
327800bd | 384 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 385 | case TCP_CLOSE_WAIT: |
327800bd | 386 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 387 | con_sock_state_closing(con); |
c9ffc77a | 388 | con_flag_set(con, CON_FLAG_SOCK_CLOSED); |
d65c9e0b | 389 | queue_con(con); |
31b8006e SW |
390 | break; |
391 | case TCP_ESTABLISHED: | |
327800bd | 392 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 393 | con_sock_state_connected(con); |
31b8006e SW |
394 | queue_con(con); |
395 | break; | |
d3002b97 AE |
396 | default: /* Everything else is uninteresting */ |
397 | break; | |
31b8006e SW |
398 | } |
399 | } | |
400 | ||
401 | /* | |
402 | * set up socket callbacks | |
403 | */ | |
404 | static void set_sock_callbacks(struct socket *sock, | |
405 | struct ceph_connection *con) | |
406 | { | |
407 | struct sock *sk = sock->sk; | |
bd406145 | 408 | sk->sk_user_data = con; |
327800bd AE |
409 | sk->sk_data_ready = ceph_sock_data_ready; |
410 | sk->sk_write_space = ceph_sock_write_space; | |
411 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
412 | } |
413 | ||
414 | ||
415 | /* | |
416 | * socket helpers | |
417 | */ | |
418 | ||
419 | /* | |
420 | * initiate connection to a remote socket. | |
421 | */ | |
41617d0c | 422 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 423 | { |
f91d3471 | 424 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
425 | struct socket *sock; |
426 | int ret; | |
427 | ||
428 | BUG_ON(con->sock); | |
f91d3471 SW |
429 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
430 | IPPROTO_TCP, &sock); | |
31b8006e | 431 | if (ret) |
41617d0c | 432 | return ret; |
31b8006e SW |
433 | sock->sk->sk_allocation = GFP_NOFS; |
434 | ||
a6a5349d SW |
435 | #ifdef CONFIG_LOCKDEP |
436 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
437 | #endif | |
438 | ||
31b8006e SW |
439 | set_sock_callbacks(sock, con); |
440 | ||
3d14c5d2 | 441 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 442 | |
89a86be0 | 443 | con_sock_state_connecting(con); |
f91d3471 SW |
444 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
445 | O_NONBLOCK); | |
31b8006e SW |
446 | if (ret == -EINPROGRESS) { |
447 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 448 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 449 | sock->sk->sk_state); |
a5bc3129 | 450 | } else if (ret < 0) { |
31b8006e | 451 | pr_err("connect %s error %d\n", |
3d14c5d2 | 452 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 453 | sock_release(sock); |
31b8006e | 454 | con->error_msg = "connect error"; |
31b8006e | 455 | |
41617d0c | 456 | return ret; |
a5bc3129 AE |
457 | } |
458 | con->sock = sock; | |
41617d0c | 459 | return 0; |
31b8006e SW |
460 | } |
461 | ||
462 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
463 | { | |
464 | struct kvec iov = {buf, len}; | |
465 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 466 | int r; |
31b8006e | 467 | |
98bdb0aa SW |
468 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
469 | if (r == -EAGAIN) | |
470 | r = 0; | |
471 | return r; | |
31b8006e SW |
472 | } |
473 | ||
474 | /* | |
475 | * write something. @more is true if caller will be sending more data | |
476 | * shortly. | |
477 | */ | |
478 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
479 | size_t kvlen, size_t len, int more) | |
480 | { | |
481 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 482 | int r; |
31b8006e SW |
483 | |
484 | if (more) | |
485 | msg.msg_flags |= MSG_MORE; | |
486 | else | |
487 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
488 | ||
42961d23 SW |
489 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
490 | if (r == -EAGAIN) | |
491 | r = 0; | |
492 | return r; | |
31b8006e SW |
493 | } |
494 | ||
31739139 | 495 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
e1dcb128 | 496 | int offset, size_t size, bool more) |
31739139 AE |
497 | { |
498 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
499 | int ret; | |
500 | ||
501 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
502 | if (ret == -EAGAIN) | |
503 | ret = 0; | |
504 | ||
505 | return ret; | |
506 | } | |
507 | ||
31b8006e SW |
508 | |
509 | /* | |
510 | * Shutdown/close the socket for the given connection. | |
511 | */ | |
512 | static int con_close_socket(struct ceph_connection *con) | |
513 | { | |
8007b8d6 | 514 | int rc = 0; |
31b8006e SW |
515 | |
516 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
8007b8d6 SW |
517 | if (con->sock) { |
518 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
519 | sock_release(con->sock); | |
520 | con->sock = NULL; | |
521 | } | |
456ea468 AE |
522 | |
523 | /* | |
4a861692 | 524 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
525 | * independent of the connection mutex, and we could have |
526 | * received a socket close event before we had the chance to | |
527 | * shut the socket down. | |
528 | */ | |
c9ffc77a | 529 | con_flag_clear(con, CON_FLAG_SOCK_CLOSED); |
8007b8d6 | 530 | |
ce2c8903 | 531 | con_sock_state_closed(con); |
31b8006e SW |
532 | return rc; |
533 | } | |
534 | ||
535 | /* | |
536 | * Reset a connection. Discard all incoming and outgoing messages | |
537 | * and clear *_seq state. | |
538 | */ | |
539 | static void ceph_msg_remove(struct ceph_msg *msg) | |
540 | { | |
541 | list_del_init(&msg->list_head); | |
38941f80 | 542 | BUG_ON(msg->con == NULL); |
36eb71aa | 543 | msg->con->ops->put(msg->con); |
38941f80 AE |
544 | msg->con = NULL; |
545 | ||
31b8006e SW |
546 | ceph_msg_put(msg); |
547 | } | |
548 | static void ceph_msg_remove_list(struct list_head *head) | |
549 | { | |
550 | while (!list_empty(head)) { | |
551 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
552 | list_head); | |
553 | ceph_msg_remove(msg); | |
554 | } | |
555 | } | |
556 | ||
557 | static void reset_connection(struct ceph_connection *con) | |
558 | { | |
559 | /* reset connection, out_queue, msg_ and connect_seq */ | |
560 | /* discard existing out_queue and msg_seq */ | |
0fa6ebc6 | 561 | dout("reset_connection %p\n", con); |
31b8006e SW |
562 | ceph_msg_remove_list(&con->out_queue); |
563 | ceph_msg_remove_list(&con->out_sent); | |
564 | ||
cf3e5c40 | 565 | if (con->in_msg) { |
38941f80 AE |
566 | BUG_ON(con->in_msg->con != con); |
567 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
568 | ceph_msg_put(con->in_msg); |
569 | con->in_msg = NULL; | |
36eb71aa | 570 | con->ops->put(con); |
cf3e5c40 SW |
571 | } |
572 | ||
31b8006e SW |
573 | con->connect_seq = 0; |
574 | con->out_seq = 0; | |
c86a2930 SW |
575 | if (con->out_msg) { |
576 | ceph_msg_put(con->out_msg); | |
577 | con->out_msg = NULL; | |
578 | } | |
31b8006e | 579 | con->in_seq = 0; |
0e0d5e0c | 580 | con->in_seq_acked = 0; |
31b8006e SW |
581 | } |
582 | ||
583 | /* | |
584 | * mark a peer down. drop any open connections. | |
585 | */ | |
586 | void ceph_con_close(struct ceph_connection *con) | |
587 | { | |
8c50c817 | 588 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
589 | dout("con_close %p peer %s\n", con, |
590 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
8dacc7da | 591 | con->state = CON_STATE_CLOSED; |
a5988c49 | 592 | |
c9ffc77a AE |
593 | con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */ |
594 | con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING); | |
595 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); | |
596 | con_flag_clear(con, CON_FLAG_BACKOFF); | |
a5988c49 | 597 | |
31b8006e | 598 | reset_connection(con); |
6f2bc3ff | 599 | con->peer_global_seq = 0; |
91e45ce3 | 600 | cancel_delayed_work(&con->work); |
ee76e073 | 601 | con_close_socket(con); |
ec302645 | 602 | mutex_unlock(&con->mutex); |
31b8006e | 603 | } |
3d14c5d2 | 604 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 605 | |
31b8006e SW |
606 | /* |
607 | * Reopen a closed connection, with a new peer address. | |
608 | */ | |
b7a9e5dd SW |
609 | void ceph_con_open(struct ceph_connection *con, |
610 | __u8 entity_type, __u64 entity_num, | |
611 | struct ceph_entity_addr *addr) | |
31b8006e | 612 | { |
5469155f | 613 | mutex_lock(&con->mutex); |
3d14c5d2 | 614 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
8dacc7da | 615 | |
122070a2 | 616 | WARN_ON(con->state != CON_STATE_CLOSED); |
8dacc7da | 617 | con->state = CON_STATE_PREOPEN; |
a5988c49 | 618 | |
b7a9e5dd SW |
619 | con->peer_name.type = (__u8) entity_type; |
620 | con->peer_name.num = cpu_to_le64(entity_num); | |
621 | ||
31b8006e | 622 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 623 | con->delay = 0; /* reset backoff memory */ |
5469155f | 624 | mutex_unlock(&con->mutex); |
31b8006e SW |
625 | queue_con(con); |
626 | } | |
3d14c5d2 | 627 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 628 | |
87b315a5 SW |
629 | /* |
630 | * return true if this connection ever successfully opened | |
631 | */ | |
632 | bool ceph_con_opened(struct ceph_connection *con) | |
633 | { | |
634 | return con->connect_seq > 0; | |
635 | } | |
636 | ||
31b8006e SW |
637 | /* |
638 | * initialize a new connection. | |
639 | */ | |
1bfd89f4 AE |
640 | void ceph_con_init(struct ceph_connection *con, void *private, |
641 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 642 | struct ceph_messenger *msgr) |
31b8006e SW |
643 | { |
644 | dout("con_init %p\n", con); | |
645 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
646 | con->private = private; |
647 | con->ops = ops; | |
31b8006e | 648 | con->msgr = msgr; |
ce2c8903 AE |
649 | |
650 | con_sock_state_init(con); | |
651 | ||
ec302645 | 652 | mutex_init(&con->mutex); |
31b8006e SW |
653 | INIT_LIST_HEAD(&con->out_queue); |
654 | INIT_LIST_HEAD(&con->out_sent); | |
655 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 | 656 | |
8dacc7da | 657 | con->state = CON_STATE_CLOSED; |
31b8006e | 658 | } |
3d14c5d2 | 659 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
660 | |
661 | ||
662 | /* | |
663 | * We maintain a global counter to order connection attempts. Get | |
664 | * a unique seq greater than @gt. | |
665 | */ | |
666 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
667 | { | |
668 | u32 ret; | |
669 | ||
670 | spin_lock(&msgr->global_seq_lock); | |
671 | if (msgr->global_seq < gt) | |
672 | msgr->global_seq = gt; | |
673 | ret = ++msgr->global_seq; | |
674 | spin_unlock(&msgr->global_seq_lock); | |
675 | return ret; | |
676 | } | |
677 | ||
e2200423 | 678 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
679 | { |
680 | con->out_kvec_left = 0; | |
681 | con->out_kvec_bytes = 0; | |
682 | con->out_kvec_cur = &con->out_kvec[0]; | |
683 | } | |
684 | ||
e2200423 | 685 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
686 | size_t size, void *data) |
687 | { | |
688 | int index; | |
689 | ||
690 | index = con->out_kvec_left; | |
691 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
692 | ||
693 | con->out_kvec[index].iov_len = size; | |
694 | con->out_kvec[index].iov_base = data; | |
695 | con->out_kvec_left++; | |
696 | con->out_kvec_bytes += size; | |
697 | } | |
31b8006e | 698 | |
df6ad1f9 | 699 | #ifdef CONFIG_BLOCK |
07c09b72 AE |
700 | static void init_bio_iter(struct bio *bio, struct bio **bio_iter, |
701 | unsigned int *bio_seg) | |
df6ad1f9 AE |
702 | { |
703 | if (!bio) { | |
07c09b72 AE |
704 | *bio_iter = NULL; |
705 | *bio_seg = 0; | |
df6ad1f9 AE |
706 | return; |
707 | } | |
07c09b72 AE |
708 | *bio_iter = bio; |
709 | *bio_seg = (unsigned int) bio->bi_idx; | |
df6ad1f9 AE |
710 | } |
711 | ||
07c09b72 | 712 | static void iter_bio_next(struct bio **bio_iter, unsigned int *seg) |
df6ad1f9 AE |
713 | { |
714 | if (*bio_iter == NULL) | |
715 | return; | |
716 | ||
717 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
718 | ||
719 | (*seg)++; | |
720 | if (*seg == (*bio_iter)->bi_vcnt) | |
721 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
722 | } | |
723 | #endif | |
724 | ||
78625051 AE |
725 | static void prepare_message_data(struct ceph_msg *msg, |
726 | struct ceph_msg_pos *msg_pos) | |
739c905b | 727 | { |
739c905b AE |
728 | BUG_ON(!msg); |
729 | BUG_ON(!msg->hdr.data_len); | |
730 | ||
731 | /* initialize page iterator */ | |
bae6acd9 | 732 | msg_pos->page = 0; |
739c905b | 733 | if (msg->pages) |
bae6acd9 | 734 | msg_pos->page_pos = msg->page_alignment; |
739c905b | 735 | else |
bae6acd9 | 736 | msg_pos->page_pos = 0; |
572c588e | 737 | #ifdef CONFIG_BLOCK |
abdaa6a8 | 738 | if (msg->bio) |
572c588e AE |
739 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); |
740 | #endif | |
bae6acd9 AE |
741 | msg_pos->data_pos = 0; |
742 | msg_pos->did_page_crc = false; | |
739c905b AE |
743 | } |
744 | ||
31b8006e SW |
745 | /* |
746 | * Prepare footer for currently outgoing message, and finish things | |
747 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
748 | */ | |
859eb799 | 749 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
750 | { |
751 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 752 | int v = con->out_kvec_left; |
31b8006e | 753 | |
fd154f3c AE |
754 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
755 | ||
31b8006e SW |
756 | dout("prepare_write_message_footer %p\n", con); |
757 | con->out_kvec_is_msg = true; | |
758 | con->out_kvec[v].iov_base = &m->footer; | |
759 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
760 | con->out_kvec_bytes += sizeof(m->footer); | |
761 | con->out_kvec_left++; | |
762 | con->out_more = m->more_to_follow; | |
c86a2930 | 763 | con->out_msg_done = true; |
31b8006e SW |
764 | } |
765 | ||
766 | /* | |
767 | * Prepare headers for the next outgoing message. | |
768 | */ | |
769 | static void prepare_write_message(struct ceph_connection *con) | |
770 | { | |
771 | struct ceph_msg *m; | |
a9a0c51a | 772 | u32 crc; |
31b8006e | 773 | |
e2200423 | 774 | con_out_kvec_reset(con); |
31b8006e | 775 | con->out_kvec_is_msg = true; |
c86a2930 | 776 | con->out_msg_done = false; |
31b8006e SW |
777 | |
778 | /* Sneak an ack in there first? If we can get it into the same | |
779 | * TCP packet that's a good thing. */ | |
780 | if (con->in_seq > con->in_seq_acked) { | |
781 | con->in_seq_acked = con->in_seq; | |
e2200423 | 782 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 783 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 784 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 785 | &con->out_temp_ack); |
31b8006e SW |
786 | } |
787 | ||
38941f80 | 788 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 789 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 790 | con->out_msg = m; |
38941f80 | 791 | BUG_ON(m->con != con); |
4cf9d544 SW |
792 | |
793 | /* put message on sent list */ | |
794 | ceph_msg_get(m); | |
795 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 796 | |
e84346b7 SW |
797 | /* |
798 | * only assign outgoing seq # if we haven't sent this message | |
799 | * yet. if it is requeued, resend with it's original seq. | |
800 | */ | |
801 | if (m->needs_out_seq) { | |
802 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
803 | m->needs_out_seq = false; | |
804 | } | |
31b8006e | 805 | |
4a73ef27 | 806 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n", |
31b8006e SW |
807 | m, con->out_seq, le16_to_cpu(m->hdr.type), |
808 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
4a73ef27 | 809 | le32_to_cpu(m->hdr.data_len), m->length); |
31b8006e SW |
810 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); |
811 | ||
812 | /* tag + hdr + front + middle */ | |
e2200423 AE |
813 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
814 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
815 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 816 | |
31b8006e | 817 | if (m->middle) |
e2200423 | 818 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 819 | m->middle->vec.iov_base); |
31b8006e SW |
820 | |
821 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
822 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
823 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 824 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
825 | |
826 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
827 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
828 | if (m->middle) { | |
829 | crc = crc32c(0, m->middle->vec.iov_base, | |
830 | m->middle->vec.iov_len); | |
831 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
832 | } else | |
31b8006e | 833 | con->out_msg->footer.middle_crc = 0; |
739c905b | 834 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
835 | le32_to_cpu(con->out_msg->footer.front_crc), |
836 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
837 | ||
838 | /* is there a data payload? */ | |
739c905b | 839 | con->out_msg->footer.data_crc = 0; |
78625051 AE |
840 | if (m->hdr.data_len) { |
841 | prepare_message_data(con->out_msg, &con->out_msg_pos); | |
842 | con->out_more = 1; /* data + footer will follow */ | |
843 | } else { | |
31b8006e | 844 | /* no, queue up footer too and be done */ |
859eb799 | 845 | prepare_write_message_footer(con); |
78625051 | 846 | } |
31b8006e | 847 | |
c9ffc77a | 848 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
849 | } |
850 | ||
851 | /* | |
852 | * Prepare an ack. | |
853 | */ | |
854 | static void prepare_write_ack(struct ceph_connection *con) | |
855 | { | |
856 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
857 | con->in_seq_acked, con->in_seq); | |
858 | con->in_seq_acked = con->in_seq; | |
859 | ||
e2200423 | 860 | con_out_kvec_reset(con); |
859eb799 | 861 | |
e2200423 | 862 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 863 | |
31b8006e | 864 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 865 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
866 | &con->out_temp_ack); |
867 | ||
31b8006e | 868 | con->out_more = 1; /* more will follow.. eventually.. */ |
c9ffc77a | 869 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
870 | } |
871 | ||
872 | /* | |
873 | * Prepare to write keepalive byte. | |
874 | */ | |
875 | static void prepare_write_keepalive(struct ceph_connection *con) | |
876 | { | |
877 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
878 | con_out_kvec_reset(con); |
879 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
c9ffc77a | 880 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
881 | } |
882 | ||
883 | /* | |
884 | * Connection negotiation. | |
885 | */ | |
886 | ||
dac1e716 AE |
887 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
888 | int *auth_proto) | |
4e7a5dcd | 889 | { |
a3530df3 | 890 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
891 | |
892 | if (!con->ops->get_authorizer) { | |
893 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
894 | con->out_connect.authorizer_len = 0; | |
729796be | 895 | return NULL; |
b1c6b980 AE |
896 | } |
897 | ||
898 | /* Can't hold the mutex while getting authorizer */ | |
ec302645 | 899 | mutex_unlock(&con->mutex); |
dac1e716 | 900 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
ec302645 | 901 | mutex_lock(&con->mutex); |
4e7a5dcd | 902 | |
a3530df3 | 903 | if (IS_ERR(auth)) |
729796be | 904 | return auth; |
8dacc7da | 905 | if (con->state != CON_STATE_NEGOTIATING) |
729796be | 906 | return ERR_PTR(-EAGAIN); |
0da5d703 | 907 | |
8f43fb53 AE |
908 | con->auth_reply_buf = auth->authorizer_reply_buf; |
909 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
729796be | 910 | return auth; |
4e7a5dcd SW |
911 | } |
912 | ||
31b8006e SW |
913 | /* |
914 | * We connected to a peer and are saying hello. | |
915 | */ | |
e825a66d | 916 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 917 | { |
e2200423 AE |
918 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
919 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 920 | &con->msgr->my_enc_addr); |
eed0ef2c | 921 | |
eed0ef2c | 922 | con->out_more = 0; |
c9ffc77a | 923 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
eed0ef2c SW |
924 | } |
925 | ||
e825a66d | 926 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 927 | { |
95c96174 | 928 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 929 | int proto; |
dac1e716 | 930 | int auth_proto; |
729796be | 931 | struct ceph_auth_handshake *auth; |
31b8006e SW |
932 | |
933 | switch (con->peer_name.type) { | |
934 | case CEPH_ENTITY_TYPE_MON: | |
935 | proto = CEPH_MONC_PROTOCOL; | |
936 | break; | |
937 | case CEPH_ENTITY_TYPE_OSD: | |
938 | proto = CEPH_OSDC_PROTOCOL; | |
939 | break; | |
940 | case CEPH_ENTITY_TYPE_MDS: | |
941 | proto = CEPH_MDSC_PROTOCOL; | |
942 | break; | |
943 | default: | |
944 | BUG(); | |
945 | } | |
946 | ||
947 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
948 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 949 | |
e825a66d | 950 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
951 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
952 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
953 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
954 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
955 | con->out_connect.flags = 0; | |
31b8006e | 956 | |
dac1e716 AE |
957 | auth_proto = CEPH_AUTH_UNKNOWN; |
958 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
959 | if (IS_ERR(auth)) |
960 | return PTR_ERR(auth); | |
3da54776 | 961 | |
dac1e716 | 962 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
963 | con->out_connect.authorizer_len = auth ? |
964 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
965 | ||
e2200423 | 966 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
967 | &con->out_connect); |
968 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 969 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 970 | auth->authorizer_buf); |
859eb799 | 971 | |
31b8006e | 972 | con->out_more = 0; |
c9ffc77a | 973 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
4e7a5dcd | 974 | |
e10c758e | 975 | return 0; |
31b8006e SW |
976 | } |
977 | ||
31b8006e SW |
978 | /* |
979 | * write as much of pending kvecs to the socket as we can. | |
980 | * 1 -> done | |
981 | * 0 -> socket full, but more to do | |
982 | * <0 -> error | |
983 | */ | |
984 | static int write_partial_kvec(struct ceph_connection *con) | |
985 | { | |
986 | int ret; | |
987 | ||
988 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
989 | while (con->out_kvec_bytes > 0) { | |
990 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
991 | con->out_kvec_left, con->out_kvec_bytes, | |
992 | con->out_more); | |
993 | if (ret <= 0) | |
994 | goto out; | |
995 | con->out_kvec_bytes -= ret; | |
996 | if (con->out_kvec_bytes == 0) | |
997 | break; /* done */ | |
f42299e6 AE |
998 | |
999 | /* account for full iov entries consumed */ | |
1000 | while (ret >= con->out_kvec_cur->iov_len) { | |
1001 | BUG_ON(!con->out_kvec_left); | |
1002 | ret -= con->out_kvec_cur->iov_len; | |
1003 | con->out_kvec_cur++; | |
1004 | con->out_kvec_left--; | |
1005 | } | |
1006 | /* and for a partially-consumed entry */ | |
1007 | if (ret) { | |
1008 | con->out_kvec_cur->iov_len -= ret; | |
1009 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
1010 | } |
1011 | } | |
1012 | con->out_kvec_left = 0; | |
1013 | con->out_kvec_is_msg = false; | |
1014 | ret = 1; | |
1015 | out: | |
1016 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
1017 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
1018 | return ret; /* done! */ | |
1019 | } | |
1020 | ||
84ca8fc8 AE |
1021 | static void out_msg_pos_next(struct ceph_connection *con, struct page *page, |
1022 | size_t len, size_t sent, bool in_trail) | |
68b4476b | 1023 | { |
84ca8fc8 | 1024 | struct ceph_msg *msg = con->out_msg; |
bae6acd9 | 1025 | struct ceph_msg_pos *msg_pos = &con->out_msg_pos; |
68b4476b | 1026 | |
84ca8fc8 AE |
1027 | BUG_ON(!msg); |
1028 | BUG_ON(!sent); | |
68b4476b | 1029 | |
bae6acd9 AE |
1030 | msg_pos->data_pos += sent; |
1031 | msg_pos->page_pos += sent; | |
5821bd8c AE |
1032 | if (sent < len) |
1033 | return; | |
68b4476b | 1034 | |
5821bd8c | 1035 | BUG_ON(sent != len); |
bae6acd9 AE |
1036 | msg_pos->page_pos = 0; |
1037 | msg_pos->page++; | |
1038 | msg_pos->did_page_crc = false; | |
5821bd8c | 1039 | if (in_trail) |
35c7bfbc | 1040 | list_rotate_left(&msg->trail->head); |
5821bd8c | 1041 | else if (msg->pagelist) |
35c7bfbc | 1042 | list_rotate_left(&msg->pagelist->head); |
84ca8fc8 | 1043 | #ifdef CONFIG_BLOCK |
5821bd8c AE |
1044 | else if (msg->bio) |
1045 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
68b4476b | 1046 | #endif |
84ca8fc8 | 1047 | } |
68b4476b | 1048 | |
e788182f AE |
1049 | static void in_msg_pos_next(struct ceph_connection *con, size_t len, |
1050 | size_t received) | |
1051 | { | |
1052 | struct ceph_msg *msg = con->in_msg; | |
bae6acd9 | 1053 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; |
e788182f AE |
1054 | |
1055 | BUG_ON(!msg); | |
1056 | BUG_ON(!received); | |
1057 | ||
bae6acd9 AE |
1058 | msg_pos->data_pos += received; |
1059 | msg_pos->page_pos += received; | |
e788182f AE |
1060 | if (received < len) |
1061 | return; | |
1062 | ||
1063 | BUG_ON(received != len); | |
bae6acd9 AE |
1064 | msg_pos->page_pos = 0; |
1065 | msg_pos->page++; | |
e788182f AE |
1066 | #ifdef CONFIG_BLOCK |
1067 | if (msg->bio) | |
1068 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
1069 | #endif /* CONFIG_BLOCK */ | |
1070 | } | |
1071 | ||
31b8006e SW |
1072 | /* |
1073 | * Write as much message data payload as we can. If we finish, queue | |
1074 | * up the footer. | |
1075 | * 1 -> done, footer is now queued in out_kvec[]. | |
1076 | * 0 -> socket full, but more to do | |
1077 | * <0 -> error | |
1078 | */ | |
1079 | static int write_partial_msg_pages(struct ceph_connection *con) | |
1080 | { | |
1081 | struct ceph_msg *msg = con->out_msg; | |
bae6acd9 | 1082 | struct ceph_msg_pos *msg_pos = &con->out_msg_pos; |
95c96174 | 1083 | unsigned int data_len = le32_to_cpu(msg->hdr.data_len); |
31b8006e | 1084 | size_t len; |
37675b0f | 1085 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 1086 | int ret; |
68b4476b | 1087 | int total_max_write; |
84ca8fc8 | 1088 | bool in_trail = false; |
5821bd8c AE |
1089 | const size_t trail_len = (msg->trail ? msg->trail->length : 0); |
1090 | const size_t trail_off = data_len - trail_len; | |
31b8006e | 1091 | |
4a73ef27 | 1092 | dout("write_partial_msg_pages %p msg %p page %d offset %d\n", |
bae6acd9 | 1093 | con, msg, msg_pos->page, msg_pos->page_pos); |
31b8006e | 1094 | |
5821bd8c AE |
1095 | /* |
1096 | * Iterate through each page that contains data to be | |
1097 | * written, and send as much as possible for each. | |
1098 | * | |
1099 | * If we are calculating the data crc (the default), we will | |
1100 | * need to map the page. If we have no pages, they have | |
1101 | * been revoked, so use the zero page. | |
1102 | */ | |
bae6acd9 | 1103 | while (data_len > msg_pos->data_pos) { |
31b8006e | 1104 | struct page *page = NULL; |
68b4476b | 1105 | int max_write = PAGE_SIZE; |
9bd19663 | 1106 | int bio_offset = 0; |
68b4476b | 1107 | |
bae6acd9 | 1108 | in_trail = in_trail || msg_pos->data_pos >= trail_off; |
5821bd8c | 1109 | if (!in_trail) |
bae6acd9 | 1110 | total_max_write = trail_off - msg_pos->data_pos; |
68b4476b | 1111 | |
5821bd8c | 1112 | if (in_trail) { |
bae6acd9 | 1113 | total_max_write = data_len - msg_pos->data_pos; |
68b4476b YS |
1114 | |
1115 | page = list_first_entry(&msg->trail->head, | |
1116 | struct page, lru); | |
68b4476b | 1117 | } else if (msg->pages) { |
bae6acd9 | 1118 | page = msg->pages[msg_pos->page]; |
58bb3b37 SW |
1119 | } else if (msg->pagelist) { |
1120 | page = list_first_entry(&msg->pagelist->head, | |
1121 | struct page, lru); | |
68b4476b YS |
1122 | #ifdef CONFIG_BLOCK |
1123 | } else if (msg->bio) { | |
1124 | struct bio_vec *bv; | |
1125 | ||
1126 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
1127 | page = bv->bv_page; | |
9bd19663 | 1128 | bio_offset = bv->bv_offset; |
68b4476b YS |
1129 | max_write = bv->bv_len; |
1130 | #endif | |
31b8006e | 1131 | } else { |
57666519 | 1132 | page = zero_page; |
31b8006e | 1133 | } |
bae6acd9 | 1134 | len = min_t(int, max_write - msg_pos->page_pos, |
68b4476b YS |
1135 | total_max_write); |
1136 | ||
bae6acd9 | 1137 | if (do_datacrc && !msg_pos->did_page_crc) { |
9bd19663 | 1138 | void *base; |
5821bd8c | 1139 | u32 crc = le32_to_cpu(msg->footer.data_crc); |
8d63e318 | 1140 | char *kaddr; |
31b8006e | 1141 | |
8d63e318 | 1142 | kaddr = kmap(page); |
31b8006e | 1143 | BUG_ON(kaddr == NULL); |
bae6acd9 | 1144 | base = kaddr + msg_pos->page_pos + bio_offset; |
5821bd8c | 1145 | crc = crc32c(crc, base, len); |
5ce765a5 | 1146 | kunmap(page); |
84ca8fc8 | 1147 | msg->footer.data_crc = cpu_to_le32(crc); |
bae6acd9 | 1148 | msg_pos->did_page_crc = true; |
31b8006e | 1149 | } |
e36b13cc | 1150 | ret = ceph_tcp_sendpage(con->sock, page, |
bae6acd9 | 1151 | msg_pos->page_pos + bio_offset, |
e1dcb128 | 1152 | len, true); |
31b8006e SW |
1153 | if (ret <= 0) |
1154 | goto out; | |
1155 | ||
84ca8fc8 | 1156 | out_msg_pos_next(con, page, len, (size_t) ret, in_trail); |
31b8006e SW |
1157 | } |
1158 | ||
1159 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
1160 | ||
1161 | /* prepare and queue up footer, too */ | |
37675b0f | 1162 | if (!do_datacrc) |
84ca8fc8 | 1163 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1164 | con_out_kvec_reset(con); |
859eb799 | 1165 | prepare_write_message_footer(con); |
31b8006e SW |
1166 | ret = 1; |
1167 | out: | |
1168 | return ret; | |
1169 | } | |
1170 | ||
1171 | /* | |
1172 | * write some zeros | |
1173 | */ | |
1174 | static int write_partial_skip(struct ceph_connection *con) | |
1175 | { | |
1176 | int ret; | |
1177 | ||
1178 | while (con->out_skip > 0) { | |
31739139 | 1179 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1180 | |
e1dcb128 | 1181 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true); |
31b8006e SW |
1182 | if (ret <= 0) |
1183 | goto out; | |
1184 | con->out_skip -= ret; | |
1185 | } | |
1186 | ret = 1; | |
1187 | out: | |
1188 | return ret; | |
1189 | } | |
1190 | ||
1191 | /* | |
1192 | * Prepare to read connection handshake, or an ack. | |
1193 | */ | |
eed0ef2c SW |
1194 | static void prepare_read_banner(struct ceph_connection *con) |
1195 | { | |
1196 | dout("prepare_read_banner %p\n", con); | |
1197 | con->in_base_pos = 0; | |
1198 | } | |
1199 | ||
31b8006e SW |
1200 | static void prepare_read_connect(struct ceph_connection *con) |
1201 | { | |
1202 | dout("prepare_read_connect %p\n", con); | |
1203 | con->in_base_pos = 0; | |
1204 | } | |
1205 | ||
1206 | static void prepare_read_ack(struct ceph_connection *con) | |
1207 | { | |
1208 | dout("prepare_read_ack %p\n", con); | |
1209 | con->in_base_pos = 0; | |
1210 | } | |
1211 | ||
1212 | static void prepare_read_tag(struct ceph_connection *con) | |
1213 | { | |
1214 | dout("prepare_read_tag %p\n", con); | |
1215 | con->in_base_pos = 0; | |
1216 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1217 | } | |
1218 | ||
1219 | /* | |
1220 | * Prepare to read a message. | |
1221 | */ | |
1222 | static int prepare_read_message(struct ceph_connection *con) | |
1223 | { | |
1224 | dout("prepare_read_message %p\n", con); | |
1225 | BUG_ON(con->in_msg != NULL); | |
1226 | con->in_base_pos = 0; | |
1227 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1228 | return 0; | |
1229 | } | |
1230 | ||
1231 | ||
1232 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1233 | int end, int size, void *object) |
31b8006e | 1234 | { |
e6cee71f AE |
1235 | while (con->in_base_pos < end) { |
1236 | int left = end - con->in_base_pos; | |
31b8006e SW |
1237 | int have = size - left; |
1238 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1239 | if (ret <= 0) | |
1240 | return ret; | |
1241 | con->in_base_pos += ret; | |
1242 | } | |
1243 | return 1; | |
1244 | } | |
1245 | ||
1246 | ||
1247 | /* | |
1248 | * Read all or part of the connect-side handshake on a new connection | |
1249 | */ | |
eed0ef2c | 1250 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1251 | { |
fd51653f AE |
1252 | int size; |
1253 | int end; | |
1254 | int ret; | |
31b8006e | 1255 | |
eed0ef2c | 1256 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1257 | |
1258 | /* peer's banner */ | |
fd51653f AE |
1259 | size = strlen(CEPH_BANNER); |
1260 | end = size; | |
1261 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1262 | if (ret <= 0) |
1263 | goto out; | |
fd51653f AE |
1264 | |
1265 | size = sizeof (con->actual_peer_addr); | |
1266 | end += size; | |
1267 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1268 | if (ret <= 0) |
1269 | goto out; | |
fd51653f AE |
1270 | |
1271 | size = sizeof (con->peer_addr_for_me); | |
1272 | end += size; | |
1273 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1274 | if (ret <= 0) |
1275 | goto out; | |
fd51653f | 1276 | |
eed0ef2c SW |
1277 | out: |
1278 | return ret; | |
1279 | } | |
1280 | ||
1281 | static int read_partial_connect(struct ceph_connection *con) | |
1282 | { | |
fd51653f AE |
1283 | int size; |
1284 | int end; | |
1285 | int ret; | |
eed0ef2c SW |
1286 | |
1287 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1288 | ||
fd51653f AE |
1289 | size = sizeof (con->in_reply); |
1290 | end = size; | |
1291 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1292 | if (ret <= 0) |
1293 | goto out; | |
fd51653f AE |
1294 | |
1295 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1296 | end += size; | |
1297 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1298 | if (ret <= 0) |
1299 | goto out; | |
31b8006e | 1300 | |
4e7a5dcd SW |
1301 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1302 | con, (int)con->in_reply.tag, | |
1303 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1304 | le32_to_cpu(con->in_reply.global_seq)); |
1305 | out: | |
1306 | return ret; | |
eed0ef2c | 1307 | |
31b8006e SW |
1308 | } |
1309 | ||
1310 | /* | |
1311 | * Verify the hello banner looks okay. | |
1312 | */ | |
1313 | static int verify_hello(struct ceph_connection *con) | |
1314 | { | |
1315 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1316 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1317 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1318 | con->error_msg = "protocol error, bad banner"; |
1319 | return -1; | |
1320 | } | |
1321 | return 0; | |
1322 | } | |
1323 | ||
1324 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1325 | { | |
1326 | switch (ss->ss_family) { | |
1327 | case AF_INET: | |
1328 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1329 | case AF_INET6: | |
1330 | return | |
1331 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1332 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1333 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1334 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1335 | } | |
1336 | return false; | |
1337 | } | |
1338 | ||
1339 | static int addr_port(struct sockaddr_storage *ss) | |
1340 | { | |
1341 | switch (ss->ss_family) { | |
1342 | case AF_INET: | |
f28bcfbe | 1343 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1344 | case AF_INET6: |
f28bcfbe | 1345 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1346 | } |
1347 | return 0; | |
1348 | } | |
1349 | ||
1350 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1351 | { | |
1352 | switch (ss->ss_family) { | |
1353 | case AF_INET: | |
1354 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1355 | break; |
31b8006e SW |
1356 | case AF_INET6: |
1357 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1358 | break; |
31b8006e SW |
1359 | } |
1360 | } | |
1361 | ||
ee3b56f2 NW |
1362 | /* |
1363 | * Unlike other *_pton function semantics, zero indicates success. | |
1364 | */ | |
1365 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1366 | char delim, const char **ipend) | |
1367 | { | |
99f0f3b2 AE |
1368 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1369 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1370 | |
1371 | memset(ss, 0, sizeof(*ss)); | |
1372 | ||
1373 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1374 | ss->ss_family = AF_INET; | |
1375 | return 0; | |
1376 | } | |
1377 | ||
1378 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1379 | ss->ss_family = AF_INET6; | |
1380 | return 0; | |
1381 | } | |
1382 | ||
1383 | return -EINVAL; | |
1384 | } | |
1385 | ||
1386 | /* | |
1387 | * Extract hostname string and resolve using kernel DNS facility. | |
1388 | */ | |
1389 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1390 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1391 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1392 | { | |
1393 | const char *end, *delim_p; | |
1394 | char *colon_p, *ip_addr = NULL; | |
1395 | int ip_len, ret; | |
1396 | ||
1397 | /* | |
1398 | * The end of the hostname occurs immediately preceding the delimiter or | |
1399 | * the port marker (':') where the delimiter takes precedence. | |
1400 | */ | |
1401 | delim_p = memchr(name, delim, namelen); | |
1402 | colon_p = memchr(name, ':', namelen); | |
1403 | ||
1404 | if (delim_p && colon_p) | |
1405 | end = delim_p < colon_p ? delim_p : colon_p; | |
1406 | else if (!delim_p && colon_p) | |
1407 | end = colon_p; | |
1408 | else { | |
1409 | end = delim_p; | |
1410 | if (!end) /* case: hostname:/ */ | |
1411 | end = name + namelen; | |
1412 | } | |
1413 | ||
1414 | if (end <= name) | |
1415 | return -EINVAL; | |
1416 | ||
1417 | /* do dns_resolve upcall */ | |
1418 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1419 | if (ip_len > 0) | |
1420 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1421 | else | |
1422 | ret = -ESRCH; | |
1423 | ||
1424 | kfree(ip_addr); | |
1425 | ||
1426 | *ipend = end; | |
1427 | ||
1428 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1429 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1430 | ||
1431 | return ret; | |
1432 | } | |
1433 | #else | |
1434 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1435 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1436 | { | |
1437 | return -EINVAL; | |
1438 | } | |
1439 | #endif | |
1440 | ||
1441 | /* | |
1442 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1443 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1444 | */ | |
1445 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1446 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1447 | { | |
1448 | int ret; | |
1449 | ||
1450 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1451 | if (ret) | |
1452 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1453 | ||
1454 | return ret; | |
1455 | } | |
1456 | ||
31b8006e SW |
1457 | /* |
1458 | * Parse an ip[:port] list into an addr array. Use the default | |
1459 | * monitor port if a port isn't specified. | |
1460 | */ | |
1461 | int ceph_parse_ips(const char *c, const char *end, | |
1462 | struct ceph_entity_addr *addr, | |
1463 | int max_count, int *count) | |
1464 | { | |
ee3b56f2 | 1465 | int i, ret = -EINVAL; |
31b8006e SW |
1466 | const char *p = c; |
1467 | ||
1468 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1469 | for (i = 0; i < max_count; i++) { | |
1470 | const char *ipend; | |
1471 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1472 | int port; |
39139f64 SW |
1473 | char delim = ','; |
1474 | ||
1475 | if (*p == '[') { | |
1476 | delim = ']'; | |
1477 | p++; | |
1478 | } | |
31b8006e | 1479 | |
ee3b56f2 NW |
1480 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1481 | if (ret) | |
31b8006e | 1482 | goto bad; |
ee3b56f2 NW |
1483 | ret = -EINVAL; |
1484 | ||
31b8006e SW |
1485 | p = ipend; |
1486 | ||
39139f64 SW |
1487 | if (delim == ']') { |
1488 | if (*p != ']') { | |
1489 | dout("missing matching ']'\n"); | |
1490 | goto bad; | |
1491 | } | |
1492 | p++; | |
1493 | } | |
1494 | ||
31b8006e SW |
1495 | /* port? */ |
1496 | if (p < end && *p == ':') { | |
1497 | port = 0; | |
1498 | p++; | |
1499 | while (p < end && *p >= '0' && *p <= '9') { | |
1500 | port = (port * 10) + (*p - '0'); | |
1501 | p++; | |
1502 | } | |
1503 | if (port > 65535 || port == 0) | |
1504 | goto bad; | |
1505 | } else { | |
1506 | port = CEPH_MON_PORT; | |
1507 | } | |
1508 | ||
1509 | addr_set_port(ss, port); | |
1510 | ||
3d14c5d2 | 1511 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1512 | |
1513 | if (p == end) | |
1514 | break; | |
1515 | if (*p != ',') | |
1516 | goto bad; | |
1517 | p++; | |
1518 | } | |
1519 | ||
1520 | if (p != end) | |
1521 | goto bad; | |
1522 | ||
1523 | if (count) | |
1524 | *count = i + 1; | |
1525 | return 0; | |
1526 | ||
1527 | bad: | |
39139f64 | 1528 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1529 | return ret; |
31b8006e | 1530 | } |
3d14c5d2 | 1531 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1532 | |
eed0ef2c | 1533 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1534 | { |
eed0ef2c | 1535 | dout("process_banner on %p\n", con); |
31b8006e SW |
1536 | |
1537 | if (verify_hello(con) < 0) | |
1538 | return -1; | |
1539 | ||
63f2d211 SW |
1540 | ceph_decode_addr(&con->actual_peer_addr); |
1541 | ceph_decode_addr(&con->peer_addr_for_me); | |
1542 | ||
31b8006e SW |
1543 | /* |
1544 | * Make sure the other end is who we wanted. note that the other | |
1545 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1546 | * them the benefit of the doubt. | |
1547 | */ | |
103e2d3a SW |
1548 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1549 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1550 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1551 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1552 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1553 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1554 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1555 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1556 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1557 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1558 | return -1; |
1559 | } | |
1560 | ||
1561 | /* | |
1562 | * did we learn our address? | |
1563 | */ | |
1564 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1565 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1566 | ||
1567 | memcpy(&con->msgr->inst.addr.in_addr, | |
1568 | &con->peer_addr_for_me.in_addr, | |
1569 | sizeof(con->peer_addr_for_me.in_addr)); | |
1570 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1571 | encode_my_addr(con->msgr); |
eed0ef2c | 1572 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1573 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1574 | } |
1575 | ||
eed0ef2c SW |
1576 | return 0; |
1577 | } | |
1578 | ||
1579 | static int process_connect(struct ceph_connection *con) | |
1580 | { | |
3d14c5d2 YS |
1581 | u64 sup_feat = con->msgr->supported_features; |
1582 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1583 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1584 | int ret; |
04a419f9 | 1585 | |
eed0ef2c SW |
1586 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1587 | ||
31b8006e | 1588 | switch (con->in_reply.tag) { |
04a419f9 SW |
1589 | case CEPH_MSGR_TAG_FEATURES: |
1590 | pr_err("%s%lld %s feature set mismatch," | |
1591 | " my %llx < server's %llx, missing %llx\n", | |
1592 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1593 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1594 | sup_feat, server_feat, server_feat & ~sup_feat); |
1595 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 1596 | reset_connection(con); |
04a419f9 SW |
1597 | return -1; |
1598 | ||
31b8006e | 1599 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1600 | pr_err("%s%lld %s protocol version mismatch," |
1601 | " my %d != server's %d\n", | |
1602 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1603 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1604 | le32_to_cpu(con->out_connect.protocol_version), |
1605 | le32_to_cpu(con->in_reply.protocol_version)); | |
1606 | con->error_msg = "protocol version mismatch"; | |
0fa6ebc6 | 1607 | reset_connection(con); |
31b8006e SW |
1608 | return -1; |
1609 | ||
4e7a5dcd SW |
1610 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1611 | con->auth_retry++; | |
1612 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1613 | con->auth_retry); | |
1614 | if (con->auth_retry == 2) { | |
1615 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1616 | return -1; |
1617 | } | |
1618 | con->auth_retry = 1; | |
6d4221b5 | 1619 | con_out_kvec_reset(con); |
e825a66d | 1620 | ret = prepare_write_connect(con); |
0da5d703 SW |
1621 | if (ret < 0) |
1622 | return ret; | |
63733a0f | 1623 | prepare_read_connect(con); |
4e7a5dcd | 1624 | break; |
31b8006e SW |
1625 | |
1626 | case CEPH_MSGR_TAG_RESETSESSION: | |
1627 | /* | |
1628 | * If we connected with a large connect_seq but the peer | |
1629 | * has no record of a session with us (no connection, or | |
1630 | * connect_seq == 0), they will send RESETSESION to indicate | |
1631 | * that they must have reset their session, and may have | |
1632 | * dropped messages. | |
1633 | */ | |
1634 | dout("process_connect got RESET peer seq %u\n", | |
5bdca4e0 | 1635 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
1636 | pr_err("%s%lld %s connection reset\n", |
1637 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1638 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1639 | reset_connection(con); |
6d4221b5 | 1640 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1641 | ret = prepare_write_connect(con); |
1642 | if (ret < 0) | |
1643 | return ret; | |
31b8006e SW |
1644 | prepare_read_connect(con); |
1645 | ||
1646 | /* Tell ceph about it. */ | |
ec302645 | 1647 | mutex_unlock(&con->mutex); |
31b8006e SW |
1648 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1649 | if (con->ops->peer_reset) | |
1650 | con->ops->peer_reset(con); | |
ec302645 | 1651 | mutex_lock(&con->mutex); |
8dacc7da | 1652 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 1653 | return -EAGAIN; |
31b8006e SW |
1654 | break; |
1655 | ||
1656 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1657 | /* | |
1658 | * If we sent a smaller connect_seq than the peer has, try | |
1659 | * again with a larger value. | |
1660 | */ | |
5bdca4e0 | 1661 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 1662 | le32_to_cpu(con->out_connect.connect_seq), |
5bdca4e0 SW |
1663 | le32_to_cpu(con->in_reply.connect_seq)); |
1664 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
6d4221b5 | 1665 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1666 | ret = prepare_write_connect(con); |
1667 | if (ret < 0) | |
1668 | return ret; | |
31b8006e SW |
1669 | prepare_read_connect(con); |
1670 | break; | |
1671 | ||
1672 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1673 | /* | |
1674 | * If we sent a smaller global_seq than the peer has, try | |
1675 | * again with a larger value. | |
1676 | */ | |
eed0ef2c | 1677 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 1678 | con->peer_global_seq, |
5bdca4e0 | 1679 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 1680 | get_global_seq(con->msgr, |
5bdca4e0 | 1681 | le32_to_cpu(con->in_reply.global_seq)); |
6d4221b5 | 1682 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1683 | ret = prepare_write_connect(con); |
1684 | if (ret < 0) | |
1685 | return ret; | |
31b8006e SW |
1686 | prepare_read_connect(con); |
1687 | break; | |
1688 | ||
1689 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1690 | if (req_feat & ~server_feat) { |
1691 | pr_err("%s%lld %s protocol feature mismatch," | |
1692 | " my required %llx > server's %llx, need %llx\n", | |
1693 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1694 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1695 | req_feat, server_feat, req_feat & ~server_feat); |
1696 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 1697 | reset_connection(con); |
04a419f9 SW |
1698 | return -1; |
1699 | } | |
8dacc7da | 1700 | |
122070a2 | 1701 | WARN_ON(con->state != CON_STATE_NEGOTIATING); |
8dacc7da SW |
1702 | con->state = CON_STATE_OPEN; |
1703 | ||
31b8006e SW |
1704 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1705 | con->connect_seq++; | |
aba558e2 | 1706 | con->peer_features = server_feat; |
31b8006e SW |
1707 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1708 | con->peer_global_seq, | |
1709 | le32_to_cpu(con->in_reply.connect_seq), | |
1710 | con->connect_seq); | |
1711 | WARN_ON(con->connect_seq != | |
1712 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1713 | |
1714 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
c9ffc77a | 1715 | con_flag_set(con, CON_FLAG_LOSSYTX); |
92ac41d0 | 1716 | |
85effe18 | 1717 | con->delay = 0; /* reset backoff memory */ |
92ac41d0 | 1718 | |
31b8006e SW |
1719 | prepare_read_tag(con); |
1720 | break; | |
1721 | ||
1722 | case CEPH_MSGR_TAG_WAIT: | |
1723 | /* | |
1724 | * If there is a connection race (we are opening | |
1725 | * connections to each other), one of us may just have | |
1726 | * to WAIT. This shouldn't happen if we are the | |
1727 | * client. | |
1728 | */ | |
04177882 SW |
1729 | pr_err("process_connect got WAIT as client\n"); |
1730 | con->error_msg = "protocol error, got WAIT as client"; | |
1731 | return -1; | |
31b8006e SW |
1732 | |
1733 | default: | |
1734 | pr_err("connect protocol error, will retry\n"); | |
1735 | con->error_msg = "protocol error, garbage tag during connect"; | |
1736 | return -1; | |
1737 | } | |
1738 | return 0; | |
1739 | } | |
1740 | ||
1741 | ||
1742 | /* | |
1743 | * read (part of) an ack | |
1744 | */ | |
1745 | static int read_partial_ack(struct ceph_connection *con) | |
1746 | { | |
fd51653f AE |
1747 | int size = sizeof (con->in_temp_ack); |
1748 | int end = size; | |
31b8006e | 1749 | |
fd51653f | 1750 | return read_partial(con, end, size, &con->in_temp_ack); |
31b8006e SW |
1751 | } |
1752 | ||
1753 | ||
1754 | /* | |
1755 | * We can finally discard anything that's been acked. | |
1756 | */ | |
1757 | static void process_ack(struct ceph_connection *con) | |
1758 | { | |
1759 | struct ceph_msg *m; | |
1760 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1761 | u64 seq; | |
1762 | ||
31b8006e SW |
1763 | while (!list_empty(&con->out_sent)) { |
1764 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1765 | list_head); | |
1766 | seq = le64_to_cpu(m->hdr.seq); | |
1767 | if (seq > ack) | |
1768 | break; | |
1769 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1770 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1771 | m->ack_stamp = jiffies; |
31b8006e SW |
1772 | ceph_msg_remove(m); |
1773 | } | |
31b8006e SW |
1774 | prepare_read_tag(con); |
1775 | } | |
1776 | ||
1777 | ||
1778 | ||
1779 | ||
2450418c | 1780 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1781 | struct kvec *section, |
1782 | unsigned int sec_len, u32 *crc) | |
2450418c | 1783 | { |
68b4476b | 1784 | int ret, left; |
2450418c YS |
1785 | |
1786 | BUG_ON(!section); | |
1787 | ||
1788 | while (section->iov_len < sec_len) { | |
1789 | BUG_ON(section->iov_base == NULL); | |
1790 | left = sec_len - section->iov_len; | |
1791 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1792 | section->iov_len, left); | |
1793 | if (ret <= 0) | |
1794 | return ret; | |
1795 | section->iov_len += ret; | |
2450418c | 1796 | } |
fe3ad593 AE |
1797 | if (section->iov_len == sec_len) |
1798 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1799 | |
2450418c YS |
1800 | return 1; |
1801 | } | |
31b8006e | 1802 | |
4740a623 | 1803 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
68b4476b YS |
1804 | |
1805 | static int read_partial_message_pages(struct ceph_connection *con, | |
1806 | struct page **pages, | |
95c96174 | 1807 | unsigned int data_len, bool do_datacrc) |
68b4476b | 1808 | { |
bae6acd9 | 1809 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; |
e788182f | 1810 | struct page *page; |
68b4476b YS |
1811 | void *p; |
1812 | int ret; | |
1813 | int left; | |
1814 | ||
bae6acd9 AE |
1815 | left = min((int)(data_len - msg_pos->data_pos), |
1816 | (int)(PAGE_SIZE - msg_pos->page_pos)); | |
68b4476b YS |
1817 | /* (page) data */ |
1818 | BUG_ON(pages == NULL); | |
bae6acd9 | 1819 | page = pages[msg_pos->page]; |
e788182f | 1820 | p = kmap(page); |
bae6acd9 | 1821 | ret = ceph_tcp_recvmsg(con->sock, p + msg_pos->page_pos, left); |
bca064d2 | 1822 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1823 | con->in_data_crc = |
1824 | crc32c(con->in_data_crc, | |
bae6acd9 | 1825 | p + msg_pos->page_pos, ret); |
e788182f | 1826 | kunmap(page); |
68b4476b YS |
1827 | if (ret <= 0) |
1828 | return ret; | |
e788182f AE |
1829 | |
1830 | in_msg_pos_next(con, left, ret); | |
68b4476b YS |
1831 | |
1832 | return ret; | |
1833 | } | |
1834 | ||
1835 | #ifdef CONFIG_BLOCK | |
1836 | static int read_partial_message_bio(struct ceph_connection *con, | |
95c96174 | 1837 | unsigned int data_len, bool do_datacrc) |
68b4476b | 1838 | { |
b3d56fab | 1839 | struct ceph_msg *msg = con->in_msg; |
bae6acd9 | 1840 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; |
b3d56fab | 1841 | struct bio_vec *bv; |
e788182f | 1842 | struct page *page; |
68b4476b YS |
1843 | void *p; |
1844 | int ret, left; | |
1845 | ||
b3d56fab AE |
1846 | BUG_ON(!msg); |
1847 | BUG_ON(!msg->bio_iter); | |
1848 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
e788182f | 1849 | |
bae6acd9 AE |
1850 | left = min((int)(data_len - msg_pos->data_pos), |
1851 | (int)(bv->bv_len - msg_pos->page_pos)); | |
68b4476b | 1852 | |
e788182f AE |
1853 | page = bv->bv_page; |
1854 | p = kmap(page) + bv->bv_offset; | |
68b4476b | 1855 | |
bae6acd9 | 1856 | ret = ceph_tcp_recvmsg(con->sock, p + msg_pos->page_pos, left); |
bca064d2 | 1857 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1858 | con->in_data_crc = |
1859 | crc32c(con->in_data_crc, | |
bae6acd9 | 1860 | p + msg_pos->page_pos, ret); |
e788182f | 1861 | kunmap(page); |
68b4476b YS |
1862 | if (ret <= 0) |
1863 | return ret; | |
e788182f AE |
1864 | |
1865 | in_msg_pos_next(con, left, ret); | |
68b4476b YS |
1866 | |
1867 | return ret; | |
1868 | } | |
1869 | #endif | |
1870 | ||
31b8006e SW |
1871 | /* |
1872 | * read (part of) a message. | |
1873 | */ | |
1874 | static int read_partial_message(struct ceph_connection *con) | |
1875 | { | |
1876 | struct ceph_msg *m = con->in_msg; | |
bae6acd9 | 1877 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; |
fd51653f AE |
1878 | int size; |
1879 | int end; | |
31b8006e | 1880 | int ret; |
95c96174 | 1881 | unsigned int front_len, middle_len, data_len; |
37675b0f | 1882 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 1883 | u64 seq; |
fe3ad593 | 1884 | u32 crc; |
31b8006e SW |
1885 | |
1886 | dout("read_partial_message con %p msg %p\n", con, m); | |
1887 | ||
1888 | /* header */ | |
fd51653f AE |
1889 | size = sizeof (con->in_hdr); |
1890 | end = size; | |
1891 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1892 | if (ret <= 0) |
1893 | return ret; | |
fe3ad593 AE |
1894 | |
1895 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1896 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1897 | pr_err("read_partial_message bad hdr " | |
1898 | " crc %u != expected %u\n", | |
1899 | crc, con->in_hdr.crc); | |
1900 | return -EBADMSG; | |
1901 | } | |
1902 | ||
31b8006e SW |
1903 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1904 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1905 | return -EIO; | |
1906 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
7b11ba37 | 1907 | if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN) |
31b8006e SW |
1908 | return -EIO; |
1909 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1910 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1911 | return -EIO; | |
1912 | ||
ae18756b SW |
1913 | /* verify seq# */ |
1914 | seq = le64_to_cpu(con->in_hdr.seq); | |
1915 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1916 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1917 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1918 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1919 | seq, con->in_seq + 1); |
1920 | con->in_base_pos = -front_len - middle_len - data_len - | |
1921 | sizeof(m->footer); | |
1922 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1923 | return 0; |
1924 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1925 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1926 | seq, con->in_seq + 1); | |
1927 | con->error_msg = "bad message sequence # for incoming message"; | |
1928 | return -EBADMSG; | |
1929 | } | |
1930 | ||
31b8006e SW |
1931 | /* allocate message? */ |
1932 | if (!con->in_msg) { | |
4740a623 SW |
1933 | int skip = 0; |
1934 | ||
31b8006e | 1935 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
6ebc8b32 | 1936 | front_len, data_len); |
4740a623 SW |
1937 | ret = ceph_con_in_msg_alloc(con, &skip); |
1938 | if (ret < 0) | |
1939 | return ret; | |
2450418c | 1940 | if (skip) { |
31b8006e | 1941 | /* skip this message */ |
a79832f2 | 1942 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1943 | BUG_ON(con->in_msg); |
31b8006e SW |
1944 | con->in_base_pos = -front_len - middle_len - data_len - |
1945 | sizeof(m->footer); | |
1946 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1947 | con->in_seq++; |
31b8006e SW |
1948 | return 0; |
1949 | } | |
38941f80 | 1950 | |
4740a623 | 1951 | BUG_ON(!con->in_msg); |
38941f80 | 1952 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
1953 | m = con->in_msg; |
1954 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1955 | if (m->middle) |
1956 | m->middle->vec.iov_len = 0; | |
9d7f0f13 | 1957 | |
78625051 | 1958 | /* prepare for data payload, if any */ |
a4107026 | 1959 | |
78625051 AE |
1960 | if (data_len) |
1961 | prepare_message_data(con->in_msg, &con->in_msg_pos); | |
31b8006e SW |
1962 | } |
1963 | ||
1964 | /* front */ | |
2450418c YS |
1965 | ret = read_partial_message_section(con, &m->front, front_len, |
1966 | &con->in_front_crc); | |
1967 | if (ret <= 0) | |
1968 | return ret; | |
31b8006e SW |
1969 | |
1970 | /* middle */ | |
2450418c | 1971 | if (m->middle) { |
213c99ee SW |
1972 | ret = read_partial_message_section(con, &m->middle->vec, |
1973 | middle_len, | |
2450418c | 1974 | &con->in_middle_crc); |
31b8006e SW |
1975 | if (ret <= 0) |
1976 | return ret; | |
31b8006e SW |
1977 | } |
1978 | ||
1979 | /* (page) data */ | |
bae6acd9 | 1980 | while (msg_pos->data_pos < data_len) { |
68b4476b YS |
1981 | if (m->pages) { |
1982 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1983 | data_len, do_datacrc); |
68b4476b YS |
1984 | if (ret <= 0) |
1985 | return ret; | |
1986 | #ifdef CONFIG_BLOCK | |
1987 | } else if (m->bio) { | |
68b4476b | 1988 | ret = read_partial_message_bio(con, |
bca064d2 | 1989 | data_len, do_datacrc); |
68b4476b YS |
1990 | if (ret <= 0) |
1991 | return ret; | |
1992 | #endif | |
1993 | } else { | |
1994 | BUG_ON(1); | |
31b8006e SW |
1995 | } |
1996 | } | |
1997 | ||
31b8006e | 1998 | /* footer */ |
fd51653f AE |
1999 | size = sizeof (m->footer); |
2000 | end += size; | |
2001 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
2002 | if (ret <= 0) |
2003 | return ret; | |
2004 | ||
31b8006e SW |
2005 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
2006 | m, front_len, m->footer.front_crc, middle_len, | |
2007 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
2008 | ||
2009 | /* crc ok? */ | |
2010 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
2011 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
2012 | m, con->in_front_crc, m->footer.front_crc); | |
2013 | return -EBADMSG; | |
2014 | } | |
2015 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
2016 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
2017 | m, con->in_middle_crc, m->footer.middle_crc); | |
2018 | return -EBADMSG; | |
2019 | } | |
bca064d2 | 2020 | if (do_datacrc && |
31b8006e SW |
2021 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
2022 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
2023 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
2024 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
2025 | return -EBADMSG; | |
2026 | } | |
2027 | ||
2028 | return 1; /* done! */ | |
2029 | } | |
2030 | ||
2031 | /* | |
2032 | * Process message. This happens in the worker thread. The callback should | |
2033 | * be careful not to do anything that waits on other incoming messages or it | |
2034 | * may deadlock. | |
2035 | */ | |
2036 | static void process_message(struct ceph_connection *con) | |
2037 | { | |
5e095e8b | 2038 | struct ceph_msg *msg; |
31b8006e | 2039 | |
38941f80 AE |
2040 | BUG_ON(con->in_msg->con != con); |
2041 | con->in_msg->con = NULL; | |
5e095e8b | 2042 | msg = con->in_msg; |
31b8006e | 2043 | con->in_msg = NULL; |
36eb71aa | 2044 | con->ops->put(con); |
31b8006e SW |
2045 | |
2046 | /* if first message, set peer_name */ | |
2047 | if (con->peer_name.type == 0) | |
dbad185d | 2048 | con->peer_name = msg->hdr.src; |
31b8006e | 2049 | |
31b8006e | 2050 | con->in_seq++; |
ec302645 | 2051 | mutex_unlock(&con->mutex); |
31b8006e SW |
2052 | |
2053 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
2054 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 2055 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
2056 | le16_to_cpu(msg->hdr.type), |
2057 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2058 | le32_to_cpu(msg->hdr.front_len), | |
2059 | le32_to_cpu(msg->hdr.data_len), | |
2060 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
2061 | con->ops->dispatch(con, msg); | |
ec302645 SW |
2062 | |
2063 | mutex_lock(&con->mutex); | |
31b8006e SW |
2064 | } |
2065 | ||
2066 | ||
2067 | /* | |
2068 | * Write something to the socket. Called in a worker thread when the | |
2069 | * socket appears to be writeable and we have something ready to send. | |
2070 | */ | |
2071 | static int try_write(struct ceph_connection *con) | |
2072 | { | |
31b8006e SW |
2073 | int ret = 1; |
2074 | ||
d59315ca | 2075 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 2076 | |
31b8006e SW |
2077 | more: |
2078 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
2079 | ||
2080 | /* open the socket first? */ | |
8dacc7da SW |
2081 | if (con->state == CON_STATE_PREOPEN) { |
2082 | BUG_ON(con->sock); | |
2083 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2084 | |
e2200423 | 2085 | con_out_kvec_reset(con); |
e825a66d | 2086 | prepare_write_banner(con); |
eed0ef2c | 2087 | prepare_read_banner(con); |
31b8006e | 2088 | |
cf3e5c40 | 2089 | BUG_ON(con->in_msg); |
31b8006e SW |
2090 | con->in_tag = CEPH_MSGR_TAG_READY; |
2091 | dout("try_write initiating connect on %p new state %lu\n", | |
2092 | con, con->state); | |
41617d0c AE |
2093 | ret = ceph_tcp_connect(con); |
2094 | if (ret < 0) { | |
31b8006e | 2095 | con->error_msg = "connect error"; |
31b8006e SW |
2096 | goto out; |
2097 | } | |
2098 | } | |
2099 | ||
2100 | more_kvec: | |
2101 | /* kvec data queued? */ | |
2102 | if (con->out_skip) { | |
2103 | ret = write_partial_skip(con); | |
2104 | if (ret <= 0) | |
42961d23 | 2105 | goto out; |
31b8006e SW |
2106 | } |
2107 | if (con->out_kvec_left) { | |
2108 | ret = write_partial_kvec(con); | |
2109 | if (ret <= 0) | |
42961d23 | 2110 | goto out; |
31b8006e SW |
2111 | } |
2112 | ||
2113 | /* msg pages? */ | |
2114 | if (con->out_msg) { | |
c86a2930 SW |
2115 | if (con->out_msg_done) { |
2116 | ceph_msg_put(con->out_msg); | |
2117 | con->out_msg = NULL; /* we're done with this one */ | |
2118 | goto do_next; | |
2119 | } | |
2120 | ||
31b8006e SW |
2121 | ret = write_partial_msg_pages(con); |
2122 | if (ret == 1) | |
2123 | goto more_kvec; /* we need to send the footer, too! */ | |
2124 | if (ret == 0) | |
42961d23 | 2125 | goto out; |
31b8006e SW |
2126 | if (ret < 0) { |
2127 | dout("try_write write_partial_msg_pages err %d\n", | |
2128 | ret); | |
42961d23 | 2129 | goto out; |
31b8006e SW |
2130 | } |
2131 | } | |
2132 | ||
c86a2930 | 2133 | do_next: |
8dacc7da | 2134 | if (con->state == CON_STATE_OPEN) { |
31b8006e SW |
2135 | /* is anything else pending? */ |
2136 | if (!list_empty(&con->out_queue)) { | |
2137 | prepare_write_message(con); | |
2138 | goto more; | |
2139 | } | |
2140 | if (con->in_seq > con->in_seq_acked) { | |
2141 | prepare_write_ack(con); | |
2142 | goto more; | |
2143 | } | |
c9ffc77a | 2144 | if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) { |
31b8006e SW |
2145 | prepare_write_keepalive(con); |
2146 | goto more; | |
2147 | } | |
2148 | } | |
2149 | ||
2150 | /* Nothing to do! */ | |
c9ffc77a | 2151 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
31b8006e | 2152 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2153 | ret = 0; |
2154 | out: | |
42961d23 | 2155 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2156 | return ret; |
2157 | } | |
2158 | ||
2159 | ||
2160 | ||
2161 | /* | |
2162 | * Read what we can from the socket. | |
2163 | */ | |
2164 | static int try_read(struct ceph_connection *con) | |
2165 | { | |
31b8006e SW |
2166 | int ret = -1; |
2167 | ||
8dacc7da SW |
2168 | more: |
2169 | dout("try_read start on %p state %lu\n", con, con->state); | |
2170 | if (con->state != CON_STATE_CONNECTING && | |
2171 | con->state != CON_STATE_NEGOTIATING && | |
2172 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2173 | return 0; |
2174 | ||
8dacc7da | 2175 | BUG_ON(!con->sock); |
ec302645 | 2176 | |
31b8006e SW |
2177 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2178 | con->in_base_pos); | |
0da5d703 | 2179 | |
8dacc7da | 2180 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2181 | dout("try_read connecting\n"); |
2182 | ret = read_partial_banner(con); | |
2183 | if (ret <= 0) | |
ab166d5a | 2184 | goto out; |
7593af92 AE |
2185 | ret = process_banner(con); |
2186 | if (ret < 0) | |
2187 | goto out; | |
2188 | ||
8dacc7da | 2189 | con->state = CON_STATE_NEGOTIATING; |
7593af92 | 2190 | |
6d4221b5 JS |
2191 | /* |
2192 | * Received banner is good, exchange connection info. | |
2193 | * Do not reset out_kvec, as sending our banner raced | |
2194 | * with receiving peer banner after connect completed. | |
2195 | */ | |
7593af92 AE |
2196 | ret = prepare_write_connect(con); |
2197 | if (ret < 0) | |
2198 | goto out; | |
2199 | prepare_read_connect(con); | |
2200 | ||
2201 | /* Send connection info before awaiting response */ | |
0da5d703 SW |
2202 | goto out; |
2203 | } | |
2204 | ||
8dacc7da | 2205 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2206 | dout("try_read negotiating\n"); |
31b8006e SW |
2207 | ret = read_partial_connect(con); |
2208 | if (ret <= 0) | |
31b8006e | 2209 | goto out; |
98bdb0aa SW |
2210 | ret = process_connect(con); |
2211 | if (ret < 0) | |
2212 | goto out; | |
31b8006e SW |
2213 | goto more; |
2214 | } | |
2215 | ||
122070a2 | 2216 | WARN_ON(con->state != CON_STATE_OPEN); |
8dacc7da | 2217 | |
31b8006e SW |
2218 | if (con->in_base_pos < 0) { |
2219 | /* | |
2220 | * skipping + discarding content. | |
2221 | * | |
2222 | * FIXME: there must be a better way to do this! | |
2223 | */ | |
84495f49 AE |
2224 | static char buf[SKIP_BUF_SIZE]; |
2225 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2226 | ||
31b8006e SW |
2227 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2228 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2229 | if (ret <= 0) | |
98bdb0aa | 2230 | goto out; |
31b8006e SW |
2231 | con->in_base_pos += ret; |
2232 | if (con->in_base_pos) | |
2233 | goto more; | |
2234 | } | |
2235 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2236 | /* | |
2237 | * what's next? | |
2238 | */ | |
2239 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2240 | if (ret <= 0) | |
98bdb0aa | 2241 | goto out; |
31b8006e SW |
2242 | dout("try_read got tag %d\n", (int)con->in_tag); |
2243 | switch (con->in_tag) { | |
2244 | case CEPH_MSGR_TAG_MSG: | |
2245 | prepare_read_message(con); | |
2246 | break; | |
2247 | case CEPH_MSGR_TAG_ACK: | |
2248 | prepare_read_ack(con); | |
2249 | break; | |
2250 | case CEPH_MSGR_TAG_CLOSE: | |
8dacc7da SW |
2251 | con_close_socket(con); |
2252 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2253 | goto out; |
31b8006e SW |
2254 | default: |
2255 | goto bad_tag; | |
2256 | } | |
2257 | } | |
2258 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2259 | ret = read_partial_message(con); | |
2260 | if (ret <= 0) { | |
2261 | switch (ret) { | |
2262 | case -EBADMSG: | |
2263 | con->error_msg = "bad crc"; | |
2264 | ret = -EIO; | |
98bdb0aa | 2265 | break; |
31b8006e SW |
2266 | case -EIO: |
2267 | con->error_msg = "io error"; | |
98bdb0aa | 2268 | break; |
31b8006e | 2269 | } |
98bdb0aa | 2270 | goto out; |
31b8006e SW |
2271 | } |
2272 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2273 | goto more; | |
2274 | process_message(con); | |
7b862e07 SW |
2275 | if (con->state == CON_STATE_OPEN) |
2276 | prepare_read_tag(con); | |
31b8006e SW |
2277 | goto more; |
2278 | } | |
2279 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2280 | ret = read_partial_ack(con); | |
2281 | if (ret <= 0) | |
98bdb0aa | 2282 | goto out; |
31b8006e SW |
2283 | process_ack(con); |
2284 | goto more; | |
2285 | } | |
2286 | ||
31b8006e | 2287 | out: |
98bdb0aa | 2288 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2289 | return ret; |
2290 | ||
2291 | bad_tag: | |
2292 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2293 | con->error_msg = "protocol error, garbage tag"; | |
2294 | ret = -1; | |
2295 | goto out; | |
2296 | } | |
2297 | ||
2298 | ||
2299 | /* | |
802c6d96 AE |
2300 | * Atomically queue work on a connection after the specified delay. |
2301 | * Bump @con reference to avoid races with connection teardown. | |
2302 | * Returns 0 if work was queued, or an error code otherwise. | |
31b8006e | 2303 | */ |
802c6d96 | 2304 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
31b8006e | 2305 | { |
31b8006e | 2306 | if (!con->ops->get(con)) { |
802c6d96 AE |
2307 | dout("%s %p ref count 0\n", __func__, con); |
2308 | ||
2309 | return -ENOENT; | |
31b8006e SW |
2310 | } |
2311 | ||
802c6d96 AE |
2312 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
2313 | dout("%s %p - already queued\n", __func__, con); | |
31b8006e | 2314 | con->ops->put(con); |
802c6d96 AE |
2315 | |
2316 | return -EBUSY; | |
31b8006e | 2317 | } |
802c6d96 AE |
2318 | |
2319 | dout("%s %p %lu\n", __func__, con, delay); | |
2320 | ||
2321 | return 0; | |
2322 | } | |
2323 | ||
2324 | static void queue_con(struct ceph_connection *con) | |
2325 | { | |
2326 | (void) queue_con_delay(con, 0); | |
31b8006e SW |
2327 | } |
2328 | ||
7bb21d68 AE |
2329 | static bool con_sock_closed(struct ceph_connection *con) |
2330 | { | |
c9ffc77a | 2331 | if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED)) |
7bb21d68 AE |
2332 | return false; |
2333 | ||
2334 | #define CASE(x) \ | |
2335 | case CON_STATE_ ## x: \ | |
2336 | con->error_msg = "socket closed (con state " #x ")"; \ | |
2337 | break; | |
2338 | ||
2339 | switch (con->state) { | |
2340 | CASE(CLOSED); | |
2341 | CASE(PREOPEN); | |
2342 | CASE(CONNECTING); | |
2343 | CASE(NEGOTIATING); | |
2344 | CASE(OPEN); | |
2345 | CASE(STANDBY); | |
2346 | default: | |
2347 | pr_warning("%s con %p unrecognized state %lu\n", | |
2348 | __func__, con, con->state); | |
2349 | con->error_msg = "unrecognized con state"; | |
2350 | BUG(); | |
2351 | break; | |
2352 | } | |
2353 | #undef CASE | |
2354 | ||
2355 | return true; | |
2356 | } | |
2357 | ||
f20a39fd AE |
2358 | static bool con_backoff(struct ceph_connection *con) |
2359 | { | |
2360 | int ret; | |
2361 | ||
2362 | if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF)) | |
2363 | return false; | |
2364 | ||
2365 | ret = queue_con_delay(con, round_jiffies_relative(con->delay)); | |
2366 | if (ret) { | |
2367 | dout("%s: con %p FAILED to back off %lu\n", __func__, | |
2368 | con, con->delay); | |
2369 | BUG_ON(ret == -ENOENT); | |
2370 | con_flag_set(con, CON_FLAG_BACKOFF); | |
2371 | } | |
2372 | ||
2373 | return true; | |
2374 | } | |
2375 | ||
93209264 AE |
2376 | /* Finish fault handling; con->mutex must *not* be held here */ |
2377 | ||
2378 | static void con_fault_finish(struct ceph_connection *con) | |
2379 | { | |
2380 | /* | |
2381 | * in case we faulted due to authentication, invalidate our | |
2382 | * current tickets so that we can get new ones. | |
2383 | */ | |
2384 | if (con->auth_retry && con->ops->invalidate_authorizer) { | |
2385 | dout("calling invalidate_authorizer()\n"); | |
2386 | con->ops->invalidate_authorizer(con); | |
2387 | } | |
2388 | ||
2389 | if (con->ops->fault) | |
2390 | con->ops->fault(con); | |
2391 | } | |
2392 | ||
31b8006e SW |
2393 | /* |
2394 | * Do some work on a connection. Drop a connection ref when we're done. | |
2395 | */ | |
2396 | static void con_work(struct work_struct *work) | |
2397 | { | |
2398 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2399 | work.work); | |
49659416 | 2400 | bool fault; |
31b8006e | 2401 | |
9dd4658d | 2402 | mutex_lock(&con->mutex); |
49659416 AE |
2403 | while (true) { |
2404 | int ret; | |
31b8006e | 2405 | |
49659416 AE |
2406 | if ((fault = con_sock_closed(con))) { |
2407 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); | |
2408 | break; | |
2409 | } | |
2410 | if (con_backoff(con)) { | |
2411 | dout("%s: con %p BACKOFF\n", __func__, con); | |
2412 | break; | |
2413 | } | |
2414 | if (con->state == CON_STATE_STANDBY) { | |
2415 | dout("%s: con %p STANDBY\n", __func__, con); | |
2416 | break; | |
2417 | } | |
2418 | if (con->state == CON_STATE_CLOSED) { | |
2419 | dout("%s: con %p CLOSED\n", __func__, con); | |
2420 | BUG_ON(con->sock); | |
2421 | break; | |
2422 | } | |
2423 | if (con->state == CON_STATE_PREOPEN) { | |
2424 | dout("%s: con %p PREOPEN\n", __func__, con); | |
2425 | BUG_ON(con->sock); | |
2426 | } | |
0da5d703 | 2427 | |
49659416 AE |
2428 | ret = try_read(con); |
2429 | if (ret < 0) { | |
2430 | if (ret == -EAGAIN) | |
2431 | continue; | |
2432 | con->error_msg = "socket error on read"; | |
2433 | fault = true; | |
2434 | break; | |
2435 | } | |
2436 | ||
2437 | ret = try_write(con); | |
2438 | if (ret < 0) { | |
2439 | if (ret == -EAGAIN) | |
2440 | continue; | |
2441 | con->error_msg = "socket error on write"; | |
2442 | fault = true; | |
2443 | } | |
2444 | ||
2445 | break; /* If we make it to here, we're done */ | |
3a140a0d | 2446 | } |
b6e7b6a1 AE |
2447 | if (fault) |
2448 | con_fault(con); | |
9dd4658d | 2449 | mutex_unlock(&con->mutex); |
0da5d703 | 2450 | |
b6e7b6a1 AE |
2451 | if (fault) |
2452 | con_fault_finish(con); | |
2453 | ||
2454 | con->ops->put(con); | |
31b8006e SW |
2455 | } |
2456 | ||
31b8006e SW |
2457 | /* |
2458 | * Generic error/fault handler. A retry mechanism is used with | |
2459 | * exponential backoff | |
2460 | */ | |
93209264 | 2461 | static void con_fault(struct ceph_connection *con) |
31b8006e | 2462 | { |
28362986 | 2463 | pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2464 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2465 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2466 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2467 | |
122070a2 | 2468 | WARN_ON(con->state != CON_STATE_CONNECTING && |
8dacc7da SW |
2469 | con->state != CON_STATE_NEGOTIATING && |
2470 | con->state != CON_STATE_OPEN); | |
ec302645 | 2471 | |
31b8006e | 2472 | con_close_socket(con); |
5e095e8b | 2473 | |
c9ffc77a | 2474 | if (con_flag_test(con, CON_FLAG_LOSSYTX)) { |
8dacc7da SW |
2475 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
2476 | con->state = CON_STATE_CLOSED; | |
93209264 | 2477 | return; |
3b5ede07 SW |
2478 | } |
2479 | ||
5e095e8b | 2480 | if (con->in_msg) { |
38941f80 AE |
2481 | BUG_ON(con->in_msg->con != con); |
2482 | con->in_msg->con = NULL; | |
5e095e8b SW |
2483 | ceph_msg_put(con->in_msg); |
2484 | con->in_msg = NULL; | |
36eb71aa | 2485 | con->ops->put(con); |
5e095e8b | 2486 | } |
31b8006e | 2487 | |
e80a52d1 SW |
2488 | /* Requeue anything that hasn't been acked */ |
2489 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2490 | |
e76661d0 SW |
2491 | /* If there are no messages queued or keepalive pending, place |
2492 | * the connection in a STANDBY state */ | |
2493 | if (list_empty(&con->out_queue) && | |
c9ffc77a | 2494 | !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) { |
e00de341 | 2495 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
c9ffc77a | 2496 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
8dacc7da | 2497 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
2498 | } else { |
2499 | /* retry after a delay. */ | |
8dacc7da | 2500 | con->state = CON_STATE_PREOPEN; |
e80a52d1 SW |
2501 | if (con->delay == 0) |
2502 | con->delay = BASE_DELAY_INTERVAL; | |
2503 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2504 | con->delay *= 2; | |
c9ffc77a | 2505 | con_flag_set(con, CON_FLAG_BACKOFF); |
8618e30b | 2506 | queue_con(con); |
31b8006e | 2507 | } |
31b8006e SW |
2508 | } |
2509 | ||
2510 | ||
2511 | ||
2512 | /* | |
15d9882c | 2513 | * initialize a new messenger instance |
31b8006e | 2514 | */ |
15d9882c AE |
2515 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2516 | struct ceph_entity_addr *myaddr, | |
2517 | u32 supported_features, | |
2518 | u32 required_features, | |
2519 | bool nocrc) | |
31b8006e | 2520 | { |
3d14c5d2 YS |
2521 | msgr->supported_features = supported_features; |
2522 | msgr->required_features = required_features; | |
2523 | ||
31b8006e SW |
2524 | spin_lock_init(&msgr->global_seq_lock); |
2525 | ||
31b8006e SW |
2526 | if (myaddr) |
2527 | msgr->inst.addr = *myaddr; | |
2528 | ||
2529 | /* select a random nonce */ | |
ac8839d7 | 2530 | msgr->inst.addr.type = 0; |
103e2d3a | 2531 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2532 | encode_my_addr(msgr); |
15d9882c | 2533 | msgr->nocrc = nocrc; |
31b8006e | 2534 | |
a2a32584 | 2535 | atomic_set(&msgr->stopping, 0); |
31b8006e | 2536 | |
15d9882c | 2537 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2538 | } |
15d9882c | 2539 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2540 | |
e00de341 SW |
2541 | static void clear_standby(struct ceph_connection *con) |
2542 | { | |
2543 | /* come back from STANDBY? */ | |
8dacc7da | 2544 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 2545 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 2546 | con->state = CON_STATE_PREOPEN; |
e00de341 | 2547 | con->connect_seq++; |
c9ffc77a AE |
2548 | WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING)); |
2549 | WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)); | |
e00de341 SW |
2550 | } |
2551 | } | |
2552 | ||
31b8006e SW |
2553 | /* |
2554 | * Queue up an outgoing message on the given connection. | |
2555 | */ | |
2556 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2557 | { | |
31b8006e | 2558 | /* set src+dst */ |
dbad185d | 2559 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 2560 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
2561 | msg->needs_out_seq = true; |
2562 | ||
ec302645 | 2563 | mutex_lock(&con->mutex); |
92ce034b | 2564 | |
8dacc7da | 2565 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
2566 | dout("con_send %p closed, dropping %p\n", con, msg); |
2567 | ceph_msg_put(msg); | |
2568 | mutex_unlock(&con->mutex); | |
2569 | return; | |
2570 | } | |
2571 | ||
38941f80 | 2572 | BUG_ON(msg->con != NULL); |
36eb71aa | 2573 | msg->con = con->ops->get(con); |
92ce034b AE |
2574 | BUG_ON(msg->con == NULL); |
2575 | ||
31b8006e SW |
2576 | BUG_ON(!list_empty(&msg->list_head)); |
2577 | list_add_tail(&msg->list_head, &con->out_queue); | |
2578 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2579 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2580 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2581 | le32_to_cpu(msg->hdr.front_len), | |
2582 | le32_to_cpu(msg->hdr.middle_len), | |
2583 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
2584 | |
2585 | clear_standby(con); | |
ec302645 | 2586 | mutex_unlock(&con->mutex); |
31b8006e SW |
2587 | |
2588 | /* if there wasn't anything waiting to send before, queue | |
2589 | * new work */ | |
c9ffc77a | 2590 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
31b8006e SW |
2591 | queue_con(con); |
2592 | } | |
3d14c5d2 | 2593 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2594 | |
2595 | /* | |
2596 | * Revoke a message that was previously queued for send | |
2597 | */ | |
6740a845 | 2598 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2599 | { |
6740a845 AE |
2600 | struct ceph_connection *con = msg->con; |
2601 | ||
2602 | if (!con) | |
2603 | return; /* Message not in our possession */ | |
2604 | ||
ec302645 | 2605 | mutex_lock(&con->mutex); |
31b8006e | 2606 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2607 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2608 | list_del_init(&msg->list_head); |
38941f80 | 2609 | BUG_ON(msg->con == NULL); |
36eb71aa | 2610 | msg->con->ops->put(msg->con); |
38941f80 | 2611 | msg->con = NULL; |
31b8006e | 2612 | msg->hdr.seq = 0; |
38941f80 | 2613 | |
31b8006e | 2614 | ceph_msg_put(msg); |
ed98adad SW |
2615 | } |
2616 | if (con->out_msg == msg) { | |
38941f80 | 2617 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2618 | con->out_msg = NULL; |
31b8006e SW |
2619 | if (con->out_kvec_is_msg) { |
2620 | con->out_skip = con->out_kvec_bytes; | |
2621 | con->out_kvec_is_msg = false; | |
2622 | } | |
ed98adad | 2623 | msg->hdr.seq = 0; |
92ce034b AE |
2624 | |
2625 | ceph_msg_put(msg); | |
31b8006e | 2626 | } |
ec302645 | 2627 | mutex_unlock(&con->mutex); |
31b8006e SW |
2628 | } |
2629 | ||
350b1c32 | 2630 | /* |
0d59ab81 | 2631 | * Revoke a message that we may be reading data into |
350b1c32 | 2632 | */ |
8921d114 | 2633 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2634 | { |
8921d114 AE |
2635 | struct ceph_connection *con; |
2636 | ||
2637 | BUG_ON(msg == NULL); | |
2638 | if (!msg->con) { | |
2639 | dout("%s msg %p null con\n", __func__, msg); | |
2640 | ||
2641 | return; /* Message not in our possession */ | |
2642 | } | |
2643 | ||
2644 | con = msg->con; | |
350b1c32 | 2645 | mutex_lock(&con->mutex); |
8921d114 | 2646 | if (con->in_msg == msg) { |
95c96174 ED |
2647 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
2648 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
2649 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
2650 | |
2651 | /* skip rest of message */ | |
8921d114 AE |
2652 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
2653 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 2654 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
2655 | front_len - |
2656 | middle_len - | |
2657 | data_len - | |
350b1c32 | 2658 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2659 | ceph_msg_put(con->in_msg); |
2660 | con->in_msg = NULL; | |
2661 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2662 | con->in_seq++; |
350b1c32 | 2663 | } else { |
8921d114 AE |
2664 | dout("%s %p in_msg %p msg %p no-op\n", |
2665 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
2666 | } |
2667 | mutex_unlock(&con->mutex); | |
2668 | } | |
2669 | ||
31b8006e SW |
2670 | /* |
2671 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2672 | */ | |
2673 | void ceph_con_keepalive(struct ceph_connection *con) | |
2674 | { | |
e00de341 | 2675 | dout("con_keepalive %p\n", con); |
00650931 | 2676 | mutex_lock(&con->mutex); |
e00de341 | 2677 | clear_standby(con); |
00650931 | 2678 | mutex_unlock(&con->mutex); |
c9ffc77a AE |
2679 | if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 && |
2680 | con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) | |
31b8006e SW |
2681 | queue_con(con); |
2682 | } | |
3d14c5d2 | 2683 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e | 2684 | |
02afca6c | 2685 | void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages, |
f1baeb2b | 2686 | size_t length, size_t alignment) |
02afca6c | 2687 | { |
07aa1558 AE |
2688 | BUG_ON(!pages); |
2689 | BUG_ON(!length); | |
2690 | BUG_ON(msg->pages); | |
2691 | BUG_ON(msg->length); | |
02afca6c AE |
2692 | |
2693 | msg->pages = pages; | |
4a73ef27 | 2694 | msg->length = length; |
02afca6c AE |
2695 | msg->page_alignment = alignment & ~PAGE_MASK; |
2696 | } | |
2697 | EXPORT_SYMBOL(ceph_msg_data_set_pages); | |
31b8006e | 2698 | |
27fa8385 AE |
2699 | void ceph_msg_data_set_pagelist(struct ceph_msg *msg, |
2700 | struct ceph_pagelist *pagelist) | |
2701 | { | |
07aa1558 AE |
2702 | BUG_ON(!pagelist); |
2703 | BUG_ON(!pagelist->length); | |
2704 | BUG_ON(msg->pagelist); | |
27fa8385 AE |
2705 | |
2706 | msg->pagelist = pagelist; | |
2707 | } | |
2708 | EXPORT_SYMBOL(ceph_msg_data_set_pagelist); | |
2709 | ||
2710 | void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio) | |
2711 | { | |
07aa1558 AE |
2712 | BUG_ON(!bio); |
2713 | BUG_ON(msg->bio); | |
27fa8385 AE |
2714 | |
2715 | msg->bio = bio; | |
2716 | } | |
2717 | EXPORT_SYMBOL(ceph_msg_data_set_bio); | |
2718 | ||
2719 | void ceph_msg_data_set_trail(struct ceph_msg *msg, struct ceph_pagelist *trail) | |
2720 | { | |
07aa1558 AE |
2721 | BUG_ON(!trail); |
2722 | BUG_ON(!trail->length); | |
2723 | BUG_ON(msg->trail); | |
27fa8385 AE |
2724 | |
2725 | msg->trail = trail; | |
2726 | } | |
2727 | EXPORT_SYMBOL(ceph_msg_data_set_trail); | |
2728 | ||
31b8006e SW |
2729 | /* |
2730 | * construct a new message with given type, size | |
2731 | * the new msg has a ref count of 1. | |
2732 | */ | |
b61c2763 SW |
2733 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2734 | bool can_fail) | |
31b8006e SW |
2735 | { |
2736 | struct ceph_msg *m; | |
2737 | ||
9516e45b | 2738 | m = kzalloc(sizeof(*m), flags); |
31b8006e SW |
2739 | if (m == NULL) |
2740 | goto out; | |
31b8006e SW |
2741 | |
2742 | m->hdr.type = cpu_to_le16(type); | |
45c6ceb5 | 2743 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
31b8006e | 2744 | m->hdr.front_len = cpu_to_le32(front_len); |
ca20892d | 2745 | |
9516e45b AE |
2746 | INIT_LIST_HEAD(&m->list_head); |
2747 | kref_init(&m->kref); | |
ca20892d | 2748 | |
31b8006e | 2749 | /* front */ |
9516e45b | 2750 | m->front_max = front_len; |
31b8006e SW |
2751 | if (front_len) { |
2752 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2753 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2754 | PAGE_KERNEL); |
2755 | m->front_is_vmalloc = true; | |
2756 | } else { | |
34d23762 | 2757 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2758 | } |
2759 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2760 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2761 | front_len); |
2762 | goto out2; | |
2763 | } | |
2764 | } else { | |
2765 | m->front.iov_base = NULL; | |
2766 | } | |
2767 | m->front.iov_len = front_len; | |
2768 | ||
bb257664 | 2769 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2770 | return m; |
2771 | ||
2772 | out2: | |
2773 | ceph_msg_put(m); | |
2774 | out: | |
b61c2763 SW |
2775 | if (!can_fail) { |
2776 | pr_err("msg_new can't create type %d front %d\n", type, | |
2777 | front_len); | |
f0ed1b7c | 2778 | WARN_ON(1); |
b61c2763 SW |
2779 | } else { |
2780 | dout("msg_new can't create type %d front %d\n", type, | |
2781 | front_len); | |
2782 | } | |
a79832f2 | 2783 | return NULL; |
31b8006e | 2784 | } |
3d14c5d2 | 2785 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2786 | |
31b8006e SW |
2787 | /* |
2788 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2789 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2790 | * per-type header in the front and then gracefully fail (i.e., | |
2791 | * propagate the error to the caller based on info in the front) when | |
2792 | * the middle is too large. | |
2793 | */ | |
2450418c | 2794 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2795 | { |
2796 | int type = le16_to_cpu(msg->hdr.type); | |
2797 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2798 | ||
2799 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2800 | ceph_msg_type_name(type), middle_len); | |
2801 | BUG_ON(!middle_len); | |
2802 | BUG_ON(msg->middle); | |
2803 | ||
b6c1d5b8 | 2804 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2805 | if (!msg->middle) |
2806 | return -ENOMEM; | |
2807 | return 0; | |
2808 | } | |
2809 | ||
2450418c | 2810 | /* |
1c20f2d2 AE |
2811 | * Allocate a message for receiving an incoming message on a |
2812 | * connection, and save the result in con->in_msg. Uses the | |
2813 | * connection's private alloc_msg op if available. | |
2814 | * | |
4740a623 SW |
2815 | * Returns 0 on success, or a negative error code. |
2816 | * | |
2817 | * On success, if we set *skip = 1: | |
2818 | * - the next message should be skipped and ignored. | |
2819 | * - con->in_msg == NULL | |
2820 | * or if we set *skip = 0: | |
2821 | * - con->in_msg is non-null. | |
2822 | * On error (ENOMEM, EAGAIN, ...), | |
2823 | * - con->in_msg == NULL | |
2450418c | 2824 | */ |
4740a623 | 2825 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
2450418c | 2826 | { |
4740a623 | 2827 | struct ceph_msg_header *hdr = &con->in_hdr; |
2450418c | 2828 | int middle_len = le32_to_cpu(hdr->middle_len); |
1d866d1c | 2829 | struct ceph_msg *msg; |
4740a623 | 2830 | int ret = 0; |
2450418c | 2831 | |
1c20f2d2 | 2832 | BUG_ON(con->in_msg != NULL); |
53ded495 | 2833 | BUG_ON(!con->ops->alloc_msg); |
2450418c | 2834 | |
53ded495 AE |
2835 | mutex_unlock(&con->mutex); |
2836 | msg = con->ops->alloc_msg(con, hdr, skip); | |
2837 | mutex_lock(&con->mutex); | |
2838 | if (con->state != CON_STATE_OPEN) { | |
2839 | if (msg) | |
1d866d1c | 2840 | ceph_msg_put(msg); |
53ded495 AE |
2841 | return -EAGAIN; |
2842 | } | |
4137577a AE |
2843 | if (msg) { |
2844 | BUG_ON(*skip); | |
2845 | con->in_msg = msg; | |
36eb71aa | 2846 | con->in_msg->con = con->ops->get(con); |
92ce034b | 2847 | BUG_ON(con->in_msg->con == NULL); |
4137577a AE |
2848 | } else { |
2849 | /* | |
2850 | * Null message pointer means either we should skip | |
2851 | * this message or we couldn't allocate memory. The | |
2852 | * former is not an error. | |
2853 | */ | |
2854 | if (*skip) | |
2855 | return 0; | |
2856 | con->error_msg = "error allocating memory for incoming message"; | |
2857 | ||
53ded495 | 2858 | return -ENOMEM; |
2450418c | 2859 | } |
1c20f2d2 | 2860 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2861 | |
1c20f2d2 AE |
2862 | if (middle_len && !con->in_msg->middle) { |
2863 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 2864 | if (ret < 0) { |
1c20f2d2 AE |
2865 | ceph_msg_put(con->in_msg); |
2866 | con->in_msg = NULL; | |
2450418c YS |
2867 | } |
2868 | } | |
9d7f0f13 | 2869 | |
4740a623 | 2870 | return ret; |
2450418c YS |
2871 | } |
2872 | ||
31b8006e SW |
2873 | |
2874 | /* | |
2875 | * Free a generically kmalloc'd message. | |
2876 | */ | |
2877 | void ceph_msg_kfree(struct ceph_msg *m) | |
2878 | { | |
2879 | dout("msg_kfree %p\n", m); | |
2880 | if (m->front_is_vmalloc) | |
2881 | vfree(m->front.iov_base); | |
2882 | else | |
2883 | kfree(m->front.iov_base); | |
2884 | kfree(m); | |
2885 | } | |
2886 | ||
2887 | /* | |
2888 | * Drop a msg ref. Destroy as needed. | |
2889 | */ | |
c2e552e7 SW |
2890 | void ceph_msg_last_put(struct kref *kref) |
2891 | { | |
2892 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2893 | |
c2e552e7 SW |
2894 | dout("ceph_msg_put last one on %p\n", m); |
2895 | WARN_ON(!list_empty(&m->list_head)); | |
2896 | ||
2897 | /* drop middle, data, if any */ | |
2898 | if (m->middle) { | |
2899 | ceph_buffer_put(m->middle); | |
2900 | m->middle = NULL; | |
31b8006e | 2901 | } |
4a73ef27 | 2902 | m->length = 0; |
c2e552e7 SW |
2903 | m->pages = NULL; |
2904 | ||
58bb3b37 SW |
2905 | if (m->pagelist) { |
2906 | ceph_pagelist_release(m->pagelist); | |
2907 | kfree(m->pagelist); | |
2908 | m->pagelist = NULL; | |
2909 | } | |
2910 | ||
68b4476b YS |
2911 | m->trail = NULL; |
2912 | ||
c2e552e7 SW |
2913 | if (m->pool) |
2914 | ceph_msgpool_put(m->pool, m); | |
2915 | else | |
2916 | ceph_msg_kfree(m); | |
31b8006e | 2917 | } |
3d14c5d2 | 2918 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2919 | |
2920 | void ceph_msg_dump(struct ceph_msg *msg) | |
2921 | { | |
4a73ef27 AE |
2922 | pr_debug("msg_dump %p (front_max %d length %zd)\n", msg, |
2923 | msg->front_max, msg->length); | |
9ec7cab1 SW |
2924 | print_hex_dump(KERN_DEBUG, "header: ", |
2925 | DUMP_PREFIX_OFFSET, 16, 1, | |
2926 | &msg->hdr, sizeof(msg->hdr), true); | |
2927 | print_hex_dump(KERN_DEBUG, " front: ", | |
2928 | DUMP_PREFIX_OFFSET, 16, 1, | |
2929 | msg->front.iov_base, msg->front.iov_len, true); | |
2930 | if (msg->middle) | |
2931 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2932 | DUMP_PREFIX_OFFSET, 16, 1, | |
2933 | msg->middle->vec.iov_base, | |
2934 | msg->middle->vec.iov_len, true); | |
2935 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2936 | DUMP_PREFIX_OFFSET, 16, 1, | |
2937 | &msg->footer, sizeof(msg->footer), true); | |
2938 | } | |
3d14c5d2 | 2939 | EXPORT_SYMBOL(ceph_msg_dump); |