]>
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 | ||
2b3e0c90 | 18 | #include <linux/ceph/ceph_features.h> |
3d14c5d2 YS |
19 | #include <linux/ceph/libceph.h> |
20 | #include <linux/ceph/messenger.h> | |
21 | #include <linux/ceph/decode.h> | |
22 | #include <linux/ceph/pagelist.h> | |
bc3b2d7f | 23 | #include <linux/export.h> |
31b8006e | 24 | |
fe38a2b6 AE |
25 | #define list_entry_next(pos, member) \ |
26 | list_entry(pos->member.next, typeof(*pos), member) | |
27 | ||
31b8006e SW |
28 | /* |
29 | * Ceph uses the messenger to exchange ceph_msg messages with other | |
30 | * hosts in the system. The messenger provides ordered and reliable | |
31 | * delivery. We tolerate TCP disconnects by reconnecting (with | |
32 | * exponential backoff) in the case of a fault (disconnection, bad | |
33 | * crc, protocol error). Acks allow sent messages to be discarded by | |
34 | * the sender. | |
35 | */ | |
36 | ||
bc18f4b1 AE |
37 | /* |
38 | * We track the state of the socket on a given connection using | |
39 | * values defined below. The transition to a new socket state is | |
40 | * handled by a function which verifies we aren't coming from an | |
41 | * unexpected state. | |
42 | * | |
43 | * -------- | |
44 | * | NEW* | transient initial state | |
45 | * -------- | |
46 | * | con_sock_state_init() | |
47 | * v | |
48 | * ---------- | |
49 | * | CLOSED | initialized, but no socket (and no | |
50 | * ---------- TCP connection) | |
51 | * ^ \ | |
52 | * | \ con_sock_state_connecting() | |
53 | * | ---------------------- | |
54 | * | \ | |
55 | * + con_sock_state_closed() \ | |
fbb85a47 SW |
56 | * |+--------------------------- \ |
57 | * | \ \ \ | |
58 | * | ----------- \ \ | |
59 | * | | CLOSING | socket event; \ \ | |
60 | * | ----------- await close \ \ | |
61 | * | ^ \ | | |
62 | * | | \ | | |
63 | * | + con_sock_state_closing() \ | | |
64 | * | / \ | | | |
65 | * | / --------------- | | | |
66 | * | / \ v v | |
bc18f4b1 AE |
67 | * | / -------------- |
68 | * | / -----------------| CONNECTING | socket created, TCP | |
69 | * | | / -------------- connect initiated | |
70 | * | | | con_sock_state_connected() | |
71 | * | | v | |
72 | * ------------- | |
73 | * | CONNECTED | TCP connection established | |
74 | * ------------- | |
75 | * | |
76 | * State values for ceph_connection->sock_state; NEW is assumed to be 0. | |
77 | */ | |
ce2c8903 AE |
78 | |
79 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ | |
80 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ | |
81 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ | |
82 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ | |
83 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ | |
84 | ||
8dacc7da SW |
85 | /* |
86 | * connection states | |
87 | */ | |
88 | #define CON_STATE_CLOSED 1 /* -> PREOPEN */ | |
89 | #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */ | |
90 | #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */ | |
91 | #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */ | |
92 | #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */ | |
93 | #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */ | |
94 | ||
4a861692 SW |
95 | /* |
96 | * ceph_connection flag bits | |
97 | */ | |
98 | #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop | |
99 | * messages on errors */ | |
100 | #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */ | |
101 | #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */ | |
102 | #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */ | |
103 | #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */ | |
8dacc7da | 104 | |
c9ffc77a AE |
105 | static bool con_flag_valid(unsigned long con_flag) |
106 | { | |
107 | switch (con_flag) { | |
108 | case CON_FLAG_LOSSYTX: | |
109 | case CON_FLAG_KEEPALIVE_PENDING: | |
110 | case CON_FLAG_WRITE_PENDING: | |
111 | case CON_FLAG_SOCK_CLOSED: | |
112 | case CON_FLAG_BACKOFF: | |
113 | return true; | |
114 | default: | |
115 | return false; | |
116 | } | |
117 | } | |
118 | ||
119 | static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag) | |
120 | { | |
121 | BUG_ON(!con_flag_valid(con_flag)); | |
122 | ||
123 | clear_bit(con_flag, &con->flags); | |
124 | } | |
125 | ||
126 | static void con_flag_set(struct ceph_connection *con, unsigned long con_flag) | |
127 | { | |
128 | BUG_ON(!con_flag_valid(con_flag)); | |
129 | ||
130 | set_bit(con_flag, &con->flags); | |
131 | } | |
132 | ||
133 | static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag) | |
134 | { | |
135 | BUG_ON(!con_flag_valid(con_flag)); | |
136 | ||
137 | return test_bit(con_flag, &con->flags); | |
138 | } | |
139 | ||
140 | static bool con_flag_test_and_clear(struct ceph_connection *con, | |
141 | unsigned long con_flag) | |
142 | { | |
143 | BUG_ON(!con_flag_valid(con_flag)); | |
144 | ||
145 | return test_and_clear_bit(con_flag, &con->flags); | |
146 | } | |
147 | ||
148 | static bool con_flag_test_and_set(struct ceph_connection *con, | |
149 | unsigned long con_flag) | |
150 | { | |
151 | BUG_ON(!con_flag_valid(con_flag)); | |
152 | ||
153 | return test_and_set_bit(con_flag, &con->flags); | |
154 | } | |
155 | ||
e3d5d638 AE |
156 | /* Slab caches for frequently-allocated structures */ |
157 | ||
158 | static struct kmem_cache *ceph_msg_cache; | |
81b36be4 | 159 | static struct kmem_cache *ceph_msg_data_cache; |
e3d5d638 | 160 | |
31b8006e SW |
161 | /* static tag bytes (protocol control messages) */ |
162 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
163 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
164 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
165 | ||
a6a5349d SW |
166 | #ifdef CONFIG_LOCKDEP |
167 | static struct lock_class_key socket_class; | |
168 | #endif | |
169 | ||
84495f49 AE |
170 | /* |
171 | * When skipping (ignoring) a block of input we read it into a "skip | |
172 | * buffer," which is this many bytes in size. | |
173 | */ | |
174 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
175 | |
176 | static void queue_con(struct ceph_connection *con); | |
37ab77ac | 177 | static void cancel_con(struct ceph_connection *con); |
31b8006e | 178 | static void con_work(struct work_struct *); |
93209264 | 179 | static void con_fault(struct ceph_connection *con); |
31b8006e | 180 | |
31b8006e | 181 | /* |
f64a9317 AE |
182 | * Nicely render a sockaddr as a string. An array of formatted |
183 | * strings is used, to approximate reentrancy. | |
31b8006e | 184 | */ |
f64a9317 AE |
185 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
186 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
187 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
188 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
189 | ||
190 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
191 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 192 | |
57666519 | 193 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 194 | |
3d14c5d2 | 195 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
196 | { |
197 | int i; | |
198 | char *s; | |
99f0f3b2 AE |
199 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
200 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 201 | |
f64a9317 | 202 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
203 | s = addr_str[i]; |
204 | ||
205 | switch (ss->ss_family) { | |
206 | case AF_INET: | |
bd406145 AE |
207 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
208 | ntohs(in4->sin_port)); | |
31b8006e SW |
209 | break; |
210 | ||
211 | case AF_INET6: | |
bd406145 AE |
212 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
213 | ntohs(in6->sin6_port)); | |
31b8006e SW |
214 | break; |
215 | ||
216 | default: | |
d3002b97 AE |
217 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
218 | ss->ss_family); | |
31b8006e SW |
219 | } |
220 | ||
221 | return s; | |
222 | } | |
3d14c5d2 | 223 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 224 | |
63f2d211 SW |
225 | static void encode_my_addr(struct ceph_messenger *msgr) |
226 | { | |
227 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
228 | ceph_encode_addr(&msgr->my_enc_addr); | |
229 | } | |
230 | ||
31b8006e SW |
231 | /* |
232 | * work queue for all reading and writing to/from the socket. | |
233 | */ | |
e0f43c94 | 234 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 235 | |
e3d5d638 AE |
236 | static int ceph_msgr_slab_init(void) |
237 | { | |
238 | BUG_ON(ceph_msg_cache); | |
239 | ceph_msg_cache = kmem_cache_create("ceph_msg", | |
240 | sizeof (struct ceph_msg), | |
241 | __alignof__(struct ceph_msg), 0, NULL); | |
81b36be4 AE |
242 | |
243 | if (!ceph_msg_cache) | |
244 | return -ENOMEM; | |
245 | ||
246 | BUG_ON(ceph_msg_data_cache); | |
247 | ceph_msg_data_cache = kmem_cache_create("ceph_msg_data", | |
248 | sizeof (struct ceph_msg_data), | |
249 | __alignof__(struct ceph_msg_data), | |
250 | 0, NULL); | |
251 | if (ceph_msg_data_cache) | |
252 | return 0; | |
253 | ||
254 | kmem_cache_destroy(ceph_msg_cache); | |
255 | ceph_msg_cache = NULL; | |
256 | ||
257 | return -ENOMEM; | |
e3d5d638 AE |
258 | } |
259 | ||
260 | static void ceph_msgr_slab_exit(void) | |
261 | { | |
81b36be4 AE |
262 | BUG_ON(!ceph_msg_data_cache); |
263 | kmem_cache_destroy(ceph_msg_data_cache); | |
264 | ceph_msg_data_cache = NULL; | |
265 | ||
e3d5d638 AE |
266 | BUG_ON(!ceph_msg_cache); |
267 | kmem_cache_destroy(ceph_msg_cache); | |
268 | ceph_msg_cache = NULL; | |
269 | } | |
270 | ||
15417167 | 271 | static void _ceph_msgr_exit(void) |
6173d1f0 | 272 | { |
d3002b97 | 273 | if (ceph_msgr_wq) { |
6173d1f0 | 274 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
275 | ceph_msgr_wq = NULL; |
276 | } | |
6173d1f0 | 277 | |
e3d5d638 AE |
278 | ceph_msgr_slab_exit(); |
279 | ||
6173d1f0 AE |
280 | BUG_ON(zero_page == NULL); |
281 | kunmap(zero_page); | |
282 | page_cache_release(zero_page); | |
283 | zero_page = NULL; | |
284 | } | |
285 | ||
3d14c5d2 | 286 | int ceph_msgr_init(void) |
31b8006e | 287 | { |
57666519 AE |
288 | BUG_ON(zero_page != NULL); |
289 | zero_page = ZERO_PAGE(0); | |
290 | page_cache_get(zero_page); | |
291 | ||
e3d5d638 AE |
292 | if (ceph_msgr_slab_init()) |
293 | return -ENOMEM; | |
294 | ||
f9865f06 ID |
295 | /* |
296 | * The number of active work items is limited by the number of | |
297 | * connections, so leave @max_active at default. | |
298 | */ | |
299 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0); | |
6173d1f0 AE |
300 | if (ceph_msgr_wq) |
301 | return 0; | |
57666519 | 302 | |
6173d1f0 AE |
303 | pr_err("msgr_init failed to create workqueue\n"); |
304 | _ceph_msgr_exit(); | |
57666519 | 305 | |
6173d1f0 | 306 | return -ENOMEM; |
31b8006e | 307 | } |
3d14c5d2 | 308 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
309 | |
310 | void ceph_msgr_exit(void) | |
311 | { | |
57666519 | 312 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 313 | |
6173d1f0 | 314 | _ceph_msgr_exit(); |
31b8006e | 315 | } |
3d14c5d2 | 316 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 317 | |
cd84db6e | 318 | void ceph_msgr_flush(void) |
a922d38f SW |
319 | { |
320 | flush_workqueue(ceph_msgr_wq); | |
321 | } | |
3d14c5d2 | 322 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 323 | |
ce2c8903 AE |
324 | /* Connection socket state transition functions */ |
325 | ||
326 | static void con_sock_state_init(struct ceph_connection *con) | |
327 | { | |
328 | int old_state; | |
329 | ||
330 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
331 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
332 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
333 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
334 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 AE |
335 | } |
336 | ||
337 | static void con_sock_state_connecting(struct ceph_connection *con) | |
338 | { | |
339 | int old_state; | |
340 | ||
341 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
342 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
343 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
344 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
345 | CON_SOCK_STATE_CONNECTING); | |
ce2c8903 AE |
346 | } |
347 | ||
348 | static void con_sock_state_connected(struct ceph_connection *con) | |
349 | { | |
350 | int old_state; | |
351 | ||
352 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
353 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
354 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
355 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
356 | CON_SOCK_STATE_CONNECTED); | |
ce2c8903 AE |
357 | } |
358 | ||
359 | static void con_sock_state_closing(struct ceph_connection *con) | |
360 | { | |
361 | int old_state; | |
362 | ||
363 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
364 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
365 | old_state != CON_SOCK_STATE_CONNECTED && | |
366 | old_state != CON_SOCK_STATE_CLOSING)) | |
367 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
368 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
369 | CON_SOCK_STATE_CLOSING); | |
ce2c8903 AE |
370 | } |
371 | ||
372 | static void con_sock_state_closed(struct ceph_connection *con) | |
373 | { | |
374 | int old_state; | |
375 | ||
376 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
377 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 | 378 | old_state != CON_SOCK_STATE_CLOSING && |
8007b8d6 SW |
379 | old_state != CON_SOCK_STATE_CONNECTING && |
380 | old_state != CON_SOCK_STATE_CLOSED)) | |
ce2c8903 | 381 | printk("%s: unexpected old state %d\n", __func__, old_state); |
8007b8d6 SW |
382 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
383 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 | 384 | } |
a922d38f | 385 | |
31b8006e SW |
386 | /* |
387 | * socket callback functions | |
388 | */ | |
389 | ||
390 | /* data available on socket, or listen socket received a connect */ | |
676d2369 | 391 | static void ceph_sock_data_ready(struct sock *sk) |
31b8006e | 392 | { |
bd406145 | 393 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
394 | if (atomic_read(&con->msgr->stopping)) { |
395 | return; | |
396 | } | |
bd406145 | 397 | |
31b8006e | 398 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 399 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
400 | con, con->state); |
401 | queue_con(con); | |
402 | } | |
403 | } | |
404 | ||
405 | /* socket has buffer space for writing */ | |
327800bd | 406 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 407 | { |
d3002b97 | 408 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 409 | |
182fac26 JS |
410 | /* only queue to workqueue if there is data we want to write, |
411 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 412 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
413 | * doesn't get called again until try_write() fills the socket |
414 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
415 | * and net/core/stream.c:sk_stream_write_space(). | |
416 | */ | |
c9ffc77a | 417 | if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) { |
64dc6130 | 418 | if (sk_stream_is_writeable(sk)) { |
327800bd | 419 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
420 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
421 | queue_con(con); | |
422 | } | |
31b8006e | 423 | } else { |
327800bd | 424 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 425 | } |
31b8006e SW |
426 | } |
427 | ||
428 | /* socket's state has changed */ | |
327800bd | 429 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 430 | { |
bd406145 | 431 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 432 | |
327800bd | 433 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
434 | con, con->state, sk->sk_state); |
435 | ||
31b8006e SW |
436 | switch (sk->sk_state) { |
437 | case TCP_CLOSE: | |
327800bd | 438 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 439 | case TCP_CLOSE_WAIT: |
327800bd | 440 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 441 | con_sock_state_closing(con); |
c9ffc77a | 442 | con_flag_set(con, CON_FLAG_SOCK_CLOSED); |
d65c9e0b | 443 | queue_con(con); |
31b8006e SW |
444 | break; |
445 | case TCP_ESTABLISHED: | |
327800bd | 446 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 447 | con_sock_state_connected(con); |
31b8006e SW |
448 | queue_con(con); |
449 | break; | |
d3002b97 AE |
450 | default: /* Everything else is uninteresting */ |
451 | break; | |
31b8006e SW |
452 | } |
453 | } | |
454 | ||
455 | /* | |
456 | * set up socket callbacks | |
457 | */ | |
458 | static void set_sock_callbacks(struct socket *sock, | |
459 | struct ceph_connection *con) | |
460 | { | |
461 | struct sock *sk = sock->sk; | |
bd406145 | 462 | sk->sk_user_data = con; |
327800bd AE |
463 | sk->sk_data_ready = ceph_sock_data_ready; |
464 | sk->sk_write_space = ceph_sock_write_space; | |
465 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
466 | } |
467 | ||
468 | ||
469 | /* | |
470 | * socket helpers | |
471 | */ | |
472 | ||
473 | /* | |
474 | * initiate connection to a remote socket. | |
475 | */ | |
41617d0c | 476 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 477 | { |
f91d3471 | 478 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
479 | struct socket *sock; |
480 | int ret; | |
481 | ||
482 | BUG_ON(con->sock); | |
f91d3471 SW |
483 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
484 | IPPROTO_TCP, &sock); | |
31b8006e | 485 | if (ret) |
41617d0c | 486 | return ret; |
89baaa57 | 487 | sock->sk->sk_allocation = GFP_NOFS | __GFP_MEMALLOC; |
31b8006e | 488 | |
a6a5349d SW |
489 | #ifdef CONFIG_LOCKDEP |
490 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
491 | #endif | |
492 | ||
31b8006e SW |
493 | set_sock_callbacks(sock, con); |
494 | ||
3d14c5d2 | 495 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 496 | |
89a86be0 | 497 | con_sock_state_connecting(con); |
f91d3471 SW |
498 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
499 | O_NONBLOCK); | |
31b8006e SW |
500 | if (ret == -EINPROGRESS) { |
501 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 502 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 503 | sock->sk->sk_state); |
a5bc3129 | 504 | } else if (ret < 0) { |
31b8006e | 505 | pr_err("connect %s error %d\n", |
3d14c5d2 | 506 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 507 | sock_release(sock); |
31b8006e | 508 | con->error_msg = "connect error"; |
31b8006e | 509 | |
41617d0c | 510 | return ret; |
a5bc3129 | 511 | } |
89baaa57 MC |
512 | |
513 | sk_set_memalloc(sock->sk); | |
514 | ||
a5bc3129 | 515 | con->sock = sock; |
41617d0c | 516 | return 0; |
31b8006e SW |
517 | } |
518 | ||
519 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
520 | { | |
521 | struct kvec iov = {buf, len}; | |
522 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 523 | int r; |
31b8006e | 524 | |
98bdb0aa SW |
525 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
526 | if (r == -EAGAIN) | |
527 | r = 0; | |
528 | return r; | |
31b8006e SW |
529 | } |
530 | ||
afb3d90e AE |
531 | static int ceph_tcp_recvpage(struct socket *sock, struct page *page, |
532 | int page_offset, size_t length) | |
533 | { | |
534 | void *kaddr; | |
535 | int ret; | |
536 | ||
537 | BUG_ON(page_offset + length > PAGE_SIZE); | |
538 | ||
539 | kaddr = kmap(page); | |
540 | BUG_ON(!kaddr); | |
541 | ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length); | |
542 | kunmap(page); | |
543 | ||
544 | return ret; | |
545 | } | |
546 | ||
31b8006e SW |
547 | /* |
548 | * write something. @more is true if caller will be sending more data | |
549 | * shortly. | |
550 | */ | |
551 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
552 | size_t kvlen, size_t len, int more) | |
553 | { | |
554 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 555 | int r; |
31b8006e SW |
556 | |
557 | if (more) | |
558 | msg.msg_flags |= MSG_MORE; | |
559 | else | |
560 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
561 | ||
42961d23 SW |
562 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
563 | if (r == -EAGAIN) | |
564 | r = 0; | |
565 | return r; | |
31b8006e SW |
566 | } |
567 | ||
178eda29 | 568 | static int __ceph_tcp_sendpage(struct socket *sock, struct page *page, |
e1dcb128 | 569 | int offset, size_t size, bool more) |
31739139 AE |
570 | { |
571 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
572 | int ret; | |
573 | ||
574 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
575 | if (ret == -EAGAIN) | |
576 | ret = 0; | |
577 | ||
578 | return ret; | |
579 | } | |
580 | ||
178eda29 CC |
581 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
582 | int offset, size_t size, bool more) | |
583 | { | |
584 | int ret; | |
585 | struct kvec iov; | |
586 | ||
587 | /* sendpage cannot properly handle pages with page_count == 0, | |
588 | * we need to fallback to sendmsg if that's the case */ | |
589 | if (page_count(page) >= 1) | |
590 | return __ceph_tcp_sendpage(sock, page, offset, size, more); | |
591 | ||
592 | iov.iov_base = kmap(page) + offset; | |
593 | iov.iov_len = size; | |
594 | ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more); | |
595 | kunmap(page); | |
596 | ||
597 | return ret; | |
598 | } | |
31b8006e SW |
599 | |
600 | /* | |
601 | * Shutdown/close the socket for the given connection. | |
602 | */ | |
603 | static int con_close_socket(struct ceph_connection *con) | |
604 | { | |
8007b8d6 | 605 | int rc = 0; |
31b8006e SW |
606 | |
607 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
8007b8d6 SW |
608 | if (con->sock) { |
609 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
610 | sock_release(con->sock); | |
611 | con->sock = NULL; | |
612 | } | |
456ea468 AE |
613 | |
614 | /* | |
4a861692 | 615 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
616 | * independent of the connection mutex, and we could have |
617 | * received a socket close event before we had the chance to | |
618 | * shut the socket down. | |
619 | */ | |
c9ffc77a | 620 | con_flag_clear(con, CON_FLAG_SOCK_CLOSED); |
8007b8d6 | 621 | |
ce2c8903 | 622 | con_sock_state_closed(con); |
31b8006e SW |
623 | return rc; |
624 | } | |
625 | ||
626 | /* | |
627 | * Reset a connection. Discard all incoming and outgoing messages | |
628 | * and clear *_seq state. | |
629 | */ | |
630 | static void ceph_msg_remove(struct ceph_msg *msg) | |
631 | { | |
632 | list_del_init(&msg->list_head); | |
38941f80 | 633 | BUG_ON(msg->con == NULL); |
36eb71aa | 634 | msg->con->ops->put(msg->con); |
38941f80 AE |
635 | msg->con = NULL; |
636 | ||
31b8006e SW |
637 | ceph_msg_put(msg); |
638 | } | |
639 | static void ceph_msg_remove_list(struct list_head *head) | |
640 | { | |
641 | while (!list_empty(head)) { | |
642 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
643 | list_head); | |
644 | ceph_msg_remove(msg); | |
645 | } | |
646 | } | |
647 | ||
648 | static void reset_connection(struct ceph_connection *con) | |
649 | { | |
650 | /* reset connection, out_queue, msg_ and connect_seq */ | |
651 | /* discard existing out_queue and msg_seq */ | |
0fa6ebc6 | 652 | dout("reset_connection %p\n", con); |
31b8006e SW |
653 | ceph_msg_remove_list(&con->out_queue); |
654 | ceph_msg_remove_list(&con->out_sent); | |
655 | ||
cf3e5c40 | 656 | if (con->in_msg) { |
38941f80 AE |
657 | BUG_ON(con->in_msg->con != con); |
658 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
659 | ceph_msg_put(con->in_msg); |
660 | con->in_msg = NULL; | |
36eb71aa | 661 | con->ops->put(con); |
cf3e5c40 SW |
662 | } |
663 | ||
31b8006e SW |
664 | con->connect_seq = 0; |
665 | con->out_seq = 0; | |
c86a2930 SW |
666 | if (con->out_msg) { |
667 | ceph_msg_put(con->out_msg); | |
668 | con->out_msg = NULL; | |
669 | } | |
31b8006e | 670 | con->in_seq = 0; |
0e0d5e0c | 671 | con->in_seq_acked = 0; |
31b8006e SW |
672 | } |
673 | ||
674 | /* | |
675 | * mark a peer down. drop any open connections. | |
676 | */ | |
677 | void ceph_con_close(struct ceph_connection *con) | |
678 | { | |
8c50c817 | 679 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
680 | dout("con_close %p peer %s\n", con, |
681 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
8dacc7da | 682 | con->state = CON_STATE_CLOSED; |
a5988c49 | 683 | |
c9ffc77a AE |
684 | con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */ |
685 | con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING); | |
686 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); | |
687 | con_flag_clear(con, CON_FLAG_BACKOFF); | |
a5988c49 | 688 | |
31b8006e | 689 | reset_connection(con); |
6f2bc3ff | 690 | con->peer_global_seq = 0; |
37ab77ac | 691 | cancel_con(con); |
ee76e073 | 692 | con_close_socket(con); |
ec302645 | 693 | mutex_unlock(&con->mutex); |
31b8006e | 694 | } |
3d14c5d2 | 695 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 696 | |
31b8006e SW |
697 | /* |
698 | * Reopen a closed connection, with a new peer address. | |
699 | */ | |
b7a9e5dd SW |
700 | void ceph_con_open(struct ceph_connection *con, |
701 | __u8 entity_type, __u64 entity_num, | |
702 | struct ceph_entity_addr *addr) | |
31b8006e | 703 | { |
5469155f | 704 | mutex_lock(&con->mutex); |
3d14c5d2 | 705 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
8dacc7da | 706 | |
122070a2 | 707 | WARN_ON(con->state != CON_STATE_CLOSED); |
8dacc7da | 708 | con->state = CON_STATE_PREOPEN; |
a5988c49 | 709 | |
b7a9e5dd SW |
710 | con->peer_name.type = (__u8) entity_type; |
711 | con->peer_name.num = cpu_to_le64(entity_num); | |
712 | ||
31b8006e | 713 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 714 | con->delay = 0; /* reset backoff memory */ |
5469155f | 715 | mutex_unlock(&con->mutex); |
31b8006e SW |
716 | queue_con(con); |
717 | } | |
3d14c5d2 | 718 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 719 | |
87b315a5 SW |
720 | /* |
721 | * return true if this connection ever successfully opened | |
722 | */ | |
723 | bool ceph_con_opened(struct ceph_connection *con) | |
724 | { | |
725 | return con->connect_seq > 0; | |
726 | } | |
727 | ||
31b8006e SW |
728 | /* |
729 | * initialize a new connection. | |
730 | */ | |
1bfd89f4 AE |
731 | void ceph_con_init(struct ceph_connection *con, void *private, |
732 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 733 | struct ceph_messenger *msgr) |
31b8006e SW |
734 | { |
735 | dout("con_init %p\n", con); | |
736 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
737 | con->private = private; |
738 | con->ops = ops; | |
31b8006e | 739 | con->msgr = msgr; |
ce2c8903 AE |
740 | |
741 | con_sock_state_init(con); | |
742 | ||
ec302645 | 743 | mutex_init(&con->mutex); |
31b8006e SW |
744 | INIT_LIST_HEAD(&con->out_queue); |
745 | INIT_LIST_HEAD(&con->out_sent); | |
746 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 | 747 | |
8dacc7da | 748 | con->state = CON_STATE_CLOSED; |
31b8006e | 749 | } |
3d14c5d2 | 750 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
751 | |
752 | ||
753 | /* | |
754 | * We maintain a global counter to order connection attempts. Get | |
755 | * a unique seq greater than @gt. | |
756 | */ | |
757 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
758 | { | |
759 | u32 ret; | |
760 | ||
761 | spin_lock(&msgr->global_seq_lock); | |
762 | if (msgr->global_seq < gt) | |
763 | msgr->global_seq = gt; | |
764 | ret = ++msgr->global_seq; | |
765 | spin_unlock(&msgr->global_seq_lock); | |
766 | return ret; | |
767 | } | |
768 | ||
e2200423 | 769 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
770 | { |
771 | con->out_kvec_left = 0; | |
772 | con->out_kvec_bytes = 0; | |
773 | con->out_kvec_cur = &con->out_kvec[0]; | |
774 | } | |
775 | ||
e2200423 | 776 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
777 | size_t size, void *data) |
778 | { | |
779 | int index; | |
780 | ||
781 | index = con->out_kvec_left; | |
782 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
783 | ||
784 | con->out_kvec[index].iov_len = size; | |
785 | con->out_kvec[index].iov_base = data; | |
786 | con->out_kvec_left++; | |
787 | con->out_kvec_bytes += size; | |
788 | } | |
31b8006e | 789 | |
df6ad1f9 | 790 | #ifdef CONFIG_BLOCK |
6aaa4511 AE |
791 | |
792 | /* | |
793 | * For a bio data item, a piece is whatever remains of the next | |
794 | * entry in the current bio iovec, or the first entry in the next | |
795 | * bio in the list. | |
796 | */ | |
8ae4f4f5 | 797 | static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 798 | size_t length) |
6aaa4511 | 799 | { |
8ae4f4f5 | 800 | struct ceph_msg_data *data = cursor->data; |
6aaa4511 AE |
801 | struct bio *bio; |
802 | ||
803 | BUG_ON(data->type != CEPH_MSG_DATA_BIO); | |
804 | ||
805 | bio = data->bio; | |
806 | BUG_ON(!bio); | |
6aaa4511 | 807 | |
ca8b3a69 | 808 | cursor->resid = min(length, data->bio_length); |
6aaa4511 | 809 | cursor->bio = bio; |
f38a5181 KO |
810 | cursor->bvec_iter = bio->bi_iter; |
811 | cursor->last_piece = | |
812 | cursor->resid <= bio_iter_len(bio, cursor->bvec_iter); | |
6aaa4511 AE |
813 | } |
814 | ||
8ae4f4f5 | 815 | static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor, |
6aaa4511 AE |
816 | size_t *page_offset, |
817 | size_t *length) | |
818 | { | |
8ae4f4f5 | 819 | struct ceph_msg_data *data = cursor->data; |
6aaa4511 | 820 | struct bio *bio; |
f38a5181 | 821 | struct bio_vec bio_vec; |
6aaa4511 AE |
822 | |
823 | BUG_ON(data->type != CEPH_MSG_DATA_BIO); | |
824 | ||
825 | bio = cursor->bio; | |
826 | BUG_ON(!bio); | |
827 | ||
f38a5181 | 828 | bio_vec = bio_iter_iovec(bio, cursor->bvec_iter); |
6aaa4511 | 829 | |
f38a5181 | 830 | *page_offset = (size_t) bio_vec.bv_offset; |
6aaa4511 | 831 | BUG_ON(*page_offset >= PAGE_SIZE); |
25aff7c5 AE |
832 | if (cursor->last_piece) /* pagelist offset is always 0 */ |
833 | *length = cursor->resid; | |
834 | else | |
f38a5181 | 835 | *length = (size_t) bio_vec.bv_len; |
25aff7c5 | 836 | BUG_ON(*length > cursor->resid); |
5df521b1 | 837 | BUG_ON(*page_offset + *length > PAGE_SIZE); |
6aaa4511 | 838 | |
f38a5181 | 839 | return bio_vec.bv_page; |
6aaa4511 AE |
840 | } |
841 | ||
8ae4f4f5 AE |
842 | static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor, |
843 | size_t bytes) | |
6aaa4511 | 844 | { |
6aaa4511 | 845 | struct bio *bio; |
f38a5181 | 846 | struct bio_vec bio_vec; |
6aaa4511 | 847 | |
8ae4f4f5 | 848 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO); |
6aaa4511 AE |
849 | |
850 | bio = cursor->bio; | |
851 | BUG_ON(!bio); | |
852 | ||
f38a5181 | 853 | bio_vec = bio_iter_iovec(bio, cursor->bvec_iter); |
6aaa4511 AE |
854 | |
855 | /* Advance the cursor offset */ | |
856 | ||
25aff7c5 AE |
857 | BUG_ON(cursor->resid < bytes); |
858 | cursor->resid -= bytes; | |
f38a5181 KO |
859 | |
860 | bio_advance_iter(bio, &cursor->bvec_iter, bytes); | |
861 | ||
862 | if (bytes < bio_vec.bv_len) | |
6aaa4511 AE |
863 | return false; /* more bytes to process in this segment */ |
864 | ||
865 | /* Move on to the next segment, and possibly the next bio */ | |
866 | ||
f38a5181 | 867 | if (!cursor->bvec_iter.bi_size) { |
6aaa4511 | 868 | bio = bio->bi_next; |
0ec1d15e ID |
869 | cursor->bio = bio; |
870 | if (bio) | |
871 | cursor->bvec_iter = bio->bi_iter; | |
872 | else | |
873 | memset(&cursor->bvec_iter, 0, | |
874 | sizeof(cursor->bvec_iter)); | |
6aaa4511 | 875 | } |
6aaa4511 | 876 | |
25aff7c5 AE |
877 | if (!cursor->last_piece) { |
878 | BUG_ON(!cursor->resid); | |
879 | BUG_ON(!bio); | |
880 | /* A short read is OK, so use <= rather than == */ | |
f38a5181 | 881 | if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter)) |
6aaa4511 | 882 | cursor->last_piece = true; |
25aff7c5 | 883 | } |
6aaa4511 AE |
884 | |
885 | return true; | |
886 | } | |
ea96571f | 887 | #endif /* CONFIG_BLOCK */ |
df6ad1f9 | 888 | |
e766d7b5 AE |
889 | /* |
890 | * For a page array, a piece comes from the first page in the array | |
891 | * that has not already been fully consumed. | |
892 | */ | |
8ae4f4f5 | 893 | static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 894 | size_t length) |
e766d7b5 | 895 | { |
8ae4f4f5 | 896 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
897 | int page_count; |
898 | ||
899 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
900 | ||
901 | BUG_ON(!data->pages); | |
902 | BUG_ON(!data->length); | |
903 | ||
ca8b3a69 | 904 | cursor->resid = min(length, data->length); |
e766d7b5 | 905 | page_count = calc_pages_for(data->alignment, (u64)data->length); |
e766d7b5 AE |
906 | cursor->page_offset = data->alignment & ~PAGE_MASK; |
907 | cursor->page_index = 0; | |
56fc5659 AE |
908 | BUG_ON(page_count > (int)USHRT_MAX); |
909 | cursor->page_count = (unsigned short)page_count; | |
910 | BUG_ON(length > SIZE_MAX - cursor->page_offset); | |
5f740d7e | 911 | cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
912 | } |
913 | ||
8ae4f4f5 AE |
914 | static struct page * |
915 | ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor, | |
916 | size_t *page_offset, size_t *length) | |
e766d7b5 | 917 | { |
8ae4f4f5 | 918 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
919 | |
920 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
921 | ||
922 | BUG_ON(cursor->page_index >= cursor->page_count); | |
923 | BUG_ON(cursor->page_offset >= PAGE_SIZE); | |
e766d7b5 AE |
924 | |
925 | *page_offset = cursor->page_offset; | |
25aff7c5 | 926 | if (cursor->last_piece) |
e766d7b5 | 927 | *length = cursor->resid; |
25aff7c5 | 928 | else |
e766d7b5 | 929 | *length = PAGE_SIZE - *page_offset; |
e766d7b5 AE |
930 | |
931 | return data->pages[cursor->page_index]; | |
932 | } | |
933 | ||
8ae4f4f5 | 934 | static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor, |
e766d7b5 AE |
935 | size_t bytes) |
936 | { | |
8ae4f4f5 | 937 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES); |
e766d7b5 AE |
938 | |
939 | BUG_ON(cursor->page_offset + bytes > PAGE_SIZE); | |
e766d7b5 AE |
940 | |
941 | /* Advance the cursor page offset */ | |
942 | ||
943 | cursor->resid -= bytes; | |
5df521b1 AE |
944 | cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK; |
945 | if (!bytes || cursor->page_offset) | |
e766d7b5 AE |
946 | return false; /* more bytes to process in the current page */ |
947 | ||
d90deda6 YZ |
948 | if (!cursor->resid) |
949 | return false; /* no more data */ | |
950 | ||
5df521b1 | 951 | /* Move on to the next page; offset is already at 0 */ |
e766d7b5 AE |
952 | |
953 | BUG_ON(cursor->page_index >= cursor->page_count); | |
e766d7b5 | 954 | cursor->page_index++; |
25aff7c5 | 955 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
956 | |
957 | return true; | |
958 | } | |
959 | ||
fe38a2b6 | 960 | /* |
dd236fcb AE |
961 | * For a pagelist, a piece is whatever remains to be consumed in the |
962 | * first page in the list, or the front of the next page. | |
fe38a2b6 | 963 | */ |
8ae4f4f5 AE |
964 | static void |
965 | ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor, | |
25aff7c5 | 966 | size_t length) |
fe38a2b6 | 967 | { |
8ae4f4f5 | 968 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
969 | struct ceph_pagelist *pagelist; |
970 | struct page *page; | |
971 | ||
dd236fcb | 972 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); |
fe38a2b6 AE |
973 | |
974 | pagelist = data->pagelist; | |
975 | BUG_ON(!pagelist); | |
25aff7c5 AE |
976 | |
977 | if (!length) | |
fe38a2b6 AE |
978 | return; /* pagelist can be assigned but empty */ |
979 | ||
980 | BUG_ON(list_empty(&pagelist->head)); | |
981 | page = list_first_entry(&pagelist->head, struct page, lru); | |
982 | ||
ca8b3a69 | 983 | cursor->resid = min(length, pagelist->length); |
fe38a2b6 AE |
984 | cursor->page = page; |
985 | cursor->offset = 0; | |
a51b272e | 986 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
987 | } |
988 | ||
8ae4f4f5 AE |
989 | static struct page * |
990 | ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor, | |
991 | size_t *page_offset, size_t *length) | |
fe38a2b6 | 992 | { |
8ae4f4f5 | 993 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 | 994 | struct ceph_pagelist *pagelist; |
fe38a2b6 AE |
995 | |
996 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
997 | ||
998 | pagelist = data->pagelist; | |
999 | BUG_ON(!pagelist); | |
1000 | ||
1001 | BUG_ON(!cursor->page); | |
25aff7c5 | 1002 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); |
fe38a2b6 | 1003 | |
5df521b1 | 1004 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 | 1005 | *page_offset = cursor->offset & ~PAGE_MASK; |
5df521b1 | 1006 | if (cursor->last_piece) |
25aff7c5 AE |
1007 | *length = cursor->resid; |
1008 | else | |
1009 | *length = PAGE_SIZE - *page_offset; | |
fe38a2b6 | 1010 | |
8ae4f4f5 | 1011 | return cursor->page; |
fe38a2b6 AE |
1012 | } |
1013 | ||
8ae4f4f5 | 1014 | static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor, |
dd236fcb | 1015 | size_t bytes) |
fe38a2b6 | 1016 | { |
8ae4f4f5 | 1017 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
1018 | struct ceph_pagelist *pagelist; |
1019 | ||
1020 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
1021 | ||
1022 | pagelist = data->pagelist; | |
1023 | BUG_ON(!pagelist); | |
25aff7c5 AE |
1024 | |
1025 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); | |
fe38a2b6 AE |
1026 | BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE); |
1027 | ||
1028 | /* Advance the cursor offset */ | |
1029 | ||
25aff7c5 | 1030 | cursor->resid -= bytes; |
fe38a2b6 | 1031 | cursor->offset += bytes; |
5df521b1 | 1032 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 AE |
1033 | if (!bytes || cursor->offset & ~PAGE_MASK) |
1034 | return false; /* more bytes to process in the current page */ | |
1035 | ||
d90deda6 YZ |
1036 | if (!cursor->resid) |
1037 | return false; /* no more data */ | |
1038 | ||
fe38a2b6 AE |
1039 | /* Move on to the next page */ |
1040 | ||
1041 | BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head)); | |
1042 | cursor->page = list_entry_next(cursor->page, lru); | |
25aff7c5 | 1043 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
1044 | |
1045 | return true; | |
1046 | } | |
1047 | ||
dd236fcb AE |
1048 | /* |
1049 | * Message data is handled (sent or received) in pieces, where each | |
1050 | * piece resides on a single page. The network layer might not | |
1051 | * consume an entire piece at once. A data item's cursor keeps | |
1052 | * track of which piece is next to process and how much remains to | |
1053 | * be processed in that piece. It also tracks whether the current | |
1054 | * piece is the last one in the data item. | |
1055 | */ | |
ca8b3a69 | 1056 | static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor) |
dd236fcb | 1057 | { |
ca8b3a69 | 1058 | size_t length = cursor->total_resid; |
8ae4f4f5 | 1059 | |
8ae4f4f5 | 1060 | switch (cursor->data->type) { |
dd236fcb | 1061 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1062 | ceph_msg_data_pagelist_cursor_init(cursor, length); |
dd236fcb | 1063 | break; |
e766d7b5 | 1064 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1065 | ceph_msg_data_pages_cursor_init(cursor, length); |
e766d7b5 | 1066 | break; |
dd236fcb AE |
1067 | #ifdef CONFIG_BLOCK |
1068 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1069 | ceph_msg_data_bio_cursor_init(cursor, length); |
6aaa4511 | 1070 | break; |
dd236fcb | 1071 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1072 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1073 | default: |
1074 | /* BUG(); */ | |
1075 | break; | |
1076 | } | |
8ae4f4f5 | 1077 | cursor->need_crc = true; |
dd236fcb AE |
1078 | } |
1079 | ||
ca8b3a69 AE |
1080 | static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length) |
1081 | { | |
1082 | struct ceph_msg_data_cursor *cursor = &msg->cursor; | |
1083 | struct ceph_msg_data *data; | |
1084 | ||
1085 | BUG_ON(!length); | |
1086 | BUG_ON(length > msg->data_length); | |
1087 | BUG_ON(list_empty(&msg->data)); | |
1088 | ||
ca8b3a69 AE |
1089 | cursor->data_head = &msg->data; |
1090 | cursor->total_resid = length; | |
1091 | data = list_first_entry(&msg->data, struct ceph_msg_data, links); | |
1092 | cursor->data = data; | |
1093 | ||
1094 | __ceph_msg_data_cursor_init(cursor); | |
1095 | } | |
1096 | ||
dd236fcb AE |
1097 | /* |
1098 | * Return the page containing the next piece to process for a given | |
1099 | * data item, and supply the page offset and length of that piece. | |
1100 | * Indicate whether this is the last piece in this data item. | |
1101 | */ | |
8ae4f4f5 AE |
1102 | static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor, |
1103 | size_t *page_offset, size_t *length, | |
dd236fcb AE |
1104 | bool *last_piece) |
1105 | { | |
1106 | struct page *page; | |
1107 | ||
8ae4f4f5 | 1108 | switch (cursor->data->type) { |
dd236fcb | 1109 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1110 | page = ceph_msg_data_pagelist_next(cursor, page_offset, length); |
dd236fcb | 1111 | break; |
e766d7b5 | 1112 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1113 | page = ceph_msg_data_pages_next(cursor, page_offset, length); |
e766d7b5 | 1114 | break; |
dd236fcb AE |
1115 | #ifdef CONFIG_BLOCK |
1116 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1117 | page = ceph_msg_data_bio_next(cursor, page_offset, length); |
6aaa4511 | 1118 | break; |
dd236fcb | 1119 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1120 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1121 | default: |
1122 | page = NULL; | |
1123 | break; | |
1124 | } | |
1125 | BUG_ON(!page); | |
1126 | BUG_ON(*page_offset + *length > PAGE_SIZE); | |
1127 | BUG_ON(!*length); | |
1128 | if (last_piece) | |
8ae4f4f5 | 1129 | *last_piece = cursor->last_piece; |
dd236fcb AE |
1130 | |
1131 | return page; | |
1132 | } | |
1133 | ||
1134 | /* | |
1135 | * Returns true if the result moves the cursor on to the next piece | |
1136 | * of the data item. | |
1137 | */ | |
8ae4f4f5 AE |
1138 | static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, |
1139 | size_t bytes) | |
dd236fcb AE |
1140 | { |
1141 | bool new_piece; | |
1142 | ||
25aff7c5 | 1143 | BUG_ON(bytes > cursor->resid); |
8ae4f4f5 | 1144 | switch (cursor->data->type) { |
dd236fcb | 1145 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1146 | new_piece = ceph_msg_data_pagelist_advance(cursor, bytes); |
dd236fcb | 1147 | break; |
e766d7b5 | 1148 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1149 | new_piece = ceph_msg_data_pages_advance(cursor, bytes); |
e766d7b5 | 1150 | break; |
dd236fcb AE |
1151 | #ifdef CONFIG_BLOCK |
1152 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1153 | new_piece = ceph_msg_data_bio_advance(cursor, bytes); |
6aaa4511 | 1154 | break; |
dd236fcb | 1155 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1156 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1157 | default: |
1158 | BUG(); | |
1159 | break; | |
1160 | } | |
ca8b3a69 | 1161 | cursor->total_resid -= bytes; |
dd236fcb | 1162 | |
ca8b3a69 AE |
1163 | if (!cursor->resid && cursor->total_resid) { |
1164 | WARN_ON(!cursor->last_piece); | |
1165 | BUG_ON(list_is_last(&cursor->data->links, cursor->data_head)); | |
1166 | cursor->data = list_entry_next(cursor->data, links); | |
1167 | __ceph_msg_data_cursor_init(cursor); | |
a51b272e | 1168 | new_piece = true; |
ca8b3a69 | 1169 | } |
a51b272e | 1170 | cursor->need_crc = new_piece; |
ca8b3a69 | 1171 | |
dd236fcb AE |
1172 | return new_piece; |
1173 | } | |
1174 | ||
98fa5dd8 | 1175 | static void prepare_message_data(struct ceph_msg *msg, u32 data_len) |
739c905b | 1176 | { |
739c905b | 1177 | BUG_ON(!msg); |
25aff7c5 | 1178 | BUG_ON(!data_len); |
739c905b | 1179 | |
4c59b4a2 | 1180 | /* Initialize data cursor */ |
fe38a2b6 | 1181 | |
8ae4f4f5 | 1182 | ceph_msg_data_cursor_init(msg, (size_t)data_len); |
739c905b AE |
1183 | } |
1184 | ||
31b8006e SW |
1185 | /* |
1186 | * Prepare footer for currently outgoing message, and finish things | |
1187 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
1188 | */ | |
859eb799 | 1189 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
1190 | { |
1191 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 1192 | int v = con->out_kvec_left; |
31b8006e | 1193 | |
fd154f3c AE |
1194 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
1195 | ||
31b8006e SW |
1196 | dout("prepare_write_message_footer %p\n", con); |
1197 | con->out_kvec_is_msg = true; | |
1198 | con->out_kvec[v].iov_base = &m->footer; | |
33d07337 YZ |
1199 | if (con->peer_features & CEPH_FEATURE_MSG_AUTH) { |
1200 | if (con->ops->sign_message) | |
1201 | con->ops->sign_message(con, m); | |
1202 | else | |
1203 | m->footer.sig = 0; | |
1204 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
1205 | con->out_kvec_bytes += sizeof(m->footer); | |
1206 | } else { | |
1207 | m->old_footer.flags = m->footer.flags; | |
1208 | con->out_kvec[v].iov_len = sizeof(m->old_footer); | |
1209 | con->out_kvec_bytes += sizeof(m->old_footer); | |
1210 | } | |
31b8006e SW |
1211 | con->out_kvec_left++; |
1212 | con->out_more = m->more_to_follow; | |
c86a2930 | 1213 | con->out_msg_done = true; |
31b8006e SW |
1214 | } |
1215 | ||
1216 | /* | |
1217 | * Prepare headers for the next outgoing message. | |
1218 | */ | |
1219 | static void prepare_write_message(struct ceph_connection *con) | |
1220 | { | |
1221 | struct ceph_msg *m; | |
a9a0c51a | 1222 | u32 crc; |
31b8006e | 1223 | |
e2200423 | 1224 | con_out_kvec_reset(con); |
31b8006e | 1225 | con->out_kvec_is_msg = true; |
c86a2930 | 1226 | con->out_msg_done = false; |
31b8006e SW |
1227 | |
1228 | /* Sneak an ack in there first? If we can get it into the same | |
1229 | * TCP packet that's a good thing. */ | |
1230 | if (con->in_seq > con->in_seq_acked) { | |
1231 | con->in_seq_acked = con->in_seq; | |
e2200423 | 1232 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 1233 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1234 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 1235 | &con->out_temp_ack); |
31b8006e SW |
1236 | } |
1237 | ||
38941f80 | 1238 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 1239 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 1240 | con->out_msg = m; |
38941f80 | 1241 | BUG_ON(m->con != con); |
4cf9d544 SW |
1242 | |
1243 | /* put message on sent list */ | |
1244 | ceph_msg_get(m); | |
1245 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 1246 | |
e84346b7 SW |
1247 | /* |
1248 | * only assign outgoing seq # if we haven't sent this message | |
1249 | * yet. if it is requeued, resend with it's original seq. | |
1250 | */ | |
1251 | if (m->needs_out_seq) { | |
1252 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
1253 | m->needs_out_seq = false; | |
1254 | } | |
98fa5dd8 | 1255 | WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len)); |
31b8006e | 1256 | |
98fa5dd8 | 1257 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n", |
31b8006e SW |
1258 | m, con->out_seq, le16_to_cpu(m->hdr.type), |
1259 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
98fa5dd8 | 1260 | m->data_length); |
31b8006e SW |
1261 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); |
1262 | ||
1263 | /* tag + hdr + front + middle */ | |
e2200423 AE |
1264 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
1265 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
1266 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 1267 | |
31b8006e | 1268 | if (m->middle) |
e2200423 | 1269 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 1270 | m->middle->vec.iov_base); |
31b8006e SW |
1271 | |
1272 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
1273 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
1274 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 1275 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
1276 | |
1277 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
1278 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
1279 | if (m->middle) { | |
1280 | crc = crc32c(0, m->middle->vec.iov_base, | |
1281 | m->middle->vec.iov_len); | |
1282 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
1283 | } else | |
31b8006e | 1284 | con->out_msg->footer.middle_crc = 0; |
739c905b | 1285 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
1286 | le32_to_cpu(con->out_msg->footer.front_crc), |
1287 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
1288 | ||
1289 | /* is there a data payload? */ | |
739c905b | 1290 | con->out_msg->footer.data_crc = 0; |
98fa5dd8 AE |
1291 | if (m->data_length) { |
1292 | prepare_message_data(con->out_msg, m->data_length); | |
78625051 AE |
1293 | con->out_more = 1; /* data + footer will follow */ |
1294 | } else { | |
31b8006e | 1295 | /* no, queue up footer too and be done */ |
859eb799 | 1296 | prepare_write_message_footer(con); |
78625051 | 1297 | } |
31b8006e | 1298 | |
c9ffc77a | 1299 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1300 | } |
1301 | ||
1302 | /* | |
1303 | * Prepare an ack. | |
1304 | */ | |
1305 | static void prepare_write_ack(struct ceph_connection *con) | |
1306 | { | |
1307 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
1308 | con->in_seq_acked, con->in_seq); | |
1309 | con->in_seq_acked = con->in_seq; | |
1310 | ||
e2200423 | 1311 | con_out_kvec_reset(con); |
859eb799 | 1312 | |
e2200423 | 1313 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 1314 | |
31b8006e | 1315 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1316 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
1317 | &con->out_temp_ack); |
1318 | ||
31b8006e | 1319 | con->out_more = 1; /* more will follow.. eventually.. */ |
c9ffc77a | 1320 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1321 | } |
1322 | ||
3a23083b SW |
1323 | /* |
1324 | * Prepare to share the seq during handshake | |
1325 | */ | |
1326 | static void prepare_write_seq(struct ceph_connection *con) | |
1327 | { | |
1328 | dout("prepare_write_seq %p %llu -> %llu\n", con, | |
1329 | con->in_seq_acked, con->in_seq); | |
1330 | con->in_seq_acked = con->in_seq; | |
1331 | ||
1332 | con_out_kvec_reset(con); | |
1333 | ||
1334 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); | |
1335 | con_out_kvec_add(con, sizeof (con->out_temp_ack), | |
1336 | &con->out_temp_ack); | |
1337 | ||
1338 | con_flag_set(con, CON_FLAG_WRITE_PENDING); | |
1339 | } | |
1340 | ||
31b8006e SW |
1341 | /* |
1342 | * Prepare to write keepalive byte. | |
1343 | */ | |
1344 | static void prepare_write_keepalive(struct ceph_connection *con) | |
1345 | { | |
1346 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
1347 | con_out_kvec_reset(con); |
1348 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
c9ffc77a | 1349 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1350 | } |
1351 | ||
1352 | /* | |
1353 | * Connection negotiation. | |
1354 | */ | |
1355 | ||
dac1e716 AE |
1356 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
1357 | int *auth_proto) | |
4e7a5dcd | 1358 | { |
a3530df3 | 1359 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
1360 | |
1361 | if (!con->ops->get_authorizer) { | |
1362 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
1363 | con->out_connect.authorizer_len = 0; | |
729796be | 1364 | return NULL; |
b1c6b980 AE |
1365 | } |
1366 | ||
1367 | /* Can't hold the mutex while getting authorizer */ | |
ec302645 | 1368 | mutex_unlock(&con->mutex); |
dac1e716 | 1369 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
ec302645 | 1370 | mutex_lock(&con->mutex); |
4e7a5dcd | 1371 | |
a3530df3 | 1372 | if (IS_ERR(auth)) |
729796be | 1373 | return auth; |
8dacc7da | 1374 | if (con->state != CON_STATE_NEGOTIATING) |
729796be | 1375 | return ERR_PTR(-EAGAIN); |
0da5d703 | 1376 | |
8f43fb53 AE |
1377 | con->auth_reply_buf = auth->authorizer_reply_buf; |
1378 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
729796be | 1379 | return auth; |
4e7a5dcd SW |
1380 | } |
1381 | ||
31b8006e SW |
1382 | /* |
1383 | * We connected to a peer and are saying hello. | |
1384 | */ | |
e825a66d | 1385 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 1386 | { |
e2200423 AE |
1387 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
1388 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 1389 | &con->msgr->my_enc_addr); |
eed0ef2c | 1390 | |
eed0ef2c | 1391 | con->out_more = 0; |
c9ffc77a | 1392 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
eed0ef2c SW |
1393 | } |
1394 | ||
e825a66d | 1395 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 1396 | { |
95c96174 | 1397 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 1398 | int proto; |
dac1e716 | 1399 | int auth_proto; |
729796be | 1400 | struct ceph_auth_handshake *auth; |
31b8006e SW |
1401 | |
1402 | switch (con->peer_name.type) { | |
1403 | case CEPH_ENTITY_TYPE_MON: | |
1404 | proto = CEPH_MONC_PROTOCOL; | |
1405 | break; | |
1406 | case CEPH_ENTITY_TYPE_OSD: | |
1407 | proto = CEPH_OSDC_PROTOCOL; | |
1408 | break; | |
1409 | case CEPH_ENTITY_TYPE_MDS: | |
1410 | proto = CEPH_MDSC_PROTOCOL; | |
1411 | break; | |
1412 | default: | |
1413 | BUG(); | |
1414 | } | |
1415 | ||
1416 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
1417 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 1418 | |
e825a66d | 1419 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
1420 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
1421 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
1422 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
1423 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
1424 | con->out_connect.flags = 0; | |
31b8006e | 1425 | |
dac1e716 AE |
1426 | auth_proto = CEPH_AUTH_UNKNOWN; |
1427 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
1428 | if (IS_ERR(auth)) |
1429 | return PTR_ERR(auth); | |
3da54776 | 1430 | |
dac1e716 | 1431 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
1432 | con->out_connect.authorizer_len = auth ? |
1433 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
1434 | ||
e2200423 | 1435 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
1436 | &con->out_connect); |
1437 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 1438 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 1439 | auth->authorizer_buf); |
859eb799 | 1440 | |
31b8006e | 1441 | con->out_more = 0; |
c9ffc77a | 1442 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
4e7a5dcd | 1443 | |
e10c758e | 1444 | return 0; |
31b8006e SW |
1445 | } |
1446 | ||
31b8006e SW |
1447 | /* |
1448 | * write as much of pending kvecs to the socket as we can. | |
1449 | * 1 -> done | |
1450 | * 0 -> socket full, but more to do | |
1451 | * <0 -> error | |
1452 | */ | |
1453 | static int write_partial_kvec(struct ceph_connection *con) | |
1454 | { | |
1455 | int ret; | |
1456 | ||
1457 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
1458 | while (con->out_kvec_bytes > 0) { | |
1459 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
1460 | con->out_kvec_left, con->out_kvec_bytes, | |
1461 | con->out_more); | |
1462 | if (ret <= 0) | |
1463 | goto out; | |
1464 | con->out_kvec_bytes -= ret; | |
1465 | if (con->out_kvec_bytes == 0) | |
1466 | break; /* done */ | |
f42299e6 AE |
1467 | |
1468 | /* account for full iov entries consumed */ | |
1469 | while (ret >= con->out_kvec_cur->iov_len) { | |
1470 | BUG_ON(!con->out_kvec_left); | |
1471 | ret -= con->out_kvec_cur->iov_len; | |
1472 | con->out_kvec_cur++; | |
1473 | con->out_kvec_left--; | |
1474 | } | |
1475 | /* and for a partially-consumed entry */ | |
1476 | if (ret) { | |
1477 | con->out_kvec_cur->iov_len -= ret; | |
1478 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
1479 | } |
1480 | } | |
1481 | con->out_kvec_left = 0; | |
1482 | con->out_kvec_is_msg = false; | |
1483 | ret = 1; | |
1484 | out: | |
1485 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
1486 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
1487 | return ret; /* done! */ | |
1488 | } | |
1489 | ||
35b62808 AE |
1490 | static u32 ceph_crc32c_page(u32 crc, struct page *page, |
1491 | unsigned int page_offset, | |
1492 | unsigned int length) | |
1493 | { | |
1494 | char *kaddr; | |
1495 | ||
1496 | kaddr = kmap(page); | |
1497 | BUG_ON(kaddr == NULL); | |
1498 | crc = crc32c(crc, kaddr + page_offset, length); | |
1499 | kunmap(page); | |
1500 | ||
1501 | return crc; | |
1502 | } | |
31b8006e SW |
1503 | /* |
1504 | * Write as much message data payload as we can. If we finish, queue | |
1505 | * up the footer. | |
1506 | * 1 -> done, footer is now queued in out_kvec[]. | |
1507 | * 0 -> socket full, but more to do | |
1508 | * <0 -> error | |
1509 | */ | |
34d2d200 | 1510 | static int write_partial_message_data(struct ceph_connection *con) |
31b8006e SW |
1511 | { |
1512 | struct ceph_msg *msg = con->out_msg; | |
8ae4f4f5 | 1513 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
37675b0f | 1514 | bool do_datacrc = !con->msgr->nocrc; |
f5db90bc | 1515 | u32 crc; |
31b8006e | 1516 | |
859a35d5 | 1517 | dout("%s %p msg %p\n", __func__, con, msg); |
31b8006e | 1518 | |
5240d9f9 | 1519 | if (list_empty(&msg->data)) |
4c59b4a2 AE |
1520 | return -EINVAL; |
1521 | ||
5821bd8c AE |
1522 | /* |
1523 | * Iterate through each page that contains data to be | |
1524 | * written, and send as much as possible for each. | |
1525 | * | |
1526 | * If we are calculating the data crc (the default), we will | |
1527 | * need to map the page. If we have no pages, they have | |
1528 | * been revoked, so use the zero page. | |
1529 | */ | |
f5db90bc | 1530 | crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0; |
643c68a4 | 1531 | while (cursor->resid) { |
8a166d05 | 1532 | struct page *page; |
e387d525 AE |
1533 | size_t page_offset; |
1534 | size_t length; | |
8a166d05 | 1535 | bool last_piece; |
8ea299bc | 1536 | bool need_crc; |
f5db90bc | 1537 | int ret; |
68b4476b | 1538 | |
8ae4f4f5 | 1539 | page = ceph_msg_data_next(&msg->cursor, &page_offset, &length, |
4c59b4a2 | 1540 | &last_piece); |
e387d525 | 1541 | ret = ceph_tcp_sendpage(con->sock, page, page_offset, |
fe38a2b6 | 1542 | length, last_piece); |
f5db90bc AE |
1543 | if (ret <= 0) { |
1544 | if (do_datacrc) | |
1545 | msg->footer.data_crc = cpu_to_le32(crc); | |
31b8006e | 1546 | |
f5db90bc AE |
1547 | return ret; |
1548 | } | |
143334ff AE |
1549 | if (do_datacrc && cursor->need_crc) |
1550 | crc = ceph_crc32c_page(crc, page, page_offset, length); | |
8ae4f4f5 | 1551 | need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret); |
31b8006e SW |
1552 | } |
1553 | ||
34d2d200 | 1554 | dout("%s %p msg %p done\n", __func__, con, msg); |
31b8006e SW |
1555 | |
1556 | /* prepare and queue up footer, too */ | |
f5db90bc AE |
1557 | if (do_datacrc) |
1558 | msg->footer.data_crc = cpu_to_le32(crc); | |
1559 | else | |
84ca8fc8 | 1560 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1561 | con_out_kvec_reset(con); |
859eb799 | 1562 | prepare_write_message_footer(con); |
f5db90bc AE |
1563 | |
1564 | return 1; /* must return > 0 to indicate success */ | |
31b8006e SW |
1565 | } |
1566 | ||
1567 | /* | |
1568 | * write some zeros | |
1569 | */ | |
1570 | static int write_partial_skip(struct ceph_connection *con) | |
1571 | { | |
1572 | int ret; | |
1573 | ||
1574 | while (con->out_skip > 0) { | |
31739139 | 1575 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1576 | |
e1dcb128 | 1577 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true); |
31b8006e SW |
1578 | if (ret <= 0) |
1579 | goto out; | |
1580 | con->out_skip -= ret; | |
1581 | } | |
1582 | ret = 1; | |
1583 | out: | |
1584 | return ret; | |
1585 | } | |
1586 | ||
1587 | /* | |
1588 | * Prepare to read connection handshake, or an ack. | |
1589 | */ | |
eed0ef2c SW |
1590 | static void prepare_read_banner(struct ceph_connection *con) |
1591 | { | |
1592 | dout("prepare_read_banner %p\n", con); | |
1593 | con->in_base_pos = 0; | |
1594 | } | |
1595 | ||
31b8006e SW |
1596 | static void prepare_read_connect(struct ceph_connection *con) |
1597 | { | |
1598 | dout("prepare_read_connect %p\n", con); | |
1599 | con->in_base_pos = 0; | |
1600 | } | |
1601 | ||
1602 | static void prepare_read_ack(struct ceph_connection *con) | |
1603 | { | |
1604 | dout("prepare_read_ack %p\n", con); | |
1605 | con->in_base_pos = 0; | |
1606 | } | |
1607 | ||
3a23083b SW |
1608 | static void prepare_read_seq(struct ceph_connection *con) |
1609 | { | |
1610 | dout("prepare_read_seq %p\n", con); | |
1611 | con->in_base_pos = 0; | |
1612 | con->in_tag = CEPH_MSGR_TAG_SEQ; | |
1613 | } | |
1614 | ||
31b8006e SW |
1615 | static void prepare_read_tag(struct ceph_connection *con) |
1616 | { | |
1617 | dout("prepare_read_tag %p\n", con); | |
1618 | con->in_base_pos = 0; | |
1619 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1620 | } | |
1621 | ||
1622 | /* | |
1623 | * Prepare to read a message. | |
1624 | */ | |
1625 | static int prepare_read_message(struct ceph_connection *con) | |
1626 | { | |
1627 | dout("prepare_read_message %p\n", con); | |
1628 | BUG_ON(con->in_msg != NULL); | |
1629 | con->in_base_pos = 0; | |
1630 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1631 | return 0; | |
1632 | } | |
1633 | ||
1634 | ||
1635 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1636 | int end, int size, void *object) |
31b8006e | 1637 | { |
e6cee71f AE |
1638 | while (con->in_base_pos < end) { |
1639 | int left = end - con->in_base_pos; | |
31b8006e SW |
1640 | int have = size - left; |
1641 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1642 | if (ret <= 0) | |
1643 | return ret; | |
1644 | con->in_base_pos += ret; | |
1645 | } | |
1646 | return 1; | |
1647 | } | |
1648 | ||
1649 | ||
1650 | /* | |
1651 | * Read all or part of the connect-side handshake on a new connection | |
1652 | */ | |
eed0ef2c | 1653 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1654 | { |
fd51653f AE |
1655 | int size; |
1656 | int end; | |
1657 | int ret; | |
31b8006e | 1658 | |
eed0ef2c | 1659 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1660 | |
1661 | /* peer's banner */ | |
fd51653f AE |
1662 | size = strlen(CEPH_BANNER); |
1663 | end = size; | |
1664 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1665 | if (ret <= 0) |
1666 | goto out; | |
fd51653f AE |
1667 | |
1668 | size = sizeof (con->actual_peer_addr); | |
1669 | end += size; | |
1670 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1671 | if (ret <= 0) |
1672 | goto out; | |
fd51653f AE |
1673 | |
1674 | size = sizeof (con->peer_addr_for_me); | |
1675 | end += size; | |
1676 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1677 | if (ret <= 0) |
1678 | goto out; | |
fd51653f | 1679 | |
eed0ef2c SW |
1680 | out: |
1681 | return ret; | |
1682 | } | |
1683 | ||
1684 | static int read_partial_connect(struct ceph_connection *con) | |
1685 | { | |
fd51653f AE |
1686 | int size; |
1687 | int end; | |
1688 | int ret; | |
eed0ef2c SW |
1689 | |
1690 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1691 | ||
fd51653f AE |
1692 | size = sizeof (con->in_reply); |
1693 | end = size; | |
1694 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1695 | if (ret <= 0) |
1696 | goto out; | |
fd51653f AE |
1697 | |
1698 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1699 | end += size; | |
1700 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1701 | if (ret <= 0) |
1702 | goto out; | |
31b8006e | 1703 | |
4e7a5dcd SW |
1704 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1705 | con, (int)con->in_reply.tag, | |
1706 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1707 | le32_to_cpu(con->in_reply.global_seq)); |
1708 | out: | |
1709 | return ret; | |
eed0ef2c | 1710 | |
31b8006e SW |
1711 | } |
1712 | ||
1713 | /* | |
1714 | * Verify the hello banner looks okay. | |
1715 | */ | |
1716 | static int verify_hello(struct ceph_connection *con) | |
1717 | { | |
1718 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1719 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1720 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1721 | con->error_msg = "protocol error, bad banner"; |
1722 | return -1; | |
1723 | } | |
1724 | return 0; | |
1725 | } | |
1726 | ||
1727 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1728 | { | |
1729 | switch (ss->ss_family) { | |
1730 | case AF_INET: | |
1731 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1732 | case AF_INET6: | |
1733 | return | |
1734 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1735 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1736 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1737 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1738 | } | |
1739 | return false; | |
1740 | } | |
1741 | ||
1742 | static int addr_port(struct sockaddr_storage *ss) | |
1743 | { | |
1744 | switch (ss->ss_family) { | |
1745 | case AF_INET: | |
f28bcfbe | 1746 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1747 | case AF_INET6: |
f28bcfbe | 1748 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1749 | } |
1750 | return 0; | |
1751 | } | |
1752 | ||
1753 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1754 | { | |
1755 | switch (ss->ss_family) { | |
1756 | case AF_INET: | |
1757 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1758 | break; |
31b8006e SW |
1759 | case AF_INET6: |
1760 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1761 | break; |
31b8006e SW |
1762 | } |
1763 | } | |
1764 | ||
ee3b56f2 NW |
1765 | /* |
1766 | * Unlike other *_pton function semantics, zero indicates success. | |
1767 | */ | |
1768 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1769 | char delim, const char **ipend) | |
1770 | { | |
99f0f3b2 AE |
1771 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1772 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1773 | |
1774 | memset(ss, 0, sizeof(*ss)); | |
1775 | ||
1776 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1777 | ss->ss_family = AF_INET; | |
1778 | return 0; | |
1779 | } | |
1780 | ||
1781 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1782 | ss->ss_family = AF_INET6; | |
1783 | return 0; | |
1784 | } | |
1785 | ||
1786 | return -EINVAL; | |
1787 | } | |
1788 | ||
1789 | /* | |
1790 | * Extract hostname string and resolve using kernel DNS facility. | |
1791 | */ | |
1792 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1793 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1794 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1795 | { | |
1796 | const char *end, *delim_p; | |
1797 | char *colon_p, *ip_addr = NULL; | |
1798 | int ip_len, ret; | |
1799 | ||
1800 | /* | |
1801 | * The end of the hostname occurs immediately preceding the delimiter or | |
1802 | * the port marker (':') where the delimiter takes precedence. | |
1803 | */ | |
1804 | delim_p = memchr(name, delim, namelen); | |
1805 | colon_p = memchr(name, ':', namelen); | |
1806 | ||
1807 | if (delim_p && colon_p) | |
1808 | end = delim_p < colon_p ? delim_p : colon_p; | |
1809 | else if (!delim_p && colon_p) | |
1810 | end = colon_p; | |
1811 | else { | |
1812 | end = delim_p; | |
1813 | if (!end) /* case: hostname:/ */ | |
1814 | end = name + namelen; | |
1815 | } | |
1816 | ||
1817 | if (end <= name) | |
1818 | return -EINVAL; | |
1819 | ||
1820 | /* do dns_resolve upcall */ | |
1821 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1822 | if (ip_len > 0) | |
1823 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1824 | else | |
1825 | ret = -ESRCH; | |
1826 | ||
1827 | kfree(ip_addr); | |
1828 | ||
1829 | *ipend = end; | |
1830 | ||
1831 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1832 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1833 | ||
1834 | return ret; | |
1835 | } | |
1836 | #else | |
1837 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1838 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1839 | { | |
1840 | return -EINVAL; | |
1841 | } | |
1842 | #endif | |
1843 | ||
1844 | /* | |
1845 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1846 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1847 | */ | |
1848 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1849 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1850 | { | |
1851 | int ret; | |
1852 | ||
1853 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1854 | if (ret) | |
1855 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1856 | ||
1857 | return ret; | |
1858 | } | |
1859 | ||
31b8006e SW |
1860 | /* |
1861 | * Parse an ip[:port] list into an addr array. Use the default | |
1862 | * monitor port if a port isn't specified. | |
1863 | */ | |
1864 | int ceph_parse_ips(const char *c, const char *end, | |
1865 | struct ceph_entity_addr *addr, | |
1866 | int max_count, int *count) | |
1867 | { | |
ee3b56f2 | 1868 | int i, ret = -EINVAL; |
31b8006e SW |
1869 | const char *p = c; |
1870 | ||
1871 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1872 | for (i = 0; i < max_count; i++) { | |
1873 | const char *ipend; | |
1874 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1875 | int port; |
39139f64 SW |
1876 | char delim = ','; |
1877 | ||
1878 | if (*p == '[') { | |
1879 | delim = ']'; | |
1880 | p++; | |
1881 | } | |
31b8006e | 1882 | |
ee3b56f2 NW |
1883 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1884 | if (ret) | |
31b8006e | 1885 | goto bad; |
ee3b56f2 NW |
1886 | ret = -EINVAL; |
1887 | ||
31b8006e SW |
1888 | p = ipend; |
1889 | ||
39139f64 SW |
1890 | if (delim == ']') { |
1891 | if (*p != ']') { | |
1892 | dout("missing matching ']'\n"); | |
1893 | goto bad; | |
1894 | } | |
1895 | p++; | |
1896 | } | |
1897 | ||
31b8006e SW |
1898 | /* port? */ |
1899 | if (p < end && *p == ':') { | |
1900 | port = 0; | |
1901 | p++; | |
1902 | while (p < end && *p >= '0' && *p <= '9') { | |
1903 | port = (port * 10) + (*p - '0'); | |
1904 | p++; | |
1905 | } | |
f48db1e9 ID |
1906 | if (port == 0) |
1907 | port = CEPH_MON_PORT; | |
1908 | else if (port > 65535) | |
31b8006e SW |
1909 | goto bad; |
1910 | } else { | |
1911 | port = CEPH_MON_PORT; | |
1912 | } | |
1913 | ||
1914 | addr_set_port(ss, port); | |
1915 | ||
3d14c5d2 | 1916 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1917 | |
1918 | if (p == end) | |
1919 | break; | |
1920 | if (*p != ',') | |
1921 | goto bad; | |
1922 | p++; | |
1923 | } | |
1924 | ||
1925 | if (p != end) | |
1926 | goto bad; | |
1927 | ||
1928 | if (count) | |
1929 | *count = i + 1; | |
1930 | return 0; | |
1931 | ||
1932 | bad: | |
39139f64 | 1933 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1934 | return ret; |
31b8006e | 1935 | } |
3d14c5d2 | 1936 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1937 | |
eed0ef2c | 1938 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1939 | { |
eed0ef2c | 1940 | dout("process_banner on %p\n", con); |
31b8006e SW |
1941 | |
1942 | if (verify_hello(con) < 0) | |
1943 | return -1; | |
1944 | ||
63f2d211 SW |
1945 | ceph_decode_addr(&con->actual_peer_addr); |
1946 | ceph_decode_addr(&con->peer_addr_for_me); | |
1947 | ||
31b8006e SW |
1948 | /* |
1949 | * Make sure the other end is who we wanted. note that the other | |
1950 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1951 | * them the benefit of the doubt. | |
1952 | */ | |
103e2d3a SW |
1953 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1954 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1955 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1956 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
b9a67899 JP |
1957 | pr_warn("wrong peer, want %s/%d, got %s/%d\n", |
1958 | ceph_pr_addr(&con->peer_addr.in_addr), | |
1959 | (int)le32_to_cpu(con->peer_addr.nonce), | |
1960 | ceph_pr_addr(&con->actual_peer_addr.in_addr), | |
1961 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); | |
58bb3b37 | 1962 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1963 | return -1; |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * did we learn our address? | |
1968 | */ | |
1969 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1970 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1971 | ||
1972 | memcpy(&con->msgr->inst.addr.in_addr, | |
1973 | &con->peer_addr_for_me.in_addr, | |
1974 | sizeof(con->peer_addr_for_me.in_addr)); | |
1975 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1976 | encode_my_addr(con->msgr); |
eed0ef2c | 1977 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1978 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1979 | } |
1980 | ||
eed0ef2c SW |
1981 | return 0; |
1982 | } | |
1983 | ||
1984 | static int process_connect(struct ceph_connection *con) | |
1985 | { | |
3d14c5d2 YS |
1986 | u64 sup_feat = con->msgr->supported_features; |
1987 | u64 req_feat = con->msgr->required_features; | |
2b3e0c90 ID |
1988 | u64 server_feat = ceph_sanitize_features( |
1989 | le64_to_cpu(con->in_reply.features)); | |
0da5d703 | 1990 | int ret; |
04a419f9 | 1991 | |
eed0ef2c SW |
1992 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1993 | ||
31b8006e | 1994 | switch (con->in_reply.tag) { |
04a419f9 SW |
1995 | case CEPH_MSGR_TAG_FEATURES: |
1996 | pr_err("%s%lld %s feature set mismatch," | |
1997 | " my %llx < server's %llx, missing %llx\n", | |
1998 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1999 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
2000 | sup_feat, server_feat, server_feat & ~sup_feat); |
2001 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 2002 | reset_connection(con); |
04a419f9 SW |
2003 | return -1; |
2004 | ||
31b8006e | 2005 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
2006 | pr_err("%s%lld %s protocol version mismatch," |
2007 | " my %d != server's %d\n", | |
2008 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2009 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
2010 | le32_to_cpu(con->out_connect.protocol_version), |
2011 | le32_to_cpu(con->in_reply.protocol_version)); | |
2012 | con->error_msg = "protocol version mismatch"; | |
0fa6ebc6 | 2013 | reset_connection(con); |
31b8006e SW |
2014 | return -1; |
2015 | ||
4e7a5dcd SW |
2016 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
2017 | con->auth_retry++; | |
2018 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
2019 | con->auth_retry); | |
2020 | if (con->auth_retry == 2) { | |
2021 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
2022 | return -1; |
2023 | } | |
6d4221b5 | 2024 | con_out_kvec_reset(con); |
e825a66d | 2025 | ret = prepare_write_connect(con); |
0da5d703 SW |
2026 | if (ret < 0) |
2027 | return ret; | |
63733a0f | 2028 | prepare_read_connect(con); |
4e7a5dcd | 2029 | break; |
31b8006e SW |
2030 | |
2031 | case CEPH_MSGR_TAG_RESETSESSION: | |
2032 | /* | |
2033 | * If we connected with a large connect_seq but the peer | |
2034 | * has no record of a session with us (no connection, or | |
2035 | * connect_seq == 0), they will send RESETSESION to indicate | |
2036 | * that they must have reset their session, and may have | |
2037 | * dropped messages. | |
2038 | */ | |
2039 | dout("process_connect got RESET peer seq %u\n", | |
5bdca4e0 | 2040 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
2041 | pr_err("%s%lld %s connection reset\n", |
2042 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2043 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2044 | reset_connection(con); |
6d4221b5 | 2045 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2046 | ret = prepare_write_connect(con); |
2047 | if (ret < 0) | |
2048 | return ret; | |
31b8006e SW |
2049 | prepare_read_connect(con); |
2050 | ||
2051 | /* Tell ceph about it. */ | |
ec302645 | 2052 | mutex_unlock(&con->mutex); |
31b8006e SW |
2053 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
2054 | if (con->ops->peer_reset) | |
2055 | con->ops->peer_reset(con); | |
ec302645 | 2056 | mutex_lock(&con->mutex); |
8dacc7da | 2057 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 2058 | return -EAGAIN; |
31b8006e SW |
2059 | break; |
2060 | ||
2061 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
2062 | /* | |
2063 | * If we sent a smaller connect_seq than the peer has, try | |
2064 | * again with a larger value. | |
2065 | */ | |
5bdca4e0 | 2066 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 2067 | le32_to_cpu(con->out_connect.connect_seq), |
5bdca4e0 SW |
2068 | le32_to_cpu(con->in_reply.connect_seq)); |
2069 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
6d4221b5 | 2070 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2071 | ret = prepare_write_connect(con); |
2072 | if (ret < 0) | |
2073 | return ret; | |
31b8006e SW |
2074 | prepare_read_connect(con); |
2075 | break; | |
2076 | ||
2077 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
2078 | /* | |
2079 | * If we sent a smaller global_seq than the peer has, try | |
2080 | * again with a larger value. | |
2081 | */ | |
eed0ef2c | 2082 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 2083 | con->peer_global_seq, |
5bdca4e0 | 2084 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 2085 | get_global_seq(con->msgr, |
5bdca4e0 | 2086 | le32_to_cpu(con->in_reply.global_seq)); |
6d4221b5 | 2087 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2088 | ret = prepare_write_connect(con); |
2089 | if (ret < 0) | |
2090 | return ret; | |
31b8006e SW |
2091 | prepare_read_connect(con); |
2092 | break; | |
2093 | ||
3a23083b | 2094 | case CEPH_MSGR_TAG_SEQ: |
31b8006e | 2095 | case CEPH_MSGR_TAG_READY: |
04a419f9 SW |
2096 | if (req_feat & ~server_feat) { |
2097 | pr_err("%s%lld %s protocol feature mismatch," | |
2098 | " my required %llx > server's %llx, need %llx\n", | |
2099 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2100 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
2101 | req_feat, server_feat, req_feat & ~server_feat); |
2102 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 2103 | reset_connection(con); |
04a419f9 SW |
2104 | return -1; |
2105 | } | |
8dacc7da | 2106 | |
122070a2 | 2107 | WARN_ON(con->state != CON_STATE_NEGOTIATING); |
8dacc7da | 2108 | con->state = CON_STATE_OPEN; |
20e55c4c | 2109 | con->auth_retry = 0; /* we authenticated; clear flag */ |
31b8006e SW |
2110 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
2111 | con->connect_seq++; | |
aba558e2 | 2112 | con->peer_features = server_feat; |
31b8006e SW |
2113 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
2114 | con->peer_global_seq, | |
2115 | le32_to_cpu(con->in_reply.connect_seq), | |
2116 | con->connect_seq); | |
2117 | WARN_ON(con->connect_seq != | |
2118 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
2119 | |
2120 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
c9ffc77a | 2121 | con_flag_set(con, CON_FLAG_LOSSYTX); |
92ac41d0 | 2122 | |
85effe18 | 2123 | con->delay = 0; /* reset backoff memory */ |
92ac41d0 | 2124 | |
3a23083b SW |
2125 | if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) { |
2126 | prepare_write_seq(con); | |
2127 | prepare_read_seq(con); | |
2128 | } else { | |
2129 | prepare_read_tag(con); | |
2130 | } | |
31b8006e SW |
2131 | break; |
2132 | ||
2133 | case CEPH_MSGR_TAG_WAIT: | |
2134 | /* | |
2135 | * If there is a connection race (we are opening | |
2136 | * connections to each other), one of us may just have | |
2137 | * to WAIT. This shouldn't happen if we are the | |
2138 | * client. | |
2139 | */ | |
04177882 SW |
2140 | pr_err("process_connect got WAIT as client\n"); |
2141 | con->error_msg = "protocol error, got WAIT as client"; | |
2142 | return -1; | |
31b8006e SW |
2143 | |
2144 | default: | |
2145 | pr_err("connect protocol error, will retry\n"); | |
2146 | con->error_msg = "protocol error, garbage tag during connect"; | |
2147 | return -1; | |
2148 | } | |
2149 | return 0; | |
2150 | } | |
2151 | ||
2152 | ||
2153 | /* | |
2154 | * read (part of) an ack | |
2155 | */ | |
2156 | static int read_partial_ack(struct ceph_connection *con) | |
2157 | { | |
fd51653f AE |
2158 | int size = sizeof (con->in_temp_ack); |
2159 | int end = size; | |
31b8006e | 2160 | |
fd51653f | 2161 | return read_partial(con, end, size, &con->in_temp_ack); |
31b8006e SW |
2162 | } |
2163 | ||
31b8006e SW |
2164 | /* |
2165 | * We can finally discard anything that's been acked. | |
2166 | */ | |
2167 | static void process_ack(struct ceph_connection *con) | |
2168 | { | |
2169 | struct ceph_msg *m; | |
2170 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
2171 | u64 seq; | |
2172 | ||
31b8006e SW |
2173 | while (!list_empty(&con->out_sent)) { |
2174 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
2175 | list_head); | |
2176 | seq = le64_to_cpu(m->hdr.seq); | |
2177 | if (seq > ack) | |
2178 | break; | |
2179 | dout("got ack for seq %llu type %d at %p\n", seq, | |
2180 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 2181 | m->ack_stamp = jiffies; |
31b8006e SW |
2182 | ceph_msg_remove(m); |
2183 | } | |
31b8006e SW |
2184 | prepare_read_tag(con); |
2185 | } | |
2186 | ||
2187 | ||
2450418c | 2188 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
2189 | struct kvec *section, |
2190 | unsigned int sec_len, u32 *crc) | |
2450418c | 2191 | { |
68b4476b | 2192 | int ret, left; |
2450418c YS |
2193 | |
2194 | BUG_ON(!section); | |
2195 | ||
2196 | while (section->iov_len < sec_len) { | |
2197 | BUG_ON(section->iov_base == NULL); | |
2198 | left = sec_len - section->iov_len; | |
2199 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
2200 | section->iov_len, left); | |
2201 | if (ret <= 0) | |
2202 | return ret; | |
2203 | section->iov_len += ret; | |
2450418c | 2204 | } |
fe3ad593 AE |
2205 | if (section->iov_len == sec_len) |
2206 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 2207 | |
2450418c YS |
2208 | return 1; |
2209 | } | |
31b8006e | 2210 | |
34d2d200 AE |
2211 | static int read_partial_msg_data(struct ceph_connection *con) |
2212 | { | |
2213 | struct ceph_msg *msg = con->in_msg; | |
8ae4f4f5 | 2214 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
34d2d200 | 2215 | const bool do_datacrc = !con->msgr->nocrc; |
686be208 AE |
2216 | struct page *page; |
2217 | size_t page_offset; | |
2218 | size_t length; | |
f5db90bc | 2219 | u32 crc = 0; |
34d2d200 AE |
2220 | int ret; |
2221 | ||
2222 | BUG_ON(!msg); | |
5240d9f9 | 2223 | if (list_empty(&msg->data)) |
4c59b4a2 | 2224 | return -EIO; |
34d2d200 | 2225 | |
f5db90bc AE |
2226 | if (do_datacrc) |
2227 | crc = con->in_data_crc; | |
643c68a4 | 2228 | while (cursor->resid) { |
8ae4f4f5 | 2229 | page = ceph_msg_data_next(&msg->cursor, &page_offset, &length, |
4c59b4a2 | 2230 | NULL); |
686be208 | 2231 | ret = ceph_tcp_recvpage(con->sock, page, page_offset, length); |
f5db90bc AE |
2232 | if (ret <= 0) { |
2233 | if (do_datacrc) | |
2234 | con->in_data_crc = crc; | |
2235 | ||
686be208 | 2236 | return ret; |
f5db90bc | 2237 | } |
686be208 AE |
2238 | |
2239 | if (do_datacrc) | |
f5db90bc | 2240 | crc = ceph_crc32c_page(crc, page, page_offset, ret); |
8ae4f4f5 | 2241 | (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret); |
34d2d200 | 2242 | } |
f5db90bc AE |
2243 | if (do_datacrc) |
2244 | con->in_data_crc = crc; | |
34d2d200 AE |
2245 | |
2246 | return 1; /* must return > 0 to indicate success */ | |
2247 | } | |
2248 | ||
31b8006e SW |
2249 | /* |
2250 | * read (part of) a message. | |
2251 | */ | |
686be208 AE |
2252 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
2253 | ||
31b8006e SW |
2254 | static int read_partial_message(struct ceph_connection *con) |
2255 | { | |
2256 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
2257 | int size; |
2258 | int end; | |
31b8006e | 2259 | int ret; |
95c96174 | 2260 | unsigned int front_len, middle_len, data_len; |
37675b0f | 2261 | bool do_datacrc = !con->msgr->nocrc; |
33d07337 | 2262 | bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH); |
ae18756b | 2263 | u64 seq; |
fe3ad593 | 2264 | u32 crc; |
31b8006e SW |
2265 | |
2266 | dout("read_partial_message con %p msg %p\n", con, m); | |
2267 | ||
2268 | /* header */ | |
fd51653f AE |
2269 | size = sizeof (con->in_hdr); |
2270 | end = size; | |
2271 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
2272 | if (ret <= 0) |
2273 | return ret; | |
fe3ad593 AE |
2274 | |
2275 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
2276 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
2277 | pr_err("read_partial_message bad hdr " | |
2278 | " crc %u != expected %u\n", | |
2279 | crc, con->in_hdr.crc); | |
2280 | return -EBADMSG; | |
2281 | } | |
2282 | ||
31b8006e SW |
2283 | front_len = le32_to_cpu(con->in_hdr.front_len); |
2284 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
2285 | return -EIO; | |
2286 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
7b11ba37 | 2287 | if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN) |
31b8006e SW |
2288 | return -EIO; |
2289 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
2290 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
2291 | return -EIO; | |
2292 | ||
ae18756b SW |
2293 | /* verify seq# */ |
2294 | seq = le64_to_cpu(con->in_hdr.seq); | |
2295 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 2296 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 2297 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2298 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
2299 | seq, con->in_seq + 1); |
2300 | con->in_base_pos = -front_len - middle_len - data_len - | |
2301 | sizeof(m->footer); | |
2302 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
2303 | return 0; |
2304 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
2305 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
2306 | seq, con->in_seq + 1); | |
2307 | con->error_msg = "bad message sequence # for incoming message"; | |
2308 | return -EBADMSG; | |
2309 | } | |
2310 | ||
31b8006e SW |
2311 | /* allocate message? */ |
2312 | if (!con->in_msg) { | |
4740a623 SW |
2313 | int skip = 0; |
2314 | ||
31b8006e | 2315 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
6ebc8b32 | 2316 | front_len, data_len); |
4740a623 SW |
2317 | ret = ceph_con_in_msg_alloc(con, &skip); |
2318 | if (ret < 0) | |
2319 | return ret; | |
f759ebb9 AE |
2320 | |
2321 | BUG_ON(!con->in_msg ^ skip); | |
2322 | if (con->in_msg && data_len > con->in_msg->data_length) { | |
b9a67899 | 2323 | pr_warn("%s skipping long message (%u > %zd)\n", |
f759ebb9 AE |
2324 | __func__, data_len, con->in_msg->data_length); |
2325 | ceph_msg_put(con->in_msg); | |
2326 | con->in_msg = NULL; | |
2327 | skip = 1; | |
2328 | } | |
2450418c | 2329 | if (skip) { |
31b8006e | 2330 | /* skip this message */ |
a79832f2 | 2331 | dout("alloc_msg said skip message\n"); |
31b8006e SW |
2332 | con->in_base_pos = -front_len - middle_len - data_len - |
2333 | sizeof(m->footer); | |
2334 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2335 | con->in_seq++; |
31b8006e SW |
2336 | return 0; |
2337 | } | |
38941f80 | 2338 | |
4740a623 | 2339 | BUG_ON(!con->in_msg); |
38941f80 | 2340 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
2341 | m = con->in_msg; |
2342 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
2343 | if (m->middle) |
2344 | m->middle->vec.iov_len = 0; | |
9d7f0f13 | 2345 | |
78625051 | 2346 | /* prepare for data payload, if any */ |
a4107026 | 2347 | |
78625051 | 2348 | if (data_len) |
98fa5dd8 | 2349 | prepare_message_data(con->in_msg, data_len); |
31b8006e SW |
2350 | } |
2351 | ||
2352 | /* front */ | |
2450418c YS |
2353 | ret = read_partial_message_section(con, &m->front, front_len, |
2354 | &con->in_front_crc); | |
2355 | if (ret <= 0) | |
2356 | return ret; | |
31b8006e SW |
2357 | |
2358 | /* middle */ | |
2450418c | 2359 | if (m->middle) { |
213c99ee SW |
2360 | ret = read_partial_message_section(con, &m->middle->vec, |
2361 | middle_len, | |
2450418c | 2362 | &con->in_middle_crc); |
31b8006e SW |
2363 | if (ret <= 0) |
2364 | return ret; | |
31b8006e SW |
2365 | } |
2366 | ||
2367 | /* (page) data */ | |
34d2d200 AE |
2368 | if (data_len) { |
2369 | ret = read_partial_msg_data(con); | |
2370 | if (ret <= 0) | |
2371 | return ret; | |
31b8006e SW |
2372 | } |
2373 | ||
31b8006e | 2374 | /* footer */ |
33d07337 YZ |
2375 | if (need_sign) |
2376 | size = sizeof(m->footer); | |
2377 | else | |
2378 | size = sizeof(m->old_footer); | |
2379 | ||
fd51653f AE |
2380 | end += size; |
2381 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
2382 | if (ret <= 0) |
2383 | return ret; | |
2384 | ||
33d07337 YZ |
2385 | if (!need_sign) { |
2386 | m->footer.flags = m->old_footer.flags; | |
2387 | m->footer.sig = 0; | |
2388 | } | |
2389 | ||
31b8006e SW |
2390 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
2391 | m, front_len, m->footer.front_crc, middle_len, | |
2392 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
2393 | ||
2394 | /* crc ok? */ | |
2395 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
2396 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
2397 | m, con->in_front_crc, m->footer.front_crc); | |
2398 | return -EBADMSG; | |
2399 | } | |
2400 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
2401 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
2402 | m, con->in_middle_crc, m->footer.middle_crc); | |
2403 | return -EBADMSG; | |
2404 | } | |
bca064d2 | 2405 | if (do_datacrc && |
31b8006e SW |
2406 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
2407 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
2408 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
2409 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
2410 | return -EBADMSG; | |
2411 | } | |
2412 | ||
33d07337 YZ |
2413 | if (need_sign && con->ops->check_message_signature && |
2414 | con->ops->check_message_signature(con, m)) { | |
2415 | pr_err("read_partial_message %p signature check failed\n", m); | |
2416 | return -EBADMSG; | |
2417 | } | |
2418 | ||
31b8006e SW |
2419 | return 1; /* done! */ |
2420 | } | |
2421 | ||
2422 | /* | |
2423 | * Process message. This happens in the worker thread. The callback should | |
2424 | * be careful not to do anything that waits on other incoming messages or it | |
2425 | * may deadlock. | |
2426 | */ | |
2427 | static void process_message(struct ceph_connection *con) | |
2428 | { | |
5e095e8b | 2429 | struct ceph_msg *msg; |
31b8006e | 2430 | |
38941f80 AE |
2431 | BUG_ON(con->in_msg->con != con); |
2432 | con->in_msg->con = NULL; | |
5e095e8b | 2433 | msg = con->in_msg; |
31b8006e | 2434 | con->in_msg = NULL; |
36eb71aa | 2435 | con->ops->put(con); |
31b8006e SW |
2436 | |
2437 | /* if first message, set peer_name */ | |
2438 | if (con->peer_name.type == 0) | |
dbad185d | 2439 | con->peer_name = msg->hdr.src; |
31b8006e | 2440 | |
31b8006e | 2441 | con->in_seq++; |
ec302645 | 2442 | mutex_unlock(&con->mutex); |
31b8006e SW |
2443 | |
2444 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
2445 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 2446 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
2447 | le16_to_cpu(msg->hdr.type), |
2448 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2449 | le32_to_cpu(msg->hdr.front_len), | |
2450 | le32_to_cpu(msg->hdr.data_len), | |
2451 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
2452 | con->ops->dispatch(con, msg); | |
ec302645 SW |
2453 | |
2454 | mutex_lock(&con->mutex); | |
31b8006e SW |
2455 | } |
2456 | ||
2457 | ||
2458 | /* | |
2459 | * Write something to the socket. Called in a worker thread when the | |
2460 | * socket appears to be writeable and we have something ready to send. | |
2461 | */ | |
2462 | static int try_write(struct ceph_connection *con) | |
2463 | { | |
31b8006e SW |
2464 | int ret = 1; |
2465 | ||
d59315ca | 2466 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 2467 | |
31b8006e SW |
2468 | more: |
2469 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
2470 | ||
2471 | /* open the socket first? */ | |
8dacc7da SW |
2472 | if (con->state == CON_STATE_PREOPEN) { |
2473 | BUG_ON(con->sock); | |
2474 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2475 | |
e2200423 | 2476 | con_out_kvec_reset(con); |
e825a66d | 2477 | prepare_write_banner(con); |
eed0ef2c | 2478 | prepare_read_banner(con); |
31b8006e | 2479 | |
cf3e5c40 | 2480 | BUG_ON(con->in_msg); |
31b8006e SW |
2481 | con->in_tag = CEPH_MSGR_TAG_READY; |
2482 | dout("try_write initiating connect on %p new state %lu\n", | |
2483 | con, con->state); | |
41617d0c AE |
2484 | ret = ceph_tcp_connect(con); |
2485 | if (ret < 0) { | |
31b8006e | 2486 | con->error_msg = "connect error"; |
31b8006e SW |
2487 | goto out; |
2488 | } | |
2489 | } | |
2490 | ||
2491 | more_kvec: | |
2492 | /* kvec data queued? */ | |
2493 | if (con->out_skip) { | |
2494 | ret = write_partial_skip(con); | |
2495 | if (ret <= 0) | |
42961d23 | 2496 | goto out; |
31b8006e SW |
2497 | } |
2498 | if (con->out_kvec_left) { | |
2499 | ret = write_partial_kvec(con); | |
2500 | if (ret <= 0) | |
42961d23 | 2501 | goto out; |
31b8006e SW |
2502 | } |
2503 | ||
2504 | /* msg pages? */ | |
2505 | if (con->out_msg) { | |
c86a2930 SW |
2506 | if (con->out_msg_done) { |
2507 | ceph_msg_put(con->out_msg); | |
2508 | con->out_msg = NULL; /* we're done with this one */ | |
2509 | goto do_next; | |
2510 | } | |
2511 | ||
34d2d200 | 2512 | ret = write_partial_message_data(con); |
31b8006e SW |
2513 | if (ret == 1) |
2514 | goto more_kvec; /* we need to send the footer, too! */ | |
2515 | if (ret == 0) | |
42961d23 | 2516 | goto out; |
31b8006e | 2517 | if (ret < 0) { |
34d2d200 | 2518 | dout("try_write write_partial_message_data err %d\n", |
31b8006e | 2519 | ret); |
42961d23 | 2520 | goto out; |
31b8006e SW |
2521 | } |
2522 | } | |
2523 | ||
c86a2930 | 2524 | do_next: |
8dacc7da | 2525 | if (con->state == CON_STATE_OPEN) { |
31b8006e SW |
2526 | /* is anything else pending? */ |
2527 | if (!list_empty(&con->out_queue)) { | |
2528 | prepare_write_message(con); | |
2529 | goto more; | |
2530 | } | |
2531 | if (con->in_seq > con->in_seq_acked) { | |
2532 | prepare_write_ack(con); | |
2533 | goto more; | |
2534 | } | |
c9ffc77a | 2535 | if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) { |
31b8006e SW |
2536 | prepare_write_keepalive(con); |
2537 | goto more; | |
2538 | } | |
2539 | } | |
2540 | ||
2541 | /* Nothing to do! */ | |
c9ffc77a | 2542 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
31b8006e | 2543 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2544 | ret = 0; |
2545 | out: | |
42961d23 | 2546 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2547 | return ret; |
2548 | } | |
2549 | ||
2550 | ||
2551 | ||
2552 | /* | |
2553 | * Read what we can from the socket. | |
2554 | */ | |
2555 | static int try_read(struct ceph_connection *con) | |
2556 | { | |
31b8006e SW |
2557 | int ret = -1; |
2558 | ||
8dacc7da SW |
2559 | more: |
2560 | dout("try_read start on %p state %lu\n", con, con->state); | |
2561 | if (con->state != CON_STATE_CONNECTING && | |
2562 | con->state != CON_STATE_NEGOTIATING && | |
2563 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2564 | return 0; |
2565 | ||
8dacc7da | 2566 | BUG_ON(!con->sock); |
ec302645 | 2567 | |
31b8006e SW |
2568 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2569 | con->in_base_pos); | |
0da5d703 | 2570 | |
8dacc7da | 2571 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2572 | dout("try_read connecting\n"); |
2573 | ret = read_partial_banner(con); | |
2574 | if (ret <= 0) | |
ab166d5a | 2575 | goto out; |
7593af92 AE |
2576 | ret = process_banner(con); |
2577 | if (ret < 0) | |
2578 | goto out; | |
2579 | ||
8dacc7da | 2580 | con->state = CON_STATE_NEGOTIATING; |
7593af92 | 2581 | |
6d4221b5 JS |
2582 | /* |
2583 | * Received banner is good, exchange connection info. | |
2584 | * Do not reset out_kvec, as sending our banner raced | |
2585 | * with receiving peer banner after connect completed. | |
2586 | */ | |
7593af92 AE |
2587 | ret = prepare_write_connect(con); |
2588 | if (ret < 0) | |
2589 | goto out; | |
2590 | prepare_read_connect(con); | |
2591 | ||
2592 | /* Send connection info before awaiting response */ | |
0da5d703 SW |
2593 | goto out; |
2594 | } | |
2595 | ||
8dacc7da | 2596 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2597 | dout("try_read negotiating\n"); |
31b8006e SW |
2598 | ret = read_partial_connect(con); |
2599 | if (ret <= 0) | |
31b8006e | 2600 | goto out; |
98bdb0aa SW |
2601 | ret = process_connect(con); |
2602 | if (ret < 0) | |
2603 | goto out; | |
31b8006e SW |
2604 | goto more; |
2605 | } | |
2606 | ||
122070a2 | 2607 | WARN_ON(con->state != CON_STATE_OPEN); |
8dacc7da | 2608 | |
31b8006e SW |
2609 | if (con->in_base_pos < 0) { |
2610 | /* | |
2611 | * skipping + discarding content. | |
2612 | * | |
2613 | * FIXME: there must be a better way to do this! | |
2614 | */ | |
84495f49 AE |
2615 | static char buf[SKIP_BUF_SIZE]; |
2616 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2617 | ||
31b8006e SW |
2618 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2619 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2620 | if (ret <= 0) | |
98bdb0aa | 2621 | goto out; |
31b8006e SW |
2622 | con->in_base_pos += ret; |
2623 | if (con->in_base_pos) | |
2624 | goto more; | |
2625 | } | |
2626 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2627 | /* | |
2628 | * what's next? | |
2629 | */ | |
2630 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2631 | if (ret <= 0) | |
98bdb0aa | 2632 | goto out; |
31b8006e SW |
2633 | dout("try_read got tag %d\n", (int)con->in_tag); |
2634 | switch (con->in_tag) { | |
2635 | case CEPH_MSGR_TAG_MSG: | |
2636 | prepare_read_message(con); | |
2637 | break; | |
2638 | case CEPH_MSGR_TAG_ACK: | |
2639 | prepare_read_ack(con); | |
2640 | break; | |
2641 | case CEPH_MSGR_TAG_CLOSE: | |
8dacc7da SW |
2642 | con_close_socket(con); |
2643 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2644 | goto out; |
31b8006e SW |
2645 | default: |
2646 | goto bad_tag; | |
2647 | } | |
2648 | } | |
2649 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2650 | ret = read_partial_message(con); | |
2651 | if (ret <= 0) { | |
2652 | switch (ret) { | |
2653 | case -EBADMSG: | |
2654 | con->error_msg = "bad crc"; | |
2655 | ret = -EIO; | |
98bdb0aa | 2656 | break; |
31b8006e SW |
2657 | case -EIO: |
2658 | con->error_msg = "io error"; | |
98bdb0aa | 2659 | break; |
31b8006e | 2660 | } |
98bdb0aa | 2661 | goto out; |
31b8006e SW |
2662 | } |
2663 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2664 | goto more; | |
2665 | process_message(con); | |
7b862e07 SW |
2666 | if (con->state == CON_STATE_OPEN) |
2667 | prepare_read_tag(con); | |
31b8006e SW |
2668 | goto more; |
2669 | } | |
3a23083b SW |
2670 | if (con->in_tag == CEPH_MSGR_TAG_ACK || |
2671 | con->in_tag == CEPH_MSGR_TAG_SEQ) { | |
2672 | /* | |
2673 | * the final handshake seq exchange is semantically | |
2674 | * equivalent to an ACK | |
2675 | */ | |
31b8006e SW |
2676 | ret = read_partial_ack(con); |
2677 | if (ret <= 0) | |
98bdb0aa | 2678 | goto out; |
31b8006e SW |
2679 | process_ack(con); |
2680 | goto more; | |
2681 | } | |
2682 | ||
31b8006e | 2683 | out: |
98bdb0aa | 2684 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2685 | return ret; |
2686 | ||
2687 | bad_tag: | |
2688 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2689 | con->error_msg = "protocol error, garbage tag"; | |
2690 | ret = -1; | |
2691 | goto out; | |
2692 | } | |
2693 | ||
2694 | ||
2695 | /* | |
802c6d96 AE |
2696 | * Atomically queue work on a connection after the specified delay. |
2697 | * Bump @con reference to avoid races with connection teardown. | |
2698 | * Returns 0 if work was queued, or an error code otherwise. | |
31b8006e | 2699 | */ |
802c6d96 | 2700 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
31b8006e | 2701 | { |
31b8006e | 2702 | if (!con->ops->get(con)) { |
802c6d96 | 2703 | dout("%s %p ref count 0\n", __func__, con); |
802c6d96 | 2704 | return -ENOENT; |
31b8006e SW |
2705 | } |
2706 | ||
802c6d96 AE |
2707 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
2708 | dout("%s %p - already queued\n", __func__, con); | |
31b8006e | 2709 | con->ops->put(con); |
802c6d96 | 2710 | return -EBUSY; |
31b8006e | 2711 | } |
802c6d96 AE |
2712 | |
2713 | dout("%s %p %lu\n", __func__, con, delay); | |
802c6d96 AE |
2714 | return 0; |
2715 | } | |
2716 | ||
2717 | static void queue_con(struct ceph_connection *con) | |
2718 | { | |
2719 | (void) queue_con_delay(con, 0); | |
31b8006e SW |
2720 | } |
2721 | ||
37ab77ac ID |
2722 | static void cancel_con(struct ceph_connection *con) |
2723 | { | |
2724 | if (cancel_delayed_work(&con->work)) { | |
2725 | dout("%s %p\n", __func__, con); | |
2726 | con->ops->put(con); | |
2727 | } | |
2728 | } | |
2729 | ||
7bb21d68 AE |
2730 | static bool con_sock_closed(struct ceph_connection *con) |
2731 | { | |
c9ffc77a | 2732 | if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED)) |
7bb21d68 AE |
2733 | return false; |
2734 | ||
2735 | #define CASE(x) \ | |
2736 | case CON_STATE_ ## x: \ | |
2737 | con->error_msg = "socket closed (con state " #x ")"; \ | |
2738 | break; | |
2739 | ||
2740 | switch (con->state) { | |
2741 | CASE(CLOSED); | |
2742 | CASE(PREOPEN); | |
2743 | CASE(CONNECTING); | |
2744 | CASE(NEGOTIATING); | |
2745 | CASE(OPEN); | |
2746 | CASE(STANDBY); | |
2747 | default: | |
b9a67899 | 2748 | pr_warn("%s con %p unrecognized state %lu\n", |
7bb21d68 AE |
2749 | __func__, con, con->state); |
2750 | con->error_msg = "unrecognized con state"; | |
2751 | BUG(); | |
2752 | break; | |
2753 | } | |
2754 | #undef CASE | |
2755 | ||
2756 | return true; | |
2757 | } | |
2758 | ||
f20a39fd AE |
2759 | static bool con_backoff(struct ceph_connection *con) |
2760 | { | |
2761 | int ret; | |
2762 | ||
2763 | if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF)) | |
2764 | return false; | |
2765 | ||
2766 | ret = queue_con_delay(con, round_jiffies_relative(con->delay)); | |
2767 | if (ret) { | |
2768 | dout("%s: con %p FAILED to back off %lu\n", __func__, | |
2769 | con, con->delay); | |
2770 | BUG_ON(ret == -ENOENT); | |
2771 | con_flag_set(con, CON_FLAG_BACKOFF); | |
2772 | } | |
2773 | ||
2774 | return true; | |
2775 | } | |
2776 | ||
93209264 AE |
2777 | /* Finish fault handling; con->mutex must *not* be held here */ |
2778 | ||
2779 | static void con_fault_finish(struct ceph_connection *con) | |
2780 | { | |
2781 | /* | |
2782 | * in case we faulted due to authentication, invalidate our | |
2783 | * current tickets so that we can get new ones. | |
2784 | */ | |
2785 | if (con->auth_retry && con->ops->invalidate_authorizer) { | |
2786 | dout("calling invalidate_authorizer()\n"); | |
2787 | con->ops->invalidate_authorizer(con); | |
2788 | } | |
2789 | ||
2790 | if (con->ops->fault) | |
2791 | con->ops->fault(con); | |
2792 | } | |
2793 | ||
31b8006e SW |
2794 | /* |
2795 | * Do some work on a connection. Drop a connection ref when we're done. | |
2796 | */ | |
2797 | static void con_work(struct work_struct *work) | |
2798 | { | |
2799 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2800 | work.work); | |
89baaa57 | 2801 | unsigned long pflags = current->flags; |
49659416 | 2802 | bool fault; |
31b8006e | 2803 | |
89baaa57 MC |
2804 | current->flags |= PF_MEMALLOC; |
2805 | ||
9dd4658d | 2806 | mutex_lock(&con->mutex); |
49659416 AE |
2807 | while (true) { |
2808 | int ret; | |
31b8006e | 2809 | |
49659416 AE |
2810 | if ((fault = con_sock_closed(con))) { |
2811 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); | |
2812 | break; | |
2813 | } | |
2814 | if (con_backoff(con)) { | |
2815 | dout("%s: con %p BACKOFF\n", __func__, con); | |
2816 | break; | |
2817 | } | |
2818 | if (con->state == CON_STATE_STANDBY) { | |
2819 | dout("%s: con %p STANDBY\n", __func__, con); | |
2820 | break; | |
2821 | } | |
2822 | if (con->state == CON_STATE_CLOSED) { | |
2823 | dout("%s: con %p CLOSED\n", __func__, con); | |
2824 | BUG_ON(con->sock); | |
2825 | break; | |
2826 | } | |
2827 | if (con->state == CON_STATE_PREOPEN) { | |
2828 | dout("%s: con %p PREOPEN\n", __func__, con); | |
2829 | BUG_ON(con->sock); | |
2830 | } | |
0da5d703 | 2831 | |
49659416 AE |
2832 | ret = try_read(con); |
2833 | if (ret < 0) { | |
2834 | if (ret == -EAGAIN) | |
2835 | continue; | |
2836 | con->error_msg = "socket error on read"; | |
2837 | fault = true; | |
2838 | break; | |
2839 | } | |
2840 | ||
2841 | ret = try_write(con); | |
2842 | if (ret < 0) { | |
2843 | if (ret == -EAGAIN) | |
2844 | continue; | |
2845 | con->error_msg = "socket error on write"; | |
2846 | fault = true; | |
2847 | } | |
2848 | ||
2849 | break; /* If we make it to here, we're done */ | |
3a140a0d | 2850 | } |
b6e7b6a1 AE |
2851 | if (fault) |
2852 | con_fault(con); | |
9dd4658d | 2853 | mutex_unlock(&con->mutex); |
0da5d703 | 2854 | |
b6e7b6a1 AE |
2855 | if (fault) |
2856 | con_fault_finish(con); | |
2857 | ||
2858 | con->ops->put(con); | |
89baaa57 MC |
2859 | |
2860 | tsk_restore_flags(current, pflags, PF_MEMALLOC); | |
31b8006e SW |
2861 | } |
2862 | ||
31b8006e SW |
2863 | /* |
2864 | * Generic error/fault handler. A retry mechanism is used with | |
2865 | * exponential backoff | |
2866 | */ | |
93209264 | 2867 | static void con_fault(struct ceph_connection *con) |
31b8006e | 2868 | { |
b9a67899 JP |
2869 | pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
2870 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); | |
31b8006e | 2871 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2872 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2873 | |
122070a2 | 2874 | WARN_ON(con->state != CON_STATE_CONNECTING && |
8dacc7da SW |
2875 | con->state != CON_STATE_NEGOTIATING && |
2876 | con->state != CON_STATE_OPEN); | |
ec302645 | 2877 | |
31b8006e | 2878 | con_close_socket(con); |
5e095e8b | 2879 | |
c9ffc77a | 2880 | if (con_flag_test(con, CON_FLAG_LOSSYTX)) { |
8dacc7da SW |
2881 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
2882 | con->state = CON_STATE_CLOSED; | |
93209264 | 2883 | return; |
3b5ede07 SW |
2884 | } |
2885 | ||
5e095e8b | 2886 | if (con->in_msg) { |
38941f80 AE |
2887 | BUG_ON(con->in_msg->con != con); |
2888 | con->in_msg->con = NULL; | |
5e095e8b SW |
2889 | ceph_msg_put(con->in_msg); |
2890 | con->in_msg = NULL; | |
36eb71aa | 2891 | con->ops->put(con); |
5e095e8b | 2892 | } |
31b8006e | 2893 | |
e80a52d1 SW |
2894 | /* Requeue anything that hasn't been acked */ |
2895 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2896 | |
e76661d0 SW |
2897 | /* If there are no messages queued or keepalive pending, place |
2898 | * the connection in a STANDBY state */ | |
2899 | if (list_empty(&con->out_queue) && | |
c9ffc77a | 2900 | !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) { |
e00de341 | 2901 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
c9ffc77a | 2902 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
8dacc7da | 2903 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
2904 | } else { |
2905 | /* retry after a delay. */ | |
8dacc7da | 2906 | con->state = CON_STATE_PREOPEN; |
e80a52d1 SW |
2907 | if (con->delay == 0) |
2908 | con->delay = BASE_DELAY_INTERVAL; | |
2909 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2910 | con->delay *= 2; | |
c9ffc77a | 2911 | con_flag_set(con, CON_FLAG_BACKOFF); |
8618e30b | 2912 | queue_con(con); |
31b8006e | 2913 | } |
31b8006e SW |
2914 | } |
2915 | ||
2916 | ||
2917 | ||
2918 | /* | |
15d9882c | 2919 | * initialize a new messenger instance |
31b8006e | 2920 | */ |
15d9882c AE |
2921 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2922 | struct ceph_entity_addr *myaddr, | |
12b4629a ID |
2923 | u64 supported_features, |
2924 | u64 required_features, | |
15d9882c | 2925 | bool nocrc) |
31b8006e | 2926 | { |
3d14c5d2 YS |
2927 | msgr->supported_features = supported_features; |
2928 | msgr->required_features = required_features; | |
2929 | ||
31b8006e SW |
2930 | spin_lock_init(&msgr->global_seq_lock); |
2931 | ||
31b8006e SW |
2932 | if (myaddr) |
2933 | msgr->inst.addr = *myaddr; | |
2934 | ||
2935 | /* select a random nonce */ | |
ac8839d7 | 2936 | msgr->inst.addr.type = 0; |
103e2d3a | 2937 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2938 | encode_my_addr(msgr); |
15d9882c | 2939 | msgr->nocrc = nocrc; |
31b8006e | 2940 | |
a2a32584 | 2941 | atomic_set(&msgr->stopping, 0); |
31b8006e | 2942 | |
15d9882c | 2943 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2944 | } |
15d9882c | 2945 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2946 | |
e00de341 SW |
2947 | static void clear_standby(struct ceph_connection *con) |
2948 | { | |
2949 | /* come back from STANDBY? */ | |
8dacc7da | 2950 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 2951 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 2952 | con->state = CON_STATE_PREOPEN; |
e00de341 | 2953 | con->connect_seq++; |
c9ffc77a AE |
2954 | WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING)); |
2955 | WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)); | |
e00de341 SW |
2956 | } |
2957 | } | |
2958 | ||
31b8006e SW |
2959 | /* |
2960 | * Queue up an outgoing message on the given connection. | |
2961 | */ | |
2962 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2963 | { | |
31b8006e | 2964 | /* set src+dst */ |
dbad185d | 2965 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 2966 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
2967 | msg->needs_out_seq = true; |
2968 | ||
ec302645 | 2969 | mutex_lock(&con->mutex); |
92ce034b | 2970 | |
8dacc7da | 2971 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
2972 | dout("con_send %p closed, dropping %p\n", con, msg); |
2973 | ceph_msg_put(msg); | |
2974 | mutex_unlock(&con->mutex); | |
2975 | return; | |
2976 | } | |
2977 | ||
38941f80 | 2978 | BUG_ON(msg->con != NULL); |
36eb71aa | 2979 | msg->con = con->ops->get(con); |
92ce034b AE |
2980 | BUG_ON(msg->con == NULL); |
2981 | ||
31b8006e SW |
2982 | BUG_ON(!list_empty(&msg->list_head)); |
2983 | list_add_tail(&msg->list_head, &con->out_queue); | |
2984 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2985 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2986 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2987 | le32_to_cpu(msg->hdr.front_len), | |
2988 | le32_to_cpu(msg->hdr.middle_len), | |
2989 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
2990 | |
2991 | clear_standby(con); | |
ec302645 | 2992 | mutex_unlock(&con->mutex); |
31b8006e SW |
2993 | |
2994 | /* if there wasn't anything waiting to send before, queue | |
2995 | * new work */ | |
c9ffc77a | 2996 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
31b8006e SW |
2997 | queue_con(con); |
2998 | } | |
3d14c5d2 | 2999 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
3000 | |
3001 | /* | |
3002 | * Revoke a message that was previously queued for send | |
3003 | */ | |
6740a845 | 3004 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 3005 | { |
6740a845 AE |
3006 | struct ceph_connection *con = msg->con; |
3007 | ||
3008 | if (!con) | |
3009 | return; /* Message not in our possession */ | |
3010 | ||
ec302645 | 3011 | mutex_lock(&con->mutex); |
31b8006e | 3012 | if (!list_empty(&msg->list_head)) { |
38941f80 | 3013 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 3014 | list_del_init(&msg->list_head); |
38941f80 | 3015 | BUG_ON(msg->con == NULL); |
36eb71aa | 3016 | msg->con->ops->put(msg->con); |
38941f80 | 3017 | msg->con = NULL; |
31b8006e | 3018 | msg->hdr.seq = 0; |
38941f80 | 3019 | |
31b8006e | 3020 | ceph_msg_put(msg); |
ed98adad SW |
3021 | } |
3022 | if (con->out_msg == msg) { | |
38941f80 | 3023 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 3024 | con->out_msg = NULL; |
31b8006e SW |
3025 | if (con->out_kvec_is_msg) { |
3026 | con->out_skip = con->out_kvec_bytes; | |
3027 | con->out_kvec_is_msg = false; | |
3028 | } | |
ed98adad | 3029 | msg->hdr.seq = 0; |
92ce034b AE |
3030 | |
3031 | ceph_msg_put(msg); | |
31b8006e | 3032 | } |
ec302645 | 3033 | mutex_unlock(&con->mutex); |
31b8006e SW |
3034 | } |
3035 | ||
350b1c32 | 3036 | /* |
0d59ab81 | 3037 | * Revoke a message that we may be reading data into |
350b1c32 | 3038 | */ |
8921d114 | 3039 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 3040 | { |
8921d114 AE |
3041 | struct ceph_connection *con; |
3042 | ||
3043 | BUG_ON(msg == NULL); | |
3044 | if (!msg->con) { | |
3045 | dout("%s msg %p null con\n", __func__, msg); | |
3046 | ||
3047 | return; /* Message not in our possession */ | |
3048 | } | |
3049 | ||
3050 | con = msg->con; | |
350b1c32 | 3051 | mutex_lock(&con->mutex); |
8921d114 | 3052 | if (con->in_msg == msg) { |
95c96174 ED |
3053 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
3054 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
3055 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
3056 | |
3057 | /* skip rest of message */ | |
8921d114 AE |
3058 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
3059 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 3060 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
3061 | front_len - |
3062 | middle_len - | |
3063 | data_len - | |
350b1c32 | 3064 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
3065 | ceph_msg_put(con->in_msg); |
3066 | con->in_msg = NULL; | |
3067 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 3068 | con->in_seq++; |
350b1c32 | 3069 | } else { |
8921d114 AE |
3070 | dout("%s %p in_msg %p msg %p no-op\n", |
3071 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
3072 | } |
3073 | mutex_unlock(&con->mutex); | |
3074 | } | |
3075 | ||
31b8006e SW |
3076 | /* |
3077 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
3078 | */ | |
3079 | void ceph_con_keepalive(struct ceph_connection *con) | |
3080 | { | |
e00de341 | 3081 | dout("con_keepalive %p\n", con); |
00650931 | 3082 | mutex_lock(&con->mutex); |
e00de341 | 3083 | clear_standby(con); |
00650931 | 3084 | mutex_unlock(&con->mutex); |
c9ffc77a AE |
3085 | if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 && |
3086 | con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) | |
31b8006e SW |
3087 | queue_con(con); |
3088 | } | |
3d14c5d2 | 3089 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e | 3090 | |
6644ed7b | 3091 | static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type) |
43794509 | 3092 | { |
6644ed7b AE |
3093 | struct ceph_msg_data *data; |
3094 | ||
3095 | if (WARN_ON(!ceph_msg_data_type_valid(type))) | |
3096 | return NULL; | |
3097 | ||
81b36be4 | 3098 | data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS); |
6644ed7b AE |
3099 | if (data) |
3100 | data->type = type; | |
5240d9f9 | 3101 | INIT_LIST_HEAD(&data->links); |
6644ed7b AE |
3102 | |
3103 | return data; | |
3104 | } | |
3105 | ||
3106 | static void ceph_msg_data_destroy(struct ceph_msg_data *data) | |
3107 | { | |
3108 | if (!data) | |
3109 | return; | |
3110 | ||
5240d9f9 | 3111 | WARN_ON(!list_empty(&data->links)); |
e4339d28 | 3112 | if (data->type == CEPH_MSG_DATA_PAGELIST) |
6644ed7b | 3113 | ceph_pagelist_release(data->pagelist); |
81b36be4 | 3114 | kmem_cache_free(ceph_msg_data_cache, data); |
43794509 AE |
3115 | } |
3116 | ||
90af3602 | 3117 | void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages, |
f1baeb2b | 3118 | size_t length, size_t alignment) |
02afca6c | 3119 | { |
6644ed7b AE |
3120 | struct ceph_msg_data *data; |
3121 | ||
07aa1558 AE |
3122 | BUG_ON(!pages); |
3123 | BUG_ON(!length); | |
6644ed7b AE |
3124 | |
3125 | data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES); | |
3126 | BUG_ON(!data); | |
3127 | data->pages = pages; | |
3128 | data->length = length; | |
3129 | data->alignment = alignment & ~PAGE_MASK; | |
02afca6c | 3130 | |
5240d9f9 AE |
3131 | list_add_tail(&data->links, &msg->data); |
3132 | msg->data_length += length; | |
02afca6c | 3133 | } |
90af3602 | 3134 | EXPORT_SYMBOL(ceph_msg_data_add_pages); |
31b8006e | 3135 | |
90af3602 | 3136 | void ceph_msg_data_add_pagelist(struct ceph_msg *msg, |
27fa8385 AE |
3137 | struct ceph_pagelist *pagelist) |
3138 | { | |
6644ed7b AE |
3139 | struct ceph_msg_data *data; |
3140 | ||
07aa1558 AE |
3141 | BUG_ON(!pagelist); |
3142 | BUG_ON(!pagelist->length); | |
27fa8385 | 3143 | |
6644ed7b AE |
3144 | data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST); |
3145 | BUG_ON(!data); | |
3146 | data->pagelist = pagelist; | |
3147 | ||
5240d9f9 AE |
3148 | list_add_tail(&data->links, &msg->data); |
3149 | msg->data_length += pagelist->length; | |
27fa8385 | 3150 | } |
90af3602 | 3151 | EXPORT_SYMBOL(ceph_msg_data_add_pagelist); |
27fa8385 | 3152 | |
ea96571f | 3153 | #ifdef CONFIG_BLOCK |
90af3602 | 3154 | void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio, |
a1930804 | 3155 | size_t length) |
27fa8385 | 3156 | { |
6644ed7b AE |
3157 | struct ceph_msg_data *data; |
3158 | ||
07aa1558 | 3159 | BUG_ON(!bio); |
27fa8385 | 3160 | |
6644ed7b AE |
3161 | data = ceph_msg_data_create(CEPH_MSG_DATA_BIO); |
3162 | BUG_ON(!data); | |
3163 | data->bio = bio; | |
c851c495 | 3164 | data->bio_length = length; |
6644ed7b | 3165 | |
5240d9f9 AE |
3166 | list_add_tail(&data->links, &msg->data); |
3167 | msg->data_length += length; | |
27fa8385 | 3168 | } |
90af3602 | 3169 | EXPORT_SYMBOL(ceph_msg_data_add_bio); |
ea96571f | 3170 | #endif /* CONFIG_BLOCK */ |
27fa8385 | 3171 | |
31b8006e SW |
3172 | /* |
3173 | * construct a new message with given type, size | |
3174 | * the new msg has a ref count of 1. | |
3175 | */ | |
b61c2763 SW |
3176 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
3177 | bool can_fail) | |
31b8006e SW |
3178 | { |
3179 | struct ceph_msg *m; | |
3180 | ||
e3d5d638 | 3181 | m = kmem_cache_zalloc(ceph_msg_cache, flags); |
31b8006e SW |
3182 | if (m == NULL) |
3183 | goto out; | |
31b8006e SW |
3184 | |
3185 | m->hdr.type = cpu_to_le16(type); | |
45c6ceb5 | 3186 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
31b8006e | 3187 | m->hdr.front_len = cpu_to_le32(front_len); |
ca20892d | 3188 | |
9516e45b AE |
3189 | INIT_LIST_HEAD(&m->list_head); |
3190 | kref_init(&m->kref); | |
5240d9f9 | 3191 | INIT_LIST_HEAD(&m->data); |
ca20892d | 3192 | |
31b8006e SW |
3193 | /* front */ |
3194 | if (front_len) { | |
eeb0bed5 | 3195 | m->front.iov_base = ceph_kvmalloc(front_len, flags); |
31b8006e | 3196 | if (m->front.iov_base == NULL) { |
b61c2763 | 3197 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
3198 | front_len); |
3199 | goto out2; | |
3200 | } | |
3201 | } else { | |
3202 | m->front.iov_base = NULL; | |
3203 | } | |
f2be82b0 | 3204 | m->front_alloc_len = m->front.iov_len = front_len; |
31b8006e | 3205 | |
bb257664 | 3206 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
3207 | return m; |
3208 | ||
3209 | out2: | |
3210 | ceph_msg_put(m); | |
3211 | out: | |
b61c2763 SW |
3212 | if (!can_fail) { |
3213 | pr_err("msg_new can't create type %d front %d\n", type, | |
3214 | front_len); | |
f0ed1b7c | 3215 | WARN_ON(1); |
b61c2763 SW |
3216 | } else { |
3217 | dout("msg_new can't create type %d front %d\n", type, | |
3218 | front_len); | |
3219 | } | |
a79832f2 | 3220 | return NULL; |
31b8006e | 3221 | } |
3d14c5d2 | 3222 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 3223 | |
31b8006e SW |
3224 | /* |
3225 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
3226 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
3227 | * per-type header in the front and then gracefully fail (i.e., | |
3228 | * propagate the error to the caller based on info in the front) when | |
3229 | * the middle is too large. | |
3230 | */ | |
2450418c | 3231 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
3232 | { |
3233 | int type = le16_to_cpu(msg->hdr.type); | |
3234 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
3235 | ||
3236 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
3237 | ceph_msg_type_name(type), middle_len); | |
3238 | BUG_ON(!middle_len); | |
3239 | BUG_ON(msg->middle); | |
3240 | ||
b6c1d5b8 | 3241 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
3242 | if (!msg->middle) |
3243 | return -ENOMEM; | |
3244 | return 0; | |
3245 | } | |
3246 | ||
2450418c | 3247 | /* |
1c20f2d2 AE |
3248 | * Allocate a message for receiving an incoming message on a |
3249 | * connection, and save the result in con->in_msg. Uses the | |
3250 | * connection's private alloc_msg op if available. | |
3251 | * | |
4740a623 SW |
3252 | * Returns 0 on success, or a negative error code. |
3253 | * | |
3254 | * On success, if we set *skip = 1: | |
3255 | * - the next message should be skipped and ignored. | |
3256 | * - con->in_msg == NULL | |
3257 | * or if we set *skip = 0: | |
3258 | * - con->in_msg is non-null. | |
3259 | * On error (ENOMEM, EAGAIN, ...), | |
3260 | * - con->in_msg == NULL | |
2450418c | 3261 | */ |
4740a623 | 3262 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
2450418c | 3263 | { |
4740a623 | 3264 | struct ceph_msg_header *hdr = &con->in_hdr; |
2450418c | 3265 | int middle_len = le32_to_cpu(hdr->middle_len); |
1d866d1c | 3266 | struct ceph_msg *msg; |
4740a623 | 3267 | int ret = 0; |
2450418c | 3268 | |
1c20f2d2 | 3269 | BUG_ON(con->in_msg != NULL); |
53ded495 | 3270 | BUG_ON(!con->ops->alloc_msg); |
2450418c | 3271 | |
53ded495 AE |
3272 | mutex_unlock(&con->mutex); |
3273 | msg = con->ops->alloc_msg(con, hdr, skip); | |
3274 | mutex_lock(&con->mutex); | |
3275 | if (con->state != CON_STATE_OPEN) { | |
3276 | if (msg) | |
1d866d1c | 3277 | ceph_msg_put(msg); |
53ded495 AE |
3278 | return -EAGAIN; |
3279 | } | |
4137577a AE |
3280 | if (msg) { |
3281 | BUG_ON(*skip); | |
3282 | con->in_msg = msg; | |
36eb71aa | 3283 | con->in_msg->con = con->ops->get(con); |
92ce034b | 3284 | BUG_ON(con->in_msg->con == NULL); |
4137577a AE |
3285 | } else { |
3286 | /* | |
3287 | * Null message pointer means either we should skip | |
3288 | * this message or we couldn't allocate memory. The | |
3289 | * former is not an error. | |
3290 | */ | |
3291 | if (*skip) | |
3292 | return 0; | |
3293 | con->error_msg = "error allocating memory for incoming message"; | |
3294 | ||
53ded495 | 3295 | return -ENOMEM; |
2450418c | 3296 | } |
1c20f2d2 | 3297 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 3298 | |
1c20f2d2 AE |
3299 | if (middle_len && !con->in_msg->middle) { |
3300 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 3301 | if (ret < 0) { |
1c20f2d2 AE |
3302 | ceph_msg_put(con->in_msg); |
3303 | con->in_msg = NULL; | |
2450418c YS |
3304 | } |
3305 | } | |
9d7f0f13 | 3306 | |
4740a623 | 3307 | return ret; |
2450418c YS |
3308 | } |
3309 | ||
31b8006e SW |
3310 | |
3311 | /* | |
3312 | * Free a generically kmalloc'd message. | |
3313 | */ | |
0215e44b | 3314 | static void ceph_msg_free(struct ceph_msg *m) |
31b8006e | 3315 | { |
0215e44b | 3316 | dout("%s %p\n", __func__, m); |
4965fc38 | 3317 | kvfree(m->front.iov_base); |
e3d5d638 | 3318 | kmem_cache_free(ceph_msg_cache, m); |
31b8006e SW |
3319 | } |
3320 | ||
0215e44b | 3321 | static void ceph_msg_release(struct kref *kref) |
c2e552e7 SW |
3322 | { |
3323 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
5240d9f9 AE |
3324 | LIST_HEAD(data); |
3325 | struct list_head *links; | |
3326 | struct list_head *next; | |
31b8006e | 3327 | |
0215e44b | 3328 | dout("%s %p\n", __func__, m); |
c2e552e7 SW |
3329 | WARN_ON(!list_empty(&m->list_head)); |
3330 | ||
3331 | /* drop middle, data, if any */ | |
3332 | if (m->middle) { | |
3333 | ceph_buffer_put(m->middle); | |
3334 | m->middle = NULL; | |
31b8006e | 3335 | } |
5240d9f9 AE |
3336 | |
3337 | list_splice_init(&m->data, &data); | |
3338 | list_for_each_safe(links, next, &data) { | |
3339 | struct ceph_msg_data *data; | |
3340 | ||
3341 | data = list_entry(links, struct ceph_msg_data, links); | |
3342 | list_del_init(links); | |
3343 | ceph_msg_data_destroy(data); | |
3344 | } | |
a1930804 | 3345 | m->data_length = 0; |
58bb3b37 | 3346 | |
c2e552e7 SW |
3347 | if (m->pool) |
3348 | ceph_msgpool_put(m->pool, m); | |
3349 | else | |
0215e44b ID |
3350 | ceph_msg_free(m); |
3351 | } | |
3352 | ||
3353 | struct ceph_msg *ceph_msg_get(struct ceph_msg *msg) | |
3354 | { | |
3355 | dout("%s %p (was %d)\n", __func__, msg, | |
3356 | atomic_read(&msg->kref.refcount)); | |
3357 | kref_get(&msg->kref); | |
3358 | return msg; | |
3359 | } | |
3360 | EXPORT_SYMBOL(ceph_msg_get); | |
3361 | ||
3362 | void ceph_msg_put(struct ceph_msg *msg) | |
3363 | { | |
3364 | dout("%s %p (was %d)\n", __func__, msg, | |
3365 | atomic_read(&msg->kref.refcount)); | |
3366 | kref_put(&msg->kref, ceph_msg_release); | |
31b8006e | 3367 | } |
0215e44b | 3368 | EXPORT_SYMBOL(ceph_msg_put); |
9ec7cab1 SW |
3369 | |
3370 | void ceph_msg_dump(struct ceph_msg *msg) | |
3371 | { | |
3cea4c30 ID |
3372 | pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg, |
3373 | msg->front_alloc_len, msg->data_length); | |
9ec7cab1 SW |
3374 | print_hex_dump(KERN_DEBUG, "header: ", |
3375 | DUMP_PREFIX_OFFSET, 16, 1, | |
3376 | &msg->hdr, sizeof(msg->hdr), true); | |
3377 | print_hex_dump(KERN_DEBUG, " front: ", | |
3378 | DUMP_PREFIX_OFFSET, 16, 1, | |
3379 | msg->front.iov_base, msg->front.iov_len, true); | |
3380 | if (msg->middle) | |
3381 | print_hex_dump(KERN_DEBUG, "middle: ", | |
3382 | DUMP_PREFIX_OFFSET, 16, 1, | |
3383 | msg->middle->vec.iov_base, | |
3384 | msg->middle->vec.iov_len, true); | |
3385 | print_hex_dump(KERN_DEBUG, "footer: ", | |
3386 | DUMP_PREFIX_OFFSET, 16, 1, | |
3387 | &msg->footer, sizeof(msg->footer), true); | |
3388 | } | |
3d14c5d2 | 3389 | EXPORT_SYMBOL(ceph_msg_dump); |