]>
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> | |
757856d2 | 9 | #include <linux/nsproxy.h> |
5a0e3ad6 | 10 | #include <linux/slab.h> |
31b8006e SW |
11 | #include <linux/socket.h> |
12 | #include <linux/string.h> | |
3ebc21f7 | 13 | #ifdef CONFIG_BLOCK |
68b4476b | 14 | #include <linux/bio.h> |
3ebc21f7 | 15 | #endif /* CONFIG_BLOCK */ |
ee3b56f2 | 16 | #include <linux/dns_resolver.h> |
31b8006e SW |
17 | #include <net/tcp.h> |
18 | ||
2b3e0c90 | 19 | #include <linux/ceph/ceph_features.h> |
3d14c5d2 YS |
20 | #include <linux/ceph/libceph.h> |
21 | #include <linux/ceph/messenger.h> | |
22 | #include <linux/ceph/decode.h> | |
23 | #include <linux/ceph/pagelist.h> | |
bc3b2d7f | 24 | #include <linux/export.h> |
31b8006e | 25 | |
fe38a2b6 AE |
26 | #define list_entry_next(pos, member) \ |
27 | list_entry(pos->member.next, typeof(*pos), member) | |
28 | ||
31b8006e SW |
29 | /* |
30 | * Ceph uses the messenger to exchange ceph_msg messages with other | |
31 | * hosts in the system. The messenger provides ordered and reliable | |
32 | * delivery. We tolerate TCP disconnects by reconnecting (with | |
33 | * exponential backoff) in the case of a fault (disconnection, bad | |
34 | * crc, protocol error). Acks allow sent messages to be discarded by | |
35 | * the sender. | |
36 | */ | |
37 | ||
bc18f4b1 AE |
38 | /* |
39 | * We track the state of the socket on a given connection using | |
40 | * values defined below. The transition to a new socket state is | |
41 | * handled by a function which verifies we aren't coming from an | |
42 | * unexpected state. | |
43 | * | |
44 | * -------- | |
45 | * | NEW* | transient initial state | |
46 | * -------- | |
47 | * | con_sock_state_init() | |
48 | * v | |
49 | * ---------- | |
50 | * | CLOSED | initialized, but no socket (and no | |
51 | * ---------- TCP connection) | |
52 | * ^ \ | |
53 | * | \ con_sock_state_connecting() | |
54 | * | ---------------------- | |
55 | * | \ | |
56 | * + con_sock_state_closed() \ | |
fbb85a47 SW |
57 | * |+--------------------------- \ |
58 | * | \ \ \ | |
59 | * | ----------- \ \ | |
60 | * | | CLOSING | socket event; \ \ | |
61 | * | ----------- await close \ \ | |
62 | * | ^ \ | | |
63 | * | | \ | | |
64 | * | + con_sock_state_closing() \ | | |
65 | * | / \ | | | |
66 | * | / --------------- | | | |
67 | * | / \ v v | |
bc18f4b1 AE |
68 | * | / -------------- |
69 | * | / -----------------| CONNECTING | socket created, TCP | |
70 | * | | / -------------- connect initiated | |
71 | * | | | con_sock_state_connected() | |
72 | * | | v | |
73 | * ------------- | |
74 | * | CONNECTED | TCP connection established | |
75 | * ------------- | |
76 | * | |
77 | * State values for ceph_connection->sock_state; NEW is assumed to be 0. | |
78 | */ | |
ce2c8903 AE |
79 | |
80 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ | |
81 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ | |
82 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ | |
83 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ | |
84 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ | |
85 | ||
8dacc7da SW |
86 | /* |
87 | * connection states | |
88 | */ | |
89 | #define CON_STATE_CLOSED 1 /* -> PREOPEN */ | |
90 | #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */ | |
91 | #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */ | |
92 | #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */ | |
93 | #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */ | |
94 | #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */ | |
95 | ||
4a861692 SW |
96 | /* |
97 | * ceph_connection flag bits | |
98 | */ | |
99 | #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop | |
100 | * messages on errors */ | |
101 | #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */ | |
102 | #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */ | |
103 | #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */ | |
104 | #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */ | |
8dacc7da | 105 | |
c9ffc77a AE |
106 | static bool con_flag_valid(unsigned long con_flag) |
107 | { | |
108 | switch (con_flag) { | |
109 | case CON_FLAG_LOSSYTX: | |
110 | case CON_FLAG_KEEPALIVE_PENDING: | |
111 | case CON_FLAG_WRITE_PENDING: | |
112 | case CON_FLAG_SOCK_CLOSED: | |
113 | case CON_FLAG_BACKOFF: | |
114 | return true; | |
115 | default: | |
116 | return false; | |
117 | } | |
118 | } | |
119 | ||
120 | static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag) | |
121 | { | |
122 | BUG_ON(!con_flag_valid(con_flag)); | |
123 | ||
124 | clear_bit(con_flag, &con->flags); | |
125 | } | |
126 | ||
127 | static void con_flag_set(struct ceph_connection *con, unsigned long con_flag) | |
128 | { | |
129 | BUG_ON(!con_flag_valid(con_flag)); | |
130 | ||
131 | set_bit(con_flag, &con->flags); | |
132 | } | |
133 | ||
134 | static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag) | |
135 | { | |
136 | BUG_ON(!con_flag_valid(con_flag)); | |
137 | ||
138 | return test_bit(con_flag, &con->flags); | |
139 | } | |
140 | ||
141 | static bool con_flag_test_and_clear(struct ceph_connection *con, | |
142 | unsigned long con_flag) | |
143 | { | |
144 | BUG_ON(!con_flag_valid(con_flag)); | |
145 | ||
146 | return test_and_clear_bit(con_flag, &con->flags); | |
147 | } | |
148 | ||
149 | static bool con_flag_test_and_set(struct ceph_connection *con, | |
150 | unsigned long con_flag) | |
151 | { | |
152 | BUG_ON(!con_flag_valid(con_flag)); | |
153 | ||
154 | return test_and_set_bit(con_flag, &con->flags); | |
155 | } | |
156 | ||
e3d5d638 AE |
157 | /* Slab caches for frequently-allocated structures */ |
158 | ||
159 | static struct kmem_cache *ceph_msg_cache; | |
81b36be4 | 160 | static struct kmem_cache *ceph_msg_data_cache; |
e3d5d638 | 161 | |
31b8006e SW |
162 | /* static tag bytes (protocol control messages) */ |
163 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
164 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
165 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
8b9558aa | 166 | static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2; |
31b8006e | 167 | |
a6a5349d SW |
168 | #ifdef CONFIG_LOCKDEP |
169 | static struct lock_class_key socket_class; | |
170 | #endif | |
171 | ||
84495f49 AE |
172 | /* |
173 | * When skipping (ignoring) a block of input we read it into a "skip | |
174 | * buffer," which is this many bytes in size. | |
175 | */ | |
176 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
177 | |
178 | static void queue_con(struct ceph_connection *con); | |
37ab77ac | 179 | static void cancel_con(struct ceph_connection *con); |
68931622 | 180 | static void ceph_con_workfn(struct work_struct *); |
93209264 | 181 | static void con_fault(struct ceph_connection *con); |
31b8006e | 182 | |
31b8006e | 183 | /* |
f64a9317 AE |
184 | * Nicely render a sockaddr as a string. An array of formatted |
185 | * strings is used, to approximate reentrancy. | |
31b8006e | 186 | */ |
f64a9317 AE |
187 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
188 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
189 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
190 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
191 | ||
192 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
193 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 194 | |
57666519 | 195 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 196 | |
3d14c5d2 | 197 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
198 | { |
199 | int i; | |
200 | char *s; | |
99f0f3b2 AE |
201 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
202 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 203 | |
f64a9317 | 204 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
205 | s = addr_str[i]; |
206 | ||
207 | switch (ss->ss_family) { | |
208 | case AF_INET: | |
bd406145 AE |
209 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
210 | ntohs(in4->sin_port)); | |
31b8006e SW |
211 | break; |
212 | ||
213 | case AF_INET6: | |
bd406145 AE |
214 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
215 | ntohs(in6->sin6_port)); | |
31b8006e SW |
216 | break; |
217 | ||
218 | default: | |
d3002b97 AE |
219 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
220 | ss->ss_family); | |
31b8006e SW |
221 | } |
222 | ||
223 | return s; | |
224 | } | |
3d14c5d2 | 225 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 226 | |
63f2d211 SW |
227 | static void encode_my_addr(struct ceph_messenger *msgr) |
228 | { | |
229 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
230 | ceph_encode_addr(&msgr->my_enc_addr); | |
231 | } | |
232 | ||
31b8006e SW |
233 | /* |
234 | * work queue for all reading and writing to/from the socket. | |
235 | */ | |
e0f43c94 | 236 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 237 | |
e3d5d638 AE |
238 | static int ceph_msgr_slab_init(void) |
239 | { | |
240 | BUG_ON(ceph_msg_cache); | |
241 | ceph_msg_cache = kmem_cache_create("ceph_msg", | |
242 | sizeof (struct ceph_msg), | |
243 | __alignof__(struct ceph_msg), 0, NULL); | |
81b36be4 AE |
244 | |
245 | if (!ceph_msg_cache) | |
246 | return -ENOMEM; | |
247 | ||
248 | BUG_ON(ceph_msg_data_cache); | |
249 | ceph_msg_data_cache = kmem_cache_create("ceph_msg_data", | |
250 | sizeof (struct ceph_msg_data), | |
251 | __alignof__(struct ceph_msg_data), | |
252 | 0, NULL); | |
253 | if (ceph_msg_data_cache) | |
254 | return 0; | |
255 | ||
256 | kmem_cache_destroy(ceph_msg_cache); | |
257 | ceph_msg_cache = NULL; | |
258 | ||
259 | return -ENOMEM; | |
e3d5d638 AE |
260 | } |
261 | ||
262 | static void ceph_msgr_slab_exit(void) | |
263 | { | |
81b36be4 AE |
264 | BUG_ON(!ceph_msg_data_cache); |
265 | kmem_cache_destroy(ceph_msg_data_cache); | |
266 | ceph_msg_data_cache = NULL; | |
267 | ||
e3d5d638 AE |
268 | BUG_ON(!ceph_msg_cache); |
269 | kmem_cache_destroy(ceph_msg_cache); | |
270 | ceph_msg_cache = NULL; | |
271 | } | |
272 | ||
15417167 | 273 | static void _ceph_msgr_exit(void) |
6173d1f0 | 274 | { |
d3002b97 | 275 | if (ceph_msgr_wq) { |
6173d1f0 | 276 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
277 | ceph_msgr_wq = NULL; |
278 | } | |
6173d1f0 | 279 | |
6173d1f0 | 280 | BUG_ON(zero_page == NULL); |
6173d1f0 AE |
281 | page_cache_release(zero_page); |
282 | zero_page = NULL; | |
d920ff6f BC |
283 | |
284 | ceph_msgr_slab_exit(); | |
6173d1f0 AE |
285 | } |
286 | ||
3d14c5d2 | 287 | int ceph_msgr_init(void) |
31b8006e | 288 | { |
d920ff6f BC |
289 | if (ceph_msgr_slab_init()) |
290 | return -ENOMEM; | |
291 | ||
57666519 AE |
292 | BUG_ON(zero_page != NULL); |
293 | zero_page = ZERO_PAGE(0); | |
294 | page_cache_get(zero_page); | |
295 | ||
f9865f06 ID |
296 | /* |
297 | * The number of active work items is limited by the number of | |
298 | * connections, so leave @max_active at default. | |
299 | */ | |
300 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0); | |
6173d1f0 AE |
301 | if (ceph_msgr_wq) |
302 | return 0; | |
57666519 | 303 | |
6173d1f0 AE |
304 | pr_err("msgr_init failed to create workqueue\n"); |
305 | _ceph_msgr_exit(); | |
57666519 | 306 | |
6173d1f0 | 307 | return -ENOMEM; |
31b8006e | 308 | } |
3d14c5d2 | 309 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
310 | |
311 | void ceph_msgr_exit(void) | |
312 | { | |
57666519 | 313 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 314 | |
6173d1f0 | 315 | _ceph_msgr_exit(); |
31b8006e | 316 | } |
3d14c5d2 | 317 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 318 | |
cd84db6e | 319 | void ceph_msgr_flush(void) |
a922d38f SW |
320 | { |
321 | flush_workqueue(ceph_msgr_wq); | |
322 | } | |
3d14c5d2 | 323 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 324 | |
ce2c8903 AE |
325 | /* Connection socket state transition functions */ |
326 | ||
327 | static void con_sock_state_init(struct ceph_connection *con) | |
328 | { | |
329 | int old_state; | |
330 | ||
331 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
332 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
333 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
334 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
335 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 AE |
336 | } |
337 | ||
338 | static void con_sock_state_connecting(struct ceph_connection *con) | |
339 | { | |
340 | int old_state; | |
341 | ||
342 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
343 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
344 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
345 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
346 | CON_SOCK_STATE_CONNECTING); | |
ce2c8903 AE |
347 | } |
348 | ||
349 | static void con_sock_state_connected(struct ceph_connection *con) | |
350 | { | |
351 | int old_state; | |
352 | ||
353 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
354 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
355 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
356 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
357 | CON_SOCK_STATE_CONNECTED); | |
ce2c8903 AE |
358 | } |
359 | ||
360 | static void con_sock_state_closing(struct ceph_connection *con) | |
361 | { | |
362 | int old_state; | |
363 | ||
364 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
365 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
366 | old_state != CON_SOCK_STATE_CONNECTED && | |
367 | old_state != CON_SOCK_STATE_CLOSING)) | |
368 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
369 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
370 | CON_SOCK_STATE_CLOSING); | |
ce2c8903 AE |
371 | } |
372 | ||
373 | static void con_sock_state_closed(struct ceph_connection *con) | |
374 | { | |
375 | int old_state; | |
376 | ||
377 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
378 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 | 379 | old_state != CON_SOCK_STATE_CLOSING && |
8007b8d6 SW |
380 | old_state != CON_SOCK_STATE_CONNECTING && |
381 | old_state != CON_SOCK_STATE_CLOSED)) | |
ce2c8903 | 382 | printk("%s: unexpected old state %d\n", __func__, old_state); |
8007b8d6 SW |
383 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
384 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 | 385 | } |
a922d38f | 386 | |
31b8006e SW |
387 | /* |
388 | * socket callback functions | |
389 | */ | |
390 | ||
391 | /* data available on socket, or listen socket received a connect */ | |
676d2369 | 392 | static void ceph_sock_data_ready(struct sock *sk) |
31b8006e | 393 | { |
bd406145 | 394 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
395 | if (atomic_read(&con->msgr->stopping)) { |
396 | return; | |
397 | } | |
bd406145 | 398 | |
31b8006e | 399 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 400 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
401 | con, con->state); |
402 | queue_con(con); | |
403 | } | |
404 | } | |
405 | ||
406 | /* socket has buffer space for writing */ | |
327800bd | 407 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 408 | { |
d3002b97 | 409 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 410 | |
182fac26 JS |
411 | /* only queue to workqueue if there is data we want to write, |
412 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 413 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
414 | * doesn't get called again until try_write() fills the socket |
415 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
416 | * and net/core/stream.c:sk_stream_write_space(). | |
417 | */ | |
c9ffc77a | 418 | if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) { |
64dc6130 | 419 | if (sk_stream_is_writeable(sk)) { |
327800bd | 420 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
421 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
422 | queue_con(con); | |
423 | } | |
31b8006e | 424 | } else { |
327800bd | 425 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 426 | } |
31b8006e SW |
427 | } |
428 | ||
429 | /* socket's state has changed */ | |
327800bd | 430 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 431 | { |
bd406145 | 432 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 433 | |
327800bd | 434 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
435 | con, con->state, sk->sk_state); |
436 | ||
31b8006e SW |
437 | switch (sk->sk_state) { |
438 | case TCP_CLOSE: | |
327800bd | 439 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 440 | case TCP_CLOSE_WAIT: |
327800bd | 441 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 442 | con_sock_state_closing(con); |
c9ffc77a | 443 | con_flag_set(con, CON_FLAG_SOCK_CLOSED); |
d65c9e0b | 444 | queue_con(con); |
31b8006e SW |
445 | break; |
446 | case TCP_ESTABLISHED: | |
327800bd | 447 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 448 | con_sock_state_connected(con); |
31b8006e SW |
449 | queue_con(con); |
450 | break; | |
d3002b97 AE |
451 | default: /* Everything else is uninteresting */ |
452 | break; | |
31b8006e SW |
453 | } |
454 | } | |
455 | ||
456 | /* | |
457 | * set up socket callbacks | |
458 | */ | |
459 | static void set_sock_callbacks(struct socket *sock, | |
460 | struct ceph_connection *con) | |
461 | { | |
462 | struct sock *sk = sock->sk; | |
bd406145 | 463 | sk->sk_user_data = con; |
327800bd AE |
464 | sk->sk_data_ready = ceph_sock_data_ready; |
465 | sk->sk_write_space = ceph_sock_write_space; | |
466 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
467 | } |
468 | ||
469 | ||
470 | /* | |
471 | * socket helpers | |
472 | */ | |
473 | ||
474 | /* | |
475 | * initiate connection to a remote socket. | |
476 | */ | |
41617d0c | 477 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 478 | { |
f91d3471 | 479 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
480 | struct socket *sock; |
481 | int ret; | |
482 | ||
483 | BUG_ON(con->sock); | |
757856d2 | 484 | ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family, |
eeb1bd5c | 485 | SOCK_STREAM, IPPROTO_TCP, &sock); |
31b8006e | 486 | if (ret) |
41617d0c | 487 | return ret; |
6d7fdb0a | 488 | sock->sk->sk_allocation = GFP_NOFS; |
31b8006e | 489 | |
a6a5349d SW |
490 | #ifdef CONFIG_LOCKDEP |
491 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
492 | #endif | |
493 | ||
31b8006e SW |
494 | set_sock_callbacks(sock, con); |
495 | ||
3d14c5d2 | 496 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 497 | |
89a86be0 | 498 | con_sock_state_connecting(con); |
f91d3471 SW |
499 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
500 | O_NONBLOCK); | |
31b8006e SW |
501 | if (ret == -EINPROGRESS) { |
502 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 503 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 504 | sock->sk->sk_state); |
a5bc3129 | 505 | } else if (ret < 0) { |
31b8006e | 506 | pr_err("connect %s error %d\n", |
3d14c5d2 | 507 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 508 | sock_release(sock); |
41617d0c | 509 | return ret; |
a5bc3129 | 510 | } |
89baaa57 | 511 | |
859bff51 | 512 | if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) { |
ba988f87 CH |
513 | int optval = 1; |
514 | ||
515 | ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, | |
516 | (char *)&optval, sizeof(optval)); | |
517 | if (ret) | |
518 | pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d", | |
519 | ret); | |
520 | } | |
521 | ||
a5bc3129 | 522 | con->sock = sock; |
41617d0c | 523 | return 0; |
31b8006e SW |
524 | } |
525 | ||
526 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
527 | { | |
528 | struct kvec iov = {buf, len}; | |
529 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 530 | int r; |
31b8006e | 531 | |
98bdb0aa SW |
532 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
533 | if (r == -EAGAIN) | |
534 | r = 0; | |
535 | return r; | |
31b8006e SW |
536 | } |
537 | ||
afb3d90e AE |
538 | static int ceph_tcp_recvpage(struct socket *sock, struct page *page, |
539 | int page_offset, size_t length) | |
540 | { | |
541 | void *kaddr; | |
542 | int ret; | |
543 | ||
544 | BUG_ON(page_offset + length > PAGE_SIZE); | |
545 | ||
546 | kaddr = kmap(page); | |
547 | BUG_ON(!kaddr); | |
548 | ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length); | |
549 | kunmap(page); | |
550 | ||
551 | return ret; | |
552 | } | |
553 | ||
31b8006e SW |
554 | /* |
555 | * write something. @more is true if caller will be sending more data | |
556 | * shortly. | |
557 | */ | |
558 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
559 | size_t kvlen, size_t len, int more) | |
560 | { | |
561 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 562 | int r; |
31b8006e SW |
563 | |
564 | if (more) | |
565 | msg.msg_flags |= MSG_MORE; | |
566 | else | |
567 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
568 | ||
42961d23 SW |
569 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
570 | if (r == -EAGAIN) | |
571 | r = 0; | |
572 | return r; | |
31b8006e SW |
573 | } |
574 | ||
178eda29 | 575 | static int __ceph_tcp_sendpage(struct socket *sock, struct page *page, |
e1dcb128 | 576 | int offset, size_t size, bool more) |
31739139 AE |
577 | { |
578 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
579 | int ret; | |
580 | ||
581 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
582 | if (ret == -EAGAIN) | |
583 | ret = 0; | |
584 | ||
585 | return ret; | |
586 | } | |
587 | ||
178eda29 CC |
588 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
589 | int offset, size_t size, bool more) | |
590 | { | |
591 | int ret; | |
592 | struct kvec iov; | |
593 | ||
594 | /* sendpage cannot properly handle pages with page_count == 0, | |
595 | * we need to fallback to sendmsg if that's the case */ | |
596 | if (page_count(page) >= 1) | |
597 | return __ceph_tcp_sendpage(sock, page, offset, size, more); | |
598 | ||
599 | iov.iov_base = kmap(page) + offset; | |
600 | iov.iov_len = size; | |
601 | ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more); | |
602 | kunmap(page); | |
603 | ||
604 | return ret; | |
605 | } | |
31b8006e SW |
606 | |
607 | /* | |
608 | * Shutdown/close the socket for the given connection. | |
609 | */ | |
610 | static int con_close_socket(struct ceph_connection *con) | |
611 | { | |
8007b8d6 | 612 | int rc = 0; |
31b8006e SW |
613 | |
614 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
8007b8d6 SW |
615 | if (con->sock) { |
616 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
617 | sock_release(con->sock); | |
618 | con->sock = NULL; | |
619 | } | |
456ea468 AE |
620 | |
621 | /* | |
4a861692 | 622 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
623 | * independent of the connection mutex, and we could have |
624 | * received a socket close event before we had the chance to | |
625 | * shut the socket down. | |
626 | */ | |
c9ffc77a | 627 | con_flag_clear(con, CON_FLAG_SOCK_CLOSED); |
8007b8d6 | 628 | |
ce2c8903 | 629 | con_sock_state_closed(con); |
31b8006e SW |
630 | return rc; |
631 | } | |
632 | ||
633 | /* | |
634 | * Reset a connection. Discard all incoming and outgoing messages | |
635 | * and clear *_seq state. | |
636 | */ | |
637 | static void ceph_msg_remove(struct ceph_msg *msg) | |
638 | { | |
639 | list_del_init(&msg->list_head); | |
38941f80 | 640 | |
31b8006e SW |
641 | ceph_msg_put(msg); |
642 | } | |
643 | static void ceph_msg_remove_list(struct list_head *head) | |
644 | { | |
645 | while (!list_empty(head)) { | |
646 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
647 | list_head); | |
648 | ceph_msg_remove(msg); | |
649 | } | |
650 | } | |
651 | ||
652 | static void reset_connection(struct ceph_connection *con) | |
653 | { | |
654 | /* reset connection, out_queue, msg_ and connect_seq */ | |
655 | /* discard existing out_queue and msg_seq */ | |
0fa6ebc6 | 656 | dout("reset_connection %p\n", con); |
31b8006e SW |
657 | ceph_msg_remove_list(&con->out_queue); |
658 | ceph_msg_remove_list(&con->out_sent); | |
659 | ||
cf3e5c40 | 660 | if (con->in_msg) { |
38941f80 | 661 | BUG_ON(con->in_msg->con != con); |
cf3e5c40 SW |
662 | ceph_msg_put(con->in_msg); |
663 | con->in_msg = NULL; | |
664 | } | |
665 | ||
31b8006e SW |
666 | con->connect_seq = 0; |
667 | con->out_seq = 0; | |
c86a2930 | 668 | if (con->out_msg) { |
583d0fef | 669 | BUG_ON(con->out_msg->con != con); |
c86a2930 SW |
670 | ceph_msg_put(con->out_msg); |
671 | con->out_msg = NULL; | |
672 | } | |
31b8006e | 673 | con->in_seq = 0; |
0e0d5e0c | 674 | con->in_seq_acked = 0; |
31b8006e SW |
675 | } |
676 | ||
677 | /* | |
678 | * mark a peer down. drop any open connections. | |
679 | */ | |
680 | void ceph_con_close(struct ceph_connection *con) | |
681 | { | |
8c50c817 | 682 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
683 | dout("con_close %p peer %s\n", con, |
684 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
8dacc7da | 685 | con->state = CON_STATE_CLOSED; |
a5988c49 | 686 | |
c9ffc77a AE |
687 | con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */ |
688 | con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING); | |
689 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); | |
690 | con_flag_clear(con, CON_FLAG_BACKOFF); | |
a5988c49 | 691 | |
31b8006e | 692 | reset_connection(con); |
6f2bc3ff | 693 | con->peer_global_seq = 0; |
37ab77ac | 694 | cancel_con(con); |
ee76e073 | 695 | con_close_socket(con); |
ec302645 | 696 | mutex_unlock(&con->mutex); |
31b8006e | 697 | } |
3d14c5d2 | 698 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 699 | |
31b8006e SW |
700 | /* |
701 | * Reopen a closed connection, with a new peer address. | |
702 | */ | |
b7a9e5dd SW |
703 | void ceph_con_open(struct ceph_connection *con, |
704 | __u8 entity_type, __u64 entity_num, | |
705 | struct ceph_entity_addr *addr) | |
31b8006e | 706 | { |
5469155f | 707 | mutex_lock(&con->mutex); |
3d14c5d2 | 708 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
8dacc7da | 709 | |
122070a2 | 710 | WARN_ON(con->state != CON_STATE_CLOSED); |
8dacc7da | 711 | con->state = CON_STATE_PREOPEN; |
a5988c49 | 712 | |
b7a9e5dd SW |
713 | con->peer_name.type = (__u8) entity_type; |
714 | con->peer_name.num = cpu_to_le64(entity_num); | |
715 | ||
31b8006e | 716 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 717 | con->delay = 0; /* reset backoff memory */ |
5469155f | 718 | mutex_unlock(&con->mutex); |
31b8006e SW |
719 | queue_con(con); |
720 | } | |
3d14c5d2 | 721 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 722 | |
87b315a5 SW |
723 | /* |
724 | * return true if this connection ever successfully opened | |
725 | */ | |
726 | bool ceph_con_opened(struct ceph_connection *con) | |
727 | { | |
728 | return con->connect_seq > 0; | |
729 | } | |
730 | ||
31b8006e SW |
731 | /* |
732 | * initialize a new connection. | |
733 | */ | |
1bfd89f4 AE |
734 | void ceph_con_init(struct ceph_connection *con, void *private, |
735 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 736 | struct ceph_messenger *msgr) |
31b8006e SW |
737 | { |
738 | dout("con_init %p\n", con); | |
739 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
740 | con->private = private; |
741 | con->ops = ops; | |
31b8006e | 742 | con->msgr = msgr; |
ce2c8903 AE |
743 | |
744 | con_sock_state_init(con); | |
745 | ||
ec302645 | 746 | mutex_init(&con->mutex); |
31b8006e SW |
747 | INIT_LIST_HEAD(&con->out_queue); |
748 | INIT_LIST_HEAD(&con->out_sent); | |
68931622 | 749 | INIT_DELAYED_WORK(&con->work, ceph_con_workfn); |
a5988c49 | 750 | |
8dacc7da | 751 | con->state = CON_STATE_CLOSED; |
31b8006e | 752 | } |
3d14c5d2 | 753 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
754 | |
755 | ||
756 | /* | |
757 | * We maintain a global counter to order connection attempts. Get | |
758 | * a unique seq greater than @gt. | |
759 | */ | |
760 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
761 | { | |
762 | u32 ret; | |
763 | ||
764 | spin_lock(&msgr->global_seq_lock); | |
765 | if (msgr->global_seq < gt) | |
766 | msgr->global_seq = gt; | |
767 | ret = ++msgr->global_seq; | |
768 | spin_unlock(&msgr->global_seq_lock); | |
769 | return ret; | |
770 | } | |
771 | ||
e2200423 | 772 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
773 | { |
774 | con->out_kvec_left = 0; | |
775 | con->out_kvec_bytes = 0; | |
776 | con->out_kvec_cur = &con->out_kvec[0]; | |
777 | } | |
778 | ||
e2200423 | 779 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
780 | size_t size, void *data) |
781 | { | |
782 | int index; | |
783 | ||
784 | index = con->out_kvec_left; | |
785 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
786 | ||
787 | con->out_kvec[index].iov_len = size; | |
788 | con->out_kvec[index].iov_base = data; | |
789 | con->out_kvec_left++; | |
790 | con->out_kvec_bytes += size; | |
791 | } | |
31b8006e | 792 | |
df6ad1f9 | 793 | #ifdef CONFIG_BLOCK |
6aaa4511 AE |
794 | |
795 | /* | |
796 | * For a bio data item, a piece is whatever remains of the next | |
797 | * entry in the current bio iovec, or the first entry in the next | |
798 | * bio in the list. | |
799 | */ | |
8ae4f4f5 | 800 | static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 801 | size_t length) |
6aaa4511 | 802 | { |
8ae4f4f5 | 803 | struct ceph_msg_data *data = cursor->data; |
6aaa4511 AE |
804 | struct bio *bio; |
805 | ||
806 | BUG_ON(data->type != CEPH_MSG_DATA_BIO); | |
807 | ||
808 | bio = data->bio; | |
809 | BUG_ON(!bio); | |
6aaa4511 | 810 | |
ca8b3a69 | 811 | cursor->resid = min(length, data->bio_length); |
6aaa4511 | 812 | cursor->bio = bio; |
f38a5181 KO |
813 | cursor->bvec_iter = bio->bi_iter; |
814 | cursor->last_piece = | |
815 | cursor->resid <= bio_iter_len(bio, cursor->bvec_iter); | |
6aaa4511 AE |
816 | } |
817 | ||
8ae4f4f5 | 818 | static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor, |
6aaa4511 AE |
819 | size_t *page_offset, |
820 | size_t *length) | |
821 | { | |
8ae4f4f5 | 822 | struct ceph_msg_data *data = cursor->data; |
6aaa4511 | 823 | struct bio *bio; |
f38a5181 | 824 | struct bio_vec bio_vec; |
6aaa4511 AE |
825 | |
826 | BUG_ON(data->type != CEPH_MSG_DATA_BIO); | |
827 | ||
828 | bio = cursor->bio; | |
829 | BUG_ON(!bio); | |
830 | ||
f38a5181 | 831 | bio_vec = bio_iter_iovec(bio, cursor->bvec_iter); |
6aaa4511 | 832 | |
f38a5181 | 833 | *page_offset = (size_t) bio_vec.bv_offset; |
6aaa4511 | 834 | BUG_ON(*page_offset >= PAGE_SIZE); |
25aff7c5 AE |
835 | if (cursor->last_piece) /* pagelist offset is always 0 */ |
836 | *length = cursor->resid; | |
837 | else | |
f38a5181 | 838 | *length = (size_t) bio_vec.bv_len; |
25aff7c5 | 839 | BUG_ON(*length > cursor->resid); |
5df521b1 | 840 | BUG_ON(*page_offset + *length > PAGE_SIZE); |
6aaa4511 | 841 | |
f38a5181 | 842 | return bio_vec.bv_page; |
6aaa4511 AE |
843 | } |
844 | ||
8ae4f4f5 AE |
845 | static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor, |
846 | size_t bytes) | |
6aaa4511 | 847 | { |
6aaa4511 | 848 | struct bio *bio; |
f38a5181 | 849 | struct bio_vec bio_vec; |
6aaa4511 | 850 | |
8ae4f4f5 | 851 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO); |
6aaa4511 AE |
852 | |
853 | bio = cursor->bio; | |
854 | BUG_ON(!bio); | |
855 | ||
f38a5181 | 856 | bio_vec = bio_iter_iovec(bio, cursor->bvec_iter); |
6aaa4511 AE |
857 | |
858 | /* Advance the cursor offset */ | |
859 | ||
25aff7c5 AE |
860 | BUG_ON(cursor->resid < bytes); |
861 | cursor->resid -= bytes; | |
f38a5181 KO |
862 | |
863 | bio_advance_iter(bio, &cursor->bvec_iter, bytes); | |
864 | ||
865 | if (bytes < bio_vec.bv_len) | |
6aaa4511 AE |
866 | return false; /* more bytes to process in this segment */ |
867 | ||
868 | /* Move on to the next segment, and possibly the next bio */ | |
869 | ||
f38a5181 | 870 | if (!cursor->bvec_iter.bi_size) { |
6aaa4511 | 871 | bio = bio->bi_next; |
0ec1d15e ID |
872 | cursor->bio = bio; |
873 | if (bio) | |
874 | cursor->bvec_iter = bio->bi_iter; | |
875 | else | |
876 | memset(&cursor->bvec_iter, 0, | |
877 | sizeof(cursor->bvec_iter)); | |
6aaa4511 | 878 | } |
6aaa4511 | 879 | |
25aff7c5 AE |
880 | if (!cursor->last_piece) { |
881 | BUG_ON(!cursor->resid); | |
882 | BUG_ON(!bio); | |
883 | /* A short read is OK, so use <= rather than == */ | |
f38a5181 | 884 | if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter)) |
6aaa4511 | 885 | cursor->last_piece = true; |
25aff7c5 | 886 | } |
6aaa4511 AE |
887 | |
888 | return true; | |
889 | } | |
ea96571f | 890 | #endif /* CONFIG_BLOCK */ |
df6ad1f9 | 891 | |
e766d7b5 AE |
892 | /* |
893 | * For a page array, a piece comes from the first page in the array | |
894 | * that has not already been fully consumed. | |
895 | */ | |
8ae4f4f5 | 896 | static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 897 | size_t length) |
e766d7b5 | 898 | { |
8ae4f4f5 | 899 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
900 | int page_count; |
901 | ||
902 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
903 | ||
904 | BUG_ON(!data->pages); | |
905 | BUG_ON(!data->length); | |
906 | ||
ca8b3a69 | 907 | cursor->resid = min(length, data->length); |
e766d7b5 | 908 | page_count = calc_pages_for(data->alignment, (u64)data->length); |
e766d7b5 AE |
909 | cursor->page_offset = data->alignment & ~PAGE_MASK; |
910 | cursor->page_index = 0; | |
56fc5659 AE |
911 | BUG_ON(page_count > (int)USHRT_MAX); |
912 | cursor->page_count = (unsigned short)page_count; | |
913 | BUG_ON(length > SIZE_MAX - cursor->page_offset); | |
5f740d7e | 914 | cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
915 | } |
916 | ||
8ae4f4f5 AE |
917 | static struct page * |
918 | ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor, | |
919 | size_t *page_offset, size_t *length) | |
e766d7b5 | 920 | { |
8ae4f4f5 | 921 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
922 | |
923 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
924 | ||
925 | BUG_ON(cursor->page_index >= cursor->page_count); | |
926 | BUG_ON(cursor->page_offset >= PAGE_SIZE); | |
e766d7b5 AE |
927 | |
928 | *page_offset = cursor->page_offset; | |
25aff7c5 | 929 | if (cursor->last_piece) |
e766d7b5 | 930 | *length = cursor->resid; |
25aff7c5 | 931 | else |
e766d7b5 | 932 | *length = PAGE_SIZE - *page_offset; |
e766d7b5 AE |
933 | |
934 | return data->pages[cursor->page_index]; | |
935 | } | |
936 | ||
8ae4f4f5 | 937 | static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor, |
e766d7b5 AE |
938 | size_t bytes) |
939 | { | |
8ae4f4f5 | 940 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES); |
e766d7b5 AE |
941 | |
942 | BUG_ON(cursor->page_offset + bytes > PAGE_SIZE); | |
e766d7b5 AE |
943 | |
944 | /* Advance the cursor page offset */ | |
945 | ||
946 | cursor->resid -= bytes; | |
5df521b1 AE |
947 | cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK; |
948 | if (!bytes || cursor->page_offset) | |
e766d7b5 AE |
949 | return false; /* more bytes to process in the current page */ |
950 | ||
d90deda6 YZ |
951 | if (!cursor->resid) |
952 | return false; /* no more data */ | |
953 | ||
5df521b1 | 954 | /* Move on to the next page; offset is already at 0 */ |
e766d7b5 AE |
955 | |
956 | BUG_ON(cursor->page_index >= cursor->page_count); | |
e766d7b5 | 957 | cursor->page_index++; |
25aff7c5 | 958 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
959 | |
960 | return true; | |
961 | } | |
962 | ||
fe38a2b6 | 963 | /* |
dd236fcb AE |
964 | * For a pagelist, a piece is whatever remains to be consumed in the |
965 | * first page in the list, or the front of the next page. | |
fe38a2b6 | 966 | */ |
8ae4f4f5 AE |
967 | static void |
968 | ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor, | |
25aff7c5 | 969 | size_t length) |
fe38a2b6 | 970 | { |
8ae4f4f5 | 971 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
972 | struct ceph_pagelist *pagelist; |
973 | struct page *page; | |
974 | ||
dd236fcb | 975 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); |
fe38a2b6 AE |
976 | |
977 | pagelist = data->pagelist; | |
978 | BUG_ON(!pagelist); | |
25aff7c5 AE |
979 | |
980 | if (!length) | |
fe38a2b6 AE |
981 | return; /* pagelist can be assigned but empty */ |
982 | ||
983 | BUG_ON(list_empty(&pagelist->head)); | |
984 | page = list_first_entry(&pagelist->head, struct page, lru); | |
985 | ||
ca8b3a69 | 986 | cursor->resid = min(length, pagelist->length); |
fe38a2b6 AE |
987 | cursor->page = page; |
988 | cursor->offset = 0; | |
a51b272e | 989 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
990 | } |
991 | ||
8ae4f4f5 AE |
992 | static struct page * |
993 | ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor, | |
994 | size_t *page_offset, size_t *length) | |
fe38a2b6 | 995 | { |
8ae4f4f5 | 996 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 | 997 | struct ceph_pagelist *pagelist; |
fe38a2b6 AE |
998 | |
999 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
1000 | ||
1001 | pagelist = data->pagelist; | |
1002 | BUG_ON(!pagelist); | |
1003 | ||
1004 | BUG_ON(!cursor->page); | |
25aff7c5 | 1005 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); |
fe38a2b6 | 1006 | |
5df521b1 | 1007 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 | 1008 | *page_offset = cursor->offset & ~PAGE_MASK; |
5df521b1 | 1009 | if (cursor->last_piece) |
25aff7c5 AE |
1010 | *length = cursor->resid; |
1011 | else | |
1012 | *length = PAGE_SIZE - *page_offset; | |
fe38a2b6 | 1013 | |
8ae4f4f5 | 1014 | return cursor->page; |
fe38a2b6 AE |
1015 | } |
1016 | ||
8ae4f4f5 | 1017 | static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor, |
dd236fcb | 1018 | size_t bytes) |
fe38a2b6 | 1019 | { |
8ae4f4f5 | 1020 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
1021 | struct ceph_pagelist *pagelist; |
1022 | ||
1023 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
1024 | ||
1025 | pagelist = data->pagelist; | |
1026 | BUG_ON(!pagelist); | |
25aff7c5 AE |
1027 | |
1028 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); | |
fe38a2b6 AE |
1029 | BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE); |
1030 | ||
1031 | /* Advance the cursor offset */ | |
1032 | ||
25aff7c5 | 1033 | cursor->resid -= bytes; |
fe38a2b6 | 1034 | cursor->offset += bytes; |
5df521b1 | 1035 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 AE |
1036 | if (!bytes || cursor->offset & ~PAGE_MASK) |
1037 | return false; /* more bytes to process in the current page */ | |
1038 | ||
d90deda6 YZ |
1039 | if (!cursor->resid) |
1040 | return false; /* no more data */ | |
1041 | ||
fe38a2b6 AE |
1042 | /* Move on to the next page */ |
1043 | ||
1044 | BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head)); | |
1045 | cursor->page = list_entry_next(cursor->page, lru); | |
25aff7c5 | 1046 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
1047 | |
1048 | return true; | |
1049 | } | |
1050 | ||
dd236fcb AE |
1051 | /* |
1052 | * Message data is handled (sent or received) in pieces, where each | |
1053 | * piece resides on a single page. The network layer might not | |
1054 | * consume an entire piece at once. A data item's cursor keeps | |
1055 | * track of which piece is next to process and how much remains to | |
1056 | * be processed in that piece. It also tracks whether the current | |
1057 | * piece is the last one in the data item. | |
1058 | */ | |
ca8b3a69 | 1059 | static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor) |
dd236fcb | 1060 | { |
ca8b3a69 | 1061 | size_t length = cursor->total_resid; |
8ae4f4f5 | 1062 | |
8ae4f4f5 | 1063 | switch (cursor->data->type) { |
dd236fcb | 1064 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1065 | ceph_msg_data_pagelist_cursor_init(cursor, length); |
dd236fcb | 1066 | break; |
e766d7b5 | 1067 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1068 | ceph_msg_data_pages_cursor_init(cursor, length); |
e766d7b5 | 1069 | break; |
dd236fcb AE |
1070 | #ifdef CONFIG_BLOCK |
1071 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1072 | ceph_msg_data_bio_cursor_init(cursor, length); |
6aaa4511 | 1073 | break; |
dd236fcb | 1074 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1075 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1076 | default: |
1077 | /* BUG(); */ | |
1078 | break; | |
1079 | } | |
8ae4f4f5 | 1080 | cursor->need_crc = true; |
dd236fcb AE |
1081 | } |
1082 | ||
ca8b3a69 AE |
1083 | static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length) |
1084 | { | |
1085 | struct ceph_msg_data_cursor *cursor = &msg->cursor; | |
1086 | struct ceph_msg_data *data; | |
1087 | ||
1088 | BUG_ON(!length); | |
1089 | BUG_ON(length > msg->data_length); | |
1090 | BUG_ON(list_empty(&msg->data)); | |
1091 | ||
ca8b3a69 AE |
1092 | cursor->data_head = &msg->data; |
1093 | cursor->total_resid = length; | |
1094 | data = list_first_entry(&msg->data, struct ceph_msg_data, links); | |
1095 | cursor->data = data; | |
1096 | ||
1097 | __ceph_msg_data_cursor_init(cursor); | |
1098 | } | |
1099 | ||
dd236fcb AE |
1100 | /* |
1101 | * Return the page containing the next piece to process for a given | |
1102 | * data item, and supply the page offset and length of that piece. | |
1103 | * Indicate whether this is the last piece in this data item. | |
1104 | */ | |
8ae4f4f5 AE |
1105 | static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor, |
1106 | size_t *page_offset, size_t *length, | |
dd236fcb AE |
1107 | bool *last_piece) |
1108 | { | |
1109 | struct page *page; | |
1110 | ||
8ae4f4f5 | 1111 | switch (cursor->data->type) { |
dd236fcb | 1112 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1113 | page = ceph_msg_data_pagelist_next(cursor, page_offset, length); |
dd236fcb | 1114 | break; |
e766d7b5 | 1115 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1116 | page = ceph_msg_data_pages_next(cursor, page_offset, length); |
e766d7b5 | 1117 | break; |
dd236fcb AE |
1118 | #ifdef CONFIG_BLOCK |
1119 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1120 | page = ceph_msg_data_bio_next(cursor, page_offset, length); |
6aaa4511 | 1121 | break; |
dd236fcb | 1122 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1123 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1124 | default: |
1125 | page = NULL; | |
1126 | break; | |
1127 | } | |
1128 | BUG_ON(!page); | |
1129 | BUG_ON(*page_offset + *length > PAGE_SIZE); | |
1130 | BUG_ON(!*length); | |
1131 | if (last_piece) | |
8ae4f4f5 | 1132 | *last_piece = cursor->last_piece; |
dd236fcb AE |
1133 | |
1134 | return page; | |
1135 | } | |
1136 | ||
1137 | /* | |
1138 | * Returns true if the result moves the cursor on to the next piece | |
1139 | * of the data item. | |
1140 | */ | |
8ae4f4f5 AE |
1141 | static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, |
1142 | size_t bytes) | |
dd236fcb AE |
1143 | { |
1144 | bool new_piece; | |
1145 | ||
25aff7c5 | 1146 | BUG_ON(bytes > cursor->resid); |
8ae4f4f5 | 1147 | switch (cursor->data->type) { |
dd236fcb | 1148 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1149 | new_piece = ceph_msg_data_pagelist_advance(cursor, bytes); |
dd236fcb | 1150 | break; |
e766d7b5 | 1151 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1152 | new_piece = ceph_msg_data_pages_advance(cursor, bytes); |
e766d7b5 | 1153 | break; |
dd236fcb AE |
1154 | #ifdef CONFIG_BLOCK |
1155 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1156 | new_piece = ceph_msg_data_bio_advance(cursor, bytes); |
6aaa4511 | 1157 | break; |
dd236fcb | 1158 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1159 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1160 | default: |
1161 | BUG(); | |
1162 | break; | |
1163 | } | |
ca8b3a69 | 1164 | cursor->total_resid -= bytes; |
dd236fcb | 1165 | |
ca8b3a69 AE |
1166 | if (!cursor->resid && cursor->total_resid) { |
1167 | WARN_ON(!cursor->last_piece); | |
1168 | BUG_ON(list_is_last(&cursor->data->links, cursor->data_head)); | |
1169 | cursor->data = list_entry_next(cursor->data, links); | |
1170 | __ceph_msg_data_cursor_init(cursor); | |
a51b272e | 1171 | new_piece = true; |
ca8b3a69 | 1172 | } |
a51b272e | 1173 | cursor->need_crc = new_piece; |
ca8b3a69 | 1174 | |
dd236fcb AE |
1175 | return new_piece; |
1176 | } | |
1177 | ||
98fa5dd8 | 1178 | static void prepare_message_data(struct ceph_msg *msg, u32 data_len) |
739c905b | 1179 | { |
739c905b | 1180 | BUG_ON(!msg); |
25aff7c5 | 1181 | BUG_ON(!data_len); |
739c905b | 1182 | |
4c59b4a2 | 1183 | /* Initialize data cursor */ |
fe38a2b6 | 1184 | |
8ae4f4f5 | 1185 | ceph_msg_data_cursor_init(msg, (size_t)data_len); |
739c905b AE |
1186 | } |
1187 | ||
31b8006e SW |
1188 | /* |
1189 | * Prepare footer for currently outgoing message, and finish things | |
1190 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
1191 | */ | |
859eb799 | 1192 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
1193 | { |
1194 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 1195 | int v = con->out_kvec_left; |
31b8006e | 1196 | |
fd154f3c AE |
1197 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
1198 | ||
31b8006e SW |
1199 | dout("prepare_write_message_footer %p\n", con); |
1200 | con->out_kvec_is_msg = true; | |
1201 | con->out_kvec[v].iov_base = &m->footer; | |
33d07337 YZ |
1202 | if (con->peer_features & CEPH_FEATURE_MSG_AUTH) { |
1203 | if (con->ops->sign_message) | |
79dbd1ba | 1204 | con->ops->sign_message(m); |
33d07337 YZ |
1205 | else |
1206 | m->footer.sig = 0; | |
1207 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
1208 | con->out_kvec_bytes += sizeof(m->footer); | |
1209 | } else { | |
1210 | m->old_footer.flags = m->footer.flags; | |
1211 | con->out_kvec[v].iov_len = sizeof(m->old_footer); | |
1212 | con->out_kvec_bytes += sizeof(m->old_footer); | |
1213 | } | |
31b8006e SW |
1214 | con->out_kvec_left++; |
1215 | con->out_more = m->more_to_follow; | |
c86a2930 | 1216 | con->out_msg_done = true; |
31b8006e SW |
1217 | } |
1218 | ||
1219 | /* | |
1220 | * Prepare headers for the next outgoing message. | |
1221 | */ | |
1222 | static void prepare_write_message(struct ceph_connection *con) | |
1223 | { | |
1224 | struct ceph_msg *m; | |
a9a0c51a | 1225 | u32 crc; |
31b8006e | 1226 | |
e2200423 | 1227 | con_out_kvec_reset(con); |
31b8006e | 1228 | con->out_kvec_is_msg = true; |
c86a2930 | 1229 | con->out_msg_done = false; |
31b8006e SW |
1230 | |
1231 | /* Sneak an ack in there first? If we can get it into the same | |
1232 | * TCP packet that's a good thing. */ | |
1233 | if (con->in_seq > con->in_seq_acked) { | |
1234 | con->in_seq_acked = con->in_seq; | |
e2200423 | 1235 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 1236 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1237 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 1238 | &con->out_temp_ack); |
31b8006e SW |
1239 | } |
1240 | ||
38941f80 | 1241 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 1242 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 1243 | con->out_msg = m; |
38941f80 | 1244 | BUG_ON(m->con != con); |
4cf9d544 SW |
1245 | |
1246 | /* put message on sent list */ | |
1247 | ceph_msg_get(m); | |
1248 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 1249 | |
e84346b7 SW |
1250 | /* |
1251 | * only assign outgoing seq # if we haven't sent this message | |
1252 | * yet. if it is requeued, resend with it's original seq. | |
1253 | */ | |
1254 | if (m->needs_out_seq) { | |
1255 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
1256 | m->needs_out_seq = false; | |
1257 | } | |
98fa5dd8 | 1258 | WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len)); |
31b8006e | 1259 | |
98fa5dd8 | 1260 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n", |
31b8006e SW |
1261 | m, con->out_seq, le16_to_cpu(m->hdr.type), |
1262 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
98fa5dd8 | 1263 | m->data_length); |
31b8006e SW |
1264 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); |
1265 | ||
1266 | /* tag + hdr + front + middle */ | |
e2200423 AE |
1267 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
1268 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
1269 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 1270 | |
31b8006e | 1271 | if (m->middle) |
e2200423 | 1272 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 1273 | m->middle->vec.iov_base); |
31b8006e SW |
1274 | |
1275 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
1276 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
1277 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 1278 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
1279 | |
1280 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
1281 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
1282 | if (m->middle) { | |
1283 | crc = crc32c(0, m->middle->vec.iov_base, | |
1284 | m->middle->vec.iov_len); | |
1285 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
1286 | } else | |
31b8006e | 1287 | con->out_msg->footer.middle_crc = 0; |
739c905b | 1288 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
1289 | le32_to_cpu(con->out_msg->footer.front_crc), |
1290 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
1291 | ||
1292 | /* is there a data payload? */ | |
739c905b | 1293 | con->out_msg->footer.data_crc = 0; |
98fa5dd8 AE |
1294 | if (m->data_length) { |
1295 | prepare_message_data(con->out_msg, m->data_length); | |
78625051 AE |
1296 | con->out_more = 1; /* data + footer will follow */ |
1297 | } else { | |
31b8006e | 1298 | /* no, queue up footer too and be done */ |
859eb799 | 1299 | prepare_write_message_footer(con); |
78625051 | 1300 | } |
31b8006e | 1301 | |
c9ffc77a | 1302 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1303 | } |
1304 | ||
1305 | /* | |
1306 | * Prepare an ack. | |
1307 | */ | |
1308 | static void prepare_write_ack(struct ceph_connection *con) | |
1309 | { | |
1310 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
1311 | con->in_seq_acked, con->in_seq); | |
1312 | con->in_seq_acked = con->in_seq; | |
1313 | ||
e2200423 | 1314 | con_out_kvec_reset(con); |
859eb799 | 1315 | |
e2200423 | 1316 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 1317 | |
31b8006e | 1318 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1319 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
1320 | &con->out_temp_ack); |
1321 | ||
31b8006e | 1322 | con->out_more = 1; /* more will follow.. eventually.. */ |
c9ffc77a | 1323 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1324 | } |
1325 | ||
3a23083b SW |
1326 | /* |
1327 | * Prepare to share the seq during handshake | |
1328 | */ | |
1329 | static void prepare_write_seq(struct ceph_connection *con) | |
1330 | { | |
1331 | dout("prepare_write_seq %p %llu -> %llu\n", con, | |
1332 | con->in_seq_acked, con->in_seq); | |
1333 | con->in_seq_acked = con->in_seq; | |
1334 | ||
1335 | con_out_kvec_reset(con); | |
1336 | ||
1337 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); | |
1338 | con_out_kvec_add(con, sizeof (con->out_temp_ack), | |
1339 | &con->out_temp_ack); | |
1340 | ||
1341 | con_flag_set(con, CON_FLAG_WRITE_PENDING); | |
1342 | } | |
1343 | ||
31b8006e SW |
1344 | /* |
1345 | * Prepare to write keepalive byte. | |
1346 | */ | |
1347 | static void prepare_write_keepalive(struct ceph_connection *con) | |
1348 | { | |
1349 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 | 1350 | con_out_kvec_reset(con); |
8b9558aa | 1351 | if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) { |
7f61f545 ID |
1352 | struct timespec now = CURRENT_TIME; |
1353 | ||
8b9558aa | 1354 | con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2); |
7f61f545 ID |
1355 | ceph_encode_timespec(&con->out_temp_keepalive2, &now); |
1356 | con_out_kvec_add(con, sizeof(con->out_temp_keepalive2), | |
1357 | &con->out_temp_keepalive2); | |
8b9558aa YZ |
1358 | } else { |
1359 | con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive); | |
1360 | } | |
c9ffc77a | 1361 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1362 | } |
1363 | ||
1364 | /* | |
1365 | * Connection negotiation. | |
1366 | */ | |
1367 | ||
dac1e716 AE |
1368 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
1369 | int *auth_proto) | |
4e7a5dcd | 1370 | { |
a3530df3 | 1371 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
1372 | |
1373 | if (!con->ops->get_authorizer) { | |
1374 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
1375 | con->out_connect.authorizer_len = 0; | |
729796be | 1376 | return NULL; |
b1c6b980 AE |
1377 | } |
1378 | ||
1379 | /* Can't hold the mutex while getting authorizer */ | |
ec302645 | 1380 | mutex_unlock(&con->mutex); |
dac1e716 | 1381 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
ec302645 | 1382 | mutex_lock(&con->mutex); |
4e7a5dcd | 1383 | |
a3530df3 | 1384 | if (IS_ERR(auth)) |
729796be | 1385 | return auth; |
8dacc7da | 1386 | if (con->state != CON_STATE_NEGOTIATING) |
729796be | 1387 | return ERR_PTR(-EAGAIN); |
0da5d703 | 1388 | |
8f43fb53 AE |
1389 | con->auth_reply_buf = auth->authorizer_reply_buf; |
1390 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
729796be | 1391 | return auth; |
4e7a5dcd SW |
1392 | } |
1393 | ||
31b8006e SW |
1394 | /* |
1395 | * We connected to a peer and are saying hello. | |
1396 | */ | |
e825a66d | 1397 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 1398 | { |
e2200423 AE |
1399 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
1400 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 1401 | &con->msgr->my_enc_addr); |
eed0ef2c | 1402 | |
eed0ef2c | 1403 | con->out_more = 0; |
c9ffc77a | 1404 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
eed0ef2c SW |
1405 | } |
1406 | ||
e825a66d | 1407 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 1408 | { |
95c96174 | 1409 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 1410 | int proto; |
dac1e716 | 1411 | int auth_proto; |
729796be | 1412 | struct ceph_auth_handshake *auth; |
31b8006e SW |
1413 | |
1414 | switch (con->peer_name.type) { | |
1415 | case CEPH_ENTITY_TYPE_MON: | |
1416 | proto = CEPH_MONC_PROTOCOL; | |
1417 | break; | |
1418 | case CEPH_ENTITY_TYPE_OSD: | |
1419 | proto = CEPH_OSDC_PROTOCOL; | |
1420 | break; | |
1421 | case CEPH_ENTITY_TYPE_MDS: | |
1422 | proto = CEPH_MDSC_PROTOCOL; | |
1423 | break; | |
1424 | default: | |
1425 | BUG(); | |
1426 | } | |
1427 | ||
1428 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
1429 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 1430 | |
859bff51 ID |
1431 | con->out_connect.features = |
1432 | cpu_to_le64(from_msgr(con->msgr)->supported_features); | |
31b8006e SW |
1433 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
1434 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
1435 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
1436 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
1437 | con->out_connect.flags = 0; | |
31b8006e | 1438 | |
dac1e716 AE |
1439 | auth_proto = CEPH_AUTH_UNKNOWN; |
1440 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
1441 | if (IS_ERR(auth)) |
1442 | return PTR_ERR(auth); | |
3da54776 | 1443 | |
dac1e716 | 1444 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
1445 | con->out_connect.authorizer_len = auth ? |
1446 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
1447 | ||
e2200423 | 1448 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
1449 | &con->out_connect); |
1450 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 1451 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 1452 | auth->authorizer_buf); |
859eb799 | 1453 | |
31b8006e | 1454 | con->out_more = 0; |
c9ffc77a | 1455 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
4e7a5dcd | 1456 | |
e10c758e | 1457 | return 0; |
31b8006e SW |
1458 | } |
1459 | ||
31b8006e SW |
1460 | /* |
1461 | * write as much of pending kvecs to the socket as we can. | |
1462 | * 1 -> done | |
1463 | * 0 -> socket full, but more to do | |
1464 | * <0 -> error | |
1465 | */ | |
1466 | static int write_partial_kvec(struct ceph_connection *con) | |
1467 | { | |
1468 | int ret; | |
1469 | ||
1470 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
1471 | while (con->out_kvec_bytes > 0) { | |
1472 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
1473 | con->out_kvec_left, con->out_kvec_bytes, | |
1474 | con->out_more); | |
1475 | if (ret <= 0) | |
1476 | goto out; | |
1477 | con->out_kvec_bytes -= ret; | |
1478 | if (con->out_kvec_bytes == 0) | |
1479 | break; /* done */ | |
f42299e6 AE |
1480 | |
1481 | /* account for full iov entries consumed */ | |
1482 | while (ret >= con->out_kvec_cur->iov_len) { | |
1483 | BUG_ON(!con->out_kvec_left); | |
1484 | ret -= con->out_kvec_cur->iov_len; | |
1485 | con->out_kvec_cur++; | |
1486 | con->out_kvec_left--; | |
1487 | } | |
1488 | /* and for a partially-consumed entry */ | |
1489 | if (ret) { | |
1490 | con->out_kvec_cur->iov_len -= ret; | |
1491 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
1492 | } |
1493 | } | |
1494 | con->out_kvec_left = 0; | |
1495 | con->out_kvec_is_msg = false; | |
1496 | ret = 1; | |
1497 | out: | |
1498 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
1499 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
1500 | return ret; /* done! */ | |
1501 | } | |
1502 | ||
35b62808 AE |
1503 | static u32 ceph_crc32c_page(u32 crc, struct page *page, |
1504 | unsigned int page_offset, | |
1505 | unsigned int length) | |
1506 | { | |
1507 | char *kaddr; | |
1508 | ||
1509 | kaddr = kmap(page); | |
1510 | BUG_ON(kaddr == NULL); | |
1511 | crc = crc32c(crc, kaddr + page_offset, length); | |
1512 | kunmap(page); | |
1513 | ||
1514 | return crc; | |
1515 | } | |
31b8006e SW |
1516 | /* |
1517 | * Write as much message data payload as we can. If we finish, queue | |
1518 | * up the footer. | |
1519 | * 1 -> done, footer is now queued in out_kvec[]. | |
1520 | * 0 -> socket full, but more to do | |
1521 | * <0 -> error | |
1522 | */ | |
34d2d200 | 1523 | static int write_partial_message_data(struct ceph_connection *con) |
31b8006e SW |
1524 | { |
1525 | struct ceph_msg *msg = con->out_msg; | |
8ae4f4f5 | 1526 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
859bff51 | 1527 | bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC); |
f5db90bc | 1528 | u32 crc; |
31b8006e | 1529 | |
859a35d5 | 1530 | dout("%s %p msg %p\n", __func__, con, msg); |
31b8006e | 1531 | |
5240d9f9 | 1532 | if (list_empty(&msg->data)) |
4c59b4a2 AE |
1533 | return -EINVAL; |
1534 | ||
5821bd8c AE |
1535 | /* |
1536 | * Iterate through each page that contains data to be | |
1537 | * written, and send as much as possible for each. | |
1538 | * | |
1539 | * If we are calculating the data crc (the default), we will | |
1540 | * need to map the page. If we have no pages, they have | |
1541 | * been revoked, so use the zero page. | |
1542 | */ | |
f5db90bc | 1543 | crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0; |
643c68a4 | 1544 | while (cursor->resid) { |
8a166d05 | 1545 | struct page *page; |
e387d525 AE |
1546 | size_t page_offset; |
1547 | size_t length; | |
8a166d05 | 1548 | bool last_piece; |
8ea299bc | 1549 | bool need_crc; |
f5db90bc | 1550 | int ret; |
68b4476b | 1551 | |
343128ce SB |
1552 | page = ceph_msg_data_next(cursor, &page_offset, &length, |
1553 | &last_piece); | |
e387d525 | 1554 | ret = ceph_tcp_sendpage(con->sock, page, page_offset, |
c2cfa194 | 1555 | length, !last_piece); |
f5db90bc AE |
1556 | if (ret <= 0) { |
1557 | if (do_datacrc) | |
1558 | msg->footer.data_crc = cpu_to_le32(crc); | |
31b8006e | 1559 | |
f5db90bc AE |
1560 | return ret; |
1561 | } | |
143334ff AE |
1562 | if (do_datacrc && cursor->need_crc) |
1563 | crc = ceph_crc32c_page(crc, page, page_offset, length); | |
343128ce | 1564 | need_crc = ceph_msg_data_advance(cursor, (size_t)ret); |
31b8006e SW |
1565 | } |
1566 | ||
34d2d200 | 1567 | dout("%s %p msg %p done\n", __func__, con, msg); |
31b8006e SW |
1568 | |
1569 | /* prepare and queue up footer, too */ | |
f5db90bc AE |
1570 | if (do_datacrc) |
1571 | msg->footer.data_crc = cpu_to_le32(crc); | |
1572 | else | |
84ca8fc8 | 1573 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1574 | con_out_kvec_reset(con); |
859eb799 | 1575 | prepare_write_message_footer(con); |
f5db90bc AE |
1576 | |
1577 | return 1; /* must return > 0 to indicate success */ | |
31b8006e SW |
1578 | } |
1579 | ||
1580 | /* | |
1581 | * write some zeros | |
1582 | */ | |
1583 | static int write_partial_skip(struct ceph_connection *con) | |
1584 | { | |
1585 | int ret; | |
1586 | ||
1587 | while (con->out_skip > 0) { | |
31739139 | 1588 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1589 | |
e1dcb128 | 1590 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true); |
31b8006e SW |
1591 | if (ret <= 0) |
1592 | goto out; | |
1593 | con->out_skip -= ret; | |
1594 | } | |
1595 | ret = 1; | |
1596 | out: | |
1597 | return ret; | |
1598 | } | |
1599 | ||
1600 | /* | |
1601 | * Prepare to read connection handshake, or an ack. | |
1602 | */ | |
eed0ef2c SW |
1603 | static void prepare_read_banner(struct ceph_connection *con) |
1604 | { | |
1605 | dout("prepare_read_banner %p\n", con); | |
1606 | con->in_base_pos = 0; | |
1607 | } | |
1608 | ||
31b8006e SW |
1609 | static void prepare_read_connect(struct ceph_connection *con) |
1610 | { | |
1611 | dout("prepare_read_connect %p\n", con); | |
1612 | con->in_base_pos = 0; | |
1613 | } | |
1614 | ||
1615 | static void prepare_read_ack(struct ceph_connection *con) | |
1616 | { | |
1617 | dout("prepare_read_ack %p\n", con); | |
1618 | con->in_base_pos = 0; | |
1619 | } | |
1620 | ||
3a23083b SW |
1621 | static void prepare_read_seq(struct ceph_connection *con) |
1622 | { | |
1623 | dout("prepare_read_seq %p\n", con); | |
1624 | con->in_base_pos = 0; | |
1625 | con->in_tag = CEPH_MSGR_TAG_SEQ; | |
1626 | } | |
1627 | ||
31b8006e SW |
1628 | static void prepare_read_tag(struct ceph_connection *con) |
1629 | { | |
1630 | dout("prepare_read_tag %p\n", con); | |
1631 | con->in_base_pos = 0; | |
1632 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1633 | } | |
1634 | ||
8b9558aa YZ |
1635 | static void prepare_read_keepalive_ack(struct ceph_connection *con) |
1636 | { | |
1637 | dout("prepare_read_keepalive_ack %p\n", con); | |
1638 | con->in_base_pos = 0; | |
1639 | } | |
1640 | ||
31b8006e SW |
1641 | /* |
1642 | * Prepare to read a message. | |
1643 | */ | |
1644 | static int prepare_read_message(struct ceph_connection *con) | |
1645 | { | |
1646 | dout("prepare_read_message %p\n", con); | |
1647 | BUG_ON(con->in_msg != NULL); | |
1648 | con->in_base_pos = 0; | |
1649 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1650 | return 0; | |
1651 | } | |
1652 | ||
1653 | ||
1654 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1655 | int end, int size, void *object) |
31b8006e | 1656 | { |
e6cee71f AE |
1657 | while (con->in_base_pos < end) { |
1658 | int left = end - con->in_base_pos; | |
31b8006e SW |
1659 | int have = size - left; |
1660 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1661 | if (ret <= 0) | |
1662 | return ret; | |
1663 | con->in_base_pos += ret; | |
1664 | } | |
1665 | return 1; | |
1666 | } | |
1667 | ||
1668 | ||
1669 | /* | |
1670 | * Read all or part of the connect-side handshake on a new connection | |
1671 | */ | |
eed0ef2c | 1672 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1673 | { |
fd51653f AE |
1674 | int size; |
1675 | int end; | |
1676 | int ret; | |
31b8006e | 1677 | |
eed0ef2c | 1678 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1679 | |
1680 | /* peer's banner */ | |
fd51653f AE |
1681 | size = strlen(CEPH_BANNER); |
1682 | end = size; | |
1683 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1684 | if (ret <= 0) |
1685 | goto out; | |
fd51653f AE |
1686 | |
1687 | size = sizeof (con->actual_peer_addr); | |
1688 | end += size; | |
1689 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1690 | if (ret <= 0) |
1691 | goto out; | |
fd51653f AE |
1692 | |
1693 | size = sizeof (con->peer_addr_for_me); | |
1694 | end += size; | |
1695 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1696 | if (ret <= 0) |
1697 | goto out; | |
fd51653f | 1698 | |
eed0ef2c SW |
1699 | out: |
1700 | return ret; | |
1701 | } | |
1702 | ||
1703 | static int read_partial_connect(struct ceph_connection *con) | |
1704 | { | |
fd51653f AE |
1705 | int size; |
1706 | int end; | |
1707 | int ret; | |
eed0ef2c SW |
1708 | |
1709 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1710 | ||
fd51653f AE |
1711 | size = sizeof (con->in_reply); |
1712 | end = size; | |
1713 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1714 | if (ret <= 0) |
1715 | goto out; | |
fd51653f AE |
1716 | |
1717 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1718 | end += size; | |
1719 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1720 | if (ret <= 0) |
1721 | goto out; | |
31b8006e | 1722 | |
4e7a5dcd SW |
1723 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1724 | con, (int)con->in_reply.tag, | |
1725 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1726 | le32_to_cpu(con->in_reply.global_seq)); |
1727 | out: | |
1728 | return ret; | |
eed0ef2c | 1729 | |
31b8006e SW |
1730 | } |
1731 | ||
1732 | /* | |
1733 | * Verify the hello banner looks okay. | |
1734 | */ | |
1735 | static int verify_hello(struct ceph_connection *con) | |
1736 | { | |
1737 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1738 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1739 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1740 | con->error_msg = "protocol error, bad banner"; |
1741 | return -1; | |
1742 | } | |
1743 | return 0; | |
1744 | } | |
1745 | ||
1746 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1747 | { | |
c44bd69c ID |
1748 | struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr; |
1749 | struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr; | |
1750 | ||
31b8006e SW |
1751 | switch (ss->ss_family) { |
1752 | case AF_INET: | |
c44bd69c | 1753 | return addr->s_addr == htonl(INADDR_ANY); |
31b8006e | 1754 | case AF_INET6: |
c44bd69c ID |
1755 | return ipv6_addr_any(addr6); |
1756 | default: | |
1757 | return true; | |
31b8006e | 1758 | } |
31b8006e SW |
1759 | } |
1760 | ||
1761 | static int addr_port(struct sockaddr_storage *ss) | |
1762 | { | |
1763 | switch (ss->ss_family) { | |
1764 | case AF_INET: | |
f28bcfbe | 1765 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1766 | case AF_INET6: |
f28bcfbe | 1767 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1768 | } |
1769 | return 0; | |
1770 | } | |
1771 | ||
1772 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1773 | { | |
1774 | switch (ss->ss_family) { | |
1775 | case AF_INET: | |
1776 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1777 | break; |
31b8006e SW |
1778 | case AF_INET6: |
1779 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1780 | break; |
31b8006e SW |
1781 | } |
1782 | } | |
1783 | ||
ee3b56f2 NW |
1784 | /* |
1785 | * Unlike other *_pton function semantics, zero indicates success. | |
1786 | */ | |
1787 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1788 | char delim, const char **ipend) | |
1789 | { | |
99f0f3b2 AE |
1790 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1791 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1792 | |
1793 | memset(ss, 0, sizeof(*ss)); | |
1794 | ||
1795 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1796 | ss->ss_family = AF_INET; | |
1797 | return 0; | |
1798 | } | |
1799 | ||
1800 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1801 | ss->ss_family = AF_INET6; | |
1802 | return 0; | |
1803 | } | |
1804 | ||
1805 | return -EINVAL; | |
1806 | } | |
1807 | ||
1808 | /* | |
1809 | * Extract hostname string and resolve using kernel DNS facility. | |
1810 | */ | |
1811 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1812 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1813 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1814 | { | |
1815 | const char *end, *delim_p; | |
1816 | char *colon_p, *ip_addr = NULL; | |
1817 | int ip_len, ret; | |
1818 | ||
1819 | /* | |
1820 | * The end of the hostname occurs immediately preceding the delimiter or | |
1821 | * the port marker (':') where the delimiter takes precedence. | |
1822 | */ | |
1823 | delim_p = memchr(name, delim, namelen); | |
1824 | colon_p = memchr(name, ':', namelen); | |
1825 | ||
1826 | if (delim_p && colon_p) | |
1827 | end = delim_p < colon_p ? delim_p : colon_p; | |
1828 | else if (!delim_p && colon_p) | |
1829 | end = colon_p; | |
1830 | else { | |
1831 | end = delim_p; | |
1832 | if (!end) /* case: hostname:/ */ | |
1833 | end = name + namelen; | |
1834 | } | |
1835 | ||
1836 | if (end <= name) | |
1837 | return -EINVAL; | |
1838 | ||
1839 | /* do dns_resolve upcall */ | |
1840 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1841 | if (ip_len > 0) | |
1842 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1843 | else | |
1844 | ret = -ESRCH; | |
1845 | ||
1846 | kfree(ip_addr); | |
1847 | ||
1848 | *ipend = end; | |
1849 | ||
1850 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1851 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1852 | ||
1853 | return ret; | |
1854 | } | |
1855 | #else | |
1856 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1857 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1858 | { | |
1859 | return -EINVAL; | |
1860 | } | |
1861 | #endif | |
1862 | ||
1863 | /* | |
1864 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1865 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1866 | */ | |
1867 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1868 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1869 | { | |
1870 | int ret; | |
1871 | ||
1872 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1873 | if (ret) | |
1874 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1875 | ||
1876 | return ret; | |
1877 | } | |
1878 | ||
31b8006e SW |
1879 | /* |
1880 | * Parse an ip[:port] list into an addr array. Use the default | |
1881 | * monitor port if a port isn't specified. | |
1882 | */ | |
1883 | int ceph_parse_ips(const char *c, const char *end, | |
1884 | struct ceph_entity_addr *addr, | |
1885 | int max_count, int *count) | |
1886 | { | |
ee3b56f2 | 1887 | int i, ret = -EINVAL; |
31b8006e SW |
1888 | const char *p = c; |
1889 | ||
1890 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1891 | for (i = 0; i < max_count; i++) { | |
1892 | const char *ipend; | |
1893 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1894 | int port; |
39139f64 SW |
1895 | char delim = ','; |
1896 | ||
1897 | if (*p == '[') { | |
1898 | delim = ']'; | |
1899 | p++; | |
1900 | } | |
31b8006e | 1901 | |
ee3b56f2 NW |
1902 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1903 | if (ret) | |
31b8006e | 1904 | goto bad; |
ee3b56f2 NW |
1905 | ret = -EINVAL; |
1906 | ||
31b8006e SW |
1907 | p = ipend; |
1908 | ||
39139f64 SW |
1909 | if (delim == ']') { |
1910 | if (*p != ']') { | |
1911 | dout("missing matching ']'\n"); | |
1912 | goto bad; | |
1913 | } | |
1914 | p++; | |
1915 | } | |
1916 | ||
31b8006e SW |
1917 | /* port? */ |
1918 | if (p < end && *p == ':') { | |
1919 | port = 0; | |
1920 | p++; | |
1921 | while (p < end && *p >= '0' && *p <= '9') { | |
1922 | port = (port * 10) + (*p - '0'); | |
1923 | p++; | |
1924 | } | |
f48db1e9 ID |
1925 | if (port == 0) |
1926 | port = CEPH_MON_PORT; | |
1927 | else if (port > 65535) | |
31b8006e SW |
1928 | goto bad; |
1929 | } else { | |
1930 | port = CEPH_MON_PORT; | |
1931 | } | |
1932 | ||
1933 | addr_set_port(ss, port); | |
1934 | ||
3d14c5d2 | 1935 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1936 | |
1937 | if (p == end) | |
1938 | break; | |
1939 | if (*p != ',') | |
1940 | goto bad; | |
1941 | p++; | |
1942 | } | |
1943 | ||
1944 | if (p != end) | |
1945 | goto bad; | |
1946 | ||
1947 | if (count) | |
1948 | *count = i + 1; | |
1949 | return 0; | |
1950 | ||
1951 | bad: | |
39139f64 | 1952 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1953 | return ret; |
31b8006e | 1954 | } |
3d14c5d2 | 1955 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1956 | |
eed0ef2c | 1957 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1958 | { |
eed0ef2c | 1959 | dout("process_banner on %p\n", con); |
31b8006e SW |
1960 | |
1961 | if (verify_hello(con) < 0) | |
1962 | return -1; | |
1963 | ||
63f2d211 SW |
1964 | ceph_decode_addr(&con->actual_peer_addr); |
1965 | ceph_decode_addr(&con->peer_addr_for_me); | |
1966 | ||
31b8006e SW |
1967 | /* |
1968 | * Make sure the other end is who we wanted. note that the other | |
1969 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1970 | * them the benefit of the doubt. | |
1971 | */ | |
103e2d3a SW |
1972 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1973 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1974 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1975 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
b9a67899 JP |
1976 | pr_warn("wrong peer, want %s/%d, got %s/%d\n", |
1977 | ceph_pr_addr(&con->peer_addr.in_addr), | |
1978 | (int)le32_to_cpu(con->peer_addr.nonce), | |
1979 | ceph_pr_addr(&con->actual_peer_addr.in_addr), | |
1980 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); | |
58bb3b37 | 1981 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1982 | return -1; |
1983 | } | |
1984 | ||
1985 | /* | |
1986 | * did we learn our address? | |
1987 | */ | |
1988 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1989 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1990 | ||
1991 | memcpy(&con->msgr->inst.addr.in_addr, | |
1992 | &con->peer_addr_for_me.in_addr, | |
1993 | sizeof(con->peer_addr_for_me.in_addr)); | |
1994 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1995 | encode_my_addr(con->msgr); |
eed0ef2c | 1996 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1997 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1998 | } |
1999 | ||
eed0ef2c SW |
2000 | return 0; |
2001 | } | |
2002 | ||
2003 | static int process_connect(struct ceph_connection *con) | |
2004 | { | |
859bff51 ID |
2005 | u64 sup_feat = from_msgr(con->msgr)->supported_features; |
2006 | u64 req_feat = from_msgr(con->msgr)->required_features; | |
2b3e0c90 ID |
2007 | u64 server_feat = ceph_sanitize_features( |
2008 | le64_to_cpu(con->in_reply.features)); | |
0da5d703 | 2009 | int ret; |
04a419f9 | 2010 | |
eed0ef2c SW |
2011 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
2012 | ||
31b8006e | 2013 | switch (con->in_reply.tag) { |
04a419f9 SW |
2014 | case CEPH_MSGR_TAG_FEATURES: |
2015 | pr_err("%s%lld %s feature set mismatch," | |
2016 | " my %llx < server's %llx, missing %llx\n", | |
2017 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2018 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
2019 | sup_feat, server_feat, server_feat & ~sup_feat); |
2020 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 2021 | reset_connection(con); |
04a419f9 SW |
2022 | return -1; |
2023 | ||
31b8006e | 2024 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
2025 | pr_err("%s%lld %s protocol version mismatch," |
2026 | " my %d != server's %d\n", | |
2027 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2028 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
2029 | le32_to_cpu(con->out_connect.protocol_version), |
2030 | le32_to_cpu(con->in_reply.protocol_version)); | |
2031 | con->error_msg = "protocol version mismatch"; | |
0fa6ebc6 | 2032 | reset_connection(con); |
31b8006e SW |
2033 | return -1; |
2034 | ||
4e7a5dcd SW |
2035 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
2036 | con->auth_retry++; | |
2037 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
2038 | con->auth_retry); | |
2039 | if (con->auth_retry == 2) { | |
2040 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
2041 | return -1; |
2042 | } | |
6d4221b5 | 2043 | con_out_kvec_reset(con); |
e825a66d | 2044 | ret = prepare_write_connect(con); |
0da5d703 SW |
2045 | if (ret < 0) |
2046 | return ret; | |
63733a0f | 2047 | prepare_read_connect(con); |
4e7a5dcd | 2048 | break; |
31b8006e SW |
2049 | |
2050 | case CEPH_MSGR_TAG_RESETSESSION: | |
2051 | /* | |
2052 | * If we connected with a large connect_seq but the peer | |
2053 | * has no record of a session with us (no connection, or | |
2054 | * connect_seq == 0), they will send RESETSESION to indicate | |
2055 | * that they must have reset their session, and may have | |
2056 | * dropped messages. | |
2057 | */ | |
2058 | dout("process_connect got RESET peer seq %u\n", | |
5bdca4e0 | 2059 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
2060 | pr_err("%s%lld %s connection reset\n", |
2061 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2062 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2063 | reset_connection(con); |
6d4221b5 | 2064 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2065 | ret = prepare_write_connect(con); |
2066 | if (ret < 0) | |
2067 | return ret; | |
31b8006e SW |
2068 | prepare_read_connect(con); |
2069 | ||
2070 | /* Tell ceph about it. */ | |
ec302645 | 2071 | mutex_unlock(&con->mutex); |
31b8006e SW |
2072 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
2073 | if (con->ops->peer_reset) | |
2074 | con->ops->peer_reset(con); | |
ec302645 | 2075 | mutex_lock(&con->mutex); |
8dacc7da | 2076 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 2077 | return -EAGAIN; |
31b8006e SW |
2078 | break; |
2079 | ||
2080 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
2081 | /* | |
2082 | * If we sent a smaller connect_seq than the peer has, try | |
2083 | * again with a larger value. | |
2084 | */ | |
5bdca4e0 | 2085 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 2086 | le32_to_cpu(con->out_connect.connect_seq), |
5bdca4e0 SW |
2087 | le32_to_cpu(con->in_reply.connect_seq)); |
2088 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
6d4221b5 | 2089 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2090 | ret = prepare_write_connect(con); |
2091 | if (ret < 0) | |
2092 | return ret; | |
31b8006e SW |
2093 | prepare_read_connect(con); |
2094 | break; | |
2095 | ||
2096 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
2097 | /* | |
2098 | * If we sent a smaller global_seq than the peer has, try | |
2099 | * again with a larger value. | |
2100 | */ | |
eed0ef2c | 2101 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 2102 | con->peer_global_seq, |
5bdca4e0 | 2103 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 2104 | get_global_seq(con->msgr, |
5bdca4e0 | 2105 | le32_to_cpu(con->in_reply.global_seq)); |
6d4221b5 | 2106 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2107 | ret = prepare_write_connect(con); |
2108 | if (ret < 0) | |
2109 | return ret; | |
31b8006e SW |
2110 | prepare_read_connect(con); |
2111 | break; | |
2112 | ||
3a23083b | 2113 | case CEPH_MSGR_TAG_SEQ: |
31b8006e | 2114 | case CEPH_MSGR_TAG_READY: |
04a419f9 SW |
2115 | if (req_feat & ~server_feat) { |
2116 | pr_err("%s%lld %s protocol feature mismatch," | |
2117 | " my required %llx > server's %llx, need %llx\n", | |
2118 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2119 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
2120 | req_feat, server_feat, req_feat & ~server_feat); |
2121 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 2122 | reset_connection(con); |
04a419f9 SW |
2123 | return -1; |
2124 | } | |
8dacc7da | 2125 | |
122070a2 | 2126 | WARN_ON(con->state != CON_STATE_NEGOTIATING); |
8dacc7da | 2127 | con->state = CON_STATE_OPEN; |
20e55c4c | 2128 | con->auth_retry = 0; /* we authenticated; clear flag */ |
31b8006e SW |
2129 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
2130 | con->connect_seq++; | |
aba558e2 | 2131 | con->peer_features = server_feat; |
31b8006e SW |
2132 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
2133 | con->peer_global_seq, | |
2134 | le32_to_cpu(con->in_reply.connect_seq), | |
2135 | con->connect_seq); | |
2136 | WARN_ON(con->connect_seq != | |
2137 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
2138 | |
2139 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
c9ffc77a | 2140 | con_flag_set(con, CON_FLAG_LOSSYTX); |
92ac41d0 | 2141 | |
85effe18 | 2142 | con->delay = 0; /* reset backoff memory */ |
92ac41d0 | 2143 | |
3a23083b SW |
2144 | if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) { |
2145 | prepare_write_seq(con); | |
2146 | prepare_read_seq(con); | |
2147 | } else { | |
2148 | prepare_read_tag(con); | |
2149 | } | |
31b8006e SW |
2150 | break; |
2151 | ||
2152 | case CEPH_MSGR_TAG_WAIT: | |
2153 | /* | |
2154 | * If there is a connection race (we are opening | |
2155 | * connections to each other), one of us may just have | |
2156 | * to WAIT. This shouldn't happen if we are the | |
2157 | * client. | |
2158 | */ | |
04177882 SW |
2159 | con->error_msg = "protocol error, got WAIT as client"; |
2160 | return -1; | |
31b8006e SW |
2161 | |
2162 | default: | |
31b8006e SW |
2163 | con->error_msg = "protocol error, garbage tag during connect"; |
2164 | return -1; | |
2165 | } | |
2166 | return 0; | |
2167 | } | |
2168 | ||
2169 | ||
2170 | /* | |
2171 | * read (part of) an ack | |
2172 | */ | |
2173 | static int read_partial_ack(struct ceph_connection *con) | |
2174 | { | |
fd51653f AE |
2175 | int size = sizeof (con->in_temp_ack); |
2176 | int end = size; | |
31b8006e | 2177 | |
fd51653f | 2178 | return read_partial(con, end, size, &con->in_temp_ack); |
31b8006e SW |
2179 | } |
2180 | ||
31b8006e SW |
2181 | /* |
2182 | * We can finally discard anything that's been acked. | |
2183 | */ | |
2184 | static void process_ack(struct ceph_connection *con) | |
2185 | { | |
2186 | struct ceph_msg *m; | |
2187 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
2188 | u64 seq; | |
2189 | ||
31b8006e SW |
2190 | while (!list_empty(&con->out_sent)) { |
2191 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
2192 | list_head); | |
2193 | seq = le64_to_cpu(m->hdr.seq); | |
2194 | if (seq > ack) | |
2195 | break; | |
2196 | dout("got ack for seq %llu type %d at %p\n", seq, | |
2197 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 2198 | m->ack_stamp = jiffies; |
31b8006e SW |
2199 | ceph_msg_remove(m); |
2200 | } | |
31b8006e SW |
2201 | prepare_read_tag(con); |
2202 | } | |
2203 | ||
2204 | ||
2450418c | 2205 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
2206 | struct kvec *section, |
2207 | unsigned int sec_len, u32 *crc) | |
2450418c | 2208 | { |
68b4476b | 2209 | int ret, left; |
2450418c YS |
2210 | |
2211 | BUG_ON(!section); | |
2212 | ||
2213 | while (section->iov_len < sec_len) { | |
2214 | BUG_ON(section->iov_base == NULL); | |
2215 | left = sec_len - section->iov_len; | |
2216 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
2217 | section->iov_len, left); | |
2218 | if (ret <= 0) | |
2219 | return ret; | |
2220 | section->iov_len += ret; | |
2450418c | 2221 | } |
fe3ad593 AE |
2222 | if (section->iov_len == sec_len) |
2223 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 2224 | |
2450418c YS |
2225 | return 1; |
2226 | } | |
31b8006e | 2227 | |
34d2d200 AE |
2228 | static int read_partial_msg_data(struct ceph_connection *con) |
2229 | { | |
2230 | struct ceph_msg *msg = con->in_msg; | |
8ae4f4f5 | 2231 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
859bff51 | 2232 | bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC); |
686be208 AE |
2233 | struct page *page; |
2234 | size_t page_offset; | |
2235 | size_t length; | |
f5db90bc | 2236 | u32 crc = 0; |
34d2d200 AE |
2237 | int ret; |
2238 | ||
2239 | BUG_ON(!msg); | |
5240d9f9 | 2240 | if (list_empty(&msg->data)) |
4c59b4a2 | 2241 | return -EIO; |
34d2d200 | 2242 | |
f5db90bc AE |
2243 | if (do_datacrc) |
2244 | crc = con->in_data_crc; | |
643c68a4 | 2245 | while (cursor->resid) { |
343128ce | 2246 | page = ceph_msg_data_next(cursor, &page_offset, &length, NULL); |
686be208 | 2247 | ret = ceph_tcp_recvpage(con->sock, page, page_offset, length); |
f5db90bc AE |
2248 | if (ret <= 0) { |
2249 | if (do_datacrc) | |
2250 | con->in_data_crc = crc; | |
2251 | ||
686be208 | 2252 | return ret; |
f5db90bc | 2253 | } |
686be208 AE |
2254 | |
2255 | if (do_datacrc) | |
f5db90bc | 2256 | crc = ceph_crc32c_page(crc, page, page_offset, ret); |
343128ce | 2257 | (void) ceph_msg_data_advance(cursor, (size_t)ret); |
34d2d200 | 2258 | } |
f5db90bc AE |
2259 | if (do_datacrc) |
2260 | con->in_data_crc = crc; | |
34d2d200 AE |
2261 | |
2262 | return 1; /* must return > 0 to indicate success */ | |
2263 | } | |
2264 | ||
31b8006e SW |
2265 | /* |
2266 | * read (part of) a message. | |
2267 | */ | |
686be208 AE |
2268 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
2269 | ||
31b8006e SW |
2270 | static int read_partial_message(struct ceph_connection *con) |
2271 | { | |
2272 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
2273 | int size; |
2274 | int end; | |
31b8006e | 2275 | int ret; |
95c96174 | 2276 | unsigned int front_len, middle_len, data_len; |
859bff51 | 2277 | bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC); |
33d07337 | 2278 | bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH); |
ae18756b | 2279 | u64 seq; |
fe3ad593 | 2280 | u32 crc; |
31b8006e SW |
2281 | |
2282 | dout("read_partial_message con %p msg %p\n", con, m); | |
2283 | ||
2284 | /* header */ | |
fd51653f AE |
2285 | size = sizeof (con->in_hdr); |
2286 | end = size; | |
2287 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
2288 | if (ret <= 0) |
2289 | return ret; | |
fe3ad593 AE |
2290 | |
2291 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
2292 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
67c64eb7 | 2293 | pr_err("read_partial_message bad hdr crc %u != expected %u\n", |
fe3ad593 AE |
2294 | crc, con->in_hdr.crc); |
2295 | return -EBADMSG; | |
2296 | } | |
2297 | ||
31b8006e SW |
2298 | front_len = le32_to_cpu(con->in_hdr.front_len); |
2299 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
2300 | return -EIO; | |
2301 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
7b11ba37 | 2302 | if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN) |
31b8006e SW |
2303 | return -EIO; |
2304 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
2305 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
2306 | return -EIO; | |
2307 | ||
ae18756b SW |
2308 | /* verify seq# */ |
2309 | seq = le64_to_cpu(con->in_hdr.seq); | |
2310 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 2311 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 2312 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2313 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
2314 | seq, con->in_seq + 1); |
2315 | con->in_base_pos = -front_len - middle_len - data_len - | |
2316 | sizeof(m->footer); | |
2317 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
2318 | return 0; |
2319 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
2320 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
2321 | seq, con->in_seq + 1); | |
2322 | con->error_msg = "bad message sequence # for incoming message"; | |
67c64eb7 | 2323 | return -EBADE; |
ae18756b SW |
2324 | } |
2325 | ||
31b8006e SW |
2326 | /* allocate message? */ |
2327 | if (!con->in_msg) { | |
4740a623 SW |
2328 | int skip = 0; |
2329 | ||
31b8006e | 2330 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
6ebc8b32 | 2331 | front_len, data_len); |
4740a623 SW |
2332 | ret = ceph_con_in_msg_alloc(con, &skip); |
2333 | if (ret < 0) | |
2334 | return ret; | |
f759ebb9 AE |
2335 | |
2336 | BUG_ON(!con->in_msg ^ skip); | |
2450418c | 2337 | if (skip) { |
31b8006e | 2338 | /* skip this message */ |
a79832f2 | 2339 | dout("alloc_msg said skip message\n"); |
31b8006e SW |
2340 | con->in_base_pos = -front_len - middle_len - data_len - |
2341 | sizeof(m->footer); | |
2342 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2343 | con->in_seq++; |
31b8006e SW |
2344 | return 0; |
2345 | } | |
38941f80 | 2346 | |
4740a623 | 2347 | BUG_ON(!con->in_msg); |
38941f80 | 2348 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
2349 | m = con->in_msg; |
2350 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
2351 | if (m->middle) |
2352 | m->middle->vec.iov_len = 0; | |
9d7f0f13 | 2353 | |
78625051 | 2354 | /* prepare for data payload, if any */ |
a4107026 | 2355 | |
78625051 | 2356 | if (data_len) |
98fa5dd8 | 2357 | prepare_message_data(con->in_msg, data_len); |
31b8006e SW |
2358 | } |
2359 | ||
2360 | /* front */ | |
2450418c YS |
2361 | ret = read_partial_message_section(con, &m->front, front_len, |
2362 | &con->in_front_crc); | |
2363 | if (ret <= 0) | |
2364 | return ret; | |
31b8006e SW |
2365 | |
2366 | /* middle */ | |
2450418c | 2367 | if (m->middle) { |
213c99ee SW |
2368 | ret = read_partial_message_section(con, &m->middle->vec, |
2369 | middle_len, | |
2450418c | 2370 | &con->in_middle_crc); |
31b8006e SW |
2371 | if (ret <= 0) |
2372 | return ret; | |
31b8006e SW |
2373 | } |
2374 | ||
2375 | /* (page) data */ | |
34d2d200 AE |
2376 | if (data_len) { |
2377 | ret = read_partial_msg_data(con); | |
2378 | if (ret <= 0) | |
2379 | return ret; | |
31b8006e SW |
2380 | } |
2381 | ||
31b8006e | 2382 | /* footer */ |
33d07337 YZ |
2383 | if (need_sign) |
2384 | size = sizeof(m->footer); | |
2385 | else | |
2386 | size = sizeof(m->old_footer); | |
2387 | ||
fd51653f AE |
2388 | end += size; |
2389 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
2390 | if (ret <= 0) |
2391 | return ret; | |
2392 | ||
33d07337 YZ |
2393 | if (!need_sign) { |
2394 | m->footer.flags = m->old_footer.flags; | |
2395 | m->footer.sig = 0; | |
2396 | } | |
2397 | ||
31b8006e SW |
2398 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
2399 | m, front_len, m->footer.front_crc, middle_len, | |
2400 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
2401 | ||
2402 | /* crc ok? */ | |
2403 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
2404 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
2405 | m, con->in_front_crc, m->footer.front_crc); | |
2406 | return -EBADMSG; | |
2407 | } | |
2408 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
2409 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
2410 | m, con->in_middle_crc, m->footer.middle_crc); | |
2411 | return -EBADMSG; | |
2412 | } | |
bca064d2 | 2413 | if (do_datacrc && |
31b8006e SW |
2414 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
2415 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
2416 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
2417 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
2418 | return -EBADMSG; | |
2419 | } | |
2420 | ||
33d07337 | 2421 | if (need_sign && con->ops->check_message_signature && |
79dbd1ba | 2422 | con->ops->check_message_signature(m)) { |
33d07337 YZ |
2423 | pr_err("read_partial_message %p signature check failed\n", m); |
2424 | return -EBADMSG; | |
2425 | } | |
2426 | ||
31b8006e SW |
2427 | return 1; /* done! */ |
2428 | } | |
2429 | ||
2430 | /* | |
2431 | * Process message. This happens in the worker thread. The callback should | |
2432 | * be careful not to do anything that waits on other incoming messages or it | |
2433 | * may deadlock. | |
2434 | */ | |
2435 | static void process_message(struct ceph_connection *con) | |
2436 | { | |
583d0fef | 2437 | struct ceph_msg *msg = con->in_msg; |
31b8006e | 2438 | |
38941f80 | 2439 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
2440 | con->in_msg = NULL; |
2441 | ||
2442 | /* if first message, set peer_name */ | |
2443 | if (con->peer_name.type == 0) | |
dbad185d | 2444 | con->peer_name = msg->hdr.src; |
31b8006e | 2445 | |
31b8006e | 2446 | con->in_seq++; |
ec302645 | 2447 | mutex_unlock(&con->mutex); |
31b8006e SW |
2448 | |
2449 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
2450 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 2451 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
2452 | le16_to_cpu(msg->hdr.type), |
2453 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2454 | le32_to_cpu(msg->hdr.front_len), | |
2455 | le32_to_cpu(msg->hdr.data_len), | |
2456 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
2457 | con->ops->dispatch(con, msg); | |
ec302645 SW |
2458 | |
2459 | mutex_lock(&con->mutex); | |
31b8006e SW |
2460 | } |
2461 | ||
8b9558aa YZ |
2462 | static int read_keepalive_ack(struct ceph_connection *con) |
2463 | { | |
2464 | struct ceph_timespec ceph_ts; | |
2465 | size_t size = sizeof(ceph_ts); | |
2466 | int ret = read_partial(con, size, size, &ceph_ts); | |
2467 | if (ret <= 0) | |
2468 | return ret; | |
2469 | ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts); | |
2470 | prepare_read_tag(con); | |
2471 | return 1; | |
2472 | } | |
31b8006e SW |
2473 | |
2474 | /* | |
2475 | * Write something to the socket. Called in a worker thread when the | |
2476 | * socket appears to be writeable and we have something ready to send. | |
2477 | */ | |
2478 | static int try_write(struct ceph_connection *con) | |
2479 | { | |
31b8006e SW |
2480 | int ret = 1; |
2481 | ||
d59315ca | 2482 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 2483 | |
31b8006e SW |
2484 | more: |
2485 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
2486 | ||
2487 | /* open the socket first? */ | |
8dacc7da SW |
2488 | if (con->state == CON_STATE_PREOPEN) { |
2489 | BUG_ON(con->sock); | |
2490 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2491 | |
e2200423 | 2492 | con_out_kvec_reset(con); |
e825a66d | 2493 | prepare_write_banner(con); |
eed0ef2c | 2494 | prepare_read_banner(con); |
31b8006e | 2495 | |
cf3e5c40 | 2496 | BUG_ON(con->in_msg); |
31b8006e SW |
2497 | con->in_tag = CEPH_MSGR_TAG_READY; |
2498 | dout("try_write initiating connect on %p new state %lu\n", | |
2499 | con, con->state); | |
41617d0c AE |
2500 | ret = ceph_tcp_connect(con); |
2501 | if (ret < 0) { | |
31b8006e | 2502 | con->error_msg = "connect error"; |
31b8006e SW |
2503 | goto out; |
2504 | } | |
2505 | } | |
2506 | ||
2507 | more_kvec: | |
2508 | /* kvec data queued? */ | |
2509 | if (con->out_skip) { | |
2510 | ret = write_partial_skip(con); | |
2511 | if (ret <= 0) | |
42961d23 | 2512 | goto out; |
31b8006e SW |
2513 | } |
2514 | if (con->out_kvec_left) { | |
2515 | ret = write_partial_kvec(con); | |
2516 | if (ret <= 0) | |
42961d23 | 2517 | goto out; |
31b8006e SW |
2518 | } |
2519 | ||
2520 | /* msg pages? */ | |
2521 | if (con->out_msg) { | |
c86a2930 SW |
2522 | if (con->out_msg_done) { |
2523 | ceph_msg_put(con->out_msg); | |
2524 | con->out_msg = NULL; /* we're done with this one */ | |
2525 | goto do_next; | |
2526 | } | |
2527 | ||
34d2d200 | 2528 | ret = write_partial_message_data(con); |
31b8006e SW |
2529 | if (ret == 1) |
2530 | goto more_kvec; /* we need to send the footer, too! */ | |
2531 | if (ret == 0) | |
42961d23 | 2532 | goto out; |
31b8006e | 2533 | if (ret < 0) { |
34d2d200 | 2534 | dout("try_write write_partial_message_data err %d\n", |
31b8006e | 2535 | ret); |
42961d23 | 2536 | goto out; |
31b8006e SW |
2537 | } |
2538 | } | |
2539 | ||
c86a2930 | 2540 | do_next: |
8dacc7da | 2541 | if (con->state == CON_STATE_OPEN) { |
8b9558aa YZ |
2542 | if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) { |
2543 | prepare_write_keepalive(con); | |
2544 | goto more; | |
2545 | } | |
31b8006e SW |
2546 | /* is anything else pending? */ |
2547 | if (!list_empty(&con->out_queue)) { | |
2548 | prepare_write_message(con); | |
2549 | goto more; | |
2550 | } | |
2551 | if (con->in_seq > con->in_seq_acked) { | |
2552 | prepare_write_ack(con); | |
2553 | goto more; | |
2554 | } | |
31b8006e SW |
2555 | } |
2556 | ||
2557 | /* Nothing to do! */ | |
c9ffc77a | 2558 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
31b8006e | 2559 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2560 | ret = 0; |
2561 | out: | |
42961d23 | 2562 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2563 | return ret; |
2564 | } | |
2565 | ||
2566 | ||
2567 | ||
2568 | /* | |
2569 | * Read what we can from the socket. | |
2570 | */ | |
2571 | static int try_read(struct ceph_connection *con) | |
2572 | { | |
31b8006e SW |
2573 | int ret = -1; |
2574 | ||
8dacc7da SW |
2575 | more: |
2576 | dout("try_read start on %p state %lu\n", con, con->state); | |
2577 | if (con->state != CON_STATE_CONNECTING && | |
2578 | con->state != CON_STATE_NEGOTIATING && | |
2579 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2580 | return 0; |
2581 | ||
8dacc7da | 2582 | BUG_ON(!con->sock); |
ec302645 | 2583 | |
31b8006e SW |
2584 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2585 | con->in_base_pos); | |
0da5d703 | 2586 | |
8dacc7da | 2587 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2588 | dout("try_read connecting\n"); |
2589 | ret = read_partial_banner(con); | |
2590 | if (ret <= 0) | |
ab166d5a | 2591 | goto out; |
7593af92 AE |
2592 | ret = process_banner(con); |
2593 | if (ret < 0) | |
2594 | goto out; | |
2595 | ||
8dacc7da | 2596 | con->state = CON_STATE_NEGOTIATING; |
7593af92 | 2597 | |
6d4221b5 JS |
2598 | /* |
2599 | * Received banner is good, exchange connection info. | |
2600 | * Do not reset out_kvec, as sending our banner raced | |
2601 | * with receiving peer banner after connect completed. | |
2602 | */ | |
7593af92 AE |
2603 | ret = prepare_write_connect(con); |
2604 | if (ret < 0) | |
2605 | goto out; | |
2606 | prepare_read_connect(con); | |
2607 | ||
2608 | /* Send connection info before awaiting response */ | |
0da5d703 SW |
2609 | goto out; |
2610 | } | |
2611 | ||
8dacc7da | 2612 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2613 | dout("try_read negotiating\n"); |
31b8006e SW |
2614 | ret = read_partial_connect(con); |
2615 | if (ret <= 0) | |
31b8006e | 2616 | goto out; |
98bdb0aa SW |
2617 | ret = process_connect(con); |
2618 | if (ret < 0) | |
2619 | goto out; | |
31b8006e SW |
2620 | goto more; |
2621 | } | |
2622 | ||
122070a2 | 2623 | WARN_ON(con->state != CON_STATE_OPEN); |
8dacc7da | 2624 | |
31b8006e SW |
2625 | if (con->in_base_pos < 0) { |
2626 | /* | |
2627 | * skipping + discarding content. | |
2628 | * | |
2629 | * FIXME: there must be a better way to do this! | |
2630 | */ | |
84495f49 AE |
2631 | static char buf[SKIP_BUF_SIZE]; |
2632 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2633 | ||
31b8006e SW |
2634 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2635 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2636 | if (ret <= 0) | |
98bdb0aa | 2637 | goto out; |
31b8006e SW |
2638 | con->in_base_pos += ret; |
2639 | if (con->in_base_pos) | |
2640 | goto more; | |
2641 | } | |
2642 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2643 | /* | |
2644 | * what's next? | |
2645 | */ | |
2646 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2647 | if (ret <= 0) | |
98bdb0aa | 2648 | goto out; |
31b8006e SW |
2649 | dout("try_read got tag %d\n", (int)con->in_tag); |
2650 | switch (con->in_tag) { | |
2651 | case CEPH_MSGR_TAG_MSG: | |
2652 | prepare_read_message(con); | |
2653 | break; | |
2654 | case CEPH_MSGR_TAG_ACK: | |
2655 | prepare_read_ack(con); | |
2656 | break; | |
8b9558aa YZ |
2657 | case CEPH_MSGR_TAG_KEEPALIVE2_ACK: |
2658 | prepare_read_keepalive_ack(con); | |
2659 | break; | |
31b8006e | 2660 | case CEPH_MSGR_TAG_CLOSE: |
8dacc7da SW |
2661 | con_close_socket(con); |
2662 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2663 | goto out; |
31b8006e SW |
2664 | default: |
2665 | goto bad_tag; | |
2666 | } | |
2667 | } | |
2668 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2669 | ret = read_partial_message(con); | |
2670 | if (ret <= 0) { | |
2671 | switch (ret) { | |
2672 | case -EBADMSG: | |
a51983e4 | 2673 | con->error_msg = "bad crc/signature"; |
67c64eb7 ID |
2674 | /* fall through */ |
2675 | case -EBADE: | |
31b8006e | 2676 | ret = -EIO; |
98bdb0aa | 2677 | break; |
31b8006e SW |
2678 | case -EIO: |
2679 | con->error_msg = "io error"; | |
98bdb0aa | 2680 | break; |
31b8006e | 2681 | } |
98bdb0aa | 2682 | goto out; |
31b8006e SW |
2683 | } |
2684 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2685 | goto more; | |
2686 | process_message(con); | |
7b862e07 SW |
2687 | if (con->state == CON_STATE_OPEN) |
2688 | prepare_read_tag(con); | |
31b8006e SW |
2689 | goto more; |
2690 | } | |
3a23083b SW |
2691 | if (con->in_tag == CEPH_MSGR_TAG_ACK || |
2692 | con->in_tag == CEPH_MSGR_TAG_SEQ) { | |
2693 | /* | |
2694 | * the final handshake seq exchange is semantically | |
2695 | * equivalent to an ACK | |
2696 | */ | |
31b8006e SW |
2697 | ret = read_partial_ack(con); |
2698 | if (ret <= 0) | |
98bdb0aa | 2699 | goto out; |
31b8006e SW |
2700 | process_ack(con); |
2701 | goto more; | |
2702 | } | |
8b9558aa YZ |
2703 | if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) { |
2704 | ret = read_keepalive_ack(con); | |
2705 | if (ret <= 0) | |
2706 | goto out; | |
2707 | goto more; | |
2708 | } | |
31b8006e | 2709 | |
31b8006e | 2710 | out: |
98bdb0aa | 2711 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2712 | return ret; |
2713 | ||
2714 | bad_tag: | |
2715 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2716 | con->error_msg = "protocol error, garbage tag"; | |
2717 | ret = -1; | |
2718 | goto out; | |
2719 | } | |
2720 | ||
2721 | ||
2722 | /* | |
802c6d96 AE |
2723 | * Atomically queue work on a connection after the specified delay. |
2724 | * Bump @con reference to avoid races with connection teardown. | |
2725 | * Returns 0 if work was queued, or an error code otherwise. | |
31b8006e | 2726 | */ |
802c6d96 | 2727 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
31b8006e | 2728 | { |
31b8006e | 2729 | if (!con->ops->get(con)) { |
802c6d96 | 2730 | dout("%s %p ref count 0\n", __func__, con); |
802c6d96 | 2731 | return -ENOENT; |
31b8006e SW |
2732 | } |
2733 | ||
802c6d96 AE |
2734 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
2735 | dout("%s %p - already queued\n", __func__, con); | |
31b8006e | 2736 | con->ops->put(con); |
802c6d96 | 2737 | return -EBUSY; |
31b8006e | 2738 | } |
802c6d96 AE |
2739 | |
2740 | dout("%s %p %lu\n", __func__, con, delay); | |
802c6d96 AE |
2741 | return 0; |
2742 | } | |
2743 | ||
2744 | static void queue_con(struct ceph_connection *con) | |
2745 | { | |
2746 | (void) queue_con_delay(con, 0); | |
31b8006e SW |
2747 | } |
2748 | ||
37ab77ac ID |
2749 | static void cancel_con(struct ceph_connection *con) |
2750 | { | |
2751 | if (cancel_delayed_work(&con->work)) { | |
2752 | dout("%s %p\n", __func__, con); | |
2753 | con->ops->put(con); | |
2754 | } | |
2755 | } | |
2756 | ||
7bb21d68 AE |
2757 | static bool con_sock_closed(struct ceph_connection *con) |
2758 | { | |
c9ffc77a | 2759 | if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED)) |
7bb21d68 AE |
2760 | return false; |
2761 | ||
2762 | #define CASE(x) \ | |
2763 | case CON_STATE_ ## x: \ | |
2764 | con->error_msg = "socket closed (con state " #x ")"; \ | |
2765 | break; | |
2766 | ||
2767 | switch (con->state) { | |
2768 | CASE(CLOSED); | |
2769 | CASE(PREOPEN); | |
2770 | CASE(CONNECTING); | |
2771 | CASE(NEGOTIATING); | |
2772 | CASE(OPEN); | |
2773 | CASE(STANDBY); | |
2774 | default: | |
b9a67899 | 2775 | pr_warn("%s con %p unrecognized state %lu\n", |
7bb21d68 AE |
2776 | __func__, con, con->state); |
2777 | con->error_msg = "unrecognized con state"; | |
2778 | BUG(); | |
2779 | break; | |
2780 | } | |
2781 | #undef CASE | |
2782 | ||
2783 | return true; | |
2784 | } | |
2785 | ||
f20a39fd AE |
2786 | static bool con_backoff(struct ceph_connection *con) |
2787 | { | |
2788 | int ret; | |
2789 | ||
2790 | if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF)) | |
2791 | return false; | |
2792 | ||
2793 | ret = queue_con_delay(con, round_jiffies_relative(con->delay)); | |
2794 | if (ret) { | |
2795 | dout("%s: con %p FAILED to back off %lu\n", __func__, | |
2796 | con, con->delay); | |
2797 | BUG_ON(ret == -ENOENT); | |
2798 | con_flag_set(con, CON_FLAG_BACKOFF); | |
2799 | } | |
2800 | ||
2801 | return true; | |
2802 | } | |
2803 | ||
93209264 AE |
2804 | /* Finish fault handling; con->mutex must *not* be held here */ |
2805 | ||
2806 | static void con_fault_finish(struct ceph_connection *con) | |
2807 | { | |
2808 | /* | |
2809 | * in case we faulted due to authentication, invalidate our | |
2810 | * current tickets so that we can get new ones. | |
2811 | */ | |
2812 | if (con->auth_retry && con->ops->invalidate_authorizer) { | |
2813 | dout("calling invalidate_authorizer()\n"); | |
2814 | con->ops->invalidate_authorizer(con); | |
2815 | } | |
2816 | ||
2817 | if (con->ops->fault) | |
2818 | con->ops->fault(con); | |
2819 | } | |
2820 | ||
31b8006e SW |
2821 | /* |
2822 | * Do some work on a connection. Drop a connection ref when we're done. | |
2823 | */ | |
68931622 | 2824 | static void ceph_con_workfn(struct work_struct *work) |
31b8006e SW |
2825 | { |
2826 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2827 | work.work); | |
49659416 | 2828 | bool fault; |
31b8006e | 2829 | |
9dd4658d | 2830 | mutex_lock(&con->mutex); |
49659416 AE |
2831 | while (true) { |
2832 | int ret; | |
31b8006e | 2833 | |
49659416 AE |
2834 | if ((fault = con_sock_closed(con))) { |
2835 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); | |
2836 | break; | |
2837 | } | |
2838 | if (con_backoff(con)) { | |
2839 | dout("%s: con %p BACKOFF\n", __func__, con); | |
2840 | break; | |
2841 | } | |
2842 | if (con->state == CON_STATE_STANDBY) { | |
2843 | dout("%s: con %p STANDBY\n", __func__, con); | |
2844 | break; | |
2845 | } | |
2846 | if (con->state == CON_STATE_CLOSED) { | |
2847 | dout("%s: con %p CLOSED\n", __func__, con); | |
2848 | BUG_ON(con->sock); | |
2849 | break; | |
2850 | } | |
2851 | if (con->state == CON_STATE_PREOPEN) { | |
2852 | dout("%s: con %p PREOPEN\n", __func__, con); | |
2853 | BUG_ON(con->sock); | |
2854 | } | |
0da5d703 | 2855 | |
49659416 AE |
2856 | ret = try_read(con); |
2857 | if (ret < 0) { | |
2858 | if (ret == -EAGAIN) | |
2859 | continue; | |
67c64eb7 ID |
2860 | if (!con->error_msg) |
2861 | con->error_msg = "socket error on read"; | |
49659416 AE |
2862 | fault = true; |
2863 | break; | |
2864 | } | |
2865 | ||
2866 | ret = try_write(con); | |
2867 | if (ret < 0) { | |
2868 | if (ret == -EAGAIN) | |
2869 | continue; | |
67c64eb7 ID |
2870 | if (!con->error_msg) |
2871 | con->error_msg = "socket error on write"; | |
49659416 AE |
2872 | fault = true; |
2873 | } | |
2874 | ||
2875 | break; /* If we make it to here, we're done */ | |
3a140a0d | 2876 | } |
b6e7b6a1 AE |
2877 | if (fault) |
2878 | con_fault(con); | |
9dd4658d | 2879 | mutex_unlock(&con->mutex); |
0da5d703 | 2880 | |
b6e7b6a1 AE |
2881 | if (fault) |
2882 | con_fault_finish(con); | |
2883 | ||
2884 | con->ops->put(con); | |
31b8006e SW |
2885 | } |
2886 | ||
31b8006e SW |
2887 | /* |
2888 | * Generic error/fault handler. A retry mechanism is used with | |
2889 | * exponential backoff | |
2890 | */ | |
93209264 | 2891 | static void con_fault(struct ceph_connection *con) |
31b8006e | 2892 | { |
31b8006e | 2893 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2894 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2895 | |
67c64eb7 ID |
2896 | pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
2897 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); | |
2898 | con->error_msg = NULL; | |
2899 | ||
122070a2 | 2900 | WARN_ON(con->state != CON_STATE_CONNECTING && |
8dacc7da SW |
2901 | con->state != CON_STATE_NEGOTIATING && |
2902 | con->state != CON_STATE_OPEN); | |
ec302645 | 2903 | |
31b8006e | 2904 | con_close_socket(con); |
5e095e8b | 2905 | |
c9ffc77a | 2906 | if (con_flag_test(con, CON_FLAG_LOSSYTX)) { |
8dacc7da SW |
2907 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
2908 | con->state = CON_STATE_CLOSED; | |
93209264 | 2909 | return; |
3b5ede07 SW |
2910 | } |
2911 | ||
5e095e8b | 2912 | if (con->in_msg) { |
38941f80 | 2913 | BUG_ON(con->in_msg->con != con); |
5e095e8b SW |
2914 | ceph_msg_put(con->in_msg); |
2915 | con->in_msg = NULL; | |
2916 | } | |
31b8006e | 2917 | |
e80a52d1 SW |
2918 | /* Requeue anything that hasn't been acked */ |
2919 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2920 | |
e76661d0 SW |
2921 | /* If there are no messages queued or keepalive pending, place |
2922 | * the connection in a STANDBY state */ | |
2923 | if (list_empty(&con->out_queue) && | |
c9ffc77a | 2924 | !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) { |
e00de341 | 2925 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
c9ffc77a | 2926 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
8dacc7da | 2927 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
2928 | } else { |
2929 | /* retry after a delay. */ | |
8dacc7da | 2930 | con->state = CON_STATE_PREOPEN; |
e80a52d1 SW |
2931 | if (con->delay == 0) |
2932 | con->delay = BASE_DELAY_INTERVAL; | |
2933 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2934 | con->delay *= 2; | |
c9ffc77a | 2935 | con_flag_set(con, CON_FLAG_BACKOFF); |
8618e30b | 2936 | queue_con(con); |
31b8006e | 2937 | } |
31b8006e SW |
2938 | } |
2939 | ||
2940 | ||
2941 | ||
2942 | /* | |
15d9882c | 2943 | * initialize a new messenger instance |
31b8006e | 2944 | */ |
15d9882c | 2945 | void ceph_messenger_init(struct ceph_messenger *msgr, |
859bff51 | 2946 | struct ceph_entity_addr *myaddr) |
31b8006e | 2947 | { |
31b8006e SW |
2948 | spin_lock_init(&msgr->global_seq_lock); |
2949 | ||
31b8006e SW |
2950 | if (myaddr) |
2951 | msgr->inst.addr = *myaddr; | |
2952 | ||
2953 | /* select a random nonce */ | |
ac8839d7 | 2954 | msgr->inst.addr.type = 0; |
103e2d3a | 2955 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2956 | encode_my_addr(msgr); |
31b8006e | 2957 | |
a2a32584 | 2958 | atomic_set(&msgr->stopping, 0); |
757856d2 | 2959 | write_pnet(&msgr->net, get_net(current->nsproxy->net_ns)); |
31b8006e | 2960 | |
15d9882c | 2961 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2962 | } |
15d9882c | 2963 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2964 | |
757856d2 ID |
2965 | void ceph_messenger_fini(struct ceph_messenger *msgr) |
2966 | { | |
2967 | put_net(read_pnet(&msgr->net)); | |
2968 | } | |
2969 | EXPORT_SYMBOL(ceph_messenger_fini); | |
2970 | ||
583d0fef ID |
2971 | static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con) |
2972 | { | |
2973 | if (msg->con) | |
2974 | msg->con->ops->put(msg->con); | |
2975 | ||
2976 | msg->con = con ? con->ops->get(con) : NULL; | |
2977 | BUG_ON(msg->con != con); | |
2978 | } | |
2979 | ||
e00de341 SW |
2980 | static void clear_standby(struct ceph_connection *con) |
2981 | { | |
2982 | /* come back from STANDBY? */ | |
8dacc7da | 2983 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 2984 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 2985 | con->state = CON_STATE_PREOPEN; |
e00de341 | 2986 | con->connect_seq++; |
c9ffc77a AE |
2987 | WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING)); |
2988 | WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)); | |
e00de341 SW |
2989 | } |
2990 | } | |
2991 | ||
31b8006e SW |
2992 | /* |
2993 | * Queue up an outgoing message on the given connection. | |
2994 | */ | |
2995 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2996 | { | |
31b8006e | 2997 | /* set src+dst */ |
dbad185d | 2998 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 2999 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
3000 | msg->needs_out_seq = true; |
3001 | ||
ec302645 | 3002 | mutex_lock(&con->mutex); |
92ce034b | 3003 | |
8dacc7da | 3004 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
3005 | dout("con_send %p closed, dropping %p\n", con, msg); |
3006 | ceph_msg_put(msg); | |
3007 | mutex_unlock(&con->mutex); | |
3008 | return; | |
3009 | } | |
3010 | ||
583d0fef | 3011 | msg_con_set(msg, con); |
92ce034b | 3012 | |
31b8006e SW |
3013 | BUG_ON(!list_empty(&msg->list_head)); |
3014 | list_add_tail(&msg->list_head, &con->out_queue); | |
3015 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
3016 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
3017 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
3018 | le32_to_cpu(msg->hdr.front_len), | |
3019 | le32_to_cpu(msg->hdr.middle_len), | |
3020 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
3021 | |
3022 | clear_standby(con); | |
ec302645 | 3023 | mutex_unlock(&con->mutex); |
31b8006e SW |
3024 | |
3025 | /* if there wasn't anything waiting to send before, queue | |
3026 | * new work */ | |
c9ffc77a | 3027 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
31b8006e SW |
3028 | queue_con(con); |
3029 | } | |
3d14c5d2 | 3030 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
3031 | |
3032 | /* | |
3033 | * Revoke a message that was previously queued for send | |
3034 | */ | |
6740a845 | 3035 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 3036 | { |
6740a845 AE |
3037 | struct ceph_connection *con = msg->con; |
3038 | ||
583d0fef ID |
3039 | if (!con) { |
3040 | dout("%s msg %p null con\n", __func__, msg); | |
6740a845 | 3041 | return; /* Message not in our possession */ |
583d0fef | 3042 | } |
6740a845 | 3043 | |
ec302645 | 3044 | mutex_lock(&con->mutex); |
31b8006e | 3045 | if (!list_empty(&msg->list_head)) { |
38941f80 | 3046 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 3047 | list_del_init(&msg->list_head); |
31b8006e | 3048 | msg->hdr.seq = 0; |
38941f80 | 3049 | |
31b8006e | 3050 | ceph_msg_put(msg); |
ed98adad SW |
3051 | } |
3052 | if (con->out_msg == msg) { | |
38941f80 | 3053 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 3054 | con->out_msg = NULL; |
31b8006e SW |
3055 | if (con->out_kvec_is_msg) { |
3056 | con->out_skip = con->out_kvec_bytes; | |
3057 | con->out_kvec_is_msg = false; | |
3058 | } | |
ed98adad | 3059 | msg->hdr.seq = 0; |
92ce034b AE |
3060 | |
3061 | ceph_msg_put(msg); | |
31b8006e | 3062 | } |
ec302645 | 3063 | mutex_unlock(&con->mutex); |
31b8006e SW |
3064 | } |
3065 | ||
350b1c32 | 3066 | /* |
0d59ab81 | 3067 | * Revoke a message that we may be reading data into |
350b1c32 | 3068 | */ |
8921d114 | 3069 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 3070 | { |
583d0fef | 3071 | struct ceph_connection *con = msg->con; |
8921d114 | 3072 | |
583d0fef | 3073 | if (!con) { |
8921d114 | 3074 | dout("%s msg %p null con\n", __func__, msg); |
8921d114 AE |
3075 | return; /* Message not in our possession */ |
3076 | } | |
3077 | ||
350b1c32 | 3078 | mutex_lock(&con->mutex); |
8921d114 | 3079 | if (con->in_msg == msg) { |
95c96174 ED |
3080 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
3081 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
3082 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
3083 | |
3084 | /* skip rest of message */ | |
8921d114 AE |
3085 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
3086 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 3087 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
3088 | front_len - |
3089 | middle_len - | |
3090 | data_len - | |
350b1c32 | 3091 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
3092 | ceph_msg_put(con->in_msg); |
3093 | con->in_msg = NULL; | |
3094 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 3095 | con->in_seq++; |
350b1c32 | 3096 | } else { |
8921d114 AE |
3097 | dout("%s %p in_msg %p msg %p no-op\n", |
3098 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
3099 | } |
3100 | mutex_unlock(&con->mutex); | |
3101 | } | |
3102 | ||
31b8006e SW |
3103 | /* |
3104 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
3105 | */ | |
3106 | void ceph_con_keepalive(struct ceph_connection *con) | |
3107 | { | |
e00de341 | 3108 | dout("con_keepalive %p\n", con); |
00650931 | 3109 | mutex_lock(&con->mutex); |
e00de341 | 3110 | clear_standby(con); |
00650931 | 3111 | mutex_unlock(&con->mutex); |
c9ffc77a AE |
3112 | if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 && |
3113 | con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) | |
31b8006e SW |
3114 | queue_con(con); |
3115 | } | |
3d14c5d2 | 3116 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e | 3117 | |
8b9558aa YZ |
3118 | bool ceph_con_keepalive_expired(struct ceph_connection *con, |
3119 | unsigned long interval) | |
3120 | { | |
3121 | if (interval > 0 && | |
3122 | (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) { | |
3123 | struct timespec now = CURRENT_TIME; | |
3124 | struct timespec ts; | |
3125 | jiffies_to_timespec(interval, &ts); | |
3126 | ts = timespec_add(con->last_keepalive_ack, ts); | |
3127 | return timespec_compare(&now, &ts) >= 0; | |
3128 | } | |
3129 | return false; | |
3130 | } | |
3131 | ||
6644ed7b | 3132 | static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type) |
43794509 | 3133 | { |
6644ed7b AE |
3134 | struct ceph_msg_data *data; |
3135 | ||
3136 | if (WARN_ON(!ceph_msg_data_type_valid(type))) | |
3137 | return NULL; | |
3138 | ||
81b36be4 | 3139 | data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS); |
6644ed7b AE |
3140 | if (data) |
3141 | data->type = type; | |
5240d9f9 | 3142 | INIT_LIST_HEAD(&data->links); |
6644ed7b AE |
3143 | |
3144 | return data; | |
3145 | } | |
3146 | ||
3147 | static void ceph_msg_data_destroy(struct ceph_msg_data *data) | |
3148 | { | |
3149 | if (!data) | |
3150 | return; | |
3151 | ||
5240d9f9 | 3152 | WARN_ON(!list_empty(&data->links)); |
e4339d28 | 3153 | if (data->type == CEPH_MSG_DATA_PAGELIST) |
6644ed7b | 3154 | ceph_pagelist_release(data->pagelist); |
81b36be4 | 3155 | kmem_cache_free(ceph_msg_data_cache, data); |
43794509 AE |
3156 | } |
3157 | ||
90af3602 | 3158 | void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages, |
f1baeb2b | 3159 | size_t length, size_t alignment) |
02afca6c | 3160 | { |
6644ed7b AE |
3161 | struct ceph_msg_data *data; |
3162 | ||
07aa1558 AE |
3163 | BUG_ON(!pages); |
3164 | BUG_ON(!length); | |
6644ed7b AE |
3165 | |
3166 | data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES); | |
3167 | BUG_ON(!data); | |
3168 | data->pages = pages; | |
3169 | data->length = length; | |
3170 | data->alignment = alignment & ~PAGE_MASK; | |
02afca6c | 3171 | |
5240d9f9 AE |
3172 | list_add_tail(&data->links, &msg->data); |
3173 | msg->data_length += length; | |
02afca6c | 3174 | } |
90af3602 | 3175 | EXPORT_SYMBOL(ceph_msg_data_add_pages); |
31b8006e | 3176 | |
90af3602 | 3177 | void ceph_msg_data_add_pagelist(struct ceph_msg *msg, |
27fa8385 AE |
3178 | struct ceph_pagelist *pagelist) |
3179 | { | |
6644ed7b AE |
3180 | struct ceph_msg_data *data; |
3181 | ||
07aa1558 AE |
3182 | BUG_ON(!pagelist); |
3183 | BUG_ON(!pagelist->length); | |
27fa8385 | 3184 | |
6644ed7b AE |
3185 | data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST); |
3186 | BUG_ON(!data); | |
3187 | data->pagelist = pagelist; | |
3188 | ||
5240d9f9 AE |
3189 | list_add_tail(&data->links, &msg->data); |
3190 | msg->data_length += pagelist->length; | |
27fa8385 | 3191 | } |
90af3602 | 3192 | EXPORT_SYMBOL(ceph_msg_data_add_pagelist); |
27fa8385 | 3193 | |
ea96571f | 3194 | #ifdef CONFIG_BLOCK |
90af3602 | 3195 | void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio, |
a1930804 | 3196 | size_t length) |
27fa8385 | 3197 | { |
6644ed7b AE |
3198 | struct ceph_msg_data *data; |
3199 | ||
07aa1558 | 3200 | BUG_ON(!bio); |
27fa8385 | 3201 | |
6644ed7b AE |
3202 | data = ceph_msg_data_create(CEPH_MSG_DATA_BIO); |
3203 | BUG_ON(!data); | |
3204 | data->bio = bio; | |
c851c495 | 3205 | data->bio_length = length; |
6644ed7b | 3206 | |
5240d9f9 AE |
3207 | list_add_tail(&data->links, &msg->data); |
3208 | msg->data_length += length; | |
27fa8385 | 3209 | } |
90af3602 | 3210 | EXPORT_SYMBOL(ceph_msg_data_add_bio); |
ea96571f | 3211 | #endif /* CONFIG_BLOCK */ |
27fa8385 | 3212 | |
31b8006e SW |
3213 | /* |
3214 | * construct a new message with given type, size | |
3215 | * the new msg has a ref count of 1. | |
3216 | */ | |
b61c2763 SW |
3217 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
3218 | bool can_fail) | |
31b8006e SW |
3219 | { |
3220 | struct ceph_msg *m; | |
3221 | ||
e3d5d638 | 3222 | m = kmem_cache_zalloc(ceph_msg_cache, flags); |
31b8006e SW |
3223 | if (m == NULL) |
3224 | goto out; | |
31b8006e SW |
3225 | |
3226 | m->hdr.type = cpu_to_le16(type); | |
45c6ceb5 | 3227 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
31b8006e | 3228 | m->hdr.front_len = cpu_to_le32(front_len); |
ca20892d | 3229 | |
9516e45b AE |
3230 | INIT_LIST_HEAD(&m->list_head); |
3231 | kref_init(&m->kref); | |
5240d9f9 | 3232 | INIT_LIST_HEAD(&m->data); |
ca20892d | 3233 | |
31b8006e SW |
3234 | /* front */ |
3235 | if (front_len) { | |
eeb0bed5 | 3236 | m->front.iov_base = ceph_kvmalloc(front_len, flags); |
31b8006e | 3237 | if (m->front.iov_base == NULL) { |
b61c2763 | 3238 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
3239 | front_len); |
3240 | goto out2; | |
3241 | } | |
3242 | } else { | |
3243 | m->front.iov_base = NULL; | |
3244 | } | |
f2be82b0 | 3245 | m->front_alloc_len = m->front.iov_len = front_len; |
31b8006e | 3246 | |
bb257664 | 3247 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
3248 | return m; |
3249 | ||
3250 | out2: | |
3251 | ceph_msg_put(m); | |
3252 | out: | |
b61c2763 SW |
3253 | if (!can_fail) { |
3254 | pr_err("msg_new can't create type %d front %d\n", type, | |
3255 | front_len); | |
f0ed1b7c | 3256 | WARN_ON(1); |
b61c2763 SW |
3257 | } else { |
3258 | dout("msg_new can't create type %d front %d\n", type, | |
3259 | front_len); | |
3260 | } | |
a79832f2 | 3261 | return NULL; |
31b8006e | 3262 | } |
3d14c5d2 | 3263 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 3264 | |
31b8006e SW |
3265 | /* |
3266 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
3267 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
3268 | * per-type header in the front and then gracefully fail (i.e., | |
3269 | * propagate the error to the caller based on info in the front) when | |
3270 | * the middle is too large. | |
3271 | */ | |
2450418c | 3272 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
3273 | { |
3274 | int type = le16_to_cpu(msg->hdr.type); | |
3275 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
3276 | ||
3277 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
3278 | ceph_msg_type_name(type), middle_len); | |
3279 | BUG_ON(!middle_len); | |
3280 | BUG_ON(msg->middle); | |
3281 | ||
b6c1d5b8 | 3282 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
3283 | if (!msg->middle) |
3284 | return -ENOMEM; | |
3285 | return 0; | |
3286 | } | |
3287 | ||
2450418c | 3288 | /* |
1c20f2d2 AE |
3289 | * Allocate a message for receiving an incoming message on a |
3290 | * connection, and save the result in con->in_msg. Uses the | |
3291 | * connection's private alloc_msg op if available. | |
3292 | * | |
4740a623 SW |
3293 | * Returns 0 on success, or a negative error code. |
3294 | * | |
3295 | * On success, if we set *skip = 1: | |
3296 | * - the next message should be skipped and ignored. | |
3297 | * - con->in_msg == NULL | |
3298 | * or if we set *skip = 0: | |
3299 | * - con->in_msg is non-null. | |
3300 | * On error (ENOMEM, EAGAIN, ...), | |
3301 | * - con->in_msg == NULL | |
2450418c | 3302 | */ |
4740a623 | 3303 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
2450418c | 3304 | { |
4740a623 | 3305 | struct ceph_msg_header *hdr = &con->in_hdr; |
2450418c | 3306 | int middle_len = le32_to_cpu(hdr->middle_len); |
1d866d1c | 3307 | struct ceph_msg *msg; |
4740a623 | 3308 | int ret = 0; |
2450418c | 3309 | |
1c20f2d2 | 3310 | BUG_ON(con->in_msg != NULL); |
53ded495 | 3311 | BUG_ON(!con->ops->alloc_msg); |
2450418c | 3312 | |
53ded495 AE |
3313 | mutex_unlock(&con->mutex); |
3314 | msg = con->ops->alloc_msg(con, hdr, skip); | |
3315 | mutex_lock(&con->mutex); | |
3316 | if (con->state != CON_STATE_OPEN) { | |
3317 | if (msg) | |
1d866d1c | 3318 | ceph_msg_put(msg); |
53ded495 AE |
3319 | return -EAGAIN; |
3320 | } | |
4137577a AE |
3321 | if (msg) { |
3322 | BUG_ON(*skip); | |
583d0fef | 3323 | msg_con_set(msg, con); |
4137577a | 3324 | con->in_msg = msg; |
4137577a AE |
3325 | } else { |
3326 | /* | |
3327 | * Null message pointer means either we should skip | |
3328 | * this message or we couldn't allocate memory. The | |
3329 | * former is not an error. | |
3330 | */ | |
3331 | if (*skip) | |
3332 | return 0; | |
4137577a | 3333 | |
67c64eb7 | 3334 | con->error_msg = "error allocating memory for incoming message"; |
53ded495 | 3335 | return -ENOMEM; |
2450418c | 3336 | } |
1c20f2d2 | 3337 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 3338 | |
1c20f2d2 AE |
3339 | if (middle_len && !con->in_msg->middle) { |
3340 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 3341 | if (ret < 0) { |
1c20f2d2 AE |
3342 | ceph_msg_put(con->in_msg); |
3343 | con->in_msg = NULL; | |
2450418c YS |
3344 | } |
3345 | } | |
9d7f0f13 | 3346 | |
4740a623 | 3347 | return ret; |
2450418c YS |
3348 | } |
3349 | ||
31b8006e SW |
3350 | |
3351 | /* | |
3352 | * Free a generically kmalloc'd message. | |
3353 | */ | |
0215e44b | 3354 | static void ceph_msg_free(struct ceph_msg *m) |
31b8006e | 3355 | { |
0215e44b | 3356 | dout("%s %p\n", __func__, m); |
4965fc38 | 3357 | kvfree(m->front.iov_base); |
e3d5d638 | 3358 | kmem_cache_free(ceph_msg_cache, m); |
31b8006e SW |
3359 | } |
3360 | ||
0215e44b | 3361 | static void ceph_msg_release(struct kref *kref) |
c2e552e7 SW |
3362 | { |
3363 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
5240d9f9 AE |
3364 | LIST_HEAD(data); |
3365 | struct list_head *links; | |
3366 | struct list_head *next; | |
31b8006e | 3367 | |
0215e44b | 3368 | dout("%s %p\n", __func__, m); |
c2e552e7 SW |
3369 | WARN_ON(!list_empty(&m->list_head)); |
3370 | ||
583d0fef ID |
3371 | msg_con_set(m, NULL); |
3372 | ||
c2e552e7 SW |
3373 | /* drop middle, data, if any */ |
3374 | if (m->middle) { | |
3375 | ceph_buffer_put(m->middle); | |
3376 | m->middle = NULL; | |
31b8006e | 3377 | } |
5240d9f9 AE |
3378 | |
3379 | list_splice_init(&m->data, &data); | |
3380 | list_for_each_safe(links, next, &data) { | |
3381 | struct ceph_msg_data *data; | |
3382 | ||
3383 | data = list_entry(links, struct ceph_msg_data, links); | |
3384 | list_del_init(links); | |
3385 | ceph_msg_data_destroy(data); | |
3386 | } | |
a1930804 | 3387 | m->data_length = 0; |
58bb3b37 | 3388 | |
c2e552e7 SW |
3389 | if (m->pool) |
3390 | ceph_msgpool_put(m->pool, m); | |
3391 | else | |
0215e44b ID |
3392 | ceph_msg_free(m); |
3393 | } | |
3394 | ||
3395 | struct ceph_msg *ceph_msg_get(struct ceph_msg *msg) | |
3396 | { | |
3397 | dout("%s %p (was %d)\n", __func__, msg, | |
3398 | atomic_read(&msg->kref.refcount)); | |
3399 | kref_get(&msg->kref); | |
3400 | return msg; | |
3401 | } | |
3402 | EXPORT_SYMBOL(ceph_msg_get); | |
3403 | ||
3404 | void ceph_msg_put(struct ceph_msg *msg) | |
3405 | { | |
3406 | dout("%s %p (was %d)\n", __func__, msg, | |
3407 | atomic_read(&msg->kref.refcount)); | |
3408 | kref_put(&msg->kref, ceph_msg_release); | |
31b8006e | 3409 | } |
0215e44b | 3410 | EXPORT_SYMBOL(ceph_msg_put); |
9ec7cab1 SW |
3411 | |
3412 | void ceph_msg_dump(struct ceph_msg *msg) | |
3413 | { | |
3cea4c30 ID |
3414 | pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg, |
3415 | msg->front_alloc_len, msg->data_length); | |
9ec7cab1 SW |
3416 | print_hex_dump(KERN_DEBUG, "header: ", |
3417 | DUMP_PREFIX_OFFSET, 16, 1, | |
3418 | &msg->hdr, sizeof(msg->hdr), true); | |
3419 | print_hex_dump(KERN_DEBUG, " front: ", | |
3420 | DUMP_PREFIX_OFFSET, 16, 1, | |
3421 | msg->front.iov_base, msg->front.iov_len, true); | |
3422 | if (msg->middle) | |
3423 | print_hex_dump(KERN_DEBUG, "middle: ", | |
3424 | DUMP_PREFIX_OFFSET, 16, 1, | |
3425 | msg->middle->vec.iov_base, | |
3426 | msg->middle->vec.iov_len, true); | |
3427 | print_hex_dump(KERN_DEBUG, "footer: ", | |
3428 | DUMP_PREFIX_OFFSET, 16, 1, | |
3429 | &msg->footer, sizeof(msg->footer), true); | |
3430 | } | |
3d14c5d2 | 3431 | EXPORT_SYMBOL(ceph_msg_dump); |