]>
Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
31b8006e SW |
2 | |
3 | #include <linux/crc32c.h> | |
4 | #include <linux/ctype.h> | |
5 | #include <linux/highmem.h> | |
6 | #include <linux/inet.h> | |
7 | #include <linux/kthread.h> | |
8 | #include <linux/net.h> | |
5a0e3ad6 | 9 | #include <linux/slab.h> |
31b8006e SW |
10 | #include <linux/socket.h> |
11 | #include <linux/string.h> | |
3ebc21f7 | 12 | #ifdef CONFIG_BLOCK |
68b4476b | 13 | #include <linux/bio.h> |
3ebc21f7 | 14 | #endif /* CONFIG_BLOCK */ |
ee3b56f2 | 15 | #include <linux/dns_resolver.h> |
31b8006e SW |
16 | #include <net/tcp.h> |
17 | ||
2b3e0c90 | 18 | #include <linux/ceph/ceph_features.h> |
3d14c5d2 YS |
19 | #include <linux/ceph/libceph.h> |
20 | #include <linux/ceph/messenger.h> | |
21 | #include <linux/ceph/decode.h> | |
22 | #include <linux/ceph/pagelist.h> | |
bc3b2d7f | 23 | #include <linux/export.h> |
31b8006e | 24 | |
fe38a2b6 AE |
25 | #define list_entry_next(pos, member) \ |
26 | list_entry(pos->member.next, typeof(*pos), member) | |
27 | ||
31b8006e SW |
28 | /* |
29 | * Ceph uses the messenger to exchange ceph_msg messages with other | |
30 | * hosts in the system. The messenger provides ordered and reliable | |
31 | * delivery. We tolerate TCP disconnects by reconnecting (with | |
32 | * exponential backoff) in the case of a fault (disconnection, bad | |
33 | * crc, protocol error). Acks allow sent messages to be discarded by | |
34 | * the sender. | |
35 | */ | |
36 | ||
bc18f4b1 AE |
37 | /* |
38 | * We track the state of the socket on a given connection using | |
39 | * values defined below. The transition to a new socket state is | |
40 | * handled by a function which verifies we aren't coming from an | |
41 | * unexpected state. | |
42 | * | |
43 | * -------- | |
44 | * | NEW* | transient initial state | |
45 | * -------- | |
46 | * | con_sock_state_init() | |
47 | * v | |
48 | * ---------- | |
49 | * | CLOSED | initialized, but no socket (and no | |
50 | * ---------- TCP connection) | |
51 | * ^ \ | |
52 | * | \ con_sock_state_connecting() | |
53 | * | ---------------------- | |
54 | * | \ | |
55 | * + con_sock_state_closed() \ | |
fbb85a47 SW |
56 | * |+--------------------------- \ |
57 | * | \ \ \ | |
58 | * | ----------- \ \ | |
59 | * | | CLOSING | socket event; \ \ | |
60 | * | ----------- await close \ \ | |
61 | * | ^ \ | | |
62 | * | | \ | | |
63 | * | + con_sock_state_closing() \ | | |
64 | * | / \ | | | |
65 | * | / --------------- | | | |
66 | * | / \ v v | |
bc18f4b1 AE |
67 | * | / -------------- |
68 | * | / -----------------| CONNECTING | socket created, TCP | |
69 | * | | / -------------- connect initiated | |
70 | * | | | con_sock_state_connected() | |
71 | * | | v | |
72 | * ------------- | |
73 | * | CONNECTED | TCP connection established | |
74 | * ------------- | |
75 | * | |
76 | * State values for ceph_connection->sock_state; NEW is assumed to be 0. | |
77 | */ | |
ce2c8903 AE |
78 | |
79 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ | |
80 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ | |
81 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ | |
82 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ | |
83 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ | |
84 | ||
8dacc7da SW |
85 | /* |
86 | * connection states | |
87 | */ | |
88 | #define CON_STATE_CLOSED 1 /* -> PREOPEN */ | |
89 | #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */ | |
90 | #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */ | |
91 | #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */ | |
92 | #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */ | |
93 | #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */ | |
94 | ||
4a861692 SW |
95 | /* |
96 | * ceph_connection flag bits | |
97 | */ | |
98 | #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop | |
99 | * messages on errors */ | |
100 | #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */ | |
101 | #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */ | |
102 | #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */ | |
103 | #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */ | |
8dacc7da | 104 | |
c9ffc77a AE |
105 | static bool con_flag_valid(unsigned long con_flag) |
106 | { | |
107 | switch (con_flag) { | |
108 | case CON_FLAG_LOSSYTX: | |
109 | case CON_FLAG_KEEPALIVE_PENDING: | |
110 | case CON_FLAG_WRITE_PENDING: | |
111 | case CON_FLAG_SOCK_CLOSED: | |
112 | case CON_FLAG_BACKOFF: | |
113 | return true; | |
114 | default: | |
115 | return false; | |
116 | } | |
117 | } | |
118 | ||
119 | static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag) | |
120 | { | |
121 | BUG_ON(!con_flag_valid(con_flag)); | |
122 | ||
123 | clear_bit(con_flag, &con->flags); | |
124 | } | |
125 | ||
126 | static void con_flag_set(struct ceph_connection *con, unsigned long con_flag) | |
127 | { | |
128 | BUG_ON(!con_flag_valid(con_flag)); | |
129 | ||
130 | set_bit(con_flag, &con->flags); | |
131 | } | |
132 | ||
133 | static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag) | |
134 | { | |
135 | BUG_ON(!con_flag_valid(con_flag)); | |
136 | ||
137 | return test_bit(con_flag, &con->flags); | |
138 | } | |
139 | ||
140 | static bool con_flag_test_and_clear(struct ceph_connection *con, | |
141 | unsigned long con_flag) | |
142 | { | |
143 | BUG_ON(!con_flag_valid(con_flag)); | |
144 | ||
145 | return test_and_clear_bit(con_flag, &con->flags); | |
146 | } | |
147 | ||
148 | static bool con_flag_test_and_set(struct ceph_connection *con, | |
149 | unsigned long con_flag) | |
150 | { | |
151 | BUG_ON(!con_flag_valid(con_flag)); | |
152 | ||
153 | return test_and_set_bit(con_flag, &con->flags); | |
154 | } | |
155 | ||
e3d5d638 AE |
156 | /* Slab caches for frequently-allocated structures */ |
157 | ||
158 | static struct kmem_cache *ceph_msg_cache; | |
81b36be4 | 159 | static struct kmem_cache *ceph_msg_data_cache; |
e3d5d638 | 160 | |
31b8006e SW |
161 | /* static tag bytes (protocol control messages) */ |
162 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
163 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
164 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
165 | ||
a6a5349d SW |
166 | #ifdef CONFIG_LOCKDEP |
167 | static struct lock_class_key socket_class; | |
168 | #endif | |
169 | ||
84495f49 AE |
170 | /* |
171 | * When skipping (ignoring) a block of input we read it into a "skip | |
172 | * buffer," which is this many bytes in size. | |
173 | */ | |
174 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
175 | |
176 | static void queue_con(struct ceph_connection *con); | |
177 | static void con_work(struct work_struct *); | |
93209264 | 178 | static void con_fault(struct ceph_connection *con); |
31b8006e | 179 | |
31b8006e | 180 | /* |
f64a9317 AE |
181 | * Nicely render a sockaddr as a string. An array of formatted |
182 | * strings is used, to approximate reentrancy. | |
31b8006e | 183 | */ |
f64a9317 AE |
184 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
185 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
186 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
187 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
188 | ||
189 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
190 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 191 | |
57666519 | 192 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 193 | |
3d14c5d2 | 194 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
195 | { |
196 | int i; | |
197 | char *s; | |
99f0f3b2 AE |
198 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
199 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 200 | |
f64a9317 | 201 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
202 | s = addr_str[i]; |
203 | ||
204 | switch (ss->ss_family) { | |
205 | case AF_INET: | |
bd406145 AE |
206 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
207 | ntohs(in4->sin_port)); | |
31b8006e SW |
208 | break; |
209 | ||
210 | case AF_INET6: | |
bd406145 AE |
211 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
212 | ntohs(in6->sin6_port)); | |
31b8006e SW |
213 | break; |
214 | ||
215 | default: | |
d3002b97 AE |
216 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
217 | ss->ss_family); | |
31b8006e SW |
218 | } |
219 | ||
220 | return s; | |
221 | } | |
3d14c5d2 | 222 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 223 | |
63f2d211 SW |
224 | static void encode_my_addr(struct ceph_messenger *msgr) |
225 | { | |
226 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
227 | ceph_encode_addr(&msgr->my_enc_addr); | |
228 | } | |
229 | ||
31b8006e SW |
230 | /* |
231 | * work queue for all reading and writing to/from the socket. | |
232 | */ | |
e0f43c94 | 233 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 234 | |
e3d5d638 AE |
235 | static int ceph_msgr_slab_init(void) |
236 | { | |
237 | BUG_ON(ceph_msg_cache); | |
238 | ceph_msg_cache = kmem_cache_create("ceph_msg", | |
239 | sizeof (struct ceph_msg), | |
240 | __alignof__(struct ceph_msg), 0, NULL); | |
81b36be4 AE |
241 | |
242 | if (!ceph_msg_cache) | |
243 | return -ENOMEM; | |
244 | ||
245 | BUG_ON(ceph_msg_data_cache); | |
246 | ceph_msg_data_cache = kmem_cache_create("ceph_msg_data", | |
247 | sizeof (struct ceph_msg_data), | |
248 | __alignof__(struct ceph_msg_data), | |
249 | 0, NULL); | |
250 | if (ceph_msg_data_cache) | |
251 | return 0; | |
252 | ||
253 | kmem_cache_destroy(ceph_msg_cache); | |
254 | ceph_msg_cache = NULL; | |
255 | ||
256 | return -ENOMEM; | |
e3d5d638 AE |
257 | } |
258 | ||
259 | static void ceph_msgr_slab_exit(void) | |
260 | { | |
81b36be4 AE |
261 | BUG_ON(!ceph_msg_data_cache); |
262 | kmem_cache_destroy(ceph_msg_data_cache); | |
263 | ceph_msg_data_cache = NULL; | |
264 | ||
e3d5d638 AE |
265 | BUG_ON(!ceph_msg_cache); |
266 | kmem_cache_destroy(ceph_msg_cache); | |
267 | ceph_msg_cache = NULL; | |
268 | } | |
269 | ||
15417167 | 270 | static void _ceph_msgr_exit(void) |
6173d1f0 | 271 | { |
d3002b97 | 272 | if (ceph_msgr_wq) { |
6173d1f0 | 273 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
274 | ceph_msgr_wq = NULL; |
275 | } | |
6173d1f0 | 276 | |
e3d5d638 AE |
277 | ceph_msgr_slab_exit(); |
278 | ||
6173d1f0 AE |
279 | BUG_ON(zero_page == NULL); |
280 | kunmap(zero_page); | |
281 | page_cache_release(zero_page); | |
282 | zero_page = NULL; | |
283 | } | |
284 | ||
3d14c5d2 | 285 | int ceph_msgr_init(void) |
31b8006e | 286 | { |
57666519 AE |
287 | BUG_ON(zero_page != NULL); |
288 | zero_page = ZERO_PAGE(0); | |
289 | page_cache_get(zero_page); | |
290 | ||
e3d5d638 AE |
291 | if (ceph_msgr_slab_init()) |
292 | return -ENOMEM; | |
293 | ||
4d1829a5 | 294 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", 0, 0); |
6173d1f0 AE |
295 | if (ceph_msgr_wq) |
296 | return 0; | |
57666519 | 297 | |
6173d1f0 AE |
298 | pr_err("msgr_init failed to create workqueue\n"); |
299 | _ceph_msgr_exit(); | |
57666519 | 300 | |
6173d1f0 | 301 | return -ENOMEM; |
31b8006e | 302 | } |
3d14c5d2 | 303 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
304 | |
305 | void ceph_msgr_exit(void) | |
306 | { | |
57666519 | 307 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 308 | |
6173d1f0 | 309 | _ceph_msgr_exit(); |
31b8006e | 310 | } |
3d14c5d2 | 311 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 312 | |
cd84db6e | 313 | void ceph_msgr_flush(void) |
a922d38f SW |
314 | { |
315 | flush_workqueue(ceph_msgr_wq); | |
316 | } | |
3d14c5d2 | 317 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 318 | |
ce2c8903 AE |
319 | /* Connection socket state transition functions */ |
320 | ||
321 | static void con_sock_state_init(struct ceph_connection *con) | |
322 | { | |
323 | int old_state; | |
324 | ||
325 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
326 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
327 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
328 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
329 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 AE |
330 | } |
331 | ||
332 | static void con_sock_state_connecting(struct ceph_connection *con) | |
333 | { | |
334 | int old_state; | |
335 | ||
336 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
337 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
338 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
339 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
340 | CON_SOCK_STATE_CONNECTING); | |
ce2c8903 AE |
341 | } |
342 | ||
343 | static void con_sock_state_connected(struct ceph_connection *con) | |
344 | { | |
345 | int old_state; | |
346 | ||
347 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
348 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
349 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
350 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
351 | CON_SOCK_STATE_CONNECTED); | |
ce2c8903 AE |
352 | } |
353 | ||
354 | static void con_sock_state_closing(struct ceph_connection *con) | |
355 | { | |
356 | int old_state; | |
357 | ||
358 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
359 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
360 | old_state != CON_SOCK_STATE_CONNECTED && | |
361 | old_state != CON_SOCK_STATE_CLOSING)) | |
362 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
363 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
364 | CON_SOCK_STATE_CLOSING); | |
ce2c8903 AE |
365 | } |
366 | ||
367 | static void con_sock_state_closed(struct ceph_connection *con) | |
368 | { | |
369 | int old_state; | |
370 | ||
371 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
372 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 | 373 | old_state != CON_SOCK_STATE_CLOSING && |
8007b8d6 SW |
374 | old_state != CON_SOCK_STATE_CONNECTING && |
375 | old_state != CON_SOCK_STATE_CLOSED)) | |
ce2c8903 | 376 | printk("%s: unexpected old state %d\n", __func__, old_state); |
8007b8d6 SW |
377 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
378 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 | 379 | } |
a922d38f | 380 | |
31b8006e SW |
381 | /* |
382 | * socket callback functions | |
383 | */ | |
384 | ||
385 | /* data available on socket, or listen socket received a connect */ | |
327800bd | 386 | static void ceph_sock_data_ready(struct sock *sk, int count_unused) |
31b8006e | 387 | { |
bd406145 | 388 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
389 | if (atomic_read(&con->msgr->stopping)) { |
390 | return; | |
391 | } | |
bd406145 | 392 | |
31b8006e | 393 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 394 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
395 | con, con->state); |
396 | queue_con(con); | |
397 | } | |
398 | } | |
399 | ||
400 | /* socket has buffer space for writing */ | |
327800bd | 401 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 402 | { |
d3002b97 | 403 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 404 | |
182fac26 JS |
405 | /* only queue to workqueue if there is data we want to write, |
406 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 407 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
408 | * doesn't get called again until try_write() fills the socket |
409 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
410 | * and net/core/stream.c:sk_stream_write_space(). | |
411 | */ | |
c9ffc77a | 412 | if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) { |
64dc6130 | 413 | if (sk_stream_is_writeable(sk)) { |
327800bd | 414 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
415 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
416 | queue_con(con); | |
417 | } | |
31b8006e | 418 | } else { |
327800bd | 419 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 420 | } |
31b8006e SW |
421 | } |
422 | ||
423 | /* socket's state has changed */ | |
327800bd | 424 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 425 | { |
bd406145 | 426 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 427 | |
327800bd | 428 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
429 | con, con->state, sk->sk_state); |
430 | ||
31b8006e SW |
431 | switch (sk->sk_state) { |
432 | case TCP_CLOSE: | |
327800bd | 433 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 434 | case TCP_CLOSE_WAIT: |
327800bd | 435 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 436 | con_sock_state_closing(con); |
c9ffc77a | 437 | con_flag_set(con, CON_FLAG_SOCK_CLOSED); |
d65c9e0b | 438 | queue_con(con); |
31b8006e SW |
439 | break; |
440 | case TCP_ESTABLISHED: | |
327800bd | 441 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 442 | con_sock_state_connected(con); |
31b8006e SW |
443 | queue_con(con); |
444 | break; | |
d3002b97 AE |
445 | default: /* Everything else is uninteresting */ |
446 | break; | |
31b8006e SW |
447 | } |
448 | } | |
449 | ||
450 | /* | |
451 | * set up socket callbacks | |
452 | */ | |
453 | static void set_sock_callbacks(struct socket *sock, | |
454 | struct ceph_connection *con) | |
455 | { | |
456 | struct sock *sk = sock->sk; | |
bd406145 | 457 | sk->sk_user_data = con; |
327800bd AE |
458 | sk->sk_data_ready = ceph_sock_data_ready; |
459 | sk->sk_write_space = ceph_sock_write_space; | |
460 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
461 | } |
462 | ||
463 | ||
464 | /* | |
465 | * socket helpers | |
466 | */ | |
467 | ||
468 | /* | |
469 | * initiate connection to a remote socket. | |
470 | */ | |
41617d0c | 471 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 472 | { |
f91d3471 | 473 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
474 | struct socket *sock; |
475 | int ret; | |
476 | ||
477 | BUG_ON(con->sock); | |
f91d3471 SW |
478 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
479 | IPPROTO_TCP, &sock); | |
31b8006e | 480 | if (ret) |
41617d0c | 481 | return ret; |
31b8006e SW |
482 | sock->sk->sk_allocation = GFP_NOFS; |
483 | ||
a6a5349d SW |
484 | #ifdef CONFIG_LOCKDEP |
485 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
486 | #endif | |
487 | ||
31b8006e SW |
488 | set_sock_callbacks(sock, con); |
489 | ||
3d14c5d2 | 490 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 491 | |
89a86be0 | 492 | con_sock_state_connecting(con); |
f91d3471 SW |
493 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
494 | O_NONBLOCK); | |
31b8006e SW |
495 | if (ret == -EINPROGRESS) { |
496 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 497 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 498 | sock->sk->sk_state); |
a5bc3129 | 499 | } else if (ret < 0) { |
31b8006e | 500 | pr_err("connect %s error %d\n", |
3d14c5d2 | 501 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 502 | sock_release(sock); |
31b8006e | 503 | con->error_msg = "connect error"; |
31b8006e | 504 | |
41617d0c | 505 | return ret; |
a5bc3129 AE |
506 | } |
507 | con->sock = sock; | |
41617d0c | 508 | return 0; |
31b8006e SW |
509 | } |
510 | ||
511 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
512 | { | |
513 | struct kvec iov = {buf, len}; | |
514 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 515 | int r; |
31b8006e | 516 | |
98bdb0aa SW |
517 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
518 | if (r == -EAGAIN) | |
519 | r = 0; | |
520 | return r; | |
31b8006e SW |
521 | } |
522 | ||
afb3d90e AE |
523 | static int ceph_tcp_recvpage(struct socket *sock, struct page *page, |
524 | int page_offset, size_t length) | |
525 | { | |
526 | void *kaddr; | |
527 | int ret; | |
528 | ||
529 | BUG_ON(page_offset + length > PAGE_SIZE); | |
530 | ||
531 | kaddr = kmap(page); | |
532 | BUG_ON(!kaddr); | |
533 | ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length); | |
534 | kunmap(page); | |
535 | ||
536 | return ret; | |
537 | } | |
538 | ||
31b8006e SW |
539 | /* |
540 | * write something. @more is true if caller will be sending more data | |
541 | * shortly. | |
542 | */ | |
543 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
544 | size_t kvlen, size_t len, int more) | |
545 | { | |
546 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 547 | int r; |
31b8006e SW |
548 | |
549 | if (more) | |
550 | msg.msg_flags |= MSG_MORE; | |
551 | else | |
552 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
553 | ||
42961d23 SW |
554 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
555 | if (r == -EAGAIN) | |
556 | r = 0; | |
557 | return r; | |
31b8006e SW |
558 | } |
559 | ||
31739139 | 560 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
e1dcb128 | 561 | int offset, size_t size, bool more) |
31739139 AE |
562 | { |
563 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
564 | int ret; | |
565 | ||
566 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
567 | if (ret == -EAGAIN) | |
568 | ret = 0; | |
569 | ||
570 | return ret; | |
571 | } | |
572 | ||
31b8006e SW |
573 | |
574 | /* | |
575 | * Shutdown/close the socket for the given connection. | |
576 | */ | |
577 | static int con_close_socket(struct ceph_connection *con) | |
578 | { | |
8007b8d6 | 579 | int rc = 0; |
31b8006e SW |
580 | |
581 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
8007b8d6 SW |
582 | if (con->sock) { |
583 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
584 | sock_release(con->sock); | |
585 | con->sock = NULL; | |
586 | } | |
456ea468 AE |
587 | |
588 | /* | |
4a861692 | 589 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
590 | * independent of the connection mutex, and we could have |
591 | * received a socket close event before we had the chance to | |
592 | * shut the socket down. | |
593 | */ | |
c9ffc77a | 594 | con_flag_clear(con, CON_FLAG_SOCK_CLOSED); |
8007b8d6 | 595 | |
ce2c8903 | 596 | con_sock_state_closed(con); |
31b8006e SW |
597 | return rc; |
598 | } | |
599 | ||
600 | /* | |
601 | * Reset a connection. Discard all incoming and outgoing messages | |
602 | * and clear *_seq state. | |
603 | */ | |
604 | static void ceph_msg_remove(struct ceph_msg *msg) | |
605 | { | |
606 | list_del_init(&msg->list_head); | |
38941f80 | 607 | BUG_ON(msg->con == NULL); |
36eb71aa | 608 | msg->con->ops->put(msg->con); |
38941f80 AE |
609 | msg->con = NULL; |
610 | ||
31b8006e SW |
611 | ceph_msg_put(msg); |
612 | } | |
613 | static void ceph_msg_remove_list(struct list_head *head) | |
614 | { | |
615 | while (!list_empty(head)) { | |
616 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
617 | list_head); | |
618 | ceph_msg_remove(msg); | |
619 | } | |
620 | } | |
621 | ||
622 | static void reset_connection(struct ceph_connection *con) | |
623 | { | |
624 | /* reset connection, out_queue, msg_ and connect_seq */ | |
625 | /* discard existing out_queue and msg_seq */ | |
0fa6ebc6 | 626 | dout("reset_connection %p\n", con); |
31b8006e SW |
627 | ceph_msg_remove_list(&con->out_queue); |
628 | ceph_msg_remove_list(&con->out_sent); | |
629 | ||
cf3e5c40 | 630 | if (con->in_msg) { |
38941f80 AE |
631 | BUG_ON(con->in_msg->con != con); |
632 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
633 | ceph_msg_put(con->in_msg); |
634 | con->in_msg = NULL; | |
36eb71aa | 635 | con->ops->put(con); |
cf3e5c40 SW |
636 | } |
637 | ||
31b8006e SW |
638 | con->connect_seq = 0; |
639 | con->out_seq = 0; | |
c86a2930 SW |
640 | if (con->out_msg) { |
641 | ceph_msg_put(con->out_msg); | |
642 | con->out_msg = NULL; | |
643 | } | |
31b8006e | 644 | con->in_seq = 0; |
0e0d5e0c | 645 | con->in_seq_acked = 0; |
31b8006e SW |
646 | } |
647 | ||
648 | /* | |
649 | * mark a peer down. drop any open connections. | |
650 | */ | |
651 | void ceph_con_close(struct ceph_connection *con) | |
652 | { | |
8c50c817 | 653 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
654 | dout("con_close %p peer %s\n", con, |
655 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
8dacc7da | 656 | con->state = CON_STATE_CLOSED; |
a5988c49 | 657 | |
c9ffc77a AE |
658 | con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */ |
659 | con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING); | |
660 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); | |
661 | con_flag_clear(con, CON_FLAG_BACKOFF); | |
a5988c49 | 662 | |
31b8006e | 663 | reset_connection(con); |
6f2bc3ff | 664 | con->peer_global_seq = 0; |
91e45ce3 | 665 | cancel_delayed_work(&con->work); |
ee76e073 | 666 | con_close_socket(con); |
ec302645 | 667 | mutex_unlock(&con->mutex); |
31b8006e | 668 | } |
3d14c5d2 | 669 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 670 | |
31b8006e SW |
671 | /* |
672 | * Reopen a closed connection, with a new peer address. | |
673 | */ | |
b7a9e5dd SW |
674 | void ceph_con_open(struct ceph_connection *con, |
675 | __u8 entity_type, __u64 entity_num, | |
676 | struct ceph_entity_addr *addr) | |
31b8006e | 677 | { |
5469155f | 678 | mutex_lock(&con->mutex); |
3d14c5d2 | 679 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
8dacc7da | 680 | |
122070a2 | 681 | WARN_ON(con->state != CON_STATE_CLOSED); |
8dacc7da | 682 | con->state = CON_STATE_PREOPEN; |
a5988c49 | 683 | |
b7a9e5dd SW |
684 | con->peer_name.type = (__u8) entity_type; |
685 | con->peer_name.num = cpu_to_le64(entity_num); | |
686 | ||
31b8006e | 687 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 688 | con->delay = 0; /* reset backoff memory */ |
5469155f | 689 | mutex_unlock(&con->mutex); |
31b8006e SW |
690 | queue_con(con); |
691 | } | |
3d14c5d2 | 692 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 693 | |
87b315a5 SW |
694 | /* |
695 | * return true if this connection ever successfully opened | |
696 | */ | |
697 | bool ceph_con_opened(struct ceph_connection *con) | |
698 | { | |
699 | return con->connect_seq > 0; | |
700 | } | |
701 | ||
31b8006e SW |
702 | /* |
703 | * initialize a new connection. | |
704 | */ | |
1bfd89f4 AE |
705 | void ceph_con_init(struct ceph_connection *con, void *private, |
706 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 707 | struct ceph_messenger *msgr) |
31b8006e SW |
708 | { |
709 | dout("con_init %p\n", con); | |
710 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
711 | con->private = private; |
712 | con->ops = ops; | |
31b8006e | 713 | con->msgr = msgr; |
ce2c8903 AE |
714 | |
715 | con_sock_state_init(con); | |
716 | ||
ec302645 | 717 | mutex_init(&con->mutex); |
31b8006e SW |
718 | INIT_LIST_HEAD(&con->out_queue); |
719 | INIT_LIST_HEAD(&con->out_sent); | |
720 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 | 721 | |
8dacc7da | 722 | con->state = CON_STATE_CLOSED; |
31b8006e | 723 | } |
3d14c5d2 | 724 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
725 | |
726 | ||
727 | /* | |
728 | * We maintain a global counter to order connection attempts. Get | |
729 | * a unique seq greater than @gt. | |
730 | */ | |
731 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
732 | { | |
733 | u32 ret; | |
734 | ||
735 | spin_lock(&msgr->global_seq_lock); | |
736 | if (msgr->global_seq < gt) | |
737 | msgr->global_seq = gt; | |
738 | ret = ++msgr->global_seq; | |
739 | spin_unlock(&msgr->global_seq_lock); | |
740 | return ret; | |
741 | } | |
742 | ||
e2200423 | 743 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
744 | { |
745 | con->out_kvec_left = 0; | |
746 | con->out_kvec_bytes = 0; | |
747 | con->out_kvec_cur = &con->out_kvec[0]; | |
748 | } | |
749 | ||
e2200423 | 750 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
751 | size_t size, void *data) |
752 | { | |
753 | int index; | |
754 | ||
755 | index = con->out_kvec_left; | |
756 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
757 | ||
758 | con->out_kvec[index].iov_len = size; | |
759 | con->out_kvec[index].iov_base = data; | |
760 | con->out_kvec_left++; | |
761 | con->out_kvec_bytes += size; | |
762 | } | |
31b8006e | 763 | |
df6ad1f9 | 764 | #ifdef CONFIG_BLOCK |
6aaa4511 AE |
765 | |
766 | /* | |
767 | * For a bio data item, a piece is whatever remains of the next | |
768 | * entry in the current bio iovec, or the first entry in the next | |
769 | * bio in the list. | |
770 | */ | |
8ae4f4f5 | 771 | static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 772 | size_t length) |
6aaa4511 | 773 | { |
8ae4f4f5 | 774 | struct ceph_msg_data *data = cursor->data; |
6aaa4511 AE |
775 | struct bio *bio; |
776 | ||
777 | BUG_ON(data->type != CEPH_MSG_DATA_BIO); | |
778 | ||
779 | bio = data->bio; | |
780 | BUG_ON(!bio); | |
781 | BUG_ON(!bio->bi_vcnt); | |
6aaa4511 | 782 | |
ca8b3a69 | 783 | cursor->resid = min(length, data->bio_length); |
6aaa4511 AE |
784 | cursor->bio = bio; |
785 | cursor->vector_index = 0; | |
786 | cursor->vector_offset = 0; | |
25aff7c5 | 787 | cursor->last_piece = length <= bio->bi_io_vec[0].bv_len; |
6aaa4511 AE |
788 | } |
789 | ||
8ae4f4f5 | 790 | static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor, |
6aaa4511 AE |
791 | size_t *page_offset, |
792 | size_t *length) | |
793 | { | |
8ae4f4f5 | 794 | struct ceph_msg_data *data = cursor->data; |
6aaa4511 AE |
795 | struct bio *bio; |
796 | struct bio_vec *bio_vec; | |
797 | unsigned int index; | |
798 | ||
799 | BUG_ON(data->type != CEPH_MSG_DATA_BIO); | |
800 | ||
801 | bio = cursor->bio; | |
802 | BUG_ON(!bio); | |
803 | ||
804 | index = cursor->vector_index; | |
805 | BUG_ON(index >= (unsigned int) bio->bi_vcnt); | |
806 | ||
807 | bio_vec = &bio->bi_io_vec[index]; | |
808 | BUG_ON(cursor->vector_offset >= bio_vec->bv_len); | |
809 | *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset); | |
810 | BUG_ON(*page_offset >= PAGE_SIZE); | |
25aff7c5 AE |
811 | if (cursor->last_piece) /* pagelist offset is always 0 */ |
812 | *length = cursor->resid; | |
813 | else | |
814 | *length = (size_t) (bio_vec->bv_len - cursor->vector_offset); | |
25aff7c5 | 815 | BUG_ON(*length > cursor->resid); |
5df521b1 | 816 | BUG_ON(*page_offset + *length > PAGE_SIZE); |
6aaa4511 AE |
817 | |
818 | return bio_vec->bv_page; | |
819 | } | |
820 | ||
8ae4f4f5 AE |
821 | static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor, |
822 | size_t bytes) | |
6aaa4511 | 823 | { |
6aaa4511 AE |
824 | struct bio *bio; |
825 | struct bio_vec *bio_vec; | |
826 | unsigned int index; | |
827 | ||
8ae4f4f5 | 828 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO); |
6aaa4511 AE |
829 | |
830 | bio = cursor->bio; | |
831 | BUG_ON(!bio); | |
832 | ||
833 | index = cursor->vector_index; | |
834 | BUG_ON(index >= (unsigned int) bio->bi_vcnt); | |
835 | bio_vec = &bio->bi_io_vec[index]; | |
6aaa4511 AE |
836 | |
837 | /* Advance the cursor offset */ | |
838 | ||
25aff7c5 AE |
839 | BUG_ON(cursor->resid < bytes); |
840 | cursor->resid -= bytes; | |
6aaa4511 AE |
841 | cursor->vector_offset += bytes; |
842 | if (cursor->vector_offset < bio_vec->bv_len) | |
843 | return false; /* more bytes to process in this segment */ | |
25aff7c5 | 844 | BUG_ON(cursor->vector_offset != bio_vec->bv_len); |
6aaa4511 AE |
845 | |
846 | /* Move on to the next segment, and possibly the next bio */ | |
847 | ||
25aff7c5 | 848 | if (++index == (unsigned int) bio->bi_vcnt) { |
6aaa4511 | 849 | bio = bio->bi_next; |
25aff7c5 | 850 | index = 0; |
6aaa4511 | 851 | } |
25aff7c5 AE |
852 | cursor->bio = bio; |
853 | cursor->vector_index = index; | |
6aaa4511 AE |
854 | cursor->vector_offset = 0; |
855 | ||
25aff7c5 AE |
856 | if (!cursor->last_piece) { |
857 | BUG_ON(!cursor->resid); | |
858 | BUG_ON(!bio); | |
859 | /* A short read is OK, so use <= rather than == */ | |
860 | if (cursor->resid <= bio->bi_io_vec[index].bv_len) | |
6aaa4511 | 861 | cursor->last_piece = true; |
25aff7c5 | 862 | } |
6aaa4511 AE |
863 | |
864 | return true; | |
865 | } | |
ea96571f | 866 | #endif /* CONFIG_BLOCK */ |
df6ad1f9 | 867 | |
e766d7b5 AE |
868 | /* |
869 | * For a page array, a piece comes from the first page in the array | |
870 | * that has not already been fully consumed. | |
871 | */ | |
8ae4f4f5 | 872 | static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 873 | size_t length) |
e766d7b5 | 874 | { |
8ae4f4f5 | 875 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
876 | int page_count; |
877 | ||
878 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
879 | ||
880 | BUG_ON(!data->pages); | |
881 | BUG_ON(!data->length); | |
882 | ||
ca8b3a69 | 883 | cursor->resid = min(length, data->length); |
e766d7b5 | 884 | page_count = calc_pages_for(data->alignment, (u64)data->length); |
e766d7b5 AE |
885 | cursor->page_offset = data->alignment & ~PAGE_MASK; |
886 | cursor->page_index = 0; | |
56fc5659 AE |
887 | BUG_ON(page_count > (int)USHRT_MAX); |
888 | cursor->page_count = (unsigned short)page_count; | |
889 | BUG_ON(length > SIZE_MAX - cursor->page_offset); | |
890 | cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE; | |
e766d7b5 AE |
891 | } |
892 | ||
8ae4f4f5 AE |
893 | static struct page * |
894 | ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor, | |
895 | size_t *page_offset, size_t *length) | |
e766d7b5 | 896 | { |
8ae4f4f5 | 897 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
898 | |
899 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
900 | ||
901 | BUG_ON(cursor->page_index >= cursor->page_count); | |
902 | BUG_ON(cursor->page_offset >= PAGE_SIZE); | |
e766d7b5 AE |
903 | |
904 | *page_offset = cursor->page_offset; | |
25aff7c5 | 905 | if (cursor->last_piece) |
e766d7b5 | 906 | *length = cursor->resid; |
25aff7c5 | 907 | else |
e766d7b5 | 908 | *length = PAGE_SIZE - *page_offset; |
e766d7b5 AE |
909 | |
910 | return data->pages[cursor->page_index]; | |
911 | } | |
912 | ||
8ae4f4f5 | 913 | static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor, |
e766d7b5 AE |
914 | size_t bytes) |
915 | { | |
8ae4f4f5 | 916 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES); |
e766d7b5 AE |
917 | |
918 | BUG_ON(cursor->page_offset + bytes > PAGE_SIZE); | |
e766d7b5 AE |
919 | |
920 | /* Advance the cursor page offset */ | |
921 | ||
922 | cursor->resid -= bytes; | |
5df521b1 AE |
923 | cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK; |
924 | if (!bytes || cursor->page_offset) | |
e766d7b5 AE |
925 | return false; /* more bytes to process in the current page */ |
926 | ||
5df521b1 | 927 | /* Move on to the next page; offset is already at 0 */ |
e766d7b5 AE |
928 | |
929 | BUG_ON(cursor->page_index >= cursor->page_count); | |
e766d7b5 | 930 | cursor->page_index++; |
25aff7c5 | 931 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
932 | |
933 | return true; | |
934 | } | |
935 | ||
fe38a2b6 | 936 | /* |
dd236fcb AE |
937 | * For a pagelist, a piece is whatever remains to be consumed in the |
938 | * first page in the list, or the front of the next page. | |
fe38a2b6 | 939 | */ |
8ae4f4f5 AE |
940 | static void |
941 | ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor, | |
25aff7c5 | 942 | size_t length) |
fe38a2b6 | 943 | { |
8ae4f4f5 | 944 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
945 | struct ceph_pagelist *pagelist; |
946 | struct page *page; | |
947 | ||
dd236fcb | 948 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); |
fe38a2b6 AE |
949 | |
950 | pagelist = data->pagelist; | |
951 | BUG_ON(!pagelist); | |
25aff7c5 AE |
952 | |
953 | if (!length) | |
fe38a2b6 AE |
954 | return; /* pagelist can be assigned but empty */ |
955 | ||
956 | BUG_ON(list_empty(&pagelist->head)); | |
957 | page = list_first_entry(&pagelist->head, struct page, lru); | |
958 | ||
ca8b3a69 | 959 | cursor->resid = min(length, pagelist->length); |
fe38a2b6 AE |
960 | cursor->page = page; |
961 | cursor->offset = 0; | |
a51b272e | 962 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
963 | } |
964 | ||
8ae4f4f5 AE |
965 | static struct page * |
966 | ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor, | |
967 | size_t *page_offset, size_t *length) | |
fe38a2b6 | 968 | { |
8ae4f4f5 | 969 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 | 970 | struct ceph_pagelist *pagelist; |
fe38a2b6 AE |
971 | |
972 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
973 | ||
974 | pagelist = data->pagelist; | |
975 | BUG_ON(!pagelist); | |
976 | ||
977 | BUG_ON(!cursor->page); | |
25aff7c5 | 978 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); |
fe38a2b6 | 979 | |
5df521b1 | 980 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 | 981 | *page_offset = cursor->offset & ~PAGE_MASK; |
5df521b1 | 982 | if (cursor->last_piece) |
25aff7c5 AE |
983 | *length = cursor->resid; |
984 | else | |
985 | *length = PAGE_SIZE - *page_offset; | |
fe38a2b6 | 986 | |
8ae4f4f5 | 987 | return cursor->page; |
fe38a2b6 AE |
988 | } |
989 | ||
8ae4f4f5 | 990 | static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor, |
dd236fcb | 991 | size_t bytes) |
fe38a2b6 | 992 | { |
8ae4f4f5 | 993 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
994 | struct ceph_pagelist *pagelist; |
995 | ||
996 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
997 | ||
998 | pagelist = data->pagelist; | |
999 | BUG_ON(!pagelist); | |
25aff7c5 AE |
1000 | |
1001 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); | |
fe38a2b6 AE |
1002 | BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE); |
1003 | ||
1004 | /* Advance the cursor offset */ | |
1005 | ||
25aff7c5 | 1006 | cursor->resid -= bytes; |
fe38a2b6 | 1007 | cursor->offset += bytes; |
5df521b1 | 1008 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 AE |
1009 | if (!bytes || cursor->offset & ~PAGE_MASK) |
1010 | return false; /* more bytes to process in the current page */ | |
1011 | ||
1012 | /* Move on to the next page */ | |
1013 | ||
1014 | BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head)); | |
1015 | cursor->page = list_entry_next(cursor->page, lru); | |
25aff7c5 | 1016 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
1017 | |
1018 | return true; | |
1019 | } | |
1020 | ||
dd236fcb AE |
1021 | /* |
1022 | * Message data is handled (sent or received) in pieces, where each | |
1023 | * piece resides on a single page. The network layer might not | |
1024 | * consume an entire piece at once. A data item's cursor keeps | |
1025 | * track of which piece is next to process and how much remains to | |
1026 | * be processed in that piece. It also tracks whether the current | |
1027 | * piece is the last one in the data item. | |
1028 | */ | |
ca8b3a69 | 1029 | static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor) |
dd236fcb | 1030 | { |
ca8b3a69 | 1031 | size_t length = cursor->total_resid; |
8ae4f4f5 | 1032 | |
8ae4f4f5 | 1033 | switch (cursor->data->type) { |
dd236fcb | 1034 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1035 | ceph_msg_data_pagelist_cursor_init(cursor, length); |
dd236fcb | 1036 | break; |
e766d7b5 | 1037 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1038 | ceph_msg_data_pages_cursor_init(cursor, length); |
e766d7b5 | 1039 | break; |
dd236fcb AE |
1040 | #ifdef CONFIG_BLOCK |
1041 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1042 | ceph_msg_data_bio_cursor_init(cursor, length); |
6aaa4511 | 1043 | break; |
dd236fcb | 1044 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1045 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1046 | default: |
1047 | /* BUG(); */ | |
1048 | break; | |
1049 | } | |
8ae4f4f5 | 1050 | cursor->need_crc = true; |
dd236fcb AE |
1051 | } |
1052 | ||
ca8b3a69 AE |
1053 | static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length) |
1054 | { | |
1055 | struct ceph_msg_data_cursor *cursor = &msg->cursor; | |
1056 | struct ceph_msg_data *data; | |
1057 | ||
1058 | BUG_ON(!length); | |
1059 | BUG_ON(length > msg->data_length); | |
1060 | BUG_ON(list_empty(&msg->data)); | |
1061 | ||
ca8b3a69 AE |
1062 | cursor->data_head = &msg->data; |
1063 | cursor->total_resid = length; | |
1064 | data = list_first_entry(&msg->data, struct ceph_msg_data, links); | |
1065 | cursor->data = data; | |
1066 | ||
1067 | __ceph_msg_data_cursor_init(cursor); | |
1068 | } | |
1069 | ||
dd236fcb AE |
1070 | /* |
1071 | * Return the page containing the next piece to process for a given | |
1072 | * data item, and supply the page offset and length of that piece. | |
1073 | * Indicate whether this is the last piece in this data item. | |
1074 | */ | |
8ae4f4f5 AE |
1075 | static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor, |
1076 | size_t *page_offset, size_t *length, | |
dd236fcb AE |
1077 | bool *last_piece) |
1078 | { | |
1079 | struct page *page; | |
1080 | ||
8ae4f4f5 | 1081 | switch (cursor->data->type) { |
dd236fcb | 1082 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1083 | page = ceph_msg_data_pagelist_next(cursor, page_offset, length); |
dd236fcb | 1084 | break; |
e766d7b5 | 1085 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1086 | page = ceph_msg_data_pages_next(cursor, page_offset, length); |
e766d7b5 | 1087 | break; |
dd236fcb AE |
1088 | #ifdef CONFIG_BLOCK |
1089 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1090 | page = ceph_msg_data_bio_next(cursor, page_offset, length); |
6aaa4511 | 1091 | break; |
dd236fcb | 1092 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1093 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1094 | default: |
1095 | page = NULL; | |
1096 | break; | |
1097 | } | |
1098 | BUG_ON(!page); | |
1099 | BUG_ON(*page_offset + *length > PAGE_SIZE); | |
1100 | BUG_ON(!*length); | |
1101 | if (last_piece) | |
8ae4f4f5 | 1102 | *last_piece = cursor->last_piece; |
dd236fcb AE |
1103 | |
1104 | return page; | |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * Returns true if the result moves the cursor on to the next piece | |
1109 | * of the data item. | |
1110 | */ | |
8ae4f4f5 AE |
1111 | static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, |
1112 | size_t bytes) | |
dd236fcb AE |
1113 | { |
1114 | bool new_piece; | |
1115 | ||
25aff7c5 | 1116 | BUG_ON(bytes > cursor->resid); |
8ae4f4f5 | 1117 | switch (cursor->data->type) { |
dd236fcb | 1118 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1119 | new_piece = ceph_msg_data_pagelist_advance(cursor, bytes); |
dd236fcb | 1120 | break; |
e766d7b5 | 1121 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1122 | new_piece = ceph_msg_data_pages_advance(cursor, bytes); |
e766d7b5 | 1123 | break; |
dd236fcb AE |
1124 | #ifdef CONFIG_BLOCK |
1125 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1126 | new_piece = ceph_msg_data_bio_advance(cursor, bytes); |
6aaa4511 | 1127 | break; |
dd236fcb | 1128 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1129 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1130 | default: |
1131 | BUG(); | |
1132 | break; | |
1133 | } | |
ca8b3a69 | 1134 | cursor->total_resid -= bytes; |
dd236fcb | 1135 | |
ca8b3a69 AE |
1136 | if (!cursor->resid && cursor->total_resid) { |
1137 | WARN_ON(!cursor->last_piece); | |
1138 | BUG_ON(list_is_last(&cursor->data->links, cursor->data_head)); | |
1139 | cursor->data = list_entry_next(cursor->data, links); | |
1140 | __ceph_msg_data_cursor_init(cursor); | |
a51b272e | 1141 | new_piece = true; |
ca8b3a69 | 1142 | } |
a51b272e | 1143 | cursor->need_crc = new_piece; |
ca8b3a69 | 1144 | |
dd236fcb AE |
1145 | return new_piece; |
1146 | } | |
1147 | ||
98fa5dd8 | 1148 | static void prepare_message_data(struct ceph_msg *msg, u32 data_len) |
739c905b | 1149 | { |
739c905b | 1150 | BUG_ON(!msg); |
25aff7c5 | 1151 | BUG_ON(!data_len); |
739c905b | 1152 | |
4c59b4a2 | 1153 | /* Initialize data cursor */ |
fe38a2b6 | 1154 | |
8ae4f4f5 | 1155 | ceph_msg_data_cursor_init(msg, (size_t)data_len); |
739c905b AE |
1156 | } |
1157 | ||
31b8006e SW |
1158 | /* |
1159 | * Prepare footer for currently outgoing message, and finish things | |
1160 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
1161 | */ | |
859eb799 | 1162 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
1163 | { |
1164 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 1165 | int v = con->out_kvec_left; |
31b8006e | 1166 | |
fd154f3c AE |
1167 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
1168 | ||
31b8006e SW |
1169 | dout("prepare_write_message_footer %p\n", con); |
1170 | con->out_kvec_is_msg = true; | |
1171 | con->out_kvec[v].iov_base = &m->footer; | |
1172 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
1173 | con->out_kvec_bytes += sizeof(m->footer); | |
1174 | con->out_kvec_left++; | |
1175 | con->out_more = m->more_to_follow; | |
c86a2930 | 1176 | con->out_msg_done = true; |
31b8006e SW |
1177 | } |
1178 | ||
1179 | /* | |
1180 | * Prepare headers for the next outgoing message. | |
1181 | */ | |
1182 | static void prepare_write_message(struct ceph_connection *con) | |
1183 | { | |
1184 | struct ceph_msg *m; | |
a9a0c51a | 1185 | u32 crc; |
31b8006e | 1186 | |
e2200423 | 1187 | con_out_kvec_reset(con); |
31b8006e | 1188 | con->out_kvec_is_msg = true; |
c86a2930 | 1189 | con->out_msg_done = false; |
31b8006e SW |
1190 | |
1191 | /* Sneak an ack in there first? If we can get it into the same | |
1192 | * TCP packet that's a good thing. */ | |
1193 | if (con->in_seq > con->in_seq_acked) { | |
1194 | con->in_seq_acked = con->in_seq; | |
e2200423 | 1195 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 1196 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1197 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 1198 | &con->out_temp_ack); |
31b8006e SW |
1199 | } |
1200 | ||
38941f80 | 1201 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 1202 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 1203 | con->out_msg = m; |
38941f80 | 1204 | BUG_ON(m->con != con); |
4cf9d544 SW |
1205 | |
1206 | /* put message on sent list */ | |
1207 | ceph_msg_get(m); | |
1208 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 1209 | |
e84346b7 SW |
1210 | /* |
1211 | * only assign outgoing seq # if we haven't sent this message | |
1212 | * yet. if it is requeued, resend with it's original seq. | |
1213 | */ | |
1214 | if (m->needs_out_seq) { | |
1215 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
1216 | m->needs_out_seq = false; | |
1217 | } | |
98fa5dd8 | 1218 | WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len)); |
31b8006e | 1219 | |
98fa5dd8 | 1220 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n", |
31b8006e SW |
1221 | m, con->out_seq, le16_to_cpu(m->hdr.type), |
1222 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
98fa5dd8 | 1223 | m->data_length); |
31b8006e SW |
1224 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); |
1225 | ||
1226 | /* tag + hdr + front + middle */ | |
e2200423 AE |
1227 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
1228 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
1229 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 1230 | |
31b8006e | 1231 | if (m->middle) |
e2200423 | 1232 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 1233 | m->middle->vec.iov_base); |
31b8006e SW |
1234 | |
1235 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
1236 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
1237 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 1238 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
1239 | |
1240 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
1241 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
1242 | if (m->middle) { | |
1243 | crc = crc32c(0, m->middle->vec.iov_base, | |
1244 | m->middle->vec.iov_len); | |
1245 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
1246 | } else | |
31b8006e | 1247 | con->out_msg->footer.middle_crc = 0; |
739c905b | 1248 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
1249 | le32_to_cpu(con->out_msg->footer.front_crc), |
1250 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
1251 | ||
1252 | /* is there a data payload? */ | |
739c905b | 1253 | con->out_msg->footer.data_crc = 0; |
98fa5dd8 AE |
1254 | if (m->data_length) { |
1255 | prepare_message_data(con->out_msg, m->data_length); | |
78625051 AE |
1256 | con->out_more = 1; /* data + footer will follow */ |
1257 | } else { | |
31b8006e | 1258 | /* no, queue up footer too and be done */ |
859eb799 | 1259 | prepare_write_message_footer(con); |
78625051 | 1260 | } |
31b8006e | 1261 | |
c9ffc77a | 1262 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1263 | } |
1264 | ||
1265 | /* | |
1266 | * Prepare an ack. | |
1267 | */ | |
1268 | static void prepare_write_ack(struct ceph_connection *con) | |
1269 | { | |
1270 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
1271 | con->in_seq_acked, con->in_seq); | |
1272 | con->in_seq_acked = con->in_seq; | |
1273 | ||
e2200423 | 1274 | con_out_kvec_reset(con); |
859eb799 | 1275 | |
e2200423 | 1276 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 1277 | |
31b8006e | 1278 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1279 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
1280 | &con->out_temp_ack); |
1281 | ||
31b8006e | 1282 | con->out_more = 1; /* more will follow.. eventually.. */ |
c9ffc77a | 1283 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1284 | } |
1285 | ||
3a23083b SW |
1286 | /* |
1287 | * Prepare to share the seq during handshake | |
1288 | */ | |
1289 | static void prepare_write_seq(struct ceph_connection *con) | |
1290 | { | |
1291 | dout("prepare_write_seq %p %llu -> %llu\n", con, | |
1292 | con->in_seq_acked, con->in_seq); | |
1293 | con->in_seq_acked = con->in_seq; | |
1294 | ||
1295 | con_out_kvec_reset(con); | |
1296 | ||
1297 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); | |
1298 | con_out_kvec_add(con, sizeof (con->out_temp_ack), | |
1299 | &con->out_temp_ack); | |
1300 | ||
1301 | con_flag_set(con, CON_FLAG_WRITE_PENDING); | |
1302 | } | |
1303 | ||
31b8006e SW |
1304 | /* |
1305 | * Prepare to write keepalive byte. | |
1306 | */ | |
1307 | static void prepare_write_keepalive(struct ceph_connection *con) | |
1308 | { | |
1309 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
1310 | con_out_kvec_reset(con); |
1311 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
c9ffc77a | 1312 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1313 | } |
1314 | ||
1315 | /* | |
1316 | * Connection negotiation. | |
1317 | */ | |
1318 | ||
dac1e716 AE |
1319 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
1320 | int *auth_proto) | |
4e7a5dcd | 1321 | { |
a3530df3 | 1322 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
1323 | |
1324 | if (!con->ops->get_authorizer) { | |
1325 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
1326 | con->out_connect.authorizer_len = 0; | |
729796be | 1327 | return NULL; |
b1c6b980 AE |
1328 | } |
1329 | ||
1330 | /* Can't hold the mutex while getting authorizer */ | |
ec302645 | 1331 | mutex_unlock(&con->mutex); |
dac1e716 | 1332 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
ec302645 | 1333 | mutex_lock(&con->mutex); |
4e7a5dcd | 1334 | |
a3530df3 | 1335 | if (IS_ERR(auth)) |
729796be | 1336 | return auth; |
8dacc7da | 1337 | if (con->state != CON_STATE_NEGOTIATING) |
729796be | 1338 | return ERR_PTR(-EAGAIN); |
0da5d703 | 1339 | |
8f43fb53 AE |
1340 | con->auth_reply_buf = auth->authorizer_reply_buf; |
1341 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
729796be | 1342 | return auth; |
4e7a5dcd SW |
1343 | } |
1344 | ||
31b8006e SW |
1345 | /* |
1346 | * We connected to a peer and are saying hello. | |
1347 | */ | |
e825a66d | 1348 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 1349 | { |
e2200423 AE |
1350 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
1351 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 1352 | &con->msgr->my_enc_addr); |
eed0ef2c | 1353 | |
eed0ef2c | 1354 | con->out_more = 0; |
c9ffc77a | 1355 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
eed0ef2c SW |
1356 | } |
1357 | ||
e825a66d | 1358 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 1359 | { |
95c96174 | 1360 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 1361 | int proto; |
dac1e716 | 1362 | int auth_proto; |
729796be | 1363 | struct ceph_auth_handshake *auth; |
31b8006e SW |
1364 | |
1365 | switch (con->peer_name.type) { | |
1366 | case CEPH_ENTITY_TYPE_MON: | |
1367 | proto = CEPH_MONC_PROTOCOL; | |
1368 | break; | |
1369 | case CEPH_ENTITY_TYPE_OSD: | |
1370 | proto = CEPH_OSDC_PROTOCOL; | |
1371 | break; | |
1372 | case CEPH_ENTITY_TYPE_MDS: | |
1373 | proto = CEPH_MDSC_PROTOCOL; | |
1374 | break; | |
1375 | default: | |
1376 | BUG(); | |
1377 | } | |
1378 | ||
1379 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
1380 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 1381 | |
e825a66d | 1382 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
1383 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
1384 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
1385 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
1386 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
1387 | con->out_connect.flags = 0; | |
31b8006e | 1388 | |
dac1e716 AE |
1389 | auth_proto = CEPH_AUTH_UNKNOWN; |
1390 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
1391 | if (IS_ERR(auth)) |
1392 | return PTR_ERR(auth); | |
3da54776 | 1393 | |
dac1e716 | 1394 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
1395 | con->out_connect.authorizer_len = auth ? |
1396 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
1397 | ||
e2200423 | 1398 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
1399 | &con->out_connect); |
1400 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 1401 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 1402 | auth->authorizer_buf); |
859eb799 | 1403 | |
31b8006e | 1404 | con->out_more = 0; |
c9ffc77a | 1405 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
4e7a5dcd | 1406 | |
e10c758e | 1407 | return 0; |
31b8006e SW |
1408 | } |
1409 | ||
31b8006e SW |
1410 | /* |
1411 | * write as much of pending kvecs to the socket as we can. | |
1412 | * 1 -> done | |
1413 | * 0 -> socket full, but more to do | |
1414 | * <0 -> error | |
1415 | */ | |
1416 | static int write_partial_kvec(struct ceph_connection *con) | |
1417 | { | |
1418 | int ret; | |
1419 | ||
1420 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
1421 | while (con->out_kvec_bytes > 0) { | |
1422 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
1423 | con->out_kvec_left, con->out_kvec_bytes, | |
1424 | con->out_more); | |
1425 | if (ret <= 0) | |
1426 | goto out; | |
1427 | con->out_kvec_bytes -= ret; | |
1428 | if (con->out_kvec_bytes == 0) | |
1429 | break; /* done */ | |
f42299e6 AE |
1430 | |
1431 | /* account for full iov entries consumed */ | |
1432 | while (ret >= con->out_kvec_cur->iov_len) { | |
1433 | BUG_ON(!con->out_kvec_left); | |
1434 | ret -= con->out_kvec_cur->iov_len; | |
1435 | con->out_kvec_cur++; | |
1436 | con->out_kvec_left--; | |
1437 | } | |
1438 | /* and for a partially-consumed entry */ | |
1439 | if (ret) { | |
1440 | con->out_kvec_cur->iov_len -= ret; | |
1441 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
1442 | } |
1443 | } | |
1444 | con->out_kvec_left = 0; | |
1445 | con->out_kvec_is_msg = false; | |
1446 | ret = 1; | |
1447 | out: | |
1448 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
1449 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
1450 | return ret; /* done! */ | |
1451 | } | |
1452 | ||
35b62808 AE |
1453 | static u32 ceph_crc32c_page(u32 crc, struct page *page, |
1454 | unsigned int page_offset, | |
1455 | unsigned int length) | |
1456 | { | |
1457 | char *kaddr; | |
1458 | ||
1459 | kaddr = kmap(page); | |
1460 | BUG_ON(kaddr == NULL); | |
1461 | crc = crc32c(crc, kaddr + page_offset, length); | |
1462 | kunmap(page); | |
1463 | ||
1464 | return crc; | |
1465 | } | |
31b8006e SW |
1466 | /* |
1467 | * Write as much message data payload as we can. If we finish, queue | |
1468 | * up the footer. | |
1469 | * 1 -> done, footer is now queued in out_kvec[]. | |
1470 | * 0 -> socket full, but more to do | |
1471 | * <0 -> error | |
1472 | */ | |
34d2d200 | 1473 | static int write_partial_message_data(struct ceph_connection *con) |
31b8006e SW |
1474 | { |
1475 | struct ceph_msg *msg = con->out_msg; | |
8ae4f4f5 | 1476 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
37675b0f | 1477 | bool do_datacrc = !con->msgr->nocrc; |
f5db90bc | 1478 | u32 crc; |
31b8006e | 1479 | |
859a35d5 | 1480 | dout("%s %p msg %p\n", __func__, con, msg); |
31b8006e | 1481 | |
5240d9f9 | 1482 | if (list_empty(&msg->data)) |
4c59b4a2 AE |
1483 | return -EINVAL; |
1484 | ||
5821bd8c AE |
1485 | /* |
1486 | * Iterate through each page that contains data to be | |
1487 | * written, and send as much as possible for each. | |
1488 | * | |
1489 | * If we are calculating the data crc (the default), we will | |
1490 | * need to map the page. If we have no pages, they have | |
1491 | * been revoked, so use the zero page. | |
1492 | */ | |
f5db90bc | 1493 | crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0; |
643c68a4 | 1494 | while (cursor->resid) { |
8a166d05 | 1495 | struct page *page; |
e387d525 AE |
1496 | size_t page_offset; |
1497 | size_t length; | |
8a166d05 | 1498 | bool last_piece; |
8ea299bc | 1499 | bool need_crc; |
f5db90bc | 1500 | int ret; |
68b4476b | 1501 | |
8ae4f4f5 | 1502 | page = ceph_msg_data_next(&msg->cursor, &page_offset, &length, |
4c59b4a2 | 1503 | &last_piece); |
e387d525 | 1504 | ret = ceph_tcp_sendpage(con->sock, page, page_offset, |
fe38a2b6 | 1505 | length, last_piece); |
f5db90bc AE |
1506 | if (ret <= 0) { |
1507 | if (do_datacrc) | |
1508 | msg->footer.data_crc = cpu_to_le32(crc); | |
31b8006e | 1509 | |
f5db90bc AE |
1510 | return ret; |
1511 | } | |
143334ff AE |
1512 | if (do_datacrc && cursor->need_crc) |
1513 | crc = ceph_crc32c_page(crc, page, page_offset, length); | |
8ae4f4f5 | 1514 | need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret); |
31b8006e SW |
1515 | } |
1516 | ||
34d2d200 | 1517 | dout("%s %p msg %p done\n", __func__, con, msg); |
31b8006e SW |
1518 | |
1519 | /* prepare and queue up footer, too */ | |
f5db90bc AE |
1520 | if (do_datacrc) |
1521 | msg->footer.data_crc = cpu_to_le32(crc); | |
1522 | else | |
84ca8fc8 | 1523 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1524 | con_out_kvec_reset(con); |
859eb799 | 1525 | prepare_write_message_footer(con); |
f5db90bc AE |
1526 | |
1527 | return 1; /* must return > 0 to indicate success */ | |
31b8006e SW |
1528 | } |
1529 | ||
1530 | /* | |
1531 | * write some zeros | |
1532 | */ | |
1533 | static int write_partial_skip(struct ceph_connection *con) | |
1534 | { | |
1535 | int ret; | |
1536 | ||
1537 | while (con->out_skip > 0) { | |
31739139 | 1538 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1539 | |
e1dcb128 | 1540 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true); |
31b8006e SW |
1541 | if (ret <= 0) |
1542 | goto out; | |
1543 | con->out_skip -= ret; | |
1544 | } | |
1545 | ret = 1; | |
1546 | out: | |
1547 | return ret; | |
1548 | } | |
1549 | ||
1550 | /* | |
1551 | * Prepare to read connection handshake, or an ack. | |
1552 | */ | |
eed0ef2c SW |
1553 | static void prepare_read_banner(struct ceph_connection *con) |
1554 | { | |
1555 | dout("prepare_read_banner %p\n", con); | |
1556 | con->in_base_pos = 0; | |
1557 | } | |
1558 | ||
31b8006e SW |
1559 | static void prepare_read_connect(struct ceph_connection *con) |
1560 | { | |
1561 | dout("prepare_read_connect %p\n", con); | |
1562 | con->in_base_pos = 0; | |
1563 | } | |
1564 | ||
1565 | static void prepare_read_ack(struct ceph_connection *con) | |
1566 | { | |
1567 | dout("prepare_read_ack %p\n", con); | |
1568 | con->in_base_pos = 0; | |
1569 | } | |
1570 | ||
3a23083b SW |
1571 | static void prepare_read_seq(struct ceph_connection *con) |
1572 | { | |
1573 | dout("prepare_read_seq %p\n", con); | |
1574 | con->in_base_pos = 0; | |
1575 | con->in_tag = CEPH_MSGR_TAG_SEQ; | |
1576 | } | |
1577 | ||
31b8006e SW |
1578 | static void prepare_read_tag(struct ceph_connection *con) |
1579 | { | |
1580 | dout("prepare_read_tag %p\n", con); | |
1581 | con->in_base_pos = 0; | |
1582 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1583 | } | |
1584 | ||
1585 | /* | |
1586 | * Prepare to read a message. | |
1587 | */ | |
1588 | static int prepare_read_message(struct ceph_connection *con) | |
1589 | { | |
1590 | dout("prepare_read_message %p\n", con); | |
1591 | BUG_ON(con->in_msg != NULL); | |
1592 | con->in_base_pos = 0; | |
1593 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1594 | return 0; | |
1595 | } | |
1596 | ||
1597 | ||
1598 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1599 | int end, int size, void *object) |
31b8006e | 1600 | { |
e6cee71f AE |
1601 | while (con->in_base_pos < end) { |
1602 | int left = end - con->in_base_pos; | |
31b8006e SW |
1603 | int have = size - left; |
1604 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1605 | if (ret <= 0) | |
1606 | return ret; | |
1607 | con->in_base_pos += ret; | |
1608 | } | |
1609 | return 1; | |
1610 | } | |
1611 | ||
1612 | ||
1613 | /* | |
1614 | * Read all or part of the connect-side handshake on a new connection | |
1615 | */ | |
eed0ef2c | 1616 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1617 | { |
fd51653f AE |
1618 | int size; |
1619 | int end; | |
1620 | int ret; | |
31b8006e | 1621 | |
eed0ef2c | 1622 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1623 | |
1624 | /* peer's banner */ | |
fd51653f AE |
1625 | size = strlen(CEPH_BANNER); |
1626 | end = size; | |
1627 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1628 | if (ret <= 0) |
1629 | goto out; | |
fd51653f AE |
1630 | |
1631 | size = sizeof (con->actual_peer_addr); | |
1632 | end += size; | |
1633 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1634 | if (ret <= 0) |
1635 | goto out; | |
fd51653f AE |
1636 | |
1637 | size = sizeof (con->peer_addr_for_me); | |
1638 | end += size; | |
1639 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1640 | if (ret <= 0) |
1641 | goto out; | |
fd51653f | 1642 | |
eed0ef2c SW |
1643 | out: |
1644 | return ret; | |
1645 | } | |
1646 | ||
1647 | static int read_partial_connect(struct ceph_connection *con) | |
1648 | { | |
fd51653f AE |
1649 | int size; |
1650 | int end; | |
1651 | int ret; | |
eed0ef2c SW |
1652 | |
1653 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1654 | ||
fd51653f AE |
1655 | size = sizeof (con->in_reply); |
1656 | end = size; | |
1657 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1658 | if (ret <= 0) |
1659 | goto out; | |
fd51653f AE |
1660 | |
1661 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1662 | end += size; | |
1663 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1664 | if (ret <= 0) |
1665 | goto out; | |
31b8006e | 1666 | |
4e7a5dcd SW |
1667 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1668 | con, (int)con->in_reply.tag, | |
1669 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1670 | le32_to_cpu(con->in_reply.global_seq)); |
1671 | out: | |
1672 | return ret; | |
eed0ef2c | 1673 | |
31b8006e SW |
1674 | } |
1675 | ||
1676 | /* | |
1677 | * Verify the hello banner looks okay. | |
1678 | */ | |
1679 | static int verify_hello(struct ceph_connection *con) | |
1680 | { | |
1681 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1682 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1683 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1684 | con->error_msg = "protocol error, bad banner"; |
1685 | return -1; | |
1686 | } | |
1687 | return 0; | |
1688 | } | |
1689 | ||
1690 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1691 | { | |
1692 | switch (ss->ss_family) { | |
1693 | case AF_INET: | |
1694 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1695 | case AF_INET6: | |
1696 | return | |
1697 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1698 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1699 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1700 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1701 | } | |
1702 | return false; | |
1703 | } | |
1704 | ||
1705 | static int addr_port(struct sockaddr_storage *ss) | |
1706 | { | |
1707 | switch (ss->ss_family) { | |
1708 | case AF_INET: | |
f28bcfbe | 1709 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1710 | case AF_INET6: |
f28bcfbe | 1711 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1712 | } |
1713 | return 0; | |
1714 | } | |
1715 | ||
1716 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1717 | { | |
1718 | switch (ss->ss_family) { | |
1719 | case AF_INET: | |
1720 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1721 | break; |
31b8006e SW |
1722 | case AF_INET6: |
1723 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1724 | break; |
31b8006e SW |
1725 | } |
1726 | } | |
1727 | ||
ee3b56f2 NW |
1728 | /* |
1729 | * Unlike other *_pton function semantics, zero indicates success. | |
1730 | */ | |
1731 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1732 | char delim, const char **ipend) | |
1733 | { | |
99f0f3b2 AE |
1734 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1735 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1736 | |
1737 | memset(ss, 0, sizeof(*ss)); | |
1738 | ||
1739 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1740 | ss->ss_family = AF_INET; | |
1741 | return 0; | |
1742 | } | |
1743 | ||
1744 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1745 | ss->ss_family = AF_INET6; | |
1746 | return 0; | |
1747 | } | |
1748 | ||
1749 | return -EINVAL; | |
1750 | } | |
1751 | ||
1752 | /* | |
1753 | * Extract hostname string and resolve using kernel DNS facility. | |
1754 | */ | |
1755 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1756 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1757 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1758 | { | |
1759 | const char *end, *delim_p; | |
1760 | char *colon_p, *ip_addr = NULL; | |
1761 | int ip_len, ret; | |
1762 | ||
1763 | /* | |
1764 | * The end of the hostname occurs immediately preceding the delimiter or | |
1765 | * the port marker (':') where the delimiter takes precedence. | |
1766 | */ | |
1767 | delim_p = memchr(name, delim, namelen); | |
1768 | colon_p = memchr(name, ':', namelen); | |
1769 | ||
1770 | if (delim_p && colon_p) | |
1771 | end = delim_p < colon_p ? delim_p : colon_p; | |
1772 | else if (!delim_p && colon_p) | |
1773 | end = colon_p; | |
1774 | else { | |
1775 | end = delim_p; | |
1776 | if (!end) /* case: hostname:/ */ | |
1777 | end = name + namelen; | |
1778 | } | |
1779 | ||
1780 | if (end <= name) | |
1781 | return -EINVAL; | |
1782 | ||
1783 | /* do dns_resolve upcall */ | |
1784 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1785 | if (ip_len > 0) | |
1786 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1787 | else | |
1788 | ret = -ESRCH; | |
1789 | ||
1790 | kfree(ip_addr); | |
1791 | ||
1792 | *ipend = end; | |
1793 | ||
1794 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1795 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1796 | ||
1797 | return ret; | |
1798 | } | |
1799 | #else | |
1800 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1801 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1802 | { | |
1803 | return -EINVAL; | |
1804 | } | |
1805 | #endif | |
1806 | ||
1807 | /* | |
1808 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1809 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1810 | */ | |
1811 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1812 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1813 | { | |
1814 | int ret; | |
1815 | ||
1816 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1817 | if (ret) | |
1818 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1819 | ||
1820 | return ret; | |
1821 | } | |
1822 | ||
31b8006e SW |
1823 | /* |
1824 | * Parse an ip[:port] list into an addr array. Use the default | |
1825 | * monitor port if a port isn't specified. | |
1826 | */ | |
1827 | int ceph_parse_ips(const char *c, const char *end, | |
1828 | struct ceph_entity_addr *addr, | |
1829 | int max_count, int *count) | |
1830 | { | |
ee3b56f2 | 1831 | int i, ret = -EINVAL; |
31b8006e SW |
1832 | const char *p = c; |
1833 | ||
1834 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1835 | for (i = 0; i < max_count; i++) { | |
1836 | const char *ipend; | |
1837 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1838 | int port; |
39139f64 SW |
1839 | char delim = ','; |
1840 | ||
1841 | if (*p == '[') { | |
1842 | delim = ']'; | |
1843 | p++; | |
1844 | } | |
31b8006e | 1845 | |
ee3b56f2 NW |
1846 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1847 | if (ret) | |
31b8006e | 1848 | goto bad; |
ee3b56f2 NW |
1849 | ret = -EINVAL; |
1850 | ||
31b8006e SW |
1851 | p = ipend; |
1852 | ||
39139f64 SW |
1853 | if (delim == ']') { |
1854 | if (*p != ']') { | |
1855 | dout("missing matching ']'\n"); | |
1856 | goto bad; | |
1857 | } | |
1858 | p++; | |
1859 | } | |
1860 | ||
31b8006e SW |
1861 | /* port? */ |
1862 | if (p < end && *p == ':') { | |
1863 | port = 0; | |
1864 | p++; | |
1865 | while (p < end && *p >= '0' && *p <= '9') { | |
1866 | port = (port * 10) + (*p - '0'); | |
1867 | p++; | |
1868 | } | |
1869 | if (port > 65535 || port == 0) | |
1870 | goto bad; | |
1871 | } else { | |
1872 | port = CEPH_MON_PORT; | |
1873 | } | |
1874 | ||
1875 | addr_set_port(ss, port); | |
1876 | ||
3d14c5d2 | 1877 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1878 | |
1879 | if (p == end) | |
1880 | break; | |
1881 | if (*p != ',') | |
1882 | goto bad; | |
1883 | p++; | |
1884 | } | |
1885 | ||
1886 | if (p != end) | |
1887 | goto bad; | |
1888 | ||
1889 | if (count) | |
1890 | *count = i + 1; | |
1891 | return 0; | |
1892 | ||
1893 | bad: | |
39139f64 | 1894 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1895 | return ret; |
31b8006e | 1896 | } |
3d14c5d2 | 1897 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1898 | |
eed0ef2c | 1899 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1900 | { |
eed0ef2c | 1901 | dout("process_banner on %p\n", con); |
31b8006e SW |
1902 | |
1903 | if (verify_hello(con) < 0) | |
1904 | return -1; | |
1905 | ||
63f2d211 SW |
1906 | ceph_decode_addr(&con->actual_peer_addr); |
1907 | ceph_decode_addr(&con->peer_addr_for_me); | |
1908 | ||
31b8006e SW |
1909 | /* |
1910 | * Make sure the other end is who we wanted. note that the other | |
1911 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1912 | * them the benefit of the doubt. | |
1913 | */ | |
103e2d3a SW |
1914 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1915 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1916 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1917 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1918 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1919 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1920 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1921 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1922 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1923 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1924 | return -1; |
1925 | } | |
1926 | ||
1927 | /* | |
1928 | * did we learn our address? | |
1929 | */ | |
1930 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1931 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1932 | ||
1933 | memcpy(&con->msgr->inst.addr.in_addr, | |
1934 | &con->peer_addr_for_me.in_addr, | |
1935 | sizeof(con->peer_addr_for_me.in_addr)); | |
1936 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1937 | encode_my_addr(con->msgr); |
eed0ef2c | 1938 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1939 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1940 | } |
1941 | ||
eed0ef2c SW |
1942 | return 0; |
1943 | } | |
1944 | ||
1945 | static int process_connect(struct ceph_connection *con) | |
1946 | { | |
3d14c5d2 YS |
1947 | u64 sup_feat = con->msgr->supported_features; |
1948 | u64 req_feat = con->msgr->required_features; | |
2b3e0c90 ID |
1949 | u64 server_feat = ceph_sanitize_features( |
1950 | le64_to_cpu(con->in_reply.features)); | |
0da5d703 | 1951 | int ret; |
04a419f9 | 1952 | |
eed0ef2c SW |
1953 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1954 | ||
31b8006e | 1955 | switch (con->in_reply.tag) { |
04a419f9 SW |
1956 | case CEPH_MSGR_TAG_FEATURES: |
1957 | pr_err("%s%lld %s feature set mismatch," | |
1958 | " my %llx < server's %llx, missing %llx\n", | |
1959 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1960 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1961 | sup_feat, server_feat, server_feat & ~sup_feat); |
1962 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 1963 | reset_connection(con); |
04a419f9 SW |
1964 | return -1; |
1965 | ||
31b8006e | 1966 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1967 | pr_err("%s%lld %s protocol version mismatch," |
1968 | " my %d != server's %d\n", | |
1969 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1970 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1971 | le32_to_cpu(con->out_connect.protocol_version), |
1972 | le32_to_cpu(con->in_reply.protocol_version)); | |
1973 | con->error_msg = "protocol version mismatch"; | |
0fa6ebc6 | 1974 | reset_connection(con); |
31b8006e SW |
1975 | return -1; |
1976 | ||
4e7a5dcd SW |
1977 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1978 | con->auth_retry++; | |
1979 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1980 | con->auth_retry); | |
1981 | if (con->auth_retry == 2) { | |
1982 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1983 | return -1; |
1984 | } | |
6d4221b5 | 1985 | con_out_kvec_reset(con); |
e825a66d | 1986 | ret = prepare_write_connect(con); |
0da5d703 SW |
1987 | if (ret < 0) |
1988 | return ret; | |
63733a0f | 1989 | prepare_read_connect(con); |
4e7a5dcd | 1990 | break; |
31b8006e SW |
1991 | |
1992 | case CEPH_MSGR_TAG_RESETSESSION: | |
1993 | /* | |
1994 | * If we connected with a large connect_seq but the peer | |
1995 | * has no record of a session with us (no connection, or | |
1996 | * connect_seq == 0), they will send RESETSESION to indicate | |
1997 | * that they must have reset their session, and may have | |
1998 | * dropped messages. | |
1999 | */ | |
2000 | dout("process_connect got RESET peer seq %u\n", | |
5bdca4e0 | 2001 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
2002 | pr_err("%s%lld %s connection reset\n", |
2003 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2004 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2005 | reset_connection(con); |
6d4221b5 | 2006 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2007 | ret = prepare_write_connect(con); |
2008 | if (ret < 0) | |
2009 | return ret; | |
31b8006e SW |
2010 | prepare_read_connect(con); |
2011 | ||
2012 | /* Tell ceph about it. */ | |
ec302645 | 2013 | mutex_unlock(&con->mutex); |
31b8006e SW |
2014 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
2015 | if (con->ops->peer_reset) | |
2016 | con->ops->peer_reset(con); | |
ec302645 | 2017 | mutex_lock(&con->mutex); |
8dacc7da | 2018 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 2019 | return -EAGAIN; |
31b8006e SW |
2020 | break; |
2021 | ||
2022 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
2023 | /* | |
2024 | * If we sent a smaller connect_seq than the peer has, try | |
2025 | * again with a larger value. | |
2026 | */ | |
5bdca4e0 | 2027 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 2028 | le32_to_cpu(con->out_connect.connect_seq), |
5bdca4e0 SW |
2029 | le32_to_cpu(con->in_reply.connect_seq)); |
2030 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
6d4221b5 | 2031 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2032 | ret = prepare_write_connect(con); |
2033 | if (ret < 0) | |
2034 | return ret; | |
31b8006e SW |
2035 | prepare_read_connect(con); |
2036 | break; | |
2037 | ||
2038 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
2039 | /* | |
2040 | * If we sent a smaller global_seq than the peer has, try | |
2041 | * again with a larger value. | |
2042 | */ | |
eed0ef2c | 2043 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 2044 | con->peer_global_seq, |
5bdca4e0 | 2045 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 2046 | get_global_seq(con->msgr, |
5bdca4e0 | 2047 | le32_to_cpu(con->in_reply.global_seq)); |
6d4221b5 | 2048 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2049 | ret = prepare_write_connect(con); |
2050 | if (ret < 0) | |
2051 | return ret; | |
31b8006e SW |
2052 | prepare_read_connect(con); |
2053 | break; | |
2054 | ||
3a23083b | 2055 | case CEPH_MSGR_TAG_SEQ: |
31b8006e | 2056 | case CEPH_MSGR_TAG_READY: |
04a419f9 SW |
2057 | if (req_feat & ~server_feat) { |
2058 | pr_err("%s%lld %s protocol feature mismatch," | |
2059 | " my required %llx > server's %llx, need %llx\n", | |
2060 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2061 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
2062 | req_feat, server_feat, req_feat & ~server_feat); |
2063 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 2064 | reset_connection(con); |
04a419f9 SW |
2065 | return -1; |
2066 | } | |
8dacc7da | 2067 | |
122070a2 | 2068 | WARN_ON(con->state != CON_STATE_NEGOTIATING); |
8dacc7da | 2069 | con->state = CON_STATE_OPEN; |
20e55c4c | 2070 | con->auth_retry = 0; /* we authenticated; clear flag */ |
31b8006e SW |
2071 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
2072 | con->connect_seq++; | |
aba558e2 | 2073 | con->peer_features = server_feat; |
31b8006e SW |
2074 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
2075 | con->peer_global_seq, | |
2076 | le32_to_cpu(con->in_reply.connect_seq), | |
2077 | con->connect_seq); | |
2078 | WARN_ON(con->connect_seq != | |
2079 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
2080 | |
2081 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
c9ffc77a | 2082 | con_flag_set(con, CON_FLAG_LOSSYTX); |
92ac41d0 | 2083 | |
85effe18 | 2084 | con->delay = 0; /* reset backoff memory */ |
92ac41d0 | 2085 | |
3a23083b SW |
2086 | if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) { |
2087 | prepare_write_seq(con); | |
2088 | prepare_read_seq(con); | |
2089 | } else { | |
2090 | prepare_read_tag(con); | |
2091 | } | |
31b8006e SW |
2092 | break; |
2093 | ||
2094 | case CEPH_MSGR_TAG_WAIT: | |
2095 | /* | |
2096 | * If there is a connection race (we are opening | |
2097 | * connections to each other), one of us may just have | |
2098 | * to WAIT. This shouldn't happen if we are the | |
2099 | * client. | |
2100 | */ | |
04177882 SW |
2101 | pr_err("process_connect got WAIT as client\n"); |
2102 | con->error_msg = "protocol error, got WAIT as client"; | |
2103 | return -1; | |
31b8006e SW |
2104 | |
2105 | default: | |
2106 | pr_err("connect protocol error, will retry\n"); | |
2107 | con->error_msg = "protocol error, garbage tag during connect"; | |
2108 | return -1; | |
2109 | } | |
2110 | return 0; | |
2111 | } | |
2112 | ||
2113 | ||
2114 | /* | |
2115 | * read (part of) an ack | |
2116 | */ | |
2117 | static int read_partial_ack(struct ceph_connection *con) | |
2118 | { | |
fd51653f AE |
2119 | int size = sizeof (con->in_temp_ack); |
2120 | int end = size; | |
31b8006e | 2121 | |
fd51653f | 2122 | return read_partial(con, end, size, &con->in_temp_ack); |
31b8006e SW |
2123 | } |
2124 | ||
31b8006e SW |
2125 | /* |
2126 | * We can finally discard anything that's been acked. | |
2127 | */ | |
2128 | static void process_ack(struct ceph_connection *con) | |
2129 | { | |
2130 | struct ceph_msg *m; | |
2131 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
2132 | u64 seq; | |
2133 | ||
31b8006e SW |
2134 | while (!list_empty(&con->out_sent)) { |
2135 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
2136 | list_head); | |
2137 | seq = le64_to_cpu(m->hdr.seq); | |
2138 | if (seq > ack) | |
2139 | break; | |
2140 | dout("got ack for seq %llu type %d at %p\n", seq, | |
2141 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 2142 | m->ack_stamp = jiffies; |
31b8006e SW |
2143 | ceph_msg_remove(m); |
2144 | } | |
31b8006e SW |
2145 | prepare_read_tag(con); |
2146 | } | |
2147 | ||
2148 | ||
2450418c | 2149 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
2150 | struct kvec *section, |
2151 | unsigned int sec_len, u32 *crc) | |
2450418c | 2152 | { |
68b4476b | 2153 | int ret, left; |
2450418c YS |
2154 | |
2155 | BUG_ON(!section); | |
2156 | ||
2157 | while (section->iov_len < sec_len) { | |
2158 | BUG_ON(section->iov_base == NULL); | |
2159 | left = sec_len - section->iov_len; | |
2160 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
2161 | section->iov_len, left); | |
2162 | if (ret <= 0) | |
2163 | return ret; | |
2164 | section->iov_len += ret; | |
2450418c | 2165 | } |
fe3ad593 AE |
2166 | if (section->iov_len == sec_len) |
2167 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 2168 | |
2450418c YS |
2169 | return 1; |
2170 | } | |
31b8006e | 2171 | |
34d2d200 AE |
2172 | static int read_partial_msg_data(struct ceph_connection *con) |
2173 | { | |
2174 | struct ceph_msg *msg = con->in_msg; | |
8ae4f4f5 | 2175 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
34d2d200 | 2176 | const bool do_datacrc = !con->msgr->nocrc; |
686be208 AE |
2177 | struct page *page; |
2178 | size_t page_offset; | |
2179 | size_t length; | |
f5db90bc | 2180 | u32 crc = 0; |
34d2d200 AE |
2181 | int ret; |
2182 | ||
2183 | BUG_ON(!msg); | |
5240d9f9 | 2184 | if (list_empty(&msg->data)) |
4c59b4a2 | 2185 | return -EIO; |
34d2d200 | 2186 | |
f5db90bc AE |
2187 | if (do_datacrc) |
2188 | crc = con->in_data_crc; | |
643c68a4 | 2189 | while (cursor->resid) { |
8ae4f4f5 | 2190 | page = ceph_msg_data_next(&msg->cursor, &page_offset, &length, |
4c59b4a2 | 2191 | NULL); |
686be208 | 2192 | ret = ceph_tcp_recvpage(con->sock, page, page_offset, length); |
f5db90bc AE |
2193 | if (ret <= 0) { |
2194 | if (do_datacrc) | |
2195 | con->in_data_crc = crc; | |
2196 | ||
686be208 | 2197 | return ret; |
f5db90bc | 2198 | } |
686be208 AE |
2199 | |
2200 | if (do_datacrc) | |
f5db90bc | 2201 | crc = ceph_crc32c_page(crc, page, page_offset, ret); |
8ae4f4f5 | 2202 | (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret); |
34d2d200 | 2203 | } |
f5db90bc AE |
2204 | if (do_datacrc) |
2205 | con->in_data_crc = crc; | |
34d2d200 AE |
2206 | |
2207 | return 1; /* must return > 0 to indicate success */ | |
2208 | } | |
2209 | ||
31b8006e SW |
2210 | /* |
2211 | * read (part of) a message. | |
2212 | */ | |
686be208 AE |
2213 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
2214 | ||
31b8006e SW |
2215 | static int read_partial_message(struct ceph_connection *con) |
2216 | { | |
2217 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
2218 | int size; |
2219 | int end; | |
31b8006e | 2220 | int ret; |
95c96174 | 2221 | unsigned int front_len, middle_len, data_len; |
37675b0f | 2222 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 2223 | u64 seq; |
fe3ad593 | 2224 | u32 crc; |
31b8006e SW |
2225 | |
2226 | dout("read_partial_message con %p msg %p\n", con, m); | |
2227 | ||
2228 | /* header */ | |
fd51653f AE |
2229 | size = sizeof (con->in_hdr); |
2230 | end = size; | |
2231 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
2232 | if (ret <= 0) |
2233 | return ret; | |
fe3ad593 AE |
2234 | |
2235 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
2236 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
2237 | pr_err("read_partial_message bad hdr " | |
2238 | " crc %u != expected %u\n", | |
2239 | crc, con->in_hdr.crc); | |
2240 | return -EBADMSG; | |
2241 | } | |
2242 | ||
31b8006e SW |
2243 | front_len = le32_to_cpu(con->in_hdr.front_len); |
2244 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
2245 | return -EIO; | |
2246 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
7b11ba37 | 2247 | if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN) |
31b8006e SW |
2248 | return -EIO; |
2249 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
2250 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
2251 | return -EIO; | |
2252 | ||
ae18756b SW |
2253 | /* verify seq# */ |
2254 | seq = le64_to_cpu(con->in_hdr.seq); | |
2255 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 2256 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 2257 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2258 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
2259 | seq, con->in_seq + 1); |
2260 | con->in_base_pos = -front_len - middle_len - data_len - | |
2261 | sizeof(m->footer); | |
2262 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
2263 | return 0; |
2264 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
2265 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
2266 | seq, con->in_seq + 1); | |
2267 | con->error_msg = "bad message sequence # for incoming message"; | |
2268 | return -EBADMSG; | |
2269 | } | |
2270 | ||
31b8006e SW |
2271 | /* allocate message? */ |
2272 | if (!con->in_msg) { | |
4740a623 SW |
2273 | int skip = 0; |
2274 | ||
31b8006e | 2275 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
6ebc8b32 | 2276 | front_len, data_len); |
4740a623 SW |
2277 | ret = ceph_con_in_msg_alloc(con, &skip); |
2278 | if (ret < 0) | |
2279 | return ret; | |
f759ebb9 AE |
2280 | |
2281 | BUG_ON(!con->in_msg ^ skip); | |
2282 | if (con->in_msg && data_len > con->in_msg->data_length) { | |
2283 | pr_warning("%s skipping long message (%u > %zd)\n", | |
2284 | __func__, data_len, con->in_msg->data_length); | |
2285 | ceph_msg_put(con->in_msg); | |
2286 | con->in_msg = NULL; | |
2287 | skip = 1; | |
2288 | } | |
2450418c | 2289 | if (skip) { |
31b8006e | 2290 | /* skip this message */ |
a79832f2 | 2291 | dout("alloc_msg said skip message\n"); |
31b8006e SW |
2292 | con->in_base_pos = -front_len - middle_len - data_len - |
2293 | sizeof(m->footer); | |
2294 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2295 | con->in_seq++; |
31b8006e SW |
2296 | return 0; |
2297 | } | |
38941f80 | 2298 | |
4740a623 | 2299 | BUG_ON(!con->in_msg); |
38941f80 | 2300 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
2301 | m = con->in_msg; |
2302 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
2303 | if (m->middle) |
2304 | m->middle->vec.iov_len = 0; | |
9d7f0f13 | 2305 | |
78625051 | 2306 | /* prepare for data payload, if any */ |
a4107026 | 2307 | |
78625051 | 2308 | if (data_len) |
98fa5dd8 | 2309 | prepare_message_data(con->in_msg, data_len); |
31b8006e SW |
2310 | } |
2311 | ||
2312 | /* front */ | |
2450418c YS |
2313 | ret = read_partial_message_section(con, &m->front, front_len, |
2314 | &con->in_front_crc); | |
2315 | if (ret <= 0) | |
2316 | return ret; | |
31b8006e SW |
2317 | |
2318 | /* middle */ | |
2450418c | 2319 | if (m->middle) { |
213c99ee SW |
2320 | ret = read_partial_message_section(con, &m->middle->vec, |
2321 | middle_len, | |
2450418c | 2322 | &con->in_middle_crc); |
31b8006e SW |
2323 | if (ret <= 0) |
2324 | return ret; | |
31b8006e SW |
2325 | } |
2326 | ||
2327 | /* (page) data */ | |
34d2d200 AE |
2328 | if (data_len) { |
2329 | ret = read_partial_msg_data(con); | |
2330 | if (ret <= 0) | |
2331 | return ret; | |
31b8006e SW |
2332 | } |
2333 | ||
31b8006e | 2334 | /* footer */ |
fd51653f AE |
2335 | size = sizeof (m->footer); |
2336 | end += size; | |
2337 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
2338 | if (ret <= 0) |
2339 | return ret; | |
2340 | ||
31b8006e SW |
2341 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
2342 | m, front_len, m->footer.front_crc, middle_len, | |
2343 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
2344 | ||
2345 | /* crc ok? */ | |
2346 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
2347 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
2348 | m, con->in_front_crc, m->footer.front_crc); | |
2349 | return -EBADMSG; | |
2350 | } | |
2351 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
2352 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
2353 | m, con->in_middle_crc, m->footer.middle_crc); | |
2354 | return -EBADMSG; | |
2355 | } | |
bca064d2 | 2356 | if (do_datacrc && |
31b8006e SW |
2357 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
2358 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
2359 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
2360 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
2361 | return -EBADMSG; | |
2362 | } | |
2363 | ||
2364 | return 1; /* done! */ | |
2365 | } | |
2366 | ||
2367 | /* | |
2368 | * Process message. This happens in the worker thread. The callback should | |
2369 | * be careful not to do anything that waits on other incoming messages or it | |
2370 | * may deadlock. | |
2371 | */ | |
2372 | static void process_message(struct ceph_connection *con) | |
2373 | { | |
5e095e8b | 2374 | struct ceph_msg *msg; |
31b8006e | 2375 | |
38941f80 AE |
2376 | BUG_ON(con->in_msg->con != con); |
2377 | con->in_msg->con = NULL; | |
5e095e8b | 2378 | msg = con->in_msg; |
31b8006e | 2379 | con->in_msg = NULL; |
36eb71aa | 2380 | con->ops->put(con); |
31b8006e SW |
2381 | |
2382 | /* if first message, set peer_name */ | |
2383 | if (con->peer_name.type == 0) | |
dbad185d | 2384 | con->peer_name = msg->hdr.src; |
31b8006e | 2385 | |
31b8006e | 2386 | con->in_seq++; |
ec302645 | 2387 | mutex_unlock(&con->mutex); |
31b8006e SW |
2388 | |
2389 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
2390 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 2391 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
2392 | le16_to_cpu(msg->hdr.type), |
2393 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2394 | le32_to_cpu(msg->hdr.front_len), | |
2395 | le32_to_cpu(msg->hdr.data_len), | |
2396 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
2397 | con->ops->dispatch(con, msg); | |
ec302645 SW |
2398 | |
2399 | mutex_lock(&con->mutex); | |
31b8006e SW |
2400 | } |
2401 | ||
2402 | ||
2403 | /* | |
2404 | * Write something to the socket. Called in a worker thread when the | |
2405 | * socket appears to be writeable and we have something ready to send. | |
2406 | */ | |
2407 | static int try_write(struct ceph_connection *con) | |
2408 | { | |
31b8006e SW |
2409 | int ret = 1; |
2410 | ||
d59315ca | 2411 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 2412 | |
31b8006e SW |
2413 | more: |
2414 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
2415 | ||
2416 | /* open the socket first? */ | |
8dacc7da SW |
2417 | if (con->state == CON_STATE_PREOPEN) { |
2418 | BUG_ON(con->sock); | |
2419 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2420 | |
e2200423 | 2421 | con_out_kvec_reset(con); |
e825a66d | 2422 | prepare_write_banner(con); |
eed0ef2c | 2423 | prepare_read_banner(con); |
31b8006e | 2424 | |
cf3e5c40 | 2425 | BUG_ON(con->in_msg); |
31b8006e SW |
2426 | con->in_tag = CEPH_MSGR_TAG_READY; |
2427 | dout("try_write initiating connect on %p new state %lu\n", | |
2428 | con, con->state); | |
41617d0c AE |
2429 | ret = ceph_tcp_connect(con); |
2430 | if (ret < 0) { | |
31b8006e | 2431 | con->error_msg = "connect error"; |
31b8006e SW |
2432 | goto out; |
2433 | } | |
2434 | } | |
2435 | ||
2436 | more_kvec: | |
2437 | /* kvec data queued? */ | |
2438 | if (con->out_skip) { | |
2439 | ret = write_partial_skip(con); | |
2440 | if (ret <= 0) | |
42961d23 | 2441 | goto out; |
31b8006e SW |
2442 | } |
2443 | if (con->out_kvec_left) { | |
2444 | ret = write_partial_kvec(con); | |
2445 | if (ret <= 0) | |
42961d23 | 2446 | goto out; |
31b8006e SW |
2447 | } |
2448 | ||
2449 | /* msg pages? */ | |
2450 | if (con->out_msg) { | |
c86a2930 SW |
2451 | if (con->out_msg_done) { |
2452 | ceph_msg_put(con->out_msg); | |
2453 | con->out_msg = NULL; /* we're done with this one */ | |
2454 | goto do_next; | |
2455 | } | |
2456 | ||
34d2d200 | 2457 | ret = write_partial_message_data(con); |
31b8006e SW |
2458 | if (ret == 1) |
2459 | goto more_kvec; /* we need to send the footer, too! */ | |
2460 | if (ret == 0) | |
42961d23 | 2461 | goto out; |
31b8006e | 2462 | if (ret < 0) { |
34d2d200 | 2463 | dout("try_write write_partial_message_data err %d\n", |
31b8006e | 2464 | ret); |
42961d23 | 2465 | goto out; |
31b8006e SW |
2466 | } |
2467 | } | |
2468 | ||
c86a2930 | 2469 | do_next: |
8dacc7da | 2470 | if (con->state == CON_STATE_OPEN) { |
31b8006e SW |
2471 | /* is anything else pending? */ |
2472 | if (!list_empty(&con->out_queue)) { | |
2473 | prepare_write_message(con); | |
2474 | goto more; | |
2475 | } | |
2476 | if (con->in_seq > con->in_seq_acked) { | |
2477 | prepare_write_ack(con); | |
2478 | goto more; | |
2479 | } | |
c9ffc77a | 2480 | if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) { |
31b8006e SW |
2481 | prepare_write_keepalive(con); |
2482 | goto more; | |
2483 | } | |
2484 | } | |
2485 | ||
2486 | /* Nothing to do! */ | |
c9ffc77a | 2487 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
31b8006e | 2488 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2489 | ret = 0; |
2490 | out: | |
42961d23 | 2491 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2492 | return ret; |
2493 | } | |
2494 | ||
2495 | ||
2496 | ||
2497 | /* | |
2498 | * Read what we can from the socket. | |
2499 | */ | |
2500 | static int try_read(struct ceph_connection *con) | |
2501 | { | |
31b8006e SW |
2502 | int ret = -1; |
2503 | ||
8dacc7da SW |
2504 | more: |
2505 | dout("try_read start on %p state %lu\n", con, con->state); | |
2506 | if (con->state != CON_STATE_CONNECTING && | |
2507 | con->state != CON_STATE_NEGOTIATING && | |
2508 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2509 | return 0; |
2510 | ||
8dacc7da | 2511 | BUG_ON(!con->sock); |
ec302645 | 2512 | |
31b8006e SW |
2513 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2514 | con->in_base_pos); | |
0da5d703 | 2515 | |
8dacc7da | 2516 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2517 | dout("try_read connecting\n"); |
2518 | ret = read_partial_banner(con); | |
2519 | if (ret <= 0) | |
ab166d5a | 2520 | goto out; |
7593af92 AE |
2521 | ret = process_banner(con); |
2522 | if (ret < 0) | |
2523 | goto out; | |
2524 | ||
8dacc7da | 2525 | con->state = CON_STATE_NEGOTIATING; |
7593af92 | 2526 | |
6d4221b5 JS |
2527 | /* |
2528 | * Received banner is good, exchange connection info. | |
2529 | * Do not reset out_kvec, as sending our banner raced | |
2530 | * with receiving peer banner after connect completed. | |
2531 | */ | |
7593af92 AE |
2532 | ret = prepare_write_connect(con); |
2533 | if (ret < 0) | |
2534 | goto out; | |
2535 | prepare_read_connect(con); | |
2536 | ||
2537 | /* Send connection info before awaiting response */ | |
0da5d703 SW |
2538 | goto out; |
2539 | } | |
2540 | ||
8dacc7da | 2541 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2542 | dout("try_read negotiating\n"); |
31b8006e SW |
2543 | ret = read_partial_connect(con); |
2544 | if (ret <= 0) | |
31b8006e | 2545 | goto out; |
98bdb0aa SW |
2546 | ret = process_connect(con); |
2547 | if (ret < 0) | |
2548 | goto out; | |
31b8006e SW |
2549 | goto more; |
2550 | } | |
2551 | ||
122070a2 | 2552 | WARN_ON(con->state != CON_STATE_OPEN); |
8dacc7da | 2553 | |
31b8006e SW |
2554 | if (con->in_base_pos < 0) { |
2555 | /* | |
2556 | * skipping + discarding content. | |
2557 | * | |
2558 | * FIXME: there must be a better way to do this! | |
2559 | */ | |
84495f49 AE |
2560 | static char buf[SKIP_BUF_SIZE]; |
2561 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2562 | ||
31b8006e SW |
2563 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2564 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2565 | if (ret <= 0) | |
98bdb0aa | 2566 | goto out; |
31b8006e SW |
2567 | con->in_base_pos += ret; |
2568 | if (con->in_base_pos) | |
2569 | goto more; | |
2570 | } | |
2571 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2572 | /* | |
2573 | * what's next? | |
2574 | */ | |
2575 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2576 | if (ret <= 0) | |
98bdb0aa | 2577 | goto out; |
31b8006e SW |
2578 | dout("try_read got tag %d\n", (int)con->in_tag); |
2579 | switch (con->in_tag) { | |
2580 | case CEPH_MSGR_TAG_MSG: | |
2581 | prepare_read_message(con); | |
2582 | break; | |
2583 | case CEPH_MSGR_TAG_ACK: | |
2584 | prepare_read_ack(con); | |
2585 | break; | |
2586 | case CEPH_MSGR_TAG_CLOSE: | |
8dacc7da SW |
2587 | con_close_socket(con); |
2588 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2589 | goto out; |
31b8006e SW |
2590 | default: |
2591 | goto bad_tag; | |
2592 | } | |
2593 | } | |
2594 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2595 | ret = read_partial_message(con); | |
2596 | if (ret <= 0) { | |
2597 | switch (ret) { | |
2598 | case -EBADMSG: | |
2599 | con->error_msg = "bad crc"; | |
2600 | ret = -EIO; | |
98bdb0aa | 2601 | break; |
31b8006e SW |
2602 | case -EIO: |
2603 | con->error_msg = "io error"; | |
98bdb0aa | 2604 | break; |
31b8006e | 2605 | } |
98bdb0aa | 2606 | goto out; |
31b8006e SW |
2607 | } |
2608 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2609 | goto more; | |
2610 | process_message(con); | |
7b862e07 SW |
2611 | if (con->state == CON_STATE_OPEN) |
2612 | prepare_read_tag(con); | |
31b8006e SW |
2613 | goto more; |
2614 | } | |
3a23083b SW |
2615 | if (con->in_tag == CEPH_MSGR_TAG_ACK || |
2616 | con->in_tag == CEPH_MSGR_TAG_SEQ) { | |
2617 | /* | |
2618 | * the final handshake seq exchange is semantically | |
2619 | * equivalent to an ACK | |
2620 | */ | |
31b8006e SW |
2621 | ret = read_partial_ack(con); |
2622 | if (ret <= 0) | |
98bdb0aa | 2623 | goto out; |
31b8006e SW |
2624 | process_ack(con); |
2625 | goto more; | |
2626 | } | |
2627 | ||
31b8006e | 2628 | out: |
98bdb0aa | 2629 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2630 | return ret; |
2631 | ||
2632 | bad_tag: | |
2633 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2634 | con->error_msg = "protocol error, garbage tag"; | |
2635 | ret = -1; | |
2636 | goto out; | |
2637 | } | |
2638 | ||
2639 | ||
2640 | /* | |
802c6d96 AE |
2641 | * Atomically queue work on a connection after the specified delay. |
2642 | * Bump @con reference to avoid races with connection teardown. | |
2643 | * Returns 0 if work was queued, or an error code otherwise. | |
31b8006e | 2644 | */ |
802c6d96 | 2645 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
31b8006e | 2646 | { |
31b8006e | 2647 | if (!con->ops->get(con)) { |
802c6d96 AE |
2648 | dout("%s %p ref count 0\n", __func__, con); |
2649 | ||
2650 | return -ENOENT; | |
31b8006e SW |
2651 | } |
2652 | ||
802c6d96 AE |
2653 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
2654 | dout("%s %p - already queued\n", __func__, con); | |
31b8006e | 2655 | con->ops->put(con); |
802c6d96 AE |
2656 | |
2657 | return -EBUSY; | |
31b8006e | 2658 | } |
802c6d96 AE |
2659 | |
2660 | dout("%s %p %lu\n", __func__, con, delay); | |
2661 | ||
2662 | return 0; | |
2663 | } | |
2664 | ||
2665 | static void queue_con(struct ceph_connection *con) | |
2666 | { | |
2667 | (void) queue_con_delay(con, 0); | |
31b8006e SW |
2668 | } |
2669 | ||
7bb21d68 AE |
2670 | static bool con_sock_closed(struct ceph_connection *con) |
2671 | { | |
c9ffc77a | 2672 | if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED)) |
7bb21d68 AE |
2673 | return false; |
2674 | ||
2675 | #define CASE(x) \ | |
2676 | case CON_STATE_ ## x: \ | |
2677 | con->error_msg = "socket closed (con state " #x ")"; \ | |
2678 | break; | |
2679 | ||
2680 | switch (con->state) { | |
2681 | CASE(CLOSED); | |
2682 | CASE(PREOPEN); | |
2683 | CASE(CONNECTING); | |
2684 | CASE(NEGOTIATING); | |
2685 | CASE(OPEN); | |
2686 | CASE(STANDBY); | |
2687 | default: | |
2688 | pr_warning("%s con %p unrecognized state %lu\n", | |
2689 | __func__, con, con->state); | |
2690 | con->error_msg = "unrecognized con state"; | |
2691 | BUG(); | |
2692 | break; | |
2693 | } | |
2694 | #undef CASE | |
2695 | ||
2696 | return true; | |
2697 | } | |
2698 | ||
f20a39fd AE |
2699 | static bool con_backoff(struct ceph_connection *con) |
2700 | { | |
2701 | int ret; | |
2702 | ||
2703 | if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF)) | |
2704 | return false; | |
2705 | ||
2706 | ret = queue_con_delay(con, round_jiffies_relative(con->delay)); | |
2707 | if (ret) { | |
2708 | dout("%s: con %p FAILED to back off %lu\n", __func__, | |
2709 | con, con->delay); | |
2710 | BUG_ON(ret == -ENOENT); | |
2711 | con_flag_set(con, CON_FLAG_BACKOFF); | |
2712 | } | |
2713 | ||
2714 | return true; | |
2715 | } | |
2716 | ||
93209264 AE |
2717 | /* Finish fault handling; con->mutex must *not* be held here */ |
2718 | ||
2719 | static void con_fault_finish(struct ceph_connection *con) | |
2720 | { | |
2721 | /* | |
2722 | * in case we faulted due to authentication, invalidate our | |
2723 | * current tickets so that we can get new ones. | |
2724 | */ | |
2725 | if (con->auth_retry && con->ops->invalidate_authorizer) { | |
2726 | dout("calling invalidate_authorizer()\n"); | |
2727 | con->ops->invalidate_authorizer(con); | |
2728 | } | |
2729 | ||
2730 | if (con->ops->fault) | |
2731 | con->ops->fault(con); | |
2732 | } | |
2733 | ||
31b8006e SW |
2734 | /* |
2735 | * Do some work on a connection. Drop a connection ref when we're done. | |
2736 | */ | |
2737 | static void con_work(struct work_struct *work) | |
2738 | { | |
2739 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2740 | work.work); | |
49659416 | 2741 | bool fault; |
31b8006e | 2742 | |
9dd4658d | 2743 | mutex_lock(&con->mutex); |
49659416 AE |
2744 | while (true) { |
2745 | int ret; | |
31b8006e | 2746 | |
49659416 AE |
2747 | if ((fault = con_sock_closed(con))) { |
2748 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); | |
2749 | break; | |
2750 | } | |
2751 | if (con_backoff(con)) { | |
2752 | dout("%s: con %p BACKOFF\n", __func__, con); | |
2753 | break; | |
2754 | } | |
2755 | if (con->state == CON_STATE_STANDBY) { | |
2756 | dout("%s: con %p STANDBY\n", __func__, con); | |
2757 | break; | |
2758 | } | |
2759 | if (con->state == CON_STATE_CLOSED) { | |
2760 | dout("%s: con %p CLOSED\n", __func__, con); | |
2761 | BUG_ON(con->sock); | |
2762 | break; | |
2763 | } | |
2764 | if (con->state == CON_STATE_PREOPEN) { | |
2765 | dout("%s: con %p PREOPEN\n", __func__, con); | |
2766 | BUG_ON(con->sock); | |
2767 | } | |
0da5d703 | 2768 | |
49659416 AE |
2769 | ret = try_read(con); |
2770 | if (ret < 0) { | |
2771 | if (ret == -EAGAIN) | |
2772 | continue; | |
2773 | con->error_msg = "socket error on read"; | |
2774 | fault = true; | |
2775 | break; | |
2776 | } | |
2777 | ||
2778 | ret = try_write(con); | |
2779 | if (ret < 0) { | |
2780 | if (ret == -EAGAIN) | |
2781 | continue; | |
2782 | con->error_msg = "socket error on write"; | |
2783 | fault = true; | |
2784 | } | |
2785 | ||
2786 | break; /* If we make it to here, we're done */ | |
3a140a0d | 2787 | } |
b6e7b6a1 AE |
2788 | if (fault) |
2789 | con_fault(con); | |
9dd4658d | 2790 | mutex_unlock(&con->mutex); |
0da5d703 | 2791 | |
b6e7b6a1 AE |
2792 | if (fault) |
2793 | con_fault_finish(con); | |
2794 | ||
2795 | con->ops->put(con); | |
31b8006e SW |
2796 | } |
2797 | ||
31b8006e SW |
2798 | /* |
2799 | * Generic error/fault handler. A retry mechanism is used with | |
2800 | * exponential backoff | |
2801 | */ | |
93209264 | 2802 | static void con_fault(struct ceph_connection *con) |
31b8006e | 2803 | { |
28362986 | 2804 | pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2805 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2806 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2807 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2808 | |
122070a2 | 2809 | WARN_ON(con->state != CON_STATE_CONNECTING && |
8dacc7da SW |
2810 | con->state != CON_STATE_NEGOTIATING && |
2811 | con->state != CON_STATE_OPEN); | |
ec302645 | 2812 | |
31b8006e | 2813 | con_close_socket(con); |
5e095e8b | 2814 | |
c9ffc77a | 2815 | if (con_flag_test(con, CON_FLAG_LOSSYTX)) { |
8dacc7da SW |
2816 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
2817 | con->state = CON_STATE_CLOSED; | |
93209264 | 2818 | return; |
3b5ede07 SW |
2819 | } |
2820 | ||
5e095e8b | 2821 | if (con->in_msg) { |
38941f80 AE |
2822 | BUG_ON(con->in_msg->con != con); |
2823 | con->in_msg->con = NULL; | |
5e095e8b SW |
2824 | ceph_msg_put(con->in_msg); |
2825 | con->in_msg = NULL; | |
36eb71aa | 2826 | con->ops->put(con); |
5e095e8b | 2827 | } |
31b8006e | 2828 | |
e80a52d1 SW |
2829 | /* Requeue anything that hasn't been acked */ |
2830 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2831 | |
e76661d0 SW |
2832 | /* If there are no messages queued or keepalive pending, place |
2833 | * the connection in a STANDBY state */ | |
2834 | if (list_empty(&con->out_queue) && | |
c9ffc77a | 2835 | !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) { |
e00de341 | 2836 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
c9ffc77a | 2837 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
8dacc7da | 2838 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
2839 | } else { |
2840 | /* retry after a delay. */ | |
8dacc7da | 2841 | con->state = CON_STATE_PREOPEN; |
e80a52d1 SW |
2842 | if (con->delay == 0) |
2843 | con->delay = BASE_DELAY_INTERVAL; | |
2844 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2845 | con->delay *= 2; | |
c9ffc77a | 2846 | con_flag_set(con, CON_FLAG_BACKOFF); |
8618e30b | 2847 | queue_con(con); |
31b8006e | 2848 | } |
31b8006e SW |
2849 | } |
2850 | ||
2851 | ||
2852 | ||
2853 | /* | |
15d9882c | 2854 | * initialize a new messenger instance |
31b8006e | 2855 | */ |
15d9882c AE |
2856 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2857 | struct ceph_entity_addr *myaddr, | |
12b4629a ID |
2858 | u64 supported_features, |
2859 | u64 required_features, | |
15d9882c | 2860 | bool nocrc) |
31b8006e | 2861 | { |
3d14c5d2 YS |
2862 | msgr->supported_features = supported_features; |
2863 | msgr->required_features = required_features; | |
2864 | ||
31b8006e SW |
2865 | spin_lock_init(&msgr->global_seq_lock); |
2866 | ||
31b8006e SW |
2867 | if (myaddr) |
2868 | msgr->inst.addr = *myaddr; | |
2869 | ||
2870 | /* select a random nonce */ | |
ac8839d7 | 2871 | msgr->inst.addr.type = 0; |
103e2d3a | 2872 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2873 | encode_my_addr(msgr); |
15d9882c | 2874 | msgr->nocrc = nocrc; |
31b8006e | 2875 | |
a2a32584 | 2876 | atomic_set(&msgr->stopping, 0); |
31b8006e | 2877 | |
15d9882c | 2878 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2879 | } |
15d9882c | 2880 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2881 | |
e00de341 SW |
2882 | static void clear_standby(struct ceph_connection *con) |
2883 | { | |
2884 | /* come back from STANDBY? */ | |
8dacc7da | 2885 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 2886 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 2887 | con->state = CON_STATE_PREOPEN; |
e00de341 | 2888 | con->connect_seq++; |
c9ffc77a AE |
2889 | WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING)); |
2890 | WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)); | |
e00de341 SW |
2891 | } |
2892 | } | |
2893 | ||
31b8006e SW |
2894 | /* |
2895 | * Queue up an outgoing message on the given connection. | |
2896 | */ | |
2897 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2898 | { | |
31b8006e | 2899 | /* set src+dst */ |
dbad185d | 2900 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 2901 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
2902 | msg->needs_out_seq = true; |
2903 | ||
ec302645 | 2904 | mutex_lock(&con->mutex); |
92ce034b | 2905 | |
8dacc7da | 2906 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
2907 | dout("con_send %p closed, dropping %p\n", con, msg); |
2908 | ceph_msg_put(msg); | |
2909 | mutex_unlock(&con->mutex); | |
2910 | return; | |
2911 | } | |
2912 | ||
38941f80 | 2913 | BUG_ON(msg->con != NULL); |
36eb71aa | 2914 | msg->con = con->ops->get(con); |
92ce034b AE |
2915 | BUG_ON(msg->con == NULL); |
2916 | ||
31b8006e SW |
2917 | BUG_ON(!list_empty(&msg->list_head)); |
2918 | list_add_tail(&msg->list_head, &con->out_queue); | |
2919 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2920 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2921 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2922 | le32_to_cpu(msg->hdr.front_len), | |
2923 | le32_to_cpu(msg->hdr.middle_len), | |
2924 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
2925 | |
2926 | clear_standby(con); | |
ec302645 | 2927 | mutex_unlock(&con->mutex); |
31b8006e SW |
2928 | |
2929 | /* if there wasn't anything waiting to send before, queue | |
2930 | * new work */ | |
c9ffc77a | 2931 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
31b8006e SW |
2932 | queue_con(con); |
2933 | } | |
3d14c5d2 | 2934 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2935 | |
2936 | /* | |
2937 | * Revoke a message that was previously queued for send | |
2938 | */ | |
6740a845 | 2939 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2940 | { |
6740a845 AE |
2941 | struct ceph_connection *con = msg->con; |
2942 | ||
2943 | if (!con) | |
2944 | return; /* Message not in our possession */ | |
2945 | ||
ec302645 | 2946 | mutex_lock(&con->mutex); |
31b8006e | 2947 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2948 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2949 | list_del_init(&msg->list_head); |
38941f80 | 2950 | BUG_ON(msg->con == NULL); |
36eb71aa | 2951 | msg->con->ops->put(msg->con); |
38941f80 | 2952 | msg->con = NULL; |
31b8006e | 2953 | msg->hdr.seq = 0; |
38941f80 | 2954 | |
31b8006e | 2955 | ceph_msg_put(msg); |
ed98adad SW |
2956 | } |
2957 | if (con->out_msg == msg) { | |
38941f80 | 2958 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2959 | con->out_msg = NULL; |
31b8006e SW |
2960 | if (con->out_kvec_is_msg) { |
2961 | con->out_skip = con->out_kvec_bytes; | |
2962 | con->out_kvec_is_msg = false; | |
2963 | } | |
ed98adad | 2964 | msg->hdr.seq = 0; |
92ce034b AE |
2965 | |
2966 | ceph_msg_put(msg); | |
31b8006e | 2967 | } |
ec302645 | 2968 | mutex_unlock(&con->mutex); |
31b8006e SW |
2969 | } |
2970 | ||
350b1c32 | 2971 | /* |
0d59ab81 | 2972 | * Revoke a message that we may be reading data into |
350b1c32 | 2973 | */ |
8921d114 | 2974 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2975 | { |
8921d114 AE |
2976 | struct ceph_connection *con; |
2977 | ||
2978 | BUG_ON(msg == NULL); | |
2979 | if (!msg->con) { | |
2980 | dout("%s msg %p null con\n", __func__, msg); | |
2981 | ||
2982 | return; /* Message not in our possession */ | |
2983 | } | |
2984 | ||
2985 | con = msg->con; | |
350b1c32 | 2986 | mutex_lock(&con->mutex); |
8921d114 | 2987 | if (con->in_msg == msg) { |
95c96174 ED |
2988 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
2989 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
2990 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
2991 | |
2992 | /* skip rest of message */ | |
8921d114 AE |
2993 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
2994 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 2995 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
2996 | front_len - |
2997 | middle_len - | |
2998 | data_len - | |
350b1c32 | 2999 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
3000 | ceph_msg_put(con->in_msg); |
3001 | con->in_msg = NULL; | |
3002 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 3003 | con->in_seq++; |
350b1c32 | 3004 | } else { |
8921d114 AE |
3005 | dout("%s %p in_msg %p msg %p no-op\n", |
3006 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
3007 | } |
3008 | mutex_unlock(&con->mutex); | |
3009 | } | |
3010 | ||
31b8006e SW |
3011 | /* |
3012 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
3013 | */ | |
3014 | void ceph_con_keepalive(struct ceph_connection *con) | |
3015 | { | |
e00de341 | 3016 | dout("con_keepalive %p\n", con); |
00650931 | 3017 | mutex_lock(&con->mutex); |
e00de341 | 3018 | clear_standby(con); |
00650931 | 3019 | mutex_unlock(&con->mutex); |
c9ffc77a AE |
3020 | if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 && |
3021 | con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) | |
31b8006e SW |
3022 | queue_con(con); |
3023 | } | |
3d14c5d2 | 3024 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e | 3025 | |
6644ed7b | 3026 | static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type) |
43794509 | 3027 | { |
6644ed7b AE |
3028 | struct ceph_msg_data *data; |
3029 | ||
3030 | if (WARN_ON(!ceph_msg_data_type_valid(type))) | |
3031 | return NULL; | |
3032 | ||
81b36be4 | 3033 | data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS); |
6644ed7b AE |
3034 | if (data) |
3035 | data->type = type; | |
5240d9f9 | 3036 | INIT_LIST_HEAD(&data->links); |
6644ed7b AE |
3037 | |
3038 | return data; | |
3039 | } | |
3040 | ||
3041 | static void ceph_msg_data_destroy(struct ceph_msg_data *data) | |
3042 | { | |
3043 | if (!data) | |
3044 | return; | |
3045 | ||
5240d9f9 | 3046 | WARN_ON(!list_empty(&data->links)); |
6644ed7b AE |
3047 | if (data->type == CEPH_MSG_DATA_PAGELIST) { |
3048 | ceph_pagelist_release(data->pagelist); | |
3049 | kfree(data->pagelist); | |
3050 | } | |
81b36be4 | 3051 | kmem_cache_free(ceph_msg_data_cache, data); |
43794509 AE |
3052 | } |
3053 | ||
90af3602 | 3054 | void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages, |
f1baeb2b | 3055 | size_t length, size_t alignment) |
02afca6c | 3056 | { |
6644ed7b AE |
3057 | struct ceph_msg_data *data; |
3058 | ||
07aa1558 AE |
3059 | BUG_ON(!pages); |
3060 | BUG_ON(!length); | |
6644ed7b AE |
3061 | |
3062 | data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES); | |
3063 | BUG_ON(!data); | |
3064 | data->pages = pages; | |
3065 | data->length = length; | |
3066 | data->alignment = alignment & ~PAGE_MASK; | |
02afca6c | 3067 | |
5240d9f9 AE |
3068 | list_add_tail(&data->links, &msg->data); |
3069 | msg->data_length += length; | |
02afca6c | 3070 | } |
90af3602 | 3071 | EXPORT_SYMBOL(ceph_msg_data_add_pages); |
31b8006e | 3072 | |
90af3602 | 3073 | void ceph_msg_data_add_pagelist(struct ceph_msg *msg, |
27fa8385 AE |
3074 | struct ceph_pagelist *pagelist) |
3075 | { | |
6644ed7b AE |
3076 | struct ceph_msg_data *data; |
3077 | ||
07aa1558 AE |
3078 | BUG_ON(!pagelist); |
3079 | BUG_ON(!pagelist->length); | |
27fa8385 | 3080 | |
6644ed7b AE |
3081 | data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST); |
3082 | BUG_ON(!data); | |
3083 | data->pagelist = pagelist; | |
3084 | ||
5240d9f9 AE |
3085 | list_add_tail(&data->links, &msg->data); |
3086 | msg->data_length += pagelist->length; | |
27fa8385 | 3087 | } |
90af3602 | 3088 | EXPORT_SYMBOL(ceph_msg_data_add_pagelist); |
27fa8385 | 3089 | |
ea96571f | 3090 | #ifdef CONFIG_BLOCK |
90af3602 | 3091 | void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio, |
a1930804 | 3092 | size_t length) |
27fa8385 | 3093 | { |
6644ed7b AE |
3094 | struct ceph_msg_data *data; |
3095 | ||
07aa1558 | 3096 | BUG_ON(!bio); |
27fa8385 | 3097 | |
6644ed7b AE |
3098 | data = ceph_msg_data_create(CEPH_MSG_DATA_BIO); |
3099 | BUG_ON(!data); | |
3100 | data->bio = bio; | |
c851c495 | 3101 | data->bio_length = length; |
6644ed7b | 3102 | |
5240d9f9 AE |
3103 | list_add_tail(&data->links, &msg->data); |
3104 | msg->data_length += length; | |
27fa8385 | 3105 | } |
90af3602 | 3106 | EXPORT_SYMBOL(ceph_msg_data_add_bio); |
ea96571f | 3107 | #endif /* CONFIG_BLOCK */ |
27fa8385 | 3108 | |
31b8006e SW |
3109 | /* |
3110 | * construct a new message with given type, size | |
3111 | * the new msg has a ref count of 1. | |
3112 | */ | |
b61c2763 SW |
3113 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
3114 | bool can_fail) | |
31b8006e SW |
3115 | { |
3116 | struct ceph_msg *m; | |
3117 | ||
e3d5d638 | 3118 | m = kmem_cache_zalloc(ceph_msg_cache, flags); |
31b8006e SW |
3119 | if (m == NULL) |
3120 | goto out; | |
31b8006e SW |
3121 | |
3122 | m->hdr.type = cpu_to_le16(type); | |
45c6ceb5 | 3123 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
31b8006e | 3124 | m->hdr.front_len = cpu_to_le32(front_len); |
ca20892d | 3125 | |
9516e45b AE |
3126 | INIT_LIST_HEAD(&m->list_head); |
3127 | kref_init(&m->kref); | |
5240d9f9 | 3128 | INIT_LIST_HEAD(&m->data); |
ca20892d | 3129 | |
31b8006e | 3130 | /* front */ |
9516e45b | 3131 | m->front_max = front_len; |
31b8006e SW |
3132 | if (front_len) { |
3133 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 3134 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
3135 | PAGE_KERNEL); |
3136 | m->front_is_vmalloc = true; | |
3137 | } else { | |
34d23762 | 3138 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
3139 | } |
3140 | if (m->front.iov_base == NULL) { | |
b61c2763 | 3141 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
3142 | front_len); |
3143 | goto out2; | |
3144 | } | |
3145 | } else { | |
3146 | m->front.iov_base = NULL; | |
3147 | } | |
3148 | m->front.iov_len = front_len; | |
3149 | ||
bb257664 | 3150 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
3151 | return m; |
3152 | ||
3153 | out2: | |
3154 | ceph_msg_put(m); | |
3155 | out: | |
b61c2763 SW |
3156 | if (!can_fail) { |
3157 | pr_err("msg_new can't create type %d front %d\n", type, | |
3158 | front_len); | |
f0ed1b7c | 3159 | WARN_ON(1); |
b61c2763 SW |
3160 | } else { |
3161 | dout("msg_new can't create type %d front %d\n", type, | |
3162 | front_len); | |
3163 | } | |
a79832f2 | 3164 | return NULL; |
31b8006e | 3165 | } |
3d14c5d2 | 3166 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 3167 | |
31b8006e SW |
3168 | /* |
3169 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
3170 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
3171 | * per-type header in the front and then gracefully fail (i.e., | |
3172 | * propagate the error to the caller based on info in the front) when | |
3173 | * the middle is too large. | |
3174 | */ | |
2450418c | 3175 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
3176 | { |
3177 | int type = le16_to_cpu(msg->hdr.type); | |
3178 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
3179 | ||
3180 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
3181 | ceph_msg_type_name(type), middle_len); | |
3182 | BUG_ON(!middle_len); | |
3183 | BUG_ON(msg->middle); | |
3184 | ||
b6c1d5b8 | 3185 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
3186 | if (!msg->middle) |
3187 | return -ENOMEM; | |
3188 | return 0; | |
3189 | } | |
3190 | ||
2450418c | 3191 | /* |
1c20f2d2 AE |
3192 | * Allocate a message for receiving an incoming message on a |
3193 | * connection, and save the result in con->in_msg. Uses the | |
3194 | * connection's private alloc_msg op if available. | |
3195 | * | |
4740a623 SW |
3196 | * Returns 0 on success, or a negative error code. |
3197 | * | |
3198 | * On success, if we set *skip = 1: | |
3199 | * - the next message should be skipped and ignored. | |
3200 | * - con->in_msg == NULL | |
3201 | * or if we set *skip = 0: | |
3202 | * - con->in_msg is non-null. | |
3203 | * On error (ENOMEM, EAGAIN, ...), | |
3204 | * - con->in_msg == NULL | |
2450418c | 3205 | */ |
4740a623 | 3206 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
2450418c | 3207 | { |
4740a623 | 3208 | struct ceph_msg_header *hdr = &con->in_hdr; |
2450418c | 3209 | int middle_len = le32_to_cpu(hdr->middle_len); |
1d866d1c | 3210 | struct ceph_msg *msg; |
4740a623 | 3211 | int ret = 0; |
2450418c | 3212 | |
1c20f2d2 | 3213 | BUG_ON(con->in_msg != NULL); |
53ded495 | 3214 | BUG_ON(!con->ops->alloc_msg); |
2450418c | 3215 | |
53ded495 AE |
3216 | mutex_unlock(&con->mutex); |
3217 | msg = con->ops->alloc_msg(con, hdr, skip); | |
3218 | mutex_lock(&con->mutex); | |
3219 | if (con->state != CON_STATE_OPEN) { | |
3220 | if (msg) | |
1d866d1c | 3221 | ceph_msg_put(msg); |
53ded495 AE |
3222 | return -EAGAIN; |
3223 | } | |
4137577a AE |
3224 | if (msg) { |
3225 | BUG_ON(*skip); | |
3226 | con->in_msg = msg; | |
36eb71aa | 3227 | con->in_msg->con = con->ops->get(con); |
92ce034b | 3228 | BUG_ON(con->in_msg->con == NULL); |
4137577a AE |
3229 | } else { |
3230 | /* | |
3231 | * Null message pointer means either we should skip | |
3232 | * this message or we couldn't allocate memory. The | |
3233 | * former is not an error. | |
3234 | */ | |
3235 | if (*skip) | |
3236 | return 0; | |
3237 | con->error_msg = "error allocating memory for incoming message"; | |
3238 | ||
53ded495 | 3239 | return -ENOMEM; |
2450418c | 3240 | } |
1c20f2d2 | 3241 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 3242 | |
1c20f2d2 AE |
3243 | if (middle_len && !con->in_msg->middle) { |
3244 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 3245 | if (ret < 0) { |
1c20f2d2 AE |
3246 | ceph_msg_put(con->in_msg); |
3247 | con->in_msg = NULL; | |
2450418c YS |
3248 | } |
3249 | } | |
9d7f0f13 | 3250 | |
4740a623 | 3251 | return ret; |
2450418c YS |
3252 | } |
3253 | ||
31b8006e SW |
3254 | |
3255 | /* | |
3256 | * Free a generically kmalloc'd message. | |
3257 | */ | |
3258 | void ceph_msg_kfree(struct ceph_msg *m) | |
3259 | { | |
3260 | dout("msg_kfree %p\n", m); | |
3261 | if (m->front_is_vmalloc) | |
3262 | vfree(m->front.iov_base); | |
3263 | else | |
3264 | kfree(m->front.iov_base); | |
e3d5d638 | 3265 | kmem_cache_free(ceph_msg_cache, m); |
31b8006e SW |
3266 | } |
3267 | ||
3268 | /* | |
3269 | * Drop a msg ref. Destroy as needed. | |
3270 | */ | |
c2e552e7 SW |
3271 | void ceph_msg_last_put(struct kref *kref) |
3272 | { | |
3273 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
5240d9f9 AE |
3274 | LIST_HEAD(data); |
3275 | struct list_head *links; | |
3276 | struct list_head *next; | |
31b8006e | 3277 | |
c2e552e7 SW |
3278 | dout("ceph_msg_put last one on %p\n", m); |
3279 | WARN_ON(!list_empty(&m->list_head)); | |
3280 | ||
3281 | /* drop middle, data, if any */ | |
3282 | if (m->middle) { | |
3283 | ceph_buffer_put(m->middle); | |
3284 | m->middle = NULL; | |
31b8006e | 3285 | } |
5240d9f9 AE |
3286 | |
3287 | list_splice_init(&m->data, &data); | |
3288 | list_for_each_safe(links, next, &data) { | |
3289 | struct ceph_msg_data *data; | |
3290 | ||
3291 | data = list_entry(links, struct ceph_msg_data, links); | |
3292 | list_del_init(links); | |
3293 | ceph_msg_data_destroy(data); | |
3294 | } | |
a1930804 | 3295 | m->data_length = 0; |
58bb3b37 | 3296 | |
c2e552e7 SW |
3297 | if (m->pool) |
3298 | ceph_msgpool_put(m->pool, m); | |
3299 | else | |
3300 | ceph_msg_kfree(m); | |
31b8006e | 3301 | } |
3d14c5d2 | 3302 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
3303 | |
3304 | void ceph_msg_dump(struct ceph_msg *msg) | |
3305 | { | |
4a73ef27 | 3306 | pr_debug("msg_dump %p (front_max %d length %zd)\n", msg, |
5240d9f9 | 3307 | msg->front_max, msg->data_length); |
9ec7cab1 SW |
3308 | print_hex_dump(KERN_DEBUG, "header: ", |
3309 | DUMP_PREFIX_OFFSET, 16, 1, | |
3310 | &msg->hdr, sizeof(msg->hdr), true); | |
3311 | print_hex_dump(KERN_DEBUG, " front: ", | |
3312 | DUMP_PREFIX_OFFSET, 16, 1, | |
3313 | msg->front.iov_base, msg->front.iov_len, true); | |
3314 | if (msg->middle) | |
3315 | print_hex_dump(KERN_DEBUG, "middle: ", | |
3316 | DUMP_PREFIX_OFFSET, 16, 1, | |
3317 | msg->middle->vec.iov_base, | |
3318 | msg->middle->vec.iov_len, true); | |
3319 | print_hex_dump(KERN_DEBUG, "footer: ", | |
3320 | DUMP_PREFIX_OFFSET, 16, 1, | |
3321 | &msg->footer, sizeof(msg->footer), true); | |
3322 | } | |
3d14c5d2 | 3323 | EXPORT_SYMBOL(ceph_msg_dump); |