Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
31b8006e SW |
2 | |
3 | #include <linux/crc32c.h> | |
4 | #include <linux/ctype.h> | |
5 | #include <linux/highmem.h> | |
6 | #include <linux/inet.h> | |
7 | #include <linux/kthread.h> | |
8 | #include <linux/net.h> | |
5a0e3ad6 | 9 | #include <linux/slab.h> |
31b8006e SW |
10 | #include <linux/socket.h> |
11 | #include <linux/string.h> | |
3ebc21f7 | 12 | #ifdef CONFIG_BLOCK |
68b4476b | 13 | #include <linux/bio.h> |
3ebc21f7 | 14 | #endif /* CONFIG_BLOCK */ |
ee3b56f2 | 15 | #include <linux/dns_resolver.h> |
31b8006e SW |
16 | #include <net/tcp.h> |
17 | ||
3d14c5d2 YS |
18 | #include <linux/ceph/libceph.h> |
19 | #include <linux/ceph/messenger.h> | |
20 | #include <linux/ceph/decode.h> | |
21 | #include <linux/ceph/pagelist.h> | |
bc3b2d7f | 22 | #include <linux/export.h> |
31b8006e | 23 | |
fe38a2b6 AE |
24 | #define list_entry_next(pos, member) \ |
25 | list_entry(pos->member.next, typeof(*pos), member) | |
26 | ||
31b8006e SW |
27 | /* |
28 | * Ceph uses the messenger to exchange ceph_msg messages with other | |
29 | * hosts in the system. The messenger provides ordered and reliable | |
30 | * delivery. We tolerate TCP disconnects by reconnecting (with | |
31 | * exponential backoff) in the case of a fault (disconnection, bad | |
32 | * crc, protocol error). Acks allow sent messages to be discarded by | |
33 | * the sender. | |
34 | */ | |
35 | ||
bc18f4b1 AE |
36 | /* |
37 | * We track the state of the socket on a given connection using | |
38 | * values defined below. The transition to a new socket state is | |
39 | * handled by a function which verifies we aren't coming from an | |
40 | * unexpected state. | |
41 | * | |
42 | * -------- | |
43 | * | NEW* | transient initial state | |
44 | * -------- | |
45 | * | con_sock_state_init() | |
46 | * v | |
47 | * ---------- | |
48 | * | CLOSED | initialized, but no socket (and no | |
49 | * ---------- TCP connection) | |
50 | * ^ \ | |
51 | * | \ con_sock_state_connecting() | |
52 | * | ---------------------- | |
53 | * | \ | |
54 | * + con_sock_state_closed() \ | |
fbb85a47 SW |
55 | * |+--------------------------- \ |
56 | * | \ \ \ | |
57 | * | ----------- \ \ | |
58 | * | | CLOSING | socket event; \ \ | |
59 | * | ----------- await close \ \ | |
60 | * | ^ \ | | |
61 | * | | \ | | |
62 | * | + con_sock_state_closing() \ | | |
63 | * | / \ | | | |
64 | * | / --------------- | | | |
65 | * | / \ v v | |
bc18f4b1 AE |
66 | * | / -------------- |
67 | * | / -----------------| CONNECTING | socket created, TCP | |
68 | * | | / -------------- connect initiated | |
69 | * | | | con_sock_state_connected() | |
70 | * | | v | |
71 | * ------------- | |
72 | * | CONNECTED | TCP connection established | |
73 | * ------------- | |
74 | * | |
75 | * State values for ceph_connection->sock_state; NEW is assumed to be 0. | |
76 | */ | |
ce2c8903 AE |
77 | |
78 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ | |
79 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ | |
80 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ | |
81 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ | |
82 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ | |
83 | ||
8dacc7da SW |
84 | /* |
85 | * connection states | |
86 | */ | |
87 | #define CON_STATE_CLOSED 1 /* -> PREOPEN */ | |
88 | #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */ | |
89 | #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */ | |
90 | #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */ | |
91 | #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */ | |
92 | #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */ | |
93 | ||
4a861692 SW |
94 | /* |
95 | * ceph_connection flag bits | |
96 | */ | |
97 | #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop | |
98 | * messages on errors */ | |
99 | #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */ | |
100 | #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */ | |
101 | #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */ | |
102 | #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */ | |
8dacc7da | 103 | |
c9ffc77a AE |
104 | static bool con_flag_valid(unsigned long con_flag) |
105 | { | |
106 | switch (con_flag) { | |
107 | case CON_FLAG_LOSSYTX: | |
108 | case CON_FLAG_KEEPALIVE_PENDING: | |
109 | case CON_FLAG_WRITE_PENDING: | |
110 | case CON_FLAG_SOCK_CLOSED: | |
111 | case CON_FLAG_BACKOFF: | |
112 | return true; | |
113 | default: | |
114 | return false; | |
115 | } | |
116 | } | |
117 | ||
118 | static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag) | |
119 | { | |
120 | BUG_ON(!con_flag_valid(con_flag)); | |
121 | ||
122 | clear_bit(con_flag, &con->flags); | |
123 | } | |
124 | ||
125 | static void con_flag_set(struct ceph_connection *con, unsigned long con_flag) | |
126 | { | |
127 | BUG_ON(!con_flag_valid(con_flag)); | |
128 | ||
129 | set_bit(con_flag, &con->flags); | |
130 | } | |
131 | ||
132 | static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag) | |
133 | { | |
134 | BUG_ON(!con_flag_valid(con_flag)); | |
135 | ||
136 | return test_bit(con_flag, &con->flags); | |
137 | } | |
138 | ||
139 | static bool con_flag_test_and_clear(struct ceph_connection *con, | |
140 | unsigned long con_flag) | |
141 | { | |
142 | BUG_ON(!con_flag_valid(con_flag)); | |
143 | ||
144 | return test_and_clear_bit(con_flag, &con->flags); | |
145 | } | |
146 | ||
147 | static bool con_flag_test_and_set(struct ceph_connection *con, | |
148 | unsigned long con_flag) | |
149 | { | |
150 | BUG_ON(!con_flag_valid(con_flag)); | |
151 | ||
152 | return test_and_set_bit(con_flag, &con->flags); | |
153 | } | |
154 | ||
e3d5d638 AE |
155 | /* Slab caches for frequently-allocated structures */ |
156 | ||
157 | static struct kmem_cache *ceph_msg_cache; | |
81b36be4 | 158 | static struct kmem_cache *ceph_msg_data_cache; |
e3d5d638 | 159 | |
31b8006e SW |
160 | /* static tag bytes (protocol control messages) */ |
161 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
162 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
163 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
164 | ||
a6a5349d SW |
165 | #ifdef CONFIG_LOCKDEP |
166 | static struct lock_class_key socket_class; | |
167 | #endif | |
168 | ||
84495f49 AE |
169 | /* |
170 | * When skipping (ignoring) a block of input we read it into a "skip | |
171 | * buffer," which is this many bytes in size. | |
172 | */ | |
173 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
174 | |
175 | static void queue_con(struct ceph_connection *con); | |
176 | static void con_work(struct work_struct *); | |
93209264 | 177 | static void con_fault(struct ceph_connection *con); |
31b8006e | 178 | |
31b8006e | 179 | /* |
f64a9317 AE |
180 | * Nicely render a sockaddr as a string. An array of formatted |
181 | * strings is used, to approximate reentrancy. | |
31b8006e | 182 | */ |
f64a9317 AE |
183 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
184 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
185 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
186 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
187 | ||
188 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
189 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 190 | |
57666519 | 191 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 192 | |
3d14c5d2 | 193 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
194 | { |
195 | int i; | |
196 | char *s; | |
99f0f3b2 AE |
197 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
198 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 199 | |
f64a9317 | 200 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
201 | s = addr_str[i]; |
202 | ||
203 | switch (ss->ss_family) { | |
204 | case AF_INET: | |
bd406145 AE |
205 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
206 | ntohs(in4->sin_port)); | |
31b8006e SW |
207 | break; |
208 | ||
209 | case AF_INET6: | |
bd406145 AE |
210 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
211 | ntohs(in6->sin6_port)); | |
31b8006e SW |
212 | break; |
213 | ||
214 | default: | |
d3002b97 AE |
215 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
216 | ss->ss_family); | |
31b8006e SW |
217 | } |
218 | ||
219 | return s; | |
220 | } | |
3d14c5d2 | 221 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 222 | |
63f2d211 SW |
223 | static void encode_my_addr(struct ceph_messenger *msgr) |
224 | { | |
225 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
226 | ceph_encode_addr(&msgr->my_enc_addr); | |
227 | } | |
228 | ||
31b8006e SW |
229 | /* |
230 | * work queue for all reading and writing to/from the socket. | |
231 | */ | |
e0f43c94 | 232 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 233 | |
e3d5d638 AE |
234 | static int ceph_msgr_slab_init(void) |
235 | { | |
236 | BUG_ON(ceph_msg_cache); | |
237 | ceph_msg_cache = kmem_cache_create("ceph_msg", | |
238 | sizeof (struct ceph_msg), | |
239 | __alignof__(struct ceph_msg), 0, NULL); | |
81b36be4 AE |
240 | |
241 | if (!ceph_msg_cache) | |
242 | return -ENOMEM; | |
243 | ||
244 | BUG_ON(ceph_msg_data_cache); | |
245 | ceph_msg_data_cache = kmem_cache_create("ceph_msg_data", | |
246 | sizeof (struct ceph_msg_data), | |
247 | __alignof__(struct ceph_msg_data), | |
248 | 0, NULL); | |
249 | if (ceph_msg_data_cache) | |
250 | return 0; | |
251 | ||
252 | kmem_cache_destroy(ceph_msg_cache); | |
253 | ceph_msg_cache = NULL; | |
254 | ||
255 | return -ENOMEM; | |
e3d5d638 AE |
256 | } |
257 | ||
258 | static void ceph_msgr_slab_exit(void) | |
259 | { | |
81b36be4 AE |
260 | BUG_ON(!ceph_msg_data_cache); |
261 | kmem_cache_destroy(ceph_msg_data_cache); | |
262 | ceph_msg_data_cache = NULL; | |
263 | ||
e3d5d638 AE |
264 | BUG_ON(!ceph_msg_cache); |
265 | kmem_cache_destroy(ceph_msg_cache); | |
266 | ceph_msg_cache = NULL; | |
267 | } | |
268 | ||
15417167 | 269 | static void _ceph_msgr_exit(void) |
6173d1f0 | 270 | { |
d3002b97 | 271 | if (ceph_msgr_wq) { |
6173d1f0 | 272 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
273 | ceph_msgr_wq = NULL; |
274 | } | |
6173d1f0 | 275 | |
e3d5d638 AE |
276 | ceph_msgr_slab_exit(); |
277 | ||
6173d1f0 AE |
278 | BUG_ON(zero_page == NULL); |
279 | kunmap(zero_page); | |
280 | page_cache_release(zero_page); | |
281 | zero_page = NULL; | |
282 | } | |
283 | ||
3d14c5d2 | 284 | int ceph_msgr_init(void) |
31b8006e | 285 | { |
57666519 AE |
286 | BUG_ON(zero_page != NULL); |
287 | zero_page = ZERO_PAGE(0); | |
288 | page_cache_get(zero_page); | |
289 | ||
e3d5d638 AE |
290 | if (ceph_msgr_slab_init()) |
291 | return -ENOMEM; | |
292 | ||
f05c0daa ID |
293 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", |
294 | WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 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)) { |
182fac26 | 413 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(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 | ||
a757a4e2 | 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 | ||
a757a4e2 CC |
573 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
574 | int offset, size_t size, bool more) | |
575 | { | |
576 | int ret; | |
577 | struct kvec iov; | |
578 | ||
579 | /* sendpage cannot properly handle pages with page_count == 0, | |
580 | * we need to fallback to sendmsg if that's the case */ | |
581 | if (page_count(page) >= 1) | |
582 | return __ceph_tcp_sendpage(sock, page, offset, size, more); | |
583 | ||
584 | iov.iov_base = kmap(page) + offset; | |
585 | iov.iov_len = size; | |
586 | ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more); | |
587 | kunmap(page); | |
588 | ||
589 | return ret; | |
590 | } | |
31b8006e SW |
591 | |
592 | /* | |
593 | * Shutdown/close the socket for the given connection. | |
594 | */ | |
595 | static int con_close_socket(struct ceph_connection *con) | |
596 | { | |
8007b8d6 | 597 | int rc = 0; |
31b8006e SW |
598 | |
599 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
8007b8d6 SW |
600 | if (con->sock) { |
601 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
602 | sock_release(con->sock); | |
603 | con->sock = NULL; | |
604 | } | |
456ea468 AE |
605 | |
606 | /* | |
4a861692 | 607 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
608 | * independent of the connection mutex, and we could have |
609 | * received a socket close event before we had the chance to | |
610 | * shut the socket down. | |
611 | */ | |
c9ffc77a | 612 | con_flag_clear(con, CON_FLAG_SOCK_CLOSED); |
8007b8d6 | 613 | |
ce2c8903 | 614 | con_sock_state_closed(con); |
31b8006e SW |
615 | return rc; |
616 | } | |
617 | ||
618 | /* | |
619 | * Reset a connection. Discard all incoming and outgoing messages | |
620 | * and clear *_seq state. | |
621 | */ | |
622 | static void ceph_msg_remove(struct ceph_msg *msg) | |
623 | { | |
624 | list_del_init(&msg->list_head); | |
38941f80 | 625 | BUG_ON(msg->con == NULL); |
36eb71aa | 626 | msg->con->ops->put(msg->con); |
38941f80 AE |
627 | msg->con = NULL; |
628 | ||
31b8006e SW |
629 | ceph_msg_put(msg); |
630 | } | |
631 | static void ceph_msg_remove_list(struct list_head *head) | |
632 | { | |
633 | while (!list_empty(head)) { | |
634 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
635 | list_head); | |
636 | ceph_msg_remove(msg); | |
637 | } | |
638 | } | |
639 | ||
640 | static void reset_connection(struct ceph_connection *con) | |
641 | { | |
642 | /* reset connection, out_queue, msg_ and connect_seq */ | |
643 | /* discard existing out_queue and msg_seq */ | |
0fa6ebc6 | 644 | dout("reset_connection %p\n", con); |
31b8006e SW |
645 | ceph_msg_remove_list(&con->out_queue); |
646 | ceph_msg_remove_list(&con->out_sent); | |
647 | ||
cf3e5c40 | 648 | if (con->in_msg) { |
38941f80 AE |
649 | BUG_ON(con->in_msg->con != con); |
650 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
651 | ceph_msg_put(con->in_msg); |
652 | con->in_msg = NULL; | |
36eb71aa | 653 | con->ops->put(con); |
cf3e5c40 SW |
654 | } |
655 | ||
31b8006e SW |
656 | con->connect_seq = 0; |
657 | con->out_seq = 0; | |
c86a2930 SW |
658 | if (con->out_msg) { |
659 | ceph_msg_put(con->out_msg); | |
660 | con->out_msg = NULL; | |
661 | } | |
31b8006e | 662 | con->in_seq = 0; |
0e0d5e0c | 663 | con->in_seq_acked = 0; |
31b8006e SW |
664 | } |
665 | ||
666 | /* | |
667 | * mark a peer down. drop any open connections. | |
668 | */ | |
669 | void ceph_con_close(struct ceph_connection *con) | |
670 | { | |
8c50c817 | 671 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
672 | dout("con_close %p peer %s\n", con, |
673 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
8dacc7da | 674 | con->state = CON_STATE_CLOSED; |
a5988c49 | 675 | |
c9ffc77a AE |
676 | con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */ |
677 | con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING); | |
678 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); | |
679 | con_flag_clear(con, CON_FLAG_BACKOFF); | |
a5988c49 | 680 | |
31b8006e | 681 | reset_connection(con); |
6f2bc3ff | 682 | con->peer_global_seq = 0; |
91e45ce3 | 683 | cancel_delayed_work(&con->work); |
ee76e073 | 684 | con_close_socket(con); |
ec302645 | 685 | mutex_unlock(&con->mutex); |
31b8006e | 686 | } |
3d14c5d2 | 687 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 688 | |
31b8006e SW |
689 | /* |
690 | * Reopen a closed connection, with a new peer address. | |
691 | */ | |
b7a9e5dd SW |
692 | void ceph_con_open(struct ceph_connection *con, |
693 | __u8 entity_type, __u64 entity_num, | |
694 | struct ceph_entity_addr *addr) | |
31b8006e | 695 | { |
5469155f | 696 | mutex_lock(&con->mutex); |
3d14c5d2 | 697 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
8dacc7da | 698 | |
122070a2 | 699 | WARN_ON(con->state != CON_STATE_CLOSED); |
8dacc7da | 700 | con->state = CON_STATE_PREOPEN; |
a5988c49 | 701 | |
b7a9e5dd SW |
702 | con->peer_name.type = (__u8) entity_type; |
703 | con->peer_name.num = cpu_to_le64(entity_num); | |
704 | ||
31b8006e | 705 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 706 | con->delay = 0; /* reset backoff memory */ |
5469155f | 707 | mutex_unlock(&con->mutex); |
31b8006e SW |
708 | queue_con(con); |
709 | } | |
3d14c5d2 | 710 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 711 | |
87b315a5 SW |
712 | /* |
713 | * return true if this connection ever successfully opened | |
714 | */ | |
715 | bool ceph_con_opened(struct ceph_connection *con) | |
716 | { | |
717 | return con->connect_seq > 0; | |
718 | } | |
719 | ||
31b8006e SW |
720 | /* |
721 | * initialize a new connection. | |
722 | */ | |
1bfd89f4 AE |
723 | void ceph_con_init(struct ceph_connection *con, void *private, |
724 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 725 | struct ceph_messenger *msgr) |
31b8006e SW |
726 | { |
727 | dout("con_init %p\n", con); | |
728 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
729 | con->private = private; |
730 | con->ops = ops; | |
31b8006e | 731 | con->msgr = msgr; |
ce2c8903 AE |
732 | |
733 | con_sock_state_init(con); | |
734 | ||
ec302645 | 735 | mutex_init(&con->mutex); |
31b8006e SW |
736 | INIT_LIST_HEAD(&con->out_queue); |
737 | INIT_LIST_HEAD(&con->out_sent); | |
738 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 | 739 | |
8dacc7da | 740 | con->state = CON_STATE_CLOSED; |
31b8006e | 741 | } |
3d14c5d2 | 742 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
743 | |
744 | ||
745 | /* | |
746 | * We maintain a global counter to order connection attempts. Get | |
747 | * a unique seq greater than @gt. | |
748 | */ | |
749 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
750 | { | |
751 | u32 ret; | |
752 | ||
753 | spin_lock(&msgr->global_seq_lock); | |
754 | if (msgr->global_seq < gt) | |
755 | msgr->global_seq = gt; | |
756 | ret = ++msgr->global_seq; | |
757 | spin_unlock(&msgr->global_seq_lock); | |
758 | return ret; | |
759 | } | |
760 | ||
e2200423 | 761 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
762 | { |
763 | con->out_kvec_left = 0; | |
764 | con->out_kvec_bytes = 0; | |
765 | con->out_kvec_cur = &con->out_kvec[0]; | |
766 | } | |
767 | ||
e2200423 | 768 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
769 | size_t size, void *data) |
770 | { | |
771 | int index; | |
772 | ||
773 | index = con->out_kvec_left; | |
774 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
775 | ||
776 | con->out_kvec[index].iov_len = size; | |
777 | con->out_kvec[index].iov_base = data; | |
778 | con->out_kvec_left++; | |
779 | con->out_kvec_bytes += size; | |
780 | } | |
31b8006e | 781 | |
df6ad1f9 | 782 | #ifdef CONFIG_BLOCK |
6aaa4511 AE |
783 | |
784 | /* | |
785 | * For a bio data item, a piece is whatever remains of the next | |
786 | * entry in the current bio iovec, or the first entry in the next | |
787 | * bio in the list. | |
788 | */ | |
8ae4f4f5 | 789 | static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 790 | size_t length) |
6aaa4511 | 791 | { |
8ae4f4f5 | 792 | struct ceph_msg_data *data = cursor->data; |
6aaa4511 AE |
793 | struct bio *bio; |
794 | ||
795 | BUG_ON(data->type != CEPH_MSG_DATA_BIO); | |
796 | ||
797 | bio = data->bio; | |
798 | BUG_ON(!bio); | |
799 | BUG_ON(!bio->bi_vcnt); | |
6aaa4511 | 800 | |
ca8b3a69 | 801 | cursor->resid = min(length, data->bio_length); |
6aaa4511 AE |
802 | cursor->bio = bio; |
803 | cursor->vector_index = 0; | |
804 | cursor->vector_offset = 0; | |
25aff7c5 | 805 | cursor->last_piece = length <= bio->bi_io_vec[0].bv_len; |
6aaa4511 AE |
806 | } |
807 | ||
8ae4f4f5 | 808 | static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor, |
6aaa4511 AE |
809 | size_t *page_offset, |
810 | size_t *length) | |
811 | { | |
8ae4f4f5 | 812 | struct ceph_msg_data *data = cursor->data; |
6aaa4511 AE |
813 | struct bio *bio; |
814 | struct bio_vec *bio_vec; | |
815 | unsigned int index; | |
816 | ||
817 | BUG_ON(data->type != CEPH_MSG_DATA_BIO); | |
818 | ||
819 | bio = cursor->bio; | |
820 | BUG_ON(!bio); | |
821 | ||
822 | index = cursor->vector_index; | |
823 | BUG_ON(index >= (unsigned int) bio->bi_vcnt); | |
824 | ||
825 | bio_vec = &bio->bi_io_vec[index]; | |
826 | BUG_ON(cursor->vector_offset >= bio_vec->bv_len); | |
827 | *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset); | |
828 | BUG_ON(*page_offset >= PAGE_SIZE); | |
25aff7c5 AE |
829 | if (cursor->last_piece) /* pagelist offset is always 0 */ |
830 | *length = cursor->resid; | |
831 | else | |
832 | *length = (size_t) (bio_vec->bv_len - cursor->vector_offset); | |
25aff7c5 | 833 | BUG_ON(*length > cursor->resid); |
5df521b1 | 834 | BUG_ON(*page_offset + *length > PAGE_SIZE); |
6aaa4511 AE |
835 | |
836 | return bio_vec->bv_page; | |
837 | } | |
838 | ||
8ae4f4f5 AE |
839 | static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor, |
840 | size_t bytes) | |
6aaa4511 | 841 | { |
6aaa4511 AE |
842 | struct bio *bio; |
843 | struct bio_vec *bio_vec; | |
844 | unsigned int index; | |
845 | ||
8ae4f4f5 | 846 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO); |
6aaa4511 AE |
847 | |
848 | bio = cursor->bio; | |
849 | BUG_ON(!bio); | |
850 | ||
851 | index = cursor->vector_index; | |
852 | BUG_ON(index >= (unsigned int) bio->bi_vcnt); | |
853 | bio_vec = &bio->bi_io_vec[index]; | |
6aaa4511 AE |
854 | |
855 | /* Advance the cursor offset */ | |
856 | ||
25aff7c5 AE |
857 | BUG_ON(cursor->resid < bytes); |
858 | cursor->resid -= bytes; | |
6aaa4511 AE |
859 | cursor->vector_offset += bytes; |
860 | if (cursor->vector_offset < bio_vec->bv_len) | |
861 | return false; /* more bytes to process in this segment */ | |
25aff7c5 | 862 | BUG_ON(cursor->vector_offset != bio_vec->bv_len); |
6aaa4511 AE |
863 | |
864 | /* Move on to the next segment, and possibly the next bio */ | |
865 | ||
25aff7c5 | 866 | if (++index == (unsigned int) bio->bi_vcnt) { |
6aaa4511 | 867 | bio = bio->bi_next; |
25aff7c5 | 868 | index = 0; |
6aaa4511 | 869 | } |
25aff7c5 AE |
870 | cursor->bio = bio; |
871 | cursor->vector_index = index; | |
6aaa4511 AE |
872 | cursor->vector_offset = 0; |
873 | ||
25aff7c5 AE |
874 | if (!cursor->last_piece) { |
875 | BUG_ON(!cursor->resid); | |
876 | BUG_ON(!bio); | |
877 | /* A short read is OK, so use <= rather than == */ | |
878 | if (cursor->resid <= bio->bi_io_vec[index].bv_len) | |
6aaa4511 | 879 | cursor->last_piece = true; |
25aff7c5 | 880 | } |
6aaa4511 AE |
881 | |
882 | return true; | |
883 | } | |
ea96571f | 884 | #endif /* CONFIG_BLOCK */ |
df6ad1f9 | 885 | |
e766d7b5 AE |
886 | /* |
887 | * For a page array, a piece comes from the first page in the array | |
888 | * that has not already been fully consumed. | |
889 | */ | |
8ae4f4f5 | 890 | static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor, |
25aff7c5 | 891 | size_t length) |
e766d7b5 | 892 | { |
8ae4f4f5 | 893 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
894 | int page_count; |
895 | ||
896 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
897 | ||
898 | BUG_ON(!data->pages); | |
899 | BUG_ON(!data->length); | |
900 | ||
ca8b3a69 | 901 | cursor->resid = min(length, data->length); |
e766d7b5 | 902 | page_count = calc_pages_for(data->alignment, (u64)data->length); |
e766d7b5 AE |
903 | cursor->page_offset = data->alignment & ~PAGE_MASK; |
904 | cursor->page_index = 0; | |
56fc5659 AE |
905 | BUG_ON(page_count > (int)USHRT_MAX); |
906 | cursor->page_count = (unsigned short)page_count; | |
907 | BUG_ON(length > SIZE_MAX - cursor->page_offset); | |
a6489727 | 908 | cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
909 | } |
910 | ||
8ae4f4f5 AE |
911 | static struct page * |
912 | ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor, | |
913 | size_t *page_offset, size_t *length) | |
e766d7b5 | 914 | { |
8ae4f4f5 | 915 | struct ceph_msg_data *data = cursor->data; |
e766d7b5 AE |
916 | |
917 | BUG_ON(data->type != CEPH_MSG_DATA_PAGES); | |
918 | ||
919 | BUG_ON(cursor->page_index >= cursor->page_count); | |
920 | BUG_ON(cursor->page_offset >= PAGE_SIZE); | |
e766d7b5 AE |
921 | |
922 | *page_offset = cursor->page_offset; | |
25aff7c5 | 923 | if (cursor->last_piece) |
e766d7b5 | 924 | *length = cursor->resid; |
25aff7c5 | 925 | else |
e766d7b5 | 926 | *length = PAGE_SIZE - *page_offset; |
e766d7b5 AE |
927 | |
928 | return data->pages[cursor->page_index]; | |
929 | } | |
930 | ||
8ae4f4f5 | 931 | static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor, |
e766d7b5 AE |
932 | size_t bytes) |
933 | { | |
8ae4f4f5 | 934 | BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES); |
e766d7b5 AE |
935 | |
936 | BUG_ON(cursor->page_offset + bytes > PAGE_SIZE); | |
e766d7b5 AE |
937 | |
938 | /* Advance the cursor page offset */ | |
939 | ||
940 | cursor->resid -= bytes; | |
5df521b1 AE |
941 | cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK; |
942 | if (!bytes || cursor->page_offset) | |
e766d7b5 AE |
943 | return false; /* more bytes to process in the current page */ |
944 | ||
5df521b1 | 945 | /* Move on to the next page; offset is already at 0 */ |
e766d7b5 AE |
946 | |
947 | BUG_ON(cursor->page_index >= cursor->page_count); | |
e766d7b5 | 948 | cursor->page_index++; |
25aff7c5 | 949 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
e766d7b5 AE |
950 | |
951 | return true; | |
952 | } | |
953 | ||
fe38a2b6 | 954 | /* |
dd236fcb AE |
955 | * For a pagelist, a piece is whatever remains to be consumed in the |
956 | * first page in the list, or the front of the next page. | |
fe38a2b6 | 957 | */ |
8ae4f4f5 AE |
958 | static void |
959 | ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor, | |
25aff7c5 | 960 | size_t length) |
fe38a2b6 | 961 | { |
8ae4f4f5 | 962 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
963 | struct ceph_pagelist *pagelist; |
964 | struct page *page; | |
965 | ||
dd236fcb | 966 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); |
fe38a2b6 AE |
967 | |
968 | pagelist = data->pagelist; | |
969 | BUG_ON(!pagelist); | |
25aff7c5 AE |
970 | |
971 | if (!length) | |
fe38a2b6 AE |
972 | return; /* pagelist can be assigned but empty */ |
973 | ||
974 | BUG_ON(list_empty(&pagelist->head)); | |
975 | page = list_first_entry(&pagelist->head, struct page, lru); | |
976 | ||
ca8b3a69 | 977 | cursor->resid = min(length, pagelist->length); |
fe38a2b6 AE |
978 | cursor->page = page; |
979 | cursor->offset = 0; | |
a51b272e | 980 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
981 | } |
982 | ||
8ae4f4f5 AE |
983 | static struct page * |
984 | ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor, | |
985 | size_t *page_offset, size_t *length) | |
fe38a2b6 | 986 | { |
8ae4f4f5 | 987 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 | 988 | struct ceph_pagelist *pagelist; |
fe38a2b6 AE |
989 | |
990 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
991 | ||
992 | pagelist = data->pagelist; | |
993 | BUG_ON(!pagelist); | |
994 | ||
995 | BUG_ON(!cursor->page); | |
25aff7c5 | 996 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); |
fe38a2b6 | 997 | |
5df521b1 | 998 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 | 999 | *page_offset = cursor->offset & ~PAGE_MASK; |
5df521b1 | 1000 | if (cursor->last_piece) |
25aff7c5 AE |
1001 | *length = cursor->resid; |
1002 | else | |
1003 | *length = PAGE_SIZE - *page_offset; | |
fe38a2b6 | 1004 | |
8ae4f4f5 | 1005 | return cursor->page; |
fe38a2b6 AE |
1006 | } |
1007 | ||
8ae4f4f5 | 1008 | static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor, |
dd236fcb | 1009 | size_t bytes) |
fe38a2b6 | 1010 | { |
8ae4f4f5 | 1011 | struct ceph_msg_data *data = cursor->data; |
fe38a2b6 AE |
1012 | struct ceph_pagelist *pagelist; |
1013 | ||
1014 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
1015 | ||
1016 | pagelist = data->pagelist; | |
1017 | BUG_ON(!pagelist); | |
25aff7c5 AE |
1018 | |
1019 | BUG_ON(cursor->offset + cursor->resid != pagelist->length); | |
fe38a2b6 AE |
1020 | BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE); |
1021 | ||
1022 | /* Advance the cursor offset */ | |
1023 | ||
25aff7c5 | 1024 | cursor->resid -= bytes; |
fe38a2b6 | 1025 | cursor->offset += bytes; |
5df521b1 | 1026 | /* offset of first page in pagelist is always 0 */ |
fe38a2b6 AE |
1027 | if (!bytes || cursor->offset & ~PAGE_MASK) |
1028 | return false; /* more bytes to process in the current page */ | |
1029 | ||
1030 | /* Move on to the next page */ | |
1031 | ||
1032 | BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head)); | |
1033 | cursor->page = list_entry_next(cursor->page, lru); | |
25aff7c5 | 1034 | cursor->last_piece = cursor->resid <= PAGE_SIZE; |
fe38a2b6 AE |
1035 | |
1036 | return true; | |
1037 | } | |
1038 | ||
dd236fcb AE |
1039 | /* |
1040 | * Message data is handled (sent or received) in pieces, where each | |
1041 | * piece resides on a single page. The network layer might not | |
1042 | * consume an entire piece at once. A data item's cursor keeps | |
1043 | * track of which piece is next to process and how much remains to | |
1044 | * be processed in that piece. It also tracks whether the current | |
1045 | * piece is the last one in the data item. | |
1046 | */ | |
ca8b3a69 | 1047 | static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor) |
dd236fcb | 1048 | { |
ca8b3a69 | 1049 | size_t length = cursor->total_resid; |
8ae4f4f5 | 1050 | |
8ae4f4f5 | 1051 | switch (cursor->data->type) { |
dd236fcb | 1052 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1053 | ceph_msg_data_pagelist_cursor_init(cursor, length); |
dd236fcb | 1054 | break; |
e766d7b5 | 1055 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1056 | ceph_msg_data_pages_cursor_init(cursor, length); |
e766d7b5 | 1057 | break; |
dd236fcb AE |
1058 | #ifdef CONFIG_BLOCK |
1059 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1060 | ceph_msg_data_bio_cursor_init(cursor, length); |
6aaa4511 | 1061 | break; |
dd236fcb | 1062 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1063 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1064 | default: |
1065 | /* BUG(); */ | |
1066 | break; | |
1067 | } | |
8ae4f4f5 | 1068 | cursor->need_crc = true; |
dd236fcb AE |
1069 | } |
1070 | ||
ca8b3a69 AE |
1071 | static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length) |
1072 | { | |
1073 | struct ceph_msg_data_cursor *cursor = &msg->cursor; | |
1074 | struct ceph_msg_data *data; | |
1075 | ||
1076 | BUG_ON(!length); | |
1077 | BUG_ON(length > msg->data_length); | |
1078 | BUG_ON(list_empty(&msg->data)); | |
1079 | ||
ca8b3a69 AE |
1080 | cursor->data_head = &msg->data; |
1081 | cursor->total_resid = length; | |
1082 | data = list_first_entry(&msg->data, struct ceph_msg_data, links); | |
1083 | cursor->data = data; | |
1084 | ||
1085 | __ceph_msg_data_cursor_init(cursor); | |
1086 | } | |
1087 | ||
dd236fcb AE |
1088 | /* |
1089 | * Return the page containing the next piece to process for a given | |
1090 | * data item, and supply the page offset and length of that piece. | |
1091 | * Indicate whether this is the last piece in this data item. | |
1092 | */ | |
8ae4f4f5 AE |
1093 | static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor, |
1094 | size_t *page_offset, size_t *length, | |
dd236fcb AE |
1095 | bool *last_piece) |
1096 | { | |
1097 | struct page *page; | |
1098 | ||
8ae4f4f5 | 1099 | switch (cursor->data->type) { |
dd236fcb | 1100 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1101 | page = ceph_msg_data_pagelist_next(cursor, page_offset, length); |
dd236fcb | 1102 | break; |
e766d7b5 | 1103 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1104 | page = ceph_msg_data_pages_next(cursor, page_offset, length); |
e766d7b5 | 1105 | break; |
dd236fcb AE |
1106 | #ifdef CONFIG_BLOCK |
1107 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1108 | page = ceph_msg_data_bio_next(cursor, page_offset, length); |
6aaa4511 | 1109 | break; |
dd236fcb | 1110 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1111 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1112 | default: |
1113 | page = NULL; | |
1114 | break; | |
1115 | } | |
1116 | BUG_ON(!page); | |
1117 | BUG_ON(*page_offset + *length > PAGE_SIZE); | |
1118 | BUG_ON(!*length); | |
1119 | if (last_piece) | |
8ae4f4f5 | 1120 | *last_piece = cursor->last_piece; |
dd236fcb AE |
1121 | |
1122 | return page; | |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * Returns true if the result moves the cursor on to the next piece | |
1127 | * of the data item. | |
1128 | */ | |
8ae4f4f5 AE |
1129 | static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, |
1130 | size_t bytes) | |
dd236fcb AE |
1131 | { |
1132 | bool new_piece; | |
1133 | ||
25aff7c5 | 1134 | BUG_ON(bytes > cursor->resid); |
8ae4f4f5 | 1135 | switch (cursor->data->type) { |
dd236fcb | 1136 | case CEPH_MSG_DATA_PAGELIST: |
8ae4f4f5 | 1137 | new_piece = ceph_msg_data_pagelist_advance(cursor, bytes); |
dd236fcb | 1138 | break; |
e766d7b5 | 1139 | case CEPH_MSG_DATA_PAGES: |
8ae4f4f5 | 1140 | new_piece = ceph_msg_data_pages_advance(cursor, bytes); |
e766d7b5 | 1141 | break; |
dd236fcb AE |
1142 | #ifdef CONFIG_BLOCK |
1143 | case CEPH_MSG_DATA_BIO: | |
8ae4f4f5 | 1144 | new_piece = ceph_msg_data_bio_advance(cursor, bytes); |
6aaa4511 | 1145 | break; |
dd236fcb | 1146 | #endif /* CONFIG_BLOCK */ |
6aaa4511 | 1147 | case CEPH_MSG_DATA_NONE: |
dd236fcb AE |
1148 | default: |
1149 | BUG(); | |
1150 | break; | |
1151 | } | |
ca8b3a69 | 1152 | cursor->total_resid -= bytes; |
dd236fcb | 1153 | |
ca8b3a69 AE |
1154 | if (!cursor->resid && cursor->total_resid) { |
1155 | WARN_ON(!cursor->last_piece); | |
1156 | BUG_ON(list_is_last(&cursor->data->links, cursor->data_head)); | |
1157 | cursor->data = list_entry_next(cursor->data, links); | |
1158 | __ceph_msg_data_cursor_init(cursor); | |
a51b272e | 1159 | new_piece = true; |
ca8b3a69 | 1160 | } |
a51b272e | 1161 | cursor->need_crc = new_piece; |
ca8b3a69 | 1162 | |
dd236fcb AE |
1163 | return new_piece; |
1164 | } | |
1165 | ||
98fa5dd8 | 1166 | static void prepare_message_data(struct ceph_msg *msg, u32 data_len) |
739c905b | 1167 | { |
739c905b | 1168 | BUG_ON(!msg); |
25aff7c5 | 1169 | BUG_ON(!data_len); |
739c905b | 1170 | |
4c59b4a2 | 1171 | /* Initialize data cursor */ |
fe38a2b6 | 1172 | |
8ae4f4f5 | 1173 | ceph_msg_data_cursor_init(msg, (size_t)data_len); |
739c905b AE |
1174 | } |
1175 | ||
31b8006e SW |
1176 | /* |
1177 | * Prepare footer for currently outgoing message, and finish things | |
1178 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
1179 | */ | |
859eb799 | 1180 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
1181 | { |
1182 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 1183 | int v = con->out_kvec_left; |
31b8006e | 1184 | |
fd154f3c AE |
1185 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
1186 | ||
31b8006e SW |
1187 | dout("prepare_write_message_footer %p\n", con); |
1188 | con->out_kvec_is_msg = true; | |
1189 | con->out_kvec[v].iov_base = &m->footer; | |
1190 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
1191 | con->out_kvec_bytes += sizeof(m->footer); | |
1192 | con->out_kvec_left++; | |
1193 | con->out_more = m->more_to_follow; | |
c86a2930 | 1194 | con->out_msg_done = true; |
31b8006e SW |
1195 | } |
1196 | ||
1197 | /* | |
1198 | * Prepare headers for the next outgoing message. | |
1199 | */ | |
1200 | static void prepare_write_message(struct ceph_connection *con) | |
1201 | { | |
1202 | struct ceph_msg *m; | |
a9a0c51a | 1203 | u32 crc; |
31b8006e | 1204 | |
e2200423 | 1205 | con_out_kvec_reset(con); |
31b8006e | 1206 | con->out_kvec_is_msg = true; |
c86a2930 | 1207 | con->out_msg_done = false; |
31b8006e SW |
1208 | |
1209 | /* Sneak an ack in there first? If we can get it into the same | |
1210 | * TCP packet that's a good thing. */ | |
1211 | if (con->in_seq > con->in_seq_acked) { | |
1212 | con->in_seq_acked = con->in_seq; | |
e2200423 | 1213 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 1214 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1215 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 1216 | &con->out_temp_ack); |
31b8006e SW |
1217 | } |
1218 | ||
38941f80 | 1219 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 1220 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 1221 | con->out_msg = m; |
38941f80 | 1222 | BUG_ON(m->con != con); |
4cf9d544 SW |
1223 | |
1224 | /* put message on sent list */ | |
1225 | ceph_msg_get(m); | |
1226 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 1227 | |
e84346b7 SW |
1228 | /* |
1229 | * only assign outgoing seq # if we haven't sent this message | |
1230 | * yet. if it is requeued, resend with it's original seq. | |
1231 | */ | |
1232 | if (m->needs_out_seq) { | |
1233 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
1234 | m->needs_out_seq = false; | |
1235 | } | |
98fa5dd8 | 1236 | WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len)); |
31b8006e | 1237 | |
98fa5dd8 | 1238 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n", |
31b8006e SW |
1239 | m, con->out_seq, le16_to_cpu(m->hdr.type), |
1240 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
98fa5dd8 | 1241 | m->data_length); |
31b8006e SW |
1242 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); |
1243 | ||
1244 | /* tag + hdr + front + middle */ | |
e2200423 AE |
1245 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
1246 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
1247 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 1248 | |
31b8006e | 1249 | if (m->middle) |
e2200423 | 1250 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 1251 | m->middle->vec.iov_base); |
31b8006e SW |
1252 | |
1253 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
1254 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
1255 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 1256 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
1257 | |
1258 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
1259 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
1260 | if (m->middle) { | |
1261 | crc = crc32c(0, m->middle->vec.iov_base, | |
1262 | m->middle->vec.iov_len); | |
1263 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
1264 | } else | |
31b8006e | 1265 | con->out_msg->footer.middle_crc = 0; |
739c905b | 1266 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
1267 | le32_to_cpu(con->out_msg->footer.front_crc), |
1268 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
1269 | ||
1270 | /* is there a data payload? */ | |
739c905b | 1271 | con->out_msg->footer.data_crc = 0; |
98fa5dd8 AE |
1272 | if (m->data_length) { |
1273 | prepare_message_data(con->out_msg, m->data_length); | |
78625051 AE |
1274 | con->out_more = 1; /* data + footer will follow */ |
1275 | } else { | |
31b8006e | 1276 | /* no, queue up footer too and be done */ |
859eb799 | 1277 | prepare_write_message_footer(con); |
78625051 | 1278 | } |
31b8006e | 1279 | |
c9ffc77a | 1280 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1281 | } |
1282 | ||
1283 | /* | |
1284 | * Prepare an ack. | |
1285 | */ | |
1286 | static void prepare_write_ack(struct ceph_connection *con) | |
1287 | { | |
1288 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
1289 | con->in_seq_acked, con->in_seq); | |
1290 | con->in_seq_acked = con->in_seq; | |
1291 | ||
e2200423 | 1292 | con_out_kvec_reset(con); |
859eb799 | 1293 | |
e2200423 | 1294 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 1295 | |
31b8006e | 1296 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 1297 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
1298 | &con->out_temp_ack); |
1299 | ||
31b8006e | 1300 | con->out_more = 1; /* more will follow.. eventually.. */ |
c9ffc77a | 1301 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1302 | } |
1303 | ||
3a23083b SW |
1304 | /* |
1305 | * Prepare to share the seq during handshake | |
1306 | */ | |
1307 | static void prepare_write_seq(struct ceph_connection *con) | |
1308 | { | |
1309 | dout("prepare_write_seq %p %llu -> %llu\n", con, | |
1310 | con->in_seq_acked, con->in_seq); | |
1311 | con->in_seq_acked = con->in_seq; | |
1312 | ||
1313 | con_out_kvec_reset(con); | |
1314 | ||
1315 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); | |
1316 | con_out_kvec_add(con, sizeof (con->out_temp_ack), | |
1317 | &con->out_temp_ack); | |
1318 | ||
1319 | con_flag_set(con, CON_FLAG_WRITE_PENDING); | |
1320 | } | |
1321 | ||
31b8006e SW |
1322 | /* |
1323 | * Prepare to write keepalive byte. | |
1324 | */ | |
1325 | static void prepare_write_keepalive(struct ceph_connection *con) | |
1326 | { | |
1327 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
1328 | con_out_kvec_reset(con); |
1329 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
c9ffc77a | 1330 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1331 | } |
1332 | ||
1333 | /* | |
1334 | * Connection negotiation. | |
1335 | */ | |
1336 | ||
dac1e716 AE |
1337 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
1338 | int *auth_proto) | |
4e7a5dcd | 1339 | { |
a3530df3 | 1340 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
1341 | |
1342 | if (!con->ops->get_authorizer) { | |
1343 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
1344 | con->out_connect.authorizer_len = 0; | |
729796be | 1345 | return NULL; |
b1c6b980 AE |
1346 | } |
1347 | ||
1348 | /* Can't hold the mutex while getting authorizer */ | |
ec302645 | 1349 | mutex_unlock(&con->mutex); |
dac1e716 | 1350 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
ec302645 | 1351 | mutex_lock(&con->mutex); |
4e7a5dcd | 1352 | |
a3530df3 | 1353 | if (IS_ERR(auth)) |
729796be | 1354 | return auth; |
8dacc7da | 1355 | if (con->state != CON_STATE_NEGOTIATING) |
729796be | 1356 | return ERR_PTR(-EAGAIN); |
0da5d703 | 1357 | |
8f43fb53 AE |
1358 | con->auth_reply_buf = auth->authorizer_reply_buf; |
1359 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
729796be | 1360 | return auth; |
4e7a5dcd SW |
1361 | } |
1362 | ||
31b8006e SW |
1363 | /* |
1364 | * We connected to a peer and are saying hello. | |
1365 | */ | |
e825a66d | 1366 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 1367 | { |
e2200423 AE |
1368 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
1369 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 1370 | &con->msgr->my_enc_addr); |
eed0ef2c | 1371 | |
eed0ef2c | 1372 | con->out_more = 0; |
c9ffc77a | 1373 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
eed0ef2c SW |
1374 | } |
1375 | ||
e825a66d | 1376 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 1377 | { |
95c96174 | 1378 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 1379 | int proto; |
dac1e716 | 1380 | int auth_proto; |
729796be | 1381 | struct ceph_auth_handshake *auth; |
31b8006e SW |
1382 | |
1383 | switch (con->peer_name.type) { | |
1384 | case CEPH_ENTITY_TYPE_MON: | |
1385 | proto = CEPH_MONC_PROTOCOL; | |
1386 | break; | |
1387 | case CEPH_ENTITY_TYPE_OSD: | |
1388 | proto = CEPH_OSDC_PROTOCOL; | |
1389 | break; | |
1390 | case CEPH_ENTITY_TYPE_MDS: | |
1391 | proto = CEPH_MDSC_PROTOCOL; | |
1392 | break; | |
1393 | default: | |
1394 | BUG(); | |
1395 | } | |
1396 | ||
1397 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
1398 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 1399 | |
e825a66d | 1400 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
1401 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
1402 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
1403 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
1404 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
1405 | con->out_connect.flags = 0; | |
31b8006e | 1406 | |
dac1e716 AE |
1407 | auth_proto = CEPH_AUTH_UNKNOWN; |
1408 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
1409 | if (IS_ERR(auth)) |
1410 | return PTR_ERR(auth); | |
3da54776 | 1411 | |
dac1e716 | 1412 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
1413 | con->out_connect.authorizer_len = auth ? |
1414 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
1415 | ||
e2200423 | 1416 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
1417 | &con->out_connect); |
1418 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 1419 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 1420 | auth->authorizer_buf); |
859eb799 | 1421 | |
31b8006e | 1422 | con->out_more = 0; |
c9ffc77a | 1423 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
4e7a5dcd | 1424 | |
e10c758e | 1425 | return 0; |
31b8006e SW |
1426 | } |
1427 | ||
31b8006e SW |
1428 | /* |
1429 | * write as much of pending kvecs to the socket as we can. | |
1430 | * 1 -> done | |
1431 | * 0 -> socket full, but more to do | |
1432 | * <0 -> error | |
1433 | */ | |
1434 | static int write_partial_kvec(struct ceph_connection *con) | |
1435 | { | |
1436 | int ret; | |
1437 | ||
1438 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
1439 | while (con->out_kvec_bytes > 0) { | |
1440 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
1441 | con->out_kvec_left, con->out_kvec_bytes, | |
1442 | con->out_more); | |
1443 | if (ret <= 0) | |
1444 | goto out; | |
1445 | con->out_kvec_bytes -= ret; | |
1446 | if (con->out_kvec_bytes == 0) | |
1447 | break; /* done */ | |
f42299e6 AE |
1448 | |
1449 | /* account for full iov entries consumed */ | |
1450 | while (ret >= con->out_kvec_cur->iov_len) { | |
1451 | BUG_ON(!con->out_kvec_left); | |
1452 | ret -= con->out_kvec_cur->iov_len; | |
1453 | con->out_kvec_cur++; | |
1454 | con->out_kvec_left--; | |
1455 | } | |
1456 | /* and for a partially-consumed entry */ | |
1457 | if (ret) { | |
1458 | con->out_kvec_cur->iov_len -= ret; | |
1459 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
1460 | } |
1461 | } | |
1462 | con->out_kvec_left = 0; | |
1463 | con->out_kvec_is_msg = false; | |
1464 | ret = 1; | |
1465 | out: | |
1466 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
1467 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
1468 | return ret; /* done! */ | |
1469 | } | |
1470 | ||
35b62808 AE |
1471 | static u32 ceph_crc32c_page(u32 crc, struct page *page, |
1472 | unsigned int page_offset, | |
1473 | unsigned int length) | |
1474 | { | |
1475 | char *kaddr; | |
1476 | ||
1477 | kaddr = kmap(page); | |
1478 | BUG_ON(kaddr == NULL); | |
1479 | crc = crc32c(crc, kaddr + page_offset, length); | |
1480 | kunmap(page); | |
1481 | ||
1482 | return crc; | |
1483 | } | |
31b8006e SW |
1484 | /* |
1485 | * Write as much message data payload as we can. If we finish, queue | |
1486 | * up the footer. | |
1487 | * 1 -> done, footer is now queued in out_kvec[]. | |
1488 | * 0 -> socket full, but more to do | |
1489 | * <0 -> error | |
1490 | */ | |
34d2d200 | 1491 | static int write_partial_message_data(struct ceph_connection *con) |
31b8006e SW |
1492 | { |
1493 | struct ceph_msg *msg = con->out_msg; | |
8ae4f4f5 | 1494 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
37675b0f | 1495 | bool do_datacrc = !con->msgr->nocrc; |
f5db90bc | 1496 | u32 crc; |
31b8006e | 1497 | |
859a35d5 | 1498 | dout("%s %p msg %p\n", __func__, con, msg); |
31b8006e | 1499 | |
5240d9f9 | 1500 | if (list_empty(&msg->data)) |
4c59b4a2 AE |
1501 | return -EINVAL; |
1502 | ||
5821bd8c AE |
1503 | /* |
1504 | * Iterate through each page that contains data to be | |
1505 | * written, and send as much as possible for each. | |
1506 | * | |
1507 | * If we are calculating the data crc (the default), we will | |
1508 | * need to map the page. If we have no pages, they have | |
1509 | * been revoked, so use the zero page. | |
1510 | */ | |
f5db90bc | 1511 | crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0; |
643c68a4 | 1512 | while (cursor->resid) { |
8a166d05 | 1513 | struct page *page; |
e387d525 AE |
1514 | size_t page_offset; |
1515 | size_t length; | |
8a166d05 | 1516 | bool last_piece; |
8ea299bc | 1517 | bool need_crc; |
f5db90bc | 1518 | int ret; |
68b4476b | 1519 | |
8ae4f4f5 | 1520 | page = ceph_msg_data_next(&msg->cursor, &page_offset, &length, |
4c59b4a2 | 1521 | &last_piece); |
e387d525 | 1522 | ret = ceph_tcp_sendpage(con->sock, page, page_offset, |
fe38a2b6 | 1523 | length, last_piece); |
f5db90bc AE |
1524 | if (ret <= 0) { |
1525 | if (do_datacrc) | |
1526 | msg->footer.data_crc = cpu_to_le32(crc); | |
31b8006e | 1527 | |
f5db90bc AE |
1528 | return ret; |
1529 | } | |
143334ff AE |
1530 | if (do_datacrc && cursor->need_crc) |
1531 | crc = ceph_crc32c_page(crc, page, page_offset, length); | |
8ae4f4f5 | 1532 | need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret); |
31b8006e SW |
1533 | } |
1534 | ||
34d2d200 | 1535 | dout("%s %p msg %p done\n", __func__, con, msg); |
31b8006e SW |
1536 | |
1537 | /* prepare and queue up footer, too */ | |
f5db90bc AE |
1538 | if (do_datacrc) |
1539 | msg->footer.data_crc = cpu_to_le32(crc); | |
1540 | else | |
84ca8fc8 | 1541 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1542 | con_out_kvec_reset(con); |
859eb799 | 1543 | prepare_write_message_footer(con); |
f5db90bc AE |
1544 | |
1545 | return 1; /* must return > 0 to indicate success */ | |
31b8006e SW |
1546 | } |
1547 | ||
1548 | /* | |
1549 | * write some zeros | |
1550 | */ | |
1551 | static int write_partial_skip(struct ceph_connection *con) | |
1552 | { | |
1553 | int ret; | |
1554 | ||
1555 | while (con->out_skip > 0) { | |
31739139 | 1556 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1557 | |
e1dcb128 | 1558 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true); |
31b8006e SW |
1559 | if (ret <= 0) |
1560 | goto out; | |
1561 | con->out_skip -= ret; | |
1562 | } | |
1563 | ret = 1; | |
1564 | out: | |
1565 | return ret; | |
1566 | } | |
1567 | ||
1568 | /* | |
1569 | * Prepare to read connection handshake, or an ack. | |
1570 | */ | |
eed0ef2c SW |
1571 | static void prepare_read_banner(struct ceph_connection *con) |
1572 | { | |
1573 | dout("prepare_read_banner %p\n", con); | |
1574 | con->in_base_pos = 0; | |
1575 | } | |
1576 | ||
31b8006e SW |
1577 | static void prepare_read_connect(struct ceph_connection *con) |
1578 | { | |
1579 | dout("prepare_read_connect %p\n", con); | |
1580 | con->in_base_pos = 0; | |
1581 | } | |
1582 | ||
1583 | static void prepare_read_ack(struct ceph_connection *con) | |
1584 | { | |
1585 | dout("prepare_read_ack %p\n", con); | |
1586 | con->in_base_pos = 0; | |
1587 | } | |
1588 | ||
3a23083b SW |
1589 | static void prepare_read_seq(struct ceph_connection *con) |
1590 | { | |
1591 | dout("prepare_read_seq %p\n", con); | |
1592 | con->in_base_pos = 0; | |
1593 | con->in_tag = CEPH_MSGR_TAG_SEQ; | |
1594 | } | |
1595 | ||
31b8006e SW |
1596 | static void prepare_read_tag(struct ceph_connection *con) |
1597 | { | |
1598 | dout("prepare_read_tag %p\n", con); | |
1599 | con->in_base_pos = 0; | |
1600 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1601 | } | |
1602 | ||
1603 | /* | |
1604 | * Prepare to read a message. | |
1605 | */ | |
1606 | static int prepare_read_message(struct ceph_connection *con) | |
1607 | { | |
1608 | dout("prepare_read_message %p\n", con); | |
1609 | BUG_ON(con->in_msg != NULL); | |
1610 | con->in_base_pos = 0; | |
1611 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1612 | return 0; | |
1613 | } | |
1614 | ||
1615 | ||
1616 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1617 | int end, int size, void *object) |
31b8006e | 1618 | { |
e6cee71f AE |
1619 | while (con->in_base_pos < end) { |
1620 | int left = end - con->in_base_pos; | |
31b8006e SW |
1621 | int have = size - left; |
1622 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1623 | if (ret <= 0) | |
1624 | return ret; | |
1625 | con->in_base_pos += ret; | |
1626 | } | |
1627 | return 1; | |
1628 | } | |
1629 | ||
1630 | ||
1631 | /* | |
1632 | * Read all or part of the connect-side handshake on a new connection | |
1633 | */ | |
eed0ef2c | 1634 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1635 | { |
fd51653f AE |
1636 | int size; |
1637 | int end; | |
1638 | int ret; | |
31b8006e | 1639 | |
eed0ef2c | 1640 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1641 | |
1642 | /* peer's banner */ | |
fd51653f AE |
1643 | size = strlen(CEPH_BANNER); |
1644 | end = size; | |
1645 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1646 | if (ret <= 0) |
1647 | goto out; | |
fd51653f AE |
1648 | |
1649 | size = sizeof (con->actual_peer_addr); | |
1650 | end += size; | |
1651 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1652 | if (ret <= 0) |
1653 | goto out; | |
fd51653f AE |
1654 | |
1655 | size = sizeof (con->peer_addr_for_me); | |
1656 | end += size; | |
1657 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1658 | if (ret <= 0) |
1659 | goto out; | |
fd51653f | 1660 | |
eed0ef2c SW |
1661 | out: |
1662 | return ret; | |
1663 | } | |
1664 | ||
1665 | static int read_partial_connect(struct ceph_connection *con) | |
1666 | { | |
fd51653f AE |
1667 | int size; |
1668 | int end; | |
1669 | int ret; | |
eed0ef2c SW |
1670 | |
1671 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1672 | ||
fd51653f AE |
1673 | size = sizeof (con->in_reply); |
1674 | end = size; | |
1675 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1676 | if (ret <= 0) |
1677 | goto out; | |
fd51653f AE |
1678 | |
1679 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1680 | end += size; | |
1681 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1682 | if (ret <= 0) |
1683 | goto out; | |
31b8006e | 1684 | |
4e7a5dcd SW |
1685 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1686 | con, (int)con->in_reply.tag, | |
1687 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1688 | le32_to_cpu(con->in_reply.global_seq)); |
1689 | out: | |
1690 | return ret; | |
eed0ef2c | 1691 | |
31b8006e SW |
1692 | } |
1693 | ||
1694 | /* | |
1695 | * Verify the hello banner looks okay. | |
1696 | */ | |
1697 | static int verify_hello(struct ceph_connection *con) | |
1698 | { | |
1699 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1700 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1701 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1702 | con->error_msg = "protocol error, bad banner"; |
1703 | return -1; | |
1704 | } | |
1705 | return 0; | |
1706 | } | |
1707 | ||
1708 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1709 | { | |
1710 | switch (ss->ss_family) { | |
1711 | case AF_INET: | |
1712 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1713 | case AF_INET6: | |
1714 | return | |
1715 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1716 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1717 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1718 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1719 | } | |
1720 | return false; | |
1721 | } | |
1722 | ||
1723 | static int addr_port(struct sockaddr_storage *ss) | |
1724 | { | |
1725 | switch (ss->ss_family) { | |
1726 | case AF_INET: | |
f28bcfbe | 1727 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1728 | case AF_INET6: |
f28bcfbe | 1729 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1730 | } |
1731 | return 0; | |
1732 | } | |
1733 | ||
1734 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1735 | { | |
1736 | switch (ss->ss_family) { | |
1737 | case AF_INET: | |
1738 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1739 | break; |
31b8006e SW |
1740 | case AF_INET6: |
1741 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1742 | break; |
31b8006e SW |
1743 | } |
1744 | } | |
1745 | ||
ee3b56f2 NW |
1746 | /* |
1747 | * Unlike other *_pton function semantics, zero indicates success. | |
1748 | */ | |
1749 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1750 | char delim, const char **ipend) | |
1751 | { | |
99f0f3b2 AE |
1752 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1753 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1754 | |
1755 | memset(ss, 0, sizeof(*ss)); | |
1756 | ||
1757 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1758 | ss->ss_family = AF_INET; | |
1759 | return 0; | |
1760 | } | |
1761 | ||
1762 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1763 | ss->ss_family = AF_INET6; | |
1764 | return 0; | |
1765 | } | |
1766 | ||
1767 | return -EINVAL; | |
1768 | } | |
1769 | ||
1770 | /* | |
1771 | * Extract hostname string and resolve using kernel DNS facility. | |
1772 | */ | |
1773 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1774 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1775 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1776 | { | |
1777 | const char *end, *delim_p; | |
1778 | char *colon_p, *ip_addr = NULL; | |
1779 | int ip_len, ret; | |
1780 | ||
1781 | /* | |
1782 | * The end of the hostname occurs immediately preceding the delimiter or | |
1783 | * the port marker (':') where the delimiter takes precedence. | |
1784 | */ | |
1785 | delim_p = memchr(name, delim, namelen); | |
1786 | colon_p = memchr(name, ':', namelen); | |
1787 | ||
1788 | if (delim_p && colon_p) | |
1789 | end = delim_p < colon_p ? delim_p : colon_p; | |
1790 | else if (!delim_p && colon_p) | |
1791 | end = colon_p; | |
1792 | else { | |
1793 | end = delim_p; | |
1794 | if (!end) /* case: hostname:/ */ | |
1795 | end = name + namelen; | |
1796 | } | |
1797 | ||
1798 | if (end <= name) | |
1799 | return -EINVAL; | |
1800 | ||
1801 | /* do dns_resolve upcall */ | |
1802 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1803 | if (ip_len > 0) | |
1804 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1805 | else | |
1806 | ret = -ESRCH; | |
1807 | ||
1808 | kfree(ip_addr); | |
1809 | ||
1810 | *ipend = end; | |
1811 | ||
1812 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1813 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1814 | ||
1815 | return ret; | |
1816 | } | |
1817 | #else | |
1818 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1819 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1820 | { | |
1821 | return -EINVAL; | |
1822 | } | |
1823 | #endif | |
1824 | ||
1825 | /* | |
1826 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1827 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1828 | */ | |
1829 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1830 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1831 | { | |
1832 | int ret; | |
1833 | ||
1834 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1835 | if (ret) | |
1836 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1837 | ||
1838 | return ret; | |
1839 | } | |
1840 | ||
31b8006e SW |
1841 | /* |
1842 | * Parse an ip[:port] list into an addr array. Use the default | |
1843 | * monitor port if a port isn't specified. | |
1844 | */ | |
1845 | int ceph_parse_ips(const char *c, const char *end, | |
1846 | struct ceph_entity_addr *addr, | |
1847 | int max_count, int *count) | |
1848 | { | |
ee3b56f2 | 1849 | int i, ret = -EINVAL; |
31b8006e SW |
1850 | const char *p = c; |
1851 | ||
1852 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1853 | for (i = 0; i < max_count; i++) { | |
1854 | const char *ipend; | |
1855 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1856 | int port; |
39139f64 SW |
1857 | char delim = ','; |
1858 | ||
1859 | if (*p == '[') { | |
1860 | delim = ']'; | |
1861 | p++; | |
1862 | } | |
31b8006e | 1863 | |
ee3b56f2 NW |
1864 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1865 | if (ret) | |
31b8006e | 1866 | goto bad; |
ee3b56f2 NW |
1867 | ret = -EINVAL; |
1868 | ||
31b8006e SW |
1869 | p = ipend; |
1870 | ||
39139f64 SW |
1871 | if (delim == ']') { |
1872 | if (*p != ']') { | |
1873 | dout("missing matching ']'\n"); | |
1874 | goto bad; | |
1875 | } | |
1876 | p++; | |
1877 | } | |
1878 | ||
31b8006e SW |
1879 | /* port? */ |
1880 | if (p < end && *p == ':') { | |
1881 | port = 0; | |
1882 | p++; | |
1883 | while (p < end && *p >= '0' && *p <= '9') { | |
1884 | port = (port * 10) + (*p - '0'); | |
1885 | p++; | |
1886 | } | |
1887 | if (port > 65535 || port == 0) | |
1888 | goto bad; | |
1889 | } else { | |
1890 | port = CEPH_MON_PORT; | |
1891 | } | |
1892 | ||
1893 | addr_set_port(ss, port); | |
1894 | ||
3d14c5d2 | 1895 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1896 | |
1897 | if (p == end) | |
1898 | break; | |
1899 | if (*p != ',') | |
1900 | goto bad; | |
1901 | p++; | |
1902 | } | |
1903 | ||
1904 | if (p != end) | |
1905 | goto bad; | |
1906 | ||
1907 | if (count) | |
1908 | *count = i + 1; | |
1909 | return 0; | |
1910 | ||
1911 | bad: | |
39139f64 | 1912 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1913 | return ret; |
31b8006e | 1914 | } |
3d14c5d2 | 1915 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1916 | |
eed0ef2c | 1917 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1918 | { |
eed0ef2c | 1919 | dout("process_banner on %p\n", con); |
31b8006e SW |
1920 | |
1921 | if (verify_hello(con) < 0) | |
1922 | return -1; | |
1923 | ||
63f2d211 SW |
1924 | ceph_decode_addr(&con->actual_peer_addr); |
1925 | ceph_decode_addr(&con->peer_addr_for_me); | |
1926 | ||
31b8006e SW |
1927 | /* |
1928 | * Make sure the other end is who we wanted. note that the other | |
1929 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1930 | * them the benefit of the doubt. | |
1931 | */ | |
103e2d3a SW |
1932 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1933 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1934 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1935 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1936 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1937 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1938 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1939 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1940 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1941 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1942 | return -1; |
1943 | } | |
1944 | ||
1945 | /* | |
1946 | * did we learn our address? | |
1947 | */ | |
1948 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1949 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1950 | ||
1951 | memcpy(&con->msgr->inst.addr.in_addr, | |
1952 | &con->peer_addr_for_me.in_addr, | |
1953 | sizeof(con->peer_addr_for_me.in_addr)); | |
1954 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1955 | encode_my_addr(con->msgr); |
eed0ef2c | 1956 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1957 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1958 | } |
1959 | ||
eed0ef2c SW |
1960 | return 0; |
1961 | } | |
1962 | ||
1963 | static int process_connect(struct ceph_connection *con) | |
1964 | { | |
3d14c5d2 YS |
1965 | u64 sup_feat = con->msgr->supported_features; |
1966 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1967 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1968 | int ret; |
04a419f9 | 1969 | |
eed0ef2c SW |
1970 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1971 | ||
31b8006e | 1972 | switch (con->in_reply.tag) { |
04a419f9 SW |
1973 | case CEPH_MSGR_TAG_FEATURES: |
1974 | pr_err("%s%lld %s feature set mismatch," | |
1975 | " my %llx < server's %llx, missing %llx\n", | |
1976 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1977 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1978 | sup_feat, server_feat, server_feat & ~sup_feat); |
1979 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 1980 | reset_connection(con); |
04a419f9 SW |
1981 | return -1; |
1982 | ||
31b8006e | 1983 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1984 | pr_err("%s%lld %s protocol version mismatch," |
1985 | " my %d != server's %d\n", | |
1986 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1987 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1988 | le32_to_cpu(con->out_connect.protocol_version), |
1989 | le32_to_cpu(con->in_reply.protocol_version)); | |
1990 | con->error_msg = "protocol version mismatch"; | |
0fa6ebc6 | 1991 | reset_connection(con); |
31b8006e SW |
1992 | return -1; |
1993 | ||
4e7a5dcd SW |
1994 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1995 | con->auth_retry++; | |
1996 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1997 | con->auth_retry); | |
1998 | if (con->auth_retry == 2) { | |
1999 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
2000 | return -1; |
2001 | } | |
6d4221b5 | 2002 | con_out_kvec_reset(con); |
e825a66d | 2003 | ret = prepare_write_connect(con); |
0da5d703 SW |
2004 | if (ret < 0) |
2005 | return ret; | |
63733a0f | 2006 | prepare_read_connect(con); |
4e7a5dcd | 2007 | break; |
31b8006e SW |
2008 | |
2009 | case CEPH_MSGR_TAG_RESETSESSION: | |
2010 | /* | |
2011 | * If we connected with a large connect_seq but the peer | |
2012 | * has no record of a session with us (no connection, or | |
2013 | * connect_seq == 0), they will send RESETSESION to indicate | |
2014 | * that they must have reset their session, and may have | |
2015 | * dropped messages. | |
2016 | */ | |
2017 | dout("process_connect got RESET peer seq %u\n", | |
5bdca4e0 | 2018 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
2019 | pr_err("%s%lld %s connection reset\n", |
2020 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2021 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2022 | reset_connection(con); |
6d4221b5 | 2023 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2024 | ret = prepare_write_connect(con); |
2025 | if (ret < 0) | |
2026 | return ret; | |
31b8006e SW |
2027 | prepare_read_connect(con); |
2028 | ||
2029 | /* Tell ceph about it. */ | |
ec302645 | 2030 | mutex_unlock(&con->mutex); |
31b8006e SW |
2031 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
2032 | if (con->ops->peer_reset) | |
2033 | con->ops->peer_reset(con); | |
ec302645 | 2034 | mutex_lock(&con->mutex); |
8dacc7da | 2035 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 2036 | return -EAGAIN; |
31b8006e SW |
2037 | break; |
2038 | ||
2039 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
2040 | /* | |
2041 | * If we sent a smaller connect_seq than the peer has, try | |
2042 | * again with a larger value. | |
2043 | */ | |
5bdca4e0 | 2044 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 2045 | le32_to_cpu(con->out_connect.connect_seq), |
5bdca4e0 SW |
2046 | le32_to_cpu(con->in_reply.connect_seq)); |
2047 | con->connect_seq = le32_to_cpu(con->in_reply.connect_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 | ||
2055 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
2056 | /* | |
2057 | * If we sent a smaller global_seq than the peer has, try | |
2058 | * again with a larger value. | |
2059 | */ | |
eed0ef2c | 2060 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 2061 | con->peer_global_seq, |
5bdca4e0 | 2062 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 2063 | get_global_seq(con->msgr, |
5bdca4e0 | 2064 | le32_to_cpu(con->in_reply.global_seq)); |
6d4221b5 | 2065 | con_out_kvec_reset(con); |
5a0f8fdd AE |
2066 | ret = prepare_write_connect(con); |
2067 | if (ret < 0) | |
2068 | return ret; | |
31b8006e SW |
2069 | prepare_read_connect(con); |
2070 | break; | |
2071 | ||
3a23083b | 2072 | case CEPH_MSGR_TAG_SEQ: |
31b8006e | 2073 | case CEPH_MSGR_TAG_READY: |
04a419f9 SW |
2074 | if (req_feat & ~server_feat) { |
2075 | pr_err("%s%lld %s protocol feature mismatch," | |
2076 | " my required %llx > server's %llx, need %llx\n", | |
2077 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2078 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
2079 | req_feat, server_feat, req_feat & ~server_feat); |
2080 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 2081 | reset_connection(con); |
04a419f9 SW |
2082 | return -1; |
2083 | } | |
8dacc7da | 2084 | |
122070a2 | 2085 | WARN_ON(con->state != CON_STATE_NEGOTIATING); |
8dacc7da | 2086 | con->state = CON_STATE_OPEN; |
20e55c4c | 2087 | con->auth_retry = 0; /* we authenticated; clear flag */ |
31b8006e SW |
2088 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
2089 | con->connect_seq++; | |
aba558e2 | 2090 | con->peer_features = server_feat; |
31b8006e SW |
2091 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
2092 | con->peer_global_seq, | |
2093 | le32_to_cpu(con->in_reply.connect_seq), | |
2094 | con->connect_seq); | |
2095 | WARN_ON(con->connect_seq != | |
2096 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
2097 | |
2098 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
c9ffc77a | 2099 | con_flag_set(con, CON_FLAG_LOSSYTX); |
92ac41d0 | 2100 | |
85effe18 | 2101 | con->delay = 0; /* reset backoff memory */ |
92ac41d0 | 2102 | |
3a23083b SW |
2103 | if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) { |
2104 | prepare_write_seq(con); | |
2105 | prepare_read_seq(con); | |
2106 | } else { | |
2107 | prepare_read_tag(con); | |
2108 | } | |
31b8006e SW |
2109 | break; |
2110 | ||
2111 | case CEPH_MSGR_TAG_WAIT: | |
2112 | /* | |
2113 | * If there is a connection race (we are opening | |
2114 | * connections to each other), one of us may just have | |
2115 | * to WAIT. This shouldn't happen if we are the | |
2116 | * client. | |
2117 | */ | |
04177882 SW |
2118 | pr_err("process_connect got WAIT as client\n"); |
2119 | con->error_msg = "protocol error, got WAIT as client"; | |
2120 | return -1; | |
31b8006e SW |
2121 | |
2122 | default: | |
2123 | pr_err("connect protocol error, will retry\n"); | |
2124 | con->error_msg = "protocol error, garbage tag during connect"; | |
2125 | return -1; | |
2126 | } | |
2127 | return 0; | |
2128 | } | |
2129 | ||
2130 | ||
2131 | /* | |
2132 | * read (part of) an ack | |
2133 | */ | |
2134 | static int read_partial_ack(struct ceph_connection *con) | |
2135 | { | |
fd51653f AE |
2136 | int size = sizeof (con->in_temp_ack); |
2137 | int end = size; | |
31b8006e | 2138 | |
fd51653f | 2139 | return read_partial(con, end, size, &con->in_temp_ack); |
31b8006e SW |
2140 | } |
2141 | ||
31b8006e SW |
2142 | /* |
2143 | * We can finally discard anything that's been acked. | |
2144 | */ | |
2145 | static void process_ack(struct ceph_connection *con) | |
2146 | { | |
2147 | struct ceph_msg *m; | |
2148 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
2149 | u64 seq; | |
2150 | ||
31b8006e SW |
2151 | while (!list_empty(&con->out_sent)) { |
2152 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
2153 | list_head); | |
2154 | seq = le64_to_cpu(m->hdr.seq); | |
2155 | if (seq > ack) | |
2156 | break; | |
2157 | dout("got ack for seq %llu type %d at %p\n", seq, | |
2158 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 2159 | m->ack_stamp = jiffies; |
31b8006e SW |
2160 | ceph_msg_remove(m); |
2161 | } | |
31b8006e SW |
2162 | prepare_read_tag(con); |
2163 | } | |
2164 | ||
2165 | ||
2450418c | 2166 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
2167 | struct kvec *section, |
2168 | unsigned int sec_len, u32 *crc) | |
2450418c | 2169 | { |
68b4476b | 2170 | int ret, left; |
2450418c YS |
2171 | |
2172 | BUG_ON(!section); | |
2173 | ||
2174 | while (section->iov_len < sec_len) { | |
2175 | BUG_ON(section->iov_base == NULL); | |
2176 | left = sec_len - section->iov_len; | |
2177 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
2178 | section->iov_len, left); | |
2179 | if (ret <= 0) | |
2180 | return ret; | |
2181 | section->iov_len += ret; | |
2450418c | 2182 | } |
fe3ad593 AE |
2183 | if (section->iov_len == sec_len) |
2184 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 2185 | |
2450418c YS |
2186 | return 1; |
2187 | } | |
31b8006e | 2188 | |
34d2d200 AE |
2189 | static int read_partial_msg_data(struct ceph_connection *con) |
2190 | { | |
2191 | struct ceph_msg *msg = con->in_msg; | |
8ae4f4f5 | 2192 | struct ceph_msg_data_cursor *cursor = &msg->cursor; |
34d2d200 | 2193 | const bool do_datacrc = !con->msgr->nocrc; |
686be208 AE |
2194 | struct page *page; |
2195 | size_t page_offset; | |
2196 | size_t length; | |
f5db90bc | 2197 | u32 crc = 0; |
34d2d200 AE |
2198 | int ret; |
2199 | ||
2200 | BUG_ON(!msg); | |
5240d9f9 | 2201 | if (list_empty(&msg->data)) |
4c59b4a2 | 2202 | return -EIO; |
34d2d200 | 2203 | |
f5db90bc AE |
2204 | if (do_datacrc) |
2205 | crc = con->in_data_crc; | |
643c68a4 | 2206 | while (cursor->resid) { |
8ae4f4f5 | 2207 | page = ceph_msg_data_next(&msg->cursor, &page_offset, &length, |
4c59b4a2 | 2208 | NULL); |
686be208 | 2209 | ret = ceph_tcp_recvpage(con->sock, page, page_offset, length); |
f5db90bc AE |
2210 | if (ret <= 0) { |
2211 | if (do_datacrc) | |
2212 | con->in_data_crc = crc; | |
2213 | ||
686be208 | 2214 | return ret; |
f5db90bc | 2215 | } |
686be208 AE |
2216 | |
2217 | if (do_datacrc) | |
f5db90bc | 2218 | crc = ceph_crc32c_page(crc, page, page_offset, ret); |
8ae4f4f5 | 2219 | (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret); |
34d2d200 | 2220 | } |
f5db90bc AE |
2221 | if (do_datacrc) |
2222 | con->in_data_crc = crc; | |
34d2d200 AE |
2223 | |
2224 | return 1; /* must return > 0 to indicate success */ | |
2225 | } | |
2226 | ||
31b8006e SW |
2227 | /* |
2228 | * read (part of) a message. | |
2229 | */ | |
686be208 AE |
2230 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
2231 | ||
31b8006e SW |
2232 | static int read_partial_message(struct ceph_connection *con) |
2233 | { | |
2234 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
2235 | int size; |
2236 | int end; | |
31b8006e | 2237 | int ret; |
95c96174 | 2238 | unsigned int front_len, middle_len, data_len; |
37675b0f | 2239 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 2240 | u64 seq; |
fe3ad593 | 2241 | u32 crc; |
31b8006e SW |
2242 | |
2243 | dout("read_partial_message con %p msg %p\n", con, m); | |
2244 | ||
2245 | /* header */ | |
fd51653f AE |
2246 | size = sizeof (con->in_hdr); |
2247 | end = size; | |
2248 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
2249 | if (ret <= 0) |
2250 | return ret; | |
fe3ad593 AE |
2251 | |
2252 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
2253 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
2254 | pr_err("read_partial_message bad hdr " | |
2255 | " crc %u != expected %u\n", | |
2256 | crc, con->in_hdr.crc); | |
2257 | return -EBADMSG; | |
2258 | } | |
2259 | ||
31b8006e SW |
2260 | front_len = le32_to_cpu(con->in_hdr.front_len); |
2261 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
2262 | return -EIO; | |
2263 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
7b11ba37 | 2264 | if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN) |
31b8006e SW |
2265 | return -EIO; |
2266 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
2267 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
2268 | return -EIO; | |
2269 | ||
ae18756b SW |
2270 | /* verify seq# */ |
2271 | seq = le64_to_cpu(con->in_hdr.seq); | |
2272 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 2273 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 2274 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2275 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
2276 | seq, con->in_seq + 1); |
2277 | con->in_base_pos = -front_len - middle_len - data_len - | |
2278 | sizeof(m->footer); | |
2279 | con->in_tag = CEPH_MSGR_TAG_READY; | |
636a9c8a | 2280 | return 1; |
ae18756b SW |
2281 | } else if ((s64)seq - (s64)con->in_seq > 1) { |
2282 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
2283 | seq, con->in_seq + 1); | |
2284 | con->error_msg = "bad message sequence # for incoming message"; | |
2285 | return -EBADMSG; | |
2286 | } | |
2287 | ||
31b8006e SW |
2288 | /* allocate message? */ |
2289 | if (!con->in_msg) { | |
4740a623 SW |
2290 | int skip = 0; |
2291 | ||
31b8006e | 2292 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
6ebc8b32 | 2293 | front_len, data_len); |
4740a623 SW |
2294 | ret = ceph_con_in_msg_alloc(con, &skip); |
2295 | if (ret < 0) | |
2296 | return ret; | |
f759ebb9 AE |
2297 | |
2298 | BUG_ON(!con->in_msg ^ skip); | |
2299 | if (con->in_msg && data_len > con->in_msg->data_length) { | |
2300 | pr_warning("%s skipping long message (%u > %zd)\n", | |
2301 | __func__, data_len, con->in_msg->data_length); | |
2302 | ceph_msg_put(con->in_msg); | |
2303 | con->in_msg = NULL; | |
2304 | skip = 1; | |
2305 | } | |
2450418c | 2306 | if (skip) { |
31b8006e | 2307 | /* skip this message */ |
a79832f2 | 2308 | dout("alloc_msg said skip message\n"); |
31b8006e SW |
2309 | con->in_base_pos = -front_len - middle_len - data_len - |
2310 | sizeof(m->footer); | |
2311 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2312 | con->in_seq++; |
636a9c8a | 2313 | return 1; |
31b8006e | 2314 | } |
38941f80 | 2315 | |
4740a623 | 2316 | BUG_ON(!con->in_msg); |
38941f80 | 2317 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
2318 | m = con->in_msg; |
2319 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
2320 | if (m->middle) |
2321 | m->middle->vec.iov_len = 0; | |
9d7f0f13 | 2322 | |
78625051 | 2323 | /* prepare for data payload, if any */ |
a4107026 | 2324 | |
78625051 | 2325 | if (data_len) |
98fa5dd8 | 2326 | prepare_message_data(con->in_msg, data_len); |
31b8006e SW |
2327 | } |
2328 | ||
2329 | /* front */ | |
2450418c YS |
2330 | ret = read_partial_message_section(con, &m->front, front_len, |
2331 | &con->in_front_crc); | |
2332 | if (ret <= 0) | |
2333 | return ret; | |
31b8006e SW |
2334 | |
2335 | /* middle */ | |
2450418c | 2336 | if (m->middle) { |
213c99ee SW |
2337 | ret = read_partial_message_section(con, &m->middle->vec, |
2338 | middle_len, | |
2450418c | 2339 | &con->in_middle_crc); |
31b8006e SW |
2340 | if (ret <= 0) |
2341 | return ret; | |
31b8006e SW |
2342 | } |
2343 | ||
2344 | /* (page) data */ | |
34d2d200 AE |
2345 | if (data_len) { |
2346 | ret = read_partial_msg_data(con); | |
2347 | if (ret <= 0) | |
2348 | return ret; | |
31b8006e SW |
2349 | } |
2350 | ||
31b8006e | 2351 | /* footer */ |
fd51653f AE |
2352 | size = sizeof (m->footer); |
2353 | end += size; | |
2354 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
2355 | if (ret <= 0) |
2356 | return ret; | |
2357 | ||
31b8006e SW |
2358 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
2359 | m, front_len, m->footer.front_crc, middle_len, | |
2360 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
2361 | ||
2362 | /* crc ok? */ | |
2363 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
2364 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
2365 | m, con->in_front_crc, m->footer.front_crc); | |
2366 | return -EBADMSG; | |
2367 | } | |
2368 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
2369 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
2370 | m, con->in_middle_crc, m->footer.middle_crc); | |
2371 | return -EBADMSG; | |
2372 | } | |
bca064d2 | 2373 | if (do_datacrc && |
31b8006e SW |
2374 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
2375 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
2376 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
2377 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
2378 | return -EBADMSG; | |
2379 | } | |
2380 | ||
2381 | return 1; /* done! */ | |
2382 | } | |
2383 | ||
2384 | /* | |
2385 | * Process message. This happens in the worker thread. The callback should | |
2386 | * be careful not to do anything that waits on other incoming messages or it | |
2387 | * may deadlock. | |
2388 | */ | |
2389 | static void process_message(struct ceph_connection *con) | |
2390 | { | |
5e095e8b | 2391 | struct ceph_msg *msg; |
31b8006e | 2392 | |
38941f80 AE |
2393 | BUG_ON(con->in_msg->con != con); |
2394 | con->in_msg->con = NULL; | |
5e095e8b | 2395 | msg = con->in_msg; |
31b8006e | 2396 | con->in_msg = NULL; |
36eb71aa | 2397 | con->ops->put(con); |
31b8006e SW |
2398 | |
2399 | /* if first message, set peer_name */ | |
2400 | if (con->peer_name.type == 0) | |
dbad185d | 2401 | con->peer_name = msg->hdr.src; |
31b8006e | 2402 | |
31b8006e | 2403 | con->in_seq++; |
ec302645 | 2404 | mutex_unlock(&con->mutex); |
31b8006e SW |
2405 | |
2406 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
2407 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 2408 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
2409 | le16_to_cpu(msg->hdr.type), |
2410 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2411 | le32_to_cpu(msg->hdr.front_len), | |
2412 | le32_to_cpu(msg->hdr.data_len), | |
2413 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
2414 | con->ops->dispatch(con, msg); | |
ec302645 SW |
2415 | |
2416 | mutex_lock(&con->mutex); | |
31b8006e SW |
2417 | } |
2418 | ||
2419 | ||
2420 | /* | |
2421 | * Write something to the socket. Called in a worker thread when the | |
2422 | * socket appears to be writeable and we have something ready to send. | |
2423 | */ | |
2424 | static int try_write(struct ceph_connection *con) | |
2425 | { | |
31b8006e SW |
2426 | int ret = 1; |
2427 | ||
d59315ca | 2428 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 2429 | |
31b8006e SW |
2430 | more: |
2431 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
2432 | ||
2433 | /* open the socket first? */ | |
8dacc7da SW |
2434 | if (con->state == CON_STATE_PREOPEN) { |
2435 | BUG_ON(con->sock); | |
2436 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2437 | |
e2200423 | 2438 | con_out_kvec_reset(con); |
e825a66d | 2439 | prepare_write_banner(con); |
eed0ef2c | 2440 | prepare_read_banner(con); |
31b8006e | 2441 | |
cf3e5c40 | 2442 | BUG_ON(con->in_msg); |
31b8006e SW |
2443 | con->in_tag = CEPH_MSGR_TAG_READY; |
2444 | dout("try_write initiating connect on %p new state %lu\n", | |
2445 | con, con->state); | |
41617d0c AE |
2446 | ret = ceph_tcp_connect(con); |
2447 | if (ret < 0) { | |
31b8006e | 2448 | con->error_msg = "connect error"; |
31b8006e SW |
2449 | goto out; |
2450 | } | |
2451 | } | |
2452 | ||
2453 | more_kvec: | |
2454 | /* kvec data queued? */ | |
2455 | if (con->out_skip) { | |
2456 | ret = write_partial_skip(con); | |
2457 | if (ret <= 0) | |
42961d23 | 2458 | goto out; |
31b8006e SW |
2459 | } |
2460 | if (con->out_kvec_left) { | |
2461 | ret = write_partial_kvec(con); | |
2462 | if (ret <= 0) | |
42961d23 | 2463 | goto out; |
31b8006e SW |
2464 | } |
2465 | ||
2466 | /* msg pages? */ | |
2467 | if (con->out_msg) { | |
c86a2930 SW |
2468 | if (con->out_msg_done) { |
2469 | ceph_msg_put(con->out_msg); | |
2470 | con->out_msg = NULL; /* we're done with this one */ | |
2471 | goto do_next; | |
2472 | } | |
2473 | ||
34d2d200 | 2474 | ret = write_partial_message_data(con); |
31b8006e SW |
2475 | if (ret == 1) |
2476 | goto more_kvec; /* we need to send the footer, too! */ | |
2477 | if (ret == 0) | |
42961d23 | 2478 | goto out; |
31b8006e | 2479 | if (ret < 0) { |
34d2d200 | 2480 | dout("try_write write_partial_message_data err %d\n", |
31b8006e | 2481 | ret); |
42961d23 | 2482 | goto out; |
31b8006e SW |
2483 | } |
2484 | } | |
2485 | ||
c86a2930 | 2486 | do_next: |
8dacc7da | 2487 | if (con->state == CON_STATE_OPEN) { |
31b8006e SW |
2488 | /* is anything else pending? */ |
2489 | if (!list_empty(&con->out_queue)) { | |
2490 | prepare_write_message(con); | |
2491 | goto more; | |
2492 | } | |
2493 | if (con->in_seq > con->in_seq_acked) { | |
2494 | prepare_write_ack(con); | |
2495 | goto more; | |
2496 | } | |
c9ffc77a | 2497 | if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) { |
31b8006e SW |
2498 | prepare_write_keepalive(con); |
2499 | goto more; | |
2500 | } | |
2501 | } | |
2502 | ||
2503 | /* Nothing to do! */ | |
c9ffc77a | 2504 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
31b8006e | 2505 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2506 | ret = 0; |
2507 | out: | |
42961d23 | 2508 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2509 | return ret; |
2510 | } | |
2511 | ||
2512 | ||
2513 | ||
2514 | /* | |
2515 | * Read what we can from the socket. | |
2516 | */ | |
2517 | static int try_read(struct ceph_connection *con) | |
2518 | { | |
31b8006e SW |
2519 | int ret = -1; |
2520 | ||
8dacc7da SW |
2521 | more: |
2522 | dout("try_read start on %p state %lu\n", con, con->state); | |
2523 | if (con->state != CON_STATE_CONNECTING && | |
2524 | con->state != CON_STATE_NEGOTIATING && | |
2525 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2526 | return 0; |
2527 | ||
8dacc7da | 2528 | BUG_ON(!con->sock); |
ec302645 | 2529 | |
31b8006e SW |
2530 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2531 | con->in_base_pos); | |
0da5d703 | 2532 | |
8dacc7da | 2533 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2534 | dout("try_read connecting\n"); |
2535 | ret = read_partial_banner(con); | |
2536 | if (ret <= 0) | |
ab166d5a | 2537 | goto out; |
7593af92 AE |
2538 | ret = process_banner(con); |
2539 | if (ret < 0) | |
2540 | goto out; | |
2541 | ||
8dacc7da | 2542 | con->state = CON_STATE_NEGOTIATING; |
7593af92 | 2543 | |
6d4221b5 JS |
2544 | /* |
2545 | * Received banner is good, exchange connection info. | |
2546 | * Do not reset out_kvec, as sending our banner raced | |
2547 | * with receiving peer banner after connect completed. | |
2548 | */ | |
7593af92 AE |
2549 | ret = prepare_write_connect(con); |
2550 | if (ret < 0) | |
2551 | goto out; | |
2552 | prepare_read_connect(con); | |
2553 | ||
2554 | /* Send connection info before awaiting response */ | |
0da5d703 SW |
2555 | goto out; |
2556 | } | |
2557 | ||
8dacc7da | 2558 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2559 | dout("try_read negotiating\n"); |
31b8006e SW |
2560 | ret = read_partial_connect(con); |
2561 | if (ret <= 0) | |
31b8006e | 2562 | goto out; |
98bdb0aa SW |
2563 | ret = process_connect(con); |
2564 | if (ret < 0) | |
2565 | goto out; | |
31b8006e SW |
2566 | goto more; |
2567 | } | |
2568 | ||
122070a2 | 2569 | WARN_ON(con->state != CON_STATE_OPEN); |
8dacc7da | 2570 | |
31b8006e SW |
2571 | if (con->in_base_pos < 0) { |
2572 | /* | |
2573 | * skipping + discarding content. | |
2574 | * | |
2575 | * FIXME: there must be a better way to do this! | |
2576 | */ | |
84495f49 AE |
2577 | static char buf[SKIP_BUF_SIZE]; |
2578 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2579 | ||
31b8006e SW |
2580 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2581 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2582 | if (ret <= 0) | |
98bdb0aa | 2583 | goto out; |
31b8006e SW |
2584 | con->in_base_pos += ret; |
2585 | if (con->in_base_pos) | |
2586 | goto more; | |
2587 | } | |
2588 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2589 | /* | |
2590 | * what's next? | |
2591 | */ | |
2592 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2593 | if (ret <= 0) | |
98bdb0aa | 2594 | goto out; |
31b8006e SW |
2595 | dout("try_read got tag %d\n", (int)con->in_tag); |
2596 | switch (con->in_tag) { | |
2597 | case CEPH_MSGR_TAG_MSG: | |
2598 | prepare_read_message(con); | |
2599 | break; | |
2600 | case CEPH_MSGR_TAG_ACK: | |
2601 | prepare_read_ack(con); | |
2602 | break; | |
2603 | case CEPH_MSGR_TAG_CLOSE: | |
8dacc7da SW |
2604 | con_close_socket(con); |
2605 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2606 | goto out; |
31b8006e SW |
2607 | default: |
2608 | goto bad_tag; | |
2609 | } | |
2610 | } | |
2611 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2612 | ret = read_partial_message(con); | |
2613 | if (ret <= 0) { | |
2614 | switch (ret) { | |
2615 | case -EBADMSG: | |
2616 | con->error_msg = "bad crc"; | |
2617 | ret = -EIO; | |
98bdb0aa | 2618 | break; |
31b8006e SW |
2619 | case -EIO: |
2620 | con->error_msg = "io error"; | |
98bdb0aa | 2621 | break; |
31b8006e | 2622 | } |
98bdb0aa | 2623 | goto out; |
31b8006e SW |
2624 | } |
2625 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2626 | goto more; | |
2627 | process_message(con); | |
7b862e07 SW |
2628 | if (con->state == CON_STATE_OPEN) |
2629 | prepare_read_tag(con); | |
31b8006e SW |
2630 | goto more; |
2631 | } | |
3a23083b SW |
2632 | if (con->in_tag == CEPH_MSGR_TAG_ACK || |
2633 | con->in_tag == CEPH_MSGR_TAG_SEQ) { | |
2634 | /* | |
2635 | * the final handshake seq exchange is semantically | |
2636 | * equivalent to an ACK | |
2637 | */ | |
31b8006e SW |
2638 | ret = read_partial_ack(con); |
2639 | if (ret <= 0) | |
98bdb0aa | 2640 | goto out; |
31b8006e SW |
2641 | process_ack(con); |
2642 | goto more; | |
2643 | } | |
2644 | ||
31b8006e | 2645 | out: |
98bdb0aa | 2646 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2647 | return ret; |
2648 | ||
2649 | bad_tag: | |
2650 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2651 | con->error_msg = "protocol error, garbage tag"; | |
2652 | ret = -1; | |
2653 | goto out; | |
2654 | } | |
2655 | ||
2656 | ||
2657 | /* | |
802c6d96 AE |
2658 | * Atomically queue work on a connection after the specified delay. |
2659 | * Bump @con reference to avoid races with connection teardown. | |
2660 | * Returns 0 if work was queued, or an error code otherwise. | |
31b8006e | 2661 | */ |
802c6d96 | 2662 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
31b8006e | 2663 | { |
31b8006e | 2664 | if (!con->ops->get(con)) { |
802c6d96 AE |
2665 | dout("%s %p ref count 0\n", __func__, con); |
2666 | ||
2667 | return -ENOENT; | |
31b8006e SW |
2668 | } |
2669 | ||
802c6d96 AE |
2670 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
2671 | dout("%s %p - already queued\n", __func__, con); | |
31b8006e | 2672 | con->ops->put(con); |
802c6d96 AE |
2673 | |
2674 | return -EBUSY; | |
31b8006e | 2675 | } |
802c6d96 AE |
2676 | |
2677 | dout("%s %p %lu\n", __func__, con, delay); | |
2678 | ||
2679 | return 0; | |
2680 | } | |
2681 | ||
2682 | static void queue_con(struct ceph_connection *con) | |
2683 | { | |
2684 | (void) queue_con_delay(con, 0); | |
31b8006e SW |
2685 | } |
2686 | ||
7bb21d68 AE |
2687 | static bool con_sock_closed(struct ceph_connection *con) |
2688 | { | |
c9ffc77a | 2689 | if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED)) |
7bb21d68 AE |
2690 | return false; |
2691 | ||
2692 | #define CASE(x) \ | |
2693 | case CON_STATE_ ## x: \ | |
2694 | con->error_msg = "socket closed (con state " #x ")"; \ | |
2695 | break; | |
2696 | ||
2697 | switch (con->state) { | |
2698 | CASE(CLOSED); | |
2699 | CASE(PREOPEN); | |
2700 | CASE(CONNECTING); | |
2701 | CASE(NEGOTIATING); | |
2702 | CASE(OPEN); | |
2703 | CASE(STANDBY); | |
2704 | default: | |
2705 | pr_warning("%s con %p unrecognized state %lu\n", | |
2706 | __func__, con, con->state); | |
2707 | con->error_msg = "unrecognized con state"; | |
2708 | BUG(); | |
2709 | break; | |
2710 | } | |
2711 | #undef CASE | |
2712 | ||
2713 | return true; | |
2714 | } | |
2715 | ||
f20a39fd AE |
2716 | static bool con_backoff(struct ceph_connection *con) |
2717 | { | |
2718 | int ret; | |
2719 | ||
2720 | if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF)) | |
2721 | return false; | |
2722 | ||
2723 | ret = queue_con_delay(con, round_jiffies_relative(con->delay)); | |
2724 | if (ret) { | |
2725 | dout("%s: con %p FAILED to back off %lu\n", __func__, | |
2726 | con, con->delay); | |
2727 | BUG_ON(ret == -ENOENT); | |
2728 | con_flag_set(con, CON_FLAG_BACKOFF); | |
2729 | } | |
2730 | ||
2731 | return true; | |
2732 | } | |
2733 | ||
93209264 AE |
2734 | /* Finish fault handling; con->mutex must *not* be held here */ |
2735 | ||
2736 | static void con_fault_finish(struct ceph_connection *con) | |
2737 | { | |
2738 | /* | |
2739 | * in case we faulted due to authentication, invalidate our | |
2740 | * current tickets so that we can get new ones. | |
2741 | */ | |
2742 | if (con->auth_retry && con->ops->invalidate_authorizer) { | |
2743 | dout("calling invalidate_authorizer()\n"); | |
2744 | con->ops->invalidate_authorizer(con); | |
2745 | } | |
2746 | ||
2747 | if (con->ops->fault) | |
2748 | con->ops->fault(con); | |
2749 | } | |
2750 | ||
31b8006e SW |
2751 | /* |
2752 | * Do some work on a connection. Drop a connection ref when we're done. | |
2753 | */ | |
2754 | static void con_work(struct work_struct *work) | |
2755 | { | |
2756 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2757 | work.work); | |
49659416 | 2758 | bool fault; |
31b8006e | 2759 | |
9dd4658d | 2760 | mutex_lock(&con->mutex); |
49659416 AE |
2761 | while (true) { |
2762 | int ret; | |
31b8006e | 2763 | |
49659416 AE |
2764 | if ((fault = con_sock_closed(con))) { |
2765 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); | |
2766 | break; | |
2767 | } | |
2768 | if (con_backoff(con)) { | |
2769 | dout("%s: con %p BACKOFF\n", __func__, con); | |
2770 | break; | |
2771 | } | |
2772 | if (con->state == CON_STATE_STANDBY) { | |
2773 | dout("%s: con %p STANDBY\n", __func__, con); | |
2774 | break; | |
2775 | } | |
2776 | if (con->state == CON_STATE_CLOSED) { | |
2777 | dout("%s: con %p CLOSED\n", __func__, con); | |
2778 | BUG_ON(con->sock); | |
2779 | break; | |
2780 | } | |
2781 | if (con->state == CON_STATE_PREOPEN) { | |
2782 | dout("%s: con %p PREOPEN\n", __func__, con); | |
2783 | BUG_ON(con->sock); | |
2784 | } | |
0da5d703 | 2785 | |
49659416 AE |
2786 | ret = try_read(con); |
2787 | if (ret < 0) { | |
2788 | if (ret == -EAGAIN) | |
2789 | continue; | |
2790 | con->error_msg = "socket error on read"; | |
2791 | fault = true; | |
2792 | break; | |
2793 | } | |
2794 | ||
2795 | ret = try_write(con); | |
2796 | if (ret < 0) { | |
2797 | if (ret == -EAGAIN) | |
2798 | continue; | |
2799 | con->error_msg = "socket error on write"; | |
2800 | fault = true; | |
2801 | } | |
2802 | ||
2803 | break; /* If we make it to here, we're done */ | |
3a140a0d | 2804 | } |
b6e7b6a1 AE |
2805 | if (fault) |
2806 | con_fault(con); | |
9dd4658d | 2807 | mutex_unlock(&con->mutex); |
0da5d703 | 2808 | |
b6e7b6a1 AE |
2809 | if (fault) |
2810 | con_fault_finish(con); | |
2811 | ||
2812 | con->ops->put(con); | |
31b8006e SW |
2813 | } |
2814 | ||
31b8006e SW |
2815 | /* |
2816 | * Generic error/fault handler. A retry mechanism is used with | |
2817 | * exponential backoff | |
2818 | */ | |
93209264 | 2819 | static void con_fault(struct ceph_connection *con) |
31b8006e | 2820 | { |
28362986 | 2821 | pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2822 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2823 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2824 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2825 | |
122070a2 | 2826 | WARN_ON(con->state != CON_STATE_CONNECTING && |
8dacc7da SW |
2827 | con->state != CON_STATE_NEGOTIATING && |
2828 | con->state != CON_STATE_OPEN); | |
ec302645 | 2829 | |
31b8006e | 2830 | con_close_socket(con); |
5e095e8b | 2831 | |
c9ffc77a | 2832 | if (con_flag_test(con, CON_FLAG_LOSSYTX)) { |
8dacc7da SW |
2833 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
2834 | con->state = CON_STATE_CLOSED; | |
93209264 | 2835 | return; |
3b5ede07 SW |
2836 | } |
2837 | ||
5e095e8b | 2838 | if (con->in_msg) { |
38941f80 AE |
2839 | BUG_ON(con->in_msg->con != con); |
2840 | con->in_msg->con = NULL; | |
5e095e8b SW |
2841 | ceph_msg_put(con->in_msg); |
2842 | con->in_msg = NULL; | |
36eb71aa | 2843 | con->ops->put(con); |
5e095e8b | 2844 | } |
31b8006e | 2845 | |
e80a52d1 SW |
2846 | /* Requeue anything that hasn't been acked */ |
2847 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2848 | |
e76661d0 SW |
2849 | /* If there are no messages queued or keepalive pending, place |
2850 | * the connection in a STANDBY state */ | |
2851 | if (list_empty(&con->out_queue) && | |
c9ffc77a | 2852 | !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) { |
e00de341 | 2853 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
c9ffc77a | 2854 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
8dacc7da | 2855 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
2856 | } else { |
2857 | /* retry after a delay. */ | |
8dacc7da | 2858 | con->state = CON_STATE_PREOPEN; |
e80a52d1 SW |
2859 | if (con->delay == 0) |
2860 | con->delay = BASE_DELAY_INTERVAL; | |
2861 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2862 | con->delay *= 2; | |
c9ffc77a | 2863 | con_flag_set(con, CON_FLAG_BACKOFF); |
8618e30b | 2864 | queue_con(con); |
31b8006e | 2865 | } |
31b8006e SW |
2866 | } |
2867 | ||
2868 | ||
2869 | ||
2870 | /* | |
15d9882c | 2871 | * initialize a new messenger instance |
31b8006e | 2872 | */ |
15d9882c AE |
2873 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2874 | struct ceph_entity_addr *myaddr, | |
2875 | u32 supported_features, | |
2876 | u32 required_features, | |
2877 | bool nocrc) | |
31b8006e | 2878 | { |
3d14c5d2 YS |
2879 | msgr->supported_features = supported_features; |
2880 | msgr->required_features = required_features; | |
2881 | ||
31b8006e SW |
2882 | spin_lock_init(&msgr->global_seq_lock); |
2883 | ||
31b8006e SW |
2884 | if (myaddr) |
2885 | msgr->inst.addr = *myaddr; | |
2886 | ||
2887 | /* select a random nonce */ | |
ac8839d7 | 2888 | msgr->inst.addr.type = 0; |
103e2d3a | 2889 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2890 | encode_my_addr(msgr); |
15d9882c | 2891 | msgr->nocrc = nocrc; |
31b8006e | 2892 | |
a2a32584 | 2893 | atomic_set(&msgr->stopping, 0); |
31b8006e | 2894 | |
15d9882c | 2895 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2896 | } |
15d9882c | 2897 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2898 | |
e00de341 SW |
2899 | static void clear_standby(struct ceph_connection *con) |
2900 | { | |
2901 | /* come back from STANDBY? */ | |
8dacc7da | 2902 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 2903 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 2904 | con->state = CON_STATE_PREOPEN; |
e00de341 | 2905 | con->connect_seq++; |
c9ffc77a AE |
2906 | WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING)); |
2907 | WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)); | |
e00de341 SW |
2908 | } |
2909 | } | |
2910 | ||
31b8006e SW |
2911 | /* |
2912 | * Queue up an outgoing message on the given connection. | |
2913 | */ | |
2914 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2915 | { | |
31b8006e | 2916 | /* set src+dst */ |
dbad185d | 2917 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 2918 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
2919 | msg->needs_out_seq = true; |
2920 | ||
ec302645 | 2921 | mutex_lock(&con->mutex); |
92ce034b | 2922 | |
8dacc7da | 2923 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
2924 | dout("con_send %p closed, dropping %p\n", con, msg); |
2925 | ceph_msg_put(msg); | |
2926 | mutex_unlock(&con->mutex); | |
2927 | return; | |
2928 | } | |
2929 | ||
38941f80 | 2930 | BUG_ON(msg->con != NULL); |
36eb71aa | 2931 | msg->con = con->ops->get(con); |
92ce034b AE |
2932 | BUG_ON(msg->con == NULL); |
2933 | ||
31b8006e SW |
2934 | BUG_ON(!list_empty(&msg->list_head)); |
2935 | list_add_tail(&msg->list_head, &con->out_queue); | |
2936 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2937 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2938 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2939 | le32_to_cpu(msg->hdr.front_len), | |
2940 | le32_to_cpu(msg->hdr.middle_len), | |
2941 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
2942 | |
2943 | clear_standby(con); | |
ec302645 | 2944 | mutex_unlock(&con->mutex); |
31b8006e SW |
2945 | |
2946 | /* if there wasn't anything waiting to send before, queue | |
2947 | * new work */ | |
c9ffc77a | 2948 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
31b8006e SW |
2949 | queue_con(con); |
2950 | } | |
3d14c5d2 | 2951 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2952 | |
2953 | /* | |
2954 | * Revoke a message that was previously queued for send | |
2955 | */ | |
6740a845 | 2956 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2957 | { |
6740a845 AE |
2958 | struct ceph_connection *con = msg->con; |
2959 | ||
2960 | if (!con) | |
2961 | return; /* Message not in our possession */ | |
2962 | ||
ec302645 | 2963 | mutex_lock(&con->mutex); |
31b8006e | 2964 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2965 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2966 | list_del_init(&msg->list_head); |
38941f80 | 2967 | BUG_ON(msg->con == NULL); |
36eb71aa | 2968 | msg->con->ops->put(msg->con); |
38941f80 | 2969 | msg->con = NULL; |
31b8006e | 2970 | msg->hdr.seq = 0; |
38941f80 | 2971 | |
31b8006e | 2972 | ceph_msg_put(msg); |
ed98adad SW |
2973 | } |
2974 | if (con->out_msg == msg) { | |
38941f80 | 2975 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2976 | con->out_msg = NULL; |
31b8006e SW |
2977 | if (con->out_kvec_is_msg) { |
2978 | con->out_skip = con->out_kvec_bytes; | |
2979 | con->out_kvec_is_msg = false; | |
2980 | } | |
ed98adad | 2981 | msg->hdr.seq = 0; |
92ce034b AE |
2982 | |
2983 | ceph_msg_put(msg); | |
31b8006e | 2984 | } |
ec302645 | 2985 | mutex_unlock(&con->mutex); |
31b8006e SW |
2986 | } |
2987 | ||
350b1c32 | 2988 | /* |
0d59ab81 | 2989 | * Revoke a message that we may be reading data into |
350b1c32 | 2990 | */ |
8921d114 | 2991 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2992 | { |
8921d114 AE |
2993 | struct ceph_connection *con; |
2994 | ||
2995 | BUG_ON(msg == NULL); | |
2996 | if (!msg->con) { | |
2997 | dout("%s msg %p null con\n", __func__, msg); | |
2998 | ||
2999 | return; /* Message not in our possession */ | |
3000 | } | |
3001 | ||
3002 | con = msg->con; | |
350b1c32 | 3003 | mutex_lock(&con->mutex); |
8921d114 | 3004 | if (con->in_msg == msg) { |
95c96174 ED |
3005 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
3006 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
3007 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
3008 | |
3009 | /* skip rest of message */ | |
8921d114 AE |
3010 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
3011 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 3012 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
3013 | front_len - |
3014 | middle_len - | |
3015 | data_len - | |
350b1c32 | 3016 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
3017 | ceph_msg_put(con->in_msg); |
3018 | con->in_msg = NULL; | |
3019 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 3020 | con->in_seq++; |
350b1c32 | 3021 | } else { |
8921d114 AE |
3022 | dout("%s %p in_msg %p msg %p no-op\n", |
3023 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
3024 | } |
3025 | mutex_unlock(&con->mutex); | |
3026 | } | |
3027 | ||
31b8006e SW |
3028 | /* |
3029 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
3030 | */ | |
3031 | void ceph_con_keepalive(struct ceph_connection *con) | |
3032 | { | |
e00de341 | 3033 | dout("con_keepalive %p\n", con); |
00650931 | 3034 | mutex_lock(&con->mutex); |
e00de341 | 3035 | clear_standby(con); |
00650931 | 3036 | mutex_unlock(&con->mutex); |
c9ffc77a AE |
3037 | if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 && |
3038 | con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) | |
31b8006e SW |
3039 | queue_con(con); |
3040 | } | |
3d14c5d2 | 3041 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e | 3042 | |
6644ed7b | 3043 | static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type) |
43794509 | 3044 | { |
6644ed7b AE |
3045 | struct ceph_msg_data *data; |
3046 | ||
3047 | if (WARN_ON(!ceph_msg_data_type_valid(type))) | |
3048 | return NULL; | |
3049 | ||
81b36be4 | 3050 | data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS); |
6644ed7b AE |
3051 | if (data) |
3052 | data->type = type; | |
5240d9f9 | 3053 | INIT_LIST_HEAD(&data->links); |
6644ed7b AE |
3054 | |
3055 | return data; | |
3056 | } | |
3057 | ||
3058 | static void ceph_msg_data_destroy(struct ceph_msg_data *data) | |
3059 | { | |
3060 | if (!data) | |
3061 | return; | |
3062 | ||
5240d9f9 | 3063 | WARN_ON(!list_empty(&data->links)); |
6644ed7b AE |
3064 | if (data->type == CEPH_MSG_DATA_PAGELIST) { |
3065 | ceph_pagelist_release(data->pagelist); | |
3066 | kfree(data->pagelist); | |
3067 | } | |
81b36be4 | 3068 | kmem_cache_free(ceph_msg_data_cache, data); |
43794509 AE |
3069 | } |
3070 | ||
90af3602 | 3071 | void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages, |
f1baeb2b | 3072 | size_t length, size_t alignment) |
02afca6c | 3073 | { |
6644ed7b AE |
3074 | struct ceph_msg_data *data; |
3075 | ||
07aa1558 AE |
3076 | BUG_ON(!pages); |
3077 | BUG_ON(!length); | |
6644ed7b AE |
3078 | |
3079 | data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES); | |
3080 | BUG_ON(!data); | |
3081 | data->pages = pages; | |
3082 | data->length = length; | |
3083 | data->alignment = alignment & ~PAGE_MASK; | |
02afca6c | 3084 | |
5240d9f9 AE |
3085 | list_add_tail(&data->links, &msg->data); |
3086 | msg->data_length += length; | |
02afca6c | 3087 | } |
90af3602 | 3088 | EXPORT_SYMBOL(ceph_msg_data_add_pages); |
31b8006e | 3089 | |
90af3602 | 3090 | void ceph_msg_data_add_pagelist(struct ceph_msg *msg, |
27fa8385 AE |
3091 | struct ceph_pagelist *pagelist) |
3092 | { | |
6644ed7b AE |
3093 | struct ceph_msg_data *data; |
3094 | ||
07aa1558 AE |
3095 | BUG_ON(!pagelist); |
3096 | BUG_ON(!pagelist->length); | |
27fa8385 | 3097 | |
6644ed7b AE |
3098 | data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST); |
3099 | BUG_ON(!data); | |
3100 | data->pagelist = pagelist; | |
3101 | ||
5240d9f9 AE |
3102 | list_add_tail(&data->links, &msg->data); |
3103 | msg->data_length += pagelist->length; | |
27fa8385 | 3104 | } |
90af3602 | 3105 | EXPORT_SYMBOL(ceph_msg_data_add_pagelist); |
27fa8385 | 3106 | |
ea96571f | 3107 | #ifdef CONFIG_BLOCK |
90af3602 | 3108 | void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio, |
a1930804 | 3109 | size_t length) |
27fa8385 | 3110 | { |
6644ed7b AE |
3111 | struct ceph_msg_data *data; |
3112 | ||
07aa1558 | 3113 | BUG_ON(!bio); |
27fa8385 | 3114 | |
6644ed7b AE |
3115 | data = ceph_msg_data_create(CEPH_MSG_DATA_BIO); |
3116 | BUG_ON(!data); | |
3117 | data->bio = bio; | |
c851c495 | 3118 | data->bio_length = length; |
6644ed7b | 3119 | |
5240d9f9 AE |
3120 | list_add_tail(&data->links, &msg->data); |
3121 | msg->data_length += length; | |
27fa8385 | 3122 | } |
90af3602 | 3123 | EXPORT_SYMBOL(ceph_msg_data_add_bio); |
ea96571f | 3124 | #endif /* CONFIG_BLOCK */ |
27fa8385 | 3125 | |
31b8006e SW |
3126 | /* |
3127 | * construct a new message with given type, size | |
3128 | * the new msg has a ref count of 1. | |
3129 | */ | |
b61c2763 SW |
3130 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
3131 | bool can_fail) | |
31b8006e SW |
3132 | { |
3133 | struct ceph_msg *m; | |
3134 | ||
e3d5d638 | 3135 | m = kmem_cache_zalloc(ceph_msg_cache, flags); |
31b8006e SW |
3136 | if (m == NULL) |
3137 | goto out; | |
31b8006e SW |
3138 | |
3139 | m->hdr.type = cpu_to_le16(type); | |
45c6ceb5 | 3140 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
31b8006e | 3141 | m->hdr.front_len = cpu_to_le32(front_len); |
ca20892d | 3142 | |
9516e45b AE |
3143 | INIT_LIST_HEAD(&m->list_head); |
3144 | kref_init(&m->kref); | |
5240d9f9 | 3145 | INIT_LIST_HEAD(&m->data); |
ca20892d | 3146 | |
31b8006e | 3147 | /* front */ |
842a5780 | 3148 | m->front_alloc_len = front_len; |
31b8006e SW |
3149 | if (front_len) { |
3150 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 3151 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
3152 | PAGE_KERNEL); |
3153 | m->front_is_vmalloc = true; | |
3154 | } else { | |
34d23762 | 3155 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
3156 | } |
3157 | if (m->front.iov_base == NULL) { | |
b61c2763 | 3158 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
3159 | front_len); |
3160 | goto out2; | |
3161 | } | |
3162 | } else { | |
3163 | m->front.iov_base = NULL; | |
3164 | } | |
3165 | m->front.iov_len = front_len; | |
3166 | ||
bb257664 | 3167 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
3168 | return m; |
3169 | ||
3170 | out2: | |
3171 | ceph_msg_put(m); | |
3172 | out: | |
b61c2763 SW |
3173 | if (!can_fail) { |
3174 | pr_err("msg_new can't create type %d front %d\n", type, | |
3175 | front_len); | |
f0ed1b7c | 3176 | WARN_ON(1); |
b61c2763 SW |
3177 | } else { |
3178 | dout("msg_new can't create type %d front %d\n", type, | |
3179 | front_len); | |
3180 | } | |
a79832f2 | 3181 | return NULL; |
31b8006e | 3182 | } |
3d14c5d2 | 3183 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 3184 | |
31b8006e SW |
3185 | /* |
3186 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
3187 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
3188 | * per-type header in the front and then gracefully fail (i.e., | |
3189 | * propagate the error to the caller based on info in the front) when | |
3190 | * the middle is too large. | |
3191 | */ | |
2450418c | 3192 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
3193 | { |
3194 | int type = le16_to_cpu(msg->hdr.type); | |
3195 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
3196 | ||
3197 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
3198 | ceph_msg_type_name(type), middle_len); | |
3199 | BUG_ON(!middle_len); | |
3200 | BUG_ON(msg->middle); | |
3201 | ||
b6c1d5b8 | 3202 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
3203 | if (!msg->middle) |
3204 | return -ENOMEM; | |
3205 | return 0; | |
3206 | } | |
3207 | ||
2450418c | 3208 | /* |
1c20f2d2 AE |
3209 | * Allocate a message for receiving an incoming message on a |
3210 | * connection, and save the result in con->in_msg. Uses the | |
3211 | * connection's private alloc_msg op if available. | |
3212 | * | |
4740a623 SW |
3213 | * Returns 0 on success, or a negative error code. |
3214 | * | |
3215 | * On success, if we set *skip = 1: | |
3216 | * - the next message should be skipped and ignored. | |
3217 | * - con->in_msg == NULL | |
3218 | * or if we set *skip = 0: | |
3219 | * - con->in_msg is non-null. | |
3220 | * On error (ENOMEM, EAGAIN, ...), | |
3221 | * - con->in_msg == NULL | |
2450418c | 3222 | */ |
4740a623 | 3223 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
2450418c | 3224 | { |
4740a623 | 3225 | struct ceph_msg_header *hdr = &con->in_hdr; |
2450418c | 3226 | int middle_len = le32_to_cpu(hdr->middle_len); |
1d866d1c | 3227 | struct ceph_msg *msg; |
4740a623 | 3228 | int ret = 0; |
2450418c | 3229 | |
1c20f2d2 | 3230 | BUG_ON(con->in_msg != NULL); |
53ded495 | 3231 | BUG_ON(!con->ops->alloc_msg); |
2450418c | 3232 | |
53ded495 AE |
3233 | mutex_unlock(&con->mutex); |
3234 | msg = con->ops->alloc_msg(con, hdr, skip); | |
3235 | mutex_lock(&con->mutex); | |
3236 | if (con->state != CON_STATE_OPEN) { | |
3237 | if (msg) | |
1d866d1c | 3238 | ceph_msg_put(msg); |
53ded495 AE |
3239 | return -EAGAIN; |
3240 | } | |
4137577a AE |
3241 | if (msg) { |
3242 | BUG_ON(*skip); | |
3243 | con->in_msg = msg; | |
36eb71aa | 3244 | con->in_msg->con = con->ops->get(con); |
92ce034b | 3245 | BUG_ON(con->in_msg->con == NULL); |
4137577a AE |
3246 | } else { |
3247 | /* | |
3248 | * Null message pointer means either we should skip | |
3249 | * this message or we couldn't allocate memory. The | |
3250 | * former is not an error. | |
3251 | */ | |
3252 | if (*skip) | |
3253 | return 0; | |
3254 | con->error_msg = "error allocating memory for incoming message"; | |
3255 | ||
53ded495 | 3256 | return -ENOMEM; |
2450418c | 3257 | } |
1c20f2d2 | 3258 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 3259 | |
1c20f2d2 AE |
3260 | if (middle_len && !con->in_msg->middle) { |
3261 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 3262 | if (ret < 0) { |
1c20f2d2 AE |
3263 | ceph_msg_put(con->in_msg); |
3264 | con->in_msg = NULL; | |
2450418c YS |
3265 | } |
3266 | } | |
9d7f0f13 | 3267 | |
4740a623 | 3268 | return ret; |
2450418c YS |
3269 | } |
3270 | ||
31b8006e SW |
3271 | |
3272 | /* | |
3273 | * Free a generically kmalloc'd message. | |
3274 | */ | |
3275 | void ceph_msg_kfree(struct ceph_msg *m) | |
3276 | { | |
3277 | dout("msg_kfree %p\n", m); | |
3278 | if (m->front_is_vmalloc) | |
3279 | vfree(m->front.iov_base); | |
3280 | else | |
3281 | kfree(m->front.iov_base); | |
e3d5d638 | 3282 | kmem_cache_free(ceph_msg_cache, m); |
31b8006e SW |
3283 | } |
3284 | ||
3285 | /* | |
3286 | * Drop a msg ref. Destroy as needed. | |
3287 | */ | |
c2e552e7 SW |
3288 | void ceph_msg_last_put(struct kref *kref) |
3289 | { | |
3290 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
5240d9f9 AE |
3291 | LIST_HEAD(data); |
3292 | struct list_head *links; | |
3293 | struct list_head *next; | |
31b8006e | 3294 | |
c2e552e7 SW |
3295 | dout("ceph_msg_put last one on %p\n", m); |
3296 | WARN_ON(!list_empty(&m->list_head)); | |
3297 | ||
3298 | /* drop middle, data, if any */ | |
3299 | if (m->middle) { | |
3300 | ceph_buffer_put(m->middle); | |
3301 | m->middle = NULL; | |
31b8006e | 3302 | } |
5240d9f9 AE |
3303 | |
3304 | list_splice_init(&m->data, &data); | |
3305 | list_for_each_safe(links, next, &data) { | |
3306 | struct ceph_msg_data *data; | |
3307 | ||
3308 | data = list_entry(links, struct ceph_msg_data, links); | |
3309 | list_del_init(links); | |
3310 | ceph_msg_data_destroy(data); | |
3311 | } | |
a1930804 | 3312 | m->data_length = 0; |
58bb3b37 | 3313 | |
c2e552e7 SW |
3314 | if (m->pool) |
3315 | ceph_msgpool_put(m->pool, m); | |
3316 | else | |
3317 | ceph_msg_kfree(m); | |
31b8006e | 3318 | } |
3d14c5d2 | 3319 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
3320 | |
3321 | void ceph_msg_dump(struct ceph_msg *msg) | |
3322 | { | |
842a5780 ID |
3323 | pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg, |
3324 | msg->front_alloc_len, msg->data_length); | |
9ec7cab1 SW |
3325 | print_hex_dump(KERN_DEBUG, "header: ", |
3326 | DUMP_PREFIX_OFFSET, 16, 1, | |
3327 | &msg->hdr, sizeof(msg->hdr), true); | |
3328 | print_hex_dump(KERN_DEBUG, " front: ", | |
3329 | DUMP_PREFIX_OFFSET, 16, 1, | |
3330 | msg->front.iov_base, msg->front.iov_len, true); | |
3331 | if (msg->middle) | |
3332 | print_hex_dump(KERN_DEBUG, "middle: ", | |
3333 | DUMP_PREFIX_OFFSET, 16, 1, | |
3334 | msg->middle->vec.iov_base, | |
3335 | msg->middle->vec.iov_len, true); | |
3336 | print_hex_dump(KERN_DEBUG, "footer: ", | |
3337 | DUMP_PREFIX_OFFSET, 16, 1, | |
3338 | &msg->footer, sizeof(msg->footer), true); | |
3339 | } | |
3d14c5d2 | 3340 | EXPORT_SYMBOL(ceph_msg_dump); |