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> | |
68b4476b YS |
12 | #include <linux/bio.h> |
13 | #include <linux/blkdev.h> | |
ee3b56f2 | 14 | #include <linux/dns_resolver.h> |
31b8006e SW |
15 | #include <net/tcp.h> |
16 | ||
3d14c5d2 YS |
17 | #include <linux/ceph/libceph.h> |
18 | #include <linux/ceph/messenger.h> | |
19 | #include <linux/ceph/decode.h> | |
20 | #include <linux/ceph/pagelist.h> | |
bc3b2d7f | 21 | #include <linux/export.h> |
31b8006e SW |
22 | |
23 | /* | |
24 | * Ceph uses the messenger to exchange ceph_msg messages with other | |
25 | * hosts in the system. The messenger provides ordered and reliable | |
26 | * delivery. We tolerate TCP disconnects by reconnecting (with | |
27 | * exponential backoff) in the case of a fault (disconnection, bad | |
28 | * crc, protocol error). Acks allow sent messages to be discarded by | |
29 | * the sender. | |
30 | */ | |
31 | ||
bc18f4b1 AE |
32 | /* |
33 | * We track the state of the socket on a given connection using | |
34 | * values defined below. The transition to a new socket state is | |
35 | * handled by a function which verifies we aren't coming from an | |
36 | * unexpected state. | |
37 | * | |
38 | * -------- | |
39 | * | NEW* | transient initial state | |
40 | * -------- | |
41 | * | con_sock_state_init() | |
42 | * v | |
43 | * ---------- | |
44 | * | CLOSED | initialized, but no socket (and no | |
45 | * ---------- TCP connection) | |
46 | * ^ \ | |
47 | * | \ con_sock_state_connecting() | |
48 | * | ---------------------- | |
49 | * | \ | |
50 | * + con_sock_state_closed() \ | |
fbb85a47 SW |
51 | * |+--------------------------- \ |
52 | * | \ \ \ | |
53 | * | ----------- \ \ | |
54 | * | | CLOSING | socket event; \ \ | |
55 | * | ----------- await close \ \ | |
56 | * | ^ \ | | |
57 | * | | \ | | |
58 | * | + con_sock_state_closing() \ | | |
59 | * | / \ | | | |
60 | * | / --------------- | | | |
61 | * | / \ v v | |
bc18f4b1 AE |
62 | * | / -------------- |
63 | * | / -----------------| CONNECTING | socket created, TCP | |
64 | * | | / -------------- connect initiated | |
65 | * | | | con_sock_state_connected() | |
66 | * | | v | |
67 | * ------------- | |
68 | * | CONNECTED | TCP connection established | |
69 | * ------------- | |
70 | * | |
71 | * State values for ceph_connection->sock_state; NEW is assumed to be 0. | |
72 | */ | |
ce2c8903 AE |
73 | |
74 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ | |
75 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ | |
76 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ | |
77 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ | |
78 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ | |
79 | ||
31b8006e SW |
80 | /* static tag bytes (protocol control messages) */ |
81 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
82 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
83 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
84 | ||
a6a5349d SW |
85 | #ifdef CONFIG_LOCKDEP |
86 | static struct lock_class_key socket_class; | |
87 | #endif | |
88 | ||
84495f49 AE |
89 | /* |
90 | * When skipping (ignoring) a block of input we read it into a "skip | |
91 | * buffer," which is this many bytes in size. | |
92 | */ | |
93 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
94 | |
95 | static void queue_con(struct ceph_connection *con); | |
96 | static void con_work(struct work_struct *); | |
97 | static void ceph_fault(struct ceph_connection *con); | |
98 | ||
31b8006e | 99 | /* |
f64a9317 AE |
100 | * Nicely render a sockaddr as a string. An array of formatted |
101 | * strings is used, to approximate reentrancy. | |
31b8006e | 102 | */ |
f64a9317 AE |
103 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
104 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
105 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
106 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
107 | ||
108 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
109 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 110 | |
57666519 | 111 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 112 | |
3d14c5d2 | 113 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
114 | { |
115 | int i; | |
116 | char *s; | |
99f0f3b2 AE |
117 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
118 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 119 | |
f64a9317 | 120 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
121 | s = addr_str[i]; |
122 | ||
123 | switch (ss->ss_family) { | |
124 | case AF_INET: | |
bd406145 AE |
125 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
126 | ntohs(in4->sin_port)); | |
31b8006e SW |
127 | break; |
128 | ||
129 | case AF_INET6: | |
bd406145 AE |
130 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
131 | ntohs(in6->sin6_port)); | |
31b8006e SW |
132 | break; |
133 | ||
134 | default: | |
d3002b97 AE |
135 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
136 | ss->ss_family); | |
31b8006e SW |
137 | } |
138 | ||
139 | return s; | |
140 | } | |
3d14c5d2 | 141 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 142 | |
63f2d211 SW |
143 | static void encode_my_addr(struct ceph_messenger *msgr) |
144 | { | |
145 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
146 | ceph_encode_addr(&msgr->my_enc_addr); | |
147 | } | |
148 | ||
31b8006e SW |
149 | /* |
150 | * work queue for all reading and writing to/from the socket. | |
151 | */ | |
e0f43c94 | 152 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 153 | |
6173d1f0 AE |
154 | void _ceph_msgr_exit(void) |
155 | { | |
d3002b97 | 156 | if (ceph_msgr_wq) { |
6173d1f0 | 157 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
158 | ceph_msgr_wq = NULL; |
159 | } | |
6173d1f0 | 160 | |
6173d1f0 AE |
161 | BUG_ON(zero_page == NULL); |
162 | kunmap(zero_page); | |
163 | page_cache_release(zero_page); | |
164 | zero_page = NULL; | |
165 | } | |
166 | ||
3d14c5d2 | 167 | int ceph_msgr_init(void) |
31b8006e | 168 | { |
57666519 AE |
169 | BUG_ON(zero_page != NULL); |
170 | zero_page = ZERO_PAGE(0); | |
171 | page_cache_get(zero_page); | |
172 | ||
f363e45f | 173 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0); |
6173d1f0 AE |
174 | if (ceph_msgr_wq) |
175 | return 0; | |
57666519 | 176 | |
6173d1f0 AE |
177 | pr_err("msgr_init failed to create workqueue\n"); |
178 | _ceph_msgr_exit(); | |
57666519 | 179 | |
6173d1f0 | 180 | return -ENOMEM; |
31b8006e | 181 | } |
3d14c5d2 | 182 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
183 | |
184 | void ceph_msgr_exit(void) | |
185 | { | |
57666519 | 186 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 187 | |
6173d1f0 | 188 | _ceph_msgr_exit(); |
31b8006e | 189 | } |
3d14c5d2 | 190 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 191 | |
cd84db6e | 192 | void ceph_msgr_flush(void) |
a922d38f SW |
193 | { |
194 | flush_workqueue(ceph_msgr_wq); | |
195 | } | |
3d14c5d2 | 196 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 197 | |
ce2c8903 AE |
198 | /* Connection socket state transition functions */ |
199 | ||
200 | static void con_sock_state_init(struct ceph_connection *con) | |
201 | { | |
202 | int old_state; | |
203 | ||
204 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
205 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
206 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
207 | } | |
208 | ||
209 | static void con_sock_state_connecting(struct ceph_connection *con) | |
210 | { | |
211 | int old_state; | |
212 | ||
213 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
214 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
215 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
216 | } | |
217 | ||
218 | static void con_sock_state_connected(struct ceph_connection *con) | |
219 | { | |
220 | int old_state; | |
221 | ||
222 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
223 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
224 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
225 | } | |
226 | ||
227 | static void con_sock_state_closing(struct ceph_connection *con) | |
228 | { | |
229 | int old_state; | |
230 | ||
231 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
232 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
233 | old_state != CON_SOCK_STATE_CONNECTED && | |
234 | old_state != CON_SOCK_STATE_CLOSING)) | |
235 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
236 | } | |
237 | ||
238 | static void con_sock_state_closed(struct ceph_connection *con) | |
239 | { | |
240 | int old_state; | |
241 | ||
242 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
243 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 SW |
244 | old_state != CON_SOCK_STATE_CLOSING && |
245 | old_state != CON_SOCK_STATE_CONNECTING)) | |
ce2c8903 AE |
246 | printk("%s: unexpected old state %d\n", __func__, old_state); |
247 | } | |
a922d38f | 248 | |
31b8006e SW |
249 | /* |
250 | * socket callback functions | |
251 | */ | |
252 | ||
253 | /* data available on socket, or listen socket received a connect */ | |
327800bd | 254 | static void ceph_sock_data_ready(struct sock *sk, int count_unused) |
31b8006e | 255 | { |
bd406145 AE |
256 | struct ceph_connection *con = sk->sk_user_data; |
257 | ||
31b8006e | 258 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 259 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
260 | con, con->state); |
261 | queue_con(con); | |
262 | } | |
263 | } | |
264 | ||
265 | /* socket has buffer space for writing */ | |
327800bd | 266 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 267 | { |
d3002b97 | 268 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 269 | |
182fac26 JS |
270 | /* only queue to workqueue if there is data we want to write, |
271 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 272 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
273 | * doesn't get called again until try_write() fills the socket |
274 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
275 | * and net/core/stream.c:sk_stream_write_space(). | |
276 | */ | |
928443cd | 277 | if (test_bit(WRITE_PENDING, &con->flags)) { |
182fac26 | 278 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
327800bd | 279 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
280 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
281 | queue_con(con); | |
282 | } | |
31b8006e | 283 | } else { |
327800bd | 284 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 285 | } |
31b8006e SW |
286 | } |
287 | ||
288 | /* socket's state has changed */ | |
327800bd | 289 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 290 | { |
bd406145 | 291 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 292 | |
327800bd | 293 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
294 | con, con->state, sk->sk_state); |
295 | ||
296 | if (test_bit(CLOSED, &con->state)) | |
297 | return; | |
298 | ||
299 | switch (sk->sk_state) { | |
300 | case TCP_CLOSE: | |
327800bd | 301 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 302 | case TCP_CLOSE_WAIT: |
327800bd | 303 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 304 | con_sock_state_closing(con); |
d65c9e0b AE |
305 | set_bit(SOCK_CLOSED, &con->flags); |
306 | queue_con(con); | |
31b8006e SW |
307 | break; |
308 | case TCP_ESTABLISHED: | |
327800bd | 309 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 310 | con_sock_state_connected(con); |
31b8006e SW |
311 | queue_con(con); |
312 | break; | |
d3002b97 AE |
313 | default: /* Everything else is uninteresting */ |
314 | break; | |
31b8006e SW |
315 | } |
316 | } | |
317 | ||
318 | /* | |
319 | * set up socket callbacks | |
320 | */ | |
321 | static void set_sock_callbacks(struct socket *sock, | |
322 | struct ceph_connection *con) | |
323 | { | |
324 | struct sock *sk = sock->sk; | |
bd406145 | 325 | sk->sk_user_data = con; |
327800bd AE |
326 | sk->sk_data_ready = ceph_sock_data_ready; |
327 | sk->sk_write_space = ceph_sock_write_space; | |
328 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
329 | } |
330 | ||
331 | ||
332 | /* | |
333 | * socket helpers | |
334 | */ | |
335 | ||
336 | /* | |
337 | * initiate connection to a remote socket. | |
338 | */ | |
41617d0c | 339 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 340 | { |
f91d3471 | 341 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
342 | struct socket *sock; |
343 | int ret; | |
344 | ||
345 | BUG_ON(con->sock); | |
f91d3471 SW |
346 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
347 | IPPROTO_TCP, &sock); | |
31b8006e | 348 | if (ret) |
41617d0c | 349 | return ret; |
31b8006e SW |
350 | sock->sk->sk_allocation = GFP_NOFS; |
351 | ||
a6a5349d SW |
352 | #ifdef CONFIG_LOCKDEP |
353 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
354 | #endif | |
355 | ||
31b8006e SW |
356 | set_sock_callbacks(sock, con); |
357 | ||
3d14c5d2 | 358 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 359 | |
89a86be0 | 360 | con_sock_state_connecting(con); |
f91d3471 SW |
361 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
362 | O_NONBLOCK); | |
31b8006e SW |
363 | if (ret == -EINPROGRESS) { |
364 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 365 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 366 | sock->sk->sk_state); |
a5bc3129 | 367 | } else if (ret < 0) { |
31b8006e | 368 | pr_err("connect %s error %d\n", |
3d14c5d2 | 369 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 370 | sock_release(sock); |
31b8006e | 371 | con->error_msg = "connect error"; |
31b8006e | 372 | |
41617d0c | 373 | return ret; |
a5bc3129 AE |
374 | } |
375 | con->sock = sock; | |
41617d0c | 376 | return 0; |
31b8006e SW |
377 | } |
378 | ||
379 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
380 | { | |
381 | struct kvec iov = {buf, len}; | |
382 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 383 | int r; |
31b8006e | 384 | |
98bdb0aa SW |
385 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
386 | if (r == -EAGAIN) | |
387 | r = 0; | |
388 | return r; | |
31b8006e SW |
389 | } |
390 | ||
391 | /* | |
392 | * write something. @more is true if caller will be sending more data | |
393 | * shortly. | |
394 | */ | |
395 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
396 | size_t kvlen, size_t len, int more) | |
397 | { | |
398 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 399 | int r; |
31b8006e SW |
400 | |
401 | if (more) | |
402 | msg.msg_flags |= MSG_MORE; | |
403 | else | |
404 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
405 | ||
42961d23 SW |
406 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
407 | if (r == -EAGAIN) | |
408 | r = 0; | |
409 | return r; | |
31b8006e SW |
410 | } |
411 | ||
31739139 AE |
412 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
413 | int offset, size_t size, int more) | |
414 | { | |
415 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
416 | int ret; | |
417 | ||
418 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
419 | if (ret == -EAGAIN) | |
420 | ret = 0; | |
421 | ||
422 | return ret; | |
423 | } | |
424 | ||
31b8006e SW |
425 | |
426 | /* | |
427 | * Shutdown/close the socket for the given connection. | |
428 | */ | |
429 | static int con_close_socket(struct ceph_connection *con) | |
430 | { | |
431 | int rc; | |
432 | ||
433 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
434 | if (!con->sock) | |
435 | return 0; | |
31b8006e SW |
436 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); |
437 | sock_release(con->sock); | |
438 | con->sock = NULL; | |
456ea468 AE |
439 | |
440 | /* | |
441 | * Forcibly clear the SOCK_CLOSE flag. It gets set | |
442 | * independent of the connection mutex, and we could have | |
443 | * received a socket close event before we had the chance to | |
444 | * shut the socket down. | |
445 | */ | |
a8d00e3c | 446 | clear_bit(SOCK_CLOSED, &con->flags); |
ce2c8903 | 447 | con_sock_state_closed(con); |
31b8006e SW |
448 | return rc; |
449 | } | |
450 | ||
451 | /* | |
452 | * Reset a connection. Discard all incoming and outgoing messages | |
453 | * and clear *_seq state. | |
454 | */ | |
455 | static void ceph_msg_remove(struct ceph_msg *msg) | |
456 | { | |
457 | list_del_init(&msg->list_head); | |
38941f80 | 458 | BUG_ON(msg->con == NULL); |
36eb71aa | 459 | msg->con->ops->put(msg->con); |
38941f80 AE |
460 | msg->con = NULL; |
461 | ||
31b8006e SW |
462 | ceph_msg_put(msg); |
463 | } | |
464 | static void ceph_msg_remove_list(struct list_head *head) | |
465 | { | |
466 | while (!list_empty(head)) { | |
467 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
468 | list_head); | |
469 | ceph_msg_remove(msg); | |
470 | } | |
471 | } | |
472 | ||
473 | static void reset_connection(struct ceph_connection *con) | |
474 | { | |
475 | /* reset connection, out_queue, msg_ and connect_seq */ | |
476 | /* discard existing out_queue and msg_seq */ | |
31b8006e SW |
477 | ceph_msg_remove_list(&con->out_queue); |
478 | ceph_msg_remove_list(&con->out_sent); | |
479 | ||
cf3e5c40 | 480 | if (con->in_msg) { |
38941f80 AE |
481 | BUG_ON(con->in_msg->con != con); |
482 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
483 | ceph_msg_put(con->in_msg); |
484 | con->in_msg = NULL; | |
36eb71aa | 485 | con->ops->put(con); |
cf3e5c40 SW |
486 | } |
487 | ||
31b8006e SW |
488 | con->connect_seq = 0; |
489 | con->out_seq = 0; | |
c86a2930 SW |
490 | if (con->out_msg) { |
491 | ceph_msg_put(con->out_msg); | |
492 | con->out_msg = NULL; | |
493 | } | |
31b8006e | 494 | con->in_seq = 0; |
0e0d5e0c | 495 | con->in_seq_acked = 0; |
31b8006e SW |
496 | } |
497 | ||
498 | /* | |
499 | * mark a peer down. drop any open connections. | |
500 | */ | |
501 | void ceph_con_close(struct ceph_connection *con) | |
502 | { | |
3d14c5d2 YS |
503 | dout("con_close %p peer %s\n", con, |
504 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
a5988c49 | 505 | clear_bit(NEGOTIATING, &con->state); |
bb9e6bba | 506 | clear_bit(CONNECTING, &con->state); |
e27947c7 | 507 | clear_bit(CONNECTED, &con->state); |
31b8006e | 508 | clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */ |
a5988c49 AE |
509 | set_bit(CLOSED, &con->state); |
510 | ||
928443cd AE |
511 | clear_bit(LOSSYTX, &con->flags); /* so we retry next connect */ |
512 | clear_bit(KEEPALIVE_PENDING, &con->flags); | |
513 | clear_bit(WRITE_PENDING, &con->flags); | |
a5988c49 | 514 | |
ec302645 | 515 | mutex_lock(&con->mutex); |
31b8006e | 516 | reset_connection(con); |
6f2bc3ff | 517 | con->peer_global_seq = 0; |
91e45ce3 | 518 | cancel_delayed_work(&con->work); |
ec302645 | 519 | mutex_unlock(&con->mutex); |
31b8006e SW |
520 | queue_con(con); |
521 | } | |
3d14c5d2 | 522 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 523 | |
31b8006e SW |
524 | /* |
525 | * Reopen a closed connection, with a new peer address. | |
526 | */ | |
b7a9e5dd SW |
527 | void ceph_con_open(struct ceph_connection *con, |
528 | __u8 entity_type, __u64 entity_num, | |
529 | struct ceph_entity_addr *addr) | |
31b8006e | 530 | { |
3d14c5d2 | 531 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
31b8006e | 532 | set_bit(OPENING, &con->state); |
a5988c49 AE |
533 | WARN_ON(!test_and_clear_bit(CLOSED, &con->state)); |
534 | ||
b7a9e5dd SW |
535 | con->peer_name.type = (__u8) entity_type; |
536 | con->peer_name.num = cpu_to_le64(entity_num); | |
537 | ||
31b8006e | 538 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 539 | con->delay = 0; /* reset backoff memory */ |
31b8006e SW |
540 | queue_con(con); |
541 | } | |
3d14c5d2 | 542 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 543 | |
87b315a5 SW |
544 | /* |
545 | * return true if this connection ever successfully opened | |
546 | */ | |
547 | bool ceph_con_opened(struct ceph_connection *con) | |
548 | { | |
549 | return con->connect_seq > 0; | |
550 | } | |
551 | ||
31b8006e SW |
552 | /* |
553 | * initialize a new connection. | |
554 | */ | |
1bfd89f4 AE |
555 | void ceph_con_init(struct ceph_connection *con, void *private, |
556 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 557 | struct ceph_messenger *msgr) |
31b8006e SW |
558 | { |
559 | dout("con_init %p\n", con); | |
560 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
561 | con->private = private; |
562 | con->ops = ops; | |
31b8006e | 563 | con->msgr = msgr; |
ce2c8903 AE |
564 | |
565 | con_sock_state_init(con); | |
566 | ||
ec302645 | 567 | mutex_init(&con->mutex); |
31b8006e SW |
568 | INIT_LIST_HEAD(&con->out_queue); |
569 | INIT_LIST_HEAD(&con->out_sent); | |
570 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 AE |
571 | |
572 | set_bit(CLOSED, &con->state); | |
31b8006e | 573 | } |
3d14c5d2 | 574 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
575 | |
576 | ||
577 | /* | |
578 | * We maintain a global counter to order connection attempts. Get | |
579 | * a unique seq greater than @gt. | |
580 | */ | |
581 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
582 | { | |
583 | u32 ret; | |
584 | ||
585 | spin_lock(&msgr->global_seq_lock); | |
586 | if (msgr->global_seq < gt) | |
587 | msgr->global_seq = gt; | |
588 | ret = ++msgr->global_seq; | |
589 | spin_unlock(&msgr->global_seq_lock); | |
590 | return ret; | |
591 | } | |
592 | ||
e2200423 | 593 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
594 | { |
595 | con->out_kvec_left = 0; | |
596 | con->out_kvec_bytes = 0; | |
597 | con->out_kvec_cur = &con->out_kvec[0]; | |
598 | } | |
599 | ||
e2200423 | 600 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
601 | size_t size, void *data) |
602 | { | |
603 | int index; | |
604 | ||
605 | index = con->out_kvec_left; | |
606 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
607 | ||
608 | con->out_kvec[index].iov_len = size; | |
609 | con->out_kvec[index].iov_base = data; | |
610 | con->out_kvec_left++; | |
611 | con->out_kvec_bytes += size; | |
612 | } | |
31b8006e | 613 | |
df6ad1f9 AE |
614 | #ifdef CONFIG_BLOCK |
615 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
616 | { | |
617 | if (!bio) { | |
618 | *iter = NULL; | |
619 | *seg = 0; | |
620 | return; | |
621 | } | |
622 | *iter = bio; | |
623 | *seg = bio->bi_idx; | |
624 | } | |
625 | ||
626 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
627 | { | |
628 | if (*bio_iter == NULL) | |
629 | return; | |
630 | ||
631 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
632 | ||
633 | (*seg)++; | |
634 | if (*seg == (*bio_iter)->bi_vcnt) | |
635 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
636 | } | |
637 | #endif | |
638 | ||
739c905b AE |
639 | static void prepare_write_message_data(struct ceph_connection *con) |
640 | { | |
641 | struct ceph_msg *msg = con->out_msg; | |
642 | ||
643 | BUG_ON(!msg); | |
644 | BUG_ON(!msg->hdr.data_len); | |
645 | ||
646 | /* initialize page iterator */ | |
647 | con->out_msg_pos.page = 0; | |
648 | if (msg->pages) | |
649 | con->out_msg_pos.page_pos = msg->page_alignment; | |
650 | else | |
651 | con->out_msg_pos.page_pos = 0; | |
572c588e | 652 | #ifdef CONFIG_BLOCK |
abdaa6a8 | 653 | if (msg->bio) |
572c588e AE |
654 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); |
655 | #endif | |
739c905b AE |
656 | con->out_msg_pos.data_pos = 0; |
657 | con->out_msg_pos.did_page_crc = false; | |
658 | con->out_more = 1; /* data + footer will follow */ | |
659 | } | |
660 | ||
31b8006e SW |
661 | /* |
662 | * Prepare footer for currently outgoing message, and finish things | |
663 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
664 | */ | |
859eb799 | 665 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
666 | { |
667 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 668 | int v = con->out_kvec_left; |
31b8006e | 669 | |
fd154f3c AE |
670 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
671 | ||
31b8006e SW |
672 | dout("prepare_write_message_footer %p\n", con); |
673 | con->out_kvec_is_msg = true; | |
674 | con->out_kvec[v].iov_base = &m->footer; | |
675 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
676 | con->out_kvec_bytes += sizeof(m->footer); | |
677 | con->out_kvec_left++; | |
678 | con->out_more = m->more_to_follow; | |
c86a2930 | 679 | con->out_msg_done = true; |
31b8006e SW |
680 | } |
681 | ||
682 | /* | |
683 | * Prepare headers for the next outgoing message. | |
684 | */ | |
685 | static void prepare_write_message(struct ceph_connection *con) | |
686 | { | |
687 | struct ceph_msg *m; | |
a9a0c51a | 688 | u32 crc; |
31b8006e | 689 | |
e2200423 | 690 | con_out_kvec_reset(con); |
31b8006e | 691 | con->out_kvec_is_msg = true; |
c86a2930 | 692 | con->out_msg_done = false; |
31b8006e SW |
693 | |
694 | /* Sneak an ack in there first? If we can get it into the same | |
695 | * TCP packet that's a good thing. */ | |
696 | if (con->in_seq > con->in_seq_acked) { | |
697 | con->in_seq_acked = con->in_seq; | |
e2200423 | 698 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 699 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 700 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 701 | &con->out_temp_ack); |
31b8006e SW |
702 | } |
703 | ||
38941f80 | 704 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 705 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 706 | con->out_msg = m; |
38941f80 | 707 | BUG_ON(m->con != con); |
4cf9d544 SW |
708 | |
709 | /* put message on sent list */ | |
710 | ceph_msg_get(m); | |
711 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 712 | |
e84346b7 SW |
713 | /* |
714 | * only assign outgoing seq # if we haven't sent this message | |
715 | * yet. if it is requeued, resend with it's original seq. | |
716 | */ | |
717 | if (m->needs_out_seq) { | |
718 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
719 | m->needs_out_seq = false; | |
720 | } | |
31b8006e SW |
721 | |
722 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
723 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
724 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
725 | le32_to_cpu(m->hdr.data_len), | |
726 | m->nr_pages); | |
727 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); | |
728 | ||
729 | /* tag + hdr + front + middle */ | |
e2200423 AE |
730 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
731 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
732 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 733 | |
31b8006e | 734 | if (m->middle) |
e2200423 | 735 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 736 | m->middle->vec.iov_base); |
31b8006e SW |
737 | |
738 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
739 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
740 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 741 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
742 | |
743 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
744 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
745 | if (m->middle) { | |
746 | crc = crc32c(0, m->middle->vec.iov_base, | |
747 | m->middle->vec.iov_len); | |
748 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
749 | } else | |
31b8006e | 750 | con->out_msg->footer.middle_crc = 0; |
739c905b | 751 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
752 | le32_to_cpu(con->out_msg->footer.front_crc), |
753 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
754 | ||
755 | /* is there a data payload? */ | |
739c905b AE |
756 | con->out_msg->footer.data_crc = 0; |
757 | if (m->hdr.data_len) | |
758 | prepare_write_message_data(con); | |
759 | else | |
31b8006e | 760 | /* no, queue up footer too and be done */ |
859eb799 | 761 | prepare_write_message_footer(con); |
31b8006e | 762 | |
928443cd | 763 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
764 | } |
765 | ||
766 | /* | |
767 | * Prepare an ack. | |
768 | */ | |
769 | static void prepare_write_ack(struct ceph_connection *con) | |
770 | { | |
771 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
772 | con->in_seq_acked, con->in_seq); | |
773 | con->in_seq_acked = con->in_seq; | |
774 | ||
e2200423 | 775 | con_out_kvec_reset(con); |
859eb799 | 776 | |
e2200423 | 777 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 778 | |
31b8006e | 779 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 780 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
781 | &con->out_temp_ack); |
782 | ||
31b8006e | 783 | con->out_more = 1; /* more will follow.. eventually.. */ |
928443cd | 784 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
785 | } |
786 | ||
787 | /* | |
788 | * Prepare to write keepalive byte. | |
789 | */ | |
790 | static void prepare_write_keepalive(struct ceph_connection *con) | |
791 | { | |
792 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
793 | con_out_kvec_reset(con); |
794 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
928443cd | 795 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
796 | } |
797 | ||
798 | /* | |
799 | * Connection negotiation. | |
800 | */ | |
801 | ||
dac1e716 AE |
802 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
803 | int *auth_proto) | |
4e7a5dcd | 804 | { |
a3530df3 | 805 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
806 | |
807 | if (!con->ops->get_authorizer) { | |
808 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
809 | con->out_connect.authorizer_len = 0; | |
810 | ||
729796be | 811 | return NULL; |
b1c6b980 AE |
812 | } |
813 | ||
814 | /* Can't hold the mutex while getting authorizer */ | |
4e7a5dcd | 815 | |
ec302645 | 816 | mutex_unlock(&con->mutex); |
b1c6b980 | 817 | |
dac1e716 | 818 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
8f43fb53 | 819 | |
ec302645 | 820 | mutex_lock(&con->mutex); |
4e7a5dcd | 821 | |
a3530df3 | 822 | if (IS_ERR(auth)) |
729796be | 823 | return auth; |
928443cd | 824 | if (test_bit(CLOSED, &con->state) || test_bit(OPENING, &con->flags)) |
729796be | 825 | return ERR_PTR(-EAGAIN); |
0da5d703 | 826 | |
8f43fb53 AE |
827 | con->auth_reply_buf = auth->authorizer_reply_buf; |
828 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
829 | ||
859eb799 | 830 | |
729796be | 831 | return auth; |
4e7a5dcd SW |
832 | } |
833 | ||
31b8006e SW |
834 | /* |
835 | * We connected to a peer and are saying hello. | |
836 | */ | |
e825a66d | 837 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 838 | { |
e2200423 AE |
839 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
840 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 841 | &con->msgr->my_enc_addr); |
eed0ef2c | 842 | |
eed0ef2c | 843 | con->out_more = 0; |
928443cd | 844 | set_bit(WRITE_PENDING, &con->flags); |
eed0ef2c SW |
845 | } |
846 | ||
e825a66d | 847 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 848 | { |
95c96174 | 849 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 850 | int proto; |
dac1e716 | 851 | int auth_proto; |
729796be | 852 | struct ceph_auth_handshake *auth; |
31b8006e SW |
853 | |
854 | switch (con->peer_name.type) { | |
855 | case CEPH_ENTITY_TYPE_MON: | |
856 | proto = CEPH_MONC_PROTOCOL; | |
857 | break; | |
858 | case CEPH_ENTITY_TYPE_OSD: | |
859 | proto = CEPH_OSDC_PROTOCOL; | |
860 | break; | |
861 | case CEPH_ENTITY_TYPE_MDS: | |
862 | proto = CEPH_MDSC_PROTOCOL; | |
863 | break; | |
864 | default: | |
865 | BUG(); | |
866 | } | |
867 | ||
868 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
869 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 870 | |
e825a66d | 871 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
872 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
873 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
874 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
875 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
876 | con->out_connect.flags = 0; | |
31b8006e | 877 | |
dac1e716 AE |
878 | auth_proto = CEPH_AUTH_UNKNOWN; |
879 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
880 | if (IS_ERR(auth)) |
881 | return PTR_ERR(auth); | |
3da54776 | 882 | |
dac1e716 | 883 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
884 | con->out_connect.authorizer_len = auth ? |
885 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
886 | ||
ab166d5a | 887 | con_out_kvec_reset(con); |
e2200423 | 888 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
889 | &con->out_connect); |
890 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 891 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 892 | auth->authorizer_buf); |
859eb799 | 893 | |
31b8006e | 894 | con->out_more = 0; |
928443cd | 895 | set_bit(WRITE_PENDING, &con->flags); |
4e7a5dcd | 896 | |
e10c758e | 897 | return 0; |
31b8006e SW |
898 | } |
899 | ||
31b8006e SW |
900 | /* |
901 | * write as much of pending kvecs to the socket as we can. | |
902 | * 1 -> done | |
903 | * 0 -> socket full, but more to do | |
904 | * <0 -> error | |
905 | */ | |
906 | static int write_partial_kvec(struct ceph_connection *con) | |
907 | { | |
908 | int ret; | |
909 | ||
910 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
911 | while (con->out_kvec_bytes > 0) { | |
912 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
913 | con->out_kvec_left, con->out_kvec_bytes, | |
914 | con->out_more); | |
915 | if (ret <= 0) | |
916 | goto out; | |
917 | con->out_kvec_bytes -= ret; | |
918 | if (con->out_kvec_bytes == 0) | |
919 | break; /* done */ | |
f42299e6 AE |
920 | |
921 | /* account for full iov entries consumed */ | |
922 | while (ret >= con->out_kvec_cur->iov_len) { | |
923 | BUG_ON(!con->out_kvec_left); | |
924 | ret -= con->out_kvec_cur->iov_len; | |
925 | con->out_kvec_cur++; | |
926 | con->out_kvec_left--; | |
927 | } | |
928 | /* and for a partially-consumed entry */ | |
929 | if (ret) { | |
930 | con->out_kvec_cur->iov_len -= ret; | |
931 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
932 | } |
933 | } | |
934 | con->out_kvec_left = 0; | |
935 | con->out_kvec_is_msg = false; | |
936 | ret = 1; | |
937 | out: | |
938 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
939 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
940 | return ret; /* done! */ | |
941 | } | |
942 | ||
84ca8fc8 AE |
943 | static void out_msg_pos_next(struct ceph_connection *con, struct page *page, |
944 | size_t len, size_t sent, bool in_trail) | |
945 | { | |
946 | struct ceph_msg *msg = con->out_msg; | |
947 | ||
948 | BUG_ON(!msg); | |
949 | BUG_ON(!sent); | |
950 | ||
951 | con->out_msg_pos.data_pos += sent; | |
952 | con->out_msg_pos.page_pos += sent; | |
5821bd8c AE |
953 | if (sent < len) |
954 | return; | |
955 | ||
956 | BUG_ON(sent != len); | |
957 | con->out_msg_pos.page_pos = 0; | |
958 | con->out_msg_pos.page++; | |
959 | con->out_msg_pos.did_page_crc = false; | |
960 | if (in_trail) | |
961 | list_move_tail(&page->lru, | |
962 | &msg->trail->head); | |
963 | else if (msg->pagelist) | |
964 | list_move_tail(&page->lru, | |
965 | &msg->pagelist->head); | |
84ca8fc8 | 966 | #ifdef CONFIG_BLOCK |
5821bd8c AE |
967 | else if (msg->bio) |
968 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
84ca8fc8 | 969 | #endif |
84ca8fc8 AE |
970 | } |
971 | ||
31b8006e SW |
972 | /* |
973 | * Write as much message data payload as we can. If we finish, queue | |
974 | * up the footer. | |
975 | * 1 -> done, footer is now queued in out_kvec[]. | |
976 | * 0 -> socket full, but more to do | |
977 | * <0 -> error | |
978 | */ | |
979 | static int write_partial_msg_pages(struct ceph_connection *con) | |
980 | { | |
981 | struct ceph_msg *msg = con->out_msg; | |
95c96174 | 982 | unsigned int data_len = le32_to_cpu(msg->hdr.data_len); |
31b8006e | 983 | size_t len; |
37675b0f | 984 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 985 | int ret; |
68b4476b | 986 | int total_max_write; |
84ca8fc8 | 987 | bool in_trail = false; |
5821bd8c AE |
988 | const size_t trail_len = (msg->trail ? msg->trail->length : 0); |
989 | const size_t trail_off = data_len - trail_len; | |
31b8006e SW |
990 | |
991 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
84ca8fc8 | 992 | con, msg, con->out_msg_pos.page, msg->nr_pages, |
31b8006e SW |
993 | con->out_msg_pos.page_pos); |
994 | ||
5821bd8c AE |
995 | /* |
996 | * Iterate through each page that contains data to be | |
997 | * written, and send as much as possible for each. | |
998 | * | |
999 | * If we are calculating the data crc (the default), we will | |
1000 | * need to map the page. If we have no pages, they have | |
1001 | * been revoked, so use the zero page. | |
1002 | */ | |
68b4476b | 1003 | while (data_len > con->out_msg_pos.data_pos) { |
31b8006e | 1004 | struct page *page = NULL; |
68b4476b | 1005 | int max_write = PAGE_SIZE; |
9bd19663 | 1006 | int bio_offset = 0; |
68b4476b | 1007 | |
5821bd8c AE |
1008 | in_trail = in_trail || con->out_msg_pos.data_pos >= trail_off; |
1009 | if (!in_trail) | |
1010 | total_max_write = trail_off - con->out_msg_pos.data_pos; | |
68b4476b | 1011 | |
5821bd8c | 1012 | if (in_trail) { |
68b4476b YS |
1013 | total_max_write = data_len - con->out_msg_pos.data_pos; |
1014 | ||
1015 | page = list_first_entry(&msg->trail->head, | |
1016 | struct page, lru); | |
68b4476b | 1017 | } else if (msg->pages) { |
31b8006e | 1018 | page = msg->pages[con->out_msg_pos.page]; |
58bb3b37 SW |
1019 | } else if (msg->pagelist) { |
1020 | page = list_first_entry(&msg->pagelist->head, | |
1021 | struct page, lru); | |
68b4476b YS |
1022 | #ifdef CONFIG_BLOCK |
1023 | } else if (msg->bio) { | |
1024 | struct bio_vec *bv; | |
1025 | ||
1026 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
1027 | page = bv->bv_page; | |
9bd19663 | 1028 | bio_offset = bv->bv_offset; |
68b4476b YS |
1029 | max_write = bv->bv_len; |
1030 | #endif | |
31b8006e | 1031 | } else { |
57666519 | 1032 | page = zero_page; |
31b8006e | 1033 | } |
68b4476b YS |
1034 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
1035 | total_max_write); | |
1036 | ||
37675b0f | 1037 | if (do_datacrc && !con->out_msg_pos.did_page_crc) { |
9bd19663 | 1038 | void *base; |
5821bd8c | 1039 | u32 crc = le32_to_cpu(msg->footer.data_crc); |
8d63e318 | 1040 | char *kaddr; |
31b8006e | 1041 | |
8d63e318 | 1042 | kaddr = kmap(page); |
31b8006e | 1043 | BUG_ON(kaddr == NULL); |
9bd19663 | 1044 | base = kaddr + con->out_msg_pos.page_pos + bio_offset; |
5821bd8c | 1045 | crc = crc32c(crc, base, len); |
84ca8fc8 | 1046 | msg->footer.data_crc = cpu_to_le32(crc); |
bca064d2 | 1047 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 1048 | } |
e36b13cc | 1049 | ret = ceph_tcp_sendpage(con->sock, page, |
9bd19663 | 1050 | con->out_msg_pos.page_pos + bio_offset, |
e36b13cc | 1051 | len, 1); |
31b8006e | 1052 | |
0cdf9e60 | 1053 | if (do_datacrc) |
31b8006e SW |
1054 | kunmap(page); |
1055 | ||
1056 | if (ret <= 0) | |
1057 | goto out; | |
1058 | ||
84ca8fc8 | 1059 | out_msg_pos_next(con, page, len, (size_t) ret, in_trail); |
31b8006e SW |
1060 | } |
1061 | ||
1062 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
1063 | ||
1064 | /* prepare and queue up footer, too */ | |
37675b0f | 1065 | if (!do_datacrc) |
84ca8fc8 | 1066 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1067 | con_out_kvec_reset(con); |
859eb799 | 1068 | prepare_write_message_footer(con); |
31b8006e SW |
1069 | ret = 1; |
1070 | out: | |
1071 | return ret; | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * write some zeros | |
1076 | */ | |
1077 | static int write_partial_skip(struct ceph_connection *con) | |
1078 | { | |
1079 | int ret; | |
1080 | ||
1081 | while (con->out_skip > 0) { | |
31739139 | 1082 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1083 | |
31739139 | 1084 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1); |
31b8006e SW |
1085 | if (ret <= 0) |
1086 | goto out; | |
1087 | con->out_skip -= ret; | |
1088 | } | |
1089 | ret = 1; | |
1090 | out: | |
1091 | return ret; | |
1092 | } | |
1093 | ||
1094 | /* | |
1095 | * Prepare to read connection handshake, or an ack. | |
1096 | */ | |
eed0ef2c SW |
1097 | static void prepare_read_banner(struct ceph_connection *con) |
1098 | { | |
1099 | dout("prepare_read_banner %p\n", con); | |
1100 | con->in_base_pos = 0; | |
1101 | } | |
1102 | ||
31b8006e SW |
1103 | static void prepare_read_connect(struct ceph_connection *con) |
1104 | { | |
1105 | dout("prepare_read_connect %p\n", con); | |
1106 | con->in_base_pos = 0; | |
1107 | } | |
1108 | ||
1109 | static void prepare_read_ack(struct ceph_connection *con) | |
1110 | { | |
1111 | dout("prepare_read_ack %p\n", con); | |
1112 | con->in_base_pos = 0; | |
1113 | } | |
1114 | ||
1115 | static void prepare_read_tag(struct ceph_connection *con) | |
1116 | { | |
1117 | dout("prepare_read_tag %p\n", con); | |
1118 | con->in_base_pos = 0; | |
1119 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1120 | } | |
1121 | ||
1122 | /* | |
1123 | * Prepare to read a message. | |
1124 | */ | |
1125 | static int prepare_read_message(struct ceph_connection *con) | |
1126 | { | |
1127 | dout("prepare_read_message %p\n", con); | |
1128 | BUG_ON(con->in_msg != NULL); | |
1129 | con->in_base_pos = 0; | |
1130 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1131 | return 0; | |
1132 | } | |
1133 | ||
1134 | ||
1135 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1136 | int end, int size, void *object) |
31b8006e | 1137 | { |
e6cee71f AE |
1138 | while (con->in_base_pos < end) { |
1139 | int left = end - con->in_base_pos; | |
31b8006e SW |
1140 | int have = size - left; |
1141 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1142 | if (ret <= 0) | |
1143 | return ret; | |
1144 | con->in_base_pos += ret; | |
1145 | } | |
1146 | return 1; | |
1147 | } | |
1148 | ||
1149 | ||
1150 | /* | |
1151 | * Read all or part of the connect-side handshake on a new connection | |
1152 | */ | |
eed0ef2c | 1153 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1154 | { |
fd51653f AE |
1155 | int size; |
1156 | int end; | |
1157 | int ret; | |
31b8006e | 1158 | |
eed0ef2c | 1159 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1160 | |
1161 | /* peer's banner */ | |
fd51653f AE |
1162 | size = strlen(CEPH_BANNER); |
1163 | end = size; | |
1164 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1165 | if (ret <= 0) |
1166 | goto out; | |
fd51653f AE |
1167 | |
1168 | size = sizeof (con->actual_peer_addr); | |
1169 | end += size; | |
1170 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1171 | if (ret <= 0) |
1172 | goto out; | |
fd51653f AE |
1173 | |
1174 | size = sizeof (con->peer_addr_for_me); | |
1175 | end += size; | |
1176 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1177 | if (ret <= 0) |
1178 | goto out; | |
fd51653f | 1179 | |
eed0ef2c SW |
1180 | out: |
1181 | return ret; | |
1182 | } | |
1183 | ||
1184 | static int read_partial_connect(struct ceph_connection *con) | |
1185 | { | |
fd51653f AE |
1186 | int size; |
1187 | int end; | |
1188 | int ret; | |
eed0ef2c SW |
1189 | |
1190 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1191 | ||
fd51653f AE |
1192 | size = sizeof (con->in_reply); |
1193 | end = size; | |
1194 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1195 | if (ret <= 0) |
1196 | goto out; | |
fd51653f AE |
1197 | |
1198 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1199 | end += size; | |
1200 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1201 | if (ret <= 0) |
1202 | goto out; | |
31b8006e | 1203 | |
4e7a5dcd SW |
1204 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1205 | con, (int)con->in_reply.tag, | |
1206 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1207 | le32_to_cpu(con->in_reply.global_seq)); |
1208 | out: | |
1209 | return ret; | |
eed0ef2c | 1210 | |
31b8006e SW |
1211 | } |
1212 | ||
1213 | /* | |
1214 | * Verify the hello banner looks okay. | |
1215 | */ | |
1216 | static int verify_hello(struct ceph_connection *con) | |
1217 | { | |
1218 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1219 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1220 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1221 | con->error_msg = "protocol error, bad banner"; |
1222 | return -1; | |
1223 | } | |
1224 | return 0; | |
1225 | } | |
1226 | ||
1227 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1228 | { | |
1229 | switch (ss->ss_family) { | |
1230 | case AF_INET: | |
1231 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1232 | case AF_INET6: | |
1233 | return | |
1234 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1235 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1236 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1237 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1238 | } | |
1239 | return false; | |
1240 | } | |
1241 | ||
1242 | static int addr_port(struct sockaddr_storage *ss) | |
1243 | { | |
1244 | switch (ss->ss_family) { | |
1245 | case AF_INET: | |
f28bcfbe | 1246 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1247 | case AF_INET6: |
f28bcfbe | 1248 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1249 | } |
1250 | return 0; | |
1251 | } | |
1252 | ||
1253 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1254 | { | |
1255 | switch (ss->ss_family) { | |
1256 | case AF_INET: | |
1257 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1258 | break; |
31b8006e SW |
1259 | case AF_INET6: |
1260 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1261 | break; |
31b8006e SW |
1262 | } |
1263 | } | |
1264 | ||
ee3b56f2 NW |
1265 | /* |
1266 | * Unlike other *_pton function semantics, zero indicates success. | |
1267 | */ | |
1268 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1269 | char delim, const char **ipend) | |
1270 | { | |
99f0f3b2 AE |
1271 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1272 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1273 | |
1274 | memset(ss, 0, sizeof(*ss)); | |
1275 | ||
1276 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1277 | ss->ss_family = AF_INET; | |
1278 | return 0; | |
1279 | } | |
1280 | ||
1281 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1282 | ss->ss_family = AF_INET6; | |
1283 | return 0; | |
1284 | } | |
1285 | ||
1286 | return -EINVAL; | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * Extract hostname string and resolve using kernel DNS facility. | |
1291 | */ | |
1292 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1293 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1294 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1295 | { | |
1296 | const char *end, *delim_p; | |
1297 | char *colon_p, *ip_addr = NULL; | |
1298 | int ip_len, ret; | |
1299 | ||
1300 | /* | |
1301 | * The end of the hostname occurs immediately preceding the delimiter or | |
1302 | * the port marker (':') where the delimiter takes precedence. | |
1303 | */ | |
1304 | delim_p = memchr(name, delim, namelen); | |
1305 | colon_p = memchr(name, ':', namelen); | |
1306 | ||
1307 | if (delim_p && colon_p) | |
1308 | end = delim_p < colon_p ? delim_p : colon_p; | |
1309 | else if (!delim_p && colon_p) | |
1310 | end = colon_p; | |
1311 | else { | |
1312 | end = delim_p; | |
1313 | if (!end) /* case: hostname:/ */ | |
1314 | end = name + namelen; | |
1315 | } | |
1316 | ||
1317 | if (end <= name) | |
1318 | return -EINVAL; | |
1319 | ||
1320 | /* do dns_resolve upcall */ | |
1321 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1322 | if (ip_len > 0) | |
1323 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1324 | else | |
1325 | ret = -ESRCH; | |
1326 | ||
1327 | kfree(ip_addr); | |
1328 | ||
1329 | *ipend = end; | |
1330 | ||
1331 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1332 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1333 | ||
1334 | return ret; | |
1335 | } | |
1336 | #else | |
1337 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1338 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1339 | { | |
1340 | return -EINVAL; | |
1341 | } | |
1342 | #endif | |
1343 | ||
1344 | /* | |
1345 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1346 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1347 | */ | |
1348 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1349 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1350 | { | |
1351 | int ret; | |
1352 | ||
1353 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1354 | if (ret) | |
1355 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1356 | ||
1357 | return ret; | |
1358 | } | |
1359 | ||
31b8006e SW |
1360 | /* |
1361 | * Parse an ip[:port] list into an addr array. Use the default | |
1362 | * monitor port if a port isn't specified. | |
1363 | */ | |
1364 | int ceph_parse_ips(const char *c, const char *end, | |
1365 | struct ceph_entity_addr *addr, | |
1366 | int max_count, int *count) | |
1367 | { | |
ee3b56f2 | 1368 | int i, ret = -EINVAL; |
31b8006e SW |
1369 | const char *p = c; |
1370 | ||
1371 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1372 | for (i = 0; i < max_count; i++) { | |
1373 | const char *ipend; | |
1374 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1375 | int port; |
39139f64 SW |
1376 | char delim = ','; |
1377 | ||
1378 | if (*p == '[') { | |
1379 | delim = ']'; | |
1380 | p++; | |
1381 | } | |
31b8006e | 1382 | |
ee3b56f2 NW |
1383 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1384 | if (ret) | |
31b8006e | 1385 | goto bad; |
ee3b56f2 NW |
1386 | ret = -EINVAL; |
1387 | ||
31b8006e SW |
1388 | p = ipend; |
1389 | ||
39139f64 SW |
1390 | if (delim == ']') { |
1391 | if (*p != ']') { | |
1392 | dout("missing matching ']'\n"); | |
1393 | goto bad; | |
1394 | } | |
1395 | p++; | |
1396 | } | |
1397 | ||
31b8006e SW |
1398 | /* port? */ |
1399 | if (p < end && *p == ':') { | |
1400 | port = 0; | |
1401 | p++; | |
1402 | while (p < end && *p >= '0' && *p <= '9') { | |
1403 | port = (port * 10) + (*p - '0'); | |
1404 | p++; | |
1405 | } | |
1406 | if (port > 65535 || port == 0) | |
1407 | goto bad; | |
1408 | } else { | |
1409 | port = CEPH_MON_PORT; | |
1410 | } | |
1411 | ||
1412 | addr_set_port(ss, port); | |
1413 | ||
3d14c5d2 | 1414 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1415 | |
1416 | if (p == end) | |
1417 | break; | |
1418 | if (*p != ',') | |
1419 | goto bad; | |
1420 | p++; | |
1421 | } | |
1422 | ||
1423 | if (p != end) | |
1424 | goto bad; | |
1425 | ||
1426 | if (count) | |
1427 | *count = i + 1; | |
1428 | return 0; | |
1429 | ||
1430 | bad: | |
39139f64 | 1431 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1432 | return ret; |
31b8006e | 1433 | } |
3d14c5d2 | 1434 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1435 | |
eed0ef2c | 1436 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1437 | { |
eed0ef2c | 1438 | dout("process_banner on %p\n", con); |
31b8006e SW |
1439 | |
1440 | if (verify_hello(con) < 0) | |
1441 | return -1; | |
1442 | ||
63f2d211 SW |
1443 | ceph_decode_addr(&con->actual_peer_addr); |
1444 | ceph_decode_addr(&con->peer_addr_for_me); | |
1445 | ||
31b8006e SW |
1446 | /* |
1447 | * Make sure the other end is who we wanted. note that the other | |
1448 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1449 | * them the benefit of the doubt. | |
1450 | */ | |
103e2d3a SW |
1451 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1452 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1453 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1454 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1455 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1456 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1457 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1458 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1459 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1460 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1461 | return -1; |
1462 | } | |
1463 | ||
1464 | /* | |
1465 | * did we learn our address? | |
1466 | */ | |
1467 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1468 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1469 | ||
1470 | memcpy(&con->msgr->inst.addr.in_addr, | |
1471 | &con->peer_addr_for_me.in_addr, | |
1472 | sizeof(con->peer_addr_for_me.in_addr)); | |
1473 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1474 | encode_my_addr(con->msgr); |
eed0ef2c | 1475 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1476 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1477 | } |
1478 | ||
eed0ef2c SW |
1479 | return 0; |
1480 | } | |
1481 | ||
04a419f9 SW |
1482 | static void fail_protocol(struct ceph_connection *con) |
1483 | { | |
1484 | reset_connection(con); | |
1485 | set_bit(CLOSED, &con->state); /* in case there's queued work */ | |
04a419f9 SW |
1486 | } |
1487 | ||
eed0ef2c SW |
1488 | static int process_connect(struct ceph_connection *con) |
1489 | { | |
3d14c5d2 YS |
1490 | u64 sup_feat = con->msgr->supported_features; |
1491 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1492 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1493 | int ret; |
04a419f9 | 1494 | |
eed0ef2c SW |
1495 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1496 | ||
31b8006e | 1497 | switch (con->in_reply.tag) { |
04a419f9 SW |
1498 | case CEPH_MSGR_TAG_FEATURES: |
1499 | pr_err("%s%lld %s feature set mismatch," | |
1500 | " my %llx < server's %llx, missing %llx\n", | |
1501 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1502 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1503 | sup_feat, server_feat, server_feat & ~sup_feat); |
1504 | con->error_msg = "missing required protocol features"; | |
1505 | fail_protocol(con); | |
1506 | return -1; | |
1507 | ||
31b8006e | 1508 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1509 | pr_err("%s%lld %s protocol version mismatch," |
1510 | " my %d != server's %d\n", | |
1511 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1512 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1513 | le32_to_cpu(con->out_connect.protocol_version), |
1514 | le32_to_cpu(con->in_reply.protocol_version)); | |
1515 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1516 | fail_protocol(con); |
31b8006e SW |
1517 | return -1; |
1518 | ||
4e7a5dcd SW |
1519 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1520 | con->auth_retry++; | |
1521 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1522 | con->auth_retry); | |
1523 | if (con->auth_retry == 2) { | |
1524 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1525 | return -1; |
1526 | } | |
1527 | con->auth_retry = 1; | |
e825a66d | 1528 | ret = prepare_write_connect(con); |
0da5d703 SW |
1529 | if (ret < 0) |
1530 | return ret; | |
63733a0f | 1531 | prepare_read_connect(con); |
4e7a5dcd | 1532 | break; |
31b8006e SW |
1533 | |
1534 | case CEPH_MSGR_TAG_RESETSESSION: | |
1535 | /* | |
1536 | * If we connected with a large connect_seq but the peer | |
1537 | * has no record of a session with us (no connection, or | |
1538 | * connect_seq == 0), they will send RESETSESION to indicate | |
1539 | * that they must have reset their session, and may have | |
1540 | * dropped messages. | |
1541 | */ | |
1542 | dout("process_connect got RESET peer seq %u\n", | |
a16cb1f7 | 1543 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
1544 | pr_err("%s%lld %s connection reset\n", |
1545 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1546 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1547 | reset_connection(con); |
5a0f8fdd AE |
1548 | ret = prepare_write_connect(con); |
1549 | if (ret < 0) | |
1550 | return ret; | |
31b8006e SW |
1551 | prepare_read_connect(con); |
1552 | ||
1553 | /* Tell ceph about it. */ | |
ec302645 | 1554 | mutex_unlock(&con->mutex); |
31b8006e SW |
1555 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1556 | if (con->ops->peer_reset) | |
1557 | con->ops->peer_reset(con); | |
ec302645 | 1558 | mutex_lock(&con->mutex); |
0da5d703 SW |
1559 | if (test_bit(CLOSED, &con->state) || |
1560 | test_bit(OPENING, &con->state)) | |
1561 | return -EAGAIN; | |
31b8006e SW |
1562 | break; |
1563 | ||
1564 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1565 | /* | |
1566 | * If we sent a smaller connect_seq than the peer has, try | |
1567 | * again with a larger value. | |
1568 | */ | |
a16cb1f7 | 1569 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 1570 | le32_to_cpu(con->out_connect.connect_seq), |
a16cb1f7 SW |
1571 | le32_to_cpu(con->in_reply.connect_seq)); |
1572 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
5a0f8fdd AE |
1573 | ret = prepare_write_connect(con); |
1574 | if (ret < 0) | |
1575 | return ret; | |
31b8006e SW |
1576 | prepare_read_connect(con); |
1577 | break; | |
1578 | ||
1579 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1580 | /* | |
1581 | * If we sent a smaller global_seq than the peer has, try | |
1582 | * again with a larger value. | |
1583 | */ | |
eed0ef2c | 1584 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 1585 | con->peer_global_seq, |
a16cb1f7 | 1586 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 1587 | get_global_seq(con->msgr, |
a16cb1f7 | 1588 | le32_to_cpu(con->in_reply.global_seq)); |
5a0f8fdd AE |
1589 | ret = prepare_write_connect(con); |
1590 | if (ret < 0) | |
1591 | return ret; | |
31b8006e SW |
1592 | prepare_read_connect(con); |
1593 | break; | |
1594 | ||
1595 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1596 | if (req_feat & ~server_feat) { |
1597 | pr_err("%s%lld %s protocol feature mismatch," | |
1598 | " my required %llx > server's %llx, need %llx\n", | |
1599 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1600 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1601 | req_feat, server_feat, req_feat & ~server_feat); |
1602 | con->error_msg = "missing required protocol features"; | |
1603 | fail_protocol(con); | |
1604 | return -1; | |
1605 | } | |
3ec50d18 | 1606 | clear_bit(NEGOTIATING, &con->state); |
e27947c7 | 1607 | set_bit(CONNECTED, &con->state); |
31b8006e SW |
1608 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1609 | con->connect_seq++; | |
aba558e2 | 1610 | con->peer_features = server_feat; |
31b8006e SW |
1611 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1612 | con->peer_global_seq, | |
1613 | le32_to_cpu(con->in_reply.connect_seq), | |
1614 | con->connect_seq); | |
1615 | WARN_ON(con->connect_seq != | |
1616 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1617 | |
1618 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
928443cd | 1619 | set_bit(LOSSYTX, &con->flags); |
92ac41d0 | 1620 | |
31b8006e SW |
1621 | prepare_read_tag(con); |
1622 | break; | |
1623 | ||
1624 | case CEPH_MSGR_TAG_WAIT: | |
1625 | /* | |
1626 | * If there is a connection race (we are opening | |
1627 | * connections to each other), one of us may just have | |
1628 | * to WAIT. This shouldn't happen if we are the | |
1629 | * client. | |
1630 | */ | |
04177882 SW |
1631 | pr_err("process_connect got WAIT as client\n"); |
1632 | con->error_msg = "protocol error, got WAIT as client"; | |
1633 | return -1; | |
31b8006e SW |
1634 | |
1635 | default: | |
1636 | pr_err("connect protocol error, will retry\n"); | |
1637 | con->error_msg = "protocol error, garbage tag during connect"; | |
1638 | return -1; | |
1639 | } | |
1640 | return 0; | |
1641 | } | |
1642 | ||
1643 | ||
1644 | /* | |
1645 | * read (part of) an ack | |
1646 | */ | |
1647 | static int read_partial_ack(struct ceph_connection *con) | |
1648 | { | |
fd51653f AE |
1649 | int size = sizeof (con->in_temp_ack); |
1650 | int end = size; | |
1651 | ||
1652 | return read_partial(con, end, size, &con->in_temp_ack); | |
31b8006e SW |
1653 | } |
1654 | ||
1655 | ||
1656 | /* | |
1657 | * We can finally discard anything that's been acked. | |
1658 | */ | |
1659 | static void process_ack(struct ceph_connection *con) | |
1660 | { | |
1661 | struct ceph_msg *m; | |
1662 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1663 | u64 seq; | |
1664 | ||
31b8006e SW |
1665 | while (!list_empty(&con->out_sent)) { |
1666 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1667 | list_head); | |
1668 | seq = le64_to_cpu(m->hdr.seq); | |
1669 | if (seq > ack) | |
1670 | break; | |
1671 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1672 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1673 | m->ack_stamp = jiffies; |
31b8006e SW |
1674 | ceph_msg_remove(m); |
1675 | } | |
31b8006e SW |
1676 | prepare_read_tag(con); |
1677 | } | |
1678 | ||
1679 | ||
1680 | ||
1681 | ||
2450418c | 1682 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1683 | struct kvec *section, |
1684 | unsigned int sec_len, u32 *crc) | |
2450418c | 1685 | { |
68b4476b | 1686 | int ret, left; |
2450418c YS |
1687 | |
1688 | BUG_ON(!section); | |
1689 | ||
1690 | while (section->iov_len < sec_len) { | |
1691 | BUG_ON(section->iov_base == NULL); | |
1692 | left = sec_len - section->iov_len; | |
1693 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1694 | section->iov_len, left); | |
1695 | if (ret <= 0) | |
1696 | return ret; | |
1697 | section->iov_len += ret; | |
2450418c | 1698 | } |
fe3ad593 AE |
1699 | if (section->iov_len == sec_len) |
1700 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1701 | |
2450418c YS |
1702 | return 1; |
1703 | } | |
31b8006e | 1704 | |
1c20f2d2 AE |
1705 | static bool ceph_con_in_msg_alloc(struct ceph_connection *con, |
1706 | struct ceph_msg_header *hdr); | |
68b4476b YS |
1707 | |
1708 | ||
1709 | static int read_partial_message_pages(struct ceph_connection *con, | |
1710 | struct page **pages, | |
95c96174 | 1711 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1712 | { |
1713 | void *p; | |
1714 | int ret; | |
1715 | int left; | |
1716 | ||
1717 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1718 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1719 | /* (page) data */ | |
1720 | BUG_ON(pages == NULL); | |
1721 | p = kmap(pages[con->in_msg_pos.page]); | |
1722 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1723 | left); | |
bca064d2 | 1724 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1725 | con->in_data_crc = |
1726 | crc32c(con->in_data_crc, | |
1727 | p + con->in_msg_pos.page_pos, ret); | |
1728 | kunmap(pages[con->in_msg_pos.page]); | |
1729 | if (ret <= 0) | |
1730 | return ret; | |
1731 | con->in_msg_pos.data_pos += ret; | |
1732 | con->in_msg_pos.page_pos += ret; | |
1733 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1734 | con->in_msg_pos.page_pos = 0; | |
1735 | con->in_msg_pos.page++; | |
1736 | } | |
1737 | ||
1738 | return ret; | |
1739 | } | |
1740 | ||
1741 | #ifdef CONFIG_BLOCK | |
1742 | static int read_partial_message_bio(struct ceph_connection *con, | |
1743 | struct bio **bio_iter, int *bio_seg, | |
95c96174 | 1744 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1745 | { |
1746 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1747 | void *p; | |
1748 | int ret, left; | |
1749 | ||
68b4476b YS |
1750 | left = min((int)(data_len - con->in_msg_pos.data_pos), |
1751 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1752 | ||
1753 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1754 | ||
1755 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1756 | left); | |
bca064d2 | 1757 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1758 | con->in_data_crc = |
1759 | crc32c(con->in_data_crc, | |
1760 | p + con->in_msg_pos.page_pos, ret); | |
1761 | kunmap(bv->bv_page); | |
1762 | if (ret <= 0) | |
1763 | return ret; | |
1764 | con->in_msg_pos.data_pos += ret; | |
1765 | con->in_msg_pos.page_pos += ret; | |
1766 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1767 | con->in_msg_pos.page_pos = 0; | |
1768 | iter_bio_next(bio_iter, bio_seg); | |
1769 | } | |
1770 | ||
1771 | return ret; | |
1772 | } | |
1773 | #endif | |
1774 | ||
31b8006e SW |
1775 | /* |
1776 | * read (part of) a message. | |
1777 | */ | |
1778 | static int read_partial_message(struct ceph_connection *con) | |
1779 | { | |
1780 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1781 | int size; |
1782 | int end; | |
31b8006e | 1783 | int ret; |
95c96174 | 1784 | unsigned int front_len, middle_len, data_len; |
37675b0f | 1785 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 1786 | u64 seq; |
fe3ad593 | 1787 | u32 crc; |
31b8006e SW |
1788 | |
1789 | dout("read_partial_message con %p msg %p\n", con, m); | |
1790 | ||
1791 | /* header */ | |
fd51653f AE |
1792 | size = sizeof (con->in_hdr); |
1793 | end = size; | |
1794 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1795 | if (ret <= 0) |
1796 | return ret; | |
fe3ad593 AE |
1797 | |
1798 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1799 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1800 | pr_err("read_partial_message bad hdr " | |
1801 | " crc %u != expected %u\n", | |
1802 | crc, con->in_hdr.crc); | |
1803 | return -EBADMSG; | |
1804 | } | |
1805 | ||
31b8006e SW |
1806 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1807 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1808 | return -EIO; | |
1809 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1810 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1811 | return -EIO; | |
1812 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1813 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1814 | return -EIO; | |
1815 | ||
ae18756b SW |
1816 | /* verify seq# */ |
1817 | seq = le64_to_cpu(con->in_hdr.seq); | |
1818 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1819 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1820 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1821 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1822 | seq, con->in_seq + 1); |
1823 | con->in_base_pos = -front_len - middle_len - data_len - | |
1824 | sizeof(m->footer); | |
1825 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1826 | return 0; |
1827 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1828 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1829 | seq, con->in_seq + 1); | |
1830 | con->error_msg = "bad message sequence # for incoming message"; | |
1831 | return -EBADMSG; | |
1832 | } | |
1833 | ||
31b8006e SW |
1834 | /* allocate message? */ |
1835 | if (!con->in_msg) { | |
1836 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1837 | con->in_hdr.front_len, con->in_hdr.data_len); | |
1c20f2d2 | 1838 | if (ceph_con_in_msg_alloc(con, &con->in_hdr)) { |
31b8006e | 1839 | /* skip this message */ |
a79832f2 | 1840 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1841 | BUG_ON(con->in_msg); |
31b8006e SW |
1842 | con->in_base_pos = -front_len - middle_len - data_len - |
1843 | sizeof(m->footer); | |
1844 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1845 | con->in_seq++; |
31b8006e SW |
1846 | return 0; |
1847 | } | |
a79832f2 | 1848 | if (!con->in_msg) { |
5b3a4db3 SW |
1849 | con->error_msg = |
1850 | "error allocating memory for incoming message"; | |
a79832f2 | 1851 | return -ENOMEM; |
31b8006e | 1852 | } |
38941f80 AE |
1853 | |
1854 | BUG_ON(con->in_msg->con != con); | |
31b8006e SW |
1855 | m = con->in_msg; |
1856 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1857 | if (m->middle) |
1858 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1859 | |
1860 | con->in_msg_pos.page = 0; | |
68b4476b | 1861 | if (m->pages) |
c5c6b19d | 1862 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1863 | else |
1864 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1865 | con->in_msg_pos.data_pos = 0; |
31b8006e SW |
1866 | } |
1867 | ||
1868 | /* front */ | |
2450418c YS |
1869 | ret = read_partial_message_section(con, &m->front, front_len, |
1870 | &con->in_front_crc); | |
1871 | if (ret <= 0) | |
1872 | return ret; | |
31b8006e SW |
1873 | |
1874 | /* middle */ | |
2450418c | 1875 | if (m->middle) { |
213c99ee SW |
1876 | ret = read_partial_message_section(con, &m->middle->vec, |
1877 | middle_len, | |
2450418c | 1878 | &con->in_middle_crc); |
31b8006e SW |
1879 | if (ret <= 0) |
1880 | return ret; | |
31b8006e | 1881 | } |
68b4476b YS |
1882 | #ifdef CONFIG_BLOCK |
1883 | if (m->bio && !m->bio_iter) | |
1884 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1885 | #endif | |
31b8006e SW |
1886 | |
1887 | /* (page) data */ | |
31b8006e | 1888 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1889 | if (m->pages) { |
1890 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1891 | data_len, do_datacrc); |
68b4476b YS |
1892 | if (ret <= 0) |
1893 | return ret; | |
1894 | #ifdef CONFIG_BLOCK | |
1895 | } else if (m->bio) { | |
1896 | ||
1897 | ret = read_partial_message_bio(con, | |
1898 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1899 | data_len, do_datacrc); |
68b4476b YS |
1900 | if (ret <= 0) |
1901 | return ret; | |
1902 | #endif | |
1903 | } else { | |
1904 | BUG_ON(1); | |
31b8006e SW |
1905 | } |
1906 | } | |
1907 | ||
31b8006e | 1908 | /* footer */ |
fd51653f AE |
1909 | size = sizeof (m->footer); |
1910 | end += size; | |
1911 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
1912 | if (ret <= 0) |
1913 | return ret; | |
1914 | ||
31b8006e SW |
1915 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
1916 | m, front_len, m->footer.front_crc, middle_len, | |
1917 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1918 | ||
1919 | /* crc ok? */ | |
1920 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1921 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1922 | m, con->in_front_crc, m->footer.front_crc); | |
1923 | return -EBADMSG; | |
1924 | } | |
1925 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1926 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1927 | m, con->in_middle_crc, m->footer.middle_crc); | |
1928 | return -EBADMSG; | |
1929 | } | |
bca064d2 | 1930 | if (do_datacrc && |
31b8006e SW |
1931 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
1932 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1933 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1934 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1935 | return -EBADMSG; | |
1936 | } | |
1937 | ||
1938 | return 1; /* done! */ | |
1939 | } | |
1940 | ||
1941 | /* | |
1942 | * Process message. This happens in the worker thread. The callback should | |
1943 | * be careful not to do anything that waits on other incoming messages or it | |
1944 | * may deadlock. | |
1945 | */ | |
1946 | static void process_message(struct ceph_connection *con) | |
1947 | { | |
5e095e8b | 1948 | struct ceph_msg *msg; |
31b8006e | 1949 | |
38941f80 AE |
1950 | BUG_ON(con->in_msg->con != con); |
1951 | con->in_msg->con = NULL; | |
5e095e8b | 1952 | msg = con->in_msg; |
31b8006e | 1953 | con->in_msg = NULL; |
36eb71aa | 1954 | con->ops->put(con); |
31b8006e SW |
1955 | |
1956 | /* if first message, set peer_name */ | |
1957 | if (con->peer_name.type == 0) | |
dbad185d | 1958 | con->peer_name = msg->hdr.src; |
31b8006e | 1959 | |
31b8006e | 1960 | con->in_seq++; |
ec302645 | 1961 | mutex_unlock(&con->mutex); |
31b8006e SW |
1962 | |
1963 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1964 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1965 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1966 | le16_to_cpu(msg->hdr.type), |
1967 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1968 | le32_to_cpu(msg->hdr.front_len), | |
1969 | le32_to_cpu(msg->hdr.data_len), | |
1970 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1971 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1972 | |
1973 | mutex_lock(&con->mutex); | |
31b8006e SW |
1974 | prepare_read_tag(con); |
1975 | } | |
1976 | ||
1977 | ||
1978 | /* | |
1979 | * Write something to the socket. Called in a worker thread when the | |
1980 | * socket appears to be writeable and we have something ready to send. | |
1981 | */ | |
1982 | static int try_write(struct ceph_connection *con) | |
1983 | { | |
31b8006e SW |
1984 | int ret = 1; |
1985 | ||
d59315ca | 1986 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 1987 | |
31b8006e SW |
1988 | more: |
1989 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
1990 | ||
1991 | /* open the socket first? */ | |
1992 | if (con->sock == NULL) { | |
a5988c49 AE |
1993 | set_bit(CONNECTING, &con->state); |
1994 | ||
e2200423 | 1995 | con_out_kvec_reset(con); |
e825a66d | 1996 | prepare_write_banner(con); |
eed0ef2c | 1997 | prepare_read_banner(con); |
31b8006e | 1998 | |
cf3e5c40 | 1999 | BUG_ON(con->in_msg); |
31b8006e SW |
2000 | con->in_tag = CEPH_MSGR_TAG_READY; |
2001 | dout("try_write initiating connect on %p new state %lu\n", | |
2002 | con, con->state); | |
41617d0c AE |
2003 | ret = ceph_tcp_connect(con); |
2004 | if (ret < 0) { | |
31b8006e | 2005 | con->error_msg = "connect error"; |
31b8006e SW |
2006 | goto out; |
2007 | } | |
2008 | } | |
2009 | ||
2010 | more_kvec: | |
2011 | /* kvec data queued? */ | |
2012 | if (con->out_skip) { | |
2013 | ret = write_partial_skip(con); | |
2014 | if (ret <= 0) | |
42961d23 | 2015 | goto out; |
31b8006e SW |
2016 | } |
2017 | if (con->out_kvec_left) { | |
2018 | ret = write_partial_kvec(con); | |
2019 | if (ret <= 0) | |
42961d23 | 2020 | goto out; |
31b8006e SW |
2021 | } |
2022 | ||
2023 | /* msg pages? */ | |
2024 | if (con->out_msg) { | |
c86a2930 SW |
2025 | if (con->out_msg_done) { |
2026 | ceph_msg_put(con->out_msg); | |
2027 | con->out_msg = NULL; /* we're done with this one */ | |
2028 | goto do_next; | |
2029 | } | |
2030 | ||
31b8006e SW |
2031 | ret = write_partial_msg_pages(con); |
2032 | if (ret == 1) | |
2033 | goto more_kvec; /* we need to send the footer, too! */ | |
2034 | if (ret == 0) | |
42961d23 | 2035 | goto out; |
31b8006e SW |
2036 | if (ret < 0) { |
2037 | dout("try_write write_partial_msg_pages err %d\n", | |
2038 | ret); | |
42961d23 | 2039 | goto out; |
31b8006e SW |
2040 | } |
2041 | } | |
2042 | ||
c86a2930 | 2043 | do_next: |
7593af92 AE |
2044 | if (!test_bit(CONNECTING, &con->state) && |
2045 | !test_bit(NEGOTIATING, &con->state)) { | |
31b8006e SW |
2046 | /* is anything else pending? */ |
2047 | if (!list_empty(&con->out_queue)) { | |
2048 | prepare_write_message(con); | |
2049 | goto more; | |
2050 | } | |
2051 | if (con->in_seq > con->in_seq_acked) { | |
2052 | prepare_write_ack(con); | |
2053 | goto more; | |
2054 | } | |
928443cd | 2055 | if (test_and_clear_bit(KEEPALIVE_PENDING, &con->flags)) { |
31b8006e SW |
2056 | prepare_write_keepalive(con); |
2057 | goto more; | |
2058 | } | |
2059 | } | |
2060 | ||
2061 | /* Nothing to do! */ | |
928443cd | 2062 | clear_bit(WRITE_PENDING, &con->flags); |
31b8006e | 2063 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2064 | ret = 0; |
2065 | out: | |
42961d23 | 2066 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2067 | return ret; |
2068 | } | |
2069 | ||
2070 | ||
2071 | ||
2072 | /* | |
2073 | * Read what we can from the socket. | |
2074 | */ | |
2075 | static int try_read(struct ceph_connection *con) | |
2076 | { | |
31b8006e SW |
2077 | int ret = -1; |
2078 | ||
2079 | if (!con->sock) | |
2080 | return 0; | |
2081 | ||
2082 | if (test_bit(STANDBY, &con->state)) | |
2083 | return 0; | |
2084 | ||
2085 | dout("try_read start on %p\n", con); | |
ec302645 | 2086 | |
31b8006e SW |
2087 | more: |
2088 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, | |
2089 | con->in_base_pos); | |
0da5d703 SW |
2090 | |
2091 | /* | |
2092 | * process_connect and process_message drop and re-take | |
2093 | * con->mutex. make sure we handle a racing close or reopen. | |
2094 | */ | |
2095 | if (test_bit(CLOSED, &con->state) || | |
2096 | test_bit(OPENING, &con->state)) { | |
2097 | ret = -EAGAIN; | |
2098 | goto out; | |
2099 | } | |
2100 | ||
31b8006e | 2101 | if (test_bit(CONNECTING, &con->state)) { |
7593af92 AE |
2102 | dout("try_read connecting\n"); |
2103 | ret = read_partial_banner(con); | |
2104 | if (ret <= 0) | |
ab166d5a | 2105 | goto out; |
7593af92 AE |
2106 | ret = process_banner(con); |
2107 | if (ret < 0) | |
2108 | goto out; | |
2109 | ||
2110 | clear_bit(CONNECTING, &con->state); | |
2111 | set_bit(NEGOTIATING, &con->state); | |
2112 | ||
2113 | /* Banner is good, exchange connection info */ | |
2114 | ret = prepare_write_connect(con); | |
2115 | if (ret < 0) | |
2116 | goto out; | |
2117 | prepare_read_connect(con); | |
2118 | ||
2119 | /* Send connection info before awaiting response */ | |
2120 | goto out; | |
2121 | } | |
2122 | ||
2123 | if (test_bit(NEGOTIATING, &con->state)) { | |
2124 | dout("try_read negotiating\n"); | |
31b8006e SW |
2125 | ret = read_partial_connect(con); |
2126 | if (ret <= 0) | |
31b8006e | 2127 | goto out; |
98bdb0aa SW |
2128 | ret = process_connect(con); |
2129 | if (ret < 0) | |
2130 | goto out; | |
31b8006e SW |
2131 | goto more; |
2132 | } | |
2133 | ||
2134 | if (con->in_base_pos < 0) { | |
2135 | /* | |
2136 | * skipping + discarding content. | |
2137 | * | |
2138 | * FIXME: there must be a better way to do this! | |
2139 | */ | |
84495f49 AE |
2140 | static char buf[SKIP_BUF_SIZE]; |
2141 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2142 | ||
31b8006e SW |
2143 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2144 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2145 | if (ret <= 0) | |
98bdb0aa | 2146 | goto out; |
31b8006e SW |
2147 | con->in_base_pos += ret; |
2148 | if (con->in_base_pos) | |
2149 | goto more; | |
2150 | } | |
2151 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2152 | /* | |
2153 | * what's next? | |
2154 | */ | |
2155 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2156 | if (ret <= 0) | |
98bdb0aa | 2157 | goto out; |
31b8006e SW |
2158 | dout("try_read got tag %d\n", (int)con->in_tag); |
2159 | switch (con->in_tag) { | |
2160 | case CEPH_MSGR_TAG_MSG: | |
2161 | prepare_read_message(con); | |
2162 | break; | |
2163 | case CEPH_MSGR_TAG_ACK: | |
2164 | prepare_read_ack(con); | |
2165 | break; | |
2166 | case CEPH_MSGR_TAG_CLOSE: | |
e27947c7 | 2167 | clear_bit(CONNECTED, &con->state); |
31b8006e | 2168 | set_bit(CLOSED, &con->state); /* fixme */ |
98bdb0aa | 2169 | goto out; |
31b8006e SW |
2170 | default: |
2171 | goto bad_tag; | |
2172 | } | |
2173 | } | |
2174 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2175 | ret = read_partial_message(con); | |
2176 | if (ret <= 0) { | |
2177 | switch (ret) { | |
2178 | case -EBADMSG: | |
2179 | con->error_msg = "bad crc"; | |
2180 | ret = -EIO; | |
98bdb0aa | 2181 | break; |
31b8006e SW |
2182 | case -EIO: |
2183 | con->error_msg = "io error"; | |
98bdb0aa | 2184 | break; |
31b8006e | 2185 | } |
98bdb0aa | 2186 | goto out; |
31b8006e SW |
2187 | } |
2188 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2189 | goto more; | |
2190 | process_message(con); | |
2191 | goto more; | |
2192 | } | |
2193 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2194 | ret = read_partial_ack(con); | |
2195 | if (ret <= 0) | |
98bdb0aa | 2196 | goto out; |
31b8006e SW |
2197 | process_ack(con); |
2198 | goto more; | |
2199 | } | |
2200 | ||
31b8006e | 2201 | out: |
98bdb0aa | 2202 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2203 | return ret; |
2204 | ||
2205 | bad_tag: | |
2206 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2207 | con->error_msg = "protocol error, garbage tag"; | |
2208 | ret = -1; | |
2209 | goto out; | |
2210 | } | |
2211 | ||
2212 | ||
2213 | /* | |
2214 | * Atomically queue work on a connection. Bump @con reference to | |
2215 | * avoid races with connection teardown. | |
31b8006e SW |
2216 | */ |
2217 | static void queue_con(struct ceph_connection *con) | |
2218 | { | |
31b8006e SW |
2219 | if (!con->ops->get(con)) { |
2220 | dout("queue_con %p ref count 0\n", con); | |
2221 | return; | |
2222 | } | |
2223 | ||
f363e45f | 2224 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2225 | dout("queue_con %p - already queued\n", con); |
2226 | con->ops->put(con); | |
2227 | } else { | |
2228 | dout("queue_con %p\n", con); | |
2229 | } | |
2230 | } | |
2231 | ||
2232 | /* | |
2233 | * Do some work on a connection. Drop a connection ref when we're done. | |
2234 | */ | |
2235 | static void con_work(struct work_struct *work) | |
2236 | { | |
2237 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2238 | work.work); | |
0da5d703 | 2239 | int ret; |
31b8006e | 2240 | |
9dd4658d | 2241 | mutex_lock(&con->mutex); |
0da5d703 | 2242 | restart: |
188048bc | 2243 | if (test_and_clear_bit(SOCK_CLOSED, &con->flags)) { |
e27947c7 AE |
2244 | if (test_and_clear_bit(CONNECTED, &con->state)) |
2245 | con->error_msg = "socket closed"; | |
7593af92 AE |
2246 | else if (test_and_clear_bit(NEGOTIATING, &con->state)) |
2247 | con->error_msg = "negotiation failed"; | |
2248 | else if (test_and_clear_bit(CONNECTING, &con->state)) | |
188048bc | 2249 | con->error_msg = "connection failed"; |
7593af92 | 2250 | else |
e27947c7 | 2251 | con->error_msg = "unrecognized con state"; |
188048bc AE |
2252 | goto fault; |
2253 | } | |
2254 | ||
928443cd | 2255 | if (test_and_clear_bit(BACKOFF, &con->flags)) { |
60bf8bf8 SW |
2256 | dout("con_work %p backing off\n", con); |
2257 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2258 | round_jiffies_relative(con->delay))) { | |
2259 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2260 | mutex_unlock(&con->mutex); | |
2261 | return; | |
2262 | } else { | |
2263 | con->ops->put(con); | |
2264 | dout("con_work %p FAILED to back off %lu\n", con, | |
2265 | con->delay); | |
2266 | } | |
2267 | } | |
9dd4658d | 2268 | |
e00de341 SW |
2269 | if (test_bit(STANDBY, &con->state)) { |
2270 | dout("con_work %p STANDBY\n", con); | |
2271 | goto done; | |
2272 | } | |
31b8006e SW |
2273 | if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */ |
2274 | dout("con_work CLOSED\n"); | |
2275 | con_close_socket(con); | |
2276 | goto done; | |
2277 | } | |
2278 | if (test_and_clear_bit(OPENING, &con->state)) { | |
2279 | /* reopen w/ new peer */ | |
2280 | dout("con_work OPENING\n"); | |
2281 | con_close_socket(con); | |
2282 | } | |
2283 | ||
0da5d703 SW |
2284 | ret = try_read(con); |
2285 | if (ret == -EAGAIN) | |
2286 | goto restart; | |
2287 | if (ret < 0) | |
2288 | goto fault; | |
2289 | ||
2290 | ret = try_write(con); | |
2291 | if (ret == -EAGAIN) | |
2292 | goto restart; | |
2293 | if (ret < 0) | |
2294 | goto fault; | |
31b8006e SW |
2295 | |
2296 | done: | |
9dd4658d | 2297 | mutex_unlock(&con->mutex); |
9dd4658d | 2298 | done_unlocked: |
31b8006e | 2299 | con->ops->put(con); |
0da5d703 SW |
2300 | return; |
2301 | ||
2302 | fault: | |
2303 | mutex_unlock(&con->mutex); | |
2304 | ceph_fault(con); /* error/fault path */ | |
2305 | goto done_unlocked; | |
31b8006e SW |
2306 | } |
2307 | ||
2308 | ||
2309 | /* | |
2310 | * Generic error/fault handler. A retry mechanism is used with | |
2311 | * exponential backoff | |
2312 | */ | |
2313 | static void ceph_fault(struct ceph_connection *con) | |
2314 | { | |
2315 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2316 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2317 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2318 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2319 | |
928443cd | 2320 | if (test_bit(LOSSYTX, &con->flags)) { |
31b8006e SW |
2321 | dout("fault on LOSSYTX channel\n"); |
2322 | goto out; | |
2323 | } | |
2324 | ||
ec302645 | 2325 | mutex_lock(&con->mutex); |
91e45ce3 SW |
2326 | if (test_bit(CLOSED, &con->state)) |
2327 | goto out_unlock; | |
ec302645 | 2328 | |
31b8006e | 2329 | con_close_socket(con); |
5e095e8b SW |
2330 | |
2331 | if (con->in_msg) { | |
38941f80 AE |
2332 | BUG_ON(con->in_msg->con != con); |
2333 | con->in_msg->con = NULL; | |
5e095e8b SW |
2334 | ceph_msg_put(con->in_msg); |
2335 | con->in_msg = NULL; | |
36eb71aa | 2336 | con->ops->put(con); |
5e095e8b | 2337 | } |
31b8006e | 2338 | |
e80a52d1 SW |
2339 | /* Requeue anything that hasn't been acked */ |
2340 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2341 | |
e76661d0 SW |
2342 | /* If there are no messages queued or keepalive pending, place |
2343 | * the connection in a STANDBY state */ | |
2344 | if (list_empty(&con->out_queue) && | |
928443cd | 2345 | !test_bit(KEEPALIVE_PENDING, &con->flags)) { |
e00de341 | 2346 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
928443cd | 2347 | clear_bit(WRITE_PENDING, &con->flags); |
31b8006e | 2348 | set_bit(STANDBY, &con->state); |
e80a52d1 SW |
2349 | } else { |
2350 | /* retry after a delay. */ | |
2351 | if (con->delay == 0) | |
2352 | con->delay = BASE_DELAY_INTERVAL; | |
2353 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2354 | con->delay *= 2; | |
e80a52d1 SW |
2355 | con->ops->get(con); |
2356 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2357 | round_jiffies_relative(con->delay))) { |
2358 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2359 | } else { | |
e80a52d1 | 2360 | con->ops->put(con); |
60bf8bf8 SW |
2361 | dout("fault failed to queue %p delay %lu, backoff\n", |
2362 | con, con->delay); | |
2363 | /* | |
2364 | * In many cases we see a socket state change | |
2365 | * while con_work is running and end up | |
2366 | * queuing (non-delayed) work, such that we | |
2367 | * can't backoff with a delay. Set a flag so | |
2368 | * that when con_work restarts we schedule the | |
2369 | * delay then. | |
2370 | */ | |
928443cd | 2371 | set_bit(BACKOFF, &con->flags); |
60bf8bf8 | 2372 | } |
31b8006e SW |
2373 | } |
2374 | ||
91e45ce3 SW |
2375 | out_unlock: |
2376 | mutex_unlock(&con->mutex); | |
31b8006e | 2377 | out: |
161fd65a SW |
2378 | /* |
2379 | * in case we faulted due to authentication, invalidate our | |
2380 | * current tickets so that we can get new ones. | |
213c99ee | 2381 | */ |
161fd65a SW |
2382 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2383 | dout("calling invalidate_authorizer()\n"); | |
2384 | con->ops->invalidate_authorizer(con); | |
2385 | } | |
2386 | ||
31b8006e SW |
2387 | if (con->ops->fault) |
2388 | con->ops->fault(con); | |
2389 | } | |
2390 | ||
2391 | ||
2392 | ||
2393 | /* | |
15d9882c | 2394 | * initialize a new messenger instance |
31b8006e | 2395 | */ |
15d9882c AE |
2396 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2397 | struct ceph_entity_addr *myaddr, | |
2398 | u32 supported_features, | |
2399 | u32 required_features, | |
2400 | bool nocrc) | |
31b8006e | 2401 | { |
3d14c5d2 YS |
2402 | msgr->supported_features = supported_features; |
2403 | msgr->required_features = required_features; | |
2404 | ||
31b8006e SW |
2405 | spin_lock_init(&msgr->global_seq_lock); |
2406 | ||
31b8006e SW |
2407 | if (myaddr) |
2408 | msgr->inst.addr = *myaddr; | |
2409 | ||
2410 | /* select a random nonce */ | |
ac8839d7 | 2411 | msgr->inst.addr.type = 0; |
103e2d3a | 2412 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2413 | encode_my_addr(msgr); |
15d9882c | 2414 | msgr->nocrc = nocrc; |
31b8006e | 2415 | |
15d9882c | 2416 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2417 | } |
15d9882c | 2418 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2419 | |
e00de341 SW |
2420 | static void clear_standby(struct ceph_connection *con) |
2421 | { | |
2422 | /* come back from STANDBY? */ | |
2423 | if (test_and_clear_bit(STANDBY, &con->state)) { | |
2424 | mutex_lock(&con->mutex); | |
2425 | dout("clear_standby %p and ++connect_seq\n", con); | |
2426 | con->connect_seq++; | |
928443cd AE |
2427 | WARN_ON(test_bit(WRITE_PENDING, &con->flags)); |
2428 | WARN_ON(test_bit(KEEPALIVE_PENDING, &con->flags)); | |
e00de341 SW |
2429 | mutex_unlock(&con->mutex); |
2430 | } | |
2431 | } | |
2432 | ||
31b8006e SW |
2433 | /* |
2434 | * Queue up an outgoing message on the given connection. | |
2435 | */ | |
2436 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2437 | { | |
2438 | if (test_bit(CLOSED, &con->state)) { | |
2439 | dout("con_send %p closed, dropping %p\n", con, msg); | |
2440 | ceph_msg_put(msg); | |
2441 | return; | |
2442 | } | |
2443 | ||
2444 | /* set src+dst */ | |
dbad185d | 2445 | msg->hdr.src = con->msgr->inst.name; |
31b8006e | 2446 | |
3ca02ef9 SW |
2447 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
2448 | ||
e84346b7 SW |
2449 | msg->needs_out_seq = true; |
2450 | ||
31b8006e | 2451 | /* queue */ |
ec302645 | 2452 | mutex_lock(&con->mutex); |
92ce034b | 2453 | |
38941f80 | 2454 | BUG_ON(msg->con != NULL); |
36eb71aa | 2455 | msg->con = con->ops->get(con); |
92ce034b AE |
2456 | BUG_ON(msg->con == NULL); |
2457 | ||
31b8006e SW |
2458 | BUG_ON(!list_empty(&msg->list_head)); |
2459 | list_add_tail(&msg->list_head, &con->out_queue); | |
2460 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2461 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2462 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2463 | le32_to_cpu(msg->hdr.front_len), | |
2464 | le32_to_cpu(msg->hdr.middle_len), | |
2465 | le32_to_cpu(msg->hdr.data_len)); | |
ec302645 | 2466 | mutex_unlock(&con->mutex); |
31b8006e SW |
2467 | |
2468 | /* if there wasn't anything waiting to send before, queue | |
2469 | * new work */ | |
e00de341 | 2470 | clear_standby(con); |
928443cd | 2471 | if (test_and_set_bit(WRITE_PENDING, &con->flags) == 0) |
31b8006e SW |
2472 | queue_con(con); |
2473 | } | |
3d14c5d2 | 2474 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2475 | |
2476 | /* | |
2477 | * Revoke a message that was previously queued for send | |
2478 | */ | |
6740a845 | 2479 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2480 | { |
6740a845 AE |
2481 | struct ceph_connection *con = msg->con; |
2482 | ||
2483 | if (!con) | |
2484 | return; /* Message not in our possession */ | |
2485 | ||
ec302645 | 2486 | mutex_lock(&con->mutex); |
31b8006e | 2487 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2488 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2489 | list_del_init(&msg->list_head); |
38941f80 | 2490 | BUG_ON(msg->con == NULL); |
36eb71aa | 2491 | msg->con->ops->put(msg->con); |
38941f80 | 2492 | msg->con = NULL; |
92ce034b | 2493 | msg->hdr.seq = 0; |
38941f80 | 2494 | |
31b8006e | 2495 | ceph_msg_put(msg); |
ed98adad SW |
2496 | } |
2497 | if (con->out_msg == msg) { | |
38941f80 | 2498 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2499 | con->out_msg = NULL; |
31b8006e SW |
2500 | if (con->out_kvec_is_msg) { |
2501 | con->out_skip = con->out_kvec_bytes; | |
2502 | con->out_kvec_is_msg = false; | |
2503 | } | |
ed98adad | 2504 | msg->hdr.seq = 0; |
92ce034b AE |
2505 | |
2506 | ceph_msg_put(msg); | |
31b8006e | 2507 | } |
ec302645 | 2508 | mutex_unlock(&con->mutex); |
31b8006e SW |
2509 | } |
2510 | ||
350b1c32 | 2511 | /* |
0d59ab81 | 2512 | * Revoke a message that we may be reading data into |
350b1c32 | 2513 | */ |
8921d114 | 2514 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2515 | { |
8921d114 AE |
2516 | struct ceph_connection *con; |
2517 | ||
2518 | BUG_ON(msg == NULL); | |
2519 | if (!msg->con) { | |
2520 | dout("%s msg %p null con\n", __func__, msg); | |
2521 | ||
2522 | return; /* Message not in our possession */ | |
2523 | } | |
2524 | ||
2525 | con = msg->con; | |
350b1c32 | 2526 | mutex_lock(&con->mutex); |
8921d114 | 2527 | if (con->in_msg == msg) { |
95c96174 ED |
2528 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
2529 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
2530 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
2531 | |
2532 | /* skip rest of message */ | |
8921d114 AE |
2533 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
2534 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 2535 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
2536 | front_len - |
2537 | middle_len - | |
2538 | data_len - | |
350b1c32 | 2539 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2540 | ceph_msg_put(con->in_msg); |
2541 | con->in_msg = NULL; | |
2542 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2543 | con->in_seq++; |
350b1c32 | 2544 | } else { |
8921d114 AE |
2545 | dout("%s %p in_msg %p msg %p no-op\n", |
2546 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
2547 | } |
2548 | mutex_unlock(&con->mutex); | |
2549 | } | |
2550 | ||
31b8006e SW |
2551 | /* |
2552 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2553 | */ | |
2554 | void ceph_con_keepalive(struct ceph_connection *con) | |
2555 | { | |
e00de341 SW |
2556 | dout("con_keepalive %p\n", con); |
2557 | clear_standby(con); | |
928443cd AE |
2558 | if (test_and_set_bit(KEEPALIVE_PENDING, &con->flags) == 0 && |
2559 | test_and_set_bit(WRITE_PENDING, &con->flags) == 0) | |
31b8006e SW |
2560 | queue_con(con); |
2561 | } | |
3d14c5d2 | 2562 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2563 | |
2564 | ||
2565 | /* | |
2566 | * construct a new message with given type, size | |
2567 | * the new msg has a ref count of 1. | |
2568 | */ | |
b61c2763 SW |
2569 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2570 | bool can_fail) | |
31b8006e SW |
2571 | { |
2572 | struct ceph_msg *m; | |
2573 | ||
34d23762 | 2574 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2575 | if (m == NULL) |
2576 | goto out; | |
c2e552e7 | 2577 | kref_init(&m->kref); |
38941f80 AE |
2578 | |
2579 | m->con = NULL; | |
31b8006e SW |
2580 | INIT_LIST_HEAD(&m->list_head); |
2581 | ||
45c6ceb5 | 2582 | m->hdr.tid = 0; |
31b8006e | 2583 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2584 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2585 | m->hdr.version = 0; | |
31b8006e SW |
2586 | m->hdr.front_len = cpu_to_le32(front_len); |
2587 | m->hdr.middle_len = 0; | |
bb257664 SW |
2588 | m->hdr.data_len = 0; |
2589 | m->hdr.data_off = 0; | |
45c6ceb5 | 2590 | m->hdr.reserved = 0; |
31b8006e SW |
2591 | m->footer.front_crc = 0; |
2592 | m->footer.middle_crc = 0; | |
2593 | m->footer.data_crc = 0; | |
45c6ceb5 | 2594 | m->footer.flags = 0; |
31b8006e SW |
2595 | m->front_max = front_len; |
2596 | m->front_is_vmalloc = false; | |
2597 | m->more_to_follow = false; | |
c0d5f9db | 2598 | m->ack_stamp = 0; |
31b8006e SW |
2599 | m->pool = NULL; |
2600 | ||
ca20892d HC |
2601 | /* middle */ |
2602 | m->middle = NULL; | |
2603 | ||
2604 | /* data */ | |
2605 | m->nr_pages = 0; | |
2606 | m->page_alignment = 0; | |
2607 | m->pages = NULL; | |
2608 | m->pagelist = NULL; | |
2609 | m->bio = NULL; | |
2610 | m->bio_iter = NULL; | |
2611 | m->bio_seg = 0; | |
2612 | m->trail = NULL; | |
2613 | ||
31b8006e SW |
2614 | /* front */ |
2615 | if (front_len) { | |
2616 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2617 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2618 | PAGE_KERNEL); |
2619 | m->front_is_vmalloc = true; | |
2620 | } else { | |
34d23762 | 2621 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2622 | } |
2623 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2624 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2625 | front_len); |
2626 | goto out2; | |
2627 | } | |
2628 | } else { | |
2629 | m->front.iov_base = NULL; | |
2630 | } | |
2631 | m->front.iov_len = front_len; | |
2632 | ||
bb257664 | 2633 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2634 | return m; |
2635 | ||
2636 | out2: | |
2637 | ceph_msg_put(m); | |
2638 | out: | |
b61c2763 SW |
2639 | if (!can_fail) { |
2640 | pr_err("msg_new can't create type %d front %d\n", type, | |
2641 | front_len); | |
f0ed1b7c | 2642 | WARN_ON(1); |
b61c2763 SW |
2643 | } else { |
2644 | dout("msg_new can't create type %d front %d\n", type, | |
2645 | front_len); | |
2646 | } | |
a79832f2 | 2647 | return NULL; |
31b8006e | 2648 | } |
3d14c5d2 | 2649 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2650 | |
31b8006e SW |
2651 | /* |
2652 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2653 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2654 | * per-type header in the front and then gracefully fail (i.e., | |
2655 | * propagate the error to the caller based on info in the front) when | |
2656 | * the middle is too large. | |
2657 | */ | |
2450418c | 2658 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2659 | { |
2660 | int type = le16_to_cpu(msg->hdr.type); | |
2661 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2662 | ||
2663 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2664 | ceph_msg_type_name(type), middle_len); | |
2665 | BUG_ON(!middle_len); | |
2666 | BUG_ON(msg->middle); | |
2667 | ||
b6c1d5b8 | 2668 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2669 | if (!msg->middle) |
2670 | return -ENOMEM; | |
2671 | return 0; | |
2672 | } | |
2673 | ||
2450418c | 2674 | /* |
1c20f2d2 AE |
2675 | * Allocate a message for receiving an incoming message on a |
2676 | * connection, and save the result in con->in_msg. Uses the | |
2677 | * connection's private alloc_msg op if available. | |
2678 | * | |
2679 | * Returns true if the message should be skipped, false otherwise. | |
2680 | * If true is returned (skip message), con->in_msg will be NULL. | |
2681 | * If false is returned, con->in_msg will contain a pointer to the | |
2682 | * newly-allocated message, or NULL in case of memory exhaustion. | |
2450418c | 2683 | */ |
1c20f2d2 AE |
2684 | static bool ceph_con_in_msg_alloc(struct ceph_connection *con, |
2685 | struct ceph_msg_header *hdr) | |
2450418c YS |
2686 | { |
2687 | int type = le16_to_cpu(hdr->type); | |
2688 | int front_len = le32_to_cpu(hdr->front_len); | |
2689 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2450418c YS |
2690 | int ret; |
2691 | ||
1c20f2d2 AE |
2692 | BUG_ON(con->in_msg != NULL); |
2693 | ||
2450418c | 2694 | if (con->ops->alloc_msg) { |
1c20f2d2 AE |
2695 | int skip = 0; |
2696 | ||
0547a9b3 | 2697 | mutex_unlock(&con->mutex); |
1c20f2d2 | 2698 | con->in_msg = con->ops->alloc_msg(con, hdr, &skip); |
0547a9b3 | 2699 | mutex_lock(&con->mutex); |
92ce034b | 2700 | if (con->in_msg) { |
36eb71aa | 2701 | con->in_msg->con = con->ops->get(con); |
92ce034b AE |
2702 | BUG_ON(con->in_msg->con == NULL); |
2703 | } | |
1c20f2d2 AE |
2704 | if (skip) |
2705 | con->in_msg = NULL; | |
2706 | ||
2707 | if (!con->in_msg) | |
2708 | return skip != 0; | |
2450418c | 2709 | } |
1c20f2d2 AE |
2710 | if (!con->in_msg) { |
2711 | con->in_msg = ceph_msg_new(type, front_len, GFP_NOFS, false); | |
2712 | if (!con->in_msg) { | |
2450418c YS |
2713 | pr_err("unable to allocate msg type %d len %d\n", |
2714 | type, front_len); | |
1c20f2d2 | 2715 | return false; |
2450418c | 2716 | } |
36eb71aa | 2717 | con->in_msg->con = con->ops->get(con); |
92ce034b | 2718 | BUG_ON(con->in_msg->con == NULL); |
1c20f2d2 | 2719 | con->in_msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2720 | } |
1c20f2d2 | 2721 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2722 | |
1c20f2d2 AE |
2723 | if (middle_len && !con->in_msg->middle) { |
2724 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 2725 | if (ret < 0) { |
1c20f2d2 AE |
2726 | ceph_msg_put(con->in_msg); |
2727 | con->in_msg = NULL; | |
2450418c YS |
2728 | } |
2729 | } | |
9d7f0f13 | 2730 | |
1c20f2d2 | 2731 | return false; |
2450418c YS |
2732 | } |
2733 | ||
31b8006e SW |
2734 | |
2735 | /* | |
2736 | * Free a generically kmalloc'd message. | |
2737 | */ | |
2738 | void ceph_msg_kfree(struct ceph_msg *m) | |
2739 | { | |
2740 | dout("msg_kfree %p\n", m); | |
2741 | if (m->front_is_vmalloc) | |
2742 | vfree(m->front.iov_base); | |
2743 | else | |
2744 | kfree(m->front.iov_base); | |
2745 | kfree(m); | |
2746 | } | |
2747 | ||
2748 | /* | |
2749 | * Drop a msg ref. Destroy as needed. | |
2750 | */ | |
c2e552e7 SW |
2751 | void ceph_msg_last_put(struct kref *kref) |
2752 | { | |
2753 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2754 | |
c2e552e7 SW |
2755 | dout("ceph_msg_put last one on %p\n", m); |
2756 | WARN_ON(!list_empty(&m->list_head)); | |
2757 | ||
2758 | /* drop middle, data, if any */ | |
2759 | if (m->middle) { | |
2760 | ceph_buffer_put(m->middle); | |
2761 | m->middle = NULL; | |
31b8006e | 2762 | } |
c2e552e7 SW |
2763 | m->nr_pages = 0; |
2764 | m->pages = NULL; | |
2765 | ||
58bb3b37 SW |
2766 | if (m->pagelist) { |
2767 | ceph_pagelist_release(m->pagelist); | |
2768 | kfree(m->pagelist); | |
2769 | m->pagelist = NULL; | |
2770 | } | |
2771 | ||
68b4476b YS |
2772 | m->trail = NULL; |
2773 | ||
c2e552e7 SW |
2774 | if (m->pool) |
2775 | ceph_msgpool_put(m->pool, m); | |
2776 | else | |
2777 | ceph_msg_kfree(m); | |
31b8006e | 2778 | } |
3d14c5d2 | 2779 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2780 | |
2781 | void ceph_msg_dump(struct ceph_msg *msg) | |
2782 | { | |
2783 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2784 | msg->front_max, msg->nr_pages); | |
2785 | print_hex_dump(KERN_DEBUG, "header: ", | |
2786 | DUMP_PREFIX_OFFSET, 16, 1, | |
2787 | &msg->hdr, sizeof(msg->hdr), true); | |
2788 | print_hex_dump(KERN_DEBUG, " front: ", | |
2789 | DUMP_PREFIX_OFFSET, 16, 1, | |
2790 | msg->front.iov_base, msg->front.iov_len, true); | |
2791 | if (msg->middle) | |
2792 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2793 | DUMP_PREFIX_OFFSET, 16, 1, | |
2794 | msg->middle->vec.iov_base, | |
2795 | msg->middle->vec.iov_len, true); | |
2796 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2797 | DUMP_PREFIX_OFFSET, 16, 1, | |
2798 | &msg->footer, sizeof(msg->footer), true); | |
2799 | } | |
3d14c5d2 | 2800 | EXPORT_SYMBOL(ceph_msg_dump); |