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 | 256 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
257 | if (atomic_read(&con->msgr->stopping)) { |
258 | return; | |
259 | } | |
bd406145 | 260 | |
31b8006e | 261 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 262 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
263 | con, con->state); |
264 | queue_con(con); | |
265 | } | |
266 | } | |
267 | ||
268 | /* socket has buffer space for writing */ | |
327800bd | 269 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 270 | { |
d3002b97 | 271 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 272 | |
182fac26 JS |
273 | /* only queue to workqueue if there is data we want to write, |
274 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 275 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
276 | * doesn't get called again until try_write() fills the socket |
277 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
278 | * and net/core/stream.c:sk_stream_write_space(). | |
279 | */ | |
928443cd | 280 | if (test_bit(WRITE_PENDING, &con->flags)) { |
182fac26 | 281 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
327800bd | 282 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
283 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
284 | queue_con(con); | |
285 | } | |
31b8006e | 286 | } else { |
327800bd | 287 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 288 | } |
31b8006e SW |
289 | } |
290 | ||
291 | /* socket's state has changed */ | |
327800bd | 292 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 293 | { |
bd406145 | 294 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 295 | |
327800bd | 296 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
297 | con, con->state, sk->sk_state); |
298 | ||
31b8006e SW |
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 | { | |
8c50c817 | 503 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
504 | dout("con_close %p peer %s\n", con, |
505 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
a5988c49 | 506 | clear_bit(NEGOTIATING, &con->state); |
bb9e6bba | 507 | clear_bit(CONNECTING, &con->state); |
e27947c7 | 508 | clear_bit(CONNECTED, &con->state); |
31b8006e | 509 | clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */ |
a5988c49 AE |
510 | set_bit(CLOSED, &con->state); |
511 | ||
928443cd AE |
512 | clear_bit(LOSSYTX, &con->flags); /* so we retry next connect */ |
513 | clear_bit(KEEPALIVE_PENDING, &con->flags); | |
514 | clear_bit(WRITE_PENDING, &con->flags); | |
a5988c49 | 515 | |
31b8006e | 516 | reset_connection(con); |
6f2bc3ff | 517 | con->peer_global_seq = 0; |
91e45ce3 | 518 | cancel_delayed_work(&con->work); |
ee76e073 | 519 | con_close_socket(con); |
ec302645 | 520 | mutex_unlock(&con->mutex); |
31b8006e | 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 | { |
5469155f | 531 | mutex_lock(&con->mutex); |
3d14c5d2 | 532 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
31b8006e | 533 | set_bit(OPENING, &con->state); |
a5988c49 AE |
534 | WARN_ON(!test_and_clear_bit(CLOSED, &con->state)); |
535 | ||
b7a9e5dd SW |
536 | con->peer_name.type = (__u8) entity_type; |
537 | con->peer_name.num = cpu_to_le64(entity_num); | |
538 | ||
31b8006e | 539 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 540 | con->delay = 0; /* reset backoff memory */ |
5469155f | 541 | mutex_unlock(&con->mutex); |
31b8006e SW |
542 | queue_con(con); |
543 | } | |
3d14c5d2 | 544 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 545 | |
87b315a5 SW |
546 | /* |
547 | * return true if this connection ever successfully opened | |
548 | */ | |
549 | bool ceph_con_opened(struct ceph_connection *con) | |
550 | { | |
551 | return con->connect_seq > 0; | |
552 | } | |
553 | ||
31b8006e SW |
554 | /* |
555 | * initialize a new connection. | |
556 | */ | |
1bfd89f4 AE |
557 | void ceph_con_init(struct ceph_connection *con, void *private, |
558 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 559 | struct ceph_messenger *msgr) |
31b8006e SW |
560 | { |
561 | dout("con_init %p\n", con); | |
562 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
563 | con->private = private; |
564 | con->ops = ops; | |
31b8006e | 565 | con->msgr = msgr; |
ce2c8903 AE |
566 | |
567 | con_sock_state_init(con); | |
568 | ||
ec302645 | 569 | mutex_init(&con->mutex); |
31b8006e SW |
570 | INIT_LIST_HEAD(&con->out_queue); |
571 | INIT_LIST_HEAD(&con->out_sent); | |
572 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 AE |
573 | |
574 | set_bit(CLOSED, &con->state); | |
31b8006e | 575 | } |
3d14c5d2 | 576 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
577 | |
578 | ||
579 | /* | |
580 | * We maintain a global counter to order connection attempts. Get | |
581 | * a unique seq greater than @gt. | |
582 | */ | |
583 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
584 | { | |
585 | u32 ret; | |
586 | ||
587 | spin_lock(&msgr->global_seq_lock); | |
588 | if (msgr->global_seq < gt) | |
589 | msgr->global_seq = gt; | |
590 | ret = ++msgr->global_seq; | |
591 | spin_unlock(&msgr->global_seq_lock); | |
592 | return ret; | |
593 | } | |
594 | ||
e2200423 | 595 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
596 | { |
597 | con->out_kvec_left = 0; | |
598 | con->out_kvec_bytes = 0; | |
599 | con->out_kvec_cur = &con->out_kvec[0]; | |
600 | } | |
601 | ||
e2200423 | 602 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
603 | size_t size, void *data) |
604 | { | |
605 | int index; | |
606 | ||
607 | index = con->out_kvec_left; | |
608 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
609 | ||
610 | con->out_kvec[index].iov_len = size; | |
611 | con->out_kvec[index].iov_base = data; | |
612 | con->out_kvec_left++; | |
613 | con->out_kvec_bytes += size; | |
614 | } | |
31b8006e | 615 | |
df6ad1f9 AE |
616 | #ifdef CONFIG_BLOCK |
617 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
618 | { | |
619 | if (!bio) { | |
620 | *iter = NULL; | |
621 | *seg = 0; | |
622 | return; | |
623 | } | |
624 | *iter = bio; | |
625 | *seg = bio->bi_idx; | |
626 | } | |
627 | ||
628 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
629 | { | |
630 | if (*bio_iter == NULL) | |
631 | return; | |
632 | ||
633 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
634 | ||
635 | (*seg)++; | |
636 | if (*seg == (*bio_iter)->bi_vcnt) | |
637 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
638 | } | |
639 | #endif | |
640 | ||
739c905b AE |
641 | static void prepare_write_message_data(struct ceph_connection *con) |
642 | { | |
643 | struct ceph_msg *msg = con->out_msg; | |
644 | ||
645 | BUG_ON(!msg); | |
646 | BUG_ON(!msg->hdr.data_len); | |
647 | ||
648 | /* initialize page iterator */ | |
649 | con->out_msg_pos.page = 0; | |
650 | if (msg->pages) | |
651 | con->out_msg_pos.page_pos = msg->page_alignment; | |
652 | else | |
653 | con->out_msg_pos.page_pos = 0; | |
572c588e | 654 | #ifdef CONFIG_BLOCK |
abdaa6a8 | 655 | if (msg->bio) |
572c588e AE |
656 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); |
657 | #endif | |
739c905b AE |
658 | con->out_msg_pos.data_pos = 0; |
659 | con->out_msg_pos.did_page_crc = false; | |
660 | con->out_more = 1; /* data + footer will follow */ | |
661 | } | |
662 | ||
31b8006e SW |
663 | /* |
664 | * Prepare footer for currently outgoing message, and finish things | |
665 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
666 | */ | |
859eb799 | 667 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
668 | { |
669 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 670 | int v = con->out_kvec_left; |
31b8006e | 671 | |
fd154f3c AE |
672 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
673 | ||
31b8006e SW |
674 | dout("prepare_write_message_footer %p\n", con); |
675 | con->out_kvec_is_msg = true; | |
676 | con->out_kvec[v].iov_base = &m->footer; | |
677 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
678 | con->out_kvec_bytes += sizeof(m->footer); | |
679 | con->out_kvec_left++; | |
680 | con->out_more = m->more_to_follow; | |
c86a2930 | 681 | con->out_msg_done = true; |
31b8006e SW |
682 | } |
683 | ||
684 | /* | |
685 | * Prepare headers for the next outgoing message. | |
686 | */ | |
687 | static void prepare_write_message(struct ceph_connection *con) | |
688 | { | |
689 | struct ceph_msg *m; | |
a9a0c51a | 690 | u32 crc; |
31b8006e | 691 | |
e2200423 | 692 | con_out_kvec_reset(con); |
31b8006e | 693 | con->out_kvec_is_msg = true; |
c86a2930 | 694 | con->out_msg_done = false; |
31b8006e SW |
695 | |
696 | /* Sneak an ack in there first? If we can get it into the same | |
697 | * TCP packet that's a good thing. */ | |
698 | if (con->in_seq > con->in_seq_acked) { | |
699 | con->in_seq_acked = con->in_seq; | |
e2200423 | 700 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 701 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 702 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 703 | &con->out_temp_ack); |
31b8006e SW |
704 | } |
705 | ||
38941f80 | 706 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 707 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 708 | con->out_msg = m; |
38941f80 | 709 | BUG_ON(m->con != con); |
4cf9d544 SW |
710 | |
711 | /* put message on sent list */ | |
712 | ceph_msg_get(m); | |
713 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 714 | |
e84346b7 SW |
715 | /* |
716 | * only assign outgoing seq # if we haven't sent this message | |
717 | * yet. if it is requeued, resend with it's original seq. | |
718 | */ | |
719 | if (m->needs_out_seq) { | |
720 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
721 | m->needs_out_seq = false; | |
722 | } | |
31b8006e SW |
723 | |
724 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
725 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
726 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
727 | le32_to_cpu(m->hdr.data_len), | |
728 | m->nr_pages); | |
729 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); | |
730 | ||
731 | /* tag + hdr + front + middle */ | |
e2200423 AE |
732 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
733 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
734 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 735 | |
31b8006e | 736 | if (m->middle) |
e2200423 | 737 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 738 | m->middle->vec.iov_base); |
31b8006e SW |
739 | |
740 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
741 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
742 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 743 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
744 | |
745 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
746 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
747 | if (m->middle) { | |
748 | crc = crc32c(0, m->middle->vec.iov_base, | |
749 | m->middle->vec.iov_len); | |
750 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
751 | } else | |
31b8006e | 752 | con->out_msg->footer.middle_crc = 0; |
739c905b | 753 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
754 | le32_to_cpu(con->out_msg->footer.front_crc), |
755 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
756 | ||
757 | /* is there a data payload? */ | |
739c905b AE |
758 | con->out_msg->footer.data_crc = 0; |
759 | if (m->hdr.data_len) | |
760 | prepare_write_message_data(con); | |
761 | else | |
31b8006e | 762 | /* no, queue up footer too and be done */ |
859eb799 | 763 | prepare_write_message_footer(con); |
31b8006e | 764 | |
928443cd | 765 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
766 | } |
767 | ||
768 | /* | |
769 | * Prepare an ack. | |
770 | */ | |
771 | static void prepare_write_ack(struct ceph_connection *con) | |
772 | { | |
773 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
774 | con->in_seq_acked, con->in_seq); | |
775 | con->in_seq_acked = con->in_seq; | |
776 | ||
e2200423 | 777 | con_out_kvec_reset(con); |
859eb799 | 778 | |
e2200423 | 779 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 780 | |
31b8006e | 781 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 782 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
783 | &con->out_temp_ack); |
784 | ||
31b8006e | 785 | con->out_more = 1; /* more will follow.. eventually.. */ |
928443cd | 786 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
787 | } |
788 | ||
789 | /* | |
790 | * Prepare to write keepalive byte. | |
791 | */ | |
792 | static void prepare_write_keepalive(struct ceph_connection *con) | |
793 | { | |
794 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
795 | con_out_kvec_reset(con); |
796 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
928443cd | 797 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
798 | } |
799 | ||
800 | /* | |
801 | * Connection negotiation. | |
802 | */ | |
803 | ||
dac1e716 AE |
804 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
805 | int *auth_proto) | |
4e7a5dcd | 806 | { |
a3530df3 | 807 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
808 | |
809 | if (!con->ops->get_authorizer) { | |
810 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
811 | con->out_connect.authorizer_len = 0; | |
812 | ||
729796be | 813 | return NULL; |
b1c6b980 AE |
814 | } |
815 | ||
816 | /* Can't hold the mutex while getting authorizer */ | |
4e7a5dcd | 817 | |
ec302645 | 818 | mutex_unlock(&con->mutex); |
b1c6b980 | 819 | |
dac1e716 | 820 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
8f43fb53 | 821 | |
ec302645 | 822 | mutex_lock(&con->mutex); |
4e7a5dcd | 823 | |
a3530df3 | 824 | if (IS_ERR(auth)) |
729796be | 825 | return auth; |
928443cd | 826 | if (test_bit(CLOSED, &con->state) || test_bit(OPENING, &con->flags)) |
729796be | 827 | return ERR_PTR(-EAGAIN); |
0da5d703 | 828 | |
8f43fb53 AE |
829 | con->auth_reply_buf = auth->authorizer_reply_buf; |
830 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
831 | ||
859eb799 | 832 | |
729796be | 833 | return auth; |
4e7a5dcd SW |
834 | } |
835 | ||
31b8006e SW |
836 | /* |
837 | * We connected to a peer and are saying hello. | |
838 | */ | |
e825a66d | 839 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 840 | { |
e2200423 AE |
841 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
842 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 843 | &con->msgr->my_enc_addr); |
eed0ef2c | 844 | |
eed0ef2c | 845 | con->out_more = 0; |
928443cd | 846 | set_bit(WRITE_PENDING, &con->flags); |
eed0ef2c SW |
847 | } |
848 | ||
e825a66d | 849 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 850 | { |
95c96174 | 851 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 852 | int proto; |
dac1e716 | 853 | int auth_proto; |
729796be | 854 | struct ceph_auth_handshake *auth; |
31b8006e SW |
855 | |
856 | switch (con->peer_name.type) { | |
857 | case CEPH_ENTITY_TYPE_MON: | |
858 | proto = CEPH_MONC_PROTOCOL; | |
859 | break; | |
860 | case CEPH_ENTITY_TYPE_OSD: | |
861 | proto = CEPH_OSDC_PROTOCOL; | |
862 | break; | |
863 | case CEPH_ENTITY_TYPE_MDS: | |
864 | proto = CEPH_MDSC_PROTOCOL; | |
865 | break; | |
866 | default: | |
867 | BUG(); | |
868 | } | |
869 | ||
870 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
871 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 872 | |
e825a66d | 873 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
874 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
875 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
876 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
877 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
878 | con->out_connect.flags = 0; | |
31b8006e | 879 | |
dac1e716 AE |
880 | auth_proto = CEPH_AUTH_UNKNOWN; |
881 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
882 | if (IS_ERR(auth)) |
883 | return PTR_ERR(auth); | |
3da54776 | 884 | |
dac1e716 | 885 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
886 | con->out_connect.authorizer_len = auth ? |
887 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
888 | ||
ab166d5a | 889 | con_out_kvec_reset(con); |
e2200423 | 890 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
891 | &con->out_connect); |
892 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 893 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 894 | auth->authorizer_buf); |
859eb799 | 895 | |
31b8006e | 896 | con->out_more = 0; |
928443cd | 897 | set_bit(WRITE_PENDING, &con->flags); |
4e7a5dcd | 898 | |
e10c758e | 899 | return 0; |
31b8006e SW |
900 | } |
901 | ||
31b8006e SW |
902 | /* |
903 | * write as much of pending kvecs to the socket as we can. | |
904 | * 1 -> done | |
905 | * 0 -> socket full, but more to do | |
906 | * <0 -> error | |
907 | */ | |
908 | static int write_partial_kvec(struct ceph_connection *con) | |
909 | { | |
910 | int ret; | |
911 | ||
912 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
913 | while (con->out_kvec_bytes > 0) { | |
914 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
915 | con->out_kvec_left, con->out_kvec_bytes, | |
916 | con->out_more); | |
917 | if (ret <= 0) | |
918 | goto out; | |
919 | con->out_kvec_bytes -= ret; | |
920 | if (con->out_kvec_bytes == 0) | |
921 | break; /* done */ | |
f42299e6 AE |
922 | |
923 | /* account for full iov entries consumed */ | |
924 | while (ret >= con->out_kvec_cur->iov_len) { | |
925 | BUG_ON(!con->out_kvec_left); | |
926 | ret -= con->out_kvec_cur->iov_len; | |
927 | con->out_kvec_cur++; | |
928 | con->out_kvec_left--; | |
929 | } | |
930 | /* and for a partially-consumed entry */ | |
931 | if (ret) { | |
932 | con->out_kvec_cur->iov_len -= ret; | |
933 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
934 | } |
935 | } | |
936 | con->out_kvec_left = 0; | |
937 | con->out_kvec_is_msg = false; | |
938 | ret = 1; | |
939 | out: | |
940 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
941 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
942 | return ret; /* done! */ | |
943 | } | |
944 | ||
84ca8fc8 AE |
945 | static void out_msg_pos_next(struct ceph_connection *con, struct page *page, |
946 | size_t len, size_t sent, bool in_trail) | |
947 | { | |
948 | struct ceph_msg *msg = con->out_msg; | |
949 | ||
950 | BUG_ON(!msg); | |
951 | BUG_ON(!sent); | |
952 | ||
953 | con->out_msg_pos.data_pos += sent; | |
954 | con->out_msg_pos.page_pos += sent; | |
5821bd8c AE |
955 | if (sent < len) |
956 | return; | |
957 | ||
958 | BUG_ON(sent != len); | |
959 | con->out_msg_pos.page_pos = 0; | |
960 | con->out_msg_pos.page++; | |
961 | con->out_msg_pos.did_page_crc = false; | |
962 | if (in_trail) | |
963 | list_move_tail(&page->lru, | |
964 | &msg->trail->head); | |
965 | else if (msg->pagelist) | |
966 | list_move_tail(&page->lru, | |
967 | &msg->pagelist->head); | |
84ca8fc8 | 968 | #ifdef CONFIG_BLOCK |
5821bd8c AE |
969 | else if (msg->bio) |
970 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
84ca8fc8 | 971 | #endif |
84ca8fc8 AE |
972 | } |
973 | ||
31b8006e SW |
974 | /* |
975 | * Write as much message data payload as we can. If we finish, queue | |
976 | * up the footer. | |
977 | * 1 -> done, footer is now queued in out_kvec[]. | |
978 | * 0 -> socket full, but more to do | |
979 | * <0 -> error | |
980 | */ | |
981 | static int write_partial_msg_pages(struct ceph_connection *con) | |
982 | { | |
983 | struct ceph_msg *msg = con->out_msg; | |
95c96174 | 984 | unsigned int data_len = le32_to_cpu(msg->hdr.data_len); |
31b8006e | 985 | size_t len; |
37675b0f | 986 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 987 | int ret; |
68b4476b | 988 | int total_max_write; |
84ca8fc8 | 989 | bool in_trail = false; |
5821bd8c AE |
990 | const size_t trail_len = (msg->trail ? msg->trail->length : 0); |
991 | const size_t trail_off = data_len - trail_len; | |
31b8006e SW |
992 | |
993 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
84ca8fc8 | 994 | con, msg, con->out_msg_pos.page, msg->nr_pages, |
31b8006e SW |
995 | con->out_msg_pos.page_pos); |
996 | ||
5821bd8c AE |
997 | /* |
998 | * Iterate through each page that contains data to be | |
999 | * written, and send as much as possible for each. | |
1000 | * | |
1001 | * If we are calculating the data crc (the default), we will | |
1002 | * need to map the page. If we have no pages, they have | |
1003 | * been revoked, so use the zero page. | |
1004 | */ | |
68b4476b | 1005 | while (data_len > con->out_msg_pos.data_pos) { |
31b8006e | 1006 | struct page *page = NULL; |
68b4476b | 1007 | int max_write = PAGE_SIZE; |
9bd19663 | 1008 | int bio_offset = 0; |
68b4476b | 1009 | |
5821bd8c AE |
1010 | in_trail = in_trail || con->out_msg_pos.data_pos >= trail_off; |
1011 | if (!in_trail) | |
1012 | total_max_write = trail_off - con->out_msg_pos.data_pos; | |
68b4476b | 1013 | |
5821bd8c | 1014 | if (in_trail) { |
68b4476b YS |
1015 | total_max_write = data_len - con->out_msg_pos.data_pos; |
1016 | ||
1017 | page = list_first_entry(&msg->trail->head, | |
1018 | struct page, lru); | |
68b4476b | 1019 | } else if (msg->pages) { |
31b8006e | 1020 | page = msg->pages[con->out_msg_pos.page]; |
58bb3b37 SW |
1021 | } else if (msg->pagelist) { |
1022 | page = list_first_entry(&msg->pagelist->head, | |
1023 | struct page, lru); | |
68b4476b YS |
1024 | #ifdef CONFIG_BLOCK |
1025 | } else if (msg->bio) { | |
1026 | struct bio_vec *bv; | |
1027 | ||
1028 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
1029 | page = bv->bv_page; | |
9bd19663 | 1030 | bio_offset = bv->bv_offset; |
68b4476b YS |
1031 | max_write = bv->bv_len; |
1032 | #endif | |
31b8006e | 1033 | } else { |
57666519 | 1034 | page = zero_page; |
31b8006e | 1035 | } |
68b4476b YS |
1036 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
1037 | total_max_write); | |
1038 | ||
37675b0f | 1039 | if (do_datacrc && !con->out_msg_pos.did_page_crc) { |
9bd19663 | 1040 | void *base; |
5821bd8c | 1041 | u32 crc = le32_to_cpu(msg->footer.data_crc); |
8d63e318 | 1042 | char *kaddr; |
31b8006e | 1043 | |
8d63e318 | 1044 | kaddr = kmap(page); |
31b8006e | 1045 | BUG_ON(kaddr == NULL); |
9bd19663 | 1046 | base = kaddr + con->out_msg_pos.page_pos + bio_offset; |
5821bd8c | 1047 | crc = crc32c(crc, base, len); |
84ca8fc8 | 1048 | msg->footer.data_crc = cpu_to_le32(crc); |
bca064d2 | 1049 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 1050 | } |
e36b13cc | 1051 | ret = ceph_tcp_sendpage(con->sock, page, |
9bd19663 | 1052 | con->out_msg_pos.page_pos + bio_offset, |
e36b13cc | 1053 | len, 1); |
31b8006e | 1054 | |
0cdf9e60 | 1055 | if (do_datacrc) |
31b8006e SW |
1056 | kunmap(page); |
1057 | ||
1058 | if (ret <= 0) | |
1059 | goto out; | |
1060 | ||
84ca8fc8 | 1061 | out_msg_pos_next(con, page, len, (size_t) ret, in_trail); |
31b8006e SW |
1062 | } |
1063 | ||
1064 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
1065 | ||
1066 | /* prepare and queue up footer, too */ | |
37675b0f | 1067 | if (!do_datacrc) |
84ca8fc8 | 1068 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1069 | con_out_kvec_reset(con); |
859eb799 | 1070 | prepare_write_message_footer(con); |
31b8006e SW |
1071 | ret = 1; |
1072 | out: | |
1073 | return ret; | |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * write some zeros | |
1078 | */ | |
1079 | static int write_partial_skip(struct ceph_connection *con) | |
1080 | { | |
1081 | int ret; | |
1082 | ||
1083 | while (con->out_skip > 0) { | |
31739139 | 1084 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1085 | |
31739139 | 1086 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1); |
31b8006e SW |
1087 | if (ret <= 0) |
1088 | goto out; | |
1089 | con->out_skip -= ret; | |
1090 | } | |
1091 | ret = 1; | |
1092 | out: | |
1093 | return ret; | |
1094 | } | |
1095 | ||
1096 | /* | |
1097 | * Prepare to read connection handshake, or an ack. | |
1098 | */ | |
eed0ef2c SW |
1099 | static void prepare_read_banner(struct ceph_connection *con) |
1100 | { | |
1101 | dout("prepare_read_banner %p\n", con); | |
1102 | con->in_base_pos = 0; | |
1103 | } | |
1104 | ||
31b8006e SW |
1105 | static void prepare_read_connect(struct ceph_connection *con) |
1106 | { | |
1107 | dout("prepare_read_connect %p\n", con); | |
1108 | con->in_base_pos = 0; | |
1109 | } | |
1110 | ||
1111 | static void prepare_read_ack(struct ceph_connection *con) | |
1112 | { | |
1113 | dout("prepare_read_ack %p\n", con); | |
1114 | con->in_base_pos = 0; | |
1115 | } | |
1116 | ||
1117 | static void prepare_read_tag(struct ceph_connection *con) | |
1118 | { | |
1119 | dout("prepare_read_tag %p\n", con); | |
1120 | con->in_base_pos = 0; | |
1121 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * Prepare to read a message. | |
1126 | */ | |
1127 | static int prepare_read_message(struct ceph_connection *con) | |
1128 | { | |
1129 | dout("prepare_read_message %p\n", con); | |
1130 | BUG_ON(con->in_msg != NULL); | |
1131 | con->in_base_pos = 0; | |
1132 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1133 | return 0; | |
1134 | } | |
1135 | ||
1136 | ||
1137 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1138 | int end, int size, void *object) |
31b8006e | 1139 | { |
e6cee71f AE |
1140 | while (con->in_base_pos < end) { |
1141 | int left = end - con->in_base_pos; | |
31b8006e SW |
1142 | int have = size - left; |
1143 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1144 | if (ret <= 0) | |
1145 | return ret; | |
1146 | con->in_base_pos += ret; | |
1147 | } | |
1148 | return 1; | |
1149 | } | |
1150 | ||
1151 | ||
1152 | /* | |
1153 | * Read all or part of the connect-side handshake on a new connection | |
1154 | */ | |
eed0ef2c | 1155 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1156 | { |
fd51653f AE |
1157 | int size; |
1158 | int end; | |
1159 | int ret; | |
31b8006e | 1160 | |
eed0ef2c | 1161 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1162 | |
1163 | /* peer's banner */ | |
fd51653f AE |
1164 | size = strlen(CEPH_BANNER); |
1165 | end = size; | |
1166 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1167 | if (ret <= 0) |
1168 | goto out; | |
fd51653f AE |
1169 | |
1170 | size = sizeof (con->actual_peer_addr); | |
1171 | end += size; | |
1172 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1173 | if (ret <= 0) |
1174 | goto out; | |
fd51653f AE |
1175 | |
1176 | size = sizeof (con->peer_addr_for_me); | |
1177 | end += size; | |
1178 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1179 | if (ret <= 0) |
1180 | goto out; | |
fd51653f | 1181 | |
eed0ef2c SW |
1182 | out: |
1183 | return ret; | |
1184 | } | |
1185 | ||
1186 | static int read_partial_connect(struct ceph_connection *con) | |
1187 | { | |
fd51653f AE |
1188 | int size; |
1189 | int end; | |
1190 | int ret; | |
eed0ef2c SW |
1191 | |
1192 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1193 | ||
fd51653f AE |
1194 | size = sizeof (con->in_reply); |
1195 | end = size; | |
1196 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1197 | if (ret <= 0) |
1198 | goto out; | |
fd51653f AE |
1199 | |
1200 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1201 | end += size; | |
1202 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1203 | if (ret <= 0) |
1204 | goto out; | |
31b8006e | 1205 | |
4e7a5dcd SW |
1206 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1207 | con, (int)con->in_reply.tag, | |
1208 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1209 | le32_to_cpu(con->in_reply.global_seq)); |
1210 | out: | |
1211 | return ret; | |
eed0ef2c | 1212 | |
31b8006e SW |
1213 | } |
1214 | ||
1215 | /* | |
1216 | * Verify the hello banner looks okay. | |
1217 | */ | |
1218 | static int verify_hello(struct ceph_connection *con) | |
1219 | { | |
1220 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1221 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1222 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1223 | con->error_msg = "protocol error, bad banner"; |
1224 | return -1; | |
1225 | } | |
1226 | return 0; | |
1227 | } | |
1228 | ||
1229 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1230 | { | |
1231 | switch (ss->ss_family) { | |
1232 | case AF_INET: | |
1233 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1234 | case AF_INET6: | |
1235 | return | |
1236 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1237 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1238 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1239 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1240 | } | |
1241 | return false; | |
1242 | } | |
1243 | ||
1244 | static int addr_port(struct sockaddr_storage *ss) | |
1245 | { | |
1246 | switch (ss->ss_family) { | |
1247 | case AF_INET: | |
f28bcfbe | 1248 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1249 | case AF_INET6: |
f28bcfbe | 1250 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1251 | } |
1252 | return 0; | |
1253 | } | |
1254 | ||
1255 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1256 | { | |
1257 | switch (ss->ss_family) { | |
1258 | case AF_INET: | |
1259 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1260 | break; |
31b8006e SW |
1261 | case AF_INET6: |
1262 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1263 | break; |
31b8006e SW |
1264 | } |
1265 | } | |
1266 | ||
ee3b56f2 NW |
1267 | /* |
1268 | * Unlike other *_pton function semantics, zero indicates success. | |
1269 | */ | |
1270 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1271 | char delim, const char **ipend) | |
1272 | { | |
99f0f3b2 AE |
1273 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1274 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1275 | |
1276 | memset(ss, 0, sizeof(*ss)); | |
1277 | ||
1278 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1279 | ss->ss_family = AF_INET; | |
1280 | return 0; | |
1281 | } | |
1282 | ||
1283 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1284 | ss->ss_family = AF_INET6; | |
1285 | return 0; | |
1286 | } | |
1287 | ||
1288 | return -EINVAL; | |
1289 | } | |
1290 | ||
1291 | /* | |
1292 | * Extract hostname string and resolve using kernel DNS facility. | |
1293 | */ | |
1294 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1295 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1296 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1297 | { | |
1298 | const char *end, *delim_p; | |
1299 | char *colon_p, *ip_addr = NULL; | |
1300 | int ip_len, ret; | |
1301 | ||
1302 | /* | |
1303 | * The end of the hostname occurs immediately preceding the delimiter or | |
1304 | * the port marker (':') where the delimiter takes precedence. | |
1305 | */ | |
1306 | delim_p = memchr(name, delim, namelen); | |
1307 | colon_p = memchr(name, ':', namelen); | |
1308 | ||
1309 | if (delim_p && colon_p) | |
1310 | end = delim_p < colon_p ? delim_p : colon_p; | |
1311 | else if (!delim_p && colon_p) | |
1312 | end = colon_p; | |
1313 | else { | |
1314 | end = delim_p; | |
1315 | if (!end) /* case: hostname:/ */ | |
1316 | end = name + namelen; | |
1317 | } | |
1318 | ||
1319 | if (end <= name) | |
1320 | return -EINVAL; | |
1321 | ||
1322 | /* do dns_resolve upcall */ | |
1323 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1324 | if (ip_len > 0) | |
1325 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1326 | else | |
1327 | ret = -ESRCH; | |
1328 | ||
1329 | kfree(ip_addr); | |
1330 | ||
1331 | *ipend = end; | |
1332 | ||
1333 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1334 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1335 | ||
1336 | return ret; | |
1337 | } | |
1338 | #else | |
1339 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1340 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1341 | { | |
1342 | return -EINVAL; | |
1343 | } | |
1344 | #endif | |
1345 | ||
1346 | /* | |
1347 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1348 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1349 | */ | |
1350 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1351 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1352 | { | |
1353 | int ret; | |
1354 | ||
1355 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1356 | if (ret) | |
1357 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1358 | ||
1359 | return ret; | |
1360 | } | |
1361 | ||
31b8006e SW |
1362 | /* |
1363 | * Parse an ip[:port] list into an addr array. Use the default | |
1364 | * monitor port if a port isn't specified. | |
1365 | */ | |
1366 | int ceph_parse_ips(const char *c, const char *end, | |
1367 | struct ceph_entity_addr *addr, | |
1368 | int max_count, int *count) | |
1369 | { | |
ee3b56f2 | 1370 | int i, ret = -EINVAL; |
31b8006e SW |
1371 | const char *p = c; |
1372 | ||
1373 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1374 | for (i = 0; i < max_count; i++) { | |
1375 | const char *ipend; | |
1376 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1377 | int port; |
39139f64 SW |
1378 | char delim = ','; |
1379 | ||
1380 | if (*p == '[') { | |
1381 | delim = ']'; | |
1382 | p++; | |
1383 | } | |
31b8006e | 1384 | |
ee3b56f2 NW |
1385 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1386 | if (ret) | |
31b8006e | 1387 | goto bad; |
ee3b56f2 NW |
1388 | ret = -EINVAL; |
1389 | ||
31b8006e SW |
1390 | p = ipend; |
1391 | ||
39139f64 SW |
1392 | if (delim == ']') { |
1393 | if (*p != ']') { | |
1394 | dout("missing matching ']'\n"); | |
1395 | goto bad; | |
1396 | } | |
1397 | p++; | |
1398 | } | |
1399 | ||
31b8006e SW |
1400 | /* port? */ |
1401 | if (p < end && *p == ':') { | |
1402 | port = 0; | |
1403 | p++; | |
1404 | while (p < end && *p >= '0' && *p <= '9') { | |
1405 | port = (port * 10) + (*p - '0'); | |
1406 | p++; | |
1407 | } | |
1408 | if (port > 65535 || port == 0) | |
1409 | goto bad; | |
1410 | } else { | |
1411 | port = CEPH_MON_PORT; | |
1412 | } | |
1413 | ||
1414 | addr_set_port(ss, port); | |
1415 | ||
3d14c5d2 | 1416 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1417 | |
1418 | if (p == end) | |
1419 | break; | |
1420 | if (*p != ',') | |
1421 | goto bad; | |
1422 | p++; | |
1423 | } | |
1424 | ||
1425 | if (p != end) | |
1426 | goto bad; | |
1427 | ||
1428 | if (count) | |
1429 | *count = i + 1; | |
1430 | return 0; | |
1431 | ||
1432 | bad: | |
39139f64 | 1433 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1434 | return ret; |
31b8006e | 1435 | } |
3d14c5d2 | 1436 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1437 | |
eed0ef2c | 1438 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1439 | { |
eed0ef2c | 1440 | dout("process_banner on %p\n", con); |
31b8006e SW |
1441 | |
1442 | if (verify_hello(con) < 0) | |
1443 | return -1; | |
1444 | ||
63f2d211 SW |
1445 | ceph_decode_addr(&con->actual_peer_addr); |
1446 | ceph_decode_addr(&con->peer_addr_for_me); | |
1447 | ||
31b8006e SW |
1448 | /* |
1449 | * Make sure the other end is who we wanted. note that the other | |
1450 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1451 | * them the benefit of the doubt. | |
1452 | */ | |
103e2d3a SW |
1453 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1454 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1455 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1456 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1457 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1458 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1459 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1460 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1461 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1462 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1463 | return -1; |
1464 | } | |
1465 | ||
1466 | /* | |
1467 | * did we learn our address? | |
1468 | */ | |
1469 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1470 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1471 | ||
1472 | memcpy(&con->msgr->inst.addr.in_addr, | |
1473 | &con->peer_addr_for_me.in_addr, | |
1474 | sizeof(con->peer_addr_for_me.in_addr)); | |
1475 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1476 | encode_my_addr(con->msgr); |
eed0ef2c | 1477 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1478 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1479 | } |
1480 | ||
eed0ef2c SW |
1481 | return 0; |
1482 | } | |
1483 | ||
04a419f9 SW |
1484 | static void fail_protocol(struct ceph_connection *con) |
1485 | { | |
1486 | reset_connection(con); | |
1487 | set_bit(CLOSED, &con->state); /* in case there's queued work */ | |
04a419f9 SW |
1488 | } |
1489 | ||
eed0ef2c SW |
1490 | static int process_connect(struct ceph_connection *con) |
1491 | { | |
3d14c5d2 YS |
1492 | u64 sup_feat = con->msgr->supported_features; |
1493 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1494 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1495 | int ret; |
04a419f9 | 1496 | |
eed0ef2c SW |
1497 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1498 | ||
31b8006e | 1499 | switch (con->in_reply.tag) { |
04a419f9 SW |
1500 | case CEPH_MSGR_TAG_FEATURES: |
1501 | pr_err("%s%lld %s feature set mismatch," | |
1502 | " my %llx < server's %llx, missing %llx\n", | |
1503 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1504 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1505 | sup_feat, server_feat, server_feat & ~sup_feat); |
1506 | con->error_msg = "missing required protocol features"; | |
1507 | fail_protocol(con); | |
1508 | return -1; | |
1509 | ||
31b8006e | 1510 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1511 | pr_err("%s%lld %s protocol version mismatch," |
1512 | " my %d != server's %d\n", | |
1513 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1514 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1515 | le32_to_cpu(con->out_connect.protocol_version), |
1516 | le32_to_cpu(con->in_reply.protocol_version)); | |
1517 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1518 | fail_protocol(con); |
31b8006e SW |
1519 | return -1; |
1520 | ||
4e7a5dcd SW |
1521 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1522 | con->auth_retry++; | |
1523 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1524 | con->auth_retry); | |
1525 | if (con->auth_retry == 2) { | |
1526 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1527 | return -1; |
1528 | } | |
1529 | con->auth_retry = 1; | |
e825a66d | 1530 | ret = prepare_write_connect(con); |
0da5d703 SW |
1531 | if (ret < 0) |
1532 | return ret; | |
63733a0f | 1533 | prepare_read_connect(con); |
4e7a5dcd | 1534 | break; |
31b8006e SW |
1535 | |
1536 | case CEPH_MSGR_TAG_RESETSESSION: | |
1537 | /* | |
1538 | * If we connected with a large connect_seq but the peer | |
1539 | * has no record of a session with us (no connection, or | |
1540 | * connect_seq == 0), they will send RESETSESION to indicate | |
1541 | * that they must have reset their session, and may have | |
1542 | * dropped messages. | |
1543 | */ | |
1544 | dout("process_connect got RESET peer seq %u\n", | |
a16cb1f7 | 1545 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
1546 | pr_err("%s%lld %s connection reset\n", |
1547 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1548 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1549 | reset_connection(con); |
5a0f8fdd AE |
1550 | ret = prepare_write_connect(con); |
1551 | if (ret < 0) | |
1552 | return ret; | |
31b8006e SW |
1553 | prepare_read_connect(con); |
1554 | ||
1555 | /* Tell ceph about it. */ | |
ec302645 | 1556 | mutex_unlock(&con->mutex); |
31b8006e SW |
1557 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1558 | if (con->ops->peer_reset) | |
1559 | con->ops->peer_reset(con); | |
ec302645 | 1560 | mutex_lock(&con->mutex); |
0da5d703 SW |
1561 | if (test_bit(CLOSED, &con->state) || |
1562 | test_bit(OPENING, &con->state)) | |
1563 | return -EAGAIN; | |
31b8006e SW |
1564 | break; |
1565 | ||
1566 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1567 | /* | |
1568 | * If we sent a smaller connect_seq than the peer has, try | |
1569 | * again with a larger value. | |
1570 | */ | |
a16cb1f7 | 1571 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 1572 | le32_to_cpu(con->out_connect.connect_seq), |
a16cb1f7 SW |
1573 | le32_to_cpu(con->in_reply.connect_seq)); |
1574 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
5a0f8fdd AE |
1575 | ret = prepare_write_connect(con); |
1576 | if (ret < 0) | |
1577 | return ret; | |
31b8006e SW |
1578 | prepare_read_connect(con); |
1579 | break; | |
1580 | ||
1581 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1582 | /* | |
1583 | * If we sent a smaller global_seq than the peer has, try | |
1584 | * again with a larger value. | |
1585 | */ | |
eed0ef2c | 1586 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 1587 | con->peer_global_seq, |
a16cb1f7 | 1588 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 1589 | get_global_seq(con->msgr, |
a16cb1f7 | 1590 | le32_to_cpu(con->in_reply.global_seq)); |
5a0f8fdd AE |
1591 | ret = prepare_write_connect(con); |
1592 | if (ret < 0) | |
1593 | return ret; | |
31b8006e SW |
1594 | prepare_read_connect(con); |
1595 | break; | |
1596 | ||
1597 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1598 | if (req_feat & ~server_feat) { |
1599 | pr_err("%s%lld %s protocol feature mismatch," | |
1600 | " my required %llx > server's %llx, need %llx\n", | |
1601 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1602 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1603 | req_feat, server_feat, req_feat & ~server_feat); |
1604 | con->error_msg = "missing required protocol features"; | |
1605 | fail_protocol(con); | |
1606 | return -1; | |
1607 | } | |
3ec50d18 | 1608 | clear_bit(NEGOTIATING, &con->state); |
e27947c7 | 1609 | set_bit(CONNECTED, &con->state); |
31b8006e SW |
1610 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1611 | con->connect_seq++; | |
aba558e2 | 1612 | con->peer_features = server_feat; |
31b8006e SW |
1613 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1614 | con->peer_global_seq, | |
1615 | le32_to_cpu(con->in_reply.connect_seq), | |
1616 | con->connect_seq); | |
1617 | WARN_ON(con->connect_seq != | |
1618 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1619 | |
1620 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
928443cd | 1621 | set_bit(LOSSYTX, &con->flags); |
92ac41d0 | 1622 | |
85effe18 SW |
1623 | con->delay = 0; /* reset backoff memory */ |
1624 | ||
31b8006e SW |
1625 | prepare_read_tag(con); |
1626 | break; | |
1627 | ||
1628 | case CEPH_MSGR_TAG_WAIT: | |
1629 | /* | |
1630 | * If there is a connection race (we are opening | |
1631 | * connections to each other), one of us may just have | |
1632 | * to WAIT. This shouldn't happen if we are the | |
1633 | * client. | |
1634 | */ | |
04177882 SW |
1635 | pr_err("process_connect got WAIT as client\n"); |
1636 | con->error_msg = "protocol error, got WAIT as client"; | |
1637 | return -1; | |
31b8006e SW |
1638 | |
1639 | default: | |
1640 | pr_err("connect protocol error, will retry\n"); | |
1641 | con->error_msg = "protocol error, garbage tag during connect"; | |
1642 | return -1; | |
1643 | } | |
1644 | return 0; | |
1645 | } | |
1646 | ||
1647 | ||
1648 | /* | |
1649 | * read (part of) an ack | |
1650 | */ | |
1651 | static int read_partial_ack(struct ceph_connection *con) | |
1652 | { | |
fd51653f AE |
1653 | int size = sizeof (con->in_temp_ack); |
1654 | int end = size; | |
1655 | ||
1656 | return read_partial(con, end, size, &con->in_temp_ack); | |
31b8006e SW |
1657 | } |
1658 | ||
1659 | ||
1660 | /* | |
1661 | * We can finally discard anything that's been acked. | |
1662 | */ | |
1663 | static void process_ack(struct ceph_connection *con) | |
1664 | { | |
1665 | struct ceph_msg *m; | |
1666 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1667 | u64 seq; | |
1668 | ||
31b8006e SW |
1669 | while (!list_empty(&con->out_sent)) { |
1670 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1671 | list_head); | |
1672 | seq = le64_to_cpu(m->hdr.seq); | |
1673 | if (seq > ack) | |
1674 | break; | |
1675 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1676 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1677 | m->ack_stamp = jiffies; |
31b8006e SW |
1678 | ceph_msg_remove(m); |
1679 | } | |
31b8006e SW |
1680 | prepare_read_tag(con); |
1681 | } | |
1682 | ||
1683 | ||
1684 | ||
1685 | ||
2450418c | 1686 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1687 | struct kvec *section, |
1688 | unsigned int sec_len, u32 *crc) | |
2450418c | 1689 | { |
68b4476b | 1690 | int ret, left; |
2450418c YS |
1691 | |
1692 | BUG_ON(!section); | |
1693 | ||
1694 | while (section->iov_len < sec_len) { | |
1695 | BUG_ON(section->iov_base == NULL); | |
1696 | left = sec_len - section->iov_len; | |
1697 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1698 | section->iov_len, left); | |
1699 | if (ret <= 0) | |
1700 | return ret; | |
1701 | section->iov_len += ret; | |
2450418c | 1702 | } |
fe3ad593 AE |
1703 | if (section->iov_len == sec_len) |
1704 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1705 | |
2450418c YS |
1706 | return 1; |
1707 | } | |
31b8006e | 1708 | |
1c20f2d2 AE |
1709 | static bool ceph_con_in_msg_alloc(struct ceph_connection *con, |
1710 | struct ceph_msg_header *hdr); | |
68b4476b YS |
1711 | |
1712 | ||
1713 | static int read_partial_message_pages(struct ceph_connection *con, | |
1714 | struct page **pages, | |
95c96174 | 1715 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1716 | { |
1717 | void *p; | |
1718 | int ret; | |
1719 | int left; | |
1720 | ||
1721 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1722 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1723 | /* (page) data */ | |
1724 | BUG_ON(pages == NULL); | |
1725 | p = kmap(pages[con->in_msg_pos.page]); | |
1726 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1727 | left); | |
bca064d2 | 1728 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1729 | con->in_data_crc = |
1730 | crc32c(con->in_data_crc, | |
1731 | p + con->in_msg_pos.page_pos, ret); | |
1732 | kunmap(pages[con->in_msg_pos.page]); | |
1733 | if (ret <= 0) | |
1734 | return ret; | |
1735 | con->in_msg_pos.data_pos += ret; | |
1736 | con->in_msg_pos.page_pos += ret; | |
1737 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1738 | con->in_msg_pos.page_pos = 0; | |
1739 | con->in_msg_pos.page++; | |
1740 | } | |
1741 | ||
1742 | return ret; | |
1743 | } | |
1744 | ||
1745 | #ifdef CONFIG_BLOCK | |
1746 | static int read_partial_message_bio(struct ceph_connection *con, | |
1747 | struct bio **bio_iter, int *bio_seg, | |
95c96174 | 1748 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1749 | { |
1750 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1751 | void *p; | |
1752 | int ret, left; | |
1753 | ||
68b4476b YS |
1754 | left = min((int)(data_len - con->in_msg_pos.data_pos), |
1755 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1756 | ||
1757 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1758 | ||
1759 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1760 | left); | |
bca064d2 | 1761 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1762 | con->in_data_crc = |
1763 | crc32c(con->in_data_crc, | |
1764 | p + con->in_msg_pos.page_pos, ret); | |
1765 | kunmap(bv->bv_page); | |
1766 | if (ret <= 0) | |
1767 | return ret; | |
1768 | con->in_msg_pos.data_pos += ret; | |
1769 | con->in_msg_pos.page_pos += ret; | |
1770 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1771 | con->in_msg_pos.page_pos = 0; | |
1772 | iter_bio_next(bio_iter, bio_seg); | |
1773 | } | |
1774 | ||
1775 | return ret; | |
1776 | } | |
1777 | #endif | |
1778 | ||
31b8006e SW |
1779 | /* |
1780 | * read (part of) a message. | |
1781 | */ | |
1782 | static int read_partial_message(struct ceph_connection *con) | |
1783 | { | |
1784 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1785 | int size; |
1786 | int end; | |
31b8006e | 1787 | int ret; |
95c96174 | 1788 | unsigned int front_len, middle_len, data_len; |
37675b0f | 1789 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 1790 | u64 seq; |
fe3ad593 | 1791 | u32 crc; |
31b8006e SW |
1792 | |
1793 | dout("read_partial_message con %p msg %p\n", con, m); | |
1794 | ||
1795 | /* header */ | |
fd51653f AE |
1796 | size = sizeof (con->in_hdr); |
1797 | end = size; | |
1798 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1799 | if (ret <= 0) |
1800 | return ret; | |
fe3ad593 AE |
1801 | |
1802 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1803 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1804 | pr_err("read_partial_message bad hdr " | |
1805 | " crc %u != expected %u\n", | |
1806 | crc, con->in_hdr.crc); | |
1807 | return -EBADMSG; | |
1808 | } | |
1809 | ||
31b8006e SW |
1810 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1811 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1812 | return -EIO; | |
1813 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1814 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1815 | return -EIO; | |
1816 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1817 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1818 | return -EIO; | |
1819 | ||
ae18756b SW |
1820 | /* verify seq# */ |
1821 | seq = le64_to_cpu(con->in_hdr.seq); | |
1822 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1823 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1824 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1825 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1826 | seq, con->in_seq + 1); |
1827 | con->in_base_pos = -front_len - middle_len - data_len - | |
1828 | sizeof(m->footer); | |
1829 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1830 | return 0; |
1831 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1832 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1833 | seq, con->in_seq + 1); | |
1834 | con->error_msg = "bad message sequence # for incoming message"; | |
1835 | return -EBADMSG; | |
1836 | } | |
1837 | ||
31b8006e SW |
1838 | /* allocate message? */ |
1839 | if (!con->in_msg) { | |
1840 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1841 | con->in_hdr.front_len, con->in_hdr.data_len); | |
1c20f2d2 | 1842 | if (ceph_con_in_msg_alloc(con, &con->in_hdr)) { |
31b8006e | 1843 | /* skip this message */ |
a79832f2 | 1844 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1845 | BUG_ON(con->in_msg); |
31b8006e SW |
1846 | con->in_base_pos = -front_len - middle_len - data_len - |
1847 | sizeof(m->footer); | |
1848 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1849 | con->in_seq++; |
31b8006e SW |
1850 | return 0; |
1851 | } | |
a79832f2 | 1852 | if (!con->in_msg) { |
5b3a4db3 SW |
1853 | con->error_msg = |
1854 | "error allocating memory for incoming message"; | |
a79832f2 | 1855 | return -ENOMEM; |
31b8006e | 1856 | } |
38941f80 AE |
1857 | |
1858 | BUG_ON(con->in_msg->con != con); | |
31b8006e SW |
1859 | m = con->in_msg; |
1860 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1861 | if (m->middle) |
1862 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1863 | |
1864 | con->in_msg_pos.page = 0; | |
68b4476b | 1865 | if (m->pages) |
c5c6b19d | 1866 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1867 | else |
1868 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1869 | con->in_msg_pos.data_pos = 0; |
a4107026 SW |
1870 | |
1871 | #ifdef CONFIG_BLOCK | |
1872 | if (m->bio) | |
1873 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1874 | #endif | |
31b8006e SW |
1875 | } |
1876 | ||
1877 | /* front */ | |
2450418c YS |
1878 | ret = read_partial_message_section(con, &m->front, front_len, |
1879 | &con->in_front_crc); | |
1880 | if (ret <= 0) | |
1881 | return ret; | |
31b8006e SW |
1882 | |
1883 | /* middle */ | |
2450418c | 1884 | if (m->middle) { |
213c99ee SW |
1885 | ret = read_partial_message_section(con, &m->middle->vec, |
1886 | middle_len, | |
2450418c | 1887 | &con->in_middle_crc); |
31b8006e SW |
1888 | if (ret <= 0) |
1889 | return ret; | |
31b8006e SW |
1890 | } |
1891 | ||
1892 | /* (page) data */ | |
31b8006e | 1893 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1894 | if (m->pages) { |
1895 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1896 | data_len, do_datacrc); |
68b4476b YS |
1897 | if (ret <= 0) |
1898 | return ret; | |
1899 | #ifdef CONFIG_BLOCK | |
1900 | } else if (m->bio) { | |
a4107026 | 1901 | BUG_ON(!m->bio_iter); |
68b4476b YS |
1902 | ret = read_partial_message_bio(con, |
1903 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1904 | data_len, do_datacrc); |
68b4476b YS |
1905 | if (ret <= 0) |
1906 | return ret; | |
1907 | #endif | |
1908 | } else { | |
1909 | BUG_ON(1); | |
31b8006e SW |
1910 | } |
1911 | } | |
1912 | ||
31b8006e | 1913 | /* footer */ |
fd51653f AE |
1914 | size = sizeof (m->footer); |
1915 | end += size; | |
1916 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
1917 | if (ret <= 0) |
1918 | return ret; | |
1919 | ||
31b8006e SW |
1920 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
1921 | m, front_len, m->footer.front_crc, middle_len, | |
1922 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1923 | ||
1924 | /* crc ok? */ | |
1925 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1926 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1927 | m, con->in_front_crc, m->footer.front_crc); | |
1928 | return -EBADMSG; | |
1929 | } | |
1930 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1931 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1932 | m, con->in_middle_crc, m->footer.middle_crc); | |
1933 | return -EBADMSG; | |
1934 | } | |
bca064d2 | 1935 | if (do_datacrc && |
31b8006e SW |
1936 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
1937 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1938 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1939 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1940 | return -EBADMSG; | |
1941 | } | |
1942 | ||
1943 | return 1; /* done! */ | |
1944 | } | |
1945 | ||
1946 | /* | |
1947 | * Process message. This happens in the worker thread. The callback should | |
1948 | * be careful not to do anything that waits on other incoming messages or it | |
1949 | * may deadlock. | |
1950 | */ | |
1951 | static void process_message(struct ceph_connection *con) | |
1952 | { | |
5e095e8b | 1953 | struct ceph_msg *msg; |
31b8006e | 1954 | |
38941f80 AE |
1955 | BUG_ON(con->in_msg->con != con); |
1956 | con->in_msg->con = NULL; | |
5e095e8b | 1957 | msg = con->in_msg; |
31b8006e | 1958 | con->in_msg = NULL; |
36eb71aa | 1959 | con->ops->put(con); |
31b8006e SW |
1960 | |
1961 | /* if first message, set peer_name */ | |
1962 | if (con->peer_name.type == 0) | |
dbad185d | 1963 | con->peer_name = msg->hdr.src; |
31b8006e | 1964 | |
31b8006e | 1965 | con->in_seq++; |
ec302645 | 1966 | mutex_unlock(&con->mutex); |
31b8006e SW |
1967 | |
1968 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1969 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1970 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1971 | le16_to_cpu(msg->hdr.type), |
1972 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1973 | le32_to_cpu(msg->hdr.front_len), | |
1974 | le32_to_cpu(msg->hdr.data_len), | |
1975 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1976 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1977 | |
1978 | mutex_lock(&con->mutex); | |
31b8006e SW |
1979 | prepare_read_tag(con); |
1980 | } | |
1981 | ||
1982 | ||
1983 | /* | |
1984 | * Write something to the socket. Called in a worker thread when the | |
1985 | * socket appears to be writeable and we have something ready to send. | |
1986 | */ | |
1987 | static int try_write(struct ceph_connection *con) | |
1988 | { | |
31b8006e SW |
1989 | int ret = 1; |
1990 | ||
d59315ca | 1991 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 1992 | |
31b8006e SW |
1993 | more: |
1994 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
1995 | ||
1996 | /* open the socket first? */ | |
1997 | if (con->sock == NULL) { | |
a5988c49 AE |
1998 | set_bit(CONNECTING, &con->state); |
1999 | ||
e2200423 | 2000 | con_out_kvec_reset(con); |
e825a66d | 2001 | prepare_write_banner(con); |
eed0ef2c | 2002 | prepare_read_banner(con); |
31b8006e | 2003 | |
cf3e5c40 | 2004 | BUG_ON(con->in_msg); |
31b8006e SW |
2005 | con->in_tag = CEPH_MSGR_TAG_READY; |
2006 | dout("try_write initiating connect on %p new state %lu\n", | |
2007 | con, con->state); | |
41617d0c AE |
2008 | ret = ceph_tcp_connect(con); |
2009 | if (ret < 0) { | |
31b8006e | 2010 | con->error_msg = "connect error"; |
31b8006e SW |
2011 | goto out; |
2012 | } | |
2013 | } | |
2014 | ||
2015 | more_kvec: | |
2016 | /* kvec data queued? */ | |
2017 | if (con->out_skip) { | |
2018 | ret = write_partial_skip(con); | |
2019 | if (ret <= 0) | |
42961d23 | 2020 | goto out; |
31b8006e SW |
2021 | } |
2022 | if (con->out_kvec_left) { | |
2023 | ret = write_partial_kvec(con); | |
2024 | if (ret <= 0) | |
42961d23 | 2025 | goto out; |
31b8006e SW |
2026 | } |
2027 | ||
2028 | /* msg pages? */ | |
2029 | if (con->out_msg) { | |
c86a2930 SW |
2030 | if (con->out_msg_done) { |
2031 | ceph_msg_put(con->out_msg); | |
2032 | con->out_msg = NULL; /* we're done with this one */ | |
2033 | goto do_next; | |
2034 | } | |
2035 | ||
31b8006e SW |
2036 | ret = write_partial_msg_pages(con); |
2037 | if (ret == 1) | |
2038 | goto more_kvec; /* we need to send the footer, too! */ | |
2039 | if (ret == 0) | |
42961d23 | 2040 | goto out; |
31b8006e SW |
2041 | if (ret < 0) { |
2042 | dout("try_write write_partial_msg_pages err %d\n", | |
2043 | ret); | |
42961d23 | 2044 | goto out; |
31b8006e SW |
2045 | } |
2046 | } | |
2047 | ||
c86a2930 | 2048 | do_next: |
7593af92 AE |
2049 | if (!test_bit(CONNECTING, &con->state) && |
2050 | !test_bit(NEGOTIATING, &con->state)) { | |
31b8006e SW |
2051 | /* is anything else pending? */ |
2052 | if (!list_empty(&con->out_queue)) { | |
2053 | prepare_write_message(con); | |
2054 | goto more; | |
2055 | } | |
2056 | if (con->in_seq > con->in_seq_acked) { | |
2057 | prepare_write_ack(con); | |
2058 | goto more; | |
2059 | } | |
928443cd | 2060 | if (test_and_clear_bit(KEEPALIVE_PENDING, &con->flags)) { |
31b8006e SW |
2061 | prepare_write_keepalive(con); |
2062 | goto more; | |
2063 | } | |
2064 | } | |
2065 | ||
2066 | /* Nothing to do! */ | |
928443cd | 2067 | clear_bit(WRITE_PENDING, &con->flags); |
31b8006e | 2068 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2069 | ret = 0; |
2070 | out: | |
42961d23 | 2071 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2072 | return ret; |
2073 | } | |
2074 | ||
2075 | ||
2076 | ||
2077 | /* | |
2078 | * Read what we can from the socket. | |
2079 | */ | |
2080 | static int try_read(struct ceph_connection *con) | |
2081 | { | |
31b8006e SW |
2082 | int ret = -1; |
2083 | ||
2084 | if (!con->sock) | |
2085 | return 0; | |
2086 | ||
2087 | if (test_bit(STANDBY, &con->state)) | |
2088 | return 0; | |
2089 | ||
2090 | dout("try_read start on %p\n", con); | |
ec302645 | 2091 | |
31b8006e SW |
2092 | more: |
2093 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, | |
2094 | con->in_base_pos); | |
0da5d703 SW |
2095 | |
2096 | /* | |
2097 | * process_connect and process_message drop and re-take | |
2098 | * con->mutex. make sure we handle a racing close or reopen. | |
2099 | */ | |
2100 | if (test_bit(CLOSED, &con->state) || | |
2101 | test_bit(OPENING, &con->state)) { | |
2102 | ret = -EAGAIN; | |
2103 | goto out; | |
2104 | } | |
2105 | ||
31b8006e | 2106 | if (test_bit(CONNECTING, &con->state)) { |
7593af92 AE |
2107 | dout("try_read connecting\n"); |
2108 | ret = read_partial_banner(con); | |
2109 | if (ret <= 0) | |
ab166d5a | 2110 | goto out; |
7593af92 AE |
2111 | ret = process_banner(con); |
2112 | if (ret < 0) | |
2113 | goto out; | |
2114 | ||
2115 | clear_bit(CONNECTING, &con->state); | |
2116 | set_bit(NEGOTIATING, &con->state); | |
2117 | ||
2118 | /* Banner is good, exchange connection info */ | |
2119 | ret = prepare_write_connect(con); | |
2120 | if (ret < 0) | |
2121 | goto out; | |
2122 | prepare_read_connect(con); | |
2123 | ||
2124 | /* Send connection info before awaiting response */ | |
2125 | goto out; | |
2126 | } | |
2127 | ||
2128 | if (test_bit(NEGOTIATING, &con->state)) { | |
2129 | dout("try_read negotiating\n"); | |
31b8006e SW |
2130 | ret = read_partial_connect(con); |
2131 | if (ret <= 0) | |
31b8006e | 2132 | goto out; |
98bdb0aa SW |
2133 | ret = process_connect(con); |
2134 | if (ret < 0) | |
2135 | goto out; | |
31b8006e SW |
2136 | goto more; |
2137 | } | |
2138 | ||
2139 | if (con->in_base_pos < 0) { | |
2140 | /* | |
2141 | * skipping + discarding content. | |
2142 | * | |
2143 | * FIXME: there must be a better way to do this! | |
2144 | */ | |
84495f49 AE |
2145 | static char buf[SKIP_BUF_SIZE]; |
2146 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2147 | ||
31b8006e SW |
2148 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2149 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2150 | if (ret <= 0) | |
98bdb0aa | 2151 | goto out; |
31b8006e SW |
2152 | con->in_base_pos += ret; |
2153 | if (con->in_base_pos) | |
2154 | goto more; | |
2155 | } | |
2156 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2157 | /* | |
2158 | * what's next? | |
2159 | */ | |
2160 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2161 | if (ret <= 0) | |
98bdb0aa | 2162 | goto out; |
31b8006e SW |
2163 | dout("try_read got tag %d\n", (int)con->in_tag); |
2164 | switch (con->in_tag) { | |
2165 | case CEPH_MSGR_TAG_MSG: | |
2166 | prepare_read_message(con); | |
2167 | break; | |
2168 | case CEPH_MSGR_TAG_ACK: | |
2169 | prepare_read_ack(con); | |
2170 | break; | |
2171 | case CEPH_MSGR_TAG_CLOSE: | |
e27947c7 | 2172 | clear_bit(CONNECTED, &con->state); |
31b8006e | 2173 | set_bit(CLOSED, &con->state); /* fixme */ |
98bdb0aa | 2174 | goto out; |
31b8006e SW |
2175 | default: |
2176 | goto bad_tag; | |
2177 | } | |
2178 | } | |
2179 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2180 | ret = read_partial_message(con); | |
2181 | if (ret <= 0) { | |
2182 | switch (ret) { | |
2183 | case -EBADMSG: | |
2184 | con->error_msg = "bad crc"; | |
2185 | ret = -EIO; | |
98bdb0aa | 2186 | break; |
31b8006e SW |
2187 | case -EIO: |
2188 | con->error_msg = "io error"; | |
98bdb0aa | 2189 | break; |
31b8006e | 2190 | } |
98bdb0aa | 2191 | goto out; |
31b8006e SW |
2192 | } |
2193 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2194 | goto more; | |
2195 | process_message(con); | |
2196 | goto more; | |
2197 | } | |
2198 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2199 | ret = read_partial_ack(con); | |
2200 | if (ret <= 0) | |
98bdb0aa | 2201 | goto out; |
31b8006e SW |
2202 | process_ack(con); |
2203 | goto more; | |
2204 | } | |
2205 | ||
31b8006e | 2206 | out: |
98bdb0aa | 2207 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2208 | return ret; |
2209 | ||
2210 | bad_tag: | |
2211 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2212 | con->error_msg = "protocol error, garbage tag"; | |
2213 | ret = -1; | |
2214 | goto out; | |
2215 | } | |
2216 | ||
2217 | ||
2218 | /* | |
2219 | * Atomically queue work on a connection. Bump @con reference to | |
2220 | * avoid races with connection teardown. | |
31b8006e SW |
2221 | */ |
2222 | static void queue_con(struct ceph_connection *con) | |
2223 | { | |
31b8006e SW |
2224 | if (!con->ops->get(con)) { |
2225 | dout("queue_con %p ref count 0\n", con); | |
2226 | return; | |
2227 | } | |
2228 | ||
f363e45f | 2229 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2230 | dout("queue_con %p - already queued\n", con); |
2231 | con->ops->put(con); | |
2232 | } else { | |
2233 | dout("queue_con %p\n", con); | |
2234 | } | |
2235 | } | |
2236 | ||
2237 | /* | |
2238 | * Do some work on a connection. Drop a connection ref when we're done. | |
2239 | */ | |
2240 | static void con_work(struct work_struct *work) | |
2241 | { | |
2242 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2243 | work.work); | |
0da5d703 | 2244 | int ret; |
31b8006e | 2245 | |
9dd4658d | 2246 | mutex_lock(&con->mutex); |
0da5d703 | 2247 | restart: |
188048bc | 2248 | if (test_and_clear_bit(SOCK_CLOSED, &con->flags)) { |
e27947c7 AE |
2249 | if (test_and_clear_bit(CONNECTED, &con->state)) |
2250 | con->error_msg = "socket closed"; | |
7593af92 AE |
2251 | else if (test_and_clear_bit(NEGOTIATING, &con->state)) |
2252 | con->error_msg = "negotiation failed"; | |
2253 | else if (test_and_clear_bit(CONNECTING, &con->state)) | |
188048bc | 2254 | con->error_msg = "connection failed"; |
7593af92 | 2255 | else |
e27947c7 | 2256 | con->error_msg = "unrecognized con state"; |
188048bc AE |
2257 | goto fault; |
2258 | } | |
2259 | ||
928443cd | 2260 | if (test_and_clear_bit(BACKOFF, &con->flags)) { |
60bf8bf8 SW |
2261 | dout("con_work %p backing off\n", con); |
2262 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2263 | round_jiffies_relative(con->delay))) { | |
2264 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2265 | mutex_unlock(&con->mutex); | |
2266 | return; | |
2267 | } else { | |
2268 | con->ops->put(con); | |
2269 | dout("con_work %p FAILED to back off %lu\n", con, | |
2270 | con->delay); | |
2271 | } | |
2272 | } | |
9dd4658d | 2273 | |
e00de341 SW |
2274 | if (test_bit(STANDBY, &con->state)) { |
2275 | dout("con_work %p STANDBY\n", con); | |
2276 | goto done; | |
2277 | } | |
2e8cb100 SW |
2278 | if (test_bit(CLOSED, &con->state)) { |
2279 | dout("con_work %p CLOSED\n", con); | |
2280 | BUG_ON(con->sock); | |
31b8006e SW |
2281 | goto done; |
2282 | } | |
2283 | if (test_and_clear_bit(OPENING, &con->state)) { | |
2284 | /* reopen w/ new peer */ | |
2285 | dout("con_work OPENING\n"); | |
2e8cb100 | 2286 | BUG_ON(con->sock); |
31b8006e SW |
2287 | } |
2288 | ||
0da5d703 SW |
2289 | ret = try_read(con); |
2290 | if (ret == -EAGAIN) | |
2291 | goto restart; | |
3a140a0d SW |
2292 | if (ret < 0) { |
2293 | con->error_msg = "socket error on read"; | |
0da5d703 | 2294 | goto fault; |
3a140a0d | 2295 | } |
0da5d703 SW |
2296 | |
2297 | ret = try_write(con); | |
2298 | if (ret == -EAGAIN) | |
2299 | goto restart; | |
3a140a0d SW |
2300 | if (ret < 0) { |
2301 | con->error_msg = "socket error on write"; | |
0da5d703 | 2302 | goto fault; |
3a140a0d | 2303 | } |
31b8006e SW |
2304 | |
2305 | done: | |
9dd4658d | 2306 | mutex_unlock(&con->mutex); |
9dd4658d | 2307 | done_unlocked: |
31b8006e | 2308 | con->ops->put(con); |
0da5d703 SW |
2309 | return; |
2310 | ||
2311 | fault: | |
2312 | mutex_unlock(&con->mutex); | |
2313 | ceph_fault(con); /* error/fault path */ | |
2314 | goto done_unlocked; | |
31b8006e SW |
2315 | } |
2316 | ||
2317 | ||
2318 | /* | |
2319 | * Generic error/fault handler. A retry mechanism is used with | |
2320 | * exponential backoff | |
2321 | */ | |
2322 | static void ceph_fault(struct ceph_connection *con) | |
2323 | { | |
3b5ede07 SW |
2324 | mutex_lock(&con->mutex); |
2325 | ||
31b8006e | 2326 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2327 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2328 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2329 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2330 | |
91e45ce3 SW |
2331 | if (test_bit(CLOSED, &con->state)) |
2332 | goto out_unlock; | |
ec302645 | 2333 | |
31b8006e | 2334 | con_close_socket(con); |
5e095e8b | 2335 | |
3b5ede07 SW |
2336 | if (test_bit(LOSSYTX, &con->flags)) { |
2337 | dout("fault on LOSSYTX channel\n"); | |
2338 | goto out_unlock; | |
2339 | } | |
2340 | ||
5e095e8b | 2341 | if (con->in_msg) { |
38941f80 AE |
2342 | BUG_ON(con->in_msg->con != con); |
2343 | con->in_msg->con = NULL; | |
5e095e8b SW |
2344 | ceph_msg_put(con->in_msg); |
2345 | con->in_msg = NULL; | |
36eb71aa | 2346 | con->ops->put(con); |
5e095e8b | 2347 | } |
31b8006e | 2348 | |
e80a52d1 SW |
2349 | /* Requeue anything that hasn't been acked */ |
2350 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2351 | |
e76661d0 SW |
2352 | /* If there are no messages queued or keepalive pending, place |
2353 | * the connection in a STANDBY state */ | |
2354 | if (list_empty(&con->out_queue) && | |
928443cd | 2355 | !test_bit(KEEPALIVE_PENDING, &con->flags)) { |
e00de341 | 2356 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
928443cd | 2357 | clear_bit(WRITE_PENDING, &con->flags); |
31b8006e | 2358 | set_bit(STANDBY, &con->state); |
e80a52d1 SW |
2359 | } else { |
2360 | /* retry after a delay. */ | |
2361 | if (con->delay == 0) | |
2362 | con->delay = BASE_DELAY_INTERVAL; | |
2363 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2364 | con->delay *= 2; | |
e80a52d1 SW |
2365 | con->ops->get(con); |
2366 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2367 | round_jiffies_relative(con->delay))) { |
2368 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2369 | } else { | |
e80a52d1 | 2370 | con->ops->put(con); |
60bf8bf8 SW |
2371 | dout("fault failed to queue %p delay %lu, backoff\n", |
2372 | con, con->delay); | |
2373 | /* | |
2374 | * In many cases we see a socket state change | |
2375 | * while con_work is running and end up | |
2376 | * queuing (non-delayed) work, such that we | |
2377 | * can't backoff with a delay. Set a flag so | |
2378 | * that when con_work restarts we schedule the | |
2379 | * delay then. | |
2380 | */ | |
928443cd | 2381 | set_bit(BACKOFF, &con->flags); |
60bf8bf8 | 2382 | } |
31b8006e SW |
2383 | } |
2384 | ||
91e45ce3 SW |
2385 | out_unlock: |
2386 | mutex_unlock(&con->mutex); | |
161fd65a SW |
2387 | /* |
2388 | * in case we faulted due to authentication, invalidate our | |
2389 | * current tickets so that we can get new ones. | |
213c99ee | 2390 | */ |
161fd65a SW |
2391 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2392 | dout("calling invalidate_authorizer()\n"); | |
2393 | con->ops->invalidate_authorizer(con); | |
2394 | } | |
2395 | ||
31b8006e SW |
2396 | if (con->ops->fault) |
2397 | con->ops->fault(con); | |
2398 | } | |
2399 | ||
2400 | ||
2401 | ||
2402 | /* | |
15d9882c | 2403 | * initialize a new messenger instance |
31b8006e | 2404 | */ |
15d9882c AE |
2405 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2406 | struct ceph_entity_addr *myaddr, | |
2407 | u32 supported_features, | |
2408 | u32 required_features, | |
2409 | bool nocrc) | |
31b8006e | 2410 | { |
3d14c5d2 YS |
2411 | msgr->supported_features = supported_features; |
2412 | msgr->required_features = required_features; | |
2413 | ||
31b8006e SW |
2414 | spin_lock_init(&msgr->global_seq_lock); |
2415 | ||
31b8006e SW |
2416 | if (myaddr) |
2417 | msgr->inst.addr = *myaddr; | |
2418 | ||
2419 | /* select a random nonce */ | |
ac8839d7 | 2420 | msgr->inst.addr.type = 0; |
103e2d3a | 2421 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2422 | encode_my_addr(msgr); |
15d9882c | 2423 | msgr->nocrc = nocrc; |
31b8006e | 2424 | |
a2a32584 GH |
2425 | atomic_set(&msgr->stopping, 0); |
2426 | ||
15d9882c | 2427 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2428 | } |
15d9882c | 2429 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2430 | |
e00de341 SW |
2431 | static void clear_standby(struct ceph_connection *con) |
2432 | { | |
2433 | /* come back from STANDBY? */ | |
2434 | if (test_and_clear_bit(STANDBY, &con->state)) { | |
e00de341 SW |
2435 | dout("clear_standby %p and ++connect_seq\n", con); |
2436 | con->connect_seq++; | |
928443cd AE |
2437 | WARN_ON(test_bit(WRITE_PENDING, &con->flags)); |
2438 | WARN_ON(test_bit(KEEPALIVE_PENDING, &con->flags)); | |
e00de341 SW |
2439 | } |
2440 | } | |
2441 | ||
31b8006e SW |
2442 | /* |
2443 | * Queue up an outgoing message on the given connection. | |
2444 | */ | |
2445 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2446 | { | |
31b8006e | 2447 | /* set src+dst */ |
dbad185d | 2448 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 2449 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
2450 | msg->needs_out_seq = true; |
2451 | ||
ec302645 | 2452 | mutex_lock(&con->mutex); |
92ce034b | 2453 | |
a59b55a6 SW |
2454 | if (test_bit(CLOSED, &con->state)) { |
2455 | dout("con_send %p closed, dropping %p\n", con, msg); | |
2456 | ceph_msg_put(msg); | |
2457 | mutex_unlock(&con->mutex); | |
2458 | return; | |
2459 | } | |
2460 | ||
38941f80 | 2461 | BUG_ON(msg->con != NULL); |
36eb71aa | 2462 | msg->con = con->ops->get(con); |
92ce034b AE |
2463 | BUG_ON(msg->con == NULL); |
2464 | ||
31b8006e SW |
2465 | BUG_ON(!list_empty(&msg->list_head)); |
2466 | list_add_tail(&msg->list_head, &con->out_queue); | |
2467 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2468 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2469 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2470 | le32_to_cpu(msg->hdr.front_len), | |
2471 | le32_to_cpu(msg->hdr.middle_len), | |
2472 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
2473 | |
2474 | clear_standby(con); | |
ec302645 | 2475 | mutex_unlock(&con->mutex); |
31b8006e SW |
2476 | |
2477 | /* if there wasn't anything waiting to send before, queue | |
2478 | * new work */ | |
928443cd | 2479 | if (test_and_set_bit(WRITE_PENDING, &con->flags) == 0) |
31b8006e SW |
2480 | queue_con(con); |
2481 | } | |
3d14c5d2 | 2482 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2483 | |
2484 | /* | |
2485 | * Revoke a message that was previously queued for send | |
2486 | */ | |
6740a845 | 2487 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2488 | { |
6740a845 AE |
2489 | struct ceph_connection *con = msg->con; |
2490 | ||
2491 | if (!con) | |
2492 | return; /* Message not in our possession */ | |
2493 | ||
ec302645 | 2494 | mutex_lock(&con->mutex); |
31b8006e | 2495 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2496 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2497 | list_del_init(&msg->list_head); |
38941f80 | 2498 | BUG_ON(msg->con == NULL); |
36eb71aa | 2499 | msg->con->ops->put(msg->con); |
38941f80 | 2500 | msg->con = NULL; |
92ce034b | 2501 | msg->hdr.seq = 0; |
38941f80 | 2502 | |
31b8006e | 2503 | ceph_msg_put(msg); |
ed98adad SW |
2504 | } |
2505 | if (con->out_msg == msg) { | |
38941f80 | 2506 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2507 | con->out_msg = NULL; |
31b8006e SW |
2508 | if (con->out_kvec_is_msg) { |
2509 | con->out_skip = con->out_kvec_bytes; | |
2510 | con->out_kvec_is_msg = false; | |
2511 | } | |
ed98adad | 2512 | msg->hdr.seq = 0; |
92ce034b AE |
2513 | |
2514 | ceph_msg_put(msg); | |
31b8006e | 2515 | } |
ec302645 | 2516 | mutex_unlock(&con->mutex); |
31b8006e SW |
2517 | } |
2518 | ||
350b1c32 | 2519 | /* |
0d59ab81 | 2520 | * Revoke a message that we may be reading data into |
350b1c32 | 2521 | */ |
8921d114 | 2522 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2523 | { |
8921d114 AE |
2524 | struct ceph_connection *con; |
2525 | ||
2526 | BUG_ON(msg == NULL); | |
2527 | if (!msg->con) { | |
2528 | dout("%s msg %p null con\n", __func__, msg); | |
2529 | ||
2530 | return; /* Message not in our possession */ | |
2531 | } | |
2532 | ||
2533 | con = msg->con; | |
350b1c32 | 2534 | mutex_lock(&con->mutex); |
8921d114 | 2535 | if (con->in_msg == msg) { |
95c96174 ED |
2536 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
2537 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
2538 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
2539 | |
2540 | /* skip rest of message */ | |
8921d114 AE |
2541 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
2542 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 2543 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
2544 | front_len - |
2545 | middle_len - | |
2546 | data_len - | |
350b1c32 | 2547 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2548 | ceph_msg_put(con->in_msg); |
2549 | con->in_msg = NULL; | |
2550 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2551 | con->in_seq++; |
350b1c32 | 2552 | } else { |
8921d114 AE |
2553 | dout("%s %p in_msg %p msg %p no-op\n", |
2554 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
2555 | } |
2556 | mutex_unlock(&con->mutex); | |
2557 | } | |
2558 | ||
31b8006e SW |
2559 | /* |
2560 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2561 | */ | |
2562 | void ceph_con_keepalive(struct ceph_connection *con) | |
2563 | { | |
e00de341 | 2564 | dout("con_keepalive %p\n", con); |
00650931 | 2565 | mutex_lock(&con->mutex); |
e00de341 | 2566 | clear_standby(con); |
00650931 | 2567 | mutex_unlock(&con->mutex); |
928443cd AE |
2568 | if (test_and_set_bit(KEEPALIVE_PENDING, &con->flags) == 0 && |
2569 | test_and_set_bit(WRITE_PENDING, &con->flags) == 0) | |
31b8006e SW |
2570 | queue_con(con); |
2571 | } | |
3d14c5d2 | 2572 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2573 | |
2574 | ||
2575 | /* | |
2576 | * construct a new message with given type, size | |
2577 | * the new msg has a ref count of 1. | |
2578 | */ | |
b61c2763 SW |
2579 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2580 | bool can_fail) | |
31b8006e SW |
2581 | { |
2582 | struct ceph_msg *m; | |
2583 | ||
34d23762 | 2584 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2585 | if (m == NULL) |
2586 | goto out; | |
c2e552e7 | 2587 | kref_init(&m->kref); |
38941f80 AE |
2588 | |
2589 | m->con = NULL; | |
31b8006e SW |
2590 | INIT_LIST_HEAD(&m->list_head); |
2591 | ||
45c6ceb5 | 2592 | m->hdr.tid = 0; |
31b8006e | 2593 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2594 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2595 | m->hdr.version = 0; | |
31b8006e SW |
2596 | m->hdr.front_len = cpu_to_le32(front_len); |
2597 | m->hdr.middle_len = 0; | |
bb257664 SW |
2598 | m->hdr.data_len = 0; |
2599 | m->hdr.data_off = 0; | |
45c6ceb5 | 2600 | m->hdr.reserved = 0; |
31b8006e SW |
2601 | m->footer.front_crc = 0; |
2602 | m->footer.middle_crc = 0; | |
2603 | m->footer.data_crc = 0; | |
45c6ceb5 | 2604 | m->footer.flags = 0; |
31b8006e SW |
2605 | m->front_max = front_len; |
2606 | m->front_is_vmalloc = false; | |
2607 | m->more_to_follow = false; | |
c0d5f9db | 2608 | m->ack_stamp = 0; |
31b8006e SW |
2609 | m->pool = NULL; |
2610 | ||
ca20892d HC |
2611 | /* middle */ |
2612 | m->middle = NULL; | |
2613 | ||
2614 | /* data */ | |
2615 | m->nr_pages = 0; | |
2616 | m->page_alignment = 0; | |
2617 | m->pages = NULL; | |
2618 | m->pagelist = NULL; | |
2619 | m->bio = NULL; | |
2620 | m->bio_iter = NULL; | |
2621 | m->bio_seg = 0; | |
2622 | m->trail = NULL; | |
2623 | ||
31b8006e SW |
2624 | /* front */ |
2625 | if (front_len) { | |
2626 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2627 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2628 | PAGE_KERNEL); |
2629 | m->front_is_vmalloc = true; | |
2630 | } else { | |
34d23762 | 2631 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2632 | } |
2633 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2634 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2635 | front_len); |
2636 | goto out2; | |
2637 | } | |
2638 | } else { | |
2639 | m->front.iov_base = NULL; | |
2640 | } | |
2641 | m->front.iov_len = front_len; | |
2642 | ||
bb257664 | 2643 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2644 | return m; |
2645 | ||
2646 | out2: | |
2647 | ceph_msg_put(m); | |
2648 | out: | |
b61c2763 SW |
2649 | if (!can_fail) { |
2650 | pr_err("msg_new can't create type %d front %d\n", type, | |
2651 | front_len); | |
f0ed1b7c | 2652 | WARN_ON(1); |
b61c2763 SW |
2653 | } else { |
2654 | dout("msg_new can't create type %d front %d\n", type, | |
2655 | front_len); | |
2656 | } | |
a79832f2 | 2657 | return NULL; |
31b8006e | 2658 | } |
3d14c5d2 | 2659 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2660 | |
31b8006e SW |
2661 | /* |
2662 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2663 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2664 | * per-type header in the front and then gracefully fail (i.e., | |
2665 | * propagate the error to the caller based on info in the front) when | |
2666 | * the middle is too large. | |
2667 | */ | |
2450418c | 2668 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2669 | { |
2670 | int type = le16_to_cpu(msg->hdr.type); | |
2671 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2672 | ||
2673 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2674 | ceph_msg_type_name(type), middle_len); | |
2675 | BUG_ON(!middle_len); | |
2676 | BUG_ON(msg->middle); | |
2677 | ||
b6c1d5b8 | 2678 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2679 | if (!msg->middle) |
2680 | return -ENOMEM; | |
2681 | return 0; | |
2682 | } | |
2683 | ||
2450418c | 2684 | /* |
1c20f2d2 AE |
2685 | * Allocate a message for receiving an incoming message on a |
2686 | * connection, and save the result in con->in_msg. Uses the | |
2687 | * connection's private alloc_msg op if available. | |
2688 | * | |
2689 | * Returns true if the message should be skipped, false otherwise. | |
2690 | * If true is returned (skip message), con->in_msg will be NULL. | |
2691 | * If false is returned, con->in_msg will contain a pointer to the | |
2692 | * newly-allocated message, or NULL in case of memory exhaustion. | |
2450418c | 2693 | */ |
1c20f2d2 AE |
2694 | static bool ceph_con_in_msg_alloc(struct ceph_connection *con, |
2695 | struct ceph_msg_header *hdr) | |
2450418c YS |
2696 | { |
2697 | int type = le16_to_cpu(hdr->type); | |
2698 | int front_len = le32_to_cpu(hdr->front_len); | |
2699 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2450418c YS |
2700 | int ret; |
2701 | ||
1c20f2d2 AE |
2702 | BUG_ON(con->in_msg != NULL); |
2703 | ||
2450418c | 2704 | if (con->ops->alloc_msg) { |
1c20f2d2 AE |
2705 | int skip = 0; |
2706 | ||
0547a9b3 | 2707 | mutex_unlock(&con->mutex); |
1c20f2d2 | 2708 | con->in_msg = con->ops->alloc_msg(con, hdr, &skip); |
0547a9b3 | 2709 | mutex_lock(&con->mutex); |
92ce034b | 2710 | if (con->in_msg) { |
36eb71aa | 2711 | con->in_msg->con = con->ops->get(con); |
92ce034b AE |
2712 | BUG_ON(con->in_msg->con == NULL); |
2713 | } | |
1c20f2d2 AE |
2714 | if (skip) |
2715 | con->in_msg = NULL; | |
2716 | ||
2717 | if (!con->in_msg) | |
2718 | return skip != 0; | |
2450418c | 2719 | } |
1c20f2d2 AE |
2720 | if (!con->in_msg) { |
2721 | con->in_msg = ceph_msg_new(type, front_len, GFP_NOFS, false); | |
2722 | if (!con->in_msg) { | |
2450418c YS |
2723 | pr_err("unable to allocate msg type %d len %d\n", |
2724 | type, front_len); | |
1c20f2d2 | 2725 | return false; |
2450418c | 2726 | } |
36eb71aa | 2727 | con->in_msg->con = con->ops->get(con); |
92ce034b | 2728 | BUG_ON(con->in_msg->con == NULL); |
1c20f2d2 | 2729 | con->in_msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2730 | } |
1c20f2d2 | 2731 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2732 | |
1c20f2d2 AE |
2733 | if (middle_len && !con->in_msg->middle) { |
2734 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 2735 | if (ret < 0) { |
1c20f2d2 AE |
2736 | ceph_msg_put(con->in_msg); |
2737 | con->in_msg = NULL; | |
2450418c YS |
2738 | } |
2739 | } | |
9d7f0f13 | 2740 | |
1c20f2d2 | 2741 | return false; |
2450418c YS |
2742 | } |
2743 | ||
31b8006e SW |
2744 | |
2745 | /* | |
2746 | * Free a generically kmalloc'd message. | |
2747 | */ | |
2748 | void ceph_msg_kfree(struct ceph_msg *m) | |
2749 | { | |
2750 | dout("msg_kfree %p\n", m); | |
2751 | if (m->front_is_vmalloc) | |
2752 | vfree(m->front.iov_base); | |
2753 | else | |
2754 | kfree(m->front.iov_base); | |
2755 | kfree(m); | |
2756 | } | |
2757 | ||
2758 | /* | |
2759 | * Drop a msg ref. Destroy as needed. | |
2760 | */ | |
c2e552e7 SW |
2761 | void ceph_msg_last_put(struct kref *kref) |
2762 | { | |
2763 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2764 | |
c2e552e7 SW |
2765 | dout("ceph_msg_put last one on %p\n", m); |
2766 | WARN_ON(!list_empty(&m->list_head)); | |
2767 | ||
2768 | /* drop middle, data, if any */ | |
2769 | if (m->middle) { | |
2770 | ceph_buffer_put(m->middle); | |
2771 | m->middle = NULL; | |
31b8006e | 2772 | } |
c2e552e7 SW |
2773 | m->nr_pages = 0; |
2774 | m->pages = NULL; | |
2775 | ||
58bb3b37 SW |
2776 | if (m->pagelist) { |
2777 | ceph_pagelist_release(m->pagelist); | |
2778 | kfree(m->pagelist); | |
2779 | m->pagelist = NULL; | |
2780 | } | |
2781 | ||
68b4476b YS |
2782 | m->trail = NULL; |
2783 | ||
c2e552e7 SW |
2784 | if (m->pool) |
2785 | ceph_msgpool_put(m->pool, m); | |
2786 | else | |
2787 | ceph_msg_kfree(m); | |
31b8006e | 2788 | } |
3d14c5d2 | 2789 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2790 | |
2791 | void ceph_msg_dump(struct ceph_msg *msg) | |
2792 | { | |
2793 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2794 | msg->front_max, msg->nr_pages); | |
2795 | print_hex_dump(KERN_DEBUG, "header: ", | |
2796 | DUMP_PREFIX_OFFSET, 16, 1, | |
2797 | &msg->hdr, sizeof(msg->hdr), true); | |
2798 | print_hex_dump(KERN_DEBUG, " front: ", | |
2799 | DUMP_PREFIX_OFFSET, 16, 1, | |
2800 | msg->front.iov_base, msg->front.iov_len, true); | |
2801 | if (msg->middle) | |
2802 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2803 | DUMP_PREFIX_OFFSET, 16, 1, | |
2804 | msg->middle->vec.iov_base, | |
2805 | msg->middle->vec.iov_len, true); | |
2806 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2807 | DUMP_PREFIX_OFFSET, 16, 1, | |
2808 | &msg->footer, sizeof(msg->footer), true); | |
2809 | } | |
3d14c5d2 | 2810 | EXPORT_SYMBOL(ceph_msg_dump); |