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 | ||
299 | if (test_bit(CLOSED, &con->state)) | |
300 | return; | |
301 | ||
302 | switch (sk->sk_state) { | |
303 | case TCP_CLOSE: | |
327800bd | 304 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 305 | case TCP_CLOSE_WAIT: |
327800bd | 306 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 307 | con_sock_state_closing(con); |
d65c9e0b AE |
308 | set_bit(SOCK_CLOSED, &con->flags); |
309 | queue_con(con); | |
31b8006e SW |
310 | break; |
311 | case TCP_ESTABLISHED: | |
327800bd | 312 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 313 | con_sock_state_connected(con); |
31b8006e SW |
314 | queue_con(con); |
315 | break; | |
d3002b97 AE |
316 | default: /* Everything else is uninteresting */ |
317 | break; | |
31b8006e SW |
318 | } |
319 | } | |
320 | ||
321 | /* | |
322 | * set up socket callbacks | |
323 | */ | |
324 | static void set_sock_callbacks(struct socket *sock, | |
325 | struct ceph_connection *con) | |
326 | { | |
327 | struct sock *sk = sock->sk; | |
bd406145 | 328 | sk->sk_user_data = con; |
327800bd AE |
329 | sk->sk_data_ready = ceph_sock_data_ready; |
330 | sk->sk_write_space = ceph_sock_write_space; | |
331 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
332 | } |
333 | ||
334 | ||
335 | /* | |
336 | * socket helpers | |
337 | */ | |
338 | ||
339 | /* | |
340 | * initiate connection to a remote socket. | |
341 | */ | |
41617d0c | 342 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 343 | { |
f91d3471 | 344 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
345 | struct socket *sock; |
346 | int ret; | |
347 | ||
348 | BUG_ON(con->sock); | |
f91d3471 SW |
349 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
350 | IPPROTO_TCP, &sock); | |
31b8006e | 351 | if (ret) |
41617d0c | 352 | return ret; |
31b8006e SW |
353 | sock->sk->sk_allocation = GFP_NOFS; |
354 | ||
a6a5349d SW |
355 | #ifdef CONFIG_LOCKDEP |
356 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
357 | #endif | |
358 | ||
31b8006e SW |
359 | set_sock_callbacks(sock, con); |
360 | ||
3d14c5d2 | 361 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 362 | |
89a86be0 | 363 | con_sock_state_connecting(con); |
f91d3471 SW |
364 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
365 | O_NONBLOCK); | |
31b8006e SW |
366 | if (ret == -EINPROGRESS) { |
367 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 368 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 369 | sock->sk->sk_state); |
a5bc3129 | 370 | } else if (ret < 0) { |
31b8006e | 371 | pr_err("connect %s error %d\n", |
3d14c5d2 | 372 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 373 | sock_release(sock); |
31b8006e | 374 | con->error_msg = "connect error"; |
31b8006e | 375 | |
41617d0c | 376 | return ret; |
a5bc3129 AE |
377 | } |
378 | con->sock = sock; | |
41617d0c | 379 | return 0; |
31b8006e SW |
380 | } |
381 | ||
382 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
383 | { | |
384 | struct kvec iov = {buf, len}; | |
385 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 386 | int r; |
31b8006e | 387 | |
98bdb0aa SW |
388 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
389 | if (r == -EAGAIN) | |
390 | r = 0; | |
391 | return r; | |
31b8006e SW |
392 | } |
393 | ||
394 | /* | |
395 | * write something. @more is true if caller will be sending more data | |
396 | * shortly. | |
397 | */ | |
398 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
399 | size_t kvlen, size_t len, int more) | |
400 | { | |
401 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 402 | int r; |
31b8006e SW |
403 | |
404 | if (more) | |
405 | msg.msg_flags |= MSG_MORE; | |
406 | else | |
407 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
408 | ||
42961d23 SW |
409 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
410 | if (r == -EAGAIN) | |
411 | r = 0; | |
412 | return r; | |
31b8006e SW |
413 | } |
414 | ||
31739139 AE |
415 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
416 | int offset, size_t size, int more) | |
417 | { | |
418 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
419 | int ret; | |
420 | ||
421 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
422 | if (ret == -EAGAIN) | |
423 | ret = 0; | |
424 | ||
425 | return ret; | |
426 | } | |
427 | ||
31b8006e SW |
428 | |
429 | /* | |
430 | * Shutdown/close the socket for the given connection. | |
431 | */ | |
432 | static int con_close_socket(struct ceph_connection *con) | |
433 | { | |
434 | int rc; | |
435 | ||
436 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
437 | if (!con->sock) | |
438 | return 0; | |
31b8006e SW |
439 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); |
440 | sock_release(con->sock); | |
441 | con->sock = NULL; | |
456ea468 AE |
442 | |
443 | /* | |
444 | * Forcibly clear the SOCK_CLOSE flag. It gets set | |
445 | * independent of the connection mutex, and we could have | |
446 | * received a socket close event before we had the chance to | |
447 | * shut the socket down. | |
448 | */ | |
a8d00e3c | 449 | clear_bit(SOCK_CLOSED, &con->flags); |
ce2c8903 | 450 | con_sock_state_closed(con); |
31b8006e SW |
451 | return rc; |
452 | } | |
453 | ||
454 | /* | |
455 | * Reset a connection. Discard all incoming and outgoing messages | |
456 | * and clear *_seq state. | |
457 | */ | |
458 | static void ceph_msg_remove(struct ceph_msg *msg) | |
459 | { | |
460 | list_del_init(&msg->list_head); | |
38941f80 | 461 | BUG_ON(msg->con == NULL); |
36eb71aa | 462 | msg->con->ops->put(msg->con); |
38941f80 AE |
463 | msg->con = NULL; |
464 | ||
31b8006e SW |
465 | ceph_msg_put(msg); |
466 | } | |
467 | static void ceph_msg_remove_list(struct list_head *head) | |
468 | { | |
469 | while (!list_empty(head)) { | |
470 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
471 | list_head); | |
472 | ceph_msg_remove(msg); | |
473 | } | |
474 | } | |
475 | ||
476 | static void reset_connection(struct ceph_connection *con) | |
477 | { | |
478 | /* reset connection, out_queue, msg_ and connect_seq */ | |
479 | /* discard existing out_queue and msg_seq */ | |
31b8006e SW |
480 | ceph_msg_remove_list(&con->out_queue); |
481 | ceph_msg_remove_list(&con->out_sent); | |
482 | ||
cf3e5c40 | 483 | if (con->in_msg) { |
38941f80 AE |
484 | BUG_ON(con->in_msg->con != con); |
485 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
486 | ceph_msg_put(con->in_msg); |
487 | con->in_msg = NULL; | |
36eb71aa | 488 | con->ops->put(con); |
cf3e5c40 SW |
489 | } |
490 | ||
31b8006e SW |
491 | con->connect_seq = 0; |
492 | con->out_seq = 0; | |
c86a2930 SW |
493 | if (con->out_msg) { |
494 | ceph_msg_put(con->out_msg); | |
495 | con->out_msg = NULL; | |
496 | } | |
31b8006e | 497 | con->in_seq = 0; |
0e0d5e0c | 498 | con->in_seq_acked = 0; |
31b8006e SW |
499 | } |
500 | ||
501 | /* | |
502 | * mark a peer down. drop any open connections. | |
503 | */ | |
504 | void ceph_con_close(struct ceph_connection *con) | |
505 | { | |
8c50c817 | 506 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
507 | dout("con_close %p peer %s\n", con, |
508 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
a5988c49 | 509 | clear_bit(NEGOTIATING, &con->state); |
bb9e6bba | 510 | clear_bit(CONNECTING, &con->state); |
e27947c7 | 511 | clear_bit(CONNECTED, &con->state); |
31b8006e | 512 | clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */ |
a5988c49 AE |
513 | set_bit(CLOSED, &con->state); |
514 | ||
928443cd AE |
515 | clear_bit(LOSSYTX, &con->flags); /* so we retry next connect */ |
516 | clear_bit(KEEPALIVE_PENDING, &con->flags); | |
517 | clear_bit(WRITE_PENDING, &con->flags); | |
a5988c49 | 518 | |
31b8006e | 519 | reset_connection(con); |
6f2bc3ff | 520 | con->peer_global_seq = 0; |
91e45ce3 | 521 | cancel_delayed_work(&con->work); |
ec302645 | 522 | mutex_unlock(&con->mutex); |
8c50c817 SW |
523 | |
524 | /* | |
525 | * We cannot close the socket directly from here because the | |
526 | * work threads use it without holding the mutex. Instead, let | |
527 | * con_work() do it. | |
528 | */ | |
31b8006e SW |
529 | queue_con(con); |
530 | } | |
3d14c5d2 | 531 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 532 | |
31b8006e SW |
533 | /* |
534 | * Reopen a closed connection, with a new peer address. | |
535 | */ | |
b7a9e5dd SW |
536 | void ceph_con_open(struct ceph_connection *con, |
537 | __u8 entity_type, __u64 entity_num, | |
538 | struct ceph_entity_addr *addr) | |
31b8006e | 539 | { |
3d14c5d2 | 540 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
31b8006e | 541 | set_bit(OPENING, &con->state); |
a5988c49 AE |
542 | WARN_ON(!test_and_clear_bit(CLOSED, &con->state)); |
543 | ||
b7a9e5dd SW |
544 | con->peer_name.type = (__u8) entity_type; |
545 | con->peer_name.num = cpu_to_le64(entity_num); | |
546 | ||
31b8006e | 547 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 548 | con->delay = 0; /* reset backoff memory */ |
31b8006e SW |
549 | queue_con(con); |
550 | } | |
3d14c5d2 | 551 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 552 | |
87b315a5 SW |
553 | /* |
554 | * return true if this connection ever successfully opened | |
555 | */ | |
556 | bool ceph_con_opened(struct ceph_connection *con) | |
557 | { | |
558 | return con->connect_seq > 0; | |
559 | } | |
560 | ||
31b8006e SW |
561 | /* |
562 | * initialize a new connection. | |
563 | */ | |
1bfd89f4 AE |
564 | void ceph_con_init(struct ceph_connection *con, void *private, |
565 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 566 | struct ceph_messenger *msgr) |
31b8006e SW |
567 | { |
568 | dout("con_init %p\n", con); | |
569 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
570 | con->private = private; |
571 | con->ops = ops; | |
31b8006e | 572 | con->msgr = msgr; |
ce2c8903 AE |
573 | |
574 | con_sock_state_init(con); | |
575 | ||
ec302645 | 576 | mutex_init(&con->mutex); |
31b8006e SW |
577 | INIT_LIST_HEAD(&con->out_queue); |
578 | INIT_LIST_HEAD(&con->out_sent); | |
579 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 AE |
580 | |
581 | set_bit(CLOSED, &con->state); | |
31b8006e | 582 | } |
3d14c5d2 | 583 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
584 | |
585 | ||
586 | /* | |
587 | * We maintain a global counter to order connection attempts. Get | |
588 | * a unique seq greater than @gt. | |
589 | */ | |
590 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
591 | { | |
592 | u32 ret; | |
593 | ||
594 | spin_lock(&msgr->global_seq_lock); | |
595 | if (msgr->global_seq < gt) | |
596 | msgr->global_seq = gt; | |
597 | ret = ++msgr->global_seq; | |
598 | spin_unlock(&msgr->global_seq_lock); | |
599 | return ret; | |
600 | } | |
601 | ||
e2200423 | 602 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
603 | { |
604 | con->out_kvec_left = 0; | |
605 | con->out_kvec_bytes = 0; | |
606 | con->out_kvec_cur = &con->out_kvec[0]; | |
607 | } | |
608 | ||
e2200423 | 609 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
610 | size_t size, void *data) |
611 | { | |
612 | int index; | |
613 | ||
614 | index = con->out_kvec_left; | |
615 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
616 | ||
617 | con->out_kvec[index].iov_len = size; | |
618 | con->out_kvec[index].iov_base = data; | |
619 | con->out_kvec_left++; | |
620 | con->out_kvec_bytes += size; | |
621 | } | |
31b8006e | 622 | |
df6ad1f9 AE |
623 | #ifdef CONFIG_BLOCK |
624 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
625 | { | |
626 | if (!bio) { | |
627 | *iter = NULL; | |
628 | *seg = 0; | |
629 | return; | |
630 | } | |
631 | *iter = bio; | |
632 | *seg = bio->bi_idx; | |
633 | } | |
634 | ||
635 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
636 | { | |
637 | if (*bio_iter == NULL) | |
638 | return; | |
639 | ||
640 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
641 | ||
642 | (*seg)++; | |
643 | if (*seg == (*bio_iter)->bi_vcnt) | |
644 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
645 | } | |
646 | #endif | |
647 | ||
739c905b AE |
648 | static void prepare_write_message_data(struct ceph_connection *con) |
649 | { | |
650 | struct ceph_msg *msg = con->out_msg; | |
651 | ||
652 | BUG_ON(!msg); | |
653 | BUG_ON(!msg->hdr.data_len); | |
654 | ||
655 | /* initialize page iterator */ | |
656 | con->out_msg_pos.page = 0; | |
657 | if (msg->pages) | |
658 | con->out_msg_pos.page_pos = msg->page_alignment; | |
659 | else | |
660 | con->out_msg_pos.page_pos = 0; | |
572c588e | 661 | #ifdef CONFIG_BLOCK |
abdaa6a8 | 662 | if (msg->bio) |
572c588e AE |
663 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); |
664 | #endif | |
739c905b AE |
665 | con->out_msg_pos.data_pos = 0; |
666 | con->out_msg_pos.did_page_crc = false; | |
667 | con->out_more = 1; /* data + footer will follow */ | |
668 | } | |
669 | ||
31b8006e SW |
670 | /* |
671 | * Prepare footer for currently outgoing message, and finish things | |
672 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
673 | */ | |
859eb799 | 674 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
675 | { |
676 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 677 | int v = con->out_kvec_left; |
31b8006e | 678 | |
fd154f3c AE |
679 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
680 | ||
31b8006e SW |
681 | dout("prepare_write_message_footer %p\n", con); |
682 | con->out_kvec_is_msg = true; | |
683 | con->out_kvec[v].iov_base = &m->footer; | |
684 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
685 | con->out_kvec_bytes += sizeof(m->footer); | |
686 | con->out_kvec_left++; | |
687 | con->out_more = m->more_to_follow; | |
c86a2930 | 688 | con->out_msg_done = true; |
31b8006e SW |
689 | } |
690 | ||
691 | /* | |
692 | * Prepare headers for the next outgoing message. | |
693 | */ | |
694 | static void prepare_write_message(struct ceph_connection *con) | |
695 | { | |
696 | struct ceph_msg *m; | |
a9a0c51a | 697 | u32 crc; |
31b8006e | 698 | |
e2200423 | 699 | con_out_kvec_reset(con); |
31b8006e | 700 | con->out_kvec_is_msg = true; |
c86a2930 | 701 | con->out_msg_done = false; |
31b8006e SW |
702 | |
703 | /* Sneak an ack in there first? If we can get it into the same | |
704 | * TCP packet that's a good thing. */ | |
705 | if (con->in_seq > con->in_seq_acked) { | |
706 | con->in_seq_acked = con->in_seq; | |
e2200423 | 707 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 708 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 709 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 710 | &con->out_temp_ack); |
31b8006e SW |
711 | } |
712 | ||
38941f80 | 713 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 714 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 715 | con->out_msg = m; |
38941f80 | 716 | BUG_ON(m->con != con); |
4cf9d544 SW |
717 | |
718 | /* put message on sent list */ | |
719 | ceph_msg_get(m); | |
720 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 721 | |
e84346b7 SW |
722 | /* |
723 | * only assign outgoing seq # if we haven't sent this message | |
724 | * yet. if it is requeued, resend with it's original seq. | |
725 | */ | |
726 | if (m->needs_out_seq) { | |
727 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
728 | m->needs_out_seq = false; | |
729 | } | |
31b8006e SW |
730 | |
731 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
732 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
733 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
734 | le32_to_cpu(m->hdr.data_len), | |
735 | m->nr_pages); | |
736 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); | |
737 | ||
738 | /* tag + hdr + front + middle */ | |
e2200423 AE |
739 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
740 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
741 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 742 | |
31b8006e | 743 | if (m->middle) |
e2200423 | 744 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 745 | m->middle->vec.iov_base); |
31b8006e SW |
746 | |
747 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
748 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
749 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 750 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
751 | |
752 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
753 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
754 | if (m->middle) { | |
755 | crc = crc32c(0, m->middle->vec.iov_base, | |
756 | m->middle->vec.iov_len); | |
757 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
758 | } else | |
31b8006e | 759 | con->out_msg->footer.middle_crc = 0; |
739c905b | 760 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
761 | le32_to_cpu(con->out_msg->footer.front_crc), |
762 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
763 | ||
764 | /* is there a data payload? */ | |
739c905b AE |
765 | con->out_msg->footer.data_crc = 0; |
766 | if (m->hdr.data_len) | |
767 | prepare_write_message_data(con); | |
768 | else | |
31b8006e | 769 | /* no, queue up footer too and be done */ |
859eb799 | 770 | prepare_write_message_footer(con); |
31b8006e | 771 | |
928443cd | 772 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
773 | } |
774 | ||
775 | /* | |
776 | * Prepare an ack. | |
777 | */ | |
778 | static void prepare_write_ack(struct ceph_connection *con) | |
779 | { | |
780 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
781 | con->in_seq_acked, con->in_seq); | |
782 | con->in_seq_acked = con->in_seq; | |
783 | ||
e2200423 | 784 | con_out_kvec_reset(con); |
859eb799 | 785 | |
e2200423 | 786 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 787 | |
31b8006e | 788 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 789 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
790 | &con->out_temp_ack); |
791 | ||
31b8006e | 792 | con->out_more = 1; /* more will follow.. eventually.. */ |
928443cd | 793 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
794 | } |
795 | ||
796 | /* | |
797 | * Prepare to write keepalive byte. | |
798 | */ | |
799 | static void prepare_write_keepalive(struct ceph_connection *con) | |
800 | { | |
801 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
802 | con_out_kvec_reset(con); |
803 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
928443cd | 804 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
805 | } |
806 | ||
807 | /* | |
808 | * Connection negotiation. | |
809 | */ | |
810 | ||
dac1e716 AE |
811 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
812 | int *auth_proto) | |
4e7a5dcd | 813 | { |
a3530df3 | 814 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
815 | |
816 | if (!con->ops->get_authorizer) { | |
817 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
818 | con->out_connect.authorizer_len = 0; | |
819 | ||
729796be | 820 | return NULL; |
b1c6b980 AE |
821 | } |
822 | ||
823 | /* Can't hold the mutex while getting authorizer */ | |
4e7a5dcd | 824 | |
ec302645 | 825 | mutex_unlock(&con->mutex); |
b1c6b980 | 826 | |
dac1e716 | 827 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
8f43fb53 | 828 | |
ec302645 | 829 | mutex_lock(&con->mutex); |
4e7a5dcd | 830 | |
a3530df3 | 831 | if (IS_ERR(auth)) |
729796be | 832 | return auth; |
928443cd | 833 | if (test_bit(CLOSED, &con->state) || test_bit(OPENING, &con->flags)) |
729796be | 834 | return ERR_PTR(-EAGAIN); |
0da5d703 | 835 | |
8f43fb53 AE |
836 | con->auth_reply_buf = auth->authorizer_reply_buf; |
837 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
838 | ||
859eb799 | 839 | |
729796be | 840 | return auth; |
4e7a5dcd SW |
841 | } |
842 | ||
31b8006e SW |
843 | /* |
844 | * We connected to a peer and are saying hello. | |
845 | */ | |
e825a66d | 846 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 847 | { |
e2200423 AE |
848 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
849 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 850 | &con->msgr->my_enc_addr); |
eed0ef2c | 851 | |
eed0ef2c | 852 | con->out_more = 0; |
928443cd | 853 | set_bit(WRITE_PENDING, &con->flags); |
eed0ef2c SW |
854 | } |
855 | ||
e825a66d | 856 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 857 | { |
95c96174 | 858 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 859 | int proto; |
dac1e716 | 860 | int auth_proto; |
729796be | 861 | struct ceph_auth_handshake *auth; |
31b8006e SW |
862 | |
863 | switch (con->peer_name.type) { | |
864 | case CEPH_ENTITY_TYPE_MON: | |
865 | proto = CEPH_MONC_PROTOCOL; | |
866 | break; | |
867 | case CEPH_ENTITY_TYPE_OSD: | |
868 | proto = CEPH_OSDC_PROTOCOL; | |
869 | break; | |
870 | case CEPH_ENTITY_TYPE_MDS: | |
871 | proto = CEPH_MDSC_PROTOCOL; | |
872 | break; | |
873 | default: | |
874 | BUG(); | |
875 | } | |
876 | ||
877 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
878 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 879 | |
e825a66d | 880 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
881 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
882 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
883 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
884 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
885 | con->out_connect.flags = 0; | |
31b8006e | 886 | |
dac1e716 AE |
887 | auth_proto = CEPH_AUTH_UNKNOWN; |
888 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
889 | if (IS_ERR(auth)) |
890 | return PTR_ERR(auth); | |
3da54776 | 891 | |
dac1e716 | 892 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
893 | con->out_connect.authorizer_len = auth ? |
894 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
895 | ||
ab166d5a | 896 | con_out_kvec_reset(con); |
e2200423 | 897 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
898 | &con->out_connect); |
899 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 900 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 901 | auth->authorizer_buf); |
859eb799 | 902 | |
31b8006e | 903 | con->out_more = 0; |
928443cd | 904 | set_bit(WRITE_PENDING, &con->flags); |
4e7a5dcd | 905 | |
e10c758e | 906 | return 0; |
31b8006e SW |
907 | } |
908 | ||
31b8006e SW |
909 | /* |
910 | * write as much of pending kvecs to the socket as we can. | |
911 | * 1 -> done | |
912 | * 0 -> socket full, but more to do | |
913 | * <0 -> error | |
914 | */ | |
915 | static int write_partial_kvec(struct ceph_connection *con) | |
916 | { | |
917 | int ret; | |
918 | ||
919 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
920 | while (con->out_kvec_bytes > 0) { | |
921 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
922 | con->out_kvec_left, con->out_kvec_bytes, | |
923 | con->out_more); | |
924 | if (ret <= 0) | |
925 | goto out; | |
926 | con->out_kvec_bytes -= ret; | |
927 | if (con->out_kvec_bytes == 0) | |
928 | break; /* done */ | |
f42299e6 AE |
929 | |
930 | /* account for full iov entries consumed */ | |
931 | while (ret >= con->out_kvec_cur->iov_len) { | |
932 | BUG_ON(!con->out_kvec_left); | |
933 | ret -= con->out_kvec_cur->iov_len; | |
934 | con->out_kvec_cur++; | |
935 | con->out_kvec_left--; | |
936 | } | |
937 | /* and for a partially-consumed entry */ | |
938 | if (ret) { | |
939 | con->out_kvec_cur->iov_len -= ret; | |
940 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
941 | } |
942 | } | |
943 | con->out_kvec_left = 0; | |
944 | con->out_kvec_is_msg = false; | |
945 | ret = 1; | |
946 | out: | |
947 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
948 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
949 | return ret; /* done! */ | |
950 | } | |
951 | ||
84ca8fc8 AE |
952 | static void out_msg_pos_next(struct ceph_connection *con, struct page *page, |
953 | size_t len, size_t sent, bool in_trail) | |
954 | { | |
955 | struct ceph_msg *msg = con->out_msg; | |
956 | ||
957 | BUG_ON(!msg); | |
958 | BUG_ON(!sent); | |
959 | ||
960 | con->out_msg_pos.data_pos += sent; | |
961 | con->out_msg_pos.page_pos += sent; | |
5821bd8c AE |
962 | if (sent < len) |
963 | return; | |
964 | ||
965 | BUG_ON(sent != len); | |
966 | con->out_msg_pos.page_pos = 0; | |
967 | con->out_msg_pos.page++; | |
968 | con->out_msg_pos.did_page_crc = false; | |
969 | if (in_trail) | |
970 | list_move_tail(&page->lru, | |
971 | &msg->trail->head); | |
972 | else if (msg->pagelist) | |
973 | list_move_tail(&page->lru, | |
974 | &msg->pagelist->head); | |
84ca8fc8 | 975 | #ifdef CONFIG_BLOCK |
5821bd8c AE |
976 | else if (msg->bio) |
977 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
84ca8fc8 | 978 | #endif |
84ca8fc8 AE |
979 | } |
980 | ||
31b8006e SW |
981 | /* |
982 | * Write as much message data payload as we can. If we finish, queue | |
983 | * up the footer. | |
984 | * 1 -> done, footer is now queued in out_kvec[]. | |
985 | * 0 -> socket full, but more to do | |
986 | * <0 -> error | |
987 | */ | |
988 | static int write_partial_msg_pages(struct ceph_connection *con) | |
989 | { | |
990 | struct ceph_msg *msg = con->out_msg; | |
95c96174 | 991 | unsigned int data_len = le32_to_cpu(msg->hdr.data_len); |
31b8006e | 992 | size_t len; |
37675b0f | 993 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 994 | int ret; |
68b4476b | 995 | int total_max_write; |
84ca8fc8 | 996 | bool in_trail = false; |
5821bd8c AE |
997 | const size_t trail_len = (msg->trail ? msg->trail->length : 0); |
998 | const size_t trail_off = data_len - trail_len; | |
31b8006e SW |
999 | |
1000 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
84ca8fc8 | 1001 | con, msg, con->out_msg_pos.page, msg->nr_pages, |
31b8006e SW |
1002 | con->out_msg_pos.page_pos); |
1003 | ||
5821bd8c AE |
1004 | /* |
1005 | * Iterate through each page that contains data to be | |
1006 | * written, and send as much as possible for each. | |
1007 | * | |
1008 | * If we are calculating the data crc (the default), we will | |
1009 | * need to map the page. If we have no pages, they have | |
1010 | * been revoked, so use the zero page. | |
1011 | */ | |
68b4476b | 1012 | while (data_len > con->out_msg_pos.data_pos) { |
31b8006e | 1013 | struct page *page = NULL; |
68b4476b | 1014 | int max_write = PAGE_SIZE; |
9bd19663 | 1015 | int bio_offset = 0; |
68b4476b | 1016 | |
5821bd8c AE |
1017 | in_trail = in_trail || con->out_msg_pos.data_pos >= trail_off; |
1018 | if (!in_trail) | |
1019 | total_max_write = trail_off - con->out_msg_pos.data_pos; | |
68b4476b | 1020 | |
5821bd8c | 1021 | if (in_trail) { |
68b4476b YS |
1022 | total_max_write = data_len - con->out_msg_pos.data_pos; |
1023 | ||
1024 | page = list_first_entry(&msg->trail->head, | |
1025 | struct page, lru); | |
68b4476b | 1026 | } else if (msg->pages) { |
31b8006e | 1027 | page = msg->pages[con->out_msg_pos.page]; |
58bb3b37 SW |
1028 | } else if (msg->pagelist) { |
1029 | page = list_first_entry(&msg->pagelist->head, | |
1030 | struct page, lru); | |
68b4476b YS |
1031 | #ifdef CONFIG_BLOCK |
1032 | } else if (msg->bio) { | |
1033 | struct bio_vec *bv; | |
1034 | ||
1035 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
1036 | page = bv->bv_page; | |
9bd19663 | 1037 | bio_offset = bv->bv_offset; |
68b4476b YS |
1038 | max_write = bv->bv_len; |
1039 | #endif | |
31b8006e | 1040 | } else { |
57666519 | 1041 | page = zero_page; |
31b8006e | 1042 | } |
68b4476b YS |
1043 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
1044 | total_max_write); | |
1045 | ||
37675b0f | 1046 | if (do_datacrc && !con->out_msg_pos.did_page_crc) { |
9bd19663 | 1047 | void *base; |
5821bd8c | 1048 | u32 crc = le32_to_cpu(msg->footer.data_crc); |
8d63e318 | 1049 | char *kaddr; |
31b8006e | 1050 | |
8d63e318 | 1051 | kaddr = kmap(page); |
31b8006e | 1052 | BUG_ON(kaddr == NULL); |
9bd19663 | 1053 | base = kaddr + con->out_msg_pos.page_pos + bio_offset; |
5821bd8c | 1054 | crc = crc32c(crc, base, len); |
84ca8fc8 | 1055 | msg->footer.data_crc = cpu_to_le32(crc); |
bca064d2 | 1056 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 1057 | } |
e36b13cc | 1058 | ret = ceph_tcp_sendpage(con->sock, page, |
9bd19663 | 1059 | con->out_msg_pos.page_pos + bio_offset, |
e36b13cc | 1060 | len, 1); |
31b8006e | 1061 | |
0cdf9e60 | 1062 | if (do_datacrc) |
31b8006e SW |
1063 | kunmap(page); |
1064 | ||
1065 | if (ret <= 0) | |
1066 | goto out; | |
1067 | ||
84ca8fc8 | 1068 | out_msg_pos_next(con, page, len, (size_t) ret, in_trail); |
31b8006e SW |
1069 | } |
1070 | ||
1071 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
1072 | ||
1073 | /* prepare and queue up footer, too */ | |
37675b0f | 1074 | if (!do_datacrc) |
84ca8fc8 | 1075 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1076 | con_out_kvec_reset(con); |
859eb799 | 1077 | prepare_write_message_footer(con); |
31b8006e SW |
1078 | ret = 1; |
1079 | out: | |
1080 | return ret; | |
1081 | } | |
1082 | ||
1083 | /* | |
1084 | * write some zeros | |
1085 | */ | |
1086 | static int write_partial_skip(struct ceph_connection *con) | |
1087 | { | |
1088 | int ret; | |
1089 | ||
1090 | while (con->out_skip > 0) { | |
31739139 | 1091 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1092 | |
31739139 | 1093 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1); |
31b8006e SW |
1094 | if (ret <= 0) |
1095 | goto out; | |
1096 | con->out_skip -= ret; | |
1097 | } | |
1098 | ret = 1; | |
1099 | out: | |
1100 | return ret; | |
1101 | } | |
1102 | ||
1103 | /* | |
1104 | * Prepare to read connection handshake, or an ack. | |
1105 | */ | |
eed0ef2c SW |
1106 | static void prepare_read_banner(struct ceph_connection *con) |
1107 | { | |
1108 | dout("prepare_read_banner %p\n", con); | |
1109 | con->in_base_pos = 0; | |
1110 | } | |
1111 | ||
31b8006e SW |
1112 | static void prepare_read_connect(struct ceph_connection *con) |
1113 | { | |
1114 | dout("prepare_read_connect %p\n", con); | |
1115 | con->in_base_pos = 0; | |
1116 | } | |
1117 | ||
1118 | static void prepare_read_ack(struct ceph_connection *con) | |
1119 | { | |
1120 | dout("prepare_read_ack %p\n", con); | |
1121 | con->in_base_pos = 0; | |
1122 | } | |
1123 | ||
1124 | static void prepare_read_tag(struct ceph_connection *con) | |
1125 | { | |
1126 | dout("prepare_read_tag %p\n", con); | |
1127 | con->in_base_pos = 0; | |
1128 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * Prepare to read a message. | |
1133 | */ | |
1134 | static int prepare_read_message(struct ceph_connection *con) | |
1135 | { | |
1136 | dout("prepare_read_message %p\n", con); | |
1137 | BUG_ON(con->in_msg != NULL); | |
1138 | con->in_base_pos = 0; | |
1139 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1140 | return 0; | |
1141 | } | |
1142 | ||
1143 | ||
1144 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1145 | int end, int size, void *object) |
31b8006e | 1146 | { |
e6cee71f AE |
1147 | while (con->in_base_pos < end) { |
1148 | int left = end - con->in_base_pos; | |
31b8006e SW |
1149 | int have = size - left; |
1150 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1151 | if (ret <= 0) | |
1152 | return ret; | |
1153 | con->in_base_pos += ret; | |
1154 | } | |
1155 | return 1; | |
1156 | } | |
1157 | ||
1158 | ||
1159 | /* | |
1160 | * Read all or part of the connect-side handshake on a new connection | |
1161 | */ | |
eed0ef2c | 1162 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1163 | { |
fd51653f AE |
1164 | int size; |
1165 | int end; | |
1166 | int ret; | |
31b8006e | 1167 | |
eed0ef2c | 1168 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1169 | |
1170 | /* peer's banner */ | |
fd51653f AE |
1171 | size = strlen(CEPH_BANNER); |
1172 | end = size; | |
1173 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1174 | if (ret <= 0) |
1175 | goto out; | |
fd51653f AE |
1176 | |
1177 | size = sizeof (con->actual_peer_addr); | |
1178 | end += size; | |
1179 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1180 | if (ret <= 0) |
1181 | goto out; | |
fd51653f AE |
1182 | |
1183 | size = sizeof (con->peer_addr_for_me); | |
1184 | end += size; | |
1185 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1186 | if (ret <= 0) |
1187 | goto out; | |
fd51653f | 1188 | |
eed0ef2c SW |
1189 | out: |
1190 | return ret; | |
1191 | } | |
1192 | ||
1193 | static int read_partial_connect(struct ceph_connection *con) | |
1194 | { | |
fd51653f AE |
1195 | int size; |
1196 | int end; | |
1197 | int ret; | |
eed0ef2c SW |
1198 | |
1199 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1200 | ||
fd51653f AE |
1201 | size = sizeof (con->in_reply); |
1202 | end = size; | |
1203 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1204 | if (ret <= 0) |
1205 | goto out; | |
fd51653f AE |
1206 | |
1207 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1208 | end += size; | |
1209 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1210 | if (ret <= 0) |
1211 | goto out; | |
31b8006e | 1212 | |
4e7a5dcd SW |
1213 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1214 | con, (int)con->in_reply.tag, | |
1215 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1216 | le32_to_cpu(con->in_reply.global_seq)); |
1217 | out: | |
1218 | return ret; | |
eed0ef2c | 1219 | |
31b8006e SW |
1220 | } |
1221 | ||
1222 | /* | |
1223 | * Verify the hello banner looks okay. | |
1224 | */ | |
1225 | static int verify_hello(struct ceph_connection *con) | |
1226 | { | |
1227 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1228 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1229 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1230 | con->error_msg = "protocol error, bad banner"; |
1231 | return -1; | |
1232 | } | |
1233 | return 0; | |
1234 | } | |
1235 | ||
1236 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1237 | { | |
1238 | switch (ss->ss_family) { | |
1239 | case AF_INET: | |
1240 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1241 | case AF_INET6: | |
1242 | return | |
1243 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1244 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1245 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1246 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1247 | } | |
1248 | return false; | |
1249 | } | |
1250 | ||
1251 | static int addr_port(struct sockaddr_storage *ss) | |
1252 | { | |
1253 | switch (ss->ss_family) { | |
1254 | case AF_INET: | |
f28bcfbe | 1255 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1256 | case AF_INET6: |
f28bcfbe | 1257 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1258 | } |
1259 | return 0; | |
1260 | } | |
1261 | ||
1262 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1263 | { | |
1264 | switch (ss->ss_family) { | |
1265 | case AF_INET: | |
1266 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1267 | break; |
31b8006e SW |
1268 | case AF_INET6: |
1269 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1270 | break; |
31b8006e SW |
1271 | } |
1272 | } | |
1273 | ||
ee3b56f2 NW |
1274 | /* |
1275 | * Unlike other *_pton function semantics, zero indicates success. | |
1276 | */ | |
1277 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1278 | char delim, const char **ipend) | |
1279 | { | |
99f0f3b2 AE |
1280 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1281 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1282 | |
1283 | memset(ss, 0, sizeof(*ss)); | |
1284 | ||
1285 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1286 | ss->ss_family = AF_INET; | |
1287 | return 0; | |
1288 | } | |
1289 | ||
1290 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1291 | ss->ss_family = AF_INET6; | |
1292 | return 0; | |
1293 | } | |
1294 | ||
1295 | return -EINVAL; | |
1296 | } | |
1297 | ||
1298 | /* | |
1299 | * Extract hostname string and resolve using kernel DNS facility. | |
1300 | */ | |
1301 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1302 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1303 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1304 | { | |
1305 | const char *end, *delim_p; | |
1306 | char *colon_p, *ip_addr = NULL; | |
1307 | int ip_len, ret; | |
1308 | ||
1309 | /* | |
1310 | * The end of the hostname occurs immediately preceding the delimiter or | |
1311 | * the port marker (':') where the delimiter takes precedence. | |
1312 | */ | |
1313 | delim_p = memchr(name, delim, namelen); | |
1314 | colon_p = memchr(name, ':', namelen); | |
1315 | ||
1316 | if (delim_p && colon_p) | |
1317 | end = delim_p < colon_p ? delim_p : colon_p; | |
1318 | else if (!delim_p && colon_p) | |
1319 | end = colon_p; | |
1320 | else { | |
1321 | end = delim_p; | |
1322 | if (!end) /* case: hostname:/ */ | |
1323 | end = name + namelen; | |
1324 | } | |
1325 | ||
1326 | if (end <= name) | |
1327 | return -EINVAL; | |
1328 | ||
1329 | /* do dns_resolve upcall */ | |
1330 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1331 | if (ip_len > 0) | |
1332 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1333 | else | |
1334 | ret = -ESRCH; | |
1335 | ||
1336 | kfree(ip_addr); | |
1337 | ||
1338 | *ipend = end; | |
1339 | ||
1340 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1341 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1342 | ||
1343 | return ret; | |
1344 | } | |
1345 | #else | |
1346 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1347 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1348 | { | |
1349 | return -EINVAL; | |
1350 | } | |
1351 | #endif | |
1352 | ||
1353 | /* | |
1354 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1355 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1356 | */ | |
1357 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1358 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1359 | { | |
1360 | int ret; | |
1361 | ||
1362 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1363 | if (ret) | |
1364 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1365 | ||
1366 | return ret; | |
1367 | } | |
1368 | ||
31b8006e SW |
1369 | /* |
1370 | * Parse an ip[:port] list into an addr array. Use the default | |
1371 | * monitor port if a port isn't specified. | |
1372 | */ | |
1373 | int ceph_parse_ips(const char *c, const char *end, | |
1374 | struct ceph_entity_addr *addr, | |
1375 | int max_count, int *count) | |
1376 | { | |
ee3b56f2 | 1377 | int i, ret = -EINVAL; |
31b8006e SW |
1378 | const char *p = c; |
1379 | ||
1380 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1381 | for (i = 0; i < max_count; i++) { | |
1382 | const char *ipend; | |
1383 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1384 | int port; |
39139f64 SW |
1385 | char delim = ','; |
1386 | ||
1387 | if (*p == '[') { | |
1388 | delim = ']'; | |
1389 | p++; | |
1390 | } | |
31b8006e | 1391 | |
ee3b56f2 NW |
1392 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1393 | if (ret) | |
31b8006e | 1394 | goto bad; |
ee3b56f2 NW |
1395 | ret = -EINVAL; |
1396 | ||
31b8006e SW |
1397 | p = ipend; |
1398 | ||
39139f64 SW |
1399 | if (delim == ']') { |
1400 | if (*p != ']') { | |
1401 | dout("missing matching ']'\n"); | |
1402 | goto bad; | |
1403 | } | |
1404 | p++; | |
1405 | } | |
1406 | ||
31b8006e SW |
1407 | /* port? */ |
1408 | if (p < end && *p == ':') { | |
1409 | port = 0; | |
1410 | p++; | |
1411 | while (p < end && *p >= '0' && *p <= '9') { | |
1412 | port = (port * 10) + (*p - '0'); | |
1413 | p++; | |
1414 | } | |
1415 | if (port > 65535 || port == 0) | |
1416 | goto bad; | |
1417 | } else { | |
1418 | port = CEPH_MON_PORT; | |
1419 | } | |
1420 | ||
1421 | addr_set_port(ss, port); | |
1422 | ||
3d14c5d2 | 1423 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1424 | |
1425 | if (p == end) | |
1426 | break; | |
1427 | if (*p != ',') | |
1428 | goto bad; | |
1429 | p++; | |
1430 | } | |
1431 | ||
1432 | if (p != end) | |
1433 | goto bad; | |
1434 | ||
1435 | if (count) | |
1436 | *count = i + 1; | |
1437 | return 0; | |
1438 | ||
1439 | bad: | |
39139f64 | 1440 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1441 | return ret; |
31b8006e | 1442 | } |
3d14c5d2 | 1443 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1444 | |
eed0ef2c | 1445 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1446 | { |
eed0ef2c | 1447 | dout("process_banner on %p\n", con); |
31b8006e SW |
1448 | |
1449 | if (verify_hello(con) < 0) | |
1450 | return -1; | |
1451 | ||
63f2d211 SW |
1452 | ceph_decode_addr(&con->actual_peer_addr); |
1453 | ceph_decode_addr(&con->peer_addr_for_me); | |
1454 | ||
31b8006e SW |
1455 | /* |
1456 | * Make sure the other end is who we wanted. note that the other | |
1457 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1458 | * them the benefit of the doubt. | |
1459 | */ | |
103e2d3a SW |
1460 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1461 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1462 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1463 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1464 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1465 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1466 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1467 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1468 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1469 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1470 | return -1; |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * did we learn our address? | |
1475 | */ | |
1476 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1477 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1478 | ||
1479 | memcpy(&con->msgr->inst.addr.in_addr, | |
1480 | &con->peer_addr_for_me.in_addr, | |
1481 | sizeof(con->peer_addr_for_me.in_addr)); | |
1482 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1483 | encode_my_addr(con->msgr); |
eed0ef2c | 1484 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1485 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1486 | } |
1487 | ||
eed0ef2c SW |
1488 | return 0; |
1489 | } | |
1490 | ||
04a419f9 SW |
1491 | static void fail_protocol(struct ceph_connection *con) |
1492 | { | |
1493 | reset_connection(con); | |
1494 | set_bit(CLOSED, &con->state); /* in case there's queued work */ | |
04a419f9 SW |
1495 | } |
1496 | ||
eed0ef2c SW |
1497 | static int process_connect(struct ceph_connection *con) |
1498 | { | |
3d14c5d2 YS |
1499 | u64 sup_feat = con->msgr->supported_features; |
1500 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1501 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1502 | int ret; |
04a419f9 | 1503 | |
eed0ef2c SW |
1504 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1505 | ||
31b8006e | 1506 | switch (con->in_reply.tag) { |
04a419f9 SW |
1507 | case CEPH_MSGR_TAG_FEATURES: |
1508 | pr_err("%s%lld %s feature set mismatch," | |
1509 | " my %llx < server's %llx, missing %llx\n", | |
1510 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1511 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1512 | sup_feat, server_feat, server_feat & ~sup_feat); |
1513 | con->error_msg = "missing required protocol features"; | |
1514 | fail_protocol(con); | |
1515 | return -1; | |
1516 | ||
31b8006e | 1517 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1518 | pr_err("%s%lld %s protocol version mismatch," |
1519 | " my %d != server's %d\n", | |
1520 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1521 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1522 | le32_to_cpu(con->out_connect.protocol_version), |
1523 | le32_to_cpu(con->in_reply.protocol_version)); | |
1524 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1525 | fail_protocol(con); |
31b8006e SW |
1526 | return -1; |
1527 | ||
4e7a5dcd SW |
1528 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1529 | con->auth_retry++; | |
1530 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1531 | con->auth_retry); | |
1532 | if (con->auth_retry == 2) { | |
1533 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1534 | return -1; |
1535 | } | |
1536 | con->auth_retry = 1; | |
e825a66d | 1537 | ret = prepare_write_connect(con); |
0da5d703 SW |
1538 | if (ret < 0) |
1539 | return ret; | |
63733a0f | 1540 | prepare_read_connect(con); |
4e7a5dcd | 1541 | break; |
31b8006e SW |
1542 | |
1543 | case CEPH_MSGR_TAG_RESETSESSION: | |
1544 | /* | |
1545 | * If we connected with a large connect_seq but the peer | |
1546 | * has no record of a session with us (no connection, or | |
1547 | * connect_seq == 0), they will send RESETSESION to indicate | |
1548 | * that they must have reset their session, and may have | |
1549 | * dropped messages. | |
1550 | */ | |
1551 | dout("process_connect got RESET peer seq %u\n", | |
a16cb1f7 | 1552 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
1553 | pr_err("%s%lld %s connection reset\n", |
1554 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1555 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1556 | reset_connection(con); |
5a0f8fdd AE |
1557 | ret = prepare_write_connect(con); |
1558 | if (ret < 0) | |
1559 | return ret; | |
31b8006e SW |
1560 | prepare_read_connect(con); |
1561 | ||
1562 | /* Tell ceph about it. */ | |
ec302645 | 1563 | mutex_unlock(&con->mutex); |
31b8006e SW |
1564 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1565 | if (con->ops->peer_reset) | |
1566 | con->ops->peer_reset(con); | |
ec302645 | 1567 | mutex_lock(&con->mutex); |
0da5d703 SW |
1568 | if (test_bit(CLOSED, &con->state) || |
1569 | test_bit(OPENING, &con->state)) | |
1570 | return -EAGAIN; | |
31b8006e SW |
1571 | break; |
1572 | ||
1573 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1574 | /* | |
1575 | * If we sent a smaller connect_seq than the peer has, try | |
1576 | * again with a larger value. | |
1577 | */ | |
a16cb1f7 | 1578 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 1579 | le32_to_cpu(con->out_connect.connect_seq), |
a16cb1f7 SW |
1580 | le32_to_cpu(con->in_reply.connect_seq)); |
1581 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
5a0f8fdd AE |
1582 | ret = prepare_write_connect(con); |
1583 | if (ret < 0) | |
1584 | return ret; | |
31b8006e SW |
1585 | prepare_read_connect(con); |
1586 | break; | |
1587 | ||
1588 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1589 | /* | |
1590 | * If we sent a smaller global_seq than the peer has, try | |
1591 | * again with a larger value. | |
1592 | */ | |
eed0ef2c | 1593 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 1594 | con->peer_global_seq, |
a16cb1f7 | 1595 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 1596 | get_global_seq(con->msgr, |
a16cb1f7 | 1597 | le32_to_cpu(con->in_reply.global_seq)); |
5a0f8fdd AE |
1598 | ret = prepare_write_connect(con); |
1599 | if (ret < 0) | |
1600 | return ret; | |
31b8006e SW |
1601 | prepare_read_connect(con); |
1602 | break; | |
1603 | ||
1604 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1605 | if (req_feat & ~server_feat) { |
1606 | pr_err("%s%lld %s protocol feature mismatch," | |
1607 | " my required %llx > server's %llx, need %llx\n", | |
1608 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1609 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1610 | req_feat, server_feat, req_feat & ~server_feat); |
1611 | con->error_msg = "missing required protocol features"; | |
1612 | fail_protocol(con); | |
1613 | return -1; | |
1614 | } | |
3ec50d18 | 1615 | clear_bit(NEGOTIATING, &con->state); |
e27947c7 | 1616 | set_bit(CONNECTED, &con->state); |
31b8006e SW |
1617 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1618 | con->connect_seq++; | |
aba558e2 | 1619 | con->peer_features = server_feat; |
31b8006e SW |
1620 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1621 | con->peer_global_seq, | |
1622 | le32_to_cpu(con->in_reply.connect_seq), | |
1623 | con->connect_seq); | |
1624 | WARN_ON(con->connect_seq != | |
1625 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1626 | |
1627 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
928443cd | 1628 | set_bit(LOSSYTX, &con->flags); |
92ac41d0 | 1629 | |
31b8006e SW |
1630 | prepare_read_tag(con); |
1631 | break; | |
1632 | ||
1633 | case CEPH_MSGR_TAG_WAIT: | |
1634 | /* | |
1635 | * If there is a connection race (we are opening | |
1636 | * connections to each other), one of us may just have | |
1637 | * to WAIT. This shouldn't happen if we are the | |
1638 | * client. | |
1639 | */ | |
04177882 SW |
1640 | pr_err("process_connect got WAIT as client\n"); |
1641 | con->error_msg = "protocol error, got WAIT as client"; | |
1642 | return -1; | |
31b8006e SW |
1643 | |
1644 | default: | |
1645 | pr_err("connect protocol error, will retry\n"); | |
1646 | con->error_msg = "protocol error, garbage tag during connect"; | |
1647 | return -1; | |
1648 | } | |
1649 | return 0; | |
1650 | } | |
1651 | ||
1652 | ||
1653 | /* | |
1654 | * read (part of) an ack | |
1655 | */ | |
1656 | static int read_partial_ack(struct ceph_connection *con) | |
1657 | { | |
fd51653f AE |
1658 | int size = sizeof (con->in_temp_ack); |
1659 | int end = size; | |
1660 | ||
1661 | return read_partial(con, end, size, &con->in_temp_ack); | |
31b8006e SW |
1662 | } |
1663 | ||
1664 | ||
1665 | /* | |
1666 | * We can finally discard anything that's been acked. | |
1667 | */ | |
1668 | static void process_ack(struct ceph_connection *con) | |
1669 | { | |
1670 | struct ceph_msg *m; | |
1671 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1672 | u64 seq; | |
1673 | ||
31b8006e SW |
1674 | while (!list_empty(&con->out_sent)) { |
1675 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1676 | list_head); | |
1677 | seq = le64_to_cpu(m->hdr.seq); | |
1678 | if (seq > ack) | |
1679 | break; | |
1680 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1681 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1682 | m->ack_stamp = jiffies; |
31b8006e SW |
1683 | ceph_msg_remove(m); |
1684 | } | |
31b8006e SW |
1685 | prepare_read_tag(con); |
1686 | } | |
1687 | ||
1688 | ||
1689 | ||
1690 | ||
2450418c | 1691 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1692 | struct kvec *section, |
1693 | unsigned int sec_len, u32 *crc) | |
2450418c | 1694 | { |
68b4476b | 1695 | int ret, left; |
2450418c YS |
1696 | |
1697 | BUG_ON(!section); | |
1698 | ||
1699 | while (section->iov_len < sec_len) { | |
1700 | BUG_ON(section->iov_base == NULL); | |
1701 | left = sec_len - section->iov_len; | |
1702 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1703 | section->iov_len, left); | |
1704 | if (ret <= 0) | |
1705 | return ret; | |
1706 | section->iov_len += ret; | |
2450418c | 1707 | } |
fe3ad593 AE |
1708 | if (section->iov_len == sec_len) |
1709 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1710 | |
2450418c YS |
1711 | return 1; |
1712 | } | |
31b8006e | 1713 | |
1c20f2d2 AE |
1714 | static bool ceph_con_in_msg_alloc(struct ceph_connection *con, |
1715 | struct ceph_msg_header *hdr); | |
68b4476b YS |
1716 | |
1717 | ||
1718 | static int read_partial_message_pages(struct ceph_connection *con, | |
1719 | struct page **pages, | |
95c96174 | 1720 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1721 | { |
1722 | void *p; | |
1723 | int ret; | |
1724 | int left; | |
1725 | ||
1726 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1727 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1728 | /* (page) data */ | |
1729 | BUG_ON(pages == NULL); | |
1730 | p = kmap(pages[con->in_msg_pos.page]); | |
1731 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1732 | left); | |
bca064d2 | 1733 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1734 | con->in_data_crc = |
1735 | crc32c(con->in_data_crc, | |
1736 | p + con->in_msg_pos.page_pos, ret); | |
1737 | kunmap(pages[con->in_msg_pos.page]); | |
1738 | if (ret <= 0) | |
1739 | return ret; | |
1740 | con->in_msg_pos.data_pos += ret; | |
1741 | con->in_msg_pos.page_pos += ret; | |
1742 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1743 | con->in_msg_pos.page_pos = 0; | |
1744 | con->in_msg_pos.page++; | |
1745 | } | |
1746 | ||
1747 | return ret; | |
1748 | } | |
1749 | ||
1750 | #ifdef CONFIG_BLOCK | |
1751 | static int read_partial_message_bio(struct ceph_connection *con, | |
1752 | struct bio **bio_iter, int *bio_seg, | |
95c96174 | 1753 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1754 | { |
1755 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1756 | void *p; | |
1757 | int ret, left; | |
1758 | ||
68b4476b YS |
1759 | left = min((int)(data_len - con->in_msg_pos.data_pos), |
1760 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1761 | ||
1762 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1763 | ||
1764 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1765 | left); | |
bca064d2 | 1766 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1767 | con->in_data_crc = |
1768 | crc32c(con->in_data_crc, | |
1769 | p + con->in_msg_pos.page_pos, ret); | |
1770 | kunmap(bv->bv_page); | |
1771 | if (ret <= 0) | |
1772 | return ret; | |
1773 | con->in_msg_pos.data_pos += ret; | |
1774 | con->in_msg_pos.page_pos += ret; | |
1775 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1776 | con->in_msg_pos.page_pos = 0; | |
1777 | iter_bio_next(bio_iter, bio_seg); | |
1778 | } | |
1779 | ||
1780 | return ret; | |
1781 | } | |
1782 | #endif | |
1783 | ||
31b8006e SW |
1784 | /* |
1785 | * read (part of) a message. | |
1786 | */ | |
1787 | static int read_partial_message(struct ceph_connection *con) | |
1788 | { | |
1789 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1790 | int size; |
1791 | int end; | |
31b8006e | 1792 | int ret; |
95c96174 | 1793 | unsigned int front_len, middle_len, data_len; |
37675b0f | 1794 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 1795 | u64 seq; |
fe3ad593 | 1796 | u32 crc; |
31b8006e SW |
1797 | |
1798 | dout("read_partial_message con %p msg %p\n", con, m); | |
1799 | ||
1800 | /* header */ | |
fd51653f AE |
1801 | size = sizeof (con->in_hdr); |
1802 | end = size; | |
1803 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1804 | if (ret <= 0) |
1805 | return ret; | |
fe3ad593 AE |
1806 | |
1807 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1808 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1809 | pr_err("read_partial_message bad hdr " | |
1810 | " crc %u != expected %u\n", | |
1811 | crc, con->in_hdr.crc); | |
1812 | return -EBADMSG; | |
1813 | } | |
1814 | ||
31b8006e SW |
1815 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1816 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1817 | return -EIO; | |
1818 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1819 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1820 | return -EIO; | |
1821 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1822 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1823 | return -EIO; | |
1824 | ||
ae18756b SW |
1825 | /* verify seq# */ |
1826 | seq = le64_to_cpu(con->in_hdr.seq); | |
1827 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1828 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1829 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1830 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1831 | seq, con->in_seq + 1); |
1832 | con->in_base_pos = -front_len - middle_len - data_len - | |
1833 | sizeof(m->footer); | |
1834 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1835 | return 0; |
1836 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1837 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1838 | seq, con->in_seq + 1); | |
1839 | con->error_msg = "bad message sequence # for incoming message"; | |
1840 | return -EBADMSG; | |
1841 | } | |
1842 | ||
31b8006e SW |
1843 | /* allocate message? */ |
1844 | if (!con->in_msg) { | |
1845 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1846 | con->in_hdr.front_len, con->in_hdr.data_len); | |
1c20f2d2 | 1847 | if (ceph_con_in_msg_alloc(con, &con->in_hdr)) { |
31b8006e | 1848 | /* skip this message */ |
a79832f2 | 1849 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1850 | BUG_ON(con->in_msg); |
31b8006e SW |
1851 | con->in_base_pos = -front_len - middle_len - data_len - |
1852 | sizeof(m->footer); | |
1853 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1854 | con->in_seq++; |
31b8006e SW |
1855 | return 0; |
1856 | } | |
a79832f2 | 1857 | if (!con->in_msg) { |
5b3a4db3 SW |
1858 | con->error_msg = |
1859 | "error allocating memory for incoming message"; | |
a79832f2 | 1860 | return -ENOMEM; |
31b8006e | 1861 | } |
38941f80 AE |
1862 | |
1863 | BUG_ON(con->in_msg->con != con); | |
31b8006e SW |
1864 | m = con->in_msg; |
1865 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1866 | if (m->middle) |
1867 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1868 | |
1869 | con->in_msg_pos.page = 0; | |
68b4476b | 1870 | if (m->pages) |
c5c6b19d | 1871 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1872 | else |
1873 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1874 | con->in_msg_pos.data_pos = 0; |
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 | 1890 | } |
68b4476b YS |
1891 | #ifdef CONFIG_BLOCK |
1892 | if (m->bio && !m->bio_iter) | |
1893 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1894 | #endif | |
31b8006e SW |
1895 | |
1896 | /* (page) data */ | |
31b8006e | 1897 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1898 | if (m->pages) { |
1899 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1900 | data_len, do_datacrc); |
68b4476b YS |
1901 | if (ret <= 0) |
1902 | return ret; | |
1903 | #ifdef CONFIG_BLOCK | |
1904 | } else if (m->bio) { | |
1905 | ||
1906 | ret = read_partial_message_bio(con, | |
1907 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1908 | data_len, do_datacrc); |
68b4476b YS |
1909 | if (ret <= 0) |
1910 | return ret; | |
1911 | #endif | |
1912 | } else { | |
1913 | BUG_ON(1); | |
31b8006e SW |
1914 | } |
1915 | } | |
1916 | ||
31b8006e | 1917 | /* footer */ |
fd51653f AE |
1918 | size = sizeof (m->footer); |
1919 | end += size; | |
1920 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
1921 | if (ret <= 0) |
1922 | return ret; | |
1923 | ||
31b8006e SW |
1924 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
1925 | m, front_len, m->footer.front_crc, middle_len, | |
1926 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1927 | ||
1928 | /* crc ok? */ | |
1929 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1930 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1931 | m, con->in_front_crc, m->footer.front_crc); | |
1932 | return -EBADMSG; | |
1933 | } | |
1934 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1935 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1936 | m, con->in_middle_crc, m->footer.middle_crc); | |
1937 | return -EBADMSG; | |
1938 | } | |
bca064d2 | 1939 | if (do_datacrc && |
31b8006e SW |
1940 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
1941 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1942 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1943 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1944 | return -EBADMSG; | |
1945 | } | |
1946 | ||
1947 | return 1; /* done! */ | |
1948 | } | |
1949 | ||
1950 | /* | |
1951 | * Process message. This happens in the worker thread. The callback should | |
1952 | * be careful not to do anything that waits on other incoming messages or it | |
1953 | * may deadlock. | |
1954 | */ | |
1955 | static void process_message(struct ceph_connection *con) | |
1956 | { | |
5e095e8b | 1957 | struct ceph_msg *msg; |
31b8006e | 1958 | |
38941f80 AE |
1959 | BUG_ON(con->in_msg->con != con); |
1960 | con->in_msg->con = NULL; | |
5e095e8b | 1961 | msg = con->in_msg; |
31b8006e | 1962 | con->in_msg = NULL; |
36eb71aa | 1963 | con->ops->put(con); |
31b8006e SW |
1964 | |
1965 | /* if first message, set peer_name */ | |
1966 | if (con->peer_name.type == 0) | |
dbad185d | 1967 | con->peer_name = msg->hdr.src; |
31b8006e | 1968 | |
31b8006e | 1969 | con->in_seq++; |
ec302645 | 1970 | mutex_unlock(&con->mutex); |
31b8006e SW |
1971 | |
1972 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1973 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1974 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1975 | le16_to_cpu(msg->hdr.type), |
1976 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1977 | le32_to_cpu(msg->hdr.front_len), | |
1978 | le32_to_cpu(msg->hdr.data_len), | |
1979 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1980 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1981 | |
1982 | mutex_lock(&con->mutex); | |
31b8006e SW |
1983 | prepare_read_tag(con); |
1984 | } | |
1985 | ||
1986 | ||
1987 | /* | |
1988 | * Write something to the socket. Called in a worker thread when the | |
1989 | * socket appears to be writeable and we have something ready to send. | |
1990 | */ | |
1991 | static int try_write(struct ceph_connection *con) | |
1992 | { | |
31b8006e SW |
1993 | int ret = 1; |
1994 | ||
d59315ca | 1995 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 1996 | |
31b8006e SW |
1997 | more: |
1998 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
1999 | ||
2000 | /* open the socket first? */ | |
2001 | if (con->sock == NULL) { | |
a5988c49 AE |
2002 | set_bit(CONNECTING, &con->state); |
2003 | ||
e2200423 | 2004 | con_out_kvec_reset(con); |
e825a66d | 2005 | prepare_write_banner(con); |
eed0ef2c | 2006 | prepare_read_banner(con); |
31b8006e | 2007 | |
cf3e5c40 | 2008 | BUG_ON(con->in_msg); |
31b8006e SW |
2009 | con->in_tag = CEPH_MSGR_TAG_READY; |
2010 | dout("try_write initiating connect on %p new state %lu\n", | |
2011 | con, con->state); | |
41617d0c AE |
2012 | ret = ceph_tcp_connect(con); |
2013 | if (ret < 0) { | |
31b8006e | 2014 | con->error_msg = "connect error"; |
31b8006e SW |
2015 | goto out; |
2016 | } | |
2017 | } | |
2018 | ||
2019 | more_kvec: | |
2020 | /* kvec data queued? */ | |
2021 | if (con->out_skip) { | |
2022 | ret = write_partial_skip(con); | |
2023 | if (ret <= 0) | |
42961d23 | 2024 | goto out; |
31b8006e SW |
2025 | } |
2026 | if (con->out_kvec_left) { | |
2027 | ret = write_partial_kvec(con); | |
2028 | if (ret <= 0) | |
42961d23 | 2029 | goto out; |
31b8006e SW |
2030 | } |
2031 | ||
2032 | /* msg pages? */ | |
2033 | if (con->out_msg) { | |
c86a2930 SW |
2034 | if (con->out_msg_done) { |
2035 | ceph_msg_put(con->out_msg); | |
2036 | con->out_msg = NULL; /* we're done with this one */ | |
2037 | goto do_next; | |
2038 | } | |
2039 | ||
31b8006e SW |
2040 | ret = write_partial_msg_pages(con); |
2041 | if (ret == 1) | |
2042 | goto more_kvec; /* we need to send the footer, too! */ | |
2043 | if (ret == 0) | |
42961d23 | 2044 | goto out; |
31b8006e SW |
2045 | if (ret < 0) { |
2046 | dout("try_write write_partial_msg_pages err %d\n", | |
2047 | ret); | |
42961d23 | 2048 | goto out; |
31b8006e SW |
2049 | } |
2050 | } | |
2051 | ||
c86a2930 | 2052 | do_next: |
7593af92 AE |
2053 | if (!test_bit(CONNECTING, &con->state) && |
2054 | !test_bit(NEGOTIATING, &con->state)) { | |
31b8006e SW |
2055 | /* is anything else pending? */ |
2056 | if (!list_empty(&con->out_queue)) { | |
2057 | prepare_write_message(con); | |
2058 | goto more; | |
2059 | } | |
2060 | if (con->in_seq > con->in_seq_acked) { | |
2061 | prepare_write_ack(con); | |
2062 | goto more; | |
2063 | } | |
928443cd | 2064 | if (test_and_clear_bit(KEEPALIVE_PENDING, &con->flags)) { |
31b8006e SW |
2065 | prepare_write_keepalive(con); |
2066 | goto more; | |
2067 | } | |
2068 | } | |
2069 | ||
2070 | /* Nothing to do! */ | |
928443cd | 2071 | clear_bit(WRITE_PENDING, &con->flags); |
31b8006e | 2072 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2073 | ret = 0; |
2074 | out: | |
42961d23 | 2075 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2076 | return ret; |
2077 | } | |
2078 | ||
2079 | ||
2080 | ||
2081 | /* | |
2082 | * Read what we can from the socket. | |
2083 | */ | |
2084 | static int try_read(struct ceph_connection *con) | |
2085 | { | |
31b8006e SW |
2086 | int ret = -1; |
2087 | ||
2088 | if (!con->sock) | |
2089 | return 0; | |
2090 | ||
2091 | if (test_bit(STANDBY, &con->state)) | |
2092 | return 0; | |
2093 | ||
2094 | dout("try_read start on %p\n", con); | |
ec302645 | 2095 | |
31b8006e SW |
2096 | more: |
2097 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, | |
2098 | con->in_base_pos); | |
0da5d703 SW |
2099 | |
2100 | /* | |
2101 | * process_connect and process_message drop and re-take | |
2102 | * con->mutex. make sure we handle a racing close or reopen. | |
2103 | */ | |
2104 | if (test_bit(CLOSED, &con->state) || | |
2105 | test_bit(OPENING, &con->state)) { | |
2106 | ret = -EAGAIN; | |
2107 | goto out; | |
2108 | } | |
2109 | ||
31b8006e | 2110 | if (test_bit(CONNECTING, &con->state)) { |
7593af92 AE |
2111 | dout("try_read connecting\n"); |
2112 | ret = read_partial_banner(con); | |
2113 | if (ret <= 0) | |
ab166d5a | 2114 | goto out; |
7593af92 AE |
2115 | ret = process_banner(con); |
2116 | if (ret < 0) | |
2117 | goto out; | |
2118 | ||
2119 | clear_bit(CONNECTING, &con->state); | |
2120 | set_bit(NEGOTIATING, &con->state); | |
2121 | ||
2122 | /* Banner is good, exchange connection info */ | |
2123 | ret = prepare_write_connect(con); | |
2124 | if (ret < 0) | |
2125 | goto out; | |
2126 | prepare_read_connect(con); | |
2127 | ||
2128 | /* Send connection info before awaiting response */ | |
2129 | goto out; | |
2130 | } | |
2131 | ||
2132 | if (test_bit(NEGOTIATING, &con->state)) { | |
2133 | dout("try_read negotiating\n"); | |
31b8006e SW |
2134 | ret = read_partial_connect(con); |
2135 | if (ret <= 0) | |
31b8006e | 2136 | goto out; |
98bdb0aa SW |
2137 | ret = process_connect(con); |
2138 | if (ret < 0) | |
2139 | goto out; | |
31b8006e SW |
2140 | goto more; |
2141 | } | |
2142 | ||
2143 | if (con->in_base_pos < 0) { | |
2144 | /* | |
2145 | * skipping + discarding content. | |
2146 | * | |
2147 | * FIXME: there must be a better way to do this! | |
2148 | */ | |
84495f49 AE |
2149 | static char buf[SKIP_BUF_SIZE]; |
2150 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2151 | ||
31b8006e SW |
2152 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2153 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2154 | if (ret <= 0) | |
98bdb0aa | 2155 | goto out; |
31b8006e SW |
2156 | con->in_base_pos += ret; |
2157 | if (con->in_base_pos) | |
2158 | goto more; | |
2159 | } | |
2160 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2161 | /* | |
2162 | * what's next? | |
2163 | */ | |
2164 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2165 | if (ret <= 0) | |
98bdb0aa | 2166 | goto out; |
31b8006e SW |
2167 | dout("try_read got tag %d\n", (int)con->in_tag); |
2168 | switch (con->in_tag) { | |
2169 | case CEPH_MSGR_TAG_MSG: | |
2170 | prepare_read_message(con); | |
2171 | break; | |
2172 | case CEPH_MSGR_TAG_ACK: | |
2173 | prepare_read_ack(con); | |
2174 | break; | |
2175 | case CEPH_MSGR_TAG_CLOSE: | |
e27947c7 | 2176 | clear_bit(CONNECTED, &con->state); |
31b8006e | 2177 | set_bit(CLOSED, &con->state); /* fixme */ |
98bdb0aa | 2178 | goto out; |
31b8006e SW |
2179 | default: |
2180 | goto bad_tag; | |
2181 | } | |
2182 | } | |
2183 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2184 | ret = read_partial_message(con); | |
2185 | if (ret <= 0) { | |
2186 | switch (ret) { | |
2187 | case -EBADMSG: | |
2188 | con->error_msg = "bad crc"; | |
2189 | ret = -EIO; | |
98bdb0aa | 2190 | break; |
31b8006e SW |
2191 | case -EIO: |
2192 | con->error_msg = "io error"; | |
98bdb0aa | 2193 | break; |
31b8006e | 2194 | } |
98bdb0aa | 2195 | goto out; |
31b8006e SW |
2196 | } |
2197 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2198 | goto more; | |
2199 | process_message(con); | |
2200 | goto more; | |
2201 | } | |
2202 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2203 | ret = read_partial_ack(con); | |
2204 | if (ret <= 0) | |
98bdb0aa | 2205 | goto out; |
31b8006e SW |
2206 | process_ack(con); |
2207 | goto more; | |
2208 | } | |
2209 | ||
31b8006e | 2210 | out: |
98bdb0aa | 2211 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2212 | return ret; |
2213 | ||
2214 | bad_tag: | |
2215 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2216 | con->error_msg = "protocol error, garbage tag"; | |
2217 | ret = -1; | |
2218 | goto out; | |
2219 | } | |
2220 | ||
2221 | ||
2222 | /* | |
2223 | * Atomically queue work on a connection. Bump @con reference to | |
2224 | * avoid races with connection teardown. | |
31b8006e SW |
2225 | */ |
2226 | static void queue_con(struct ceph_connection *con) | |
2227 | { | |
31b8006e SW |
2228 | if (!con->ops->get(con)) { |
2229 | dout("queue_con %p ref count 0\n", con); | |
2230 | return; | |
2231 | } | |
2232 | ||
f363e45f | 2233 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2234 | dout("queue_con %p - already queued\n", con); |
2235 | con->ops->put(con); | |
2236 | } else { | |
2237 | dout("queue_con %p\n", con); | |
2238 | } | |
2239 | } | |
2240 | ||
2241 | /* | |
2242 | * Do some work on a connection. Drop a connection ref when we're done. | |
2243 | */ | |
2244 | static void con_work(struct work_struct *work) | |
2245 | { | |
2246 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2247 | work.work); | |
0da5d703 | 2248 | int ret; |
31b8006e | 2249 | |
9dd4658d | 2250 | mutex_lock(&con->mutex); |
0da5d703 | 2251 | restart: |
188048bc | 2252 | if (test_and_clear_bit(SOCK_CLOSED, &con->flags)) { |
e27947c7 AE |
2253 | if (test_and_clear_bit(CONNECTED, &con->state)) |
2254 | con->error_msg = "socket closed"; | |
7593af92 AE |
2255 | else if (test_and_clear_bit(NEGOTIATING, &con->state)) |
2256 | con->error_msg = "negotiation failed"; | |
2257 | else if (test_and_clear_bit(CONNECTING, &con->state)) | |
188048bc | 2258 | con->error_msg = "connection failed"; |
7593af92 | 2259 | else |
e27947c7 | 2260 | con->error_msg = "unrecognized con state"; |
188048bc AE |
2261 | goto fault; |
2262 | } | |
2263 | ||
928443cd | 2264 | if (test_and_clear_bit(BACKOFF, &con->flags)) { |
60bf8bf8 SW |
2265 | dout("con_work %p backing off\n", con); |
2266 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2267 | round_jiffies_relative(con->delay))) { | |
2268 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2269 | mutex_unlock(&con->mutex); | |
2270 | return; | |
2271 | } else { | |
2272 | con->ops->put(con); | |
2273 | dout("con_work %p FAILED to back off %lu\n", con, | |
2274 | con->delay); | |
2275 | } | |
2276 | } | |
9dd4658d | 2277 | |
e00de341 SW |
2278 | if (test_bit(STANDBY, &con->state)) { |
2279 | dout("con_work %p STANDBY\n", con); | |
2280 | goto done; | |
2281 | } | |
31b8006e SW |
2282 | if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */ |
2283 | dout("con_work CLOSED\n"); | |
2284 | con_close_socket(con); | |
2285 | goto done; | |
2286 | } | |
2287 | if (test_and_clear_bit(OPENING, &con->state)) { | |
2288 | /* reopen w/ new peer */ | |
2289 | dout("con_work OPENING\n"); | |
2290 | con_close_socket(con); | |
2291 | } | |
2292 | ||
0da5d703 SW |
2293 | ret = try_read(con); |
2294 | if (ret == -EAGAIN) | |
2295 | goto restart; | |
3a140a0d SW |
2296 | if (ret < 0) { |
2297 | con->error_msg = "socket error on read"; | |
0da5d703 | 2298 | goto fault; |
3a140a0d | 2299 | } |
0da5d703 SW |
2300 | |
2301 | ret = try_write(con); | |
2302 | if (ret == -EAGAIN) | |
2303 | goto restart; | |
3a140a0d SW |
2304 | if (ret < 0) { |
2305 | con->error_msg = "socket error on write"; | |
0da5d703 | 2306 | goto fault; |
3a140a0d | 2307 | } |
31b8006e SW |
2308 | |
2309 | done: | |
9dd4658d | 2310 | mutex_unlock(&con->mutex); |
9dd4658d | 2311 | done_unlocked: |
31b8006e | 2312 | con->ops->put(con); |
0da5d703 SW |
2313 | return; |
2314 | ||
2315 | fault: | |
2316 | mutex_unlock(&con->mutex); | |
2317 | ceph_fault(con); /* error/fault path */ | |
2318 | goto done_unlocked; | |
31b8006e SW |
2319 | } |
2320 | ||
2321 | ||
2322 | /* | |
2323 | * Generic error/fault handler. A retry mechanism is used with | |
2324 | * exponential backoff | |
2325 | */ | |
2326 | static void ceph_fault(struct ceph_connection *con) | |
2327 | { | |
2328 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2329 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2330 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2331 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2332 | |
928443cd | 2333 | if (test_bit(LOSSYTX, &con->flags)) { |
31b8006e SW |
2334 | dout("fault on LOSSYTX channel\n"); |
2335 | goto out; | |
2336 | } | |
2337 | ||
ec302645 | 2338 | mutex_lock(&con->mutex); |
91e45ce3 SW |
2339 | if (test_bit(CLOSED, &con->state)) |
2340 | goto out_unlock; | |
ec302645 | 2341 | |
31b8006e | 2342 | con_close_socket(con); |
5e095e8b SW |
2343 | |
2344 | if (con->in_msg) { | |
38941f80 AE |
2345 | BUG_ON(con->in_msg->con != con); |
2346 | con->in_msg->con = NULL; | |
5e095e8b SW |
2347 | ceph_msg_put(con->in_msg); |
2348 | con->in_msg = NULL; | |
36eb71aa | 2349 | con->ops->put(con); |
5e095e8b | 2350 | } |
31b8006e | 2351 | |
e80a52d1 SW |
2352 | /* Requeue anything that hasn't been acked */ |
2353 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2354 | |
e76661d0 SW |
2355 | /* If there are no messages queued or keepalive pending, place |
2356 | * the connection in a STANDBY state */ | |
2357 | if (list_empty(&con->out_queue) && | |
928443cd | 2358 | !test_bit(KEEPALIVE_PENDING, &con->flags)) { |
e00de341 | 2359 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
928443cd | 2360 | clear_bit(WRITE_PENDING, &con->flags); |
31b8006e | 2361 | set_bit(STANDBY, &con->state); |
e80a52d1 SW |
2362 | } else { |
2363 | /* retry after a delay. */ | |
2364 | if (con->delay == 0) | |
2365 | con->delay = BASE_DELAY_INTERVAL; | |
2366 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2367 | con->delay *= 2; | |
e80a52d1 SW |
2368 | con->ops->get(con); |
2369 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2370 | round_jiffies_relative(con->delay))) { |
2371 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2372 | } else { | |
e80a52d1 | 2373 | con->ops->put(con); |
60bf8bf8 SW |
2374 | dout("fault failed to queue %p delay %lu, backoff\n", |
2375 | con, con->delay); | |
2376 | /* | |
2377 | * In many cases we see a socket state change | |
2378 | * while con_work is running and end up | |
2379 | * queuing (non-delayed) work, such that we | |
2380 | * can't backoff with a delay. Set a flag so | |
2381 | * that when con_work restarts we schedule the | |
2382 | * delay then. | |
2383 | */ | |
928443cd | 2384 | set_bit(BACKOFF, &con->flags); |
60bf8bf8 | 2385 | } |
31b8006e SW |
2386 | } |
2387 | ||
91e45ce3 SW |
2388 | out_unlock: |
2389 | mutex_unlock(&con->mutex); | |
31b8006e | 2390 | out: |
161fd65a SW |
2391 | /* |
2392 | * in case we faulted due to authentication, invalidate our | |
2393 | * current tickets so that we can get new ones. | |
213c99ee | 2394 | */ |
161fd65a SW |
2395 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2396 | dout("calling invalidate_authorizer()\n"); | |
2397 | con->ops->invalidate_authorizer(con); | |
2398 | } | |
2399 | ||
31b8006e SW |
2400 | if (con->ops->fault) |
2401 | con->ops->fault(con); | |
2402 | } | |
2403 | ||
2404 | ||
2405 | ||
2406 | /* | |
15d9882c | 2407 | * initialize a new messenger instance |
31b8006e | 2408 | */ |
15d9882c AE |
2409 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2410 | struct ceph_entity_addr *myaddr, | |
2411 | u32 supported_features, | |
2412 | u32 required_features, | |
2413 | bool nocrc) | |
31b8006e | 2414 | { |
3d14c5d2 YS |
2415 | msgr->supported_features = supported_features; |
2416 | msgr->required_features = required_features; | |
2417 | ||
31b8006e SW |
2418 | spin_lock_init(&msgr->global_seq_lock); |
2419 | ||
31b8006e SW |
2420 | if (myaddr) |
2421 | msgr->inst.addr = *myaddr; | |
2422 | ||
2423 | /* select a random nonce */ | |
ac8839d7 | 2424 | msgr->inst.addr.type = 0; |
103e2d3a | 2425 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2426 | encode_my_addr(msgr); |
15d9882c | 2427 | msgr->nocrc = nocrc; |
31b8006e | 2428 | |
a2a32584 GH |
2429 | atomic_set(&msgr->stopping, 0); |
2430 | ||
15d9882c | 2431 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2432 | } |
15d9882c | 2433 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2434 | |
e00de341 SW |
2435 | static void clear_standby(struct ceph_connection *con) |
2436 | { | |
2437 | /* come back from STANDBY? */ | |
2438 | if (test_and_clear_bit(STANDBY, &con->state)) { | |
2439 | mutex_lock(&con->mutex); | |
2440 | dout("clear_standby %p and ++connect_seq\n", con); | |
2441 | con->connect_seq++; | |
928443cd AE |
2442 | WARN_ON(test_bit(WRITE_PENDING, &con->flags)); |
2443 | WARN_ON(test_bit(KEEPALIVE_PENDING, &con->flags)); | |
e00de341 SW |
2444 | mutex_unlock(&con->mutex); |
2445 | } | |
2446 | } | |
2447 | ||
31b8006e SW |
2448 | /* |
2449 | * Queue up an outgoing message on the given connection. | |
2450 | */ | |
2451 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2452 | { | |
2453 | if (test_bit(CLOSED, &con->state)) { | |
2454 | dout("con_send %p closed, dropping %p\n", con, msg); | |
2455 | ceph_msg_put(msg); | |
2456 | return; | |
2457 | } | |
2458 | ||
2459 | /* set src+dst */ | |
dbad185d | 2460 | msg->hdr.src = con->msgr->inst.name; |
31b8006e | 2461 | |
3ca02ef9 SW |
2462 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
2463 | ||
e84346b7 SW |
2464 | msg->needs_out_seq = true; |
2465 | ||
31b8006e | 2466 | /* queue */ |
ec302645 | 2467 | mutex_lock(&con->mutex); |
92ce034b | 2468 | |
38941f80 | 2469 | BUG_ON(msg->con != NULL); |
36eb71aa | 2470 | msg->con = con->ops->get(con); |
92ce034b AE |
2471 | BUG_ON(msg->con == NULL); |
2472 | ||
31b8006e SW |
2473 | BUG_ON(!list_empty(&msg->list_head)); |
2474 | list_add_tail(&msg->list_head, &con->out_queue); | |
2475 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2476 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2477 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2478 | le32_to_cpu(msg->hdr.front_len), | |
2479 | le32_to_cpu(msg->hdr.middle_len), | |
2480 | le32_to_cpu(msg->hdr.data_len)); | |
ec302645 | 2481 | mutex_unlock(&con->mutex); |
31b8006e SW |
2482 | |
2483 | /* if there wasn't anything waiting to send before, queue | |
2484 | * new work */ | |
e00de341 | 2485 | clear_standby(con); |
928443cd | 2486 | if (test_and_set_bit(WRITE_PENDING, &con->flags) == 0) |
31b8006e SW |
2487 | queue_con(con); |
2488 | } | |
3d14c5d2 | 2489 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2490 | |
2491 | /* | |
2492 | * Revoke a message that was previously queued for send | |
2493 | */ | |
6740a845 | 2494 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2495 | { |
6740a845 AE |
2496 | struct ceph_connection *con = msg->con; |
2497 | ||
2498 | if (!con) | |
2499 | return; /* Message not in our possession */ | |
2500 | ||
ec302645 | 2501 | mutex_lock(&con->mutex); |
31b8006e | 2502 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2503 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2504 | list_del_init(&msg->list_head); |
38941f80 | 2505 | BUG_ON(msg->con == NULL); |
36eb71aa | 2506 | msg->con->ops->put(msg->con); |
38941f80 | 2507 | msg->con = NULL; |
92ce034b | 2508 | msg->hdr.seq = 0; |
38941f80 | 2509 | |
31b8006e | 2510 | ceph_msg_put(msg); |
ed98adad SW |
2511 | } |
2512 | if (con->out_msg == msg) { | |
38941f80 | 2513 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2514 | con->out_msg = NULL; |
31b8006e SW |
2515 | if (con->out_kvec_is_msg) { |
2516 | con->out_skip = con->out_kvec_bytes; | |
2517 | con->out_kvec_is_msg = false; | |
2518 | } | |
ed98adad | 2519 | msg->hdr.seq = 0; |
92ce034b AE |
2520 | |
2521 | ceph_msg_put(msg); | |
31b8006e | 2522 | } |
ec302645 | 2523 | mutex_unlock(&con->mutex); |
31b8006e SW |
2524 | } |
2525 | ||
350b1c32 | 2526 | /* |
0d59ab81 | 2527 | * Revoke a message that we may be reading data into |
350b1c32 | 2528 | */ |
8921d114 | 2529 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2530 | { |
8921d114 AE |
2531 | struct ceph_connection *con; |
2532 | ||
2533 | BUG_ON(msg == NULL); | |
2534 | if (!msg->con) { | |
2535 | dout("%s msg %p null con\n", __func__, msg); | |
2536 | ||
2537 | return; /* Message not in our possession */ | |
2538 | } | |
2539 | ||
2540 | con = msg->con; | |
350b1c32 | 2541 | mutex_lock(&con->mutex); |
8921d114 | 2542 | if (con->in_msg == msg) { |
95c96174 ED |
2543 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
2544 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
2545 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
2546 | |
2547 | /* skip rest of message */ | |
8921d114 AE |
2548 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
2549 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 2550 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
2551 | front_len - |
2552 | middle_len - | |
2553 | data_len - | |
350b1c32 | 2554 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2555 | ceph_msg_put(con->in_msg); |
2556 | con->in_msg = NULL; | |
2557 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2558 | con->in_seq++; |
350b1c32 | 2559 | } else { |
8921d114 AE |
2560 | dout("%s %p in_msg %p msg %p no-op\n", |
2561 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
2562 | } |
2563 | mutex_unlock(&con->mutex); | |
2564 | } | |
2565 | ||
31b8006e SW |
2566 | /* |
2567 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2568 | */ | |
2569 | void ceph_con_keepalive(struct ceph_connection *con) | |
2570 | { | |
e00de341 SW |
2571 | dout("con_keepalive %p\n", con); |
2572 | clear_standby(con); | |
928443cd AE |
2573 | if (test_and_set_bit(KEEPALIVE_PENDING, &con->flags) == 0 && |
2574 | test_and_set_bit(WRITE_PENDING, &con->flags) == 0) | |
31b8006e SW |
2575 | queue_con(con); |
2576 | } | |
3d14c5d2 | 2577 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2578 | |
2579 | ||
2580 | /* | |
2581 | * construct a new message with given type, size | |
2582 | * the new msg has a ref count of 1. | |
2583 | */ | |
b61c2763 SW |
2584 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2585 | bool can_fail) | |
31b8006e SW |
2586 | { |
2587 | struct ceph_msg *m; | |
2588 | ||
34d23762 | 2589 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2590 | if (m == NULL) |
2591 | goto out; | |
c2e552e7 | 2592 | kref_init(&m->kref); |
38941f80 AE |
2593 | |
2594 | m->con = NULL; | |
31b8006e SW |
2595 | INIT_LIST_HEAD(&m->list_head); |
2596 | ||
45c6ceb5 | 2597 | m->hdr.tid = 0; |
31b8006e | 2598 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2599 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2600 | m->hdr.version = 0; | |
31b8006e SW |
2601 | m->hdr.front_len = cpu_to_le32(front_len); |
2602 | m->hdr.middle_len = 0; | |
bb257664 SW |
2603 | m->hdr.data_len = 0; |
2604 | m->hdr.data_off = 0; | |
45c6ceb5 | 2605 | m->hdr.reserved = 0; |
31b8006e SW |
2606 | m->footer.front_crc = 0; |
2607 | m->footer.middle_crc = 0; | |
2608 | m->footer.data_crc = 0; | |
45c6ceb5 | 2609 | m->footer.flags = 0; |
31b8006e SW |
2610 | m->front_max = front_len; |
2611 | m->front_is_vmalloc = false; | |
2612 | m->more_to_follow = false; | |
c0d5f9db | 2613 | m->ack_stamp = 0; |
31b8006e SW |
2614 | m->pool = NULL; |
2615 | ||
ca20892d HC |
2616 | /* middle */ |
2617 | m->middle = NULL; | |
2618 | ||
2619 | /* data */ | |
2620 | m->nr_pages = 0; | |
2621 | m->page_alignment = 0; | |
2622 | m->pages = NULL; | |
2623 | m->pagelist = NULL; | |
2624 | m->bio = NULL; | |
2625 | m->bio_iter = NULL; | |
2626 | m->bio_seg = 0; | |
2627 | m->trail = NULL; | |
2628 | ||
31b8006e SW |
2629 | /* front */ |
2630 | if (front_len) { | |
2631 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2632 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2633 | PAGE_KERNEL); |
2634 | m->front_is_vmalloc = true; | |
2635 | } else { | |
34d23762 | 2636 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2637 | } |
2638 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2639 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2640 | front_len); |
2641 | goto out2; | |
2642 | } | |
2643 | } else { | |
2644 | m->front.iov_base = NULL; | |
2645 | } | |
2646 | m->front.iov_len = front_len; | |
2647 | ||
bb257664 | 2648 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2649 | return m; |
2650 | ||
2651 | out2: | |
2652 | ceph_msg_put(m); | |
2653 | out: | |
b61c2763 SW |
2654 | if (!can_fail) { |
2655 | pr_err("msg_new can't create type %d front %d\n", type, | |
2656 | front_len); | |
f0ed1b7c | 2657 | WARN_ON(1); |
b61c2763 SW |
2658 | } else { |
2659 | dout("msg_new can't create type %d front %d\n", type, | |
2660 | front_len); | |
2661 | } | |
a79832f2 | 2662 | return NULL; |
31b8006e | 2663 | } |
3d14c5d2 | 2664 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2665 | |
31b8006e SW |
2666 | /* |
2667 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2668 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2669 | * per-type header in the front and then gracefully fail (i.e., | |
2670 | * propagate the error to the caller based on info in the front) when | |
2671 | * the middle is too large. | |
2672 | */ | |
2450418c | 2673 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2674 | { |
2675 | int type = le16_to_cpu(msg->hdr.type); | |
2676 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2677 | ||
2678 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2679 | ceph_msg_type_name(type), middle_len); | |
2680 | BUG_ON(!middle_len); | |
2681 | BUG_ON(msg->middle); | |
2682 | ||
b6c1d5b8 | 2683 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2684 | if (!msg->middle) |
2685 | return -ENOMEM; | |
2686 | return 0; | |
2687 | } | |
2688 | ||
2450418c | 2689 | /* |
1c20f2d2 AE |
2690 | * Allocate a message for receiving an incoming message on a |
2691 | * connection, and save the result in con->in_msg. Uses the | |
2692 | * connection's private alloc_msg op if available. | |
2693 | * | |
2694 | * Returns true if the message should be skipped, false otherwise. | |
2695 | * If true is returned (skip message), con->in_msg will be NULL. | |
2696 | * If false is returned, con->in_msg will contain a pointer to the | |
2697 | * newly-allocated message, or NULL in case of memory exhaustion. | |
2450418c | 2698 | */ |
1c20f2d2 AE |
2699 | static bool ceph_con_in_msg_alloc(struct ceph_connection *con, |
2700 | struct ceph_msg_header *hdr) | |
2450418c YS |
2701 | { |
2702 | int type = le16_to_cpu(hdr->type); | |
2703 | int front_len = le32_to_cpu(hdr->front_len); | |
2704 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2450418c YS |
2705 | int ret; |
2706 | ||
1c20f2d2 AE |
2707 | BUG_ON(con->in_msg != NULL); |
2708 | ||
2450418c | 2709 | if (con->ops->alloc_msg) { |
1c20f2d2 AE |
2710 | int skip = 0; |
2711 | ||
0547a9b3 | 2712 | mutex_unlock(&con->mutex); |
1c20f2d2 | 2713 | con->in_msg = con->ops->alloc_msg(con, hdr, &skip); |
0547a9b3 | 2714 | mutex_lock(&con->mutex); |
92ce034b | 2715 | if (con->in_msg) { |
36eb71aa | 2716 | con->in_msg->con = con->ops->get(con); |
92ce034b AE |
2717 | BUG_ON(con->in_msg->con == NULL); |
2718 | } | |
1c20f2d2 AE |
2719 | if (skip) |
2720 | con->in_msg = NULL; | |
2721 | ||
2722 | if (!con->in_msg) | |
2723 | return skip != 0; | |
2450418c | 2724 | } |
1c20f2d2 AE |
2725 | if (!con->in_msg) { |
2726 | con->in_msg = ceph_msg_new(type, front_len, GFP_NOFS, false); | |
2727 | if (!con->in_msg) { | |
2450418c YS |
2728 | pr_err("unable to allocate msg type %d len %d\n", |
2729 | type, front_len); | |
1c20f2d2 | 2730 | return false; |
2450418c | 2731 | } |
36eb71aa | 2732 | con->in_msg->con = con->ops->get(con); |
92ce034b | 2733 | BUG_ON(con->in_msg->con == NULL); |
1c20f2d2 | 2734 | con->in_msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2735 | } |
1c20f2d2 | 2736 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2737 | |
1c20f2d2 AE |
2738 | if (middle_len && !con->in_msg->middle) { |
2739 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 2740 | if (ret < 0) { |
1c20f2d2 AE |
2741 | ceph_msg_put(con->in_msg); |
2742 | con->in_msg = NULL; | |
2450418c YS |
2743 | } |
2744 | } | |
9d7f0f13 | 2745 | |
1c20f2d2 | 2746 | return false; |
2450418c YS |
2747 | } |
2748 | ||
31b8006e SW |
2749 | |
2750 | /* | |
2751 | * Free a generically kmalloc'd message. | |
2752 | */ | |
2753 | void ceph_msg_kfree(struct ceph_msg *m) | |
2754 | { | |
2755 | dout("msg_kfree %p\n", m); | |
2756 | if (m->front_is_vmalloc) | |
2757 | vfree(m->front.iov_base); | |
2758 | else | |
2759 | kfree(m->front.iov_base); | |
2760 | kfree(m); | |
2761 | } | |
2762 | ||
2763 | /* | |
2764 | * Drop a msg ref. Destroy as needed. | |
2765 | */ | |
c2e552e7 SW |
2766 | void ceph_msg_last_put(struct kref *kref) |
2767 | { | |
2768 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2769 | |
c2e552e7 SW |
2770 | dout("ceph_msg_put last one on %p\n", m); |
2771 | WARN_ON(!list_empty(&m->list_head)); | |
2772 | ||
2773 | /* drop middle, data, if any */ | |
2774 | if (m->middle) { | |
2775 | ceph_buffer_put(m->middle); | |
2776 | m->middle = NULL; | |
31b8006e | 2777 | } |
c2e552e7 SW |
2778 | m->nr_pages = 0; |
2779 | m->pages = NULL; | |
2780 | ||
58bb3b37 SW |
2781 | if (m->pagelist) { |
2782 | ceph_pagelist_release(m->pagelist); | |
2783 | kfree(m->pagelist); | |
2784 | m->pagelist = NULL; | |
2785 | } | |
2786 | ||
68b4476b YS |
2787 | m->trail = NULL; |
2788 | ||
c2e552e7 SW |
2789 | if (m->pool) |
2790 | ceph_msgpool_put(m->pool, m); | |
2791 | else | |
2792 | ceph_msg_kfree(m); | |
31b8006e | 2793 | } |
3d14c5d2 | 2794 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2795 | |
2796 | void ceph_msg_dump(struct ceph_msg *msg) | |
2797 | { | |
2798 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2799 | msg->front_max, msg->nr_pages); | |
2800 | print_hex_dump(KERN_DEBUG, "header: ", | |
2801 | DUMP_PREFIX_OFFSET, 16, 1, | |
2802 | &msg->hdr, sizeof(msg->hdr), true); | |
2803 | print_hex_dump(KERN_DEBUG, " front: ", | |
2804 | DUMP_PREFIX_OFFSET, 16, 1, | |
2805 | msg->front.iov_base, msg->front.iov_len, true); | |
2806 | if (msg->middle) | |
2807 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2808 | DUMP_PREFIX_OFFSET, 16, 1, | |
2809 | msg->middle->vec.iov_base, | |
2810 | msg->middle->vec.iov_len, true); | |
2811 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2812 | DUMP_PREFIX_OFFSET, 16, 1, | |
2813 | &msg->footer, sizeof(msg->footer), true); | |
2814 | } | |
3d14c5d2 | 2815 | EXPORT_SYMBOL(ceph_msg_dump); |