]>
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> | |
31b8006e SW |
14 | #include <net/tcp.h> |
15 | ||
3d14c5d2 YS |
16 | #include <linux/ceph/libceph.h> |
17 | #include <linux/ceph/messenger.h> | |
18 | #include <linux/ceph/decode.h> | |
19 | #include <linux/ceph/pagelist.h> | |
31b8006e SW |
20 | |
21 | /* | |
22 | * Ceph uses the messenger to exchange ceph_msg messages with other | |
23 | * hosts in the system. The messenger provides ordered and reliable | |
24 | * delivery. We tolerate TCP disconnects by reconnecting (with | |
25 | * exponential backoff) in the case of a fault (disconnection, bad | |
26 | * crc, protocol error). Acks allow sent messages to be discarded by | |
27 | * the sender. | |
28 | */ | |
29 | ||
30 | /* static tag bytes (protocol control messages) */ | |
31 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
32 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
33 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
34 | ||
a6a5349d SW |
35 | #ifdef CONFIG_LOCKDEP |
36 | static struct lock_class_key socket_class; | |
37 | #endif | |
38 | ||
31b8006e SW |
39 | |
40 | static void queue_con(struct ceph_connection *con); | |
41 | static void con_work(struct work_struct *); | |
42 | static void ceph_fault(struct ceph_connection *con); | |
43 | ||
31b8006e SW |
44 | /* |
45 | * nicely render a sockaddr as a string. | |
46 | */ | |
47 | #define MAX_ADDR_STR 20 | |
d06dbaf6 SW |
48 | #define MAX_ADDR_STR_LEN 60 |
49 | static char addr_str[MAX_ADDR_STR][MAX_ADDR_STR_LEN]; | |
31b8006e SW |
50 | static DEFINE_SPINLOCK(addr_str_lock); |
51 | static int last_addr_str; | |
52 | ||
3d14c5d2 | 53 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
54 | { |
55 | int i; | |
56 | char *s; | |
57 | struct sockaddr_in *in4 = (void *)ss; | |
31b8006e SW |
58 | struct sockaddr_in6 *in6 = (void *)ss; |
59 | ||
60 | spin_lock(&addr_str_lock); | |
61 | i = last_addr_str++; | |
62 | if (last_addr_str == MAX_ADDR_STR) | |
63 | last_addr_str = 0; | |
64 | spin_unlock(&addr_str_lock); | |
65 | s = addr_str[i]; | |
66 | ||
67 | switch (ss->ss_family) { | |
68 | case AF_INET: | |
d06dbaf6 SW |
69 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%u", &in4->sin_addr, |
70 | (unsigned int)ntohs(in4->sin_port)); | |
31b8006e SW |
71 | break; |
72 | ||
73 | case AF_INET6: | |
d06dbaf6 SW |
74 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%u", &in6->sin6_addr, |
75 | (unsigned int)ntohs(in6->sin6_port)); | |
31b8006e SW |
76 | break; |
77 | ||
78 | default: | |
12a2f643 SW |
79 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %d)", |
80 | (int)ss->ss_family); | |
31b8006e SW |
81 | } |
82 | ||
83 | return s; | |
84 | } | |
3d14c5d2 | 85 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 86 | |
63f2d211 SW |
87 | static void encode_my_addr(struct ceph_messenger *msgr) |
88 | { | |
89 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
90 | ceph_encode_addr(&msgr->my_enc_addr); | |
91 | } | |
92 | ||
31b8006e SW |
93 | /* |
94 | * work queue for all reading and writing to/from the socket. | |
95 | */ | |
96 | struct workqueue_struct *ceph_msgr_wq; | |
97 | ||
3d14c5d2 | 98 | int ceph_msgr_init(void) |
31b8006e | 99 | { |
f363e45f | 100 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0); |
d96c9043 SW |
101 | if (!ceph_msgr_wq) { |
102 | pr_err("msgr_init failed to create workqueue\n"); | |
103 | return -ENOMEM; | |
31b8006e SW |
104 | } |
105 | return 0; | |
106 | } | |
3d14c5d2 | 107 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
108 | |
109 | void ceph_msgr_exit(void) | |
110 | { | |
111 | destroy_workqueue(ceph_msgr_wq); | |
112 | } | |
3d14c5d2 | 113 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 114 | |
cd84db6e | 115 | void ceph_msgr_flush(void) |
a922d38f SW |
116 | { |
117 | flush_workqueue(ceph_msgr_wq); | |
118 | } | |
3d14c5d2 | 119 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f SW |
120 | |
121 | ||
31b8006e SW |
122 | /* |
123 | * socket callback functions | |
124 | */ | |
125 | ||
126 | /* data available on socket, or listen socket received a connect */ | |
127 | static void ceph_data_ready(struct sock *sk, int count_unused) | |
128 | { | |
129 | struct ceph_connection *con = | |
130 | (struct ceph_connection *)sk->sk_user_data; | |
131 | if (sk->sk_state != TCP_CLOSE_WAIT) { | |
132 | dout("ceph_data_ready on %p state = %lu, queueing work\n", | |
133 | con, con->state); | |
134 | queue_con(con); | |
135 | } | |
136 | } | |
137 | ||
138 | /* socket has buffer space for writing */ | |
139 | static void ceph_write_space(struct sock *sk) | |
140 | { | |
141 | struct ceph_connection *con = | |
142 | (struct ceph_connection *)sk->sk_user_data; | |
143 | ||
144 | /* only queue to workqueue if there is data we want to write. */ | |
145 | if (test_bit(WRITE_PENDING, &con->state)) { | |
146 | dout("ceph_write_space %p queueing write work\n", con); | |
147 | queue_con(con); | |
148 | } else { | |
149 | dout("ceph_write_space %p nothing to write\n", con); | |
150 | } | |
151 | ||
152 | /* since we have our own write_space, clear the SOCK_NOSPACE flag */ | |
153 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
154 | } | |
155 | ||
156 | /* socket's state has changed */ | |
157 | static void ceph_state_change(struct sock *sk) | |
158 | { | |
159 | struct ceph_connection *con = | |
160 | (struct ceph_connection *)sk->sk_user_data; | |
161 | ||
162 | dout("ceph_state_change %p state = %lu sk_state = %u\n", | |
163 | con, con->state, sk->sk_state); | |
164 | ||
165 | if (test_bit(CLOSED, &con->state)) | |
166 | return; | |
167 | ||
168 | switch (sk->sk_state) { | |
169 | case TCP_CLOSE: | |
170 | dout("ceph_state_change TCP_CLOSE\n"); | |
171 | case TCP_CLOSE_WAIT: | |
172 | dout("ceph_state_change TCP_CLOSE_WAIT\n"); | |
173 | if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) { | |
174 | if (test_bit(CONNECTING, &con->state)) | |
175 | con->error_msg = "connection failed"; | |
176 | else | |
177 | con->error_msg = "socket closed"; | |
178 | queue_con(con); | |
179 | } | |
180 | break; | |
181 | case TCP_ESTABLISHED: | |
182 | dout("ceph_state_change TCP_ESTABLISHED\n"); | |
183 | queue_con(con); | |
184 | break; | |
185 | } | |
186 | } | |
187 | ||
188 | /* | |
189 | * set up socket callbacks | |
190 | */ | |
191 | static void set_sock_callbacks(struct socket *sock, | |
192 | struct ceph_connection *con) | |
193 | { | |
194 | struct sock *sk = sock->sk; | |
195 | sk->sk_user_data = (void *)con; | |
196 | sk->sk_data_ready = ceph_data_ready; | |
197 | sk->sk_write_space = ceph_write_space; | |
198 | sk->sk_state_change = ceph_state_change; | |
199 | } | |
200 | ||
201 | ||
202 | /* | |
203 | * socket helpers | |
204 | */ | |
205 | ||
206 | /* | |
207 | * initiate connection to a remote socket. | |
208 | */ | |
209 | static struct socket *ceph_tcp_connect(struct ceph_connection *con) | |
210 | { | |
f91d3471 | 211 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
212 | struct socket *sock; |
213 | int ret; | |
214 | ||
215 | BUG_ON(con->sock); | |
f91d3471 SW |
216 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
217 | IPPROTO_TCP, &sock); | |
31b8006e SW |
218 | if (ret) |
219 | return ERR_PTR(ret); | |
220 | con->sock = sock; | |
221 | sock->sk->sk_allocation = GFP_NOFS; | |
222 | ||
a6a5349d SW |
223 | #ifdef CONFIG_LOCKDEP |
224 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
225 | #endif | |
226 | ||
31b8006e SW |
227 | set_sock_callbacks(sock, con); |
228 | ||
3d14c5d2 | 229 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 230 | |
f91d3471 SW |
231 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
232 | O_NONBLOCK); | |
31b8006e SW |
233 | if (ret == -EINPROGRESS) { |
234 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 235 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
236 | sock->sk->sk_state); |
237 | ret = 0; | |
238 | } | |
239 | if (ret < 0) { | |
240 | pr_err("connect %s error %d\n", | |
3d14c5d2 | 241 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e SW |
242 | sock_release(sock); |
243 | con->sock = NULL; | |
244 | con->error_msg = "connect error"; | |
245 | } | |
246 | ||
247 | if (ret < 0) | |
248 | return ERR_PTR(ret); | |
249 | return sock; | |
250 | } | |
251 | ||
252 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
253 | { | |
254 | struct kvec iov = {buf, len}; | |
255 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 256 | int r; |
31b8006e | 257 | |
98bdb0aa SW |
258 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
259 | if (r == -EAGAIN) | |
260 | r = 0; | |
261 | return r; | |
31b8006e SW |
262 | } |
263 | ||
264 | /* | |
265 | * write something. @more is true if caller will be sending more data | |
266 | * shortly. | |
267 | */ | |
268 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
269 | size_t kvlen, size_t len, int more) | |
270 | { | |
271 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 272 | int r; |
31b8006e SW |
273 | |
274 | if (more) | |
275 | msg.msg_flags |= MSG_MORE; | |
276 | else | |
277 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
278 | ||
42961d23 SW |
279 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
280 | if (r == -EAGAIN) | |
281 | r = 0; | |
282 | return r; | |
31b8006e SW |
283 | } |
284 | ||
285 | ||
286 | /* | |
287 | * Shutdown/close the socket for the given connection. | |
288 | */ | |
289 | static int con_close_socket(struct ceph_connection *con) | |
290 | { | |
291 | int rc; | |
292 | ||
293 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
294 | if (!con->sock) | |
295 | return 0; | |
296 | set_bit(SOCK_CLOSED, &con->state); | |
297 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
298 | sock_release(con->sock); | |
299 | con->sock = NULL; | |
300 | clear_bit(SOCK_CLOSED, &con->state); | |
301 | return rc; | |
302 | } | |
303 | ||
304 | /* | |
305 | * Reset a connection. Discard all incoming and outgoing messages | |
306 | * and clear *_seq state. | |
307 | */ | |
308 | static void ceph_msg_remove(struct ceph_msg *msg) | |
309 | { | |
310 | list_del_init(&msg->list_head); | |
311 | ceph_msg_put(msg); | |
312 | } | |
313 | static void ceph_msg_remove_list(struct list_head *head) | |
314 | { | |
315 | while (!list_empty(head)) { | |
316 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
317 | list_head); | |
318 | ceph_msg_remove(msg); | |
319 | } | |
320 | } | |
321 | ||
322 | static void reset_connection(struct ceph_connection *con) | |
323 | { | |
324 | /* reset connection, out_queue, msg_ and connect_seq */ | |
325 | /* discard existing out_queue and msg_seq */ | |
31b8006e SW |
326 | ceph_msg_remove_list(&con->out_queue); |
327 | ceph_msg_remove_list(&con->out_sent); | |
328 | ||
cf3e5c40 SW |
329 | if (con->in_msg) { |
330 | ceph_msg_put(con->in_msg); | |
331 | con->in_msg = NULL; | |
332 | } | |
333 | ||
31b8006e SW |
334 | con->connect_seq = 0; |
335 | con->out_seq = 0; | |
c86a2930 SW |
336 | if (con->out_msg) { |
337 | ceph_msg_put(con->out_msg); | |
338 | con->out_msg = NULL; | |
339 | } | |
31b8006e | 340 | con->in_seq = 0; |
0e0d5e0c | 341 | con->in_seq_acked = 0; |
31b8006e SW |
342 | } |
343 | ||
344 | /* | |
345 | * mark a peer down. drop any open connections. | |
346 | */ | |
347 | void ceph_con_close(struct ceph_connection *con) | |
348 | { | |
3d14c5d2 YS |
349 | dout("con_close %p peer %s\n", con, |
350 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
31b8006e SW |
351 | set_bit(CLOSED, &con->state); /* in case there's queued work */ |
352 | clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */ | |
1679f876 SW |
353 | clear_bit(LOSSYTX, &con->state); /* so we retry next connect */ |
354 | clear_bit(KEEPALIVE_PENDING, &con->state); | |
355 | clear_bit(WRITE_PENDING, &con->state); | |
ec302645 | 356 | mutex_lock(&con->mutex); |
31b8006e | 357 | reset_connection(con); |
6f2bc3ff | 358 | con->peer_global_seq = 0; |
91e45ce3 | 359 | cancel_delayed_work(&con->work); |
ec302645 | 360 | mutex_unlock(&con->mutex); |
31b8006e SW |
361 | queue_con(con); |
362 | } | |
3d14c5d2 | 363 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 364 | |
31b8006e SW |
365 | /* |
366 | * Reopen a closed connection, with a new peer address. | |
367 | */ | |
368 | void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr) | |
369 | { | |
3d14c5d2 | 370 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
31b8006e SW |
371 | set_bit(OPENING, &con->state); |
372 | clear_bit(CLOSED, &con->state); | |
373 | memcpy(&con->peer_addr, addr, sizeof(*addr)); | |
03c677e1 | 374 | con->delay = 0; /* reset backoff memory */ |
31b8006e SW |
375 | queue_con(con); |
376 | } | |
3d14c5d2 | 377 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 378 | |
87b315a5 SW |
379 | /* |
380 | * return true if this connection ever successfully opened | |
381 | */ | |
382 | bool ceph_con_opened(struct ceph_connection *con) | |
383 | { | |
384 | return con->connect_seq > 0; | |
385 | } | |
386 | ||
31b8006e SW |
387 | /* |
388 | * generic get/put | |
389 | */ | |
390 | struct ceph_connection *ceph_con_get(struct ceph_connection *con) | |
391 | { | |
392 | dout("con_get %p nref = %d -> %d\n", con, | |
393 | atomic_read(&con->nref), atomic_read(&con->nref) + 1); | |
394 | if (atomic_inc_not_zero(&con->nref)) | |
395 | return con; | |
396 | return NULL; | |
397 | } | |
398 | ||
399 | void ceph_con_put(struct ceph_connection *con) | |
400 | { | |
401 | dout("con_put %p nref = %d -> %d\n", con, | |
402 | atomic_read(&con->nref), atomic_read(&con->nref) - 1); | |
403 | BUG_ON(atomic_read(&con->nref) == 0); | |
404 | if (atomic_dec_and_test(&con->nref)) { | |
71ececda | 405 | BUG_ON(con->sock); |
31b8006e SW |
406 | kfree(con); |
407 | } | |
408 | } | |
409 | ||
410 | /* | |
411 | * initialize a new connection. | |
412 | */ | |
413 | void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con) | |
414 | { | |
415 | dout("con_init %p\n", con); | |
416 | memset(con, 0, sizeof(*con)); | |
417 | atomic_set(&con->nref, 1); | |
418 | con->msgr = msgr; | |
ec302645 | 419 | mutex_init(&con->mutex); |
31b8006e SW |
420 | INIT_LIST_HEAD(&con->out_queue); |
421 | INIT_LIST_HEAD(&con->out_sent); | |
422 | INIT_DELAYED_WORK(&con->work, con_work); | |
423 | } | |
3d14c5d2 | 424 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
425 | |
426 | ||
427 | /* | |
428 | * We maintain a global counter to order connection attempts. Get | |
429 | * a unique seq greater than @gt. | |
430 | */ | |
431 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
432 | { | |
433 | u32 ret; | |
434 | ||
435 | spin_lock(&msgr->global_seq_lock); | |
436 | if (msgr->global_seq < gt) | |
437 | msgr->global_seq = gt; | |
438 | ret = ++msgr->global_seq; | |
439 | spin_unlock(&msgr->global_seq_lock); | |
440 | return ret; | |
441 | } | |
442 | ||
443 | ||
444 | /* | |
445 | * Prepare footer for currently outgoing message, and finish things | |
446 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
447 | */ | |
448 | static void prepare_write_message_footer(struct ceph_connection *con, int v) | |
449 | { | |
450 | struct ceph_msg *m = con->out_msg; | |
451 | ||
452 | dout("prepare_write_message_footer %p\n", con); | |
453 | con->out_kvec_is_msg = true; | |
454 | con->out_kvec[v].iov_base = &m->footer; | |
455 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
456 | con->out_kvec_bytes += sizeof(m->footer); | |
457 | con->out_kvec_left++; | |
458 | con->out_more = m->more_to_follow; | |
c86a2930 | 459 | con->out_msg_done = true; |
31b8006e SW |
460 | } |
461 | ||
462 | /* | |
463 | * Prepare headers for the next outgoing message. | |
464 | */ | |
465 | static void prepare_write_message(struct ceph_connection *con) | |
466 | { | |
467 | struct ceph_msg *m; | |
468 | int v = 0; | |
469 | ||
470 | con->out_kvec_bytes = 0; | |
471 | con->out_kvec_is_msg = true; | |
c86a2930 | 472 | con->out_msg_done = false; |
31b8006e SW |
473 | |
474 | /* Sneak an ack in there first? If we can get it into the same | |
475 | * TCP packet that's a good thing. */ | |
476 | if (con->in_seq > con->in_seq_acked) { | |
477 | con->in_seq_acked = con->in_seq; | |
478 | con->out_kvec[v].iov_base = &tag_ack; | |
479 | con->out_kvec[v++].iov_len = 1; | |
480 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); | |
481 | con->out_kvec[v].iov_base = &con->out_temp_ack; | |
482 | con->out_kvec[v++].iov_len = sizeof(con->out_temp_ack); | |
483 | con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack); | |
484 | } | |
485 | ||
31b8006e SW |
486 | m = list_first_entry(&con->out_queue, |
487 | struct ceph_msg, list_head); | |
c86a2930 | 488 | con->out_msg = m; |
b3d1dbbd | 489 | if (test_bit(LOSSYTX, &con->state)) { |
6c5d1a49 SW |
490 | list_del_init(&m->list_head); |
491 | } else { | |
b3d1dbbd SW |
492 | /* put message on sent list */ |
493 | ceph_msg_get(m); | |
494 | list_move_tail(&m->list_head, &con->out_sent); | |
b3d1dbbd | 495 | } |
31b8006e | 496 | |
e84346b7 SW |
497 | /* |
498 | * only assign outgoing seq # if we haven't sent this message | |
499 | * yet. if it is requeued, resend with it's original seq. | |
500 | */ | |
501 | if (m->needs_out_seq) { | |
502 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
503 | m->needs_out_seq = false; | |
504 | } | |
31b8006e SW |
505 | |
506 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
507 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
508 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
509 | le32_to_cpu(m->hdr.data_len), | |
510 | m->nr_pages); | |
511 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); | |
512 | ||
513 | /* tag + hdr + front + middle */ | |
514 | con->out_kvec[v].iov_base = &tag_msg; | |
515 | con->out_kvec[v++].iov_len = 1; | |
516 | con->out_kvec[v].iov_base = &m->hdr; | |
517 | con->out_kvec[v++].iov_len = sizeof(m->hdr); | |
518 | con->out_kvec[v++] = m->front; | |
519 | if (m->middle) | |
520 | con->out_kvec[v++] = m->middle->vec; | |
521 | con->out_kvec_left = v; | |
522 | con->out_kvec_bytes += 1 + sizeof(m->hdr) + m->front.iov_len + | |
523 | (m->middle ? m->middle->vec.iov_len : 0); | |
524 | con->out_kvec_cur = con->out_kvec; | |
525 | ||
526 | /* fill in crc (except data pages), footer */ | |
527 | con->out_msg->hdr.crc = | |
528 | cpu_to_le32(crc32c(0, (void *)&m->hdr, | |
529 | sizeof(m->hdr) - sizeof(m->hdr.crc))); | |
530 | con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE; | |
531 | con->out_msg->footer.front_crc = | |
532 | cpu_to_le32(crc32c(0, m->front.iov_base, m->front.iov_len)); | |
533 | if (m->middle) | |
534 | con->out_msg->footer.middle_crc = | |
535 | cpu_to_le32(crc32c(0, m->middle->vec.iov_base, | |
536 | m->middle->vec.iov_len)); | |
537 | else | |
538 | con->out_msg->footer.middle_crc = 0; | |
539 | con->out_msg->footer.data_crc = 0; | |
540 | dout("prepare_write_message front_crc %u data_crc %u\n", | |
541 | le32_to_cpu(con->out_msg->footer.front_crc), | |
542 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
543 | ||
544 | /* is there a data payload? */ | |
545 | if (le32_to_cpu(m->hdr.data_len) > 0) { | |
546 | /* initialize page iterator */ | |
547 | con->out_msg_pos.page = 0; | |
68b4476b | 548 | if (m->pages) |
c5c6b19d | 549 | con->out_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
550 | else |
551 | con->out_msg_pos.page_pos = 0; | |
31b8006e SW |
552 | con->out_msg_pos.data_pos = 0; |
553 | con->out_msg_pos.did_page_crc = 0; | |
554 | con->out_more = 1; /* data + footer will follow */ | |
555 | } else { | |
556 | /* no, queue up footer too and be done */ | |
557 | prepare_write_message_footer(con, v); | |
558 | } | |
559 | ||
560 | set_bit(WRITE_PENDING, &con->state); | |
561 | } | |
562 | ||
563 | /* | |
564 | * Prepare an ack. | |
565 | */ | |
566 | static void prepare_write_ack(struct ceph_connection *con) | |
567 | { | |
568 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
569 | con->in_seq_acked, con->in_seq); | |
570 | con->in_seq_acked = con->in_seq; | |
571 | ||
572 | con->out_kvec[0].iov_base = &tag_ack; | |
573 | con->out_kvec[0].iov_len = 1; | |
574 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); | |
575 | con->out_kvec[1].iov_base = &con->out_temp_ack; | |
576 | con->out_kvec[1].iov_len = sizeof(con->out_temp_ack); | |
577 | con->out_kvec_left = 2; | |
578 | con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack); | |
579 | con->out_kvec_cur = con->out_kvec; | |
580 | con->out_more = 1; /* more will follow.. eventually.. */ | |
581 | set_bit(WRITE_PENDING, &con->state); | |
582 | } | |
583 | ||
584 | /* | |
585 | * Prepare to write keepalive byte. | |
586 | */ | |
587 | static void prepare_write_keepalive(struct ceph_connection *con) | |
588 | { | |
589 | dout("prepare_write_keepalive %p\n", con); | |
590 | con->out_kvec[0].iov_base = &tag_keepalive; | |
591 | con->out_kvec[0].iov_len = 1; | |
592 | con->out_kvec_left = 1; | |
593 | con->out_kvec_bytes = 1; | |
594 | con->out_kvec_cur = con->out_kvec; | |
595 | set_bit(WRITE_PENDING, &con->state); | |
596 | } | |
597 | ||
598 | /* | |
599 | * Connection negotiation. | |
600 | */ | |
601 | ||
0da5d703 | 602 | static int prepare_connect_authorizer(struct ceph_connection *con) |
4e7a5dcd SW |
603 | { |
604 | void *auth_buf; | |
605 | int auth_len = 0; | |
606 | int auth_protocol = 0; | |
607 | ||
ec302645 | 608 | mutex_unlock(&con->mutex); |
4e7a5dcd SW |
609 | if (con->ops->get_authorizer) |
610 | con->ops->get_authorizer(con, &auth_buf, &auth_len, | |
611 | &auth_protocol, &con->auth_reply_buf, | |
612 | &con->auth_reply_buf_len, | |
613 | con->auth_retry); | |
ec302645 | 614 | mutex_lock(&con->mutex); |
4e7a5dcd | 615 | |
0da5d703 SW |
616 | if (test_bit(CLOSED, &con->state) || |
617 | test_bit(OPENING, &con->state)) | |
618 | return -EAGAIN; | |
619 | ||
4e7a5dcd SW |
620 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol); |
621 | con->out_connect.authorizer_len = cpu_to_le32(auth_len); | |
622 | ||
e8f54ce1 SW |
623 | if (auth_len) { |
624 | con->out_kvec[con->out_kvec_left].iov_base = auth_buf; | |
625 | con->out_kvec[con->out_kvec_left].iov_len = auth_len; | |
626 | con->out_kvec_left++; | |
627 | con->out_kvec_bytes += auth_len; | |
628 | } | |
0da5d703 | 629 | return 0; |
4e7a5dcd SW |
630 | } |
631 | ||
31b8006e SW |
632 | /* |
633 | * We connected to a peer and are saying hello. | |
634 | */ | |
eed0ef2c SW |
635 | static void prepare_write_banner(struct ceph_messenger *msgr, |
636 | struct ceph_connection *con) | |
31b8006e SW |
637 | { |
638 | int len = strlen(CEPH_BANNER); | |
eed0ef2c SW |
639 | |
640 | con->out_kvec[0].iov_base = CEPH_BANNER; | |
641 | con->out_kvec[0].iov_len = len; | |
642 | con->out_kvec[1].iov_base = &msgr->my_enc_addr; | |
643 | con->out_kvec[1].iov_len = sizeof(msgr->my_enc_addr); | |
644 | con->out_kvec_left = 2; | |
645 | con->out_kvec_bytes = len + sizeof(msgr->my_enc_addr); | |
646 | con->out_kvec_cur = con->out_kvec; | |
647 | con->out_more = 0; | |
648 | set_bit(WRITE_PENDING, &con->state); | |
649 | } | |
650 | ||
0da5d703 SW |
651 | static int prepare_write_connect(struct ceph_messenger *msgr, |
652 | struct ceph_connection *con, | |
653 | int after_banner) | |
eed0ef2c | 654 | { |
31b8006e SW |
655 | unsigned global_seq = get_global_seq(con->msgr, 0); |
656 | int proto; | |
657 | ||
658 | switch (con->peer_name.type) { | |
659 | case CEPH_ENTITY_TYPE_MON: | |
660 | proto = CEPH_MONC_PROTOCOL; | |
661 | break; | |
662 | case CEPH_ENTITY_TYPE_OSD: | |
663 | proto = CEPH_OSDC_PROTOCOL; | |
664 | break; | |
665 | case CEPH_ENTITY_TYPE_MDS: | |
666 | proto = CEPH_MDSC_PROTOCOL; | |
667 | break; | |
668 | default: | |
669 | BUG(); | |
670 | } | |
671 | ||
672 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
673 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 674 | |
3d14c5d2 | 675 | con->out_connect.features = cpu_to_le64(msgr->supported_features); |
31b8006e SW |
676 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
677 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
678 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
679 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
680 | con->out_connect.flags = 0; | |
31b8006e | 681 | |
eed0ef2c SW |
682 | if (!after_banner) { |
683 | con->out_kvec_left = 0; | |
684 | con->out_kvec_bytes = 0; | |
685 | } | |
686 | con->out_kvec[con->out_kvec_left].iov_base = &con->out_connect; | |
687 | con->out_kvec[con->out_kvec_left].iov_len = sizeof(con->out_connect); | |
688 | con->out_kvec_left++; | |
689 | con->out_kvec_bytes += sizeof(con->out_connect); | |
31b8006e SW |
690 | con->out_kvec_cur = con->out_kvec; |
691 | con->out_more = 0; | |
692 | set_bit(WRITE_PENDING, &con->state); | |
4e7a5dcd | 693 | |
0da5d703 | 694 | return prepare_connect_authorizer(con); |
31b8006e SW |
695 | } |
696 | ||
697 | ||
698 | /* | |
699 | * write as much of pending kvecs to the socket as we can. | |
700 | * 1 -> done | |
701 | * 0 -> socket full, but more to do | |
702 | * <0 -> error | |
703 | */ | |
704 | static int write_partial_kvec(struct ceph_connection *con) | |
705 | { | |
706 | int ret; | |
707 | ||
708 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
709 | while (con->out_kvec_bytes > 0) { | |
710 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
711 | con->out_kvec_left, con->out_kvec_bytes, | |
712 | con->out_more); | |
713 | if (ret <= 0) | |
714 | goto out; | |
715 | con->out_kvec_bytes -= ret; | |
716 | if (con->out_kvec_bytes == 0) | |
717 | break; /* done */ | |
718 | while (ret > 0) { | |
719 | if (ret >= con->out_kvec_cur->iov_len) { | |
720 | ret -= con->out_kvec_cur->iov_len; | |
721 | con->out_kvec_cur++; | |
722 | con->out_kvec_left--; | |
723 | } else { | |
724 | con->out_kvec_cur->iov_len -= ret; | |
725 | con->out_kvec_cur->iov_base += ret; | |
726 | ret = 0; | |
727 | break; | |
728 | } | |
729 | } | |
730 | } | |
731 | con->out_kvec_left = 0; | |
732 | con->out_kvec_is_msg = false; | |
733 | ret = 1; | |
734 | out: | |
735 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
736 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
737 | return ret; /* done! */ | |
738 | } | |
739 | ||
68b4476b YS |
740 | #ifdef CONFIG_BLOCK |
741 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
742 | { | |
743 | if (!bio) { | |
744 | *iter = NULL; | |
745 | *seg = 0; | |
746 | return; | |
747 | } | |
748 | *iter = bio; | |
749 | *seg = bio->bi_idx; | |
750 | } | |
751 | ||
752 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
753 | { | |
754 | if (*bio_iter == NULL) | |
755 | return; | |
756 | ||
757 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
758 | ||
759 | (*seg)++; | |
760 | if (*seg == (*bio_iter)->bi_vcnt) | |
761 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
762 | } | |
763 | #endif | |
764 | ||
31b8006e SW |
765 | /* |
766 | * Write as much message data payload as we can. If we finish, queue | |
767 | * up the footer. | |
768 | * 1 -> done, footer is now queued in out_kvec[]. | |
769 | * 0 -> socket full, but more to do | |
770 | * <0 -> error | |
771 | */ | |
772 | static int write_partial_msg_pages(struct ceph_connection *con) | |
773 | { | |
774 | struct ceph_msg *msg = con->out_msg; | |
775 | unsigned data_len = le32_to_cpu(msg->hdr.data_len); | |
776 | size_t len; | |
777 | int crc = con->msgr->nocrc; | |
778 | int ret; | |
68b4476b YS |
779 | int total_max_write; |
780 | int in_trail = 0; | |
781 | size_t trail_len = (msg->trail ? msg->trail->length : 0); | |
31b8006e SW |
782 | |
783 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
784 | con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages, | |
785 | con->out_msg_pos.page_pos); | |
786 | ||
68b4476b YS |
787 | #ifdef CONFIG_BLOCK |
788 | if (msg->bio && !msg->bio_iter) | |
789 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); | |
790 | #endif | |
791 | ||
792 | while (data_len > con->out_msg_pos.data_pos) { | |
31b8006e SW |
793 | struct page *page = NULL; |
794 | void *kaddr = NULL; | |
68b4476b YS |
795 | int max_write = PAGE_SIZE; |
796 | int page_shift = 0; | |
797 | ||
798 | total_max_write = data_len - trail_len - | |
799 | con->out_msg_pos.data_pos; | |
31b8006e SW |
800 | |
801 | /* | |
802 | * if we are calculating the data crc (the default), we need | |
803 | * to map the page. if our pages[] has been revoked, use the | |
804 | * zero page. | |
805 | */ | |
68b4476b YS |
806 | |
807 | /* have we reached the trail part of the data? */ | |
808 | if (con->out_msg_pos.data_pos >= data_len - trail_len) { | |
809 | in_trail = 1; | |
810 | ||
811 | total_max_write = data_len - con->out_msg_pos.data_pos; | |
812 | ||
813 | page = list_first_entry(&msg->trail->head, | |
814 | struct page, lru); | |
815 | if (crc) | |
816 | kaddr = kmap(page); | |
817 | max_write = PAGE_SIZE; | |
818 | } else if (msg->pages) { | |
31b8006e SW |
819 | page = msg->pages[con->out_msg_pos.page]; |
820 | if (crc) | |
821 | kaddr = kmap(page); | |
58bb3b37 SW |
822 | } else if (msg->pagelist) { |
823 | page = list_first_entry(&msg->pagelist->head, | |
824 | struct page, lru); | |
825 | if (crc) | |
826 | kaddr = kmap(page); | |
68b4476b YS |
827 | #ifdef CONFIG_BLOCK |
828 | } else if (msg->bio) { | |
829 | struct bio_vec *bv; | |
830 | ||
831 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
832 | page = bv->bv_page; | |
833 | page_shift = bv->bv_offset; | |
834 | if (crc) | |
835 | kaddr = kmap(page) + page_shift; | |
836 | max_write = bv->bv_len; | |
837 | #endif | |
31b8006e SW |
838 | } else { |
839 | page = con->msgr->zero_page; | |
840 | if (crc) | |
841 | kaddr = page_address(con->msgr->zero_page); | |
842 | } | |
68b4476b YS |
843 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
844 | total_max_write); | |
845 | ||
31b8006e SW |
846 | if (crc && !con->out_msg_pos.did_page_crc) { |
847 | void *base = kaddr + con->out_msg_pos.page_pos; | |
848 | u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc); | |
849 | ||
850 | BUG_ON(kaddr == NULL); | |
851 | con->out_msg->footer.data_crc = | |
852 | cpu_to_le32(crc32c(tmpcrc, base, len)); | |
853 | con->out_msg_pos.did_page_crc = 1; | |
854 | } | |
31b8006e | 855 | ret = kernel_sendpage(con->sock, page, |
68b4476b YS |
856 | con->out_msg_pos.page_pos + page_shift, |
857 | len, | |
31b8006e SW |
858 | MSG_DONTWAIT | MSG_NOSIGNAL | |
859 | MSG_MORE); | |
860 | ||
68b4476b YS |
861 | if (crc && |
862 | (msg->pages || msg->pagelist || msg->bio || in_trail)) | |
31b8006e SW |
863 | kunmap(page); |
864 | ||
42961d23 SW |
865 | if (ret == -EAGAIN) |
866 | ret = 0; | |
31b8006e SW |
867 | if (ret <= 0) |
868 | goto out; | |
869 | ||
870 | con->out_msg_pos.data_pos += ret; | |
871 | con->out_msg_pos.page_pos += ret; | |
872 | if (ret == len) { | |
873 | con->out_msg_pos.page_pos = 0; | |
874 | con->out_msg_pos.page++; | |
875 | con->out_msg_pos.did_page_crc = 0; | |
68b4476b YS |
876 | if (in_trail) |
877 | list_move_tail(&page->lru, | |
878 | &msg->trail->head); | |
879 | else if (msg->pagelist) | |
58bb3b37 SW |
880 | list_move_tail(&page->lru, |
881 | &msg->pagelist->head); | |
68b4476b YS |
882 | #ifdef CONFIG_BLOCK |
883 | else if (msg->bio) | |
884 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
885 | #endif | |
31b8006e SW |
886 | } |
887 | } | |
888 | ||
889 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
890 | ||
891 | /* prepare and queue up footer, too */ | |
892 | if (!crc) | |
893 | con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; | |
894 | con->out_kvec_bytes = 0; | |
895 | con->out_kvec_left = 0; | |
896 | con->out_kvec_cur = con->out_kvec; | |
897 | prepare_write_message_footer(con, 0); | |
898 | ret = 1; | |
899 | out: | |
900 | return ret; | |
901 | } | |
902 | ||
903 | /* | |
904 | * write some zeros | |
905 | */ | |
906 | static int write_partial_skip(struct ceph_connection *con) | |
907 | { | |
908 | int ret; | |
909 | ||
910 | while (con->out_skip > 0) { | |
911 | struct kvec iov = { | |
912 | .iov_base = page_address(con->msgr->zero_page), | |
913 | .iov_len = min(con->out_skip, (int)PAGE_CACHE_SIZE) | |
914 | }; | |
915 | ||
916 | ret = ceph_tcp_sendmsg(con->sock, &iov, 1, iov.iov_len, 1); | |
917 | if (ret <= 0) | |
918 | goto out; | |
919 | con->out_skip -= ret; | |
920 | } | |
921 | ret = 1; | |
922 | out: | |
923 | return ret; | |
924 | } | |
925 | ||
926 | /* | |
927 | * Prepare to read connection handshake, or an ack. | |
928 | */ | |
eed0ef2c SW |
929 | static void prepare_read_banner(struct ceph_connection *con) |
930 | { | |
931 | dout("prepare_read_banner %p\n", con); | |
932 | con->in_base_pos = 0; | |
933 | } | |
934 | ||
31b8006e SW |
935 | static void prepare_read_connect(struct ceph_connection *con) |
936 | { | |
937 | dout("prepare_read_connect %p\n", con); | |
938 | con->in_base_pos = 0; | |
939 | } | |
940 | ||
941 | static void prepare_read_ack(struct ceph_connection *con) | |
942 | { | |
943 | dout("prepare_read_ack %p\n", con); | |
944 | con->in_base_pos = 0; | |
945 | } | |
946 | ||
947 | static void prepare_read_tag(struct ceph_connection *con) | |
948 | { | |
949 | dout("prepare_read_tag %p\n", con); | |
950 | con->in_base_pos = 0; | |
951 | con->in_tag = CEPH_MSGR_TAG_READY; | |
952 | } | |
953 | ||
954 | /* | |
955 | * Prepare to read a message. | |
956 | */ | |
957 | static int prepare_read_message(struct ceph_connection *con) | |
958 | { | |
959 | dout("prepare_read_message %p\n", con); | |
960 | BUG_ON(con->in_msg != NULL); | |
961 | con->in_base_pos = 0; | |
962 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
963 | return 0; | |
964 | } | |
965 | ||
966 | ||
967 | static int read_partial(struct ceph_connection *con, | |
968 | int *to, int size, void *object) | |
969 | { | |
970 | *to += size; | |
971 | while (con->in_base_pos < *to) { | |
972 | int left = *to - con->in_base_pos; | |
973 | int have = size - left; | |
974 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
975 | if (ret <= 0) | |
976 | return ret; | |
977 | con->in_base_pos += ret; | |
978 | } | |
979 | return 1; | |
980 | } | |
981 | ||
982 | ||
983 | /* | |
984 | * Read all or part of the connect-side handshake on a new connection | |
985 | */ | |
eed0ef2c | 986 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e SW |
987 | { |
988 | int ret, to = 0; | |
989 | ||
eed0ef2c | 990 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
991 | |
992 | /* peer's banner */ | |
993 | ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner); | |
994 | if (ret <= 0) | |
995 | goto out; | |
996 | ret = read_partial(con, &to, sizeof(con->actual_peer_addr), | |
997 | &con->actual_peer_addr); | |
998 | if (ret <= 0) | |
999 | goto out; | |
1000 | ret = read_partial(con, &to, sizeof(con->peer_addr_for_me), | |
1001 | &con->peer_addr_for_me); | |
1002 | if (ret <= 0) | |
1003 | goto out; | |
eed0ef2c SW |
1004 | out: |
1005 | return ret; | |
1006 | } | |
1007 | ||
1008 | static int read_partial_connect(struct ceph_connection *con) | |
1009 | { | |
1010 | int ret, to = 0; | |
1011 | ||
1012 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1013 | ||
31b8006e SW |
1014 | ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply); |
1015 | if (ret <= 0) | |
1016 | goto out; | |
4e7a5dcd SW |
1017 | ret = read_partial(con, &to, le32_to_cpu(con->in_reply.authorizer_len), |
1018 | con->auth_reply_buf); | |
1019 | if (ret <= 0) | |
1020 | goto out; | |
31b8006e | 1021 | |
4e7a5dcd SW |
1022 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1023 | con, (int)con->in_reply.tag, | |
1024 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1025 | le32_to_cpu(con->in_reply.global_seq)); |
1026 | out: | |
1027 | return ret; | |
eed0ef2c | 1028 | |
31b8006e SW |
1029 | } |
1030 | ||
1031 | /* | |
1032 | * Verify the hello banner looks okay. | |
1033 | */ | |
1034 | static int verify_hello(struct ceph_connection *con) | |
1035 | { | |
1036 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1037 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1038 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1039 | con->error_msg = "protocol error, bad banner"; |
1040 | return -1; | |
1041 | } | |
1042 | return 0; | |
1043 | } | |
1044 | ||
1045 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1046 | { | |
1047 | switch (ss->ss_family) { | |
1048 | case AF_INET: | |
1049 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1050 | case AF_INET6: | |
1051 | return | |
1052 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1053 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1054 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1055 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1056 | } | |
1057 | return false; | |
1058 | } | |
1059 | ||
1060 | static int addr_port(struct sockaddr_storage *ss) | |
1061 | { | |
1062 | switch (ss->ss_family) { | |
1063 | case AF_INET: | |
f28bcfbe | 1064 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1065 | case AF_INET6: |
f28bcfbe | 1066 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1067 | } |
1068 | return 0; | |
1069 | } | |
1070 | ||
1071 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1072 | { | |
1073 | switch (ss->ss_family) { | |
1074 | case AF_INET: | |
1075 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
1076 | case AF_INET6: | |
1077 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
1078 | } | |
1079 | } | |
1080 | ||
1081 | /* | |
1082 | * Parse an ip[:port] list into an addr array. Use the default | |
1083 | * monitor port if a port isn't specified. | |
1084 | */ | |
1085 | int ceph_parse_ips(const char *c, const char *end, | |
1086 | struct ceph_entity_addr *addr, | |
1087 | int max_count, int *count) | |
1088 | { | |
1089 | int i; | |
1090 | const char *p = c; | |
1091 | ||
1092 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1093 | for (i = 0; i < max_count; i++) { | |
1094 | const char *ipend; | |
1095 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
1096 | struct sockaddr_in *in4 = (void *)ss; | |
1097 | struct sockaddr_in6 *in6 = (void *)ss; | |
1098 | int port; | |
39139f64 SW |
1099 | char delim = ','; |
1100 | ||
1101 | if (*p == '[') { | |
1102 | delim = ']'; | |
1103 | p++; | |
1104 | } | |
31b8006e SW |
1105 | |
1106 | memset(ss, 0, sizeof(*ss)); | |
1107 | if (in4_pton(p, end - p, (u8 *)&in4->sin_addr.s_addr, | |
39139f64 | 1108 | delim, &ipend)) |
31b8006e | 1109 | ss->ss_family = AF_INET; |
39139f64 SW |
1110 | else if (in6_pton(p, end - p, (u8 *)&in6->sin6_addr.s6_addr, |
1111 | delim, &ipend)) | |
31b8006e | 1112 | ss->ss_family = AF_INET6; |
39139f64 | 1113 | else |
31b8006e | 1114 | goto bad; |
31b8006e SW |
1115 | p = ipend; |
1116 | ||
39139f64 SW |
1117 | if (delim == ']') { |
1118 | if (*p != ']') { | |
1119 | dout("missing matching ']'\n"); | |
1120 | goto bad; | |
1121 | } | |
1122 | p++; | |
1123 | } | |
1124 | ||
31b8006e SW |
1125 | /* port? */ |
1126 | if (p < end && *p == ':') { | |
1127 | port = 0; | |
1128 | p++; | |
1129 | while (p < end && *p >= '0' && *p <= '9') { | |
1130 | port = (port * 10) + (*p - '0'); | |
1131 | p++; | |
1132 | } | |
1133 | if (port > 65535 || port == 0) | |
1134 | goto bad; | |
1135 | } else { | |
1136 | port = CEPH_MON_PORT; | |
1137 | } | |
1138 | ||
1139 | addr_set_port(ss, port); | |
1140 | ||
3d14c5d2 | 1141 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1142 | |
1143 | if (p == end) | |
1144 | break; | |
1145 | if (*p != ',') | |
1146 | goto bad; | |
1147 | p++; | |
1148 | } | |
1149 | ||
1150 | if (p != end) | |
1151 | goto bad; | |
1152 | ||
1153 | if (count) | |
1154 | *count = i + 1; | |
1155 | return 0; | |
1156 | ||
1157 | bad: | |
39139f64 | 1158 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
31b8006e SW |
1159 | return -EINVAL; |
1160 | } | |
3d14c5d2 | 1161 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1162 | |
eed0ef2c | 1163 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1164 | { |
eed0ef2c | 1165 | dout("process_banner on %p\n", con); |
31b8006e SW |
1166 | |
1167 | if (verify_hello(con) < 0) | |
1168 | return -1; | |
1169 | ||
63f2d211 SW |
1170 | ceph_decode_addr(&con->actual_peer_addr); |
1171 | ceph_decode_addr(&con->peer_addr_for_me); | |
1172 | ||
31b8006e SW |
1173 | /* |
1174 | * Make sure the other end is who we wanted. note that the other | |
1175 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1176 | * them the benefit of the doubt. | |
1177 | */ | |
103e2d3a SW |
1178 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1179 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1180 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1181 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1182 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1183 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1184 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1185 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1186 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1187 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1188 | return -1; |
1189 | } | |
1190 | ||
1191 | /* | |
1192 | * did we learn our address? | |
1193 | */ | |
1194 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1195 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1196 | ||
1197 | memcpy(&con->msgr->inst.addr.in_addr, | |
1198 | &con->peer_addr_for_me.in_addr, | |
1199 | sizeof(con->peer_addr_for_me.in_addr)); | |
1200 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1201 | encode_my_addr(con->msgr); |
eed0ef2c | 1202 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1203 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1204 | } |
1205 | ||
eed0ef2c SW |
1206 | set_bit(NEGOTIATING, &con->state); |
1207 | prepare_read_connect(con); | |
1208 | return 0; | |
1209 | } | |
1210 | ||
04a419f9 SW |
1211 | static void fail_protocol(struct ceph_connection *con) |
1212 | { | |
1213 | reset_connection(con); | |
1214 | set_bit(CLOSED, &con->state); /* in case there's queued work */ | |
1215 | ||
1216 | mutex_unlock(&con->mutex); | |
1217 | if (con->ops->bad_proto) | |
1218 | con->ops->bad_proto(con); | |
1219 | mutex_lock(&con->mutex); | |
1220 | } | |
1221 | ||
eed0ef2c SW |
1222 | static int process_connect(struct ceph_connection *con) |
1223 | { | |
3d14c5d2 YS |
1224 | u64 sup_feat = con->msgr->supported_features; |
1225 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1226 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1227 | int ret; |
04a419f9 | 1228 | |
eed0ef2c SW |
1229 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1230 | ||
31b8006e | 1231 | switch (con->in_reply.tag) { |
04a419f9 SW |
1232 | case CEPH_MSGR_TAG_FEATURES: |
1233 | pr_err("%s%lld %s feature set mismatch," | |
1234 | " my %llx < server's %llx, missing %llx\n", | |
1235 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1236 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1237 | sup_feat, server_feat, server_feat & ~sup_feat); |
1238 | con->error_msg = "missing required protocol features"; | |
1239 | fail_protocol(con); | |
1240 | return -1; | |
1241 | ||
31b8006e | 1242 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1243 | pr_err("%s%lld %s protocol version mismatch," |
1244 | " my %d != server's %d\n", | |
1245 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1246 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1247 | le32_to_cpu(con->out_connect.protocol_version), |
1248 | le32_to_cpu(con->in_reply.protocol_version)); | |
1249 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1250 | fail_protocol(con); |
31b8006e SW |
1251 | return -1; |
1252 | ||
4e7a5dcd SW |
1253 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1254 | con->auth_retry++; | |
1255 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1256 | con->auth_retry); | |
1257 | if (con->auth_retry == 2) { | |
1258 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1259 | return -1; |
1260 | } | |
1261 | con->auth_retry = 1; | |
0da5d703 SW |
1262 | ret = prepare_write_connect(con->msgr, con, 0); |
1263 | if (ret < 0) | |
1264 | return ret; | |
63733a0f | 1265 | prepare_read_connect(con); |
4e7a5dcd | 1266 | break; |
31b8006e SW |
1267 | |
1268 | case CEPH_MSGR_TAG_RESETSESSION: | |
1269 | /* | |
1270 | * If we connected with a large connect_seq but the peer | |
1271 | * has no record of a session with us (no connection, or | |
1272 | * connect_seq == 0), they will send RESETSESION to indicate | |
1273 | * that they must have reset their session, and may have | |
1274 | * dropped messages. | |
1275 | */ | |
1276 | dout("process_connect got RESET peer seq %u\n", | |
1277 | le32_to_cpu(con->in_connect.connect_seq)); | |
1278 | pr_err("%s%lld %s connection reset\n", | |
1279 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1280 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1281 | reset_connection(con); |
eed0ef2c | 1282 | prepare_write_connect(con->msgr, con, 0); |
31b8006e SW |
1283 | prepare_read_connect(con); |
1284 | ||
1285 | /* Tell ceph about it. */ | |
ec302645 | 1286 | mutex_unlock(&con->mutex); |
31b8006e SW |
1287 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1288 | if (con->ops->peer_reset) | |
1289 | con->ops->peer_reset(con); | |
ec302645 | 1290 | mutex_lock(&con->mutex); |
0da5d703 SW |
1291 | if (test_bit(CLOSED, &con->state) || |
1292 | test_bit(OPENING, &con->state)) | |
1293 | return -EAGAIN; | |
31b8006e SW |
1294 | break; |
1295 | ||
1296 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1297 | /* | |
1298 | * If we sent a smaller connect_seq than the peer has, try | |
1299 | * again with a larger value. | |
1300 | */ | |
1301 | dout("process_connect got RETRY my seq = %u, peer_seq = %u\n", | |
1302 | le32_to_cpu(con->out_connect.connect_seq), | |
1303 | le32_to_cpu(con->in_connect.connect_seq)); | |
1304 | con->connect_seq = le32_to_cpu(con->in_connect.connect_seq); | |
eed0ef2c | 1305 | prepare_write_connect(con->msgr, con, 0); |
31b8006e SW |
1306 | prepare_read_connect(con); |
1307 | break; | |
1308 | ||
1309 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1310 | /* | |
1311 | * If we sent a smaller global_seq than the peer has, try | |
1312 | * again with a larger value. | |
1313 | */ | |
eed0ef2c | 1314 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e SW |
1315 | con->peer_global_seq, |
1316 | le32_to_cpu(con->in_connect.global_seq)); | |
1317 | get_global_seq(con->msgr, | |
1318 | le32_to_cpu(con->in_connect.global_seq)); | |
eed0ef2c | 1319 | prepare_write_connect(con->msgr, con, 0); |
31b8006e SW |
1320 | prepare_read_connect(con); |
1321 | break; | |
1322 | ||
1323 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1324 | if (req_feat & ~server_feat) { |
1325 | pr_err("%s%lld %s protocol feature mismatch," | |
1326 | " my required %llx > server's %llx, need %llx\n", | |
1327 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1328 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1329 | req_feat, server_feat, req_feat & ~server_feat); |
1330 | con->error_msg = "missing required protocol features"; | |
1331 | fail_protocol(con); | |
1332 | return -1; | |
1333 | } | |
31b8006e | 1334 | clear_bit(CONNECTING, &con->state); |
31b8006e SW |
1335 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1336 | con->connect_seq++; | |
aba558e2 | 1337 | con->peer_features = server_feat; |
31b8006e SW |
1338 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1339 | con->peer_global_seq, | |
1340 | le32_to_cpu(con->in_reply.connect_seq), | |
1341 | con->connect_seq); | |
1342 | WARN_ON(con->connect_seq != | |
1343 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1344 | |
1345 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
1346 | set_bit(LOSSYTX, &con->state); | |
1347 | ||
31b8006e SW |
1348 | prepare_read_tag(con); |
1349 | break; | |
1350 | ||
1351 | case CEPH_MSGR_TAG_WAIT: | |
1352 | /* | |
1353 | * If there is a connection race (we are opening | |
1354 | * connections to each other), one of us may just have | |
1355 | * to WAIT. This shouldn't happen if we are the | |
1356 | * client. | |
1357 | */ | |
1358 | pr_err("process_connect peer connecting WAIT\n"); | |
1359 | ||
1360 | default: | |
1361 | pr_err("connect protocol error, will retry\n"); | |
1362 | con->error_msg = "protocol error, garbage tag during connect"; | |
1363 | return -1; | |
1364 | } | |
1365 | return 0; | |
1366 | } | |
1367 | ||
1368 | ||
1369 | /* | |
1370 | * read (part of) an ack | |
1371 | */ | |
1372 | static int read_partial_ack(struct ceph_connection *con) | |
1373 | { | |
1374 | int to = 0; | |
1375 | ||
1376 | return read_partial(con, &to, sizeof(con->in_temp_ack), | |
1377 | &con->in_temp_ack); | |
1378 | } | |
1379 | ||
1380 | ||
1381 | /* | |
1382 | * We can finally discard anything that's been acked. | |
1383 | */ | |
1384 | static void process_ack(struct ceph_connection *con) | |
1385 | { | |
1386 | struct ceph_msg *m; | |
1387 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1388 | u64 seq; | |
1389 | ||
31b8006e SW |
1390 | while (!list_empty(&con->out_sent)) { |
1391 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1392 | list_head); | |
1393 | seq = le64_to_cpu(m->hdr.seq); | |
1394 | if (seq > ack) | |
1395 | break; | |
1396 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1397 | le16_to_cpu(m->hdr.type), m); | |
1398 | ceph_msg_remove(m); | |
1399 | } | |
31b8006e SW |
1400 | prepare_read_tag(con); |
1401 | } | |
1402 | ||
1403 | ||
1404 | ||
1405 | ||
2450418c | 1406 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1407 | struct kvec *section, |
1408 | unsigned int sec_len, u32 *crc) | |
2450418c | 1409 | { |
68b4476b | 1410 | int ret, left; |
2450418c YS |
1411 | |
1412 | BUG_ON(!section); | |
1413 | ||
1414 | while (section->iov_len < sec_len) { | |
1415 | BUG_ON(section->iov_base == NULL); | |
1416 | left = sec_len - section->iov_len; | |
1417 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1418 | section->iov_len, left); | |
1419 | if (ret <= 0) | |
1420 | return ret; | |
1421 | section->iov_len += ret; | |
1422 | if (section->iov_len == sec_len) | |
1423 | *crc = crc32c(0, section->iov_base, | |
1424 | section->iov_len); | |
1425 | } | |
31b8006e | 1426 | |
2450418c YS |
1427 | return 1; |
1428 | } | |
31b8006e | 1429 | |
2450418c YS |
1430 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, |
1431 | struct ceph_msg_header *hdr, | |
1432 | int *skip); | |
68b4476b YS |
1433 | |
1434 | ||
1435 | static int read_partial_message_pages(struct ceph_connection *con, | |
1436 | struct page **pages, | |
1437 | unsigned data_len, int datacrc) | |
1438 | { | |
1439 | void *p; | |
1440 | int ret; | |
1441 | int left; | |
1442 | ||
1443 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1444 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1445 | /* (page) data */ | |
1446 | BUG_ON(pages == NULL); | |
1447 | p = kmap(pages[con->in_msg_pos.page]); | |
1448 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1449 | left); | |
1450 | if (ret > 0 && datacrc) | |
1451 | con->in_data_crc = | |
1452 | crc32c(con->in_data_crc, | |
1453 | p + con->in_msg_pos.page_pos, ret); | |
1454 | kunmap(pages[con->in_msg_pos.page]); | |
1455 | if (ret <= 0) | |
1456 | return ret; | |
1457 | con->in_msg_pos.data_pos += ret; | |
1458 | con->in_msg_pos.page_pos += ret; | |
1459 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1460 | con->in_msg_pos.page_pos = 0; | |
1461 | con->in_msg_pos.page++; | |
1462 | } | |
1463 | ||
1464 | return ret; | |
1465 | } | |
1466 | ||
1467 | #ifdef CONFIG_BLOCK | |
1468 | static int read_partial_message_bio(struct ceph_connection *con, | |
1469 | struct bio **bio_iter, int *bio_seg, | |
1470 | unsigned data_len, int datacrc) | |
1471 | { | |
1472 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1473 | void *p; | |
1474 | int ret, left; | |
1475 | ||
1476 | if (IS_ERR(bv)) | |
1477 | return PTR_ERR(bv); | |
1478 | ||
1479 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1480 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1481 | ||
1482 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1483 | ||
1484 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1485 | left); | |
1486 | if (ret > 0 && datacrc) | |
1487 | con->in_data_crc = | |
1488 | crc32c(con->in_data_crc, | |
1489 | p + con->in_msg_pos.page_pos, ret); | |
1490 | kunmap(bv->bv_page); | |
1491 | if (ret <= 0) | |
1492 | return ret; | |
1493 | con->in_msg_pos.data_pos += ret; | |
1494 | con->in_msg_pos.page_pos += ret; | |
1495 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1496 | con->in_msg_pos.page_pos = 0; | |
1497 | iter_bio_next(bio_iter, bio_seg); | |
1498 | } | |
1499 | ||
1500 | return ret; | |
1501 | } | |
1502 | #endif | |
1503 | ||
31b8006e SW |
1504 | /* |
1505 | * read (part of) a message. | |
1506 | */ | |
1507 | static int read_partial_message(struct ceph_connection *con) | |
1508 | { | |
1509 | struct ceph_msg *m = con->in_msg; | |
31b8006e | 1510 | int ret; |
9d7f0f13 | 1511 | int to, left; |
c5c6b19d | 1512 | unsigned front_len, middle_len, data_len; |
31b8006e | 1513 | int datacrc = con->msgr->nocrc; |
2450418c | 1514 | int skip; |
ae18756b | 1515 | u64 seq; |
31b8006e SW |
1516 | |
1517 | dout("read_partial_message con %p msg %p\n", con, m); | |
1518 | ||
1519 | /* header */ | |
1520 | while (con->in_base_pos < sizeof(con->in_hdr)) { | |
1521 | left = sizeof(con->in_hdr) - con->in_base_pos; | |
1522 | ret = ceph_tcp_recvmsg(con->sock, | |
1523 | (char *)&con->in_hdr + con->in_base_pos, | |
1524 | left); | |
1525 | if (ret <= 0) | |
1526 | return ret; | |
1527 | con->in_base_pos += ret; | |
1528 | if (con->in_base_pos == sizeof(con->in_hdr)) { | |
1529 | u32 crc = crc32c(0, (void *)&con->in_hdr, | |
1530 | sizeof(con->in_hdr) - sizeof(con->in_hdr.crc)); | |
1531 | if (crc != le32_to_cpu(con->in_hdr.crc)) { | |
1532 | pr_err("read_partial_message bad hdr " | |
1533 | " crc %u != expected %u\n", | |
1534 | crc, con->in_hdr.crc); | |
1535 | return -EBADMSG; | |
1536 | } | |
1537 | } | |
1538 | } | |
31b8006e SW |
1539 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1540 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1541 | return -EIO; | |
1542 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1543 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1544 | return -EIO; | |
1545 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1546 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1547 | return -EIO; | |
1548 | ||
ae18756b SW |
1549 | /* verify seq# */ |
1550 | seq = le64_to_cpu(con->in_hdr.seq); | |
1551 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1552 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1553 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1554 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1555 | seq, con->in_seq + 1); |
1556 | con->in_base_pos = -front_len - middle_len - data_len - | |
1557 | sizeof(m->footer); | |
1558 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1559 | return 0; |
1560 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1561 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1562 | seq, con->in_seq + 1); | |
1563 | con->error_msg = "bad message sequence # for incoming message"; | |
1564 | return -EBADMSG; | |
1565 | } | |
1566 | ||
31b8006e SW |
1567 | /* allocate message? */ |
1568 | if (!con->in_msg) { | |
1569 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1570 | con->in_hdr.front_len, con->in_hdr.data_len); | |
ae32be31 | 1571 | skip = 0; |
2450418c YS |
1572 | con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip); |
1573 | if (skip) { | |
31b8006e | 1574 | /* skip this message */ |
a79832f2 | 1575 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1576 | BUG_ON(con->in_msg); |
31b8006e SW |
1577 | con->in_base_pos = -front_len - middle_len - data_len - |
1578 | sizeof(m->footer); | |
1579 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1580 | con->in_seq++; |
31b8006e SW |
1581 | return 0; |
1582 | } | |
a79832f2 | 1583 | if (!con->in_msg) { |
5b3a4db3 SW |
1584 | con->error_msg = |
1585 | "error allocating memory for incoming message"; | |
a79832f2 | 1586 | return -ENOMEM; |
31b8006e SW |
1587 | } |
1588 | m = con->in_msg; | |
1589 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1590 | if (m->middle) |
1591 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1592 | |
1593 | con->in_msg_pos.page = 0; | |
68b4476b | 1594 | if (m->pages) |
c5c6b19d | 1595 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1596 | else |
1597 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1598 | con->in_msg_pos.data_pos = 0; |
31b8006e SW |
1599 | } |
1600 | ||
1601 | /* front */ | |
2450418c YS |
1602 | ret = read_partial_message_section(con, &m->front, front_len, |
1603 | &con->in_front_crc); | |
1604 | if (ret <= 0) | |
1605 | return ret; | |
31b8006e SW |
1606 | |
1607 | /* middle */ | |
2450418c | 1608 | if (m->middle) { |
213c99ee SW |
1609 | ret = read_partial_message_section(con, &m->middle->vec, |
1610 | middle_len, | |
2450418c | 1611 | &con->in_middle_crc); |
31b8006e SW |
1612 | if (ret <= 0) |
1613 | return ret; | |
31b8006e | 1614 | } |
68b4476b YS |
1615 | #ifdef CONFIG_BLOCK |
1616 | if (m->bio && !m->bio_iter) | |
1617 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1618 | #endif | |
31b8006e SW |
1619 | |
1620 | /* (page) data */ | |
31b8006e | 1621 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1622 | if (m->pages) { |
1623 | ret = read_partial_message_pages(con, m->pages, | |
1624 | data_len, datacrc); | |
1625 | if (ret <= 0) | |
1626 | return ret; | |
1627 | #ifdef CONFIG_BLOCK | |
1628 | } else if (m->bio) { | |
1629 | ||
1630 | ret = read_partial_message_bio(con, | |
1631 | &m->bio_iter, &m->bio_seg, | |
1632 | data_len, datacrc); | |
1633 | if (ret <= 0) | |
1634 | return ret; | |
1635 | #endif | |
1636 | } else { | |
1637 | BUG_ON(1); | |
31b8006e SW |
1638 | } |
1639 | } | |
1640 | ||
31b8006e SW |
1641 | /* footer */ |
1642 | to = sizeof(m->hdr) + sizeof(m->footer); | |
1643 | while (con->in_base_pos < to) { | |
1644 | left = to - con->in_base_pos; | |
1645 | ret = ceph_tcp_recvmsg(con->sock, (char *)&m->footer + | |
1646 | (con->in_base_pos - sizeof(m->hdr)), | |
1647 | left); | |
1648 | if (ret <= 0) | |
1649 | return ret; | |
1650 | con->in_base_pos += ret; | |
1651 | } | |
1652 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", | |
1653 | m, front_len, m->footer.front_crc, middle_len, | |
1654 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1655 | ||
1656 | /* crc ok? */ | |
1657 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1658 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1659 | m, con->in_front_crc, m->footer.front_crc); | |
1660 | return -EBADMSG; | |
1661 | } | |
1662 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1663 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1664 | m, con->in_middle_crc, m->footer.middle_crc); | |
1665 | return -EBADMSG; | |
1666 | } | |
1667 | if (datacrc && | |
1668 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && | |
1669 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1670 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1671 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1672 | return -EBADMSG; | |
1673 | } | |
1674 | ||
1675 | return 1; /* done! */ | |
1676 | } | |
1677 | ||
1678 | /* | |
1679 | * Process message. This happens in the worker thread. The callback should | |
1680 | * be careful not to do anything that waits on other incoming messages or it | |
1681 | * may deadlock. | |
1682 | */ | |
1683 | static void process_message(struct ceph_connection *con) | |
1684 | { | |
5e095e8b | 1685 | struct ceph_msg *msg; |
31b8006e | 1686 | |
5e095e8b | 1687 | msg = con->in_msg; |
31b8006e SW |
1688 | con->in_msg = NULL; |
1689 | ||
1690 | /* if first message, set peer_name */ | |
1691 | if (con->peer_name.type == 0) | |
dbad185d | 1692 | con->peer_name = msg->hdr.src; |
31b8006e | 1693 | |
31b8006e | 1694 | con->in_seq++; |
ec302645 | 1695 | mutex_unlock(&con->mutex); |
31b8006e SW |
1696 | |
1697 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1698 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1699 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1700 | le16_to_cpu(msg->hdr.type), |
1701 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1702 | le32_to_cpu(msg->hdr.front_len), | |
1703 | le32_to_cpu(msg->hdr.data_len), | |
1704 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1705 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1706 | |
1707 | mutex_lock(&con->mutex); | |
31b8006e SW |
1708 | prepare_read_tag(con); |
1709 | } | |
1710 | ||
1711 | ||
1712 | /* | |
1713 | * Write something to the socket. Called in a worker thread when the | |
1714 | * socket appears to be writeable and we have something ready to send. | |
1715 | */ | |
1716 | static int try_write(struct ceph_connection *con) | |
1717 | { | |
1718 | struct ceph_messenger *msgr = con->msgr; | |
1719 | int ret = 1; | |
1720 | ||
1721 | dout("try_write start %p state %lu nref %d\n", con, con->state, | |
1722 | atomic_read(&con->nref)); | |
1723 | ||
31b8006e SW |
1724 | more: |
1725 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
1726 | ||
1727 | /* open the socket first? */ | |
1728 | if (con->sock == NULL) { | |
eed0ef2c SW |
1729 | prepare_write_banner(msgr, con); |
1730 | prepare_write_connect(msgr, con, 1); | |
1731 | prepare_read_banner(con); | |
31b8006e | 1732 | set_bit(CONNECTING, &con->state); |
eed0ef2c | 1733 | clear_bit(NEGOTIATING, &con->state); |
31b8006e | 1734 | |
cf3e5c40 | 1735 | BUG_ON(con->in_msg); |
31b8006e SW |
1736 | con->in_tag = CEPH_MSGR_TAG_READY; |
1737 | dout("try_write initiating connect on %p new state %lu\n", | |
1738 | con, con->state); | |
1739 | con->sock = ceph_tcp_connect(con); | |
1740 | if (IS_ERR(con->sock)) { | |
1741 | con->sock = NULL; | |
1742 | con->error_msg = "connect error"; | |
1743 | ret = -1; | |
1744 | goto out; | |
1745 | } | |
1746 | } | |
1747 | ||
1748 | more_kvec: | |
1749 | /* kvec data queued? */ | |
1750 | if (con->out_skip) { | |
1751 | ret = write_partial_skip(con); | |
1752 | if (ret <= 0) | |
42961d23 | 1753 | goto out; |
31b8006e SW |
1754 | } |
1755 | if (con->out_kvec_left) { | |
1756 | ret = write_partial_kvec(con); | |
1757 | if (ret <= 0) | |
42961d23 | 1758 | goto out; |
31b8006e SW |
1759 | } |
1760 | ||
1761 | /* msg pages? */ | |
1762 | if (con->out_msg) { | |
c86a2930 SW |
1763 | if (con->out_msg_done) { |
1764 | ceph_msg_put(con->out_msg); | |
1765 | con->out_msg = NULL; /* we're done with this one */ | |
1766 | goto do_next; | |
1767 | } | |
1768 | ||
31b8006e SW |
1769 | ret = write_partial_msg_pages(con); |
1770 | if (ret == 1) | |
1771 | goto more_kvec; /* we need to send the footer, too! */ | |
1772 | if (ret == 0) | |
42961d23 | 1773 | goto out; |
31b8006e SW |
1774 | if (ret < 0) { |
1775 | dout("try_write write_partial_msg_pages err %d\n", | |
1776 | ret); | |
42961d23 | 1777 | goto out; |
31b8006e SW |
1778 | } |
1779 | } | |
1780 | ||
c86a2930 | 1781 | do_next: |
31b8006e SW |
1782 | if (!test_bit(CONNECTING, &con->state)) { |
1783 | /* is anything else pending? */ | |
1784 | if (!list_empty(&con->out_queue)) { | |
1785 | prepare_write_message(con); | |
1786 | goto more; | |
1787 | } | |
1788 | if (con->in_seq > con->in_seq_acked) { | |
1789 | prepare_write_ack(con); | |
1790 | goto more; | |
1791 | } | |
1792 | if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) { | |
1793 | prepare_write_keepalive(con); | |
1794 | goto more; | |
1795 | } | |
1796 | } | |
1797 | ||
1798 | /* Nothing to do! */ | |
1799 | clear_bit(WRITE_PENDING, &con->state); | |
1800 | dout("try_write nothing else to write.\n"); | |
31b8006e SW |
1801 | ret = 0; |
1802 | out: | |
42961d23 | 1803 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
1804 | return ret; |
1805 | } | |
1806 | ||
1807 | ||
1808 | ||
1809 | /* | |
1810 | * Read what we can from the socket. | |
1811 | */ | |
1812 | static int try_read(struct ceph_connection *con) | |
1813 | { | |
31b8006e SW |
1814 | int ret = -1; |
1815 | ||
1816 | if (!con->sock) | |
1817 | return 0; | |
1818 | ||
1819 | if (test_bit(STANDBY, &con->state)) | |
1820 | return 0; | |
1821 | ||
1822 | dout("try_read start on %p\n", con); | |
ec302645 | 1823 | |
31b8006e SW |
1824 | more: |
1825 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, | |
1826 | con->in_base_pos); | |
0da5d703 SW |
1827 | |
1828 | /* | |
1829 | * process_connect and process_message drop and re-take | |
1830 | * con->mutex. make sure we handle a racing close or reopen. | |
1831 | */ | |
1832 | if (test_bit(CLOSED, &con->state) || | |
1833 | test_bit(OPENING, &con->state)) { | |
1834 | ret = -EAGAIN; | |
1835 | goto out; | |
1836 | } | |
1837 | ||
31b8006e | 1838 | if (test_bit(CONNECTING, &con->state)) { |
eed0ef2c SW |
1839 | if (!test_bit(NEGOTIATING, &con->state)) { |
1840 | dout("try_read connecting\n"); | |
1841 | ret = read_partial_banner(con); | |
1842 | if (ret <= 0) | |
eed0ef2c | 1843 | goto out; |
98bdb0aa SW |
1844 | ret = process_banner(con); |
1845 | if (ret < 0) | |
1846 | goto out; | |
eed0ef2c | 1847 | } |
31b8006e SW |
1848 | ret = read_partial_connect(con); |
1849 | if (ret <= 0) | |
31b8006e | 1850 | goto out; |
98bdb0aa SW |
1851 | ret = process_connect(con); |
1852 | if (ret < 0) | |
1853 | goto out; | |
31b8006e SW |
1854 | goto more; |
1855 | } | |
1856 | ||
1857 | if (con->in_base_pos < 0) { | |
1858 | /* | |
1859 | * skipping + discarding content. | |
1860 | * | |
1861 | * FIXME: there must be a better way to do this! | |
1862 | */ | |
1863 | static char buf[1024]; | |
1864 | int skip = min(1024, -con->in_base_pos); | |
1865 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); | |
1866 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
1867 | if (ret <= 0) | |
98bdb0aa | 1868 | goto out; |
31b8006e SW |
1869 | con->in_base_pos += ret; |
1870 | if (con->in_base_pos) | |
1871 | goto more; | |
1872 | } | |
1873 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
1874 | /* | |
1875 | * what's next? | |
1876 | */ | |
1877 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
1878 | if (ret <= 0) | |
98bdb0aa | 1879 | goto out; |
31b8006e SW |
1880 | dout("try_read got tag %d\n", (int)con->in_tag); |
1881 | switch (con->in_tag) { | |
1882 | case CEPH_MSGR_TAG_MSG: | |
1883 | prepare_read_message(con); | |
1884 | break; | |
1885 | case CEPH_MSGR_TAG_ACK: | |
1886 | prepare_read_ack(con); | |
1887 | break; | |
1888 | case CEPH_MSGR_TAG_CLOSE: | |
1889 | set_bit(CLOSED, &con->state); /* fixme */ | |
98bdb0aa | 1890 | goto out; |
31b8006e SW |
1891 | default: |
1892 | goto bad_tag; | |
1893 | } | |
1894 | } | |
1895 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
1896 | ret = read_partial_message(con); | |
1897 | if (ret <= 0) { | |
1898 | switch (ret) { | |
1899 | case -EBADMSG: | |
1900 | con->error_msg = "bad crc"; | |
1901 | ret = -EIO; | |
98bdb0aa | 1902 | break; |
31b8006e SW |
1903 | case -EIO: |
1904 | con->error_msg = "io error"; | |
98bdb0aa | 1905 | break; |
31b8006e | 1906 | } |
98bdb0aa | 1907 | goto out; |
31b8006e SW |
1908 | } |
1909 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
1910 | goto more; | |
1911 | process_message(con); | |
1912 | goto more; | |
1913 | } | |
1914 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
1915 | ret = read_partial_ack(con); | |
1916 | if (ret <= 0) | |
98bdb0aa | 1917 | goto out; |
31b8006e SW |
1918 | process_ack(con); |
1919 | goto more; | |
1920 | } | |
1921 | ||
31b8006e | 1922 | out: |
98bdb0aa | 1923 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
1924 | return ret; |
1925 | ||
1926 | bad_tag: | |
1927 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
1928 | con->error_msg = "protocol error, garbage tag"; | |
1929 | ret = -1; | |
1930 | goto out; | |
1931 | } | |
1932 | ||
1933 | ||
1934 | /* | |
1935 | * Atomically queue work on a connection. Bump @con reference to | |
1936 | * avoid races with connection teardown. | |
31b8006e SW |
1937 | */ |
1938 | static void queue_con(struct ceph_connection *con) | |
1939 | { | |
1940 | if (test_bit(DEAD, &con->state)) { | |
1941 | dout("queue_con %p ignoring: DEAD\n", | |
1942 | con); | |
1943 | return; | |
1944 | } | |
1945 | ||
1946 | if (!con->ops->get(con)) { | |
1947 | dout("queue_con %p ref count 0\n", con); | |
1948 | return; | |
1949 | } | |
1950 | ||
f363e45f | 1951 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
1952 | dout("queue_con %p - already queued\n", con); |
1953 | con->ops->put(con); | |
1954 | } else { | |
1955 | dout("queue_con %p\n", con); | |
1956 | } | |
1957 | } | |
1958 | ||
1959 | /* | |
1960 | * Do some work on a connection. Drop a connection ref when we're done. | |
1961 | */ | |
1962 | static void con_work(struct work_struct *work) | |
1963 | { | |
1964 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
1965 | work.work); | |
0da5d703 | 1966 | int ret; |
31b8006e | 1967 | |
9dd4658d | 1968 | mutex_lock(&con->mutex); |
0da5d703 | 1969 | restart: |
60bf8bf8 SW |
1970 | if (test_and_clear_bit(BACKOFF, &con->state)) { |
1971 | dout("con_work %p backing off\n", con); | |
1972 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
1973 | round_jiffies_relative(con->delay))) { | |
1974 | dout("con_work %p backoff %lu\n", con, con->delay); | |
1975 | mutex_unlock(&con->mutex); | |
1976 | return; | |
1977 | } else { | |
1978 | con->ops->put(con); | |
1979 | dout("con_work %p FAILED to back off %lu\n", con, | |
1980 | con->delay); | |
1981 | } | |
1982 | } | |
9dd4658d | 1983 | |
e00de341 SW |
1984 | if (test_bit(STANDBY, &con->state)) { |
1985 | dout("con_work %p STANDBY\n", con); | |
1986 | goto done; | |
1987 | } | |
31b8006e SW |
1988 | if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */ |
1989 | dout("con_work CLOSED\n"); | |
1990 | con_close_socket(con); | |
1991 | goto done; | |
1992 | } | |
1993 | if (test_and_clear_bit(OPENING, &con->state)) { | |
1994 | /* reopen w/ new peer */ | |
1995 | dout("con_work OPENING\n"); | |
1996 | con_close_socket(con); | |
1997 | } | |
1998 | ||
0da5d703 SW |
1999 | if (test_and_clear_bit(SOCK_CLOSED, &con->state)) |
2000 | goto fault; | |
2001 | ||
2002 | ret = try_read(con); | |
2003 | if (ret == -EAGAIN) | |
2004 | goto restart; | |
2005 | if (ret < 0) | |
2006 | goto fault; | |
2007 | ||
2008 | ret = try_write(con); | |
2009 | if (ret == -EAGAIN) | |
2010 | goto restart; | |
2011 | if (ret < 0) | |
2012 | goto fault; | |
31b8006e SW |
2013 | |
2014 | done: | |
9dd4658d | 2015 | mutex_unlock(&con->mutex); |
9dd4658d | 2016 | done_unlocked: |
31b8006e | 2017 | con->ops->put(con); |
0da5d703 SW |
2018 | return; |
2019 | ||
2020 | fault: | |
2021 | mutex_unlock(&con->mutex); | |
2022 | ceph_fault(con); /* error/fault path */ | |
2023 | goto done_unlocked; | |
31b8006e SW |
2024 | } |
2025 | ||
2026 | ||
2027 | /* | |
2028 | * Generic error/fault handler. A retry mechanism is used with | |
2029 | * exponential backoff | |
2030 | */ | |
2031 | static void ceph_fault(struct ceph_connection *con) | |
2032 | { | |
2033 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2034 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2035 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2036 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
2037 | |
2038 | if (test_bit(LOSSYTX, &con->state)) { | |
2039 | dout("fault on LOSSYTX channel\n"); | |
2040 | goto out; | |
2041 | } | |
2042 | ||
ec302645 | 2043 | mutex_lock(&con->mutex); |
91e45ce3 SW |
2044 | if (test_bit(CLOSED, &con->state)) |
2045 | goto out_unlock; | |
ec302645 | 2046 | |
31b8006e | 2047 | con_close_socket(con); |
5e095e8b SW |
2048 | |
2049 | if (con->in_msg) { | |
2050 | ceph_msg_put(con->in_msg); | |
2051 | con->in_msg = NULL; | |
2052 | } | |
31b8006e | 2053 | |
e80a52d1 SW |
2054 | /* Requeue anything that hasn't been acked */ |
2055 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2056 | |
e76661d0 SW |
2057 | /* If there are no messages queued or keepalive pending, place |
2058 | * the connection in a STANDBY state */ | |
2059 | if (list_empty(&con->out_queue) && | |
2060 | !test_bit(KEEPALIVE_PENDING, &con->state)) { | |
e00de341 SW |
2061 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
2062 | clear_bit(WRITE_PENDING, &con->state); | |
31b8006e | 2063 | set_bit(STANDBY, &con->state); |
e80a52d1 SW |
2064 | } else { |
2065 | /* retry after a delay. */ | |
2066 | if (con->delay == 0) | |
2067 | con->delay = BASE_DELAY_INTERVAL; | |
2068 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2069 | con->delay *= 2; | |
e80a52d1 SW |
2070 | con->ops->get(con); |
2071 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2072 | round_jiffies_relative(con->delay))) { |
2073 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2074 | } else { | |
e80a52d1 | 2075 | con->ops->put(con); |
60bf8bf8 SW |
2076 | dout("fault failed to queue %p delay %lu, backoff\n", |
2077 | con, con->delay); | |
2078 | /* | |
2079 | * In many cases we see a socket state change | |
2080 | * while con_work is running and end up | |
2081 | * queuing (non-delayed) work, such that we | |
2082 | * can't backoff with a delay. Set a flag so | |
2083 | * that when con_work restarts we schedule the | |
2084 | * delay then. | |
2085 | */ | |
2086 | set_bit(BACKOFF, &con->state); | |
2087 | } | |
31b8006e SW |
2088 | } |
2089 | ||
91e45ce3 SW |
2090 | out_unlock: |
2091 | mutex_unlock(&con->mutex); | |
31b8006e | 2092 | out: |
161fd65a SW |
2093 | /* |
2094 | * in case we faulted due to authentication, invalidate our | |
2095 | * current tickets so that we can get new ones. | |
213c99ee | 2096 | */ |
161fd65a SW |
2097 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2098 | dout("calling invalidate_authorizer()\n"); | |
2099 | con->ops->invalidate_authorizer(con); | |
2100 | } | |
2101 | ||
31b8006e SW |
2102 | if (con->ops->fault) |
2103 | con->ops->fault(con); | |
2104 | } | |
2105 | ||
2106 | ||
2107 | ||
2108 | /* | |
2109 | * create a new messenger instance | |
2110 | */ | |
3d14c5d2 YS |
2111 | struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr, |
2112 | u32 supported_features, | |
2113 | u32 required_features) | |
31b8006e SW |
2114 | { |
2115 | struct ceph_messenger *msgr; | |
2116 | ||
2117 | msgr = kzalloc(sizeof(*msgr), GFP_KERNEL); | |
2118 | if (msgr == NULL) | |
2119 | return ERR_PTR(-ENOMEM); | |
2120 | ||
3d14c5d2 YS |
2121 | msgr->supported_features = supported_features; |
2122 | msgr->required_features = required_features; | |
2123 | ||
31b8006e SW |
2124 | spin_lock_init(&msgr->global_seq_lock); |
2125 | ||
2126 | /* the zero page is needed if a request is "canceled" while the message | |
2127 | * is being written over the socket */ | |
31459fe4 | 2128 | msgr->zero_page = __page_cache_alloc(GFP_KERNEL | __GFP_ZERO); |
31b8006e SW |
2129 | if (!msgr->zero_page) { |
2130 | kfree(msgr); | |
2131 | return ERR_PTR(-ENOMEM); | |
2132 | } | |
2133 | kmap(msgr->zero_page); | |
2134 | ||
2135 | if (myaddr) | |
2136 | msgr->inst.addr = *myaddr; | |
2137 | ||
2138 | /* select a random nonce */ | |
ac8839d7 | 2139 | msgr->inst.addr.type = 0; |
103e2d3a | 2140 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2141 | encode_my_addr(msgr); |
31b8006e SW |
2142 | |
2143 | dout("messenger_create %p\n", msgr); | |
2144 | return msgr; | |
2145 | } | |
3d14c5d2 | 2146 | EXPORT_SYMBOL(ceph_messenger_create); |
31b8006e SW |
2147 | |
2148 | void ceph_messenger_destroy(struct ceph_messenger *msgr) | |
2149 | { | |
2150 | dout("destroy %p\n", msgr); | |
2151 | kunmap(msgr->zero_page); | |
2152 | __free_page(msgr->zero_page); | |
2153 | kfree(msgr); | |
2154 | dout("destroyed messenger %p\n", msgr); | |
2155 | } | |
3d14c5d2 | 2156 | EXPORT_SYMBOL(ceph_messenger_destroy); |
31b8006e | 2157 | |
e00de341 SW |
2158 | static void clear_standby(struct ceph_connection *con) |
2159 | { | |
2160 | /* come back from STANDBY? */ | |
2161 | if (test_and_clear_bit(STANDBY, &con->state)) { | |
2162 | mutex_lock(&con->mutex); | |
2163 | dout("clear_standby %p and ++connect_seq\n", con); | |
2164 | con->connect_seq++; | |
2165 | WARN_ON(test_bit(WRITE_PENDING, &con->state)); | |
2166 | WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state)); | |
2167 | mutex_unlock(&con->mutex); | |
2168 | } | |
2169 | } | |
2170 | ||
31b8006e SW |
2171 | /* |
2172 | * Queue up an outgoing message on the given connection. | |
2173 | */ | |
2174 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2175 | { | |
2176 | if (test_bit(CLOSED, &con->state)) { | |
2177 | dout("con_send %p closed, dropping %p\n", con, msg); | |
2178 | ceph_msg_put(msg); | |
2179 | return; | |
2180 | } | |
2181 | ||
2182 | /* set src+dst */ | |
dbad185d | 2183 | msg->hdr.src = con->msgr->inst.name; |
31b8006e | 2184 | |
3ca02ef9 SW |
2185 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
2186 | ||
e84346b7 SW |
2187 | msg->needs_out_seq = true; |
2188 | ||
31b8006e | 2189 | /* queue */ |
ec302645 | 2190 | mutex_lock(&con->mutex); |
31b8006e SW |
2191 | BUG_ON(!list_empty(&msg->list_head)); |
2192 | list_add_tail(&msg->list_head, &con->out_queue); | |
2193 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2194 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2195 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2196 | le32_to_cpu(msg->hdr.front_len), | |
2197 | le32_to_cpu(msg->hdr.middle_len), | |
2198 | le32_to_cpu(msg->hdr.data_len)); | |
ec302645 | 2199 | mutex_unlock(&con->mutex); |
31b8006e SW |
2200 | |
2201 | /* if there wasn't anything waiting to send before, queue | |
2202 | * new work */ | |
e00de341 | 2203 | clear_standby(con); |
31b8006e SW |
2204 | if (test_and_set_bit(WRITE_PENDING, &con->state) == 0) |
2205 | queue_con(con); | |
2206 | } | |
3d14c5d2 | 2207 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2208 | |
2209 | /* | |
2210 | * Revoke a message that was previously queued for send | |
2211 | */ | |
2212 | void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg) | |
2213 | { | |
ec302645 | 2214 | mutex_lock(&con->mutex); |
31b8006e | 2215 | if (!list_empty(&msg->list_head)) { |
ed98adad | 2216 | dout("con_revoke %p msg %p - was on queue\n", con, msg); |
31b8006e SW |
2217 | list_del_init(&msg->list_head); |
2218 | ceph_msg_put(msg); | |
2219 | msg->hdr.seq = 0; | |
ed98adad SW |
2220 | } |
2221 | if (con->out_msg == msg) { | |
2222 | dout("con_revoke %p msg %p - was sending\n", con, msg); | |
2223 | con->out_msg = NULL; | |
31b8006e SW |
2224 | if (con->out_kvec_is_msg) { |
2225 | con->out_skip = con->out_kvec_bytes; | |
2226 | con->out_kvec_is_msg = false; | |
2227 | } | |
ed98adad SW |
2228 | ceph_msg_put(msg); |
2229 | msg->hdr.seq = 0; | |
31b8006e | 2230 | } |
ec302645 | 2231 | mutex_unlock(&con->mutex); |
31b8006e SW |
2232 | } |
2233 | ||
350b1c32 | 2234 | /* |
0d59ab81 | 2235 | * Revoke a message that we may be reading data into |
350b1c32 | 2236 | */ |
0d59ab81 | 2237 | void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg) |
350b1c32 SW |
2238 | { |
2239 | mutex_lock(&con->mutex); | |
0d59ab81 YS |
2240 | if (con->in_msg && con->in_msg == msg) { |
2241 | unsigned front_len = le32_to_cpu(con->in_hdr.front_len); | |
2242 | unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
350b1c32 SW |
2243 | unsigned data_len = le32_to_cpu(con->in_hdr.data_len); |
2244 | ||
2245 | /* skip rest of message */ | |
0d59ab81 | 2246 | dout("con_revoke_pages %p msg %p revoked\n", con, msg); |
350b1c32 SW |
2247 | con->in_base_pos = con->in_base_pos - |
2248 | sizeof(struct ceph_msg_header) - | |
0d59ab81 YS |
2249 | front_len - |
2250 | middle_len - | |
2251 | data_len - | |
350b1c32 | 2252 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2253 | ceph_msg_put(con->in_msg); |
2254 | con->in_msg = NULL; | |
2255 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2256 | con->in_seq++; |
350b1c32 SW |
2257 | } else { |
2258 | dout("con_revoke_pages %p msg %p pages %p no-op\n", | |
0d59ab81 | 2259 | con, con->in_msg, msg); |
350b1c32 SW |
2260 | } |
2261 | mutex_unlock(&con->mutex); | |
2262 | } | |
2263 | ||
31b8006e SW |
2264 | /* |
2265 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2266 | */ | |
2267 | void ceph_con_keepalive(struct ceph_connection *con) | |
2268 | { | |
e00de341 SW |
2269 | dout("con_keepalive %p\n", con); |
2270 | clear_standby(con); | |
31b8006e SW |
2271 | if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 && |
2272 | test_and_set_bit(WRITE_PENDING, &con->state) == 0) | |
2273 | queue_con(con); | |
2274 | } | |
3d14c5d2 | 2275 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2276 | |
2277 | ||
2278 | /* | |
2279 | * construct a new message with given type, size | |
2280 | * the new msg has a ref count of 1. | |
2281 | */ | |
34d23762 | 2282 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags) |
31b8006e SW |
2283 | { |
2284 | struct ceph_msg *m; | |
2285 | ||
34d23762 | 2286 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2287 | if (m == NULL) |
2288 | goto out; | |
c2e552e7 | 2289 | kref_init(&m->kref); |
31b8006e SW |
2290 | INIT_LIST_HEAD(&m->list_head); |
2291 | ||
45c6ceb5 | 2292 | m->hdr.tid = 0; |
31b8006e | 2293 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2294 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2295 | m->hdr.version = 0; | |
31b8006e SW |
2296 | m->hdr.front_len = cpu_to_le32(front_len); |
2297 | m->hdr.middle_len = 0; | |
bb257664 SW |
2298 | m->hdr.data_len = 0; |
2299 | m->hdr.data_off = 0; | |
45c6ceb5 | 2300 | m->hdr.reserved = 0; |
31b8006e SW |
2301 | m->footer.front_crc = 0; |
2302 | m->footer.middle_crc = 0; | |
2303 | m->footer.data_crc = 0; | |
45c6ceb5 | 2304 | m->footer.flags = 0; |
31b8006e SW |
2305 | m->front_max = front_len; |
2306 | m->front_is_vmalloc = false; | |
2307 | m->more_to_follow = false; | |
2308 | m->pool = NULL; | |
2309 | ||
ca20892d HC |
2310 | /* middle */ |
2311 | m->middle = NULL; | |
2312 | ||
2313 | /* data */ | |
2314 | m->nr_pages = 0; | |
2315 | m->page_alignment = 0; | |
2316 | m->pages = NULL; | |
2317 | m->pagelist = NULL; | |
2318 | m->bio = NULL; | |
2319 | m->bio_iter = NULL; | |
2320 | m->bio_seg = 0; | |
2321 | m->trail = NULL; | |
2322 | ||
31b8006e SW |
2323 | /* front */ |
2324 | if (front_len) { | |
2325 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2326 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2327 | PAGE_KERNEL); |
2328 | m->front_is_vmalloc = true; | |
2329 | } else { | |
34d23762 | 2330 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2331 | } |
2332 | if (m->front.iov_base == NULL) { | |
2333 | pr_err("msg_new can't allocate %d bytes\n", | |
2334 | front_len); | |
2335 | goto out2; | |
2336 | } | |
2337 | } else { | |
2338 | m->front.iov_base = NULL; | |
2339 | } | |
2340 | m->front.iov_len = front_len; | |
2341 | ||
bb257664 | 2342 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2343 | return m; |
2344 | ||
2345 | out2: | |
2346 | ceph_msg_put(m); | |
2347 | out: | |
bb257664 | 2348 | pr_err("msg_new can't create type %d front %d\n", type, front_len); |
a79832f2 | 2349 | return NULL; |
31b8006e | 2350 | } |
3d14c5d2 | 2351 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2352 | |
31b8006e SW |
2353 | /* |
2354 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2355 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2356 | * per-type header in the front and then gracefully fail (i.e., | |
2357 | * propagate the error to the caller based on info in the front) when | |
2358 | * the middle is too large. | |
2359 | */ | |
2450418c | 2360 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2361 | { |
2362 | int type = le16_to_cpu(msg->hdr.type); | |
2363 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2364 | ||
2365 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2366 | ceph_msg_type_name(type), middle_len); | |
2367 | BUG_ON(!middle_len); | |
2368 | BUG_ON(msg->middle); | |
2369 | ||
b6c1d5b8 | 2370 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2371 | if (!msg->middle) |
2372 | return -ENOMEM; | |
2373 | return 0; | |
2374 | } | |
2375 | ||
2450418c YS |
2376 | /* |
2377 | * Generic message allocator, for incoming messages. | |
2378 | */ | |
2379 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, | |
2380 | struct ceph_msg_header *hdr, | |
2381 | int *skip) | |
2382 | { | |
2383 | int type = le16_to_cpu(hdr->type); | |
2384 | int front_len = le32_to_cpu(hdr->front_len); | |
2385 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2386 | struct ceph_msg *msg = NULL; | |
2387 | int ret; | |
2388 | ||
2389 | if (con->ops->alloc_msg) { | |
0547a9b3 | 2390 | mutex_unlock(&con->mutex); |
2450418c | 2391 | msg = con->ops->alloc_msg(con, hdr, skip); |
0547a9b3 | 2392 | mutex_lock(&con->mutex); |
a79832f2 | 2393 | if (!msg || *skip) |
2450418c YS |
2394 | return NULL; |
2395 | } | |
2396 | if (!msg) { | |
2397 | *skip = 0; | |
34d23762 | 2398 | msg = ceph_msg_new(type, front_len, GFP_NOFS); |
2450418c YS |
2399 | if (!msg) { |
2400 | pr_err("unable to allocate msg type %d len %d\n", | |
2401 | type, front_len); | |
a79832f2 | 2402 | return NULL; |
2450418c | 2403 | } |
c5c6b19d | 2404 | msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2405 | } |
9d7f0f13 | 2406 | memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2407 | |
bb257664 | 2408 | if (middle_len && !msg->middle) { |
2450418c | 2409 | ret = ceph_alloc_middle(con, msg); |
2450418c YS |
2410 | if (ret < 0) { |
2411 | ceph_msg_put(msg); | |
a79832f2 | 2412 | return NULL; |
2450418c YS |
2413 | } |
2414 | } | |
9d7f0f13 | 2415 | |
2450418c YS |
2416 | return msg; |
2417 | } | |
2418 | ||
31b8006e SW |
2419 | |
2420 | /* | |
2421 | * Free a generically kmalloc'd message. | |
2422 | */ | |
2423 | void ceph_msg_kfree(struct ceph_msg *m) | |
2424 | { | |
2425 | dout("msg_kfree %p\n", m); | |
2426 | if (m->front_is_vmalloc) | |
2427 | vfree(m->front.iov_base); | |
2428 | else | |
2429 | kfree(m->front.iov_base); | |
2430 | kfree(m); | |
2431 | } | |
2432 | ||
2433 | /* | |
2434 | * Drop a msg ref. Destroy as needed. | |
2435 | */ | |
c2e552e7 SW |
2436 | void ceph_msg_last_put(struct kref *kref) |
2437 | { | |
2438 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2439 | |
c2e552e7 SW |
2440 | dout("ceph_msg_put last one on %p\n", m); |
2441 | WARN_ON(!list_empty(&m->list_head)); | |
2442 | ||
2443 | /* drop middle, data, if any */ | |
2444 | if (m->middle) { | |
2445 | ceph_buffer_put(m->middle); | |
2446 | m->middle = NULL; | |
31b8006e | 2447 | } |
c2e552e7 SW |
2448 | m->nr_pages = 0; |
2449 | m->pages = NULL; | |
2450 | ||
58bb3b37 SW |
2451 | if (m->pagelist) { |
2452 | ceph_pagelist_release(m->pagelist); | |
2453 | kfree(m->pagelist); | |
2454 | m->pagelist = NULL; | |
2455 | } | |
2456 | ||
68b4476b YS |
2457 | m->trail = NULL; |
2458 | ||
c2e552e7 SW |
2459 | if (m->pool) |
2460 | ceph_msgpool_put(m->pool, m); | |
2461 | else | |
2462 | ceph_msg_kfree(m); | |
31b8006e | 2463 | } |
3d14c5d2 | 2464 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2465 | |
2466 | void ceph_msg_dump(struct ceph_msg *msg) | |
2467 | { | |
2468 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2469 | msg->front_max, msg->nr_pages); | |
2470 | print_hex_dump(KERN_DEBUG, "header: ", | |
2471 | DUMP_PREFIX_OFFSET, 16, 1, | |
2472 | &msg->hdr, sizeof(msg->hdr), true); | |
2473 | print_hex_dump(KERN_DEBUG, " front: ", | |
2474 | DUMP_PREFIX_OFFSET, 16, 1, | |
2475 | msg->front.iov_base, msg->front.iov_len, true); | |
2476 | if (msg->middle) | |
2477 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2478 | DUMP_PREFIX_OFFSET, 16, 1, | |
2479 | msg->middle->vec.iov_base, | |
2480 | msg->middle->vec.iov_len, true); | |
2481 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2482 | DUMP_PREFIX_OFFSET, 16, 1, | |
2483 | &msg->footer, sizeof(msg->footer), true); | |
2484 | } | |
3d14c5d2 | 2485 | EXPORT_SYMBOL(ceph_msg_dump); |