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 | ||
32 | /* static tag bytes (protocol control messages) */ | |
33 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
34 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
35 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
36 | ||
a6a5349d SW |
37 | #ifdef CONFIG_LOCKDEP |
38 | static struct lock_class_key socket_class; | |
39 | #endif | |
40 | ||
31b8006e SW |
41 | |
42 | static void queue_con(struct ceph_connection *con); | |
43 | static void con_work(struct work_struct *); | |
44 | static void ceph_fault(struct ceph_connection *con); | |
45 | ||
31b8006e | 46 | /* |
f64a9317 AE |
47 | * Nicely render a sockaddr as a string. An array of formatted |
48 | * strings is used, to approximate reentrancy. | |
31b8006e | 49 | */ |
f64a9317 AE |
50 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
51 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
52 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
53 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
54 | ||
55 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
56 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 57 | |
57666519 AE |
58 | static struct page *zero_page; /* used in certain error cases */ |
59 | static void *zero_page_address; /* kernel virtual addr of zero_page */ | |
60 | ||
3d14c5d2 | 61 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
62 | { |
63 | int i; | |
64 | char *s; | |
99f0f3b2 AE |
65 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
66 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 67 | |
f64a9317 | 68 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
69 | s = addr_str[i]; |
70 | ||
71 | switch (ss->ss_family) { | |
72 | case AF_INET: | |
bd406145 AE |
73 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
74 | ntohs(in4->sin_port)); | |
31b8006e SW |
75 | break; |
76 | ||
77 | case AF_INET6: | |
bd406145 AE |
78 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
79 | ntohs(in6->sin6_port)); | |
31b8006e SW |
80 | break; |
81 | ||
82 | default: | |
12a2f643 SW |
83 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %d)", |
84 | (int)ss->ss_family); | |
31b8006e SW |
85 | } |
86 | ||
87 | return s; | |
88 | } | |
3d14c5d2 | 89 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 90 | |
63f2d211 SW |
91 | static void encode_my_addr(struct ceph_messenger *msgr) |
92 | { | |
93 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
94 | ceph_encode_addr(&msgr->my_enc_addr); | |
95 | } | |
96 | ||
31b8006e SW |
97 | /* |
98 | * work queue for all reading and writing to/from the socket. | |
99 | */ | |
100 | struct workqueue_struct *ceph_msgr_wq; | |
101 | ||
3d14c5d2 | 102 | int ceph_msgr_init(void) |
31b8006e | 103 | { |
57666519 AE |
104 | BUG_ON(zero_page != NULL); |
105 | zero_page = ZERO_PAGE(0); | |
106 | page_cache_get(zero_page); | |
107 | ||
108 | BUG_ON(zero_page_address != NULL); | |
109 | zero_page_address = kmap(zero_page); | |
110 | ||
f363e45f | 111 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0); |
d96c9043 SW |
112 | if (!ceph_msgr_wq) { |
113 | pr_err("msgr_init failed to create workqueue\n"); | |
57666519 AE |
114 | |
115 | zero_page_address = NULL; | |
116 | kunmap(zero_page); | |
117 | page_cache_release(zero_page); | |
118 | zero_page = NULL; | |
119 | ||
d96c9043 | 120 | return -ENOMEM; |
31b8006e | 121 | } |
57666519 | 122 | |
31b8006e SW |
123 | return 0; |
124 | } | |
3d14c5d2 | 125 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
126 | |
127 | void ceph_msgr_exit(void) | |
128 | { | |
57666519 | 129 | BUG_ON(ceph_msgr_wq == NULL); |
31b8006e | 130 | destroy_workqueue(ceph_msgr_wq); |
57666519 AE |
131 | |
132 | BUG_ON(zero_page_address == NULL); | |
133 | zero_page_address = NULL; | |
134 | ||
135 | BUG_ON(zero_page == NULL); | |
136 | kunmap(zero_page); | |
137 | page_cache_release(zero_page); | |
138 | zero_page = NULL; | |
31b8006e | 139 | } |
3d14c5d2 | 140 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 141 | |
cd84db6e | 142 | void ceph_msgr_flush(void) |
a922d38f SW |
143 | { |
144 | flush_workqueue(ceph_msgr_wq); | |
145 | } | |
3d14c5d2 | 146 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f SW |
147 | |
148 | ||
31b8006e SW |
149 | /* |
150 | * socket callback functions | |
151 | */ | |
152 | ||
153 | /* data available on socket, or listen socket received a connect */ | |
154 | static void ceph_data_ready(struct sock *sk, int count_unused) | |
155 | { | |
bd406145 AE |
156 | struct ceph_connection *con = sk->sk_user_data; |
157 | ||
31b8006e SW |
158 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
159 | dout("ceph_data_ready on %p state = %lu, queueing work\n", | |
160 | con, con->state); | |
161 | queue_con(con); | |
162 | } | |
163 | } | |
164 | ||
165 | /* socket has buffer space for writing */ | |
166 | static void ceph_write_space(struct sock *sk) | |
167 | { | |
168 | struct ceph_connection *con = | |
169 | (struct ceph_connection *)sk->sk_user_data; | |
170 | ||
182fac26 JS |
171 | /* only queue to workqueue if there is data we want to write, |
172 | * and there is sufficient space in the socket buffer to accept | |
173 | * more data. clear SOCK_NOSPACE so that ceph_write_space() | |
174 | * doesn't get called again until try_write() fills the socket | |
175 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
176 | * and net/core/stream.c:sk_stream_write_space(). | |
177 | */ | |
31b8006e | 178 | if (test_bit(WRITE_PENDING, &con->state)) { |
182fac26 JS |
179 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
180 | dout("ceph_write_space %p queueing write work\n", con); | |
181 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
182 | queue_con(con); | |
183 | } | |
31b8006e SW |
184 | } else { |
185 | dout("ceph_write_space %p nothing to write\n", con); | |
186 | } | |
31b8006e SW |
187 | } |
188 | ||
189 | /* socket's state has changed */ | |
190 | static void ceph_state_change(struct sock *sk) | |
191 | { | |
bd406145 | 192 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e SW |
193 | |
194 | dout("ceph_state_change %p state = %lu sk_state = %u\n", | |
195 | con, con->state, sk->sk_state); | |
196 | ||
197 | if (test_bit(CLOSED, &con->state)) | |
198 | return; | |
199 | ||
200 | switch (sk->sk_state) { | |
201 | case TCP_CLOSE: | |
202 | dout("ceph_state_change TCP_CLOSE\n"); | |
203 | case TCP_CLOSE_WAIT: | |
204 | dout("ceph_state_change TCP_CLOSE_WAIT\n"); | |
205 | if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) { | |
206 | if (test_bit(CONNECTING, &con->state)) | |
207 | con->error_msg = "connection failed"; | |
208 | else | |
209 | con->error_msg = "socket closed"; | |
210 | queue_con(con); | |
211 | } | |
212 | break; | |
213 | case TCP_ESTABLISHED: | |
214 | dout("ceph_state_change TCP_ESTABLISHED\n"); | |
215 | queue_con(con); | |
216 | break; | |
217 | } | |
218 | } | |
219 | ||
220 | /* | |
221 | * set up socket callbacks | |
222 | */ | |
223 | static void set_sock_callbacks(struct socket *sock, | |
224 | struct ceph_connection *con) | |
225 | { | |
226 | struct sock *sk = sock->sk; | |
bd406145 | 227 | sk->sk_user_data = con; |
31b8006e SW |
228 | sk->sk_data_ready = ceph_data_ready; |
229 | sk->sk_write_space = ceph_write_space; | |
230 | sk->sk_state_change = ceph_state_change; | |
231 | } | |
232 | ||
233 | ||
234 | /* | |
235 | * socket helpers | |
236 | */ | |
237 | ||
238 | /* | |
239 | * initiate connection to a remote socket. | |
240 | */ | |
241 | static struct socket *ceph_tcp_connect(struct ceph_connection *con) | |
242 | { | |
f91d3471 | 243 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
244 | struct socket *sock; |
245 | int ret; | |
246 | ||
247 | BUG_ON(con->sock); | |
f91d3471 SW |
248 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
249 | IPPROTO_TCP, &sock); | |
31b8006e SW |
250 | if (ret) |
251 | return ERR_PTR(ret); | |
31b8006e SW |
252 | sock->sk->sk_allocation = GFP_NOFS; |
253 | ||
a6a5349d SW |
254 | #ifdef CONFIG_LOCKDEP |
255 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
256 | #endif | |
257 | ||
31b8006e SW |
258 | set_sock_callbacks(sock, con); |
259 | ||
3d14c5d2 | 260 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 261 | |
f91d3471 SW |
262 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
263 | O_NONBLOCK); | |
31b8006e SW |
264 | if (ret == -EINPROGRESS) { |
265 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 266 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 267 | sock->sk->sk_state); |
a5bc3129 | 268 | } else if (ret < 0) { |
31b8006e | 269 | pr_err("connect %s error %d\n", |
3d14c5d2 | 270 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 271 | sock_release(sock); |
31b8006e | 272 | con->error_msg = "connect error"; |
31b8006e | 273 | |
31b8006e | 274 | return ERR_PTR(ret); |
a5bc3129 AE |
275 | } |
276 | con->sock = sock; | |
277 | ||
31b8006e SW |
278 | return sock; |
279 | } | |
280 | ||
281 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
282 | { | |
283 | struct kvec iov = {buf, len}; | |
284 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 285 | int r; |
31b8006e | 286 | |
98bdb0aa SW |
287 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
288 | if (r == -EAGAIN) | |
289 | r = 0; | |
290 | return r; | |
31b8006e SW |
291 | } |
292 | ||
293 | /* | |
294 | * write something. @more is true if caller will be sending more data | |
295 | * shortly. | |
296 | */ | |
297 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
298 | size_t kvlen, size_t len, int more) | |
299 | { | |
300 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 301 | int r; |
31b8006e SW |
302 | |
303 | if (more) | |
304 | msg.msg_flags |= MSG_MORE; | |
305 | else | |
306 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
307 | ||
42961d23 SW |
308 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
309 | if (r == -EAGAIN) | |
310 | r = 0; | |
311 | return r; | |
31b8006e SW |
312 | } |
313 | ||
314 | ||
315 | /* | |
316 | * Shutdown/close the socket for the given connection. | |
317 | */ | |
318 | static int con_close_socket(struct ceph_connection *con) | |
319 | { | |
320 | int rc; | |
321 | ||
322 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
323 | if (!con->sock) | |
324 | return 0; | |
325 | set_bit(SOCK_CLOSED, &con->state); | |
326 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
327 | sock_release(con->sock); | |
328 | con->sock = NULL; | |
329 | clear_bit(SOCK_CLOSED, &con->state); | |
330 | return rc; | |
331 | } | |
332 | ||
333 | /* | |
334 | * Reset a connection. Discard all incoming and outgoing messages | |
335 | * and clear *_seq state. | |
336 | */ | |
337 | static void ceph_msg_remove(struct ceph_msg *msg) | |
338 | { | |
339 | list_del_init(&msg->list_head); | |
340 | ceph_msg_put(msg); | |
341 | } | |
342 | static void ceph_msg_remove_list(struct list_head *head) | |
343 | { | |
344 | while (!list_empty(head)) { | |
345 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
346 | list_head); | |
347 | ceph_msg_remove(msg); | |
348 | } | |
349 | } | |
350 | ||
351 | static void reset_connection(struct ceph_connection *con) | |
352 | { | |
353 | /* reset connection, out_queue, msg_ and connect_seq */ | |
354 | /* discard existing out_queue and msg_seq */ | |
31b8006e SW |
355 | ceph_msg_remove_list(&con->out_queue); |
356 | ceph_msg_remove_list(&con->out_sent); | |
357 | ||
cf3e5c40 SW |
358 | if (con->in_msg) { |
359 | ceph_msg_put(con->in_msg); | |
360 | con->in_msg = NULL; | |
361 | } | |
362 | ||
31b8006e SW |
363 | con->connect_seq = 0; |
364 | con->out_seq = 0; | |
c86a2930 SW |
365 | if (con->out_msg) { |
366 | ceph_msg_put(con->out_msg); | |
367 | con->out_msg = NULL; | |
368 | } | |
31b8006e | 369 | con->in_seq = 0; |
0e0d5e0c | 370 | con->in_seq_acked = 0; |
31b8006e SW |
371 | } |
372 | ||
373 | /* | |
374 | * mark a peer down. drop any open connections. | |
375 | */ | |
376 | void ceph_con_close(struct ceph_connection *con) | |
377 | { | |
3d14c5d2 YS |
378 | dout("con_close %p peer %s\n", con, |
379 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
31b8006e SW |
380 | set_bit(CLOSED, &con->state); /* in case there's queued work */ |
381 | clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */ | |
1679f876 SW |
382 | clear_bit(LOSSYTX, &con->state); /* so we retry next connect */ |
383 | clear_bit(KEEPALIVE_PENDING, &con->state); | |
384 | clear_bit(WRITE_PENDING, &con->state); | |
ec302645 | 385 | mutex_lock(&con->mutex); |
31b8006e | 386 | reset_connection(con); |
6f2bc3ff | 387 | con->peer_global_seq = 0; |
91e45ce3 | 388 | cancel_delayed_work(&con->work); |
ec302645 | 389 | mutex_unlock(&con->mutex); |
31b8006e SW |
390 | queue_con(con); |
391 | } | |
3d14c5d2 | 392 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 393 | |
31b8006e SW |
394 | /* |
395 | * Reopen a closed connection, with a new peer address. | |
396 | */ | |
397 | void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr) | |
398 | { | |
3d14c5d2 | 399 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
31b8006e SW |
400 | set_bit(OPENING, &con->state); |
401 | clear_bit(CLOSED, &con->state); | |
402 | memcpy(&con->peer_addr, addr, sizeof(*addr)); | |
03c677e1 | 403 | con->delay = 0; /* reset backoff memory */ |
31b8006e SW |
404 | queue_con(con); |
405 | } | |
3d14c5d2 | 406 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 407 | |
87b315a5 SW |
408 | /* |
409 | * return true if this connection ever successfully opened | |
410 | */ | |
411 | bool ceph_con_opened(struct ceph_connection *con) | |
412 | { | |
413 | return con->connect_seq > 0; | |
414 | } | |
415 | ||
31b8006e SW |
416 | /* |
417 | * generic get/put | |
418 | */ | |
419 | struct ceph_connection *ceph_con_get(struct ceph_connection *con) | |
420 | { | |
421 | dout("con_get %p nref = %d -> %d\n", con, | |
422 | atomic_read(&con->nref), atomic_read(&con->nref) + 1); | |
423 | if (atomic_inc_not_zero(&con->nref)) | |
424 | return con; | |
425 | return NULL; | |
426 | } | |
427 | ||
428 | void ceph_con_put(struct ceph_connection *con) | |
429 | { | |
430 | dout("con_put %p nref = %d -> %d\n", con, | |
431 | atomic_read(&con->nref), atomic_read(&con->nref) - 1); | |
432 | BUG_ON(atomic_read(&con->nref) == 0); | |
433 | if (atomic_dec_and_test(&con->nref)) { | |
71ececda | 434 | BUG_ON(con->sock); |
31b8006e SW |
435 | kfree(con); |
436 | } | |
437 | } | |
438 | ||
439 | /* | |
440 | * initialize a new connection. | |
441 | */ | |
442 | void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con) | |
443 | { | |
444 | dout("con_init %p\n", con); | |
445 | memset(con, 0, sizeof(*con)); | |
446 | atomic_set(&con->nref, 1); | |
447 | con->msgr = msgr; | |
ec302645 | 448 | mutex_init(&con->mutex); |
31b8006e SW |
449 | INIT_LIST_HEAD(&con->out_queue); |
450 | INIT_LIST_HEAD(&con->out_sent); | |
451 | INIT_DELAYED_WORK(&con->work, con_work); | |
452 | } | |
3d14c5d2 | 453 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
454 | |
455 | ||
456 | /* | |
457 | * We maintain a global counter to order connection attempts. Get | |
458 | * a unique seq greater than @gt. | |
459 | */ | |
460 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
461 | { | |
462 | u32 ret; | |
463 | ||
464 | spin_lock(&msgr->global_seq_lock); | |
465 | if (msgr->global_seq < gt) | |
466 | msgr->global_seq = gt; | |
467 | ret = ++msgr->global_seq; | |
468 | spin_unlock(&msgr->global_seq_lock); | |
469 | return ret; | |
470 | } | |
471 | ||
859eb799 AE |
472 | static void ceph_con_out_kvec_reset(struct ceph_connection *con) |
473 | { | |
474 | con->out_kvec_left = 0; | |
475 | con->out_kvec_bytes = 0; | |
476 | con->out_kvec_cur = &con->out_kvec[0]; | |
477 | } | |
478 | ||
479 | static void ceph_con_out_kvec_add(struct ceph_connection *con, | |
480 | size_t size, void *data) | |
481 | { | |
482 | int index; | |
483 | ||
484 | index = con->out_kvec_left; | |
485 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
486 | ||
487 | con->out_kvec[index].iov_len = size; | |
488 | con->out_kvec[index].iov_base = data; | |
489 | con->out_kvec_left++; | |
490 | con->out_kvec_bytes += size; | |
491 | } | |
31b8006e SW |
492 | |
493 | /* | |
494 | * Prepare footer for currently outgoing message, and finish things | |
495 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
496 | */ | |
859eb799 | 497 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
498 | { |
499 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 500 | int v = con->out_kvec_left; |
31b8006e SW |
501 | |
502 | dout("prepare_write_message_footer %p\n", con); | |
503 | con->out_kvec_is_msg = true; | |
504 | con->out_kvec[v].iov_base = &m->footer; | |
505 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
506 | con->out_kvec_bytes += sizeof(m->footer); | |
507 | con->out_kvec_left++; | |
508 | con->out_more = m->more_to_follow; | |
c86a2930 | 509 | con->out_msg_done = true; |
31b8006e SW |
510 | } |
511 | ||
512 | /* | |
513 | * Prepare headers for the next outgoing message. | |
514 | */ | |
515 | static void prepare_write_message(struct ceph_connection *con) | |
516 | { | |
517 | struct ceph_msg *m; | |
31b8006e | 518 | |
859eb799 | 519 | ceph_con_out_kvec_reset(con); |
31b8006e | 520 | con->out_kvec_is_msg = true; |
c86a2930 | 521 | con->out_msg_done = false; |
31b8006e SW |
522 | |
523 | /* Sneak an ack in there first? If we can get it into the same | |
524 | * TCP packet that's a good thing. */ | |
525 | if (con->in_seq > con->in_seq_acked) { | |
526 | con->in_seq_acked = con->in_seq; | |
859eb799 | 527 | ceph_con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 528 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
859eb799 AE |
529 | ceph_con_out_kvec_add(con, sizeof (con->out_temp_ack), |
530 | &con->out_temp_ack); | |
31b8006e SW |
531 | } |
532 | ||
859eb799 | 533 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 534 | con->out_msg = m; |
4cf9d544 SW |
535 | |
536 | /* put message on sent list */ | |
537 | ceph_msg_get(m); | |
538 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 539 | |
e84346b7 SW |
540 | /* |
541 | * only assign outgoing seq # if we haven't sent this message | |
542 | * yet. if it is requeued, resend with it's original seq. | |
543 | */ | |
544 | if (m->needs_out_seq) { | |
545 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
546 | m->needs_out_seq = false; | |
547 | } | |
31b8006e SW |
548 | |
549 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
550 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
551 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
552 | le32_to_cpu(m->hdr.data_len), | |
553 | m->nr_pages); | |
554 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); | |
555 | ||
556 | /* tag + hdr + front + middle */ | |
859eb799 AE |
557 | ceph_con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
558 | ceph_con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
559 | ceph_con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
560 | ||
31b8006e | 561 | if (m->middle) |
859eb799 AE |
562 | ceph_con_out_kvec_add(con, m->middle->vec.iov_len, |
563 | m->middle->vec.iov_base); | |
31b8006e SW |
564 | |
565 | /* fill in crc (except data pages), footer */ | |
566 | con->out_msg->hdr.crc = | |
bd406145 | 567 | cpu_to_le32(crc32c(0, &m->hdr, |
31b8006e SW |
568 | sizeof(m->hdr) - sizeof(m->hdr.crc))); |
569 | con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE; | |
570 | con->out_msg->footer.front_crc = | |
571 | cpu_to_le32(crc32c(0, m->front.iov_base, m->front.iov_len)); | |
572 | if (m->middle) | |
573 | con->out_msg->footer.middle_crc = | |
574 | cpu_to_le32(crc32c(0, m->middle->vec.iov_base, | |
575 | m->middle->vec.iov_len)); | |
576 | else | |
577 | con->out_msg->footer.middle_crc = 0; | |
578 | con->out_msg->footer.data_crc = 0; | |
579 | dout("prepare_write_message front_crc %u data_crc %u\n", | |
580 | le32_to_cpu(con->out_msg->footer.front_crc), | |
581 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
582 | ||
583 | /* is there a data payload? */ | |
584 | if (le32_to_cpu(m->hdr.data_len) > 0) { | |
585 | /* initialize page iterator */ | |
586 | con->out_msg_pos.page = 0; | |
68b4476b | 587 | if (m->pages) |
c5c6b19d | 588 | con->out_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
589 | else |
590 | con->out_msg_pos.page_pos = 0; | |
31b8006e SW |
591 | con->out_msg_pos.data_pos = 0; |
592 | con->out_msg_pos.did_page_crc = 0; | |
593 | con->out_more = 1; /* data + footer will follow */ | |
594 | } else { | |
595 | /* no, queue up footer too and be done */ | |
859eb799 | 596 | prepare_write_message_footer(con); |
31b8006e SW |
597 | } |
598 | ||
599 | set_bit(WRITE_PENDING, &con->state); | |
600 | } | |
601 | ||
602 | /* | |
603 | * Prepare an ack. | |
604 | */ | |
605 | static void prepare_write_ack(struct ceph_connection *con) | |
606 | { | |
607 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
608 | con->in_seq_acked, con->in_seq); | |
609 | con->in_seq_acked = con->in_seq; | |
610 | ||
859eb799 AE |
611 | ceph_con_out_kvec_reset(con); |
612 | ||
613 | ceph_con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); | |
614 | ||
31b8006e | 615 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
859eb799 AE |
616 | ceph_con_out_kvec_add(con, sizeof (con->out_temp_ack), |
617 | &con->out_temp_ack); | |
618 | ||
31b8006e SW |
619 | con->out_more = 1; /* more will follow.. eventually.. */ |
620 | set_bit(WRITE_PENDING, &con->state); | |
621 | } | |
622 | ||
623 | /* | |
624 | * Prepare to write keepalive byte. | |
625 | */ | |
626 | static void prepare_write_keepalive(struct ceph_connection *con) | |
627 | { | |
628 | dout("prepare_write_keepalive %p\n", con); | |
859eb799 AE |
629 | ceph_con_out_kvec_reset(con); |
630 | ceph_con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
31b8006e SW |
631 | set_bit(WRITE_PENDING, &con->state); |
632 | } | |
633 | ||
634 | /* | |
635 | * Connection negotiation. | |
636 | */ | |
637 | ||
0da5d703 | 638 | static int prepare_connect_authorizer(struct ceph_connection *con) |
4e7a5dcd SW |
639 | { |
640 | void *auth_buf; | |
641 | int auth_len = 0; | |
642 | int auth_protocol = 0; | |
643 | ||
ec302645 | 644 | mutex_unlock(&con->mutex); |
4e7a5dcd SW |
645 | if (con->ops->get_authorizer) |
646 | con->ops->get_authorizer(con, &auth_buf, &auth_len, | |
647 | &auth_protocol, &con->auth_reply_buf, | |
648 | &con->auth_reply_buf_len, | |
649 | con->auth_retry); | |
ec302645 | 650 | mutex_lock(&con->mutex); |
4e7a5dcd | 651 | |
0da5d703 SW |
652 | if (test_bit(CLOSED, &con->state) || |
653 | test_bit(OPENING, &con->state)) | |
654 | return -EAGAIN; | |
655 | ||
4e7a5dcd SW |
656 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol); |
657 | con->out_connect.authorizer_len = cpu_to_le32(auth_len); | |
658 | ||
859eb799 AE |
659 | if (auth_len) |
660 | ceph_con_out_kvec_add(con, auth_len, auth_buf); | |
661 | ||
0da5d703 | 662 | return 0; |
4e7a5dcd SW |
663 | } |
664 | ||
31b8006e SW |
665 | /* |
666 | * We connected to a peer and are saying hello. | |
667 | */ | |
eed0ef2c SW |
668 | static void prepare_write_banner(struct ceph_messenger *msgr, |
669 | struct ceph_connection *con) | |
31b8006e | 670 | { |
859eb799 AE |
671 | ceph_con_out_kvec_reset(con); |
672 | ceph_con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); | |
673 | ceph_con_out_kvec_add(con, sizeof (msgr->my_enc_addr), | |
674 | &msgr->my_enc_addr); | |
eed0ef2c | 675 | |
eed0ef2c SW |
676 | con->out_more = 0; |
677 | set_bit(WRITE_PENDING, &con->state); | |
678 | } | |
679 | ||
0da5d703 SW |
680 | static int prepare_write_connect(struct ceph_messenger *msgr, |
681 | struct ceph_connection *con, | |
963be4d7 | 682 | int include_banner) |
eed0ef2c | 683 | { |
31b8006e SW |
684 | unsigned global_seq = get_global_seq(con->msgr, 0); |
685 | int proto; | |
686 | ||
687 | switch (con->peer_name.type) { | |
688 | case CEPH_ENTITY_TYPE_MON: | |
689 | proto = CEPH_MONC_PROTOCOL; | |
690 | break; | |
691 | case CEPH_ENTITY_TYPE_OSD: | |
692 | proto = CEPH_OSDC_PROTOCOL; | |
693 | break; | |
694 | case CEPH_ENTITY_TYPE_MDS: | |
695 | proto = CEPH_MDSC_PROTOCOL; | |
696 | break; | |
697 | default: | |
698 | BUG(); | |
699 | } | |
700 | ||
701 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
702 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 703 | |
3d14c5d2 | 704 | con->out_connect.features = cpu_to_le64(msgr->supported_features); |
31b8006e SW |
705 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
706 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
707 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
708 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
709 | con->out_connect.flags = 0; | |
31b8006e | 710 | |
963be4d7 AE |
711 | if (include_banner) |
712 | prepare_write_banner(msgr, con); | |
859eb799 AE |
713 | else |
714 | ceph_con_out_kvec_reset(con); | |
715 | ceph_con_out_kvec_add(con, sizeof (con->out_connect), &con->out_connect); | |
716 | ||
31b8006e SW |
717 | con->out_more = 0; |
718 | set_bit(WRITE_PENDING, &con->state); | |
4e7a5dcd | 719 | |
0da5d703 | 720 | return prepare_connect_authorizer(con); |
31b8006e SW |
721 | } |
722 | ||
31b8006e SW |
723 | /* |
724 | * write as much of pending kvecs to the socket as we can. | |
725 | * 1 -> done | |
726 | * 0 -> socket full, but more to do | |
727 | * <0 -> error | |
728 | */ | |
729 | static int write_partial_kvec(struct ceph_connection *con) | |
730 | { | |
731 | int ret; | |
732 | ||
733 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
734 | while (con->out_kvec_bytes > 0) { | |
735 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
736 | con->out_kvec_left, con->out_kvec_bytes, | |
737 | con->out_more); | |
738 | if (ret <= 0) | |
739 | goto out; | |
740 | con->out_kvec_bytes -= ret; | |
741 | if (con->out_kvec_bytes == 0) | |
742 | break; /* done */ | |
743 | while (ret > 0) { | |
744 | if (ret >= con->out_kvec_cur->iov_len) { | |
745 | ret -= con->out_kvec_cur->iov_len; | |
746 | con->out_kvec_cur++; | |
747 | con->out_kvec_left--; | |
748 | } else { | |
749 | con->out_kvec_cur->iov_len -= ret; | |
750 | con->out_kvec_cur->iov_base += ret; | |
751 | ret = 0; | |
752 | break; | |
753 | } | |
754 | } | |
755 | } | |
756 | con->out_kvec_left = 0; | |
757 | con->out_kvec_is_msg = false; | |
758 | ret = 1; | |
759 | out: | |
760 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
761 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
762 | return ret; /* done! */ | |
763 | } | |
764 | ||
68b4476b YS |
765 | #ifdef CONFIG_BLOCK |
766 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
767 | { | |
768 | if (!bio) { | |
769 | *iter = NULL; | |
770 | *seg = 0; | |
771 | return; | |
772 | } | |
773 | *iter = bio; | |
774 | *seg = bio->bi_idx; | |
775 | } | |
776 | ||
777 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
778 | { | |
779 | if (*bio_iter == NULL) | |
780 | return; | |
781 | ||
782 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
783 | ||
784 | (*seg)++; | |
785 | if (*seg == (*bio_iter)->bi_vcnt) | |
786 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
787 | } | |
788 | #endif | |
789 | ||
31b8006e SW |
790 | /* |
791 | * Write as much message data payload as we can. If we finish, queue | |
792 | * up the footer. | |
793 | * 1 -> done, footer is now queued in out_kvec[]. | |
794 | * 0 -> socket full, but more to do | |
795 | * <0 -> error | |
796 | */ | |
797 | static int write_partial_msg_pages(struct ceph_connection *con) | |
798 | { | |
799 | struct ceph_msg *msg = con->out_msg; | |
800 | unsigned data_len = le32_to_cpu(msg->hdr.data_len); | |
801 | size_t len; | |
802 | int crc = con->msgr->nocrc; | |
803 | int ret; | |
68b4476b YS |
804 | int total_max_write; |
805 | int in_trail = 0; | |
806 | size_t trail_len = (msg->trail ? msg->trail->length : 0); | |
31b8006e SW |
807 | |
808 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
809 | con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages, | |
810 | con->out_msg_pos.page_pos); | |
811 | ||
68b4476b YS |
812 | #ifdef CONFIG_BLOCK |
813 | if (msg->bio && !msg->bio_iter) | |
814 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); | |
815 | #endif | |
816 | ||
817 | while (data_len > con->out_msg_pos.data_pos) { | |
31b8006e SW |
818 | struct page *page = NULL; |
819 | void *kaddr = NULL; | |
68b4476b YS |
820 | int max_write = PAGE_SIZE; |
821 | int page_shift = 0; | |
822 | ||
823 | total_max_write = data_len - trail_len - | |
824 | con->out_msg_pos.data_pos; | |
31b8006e SW |
825 | |
826 | /* | |
827 | * if we are calculating the data crc (the default), we need | |
828 | * to map the page. if our pages[] has been revoked, use the | |
829 | * zero page. | |
830 | */ | |
68b4476b YS |
831 | |
832 | /* have we reached the trail part of the data? */ | |
833 | if (con->out_msg_pos.data_pos >= data_len - trail_len) { | |
834 | in_trail = 1; | |
835 | ||
836 | total_max_write = data_len - con->out_msg_pos.data_pos; | |
837 | ||
838 | page = list_first_entry(&msg->trail->head, | |
839 | struct page, lru); | |
840 | if (crc) | |
841 | kaddr = kmap(page); | |
842 | max_write = PAGE_SIZE; | |
843 | } else if (msg->pages) { | |
31b8006e SW |
844 | page = msg->pages[con->out_msg_pos.page]; |
845 | if (crc) | |
846 | kaddr = kmap(page); | |
58bb3b37 SW |
847 | } else if (msg->pagelist) { |
848 | page = list_first_entry(&msg->pagelist->head, | |
849 | struct page, lru); | |
850 | if (crc) | |
851 | kaddr = kmap(page); | |
68b4476b YS |
852 | #ifdef CONFIG_BLOCK |
853 | } else if (msg->bio) { | |
854 | struct bio_vec *bv; | |
855 | ||
856 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
857 | page = bv->bv_page; | |
858 | page_shift = bv->bv_offset; | |
859 | if (crc) | |
860 | kaddr = kmap(page) + page_shift; | |
861 | max_write = bv->bv_len; | |
862 | #endif | |
31b8006e | 863 | } else { |
57666519 | 864 | page = zero_page; |
31b8006e | 865 | if (crc) |
57666519 | 866 | kaddr = zero_page_address; |
31b8006e | 867 | } |
68b4476b YS |
868 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
869 | total_max_write); | |
870 | ||
31b8006e SW |
871 | if (crc && !con->out_msg_pos.did_page_crc) { |
872 | void *base = kaddr + con->out_msg_pos.page_pos; | |
873 | u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc); | |
874 | ||
875 | BUG_ON(kaddr == NULL); | |
876 | con->out_msg->footer.data_crc = | |
877 | cpu_to_le32(crc32c(tmpcrc, base, len)); | |
878 | con->out_msg_pos.did_page_crc = 1; | |
879 | } | |
31b8006e | 880 | ret = kernel_sendpage(con->sock, page, |
68b4476b YS |
881 | con->out_msg_pos.page_pos + page_shift, |
882 | len, | |
31b8006e SW |
883 | MSG_DONTWAIT | MSG_NOSIGNAL | |
884 | MSG_MORE); | |
885 | ||
68b4476b YS |
886 | if (crc && |
887 | (msg->pages || msg->pagelist || msg->bio || in_trail)) | |
31b8006e SW |
888 | kunmap(page); |
889 | ||
42961d23 SW |
890 | if (ret == -EAGAIN) |
891 | ret = 0; | |
31b8006e SW |
892 | if (ret <= 0) |
893 | goto out; | |
894 | ||
895 | con->out_msg_pos.data_pos += ret; | |
896 | con->out_msg_pos.page_pos += ret; | |
897 | if (ret == len) { | |
898 | con->out_msg_pos.page_pos = 0; | |
899 | con->out_msg_pos.page++; | |
900 | con->out_msg_pos.did_page_crc = 0; | |
68b4476b YS |
901 | if (in_trail) |
902 | list_move_tail(&page->lru, | |
903 | &msg->trail->head); | |
904 | else if (msg->pagelist) | |
58bb3b37 SW |
905 | list_move_tail(&page->lru, |
906 | &msg->pagelist->head); | |
68b4476b YS |
907 | #ifdef CONFIG_BLOCK |
908 | else if (msg->bio) | |
909 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
910 | #endif | |
31b8006e SW |
911 | } |
912 | } | |
913 | ||
914 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
915 | ||
916 | /* prepare and queue up footer, too */ | |
917 | if (!crc) | |
918 | con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; | |
859eb799 AE |
919 | ceph_con_out_kvec_reset(con); |
920 | prepare_write_message_footer(con); | |
31b8006e SW |
921 | ret = 1; |
922 | out: | |
923 | return ret; | |
924 | } | |
925 | ||
926 | /* | |
927 | * write some zeros | |
928 | */ | |
929 | static int write_partial_skip(struct ceph_connection *con) | |
930 | { | |
931 | int ret; | |
932 | ||
933 | while (con->out_skip > 0) { | |
934 | struct kvec iov = { | |
57666519 | 935 | .iov_base = zero_page_address, |
31b8006e SW |
936 | .iov_len = min(con->out_skip, (int)PAGE_CACHE_SIZE) |
937 | }; | |
938 | ||
939 | ret = ceph_tcp_sendmsg(con->sock, &iov, 1, iov.iov_len, 1); | |
940 | if (ret <= 0) | |
941 | goto out; | |
942 | con->out_skip -= ret; | |
943 | } | |
944 | ret = 1; | |
945 | out: | |
946 | return ret; | |
947 | } | |
948 | ||
949 | /* | |
950 | * Prepare to read connection handshake, or an ack. | |
951 | */ | |
eed0ef2c SW |
952 | static void prepare_read_banner(struct ceph_connection *con) |
953 | { | |
954 | dout("prepare_read_banner %p\n", con); | |
955 | con->in_base_pos = 0; | |
956 | } | |
957 | ||
31b8006e SW |
958 | static void prepare_read_connect(struct ceph_connection *con) |
959 | { | |
960 | dout("prepare_read_connect %p\n", con); | |
961 | con->in_base_pos = 0; | |
962 | } | |
963 | ||
964 | static void prepare_read_ack(struct ceph_connection *con) | |
965 | { | |
966 | dout("prepare_read_ack %p\n", con); | |
967 | con->in_base_pos = 0; | |
968 | } | |
969 | ||
970 | static void prepare_read_tag(struct ceph_connection *con) | |
971 | { | |
972 | dout("prepare_read_tag %p\n", con); | |
973 | con->in_base_pos = 0; | |
974 | con->in_tag = CEPH_MSGR_TAG_READY; | |
975 | } | |
976 | ||
977 | /* | |
978 | * Prepare to read a message. | |
979 | */ | |
980 | static int prepare_read_message(struct ceph_connection *con) | |
981 | { | |
982 | dout("prepare_read_message %p\n", con); | |
983 | BUG_ON(con->in_msg != NULL); | |
984 | con->in_base_pos = 0; | |
985 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
986 | return 0; | |
987 | } | |
988 | ||
989 | ||
990 | static int read_partial(struct ceph_connection *con, | |
991 | int *to, int size, void *object) | |
992 | { | |
993 | *to += size; | |
994 | while (con->in_base_pos < *to) { | |
995 | int left = *to - con->in_base_pos; | |
996 | int have = size - left; | |
997 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
998 | if (ret <= 0) | |
999 | return ret; | |
1000 | con->in_base_pos += ret; | |
1001 | } | |
1002 | return 1; | |
1003 | } | |
1004 | ||
1005 | ||
1006 | /* | |
1007 | * Read all or part of the connect-side handshake on a new connection | |
1008 | */ | |
eed0ef2c | 1009 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e SW |
1010 | { |
1011 | int ret, to = 0; | |
1012 | ||
eed0ef2c | 1013 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1014 | |
1015 | /* peer's banner */ | |
1016 | ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner); | |
1017 | if (ret <= 0) | |
1018 | goto out; | |
1019 | ret = read_partial(con, &to, sizeof(con->actual_peer_addr), | |
1020 | &con->actual_peer_addr); | |
1021 | if (ret <= 0) | |
1022 | goto out; | |
1023 | ret = read_partial(con, &to, sizeof(con->peer_addr_for_me), | |
1024 | &con->peer_addr_for_me); | |
1025 | if (ret <= 0) | |
1026 | goto out; | |
eed0ef2c SW |
1027 | out: |
1028 | return ret; | |
1029 | } | |
1030 | ||
1031 | static int read_partial_connect(struct ceph_connection *con) | |
1032 | { | |
1033 | int ret, to = 0; | |
1034 | ||
1035 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1036 | ||
31b8006e SW |
1037 | ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply); |
1038 | if (ret <= 0) | |
1039 | goto out; | |
4e7a5dcd SW |
1040 | ret = read_partial(con, &to, le32_to_cpu(con->in_reply.authorizer_len), |
1041 | con->auth_reply_buf); | |
1042 | if (ret <= 0) | |
1043 | goto out; | |
31b8006e | 1044 | |
4e7a5dcd SW |
1045 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1046 | con, (int)con->in_reply.tag, | |
1047 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1048 | le32_to_cpu(con->in_reply.global_seq)); |
1049 | out: | |
1050 | return ret; | |
eed0ef2c | 1051 | |
31b8006e SW |
1052 | } |
1053 | ||
1054 | /* | |
1055 | * Verify the hello banner looks okay. | |
1056 | */ | |
1057 | static int verify_hello(struct ceph_connection *con) | |
1058 | { | |
1059 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1060 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1061 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1062 | con->error_msg = "protocol error, bad banner"; |
1063 | return -1; | |
1064 | } | |
1065 | return 0; | |
1066 | } | |
1067 | ||
1068 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1069 | { | |
1070 | switch (ss->ss_family) { | |
1071 | case AF_INET: | |
1072 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1073 | case AF_INET6: | |
1074 | return | |
1075 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1076 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1077 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1078 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1079 | } | |
1080 | return false; | |
1081 | } | |
1082 | ||
1083 | static int addr_port(struct sockaddr_storage *ss) | |
1084 | { | |
1085 | switch (ss->ss_family) { | |
1086 | case AF_INET: | |
f28bcfbe | 1087 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1088 | case AF_INET6: |
f28bcfbe | 1089 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1090 | } |
1091 | return 0; | |
1092 | } | |
1093 | ||
1094 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1095 | { | |
1096 | switch (ss->ss_family) { | |
1097 | case AF_INET: | |
1098 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1099 | break; |
31b8006e SW |
1100 | case AF_INET6: |
1101 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1102 | break; |
31b8006e SW |
1103 | } |
1104 | } | |
1105 | ||
ee3b56f2 NW |
1106 | /* |
1107 | * Unlike other *_pton function semantics, zero indicates success. | |
1108 | */ | |
1109 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1110 | char delim, const char **ipend) | |
1111 | { | |
99f0f3b2 AE |
1112 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1113 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1114 | |
1115 | memset(ss, 0, sizeof(*ss)); | |
1116 | ||
1117 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1118 | ss->ss_family = AF_INET; | |
1119 | return 0; | |
1120 | } | |
1121 | ||
1122 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1123 | ss->ss_family = AF_INET6; | |
1124 | return 0; | |
1125 | } | |
1126 | ||
1127 | return -EINVAL; | |
1128 | } | |
1129 | ||
1130 | /* | |
1131 | * Extract hostname string and resolve using kernel DNS facility. | |
1132 | */ | |
1133 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1134 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1135 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1136 | { | |
1137 | const char *end, *delim_p; | |
1138 | char *colon_p, *ip_addr = NULL; | |
1139 | int ip_len, ret; | |
1140 | ||
1141 | /* | |
1142 | * The end of the hostname occurs immediately preceding the delimiter or | |
1143 | * the port marker (':') where the delimiter takes precedence. | |
1144 | */ | |
1145 | delim_p = memchr(name, delim, namelen); | |
1146 | colon_p = memchr(name, ':', namelen); | |
1147 | ||
1148 | if (delim_p && colon_p) | |
1149 | end = delim_p < colon_p ? delim_p : colon_p; | |
1150 | else if (!delim_p && colon_p) | |
1151 | end = colon_p; | |
1152 | else { | |
1153 | end = delim_p; | |
1154 | if (!end) /* case: hostname:/ */ | |
1155 | end = name + namelen; | |
1156 | } | |
1157 | ||
1158 | if (end <= name) | |
1159 | return -EINVAL; | |
1160 | ||
1161 | /* do dns_resolve upcall */ | |
1162 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1163 | if (ip_len > 0) | |
1164 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1165 | else | |
1166 | ret = -ESRCH; | |
1167 | ||
1168 | kfree(ip_addr); | |
1169 | ||
1170 | *ipend = end; | |
1171 | ||
1172 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1173 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1174 | ||
1175 | return ret; | |
1176 | } | |
1177 | #else | |
1178 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1179 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1180 | { | |
1181 | return -EINVAL; | |
1182 | } | |
1183 | #endif | |
1184 | ||
1185 | /* | |
1186 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1187 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1188 | */ | |
1189 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1190 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1191 | { | |
1192 | int ret; | |
1193 | ||
1194 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1195 | if (ret) | |
1196 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1197 | ||
1198 | return ret; | |
1199 | } | |
1200 | ||
31b8006e SW |
1201 | /* |
1202 | * Parse an ip[:port] list into an addr array. Use the default | |
1203 | * monitor port if a port isn't specified. | |
1204 | */ | |
1205 | int ceph_parse_ips(const char *c, const char *end, | |
1206 | struct ceph_entity_addr *addr, | |
1207 | int max_count, int *count) | |
1208 | { | |
ee3b56f2 | 1209 | int i, ret = -EINVAL; |
31b8006e SW |
1210 | const char *p = c; |
1211 | ||
1212 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1213 | for (i = 0; i < max_count; i++) { | |
1214 | const char *ipend; | |
1215 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1216 | int port; |
39139f64 SW |
1217 | char delim = ','; |
1218 | ||
1219 | if (*p == '[') { | |
1220 | delim = ']'; | |
1221 | p++; | |
1222 | } | |
31b8006e | 1223 | |
ee3b56f2 NW |
1224 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1225 | if (ret) | |
31b8006e | 1226 | goto bad; |
ee3b56f2 NW |
1227 | ret = -EINVAL; |
1228 | ||
31b8006e SW |
1229 | p = ipend; |
1230 | ||
39139f64 SW |
1231 | if (delim == ']') { |
1232 | if (*p != ']') { | |
1233 | dout("missing matching ']'\n"); | |
1234 | goto bad; | |
1235 | } | |
1236 | p++; | |
1237 | } | |
1238 | ||
31b8006e SW |
1239 | /* port? */ |
1240 | if (p < end && *p == ':') { | |
1241 | port = 0; | |
1242 | p++; | |
1243 | while (p < end && *p >= '0' && *p <= '9') { | |
1244 | port = (port * 10) + (*p - '0'); | |
1245 | p++; | |
1246 | } | |
1247 | if (port > 65535 || port == 0) | |
1248 | goto bad; | |
1249 | } else { | |
1250 | port = CEPH_MON_PORT; | |
1251 | } | |
1252 | ||
1253 | addr_set_port(ss, port); | |
1254 | ||
3d14c5d2 | 1255 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1256 | |
1257 | if (p == end) | |
1258 | break; | |
1259 | if (*p != ',') | |
1260 | goto bad; | |
1261 | p++; | |
1262 | } | |
1263 | ||
1264 | if (p != end) | |
1265 | goto bad; | |
1266 | ||
1267 | if (count) | |
1268 | *count = i + 1; | |
1269 | return 0; | |
1270 | ||
1271 | bad: | |
39139f64 | 1272 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1273 | return ret; |
31b8006e | 1274 | } |
3d14c5d2 | 1275 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1276 | |
eed0ef2c | 1277 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1278 | { |
eed0ef2c | 1279 | dout("process_banner on %p\n", con); |
31b8006e SW |
1280 | |
1281 | if (verify_hello(con) < 0) | |
1282 | return -1; | |
1283 | ||
63f2d211 SW |
1284 | ceph_decode_addr(&con->actual_peer_addr); |
1285 | ceph_decode_addr(&con->peer_addr_for_me); | |
1286 | ||
31b8006e SW |
1287 | /* |
1288 | * Make sure the other end is who we wanted. note that the other | |
1289 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1290 | * them the benefit of the doubt. | |
1291 | */ | |
103e2d3a SW |
1292 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1293 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1294 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1295 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1296 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1297 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1298 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1299 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1300 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1301 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1302 | return -1; |
1303 | } | |
1304 | ||
1305 | /* | |
1306 | * did we learn our address? | |
1307 | */ | |
1308 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1309 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1310 | ||
1311 | memcpy(&con->msgr->inst.addr.in_addr, | |
1312 | &con->peer_addr_for_me.in_addr, | |
1313 | sizeof(con->peer_addr_for_me.in_addr)); | |
1314 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1315 | encode_my_addr(con->msgr); |
eed0ef2c | 1316 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1317 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1318 | } |
1319 | ||
eed0ef2c SW |
1320 | set_bit(NEGOTIATING, &con->state); |
1321 | prepare_read_connect(con); | |
1322 | return 0; | |
1323 | } | |
1324 | ||
04a419f9 SW |
1325 | static void fail_protocol(struct ceph_connection *con) |
1326 | { | |
1327 | reset_connection(con); | |
1328 | set_bit(CLOSED, &con->state); /* in case there's queued work */ | |
1329 | ||
1330 | mutex_unlock(&con->mutex); | |
1331 | if (con->ops->bad_proto) | |
1332 | con->ops->bad_proto(con); | |
1333 | mutex_lock(&con->mutex); | |
1334 | } | |
1335 | ||
eed0ef2c SW |
1336 | static int process_connect(struct ceph_connection *con) |
1337 | { | |
3d14c5d2 YS |
1338 | u64 sup_feat = con->msgr->supported_features; |
1339 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1340 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1341 | int ret; |
04a419f9 | 1342 | |
eed0ef2c SW |
1343 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1344 | ||
31b8006e | 1345 | switch (con->in_reply.tag) { |
04a419f9 SW |
1346 | case CEPH_MSGR_TAG_FEATURES: |
1347 | pr_err("%s%lld %s feature set mismatch," | |
1348 | " my %llx < server's %llx, missing %llx\n", | |
1349 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1350 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1351 | sup_feat, server_feat, server_feat & ~sup_feat); |
1352 | con->error_msg = "missing required protocol features"; | |
1353 | fail_protocol(con); | |
1354 | return -1; | |
1355 | ||
31b8006e | 1356 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1357 | pr_err("%s%lld %s protocol version mismatch," |
1358 | " my %d != server's %d\n", | |
1359 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1360 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1361 | le32_to_cpu(con->out_connect.protocol_version), |
1362 | le32_to_cpu(con->in_reply.protocol_version)); | |
1363 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1364 | fail_protocol(con); |
31b8006e SW |
1365 | return -1; |
1366 | ||
4e7a5dcd SW |
1367 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1368 | con->auth_retry++; | |
1369 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1370 | con->auth_retry); | |
1371 | if (con->auth_retry == 2) { | |
1372 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1373 | return -1; |
1374 | } | |
1375 | con->auth_retry = 1; | |
0da5d703 SW |
1376 | ret = prepare_write_connect(con->msgr, con, 0); |
1377 | if (ret < 0) | |
1378 | return ret; | |
63733a0f | 1379 | prepare_read_connect(con); |
4e7a5dcd | 1380 | break; |
31b8006e SW |
1381 | |
1382 | case CEPH_MSGR_TAG_RESETSESSION: | |
1383 | /* | |
1384 | * If we connected with a large connect_seq but the peer | |
1385 | * has no record of a session with us (no connection, or | |
1386 | * connect_seq == 0), they will send RESETSESION to indicate | |
1387 | * that they must have reset their session, and may have | |
1388 | * dropped messages. | |
1389 | */ | |
1390 | dout("process_connect got RESET peer seq %u\n", | |
1391 | le32_to_cpu(con->in_connect.connect_seq)); | |
1392 | pr_err("%s%lld %s connection reset\n", | |
1393 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1394 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1395 | reset_connection(con); |
eed0ef2c | 1396 | prepare_write_connect(con->msgr, con, 0); |
31b8006e SW |
1397 | prepare_read_connect(con); |
1398 | ||
1399 | /* Tell ceph about it. */ | |
ec302645 | 1400 | mutex_unlock(&con->mutex); |
31b8006e SW |
1401 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1402 | if (con->ops->peer_reset) | |
1403 | con->ops->peer_reset(con); | |
ec302645 | 1404 | mutex_lock(&con->mutex); |
0da5d703 SW |
1405 | if (test_bit(CLOSED, &con->state) || |
1406 | test_bit(OPENING, &con->state)) | |
1407 | return -EAGAIN; | |
31b8006e SW |
1408 | break; |
1409 | ||
1410 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1411 | /* | |
1412 | * If we sent a smaller connect_seq than the peer has, try | |
1413 | * again with a larger value. | |
1414 | */ | |
1415 | dout("process_connect got RETRY my seq = %u, peer_seq = %u\n", | |
1416 | le32_to_cpu(con->out_connect.connect_seq), | |
1417 | le32_to_cpu(con->in_connect.connect_seq)); | |
1418 | con->connect_seq = le32_to_cpu(con->in_connect.connect_seq); | |
eed0ef2c | 1419 | prepare_write_connect(con->msgr, con, 0); |
31b8006e SW |
1420 | prepare_read_connect(con); |
1421 | break; | |
1422 | ||
1423 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1424 | /* | |
1425 | * If we sent a smaller global_seq than the peer has, try | |
1426 | * again with a larger value. | |
1427 | */ | |
eed0ef2c | 1428 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e SW |
1429 | con->peer_global_seq, |
1430 | le32_to_cpu(con->in_connect.global_seq)); | |
1431 | get_global_seq(con->msgr, | |
1432 | le32_to_cpu(con->in_connect.global_seq)); | |
eed0ef2c | 1433 | prepare_write_connect(con->msgr, con, 0); |
31b8006e SW |
1434 | prepare_read_connect(con); |
1435 | break; | |
1436 | ||
1437 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1438 | if (req_feat & ~server_feat) { |
1439 | pr_err("%s%lld %s protocol feature mismatch," | |
1440 | " my required %llx > server's %llx, need %llx\n", | |
1441 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1442 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1443 | req_feat, server_feat, req_feat & ~server_feat); |
1444 | con->error_msg = "missing required protocol features"; | |
1445 | fail_protocol(con); | |
1446 | return -1; | |
1447 | } | |
31b8006e | 1448 | clear_bit(CONNECTING, &con->state); |
31b8006e SW |
1449 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1450 | con->connect_seq++; | |
aba558e2 | 1451 | con->peer_features = server_feat; |
31b8006e SW |
1452 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1453 | con->peer_global_seq, | |
1454 | le32_to_cpu(con->in_reply.connect_seq), | |
1455 | con->connect_seq); | |
1456 | WARN_ON(con->connect_seq != | |
1457 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1458 | |
1459 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
1460 | set_bit(LOSSYTX, &con->state); | |
1461 | ||
31b8006e SW |
1462 | prepare_read_tag(con); |
1463 | break; | |
1464 | ||
1465 | case CEPH_MSGR_TAG_WAIT: | |
1466 | /* | |
1467 | * If there is a connection race (we are opening | |
1468 | * connections to each other), one of us may just have | |
1469 | * to WAIT. This shouldn't happen if we are the | |
1470 | * client. | |
1471 | */ | |
04177882 SW |
1472 | pr_err("process_connect got WAIT as client\n"); |
1473 | con->error_msg = "protocol error, got WAIT as client"; | |
1474 | return -1; | |
31b8006e SW |
1475 | |
1476 | default: | |
1477 | pr_err("connect protocol error, will retry\n"); | |
1478 | con->error_msg = "protocol error, garbage tag during connect"; | |
1479 | return -1; | |
1480 | } | |
1481 | return 0; | |
1482 | } | |
1483 | ||
1484 | ||
1485 | /* | |
1486 | * read (part of) an ack | |
1487 | */ | |
1488 | static int read_partial_ack(struct ceph_connection *con) | |
1489 | { | |
1490 | int to = 0; | |
1491 | ||
1492 | return read_partial(con, &to, sizeof(con->in_temp_ack), | |
1493 | &con->in_temp_ack); | |
1494 | } | |
1495 | ||
1496 | ||
1497 | /* | |
1498 | * We can finally discard anything that's been acked. | |
1499 | */ | |
1500 | static void process_ack(struct ceph_connection *con) | |
1501 | { | |
1502 | struct ceph_msg *m; | |
1503 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1504 | u64 seq; | |
1505 | ||
31b8006e SW |
1506 | while (!list_empty(&con->out_sent)) { |
1507 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1508 | list_head); | |
1509 | seq = le64_to_cpu(m->hdr.seq); | |
1510 | if (seq > ack) | |
1511 | break; | |
1512 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1513 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1514 | m->ack_stamp = jiffies; |
31b8006e SW |
1515 | ceph_msg_remove(m); |
1516 | } | |
31b8006e SW |
1517 | prepare_read_tag(con); |
1518 | } | |
1519 | ||
1520 | ||
1521 | ||
1522 | ||
2450418c | 1523 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1524 | struct kvec *section, |
1525 | unsigned int sec_len, u32 *crc) | |
2450418c | 1526 | { |
68b4476b | 1527 | int ret, left; |
2450418c YS |
1528 | |
1529 | BUG_ON(!section); | |
1530 | ||
1531 | while (section->iov_len < sec_len) { | |
1532 | BUG_ON(section->iov_base == NULL); | |
1533 | left = sec_len - section->iov_len; | |
1534 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1535 | section->iov_len, left); | |
1536 | if (ret <= 0) | |
1537 | return ret; | |
1538 | section->iov_len += ret; | |
1539 | if (section->iov_len == sec_len) | |
1540 | *crc = crc32c(0, section->iov_base, | |
1541 | section->iov_len); | |
1542 | } | |
31b8006e | 1543 | |
2450418c YS |
1544 | return 1; |
1545 | } | |
31b8006e | 1546 | |
2450418c YS |
1547 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, |
1548 | struct ceph_msg_header *hdr, | |
1549 | int *skip); | |
68b4476b YS |
1550 | |
1551 | ||
1552 | static int read_partial_message_pages(struct ceph_connection *con, | |
1553 | struct page **pages, | |
1554 | unsigned data_len, int datacrc) | |
1555 | { | |
1556 | void *p; | |
1557 | int ret; | |
1558 | int left; | |
1559 | ||
1560 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1561 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1562 | /* (page) data */ | |
1563 | BUG_ON(pages == NULL); | |
1564 | p = kmap(pages[con->in_msg_pos.page]); | |
1565 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1566 | left); | |
1567 | if (ret > 0 && datacrc) | |
1568 | con->in_data_crc = | |
1569 | crc32c(con->in_data_crc, | |
1570 | p + con->in_msg_pos.page_pos, ret); | |
1571 | kunmap(pages[con->in_msg_pos.page]); | |
1572 | if (ret <= 0) | |
1573 | return ret; | |
1574 | con->in_msg_pos.data_pos += ret; | |
1575 | con->in_msg_pos.page_pos += ret; | |
1576 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1577 | con->in_msg_pos.page_pos = 0; | |
1578 | con->in_msg_pos.page++; | |
1579 | } | |
1580 | ||
1581 | return ret; | |
1582 | } | |
1583 | ||
1584 | #ifdef CONFIG_BLOCK | |
1585 | static int read_partial_message_bio(struct ceph_connection *con, | |
1586 | struct bio **bio_iter, int *bio_seg, | |
1587 | unsigned data_len, int datacrc) | |
1588 | { | |
1589 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1590 | void *p; | |
1591 | int ret, left; | |
1592 | ||
1593 | if (IS_ERR(bv)) | |
1594 | return PTR_ERR(bv); | |
1595 | ||
1596 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1597 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1598 | ||
1599 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1600 | ||
1601 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1602 | left); | |
1603 | if (ret > 0 && datacrc) | |
1604 | con->in_data_crc = | |
1605 | crc32c(con->in_data_crc, | |
1606 | p + con->in_msg_pos.page_pos, ret); | |
1607 | kunmap(bv->bv_page); | |
1608 | if (ret <= 0) | |
1609 | return ret; | |
1610 | con->in_msg_pos.data_pos += ret; | |
1611 | con->in_msg_pos.page_pos += ret; | |
1612 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1613 | con->in_msg_pos.page_pos = 0; | |
1614 | iter_bio_next(bio_iter, bio_seg); | |
1615 | } | |
1616 | ||
1617 | return ret; | |
1618 | } | |
1619 | #endif | |
1620 | ||
31b8006e SW |
1621 | /* |
1622 | * read (part of) a message. | |
1623 | */ | |
1624 | static int read_partial_message(struct ceph_connection *con) | |
1625 | { | |
1626 | struct ceph_msg *m = con->in_msg; | |
31b8006e | 1627 | int ret; |
9d7f0f13 | 1628 | int to, left; |
c5c6b19d | 1629 | unsigned front_len, middle_len, data_len; |
31b8006e | 1630 | int datacrc = con->msgr->nocrc; |
2450418c | 1631 | int skip; |
ae18756b | 1632 | u64 seq; |
31b8006e SW |
1633 | |
1634 | dout("read_partial_message con %p msg %p\n", con, m); | |
1635 | ||
1636 | /* header */ | |
1637 | while (con->in_base_pos < sizeof(con->in_hdr)) { | |
1638 | left = sizeof(con->in_hdr) - con->in_base_pos; | |
1639 | ret = ceph_tcp_recvmsg(con->sock, | |
1640 | (char *)&con->in_hdr + con->in_base_pos, | |
1641 | left); | |
1642 | if (ret <= 0) | |
1643 | return ret; | |
1644 | con->in_base_pos += ret; | |
1645 | if (con->in_base_pos == sizeof(con->in_hdr)) { | |
bd406145 | 1646 | u32 crc = crc32c(0, &con->in_hdr, |
31b8006e SW |
1647 | sizeof(con->in_hdr) - sizeof(con->in_hdr.crc)); |
1648 | if (crc != le32_to_cpu(con->in_hdr.crc)) { | |
1649 | pr_err("read_partial_message bad hdr " | |
1650 | " crc %u != expected %u\n", | |
1651 | crc, con->in_hdr.crc); | |
1652 | return -EBADMSG; | |
1653 | } | |
1654 | } | |
1655 | } | |
31b8006e SW |
1656 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1657 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1658 | return -EIO; | |
1659 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1660 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1661 | return -EIO; | |
1662 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1663 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1664 | return -EIO; | |
1665 | ||
ae18756b SW |
1666 | /* verify seq# */ |
1667 | seq = le64_to_cpu(con->in_hdr.seq); | |
1668 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1669 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1670 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1671 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1672 | seq, con->in_seq + 1); |
1673 | con->in_base_pos = -front_len - middle_len - data_len - | |
1674 | sizeof(m->footer); | |
1675 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1676 | return 0; |
1677 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1678 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1679 | seq, con->in_seq + 1); | |
1680 | con->error_msg = "bad message sequence # for incoming message"; | |
1681 | return -EBADMSG; | |
1682 | } | |
1683 | ||
31b8006e SW |
1684 | /* allocate message? */ |
1685 | if (!con->in_msg) { | |
1686 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1687 | con->in_hdr.front_len, con->in_hdr.data_len); | |
ae32be31 | 1688 | skip = 0; |
2450418c YS |
1689 | con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip); |
1690 | if (skip) { | |
31b8006e | 1691 | /* skip this message */ |
a79832f2 | 1692 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1693 | BUG_ON(con->in_msg); |
31b8006e SW |
1694 | con->in_base_pos = -front_len - middle_len - data_len - |
1695 | sizeof(m->footer); | |
1696 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1697 | con->in_seq++; |
31b8006e SW |
1698 | return 0; |
1699 | } | |
a79832f2 | 1700 | if (!con->in_msg) { |
5b3a4db3 SW |
1701 | con->error_msg = |
1702 | "error allocating memory for incoming message"; | |
a79832f2 | 1703 | return -ENOMEM; |
31b8006e SW |
1704 | } |
1705 | m = con->in_msg; | |
1706 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1707 | if (m->middle) |
1708 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1709 | |
1710 | con->in_msg_pos.page = 0; | |
68b4476b | 1711 | if (m->pages) |
c5c6b19d | 1712 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1713 | else |
1714 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1715 | con->in_msg_pos.data_pos = 0; |
31b8006e SW |
1716 | } |
1717 | ||
1718 | /* front */ | |
2450418c YS |
1719 | ret = read_partial_message_section(con, &m->front, front_len, |
1720 | &con->in_front_crc); | |
1721 | if (ret <= 0) | |
1722 | return ret; | |
31b8006e SW |
1723 | |
1724 | /* middle */ | |
2450418c | 1725 | if (m->middle) { |
213c99ee SW |
1726 | ret = read_partial_message_section(con, &m->middle->vec, |
1727 | middle_len, | |
2450418c | 1728 | &con->in_middle_crc); |
31b8006e SW |
1729 | if (ret <= 0) |
1730 | return ret; | |
31b8006e | 1731 | } |
68b4476b YS |
1732 | #ifdef CONFIG_BLOCK |
1733 | if (m->bio && !m->bio_iter) | |
1734 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1735 | #endif | |
31b8006e SW |
1736 | |
1737 | /* (page) data */ | |
31b8006e | 1738 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1739 | if (m->pages) { |
1740 | ret = read_partial_message_pages(con, m->pages, | |
1741 | data_len, datacrc); | |
1742 | if (ret <= 0) | |
1743 | return ret; | |
1744 | #ifdef CONFIG_BLOCK | |
1745 | } else if (m->bio) { | |
1746 | ||
1747 | ret = read_partial_message_bio(con, | |
1748 | &m->bio_iter, &m->bio_seg, | |
1749 | data_len, datacrc); | |
1750 | if (ret <= 0) | |
1751 | return ret; | |
1752 | #endif | |
1753 | } else { | |
1754 | BUG_ON(1); | |
31b8006e SW |
1755 | } |
1756 | } | |
1757 | ||
31b8006e SW |
1758 | /* footer */ |
1759 | to = sizeof(m->hdr) + sizeof(m->footer); | |
1760 | while (con->in_base_pos < to) { | |
1761 | left = to - con->in_base_pos; | |
1762 | ret = ceph_tcp_recvmsg(con->sock, (char *)&m->footer + | |
1763 | (con->in_base_pos - sizeof(m->hdr)), | |
1764 | left); | |
1765 | if (ret <= 0) | |
1766 | return ret; | |
1767 | con->in_base_pos += ret; | |
1768 | } | |
1769 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", | |
1770 | m, front_len, m->footer.front_crc, middle_len, | |
1771 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1772 | ||
1773 | /* crc ok? */ | |
1774 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1775 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1776 | m, con->in_front_crc, m->footer.front_crc); | |
1777 | return -EBADMSG; | |
1778 | } | |
1779 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1780 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1781 | m, con->in_middle_crc, m->footer.middle_crc); | |
1782 | return -EBADMSG; | |
1783 | } | |
1784 | if (datacrc && | |
1785 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && | |
1786 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1787 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1788 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1789 | return -EBADMSG; | |
1790 | } | |
1791 | ||
1792 | return 1; /* done! */ | |
1793 | } | |
1794 | ||
1795 | /* | |
1796 | * Process message. This happens in the worker thread. The callback should | |
1797 | * be careful not to do anything that waits on other incoming messages or it | |
1798 | * may deadlock. | |
1799 | */ | |
1800 | static void process_message(struct ceph_connection *con) | |
1801 | { | |
5e095e8b | 1802 | struct ceph_msg *msg; |
31b8006e | 1803 | |
5e095e8b | 1804 | msg = con->in_msg; |
31b8006e SW |
1805 | con->in_msg = NULL; |
1806 | ||
1807 | /* if first message, set peer_name */ | |
1808 | if (con->peer_name.type == 0) | |
dbad185d | 1809 | con->peer_name = msg->hdr.src; |
31b8006e | 1810 | |
31b8006e | 1811 | con->in_seq++; |
ec302645 | 1812 | mutex_unlock(&con->mutex); |
31b8006e SW |
1813 | |
1814 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1815 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1816 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1817 | le16_to_cpu(msg->hdr.type), |
1818 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1819 | le32_to_cpu(msg->hdr.front_len), | |
1820 | le32_to_cpu(msg->hdr.data_len), | |
1821 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1822 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1823 | |
1824 | mutex_lock(&con->mutex); | |
31b8006e SW |
1825 | prepare_read_tag(con); |
1826 | } | |
1827 | ||
1828 | ||
1829 | /* | |
1830 | * Write something to the socket. Called in a worker thread when the | |
1831 | * socket appears to be writeable and we have something ready to send. | |
1832 | */ | |
1833 | static int try_write(struct ceph_connection *con) | |
1834 | { | |
1835 | struct ceph_messenger *msgr = con->msgr; | |
1836 | int ret = 1; | |
1837 | ||
1838 | dout("try_write start %p state %lu nref %d\n", con, con->state, | |
1839 | atomic_read(&con->nref)); | |
1840 | ||
31b8006e SW |
1841 | more: |
1842 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
1843 | ||
1844 | /* open the socket first? */ | |
1845 | if (con->sock == NULL) { | |
eed0ef2c SW |
1846 | prepare_write_connect(msgr, con, 1); |
1847 | prepare_read_banner(con); | |
31b8006e | 1848 | set_bit(CONNECTING, &con->state); |
eed0ef2c | 1849 | clear_bit(NEGOTIATING, &con->state); |
31b8006e | 1850 | |
cf3e5c40 | 1851 | BUG_ON(con->in_msg); |
31b8006e SW |
1852 | con->in_tag = CEPH_MSGR_TAG_READY; |
1853 | dout("try_write initiating connect on %p new state %lu\n", | |
1854 | con, con->state); | |
1855 | con->sock = ceph_tcp_connect(con); | |
1856 | if (IS_ERR(con->sock)) { | |
1857 | con->sock = NULL; | |
1858 | con->error_msg = "connect error"; | |
1859 | ret = -1; | |
1860 | goto out; | |
1861 | } | |
1862 | } | |
1863 | ||
1864 | more_kvec: | |
1865 | /* kvec data queued? */ | |
1866 | if (con->out_skip) { | |
1867 | ret = write_partial_skip(con); | |
1868 | if (ret <= 0) | |
42961d23 | 1869 | goto out; |
31b8006e SW |
1870 | } |
1871 | if (con->out_kvec_left) { | |
1872 | ret = write_partial_kvec(con); | |
1873 | if (ret <= 0) | |
42961d23 | 1874 | goto out; |
31b8006e SW |
1875 | } |
1876 | ||
1877 | /* msg pages? */ | |
1878 | if (con->out_msg) { | |
c86a2930 SW |
1879 | if (con->out_msg_done) { |
1880 | ceph_msg_put(con->out_msg); | |
1881 | con->out_msg = NULL; /* we're done with this one */ | |
1882 | goto do_next; | |
1883 | } | |
1884 | ||
31b8006e SW |
1885 | ret = write_partial_msg_pages(con); |
1886 | if (ret == 1) | |
1887 | goto more_kvec; /* we need to send the footer, too! */ | |
1888 | if (ret == 0) | |
42961d23 | 1889 | goto out; |
31b8006e SW |
1890 | if (ret < 0) { |
1891 | dout("try_write write_partial_msg_pages err %d\n", | |
1892 | ret); | |
42961d23 | 1893 | goto out; |
31b8006e SW |
1894 | } |
1895 | } | |
1896 | ||
c86a2930 | 1897 | do_next: |
31b8006e SW |
1898 | if (!test_bit(CONNECTING, &con->state)) { |
1899 | /* is anything else pending? */ | |
1900 | if (!list_empty(&con->out_queue)) { | |
1901 | prepare_write_message(con); | |
1902 | goto more; | |
1903 | } | |
1904 | if (con->in_seq > con->in_seq_acked) { | |
1905 | prepare_write_ack(con); | |
1906 | goto more; | |
1907 | } | |
1908 | if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) { | |
1909 | prepare_write_keepalive(con); | |
1910 | goto more; | |
1911 | } | |
1912 | } | |
1913 | ||
1914 | /* Nothing to do! */ | |
1915 | clear_bit(WRITE_PENDING, &con->state); | |
1916 | dout("try_write nothing else to write.\n"); | |
31b8006e SW |
1917 | ret = 0; |
1918 | out: | |
42961d23 | 1919 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
1920 | return ret; |
1921 | } | |
1922 | ||
1923 | ||
1924 | ||
1925 | /* | |
1926 | * Read what we can from the socket. | |
1927 | */ | |
1928 | static int try_read(struct ceph_connection *con) | |
1929 | { | |
31b8006e SW |
1930 | int ret = -1; |
1931 | ||
1932 | if (!con->sock) | |
1933 | return 0; | |
1934 | ||
1935 | if (test_bit(STANDBY, &con->state)) | |
1936 | return 0; | |
1937 | ||
1938 | dout("try_read start on %p\n", con); | |
ec302645 | 1939 | |
31b8006e SW |
1940 | more: |
1941 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, | |
1942 | con->in_base_pos); | |
0da5d703 SW |
1943 | |
1944 | /* | |
1945 | * process_connect and process_message drop and re-take | |
1946 | * con->mutex. make sure we handle a racing close or reopen. | |
1947 | */ | |
1948 | if (test_bit(CLOSED, &con->state) || | |
1949 | test_bit(OPENING, &con->state)) { | |
1950 | ret = -EAGAIN; | |
1951 | goto out; | |
1952 | } | |
1953 | ||
31b8006e | 1954 | if (test_bit(CONNECTING, &con->state)) { |
eed0ef2c SW |
1955 | if (!test_bit(NEGOTIATING, &con->state)) { |
1956 | dout("try_read connecting\n"); | |
1957 | ret = read_partial_banner(con); | |
1958 | if (ret <= 0) | |
eed0ef2c | 1959 | goto out; |
98bdb0aa SW |
1960 | ret = process_banner(con); |
1961 | if (ret < 0) | |
1962 | goto out; | |
eed0ef2c | 1963 | } |
31b8006e SW |
1964 | ret = read_partial_connect(con); |
1965 | if (ret <= 0) | |
31b8006e | 1966 | goto out; |
98bdb0aa SW |
1967 | ret = process_connect(con); |
1968 | if (ret < 0) | |
1969 | goto out; | |
31b8006e SW |
1970 | goto more; |
1971 | } | |
1972 | ||
1973 | if (con->in_base_pos < 0) { | |
1974 | /* | |
1975 | * skipping + discarding content. | |
1976 | * | |
1977 | * FIXME: there must be a better way to do this! | |
1978 | */ | |
1979 | static char buf[1024]; | |
1980 | int skip = min(1024, -con->in_base_pos); | |
1981 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); | |
1982 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
1983 | if (ret <= 0) | |
98bdb0aa | 1984 | goto out; |
31b8006e SW |
1985 | con->in_base_pos += ret; |
1986 | if (con->in_base_pos) | |
1987 | goto more; | |
1988 | } | |
1989 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
1990 | /* | |
1991 | * what's next? | |
1992 | */ | |
1993 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
1994 | if (ret <= 0) | |
98bdb0aa | 1995 | goto out; |
31b8006e SW |
1996 | dout("try_read got tag %d\n", (int)con->in_tag); |
1997 | switch (con->in_tag) { | |
1998 | case CEPH_MSGR_TAG_MSG: | |
1999 | prepare_read_message(con); | |
2000 | break; | |
2001 | case CEPH_MSGR_TAG_ACK: | |
2002 | prepare_read_ack(con); | |
2003 | break; | |
2004 | case CEPH_MSGR_TAG_CLOSE: | |
2005 | set_bit(CLOSED, &con->state); /* fixme */ | |
98bdb0aa | 2006 | goto out; |
31b8006e SW |
2007 | default: |
2008 | goto bad_tag; | |
2009 | } | |
2010 | } | |
2011 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2012 | ret = read_partial_message(con); | |
2013 | if (ret <= 0) { | |
2014 | switch (ret) { | |
2015 | case -EBADMSG: | |
2016 | con->error_msg = "bad crc"; | |
2017 | ret = -EIO; | |
98bdb0aa | 2018 | break; |
31b8006e SW |
2019 | case -EIO: |
2020 | con->error_msg = "io error"; | |
98bdb0aa | 2021 | break; |
31b8006e | 2022 | } |
98bdb0aa | 2023 | goto out; |
31b8006e SW |
2024 | } |
2025 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2026 | goto more; | |
2027 | process_message(con); | |
2028 | goto more; | |
2029 | } | |
2030 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2031 | ret = read_partial_ack(con); | |
2032 | if (ret <= 0) | |
98bdb0aa | 2033 | goto out; |
31b8006e SW |
2034 | process_ack(con); |
2035 | goto more; | |
2036 | } | |
2037 | ||
31b8006e | 2038 | out: |
98bdb0aa | 2039 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2040 | return ret; |
2041 | ||
2042 | bad_tag: | |
2043 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2044 | con->error_msg = "protocol error, garbage tag"; | |
2045 | ret = -1; | |
2046 | goto out; | |
2047 | } | |
2048 | ||
2049 | ||
2050 | /* | |
2051 | * Atomically queue work on a connection. Bump @con reference to | |
2052 | * avoid races with connection teardown. | |
31b8006e SW |
2053 | */ |
2054 | static void queue_con(struct ceph_connection *con) | |
2055 | { | |
2056 | if (test_bit(DEAD, &con->state)) { | |
2057 | dout("queue_con %p ignoring: DEAD\n", | |
2058 | con); | |
2059 | return; | |
2060 | } | |
2061 | ||
2062 | if (!con->ops->get(con)) { | |
2063 | dout("queue_con %p ref count 0\n", con); | |
2064 | return; | |
2065 | } | |
2066 | ||
f363e45f | 2067 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2068 | dout("queue_con %p - already queued\n", con); |
2069 | con->ops->put(con); | |
2070 | } else { | |
2071 | dout("queue_con %p\n", con); | |
2072 | } | |
2073 | } | |
2074 | ||
2075 | /* | |
2076 | * Do some work on a connection. Drop a connection ref when we're done. | |
2077 | */ | |
2078 | static void con_work(struct work_struct *work) | |
2079 | { | |
2080 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2081 | work.work); | |
0da5d703 | 2082 | int ret; |
31b8006e | 2083 | |
9dd4658d | 2084 | mutex_lock(&con->mutex); |
0da5d703 | 2085 | restart: |
60bf8bf8 SW |
2086 | if (test_and_clear_bit(BACKOFF, &con->state)) { |
2087 | dout("con_work %p backing off\n", con); | |
2088 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2089 | round_jiffies_relative(con->delay))) { | |
2090 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2091 | mutex_unlock(&con->mutex); | |
2092 | return; | |
2093 | } else { | |
2094 | con->ops->put(con); | |
2095 | dout("con_work %p FAILED to back off %lu\n", con, | |
2096 | con->delay); | |
2097 | } | |
2098 | } | |
9dd4658d | 2099 | |
e00de341 SW |
2100 | if (test_bit(STANDBY, &con->state)) { |
2101 | dout("con_work %p STANDBY\n", con); | |
2102 | goto done; | |
2103 | } | |
31b8006e SW |
2104 | if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */ |
2105 | dout("con_work CLOSED\n"); | |
2106 | con_close_socket(con); | |
2107 | goto done; | |
2108 | } | |
2109 | if (test_and_clear_bit(OPENING, &con->state)) { | |
2110 | /* reopen w/ new peer */ | |
2111 | dout("con_work OPENING\n"); | |
2112 | con_close_socket(con); | |
2113 | } | |
2114 | ||
0da5d703 SW |
2115 | if (test_and_clear_bit(SOCK_CLOSED, &con->state)) |
2116 | goto fault; | |
2117 | ||
2118 | ret = try_read(con); | |
2119 | if (ret == -EAGAIN) | |
2120 | goto restart; | |
2121 | if (ret < 0) | |
2122 | goto fault; | |
2123 | ||
2124 | ret = try_write(con); | |
2125 | if (ret == -EAGAIN) | |
2126 | goto restart; | |
2127 | if (ret < 0) | |
2128 | goto fault; | |
31b8006e SW |
2129 | |
2130 | done: | |
9dd4658d | 2131 | mutex_unlock(&con->mutex); |
9dd4658d | 2132 | done_unlocked: |
31b8006e | 2133 | con->ops->put(con); |
0da5d703 SW |
2134 | return; |
2135 | ||
2136 | fault: | |
2137 | mutex_unlock(&con->mutex); | |
2138 | ceph_fault(con); /* error/fault path */ | |
2139 | goto done_unlocked; | |
31b8006e SW |
2140 | } |
2141 | ||
2142 | ||
2143 | /* | |
2144 | * Generic error/fault handler. A retry mechanism is used with | |
2145 | * exponential backoff | |
2146 | */ | |
2147 | static void ceph_fault(struct ceph_connection *con) | |
2148 | { | |
2149 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2150 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2151 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2152 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
2153 | |
2154 | if (test_bit(LOSSYTX, &con->state)) { | |
2155 | dout("fault on LOSSYTX channel\n"); | |
2156 | goto out; | |
2157 | } | |
2158 | ||
ec302645 | 2159 | mutex_lock(&con->mutex); |
91e45ce3 SW |
2160 | if (test_bit(CLOSED, &con->state)) |
2161 | goto out_unlock; | |
ec302645 | 2162 | |
31b8006e | 2163 | con_close_socket(con); |
5e095e8b SW |
2164 | |
2165 | if (con->in_msg) { | |
2166 | ceph_msg_put(con->in_msg); | |
2167 | con->in_msg = NULL; | |
2168 | } | |
31b8006e | 2169 | |
e80a52d1 SW |
2170 | /* Requeue anything that hasn't been acked */ |
2171 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2172 | |
e76661d0 SW |
2173 | /* If there are no messages queued or keepalive pending, place |
2174 | * the connection in a STANDBY state */ | |
2175 | if (list_empty(&con->out_queue) && | |
2176 | !test_bit(KEEPALIVE_PENDING, &con->state)) { | |
e00de341 SW |
2177 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
2178 | clear_bit(WRITE_PENDING, &con->state); | |
31b8006e | 2179 | set_bit(STANDBY, &con->state); |
e80a52d1 SW |
2180 | } else { |
2181 | /* retry after a delay. */ | |
2182 | if (con->delay == 0) | |
2183 | con->delay = BASE_DELAY_INTERVAL; | |
2184 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2185 | con->delay *= 2; | |
e80a52d1 SW |
2186 | con->ops->get(con); |
2187 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2188 | round_jiffies_relative(con->delay))) { |
2189 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2190 | } else { | |
e80a52d1 | 2191 | con->ops->put(con); |
60bf8bf8 SW |
2192 | dout("fault failed to queue %p delay %lu, backoff\n", |
2193 | con, con->delay); | |
2194 | /* | |
2195 | * In many cases we see a socket state change | |
2196 | * while con_work is running and end up | |
2197 | * queuing (non-delayed) work, such that we | |
2198 | * can't backoff with a delay. Set a flag so | |
2199 | * that when con_work restarts we schedule the | |
2200 | * delay then. | |
2201 | */ | |
2202 | set_bit(BACKOFF, &con->state); | |
2203 | } | |
31b8006e SW |
2204 | } |
2205 | ||
91e45ce3 SW |
2206 | out_unlock: |
2207 | mutex_unlock(&con->mutex); | |
31b8006e | 2208 | out: |
161fd65a SW |
2209 | /* |
2210 | * in case we faulted due to authentication, invalidate our | |
2211 | * current tickets so that we can get new ones. | |
213c99ee | 2212 | */ |
161fd65a SW |
2213 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2214 | dout("calling invalidate_authorizer()\n"); | |
2215 | con->ops->invalidate_authorizer(con); | |
2216 | } | |
2217 | ||
31b8006e SW |
2218 | if (con->ops->fault) |
2219 | con->ops->fault(con); | |
2220 | } | |
2221 | ||
2222 | ||
2223 | ||
2224 | /* | |
2225 | * create a new messenger instance | |
2226 | */ | |
3d14c5d2 YS |
2227 | struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr, |
2228 | u32 supported_features, | |
2229 | u32 required_features) | |
31b8006e SW |
2230 | { |
2231 | struct ceph_messenger *msgr; | |
2232 | ||
2233 | msgr = kzalloc(sizeof(*msgr), GFP_KERNEL); | |
2234 | if (msgr == NULL) | |
2235 | return ERR_PTR(-ENOMEM); | |
2236 | ||
3d14c5d2 YS |
2237 | msgr->supported_features = supported_features; |
2238 | msgr->required_features = required_features; | |
2239 | ||
31b8006e SW |
2240 | spin_lock_init(&msgr->global_seq_lock); |
2241 | ||
31b8006e SW |
2242 | if (myaddr) |
2243 | msgr->inst.addr = *myaddr; | |
2244 | ||
2245 | /* select a random nonce */ | |
ac8839d7 | 2246 | msgr->inst.addr.type = 0; |
103e2d3a | 2247 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2248 | encode_my_addr(msgr); |
31b8006e SW |
2249 | |
2250 | dout("messenger_create %p\n", msgr); | |
2251 | return msgr; | |
2252 | } | |
3d14c5d2 | 2253 | EXPORT_SYMBOL(ceph_messenger_create); |
31b8006e SW |
2254 | |
2255 | void ceph_messenger_destroy(struct ceph_messenger *msgr) | |
2256 | { | |
2257 | dout("destroy %p\n", msgr); | |
31b8006e SW |
2258 | kfree(msgr); |
2259 | dout("destroyed messenger %p\n", msgr); | |
2260 | } | |
3d14c5d2 | 2261 | EXPORT_SYMBOL(ceph_messenger_destroy); |
31b8006e | 2262 | |
e00de341 SW |
2263 | static void clear_standby(struct ceph_connection *con) |
2264 | { | |
2265 | /* come back from STANDBY? */ | |
2266 | if (test_and_clear_bit(STANDBY, &con->state)) { | |
2267 | mutex_lock(&con->mutex); | |
2268 | dout("clear_standby %p and ++connect_seq\n", con); | |
2269 | con->connect_seq++; | |
2270 | WARN_ON(test_bit(WRITE_PENDING, &con->state)); | |
2271 | WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state)); | |
2272 | mutex_unlock(&con->mutex); | |
2273 | } | |
2274 | } | |
2275 | ||
31b8006e SW |
2276 | /* |
2277 | * Queue up an outgoing message on the given connection. | |
2278 | */ | |
2279 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2280 | { | |
2281 | if (test_bit(CLOSED, &con->state)) { | |
2282 | dout("con_send %p closed, dropping %p\n", con, msg); | |
2283 | ceph_msg_put(msg); | |
2284 | return; | |
2285 | } | |
2286 | ||
2287 | /* set src+dst */ | |
dbad185d | 2288 | msg->hdr.src = con->msgr->inst.name; |
31b8006e | 2289 | |
3ca02ef9 SW |
2290 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
2291 | ||
e84346b7 SW |
2292 | msg->needs_out_seq = true; |
2293 | ||
31b8006e | 2294 | /* queue */ |
ec302645 | 2295 | mutex_lock(&con->mutex); |
31b8006e SW |
2296 | BUG_ON(!list_empty(&msg->list_head)); |
2297 | list_add_tail(&msg->list_head, &con->out_queue); | |
2298 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2299 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2300 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2301 | le32_to_cpu(msg->hdr.front_len), | |
2302 | le32_to_cpu(msg->hdr.middle_len), | |
2303 | le32_to_cpu(msg->hdr.data_len)); | |
ec302645 | 2304 | mutex_unlock(&con->mutex); |
31b8006e SW |
2305 | |
2306 | /* if there wasn't anything waiting to send before, queue | |
2307 | * new work */ | |
e00de341 | 2308 | clear_standby(con); |
31b8006e SW |
2309 | if (test_and_set_bit(WRITE_PENDING, &con->state) == 0) |
2310 | queue_con(con); | |
2311 | } | |
3d14c5d2 | 2312 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2313 | |
2314 | /* | |
2315 | * Revoke a message that was previously queued for send | |
2316 | */ | |
2317 | void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg) | |
2318 | { | |
ec302645 | 2319 | mutex_lock(&con->mutex); |
31b8006e | 2320 | if (!list_empty(&msg->list_head)) { |
ed98adad | 2321 | dout("con_revoke %p msg %p - was on queue\n", con, msg); |
31b8006e SW |
2322 | list_del_init(&msg->list_head); |
2323 | ceph_msg_put(msg); | |
2324 | msg->hdr.seq = 0; | |
ed98adad SW |
2325 | } |
2326 | if (con->out_msg == msg) { | |
2327 | dout("con_revoke %p msg %p - was sending\n", con, msg); | |
2328 | con->out_msg = NULL; | |
31b8006e SW |
2329 | if (con->out_kvec_is_msg) { |
2330 | con->out_skip = con->out_kvec_bytes; | |
2331 | con->out_kvec_is_msg = false; | |
2332 | } | |
ed98adad SW |
2333 | ceph_msg_put(msg); |
2334 | msg->hdr.seq = 0; | |
31b8006e | 2335 | } |
ec302645 | 2336 | mutex_unlock(&con->mutex); |
31b8006e SW |
2337 | } |
2338 | ||
350b1c32 | 2339 | /* |
0d59ab81 | 2340 | * Revoke a message that we may be reading data into |
350b1c32 | 2341 | */ |
0d59ab81 | 2342 | void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg) |
350b1c32 SW |
2343 | { |
2344 | mutex_lock(&con->mutex); | |
0d59ab81 YS |
2345 | if (con->in_msg && con->in_msg == msg) { |
2346 | unsigned front_len = le32_to_cpu(con->in_hdr.front_len); | |
2347 | unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
350b1c32 SW |
2348 | unsigned data_len = le32_to_cpu(con->in_hdr.data_len); |
2349 | ||
2350 | /* skip rest of message */ | |
0d59ab81 | 2351 | dout("con_revoke_pages %p msg %p revoked\n", con, msg); |
350b1c32 SW |
2352 | con->in_base_pos = con->in_base_pos - |
2353 | sizeof(struct ceph_msg_header) - | |
0d59ab81 YS |
2354 | front_len - |
2355 | middle_len - | |
2356 | data_len - | |
350b1c32 | 2357 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2358 | ceph_msg_put(con->in_msg); |
2359 | con->in_msg = NULL; | |
2360 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2361 | con->in_seq++; |
350b1c32 SW |
2362 | } else { |
2363 | dout("con_revoke_pages %p msg %p pages %p no-op\n", | |
0d59ab81 | 2364 | con, con->in_msg, msg); |
350b1c32 SW |
2365 | } |
2366 | mutex_unlock(&con->mutex); | |
2367 | } | |
2368 | ||
31b8006e SW |
2369 | /* |
2370 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2371 | */ | |
2372 | void ceph_con_keepalive(struct ceph_connection *con) | |
2373 | { | |
e00de341 SW |
2374 | dout("con_keepalive %p\n", con); |
2375 | clear_standby(con); | |
31b8006e SW |
2376 | if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 && |
2377 | test_and_set_bit(WRITE_PENDING, &con->state) == 0) | |
2378 | queue_con(con); | |
2379 | } | |
3d14c5d2 | 2380 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2381 | |
2382 | ||
2383 | /* | |
2384 | * construct a new message with given type, size | |
2385 | * the new msg has a ref count of 1. | |
2386 | */ | |
b61c2763 SW |
2387 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2388 | bool can_fail) | |
31b8006e SW |
2389 | { |
2390 | struct ceph_msg *m; | |
2391 | ||
34d23762 | 2392 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2393 | if (m == NULL) |
2394 | goto out; | |
c2e552e7 | 2395 | kref_init(&m->kref); |
31b8006e SW |
2396 | INIT_LIST_HEAD(&m->list_head); |
2397 | ||
45c6ceb5 | 2398 | m->hdr.tid = 0; |
31b8006e | 2399 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2400 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2401 | m->hdr.version = 0; | |
31b8006e SW |
2402 | m->hdr.front_len = cpu_to_le32(front_len); |
2403 | m->hdr.middle_len = 0; | |
bb257664 SW |
2404 | m->hdr.data_len = 0; |
2405 | m->hdr.data_off = 0; | |
45c6ceb5 | 2406 | m->hdr.reserved = 0; |
31b8006e SW |
2407 | m->footer.front_crc = 0; |
2408 | m->footer.middle_crc = 0; | |
2409 | m->footer.data_crc = 0; | |
45c6ceb5 | 2410 | m->footer.flags = 0; |
31b8006e SW |
2411 | m->front_max = front_len; |
2412 | m->front_is_vmalloc = false; | |
2413 | m->more_to_follow = false; | |
c0d5f9db | 2414 | m->ack_stamp = 0; |
31b8006e SW |
2415 | m->pool = NULL; |
2416 | ||
ca20892d HC |
2417 | /* middle */ |
2418 | m->middle = NULL; | |
2419 | ||
2420 | /* data */ | |
2421 | m->nr_pages = 0; | |
2422 | m->page_alignment = 0; | |
2423 | m->pages = NULL; | |
2424 | m->pagelist = NULL; | |
2425 | m->bio = NULL; | |
2426 | m->bio_iter = NULL; | |
2427 | m->bio_seg = 0; | |
2428 | m->trail = NULL; | |
2429 | ||
31b8006e SW |
2430 | /* front */ |
2431 | if (front_len) { | |
2432 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2433 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2434 | PAGE_KERNEL); |
2435 | m->front_is_vmalloc = true; | |
2436 | } else { | |
34d23762 | 2437 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2438 | } |
2439 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2440 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2441 | front_len); |
2442 | goto out2; | |
2443 | } | |
2444 | } else { | |
2445 | m->front.iov_base = NULL; | |
2446 | } | |
2447 | m->front.iov_len = front_len; | |
2448 | ||
bb257664 | 2449 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2450 | return m; |
2451 | ||
2452 | out2: | |
2453 | ceph_msg_put(m); | |
2454 | out: | |
b61c2763 SW |
2455 | if (!can_fail) { |
2456 | pr_err("msg_new can't create type %d front %d\n", type, | |
2457 | front_len); | |
f0ed1b7c | 2458 | WARN_ON(1); |
b61c2763 SW |
2459 | } else { |
2460 | dout("msg_new can't create type %d front %d\n", type, | |
2461 | front_len); | |
2462 | } | |
a79832f2 | 2463 | return NULL; |
31b8006e | 2464 | } |
3d14c5d2 | 2465 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2466 | |
31b8006e SW |
2467 | /* |
2468 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2469 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2470 | * per-type header in the front and then gracefully fail (i.e., | |
2471 | * propagate the error to the caller based on info in the front) when | |
2472 | * the middle is too large. | |
2473 | */ | |
2450418c | 2474 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2475 | { |
2476 | int type = le16_to_cpu(msg->hdr.type); | |
2477 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2478 | ||
2479 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2480 | ceph_msg_type_name(type), middle_len); | |
2481 | BUG_ON(!middle_len); | |
2482 | BUG_ON(msg->middle); | |
2483 | ||
b6c1d5b8 | 2484 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2485 | if (!msg->middle) |
2486 | return -ENOMEM; | |
2487 | return 0; | |
2488 | } | |
2489 | ||
2450418c YS |
2490 | /* |
2491 | * Generic message allocator, for incoming messages. | |
2492 | */ | |
2493 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, | |
2494 | struct ceph_msg_header *hdr, | |
2495 | int *skip) | |
2496 | { | |
2497 | int type = le16_to_cpu(hdr->type); | |
2498 | int front_len = le32_to_cpu(hdr->front_len); | |
2499 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2500 | struct ceph_msg *msg = NULL; | |
2501 | int ret; | |
2502 | ||
2503 | if (con->ops->alloc_msg) { | |
0547a9b3 | 2504 | mutex_unlock(&con->mutex); |
2450418c | 2505 | msg = con->ops->alloc_msg(con, hdr, skip); |
0547a9b3 | 2506 | mutex_lock(&con->mutex); |
a79832f2 | 2507 | if (!msg || *skip) |
2450418c YS |
2508 | return NULL; |
2509 | } | |
2510 | if (!msg) { | |
2511 | *skip = 0; | |
b61c2763 | 2512 | msg = ceph_msg_new(type, front_len, GFP_NOFS, false); |
2450418c YS |
2513 | if (!msg) { |
2514 | pr_err("unable to allocate msg type %d len %d\n", | |
2515 | type, front_len); | |
a79832f2 | 2516 | return NULL; |
2450418c | 2517 | } |
c5c6b19d | 2518 | msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2519 | } |
9d7f0f13 | 2520 | memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2521 | |
bb257664 | 2522 | if (middle_len && !msg->middle) { |
2450418c | 2523 | ret = ceph_alloc_middle(con, msg); |
2450418c YS |
2524 | if (ret < 0) { |
2525 | ceph_msg_put(msg); | |
a79832f2 | 2526 | return NULL; |
2450418c YS |
2527 | } |
2528 | } | |
9d7f0f13 | 2529 | |
2450418c YS |
2530 | return msg; |
2531 | } | |
2532 | ||
31b8006e SW |
2533 | |
2534 | /* | |
2535 | * Free a generically kmalloc'd message. | |
2536 | */ | |
2537 | void ceph_msg_kfree(struct ceph_msg *m) | |
2538 | { | |
2539 | dout("msg_kfree %p\n", m); | |
2540 | if (m->front_is_vmalloc) | |
2541 | vfree(m->front.iov_base); | |
2542 | else | |
2543 | kfree(m->front.iov_base); | |
2544 | kfree(m); | |
2545 | } | |
2546 | ||
2547 | /* | |
2548 | * Drop a msg ref. Destroy as needed. | |
2549 | */ | |
c2e552e7 SW |
2550 | void ceph_msg_last_put(struct kref *kref) |
2551 | { | |
2552 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2553 | |
c2e552e7 SW |
2554 | dout("ceph_msg_put last one on %p\n", m); |
2555 | WARN_ON(!list_empty(&m->list_head)); | |
2556 | ||
2557 | /* drop middle, data, if any */ | |
2558 | if (m->middle) { | |
2559 | ceph_buffer_put(m->middle); | |
2560 | m->middle = NULL; | |
31b8006e | 2561 | } |
c2e552e7 SW |
2562 | m->nr_pages = 0; |
2563 | m->pages = NULL; | |
2564 | ||
58bb3b37 SW |
2565 | if (m->pagelist) { |
2566 | ceph_pagelist_release(m->pagelist); | |
2567 | kfree(m->pagelist); | |
2568 | m->pagelist = NULL; | |
2569 | } | |
2570 | ||
68b4476b YS |
2571 | m->trail = NULL; |
2572 | ||
c2e552e7 SW |
2573 | if (m->pool) |
2574 | ceph_msgpool_put(m->pool, m); | |
2575 | else | |
2576 | ceph_msg_kfree(m); | |
31b8006e | 2577 | } |
3d14c5d2 | 2578 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2579 | |
2580 | void ceph_msg_dump(struct ceph_msg *msg) | |
2581 | { | |
2582 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2583 | msg->front_max, msg->nr_pages); | |
2584 | print_hex_dump(KERN_DEBUG, "header: ", | |
2585 | DUMP_PREFIX_OFFSET, 16, 1, | |
2586 | &msg->hdr, sizeof(msg->hdr), true); | |
2587 | print_hex_dump(KERN_DEBUG, " front: ", | |
2588 | DUMP_PREFIX_OFFSET, 16, 1, | |
2589 | msg->front.iov_base, msg->front.iov_len, true); | |
2590 | if (msg->middle) | |
2591 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2592 | DUMP_PREFIX_OFFSET, 16, 1, | |
2593 | msg->middle->vec.iov_base, | |
2594 | msg->middle->vec.iov_len, true); | |
2595 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2596 | DUMP_PREFIX_OFFSET, 16, 1, | |
2597 | &msg->footer, sizeof(msg->footer), true); | |
2598 | } | |
3d14c5d2 | 2599 | EXPORT_SYMBOL(ceph_msg_dump); |