]>
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 | ||
84495f49 AE |
41 | /* |
42 | * When skipping (ignoring) a block of input we read it into a "skip | |
43 | * buffer," which is this many bytes in size. | |
44 | */ | |
45 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
46 | |
47 | static void queue_con(struct ceph_connection *con); | |
48 | static void con_work(struct work_struct *); | |
49 | static void ceph_fault(struct ceph_connection *con); | |
50 | ||
31b8006e | 51 | /* |
f64a9317 AE |
52 | * Nicely render a sockaddr as a string. An array of formatted |
53 | * strings is used, to approximate reentrancy. | |
31b8006e | 54 | */ |
f64a9317 AE |
55 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
56 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
57 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
58 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
59 | ||
60 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
61 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 62 | |
57666519 | 63 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 64 | |
3d14c5d2 | 65 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
66 | { |
67 | int i; | |
68 | char *s; | |
99f0f3b2 AE |
69 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
70 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 71 | |
f64a9317 | 72 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
73 | s = addr_str[i]; |
74 | ||
75 | switch (ss->ss_family) { | |
76 | case AF_INET: | |
bd406145 AE |
77 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
78 | ntohs(in4->sin_port)); | |
31b8006e SW |
79 | break; |
80 | ||
81 | case AF_INET6: | |
bd406145 AE |
82 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
83 | ntohs(in6->sin6_port)); | |
31b8006e SW |
84 | break; |
85 | ||
86 | default: | |
d3002b97 AE |
87 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
88 | ss->ss_family); | |
31b8006e SW |
89 | } |
90 | ||
91 | return s; | |
92 | } | |
3d14c5d2 | 93 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 94 | |
63f2d211 SW |
95 | static void encode_my_addr(struct ceph_messenger *msgr) |
96 | { | |
97 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
98 | ceph_encode_addr(&msgr->my_enc_addr); | |
99 | } | |
100 | ||
31b8006e SW |
101 | /* |
102 | * work queue for all reading and writing to/from the socket. | |
103 | */ | |
e0f43c94 | 104 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 105 | |
6173d1f0 AE |
106 | void _ceph_msgr_exit(void) |
107 | { | |
d3002b97 | 108 | if (ceph_msgr_wq) { |
6173d1f0 | 109 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
110 | ceph_msgr_wq = NULL; |
111 | } | |
6173d1f0 | 112 | |
6173d1f0 AE |
113 | BUG_ON(zero_page == NULL); |
114 | kunmap(zero_page); | |
115 | page_cache_release(zero_page); | |
116 | zero_page = NULL; | |
117 | } | |
118 | ||
3d14c5d2 | 119 | int ceph_msgr_init(void) |
31b8006e | 120 | { |
57666519 AE |
121 | BUG_ON(zero_page != NULL); |
122 | zero_page = ZERO_PAGE(0); | |
123 | page_cache_get(zero_page); | |
124 | ||
f363e45f | 125 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0); |
6173d1f0 AE |
126 | if (ceph_msgr_wq) |
127 | return 0; | |
57666519 | 128 | |
6173d1f0 AE |
129 | pr_err("msgr_init failed to create workqueue\n"); |
130 | _ceph_msgr_exit(); | |
57666519 | 131 | |
6173d1f0 | 132 | return -ENOMEM; |
31b8006e | 133 | } |
3d14c5d2 | 134 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
135 | |
136 | void ceph_msgr_exit(void) | |
137 | { | |
57666519 | 138 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 139 | |
6173d1f0 | 140 | _ceph_msgr_exit(); |
31b8006e | 141 | } |
3d14c5d2 | 142 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 143 | |
cd84db6e | 144 | void ceph_msgr_flush(void) |
a922d38f SW |
145 | { |
146 | flush_workqueue(ceph_msgr_wq); | |
147 | } | |
3d14c5d2 | 148 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f SW |
149 | |
150 | ||
31b8006e SW |
151 | /* |
152 | * socket callback functions | |
153 | */ | |
154 | ||
155 | /* data available on socket, or listen socket received a connect */ | |
156 | static void ceph_data_ready(struct sock *sk, int count_unused) | |
157 | { | |
bd406145 AE |
158 | struct ceph_connection *con = sk->sk_user_data; |
159 | ||
31b8006e SW |
160 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
161 | dout("ceph_data_ready on %p state = %lu, queueing work\n", | |
162 | con, con->state); | |
163 | queue_con(con); | |
164 | } | |
165 | } | |
166 | ||
167 | /* socket has buffer space for writing */ | |
168 | static void ceph_write_space(struct sock *sk) | |
169 | { | |
d3002b97 | 170 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 171 | |
182fac26 JS |
172 | /* only queue to workqueue if there is data we want to write, |
173 | * and there is sufficient space in the socket buffer to accept | |
174 | * more data. clear SOCK_NOSPACE so that ceph_write_space() | |
175 | * doesn't get called again until try_write() fills the socket | |
176 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
177 | * and net/core/stream.c:sk_stream_write_space(). | |
178 | */ | |
31b8006e | 179 | if (test_bit(WRITE_PENDING, &con->state)) { |
182fac26 JS |
180 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
181 | dout("ceph_write_space %p queueing write work\n", con); | |
182 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
183 | queue_con(con); | |
184 | } | |
31b8006e SW |
185 | } else { |
186 | dout("ceph_write_space %p nothing to write\n", con); | |
187 | } | |
31b8006e SW |
188 | } |
189 | ||
190 | /* socket's state has changed */ | |
191 | static void ceph_state_change(struct sock *sk) | |
192 | { | |
bd406145 | 193 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e SW |
194 | |
195 | dout("ceph_state_change %p state = %lu sk_state = %u\n", | |
196 | con, con->state, sk->sk_state); | |
197 | ||
198 | if (test_bit(CLOSED, &con->state)) | |
199 | return; | |
200 | ||
201 | switch (sk->sk_state) { | |
202 | case TCP_CLOSE: | |
203 | dout("ceph_state_change TCP_CLOSE\n"); | |
204 | case TCP_CLOSE_WAIT: | |
205 | dout("ceph_state_change TCP_CLOSE_WAIT\n"); | |
206 | if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) { | |
207 | if (test_bit(CONNECTING, &con->state)) | |
208 | con->error_msg = "connection failed"; | |
209 | else | |
210 | con->error_msg = "socket closed"; | |
211 | queue_con(con); | |
212 | } | |
213 | break; | |
214 | case TCP_ESTABLISHED: | |
215 | dout("ceph_state_change TCP_ESTABLISHED\n"); | |
216 | queue_con(con); | |
217 | break; | |
d3002b97 AE |
218 | default: /* Everything else is uninteresting */ |
219 | break; | |
31b8006e SW |
220 | } |
221 | } | |
222 | ||
223 | /* | |
224 | * set up socket callbacks | |
225 | */ | |
226 | static void set_sock_callbacks(struct socket *sock, | |
227 | struct ceph_connection *con) | |
228 | { | |
229 | struct sock *sk = sock->sk; | |
bd406145 | 230 | sk->sk_user_data = con; |
31b8006e SW |
231 | sk->sk_data_ready = ceph_data_ready; |
232 | sk->sk_write_space = ceph_write_space; | |
233 | sk->sk_state_change = ceph_state_change; | |
234 | } | |
235 | ||
236 | ||
237 | /* | |
238 | * socket helpers | |
239 | */ | |
240 | ||
241 | /* | |
242 | * initiate connection to a remote socket. | |
243 | */ | |
41617d0c | 244 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 245 | { |
f91d3471 | 246 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
247 | struct socket *sock; |
248 | int ret; | |
249 | ||
250 | BUG_ON(con->sock); | |
f91d3471 SW |
251 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
252 | IPPROTO_TCP, &sock); | |
31b8006e | 253 | if (ret) |
41617d0c | 254 | return ret; |
31b8006e SW |
255 | sock->sk->sk_allocation = GFP_NOFS; |
256 | ||
a6a5349d SW |
257 | #ifdef CONFIG_LOCKDEP |
258 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
259 | #endif | |
260 | ||
31b8006e SW |
261 | set_sock_callbacks(sock, con); |
262 | ||
3d14c5d2 | 263 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 264 | |
f91d3471 SW |
265 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
266 | O_NONBLOCK); | |
31b8006e SW |
267 | if (ret == -EINPROGRESS) { |
268 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 269 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 270 | sock->sk->sk_state); |
a5bc3129 | 271 | } else if (ret < 0) { |
31b8006e | 272 | pr_err("connect %s error %d\n", |
3d14c5d2 | 273 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 274 | sock_release(sock); |
31b8006e | 275 | con->error_msg = "connect error"; |
31b8006e | 276 | |
41617d0c | 277 | return ret; |
a5bc3129 AE |
278 | } |
279 | con->sock = sock; | |
280 | ||
41617d0c | 281 | return 0; |
31b8006e SW |
282 | } |
283 | ||
284 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
285 | { | |
286 | struct kvec iov = {buf, len}; | |
287 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 288 | int r; |
31b8006e | 289 | |
98bdb0aa SW |
290 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
291 | if (r == -EAGAIN) | |
292 | r = 0; | |
293 | return r; | |
31b8006e SW |
294 | } |
295 | ||
296 | /* | |
297 | * write something. @more is true if caller will be sending more data | |
298 | * shortly. | |
299 | */ | |
300 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
301 | size_t kvlen, size_t len, int more) | |
302 | { | |
303 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 304 | int r; |
31b8006e SW |
305 | |
306 | if (more) | |
307 | msg.msg_flags |= MSG_MORE; | |
308 | else | |
309 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
310 | ||
42961d23 SW |
311 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
312 | if (r == -EAGAIN) | |
313 | r = 0; | |
314 | return r; | |
31b8006e SW |
315 | } |
316 | ||
31739139 AE |
317 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
318 | int offset, size_t size, int more) | |
319 | { | |
320 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
321 | int ret; | |
322 | ||
323 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
324 | if (ret == -EAGAIN) | |
325 | ret = 0; | |
326 | ||
327 | return ret; | |
328 | } | |
329 | ||
31b8006e SW |
330 | |
331 | /* | |
332 | * Shutdown/close the socket for the given connection. | |
333 | */ | |
334 | static int con_close_socket(struct ceph_connection *con) | |
335 | { | |
336 | int rc; | |
337 | ||
338 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
339 | if (!con->sock) | |
340 | return 0; | |
341 | set_bit(SOCK_CLOSED, &con->state); | |
342 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
343 | sock_release(con->sock); | |
344 | con->sock = NULL; | |
345 | clear_bit(SOCK_CLOSED, &con->state); | |
346 | return rc; | |
347 | } | |
348 | ||
349 | /* | |
350 | * Reset a connection. Discard all incoming and outgoing messages | |
351 | * and clear *_seq state. | |
352 | */ | |
353 | static void ceph_msg_remove(struct ceph_msg *msg) | |
354 | { | |
355 | list_del_init(&msg->list_head); | |
356 | ceph_msg_put(msg); | |
357 | } | |
358 | static void ceph_msg_remove_list(struct list_head *head) | |
359 | { | |
360 | while (!list_empty(head)) { | |
361 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
362 | list_head); | |
363 | ceph_msg_remove(msg); | |
364 | } | |
365 | } | |
366 | ||
367 | static void reset_connection(struct ceph_connection *con) | |
368 | { | |
369 | /* reset connection, out_queue, msg_ and connect_seq */ | |
370 | /* discard existing out_queue and msg_seq */ | |
31b8006e SW |
371 | ceph_msg_remove_list(&con->out_queue); |
372 | ceph_msg_remove_list(&con->out_sent); | |
373 | ||
cf3e5c40 SW |
374 | if (con->in_msg) { |
375 | ceph_msg_put(con->in_msg); | |
376 | con->in_msg = NULL; | |
377 | } | |
378 | ||
31b8006e SW |
379 | con->connect_seq = 0; |
380 | con->out_seq = 0; | |
c86a2930 SW |
381 | if (con->out_msg) { |
382 | ceph_msg_put(con->out_msg); | |
383 | con->out_msg = NULL; | |
384 | } | |
31b8006e | 385 | con->in_seq = 0; |
0e0d5e0c | 386 | con->in_seq_acked = 0; |
31b8006e SW |
387 | } |
388 | ||
389 | /* | |
390 | * mark a peer down. drop any open connections. | |
391 | */ | |
392 | void ceph_con_close(struct ceph_connection *con) | |
393 | { | |
3d14c5d2 YS |
394 | dout("con_close %p peer %s\n", con, |
395 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
31b8006e SW |
396 | set_bit(CLOSED, &con->state); /* in case there's queued work */ |
397 | clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */ | |
1679f876 SW |
398 | clear_bit(LOSSYTX, &con->state); /* so we retry next connect */ |
399 | clear_bit(KEEPALIVE_PENDING, &con->state); | |
400 | clear_bit(WRITE_PENDING, &con->state); | |
ec302645 | 401 | mutex_lock(&con->mutex); |
31b8006e | 402 | reset_connection(con); |
6f2bc3ff | 403 | con->peer_global_seq = 0; |
91e45ce3 | 404 | cancel_delayed_work(&con->work); |
ec302645 | 405 | mutex_unlock(&con->mutex); |
31b8006e SW |
406 | queue_con(con); |
407 | } | |
3d14c5d2 | 408 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 409 | |
31b8006e SW |
410 | /* |
411 | * Reopen a closed connection, with a new peer address. | |
412 | */ | |
413 | void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr) | |
414 | { | |
3d14c5d2 | 415 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
31b8006e SW |
416 | set_bit(OPENING, &con->state); |
417 | clear_bit(CLOSED, &con->state); | |
418 | memcpy(&con->peer_addr, addr, sizeof(*addr)); | |
03c677e1 | 419 | con->delay = 0; /* reset backoff memory */ |
31b8006e SW |
420 | queue_con(con); |
421 | } | |
3d14c5d2 | 422 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 423 | |
87b315a5 SW |
424 | /* |
425 | * return true if this connection ever successfully opened | |
426 | */ | |
427 | bool ceph_con_opened(struct ceph_connection *con) | |
428 | { | |
429 | return con->connect_seq > 0; | |
430 | } | |
431 | ||
31b8006e SW |
432 | /* |
433 | * generic get/put | |
434 | */ | |
435 | struct ceph_connection *ceph_con_get(struct ceph_connection *con) | |
436 | { | |
d3002b97 AE |
437 | int nref = __atomic_add_unless(&con->nref, 1, 0); |
438 | ||
439 | dout("con_get %p nref = %d -> %d\n", con, nref, nref + 1); | |
440 | ||
441 | return nref ? con : NULL; | |
31b8006e SW |
442 | } |
443 | ||
444 | void ceph_con_put(struct ceph_connection *con) | |
445 | { | |
d3002b97 AE |
446 | int nref = atomic_dec_return(&con->nref); |
447 | ||
448 | BUG_ON(nref < 0); | |
449 | if (nref == 0) { | |
71ececda | 450 | BUG_ON(con->sock); |
31b8006e SW |
451 | kfree(con); |
452 | } | |
d3002b97 | 453 | dout("con_put %p nref = %d -> %d\n", con, nref + 1, nref); |
31b8006e SW |
454 | } |
455 | ||
456 | /* | |
457 | * initialize a new connection. | |
458 | */ | |
459 | void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con) | |
460 | { | |
461 | dout("con_init %p\n", con); | |
462 | memset(con, 0, sizeof(*con)); | |
463 | atomic_set(&con->nref, 1); | |
464 | con->msgr = msgr; | |
ec302645 | 465 | mutex_init(&con->mutex); |
31b8006e SW |
466 | INIT_LIST_HEAD(&con->out_queue); |
467 | INIT_LIST_HEAD(&con->out_sent); | |
468 | INIT_DELAYED_WORK(&con->work, con_work); | |
469 | } | |
3d14c5d2 | 470 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
471 | |
472 | ||
473 | /* | |
474 | * We maintain a global counter to order connection attempts. Get | |
475 | * a unique seq greater than @gt. | |
476 | */ | |
477 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
478 | { | |
479 | u32 ret; | |
480 | ||
481 | spin_lock(&msgr->global_seq_lock); | |
482 | if (msgr->global_seq < gt) | |
483 | msgr->global_seq = gt; | |
484 | ret = ++msgr->global_seq; | |
485 | spin_unlock(&msgr->global_seq_lock); | |
486 | return ret; | |
487 | } | |
488 | ||
859eb799 AE |
489 | static void ceph_con_out_kvec_reset(struct ceph_connection *con) |
490 | { | |
491 | con->out_kvec_left = 0; | |
492 | con->out_kvec_bytes = 0; | |
493 | con->out_kvec_cur = &con->out_kvec[0]; | |
494 | } | |
495 | ||
496 | static void ceph_con_out_kvec_add(struct ceph_connection *con, | |
497 | size_t size, void *data) | |
498 | { | |
499 | int index; | |
500 | ||
501 | index = con->out_kvec_left; | |
502 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
503 | ||
504 | con->out_kvec[index].iov_len = size; | |
505 | con->out_kvec[index].iov_base = data; | |
506 | con->out_kvec_left++; | |
507 | con->out_kvec_bytes += size; | |
508 | } | |
31b8006e SW |
509 | |
510 | /* | |
511 | * Prepare footer for currently outgoing message, and finish things | |
512 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
513 | */ | |
859eb799 | 514 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
515 | { |
516 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 517 | int v = con->out_kvec_left; |
31b8006e SW |
518 | |
519 | dout("prepare_write_message_footer %p\n", con); | |
520 | con->out_kvec_is_msg = true; | |
521 | con->out_kvec[v].iov_base = &m->footer; | |
522 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
523 | con->out_kvec_bytes += sizeof(m->footer); | |
524 | con->out_kvec_left++; | |
525 | con->out_more = m->more_to_follow; | |
c86a2930 | 526 | con->out_msg_done = true; |
31b8006e SW |
527 | } |
528 | ||
529 | /* | |
530 | * Prepare headers for the next outgoing message. | |
531 | */ | |
532 | static void prepare_write_message(struct ceph_connection *con) | |
533 | { | |
534 | struct ceph_msg *m; | |
a9a0c51a | 535 | u32 crc; |
31b8006e | 536 | |
859eb799 | 537 | ceph_con_out_kvec_reset(con); |
31b8006e | 538 | con->out_kvec_is_msg = true; |
c86a2930 | 539 | con->out_msg_done = false; |
31b8006e SW |
540 | |
541 | /* Sneak an ack in there first? If we can get it into the same | |
542 | * TCP packet that's a good thing. */ | |
543 | if (con->in_seq > con->in_seq_acked) { | |
544 | con->in_seq_acked = con->in_seq; | |
859eb799 | 545 | ceph_con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 546 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
859eb799 AE |
547 | ceph_con_out_kvec_add(con, sizeof (con->out_temp_ack), |
548 | &con->out_temp_ack); | |
31b8006e SW |
549 | } |
550 | ||
859eb799 | 551 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 552 | con->out_msg = m; |
4cf9d544 SW |
553 | |
554 | /* put message on sent list */ | |
555 | ceph_msg_get(m); | |
556 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 557 | |
e84346b7 SW |
558 | /* |
559 | * only assign outgoing seq # if we haven't sent this message | |
560 | * yet. if it is requeued, resend with it's original seq. | |
561 | */ | |
562 | if (m->needs_out_seq) { | |
563 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
564 | m->needs_out_seq = false; | |
565 | } | |
31b8006e SW |
566 | |
567 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
568 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
569 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
570 | le32_to_cpu(m->hdr.data_len), | |
571 | m->nr_pages); | |
572 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); | |
573 | ||
574 | /* tag + hdr + front + middle */ | |
859eb799 AE |
575 | ceph_con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
576 | ceph_con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
577 | ceph_con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
578 | ||
31b8006e | 579 | if (m->middle) |
859eb799 AE |
580 | ceph_con_out_kvec_add(con, m->middle->vec.iov_len, |
581 | m->middle->vec.iov_base); | |
31b8006e SW |
582 | |
583 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
584 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
585 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
31b8006e | 586 | con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE; |
a9a0c51a AE |
587 | |
588 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
589 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
590 | if (m->middle) { | |
591 | crc = crc32c(0, m->middle->vec.iov_base, | |
592 | m->middle->vec.iov_len); | |
593 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
594 | } else | |
31b8006e SW |
595 | con->out_msg->footer.middle_crc = 0; |
596 | con->out_msg->footer.data_crc = 0; | |
597 | dout("prepare_write_message front_crc %u data_crc %u\n", | |
598 | le32_to_cpu(con->out_msg->footer.front_crc), | |
599 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
600 | ||
601 | /* is there a data payload? */ | |
602 | if (le32_to_cpu(m->hdr.data_len) > 0) { | |
603 | /* initialize page iterator */ | |
604 | con->out_msg_pos.page = 0; | |
68b4476b | 605 | if (m->pages) |
c5c6b19d | 606 | con->out_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
607 | else |
608 | con->out_msg_pos.page_pos = 0; | |
31b8006e | 609 | con->out_msg_pos.data_pos = 0; |
bca064d2 | 610 | con->out_msg_pos.did_page_crc = false; |
31b8006e SW |
611 | con->out_more = 1; /* data + footer will follow */ |
612 | } else { | |
613 | /* no, queue up footer too and be done */ | |
859eb799 | 614 | prepare_write_message_footer(con); |
31b8006e SW |
615 | } |
616 | ||
617 | set_bit(WRITE_PENDING, &con->state); | |
618 | } | |
619 | ||
620 | /* | |
621 | * Prepare an ack. | |
622 | */ | |
623 | static void prepare_write_ack(struct ceph_connection *con) | |
624 | { | |
625 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
626 | con->in_seq_acked, con->in_seq); | |
627 | con->in_seq_acked = con->in_seq; | |
628 | ||
859eb799 AE |
629 | ceph_con_out_kvec_reset(con); |
630 | ||
631 | ceph_con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); | |
632 | ||
31b8006e | 633 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
859eb799 AE |
634 | ceph_con_out_kvec_add(con, sizeof (con->out_temp_ack), |
635 | &con->out_temp_ack); | |
636 | ||
31b8006e SW |
637 | con->out_more = 1; /* more will follow.. eventually.. */ |
638 | set_bit(WRITE_PENDING, &con->state); | |
639 | } | |
640 | ||
641 | /* | |
642 | * Prepare to write keepalive byte. | |
643 | */ | |
644 | static void prepare_write_keepalive(struct ceph_connection *con) | |
645 | { | |
646 | dout("prepare_write_keepalive %p\n", con); | |
859eb799 AE |
647 | ceph_con_out_kvec_reset(con); |
648 | ceph_con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
31b8006e SW |
649 | set_bit(WRITE_PENDING, &con->state); |
650 | } | |
651 | ||
652 | /* | |
653 | * Connection negotiation. | |
654 | */ | |
655 | ||
0da5d703 | 656 | static int prepare_connect_authorizer(struct ceph_connection *con) |
4e7a5dcd SW |
657 | { |
658 | void *auth_buf; | |
659 | int auth_len = 0; | |
660 | int auth_protocol = 0; | |
661 | ||
ec302645 | 662 | mutex_unlock(&con->mutex); |
4e7a5dcd SW |
663 | if (con->ops->get_authorizer) |
664 | con->ops->get_authorizer(con, &auth_buf, &auth_len, | |
665 | &auth_protocol, &con->auth_reply_buf, | |
666 | &con->auth_reply_buf_len, | |
667 | con->auth_retry); | |
ec302645 | 668 | mutex_lock(&con->mutex); |
4e7a5dcd | 669 | |
0da5d703 SW |
670 | if (test_bit(CLOSED, &con->state) || |
671 | test_bit(OPENING, &con->state)) | |
672 | return -EAGAIN; | |
673 | ||
4e7a5dcd SW |
674 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol); |
675 | con->out_connect.authorizer_len = cpu_to_le32(auth_len); | |
676 | ||
859eb799 AE |
677 | if (auth_len) |
678 | ceph_con_out_kvec_add(con, auth_len, auth_buf); | |
679 | ||
0da5d703 | 680 | return 0; |
4e7a5dcd SW |
681 | } |
682 | ||
31b8006e SW |
683 | /* |
684 | * We connected to a peer and are saying hello. | |
685 | */ | |
e825a66d | 686 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 687 | { |
859eb799 | 688 | ceph_con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
e825a66d AE |
689 | ceph_con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), |
690 | &con->msgr->my_enc_addr); | |
eed0ef2c | 691 | |
eed0ef2c SW |
692 | con->out_more = 0; |
693 | set_bit(WRITE_PENDING, &con->state); | |
694 | } | |
695 | ||
e825a66d | 696 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 697 | { |
31b8006e SW |
698 | unsigned global_seq = get_global_seq(con->msgr, 0); |
699 | int proto; | |
e10c758e | 700 | int ret; |
31b8006e SW |
701 | |
702 | switch (con->peer_name.type) { | |
703 | case CEPH_ENTITY_TYPE_MON: | |
704 | proto = CEPH_MONC_PROTOCOL; | |
705 | break; | |
706 | case CEPH_ENTITY_TYPE_OSD: | |
707 | proto = CEPH_OSDC_PROTOCOL; | |
708 | break; | |
709 | case CEPH_ENTITY_TYPE_MDS: | |
710 | proto = CEPH_MDSC_PROTOCOL; | |
711 | break; | |
712 | default: | |
713 | BUG(); | |
714 | } | |
715 | ||
716 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
717 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 718 | |
e825a66d | 719 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
720 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
721 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
722 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
723 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
724 | con->out_connect.flags = 0; | |
31b8006e | 725 | |
859eb799 | 726 | ceph_con_out_kvec_add(con, sizeof (con->out_connect), &con->out_connect); |
e10c758e AE |
727 | ret = prepare_connect_authorizer(con); |
728 | if (ret) | |
729 | return ret; | |
859eb799 | 730 | |
31b8006e SW |
731 | con->out_more = 0; |
732 | set_bit(WRITE_PENDING, &con->state); | |
4e7a5dcd | 733 | |
e10c758e | 734 | return 0; |
31b8006e SW |
735 | } |
736 | ||
31b8006e SW |
737 | /* |
738 | * write as much of pending kvecs to the socket as we can. | |
739 | * 1 -> done | |
740 | * 0 -> socket full, but more to do | |
741 | * <0 -> error | |
742 | */ | |
743 | static int write_partial_kvec(struct ceph_connection *con) | |
744 | { | |
745 | int ret; | |
746 | ||
747 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
748 | while (con->out_kvec_bytes > 0) { | |
749 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
750 | con->out_kvec_left, con->out_kvec_bytes, | |
751 | con->out_more); | |
752 | if (ret <= 0) | |
753 | goto out; | |
754 | con->out_kvec_bytes -= ret; | |
755 | if (con->out_kvec_bytes == 0) | |
756 | break; /* done */ | |
f42299e6 AE |
757 | |
758 | /* account for full iov entries consumed */ | |
759 | while (ret >= con->out_kvec_cur->iov_len) { | |
760 | BUG_ON(!con->out_kvec_left); | |
761 | ret -= con->out_kvec_cur->iov_len; | |
762 | con->out_kvec_cur++; | |
763 | con->out_kvec_left--; | |
764 | } | |
765 | /* and for a partially-consumed entry */ | |
766 | if (ret) { | |
767 | con->out_kvec_cur->iov_len -= ret; | |
768 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
769 | } |
770 | } | |
771 | con->out_kvec_left = 0; | |
772 | con->out_kvec_is_msg = false; | |
773 | ret = 1; | |
774 | out: | |
775 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
776 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
777 | return ret; /* done! */ | |
778 | } | |
779 | ||
68b4476b YS |
780 | #ifdef CONFIG_BLOCK |
781 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
782 | { | |
783 | if (!bio) { | |
784 | *iter = NULL; | |
785 | *seg = 0; | |
786 | return; | |
787 | } | |
788 | *iter = bio; | |
789 | *seg = bio->bi_idx; | |
790 | } | |
791 | ||
792 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
793 | { | |
794 | if (*bio_iter == NULL) | |
795 | return; | |
796 | ||
797 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
798 | ||
799 | (*seg)++; | |
800 | if (*seg == (*bio_iter)->bi_vcnt) | |
801 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
802 | } | |
803 | #endif | |
804 | ||
31b8006e SW |
805 | /* |
806 | * Write as much message data payload as we can. If we finish, queue | |
807 | * up the footer. | |
808 | * 1 -> done, footer is now queued in out_kvec[]. | |
809 | * 0 -> socket full, but more to do | |
810 | * <0 -> error | |
811 | */ | |
812 | static int write_partial_msg_pages(struct ceph_connection *con) | |
813 | { | |
814 | struct ceph_msg *msg = con->out_msg; | |
815 | unsigned data_len = le32_to_cpu(msg->hdr.data_len); | |
816 | size_t len; | |
37675b0f | 817 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 818 | int ret; |
68b4476b YS |
819 | int total_max_write; |
820 | int in_trail = 0; | |
821 | size_t trail_len = (msg->trail ? msg->trail->length : 0); | |
31b8006e SW |
822 | |
823 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
824 | con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages, | |
825 | con->out_msg_pos.page_pos); | |
826 | ||
68b4476b YS |
827 | #ifdef CONFIG_BLOCK |
828 | if (msg->bio && !msg->bio_iter) | |
829 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); | |
830 | #endif | |
831 | ||
832 | while (data_len > con->out_msg_pos.data_pos) { | |
31b8006e | 833 | struct page *page = NULL; |
68b4476b | 834 | int max_write = PAGE_SIZE; |
9bd19663 | 835 | int bio_offset = 0; |
68b4476b YS |
836 | |
837 | total_max_write = data_len - trail_len - | |
838 | con->out_msg_pos.data_pos; | |
31b8006e SW |
839 | |
840 | /* | |
841 | * if we are calculating the data crc (the default), we need | |
842 | * to map the page. if our pages[] has been revoked, use the | |
843 | * zero page. | |
844 | */ | |
68b4476b YS |
845 | |
846 | /* have we reached the trail part of the data? */ | |
847 | if (con->out_msg_pos.data_pos >= data_len - trail_len) { | |
848 | in_trail = 1; | |
849 | ||
850 | total_max_write = data_len - con->out_msg_pos.data_pos; | |
851 | ||
852 | page = list_first_entry(&msg->trail->head, | |
853 | struct page, lru); | |
68b4476b YS |
854 | max_write = PAGE_SIZE; |
855 | } else if (msg->pages) { | |
31b8006e | 856 | page = msg->pages[con->out_msg_pos.page]; |
58bb3b37 SW |
857 | } else if (msg->pagelist) { |
858 | page = list_first_entry(&msg->pagelist->head, | |
859 | struct page, lru); | |
68b4476b YS |
860 | #ifdef CONFIG_BLOCK |
861 | } else if (msg->bio) { | |
862 | struct bio_vec *bv; | |
863 | ||
864 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
865 | page = bv->bv_page; | |
9bd19663 | 866 | bio_offset = bv->bv_offset; |
68b4476b YS |
867 | max_write = bv->bv_len; |
868 | #endif | |
31b8006e | 869 | } else { |
57666519 | 870 | page = zero_page; |
31b8006e | 871 | } |
68b4476b YS |
872 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
873 | total_max_write); | |
874 | ||
37675b0f | 875 | if (do_datacrc && !con->out_msg_pos.did_page_crc) { |
9bd19663 | 876 | void *base; |
a9a0c51a | 877 | u32 crc; |
31b8006e | 878 | u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc); |
8d63e318 | 879 | char *kaddr; |
31b8006e | 880 | |
8d63e318 | 881 | kaddr = kmap(page); |
31b8006e | 882 | BUG_ON(kaddr == NULL); |
9bd19663 | 883 | base = kaddr + con->out_msg_pos.page_pos + bio_offset; |
a9a0c51a AE |
884 | crc = crc32c(tmpcrc, base, len); |
885 | con->out_msg->footer.data_crc = cpu_to_le32(crc); | |
bca064d2 | 886 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 887 | } |
e36b13cc | 888 | ret = ceph_tcp_sendpage(con->sock, page, |
9bd19663 | 889 | con->out_msg_pos.page_pos + bio_offset, |
e36b13cc | 890 | len, 1); |
31b8006e | 891 | |
0cdf9e60 | 892 | if (do_datacrc) |
31b8006e SW |
893 | kunmap(page); |
894 | ||
895 | if (ret <= 0) | |
896 | goto out; | |
897 | ||
898 | con->out_msg_pos.data_pos += ret; | |
899 | con->out_msg_pos.page_pos += ret; | |
900 | if (ret == len) { | |
901 | con->out_msg_pos.page_pos = 0; | |
902 | con->out_msg_pos.page++; | |
bca064d2 | 903 | con->out_msg_pos.did_page_crc = false; |
68b4476b YS |
904 | if (in_trail) |
905 | list_move_tail(&page->lru, | |
906 | &msg->trail->head); | |
907 | else if (msg->pagelist) | |
58bb3b37 SW |
908 | list_move_tail(&page->lru, |
909 | &msg->pagelist->head); | |
68b4476b YS |
910 | #ifdef CONFIG_BLOCK |
911 | else if (msg->bio) | |
912 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
913 | #endif | |
31b8006e SW |
914 | } |
915 | } | |
916 | ||
917 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
918 | ||
919 | /* prepare and queue up footer, too */ | |
37675b0f | 920 | if (!do_datacrc) |
31b8006e | 921 | con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
859eb799 AE |
922 | ceph_con_out_kvec_reset(con); |
923 | prepare_write_message_footer(con); | |
31b8006e SW |
924 | ret = 1; |
925 | out: | |
926 | return ret; | |
927 | } | |
928 | ||
929 | /* | |
930 | * write some zeros | |
931 | */ | |
932 | static int write_partial_skip(struct ceph_connection *con) | |
933 | { | |
934 | int ret; | |
935 | ||
936 | while (con->out_skip > 0) { | |
31739139 | 937 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 938 | |
31739139 | 939 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1); |
31b8006e SW |
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, | |
fd51653f | 991 | int end, int size, void *object) |
31b8006e | 992 | { |
e6cee71f AE |
993 | while (con->in_base_pos < end) { |
994 | int left = end - con->in_base_pos; | |
31b8006e SW |
995 | int have = size - left; |
996 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
997 | if (ret <= 0) | |
998 | return ret; | |
999 | con->in_base_pos += ret; | |
1000 | } | |
1001 | return 1; | |
1002 | } | |
1003 | ||
1004 | ||
1005 | /* | |
1006 | * Read all or part of the connect-side handshake on a new connection | |
1007 | */ | |
eed0ef2c | 1008 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1009 | { |
fd51653f AE |
1010 | int size; |
1011 | int end; | |
1012 | int ret; | |
31b8006e | 1013 | |
eed0ef2c | 1014 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1015 | |
1016 | /* peer's banner */ | |
fd51653f AE |
1017 | size = strlen(CEPH_BANNER); |
1018 | end = size; | |
1019 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1020 | if (ret <= 0) |
1021 | goto out; | |
fd51653f AE |
1022 | |
1023 | size = sizeof (con->actual_peer_addr); | |
1024 | end += size; | |
1025 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1026 | if (ret <= 0) |
1027 | goto out; | |
fd51653f AE |
1028 | |
1029 | size = sizeof (con->peer_addr_for_me); | |
1030 | end += size; | |
1031 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1032 | if (ret <= 0) |
1033 | goto out; | |
fd51653f | 1034 | |
eed0ef2c SW |
1035 | out: |
1036 | return ret; | |
1037 | } | |
1038 | ||
1039 | static int read_partial_connect(struct ceph_connection *con) | |
1040 | { | |
fd51653f AE |
1041 | int size; |
1042 | int end; | |
1043 | int ret; | |
eed0ef2c SW |
1044 | |
1045 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1046 | ||
fd51653f AE |
1047 | size = sizeof (con->in_reply); |
1048 | end = size; | |
1049 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1050 | if (ret <= 0) |
1051 | goto out; | |
fd51653f AE |
1052 | |
1053 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1054 | end += size; | |
1055 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1056 | if (ret <= 0) |
1057 | goto out; | |
31b8006e | 1058 | |
4e7a5dcd SW |
1059 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1060 | con, (int)con->in_reply.tag, | |
1061 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1062 | le32_to_cpu(con->in_reply.global_seq)); |
1063 | out: | |
1064 | return ret; | |
eed0ef2c | 1065 | |
31b8006e SW |
1066 | } |
1067 | ||
1068 | /* | |
1069 | * Verify the hello banner looks okay. | |
1070 | */ | |
1071 | static int verify_hello(struct ceph_connection *con) | |
1072 | { | |
1073 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1074 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1075 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1076 | con->error_msg = "protocol error, bad banner"; |
1077 | return -1; | |
1078 | } | |
1079 | return 0; | |
1080 | } | |
1081 | ||
1082 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1083 | { | |
1084 | switch (ss->ss_family) { | |
1085 | case AF_INET: | |
1086 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1087 | case AF_INET6: | |
1088 | return | |
1089 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1090 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1091 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1092 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1093 | } | |
1094 | return false; | |
1095 | } | |
1096 | ||
1097 | static int addr_port(struct sockaddr_storage *ss) | |
1098 | { | |
1099 | switch (ss->ss_family) { | |
1100 | case AF_INET: | |
f28bcfbe | 1101 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1102 | case AF_INET6: |
f28bcfbe | 1103 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1104 | } |
1105 | return 0; | |
1106 | } | |
1107 | ||
1108 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1109 | { | |
1110 | switch (ss->ss_family) { | |
1111 | case AF_INET: | |
1112 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1113 | break; |
31b8006e SW |
1114 | case AF_INET6: |
1115 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1116 | break; |
31b8006e SW |
1117 | } |
1118 | } | |
1119 | ||
ee3b56f2 NW |
1120 | /* |
1121 | * Unlike other *_pton function semantics, zero indicates success. | |
1122 | */ | |
1123 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1124 | char delim, const char **ipend) | |
1125 | { | |
99f0f3b2 AE |
1126 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1127 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1128 | |
1129 | memset(ss, 0, sizeof(*ss)); | |
1130 | ||
1131 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1132 | ss->ss_family = AF_INET; | |
1133 | return 0; | |
1134 | } | |
1135 | ||
1136 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1137 | ss->ss_family = AF_INET6; | |
1138 | return 0; | |
1139 | } | |
1140 | ||
1141 | return -EINVAL; | |
1142 | } | |
1143 | ||
1144 | /* | |
1145 | * Extract hostname string and resolve using kernel DNS facility. | |
1146 | */ | |
1147 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1148 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1149 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1150 | { | |
1151 | const char *end, *delim_p; | |
1152 | char *colon_p, *ip_addr = NULL; | |
1153 | int ip_len, ret; | |
1154 | ||
1155 | /* | |
1156 | * The end of the hostname occurs immediately preceding the delimiter or | |
1157 | * the port marker (':') where the delimiter takes precedence. | |
1158 | */ | |
1159 | delim_p = memchr(name, delim, namelen); | |
1160 | colon_p = memchr(name, ':', namelen); | |
1161 | ||
1162 | if (delim_p && colon_p) | |
1163 | end = delim_p < colon_p ? delim_p : colon_p; | |
1164 | else if (!delim_p && colon_p) | |
1165 | end = colon_p; | |
1166 | else { | |
1167 | end = delim_p; | |
1168 | if (!end) /* case: hostname:/ */ | |
1169 | end = name + namelen; | |
1170 | } | |
1171 | ||
1172 | if (end <= name) | |
1173 | return -EINVAL; | |
1174 | ||
1175 | /* do dns_resolve upcall */ | |
1176 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1177 | if (ip_len > 0) | |
1178 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1179 | else | |
1180 | ret = -ESRCH; | |
1181 | ||
1182 | kfree(ip_addr); | |
1183 | ||
1184 | *ipend = end; | |
1185 | ||
1186 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1187 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1188 | ||
1189 | return ret; | |
1190 | } | |
1191 | #else | |
1192 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1193 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1194 | { | |
1195 | return -EINVAL; | |
1196 | } | |
1197 | #endif | |
1198 | ||
1199 | /* | |
1200 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1201 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1202 | */ | |
1203 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1204 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1205 | { | |
1206 | int ret; | |
1207 | ||
1208 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1209 | if (ret) | |
1210 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1211 | ||
1212 | return ret; | |
1213 | } | |
1214 | ||
31b8006e SW |
1215 | /* |
1216 | * Parse an ip[:port] list into an addr array. Use the default | |
1217 | * monitor port if a port isn't specified. | |
1218 | */ | |
1219 | int ceph_parse_ips(const char *c, const char *end, | |
1220 | struct ceph_entity_addr *addr, | |
1221 | int max_count, int *count) | |
1222 | { | |
ee3b56f2 | 1223 | int i, ret = -EINVAL; |
31b8006e SW |
1224 | const char *p = c; |
1225 | ||
1226 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1227 | for (i = 0; i < max_count; i++) { | |
1228 | const char *ipend; | |
1229 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1230 | int port; |
39139f64 SW |
1231 | char delim = ','; |
1232 | ||
1233 | if (*p == '[') { | |
1234 | delim = ']'; | |
1235 | p++; | |
1236 | } | |
31b8006e | 1237 | |
ee3b56f2 NW |
1238 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1239 | if (ret) | |
31b8006e | 1240 | goto bad; |
ee3b56f2 NW |
1241 | ret = -EINVAL; |
1242 | ||
31b8006e SW |
1243 | p = ipend; |
1244 | ||
39139f64 SW |
1245 | if (delim == ']') { |
1246 | if (*p != ']') { | |
1247 | dout("missing matching ']'\n"); | |
1248 | goto bad; | |
1249 | } | |
1250 | p++; | |
1251 | } | |
1252 | ||
31b8006e SW |
1253 | /* port? */ |
1254 | if (p < end && *p == ':') { | |
1255 | port = 0; | |
1256 | p++; | |
1257 | while (p < end && *p >= '0' && *p <= '9') { | |
1258 | port = (port * 10) + (*p - '0'); | |
1259 | p++; | |
1260 | } | |
1261 | if (port > 65535 || port == 0) | |
1262 | goto bad; | |
1263 | } else { | |
1264 | port = CEPH_MON_PORT; | |
1265 | } | |
1266 | ||
1267 | addr_set_port(ss, port); | |
1268 | ||
3d14c5d2 | 1269 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1270 | |
1271 | if (p == end) | |
1272 | break; | |
1273 | if (*p != ',') | |
1274 | goto bad; | |
1275 | p++; | |
1276 | } | |
1277 | ||
1278 | if (p != end) | |
1279 | goto bad; | |
1280 | ||
1281 | if (count) | |
1282 | *count = i + 1; | |
1283 | return 0; | |
1284 | ||
1285 | bad: | |
39139f64 | 1286 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1287 | return ret; |
31b8006e | 1288 | } |
3d14c5d2 | 1289 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1290 | |
eed0ef2c | 1291 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1292 | { |
eed0ef2c | 1293 | dout("process_banner on %p\n", con); |
31b8006e SW |
1294 | |
1295 | if (verify_hello(con) < 0) | |
1296 | return -1; | |
1297 | ||
63f2d211 SW |
1298 | ceph_decode_addr(&con->actual_peer_addr); |
1299 | ceph_decode_addr(&con->peer_addr_for_me); | |
1300 | ||
31b8006e SW |
1301 | /* |
1302 | * Make sure the other end is who we wanted. note that the other | |
1303 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1304 | * them the benefit of the doubt. | |
1305 | */ | |
103e2d3a SW |
1306 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1307 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1308 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1309 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1310 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1311 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1312 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1313 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1314 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1315 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1316 | return -1; |
1317 | } | |
1318 | ||
1319 | /* | |
1320 | * did we learn our address? | |
1321 | */ | |
1322 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1323 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1324 | ||
1325 | memcpy(&con->msgr->inst.addr.in_addr, | |
1326 | &con->peer_addr_for_me.in_addr, | |
1327 | sizeof(con->peer_addr_for_me.in_addr)); | |
1328 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1329 | encode_my_addr(con->msgr); |
eed0ef2c | 1330 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1331 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1332 | } |
1333 | ||
eed0ef2c SW |
1334 | set_bit(NEGOTIATING, &con->state); |
1335 | prepare_read_connect(con); | |
1336 | return 0; | |
1337 | } | |
1338 | ||
04a419f9 SW |
1339 | static void fail_protocol(struct ceph_connection *con) |
1340 | { | |
1341 | reset_connection(con); | |
1342 | set_bit(CLOSED, &con->state); /* in case there's queued work */ | |
1343 | ||
1344 | mutex_unlock(&con->mutex); | |
1345 | if (con->ops->bad_proto) | |
1346 | con->ops->bad_proto(con); | |
1347 | mutex_lock(&con->mutex); | |
1348 | } | |
1349 | ||
eed0ef2c SW |
1350 | static int process_connect(struct ceph_connection *con) |
1351 | { | |
3d14c5d2 YS |
1352 | u64 sup_feat = con->msgr->supported_features; |
1353 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1354 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1355 | int ret; |
04a419f9 | 1356 | |
eed0ef2c SW |
1357 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1358 | ||
31b8006e | 1359 | switch (con->in_reply.tag) { |
04a419f9 SW |
1360 | case CEPH_MSGR_TAG_FEATURES: |
1361 | pr_err("%s%lld %s feature set mismatch," | |
1362 | " my %llx < server's %llx, missing %llx\n", | |
1363 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1364 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1365 | sup_feat, server_feat, server_feat & ~sup_feat); |
1366 | con->error_msg = "missing required protocol features"; | |
1367 | fail_protocol(con); | |
1368 | return -1; | |
1369 | ||
31b8006e | 1370 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1371 | pr_err("%s%lld %s protocol version mismatch," |
1372 | " my %d != server's %d\n", | |
1373 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1374 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1375 | le32_to_cpu(con->out_connect.protocol_version), |
1376 | le32_to_cpu(con->in_reply.protocol_version)); | |
1377 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1378 | fail_protocol(con); |
31b8006e SW |
1379 | return -1; |
1380 | ||
4e7a5dcd SW |
1381 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1382 | con->auth_retry++; | |
1383 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1384 | con->auth_retry); | |
1385 | if (con->auth_retry == 2) { | |
1386 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1387 | return -1; |
1388 | } | |
1389 | con->auth_retry = 1; | |
84fb3adf | 1390 | ceph_con_out_kvec_reset(con); |
e825a66d | 1391 | ret = prepare_write_connect(con); |
0da5d703 SW |
1392 | if (ret < 0) |
1393 | return ret; | |
63733a0f | 1394 | prepare_read_connect(con); |
4e7a5dcd | 1395 | break; |
31b8006e SW |
1396 | |
1397 | case CEPH_MSGR_TAG_RESETSESSION: | |
1398 | /* | |
1399 | * If we connected with a large connect_seq but the peer | |
1400 | * has no record of a session with us (no connection, or | |
1401 | * connect_seq == 0), they will send RESETSESION to indicate | |
1402 | * that they must have reset their session, and may have | |
1403 | * dropped messages. | |
1404 | */ | |
1405 | dout("process_connect got RESET peer seq %u\n", | |
1406 | le32_to_cpu(con->in_connect.connect_seq)); | |
1407 | pr_err("%s%lld %s connection reset\n", | |
1408 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1409 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1410 | reset_connection(con); |
84fb3adf | 1411 | ceph_con_out_kvec_reset(con); |
e825a66d | 1412 | prepare_write_connect(con); |
31b8006e SW |
1413 | prepare_read_connect(con); |
1414 | ||
1415 | /* Tell ceph about it. */ | |
ec302645 | 1416 | mutex_unlock(&con->mutex); |
31b8006e SW |
1417 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1418 | if (con->ops->peer_reset) | |
1419 | con->ops->peer_reset(con); | |
ec302645 | 1420 | mutex_lock(&con->mutex); |
0da5d703 SW |
1421 | if (test_bit(CLOSED, &con->state) || |
1422 | test_bit(OPENING, &con->state)) | |
1423 | return -EAGAIN; | |
31b8006e SW |
1424 | break; |
1425 | ||
1426 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1427 | /* | |
1428 | * If we sent a smaller connect_seq than the peer has, try | |
1429 | * again with a larger value. | |
1430 | */ | |
1431 | dout("process_connect got RETRY my seq = %u, peer_seq = %u\n", | |
1432 | le32_to_cpu(con->out_connect.connect_seq), | |
1433 | le32_to_cpu(con->in_connect.connect_seq)); | |
1434 | con->connect_seq = le32_to_cpu(con->in_connect.connect_seq); | |
84fb3adf | 1435 | ceph_con_out_kvec_reset(con); |
e825a66d | 1436 | prepare_write_connect(con); |
31b8006e SW |
1437 | prepare_read_connect(con); |
1438 | break; | |
1439 | ||
1440 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1441 | /* | |
1442 | * If we sent a smaller global_seq than the peer has, try | |
1443 | * again with a larger value. | |
1444 | */ | |
eed0ef2c | 1445 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e SW |
1446 | con->peer_global_seq, |
1447 | le32_to_cpu(con->in_connect.global_seq)); | |
1448 | get_global_seq(con->msgr, | |
1449 | le32_to_cpu(con->in_connect.global_seq)); | |
84fb3adf | 1450 | ceph_con_out_kvec_reset(con); |
e825a66d | 1451 | prepare_write_connect(con); |
31b8006e SW |
1452 | prepare_read_connect(con); |
1453 | break; | |
1454 | ||
1455 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1456 | if (req_feat & ~server_feat) { |
1457 | pr_err("%s%lld %s protocol feature mismatch," | |
1458 | " my required %llx > server's %llx, need %llx\n", | |
1459 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1460 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1461 | req_feat, server_feat, req_feat & ~server_feat); |
1462 | con->error_msg = "missing required protocol features"; | |
1463 | fail_protocol(con); | |
1464 | return -1; | |
1465 | } | |
31b8006e | 1466 | clear_bit(CONNECTING, &con->state); |
31b8006e SW |
1467 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1468 | con->connect_seq++; | |
aba558e2 | 1469 | con->peer_features = server_feat; |
31b8006e SW |
1470 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1471 | con->peer_global_seq, | |
1472 | le32_to_cpu(con->in_reply.connect_seq), | |
1473 | con->connect_seq); | |
1474 | WARN_ON(con->connect_seq != | |
1475 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1476 | |
1477 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
1478 | set_bit(LOSSYTX, &con->state); | |
1479 | ||
31b8006e SW |
1480 | prepare_read_tag(con); |
1481 | break; | |
1482 | ||
1483 | case CEPH_MSGR_TAG_WAIT: | |
1484 | /* | |
1485 | * If there is a connection race (we are opening | |
1486 | * connections to each other), one of us may just have | |
1487 | * to WAIT. This shouldn't happen if we are the | |
1488 | * client. | |
1489 | */ | |
04177882 SW |
1490 | pr_err("process_connect got WAIT as client\n"); |
1491 | con->error_msg = "protocol error, got WAIT as client"; | |
1492 | return -1; | |
31b8006e SW |
1493 | |
1494 | default: | |
1495 | pr_err("connect protocol error, will retry\n"); | |
1496 | con->error_msg = "protocol error, garbage tag during connect"; | |
1497 | return -1; | |
1498 | } | |
1499 | return 0; | |
1500 | } | |
1501 | ||
1502 | ||
1503 | /* | |
1504 | * read (part of) an ack | |
1505 | */ | |
1506 | static int read_partial_ack(struct ceph_connection *con) | |
1507 | { | |
fd51653f AE |
1508 | int size = sizeof (con->in_temp_ack); |
1509 | int end = size; | |
1510 | ||
1511 | return read_partial(con, end, size, &con->in_temp_ack); | |
31b8006e SW |
1512 | } |
1513 | ||
1514 | ||
1515 | /* | |
1516 | * We can finally discard anything that's been acked. | |
1517 | */ | |
1518 | static void process_ack(struct ceph_connection *con) | |
1519 | { | |
1520 | struct ceph_msg *m; | |
1521 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1522 | u64 seq; | |
1523 | ||
31b8006e SW |
1524 | while (!list_empty(&con->out_sent)) { |
1525 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1526 | list_head); | |
1527 | seq = le64_to_cpu(m->hdr.seq); | |
1528 | if (seq > ack) | |
1529 | break; | |
1530 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1531 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1532 | m->ack_stamp = jiffies; |
31b8006e SW |
1533 | ceph_msg_remove(m); |
1534 | } | |
31b8006e SW |
1535 | prepare_read_tag(con); |
1536 | } | |
1537 | ||
1538 | ||
1539 | ||
1540 | ||
2450418c | 1541 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1542 | struct kvec *section, |
1543 | unsigned int sec_len, u32 *crc) | |
2450418c | 1544 | { |
68b4476b | 1545 | int ret, left; |
2450418c YS |
1546 | |
1547 | BUG_ON(!section); | |
1548 | ||
1549 | while (section->iov_len < sec_len) { | |
1550 | BUG_ON(section->iov_base == NULL); | |
1551 | left = sec_len - section->iov_len; | |
1552 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1553 | section->iov_len, left); | |
1554 | if (ret <= 0) | |
1555 | return ret; | |
1556 | section->iov_len += ret; | |
2450418c | 1557 | } |
fe3ad593 AE |
1558 | if (section->iov_len == sec_len) |
1559 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1560 | |
2450418c YS |
1561 | return 1; |
1562 | } | |
31b8006e | 1563 | |
2450418c YS |
1564 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, |
1565 | struct ceph_msg_header *hdr, | |
1566 | int *skip); | |
68b4476b YS |
1567 | |
1568 | ||
1569 | static int read_partial_message_pages(struct ceph_connection *con, | |
1570 | struct page **pages, | |
bca064d2 | 1571 | unsigned data_len, bool do_datacrc) |
68b4476b YS |
1572 | { |
1573 | void *p; | |
1574 | int ret; | |
1575 | int left; | |
1576 | ||
1577 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1578 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1579 | /* (page) data */ | |
1580 | BUG_ON(pages == NULL); | |
1581 | p = kmap(pages[con->in_msg_pos.page]); | |
1582 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1583 | left); | |
bca064d2 | 1584 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1585 | con->in_data_crc = |
1586 | crc32c(con->in_data_crc, | |
1587 | p + con->in_msg_pos.page_pos, ret); | |
1588 | kunmap(pages[con->in_msg_pos.page]); | |
1589 | if (ret <= 0) | |
1590 | return ret; | |
1591 | con->in_msg_pos.data_pos += ret; | |
1592 | con->in_msg_pos.page_pos += ret; | |
1593 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1594 | con->in_msg_pos.page_pos = 0; | |
1595 | con->in_msg_pos.page++; | |
1596 | } | |
1597 | ||
1598 | return ret; | |
1599 | } | |
1600 | ||
1601 | #ifdef CONFIG_BLOCK | |
1602 | static int read_partial_message_bio(struct ceph_connection *con, | |
1603 | struct bio **bio_iter, int *bio_seg, | |
bca064d2 | 1604 | unsigned data_len, bool do_datacrc) |
68b4476b YS |
1605 | { |
1606 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1607 | void *p; | |
1608 | int ret, left; | |
1609 | ||
1610 | if (IS_ERR(bv)) | |
1611 | return PTR_ERR(bv); | |
1612 | ||
1613 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1614 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1615 | ||
1616 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1617 | ||
1618 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1619 | left); | |
bca064d2 | 1620 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1621 | con->in_data_crc = |
1622 | crc32c(con->in_data_crc, | |
1623 | p + con->in_msg_pos.page_pos, ret); | |
1624 | kunmap(bv->bv_page); | |
1625 | if (ret <= 0) | |
1626 | return ret; | |
1627 | con->in_msg_pos.data_pos += ret; | |
1628 | con->in_msg_pos.page_pos += ret; | |
1629 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1630 | con->in_msg_pos.page_pos = 0; | |
1631 | iter_bio_next(bio_iter, bio_seg); | |
1632 | } | |
1633 | ||
1634 | return ret; | |
1635 | } | |
1636 | #endif | |
1637 | ||
31b8006e SW |
1638 | /* |
1639 | * read (part of) a message. | |
1640 | */ | |
1641 | static int read_partial_message(struct ceph_connection *con) | |
1642 | { | |
1643 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1644 | int size; |
1645 | int end; | |
31b8006e | 1646 | int ret; |
c5c6b19d | 1647 | unsigned front_len, middle_len, data_len; |
37675b0f | 1648 | bool do_datacrc = !con->msgr->nocrc; |
2450418c | 1649 | int skip; |
ae18756b | 1650 | u64 seq; |
fe3ad593 | 1651 | u32 crc; |
31b8006e SW |
1652 | |
1653 | dout("read_partial_message con %p msg %p\n", con, m); | |
1654 | ||
1655 | /* header */ | |
fd51653f AE |
1656 | size = sizeof (con->in_hdr); |
1657 | end = size; | |
1658 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1659 | if (ret <= 0) |
1660 | return ret; | |
fe3ad593 AE |
1661 | |
1662 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1663 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1664 | pr_err("read_partial_message bad hdr " | |
1665 | " crc %u != expected %u\n", | |
1666 | crc, con->in_hdr.crc); | |
1667 | return -EBADMSG; | |
1668 | } | |
1669 | ||
31b8006e SW |
1670 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1671 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1672 | return -EIO; | |
1673 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1674 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1675 | return -EIO; | |
1676 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1677 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1678 | return -EIO; | |
1679 | ||
ae18756b SW |
1680 | /* verify seq# */ |
1681 | seq = le64_to_cpu(con->in_hdr.seq); | |
1682 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1683 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1684 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1685 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1686 | seq, con->in_seq + 1); |
1687 | con->in_base_pos = -front_len - middle_len - data_len - | |
1688 | sizeof(m->footer); | |
1689 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1690 | return 0; |
1691 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1692 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1693 | seq, con->in_seq + 1); | |
1694 | con->error_msg = "bad message sequence # for incoming message"; | |
1695 | return -EBADMSG; | |
1696 | } | |
1697 | ||
31b8006e SW |
1698 | /* allocate message? */ |
1699 | if (!con->in_msg) { | |
1700 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1701 | con->in_hdr.front_len, con->in_hdr.data_len); | |
ae32be31 | 1702 | skip = 0; |
2450418c YS |
1703 | con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip); |
1704 | if (skip) { | |
31b8006e | 1705 | /* skip this message */ |
a79832f2 | 1706 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1707 | BUG_ON(con->in_msg); |
31b8006e SW |
1708 | con->in_base_pos = -front_len - middle_len - data_len - |
1709 | sizeof(m->footer); | |
1710 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1711 | con->in_seq++; |
31b8006e SW |
1712 | return 0; |
1713 | } | |
a79832f2 | 1714 | if (!con->in_msg) { |
5b3a4db3 SW |
1715 | con->error_msg = |
1716 | "error allocating memory for incoming message"; | |
a79832f2 | 1717 | return -ENOMEM; |
31b8006e SW |
1718 | } |
1719 | m = con->in_msg; | |
1720 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1721 | if (m->middle) |
1722 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1723 | |
1724 | con->in_msg_pos.page = 0; | |
68b4476b | 1725 | if (m->pages) |
c5c6b19d | 1726 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1727 | else |
1728 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1729 | con->in_msg_pos.data_pos = 0; |
31b8006e SW |
1730 | } |
1731 | ||
1732 | /* front */ | |
2450418c YS |
1733 | ret = read_partial_message_section(con, &m->front, front_len, |
1734 | &con->in_front_crc); | |
1735 | if (ret <= 0) | |
1736 | return ret; | |
31b8006e SW |
1737 | |
1738 | /* middle */ | |
2450418c | 1739 | if (m->middle) { |
213c99ee SW |
1740 | ret = read_partial_message_section(con, &m->middle->vec, |
1741 | middle_len, | |
2450418c | 1742 | &con->in_middle_crc); |
31b8006e SW |
1743 | if (ret <= 0) |
1744 | return ret; | |
31b8006e | 1745 | } |
68b4476b YS |
1746 | #ifdef CONFIG_BLOCK |
1747 | if (m->bio && !m->bio_iter) | |
1748 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1749 | #endif | |
31b8006e SW |
1750 | |
1751 | /* (page) data */ | |
31b8006e | 1752 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1753 | if (m->pages) { |
1754 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1755 | data_len, do_datacrc); |
68b4476b YS |
1756 | if (ret <= 0) |
1757 | return ret; | |
1758 | #ifdef CONFIG_BLOCK | |
1759 | } else if (m->bio) { | |
1760 | ||
1761 | ret = read_partial_message_bio(con, | |
1762 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1763 | data_len, do_datacrc); |
68b4476b YS |
1764 | if (ret <= 0) |
1765 | return ret; | |
1766 | #endif | |
1767 | } else { | |
1768 | BUG_ON(1); | |
31b8006e SW |
1769 | } |
1770 | } | |
1771 | ||
31b8006e | 1772 | /* footer */ |
fd51653f AE |
1773 | size = sizeof (m->footer); |
1774 | end += size; | |
1775 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
1776 | if (ret <= 0) |
1777 | return ret; | |
1778 | ||
31b8006e SW |
1779 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
1780 | m, front_len, m->footer.front_crc, middle_len, | |
1781 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1782 | ||
1783 | /* crc ok? */ | |
1784 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1785 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1786 | m, con->in_front_crc, m->footer.front_crc); | |
1787 | return -EBADMSG; | |
1788 | } | |
1789 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1790 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1791 | m, con->in_middle_crc, m->footer.middle_crc); | |
1792 | return -EBADMSG; | |
1793 | } | |
bca064d2 | 1794 | if (do_datacrc && |
31b8006e SW |
1795 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
1796 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1797 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1798 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1799 | return -EBADMSG; | |
1800 | } | |
1801 | ||
1802 | return 1; /* done! */ | |
1803 | } | |
1804 | ||
1805 | /* | |
1806 | * Process message. This happens in the worker thread. The callback should | |
1807 | * be careful not to do anything that waits on other incoming messages or it | |
1808 | * may deadlock. | |
1809 | */ | |
1810 | static void process_message(struct ceph_connection *con) | |
1811 | { | |
5e095e8b | 1812 | struct ceph_msg *msg; |
31b8006e | 1813 | |
5e095e8b | 1814 | msg = con->in_msg; |
31b8006e SW |
1815 | con->in_msg = NULL; |
1816 | ||
1817 | /* if first message, set peer_name */ | |
1818 | if (con->peer_name.type == 0) | |
dbad185d | 1819 | con->peer_name = msg->hdr.src; |
31b8006e | 1820 | |
31b8006e | 1821 | con->in_seq++; |
ec302645 | 1822 | mutex_unlock(&con->mutex); |
31b8006e SW |
1823 | |
1824 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1825 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1826 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1827 | le16_to_cpu(msg->hdr.type), |
1828 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1829 | le32_to_cpu(msg->hdr.front_len), | |
1830 | le32_to_cpu(msg->hdr.data_len), | |
1831 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1832 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1833 | |
1834 | mutex_lock(&con->mutex); | |
31b8006e SW |
1835 | prepare_read_tag(con); |
1836 | } | |
1837 | ||
1838 | ||
1839 | /* | |
1840 | * Write something to the socket. Called in a worker thread when the | |
1841 | * socket appears to be writeable and we have something ready to send. | |
1842 | */ | |
1843 | static int try_write(struct ceph_connection *con) | |
1844 | { | |
31b8006e SW |
1845 | int ret = 1; |
1846 | ||
1847 | dout("try_write start %p state %lu nref %d\n", con, con->state, | |
1848 | atomic_read(&con->nref)); | |
1849 | ||
31b8006e SW |
1850 | more: |
1851 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
1852 | ||
1853 | /* open the socket first? */ | |
1854 | if (con->sock == NULL) { | |
84fb3adf | 1855 | ceph_con_out_kvec_reset(con); |
e825a66d AE |
1856 | prepare_write_banner(con); |
1857 | prepare_write_connect(con); | |
eed0ef2c | 1858 | prepare_read_banner(con); |
31b8006e | 1859 | set_bit(CONNECTING, &con->state); |
eed0ef2c | 1860 | clear_bit(NEGOTIATING, &con->state); |
31b8006e | 1861 | |
cf3e5c40 | 1862 | BUG_ON(con->in_msg); |
31b8006e SW |
1863 | con->in_tag = CEPH_MSGR_TAG_READY; |
1864 | dout("try_write initiating connect on %p new state %lu\n", | |
1865 | con, con->state); | |
41617d0c AE |
1866 | ret = ceph_tcp_connect(con); |
1867 | if (ret < 0) { | |
31b8006e | 1868 | con->error_msg = "connect error"; |
31b8006e SW |
1869 | goto out; |
1870 | } | |
1871 | } | |
1872 | ||
1873 | more_kvec: | |
1874 | /* kvec data queued? */ | |
1875 | if (con->out_skip) { | |
1876 | ret = write_partial_skip(con); | |
1877 | if (ret <= 0) | |
42961d23 | 1878 | goto out; |
31b8006e SW |
1879 | } |
1880 | if (con->out_kvec_left) { | |
1881 | ret = write_partial_kvec(con); | |
1882 | if (ret <= 0) | |
42961d23 | 1883 | goto out; |
31b8006e SW |
1884 | } |
1885 | ||
1886 | /* msg pages? */ | |
1887 | if (con->out_msg) { | |
c86a2930 SW |
1888 | if (con->out_msg_done) { |
1889 | ceph_msg_put(con->out_msg); | |
1890 | con->out_msg = NULL; /* we're done with this one */ | |
1891 | goto do_next; | |
1892 | } | |
1893 | ||
31b8006e SW |
1894 | ret = write_partial_msg_pages(con); |
1895 | if (ret == 1) | |
1896 | goto more_kvec; /* we need to send the footer, too! */ | |
1897 | if (ret == 0) | |
42961d23 | 1898 | goto out; |
31b8006e SW |
1899 | if (ret < 0) { |
1900 | dout("try_write write_partial_msg_pages err %d\n", | |
1901 | ret); | |
42961d23 | 1902 | goto out; |
31b8006e SW |
1903 | } |
1904 | } | |
1905 | ||
c86a2930 | 1906 | do_next: |
31b8006e SW |
1907 | if (!test_bit(CONNECTING, &con->state)) { |
1908 | /* is anything else pending? */ | |
1909 | if (!list_empty(&con->out_queue)) { | |
1910 | prepare_write_message(con); | |
1911 | goto more; | |
1912 | } | |
1913 | if (con->in_seq > con->in_seq_acked) { | |
1914 | prepare_write_ack(con); | |
1915 | goto more; | |
1916 | } | |
1917 | if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) { | |
1918 | prepare_write_keepalive(con); | |
1919 | goto more; | |
1920 | } | |
1921 | } | |
1922 | ||
1923 | /* Nothing to do! */ | |
1924 | clear_bit(WRITE_PENDING, &con->state); | |
1925 | dout("try_write nothing else to write.\n"); | |
31b8006e SW |
1926 | ret = 0; |
1927 | out: | |
42961d23 | 1928 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
1929 | return ret; |
1930 | } | |
1931 | ||
1932 | ||
1933 | ||
1934 | /* | |
1935 | * Read what we can from the socket. | |
1936 | */ | |
1937 | static int try_read(struct ceph_connection *con) | |
1938 | { | |
31b8006e SW |
1939 | int ret = -1; |
1940 | ||
1941 | if (!con->sock) | |
1942 | return 0; | |
1943 | ||
1944 | if (test_bit(STANDBY, &con->state)) | |
1945 | return 0; | |
1946 | ||
1947 | dout("try_read start on %p\n", con); | |
ec302645 | 1948 | |
31b8006e SW |
1949 | more: |
1950 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, | |
1951 | con->in_base_pos); | |
0da5d703 SW |
1952 | |
1953 | /* | |
1954 | * process_connect and process_message drop and re-take | |
1955 | * con->mutex. make sure we handle a racing close or reopen. | |
1956 | */ | |
1957 | if (test_bit(CLOSED, &con->state) || | |
1958 | test_bit(OPENING, &con->state)) { | |
1959 | ret = -EAGAIN; | |
1960 | goto out; | |
1961 | } | |
1962 | ||
31b8006e | 1963 | if (test_bit(CONNECTING, &con->state)) { |
eed0ef2c SW |
1964 | if (!test_bit(NEGOTIATING, &con->state)) { |
1965 | dout("try_read connecting\n"); | |
1966 | ret = read_partial_banner(con); | |
1967 | if (ret <= 0) | |
eed0ef2c | 1968 | goto out; |
98bdb0aa SW |
1969 | ret = process_banner(con); |
1970 | if (ret < 0) | |
1971 | goto out; | |
eed0ef2c | 1972 | } |
31b8006e SW |
1973 | ret = read_partial_connect(con); |
1974 | if (ret <= 0) | |
31b8006e | 1975 | goto out; |
98bdb0aa SW |
1976 | ret = process_connect(con); |
1977 | if (ret < 0) | |
1978 | goto out; | |
31b8006e SW |
1979 | goto more; |
1980 | } | |
1981 | ||
1982 | if (con->in_base_pos < 0) { | |
1983 | /* | |
1984 | * skipping + discarding content. | |
1985 | * | |
1986 | * FIXME: there must be a better way to do this! | |
1987 | */ | |
84495f49 AE |
1988 | static char buf[SKIP_BUF_SIZE]; |
1989 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
1990 | ||
31b8006e SW |
1991 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
1992 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
1993 | if (ret <= 0) | |
98bdb0aa | 1994 | goto out; |
31b8006e SW |
1995 | con->in_base_pos += ret; |
1996 | if (con->in_base_pos) | |
1997 | goto more; | |
1998 | } | |
1999 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2000 | /* | |
2001 | * what's next? | |
2002 | */ | |
2003 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2004 | if (ret <= 0) | |
98bdb0aa | 2005 | goto out; |
31b8006e SW |
2006 | dout("try_read got tag %d\n", (int)con->in_tag); |
2007 | switch (con->in_tag) { | |
2008 | case CEPH_MSGR_TAG_MSG: | |
2009 | prepare_read_message(con); | |
2010 | break; | |
2011 | case CEPH_MSGR_TAG_ACK: | |
2012 | prepare_read_ack(con); | |
2013 | break; | |
2014 | case CEPH_MSGR_TAG_CLOSE: | |
2015 | set_bit(CLOSED, &con->state); /* fixme */ | |
98bdb0aa | 2016 | goto out; |
31b8006e SW |
2017 | default: |
2018 | goto bad_tag; | |
2019 | } | |
2020 | } | |
2021 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2022 | ret = read_partial_message(con); | |
2023 | if (ret <= 0) { | |
2024 | switch (ret) { | |
2025 | case -EBADMSG: | |
2026 | con->error_msg = "bad crc"; | |
2027 | ret = -EIO; | |
98bdb0aa | 2028 | break; |
31b8006e SW |
2029 | case -EIO: |
2030 | con->error_msg = "io error"; | |
98bdb0aa | 2031 | break; |
31b8006e | 2032 | } |
98bdb0aa | 2033 | goto out; |
31b8006e SW |
2034 | } |
2035 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2036 | goto more; | |
2037 | process_message(con); | |
2038 | goto more; | |
2039 | } | |
2040 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2041 | ret = read_partial_ack(con); | |
2042 | if (ret <= 0) | |
98bdb0aa | 2043 | goto out; |
31b8006e SW |
2044 | process_ack(con); |
2045 | goto more; | |
2046 | } | |
2047 | ||
31b8006e | 2048 | out: |
98bdb0aa | 2049 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2050 | return ret; |
2051 | ||
2052 | bad_tag: | |
2053 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2054 | con->error_msg = "protocol error, garbage tag"; | |
2055 | ret = -1; | |
2056 | goto out; | |
2057 | } | |
2058 | ||
2059 | ||
2060 | /* | |
2061 | * Atomically queue work on a connection. Bump @con reference to | |
2062 | * avoid races with connection teardown. | |
31b8006e SW |
2063 | */ |
2064 | static void queue_con(struct ceph_connection *con) | |
2065 | { | |
2066 | if (test_bit(DEAD, &con->state)) { | |
2067 | dout("queue_con %p ignoring: DEAD\n", | |
2068 | con); | |
2069 | return; | |
2070 | } | |
2071 | ||
2072 | if (!con->ops->get(con)) { | |
2073 | dout("queue_con %p ref count 0\n", con); | |
2074 | return; | |
2075 | } | |
2076 | ||
f363e45f | 2077 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2078 | dout("queue_con %p - already queued\n", con); |
2079 | con->ops->put(con); | |
2080 | } else { | |
2081 | dout("queue_con %p\n", con); | |
2082 | } | |
2083 | } | |
2084 | ||
2085 | /* | |
2086 | * Do some work on a connection. Drop a connection ref when we're done. | |
2087 | */ | |
2088 | static void con_work(struct work_struct *work) | |
2089 | { | |
2090 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2091 | work.work); | |
0da5d703 | 2092 | int ret; |
31b8006e | 2093 | |
9dd4658d | 2094 | mutex_lock(&con->mutex); |
0da5d703 | 2095 | restart: |
60bf8bf8 SW |
2096 | if (test_and_clear_bit(BACKOFF, &con->state)) { |
2097 | dout("con_work %p backing off\n", con); | |
2098 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2099 | round_jiffies_relative(con->delay))) { | |
2100 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2101 | mutex_unlock(&con->mutex); | |
2102 | return; | |
2103 | } else { | |
2104 | con->ops->put(con); | |
2105 | dout("con_work %p FAILED to back off %lu\n", con, | |
2106 | con->delay); | |
2107 | } | |
2108 | } | |
9dd4658d | 2109 | |
e00de341 SW |
2110 | if (test_bit(STANDBY, &con->state)) { |
2111 | dout("con_work %p STANDBY\n", con); | |
2112 | goto done; | |
2113 | } | |
31b8006e SW |
2114 | if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */ |
2115 | dout("con_work CLOSED\n"); | |
2116 | con_close_socket(con); | |
2117 | goto done; | |
2118 | } | |
2119 | if (test_and_clear_bit(OPENING, &con->state)) { | |
2120 | /* reopen w/ new peer */ | |
2121 | dout("con_work OPENING\n"); | |
2122 | con_close_socket(con); | |
2123 | } | |
2124 | ||
0da5d703 SW |
2125 | if (test_and_clear_bit(SOCK_CLOSED, &con->state)) |
2126 | goto fault; | |
2127 | ||
2128 | ret = try_read(con); | |
2129 | if (ret == -EAGAIN) | |
2130 | goto restart; | |
2131 | if (ret < 0) | |
2132 | goto fault; | |
2133 | ||
2134 | ret = try_write(con); | |
2135 | if (ret == -EAGAIN) | |
2136 | goto restart; | |
2137 | if (ret < 0) | |
2138 | goto fault; | |
31b8006e SW |
2139 | |
2140 | done: | |
9dd4658d | 2141 | mutex_unlock(&con->mutex); |
9dd4658d | 2142 | done_unlocked: |
31b8006e | 2143 | con->ops->put(con); |
0da5d703 SW |
2144 | return; |
2145 | ||
2146 | fault: | |
2147 | mutex_unlock(&con->mutex); | |
2148 | ceph_fault(con); /* error/fault path */ | |
2149 | goto done_unlocked; | |
31b8006e SW |
2150 | } |
2151 | ||
2152 | ||
2153 | /* | |
2154 | * Generic error/fault handler. A retry mechanism is used with | |
2155 | * exponential backoff | |
2156 | */ | |
2157 | static void ceph_fault(struct ceph_connection *con) | |
2158 | { | |
2159 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2160 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2161 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2162 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
2163 | |
2164 | if (test_bit(LOSSYTX, &con->state)) { | |
2165 | dout("fault on LOSSYTX channel\n"); | |
2166 | goto out; | |
2167 | } | |
2168 | ||
ec302645 | 2169 | mutex_lock(&con->mutex); |
91e45ce3 SW |
2170 | if (test_bit(CLOSED, &con->state)) |
2171 | goto out_unlock; | |
ec302645 | 2172 | |
31b8006e | 2173 | con_close_socket(con); |
5e095e8b SW |
2174 | |
2175 | if (con->in_msg) { | |
2176 | ceph_msg_put(con->in_msg); | |
2177 | con->in_msg = NULL; | |
2178 | } | |
31b8006e | 2179 | |
e80a52d1 SW |
2180 | /* Requeue anything that hasn't been acked */ |
2181 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2182 | |
e76661d0 SW |
2183 | /* If there are no messages queued or keepalive pending, place |
2184 | * the connection in a STANDBY state */ | |
2185 | if (list_empty(&con->out_queue) && | |
2186 | !test_bit(KEEPALIVE_PENDING, &con->state)) { | |
e00de341 SW |
2187 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
2188 | clear_bit(WRITE_PENDING, &con->state); | |
31b8006e | 2189 | set_bit(STANDBY, &con->state); |
e80a52d1 SW |
2190 | } else { |
2191 | /* retry after a delay. */ | |
2192 | if (con->delay == 0) | |
2193 | con->delay = BASE_DELAY_INTERVAL; | |
2194 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2195 | con->delay *= 2; | |
e80a52d1 SW |
2196 | con->ops->get(con); |
2197 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2198 | round_jiffies_relative(con->delay))) { |
2199 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2200 | } else { | |
e80a52d1 | 2201 | con->ops->put(con); |
60bf8bf8 SW |
2202 | dout("fault failed to queue %p delay %lu, backoff\n", |
2203 | con, con->delay); | |
2204 | /* | |
2205 | * In many cases we see a socket state change | |
2206 | * while con_work is running and end up | |
2207 | * queuing (non-delayed) work, such that we | |
2208 | * can't backoff with a delay. Set a flag so | |
2209 | * that when con_work restarts we schedule the | |
2210 | * delay then. | |
2211 | */ | |
2212 | set_bit(BACKOFF, &con->state); | |
2213 | } | |
31b8006e SW |
2214 | } |
2215 | ||
91e45ce3 SW |
2216 | out_unlock: |
2217 | mutex_unlock(&con->mutex); | |
31b8006e | 2218 | out: |
161fd65a SW |
2219 | /* |
2220 | * in case we faulted due to authentication, invalidate our | |
2221 | * current tickets so that we can get new ones. | |
213c99ee | 2222 | */ |
161fd65a SW |
2223 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2224 | dout("calling invalidate_authorizer()\n"); | |
2225 | con->ops->invalidate_authorizer(con); | |
2226 | } | |
2227 | ||
31b8006e SW |
2228 | if (con->ops->fault) |
2229 | con->ops->fault(con); | |
2230 | } | |
2231 | ||
2232 | ||
2233 | ||
2234 | /* | |
2235 | * create a new messenger instance | |
2236 | */ | |
3d14c5d2 YS |
2237 | struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr, |
2238 | u32 supported_features, | |
2239 | u32 required_features) | |
31b8006e SW |
2240 | { |
2241 | struct ceph_messenger *msgr; | |
2242 | ||
2243 | msgr = kzalloc(sizeof(*msgr), GFP_KERNEL); | |
2244 | if (msgr == NULL) | |
2245 | return ERR_PTR(-ENOMEM); | |
2246 | ||
3d14c5d2 YS |
2247 | msgr->supported_features = supported_features; |
2248 | msgr->required_features = required_features; | |
2249 | ||
31b8006e SW |
2250 | spin_lock_init(&msgr->global_seq_lock); |
2251 | ||
31b8006e SW |
2252 | if (myaddr) |
2253 | msgr->inst.addr = *myaddr; | |
2254 | ||
2255 | /* select a random nonce */ | |
ac8839d7 | 2256 | msgr->inst.addr.type = 0; |
103e2d3a | 2257 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2258 | encode_my_addr(msgr); |
31b8006e SW |
2259 | |
2260 | dout("messenger_create %p\n", msgr); | |
2261 | return msgr; | |
2262 | } | |
3d14c5d2 | 2263 | EXPORT_SYMBOL(ceph_messenger_create); |
31b8006e SW |
2264 | |
2265 | void ceph_messenger_destroy(struct ceph_messenger *msgr) | |
2266 | { | |
2267 | dout("destroy %p\n", msgr); | |
31b8006e SW |
2268 | kfree(msgr); |
2269 | dout("destroyed messenger %p\n", msgr); | |
2270 | } | |
3d14c5d2 | 2271 | EXPORT_SYMBOL(ceph_messenger_destroy); |
31b8006e | 2272 | |
e00de341 SW |
2273 | static void clear_standby(struct ceph_connection *con) |
2274 | { | |
2275 | /* come back from STANDBY? */ | |
2276 | if (test_and_clear_bit(STANDBY, &con->state)) { | |
2277 | mutex_lock(&con->mutex); | |
2278 | dout("clear_standby %p and ++connect_seq\n", con); | |
2279 | con->connect_seq++; | |
2280 | WARN_ON(test_bit(WRITE_PENDING, &con->state)); | |
2281 | WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state)); | |
2282 | mutex_unlock(&con->mutex); | |
2283 | } | |
2284 | } | |
2285 | ||
31b8006e SW |
2286 | /* |
2287 | * Queue up an outgoing message on the given connection. | |
2288 | */ | |
2289 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2290 | { | |
2291 | if (test_bit(CLOSED, &con->state)) { | |
2292 | dout("con_send %p closed, dropping %p\n", con, msg); | |
2293 | ceph_msg_put(msg); | |
2294 | return; | |
2295 | } | |
2296 | ||
2297 | /* set src+dst */ | |
dbad185d | 2298 | msg->hdr.src = con->msgr->inst.name; |
31b8006e | 2299 | |
3ca02ef9 SW |
2300 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
2301 | ||
e84346b7 SW |
2302 | msg->needs_out_seq = true; |
2303 | ||
31b8006e | 2304 | /* queue */ |
ec302645 | 2305 | mutex_lock(&con->mutex); |
31b8006e SW |
2306 | BUG_ON(!list_empty(&msg->list_head)); |
2307 | list_add_tail(&msg->list_head, &con->out_queue); | |
2308 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2309 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2310 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2311 | le32_to_cpu(msg->hdr.front_len), | |
2312 | le32_to_cpu(msg->hdr.middle_len), | |
2313 | le32_to_cpu(msg->hdr.data_len)); | |
ec302645 | 2314 | mutex_unlock(&con->mutex); |
31b8006e SW |
2315 | |
2316 | /* if there wasn't anything waiting to send before, queue | |
2317 | * new work */ | |
e00de341 | 2318 | clear_standby(con); |
31b8006e SW |
2319 | if (test_and_set_bit(WRITE_PENDING, &con->state) == 0) |
2320 | queue_con(con); | |
2321 | } | |
3d14c5d2 | 2322 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2323 | |
2324 | /* | |
2325 | * Revoke a message that was previously queued for send | |
2326 | */ | |
2327 | void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg) | |
2328 | { | |
ec302645 | 2329 | mutex_lock(&con->mutex); |
31b8006e | 2330 | if (!list_empty(&msg->list_head)) { |
ed98adad | 2331 | dout("con_revoke %p msg %p - was on queue\n", con, msg); |
31b8006e SW |
2332 | list_del_init(&msg->list_head); |
2333 | ceph_msg_put(msg); | |
2334 | msg->hdr.seq = 0; | |
ed98adad SW |
2335 | } |
2336 | if (con->out_msg == msg) { | |
2337 | dout("con_revoke %p msg %p - was sending\n", con, msg); | |
2338 | con->out_msg = NULL; | |
31b8006e SW |
2339 | if (con->out_kvec_is_msg) { |
2340 | con->out_skip = con->out_kvec_bytes; | |
2341 | con->out_kvec_is_msg = false; | |
2342 | } | |
ed98adad SW |
2343 | ceph_msg_put(msg); |
2344 | msg->hdr.seq = 0; | |
31b8006e | 2345 | } |
ec302645 | 2346 | mutex_unlock(&con->mutex); |
31b8006e SW |
2347 | } |
2348 | ||
350b1c32 | 2349 | /* |
0d59ab81 | 2350 | * Revoke a message that we may be reading data into |
350b1c32 | 2351 | */ |
0d59ab81 | 2352 | void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg) |
350b1c32 SW |
2353 | { |
2354 | mutex_lock(&con->mutex); | |
0d59ab81 YS |
2355 | if (con->in_msg && con->in_msg == msg) { |
2356 | unsigned front_len = le32_to_cpu(con->in_hdr.front_len); | |
2357 | unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
350b1c32 SW |
2358 | unsigned data_len = le32_to_cpu(con->in_hdr.data_len); |
2359 | ||
2360 | /* skip rest of message */ | |
0d59ab81 | 2361 | dout("con_revoke_pages %p msg %p revoked\n", con, msg); |
350b1c32 SW |
2362 | con->in_base_pos = con->in_base_pos - |
2363 | sizeof(struct ceph_msg_header) - | |
0d59ab81 YS |
2364 | front_len - |
2365 | middle_len - | |
2366 | data_len - | |
350b1c32 | 2367 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2368 | ceph_msg_put(con->in_msg); |
2369 | con->in_msg = NULL; | |
2370 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2371 | con->in_seq++; |
350b1c32 SW |
2372 | } else { |
2373 | dout("con_revoke_pages %p msg %p pages %p no-op\n", | |
0d59ab81 | 2374 | con, con->in_msg, msg); |
350b1c32 SW |
2375 | } |
2376 | mutex_unlock(&con->mutex); | |
2377 | } | |
2378 | ||
31b8006e SW |
2379 | /* |
2380 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2381 | */ | |
2382 | void ceph_con_keepalive(struct ceph_connection *con) | |
2383 | { | |
e00de341 SW |
2384 | dout("con_keepalive %p\n", con); |
2385 | clear_standby(con); | |
31b8006e SW |
2386 | if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 && |
2387 | test_and_set_bit(WRITE_PENDING, &con->state) == 0) | |
2388 | queue_con(con); | |
2389 | } | |
3d14c5d2 | 2390 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2391 | |
2392 | ||
2393 | /* | |
2394 | * construct a new message with given type, size | |
2395 | * the new msg has a ref count of 1. | |
2396 | */ | |
b61c2763 SW |
2397 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2398 | bool can_fail) | |
31b8006e SW |
2399 | { |
2400 | struct ceph_msg *m; | |
2401 | ||
34d23762 | 2402 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2403 | if (m == NULL) |
2404 | goto out; | |
c2e552e7 | 2405 | kref_init(&m->kref); |
31b8006e SW |
2406 | INIT_LIST_HEAD(&m->list_head); |
2407 | ||
45c6ceb5 | 2408 | m->hdr.tid = 0; |
31b8006e | 2409 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2410 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2411 | m->hdr.version = 0; | |
31b8006e SW |
2412 | m->hdr.front_len = cpu_to_le32(front_len); |
2413 | m->hdr.middle_len = 0; | |
bb257664 SW |
2414 | m->hdr.data_len = 0; |
2415 | m->hdr.data_off = 0; | |
45c6ceb5 | 2416 | m->hdr.reserved = 0; |
31b8006e SW |
2417 | m->footer.front_crc = 0; |
2418 | m->footer.middle_crc = 0; | |
2419 | m->footer.data_crc = 0; | |
45c6ceb5 | 2420 | m->footer.flags = 0; |
31b8006e SW |
2421 | m->front_max = front_len; |
2422 | m->front_is_vmalloc = false; | |
2423 | m->more_to_follow = false; | |
c0d5f9db | 2424 | m->ack_stamp = 0; |
31b8006e SW |
2425 | m->pool = NULL; |
2426 | ||
ca20892d HC |
2427 | /* middle */ |
2428 | m->middle = NULL; | |
2429 | ||
2430 | /* data */ | |
2431 | m->nr_pages = 0; | |
2432 | m->page_alignment = 0; | |
2433 | m->pages = NULL; | |
2434 | m->pagelist = NULL; | |
2435 | m->bio = NULL; | |
2436 | m->bio_iter = NULL; | |
2437 | m->bio_seg = 0; | |
2438 | m->trail = NULL; | |
2439 | ||
31b8006e SW |
2440 | /* front */ |
2441 | if (front_len) { | |
2442 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2443 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2444 | PAGE_KERNEL); |
2445 | m->front_is_vmalloc = true; | |
2446 | } else { | |
34d23762 | 2447 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2448 | } |
2449 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2450 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2451 | front_len); |
2452 | goto out2; | |
2453 | } | |
2454 | } else { | |
2455 | m->front.iov_base = NULL; | |
2456 | } | |
2457 | m->front.iov_len = front_len; | |
2458 | ||
bb257664 | 2459 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2460 | return m; |
2461 | ||
2462 | out2: | |
2463 | ceph_msg_put(m); | |
2464 | out: | |
b61c2763 SW |
2465 | if (!can_fail) { |
2466 | pr_err("msg_new can't create type %d front %d\n", type, | |
2467 | front_len); | |
f0ed1b7c | 2468 | WARN_ON(1); |
b61c2763 SW |
2469 | } else { |
2470 | dout("msg_new can't create type %d front %d\n", type, | |
2471 | front_len); | |
2472 | } | |
a79832f2 | 2473 | return NULL; |
31b8006e | 2474 | } |
3d14c5d2 | 2475 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2476 | |
31b8006e SW |
2477 | /* |
2478 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2479 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2480 | * per-type header in the front and then gracefully fail (i.e., | |
2481 | * propagate the error to the caller based on info in the front) when | |
2482 | * the middle is too large. | |
2483 | */ | |
2450418c | 2484 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2485 | { |
2486 | int type = le16_to_cpu(msg->hdr.type); | |
2487 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2488 | ||
2489 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2490 | ceph_msg_type_name(type), middle_len); | |
2491 | BUG_ON(!middle_len); | |
2492 | BUG_ON(msg->middle); | |
2493 | ||
b6c1d5b8 | 2494 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2495 | if (!msg->middle) |
2496 | return -ENOMEM; | |
2497 | return 0; | |
2498 | } | |
2499 | ||
2450418c YS |
2500 | /* |
2501 | * Generic message allocator, for incoming messages. | |
2502 | */ | |
2503 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, | |
2504 | struct ceph_msg_header *hdr, | |
2505 | int *skip) | |
2506 | { | |
2507 | int type = le16_to_cpu(hdr->type); | |
2508 | int front_len = le32_to_cpu(hdr->front_len); | |
2509 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2510 | struct ceph_msg *msg = NULL; | |
2511 | int ret; | |
2512 | ||
2513 | if (con->ops->alloc_msg) { | |
0547a9b3 | 2514 | mutex_unlock(&con->mutex); |
2450418c | 2515 | msg = con->ops->alloc_msg(con, hdr, skip); |
0547a9b3 | 2516 | mutex_lock(&con->mutex); |
a79832f2 | 2517 | if (!msg || *skip) |
2450418c YS |
2518 | return NULL; |
2519 | } | |
2520 | if (!msg) { | |
2521 | *skip = 0; | |
b61c2763 | 2522 | msg = ceph_msg_new(type, front_len, GFP_NOFS, false); |
2450418c YS |
2523 | if (!msg) { |
2524 | pr_err("unable to allocate msg type %d len %d\n", | |
2525 | type, front_len); | |
a79832f2 | 2526 | return NULL; |
2450418c | 2527 | } |
c5c6b19d | 2528 | msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2529 | } |
9d7f0f13 | 2530 | memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2531 | |
bb257664 | 2532 | if (middle_len && !msg->middle) { |
2450418c | 2533 | ret = ceph_alloc_middle(con, msg); |
2450418c YS |
2534 | if (ret < 0) { |
2535 | ceph_msg_put(msg); | |
a79832f2 | 2536 | return NULL; |
2450418c YS |
2537 | } |
2538 | } | |
9d7f0f13 | 2539 | |
2450418c YS |
2540 | return msg; |
2541 | } | |
2542 | ||
31b8006e SW |
2543 | |
2544 | /* | |
2545 | * Free a generically kmalloc'd message. | |
2546 | */ | |
2547 | void ceph_msg_kfree(struct ceph_msg *m) | |
2548 | { | |
2549 | dout("msg_kfree %p\n", m); | |
2550 | if (m->front_is_vmalloc) | |
2551 | vfree(m->front.iov_base); | |
2552 | else | |
2553 | kfree(m->front.iov_base); | |
2554 | kfree(m); | |
2555 | } | |
2556 | ||
2557 | /* | |
2558 | * Drop a msg ref. Destroy as needed. | |
2559 | */ | |
c2e552e7 SW |
2560 | void ceph_msg_last_put(struct kref *kref) |
2561 | { | |
2562 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2563 | |
c2e552e7 SW |
2564 | dout("ceph_msg_put last one on %p\n", m); |
2565 | WARN_ON(!list_empty(&m->list_head)); | |
2566 | ||
2567 | /* drop middle, data, if any */ | |
2568 | if (m->middle) { | |
2569 | ceph_buffer_put(m->middle); | |
2570 | m->middle = NULL; | |
31b8006e | 2571 | } |
c2e552e7 SW |
2572 | m->nr_pages = 0; |
2573 | m->pages = NULL; | |
2574 | ||
58bb3b37 SW |
2575 | if (m->pagelist) { |
2576 | ceph_pagelist_release(m->pagelist); | |
2577 | kfree(m->pagelist); | |
2578 | m->pagelist = NULL; | |
2579 | } | |
2580 | ||
68b4476b YS |
2581 | m->trail = NULL; |
2582 | ||
c2e552e7 SW |
2583 | if (m->pool) |
2584 | ceph_msgpool_put(m->pool, m); | |
2585 | else | |
2586 | ceph_msg_kfree(m); | |
31b8006e | 2587 | } |
3d14c5d2 | 2588 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2589 | |
2590 | void ceph_msg_dump(struct ceph_msg *msg) | |
2591 | { | |
2592 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2593 | msg->front_max, msg->nr_pages); | |
2594 | print_hex_dump(KERN_DEBUG, "header: ", | |
2595 | DUMP_PREFIX_OFFSET, 16, 1, | |
2596 | &msg->hdr, sizeof(msg->hdr), true); | |
2597 | print_hex_dump(KERN_DEBUG, " front: ", | |
2598 | DUMP_PREFIX_OFFSET, 16, 1, | |
2599 | msg->front.iov_base, msg->front.iov_len, true); | |
2600 | if (msg->middle) | |
2601 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2602 | DUMP_PREFIX_OFFSET, 16, 1, | |
2603 | msg->middle->vec.iov_base, | |
2604 | msg->middle->vec.iov_len, true); | |
2605 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2606 | DUMP_PREFIX_OFFSET, 16, 1, | |
2607 | &msg->footer, sizeof(msg->footer), true); | |
2608 | } | |
3d14c5d2 | 2609 | EXPORT_SYMBOL(ceph_msg_dump); |