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; | |
b1c6b980 AE |
659 | int auth_len; |
660 | int auth_protocol; | |
a3530df3 | 661 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
662 | |
663 | if (!con->ops->get_authorizer) { | |
664 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
665 | con->out_connect.authorizer_len = 0; | |
666 | ||
667 | return 0; | |
668 | } | |
669 | ||
670 | /* Can't hold the mutex while getting authorizer */ | |
4e7a5dcd | 671 | |
ec302645 | 672 | mutex_unlock(&con->mutex); |
b1c6b980 | 673 | |
b1c6b980 | 674 | auth_protocol = CEPH_AUTH_UNKNOWN; |
8f43fb53 AE |
675 | auth = con->ops->get_authorizer(con, &auth_protocol, con->auth_retry); |
676 | ||
ec302645 | 677 | mutex_lock(&con->mutex); |
4e7a5dcd | 678 | |
a3530df3 AE |
679 | if (IS_ERR(auth)) |
680 | return PTR_ERR(auth); | |
b1c6b980 | 681 | if (test_bit(CLOSED, &con->state) || test_bit(OPENING, &con->state)) |
0da5d703 SW |
682 | return -EAGAIN; |
683 | ||
8f43fb53 AE |
684 | auth_buf = auth->authorizer_buf; |
685 | auth_len = auth->authorizer_buf_len; | |
686 | con->auth_reply_buf = auth->authorizer_reply_buf; | |
687 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
688 | ||
4e7a5dcd SW |
689 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol); |
690 | con->out_connect.authorizer_len = cpu_to_le32(auth_len); | |
691 | ||
859eb799 AE |
692 | if (auth_len) |
693 | ceph_con_out_kvec_add(con, auth_len, auth_buf); | |
694 | ||
0da5d703 | 695 | return 0; |
4e7a5dcd SW |
696 | } |
697 | ||
31b8006e SW |
698 | /* |
699 | * We connected to a peer and are saying hello. | |
700 | */ | |
e825a66d | 701 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 702 | { |
859eb799 | 703 | ceph_con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
e825a66d AE |
704 | ceph_con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), |
705 | &con->msgr->my_enc_addr); | |
eed0ef2c | 706 | |
eed0ef2c SW |
707 | con->out_more = 0; |
708 | set_bit(WRITE_PENDING, &con->state); | |
709 | } | |
710 | ||
e825a66d | 711 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 712 | { |
31b8006e SW |
713 | unsigned global_seq = get_global_seq(con->msgr, 0); |
714 | int proto; | |
e10c758e | 715 | int ret; |
31b8006e SW |
716 | |
717 | switch (con->peer_name.type) { | |
718 | case CEPH_ENTITY_TYPE_MON: | |
719 | proto = CEPH_MONC_PROTOCOL; | |
720 | break; | |
721 | case CEPH_ENTITY_TYPE_OSD: | |
722 | proto = CEPH_OSDC_PROTOCOL; | |
723 | break; | |
724 | case CEPH_ENTITY_TYPE_MDS: | |
725 | proto = CEPH_MDSC_PROTOCOL; | |
726 | break; | |
727 | default: | |
728 | BUG(); | |
729 | } | |
730 | ||
731 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
732 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 733 | |
e825a66d | 734 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
735 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
736 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
737 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
738 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
739 | con->out_connect.flags = 0; | |
31b8006e | 740 | |
859eb799 | 741 | ceph_con_out_kvec_add(con, sizeof (con->out_connect), &con->out_connect); |
e10c758e AE |
742 | ret = prepare_connect_authorizer(con); |
743 | if (ret) | |
744 | return ret; | |
859eb799 | 745 | |
31b8006e SW |
746 | con->out_more = 0; |
747 | set_bit(WRITE_PENDING, &con->state); | |
4e7a5dcd | 748 | |
e10c758e | 749 | return 0; |
31b8006e SW |
750 | } |
751 | ||
31b8006e SW |
752 | /* |
753 | * write as much of pending kvecs to the socket as we can. | |
754 | * 1 -> done | |
755 | * 0 -> socket full, but more to do | |
756 | * <0 -> error | |
757 | */ | |
758 | static int write_partial_kvec(struct ceph_connection *con) | |
759 | { | |
760 | int ret; | |
761 | ||
762 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
763 | while (con->out_kvec_bytes > 0) { | |
764 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
765 | con->out_kvec_left, con->out_kvec_bytes, | |
766 | con->out_more); | |
767 | if (ret <= 0) | |
768 | goto out; | |
769 | con->out_kvec_bytes -= ret; | |
770 | if (con->out_kvec_bytes == 0) | |
771 | break; /* done */ | |
f42299e6 AE |
772 | |
773 | /* account for full iov entries consumed */ | |
774 | while (ret >= con->out_kvec_cur->iov_len) { | |
775 | BUG_ON(!con->out_kvec_left); | |
776 | ret -= con->out_kvec_cur->iov_len; | |
777 | con->out_kvec_cur++; | |
778 | con->out_kvec_left--; | |
779 | } | |
780 | /* and for a partially-consumed entry */ | |
781 | if (ret) { | |
782 | con->out_kvec_cur->iov_len -= ret; | |
783 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
784 | } |
785 | } | |
786 | con->out_kvec_left = 0; | |
787 | con->out_kvec_is_msg = false; | |
788 | ret = 1; | |
789 | out: | |
790 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
791 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
792 | return ret; /* done! */ | |
793 | } | |
794 | ||
68b4476b YS |
795 | #ifdef CONFIG_BLOCK |
796 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
797 | { | |
798 | if (!bio) { | |
799 | *iter = NULL; | |
800 | *seg = 0; | |
801 | return; | |
802 | } | |
803 | *iter = bio; | |
804 | *seg = bio->bi_idx; | |
805 | } | |
806 | ||
807 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
808 | { | |
809 | if (*bio_iter == NULL) | |
810 | return; | |
811 | ||
812 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
813 | ||
814 | (*seg)++; | |
815 | if (*seg == (*bio_iter)->bi_vcnt) | |
816 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
817 | } | |
818 | #endif | |
819 | ||
31b8006e SW |
820 | /* |
821 | * Write as much message data payload as we can. If we finish, queue | |
822 | * up the footer. | |
823 | * 1 -> done, footer is now queued in out_kvec[]. | |
824 | * 0 -> socket full, but more to do | |
825 | * <0 -> error | |
826 | */ | |
827 | static int write_partial_msg_pages(struct ceph_connection *con) | |
828 | { | |
829 | struct ceph_msg *msg = con->out_msg; | |
830 | unsigned data_len = le32_to_cpu(msg->hdr.data_len); | |
831 | size_t len; | |
37675b0f | 832 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 833 | int ret; |
68b4476b YS |
834 | int total_max_write; |
835 | int in_trail = 0; | |
836 | size_t trail_len = (msg->trail ? msg->trail->length : 0); | |
31b8006e SW |
837 | |
838 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
839 | con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages, | |
840 | con->out_msg_pos.page_pos); | |
841 | ||
68b4476b YS |
842 | #ifdef CONFIG_BLOCK |
843 | if (msg->bio && !msg->bio_iter) | |
844 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); | |
845 | #endif | |
846 | ||
847 | while (data_len > con->out_msg_pos.data_pos) { | |
31b8006e | 848 | struct page *page = NULL; |
68b4476b | 849 | int max_write = PAGE_SIZE; |
9bd19663 | 850 | int bio_offset = 0; |
68b4476b YS |
851 | |
852 | total_max_write = data_len - trail_len - | |
853 | con->out_msg_pos.data_pos; | |
31b8006e SW |
854 | |
855 | /* | |
856 | * if we are calculating the data crc (the default), we need | |
857 | * to map the page. if our pages[] has been revoked, use the | |
858 | * zero page. | |
859 | */ | |
68b4476b YS |
860 | |
861 | /* have we reached the trail part of the data? */ | |
862 | if (con->out_msg_pos.data_pos >= data_len - trail_len) { | |
863 | in_trail = 1; | |
864 | ||
865 | total_max_write = data_len - con->out_msg_pos.data_pos; | |
866 | ||
867 | page = list_first_entry(&msg->trail->head, | |
868 | struct page, lru); | |
68b4476b YS |
869 | max_write = PAGE_SIZE; |
870 | } else if (msg->pages) { | |
31b8006e | 871 | page = msg->pages[con->out_msg_pos.page]; |
58bb3b37 SW |
872 | } else if (msg->pagelist) { |
873 | page = list_first_entry(&msg->pagelist->head, | |
874 | struct page, lru); | |
68b4476b YS |
875 | #ifdef CONFIG_BLOCK |
876 | } else if (msg->bio) { | |
877 | struct bio_vec *bv; | |
878 | ||
879 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
880 | page = bv->bv_page; | |
9bd19663 | 881 | bio_offset = bv->bv_offset; |
68b4476b YS |
882 | max_write = bv->bv_len; |
883 | #endif | |
31b8006e | 884 | } else { |
57666519 | 885 | page = zero_page; |
31b8006e | 886 | } |
68b4476b YS |
887 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
888 | total_max_write); | |
889 | ||
37675b0f | 890 | if (do_datacrc && !con->out_msg_pos.did_page_crc) { |
9bd19663 | 891 | void *base; |
a9a0c51a | 892 | u32 crc; |
31b8006e | 893 | u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc); |
8d63e318 | 894 | char *kaddr; |
31b8006e | 895 | |
8d63e318 | 896 | kaddr = kmap(page); |
31b8006e | 897 | BUG_ON(kaddr == NULL); |
9bd19663 | 898 | base = kaddr + con->out_msg_pos.page_pos + bio_offset; |
a9a0c51a AE |
899 | crc = crc32c(tmpcrc, base, len); |
900 | con->out_msg->footer.data_crc = cpu_to_le32(crc); | |
bca064d2 | 901 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 902 | } |
e36b13cc | 903 | ret = ceph_tcp_sendpage(con->sock, page, |
9bd19663 | 904 | con->out_msg_pos.page_pos + bio_offset, |
e36b13cc | 905 | len, 1); |
31b8006e | 906 | |
0cdf9e60 | 907 | if (do_datacrc) |
31b8006e SW |
908 | kunmap(page); |
909 | ||
910 | if (ret <= 0) | |
911 | goto out; | |
912 | ||
913 | con->out_msg_pos.data_pos += ret; | |
914 | con->out_msg_pos.page_pos += ret; | |
915 | if (ret == len) { | |
916 | con->out_msg_pos.page_pos = 0; | |
917 | con->out_msg_pos.page++; | |
bca064d2 | 918 | con->out_msg_pos.did_page_crc = false; |
68b4476b YS |
919 | if (in_trail) |
920 | list_move_tail(&page->lru, | |
921 | &msg->trail->head); | |
922 | else if (msg->pagelist) | |
58bb3b37 SW |
923 | list_move_tail(&page->lru, |
924 | &msg->pagelist->head); | |
68b4476b YS |
925 | #ifdef CONFIG_BLOCK |
926 | else if (msg->bio) | |
927 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
928 | #endif | |
31b8006e SW |
929 | } |
930 | } | |
931 | ||
932 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
933 | ||
934 | /* prepare and queue up footer, too */ | |
37675b0f | 935 | if (!do_datacrc) |
31b8006e | 936 | con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
859eb799 AE |
937 | ceph_con_out_kvec_reset(con); |
938 | prepare_write_message_footer(con); | |
31b8006e SW |
939 | ret = 1; |
940 | out: | |
941 | return ret; | |
942 | } | |
943 | ||
944 | /* | |
945 | * write some zeros | |
946 | */ | |
947 | static int write_partial_skip(struct ceph_connection *con) | |
948 | { | |
949 | int ret; | |
950 | ||
951 | while (con->out_skip > 0) { | |
31739139 | 952 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 953 | |
31739139 | 954 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1); |
31b8006e SW |
955 | if (ret <= 0) |
956 | goto out; | |
957 | con->out_skip -= ret; | |
958 | } | |
959 | ret = 1; | |
960 | out: | |
961 | return ret; | |
962 | } | |
963 | ||
964 | /* | |
965 | * Prepare to read connection handshake, or an ack. | |
966 | */ | |
eed0ef2c SW |
967 | static void prepare_read_banner(struct ceph_connection *con) |
968 | { | |
969 | dout("prepare_read_banner %p\n", con); | |
970 | con->in_base_pos = 0; | |
971 | } | |
972 | ||
31b8006e SW |
973 | static void prepare_read_connect(struct ceph_connection *con) |
974 | { | |
975 | dout("prepare_read_connect %p\n", con); | |
976 | con->in_base_pos = 0; | |
977 | } | |
978 | ||
979 | static void prepare_read_ack(struct ceph_connection *con) | |
980 | { | |
981 | dout("prepare_read_ack %p\n", con); | |
982 | con->in_base_pos = 0; | |
983 | } | |
984 | ||
985 | static void prepare_read_tag(struct ceph_connection *con) | |
986 | { | |
987 | dout("prepare_read_tag %p\n", con); | |
988 | con->in_base_pos = 0; | |
989 | con->in_tag = CEPH_MSGR_TAG_READY; | |
990 | } | |
991 | ||
992 | /* | |
993 | * Prepare to read a message. | |
994 | */ | |
995 | static int prepare_read_message(struct ceph_connection *con) | |
996 | { | |
997 | dout("prepare_read_message %p\n", con); | |
998 | BUG_ON(con->in_msg != NULL); | |
999 | con->in_base_pos = 0; | |
1000 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1001 | return 0; | |
1002 | } | |
1003 | ||
1004 | ||
1005 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1006 | int end, int size, void *object) |
31b8006e | 1007 | { |
e6cee71f AE |
1008 | while (con->in_base_pos < end) { |
1009 | int left = end - con->in_base_pos; | |
31b8006e SW |
1010 | int have = size - left; |
1011 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1012 | if (ret <= 0) | |
1013 | return ret; | |
1014 | con->in_base_pos += ret; | |
1015 | } | |
1016 | return 1; | |
1017 | } | |
1018 | ||
1019 | ||
1020 | /* | |
1021 | * Read all or part of the connect-side handshake on a new connection | |
1022 | */ | |
eed0ef2c | 1023 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1024 | { |
fd51653f AE |
1025 | int size; |
1026 | int end; | |
1027 | int ret; | |
31b8006e | 1028 | |
eed0ef2c | 1029 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1030 | |
1031 | /* peer's banner */ | |
fd51653f AE |
1032 | size = strlen(CEPH_BANNER); |
1033 | end = size; | |
1034 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1035 | if (ret <= 0) |
1036 | goto out; | |
fd51653f AE |
1037 | |
1038 | size = sizeof (con->actual_peer_addr); | |
1039 | end += size; | |
1040 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1041 | if (ret <= 0) |
1042 | goto out; | |
fd51653f AE |
1043 | |
1044 | size = sizeof (con->peer_addr_for_me); | |
1045 | end += size; | |
1046 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1047 | if (ret <= 0) |
1048 | goto out; | |
fd51653f | 1049 | |
eed0ef2c SW |
1050 | out: |
1051 | return ret; | |
1052 | } | |
1053 | ||
1054 | static int read_partial_connect(struct ceph_connection *con) | |
1055 | { | |
fd51653f AE |
1056 | int size; |
1057 | int end; | |
1058 | int ret; | |
eed0ef2c SW |
1059 | |
1060 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1061 | ||
fd51653f AE |
1062 | size = sizeof (con->in_reply); |
1063 | end = size; | |
1064 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1065 | if (ret <= 0) |
1066 | goto out; | |
fd51653f AE |
1067 | |
1068 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1069 | end += size; | |
1070 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1071 | if (ret <= 0) |
1072 | goto out; | |
31b8006e | 1073 | |
4e7a5dcd SW |
1074 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1075 | con, (int)con->in_reply.tag, | |
1076 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1077 | le32_to_cpu(con->in_reply.global_seq)); |
1078 | out: | |
1079 | return ret; | |
eed0ef2c | 1080 | |
31b8006e SW |
1081 | } |
1082 | ||
1083 | /* | |
1084 | * Verify the hello banner looks okay. | |
1085 | */ | |
1086 | static int verify_hello(struct ceph_connection *con) | |
1087 | { | |
1088 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1089 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1090 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1091 | con->error_msg = "protocol error, bad banner"; |
1092 | return -1; | |
1093 | } | |
1094 | return 0; | |
1095 | } | |
1096 | ||
1097 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1098 | { | |
1099 | switch (ss->ss_family) { | |
1100 | case AF_INET: | |
1101 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1102 | case AF_INET6: | |
1103 | return | |
1104 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1105 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1106 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1107 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1108 | } | |
1109 | return false; | |
1110 | } | |
1111 | ||
1112 | static int addr_port(struct sockaddr_storage *ss) | |
1113 | { | |
1114 | switch (ss->ss_family) { | |
1115 | case AF_INET: | |
f28bcfbe | 1116 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1117 | case AF_INET6: |
f28bcfbe | 1118 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1119 | } |
1120 | return 0; | |
1121 | } | |
1122 | ||
1123 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1124 | { | |
1125 | switch (ss->ss_family) { | |
1126 | case AF_INET: | |
1127 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1128 | break; |
31b8006e SW |
1129 | case AF_INET6: |
1130 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1131 | break; |
31b8006e SW |
1132 | } |
1133 | } | |
1134 | ||
ee3b56f2 NW |
1135 | /* |
1136 | * Unlike other *_pton function semantics, zero indicates success. | |
1137 | */ | |
1138 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1139 | char delim, const char **ipend) | |
1140 | { | |
99f0f3b2 AE |
1141 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1142 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1143 | |
1144 | memset(ss, 0, sizeof(*ss)); | |
1145 | ||
1146 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1147 | ss->ss_family = AF_INET; | |
1148 | return 0; | |
1149 | } | |
1150 | ||
1151 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1152 | ss->ss_family = AF_INET6; | |
1153 | return 0; | |
1154 | } | |
1155 | ||
1156 | return -EINVAL; | |
1157 | } | |
1158 | ||
1159 | /* | |
1160 | * Extract hostname string and resolve using kernel DNS facility. | |
1161 | */ | |
1162 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1163 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1164 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1165 | { | |
1166 | const char *end, *delim_p; | |
1167 | char *colon_p, *ip_addr = NULL; | |
1168 | int ip_len, ret; | |
1169 | ||
1170 | /* | |
1171 | * The end of the hostname occurs immediately preceding the delimiter or | |
1172 | * the port marker (':') where the delimiter takes precedence. | |
1173 | */ | |
1174 | delim_p = memchr(name, delim, namelen); | |
1175 | colon_p = memchr(name, ':', namelen); | |
1176 | ||
1177 | if (delim_p && colon_p) | |
1178 | end = delim_p < colon_p ? delim_p : colon_p; | |
1179 | else if (!delim_p && colon_p) | |
1180 | end = colon_p; | |
1181 | else { | |
1182 | end = delim_p; | |
1183 | if (!end) /* case: hostname:/ */ | |
1184 | end = name + namelen; | |
1185 | } | |
1186 | ||
1187 | if (end <= name) | |
1188 | return -EINVAL; | |
1189 | ||
1190 | /* do dns_resolve upcall */ | |
1191 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1192 | if (ip_len > 0) | |
1193 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1194 | else | |
1195 | ret = -ESRCH; | |
1196 | ||
1197 | kfree(ip_addr); | |
1198 | ||
1199 | *ipend = end; | |
1200 | ||
1201 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1202 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1203 | ||
1204 | return ret; | |
1205 | } | |
1206 | #else | |
1207 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1208 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1209 | { | |
1210 | return -EINVAL; | |
1211 | } | |
1212 | #endif | |
1213 | ||
1214 | /* | |
1215 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1216 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1217 | */ | |
1218 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1219 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1220 | { | |
1221 | int ret; | |
1222 | ||
1223 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1224 | if (ret) | |
1225 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1226 | ||
1227 | return ret; | |
1228 | } | |
1229 | ||
31b8006e SW |
1230 | /* |
1231 | * Parse an ip[:port] list into an addr array. Use the default | |
1232 | * monitor port if a port isn't specified. | |
1233 | */ | |
1234 | int ceph_parse_ips(const char *c, const char *end, | |
1235 | struct ceph_entity_addr *addr, | |
1236 | int max_count, int *count) | |
1237 | { | |
ee3b56f2 | 1238 | int i, ret = -EINVAL; |
31b8006e SW |
1239 | const char *p = c; |
1240 | ||
1241 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1242 | for (i = 0; i < max_count; i++) { | |
1243 | const char *ipend; | |
1244 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1245 | int port; |
39139f64 SW |
1246 | char delim = ','; |
1247 | ||
1248 | if (*p == '[') { | |
1249 | delim = ']'; | |
1250 | p++; | |
1251 | } | |
31b8006e | 1252 | |
ee3b56f2 NW |
1253 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1254 | if (ret) | |
31b8006e | 1255 | goto bad; |
ee3b56f2 NW |
1256 | ret = -EINVAL; |
1257 | ||
31b8006e SW |
1258 | p = ipend; |
1259 | ||
39139f64 SW |
1260 | if (delim == ']') { |
1261 | if (*p != ']') { | |
1262 | dout("missing matching ']'\n"); | |
1263 | goto bad; | |
1264 | } | |
1265 | p++; | |
1266 | } | |
1267 | ||
31b8006e SW |
1268 | /* port? */ |
1269 | if (p < end && *p == ':') { | |
1270 | port = 0; | |
1271 | p++; | |
1272 | while (p < end && *p >= '0' && *p <= '9') { | |
1273 | port = (port * 10) + (*p - '0'); | |
1274 | p++; | |
1275 | } | |
1276 | if (port > 65535 || port == 0) | |
1277 | goto bad; | |
1278 | } else { | |
1279 | port = CEPH_MON_PORT; | |
1280 | } | |
1281 | ||
1282 | addr_set_port(ss, port); | |
1283 | ||
3d14c5d2 | 1284 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1285 | |
1286 | if (p == end) | |
1287 | break; | |
1288 | if (*p != ',') | |
1289 | goto bad; | |
1290 | p++; | |
1291 | } | |
1292 | ||
1293 | if (p != end) | |
1294 | goto bad; | |
1295 | ||
1296 | if (count) | |
1297 | *count = i + 1; | |
1298 | return 0; | |
1299 | ||
1300 | bad: | |
39139f64 | 1301 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1302 | return ret; |
31b8006e | 1303 | } |
3d14c5d2 | 1304 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1305 | |
eed0ef2c | 1306 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1307 | { |
eed0ef2c | 1308 | dout("process_banner on %p\n", con); |
31b8006e SW |
1309 | |
1310 | if (verify_hello(con) < 0) | |
1311 | return -1; | |
1312 | ||
63f2d211 SW |
1313 | ceph_decode_addr(&con->actual_peer_addr); |
1314 | ceph_decode_addr(&con->peer_addr_for_me); | |
1315 | ||
31b8006e SW |
1316 | /* |
1317 | * Make sure the other end is who we wanted. note that the other | |
1318 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1319 | * them the benefit of the doubt. | |
1320 | */ | |
103e2d3a SW |
1321 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1322 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1323 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1324 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1325 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1326 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1327 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1328 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1329 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1330 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1331 | return -1; |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * did we learn our address? | |
1336 | */ | |
1337 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1338 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1339 | ||
1340 | memcpy(&con->msgr->inst.addr.in_addr, | |
1341 | &con->peer_addr_for_me.in_addr, | |
1342 | sizeof(con->peer_addr_for_me.in_addr)); | |
1343 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1344 | encode_my_addr(con->msgr); |
eed0ef2c | 1345 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1346 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1347 | } |
1348 | ||
eed0ef2c SW |
1349 | set_bit(NEGOTIATING, &con->state); |
1350 | prepare_read_connect(con); | |
1351 | return 0; | |
1352 | } | |
1353 | ||
04a419f9 SW |
1354 | static void fail_protocol(struct ceph_connection *con) |
1355 | { | |
1356 | reset_connection(con); | |
1357 | set_bit(CLOSED, &con->state); /* in case there's queued work */ | |
1358 | ||
1359 | mutex_unlock(&con->mutex); | |
1360 | if (con->ops->bad_proto) | |
1361 | con->ops->bad_proto(con); | |
1362 | mutex_lock(&con->mutex); | |
1363 | } | |
1364 | ||
eed0ef2c SW |
1365 | static int process_connect(struct ceph_connection *con) |
1366 | { | |
3d14c5d2 YS |
1367 | u64 sup_feat = con->msgr->supported_features; |
1368 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1369 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1370 | int ret; |
04a419f9 | 1371 | |
eed0ef2c SW |
1372 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1373 | ||
31b8006e | 1374 | switch (con->in_reply.tag) { |
04a419f9 SW |
1375 | case CEPH_MSGR_TAG_FEATURES: |
1376 | pr_err("%s%lld %s feature set mismatch," | |
1377 | " my %llx < server's %llx, missing %llx\n", | |
1378 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1379 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1380 | sup_feat, server_feat, server_feat & ~sup_feat); |
1381 | con->error_msg = "missing required protocol features"; | |
1382 | fail_protocol(con); | |
1383 | return -1; | |
1384 | ||
31b8006e | 1385 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1386 | pr_err("%s%lld %s protocol version mismatch," |
1387 | " my %d != server's %d\n", | |
1388 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1389 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1390 | le32_to_cpu(con->out_connect.protocol_version), |
1391 | le32_to_cpu(con->in_reply.protocol_version)); | |
1392 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1393 | fail_protocol(con); |
31b8006e SW |
1394 | return -1; |
1395 | ||
4e7a5dcd SW |
1396 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1397 | con->auth_retry++; | |
1398 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1399 | con->auth_retry); | |
1400 | if (con->auth_retry == 2) { | |
1401 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1402 | return -1; |
1403 | } | |
1404 | con->auth_retry = 1; | |
84fb3adf | 1405 | ceph_con_out_kvec_reset(con); |
e825a66d | 1406 | ret = prepare_write_connect(con); |
0da5d703 SW |
1407 | if (ret < 0) |
1408 | return ret; | |
63733a0f | 1409 | prepare_read_connect(con); |
4e7a5dcd | 1410 | break; |
31b8006e SW |
1411 | |
1412 | case CEPH_MSGR_TAG_RESETSESSION: | |
1413 | /* | |
1414 | * If we connected with a large connect_seq but the peer | |
1415 | * has no record of a session with us (no connection, or | |
1416 | * connect_seq == 0), they will send RESETSESION to indicate | |
1417 | * that they must have reset their session, and may have | |
1418 | * dropped messages. | |
1419 | */ | |
1420 | dout("process_connect got RESET peer seq %u\n", | |
1421 | le32_to_cpu(con->in_connect.connect_seq)); | |
1422 | pr_err("%s%lld %s connection reset\n", | |
1423 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1424 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1425 | reset_connection(con); |
84fb3adf | 1426 | ceph_con_out_kvec_reset(con); |
5a0f8fdd AE |
1427 | ret = prepare_write_connect(con); |
1428 | if (ret < 0) | |
1429 | return ret; | |
31b8006e SW |
1430 | prepare_read_connect(con); |
1431 | ||
1432 | /* Tell ceph about it. */ | |
ec302645 | 1433 | mutex_unlock(&con->mutex); |
31b8006e SW |
1434 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1435 | if (con->ops->peer_reset) | |
1436 | con->ops->peer_reset(con); | |
ec302645 | 1437 | mutex_lock(&con->mutex); |
0da5d703 SW |
1438 | if (test_bit(CLOSED, &con->state) || |
1439 | test_bit(OPENING, &con->state)) | |
1440 | return -EAGAIN; | |
31b8006e SW |
1441 | break; |
1442 | ||
1443 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1444 | /* | |
1445 | * If we sent a smaller connect_seq than the peer has, try | |
1446 | * again with a larger value. | |
1447 | */ | |
1448 | dout("process_connect got RETRY my seq = %u, peer_seq = %u\n", | |
1449 | le32_to_cpu(con->out_connect.connect_seq), | |
1450 | le32_to_cpu(con->in_connect.connect_seq)); | |
1451 | con->connect_seq = le32_to_cpu(con->in_connect.connect_seq); | |
84fb3adf | 1452 | ceph_con_out_kvec_reset(con); |
5a0f8fdd AE |
1453 | ret = prepare_write_connect(con); |
1454 | if (ret < 0) | |
1455 | return ret; | |
31b8006e SW |
1456 | prepare_read_connect(con); |
1457 | break; | |
1458 | ||
1459 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1460 | /* | |
1461 | * If we sent a smaller global_seq than the peer has, try | |
1462 | * again with a larger value. | |
1463 | */ | |
eed0ef2c | 1464 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e SW |
1465 | con->peer_global_seq, |
1466 | le32_to_cpu(con->in_connect.global_seq)); | |
1467 | get_global_seq(con->msgr, | |
1468 | le32_to_cpu(con->in_connect.global_seq)); | |
84fb3adf | 1469 | ceph_con_out_kvec_reset(con); |
5a0f8fdd AE |
1470 | ret = prepare_write_connect(con); |
1471 | if (ret < 0) | |
1472 | return ret; | |
31b8006e SW |
1473 | prepare_read_connect(con); |
1474 | break; | |
1475 | ||
1476 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1477 | if (req_feat & ~server_feat) { |
1478 | pr_err("%s%lld %s protocol feature mismatch," | |
1479 | " my required %llx > server's %llx, need %llx\n", | |
1480 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1481 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1482 | req_feat, server_feat, req_feat & ~server_feat); |
1483 | con->error_msg = "missing required protocol features"; | |
1484 | fail_protocol(con); | |
1485 | return -1; | |
1486 | } | |
31b8006e | 1487 | clear_bit(CONNECTING, &con->state); |
31b8006e SW |
1488 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1489 | con->connect_seq++; | |
aba558e2 | 1490 | con->peer_features = server_feat; |
31b8006e SW |
1491 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1492 | con->peer_global_seq, | |
1493 | le32_to_cpu(con->in_reply.connect_seq), | |
1494 | con->connect_seq); | |
1495 | WARN_ON(con->connect_seq != | |
1496 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1497 | |
1498 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
1499 | set_bit(LOSSYTX, &con->state); | |
1500 | ||
31b8006e SW |
1501 | prepare_read_tag(con); |
1502 | break; | |
1503 | ||
1504 | case CEPH_MSGR_TAG_WAIT: | |
1505 | /* | |
1506 | * If there is a connection race (we are opening | |
1507 | * connections to each other), one of us may just have | |
1508 | * to WAIT. This shouldn't happen if we are the | |
1509 | * client. | |
1510 | */ | |
04177882 SW |
1511 | pr_err("process_connect got WAIT as client\n"); |
1512 | con->error_msg = "protocol error, got WAIT as client"; | |
1513 | return -1; | |
31b8006e SW |
1514 | |
1515 | default: | |
1516 | pr_err("connect protocol error, will retry\n"); | |
1517 | con->error_msg = "protocol error, garbage tag during connect"; | |
1518 | return -1; | |
1519 | } | |
1520 | return 0; | |
1521 | } | |
1522 | ||
1523 | ||
1524 | /* | |
1525 | * read (part of) an ack | |
1526 | */ | |
1527 | static int read_partial_ack(struct ceph_connection *con) | |
1528 | { | |
fd51653f AE |
1529 | int size = sizeof (con->in_temp_ack); |
1530 | int end = size; | |
1531 | ||
1532 | return read_partial(con, end, size, &con->in_temp_ack); | |
31b8006e SW |
1533 | } |
1534 | ||
1535 | ||
1536 | /* | |
1537 | * We can finally discard anything that's been acked. | |
1538 | */ | |
1539 | static void process_ack(struct ceph_connection *con) | |
1540 | { | |
1541 | struct ceph_msg *m; | |
1542 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1543 | u64 seq; | |
1544 | ||
31b8006e SW |
1545 | while (!list_empty(&con->out_sent)) { |
1546 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1547 | list_head); | |
1548 | seq = le64_to_cpu(m->hdr.seq); | |
1549 | if (seq > ack) | |
1550 | break; | |
1551 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1552 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1553 | m->ack_stamp = jiffies; |
31b8006e SW |
1554 | ceph_msg_remove(m); |
1555 | } | |
31b8006e SW |
1556 | prepare_read_tag(con); |
1557 | } | |
1558 | ||
1559 | ||
1560 | ||
1561 | ||
2450418c | 1562 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1563 | struct kvec *section, |
1564 | unsigned int sec_len, u32 *crc) | |
2450418c | 1565 | { |
68b4476b | 1566 | int ret, left; |
2450418c YS |
1567 | |
1568 | BUG_ON(!section); | |
1569 | ||
1570 | while (section->iov_len < sec_len) { | |
1571 | BUG_ON(section->iov_base == NULL); | |
1572 | left = sec_len - section->iov_len; | |
1573 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1574 | section->iov_len, left); | |
1575 | if (ret <= 0) | |
1576 | return ret; | |
1577 | section->iov_len += ret; | |
2450418c | 1578 | } |
fe3ad593 AE |
1579 | if (section->iov_len == sec_len) |
1580 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1581 | |
2450418c YS |
1582 | return 1; |
1583 | } | |
31b8006e | 1584 | |
2450418c YS |
1585 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, |
1586 | struct ceph_msg_header *hdr, | |
1587 | int *skip); | |
68b4476b YS |
1588 | |
1589 | ||
1590 | static int read_partial_message_pages(struct ceph_connection *con, | |
1591 | struct page **pages, | |
bca064d2 | 1592 | unsigned data_len, bool do_datacrc) |
68b4476b YS |
1593 | { |
1594 | void *p; | |
1595 | int ret; | |
1596 | int left; | |
1597 | ||
1598 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1599 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1600 | /* (page) data */ | |
1601 | BUG_ON(pages == NULL); | |
1602 | p = kmap(pages[con->in_msg_pos.page]); | |
1603 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1604 | left); | |
bca064d2 | 1605 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1606 | con->in_data_crc = |
1607 | crc32c(con->in_data_crc, | |
1608 | p + con->in_msg_pos.page_pos, ret); | |
1609 | kunmap(pages[con->in_msg_pos.page]); | |
1610 | if (ret <= 0) | |
1611 | return ret; | |
1612 | con->in_msg_pos.data_pos += ret; | |
1613 | con->in_msg_pos.page_pos += ret; | |
1614 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1615 | con->in_msg_pos.page_pos = 0; | |
1616 | con->in_msg_pos.page++; | |
1617 | } | |
1618 | ||
1619 | return ret; | |
1620 | } | |
1621 | ||
1622 | #ifdef CONFIG_BLOCK | |
1623 | static int read_partial_message_bio(struct ceph_connection *con, | |
1624 | struct bio **bio_iter, int *bio_seg, | |
bca064d2 | 1625 | unsigned data_len, bool do_datacrc) |
68b4476b YS |
1626 | { |
1627 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1628 | void *p; | |
1629 | int ret, left; | |
1630 | ||
1631 | if (IS_ERR(bv)) | |
1632 | return PTR_ERR(bv); | |
1633 | ||
1634 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1635 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1636 | ||
1637 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1638 | ||
1639 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1640 | left); | |
bca064d2 | 1641 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1642 | con->in_data_crc = |
1643 | crc32c(con->in_data_crc, | |
1644 | p + con->in_msg_pos.page_pos, ret); | |
1645 | kunmap(bv->bv_page); | |
1646 | if (ret <= 0) | |
1647 | return ret; | |
1648 | con->in_msg_pos.data_pos += ret; | |
1649 | con->in_msg_pos.page_pos += ret; | |
1650 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1651 | con->in_msg_pos.page_pos = 0; | |
1652 | iter_bio_next(bio_iter, bio_seg); | |
1653 | } | |
1654 | ||
1655 | return ret; | |
1656 | } | |
1657 | #endif | |
1658 | ||
31b8006e SW |
1659 | /* |
1660 | * read (part of) a message. | |
1661 | */ | |
1662 | static int read_partial_message(struct ceph_connection *con) | |
1663 | { | |
1664 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1665 | int size; |
1666 | int end; | |
31b8006e | 1667 | int ret; |
c5c6b19d | 1668 | unsigned front_len, middle_len, data_len; |
37675b0f | 1669 | bool do_datacrc = !con->msgr->nocrc; |
2450418c | 1670 | int skip; |
ae18756b | 1671 | u64 seq; |
fe3ad593 | 1672 | u32 crc; |
31b8006e SW |
1673 | |
1674 | dout("read_partial_message con %p msg %p\n", con, m); | |
1675 | ||
1676 | /* header */ | |
fd51653f AE |
1677 | size = sizeof (con->in_hdr); |
1678 | end = size; | |
1679 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1680 | if (ret <= 0) |
1681 | return ret; | |
fe3ad593 AE |
1682 | |
1683 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1684 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1685 | pr_err("read_partial_message bad hdr " | |
1686 | " crc %u != expected %u\n", | |
1687 | crc, con->in_hdr.crc); | |
1688 | return -EBADMSG; | |
1689 | } | |
1690 | ||
31b8006e SW |
1691 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1692 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1693 | return -EIO; | |
1694 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1695 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1696 | return -EIO; | |
1697 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1698 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1699 | return -EIO; | |
1700 | ||
ae18756b SW |
1701 | /* verify seq# */ |
1702 | seq = le64_to_cpu(con->in_hdr.seq); | |
1703 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1704 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1705 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1706 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1707 | seq, con->in_seq + 1); |
1708 | con->in_base_pos = -front_len - middle_len - data_len - | |
1709 | sizeof(m->footer); | |
1710 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1711 | return 0; |
1712 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1713 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1714 | seq, con->in_seq + 1); | |
1715 | con->error_msg = "bad message sequence # for incoming message"; | |
1716 | return -EBADMSG; | |
1717 | } | |
1718 | ||
31b8006e SW |
1719 | /* allocate message? */ |
1720 | if (!con->in_msg) { | |
1721 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1722 | con->in_hdr.front_len, con->in_hdr.data_len); | |
ae32be31 | 1723 | skip = 0; |
2450418c YS |
1724 | con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip); |
1725 | if (skip) { | |
31b8006e | 1726 | /* skip this message */ |
a79832f2 | 1727 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1728 | BUG_ON(con->in_msg); |
31b8006e SW |
1729 | con->in_base_pos = -front_len - middle_len - data_len - |
1730 | sizeof(m->footer); | |
1731 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1732 | con->in_seq++; |
31b8006e SW |
1733 | return 0; |
1734 | } | |
a79832f2 | 1735 | if (!con->in_msg) { |
5b3a4db3 SW |
1736 | con->error_msg = |
1737 | "error allocating memory for incoming message"; | |
a79832f2 | 1738 | return -ENOMEM; |
31b8006e SW |
1739 | } |
1740 | m = con->in_msg; | |
1741 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1742 | if (m->middle) |
1743 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1744 | |
1745 | con->in_msg_pos.page = 0; | |
68b4476b | 1746 | if (m->pages) |
c5c6b19d | 1747 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1748 | else |
1749 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1750 | con->in_msg_pos.data_pos = 0; |
31b8006e SW |
1751 | } |
1752 | ||
1753 | /* front */ | |
2450418c YS |
1754 | ret = read_partial_message_section(con, &m->front, front_len, |
1755 | &con->in_front_crc); | |
1756 | if (ret <= 0) | |
1757 | return ret; | |
31b8006e SW |
1758 | |
1759 | /* middle */ | |
2450418c | 1760 | if (m->middle) { |
213c99ee SW |
1761 | ret = read_partial_message_section(con, &m->middle->vec, |
1762 | middle_len, | |
2450418c | 1763 | &con->in_middle_crc); |
31b8006e SW |
1764 | if (ret <= 0) |
1765 | return ret; | |
31b8006e | 1766 | } |
68b4476b YS |
1767 | #ifdef CONFIG_BLOCK |
1768 | if (m->bio && !m->bio_iter) | |
1769 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1770 | #endif | |
31b8006e SW |
1771 | |
1772 | /* (page) data */ | |
31b8006e | 1773 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1774 | if (m->pages) { |
1775 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1776 | data_len, do_datacrc); |
68b4476b YS |
1777 | if (ret <= 0) |
1778 | return ret; | |
1779 | #ifdef CONFIG_BLOCK | |
1780 | } else if (m->bio) { | |
1781 | ||
1782 | ret = read_partial_message_bio(con, | |
1783 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1784 | data_len, do_datacrc); |
68b4476b YS |
1785 | if (ret <= 0) |
1786 | return ret; | |
1787 | #endif | |
1788 | } else { | |
1789 | BUG_ON(1); | |
31b8006e SW |
1790 | } |
1791 | } | |
1792 | ||
31b8006e | 1793 | /* footer */ |
fd51653f AE |
1794 | size = sizeof (m->footer); |
1795 | end += size; | |
1796 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
1797 | if (ret <= 0) |
1798 | return ret; | |
1799 | ||
31b8006e SW |
1800 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
1801 | m, front_len, m->footer.front_crc, middle_len, | |
1802 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1803 | ||
1804 | /* crc ok? */ | |
1805 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1806 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1807 | m, con->in_front_crc, m->footer.front_crc); | |
1808 | return -EBADMSG; | |
1809 | } | |
1810 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1811 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1812 | m, con->in_middle_crc, m->footer.middle_crc); | |
1813 | return -EBADMSG; | |
1814 | } | |
bca064d2 | 1815 | if (do_datacrc && |
31b8006e SW |
1816 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
1817 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1818 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1819 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1820 | return -EBADMSG; | |
1821 | } | |
1822 | ||
1823 | return 1; /* done! */ | |
1824 | } | |
1825 | ||
1826 | /* | |
1827 | * Process message. This happens in the worker thread. The callback should | |
1828 | * be careful not to do anything that waits on other incoming messages or it | |
1829 | * may deadlock. | |
1830 | */ | |
1831 | static void process_message(struct ceph_connection *con) | |
1832 | { | |
5e095e8b | 1833 | struct ceph_msg *msg; |
31b8006e | 1834 | |
5e095e8b | 1835 | msg = con->in_msg; |
31b8006e SW |
1836 | con->in_msg = NULL; |
1837 | ||
1838 | /* if first message, set peer_name */ | |
1839 | if (con->peer_name.type == 0) | |
dbad185d | 1840 | con->peer_name = msg->hdr.src; |
31b8006e | 1841 | |
31b8006e | 1842 | con->in_seq++; |
ec302645 | 1843 | mutex_unlock(&con->mutex); |
31b8006e SW |
1844 | |
1845 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1846 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1847 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1848 | le16_to_cpu(msg->hdr.type), |
1849 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1850 | le32_to_cpu(msg->hdr.front_len), | |
1851 | le32_to_cpu(msg->hdr.data_len), | |
1852 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1853 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1854 | |
1855 | mutex_lock(&con->mutex); | |
31b8006e SW |
1856 | prepare_read_tag(con); |
1857 | } | |
1858 | ||
1859 | ||
1860 | /* | |
1861 | * Write something to the socket. Called in a worker thread when the | |
1862 | * socket appears to be writeable and we have something ready to send. | |
1863 | */ | |
1864 | static int try_write(struct ceph_connection *con) | |
1865 | { | |
31b8006e SW |
1866 | int ret = 1; |
1867 | ||
1868 | dout("try_write start %p state %lu nref %d\n", con, con->state, | |
1869 | atomic_read(&con->nref)); | |
1870 | ||
31b8006e SW |
1871 | more: |
1872 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
1873 | ||
1874 | /* open the socket first? */ | |
1875 | if (con->sock == NULL) { | |
84fb3adf | 1876 | ceph_con_out_kvec_reset(con); |
e825a66d | 1877 | prepare_write_banner(con); |
5a0f8fdd AE |
1878 | ret = prepare_write_connect(con); |
1879 | if (ret < 0) | |
1880 | goto out; | |
eed0ef2c | 1881 | prepare_read_banner(con); |
31b8006e | 1882 | set_bit(CONNECTING, &con->state); |
eed0ef2c | 1883 | clear_bit(NEGOTIATING, &con->state); |
31b8006e | 1884 | |
cf3e5c40 | 1885 | BUG_ON(con->in_msg); |
31b8006e SW |
1886 | con->in_tag = CEPH_MSGR_TAG_READY; |
1887 | dout("try_write initiating connect on %p new state %lu\n", | |
1888 | con, con->state); | |
41617d0c AE |
1889 | ret = ceph_tcp_connect(con); |
1890 | if (ret < 0) { | |
31b8006e | 1891 | con->error_msg = "connect error"; |
31b8006e SW |
1892 | goto out; |
1893 | } | |
1894 | } | |
1895 | ||
1896 | more_kvec: | |
1897 | /* kvec data queued? */ | |
1898 | if (con->out_skip) { | |
1899 | ret = write_partial_skip(con); | |
1900 | if (ret <= 0) | |
42961d23 | 1901 | goto out; |
31b8006e SW |
1902 | } |
1903 | if (con->out_kvec_left) { | |
1904 | ret = write_partial_kvec(con); | |
1905 | if (ret <= 0) | |
42961d23 | 1906 | goto out; |
31b8006e SW |
1907 | } |
1908 | ||
1909 | /* msg pages? */ | |
1910 | if (con->out_msg) { | |
c86a2930 SW |
1911 | if (con->out_msg_done) { |
1912 | ceph_msg_put(con->out_msg); | |
1913 | con->out_msg = NULL; /* we're done with this one */ | |
1914 | goto do_next; | |
1915 | } | |
1916 | ||
31b8006e SW |
1917 | ret = write_partial_msg_pages(con); |
1918 | if (ret == 1) | |
1919 | goto more_kvec; /* we need to send the footer, too! */ | |
1920 | if (ret == 0) | |
42961d23 | 1921 | goto out; |
31b8006e SW |
1922 | if (ret < 0) { |
1923 | dout("try_write write_partial_msg_pages err %d\n", | |
1924 | ret); | |
42961d23 | 1925 | goto out; |
31b8006e SW |
1926 | } |
1927 | } | |
1928 | ||
c86a2930 | 1929 | do_next: |
31b8006e SW |
1930 | if (!test_bit(CONNECTING, &con->state)) { |
1931 | /* is anything else pending? */ | |
1932 | if (!list_empty(&con->out_queue)) { | |
1933 | prepare_write_message(con); | |
1934 | goto more; | |
1935 | } | |
1936 | if (con->in_seq > con->in_seq_acked) { | |
1937 | prepare_write_ack(con); | |
1938 | goto more; | |
1939 | } | |
1940 | if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) { | |
1941 | prepare_write_keepalive(con); | |
1942 | goto more; | |
1943 | } | |
1944 | } | |
1945 | ||
1946 | /* Nothing to do! */ | |
1947 | clear_bit(WRITE_PENDING, &con->state); | |
1948 | dout("try_write nothing else to write.\n"); | |
31b8006e SW |
1949 | ret = 0; |
1950 | out: | |
42961d23 | 1951 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
1952 | return ret; |
1953 | } | |
1954 | ||
1955 | ||
1956 | ||
1957 | /* | |
1958 | * Read what we can from the socket. | |
1959 | */ | |
1960 | static int try_read(struct ceph_connection *con) | |
1961 | { | |
31b8006e SW |
1962 | int ret = -1; |
1963 | ||
1964 | if (!con->sock) | |
1965 | return 0; | |
1966 | ||
1967 | if (test_bit(STANDBY, &con->state)) | |
1968 | return 0; | |
1969 | ||
1970 | dout("try_read start on %p\n", con); | |
ec302645 | 1971 | |
31b8006e SW |
1972 | more: |
1973 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, | |
1974 | con->in_base_pos); | |
0da5d703 SW |
1975 | |
1976 | /* | |
1977 | * process_connect and process_message drop and re-take | |
1978 | * con->mutex. make sure we handle a racing close or reopen. | |
1979 | */ | |
1980 | if (test_bit(CLOSED, &con->state) || | |
1981 | test_bit(OPENING, &con->state)) { | |
1982 | ret = -EAGAIN; | |
1983 | goto out; | |
1984 | } | |
1985 | ||
31b8006e | 1986 | if (test_bit(CONNECTING, &con->state)) { |
eed0ef2c SW |
1987 | if (!test_bit(NEGOTIATING, &con->state)) { |
1988 | dout("try_read connecting\n"); | |
1989 | ret = read_partial_banner(con); | |
1990 | if (ret <= 0) | |
eed0ef2c | 1991 | goto out; |
98bdb0aa SW |
1992 | ret = process_banner(con); |
1993 | if (ret < 0) | |
1994 | goto out; | |
eed0ef2c | 1995 | } |
31b8006e SW |
1996 | ret = read_partial_connect(con); |
1997 | if (ret <= 0) | |
31b8006e | 1998 | goto out; |
98bdb0aa SW |
1999 | ret = process_connect(con); |
2000 | if (ret < 0) | |
2001 | goto out; | |
31b8006e SW |
2002 | goto more; |
2003 | } | |
2004 | ||
2005 | if (con->in_base_pos < 0) { | |
2006 | /* | |
2007 | * skipping + discarding content. | |
2008 | * | |
2009 | * FIXME: there must be a better way to do this! | |
2010 | */ | |
84495f49 AE |
2011 | static char buf[SKIP_BUF_SIZE]; |
2012 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2013 | ||
31b8006e SW |
2014 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2015 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2016 | if (ret <= 0) | |
98bdb0aa | 2017 | goto out; |
31b8006e SW |
2018 | con->in_base_pos += ret; |
2019 | if (con->in_base_pos) | |
2020 | goto more; | |
2021 | } | |
2022 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2023 | /* | |
2024 | * what's next? | |
2025 | */ | |
2026 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2027 | if (ret <= 0) | |
98bdb0aa | 2028 | goto out; |
31b8006e SW |
2029 | dout("try_read got tag %d\n", (int)con->in_tag); |
2030 | switch (con->in_tag) { | |
2031 | case CEPH_MSGR_TAG_MSG: | |
2032 | prepare_read_message(con); | |
2033 | break; | |
2034 | case CEPH_MSGR_TAG_ACK: | |
2035 | prepare_read_ack(con); | |
2036 | break; | |
2037 | case CEPH_MSGR_TAG_CLOSE: | |
2038 | set_bit(CLOSED, &con->state); /* fixme */ | |
98bdb0aa | 2039 | goto out; |
31b8006e SW |
2040 | default: |
2041 | goto bad_tag; | |
2042 | } | |
2043 | } | |
2044 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2045 | ret = read_partial_message(con); | |
2046 | if (ret <= 0) { | |
2047 | switch (ret) { | |
2048 | case -EBADMSG: | |
2049 | con->error_msg = "bad crc"; | |
2050 | ret = -EIO; | |
98bdb0aa | 2051 | break; |
31b8006e SW |
2052 | case -EIO: |
2053 | con->error_msg = "io error"; | |
98bdb0aa | 2054 | break; |
31b8006e | 2055 | } |
98bdb0aa | 2056 | goto out; |
31b8006e SW |
2057 | } |
2058 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2059 | goto more; | |
2060 | process_message(con); | |
2061 | goto more; | |
2062 | } | |
2063 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2064 | ret = read_partial_ack(con); | |
2065 | if (ret <= 0) | |
98bdb0aa | 2066 | goto out; |
31b8006e SW |
2067 | process_ack(con); |
2068 | goto more; | |
2069 | } | |
2070 | ||
31b8006e | 2071 | out: |
98bdb0aa | 2072 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2073 | return ret; |
2074 | ||
2075 | bad_tag: | |
2076 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2077 | con->error_msg = "protocol error, garbage tag"; | |
2078 | ret = -1; | |
2079 | goto out; | |
2080 | } | |
2081 | ||
2082 | ||
2083 | /* | |
2084 | * Atomically queue work on a connection. Bump @con reference to | |
2085 | * avoid races with connection teardown. | |
31b8006e SW |
2086 | */ |
2087 | static void queue_con(struct ceph_connection *con) | |
2088 | { | |
2089 | if (test_bit(DEAD, &con->state)) { | |
2090 | dout("queue_con %p ignoring: DEAD\n", | |
2091 | con); | |
2092 | return; | |
2093 | } | |
2094 | ||
2095 | if (!con->ops->get(con)) { | |
2096 | dout("queue_con %p ref count 0\n", con); | |
2097 | return; | |
2098 | } | |
2099 | ||
f363e45f | 2100 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2101 | dout("queue_con %p - already queued\n", con); |
2102 | con->ops->put(con); | |
2103 | } else { | |
2104 | dout("queue_con %p\n", con); | |
2105 | } | |
2106 | } | |
2107 | ||
2108 | /* | |
2109 | * Do some work on a connection. Drop a connection ref when we're done. | |
2110 | */ | |
2111 | static void con_work(struct work_struct *work) | |
2112 | { | |
2113 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2114 | work.work); | |
0da5d703 | 2115 | int ret; |
31b8006e | 2116 | |
9dd4658d | 2117 | mutex_lock(&con->mutex); |
0da5d703 | 2118 | restart: |
60bf8bf8 SW |
2119 | if (test_and_clear_bit(BACKOFF, &con->state)) { |
2120 | dout("con_work %p backing off\n", con); | |
2121 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2122 | round_jiffies_relative(con->delay))) { | |
2123 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2124 | mutex_unlock(&con->mutex); | |
2125 | return; | |
2126 | } else { | |
2127 | con->ops->put(con); | |
2128 | dout("con_work %p FAILED to back off %lu\n", con, | |
2129 | con->delay); | |
2130 | } | |
2131 | } | |
9dd4658d | 2132 | |
e00de341 SW |
2133 | if (test_bit(STANDBY, &con->state)) { |
2134 | dout("con_work %p STANDBY\n", con); | |
2135 | goto done; | |
2136 | } | |
31b8006e SW |
2137 | if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */ |
2138 | dout("con_work CLOSED\n"); | |
2139 | con_close_socket(con); | |
2140 | goto done; | |
2141 | } | |
2142 | if (test_and_clear_bit(OPENING, &con->state)) { | |
2143 | /* reopen w/ new peer */ | |
2144 | dout("con_work OPENING\n"); | |
2145 | con_close_socket(con); | |
2146 | } | |
2147 | ||
0da5d703 SW |
2148 | if (test_and_clear_bit(SOCK_CLOSED, &con->state)) |
2149 | goto fault; | |
2150 | ||
2151 | ret = try_read(con); | |
2152 | if (ret == -EAGAIN) | |
2153 | goto restart; | |
2154 | if (ret < 0) | |
2155 | goto fault; | |
2156 | ||
2157 | ret = try_write(con); | |
2158 | if (ret == -EAGAIN) | |
2159 | goto restart; | |
2160 | if (ret < 0) | |
2161 | goto fault; | |
31b8006e SW |
2162 | |
2163 | done: | |
9dd4658d | 2164 | mutex_unlock(&con->mutex); |
9dd4658d | 2165 | done_unlocked: |
31b8006e | 2166 | con->ops->put(con); |
0da5d703 SW |
2167 | return; |
2168 | ||
2169 | fault: | |
2170 | mutex_unlock(&con->mutex); | |
2171 | ceph_fault(con); /* error/fault path */ | |
2172 | goto done_unlocked; | |
31b8006e SW |
2173 | } |
2174 | ||
2175 | ||
2176 | /* | |
2177 | * Generic error/fault handler. A retry mechanism is used with | |
2178 | * exponential backoff | |
2179 | */ | |
2180 | static void ceph_fault(struct ceph_connection *con) | |
2181 | { | |
2182 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2183 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2184 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2185 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
2186 | |
2187 | if (test_bit(LOSSYTX, &con->state)) { | |
2188 | dout("fault on LOSSYTX channel\n"); | |
2189 | goto out; | |
2190 | } | |
2191 | ||
ec302645 | 2192 | mutex_lock(&con->mutex); |
91e45ce3 SW |
2193 | if (test_bit(CLOSED, &con->state)) |
2194 | goto out_unlock; | |
ec302645 | 2195 | |
31b8006e | 2196 | con_close_socket(con); |
5e095e8b SW |
2197 | |
2198 | if (con->in_msg) { | |
2199 | ceph_msg_put(con->in_msg); | |
2200 | con->in_msg = NULL; | |
2201 | } | |
31b8006e | 2202 | |
e80a52d1 SW |
2203 | /* Requeue anything that hasn't been acked */ |
2204 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2205 | |
e76661d0 SW |
2206 | /* If there are no messages queued or keepalive pending, place |
2207 | * the connection in a STANDBY state */ | |
2208 | if (list_empty(&con->out_queue) && | |
2209 | !test_bit(KEEPALIVE_PENDING, &con->state)) { | |
e00de341 SW |
2210 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
2211 | clear_bit(WRITE_PENDING, &con->state); | |
31b8006e | 2212 | set_bit(STANDBY, &con->state); |
e80a52d1 SW |
2213 | } else { |
2214 | /* retry after a delay. */ | |
2215 | if (con->delay == 0) | |
2216 | con->delay = BASE_DELAY_INTERVAL; | |
2217 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2218 | con->delay *= 2; | |
e80a52d1 SW |
2219 | con->ops->get(con); |
2220 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2221 | round_jiffies_relative(con->delay))) { |
2222 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2223 | } else { | |
e80a52d1 | 2224 | con->ops->put(con); |
60bf8bf8 SW |
2225 | dout("fault failed to queue %p delay %lu, backoff\n", |
2226 | con, con->delay); | |
2227 | /* | |
2228 | * In many cases we see a socket state change | |
2229 | * while con_work is running and end up | |
2230 | * queuing (non-delayed) work, such that we | |
2231 | * can't backoff with a delay. Set a flag so | |
2232 | * that when con_work restarts we schedule the | |
2233 | * delay then. | |
2234 | */ | |
2235 | set_bit(BACKOFF, &con->state); | |
2236 | } | |
31b8006e SW |
2237 | } |
2238 | ||
91e45ce3 SW |
2239 | out_unlock: |
2240 | mutex_unlock(&con->mutex); | |
31b8006e | 2241 | out: |
161fd65a SW |
2242 | /* |
2243 | * in case we faulted due to authentication, invalidate our | |
2244 | * current tickets so that we can get new ones. | |
213c99ee | 2245 | */ |
161fd65a SW |
2246 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2247 | dout("calling invalidate_authorizer()\n"); | |
2248 | con->ops->invalidate_authorizer(con); | |
2249 | } | |
2250 | ||
31b8006e SW |
2251 | if (con->ops->fault) |
2252 | con->ops->fault(con); | |
2253 | } | |
2254 | ||
2255 | ||
2256 | ||
2257 | /* | |
2258 | * create a new messenger instance | |
2259 | */ | |
3d14c5d2 YS |
2260 | struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr, |
2261 | u32 supported_features, | |
2262 | u32 required_features) | |
31b8006e SW |
2263 | { |
2264 | struct ceph_messenger *msgr; | |
2265 | ||
2266 | msgr = kzalloc(sizeof(*msgr), GFP_KERNEL); | |
2267 | if (msgr == NULL) | |
2268 | return ERR_PTR(-ENOMEM); | |
2269 | ||
3d14c5d2 YS |
2270 | msgr->supported_features = supported_features; |
2271 | msgr->required_features = required_features; | |
2272 | ||
31b8006e SW |
2273 | spin_lock_init(&msgr->global_seq_lock); |
2274 | ||
31b8006e SW |
2275 | if (myaddr) |
2276 | msgr->inst.addr = *myaddr; | |
2277 | ||
2278 | /* select a random nonce */ | |
ac8839d7 | 2279 | msgr->inst.addr.type = 0; |
103e2d3a | 2280 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2281 | encode_my_addr(msgr); |
31b8006e SW |
2282 | |
2283 | dout("messenger_create %p\n", msgr); | |
2284 | return msgr; | |
2285 | } | |
3d14c5d2 | 2286 | EXPORT_SYMBOL(ceph_messenger_create); |
31b8006e SW |
2287 | |
2288 | void ceph_messenger_destroy(struct ceph_messenger *msgr) | |
2289 | { | |
2290 | dout("destroy %p\n", msgr); | |
31b8006e SW |
2291 | kfree(msgr); |
2292 | dout("destroyed messenger %p\n", msgr); | |
2293 | } | |
3d14c5d2 | 2294 | EXPORT_SYMBOL(ceph_messenger_destroy); |
31b8006e | 2295 | |
e00de341 SW |
2296 | static void clear_standby(struct ceph_connection *con) |
2297 | { | |
2298 | /* come back from STANDBY? */ | |
2299 | if (test_and_clear_bit(STANDBY, &con->state)) { | |
2300 | mutex_lock(&con->mutex); | |
2301 | dout("clear_standby %p and ++connect_seq\n", con); | |
2302 | con->connect_seq++; | |
2303 | WARN_ON(test_bit(WRITE_PENDING, &con->state)); | |
2304 | WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state)); | |
2305 | mutex_unlock(&con->mutex); | |
2306 | } | |
2307 | } | |
2308 | ||
31b8006e SW |
2309 | /* |
2310 | * Queue up an outgoing message on the given connection. | |
2311 | */ | |
2312 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2313 | { | |
2314 | if (test_bit(CLOSED, &con->state)) { | |
2315 | dout("con_send %p closed, dropping %p\n", con, msg); | |
2316 | ceph_msg_put(msg); | |
2317 | return; | |
2318 | } | |
2319 | ||
2320 | /* set src+dst */ | |
dbad185d | 2321 | msg->hdr.src = con->msgr->inst.name; |
31b8006e | 2322 | |
3ca02ef9 SW |
2323 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
2324 | ||
e84346b7 SW |
2325 | msg->needs_out_seq = true; |
2326 | ||
31b8006e | 2327 | /* queue */ |
ec302645 | 2328 | mutex_lock(&con->mutex); |
31b8006e SW |
2329 | BUG_ON(!list_empty(&msg->list_head)); |
2330 | list_add_tail(&msg->list_head, &con->out_queue); | |
2331 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2332 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2333 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2334 | le32_to_cpu(msg->hdr.front_len), | |
2335 | le32_to_cpu(msg->hdr.middle_len), | |
2336 | le32_to_cpu(msg->hdr.data_len)); | |
ec302645 | 2337 | mutex_unlock(&con->mutex); |
31b8006e SW |
2338 | |
2339 | /* if there wasn't anything waiting to send before, queue | |
2340 | * new work */ | |
e00de341 | 2341 | clear_standby(con); |
31b8006e SW |
2342 | if (test_and_set_bit(WRITE_PENDING, &con->state) == 0) |
2343 | queue_con(con); | |
2344 | } | |
3d14c5d2 | 2345 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2346 | |
2347 | /* | |
2348 | * Revoke a message that was previously queued for send | |
2349 | */ | |
2350 | void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg) | |
2351 | { | |
ec302645 | 2352 | mutex_lock(&con->mutex); |
31b8006e | 2353 | if (!list_empty(&msg->list_head)) { |
ed98adad | 2354 | dout("con_revoke %p msg %p - was on queue\n", con, msg); |
31b8006e SW |
2355 | list_del_init(&msg->list_head); |
2356 | ceph_msg_put(msg); | |
2357 | msg->hdr.seq = 0; | |
ed98adad SW |
2358 | } |
2359 | if (con->out_msg == msg) { | |
2360 | dout("con_revoke %p msg %p - was sending\n", con, msg); | |
2361 | con->out_msg = NULL; | |
31b8006e SW |
2362 | if (con->out_kvec_is_msg) { |
2363 | con->out_skip = con->out_kvec_bytes; | |
2364 | con->out_kvec_is_msg = false; | |
2365 | } | |
ed98adad SW |
2366 | ceph_msg_put(msg); |
2367 | msg->hdr.seq = 0; | |
31b8006e | 2368 | } |
ec302645 | 2369 | mutex_unlock(&con->mutex); |
31b8006e SW |
2370 | } |
2371 | ||
350b1c32 | 2372 | /* |
0d59ab81 | 2373 | * Revoke a message that we may be reading data into |
350b1c32 | 2374 | */ |
0d59ab81 | 2375 | void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg) |
350b1c32 SW |
2376 | { |
2377 | mutex_lock(&con->mutex); | |
0d59ab81 YS |
2378 | if (con->in_msg && con->in_msg == msg) { |
2379 | unsigned front_len = le32_to_cpu(con->in_hdr.front_len); | |
2380 | unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
350b1c32 SW |
2381 | unsigned data_len = le32_to_cpu(con->in_hdr.data_len); |
2382 | ||
2383 | /* skip rest of message */ | |
0d59ab81 | 2384 | dout("con_revoke_pages %p msg %p revoked\n", con, msg); |
350b1c32 SW |
2385 | con->in_base_pos = con->in_base_pos - |
2386 | sizeof(struct ceph_msg_header) - | |
0d59ab81 YS |
2387 | front_len - |
2388 | middle_len - | |
2389 | data_len - | |
350b1c32 | 2390 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2391 | ceph_msg_put(con->in_msg); |
2392 | con->in_msg = NULL; | |
2393 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2394 | con->in_seq++; |
350b1c32 SW |
2395 | } else { |
2396 | dout("con_revoke_pages %p msg %p pages %p no-op\n", | |
0d59ab81 | 2397 | con, con->in_msg, msg); |
350b1c32 SW |
2398 | } |
2399 | mutex_unlock(&con->mutex); | |
2400 | } | |
2401 | ||
31b8006e SW |
2402 | /* |
2403 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2404 | */ | |
2405 | void ceph_con_keepalive(struct ceph_connection *con) | |
2406 | { | |
e00de341 SW |
2407 | dout("con_keepalive %p\n", con); |
2408 | clear_standby(con); | |
31b8006e SW |
2409 | if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 && |
2410 | test_and_set_bit(WRITE_PENDING, &con->state) == 0) | |
2411 | queue_con(con); | |
2412 | } | |
3d14c5d2 | 2413 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2414 | |
2415 | ||
2416 | /* | |
2417 | * construct a new message with given type, size | |
2418 | * the new msg has a ref count of 1. | |
2419 | */ | |
b61c2763 SW |
2420 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2421 | bool can_fail) | |
31b8006e SW |
2422 | { |
2423 | struct ceph_msg *m; | |
2424 | ||
34d23762 | 2425 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2426 | if (m == NULL) |
2427 | goto out; | |
c2e552e7 | 2428 | kref_init(&m->kref); |
31b8006e SW |
2429 | INIT_LIST_HEAD(&m->list_head); |
2430 | ||
45c6ceb5 | 2431 | m->hdr.tid = 0; |
31b8006e | 2432 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2433 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2434 | m->hdr.version = 0; | |
31b8006e SW |
2435 | m->hdr.front_len = cpu_to_le32(front_len); |
2436 | m->hdr.middle_len = 0; | |
bb257664 SW |
2437 | m->hdr.data_len = 0; |
2438 | m->hdr.data_off = 0; | |
45c6ceb5 | 2439 | m->hdr.reserved = 0; |
31b8006e SW |
2440 | m->footer.front_crc = 0; |
2441 | m->footer.middle_crc = 0; | |
2442 | m->footer.data_crc = 0; | |
45c6ceb5 | 2443 | m->footer.flags = 0; |
31b8006e SW |
2444 | m->front_max = front_len; |
2445 | m->front_is_vmalloc = false; | |
2446 | m->more_to_follow = false; | |
c0d5f9db | 2447 | m->ack_stamp = 0; |
31b8006e SW |
2448 | m->pool = NULL; |
2449 | ||
ca20892d HC |
2450 | /* middle */ |
2451 | m->middle = NULL; | |
2452 | ||
2453 | /* data */ | |
2454 | m->nr_pages = 0; | |
2455 | m->page_alignment = 0; | |
2456 | m->pages = NULL; | |
2457 | m->pagelist = NULL; | |
2458 | m->bio = NULL; | |
2459 | m->bio_iter = NULL; | |
2460 | m->bio_seg = 0; | |
2461 | m->trail = NULL; | |
2462 | ||
31b8006e SW |
2463 | /* front */ |
2464 | if (front_len) { | |
2465 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2466 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2467 | PAGE_KERNEL); |
2468 | m->front_is_vmalloc = true; | |
2469 | } else { | |
34d23762 | 2470 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2471 | } |
2472 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2473 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2474 | front_len); |
2475 | goto out2; | |
2476 | } | |
2477 | } else { | |
2478 | m->front.iov_base = NULL; | |
2479 | } | |
2480 | m->front.iov_len = front_len; | |
2481 | ||
bb257664 | 2482 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2483 | return m; |
2484 | ||
2485 | out2: | |
2486 | ceph_msg_put(m); | |
2487 | out: | |
b61c2763 SW |
2488 | if (!can_fail) { |
2489 | pr_err("msg_new can't create type %d front %d\n", type, | |
2490 | front_len); | |
f0ed1b7c | 2491 | WARN_ON(1); |
b61c2763 SW |
2492 | } else { |
2493 | dout("msg_new can't create type %d front %d\n", type, | |
2494 | front_len); | |
2495 | } | |
a79832f2 | 2496 | return NULL; |
31b8006e | 2497 | } |
3d14c5d2 | 2498 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2499 | |
31b8006e SW |
2500 | /* |
2501 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2502 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2503 | * per-type header in the front and then gracefully fail (i.e., | |
2504 | * propagate the error to the caller based on info in the front) when | |
2505 | * the middle is too large. | |
2506 | */ | |
2450418c | 2507 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2508 | { |
2509 | int type = le16_to_cpu(msg->hdr.type); | |
2510 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2511 | ||
2512 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2513 | ceph_msg_type_name(type), middle_len); | |
2514 | BUG_ON(!middle_len); | |
2515 | BUG_ON(msg->middle); | |
2516 | ||
b6c1d5b8 | 2517 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2518 | if (!msg->middle) |
2519 | return -ENOMEM; | |
2520 | return 0; | |
2521 | } | |
2522 | ||
2450418c YS |
2523 | /* |
2524 | * Generic message allocator, for incoming messages. | |
2525 | */ | |
2526 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, | |
2527 | struct ceph_msg_header *hdr, | |
2528 | int *skip) | |
2529 | { | |
2530 | int type = le16_to_cpu(hdr->type); | |
2531 | int front_len = le32_to_cpu(hdr->front_len); | |
2532 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2533 | struct ceph_msg *msg = NULL; | |
2534 | int ret; | |
2535 | ||
2536 | if (con->ops->alloc_msg) { | |
0547a9b3 | 2537 | mutex_unlock(&con->mutex); |
2450418c | 2538 | msg = con->ops->alloc_msg(con, hdr, skip); |
0547a9b3 | 2539 | mutex_lock(&con->mutex); |
a79832f2 | 2540 | if (!msg || *skip) |
2450418c YS |
2541 | return NULL; |
2542 | } | |
2543 | if (!msg) { | |
2544 | *skip = 0; | |
b61c2763 | 2545 | msg = ceph_msg_new(type, front_len, GFP_NOFS, false); |
2450418c YS |
2546 | if (!msg) { |
2547 | pr_err("unable to allocate msg type %d len %d\n", | |
2548 | type, front_len); | |
a79832f2 | 2549 | return NULL; |
2450418c | 2550 | } |
c5c6b19d | 2551 | msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2552 | } |
9d7f0f13 | 2553 | memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2554 | |
bb257664 | 2555 | if (middle_len && !msg->middle) { |
2450418c | 2556 | ret = ceph_alloc_middle(con, msg); |
2450418c YS |
2557 | if (ret < 0) { |
2558 | ceph_msg_put(msg); | |
a79832f2 | 2559 | return NULL; |
2450418c YS |
2560 | } |
2561 | } | |
9d7f0f13 | 2562 | |
2450418c YS |
2563 | return msg; |
2564 | } | |
2565 | ||
31b8006e SW |
2566 | |
2567 | /* | |
2568 | * Free a generically kmalloc'd message. | |
2569 | */ | |
2570 | void ceph_msg_kfree(struct ceph_msg *m) | |
2571 | { | |
2572 | dout("msg_kfree %p\n", m); | |
2573 | if (m->front_is_vmalloc) | |
2574 | vfree(m->front.iov_base); | |
2575 | else | |
2576 | kfree(m->front.iov_base); | |
2577 | kfree(m); | |
2578 | } | |
2579 | ||
2580 | /* | |
2581 | * Drop a msg ref. Destroy as needed. | |
2582 | */ | |
c2e552e7 SW |
2583 | void ceph_msg_last_put(struct kref *kref) |
2584 | { | |
2585 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2586 | |
c2e552e7 SW |
2587 | dout("ceph_msg_put last one on %p\n", m); |
2588 | WARN_ON(!list_empty(&m->list_head)); | |
2589 | ||
2590 | /* drop middle, data, if any */ | |
2591 | if (m->middle) { | |
2592 | ceph_buffer_put(m->middle); | |
2593 | m->middle = NULL; | |
31b8006e | 2594 | } |
c2e552e7 SW |
2595 | m->nr_pages = 0; |
2596 | m->pages = NULL; | |
2597 | ||
58bb3b37 SW |
2598 | if (m->pagelist) { |
2599 | ceph_pagelist_release(m->pagelist); | |
2600 | kfree(m->pagelist); | |
2601 | m->pagelist = NULL; | |
2602 | } | |
2603 | ||
68b4476b YS |
2604 | m->trail = NULL; |
2605 | ||
c2e552e7 SW |
2606 | if (m->pool) |
2607 | ceph_msgpool_put(m->pool, m); | |
2608 | else | |
2609 | ceph_msg_kfree(m); | |
31b8006e | 2610 | } |
3d14c5d2 | 2611 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2612 | |
2613 | void ceph_msg_dump(struct ceph_msg *msg) | |
2614 | { | |
2615 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2616 | msg->front_max, msg->nr_pages); | |
2617 | print_hex_dump(KERN_DEBUG, "header: ", | |
2618 | DUMP_PREFIX_OFFSET, 16, 1, | |
2619 | &msg->hdr, sizeof(msg->hdr), true); | |
2620 | print_hex_dump(KERN_DEBUG, " front: ", | |
2621 | DUMP_PREFIX_OFFSET, 16, 1, | |
2622 | msg->front.iov_base, msg->front.iov_len, true); | |
2623 | if (msg->middle) | |
2624 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2625 | DUMP_PREFIX_OFFSET, 16, 1, | |
2626 | msg->middle->vec.iov_base, | |
2627 | msg->middle->vec.iov_len, true); | |
2628 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2629 | DUMP_PREFIX_OFFSET, 16, 1, | |
2630 | &msg->footer, sizeof(msg->footer), true); | |
2631 | } | |
3d14c5d2 | 2632 | EXPORT_SYMBOL(ceph_msg_dump); |