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 */ | |
327800bd | 156 | static void ceph_sock_data_ready(struct sock *sk, int count_unused) |
31b8006e | 157 | { |
bd406145 AE |
158 | struct ceph_connection *con = sk->sk_user_data; |
159 | ||
31b8006e | 160 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 161 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
162 | con, con->state); |
163 | queue_con(con); | |
164 | } | |
165 | } | |
166 | ||
167 | /* socket has buffer space for writing */ | |
327800bd | 168 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 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 | |
327800bd | 174 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
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 | 180 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
327800bd | 181 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
182 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
183 | queue_con(con); | |
184 | } | |
31b8006e | 185 | } else { |
327800bd | 186 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 187 | } |
31b8006e SW |
188 | } |
189 | ||
190 | /* socket's state has changed */ | |
327800bd | 191 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 192 | { |
bd406145 | 193 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 194 | |
327800bd | 195 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
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: | |
327800bd | 203 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 204 | case TCP_CLOSE_WAIT: |
327800bd | 205 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
31b8006e SW |
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: | |
327800bd | 215 | dout("%s TCP_ESTABLISHED\n", __func__); |
31b8006e SW |
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; |
327800bd AE |
231 | sk->sk_data_ready = ceph_sock_data_ready; |
232 | sk->sk_write_space = ceph_sock_write_space; | |
233 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
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 | ||
e2200423 | 489 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
490 | { |
491 | con->out_kvec_left = 0; | |
492 | con->out_kvec_bytes = 0; | |
493 | con->out_kvec_cur = &con->out_kvec[0]; | |
494 | } | |
495 | ||
e2200423 | 496 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
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 | |
e2200423 | 537 | 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; | |
e2200423 | 545 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 546 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 547 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 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 */ | |
e2200423 AE |
575 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
576 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
577 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 578 | |
31b8006e | 579 | if (m->middle) |
e2200423 | 580 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 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 | ||
e2200423 | 629 | con_out_kvec_reset(con); |
859eb799 | 630 | |
e2200423 | 631 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 632 | |
31b8006e | 633 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 634 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
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); | |
e2200423 AE |
647 | con_out_kvec_reset(con); |
648 | 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 | ||
dac1e716 AE |
656 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
657 | int *auth_proto) | |
4e7a5dcd | 658 | { |
a3530df3 | 659 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
660 | |
661 | if (!con->ops->get_authorizer) { | |
662 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
663 | con->out_connect.authorizer_len = 0; | |
664 | ||
729796be | 665 | return NULL; |
b1c6b980 AE |
666 | } |
667 | ||
668 | /* Can't hold the mutex while getting authorizer */ | |
4e7a5dcd | 669 | |
ec302645 | 670 | mutex_unlock(&con->mutex); |
b1c6b980 | 671 | |
dac1e716 | 672 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
8f43fb53 | 673 | |
ec302645 | 674 | mutex_lock(&con->mutex); |
4e7a5dcd | 675 | |
a3530df3 | 676 | if (IS_ERR(auth)) |
729796be | 677 | return auth; |
b1c6b980 | 678 | if (test_bit(CLOSED, &con->state) || test_bit(OPENING, &con->state)) |
729796be | 679 | return ERR_PTR(-EAGAIN); |
0da5d703 | 680 | |
8f43fb53 AE |
681 | con->auth_reply_buf = auth->authorizer_reply_buf; |
682 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
683 | ||
859eb799 | 684 | |
729796be | 685 | return auth; |
4e7a5dcd SW |
686 | } |
687 | ||
31b8006e SW |
688 | /* |
689 | * We connected to a peer and are saying hello. | |
690 | */ | |
e825a66d | 691 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 692 | { |
e2200423 AE |
693 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
694 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 695 | &con->msgr->my_enc_addr); |
eed0ef2c | 696 | |
eed0ef2c SW |
697 | con->out_more = 0; |
698 | set_bit(WRITE_PENDING, &con->state); | |
699 | } | |
700 | ||
e825a66d | 701 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 702 | { |
31b8006e SW |
703 | unsigned global_seq = get_global_seq(con->msgr, 0); |
704 | int proto; | |
dac1e716 | 705 | int auth_proto; |
729796be | 706 | struct ceph_auth_handshake *auth; |
31b8006e SW |
707 | |
708 | switch (con->peer_name.type) { | |
709 | case CEPH_ENTITY_TYPE_MON: | |
710 | proto = CEPH_MONC_PROTOCOL; | |
711 | break; | |
712 | case CEPH_ENTITY_TYPE_OSD: | |
713 | proto = CEPH_OSDC_PROTOCOL; | |
714 | break; | |
715 | case CEPH_ENTITY_TYPE_MDS: | |
716 | proto = CEPH_MDSC_PROTOCOL; | |
717 | break; | |
718 | default: | |
719 | BUG(); | |
720 | } | |
721 | ||
722 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
723 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 724 | |
e825a66d | 725 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
726 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
727 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
728 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
729 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
730 | con->out_connect.flags = 0; | |
31b8006e | 731 | |
dac1e716 AE |
732 | auth_proto = CEPH_AUTH_UNKNOWN; |
733 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
734 | if (IS_ERR(auth)) |
735 | return PTR_ERR(auth); | |
3da54776 | 736 | |
dac1e716 | 737 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
738 | con->out_connect.authorizer_len = auth ? |
739 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
740 | ||
e2200423 | 741 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
742 | &con->out_connect); |
743 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 744 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 745 | auth->authorizer_buf); |
859eb799 | 746 | |
31b8006e SW |
747 | con->out_more = 0; |
748 | set_bit(WRITE_PENDING, &con->state); | |
4e7a5dcd | 749 | |
e10c758e | 750 | return 0; |
31b8006e SW |
751 | } |
752 | ||
31b8006e SW |
753 | /* |
754 | * write as much of pending kvecs to the socket as we can. | |
755 | * 1 -> done | |
756 | * 0 -> socket full, but more to do | |
757 | * <0 -> error | |
758 | */ | |
759 | static int write_partial_kvec(struct ceph_connection *con) | |
760 | { | |
761 | int ret; | |
762 | ||
763 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
764 | while (con->out_kvec_bytes > 0) { | |
765 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
766 | con->out_kvec_left, con->out_kvec_bytes, | |
767 | con->out_more); | |
768 | if (ret <= 0) | |
769 | goto out; | |
770 | con->out_kvec_bytes -= ret; | |
771 | if (con->out_kvec_bytes == 0) | |
772 | break; /* done */ | |
f42299e6 AE |
773 | |
774 | /* account for full iov entries consumed */ | |
775 | while (ret >= con->out_kvec_cur->iov_len) { | |
776 | BUG_ON(!con->out_kvec_left); | |
777 | ret -= con->out_kvec_cur->iov_len; | |
778 | con->out_kvec_cur++; | |
779 | con->out_kvec_left--; | |
780 | } | |
781 | /* and for a partially-consumed entry */ | |
782 | if (ret) { | |
783 | con->out_kvec_cur->iov_len -= ret; | |
784 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
785 | } |
786 | } | |
787 | con->out_kvec_left = 0; | |
788 | con->out_kvec_is_msg = false; | |
789 | ret = 1; | |
790 | out: | |
791 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
792 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
793 | return ret; /* done! */ | |
794 | } | |
795 | ||
68b4476b YS |
796 | #ifdef CONFIG_BLOCK |
797 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
798 | { | |
799 | if (!bio) { | |
800 | *iter = NULL; | |
801 | *seg = 0; | |
802 | return; | |
803 | } | |
804 | *iter = bio; | |
805 | *seg = bio->bi_idx; | |
806 | } | |
807 | ||
808 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
809 | { | |
810 | if (*bio_iter == NULL) | |
811 | return; | |
812 | ||
813 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
814 | ||
815 | (*seg)++; | |
816 | if (*seg == (*bio_iter)->bi_vcnt) | |
817 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
818 | } | |
819 | #endif | |
820 | ||
31b8006e SW |
821 | /* |
822 | * Write as much message data payload as we can. If we finish, queue | |
823 | * up the footer. | |
824 | * 1 -> done, footer is now queued in out_kvec[]. | |
825 | * 0 -> socket full, but more to do | |
826 | * <0 -> error | |
827 | */ | |
828 | static int write_partial_msg_pages(struct ceph_connection *con) | |
829 | { | |
830 | struct ceph_msg *msg = con->out_msg; | |
831 | unsigned data_len = le32_to_cpu(msg->hdr.data_len); | |
832 | size_t len; | |
37675b0f | 833 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 834 | int ret; |
68b4476b YS |
835 | int total_max_write; |
836 | int in_trail = 0; | |
837 | size_t trail_len = (msg->trail ? msg->trail->length : 0); | |
31b8006e SW |
838 | |
839 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
840 | con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages, | |
841 | con->out_msg_pos.page_pos); | |
842 | ||
68b4476b YS |
843 | #ifdef CONFIG_BLOCK |
844 | if (msg->bio && !msg->bio_iter) | |
845 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); | |
846 | #endif | |
847 | ||
848 | while (data_len > con->out_msg_pos.data_pos) { | |
31b8006e | 849 | struct page *page = NULL; |
68b4476b | 850 | int max_write = PAGE_SIZE; |
9bd19663 | 851 | int bio_offset = 0; |
68b4476b YS |
852 | |
853 | total_max_write = data_len - trail_len - | |
854 | con->out_msg_pos.data_pos; | |
31b8006e SW |
855 | |
856 | /* | |
857 | * if we are calculating the data crc (the default), we need | |
858 | * to map the page. if our pages[] has been revoked, use the | |
859 | * zero page. | |
860 | */ | |
68b4476b YS |
861 | |
862 | /* have we reached the trail part of the data? */ | |
863 | if (con->out_msg_pos.data_pos >= data_len - trail_len) { | |
864 | in_trail = 1; | |
865 | ||
866 | total_max_write = data_len - con->out_msg_pos.data_pos; | |
867 | ||
868 | page = list_first_entry(&msg->trail->head, | |
869 | struct page, lru); | |
68b4476b YS |
870 | max_write = PAGE_SIZE; |
871 | } else if (msg->pages) { | |
31b8006e | 872 | page = msg->pages[con->out_msg_pos.page]; |
58bb3b37 SW |
873 | } else if (msg->pagelist) { |
874 | page = list_first_entry(&msg->pagelist->head, | |
875 | struct page, lru); | |
68b4476b YS |
876 | #ifdef CONFIG_BLOCK |
877 | } else if (msg->bio) { | |
878 | struct bio_vec *bv; | |
879 | ||
880 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
881 | page = bv->bv_page; | |
9bd19663 | 882 | bio_offset = bv->bv_offset; |
68b4476b YS |
883 | max_write = bv->bv_len; |
884 | #endif | |
31b8006e | 885 | } else { |
57666519 | 886 | page = zero_page; |
31b8006e | 887 | } |
68b4476b YS |
888 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
889 | total_max_write); | |
890 | ||
37675b0f | 891 | if (do_datacrc && !con->out_msg_pos.did_page_crc) { |
9bd19663 | 892 | void *base; |
a9a0c51a | 893 | u32 crc; |
31b8006e | 894 | u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc); |
8d63e318 | 895 | char *kaddr; |
31b8006e | 896 | |
8d63e318 | 897 | kaddr = kmap(page); |
31b8006e | 898 | BUG_ON(kaddr == NULL); |
9bd19663 | 899 | base = kaddr + con->out_msg_pos.page_pos + bio_offset; |
a9a0c51a AE |
900 | crc = crc32c(tmpcrc, base, len); |
901 | con->out_msg->footer.data_crc = cpu_to_le32(crc); | |
bca064d2 | 902 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 903 | } |
e36b13cc | 904 | ret = ceph_tcp_sendpage(con->sock, page, |
9bd19663 | 905 | con->out_msg_pos.page_pos + bio_offset, |
e36b13cc | 906 | len, 1); |
31b8006e | 907 | |
0cdf9e60 | 908 | if (do_datacrc) |
31b8006e SW |
909 | kunmap(page); |
910 | ||
911 | if (ret <= 0) | |
912 | goto out; | |
913 | ||
914 | con->out_msg_pos.data_pos += ret; | |
915 | con->out_msg_pos.page_pos += ret; | |
916 | if (ret == len) { | |
917 | con->out_msg_pos.page_pos = 0; | |
918 | con->out_msg_pos.page++; | |
bca064d2 | 919 | con->out_msg_pos.did_page_crc = false; |
68b4476b YS |
920 | if (in_trail) |
921 | list_move_tail(&page->lru, | |
922 | &msg->trail->head); | |
923 | else if (msg->pagelist) | |
58bb3b37 SW |
924 | list_move_tail(&page->lru, |
925 | &msg->pagelist->head); | |
68b4476b YS |
926 | #ifdef CONFIG_BLOCK |
927 | else if (msg->bio) | |
928 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
929 | #endif | |
31b8006e SW |
930 | } |
931 | } | |
932 | ||
933 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
934 | ||
935 | /* prepare and queue up footer, too */ | |
37675b0f | 936 | if (!do_datacrc) |
31b8006e | 937 | con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 938 | con_out_kvec_reset(con); |
859eb799 | 939 | prepare_write_message_footer(con); |
31b8006e SW |
940 | ret = 1; |
941 | out: | |
942 | return ret; | |
943 | } | |
944 | ||
945 | /* | |
946 | * write some zeros | |
947 | */ | |
948 | static int write_partial_skip(struct ceph_connection *con) | |
949 | { | |
950 | int ret; | |
951 | ||
952 | while (con->out_skip > 0) { | |
31739139 | 953 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 954 | |
31739139 | 955 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1); |
31b8006e SW |
956 | if (ret <= 0) |
957 | goto out; | |
958 | con->out_skip -= ret; | |
959 | } | |
960 | ret = 1; | |
961 | out: | |
962 | return ret; | |
963 | } | |
964 | ||
965 | /* | |
966 | * Prepare to read connection handshake, or an ack. | |
967 | */ | |
eed0ef2c SW |
968 | static void prepare_read_banner(struct ceph_connection *con) |
969 | { | |
970 | dout("prepare_read_banner %p\n", con); | |
971 | con->in_base_pos = 0; | |
972 | } | |
973 | ||
31b8006e SW |
974 | static void prepare_read_connect(struct ceph_connection *con) |
975 | { | |
976 | dout("prepare_read_connect %p\n", con); | |
977 | con->in_base_pos = 0; | |
978 | } | |
979 | ||
980 | static void prepare_read_ack(struct ceph_connection *con) | |
981 | { | |
982 | dout("prepare_read_ack %p\n", con); | |
983 | con->in_base_pos = 0; | |
984 | } | |
985 | ||
986 | static void prepare_read_tag(struct ceph_connection *con) | |
987 | { | |
988 | dout("prepare_read_tag %p\n", con); | |
989 | con->in_base_pos = 0; | |
990 | con->in_tag = CEPH_MSGR_TAG_READY; | |
991 | } | |
992 | ||
993 | /* | |
994 | * Prepare to read a message. | |
995 | */ | |
996 | static int prepare_read_message(struct ceph_connection *con) | |
997 | { | |
998 | dout("prepare_read_message %p\n", con); | |
999 | BUG_ON(con->in_msg != NULL); | |
1000 | con->in_base_pos = 0; | |
1001 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1002 | return 0; | |
1003 | } | |
1004 | ||
1005 | ||
1006 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1007 | int end, int size, void *object) |
31b8006e | 1008 | { |
e6cee71f AE |
1009 | while (con->in_base_pos < end) { |
1010 | int left = end - con->in_base_pos; | |
31b8006e SW |
1011 | int have = size - left; |
1012 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1013 | if (ret <= 0) | |
1014 | return ret; | |
1015 | con->in_base_pos += ret; | |
1016 | } | |
1017 | return 1; | |
1018 | } | |
1019 | ||
1020 | ||
1021 | /* | |
1022 | * Read all or part of the connect-side handshake on a new connection | |
1023 | */ | |
eed0ef2c | 1024 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1025 | { |
fd51653f AE |
1026 | int size; |
1027 | int end; | |
1028 | int ret; | |
31b8006e | 1029 | |
eed0ef2c | 1030 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1031 | |
1032 | /* peer's banner */ | |
fd51653f AE |
1033 | size = strlen(CEPH_BANNER); |
1034 | end = size; | |
1035 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1036 | if (ret <= 0) |
1037 | goto out; | |
fd51653f AE |
1038 | |
1039 | size = sizeof (con->actual_peer_addr); | |
1040 | end += size; | |
1041 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1042 | if (ret <= 0) |
1043 | goto out; | |
fd51653f AE |
1044 | |
1045 | size = sizeof (con->peer_addr_for_me); | |
1046 | end += size; | |
1047 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1048 | if (ret <= 0) |
1049 | goto out; | |
fd51653f | 1050 | |
eed0ef2c SW |
1051 | out: |
1052 | return ret; | |
1053 | } | |
1054 | ||
1055 | static int read_partial_connect(struct ceph_connection *con) | |
1056 | { | |
fd51653f AE |
1057 | int size; |
1058 | int end; | |
1059 | int ret; | |
eed0ef2c SW |
1060 | |
1061 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1062 | ||
fd51653f AE |
1063 | size = sizeof (con->in_reply); |
1064 | end = size; | |
1065 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1066 | if (ret <= 0) |
1067 | goto out; | |
fd51653f AE |
1068 | |
1069 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1070 | end += size; | |
1071 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1072 | if (ret <= 0) |
1073 | goto out; | |
31b8006e | 1074 | |
4e7a5dcd SW |
1075 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1076 | con, (int)con->in_reply.tag, | |
1077 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1078 | le32_to_cpu(con->in_reply.global_seq)); |
1079 | out: | |
1080 | return ret; | |
eed0ef2c | 1081 | |
31b8006e SW |
1082 | } |
1083 | ||
1084 | /* | |
1085 | * Verify the hello banner looks okay. | |
1086 | */ | |
1087 | static int verify_hello(struct ceph_connection *con) | |
1088 | { | |
1089 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1090 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1091 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1092 | con->error_msg = "protocol error, bad banner"; |
1093 | return -1; | |
1094 | } | |
1095 | return 0; | |
1096 | } | |
1097 | ||
1098 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1099 | { | |
1100 | switch (ss->ss_family) { | |
1101 | case AF_INET: | |
1102 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1103 | case AF_INET6: | |
1104 | return | |
1105 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1106 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1107 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1108 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1109 | } | |
1110 | return false; | |
1111 | } | |
1112 | ||
1113 | static int addr_port(struct sockaddr_storage *ss) | |
1114 | { | |
1115 | switch (ss->ss_family) { | |
1116 | case AF_INET: | |
f28bcfbe | 1117 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1118 | case AF_INET6: |
f28bcfbe | 1119 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1120 | } |
1121 | return 0; | |
1122 | } | |
1123 | ||
1124 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1125 | { | |
1126 | switch (ss->ss_family) { | |
1127 | case AF_INET: | |
1128 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1129 | break; |
31b8006e SW |
1130 | case AF_INET6: |
1131 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1132 | break; |
31b8006e SW |
1133 | } |
1134 | } | |
1135 | ||
ee3b56f2 NW |
1136 | /* |
1137 | * Unlike other *_pton function semantics, zero indicates success. | |
1138 | */ | |
1139 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1140 | char delim, const char **ipend) | |
1141 | { | |
99f0f3b2 AE |
1142 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1143 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1144 | |
1145 | memset(ss, 0, sizeof(*ss)); | |
1146 | ||
1147 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1148 | ss->ss_family = AF_INET; | |
1149 | return 0; | |
1150 | } | |
1151 | ||
1152 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1153 | ss->ss_family = AF_INET6; | |
1154 | return 0; | |
1155 | } | |
1156 | ||
1157 | return -EINVAL; | |
1158 | } | |
1159 | ||
1160 | /* | |
1161 | * Extract hostname string and resolve using kernel DNS facility. | |
1162 | */ | |
1163 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1164 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1165 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1166 | { | |
1167 | const char *end, *delim_p; | |
1168 | char *colon_p, *ip_addr = NULL; | |
1169 | int ip_len, ret; | |
1170 | ||
1171 | /* | |
1172 | * The end of the hostname occurs immediately preceding the delimiter or | |
1173 | * the port marker (':') where the delimiter takes precedence. | |
1174 | */ | |
1175 | delim_p = memchr(name, delim, namelen); | |
1176 | colon_p = memchr(name, ':', namelen); | |
1177 | ||
1178 | if (delim_p && colon_p) | |
1179 | end = delim_p < colon_p ? delim_p : colon_p; | |
1180 | else if (!delim_p && colon_p) | |
1181 | end = colon_p; | |
1182 | else { | |
1183 | end = delim_p; | |
1184 | if (!end) /* case: hostname:/ */ | |
1185 | end = name + namelen; | |
1186 | } | |
1187 | ||
1188 | if (end <= name) | |
1189 | return -EINVAL; | |
1190 | ||
1191 | /* do dns_resolve upcall */ | |
1192 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1193 | if (ip_len > 0) | |
1194 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1195 | else | |
1196 | ret = -ESRCH; | |
1197 | ||
1198 | kfree(ip_addr); | |
1199 | ||
1200 | *ipend = end; | |
1201 | ||
1202 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1203 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1204 | ||
1205 | return ret; | |
1206 | } | |
1207 | #else | |
1208 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1209 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1210 | { | |
1211 | return -EINVAL; | |
1212 | } | |
1213 | #endif | |
1214 | ||
1215 | /* | |
1216 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1217 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1218 | */ | |
1219 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1220 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1221 | { | |
1222 | int ret; | |
1223 | ||
1224 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1225 | if (ret) | |
1226 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1227 | ||
1228 | return ret; | |
1229 | } | |
1230 | ||
31b8006e SW |
1231 | /* |
1232 | * Parse an ip[:port] list into an addr array. Use the default | |
1233 | * monitor port if a port isn't specified. | |
1234 | */ | |
1235 | int ceph_parse_ips(const char *c, const char *end, | |
1236 | struct ceph_entity_addr *addr, | |
1237 | int max_count, int *count) | |
1238 | { | |
ee3b56f2 | 1239 | int i, ret = -EINVAL; |
31b8006e SW |
1240 | const char *p = c; |
1241 | ||
1242 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1243 | for (i = 0; i < max_count; i++) { | |
1244 | const char *ipend; | |
1245 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1246 | int port; |
39139f64 SW |
1247 | char delim = ','; |
1248 | ||
1249 | if (*p == '[') { | |
1250 | delim = ']'; | |
1251 | p++; | |
1252 | } | |
31b8006e | 1253 | |
ee3b56f2 NW |
1254 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1255 | if (ret) | |
31b8006e | 1256 | goto bad; |
ee3b56f2 NW |
1257 | ret = -EINVAL; |
1258 | ||
31b8006e SW |
1259 | p = ipend; |
1260 | ||
39139f64 SW |
1261 | if (delim == ']') { |
1262 | if (*p != ']') { | |
1263 | dout("missing matching ']'\n"); | |
1264 | goto bad; | |
1265 | } | |
1266 | p++; | |
1267 | } | |
1268 | ||
31b8006e SW |
1269 | /* port? */ |
1270 | if (p < end && *p == ':') { | |
1271 | port = 0; | |
1272 | p++; | |
1273 | while (p < end && *p >= '0' && *p <= '9') { | |
1274 | port = (port * 10) + (*p - '0'); | |
1275 | p++; | |
1276 | } | |
1277 | if (port > 65535 || port == 0) | |
1278 | goto bad; | |
1279 | } else { | |
1280 | port = CEPH_MON_PORT; | |
1281 | } | |
1282 | ||
1283 | addr_set_port(ss, port); | |
1284 | ||
3d14c5d2 | 1285 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1286 | |
1287 | if (p == end) | |
1288 | break; | |
1289 | if (*p != ',') | |
1290 | goto bad; | |
1291 | p++; | |
1292 | } | |
1293 | ||
1294 | if (p != end) | |
1295 | goto bad; | |
1296 | ||
1297 | if (count) | |
1298 | *count = i + 1; | |
1299 | return 0; | |
1300 | ||
1301 | bad: | |
39139f64 | 1302 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1303 | return ret; |
31b8006e | 1304 | } |
3d14c5d2 | 1305 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1306 | |
eed0ef2c | 1307 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1308 | { |
eed0ef2c | 1309 | dout("process_banner on %p\n", con); |
31b8006e SW |
1310 | |
1311 | if (verify_hello(con) < 0) | |
1312 | return -1; | |
1313 | ||
63f2d211 SW |
1314 | ceph_decode_addr(&con->actual_peer_addr); |
1315 | ceph_decode_addr(&con->peer_addr_for_me); | |
1316 | ||
31b8006e SW |
1317 | /* |
1318 | * Make sure the other end is who we wanted. note that the other | |
1319 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1320 | * them the benefit of the doubt. | |
1321 | */ | |
103e2d3a SW |
1322 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1323 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1324 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1325 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1326 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1327 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1328 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1329 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1330 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1331 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1332 | return -1; |
1333 | } | |
1334 | ||
1335 | /* | |
1336 | * did we learn our address? | |
1337 | */ | |
1338 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1339 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1340 | ||
1341 | memcpy(&con->msgr->inst.addr.in_addr, | |
1342 | &con->peer_addr_for_me.in_addr, | |
1343 | sizeof(con->peer_addr_for_me.in_addr)); | |
1344 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1345 | encode_my_addr(con->msgr); |
eed0ef2c | 1346 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1347 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1348 | } |
1349 | ||
eed0ef2c SW |
1350 | set_bit(NEGOTIATING, &con->state); |
1351 | prepare_read_connect(con); | |
1352 | return 0; | |
1353 | } | |
1354 | ||
04a419f9 SW |
1355 | static void fail_protocol(struct ceph_connection *con) |
1356 | { | |
1357 | reset_connection(con); | |
1358 | set_bit(CLOSED, &con->state); /* in case there's queued work */ | |
04a419f9 SW |
1359 | } |
1360 | ||
eed0ef2c SW |
1361 | static int process_connect(struct ceph_connection *con) |
1362 | { | |
3d14c5d2 YS |
1363 | u64 sup_feat = con->msgr->supported_features; |
1364 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1365 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1366 | int ret; |
04a419f9 | 1367 | |
eed0ef2c SW |
1368 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1369 | ||
31b8006e | 1370 | switch (con->in_reply.tag) { |
04a419f9 SW |
1371 | case CEPH_MSGR_TAG_FEATURES: |
1372 | pr_err("%s%lld %s feature set mismatch," | |
1373 | " my %llx < server's %llx, missing %llx\n", | |
1374 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1375 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1376 | sup_feat, server_feat, server_feat & ~sup_feat); |
1377 | con->error_msg = "missing required protocol features"; | |
1378 | fail_protocol(con); | |
1379 | return -1; | |
1380 | ||
31b8006e | 1381 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1382 | pr_err("%s%lld %s protocol version mismatch," |
1383 | " my %d != server's %d\n", | |
1384 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1385 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1386 | le32_to_cpu(con->out_connect.protocol_version), |
1387 | le32_to_cpu(con->in_reply.protocol_version)); | |
1388 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1389 | fail_protocol(con); |
31b8006e SW |
1390 | return -1; |
1391 | ||
4e7a5dcd SW |
1392 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1393 | con->auth_retry++; | |
1394 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1395 | con->auth_retry); | |
1396 | if (con->auth_retry == 2) { | |
1397 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1398 | return -1; |
1399 | } | |
1400 | con->auth_retry = 1; | |
e2200423 | 1401 | con_out_kvec_reset(con); |
e825a66d | 1402 | ret = prepare_write_connect(con); |
0da5d703 SW |
1403 | if (ret < 0) |
1404 | return ret; | |
63733a0f | 1405 | prepare_read_connect(con); |
4e7a5dcd | 1406 | break; |
31b8006e SW |
1407 | |
1408 | case CEPH_MSGR_TAG_RESETSESSION: | |
1409 | /* | |
1410 | * If we connected with a large connect_seq but the peer | |
1411 | * has no record of a session with us (no connection, or | |
1412 | * connect_seq == 0), they will send RESETSESION to indicate | |
1413 | * that they must have reset their session, and may have | |
1414 | * dropped messages. | |
1415 | */ | |
1416 | dout("process_connect got RESET peer seq %u\n", | |
1417 | le32_to_cpu(con->in_connect.connect_seq)); | |
1418 | pr_err("%s%lld %s connection reset\n", | |
1419 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1420 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1421 | reset_connection(con); |
e2200423 | 1422 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1423 | ret = prepare_write_connect(con); |
1424 | if (ret < 0) | |
1425 | return ret; | |
31b8006e SW |
1426 | prepare_read_connect(con); |
1427 | ||
1428 | /* Tell ceph about it. */ | |
ec302645 | 1429 | mutex_unlock(&con->mutex); |
31b8006e SW |
1430 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1431 | if (con->ops->peer_reset) | |
1432 | con->ops->peer_reset(con); | |
ec302645 | 1433 | mutex_lock(&con->mutex); |
0da5d703 SW |
1434 | if (test_bit(CLOSED, &con->state) || |
1435 | test_bit(OPENING, &con->state)) | |
1436 | return -EAGAIN; | |
31b8006e SW |
1437 | break; |
1438 | ||
1439 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1440 | /* | |
1441 | * If we sent a smaller connect_seq than the peer has, try | |
1442 | * again with a larger value. | |
1443 | */ | |
1444 | dout("process_connect got RETRY my seq = %u, peer_seq = %u\n", | |
1445 | le32_to_cpu(con->out_connect.connect_seq), | |
1446 | le32_to_cpu(con->in_connect.connect_seq)); | |
1447 | con->connect_seq = le32_to_cpu(con->in_connect.connect_seq); | |
e2200423 | 1448 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1449 | ret = prepare_write_connect(con); |
1450 | if (ret < 0) | |
1451 | return ret; | |
31b8006e SW |
1452 | prepare_read_connect(con); |
1453 | break; | |
1454 | ||
1455 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1456 | /* | |
1457 | * If we sent a smaller global_seq than the peer has, try | |
1458 | * again with a larger value. | |
1459 | */ | |
eed0ef2c | 1460 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e SW |
1461 | con->peer_global_seq, |
1462 | le32_to_cpu(con->in_connect.global_seq)); | |
1463 | get_global_seq(con->msgr, | |
1464 | le32_to_cpu(con->in_connect.global_seq)); | |
e2200423 | 1465 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1466 | ret = prepare_write_connect(con); |
1467 | if (ret < 0) | |
1468 | return ret; | |
31b8006e SW |
1469 | prepare_read_connect(con); |
1470 | break; | |
1471 | ||
1472 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1473 | if (req_feat & ~server_feat) { |
1474 | pr_err("%s%lld %s protocol feature mismatch," | |
1475 | " my required %llx > server's %llx, need %llx\n", | |
1476 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1477 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1478 | req_feat, server_feat, req_feat & ~server_feat); |
1479 | con->error_msg = "missing required protocol features"; | |
1480 | fail_protocol(con); | |
1481 | return -1; | |
1482 | } | |
31b8006e | 1483 | clear_bit(CONNECTING, &con->state); |
31b8006e SW |
1484 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1485 | con->connect_seq++; | |
aba558e2 | 1486 | con->peer_features = server_feat; |
31b8006e SW |
1487 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1488 | con->peer_global_seq, | |
1489 | le32_to_cpu(con->in_reply.connect_seq), | |
1490 | con->connect_seq); | |
1491 | WARN_ON(con->connect_seq != | |
1492 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1493 | |
1494 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
1495 | set_bit(LOSSYTX, &con->state); | |
1496 | ||
31b8006e SW |
1497 | prepare_read_tag(con); |
1498 | break; | |
1499 | ||
1500 | case CEPH_MSGR_TAG_WAIT: | |
1501 | /* | |
1502 | * If there is a connection race (we are opening | |
1503 | * connections to each other), one of us may just have | |
1504 | * to WAIT. This shouldn't happen if we are the | |
1505 | * client. | |
1506 | */ | |
04177882 SW |
1507 | pr_err("process_connect got WAIT as client\n"); |
1508 | con->error_msg = "protocol error, got WAIT as client"; | |
1509 | return -1; | |
31b8006e SW |
1510 | |
1511 | default: | |
1512 | pr_err("connect protocol error, will retry\n"); | |
1513 | con->error_msg = "protocol error, garbage tag during connect"; | |
1514 | return -1; | |
1515 | } | |
1516 | return 0; | |
1517 | } | |
1518 | ||
1519 | ||
1520 | /* | |
1521 | * read (part of) an ack | |
1522 | */ | |
1523 | static int read_partial_ack(struct ceph_connection *con) | |
1524 | { | |
fd51653f AE |
1525 | int size = sizeof (con->in_temp_ack); |
1526 | int end = size; | |
1527 | ||
1528 | return read_partial(con, end, size, &con->in_temp_ack); | |
31b8006e SW |
1529 | } |
1530 | ||
1531 | ||
1532 | /* | |
1533 | * We can finally discard anything that's been acked. | |
1534 | */ | |
1535 | static void process_ack(struct ceph_connection *con) | |
1536 | { | |
1537 | struct ceph_msg *m; | |
1538 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1539 | u64 seq; | |
1540 | ||
31b8006e SW |
1541 | while (!list_empty(&con->out_sent)) { |
1542 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1543 | list_head); | |
1544 | seq = le64_to_cpu(m->hdr.seq); | |
1545 | if (seq > ack) | |
1546 | break; | |
1547 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1548 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1549 | m->ack_stamp = jiffies; |
31b8006e SW |
1550 | ceph_msg_remove(m); |
1551 | } | |
31b8006e SW |
1552 | prepare_read_tag(con); |
1553 | } | |
1554 | ||
1555 | ||
1556 | ||
1557 | ||
2450418c | 1558 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1559 | struct kvec *section, |
1560 | unsigned int sec_len, u32 *crc) | |
2450418c | 1561 | { |
68b4476b | 1562 | int ret, left; |
2450418c YS |
1563 | |
1564 | BUG_ON(!section); | |
1565 | ||
1566 | while (section->iov_len < sec_len) { | |
1567 | BUG_ON(section->iov_base == NULL); | |
1568 | left = sec_len - section->iov_len; | |
1569 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1570 | section->iov_len, left); | |
1571 | if (ret <= 0) | |
1572 | return ret; | |
1573 | section->iov_len += ret; | |
2450418c | 1574 | } |
fe3ad593 AE |
1575 | if (section->iov_len == sec_len) |
1576 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1577 | |
2450418c YS |
1578 | return 1; |
1579 | } | |
31b8006e | 1580 | |
2450418c YS |
1581 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, |
1582 | struct ceph_msg_header *hdr, | |
1583 | int *skip); | |
68b4476b YS |
1584 | |
1585 | ||
1586 | static int read_partial_message_pages(struct ceph_connection *con, | |
1587 | struct page **pages, | |
bca064d2 | 1588 | unsigned data_len, bool do_datacrc) |
68b4476b YS |
1589 | { |
1590 | void *p; | |
1591 | int ret; | |
1592 | int left; | |
1593 | ||
1594 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1595 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1596 | /* (page) data */ | |
1597 | BUG_ON(pages == NULL); | |
1598 | p = kmap(pages[con->in_msg_pos.page]); | |
1599 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1600 | left); | |
bca064d2 | 1601 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1602 | con->in_data_crc = |
1603 | crc32c(con->in_data_crc, | |
1604 | p + con->in_msg_pos.page_pos, ret); | |
1605 | kunmap(pages[con->in_msg_pos.page]); | |
1606 | if (ret <= 0) | |
1607 | return ret; | |
1608 | con->in_msg_pos.data_pos += ret; | |
1609 | con->in_msg_pos.page_pos += ret; | |
1610 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1611 | con->in_msg_pos.page_pos = 0; | |
1612 | con->in_msg_pos.page++; | |
1613 | } | |
1614 | ||
1615 | return ret; | |
1616 | } | |
1617 | ||
1618 | #ifdef CONFIG_BLOCK | |
1619 | static int read_partial_message_bio(struct ceph_connection *con, | |
1620 | struct bio **bio_iter, int *bio_seg, | |
bca064d2 | 1621 | unsigned data_len, bool do_datacrc) |
68b4476b YS |
1622 | { |
1623 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1624 | void *p; | |
1625 | int ret, left; | |
1626 | ||
1627 | if (IS_ERR(bv)) | |
1628 | return PTR_ERR(bv); | |
1629 | ||
1630 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1631 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1632 | ||
1633 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1634 | ||
1635 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1636 | left); | |
bca064d2 | 1637 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1638 | con->in_data_crc = |
1639 | crc32c(con->in_data_crc, | |
1640 | p + con->in_msg_pos.page_pos, ret); | |
1641 | kunmap(bv->bv_page); | |
1642 | if (ret <= 0) | |
1643 | return ret; | |
1644 | con->in_msg_pos.data_pos += ret; | |
1645 | con->in_msg_pos.page_pos += ret; | |
1646 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1647 | con->in_msg_pos.page_pos = 0; | |
1648 | iter_bio_next(bio_iter, bio_seg); | |
1649 | } | |
1650 | ||
1651 | return ret; | |
1652 | } | |
1653 | #endif | |
1654 | ||
31b8006e SW |
1655 | /* |
1656 | * read (part of) a message. | |
1657 | */ | |
1658 | static int read_partial_message(struct ceph_connection *con) | |
1659 | { | |
1660 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1661 | int size; |
1662 | int end; | |
31b8006e | 1663 | int ret; |
c5c6b19d | 1664 | unsigned front_len, middle_len, data_len; |
37675b0f | 1665 | bool do_datacrc = !con->msgr->nocrc; |
2450418c | 1666 | int skip; |
ae18756b | 1667 | u64 seq; |
fe3ad593 | 1668 | u32 crc; |
31b8006e SW |
1669 | |
1670 | dout("read_partial_message con %p msg %p\n", con, m); | |
1671 | ||
1672 | /* header */ | |
fd51653f AE |
1673 | size = sizeof (con->in_hdr); |
1674 | end = size; | |
1675 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1676 | if (ret <= 0) |
1677 | return ret; | |
fe3ad593 AE |
1678 | |
1679 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1680 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1681 | pr_err("read_partial_message bad hdr " | |
1682 | " crc %u != expected %u\n", | |
1683 | crc, con->in_hdr.crc); | |
1684 | return -EBADMSG; | |
1685 | } | |
1686 | ||
31b8006e SW |
1687 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1688 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1689 | return -EIO; | |
1690 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1691 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1692 | return -EIO; | |
1693 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1694 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1695 | return -EIO; | |
1696 | ||
ae18756b SW |
1697 | /* verify seq# */ |
1698 | seq = le64_to_cpu(con->in_hdr.seq); | |
1699 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1700 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1701 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1702 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1703 | seq, con->in_seq + 1); |
1704 | con->in_base_pos = -front_len - middle_len - data_len - | |
1705 | sizeof(m->footer); | |
1706 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1707 | return 0; |
1708 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1709 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1710 | seq, con->in_seq + 1); | |
1711 | con->error_msg = "bad message sequence # for incoming message"; | |
1712 | return -EBADMSG; | |
1713 | } | |
1714 | ||
31b8006e SW |
1715 | /* allocate message? */ |
1716 | if (!con->in_msg) { | |
1717 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1718 | con->in_hdr.front_len, con->in_hdr.data_len); | |
ae32be31 | 1719 | skip = 0; |
2450418c YS |
1720 | con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip); |
1721 | if (skip) { | |
31b8006e | 1722 | /* skip this message */ |
a79832f2 | 1723 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1724 | BUG_ON(con->in_msg); |
31b8006e SW |
1725 | con->in_base_pos = -front_len - middle_len - data_len - |
1726 | sizeof(m->footer); | |
1727 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1728 | con->in_seq++; |
31b8006e SW |
1729 | return 0; |
1730 | } | |
a79832f2 | 1731 | if (!con->in_msg) { |
5b3a4db3 SW |
1732 | con->error_msg = |
1733 | "error allocating memory for incoming message"; | |
a79832f2 | 1734 | return -ENOMEM; |
31b8006e SW |
1735 | } |
1736 | m = con->in_msg; | |
1737 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1738 | if (m->middle) |
1739 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1740 | |
1741 | con->in_msg_pos.page = 0; | |
68b4476b | 1742 | if (m->pages) |
c5c6b19d | 1743 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1744 | else |
1745 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1746 | con->in_msg_pos.data_pos = 0; |
31b8006e SW |
1747 | } |
1748 | ||
1749 | /* front */ | |
2450418c YS |
1750 | ret = read_partial_message_section(con, &m->front, front_len, |
1751 | &con->in_front_crc); | |
1752 | if (ret <= 0) | |
1753 | return ret; | |
31b8006e SW |
1754 | |
1755 | /* middle */ | |
2450418c | 1756 | if (m->middle) { |
213c99ee SW |
1757 | ret = read_partial_message_section(con, &m->middle->vec, |
1758 | middle_len, | |
2450418c | 1759 | &con->in_middle_crc); |
31b8006e SW |
1760 | if (ret <= 0) |
1761 | return ret; | |
31b8006e | 1762 | } |
68b4476b YS |
1763 | #ifdef CONFIG_BLOCK |
1764 | if (m->bio && !m->bio_iter) | |
1765 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1766 | #endif | |
31b8006e SW |
1767 | |
1768 | /* (page) data */ | |
31b8006e | 1769 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1770 | if (m->pages) { |
1771 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1772 | data_len, do_datacrc); |
68b4476b YS |
1773 | if (ret <= 0) |
1774 | return ret; | |
1775 | #ifdef CONFIG_BLOCK | |
1776 | } else if (m->bio) { | |
1777 | ||
1778 | ret = read_partial_message_bio(con, | |
1779 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1780 | data_len, do_datacrc); |
68b4476b YS |
1781 | if (ret <= 0) |
1782 | return ret; | |
1783 | #endif | |
1784 | } else { | |
1785 | BUG_ON(1); | |
31b8006e SW |
1786 | } |
1787 | } | |
1788 | ||
31b8006e | 1789 | /* footer */ |
fd51653f AE |
1790 | size = sizeof (m->footer); |
1791 | end += size; | |
1792 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
1793 | if (ret <= 0) |
1794 | return ret; | |
1795 | ||
31b8006e SW |
1796 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
1797 | m, front_len, m->footer.front_crc, middle_len, | |
1798 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1799 | ||
1800 | /* crc ok? */ | |
1801 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1802 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1803 | m, con->in_front_crc, m->footer.front_crc); | |
1804 | return -EBADMSG; | |
1805 | } | |
1806 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1807 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1808 | m, con->in_middle_crc, m->footer.middle_crc); | |
1809 | return -EBADMSG; | |
1810 | } | |
bca064d2 | 1811 | if (do_datacrc && |
31b8006e SW |
1812 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
1813 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1814 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1815 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1816 | return -EBADMSG; | |
1817 | } | |
1818 | ||
1819 | return 1; /* done! */ | |
1820 | } | |
1821 | ||
1822 | /* | |
1823 | * Process message. This happens in the worker thread. The callback should | |
1824 | * be careful not to do anything that waits on other incoming messages or it | |
1825 | * may deadlock. | |
1826 | */ | |
1827 | static void process_message(struct ceph_connection *con) | |
1828 | { | |
5e095e8b | 1829 | struct ceph_msg *msg; |
31b8006e | 1830 | |
5e095e8b | 1831 | msg = con->in_msg; |
31b8006e SW |
1832 | con->in_msg = NULL; |
1833 | ||
1834 | /* if first message, set peer_name */ | |
1835 | if (con->peer_name.type == 0) | |
dbad185d | 1836 | con->peer_name = msg->hdr.src; |
31b8006e | 1837 | |
31b8006e | 1838 | con->in_seq++; |
ec302645 | 1839 | mutex_unlock(&con->mutex); |
31b8006e SW |
1840 | |
1841 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1842 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1843 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1844 | le16_to_cpu(msg->hdr.type), |
1845 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1846 | le32_to_cpu(msg->hdr.front_len), | |
1847 | le32_to_cpu(msg->hdr.data_len), | |
1848 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1849 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1850 | |
1851 | mutex_lock(&con->mutex); | |
31b8006e SW |
1852 | prepare_read_tag(con); |
1853 | } | |
1854 | ||
1855 | ||
1856 | /* | |
1857 | * Write something to the socket. Called in a worker thread when the | |
1858 | * socket appears to be writeable and we have something ready to send. | |
1859 | */ | |
1860 | static int try_write(struct ceph_connection *con) | |
1861 | { | |
31b8006e SW |
1862 | int ret = 1; |
1863 | ||
1864 | dout("try_write start %p state %lu nref %d\n", con, con->state, | |
1865 | atomic_read(&con->nref)); | |
1866 | ||
31b8006e SW |
1867 | more: |
1868 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
1869 | ||
1870 | /* open the socket first? */ | |
1871 | if (con->sock == NULL) { | |
e2200423 | 1872 | con_out_kvec_reset(con); |
e825a66d | 1873 | prepare_write_banner(con); |
5a0f8fdd AE |
1874 | ret = prepare_write_connect(con); |
1875 | if (ret < 0) | |
1876 | goto out; | |
eed0ef2c | 1877 | prepare_read_banner(con); |
31b8006e | 1878 | set_bit(CONNECTING, &con->state); |
eed0ef2c | 1879 | clear_bit(NEGOTIATING, &con->state); |
31b8006e | 1880 | |
cf3e5c40 | 1881 | BUG_ON(con->in_msg); |
31b8006e SW |
1882 | con->in_tag = CEPH_MSGR_TAG_READY; |
1883 | dout("try_write initiating connect on %p new state %lu\n", | |
1884 | con, con->state); | |
41617d0c AE |
1885 | ret = ceph_tcp_connect(con); |
1886 | if (ret < 0) { | |
31b8006e | 1887 | con->error_msg = "connect error"; |
31b8006e SW |
1888 | goto out; |
1889 | } | |
1890 | } | |
1891 | ||
1892 | more_kvec: | |
1893 | /* kvec data queued? */ | |
1894 | if (con->out_skip) { | |
1895 | ret = write_partial_skip(con); | |
1896 | if (ret <= 0) | |
42961d23 | 1897 | goto out; |
31b8006e SW |
1898 | } |
1899 | if (con->out_kvec_left) { | |
1900 | ret = write_partial_kvec(con); | |
1901 | if (ret <= 0) | |
42961d23 | 1902 | goto out; |
31b8006e SW |
1903 | } |
1904 | ||
1905 | /* msg pages? */ | |
1906 | if (con->out_msg) { | |
c86a2930 SW |
1907 | if (con->out_msg_done) { |
1908 | ceph_msg_put(con->out_msg); | |
1909 | con->out_msg = NULL; /* we're done with this one */ | |
1910 | goto do_next; | |
1911 | } | |
1912 | ||
31b8006e SW |
1913 | ret = write_partial_msg_pages(con); |
1914 | if (ret == 1) | |
1915 | goto more_kvec; /* we need to send the footer, too! */ | |
1916 | if (ret == 0) | |
42961d23 | 1917 | goto out; |
31b8006e SW |
1918 | if (ret < 0) { |
1919 | dout("try_write write_partial_msg_pages err %d\n", | |
1920 | ret); | |
42961d23 | 1921 | goto out; |
31b8006e SW |
1922 | } |
1923 | } | |
1924 | ||
c86a2930 | 1925 | do_next: |
31b8006e SW |
1926 | if (!test_bit(CONNECTING, &con->state)) { |
1927 | /* is anything else pending? */ | |
1928 | if (!list_empty(&con->out_queue)) { | |
1929 | prepare_write_message(con); | |
1930 | goto more; | |
1931 | } | |
1932 | if (con->in_seq > con->in_seq_acked) { | |
1933 | prepare_write_ack(con); | |
1934 | goto more; | |
1935 | } | |
1936 | if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) { | |
1937 | prepare_write_keepalive(con); | |
1938 | goto more; | |
1939 | } | |
1940 | } | |
1941 | ||
1942 | /* Nothing to do! */ | |
1943 | clear_bit(WRITE_PENDING, &con->state); | |
1944 | dout("try_write nothing else to write.\n"); | |
31b8006e SW |
1945 | ret = 0; |
1946 | out: | |
42961d23 | 1947 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
1948 | return ret; |
1949 | } | |
1950 | ||
1951 | ||
1952 | ||
1953 | /* | |
1954 | * Read what we can from the socket. | |
1955 | */ | |
1956 | static int try_read(struct ceph_connection *con) | |
1957 | { | |
31b8006e SW |
1958 | int ret = -1; |
1959 | ||
1960 | if (!con->sock) | |
1961 | return 0; | |
1962 | ||
1963 | if (test_bit(STANDBY, &con->state)) | |
1964 | return 0; | |
1965 | ||
1966 | dout("try_read start on %p\n", con); | |
ec302645 | 1967 | |
31b8006e SW |
1968 | more: |
1969 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, | |
1970 | con->in_base_pos); | |
0da5d703 SW |
1971 | |
1972 | /* | |
1973 | * process_connect and process_message drop and re-take | |
1974 | * con->mutex. make sure we handle a racing close or reopen. | |
1975 | */ | |
1976 | if (test_bit(CLOSED, &con->state) || | |
1977 | test_bit(OPENING, &con->state)) { | |
1978 | ret = -EAGAIN; | |
1979 | goto out; | |
1980 | } | |
1981 | ||
31b8006e | 1982 | if (test_bit(CONNECTING, &con->state)) { |
eed0ef2c SW |
1983 | if (!test_bit(NEGOTIATING, &con->state)) { |
1984 | dout("try_read connecting\n"); | |
1985 | ret = read_partial_banner(con); | |
1986 | if (ret <= 0) | |
eed0ef2c | 1987 | goto out; |
98bdb0aa SW |
1988 | ret = process_banner(con); |
1989 | if (ret < 0) | |
1990 | goto out; | |
eed0ef2c | 1991 | } |
31b8006e SW |
1992 | ret = read_partial_connect(con); |
1993 | if (ret <= 0) | |
31b8006e | 1994 | goto out; |
98bdb0aa SW |
1995 | ret = process_connect(con); |
1996 | if (ret < 0) | |
1997 | goto out; | |
31b8006e SW |
1998 | goto more; |
1999 | } | |
2000 | ||
2001 | if (con->in_base_pos < 0) { | |
2002 | /* | |
2003 | * skipping + discarding content. | |
2004 | * | |
2005 | * FIXME: there must be a better way to do this! | |
2006 | */ | |
84495f49 AE |
2007 | static char buf[SKIP_BUF_SIZE]; |
2008 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2009 | ||
31b8006e SW |
2010 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2011 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2012 | if (ret <= 0) | |
98bdb0aa | 2013 | goto out; |
31b8006e SW |
2014 | con->in_base_pos += ret; |
2015 | if (con->in_base_pos) | |
2016 | goto more; | |
2017 | } | |
2018 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2019 | /* | |
2020 | * what's next? | |
2021 | */ | |
2022 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2023 | if (ret <= 0) | |
98bdb0aa | 2024 | goto out; |
31b8006e SW |
2025 | dout("try_read got tag %d\n", (int)con->in_tag); |
2026 | switch (con->in_tag) { | |
2027 | case CEPH_MSGR_TAG_MSG: | |
2028 | prepare_read_message(con); | |
2029 | break; | |
2030 | case CEPH_MSGR_TAG_ACK: | |
2031 | prepare_read_ack(con); | |
2032 | break; | |
2033 | case CEPH_MSGR_TAG_CLOSE: | |
2034 | set_bit(CLOSED, &con->state); /* fixme */ | |
98bdb0aa | 2035 | goto out; |
31b8006e SW |
2036 | default: |
2037 | goto bad_tag; | |
2038 | } | |
2039 | } | |
2040 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2041 | ret = read_partial_message(con); | |
2042 | if (ret <= 0) { | |
2043 | switch (ret) { | |
2044 | case -EBADMSG: | |
2045 | con->error_msg = "bad crc"; | |
2046 | ret = -EIO; | |
98bdb0aa | 2047 | break; |
31b8006e SW |
2048 | case -EIO: |
2049 | con->error_msg = "io error"; | |
98bdb0aa | 2050 | break; |
31b8006e | 2051 | } |
98bdb0aa | 2052 | goto out; |
31b8006e SW |
2053 | } |
2054 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2055 | goto more; | |
2056 | process_message(con); | |
2057 | goto more; | |
2058 | } | |
2059 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2060 | ret = read_partial_ack(con); | |
2061 | if (ret <= 0) | |
98bdb0aa | 2062 | goto out; |
31b8006e SW |
2063 | process_ack(con); |
2064 | goto more; | |
2065 | } | |
2066 | ||
31b8006e | 2067 | out: |
98bdb0aa | 2068 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2069 | return ret; |
2070 | ||
2071 | bad_tag: | |
2072 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2073 | con->error_msg = "protocol error, garbage tag"; | |
2074 | ret = -1; | |
2075 | goto out; | |
2076 | } | |
2077 | ||
2078 | ||
2079 | /* | |
2080 | * Atomically queue work on a connection. Bump @con reference to | |
2081 | * avoid races with connection teardown. | |
31b8006e SW |
2082 | */ |
2083 | static void queue_con(struct ceph_connection *con) | |
2084 | { | |
31b8006e SW |
2085 | if (!con->ops->get(con)) { |
2086 | dout("queue_con %p ref count 0\n", con); | |
2087 | return; | |
2088 | } | |
2089 | ||
f363e45f | 2090 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2091 | dout("queue_con %p - already queued\n", con); |
2092 | con->ops->put(con); | |
2093 | } else { | |
2094 | dout("queue_con %p\n", con); | |
2095 | } | |
2096 | } | |
2097 | ||
2098 | /* | |
2099 | * Do some work on a connection. Drop a connection ref when we're done. | |
2100 | */ | |
2101 | static void con_work(struct work_struct *work) | |
2102 | { | |
2103 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2104 | work.work); | |
0da5d703 | 2105 | int ret; |
31b8006e | 2106 | |
9dd4658d | 2107 | mutex_lock(&con->mutex); |
0da5d703 | 2108 | restart: |
60bf8bf8 SW |
2109 | if (test_and_clear_bit(BACKOFF, &con->state)) { |
2110 | dout("con_work %p backing off\n", con); | |
2111 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2112 | round_jiffies_relative(con->delay))) { | |
2113 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2114 | mutex_unlock(&con->mutex); | |
2115 | return; | |
2116 | } else { | |
2117 | con->ops->put(con); | |
2118 | dout("con_work %p FAILED to back off %lu\n", con, | |
2119 | con->delay); | |
2120 | } | |
2121 | } | |
9dd4658d | 2122 | |
e00de341 SW |
2123 | if (test_bit(STANDBY, &con->state)) { |
2124 | dout("con_work %p STANDBY\n", con); | |
2125 | goto done; | |
2126 | } | |
31b8006e SW |
2127 | if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */ |
2128 | dout("con_work CLOSED\n"); | |
2129 | con_close_socket(con); | |
2130 | goto done; | |
2131 | } | |
2132 | if (test_and_clear_bit(OPENING, &con->state)) { | |
2133 | /* reopen w/ new peer */ | |
2134 | dout("con_work OPENING\n"); | |
2135 | con_close_socket(con); | |
2136 | } | |
2137 | ||
0da5d703 SW |
2138 | if (test_and_clear_bit(SOCK_CLOSED, &con->state)) |
2139 | goto fault; | |
2140 | ||
2141 | ret = try_read(con); | |
2142 | if (ret == -EAGAIN) | |
2143 | goto restart; | |
2144 | if (ret < 0) | |
2145 | goto fault; | |
2146 | ||
2147 | ret = try_write(con); | |
2148 | if (ret == -EAGAIN) | |
2149 | goto restart; | |
2150 | if (ret < 0) | |
2151 | goto fault; | |
31b8006e SW |
2152 | |
2153 | done: | |
9dd4658d | 2154 | mutex_unlock(&con->mutex); |
9dd4658d | 2155 | done_unlocked: |
31b8006e | 2156 | con->ops->put(con); |
0da5d703 SW |
2157 | return; |
2158 | ||
2159 | fault: | |
2160 | mutex_unlock(&con->mutex); | |
2161 | ceph_fault(con); /* error/fault path */ | |
2162 | goto done_unlocked; | |
31b8006e SW |
2163 | } |
2164 | ||
2165 | ||
2166 | /* | |
2167 | * Generic error/fault handler. A retry mechanism is used with | |
2168 | * exponential backoff | |
2169 | */ | |
2170 | static void ceph_fault(struct ceph_connection *con) | |
2171 | { | |
2172 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2173 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2174 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2175 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
2176 | |
2177 | if (test_bit(LOSSYTX, &con->state)) { | |
2178 | dout("fault on LOSSYTX channel\n"); | |
2179 | goto out; | |
2180 | } | |
2181 | ||
ec302645 | 2182 | mutex_lock(&con->mutex); |
91e45ce3 SW |
2183 | if (test_bit(CLOSED, &con->state)) |
2184 | goto out_unlock; | |
ec302645 | 2185 | |
31b8006e | 2186 | con_close_socket(con); |
5e095e8b SW |
2187 | |
2188 | if (con->in_msg) { | |
2189 | ceph_msg_put(con->in_msg); | |
2190 | con->in_msg = NULL; | |
2191 | } | |
31b8006e | 2192 | |
e80a52d1 SW |
2193 | /* Requeue anything that hasn't been acked */ |
2194 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2195 | |
e76661d0 SW |
2196 | /* If there are no messages queued or keepalive pending, place |
2197 | * the connection in a STANDBY state */ | |
2198 | if (list_empty(&con->out_queue) && | |
2199 | !test_bit(KEEPALIVE_PENDING, &con->state)) { | |
e00de341 SW |
2200 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
2201 | clear_bit(WRITE_PENDING, &con->state); | |
31b8006e | 2202 | set_bit(STANDBY, &con->state); |
e80a52d1 SW |
2203 | } else { |
2204 | /* retry after a delay. */ | |
2205 | if (con->delay == 0) | |
2206 | con->delay = BASE_DELAY_INTERVAL; | |
2207 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2208 | con->delay *= 2; | |
e80a52d1 SW |
2209 | con->ops->get(con); |
2210 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2211 | round_jiffies_relative(con->delay))) { |
2212 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2213 | } else { | |
e80a52d1 | 2214 | con->ops->put(con); |
60bf8bf8 SW |
2215 | dout("fault failed to queue %p delay %lu, backoff\n", |
2216 | con, con->delay); | |
2217 | /* | |
2218 | * In many cases we see a socket state change | |
2219 | * while con_work is running and end up | |
2220 | * queuing (non-delayed) work, such that we | |
2221 | * can't backoff with a delay. Set a flag so | |
2222 | * that when con_work restarts we schedule the | |
2223 | * delay then. | |
2224 | */ | |
2225 | set_bit(BACKOFF, &con->state); | |
2226 | } | |
31b8006e SW |
2227 | } |
2228 | ||
91e45ce3 SW |
2229 | out_unlock: |
2230 | mutex_unlock(&con->mutex); | |
31b8006e | 2231 | out: |
161fd65a SW |
2232 | /* |
2233 | * in case we faulted due to authentication, invalidate our | |
2234 | * current tickets so that we can get new ones. | |
213c99ee | 2235 | */ |
161fd65a SW |
2236 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2237 | dout("calling invalidate_authorizer()\n"); | |
2238 | con->ops->invalidate_authorizer(con); | |
2239 | } | |
2240 | ||
31b8006e SW |
2241 | if (con->ops->fault) |
2242 | con->ops->fault(con); | |
2243 | } | |
2244 | ||
2245 | ||
2246 | ||
2247 | /* | |
2248 | * create a new messenger instance | |
2249 | */ | |
3d14c5d2 YS |
2250 | struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr, |
2251 | u32 supported_features, | |
2252 | u32 required_features) | |
31b8006e SW |
2253 | { |
2254 | struct ceph_messenger *msgr; | |
2255 | ||
2256 | msgr = kzalloc(sizeof(*msgr), GFP_KERNEL); | |
2257 | if (msgr == NULL) | |
2258 | return ERR_PTR(-ENOMEM); | |
2259 | ||
3d14c5d2 YS |
2260 | msgr->supported_features = supported_features; |
2261 | msgr->required_features = required_features; | |
2262 | ||
31b8006e SW |
2263 | spin_lock_init(&msgr->global_seq_lock); |
2264 | ||
31b8006e SW |
2265 | if (myaddr) |
2266 | msgr->inst.addr = *myaddr; | |
2267 | ||
2268 | /* select a random nonce */ | |
ac8839d7 | 2269 | msgr->inst.addr.type = 0; |
103e2d3a | 2270 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2271 | encode_my_addr(msgr); |
31b8006e SW |
2272 | |
2273 | dout("messenger_create %p\n", msgr); | |
2274 | return msgr; | |
2275 | } | |
3d14c5d2 | 2276 | EXPORT_SYMBOL(ceph_messenger_create); |
31b8006e SW |
2277 | |
2278 | void ceph_messenger_destroy(struct ceph_messenger *msgr) | |
2279 | { | |
2280 | dout("destroy %p\n", msgr); | |
31b8006e SW |
2281 | kfree(msgr); |
2282 | dout("destroyed messenger %p\n", msgr); | |
2283 | } | |
3d14c5d2 | 2284 | EXPORT_SYMBOL(ceph_messenger_destroy); |
31b8006e | 2285 | |
e00de341 SW |
2286 | static void clear_standby(struct ceph_connection *con) |
2287 | { | |
2288 | /* come back from STANDBY? */ | |
2289 | if (test_and_clear_bit(STANDBY, &con->state)) { | |
2290 | mutex_lock(&con->mutex); | |
2291 | dout("clear_standby %p and ++connect_seq\n", con); | |
2292 | con->connect_seq++; | |
2293 | WARN_ON(test_bit(WRITE_PENDING, &con->state)); | |
2294 | WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state)); | |
2295 | mutex_unlock(&con->mutex); | |
2296 | } | |
2297 | } | |
2298 | ||
31b8006e SW |
2299 | /* |
2300 | * Queue up an outgoing message on the given connection. | |
2301 | */ | |
2302 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2303 | { | |
2304 | if (test_bit(CLOSED, &con->state)) { | |
2305 | dout("con_send %p closed, dropping %p\n", con, msg); | |
2306 | ceph_msg_put(msg); | |
2307 | return; | |
2308 | } | |
2309 | ||
2310 | /* set src+dst */ | |
dbad185d | 2311 | msg->hdr.src = con->msgr->inst.name; |
31b8006e | 2312 | |
3ca02ef9 SW |
2313 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
2314 | ||
e84346b7 SW |
2315 | msg->needs_out_seq = true; |
2316 | ||
31b8006e | 2317 | /* queue */ |
ec302645 | 2318 | mutex_lock(&con->mutex); |
31b8006e SW |
2319 | BUG_ON(!list_empty(&msg->list_head)); |
2320 | list_add_tail(&msg->list_head, &con->out_queue); | |
2321 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2322 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2323 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2324 | le32_to_cpu(msg->hdr.front_len), | |
2325 | le32_to_cpu(msg->hdr.middle_len), | |
2326 | le32_to_cpu(msg->hdr.data_len)); | |
ec302645 | 2327 | mutex_unlock(&con->mutex); |
31b8006e SW |
2328 | |
2329 | /* if there wasn't anything waiting to send before, queue | |
2330 | * new work */ | |
e00de341 | 2331 | clear_standby(con); |
31b8006e SW |
2332 | if (test_and_set_bit(WRITE_PENDING, &con->state) == 0) |
2333 | queue_con(con); | |
2334 | } | |
3d14c5d2 | 2335 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2336 | |
2337 | /* | |
2338 | * Revoke a message that was previously queued for send | |
2339 | */ | |
2340 | void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg) | |
2341 | { | |
ec302645 | 2342 | mutex_lock(&con->mutex); |
31b8006e | 2343 | if (!list_empty(&msg->list_head)) { |
ed98adad | 2344 | dout("con_revoke %p msg %p - was on queue\n", con, msg); |
31b8006e SW |
2345 | list_del_init(&msg->list_head); |
2346 | ceph_msg_put(msg); | |
2347 | msg->hdr.seq = 0; | |
ed98adad SW |
2348 | } |
2349 | if (con->out_msg == msg) { | |
2350 | dout("con_revoke %p msg %p - was sending\n", con, msg); | |
2351 | con->out_msg = NULL; | |
31b8006e SW |
2352 | if (con->out_kvec_is_msg) { |
2353 | con->out_skip = con->out_kvec_bytes; | |
2354 | con->out_kvec_is_msg = false; | |
2355 | } | |
ed98adad SW |
2356 | ceph_msg_put(msg); |
2357 | msg->hdr.seq = 0; | |
31b8006e | 2358 | } |
ec302645 | 2359 | mutex_unlock(&con->mutex); |
31b8006e SW |
2360 | } |
2361 | ||
350b1c32 | 2362 | /* |
0d59ab81 | 2363 | * Revoke a message that we may be reading data into |
350b1c32 | 2364 | */ |
0d59ab81 | 2365 | void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg) |
350b1c32 SW |
2366 | { |
2367 | mutex_lock(&con->mutex); | |
0d59ab81 YS |
2368 | if (con->in_msg && con->in_msg == msg) { |
2369 | unsigned front_len = le32_to_cpu(con->in_hdr.front_len); | |
2370 | unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
350b1c32 SW |
2371 | unsigned data_len = le32_to_cpu(con->in_hdr.data_len); |
2372 | ||
2373 | /* skip rest of message */ | |
0d59ab81 | 2374 | dout("con_revoke_pages %p msg %p revoked\n", con, msg); |
350b1c32 SW |
2375 | con->in_base_pos = con->in_base_pos - |
2376 | sizeof(struct ceph_msg_header) - | |
0d59ab81 YS |
2377 | front_len - |
2378 | middle_len - | |
2379 | data_len - | |
350b1c32 | 2380 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2381 | ceph_msg_put(con->in_msg); |
2382 | con->in_msg = NULL; | |
2383 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2384 | con->in_seq++; |
350b1c32 SW |
2385 | } else { |
2386 | dout("con_revoke_pages %p msg %p pages %p no-op\n", | |
0d59ab81 | 2387 | con, con->in_msg, msg); |
350b1c32 SW |
2388 | } |
2389 | mutex_unlock(&con->mutex); | |
2390 | } | |
2391 | ||
31b8006e SW |
2392 | /* |
2393 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2394 | */ | |
2395 | void ceph_con_keepalive(struct ceph_connection *con) | |
2396 | { | |
e00de341 SW |
2397 | dout("con_keepalive %p\n", con); |
2398 | clear_standby(con); | |
31b8006e SW |
2399 | if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 && |
2400 | test_and_set_bit(WRITE_PENDING, &con->state) == 0) | |
2401 | queue_con(con); | |
2402 | } | |
3d14c5d2 | 2403 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2404 | |
2405 | ||
2406 | /* | |
2407 | * construct a new message with given type, size | |
2408 | * the new msg has a ref count of 1. | |
2409 | */ | |
b61c2763 SW |
2410 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2411 | bool can_fail) | |
31b8006e SW |
2412 | { |
2413 | struct ceph_msg *m; | |
2414 | ||
34d23762 | 2415 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2416 | if (m == NULL) |
2417 | goto out; | |
c2e552e7 | 2418 | kref_init(&m->kref); |
31b8006e SW |
2419 | INIT_LIST_HEAD(&m->list_head); |
2420 | ||
45c6ceb5 | 2421 | m->hdr.tid = 0; |
31b8006e | 2422 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2423 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2424 | m->hdr.version = 0; | |
31b8006e SW |
2425 | m->hdr.front_len = cpu_to_le32(front_len); |
2426 | m->hdr.middle_len = 0; | |
bb257664 SW |
2427 | m->hdr.data_len = 0; |
2428 | m->hdr.data_off = 0; | |
45c6ceb5 | 2429 | m->hdr.reserved = 0; |
31b8006e SW |
2430 | m->footer.front_crc = 0; |
2431 | m->footer.middle_crc = 0; | |
2432 | m->footer.data_crc = 0; | |
45c6ceb5 | 2433 | m->footer.flags = 0; |
31b8006e SW |
2434 | m->front_max = front_len; |
2435 | m->front_is_vmalloc = false; | |
2436 | m->more_to_follow = false; | |
c0d5f9db | 2437 | m->ack_stamp = 0; |
31b8006e SW |
2438 | m->pool = NULL; |
2439 | ||
ca20892d HC |
2440 | /* middle */ |
2441 | m->middle = NULL; | |
2442 | ||
2443 | /* data */ | |
2444 | m->nr_pages = 0; | |
2445 | m->page_alignment = 0; | |
2446 | m->pages = NULL; | |
2447 | m->pagelist = NULL; | |
2448 | m->bio = NULL; | |
2449 | m->bio_iter = NULL; | |
2450 | m->bio_seg = 0; | |
2451 | m->trail = NULL; | |
2452 | ||
31b8006e SW |
2453 | /* front */ |
2454 | if (front_len) { | |
2455 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2456 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2457 | PAGE_KERNEL); |
2458 | m->front_is_vmalloc = true; | |
2459 | } else { | |
34d23762 | 2460 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2461 | } |
2462 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2463 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2464 | front_len); |
2465 | goto out2; | |
2466 | } | |
2467 | } else { | |
2468 | m->front.iov_base = NULL; | |
2469 | } | |
2470 | m->front.iov_len = front_len; | |
2471 | ||
bb257664 | 2472 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2473 | return m; |
2474 | ||
2475 | out2: | |
2476 | ceph_msg_put(m); | |
2477 | out: | |
b61c2763 SW |
2478 | if (!can_fail) { |
2479 | pr_err("msg_new can't create type %d front %d\n", type, | |
2480 | front_len); | |
f0ed1b7c | 2481 | WARN_ON(1); |
b61c2763 SW |
2482 | } else { |
2483 | dout("msg_new can't create type %d front %d\n", type, | |
2484 | front_len); | |
2485 | } | |
a79832f2 | 2486 | return NULL; |
31b8006e | 2487 | } |
3d14c5d2 | 2488 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2489 | |
31b8006e SW |
2490 | /* |
2491 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2492 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2493 | * per-type header in the front and then gracefully fail (i.e., | |
2494 | * propagate the error to the caller based on info in the front) when | |
2495 | * the middle is too large. | |
2496 | */ | |
2450418c | 2497 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2498 | { |
2499 | int type = le16_to_cpu(msg->hdr.type); | |
2500 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2501 | ||
2502 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2503 | ceph_msg_type_name(type), middle_len); | |
2504 | BUG_ON(!middle_len); | |
2505 | BUG_ON(msg->middle); | |
2506 | ||
b6c1d5b8 | 2507 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2508 | if (!msg->middle) |
2509 | return -ENOMEM; | |
2510 | return 0; | |
2511 | } | |
2512 | ||
2450418c YS |
2513 | /* |
2514 | * Generic message allocator, for incoming messages. | |
2515 | */ | |
2516 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, | |
2517 | struct ceph_msg_header *hdr, | |
2518 | int *skip) | |
2519 | { | |
2520 | int type = le16_to_cpu(hdr->type); | |
2521 | int front_len = le32_to_cpu(hdr->front_len); | |
2522 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2523 | struct ceph_msg *msg = NULL; | |
2524 | int ret; | |
2525 | ||
2526 | if (con->ops->alloc_msg) { | |
0547a9b3 | 2527 | mutex_unlock(&con->mutex); |
2450418c | 2528 | msg = con->ops->alloc_msg(con, hdr, skip); |
0547a9b3 | 2529 | mutex_lock(&con->mutex); |
a79832f2 | 2530 | if (!msg || *skip) |
2450418c YS |
2531 | return NULL; |
2532 | } | |
2533 | if (!msg) { | |
2534 | *skip = 0; | |
b61c2763 | 2535 | msg = ceph_msg_new(type, front_len, GFP_NOFS, false); |
2450418c YS |
2536 | if (!msg) { |
2537 | pr_err("unable to allocate msg type %d len %d\n", | |
2538 | type, front_len); | |
a79832f2 | 2539 | return NULL; |
2450418c | 2540 | } |
c5c6b19d | 2541 | msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2542 | } |
9d7f0f13 | 2543 | memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2544 | |
bb257664 | 2545 | if (middle_len && !msg->middle) { |
2450418c | 2546 | ret = ceph_alloc_middle(con, msg); |
2450418c YS |
2547 | if (ret < 0) { |
2548 | ceph_msg_put(msg); | |
a79832f2 | 2549 | return NULL; |
2450418c YS |
2550 | } |
2551 | } | |
9d7f0f13 | 2552 | |
2450418c YS |
2553 | return msg; |
2554 | } | |
2555 | ||
31b8006e SW |
2556 | |
2557 | /* | |
2558 | * Free a generically kmalloc'd message. | |
2559 | */ | |
2560 | void ceph_msg_kfree(struct ceph_msg *m) | |
2561 | { | |
2562 | dout("msg_kfree %p\n", m); | |
2563 | if (m->front_is_vmalloc) | |
2564 | vfree(m->front.iov_base); | |
2565 | else | |
2566 | kfree(m->front.iov_base); | |
2567 | kfree(m); | |
2568 | } | |
2569 | ||
2570 | /* | |
2571 | * Drop a msg ref. Destroy as needed. | |
2572 | */ | |
c2e552e7 SW |
2573 | void ceph_msg_last_put(struct kref *kref) |
2574 | { | |
2575 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2576 | |
c2e552e7 SW |
2577 | dout("ceph_msg_put last one on %p\n", m); |
2578 | WARN_ON(!list_empty(&m->list_head)); | |
2579 | ||
2580 | /* drop middle, data, if any */ | |
2581 | if (m->middle) { | |
2582 | ceph_buffer_put(m->middle); | |
2583 | m->middle = NULL; | |
31b8006e | 2584 | } |
c2e552e7 SW |
2585 | m->nr_pages = 0; |
2586 | m->pages = NULL; | |
2587 | ||
58bb3b37 SW |
2588 | if (m->pagelist) { |
2589 | ceph_pagelist_release(m->pagelist); | |
2590 | kfree(m->pagelist); | |
2591 | m->pagelist = NULL; | |
2592 | } | |
2593 | ||
68b4476b YS |
2594 | m->trail = NULL; |
2595 | ||
c2e552e7 SW |
2596 | if (m->pool) |
2597 | ceph_msgpool_put(m->pool, m); | |
2598 | else | |
2599 | ceph_msg_kfree(m); | |
31b8006e | 2600 | } |
3d14c5d2 | 2601 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2602 | |
2603 | void ceph_msg_dump(struct ceph_msg *msg) | |
2604 | { | |
2605 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2606 | msg->front_max, msg->nr_pages); | |
2607 | print_hex_dump(KERN_DEBUG, "header: ", | |
2608 | DUMP_PREFIX_OFFSET, 16, 1, | |
2609 | &msg->hdr, sizeof(msg->hdr), true); | |
2610 | print_hex_dump(KERN_DEBUG, " front: ", | |
2611 | DUMP_PREFIX_OFFSET, 16, 1, | |
2612 | msg->front.iov_base, msg->front.iov_len, true); | |
2613 | if (msg->middle) | |
2614 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2615 | DUMP_PREFIX_OFFSET, 16, 1, | |
2616 | msg->middle->vec.iov_base, | |
2617 | msg->middle->vec.iov_len, true); | |
2618 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2619 | DUMP_PREFIX_OFFSET, 16, 1, | |
2620 | &msg->footer, sizeof(msg->footer), true); | |
2621 | } | |
3d14c5d2 | 2622 | EXPORT_SYMBOL(ceph_msg_dump); |