]>
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 | ||
ce2c8903 AE |
32 | /* State values for ceph_connection->sock_state; NEW is assumed to be 0 */ |
33 | ||
34 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ | |
35 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ | |
36 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ | |
37 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ | |
38 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ | |
39 | ||
31b8006e SW |
40 | /* static tag bytes (protocol control messages) */ |
41 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
42 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
43 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
44 | ||
a6a5349d SW |
45 | #ifdef CONFIG_LOCKDEP |
46 | static struct lock_class_key socket_class; | |
47 | #endif | |
48 | ||
84495f49 AE |
49 | /* |
50 | * When skipping (ignoring) a block of input we read it into a "skip | |
51 | * buffer," which is this many bytes in size. | |
52 | */ | |
53 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
54 | |
55 | static void queue_con(struct ceph_connection *con); | |
56 | static void con_work(struct work_struct *); | |
57 | static void ceph_fault(struct ceph_connection *con); | |
58 | ||
31b8006e | 59 | /* |
f64a9317 AE |
60 | * Nicely render a sockaddr as a string. An array of formatted |
61 | * strings is used, to approximate reentrancy. | |
31b8006e | 62 | */ |
f64a9317 AE |
63 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
64 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
65 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
66 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
67 | ||
68 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
69 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 70 | |
57666519 | 71 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 72 | |
3d14c5d2 | 73 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
74 | { |
75 | int i; | |
76 | char *s; | |
99f0f3b2 AE |
77 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
78 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 79 | |
f64a9317 | 80 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
81 | s = addr_str[i]; |
82 | ||
83 | switch (ss->ss_family) { | |
84 | case AF_INET: | |
bd406145 AE |
85 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
86 | ntohs(in4->sin_port)); | |
31b8006e SW |
87 | break; |
88 | ||
89 | case AF_INET6: | |
bd406145 AE |
90 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
91 | ntohs(in6->sin6_port)); | |
31b8006e SW |
92 | break; |
93 | ||
94 | default: | |
d3002b97 AE |
95 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
96 | ss->ss_family); | |
31b8006e SW |
97 | } |
98 | ||
99 | return s; | |
100 | } | |
3d14c5d2 | 101 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 102 | |
63f2d211 SW |
103 | static void encode_my_addr(struct ceph_messenger *msgr) |
104 | { | |
105 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
106 | ceph_encode_addr(&msgr->my_enc_addr); | |
107 | } | |
108 | ||
31b8006e SW |
109 | /* |
110 | * work queue for all reading and writing to/from the socket. | |
111 | */ | |
e0f43c94 | 112 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 113 | |
6173d1f0 AE |
114 | void _ceph_msgr_exit(void) |
115 | { | |
d3002b97 | 116 | if (ceph_msgr_wq) { |
6173d1f0 | 117 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
118 | ceph_msgr_wq = NULL; |
119 | } | |
6173d1f0 | 120 | |
6173d1f0 AE |
121 | BUG_ON(zero_page == NULL); |
122 | kunmap(zero_page); | |
123 | page_cache_release(zero_page); | |
124 | zero_page = NULL; | |
125 | } | |
126 | ||
3d14c5d2 | 127 | int ceph_msgr_init(void) |
31b8006e | 128 | { |
57666519 AE |
129 | BUG_ON(zero_page != NULL); |
130 | zero_page = ZERO_PAGE(0); | |
131 | page_cache_get(zero_page); | |
132 | ||
f363e45f | 133 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0); |
6173d1f0 AE |
134 | if (ceph_msgr_wq) |
135 | return 0; | |
57666519 | 136 | |
6173d1f0 AE |
137 | pr_err("msgr_init failed to create workqueue\n"); |
138 | _ceph_msgr_exit(); | |
57666519 | 139 | |
6173d1f0 | 140 | return -ENOMEM; |
31b8006e | 141 | } |
3d14c5d2 | 142 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
143 | |
144 | void ceph_msgr_exit(void) | |
145 | { | |
57666519 | 146 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 147 | |
6173d1f0 | 148 | _ceph_msgr_exit(); |
31b8006e | 149 | } |
3d14c5d2 | 150 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 151 | |
cd84db6e | 152 | void ceph_msgr_flush(void) |
a922d38f SW |
153 | { |
154 | flush_workqueue(ceph_msgr_wq); | |
155 | } | |
3d14c5d2 | 156 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 157 | |
ce2c8903 AE |
158 | /* Connection socket state transition functions */ |
159 | ||
160 | static void con_sock_state_init(struct ceph_connection *con) | |
161 | { | |
162 | int old_state; | |
163 | ||
164 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
165 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
166 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
167 | } | |
168 | ||
169 | static void con_sock_state_connecting(struct ceph_connection *con) | |
170 | { | |
171 | int old_state; | |
172 | ||
173 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
174 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
175 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
176 | } | |
177 | ||
178 | static void con_sock_state_connected(struct ceph_connection *con) | |
179 | { | |
180 | int old_state; | |
181 | ||
182 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
183 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
184 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
185 | } | |
186 | ||
187 | static void con_sock_state_closing(struct ceph_connection *con) | |
188 | { | |
189 | int old_state; | |
190 | ||
191 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
192 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
193 | old_state != CON_SOCK_STATE_CONNECTED && | |
194 | old_state != CON_SOCK_STATE_CLOSING)) | |
195 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
196 | } | |
197 | ||
198 | static void con_sock_state_closed(struct ceph_connection *con) | |
199 | { | |
200 | int old_state; | |
201 | ||
202 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
203 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
204 | old_state != CON_SOCK_STATE_CLOSING)) | |
205 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
206 | } | |
a922d38f | 207 | |
31b8006e SW |
208 | /* |
209 | * socket callback functions | |
210 | */ | |
211 | ||
212 | /* data available on socket, or listen socket received a connect */ | |
327800bd | 213 | static void ceph_sock_data_ready(struct sock *sk, int count_unused) |
31b8006e | 214 | { |
bd406145 AE |
215 | struct ceph_connection *con = sk->sk_user_data; |
216 | ||
31b8006e | 217 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 218 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
219 | con, con->state); |
220 | queue_con(con); | |
221 | } | |
222 | } | |
223 | ||
224 | /* socket has buffer space for writing */ | |
327800bd | 225 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 226 | { |
d3002b97 | 227 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 228 | |
182fac26 JS |
229 | /* only queue to workqueue if there is data we want to write, |
230 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 231 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
232 | * doesn't get called again until try_write() fills the socket |
233 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
234 | * and net/core/stream.c:sk_stream_write_space(). | |
235 | */ | |
928443cd | 236 | if (test_bit(WRITE_PENDING, &con->flags)) { |
182fac26 | 237 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
327800bd | 238 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
239 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
240 | queue_con(con); | |
241 | } | |
31b8006e | 242 | } else { |
327800bd | 243 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 244 | } |
31b8006e SW |
245 | } |
246 | ||
247 | /* socket's state has changed */ | |
327800bd | 248 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 249 | { |
bd406145 | 250 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 251 | |
327800bd | 252 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
253 | con, con->state, sk->sk_state); |
254 | ||
255 | if (test_bit(CLOSED, &con->state)) | |
256 | return; | |
257 | ||
258 | switch (sk->sk_state) { | |
259 | case TCP_CLOSE: | |
327800bd | 260 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 261 | case TCP_CLOSE_WAIT: |
327800bd | 262 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 263 | con_sock_state_closing(con); |
928443cd | 264 | if (test_and_set_bit(SOCK_CLOSED, &con->flags) == 0) { |
31b8006e SW |
265 | if (test_bit(CONNECTING, &con->state)) |
266 | con->error_msg = "connection failed"; | |
267 | else | |
268 | con->error_msg = "socket closed"; | |
269 | queue_con(con); | |
270 | } | |
271 | break; | |
272 | case TCP_ESTABLISHED: | |
327800bd | 273 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 274 | con_sock_state_connected(con); |
31b8006e SW |
275 | queue_con(con); |
276 | break; | |
d3002b97 AE |
277 | default: /* Everything else is uninteresting */ |
278 | break; | |
31b8006e SW |
279 | } |
280 | } | |
281 | ||
282 | /* | |
283 | * set up socket callbacks | |
284 | */ | |
285 | static void set_sock_callbacks(struct socket *sock, | |
286 | struct ceph_connection *con) | |
287 | { | |
288 | struct sock *sk = sock->sk; | |
bd406145 | 289 | sk->sk_user_data = con; |
327800bd AE |
290 | sk->sk_data_ready = ceph_sock_data_ready; |
291 | sk->sk_write_space = ceph_sock_write_space; | |
292 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
293 | } |
294 | ||
295 | ||
296 | /* | |
297 | * socket helpers | |
298 | */ | |
299 | ||
300 | /* | |
301 | * initiate connection to a remote socket. | |
302 | */ | |
41617d0c | 303 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 304 | { |
f91d3471 | 305 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
306 | struct socket *sock; |
307 | int ret; | |
308 | ||
309 | BUG_ON(con->sock); | |
f91d3471 SW |
310 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
311 | IPPROTO_TCP, &sock); | |
31b8006e | 312 | if (ret) |
41617d0c | 313 | return ret; |
31b8006e SW |
314 | sock->sk->sk_allocation = GFP_NOFS; |
315 | ||
a6a5349d SW |
316 | #ifdef CONFIG_LOCKDEP |
317 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
318 | #endif | |
319 | ||
31b8006e SW |
320 | set_sock_callbacks(sock, con); |
321 | ||
3d14c5d2 | 322 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 323 | |
f91d3471 SW |
324 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
325 | O_NONBLOCK); | |
31b8006e SW |
326 | if (ret == -EINPROGRESS) { |
327 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 328 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 329 | sock->sk->sk_state); |
a5bc3129 | 330 | } else if (ret < 0) { |
31b8006e | 331 | pr_err("connect %s error %d\n", |
3d14c5d2 | 332 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 333 | sock_release(sock); |
31b8006e | 334 | con->error_msg = "connect error"; |
31b8006e | 335 | |
41617d0c | 336 | return ret; |
a5bc3129 AE |
337 | } |
338 | con->sock = sock; | |
ce2c8903 | 339 | con_sock_state_connecting(con); |
a5bc3129 | 340 | |
41617d0c | 341 | return 0; |
31b8006e SW |
342 | } |
343 | ||
344 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
345 | { | |
346 | struct kvec iov = {buf, len}; | |
347 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 348 | int r; |
31b8006e | 349 | |
98bdb0aa SW |
350 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
351 | if (r == -EAGAIN) | |
352 | r = 0; | |
353 | return r; | |
31b8006e SW |
354 | } |
355 | ||
356 | /* | |
357 | * write something. @more is true if caller will be sending more data | |
358 | * shortly. | |
359 | */ | |
360 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
361 | size_t kvlen, size_t len, int more) | |
362 | { | |
363 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 364 | int r; |
31b8006e SW |
365 | |
366 | if (more) | |
367 | msg.msg_flags |= MSG_MORE; | |
368 | else | |
369 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
370 | ||
42961d23 SW |
371 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
372 | if (r == -EAGAIN) | |
373 | r = 0; | |
374 | return r; | |
31b8006e SW |
375 | } |
376 | ||
31739139 AE |
377 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
378 | int offset, size_t size, int more) | |
379 | { | |
380 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
381 | int ret; | |
382 | ||
383 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
384 | if (ret == -EAGAIN) | |
385 | ret = 0; | |
386 | ||
387 | return ret; | |
388 | } | |
389 | ||
31b8006e SW |
390 | |
391 | /* | |
392 | * Shutdown/close the socket for the given connection. | |
393 | */ | |
394 | static int con_close_socket(struct ceph_connection *con) | |
395 | { | |
396 | int rc; | |
397 | ||
398 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
399 | if (!con->sock) | |
400 | return 0; | |
401 | set_bit(SOCK_CLOSED, &con->state); | |
402 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
403 | sock_release(con->sock); | |
404 | con->sock = NULL; | |
405 | clear_bit(SOCK_CLOSED, &con->state); | |
ce2c8903 | 406 | con_sock_state_closed(con); |
31b8006e SW |
407 | return rc; |
408 | } | |
409 | ||
410 | /* | |
411 | * Reset a connection. Discard all incoming and outgoing messages | |
412 | * and clear *_seq state. | |
413 | */ | |
414 | static void ceph_msg_remove(struct ceph_msg *msg) | |
415 | { | |
416 | list_del_init(&msg->list_head); | |
417 | ceph_msg_put(msg); | |
418 | } | |
419 | static void ceph_msg_remove_list(struct list_head *head) | |
420 | { | |
421 | while (!list_empty(head)) { | |
422 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
423 | list_head); | |
424 | ceph_msg_remove(msg); | |
425 | } | |
426 | } | |
427 | ||
428 | static void reset_connection(struct ceph_connection *con) | |
429 | { | |
430 | /* reset connection, out_queue, msg_ and connect_seq */ | |
431 | /* discard existing out_queue and msg_seq */ | |
31b8006e SW |
432 | ceph_msg_remove_list(&con->out_queue); |
433 | ceph_msg_remove_list(&con->out_sent); | |
434 | ||
cf3e5c40 SW |
435 | if (con->in_msg) { |
436 | ceph_msg_put(con->in_msg); | |
437 | con->in_msg = NULL; | |
438 | } | |
439 | ||
31b8006e SW |
440 | con->connect_seq = 0; |
441 | con->out_seq = 0; | |
c86a2930 SW |
442 | if (con->out_msg) { |
443 | ceph_msg_put(con->out_msg); | |
444 | con->out_msg = NULL; | |
445 | } | |
31b8006e | 446 | con->in_seq = 0; |
0e0d5e0c | 447 | con->in_seq_acked = 0; |
31b8006e SW |
448 | } |
449 | ||
450 | /* | |
451 | * mark a peer down. drop any open connections. | |
452 | */ | |
453 | void ceph_con_close(struct ceph_connection *con) | |
454 | { | |
3d14c5d2 YS |
455 | dout("con_close %p peer %s\n", con, |
456 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
a5988c49 | 457 | clear_bit(NEGOTIATING, &con->state); |
31b8006e | 458 | clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */ |
a5988c49 AE |
459 | set_bit(CLOSED, &con->state); |
460 | ||
928443cd AE |
461 | clear_bit(LOSSYTX, &con->flags); /* so we retry next connect */ |
462 | clear_bit(KEEPALIVE_PENDING, &con->flags); | |
463 | clear_bit(WRITE_PENDING, &con->flags); | |
a5988c49 | 464 | |
ec302645 | 465 | mutex_lock(&con->mutex); |
31b8006e | 466 | reset_connection(con); |
6f2bc3ff | 467 | con->peer_global_seq = 0; |
91e45ce3 | 468 | cancel_delayed_work(&con->work); |
ec302645 | 469 | mutex_unlock(&con->mutex); |
31b8006e SW |
470 | queue_con(con); |
471 | } | |
3d14c5d2 | 472 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 473 | |
31b8006e SW |
474 | /* |
475 | * Reopen a closed connection, with a new peer address. | |
476 | */ | |
477 | void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr) | |
478 | { | |
3d14c5d2 | 479 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
31b8006e | 480 | set_bit(OPENING, &con->state); |
a5988c49 AE |
481 | WARN_ON(!test_and_clear_bit(CLOSED, &con->state)); |
482 | ||
31b8006e | 483 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 484 | con->delay = 0; /* reset backoff memory */ |
31b8006e SW |
485 | queue_con(con); |
486 | } | |
3d14c5d2 | 487 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 488 | |
87b315a5 SW |
489 | /* |
490 | * return true if this connection ever successfully opened | |
491 | */ | |
492 | bool ceph_con_opened(struct ceph_connection *con) | |
493 | { | |
494 | return con->connect_seq > 0; | |
495 | } | |
496 | ||
31b8006e SW |
497 | /* |
498 | * generic get/put | |
499 | */ | |
500 | struct ceph_connection *ceph_con_get(struct ceph_connection *con) | |
501 | { | |
d3002b97 AE |
502 | int nref = __atomic_add_unless(&con->nref, 1, 0); |
503 | ||
504 | dout("con_get %p nref = %d -> %d\n", con, nref, nref + 1); | |
505 | ||
506 | return nref ? con : NULL; | |
31b8006e SW |
507 | } |
508 | ||
509 | void ceph_con_put(struct ceph_connection *con) | |
510 | { | |
d3002b97 AE |
511 | int nref = atomic_dec_return(&con->nref); |
512 | ||
513 | BUG_ON(nref < 0); | |
514 | if (nref == 0) { | |
71ececda | 515 | BUG_ON(con->sock); |
31b8006e SW |
516 | kfree(con); |
517 | } | |
d3002b97 | 518 | dout("con_put %p nref = %d -> %d\n", con, nref + 1, nref); |
31b8006e SW |
519 | } |
520 | ||
521 | /* | |
522 | * initialize a new connection. | |
523 | */ | |
524 | void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con) | |
525 | { | |
526 | dout("con_init %p\n", con); | |
527 | memset(con, 0, sizeof(*con)); | |
528 | atomic_set(&con->nref, 1); | |
529 | con->msgr = msgr; | |
ce2c8903 AE |
530 | |
531 | con_sock_state_init(con); | |
532 | ||
ec302645 | 533 | mutex_init(&con->mutex); |
31b8006e SW |
534 | INIT_LIST_HEAD(&con->out_queue); |
535 | INIT_LIST_HEAD(&con->out_sent); | |
536 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 AE |
537 | |
538 | set_bit(CLOSED, &con->state); | |
31b8006e | 539 | } |
3d14c5d2 | 540 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
541 | |
542 | ||
543 | /* | |
544 | * We maintain a global counter to order connection attempts. Get | |
545 | * a unique seq greater than @gt. | |
546 | */ | |
547 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
548 | { | |
549 | u32 ret; | |
550 | ||
551 | spin_lock(&msgr->global_seq_lock); | |
552 | if (msgr->global_seq < gt) | |
553 | msgr->global_seq = gt; | |
554 | ret = ++msgr->global_seq; | |
555 | spin_unlock(&msgr->global_seq_lock); | |
556 | return ret; | |
557 | } | |
558 | ||
e2200423 | 559 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
560 | { |
561 | con->out_kvec_left = 0; | |
562 | con->out_kvec_bytes = 0; | |
563 | con->out_kvec_cur = &con->out_kvec[0]; | |
564 | } | |
565 | ||
e2200423 | 566 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
567 | size_t size, void *data) |
568 | { | |
569 | int index; | |
570 | ||
571 | index = con->out_kvec_left; | |
572 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
573 | ||
574 | con->out_kvec[index].iov_len = size; | |
575 | con->out_kvec[index].iov_base = data; | |
576 | con->out_kvec_left++; | |
577 | con->out_kvec_bytes += size; | |
578 | } | |
31b8006e SW |
579 | |
580 | /* | |
581 | * Prepare footer for currently outgoing message, and finish things | |
582 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
583 | */ | |
859eb799 | 584 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
585 | { |
586 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 587 | int v = con->out_kvec_left; |
31b8006e SW |
588 | |
589 | dout("prepare_write_message_footer %p\n", con); | |
590 | con->out_kvec_is_msg = true; | |
591 | con->out_kvec[v].iov_base = &m->footer; | |
592 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
593 | con->out_kvec_bytes += sizeof(m->footer); | |
594 | con->out_kvec_left++; | |
595 | con->out_more = m->more_to_follow; | |
c86a2930 | 596 | con->out_msg_done = true; |
31b8006e SW |
597 | } |
598 | ||
599 | /* | |
600 | * Prepare headers for the next outgoing message. | |
601 | */ | |
602 | static void prepare_write_message(struct ceph_connection *con) | |
603 | { | |
604 | struct ceph_msg *m; | |
a9a0c51a | 605 | u32 crc; |
31b8006e | 606 | |
e2200423 | 607 | con_out_kvec_reset(con); |
31b8006e | 608 | con->out_kvec_is_msg = true; |
c86a2930 | 609 | con->out_msg_done = false; |
31b8006e SW |
610 | |
611 | /* Sneak an ack in there first? If we can get it into the same | |
612 | * TCP packet that's a good thing. */ | |
613 | if (con->in_seq > con->in_seq_acked) { | |
614 | con->in_seq_acked = con->in_seq; | |
e2200423 | 615 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 616 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 617 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 618 | &con->out_temp_ack); |
31b8006e SW |
619 | } |
620 | ||
859eb799 | 621 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 622 | con->out_msg = m; |
4cf9d544 SW |
623 | |
624 | /* put message on sent list */ | |
625 | ceph_msg_get(m); | |
626 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 627 | |
e84346b7 SW |
628 | /* |
629 | * only assign outgoing seq # if we haven't sent this message | |
630 | * yet. if it is requeued, resend with it's original seq. | |
631 | */ | |
632 | if (m->needs_out_seq) { | |
633 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
634 | m->needs_out_seq = false; | |
635 | } | |
31b8006e SW |
636 | |
637 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
638 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
639 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
640 | le32_to_cpu(m->hdr.data_len), | |
641 | m->nr_pages); | |
642 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); | |
643 | ||
644 | /* tag + hdr + front + middle */ | |
e2200423 AE |
645 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
646 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
647 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 648 | |
31b8006e | 649 | if (m->middle) |
e2200423 | 650 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 651 | m->middle->vec.iov_base); |
31b8006e SW |
652 | |
653 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
654 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
655 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
31b8006e | 656 | con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE; |
a9a0c51a AE |
657 | |
658 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
659 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
660 | if (m->middle) { | |
661 | crc = crc32c(0, m->middle->vec.iov_base, | |
662 | m->middle->vec.iov_len); | |
663 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
664 | } else | |
31b8006e SW |
665 | con->out_msg->footer.middle_crc = 0; |
666 | con->out_msg->footer.data_crc = 0; | |
667 | dout("prepare_write_message front_crc %u data_crc %u\n", | |
668 | le32_to_cpu(con->out_msg->footer.front_crc), | |
669 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
670 | ||
671 | /* is there a data payload? */ | |
672 | if (le32_to_cpu(m->hdr.data_len) > 0) { | |
673 | /* initialize page iterator */ | |
674 | con->out_msg_pos.page = 0; | |
68b4476b | 675 | if (m->pages) |
c5c6b19d | 676 | con->out_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
677 | else |
678 | con->out_msg_pos.page_pos = 0; | |
31b8006e | 679 | con->out_msg_pos.data_pos = 0; |
bca064d2 | 680 | con->out_msg_pos.did_page_crc = false; |
31b8006e SW |
681 | con->out_more = 1; /* data + footer will follow */ |
682 | } else { | |
683 | /* no, queue up footer too and be done */ | |
859eb799 | 684 | prepare_write_message_footer(con); |
31b8006e SW |
685 | } |
686 | ||
928443cd | 687 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
688 | } |
689 | ||
690 | /* | |
691 | * Prepare an ack. | |
692 | */ | |
693 | static void prepare_write_ack(struct ceph_connection *con) | |
694 | { | |
695 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
696 | con->in_seq_acked, con->in_seq); | |
697 | con->in_seq_acked = con->in_seq; | |
698 | ||
e2200423 | 699 | con_out_kvec_reset(con); |
859eb799 | 700 | |
e2200423 | 701 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 702 | |
31b8006e | 703 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 704 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
705 | &con->out_temp_ack); |
706 | ||
31b8006e | 707 | con->out_more = 1; /* more will follow.. eventually.. */ |
928443cd | 708 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
709 | } |
710 | ||
711 | /* | |
712 | * Prepare to write keepalive byte. | |
713 | */ | |
714 | static void prepare_write_keepalive(struct ceph_connection *con) | |
715 | { | |
716 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
717 | con_out_kvec_reset(con); |
718 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
928443cd | 719 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
720 | } |
721 | ||
722 | /* | |
723 | * Connection negotiation. | |
724 | */ | |
725 | ||
dac1e716 AE |
726 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
727 | int *auth_proto) | |
4e7a5dcd | 728 | { |
a3530df3 | 729 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
730 | |
731 | if (!con->ops->get_authorizer) { | |
732 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
733 | con->out_connect.authorizer_len = 0; | |
734 | ||
729796be | 735 | return NULL; |
b1c6b980 AE |
736 | } |
737 | ||
738 | /* Can't hold the mutex while getting authorizer */ | |
4e7a5dcd | 739 | |
ec302645 | 740 | mutex_unlock(&con->mutex); |
b1c6b980 | 741 | |
dac1e716 | 742 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
8f43fb53 | 743 | |
ec302645 | 744 | mutex_lock(&con->mutex); |
4e7a5dcd | 745 | |
a3530df3 | 746 | if (IS_ERR(auth)) |
729796be | 747 | return auth; |
928443cd | 748 | if (test_bit(CLOSED, &con->state) || test_bit(OPENING, &con->flags)) |
729796be | 749 | return ERR_PTR(-EAGAIN); |
0da5d703 | 750 | |
8f43fb53 AE |
751 | con->auth_reply_buf = auth->authorizer_reply_buf; |
752 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
753 | ||
859eb799 | 754 | |
729796be | 755 | return auth; |
4e7a5dcd SW |
756 | } |
757 | ||
31b8006e SW |
758 | /* |
759 | * We connected to a peer and are saying hello. | |
760 | */ | |
e825a66d | 761 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 762 | { |
e2200423 AE |
763 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
764 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 765 | &con->msgr->my_enc_addr); |
eed0ef2c | 766 | |
eed0ef2c | 767 | con->out_more = 0; |
928443cd | 768 | set_bit(WRITE_PENDING, &con->flags); |
eed0ef2c SW |
769 | } |
770 | ||
e825a66d | 771 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 772 | { |
31b8006e SW |
773 | unsigned global_seq = get_global_seq(con->msgr, 0); |
774 | int proto; | |
dac1e716 | 775 | int auth_proto; |
729796be | 776 | struct ceph_auth_handshake *auth; |
31b8006e SW |
777 | |
778 | switch (con->peer_name.type) { | |
779 | case CEPH_ENTITY_TYPE_MON: | |
780 | proto = CEPH_MONC_PROTOCOL; | |
781 | break; | |
782 | case CEPH_ENTITY_TYPE_OSD: | |
783 | proto = CEPH_OSDC_PROTOCOL; | |
784 | break; | |
785 | case CEPH_ENTITY_TYPE_MDS: | |
786 | proto = CEPH_MDSC_PROTOCOL; | |
787 | break; | |
788 | default: | |
789 | BUG(); | |
790 | } | |
791 | ||
792 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
793 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 794 | |
e825a66d | 795 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
796 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
797 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
798 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
799 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
800 | con->out_connect.flags = 0; | |
31b8006e | 801 | |
dac1e716 AE |
802 | auth_proto = CEPH_AUTH_UNKNOWN; |
803 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
804 | if (IS_ERR(auth)) |
805 | return PTR_ERR(auth); | |
3da54776 | 806 | |
dac1e716 | 807 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
808 | con->out_connect.authorizer_len = auth ? |
809 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
810 | ||
e2200423 | 811 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
812 | &con->out_connect); |
813 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 814 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 815 | auth->authorizer_buf); |
859eb799 | 816 | |
31b8006e | 817 | con->out_more = 0; |
928443cd | 818 | set_bit(WRITE_PENDING, &con->flags); |
4e7a5dcd | 819 | |
e10c758e | 820 | return 0; |
31b8006e SW |
821 | } |
822 | ||
31b8006e SW |
823 | /* |
824 | * write as much of pending kvecs to the socket as we can. | |
825 | * 1 -> done | |
826 | * 0 -> socket full, but more to do | |
827 | * <0 -> error | |
828 | */ | |
829 | static int write_partial_kvec(struct ceph_connection *con) | |
830 | { | |
831 | int ret; | |
832 | ||
833 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
834 | while (con->out_kvec_bytes > 0) { | |
835 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
836 | con->out_kvec_left, con->out_kvec_bytes, | |
837 | con->out_more); | |
838 | if (ret <= 0) | |
839 | goto out; | |
840 | con->out_kvec_bytes -= ret; | |
841 | if (con->out_kvec_bytes == 0) | |
842 | break; /* done */ | |
f42299e6 AE |
843 | |
844 | /* account for full iov entries consumed */ | |
845 | while (ret >= con->out_kvec_cur->iov_len) { | |
846 | BUG_ON(!con->out_kvec_left); | |
847 | ret -= con->out_kvec_cur->iov_len; | |
848 | con->out_kvec_cur++; | |
849 | con->out_kvec_left--; | |
850 | } | |
851 | /* and for a partially-consumed entry */ | |
852 | if (ret) { | |
853 | con->out_kvec_cur->iov_len -= ret; | |
854 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
855 | } |
856 | } | |
857 | con->out_kvec_left = 0; | |
858 | con->out_kvec_is_msg = false; | |
859 | ret = 1; | |
860 | out: | |
861 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
862 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
863 | return ret; /* done! */ | |
864 | } | |
865 | ||
68b4476b YS |
866 | #ifdef CONFIG_BLOCK |
867 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
868 | { | |
869 | if (!bio) { | |
870 | *iter = NULL; | |
871 | *seg = 0; | |
872 | return; | |
873 | } | |
874 | *iter = bio; | |
875 | *seg = bio->bi_idx; | |
876 | } | |
877 | ||
878 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
879 | { | |
880 | if (*bio_iter == NULL) | |
881 | return; | |
882 | ||
883 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
884 | ||
885 | (*seg)++; | |
886 | if (*seg == (*bio_iter)->bi_vcnt) | |
887 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
888 | } | |
889 | #endif | |
890 | ||
31b8006e SW |
891 | /* |
892 | * Write as much message data payload as we can. If we finish, queue | |
893 | * up the footer. | |
894 | * 1 -> done, footer is now queued in out_kvec[]. | |
895 | * 0 -> socket full, but more to do | |
896 | * <0 -> error | |
897 | */ | |
898 | static int write_partial_msg_pages(struct ceph_connection *con) | |
899 | { | |
900 | struct ceph_msg *msg = con->out_msg; | |
901 | unsigned data_len = le32_to_cpu(msg->hdr.data_len); | |
902 | size_t len; | |
37675b0f | 903 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 904 | int ret; |
68b4476b YS |
905 | int total_max_write; |
906 | int in_trail = 0; | |
907 | size_t trail_len = (msg->trail ? msg->trail->length : 0); | |
31b8006e SW |
908 | |
909 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
910 | con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages, | |
911 | con->out_msg_pos.page_pos); | |
912 | ||
68b4476b YS |
913 | #ifdef CONFIG_BLOCK |
914 | if (msg->bio && !msg->bio_iter) | |
915 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); | |
916 | #endif | |
917 | ||
918 | while (data_len > con->out_msg_pos.data_pos) { | |
31b8006e | 919 | struct page *page = NULL; |
68b4476b | 920 | int max_write = PAGE_SIZE; |
9bd19663 | 921 | int bio_offset = 0; |
68b4476b YS |
922 | |
923 | total_max_write = data_len - trail_len - | |
924 | con->out_msg_pos.data_pos; | |
31b8006e SW |
925 | |
926 | /* | |
927 | * if we are calculating the data crc (the default), we need | |
928 | * to map the page. if our pages[] has been revoked, use the | |
929 | * zero page. | |
930 | */ | |
68b4476b YS |
931 | |
932 | /* have we reached the trail part of the data? */ | |
933 | if (con->out_msg_pos.data_pos >= data_len - trail_len) { | |
934 | in_trail = 1; | |
935 | ||
936 | total_max_write = data_len - con->out_msg_pos.data_pos; | |
937 | ||
938 | page = list_first_entry(&msg->trail->head, | |
939 | struct page, lru); | |
68b4476b YS |
940 | max_write = PAGE_SIZE; |
941 | } else if (msg->pages) { | |
31b8006e | 942 | page = msg->pages[con->out_msg_pos.page]; |
58bb3b37 SW |
943 | } else if (msg->pagelist) { |
944 | page = list_first_entry(&msg->pagelist->head, | |
945 | struct page, lru); | |
68b4476b YS |
946 | #ifdef CONFIG_BLOCK |
947 | } else if (msg->bio) { | |
948 | struct bio_vec *bv; | |
949 | ||
950 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
951 | page = bv->bv_page; | |
9bd19663 | 952 | bio_offset = bv->bv_offset; |
68b4476b YS |
953 | max_write = bv->bv_len; |
954 | #endif | |
31b8006e | 955 | } else { |
57666519 | 956 | page = zero_page; |
31b8006e | 957 | } |
68b4476b YS |
958 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
959 | total_max_write); | |
960 | ||
37675b0f | 961 | if (do_datacrc && !con->out_msg_pos.did_page_crc) { |
9bd19663 | 962 | void *base; |
a9a0c51a | 963 | u32 crc; |
31b8006e | 964 | u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc); |
8d63e318 | 965 | char *kaddr; |
31b8006e | 966 | |
8d63e318 | 967 | kaddr = kmap(page); |
31b8006e | 968 | BUG_ON(kaddr == NULL); |
9bd19663 | 969 | base = kaddr + con->out_msg_pos.page_pos + bio_offset; |
a9a0c51a AE |
970 | crc = crc32c(tmpcrc, base, len); |
971 | con->out_msg->footer.data_crc = cpu_to_le32(crc); | |
bca064d2 | 972 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 973 | } |
e36b13cc | 974 | ret = ceph_tcp_sendpage(con->sock, page, |
9bd19663 | 975 | con->out_msg_pos.page_pos + bio_offset, |
e36b13cc | 976 | len, 1); |
31b8006e | 977 | |
0cdf9e60 | 978 | if (do_datacrc) |
31b8006e SW |
979 | kunmap(page); |
980 | ||
981 | if (ret <= 0) | |
982 | goto out; | |
983 | ||
984 | con->out_msg_pos.data_pos += ret; | |
985 | con->out_msg_pos.page_pos += ret; | |
986 | if (ret == len) { | |
987 | con->out_msg_pos.page_pos = 0; | |
988 | con->out_msg_pos.page++; | |
bca064d2 | 989 | con->out_msg_pos.did_page_crc = false; |
68b4476b YS |
990 | if (in_trail) |
991 | list_move_tail(&page->lru, | |
992 | &msg->trail->head); | |
993 | else if (msg->pagelist) | |
58bb3b37 SW |
994 | list_move_tail(&page->lru, |
995 | &msg->pagelist->head); | |
68b4476b YS |
996 | #ifdef CONFIG_BLOCK |
997 | else if (msg->bio) | |
998 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
999 | #endif | |
31b8006e SW |
1000 | } |
1001 | } | |
1002 | ||
1003 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
1004 | ||
1005 | /* prepare and queue up footer, too */ | |
37675b0f | 1006 | if (!do_datacrc) |
31b8006e | 1007 | con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1008 | con_out_kvec_reset(con); |
859eb799 | 1009 | prepare_write_message_footer(con); |
31b8006e SW |
1010 | ret = 1; |
1011 | out: | |
1012 | return ret; | |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * write some zeros | |
1017 | */ | |
1018 | static int write_partial_skip(struct ceph_connection *con) | |
1019 | { | |
1020 | int ret; | |
1021 | ||
1022 | while (con->out_skip > 0) { | |
31739139 | 1023 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1024 | |
31739139 | 1025 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1); |
31b8006e SW |
1026 | if (ret <= 0) |
1027 | goto out; | |
1028 | con->out_skip -= ret; | |
1029 | } | |
1030 | ret = 1; | |
1031 | out: | |
1032 | return ret; | |
1033 | } | |
1034 | ||
1035 | /* | |
1036 | * Prepare to read connection handshake, or an ack. | |
1037 | */ | |
eed0ef2c SW |
1038 | static void prepare_read_banner(struct ceph_connection *con) |
1039 | { | |
1040 | dout("prepare_read_banner %p\n", con); | |
1041 | con->in_base_pos = 0; | |
1042 | } | |
1043 | ||
31b8006e SW |
1044 | static void prepare_read_connect(struct ceph_connection *con) |
1045 | { | |
1046 | dout("prepare_read_connect %p\n", con); | |
1047 | con->in_base_pos = 0; | |
1048 | } | |
1049 | ||
1050 | static void prepare_read_ack(struct ceph_connection *con) | |
1051 | { | |
1052 | dout("prepare_read_ack %p\n", con); | |
1053 | con->in_base_pos = 0; | |
1054 | } | |
1055 | ||
1056 | static void prepare_read_tag(struct ceph_connection *con) | |
1057 | { | |
1058 | dout("prepare_read_tag %p\n", con); | |
1059 | con->in_base_pos = 0; | |
1060 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1061 | } | |
1062 | ||
1063 | /* | |
1064 | * Prepare to read a message. | |
1065 | */ | |
1066 | static int prepare_read_message(struct ceph_connection *con) | |
1067 | { | |
1068 | dout("prepare_read_message %p\n", con); | |
1069 | BUG_ON(con->in_msg != NULL); | |
1070 | con->in_base_pos = 0; | |
1071 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | ||
1076 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1077 | int end, int size, void *object) |
31b8006e | 1078 | { |
e6cee71f AE |
1079 | while (con->in_base_pos < end) { |
1080 | int left = end - con->in_base_pos; | |
31b8006e SW |
1081 | int have = size - left; |
1082 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1083 | if (ret <= 0) | |
1084 | return ret; | |
1085 | con->in_base_pos += ret; | |
1086 | } | |
1087 | return 1; | |
1088 | } | |
1089 | ||
1090 | ||
1091 | /* | |
1092 | * Read all or part of the connect-side handshake on a new connection | |
1093 | */ | |
eed0ef2c | 1094 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1095 | { |
fd51653f AE |
1096 | int size; |
1097 | int end; | |
1098 | int ret; | |
31b8006e | 1099 | |
eed0ef2c | 1100 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1101 | |
1102 | /* peer's banner */ | |
fd51653f AE |
1103 | size = strlen(CEPH_BANNER); |
1104 | end = size; | |
1105 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1106 | if (ret <= 0) |
1107 | goto out; | |
fd51653f AE |
1108 | |
1109 | size = sizeof (con->actual_peer_addr); | |
1110 | end += size; | |
1111 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1112 | if (ret <= 0) |
1113 | goto out; | |
fd51653f AE |
1114 | |
1115 | size = sizeof (con->peer_addr_for_me); | |
1116 | end += size; | |
1117 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1118 | if (ret <= 0) |
1119 | goto out; | |
fd51653f | 1120 | |
eed0ef2c SW |
1121 | out: |
1122 | return ret; | |
1123 | } | |
1124 | ||
1125 | static int read_partial_connect(struct ceph_connection *con) | |
1126 | { | |
fd51653f AE |
1127 | int size; |
1128 | int end; | |
1129 | int ret; | |
eed0ef2c SW |
1130 | |
1131 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1132 | ||
fd51653f AE |
1133 | size = sizeof (con->in_reply); |
1134 | end = size; | |
1135 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1136 | if (ret <= 0) |
1137 | goto out; | |
fd51653f AE |
1138 | |
1139 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1140 | end += size; | |
1141 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1142 | if (ret <= 0) |
1143 | goto out; | |
31b8006e | 1144 | |
4e7a5dcd SW |
1145 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1146 | con, (int)con->in_reply.tag, | |
1147 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1148 | le32_to_cpu(con->in_reply.global_seq)); |
1149 | out: | |
1150 | return ret; | |
eed0ef2c | 1151 | |
31b8006e SW |
1152 | } |
1153 | ||
1154 | /* | |
1155 | * Verify the hello banner looks okay. | |
1156 | */ | |
1157 | static int verify_hello(struct ceph_connection *con) | |
1158 | { | |
1159 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1160 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1161 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1162 | con->error_msg = "protocol error, bad banner"; |
1163 | return -1; | |
1164 | } | |
1165 | return 0; | |
1166 | } | |
1167 | ||
1168 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1169 | { | |
1170 | switch (ss->ss_family) { | |
1171 | case AF_INET: | |
1172 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1173 | case AF_INET6: | |
1174 | return | |
1175 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1176 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1177 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1178 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1179 | } | |
1180 | return false; | |
1181 | } | |
1182 | ||
1183 | static int addr_port(struct sockaddr_storage *ss) | |
1184 | { | |
1185 | switch (ss->ss_family) { | |
1186 | case AF_INET: | |
f28bcfbe | 1187 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1188 | case AF_INET6: |
f28bcfbe | 1189 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1190 | } |
1191 | return 0; | |
1192 | } | |
1193 | ||
1194 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1195 | { | |
1196 | switch (ss->ss_family) { | |
1197 | case AF_INET: | |
1198 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1199 | break; |
31b8006e SW |
1200 | case AF_INET6: |
1201 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1202 | break; |
31b8006e SW |
1203 | } |
1204 | } | |
1205 | ||
ee3b56f2 NW |
1206 | /* |
1207 | * Unlike other *_pton function semantics, zero indicates success. | |
1208 | */ | |
1209 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1210 | char delim, const char **ipend) | |
1211 | { | |
99f0f3b2 AE |
1212 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1213 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1214 | |
1215 | memset(ss, 0, sizeof(*ss)); | |
1216 | ||
1217 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1218 | ss->ss_family = AF_INET; | |
1219 | return 0; | |
1220 | } | |
1221 | ||
1222 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1223 | ss->ss_family = AF_INET6; | |
1224 | return 0; | |
1225 | } | |
1226 | ||
1227 | return -EINVAL; | |
1228 | } | |
1229 | ||
1230 | /* | |
1231 | * Extract hostname string and resolve using kernel DNS facility. | |
1232 | */ | |
1233 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1234 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1235 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1236 | { | |
1237 | const char *end, *delim_p; | |
1238 | char *colon_p, *ip_addr = NULL; | |
1239 | int ip_len, ret; | |
1240 | ||
1241 | /* | |
1242 | * The end of the hostname occurs immediately preceding the delimiter or | |
1243 | * the port marker (':') where the delimiter takes precedence. | |
1244 | */ | |
1245 | delim_p = memchr(name, delim, namelen); | |
1246 | colon_p = memchr(name, ':', namelen); | |
1247 | ||
1248 | if (delim_p && colon_p) | |
1249 | end = delim_p < colon_p ? delim_p : colon_p; | |
1250 | else if (!delim_p && colon_p) | |
1251 | end = colon_p; | |
1252 | else { | |
1253 | end = delim_p; | |
1254 | if (!end) /* case: hostname:/ */ | |
1255 | end = name + namelen; | |
1256 | } | |
1257 | ||
1258 | if (end <= name) | |
1259 | return -EINVAL; | |
1260 | ||
1261 | /* do dns_resolve upcall */ | |
1262 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1263 | if (ip_len > 0) | |
1264 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1265 | else | |
1266 | ret = -ESRCH; | |
1267 | ||
1268 | kfree(ip_addr); | |
1269 | ||
1270 | *ipend = end; | |
1271 | ||
1272 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1273 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1274 | ||
1275 | return ret; | |
1276 | } | |
1277 | #else | |
1278 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1279 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1280 | { | |
1281 | return -EINVAL; | |
1282 | } | |
1283 | #endif | |
1284 | ||
1285 | /* | |
1286 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1287 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1288 | */ | |
1289 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1290 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1291 | { | |
1292 | int ret; | |
1293 | ||
1294 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1295 | if (ret) | |
1296 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1297 | ||
1298 | return ret; | |
1299 | } | |
1300 | ||
31b8006e SW |
1301 | /* |
1302 | * Parse an ip[:port] list into an addr array. Use the default | |
1303 | * monitor port if a port isn't specified. | |
1304 | */ | |
1305 | int ceph_parse_ips(const char *c, const char *end, | |
1306 | struct ceph_entity_addr *addr, | |
1307 | int max_count, int *count) | |
1308 | { | |
ee3b56f2 | 1309 | int i, ret = -EINVAL; |
31b8006e SW |
1310 | const char *p = c; |
1311 | ||
1312 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1313 | for (i = 0; i < max_count; i++) { | |
1314 | const char *ipend; | |
1315 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1316 | int port; |
39139f64 SW |
1317 | char delim = ','; |
1318 | ||
1319 | if (*p == '[') { | |
1320 | delim = ']'; | |
1321 | p++; | |
1322 | } | |
31b8006e | 1323 | |
ee3b56f2 NW |
1324 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1325 | if (ret) | |
31b8006e | 1326 | goto bad; |
ee3b56f2 NW |
1327 | ret = -EINVAL; |
1328 | ||
31b8006e SW |
1329 | p = ipend; |
1330 | ||
39139f64 SW |
1331 | if (delim == ']') { |
1332 | if (*p != ']') { | |
1333 | dout("missing matching ']'\n"); | |
1334 | goto bad; | |
1335 | } | |
1336 | p++; | |
1337 | } | |
1338 | ||
31b8006e SW |
1339 | /* port? */ |
1340 | if (p < end && *p == ':') { | |
1341 | port = 0; | |
1342 | p++; | |
1343 | while (p < end && *p >= '0' && *p <= '9') { | |
1344 | port = (port * 10) + (*p - '0'); | |
1345 | p++; | |
1346 | } | |
1347 | if (port > 65535 || port == 0) | |
1348 | goto bad; | |
1349 | } else { | |
1350 | port = CEPH_MON_PORT; | |
1351 | } | |
1352 | ||
1353 | addr_set_port(ss, port); | |
1354 | ||
3d14c5d2 | 1355 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1356 | |
1357 | if (p == end) | |
1358 | break; | |
1359 | if (*p != ',') | |
1360 | goto bad; | |
1361 | p++; | |
1362 | } | |
1363 | ||
1364 | if (p != end) | |
1365 | goto bad; | |
1366 | ||
1367 | if (count) | |
1368 | *count = i + 1; | |
1369 | return 0; | |
1370 | ||
1371 | bad: | |
39139f64 | 1372 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1373 | return ret; |
31b8006e | 1374 | } |
3d14c5d2 | 1375 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1376 | |
eed0ef2c | 1377 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1378 | { |
eed0ef2c | 1379 | dout("process_banner on %p\n", con); |
31b8006e SW |
1380 | |
1381 | if (verify_hello(con) < 0) | |
1382 | return -1; | |
1383 | ||
63f2d211 SW |
1384 | ceph_decode_addr(&con->actual_peer_addr); |
1385 | ceph_decode_addr(&con->peer_addr_for_me); | |
1386 | ||
31b8006e SW |
1387 | /* |
1388 | * Make sure the other end is who we wanted. note that the other | |
1389 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1390 | * them the benefit of the doubt. | |
1391 | */ | |
103e2d3a SW |
1392 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1393 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1394 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1395 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1396 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1397 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1398 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1399 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1400 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1401 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1402 | return -1; |
1403 | } | |
1404 | ||
1405 | /* | |
1406 | * did we learn our address? | |
1407 | */ | |
1408 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1409 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1410 | ||
1411 | memcpy(&con->msgr->inst.addr.in_addr, | |
1412 | &con->peer_addr_for_me.in_addr, | |
1413 | sizeof(con->peer_addr_for_me.in_addr)); | |
1414 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1415 | encode_my_addr(con->msgr); |
eed0ef2c | 1416 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1417 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1418 | } |
1419 | ||
eed0ef2c SW |
1420 | set_bit(NEGOTIATING, &con->state); |
1421 | prepare_read_connect(con); | |
1422 | return 0; | |
1423 | } | |
1424 | ||
04a419f9 SW |
1425 | static void fail_protocol(struct ceph_connection *con) |
1426 | { | |
1427 | reset_connection(con); | |
1428 | set_bit(CLOSED, &con->state); /* in case there's queued work */ | |
04a419f9 SW |
1429 | } |
1430 | ||
eed0ef2c SW |
1431 | static int process_connect(struct ceph_connection *con) |
1432 | { | |
3d14c5d2 YS |
1433 | u64 sup_feat = con->msgr->supported_features; |
1434 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1435 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1436 | int ret; |
04a419f9 | 1437 | |
eed0ef2c SW |
1438 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1439 | ||
31b8006e | 1440 | switch (con->in_reply.tag) { |
04a419f9 SW |
1441 | case CEPH_MSGR_TAG_FEATURES: |
1442 | pr_err("%s%lld %s feature set mismatch," | |
1443 | " my %llx < server's %llx, missing %llx\n", | |
1444 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1445 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1446 | sup_feat, server_feat, server_feat & ~sup_feat); |
1447 | con->error_msg = "missing required protocol features"; | |
1448 | fail_protocol(con); | |
1449 | return -1; | |
1450 | ||
31b8006e | 1451 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1452 | pr_err("%s%lld %s protocol version mismatch," |
1453 | " my %d != server's %d\n", | |
1454 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1455 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1456 | le32_to_cpu(con->out_connect.protocol_version), |
1457 | le32_to_cpu(con->in_reply.protocol_version)); | |
1458 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1459 | fail_protocol(con); |
31b8006e SW |
1460 | return -1; |
1461 | ||
4e7a5dcd SW |
1462 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1463 | con->auth_retry++; | |
1464 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1465 | con->auth_retry); | |
1466 | if (con->auth_retry == 2) { | |
1467 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1468 | return -1; |
1469 | } | |
1470 | con->auth_retry = 1; | |
e2200423 | 1471 | con_out_kvec_reset(con); |
e825a66d | 1472 | ret = prepare_write_connect(con); |
0da5d703 SW |
1473 | if (ret < 0) |
1474 | return ret; | |
63733a0f | 1475 | prepare_read_connect(con); |
4e7a5dcd | 1476 | break; |
31b8006e SW |
1477 | |
1478 | case CEPH_MSGR_TAG_RESETSESSION: | |
1479 | /* | |
1480 | * If we connected with a large connect_seq but the peer | |
1481 | * has no record of a session with us (no connection, or | |
1482 | * connect_seq == 0), they will send RESETSESION to indicate | |
1483 | * that they must have reset their session, and may have | |
1484 | * dropped messages. | |
1485 | */ | |
1486 | dout("process_connect got RESET peer seq %u\n", | |
1487 | le32_to_cpu(con->in_connect.connect_seq)); | |
1488 | pr_err("%s%lld %s connection reset\n", | |
1489 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1490 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1491 | reset_connection(con); |
e2200423 | 1492 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1493 | ret = prepare_write_connect(con); |
1494 | if (ret < 0) | |
1495 | return ret; | |
31b8006e SW |
1496 | prepare_read_connect(con); |
1497 | ||
1498 | /* Tell ceph about it. */ | |
ec302645 | 1499 | mutex_unlock(&con->mutex); |
31b8006e SW |
1500 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1501 | if (con->ops->peer_reset) | |
1502 | con->ops->peer_reset(con); | |
ec302645 | 1503 | mutex_lock(&con->mutex); |
0da5d703 SW |
1504 | if (test_bit(CLOSED, &con->state) || |
1505 | test_bit(OPENING, &con->state)) | |
1506 | return -EAGAIN; | |
31b8006e SW |
1507 | break; |
1508 | ||
1509 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1510 | /* | |
1511 | * If we sent a smaller connect_seq than the peer has, try | |
1512 | * again with a larger value. | |
1513 | */ | |
1514 | dout("process_connect got RETRY my seq = %u, peer_seq = %u\n", | |
1515 | le32_to_cpu(con->out_connect.connect_seq), | |
1516 | le32_to_cpu(con->in_connect.connect_seq)); | |
1517 | con->connect_seq = le32_to_cpu(con->in_connect.connect_seq); | |
e2200423 | 1518 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1519 | ret = prepare_write_connect(con); |
1520 | if (ret < 0) | |
1521 | return ret; | |
31b8006e SW |
1522 | prepare_read_connect(con); |
1523 | break; | |
1524 | ||
1525 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1526 | /* | |
1527 | * If we sent a smaller global_seq than the peer has, try | |
1528 | * again with a larger value. | |
1529 | */ | |
eed0ef2c | 1530 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e SW |
1531 | con->peer_global_seq, |
1532 | le32_to_cpu(con->in_connect.global_seq)); | |
1533 | get_global_seq(con->msgr, | |
1534 | le32_to_cpu(con->in_connect.global_seq)); | |
e2200423 | 1535 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1536 | ret = prepare_write_connect(con); |
1537 | if (ret < 0) | |
1538 | return ret; | |
31b8006e SW |
1539 | prepare_read_connect(con); |
1540 | break; | |
1541 | ||
1542 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1543 | if (req_feat & ~server_feat) { |
1544 | pr_err("%s%lld %s protocol feature mismatch," | |
1545 | " my required %llx > server's %llx, need %llx\n", | |
1546 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1547 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1548 | req_feat, server_feat, req_feat & ~server_feat); |
1549 | con->error_msg = "missing required protocol features"; | |
1550 | fail_protocol(con); | |
1551 | return -1; | |
1552 | } | |
31b8006e | 1553 | clear_bit(CONNECTING, &con->state); |
31b8006e SW |
1554 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1555 | con->connect_seq++; | |
aba558e2 | 1556 | con->peer_features = server_feat; |
31b8006e SW |
1557 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1558 | con->peer_global_seq, | |
1559 | le32_to_cpu(con->in_reply.connect_seq), | |
1560 | con->connect_seq); | |
1561 | WARN_ON(con->connect_seq != | |
1562 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1563 | |
1564 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
928443cd | 1565 | set_bit(LOSSYTX, &con->flags); |
92ac41d0 | 1566 | |
31b8006e SW |
1567 | prepare_read_tag(con); |
1568 | break; | |
1569 | ||
1570 | case CEPH_MSGR_TAG_WAIT: | |
1571 | /* | |
1572 | * If there is a connection race (we are opening | |
1573 | * connections to each other), one of us may just have | |
1574 | * to WAIT. This shouldn't happen if we are the | |
1575 | * client. | |
1576 | */ | |
04177882 SW |
1577 | pr_err("process_connect got WAIT as client\n"); |
1578 | con->error_msg = "protocol error, got WAIT as client"; | |
1579 | return -1; | |
31b8006e SW |
1580 | |
1581 | default: | |
1582 | pr_err("connect protocol error, will retry\n"); | |
1583 | con->error_msg = "protocol error, garbage tag during connect"; | |
1584 | return -1; | |
1585 | } | |
1586 | return 0; | |
1587 | } | |
1588 | ||
1589 | ||
1590 | /* | |
1591 | * read (part of) an ack | |
1592 | */ | |
1593 | static int read_partial_ack(struct ceph_connection *con) | |
1594 | { | |
fd51653f AE |
1595 | int size = sizeof (con->in_temp_ack); |
1596 | int end = size; | |
1597 | ||
1598 | return read_partial(con, end, size, &con->in_temp_ack); | |
31b8006e SW |
1599 | } |
1600 | ||
1601 | ||
1602 | /* | |
1603 | * We can finally discard anything that's been acked. | |
1604 | */ | |
1605 | static void process_ack(struct ceph_connection *con) | |
1606 | { | |
1607 | struct ceph_msg *m; | |
1608 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1609 | u64 seq; | |
1610 | ||
31b8006e SW |
1611 | while (!list_empty(&con->out_sent)) { |
1612 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1613 | list_head); | |
1614 | seq = le64_to_cpu(m->hdr.seq); | |
1615 | if (seq > ack) | |
1616 | break; | |
1617 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1618 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1619 | m->ack_stamp = jiffies; |
31b8006e SW |
1620 | ceph_msg_remove(m); |
1621 | } | |
31b8006e SW |
1622 | prepare_read_tag(con); |
1623 | } | |
1624 | ||
1625 | ||
1626 | ||
1627 | ||
2450418c | 1628 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1629 | struct kvec *section, |
1630 | unsigned int sec_len, u32 *crc) | |
2450418c | 1631 | { |
68b4476b | 1632 | int ret, left; |
2450418c YS |
1633 | |
1634 | BUG_ON(!section); | |
1635 | ||
1636 | while (section->iov_len < sec_len) { | |
1637 | BUG_ON(section->iov_base == NULL); | |
1638 | left = sec_len - section->iov_len; | |
1639 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1640 | section->iov_len, left); | |
1641 | if (ret <= 0) | |
1642 | return ret; | |
1643 | section->iov_len += ret; | |
2450418c | 1644 | } |
fe3ad593 AE |
1645 | if (section->iov_len == sec_len) |
1646 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1647 | |
2450418c YS |
1648 | return 1; |
1649 | } | |
31b8006e | 1650 | |
2450418c YS |
1651 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, |
1652 | struct ceph_msg_header *hdr, | |
1653 | int *skip); | |
68b4476b YS |
1654 | |
1655 | ||
1656 | static int read_partial_message_pages(struct ceph_connection *con, | |
1657 | struct page **pages, | |
bca064d2 | 1658 | unsigned data_len, bool do_datacrc) |
68b4476b YS |
1659 | { |
1660 | void *p; | |
1661 | int ret; | |
1662 | int left; | |
1663 | ||
1664 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1665 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1666 | /* (page) data */ | |
1667 | BUG_ON(pages == NULL); | |
1668 | p = kmap(pages[con->in_msg_pos.page]); | |
1669 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1670 | left); | |
bca064d2 | 1671 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1672 | con->in_data_crc = |
1673 | crc32c(con->in_data_crc, | |
1674 | p + con->in_msg_pos.page_pos, ret); | |
1675 | kunmap(pages[con->in_msg_pos.page]); | |
1676 | if (ret <= 0) | |
1677 | return ret; | |
1678 | con->in_msg_pos.data_pos += ret; | |
1679 | con->in_msg_pos.page_pos += ret; | |
1680 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1681 | con->in_msg_pos.page_pos = 0; | |
1682 | con->in_msg_pos.page++; | |
1683 | } | |
1684 | ||
1685 | return ret; | |
1686 | } | |
1687 | ||
1688 | #ifdef CONFIG_BLOCK | |
1689 | static int read_partial_message_bio(struct ceph_connection *con, | |
1690 | struct bio **bio_iter, int *bio_seg, | |
bca064d2 | 1691 | unsigned data_len, bool do_datacrc) |
68b4476b YS |
1692 | { |
1693 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1694 | void *p; | |
1695 | int ret, left; | |
1696 | ||
1697 | if (IS_ERR(bv)) | |
1698 | return PTR_ERR(bv); | |
1699 | ||
1700 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1701 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1702 | ||
1703 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1704 | ||
1705 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1706 | left); | |
bca064d2 | 1707 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1708 | con->in_data_crc = |
1709 | crc32c(con->in_data_crc, | |
1710 | p + con->in_msg_pos.page_pos, ret); | |
1711 | kunmap(bv->bv_page); | |
1712 | if (ret <= 0) | |
1713 | return ret; | |
1714 | con->in_msg_pos.data_pos += ret; | |
1715 | con->in_msg_pos.page_pos += ret; | |
1716 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1717 | con->in_msg_pos.page_pos = 0; | |
1718 | iter_bio_next(bio_iter, bio_seg); | |
1719 | } | |
1720 | ||
1721 | return ret; | |
1722 | } | |
1723 | #endif | |
1724 | ||
31b8006e SW |
1725 | /* |
1726 | * read (part of) a message. | |
1727 | */ | |
1728 | static int read_partial_message(struct ceph_connection *con) | |
1729 | { | |
1730 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1731 | int size; |
1732 | int end; | |
31b8006e | 1733 | int ret; |
c5c6b19d | 1734 | unsigned front_len, middle_len, data_len; |
37675b0f | 1735 | bool do_datacrc = !con->msgr->nocrc; |
2450418c | 1736 | int skip; |
ae18756b | 1737 | u64 seq; |
fe3ad593 | 1738 | u32 crc; |
31b8006e SW |
1739 | |
1740 | dout("read_partial_message con %p msg %p\n", con, m); | |
1741 | ||
1742 | /* header */ | |
fd51653f AE |
1743 | size = sizeof (con->in_hdr); |
1744 | end = size; | |
1745 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1746 | if (ret <= 0) |
1747 | return ret; | |
fe3ad593 AE |
1748 | |
1749 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1750 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1751 | pr_err("read_partial_message bad hdr " | |
1752 | " crc %u != expected %u\n", | |
1753 | crc, con->in_hdr.crc); | |
1754 | return -EBADMSG; | |
1755 | } | |
1756 | ||
31b8006e SW |
1757 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1758 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1759 | return -EIO; | |
1760 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1761 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1762 | return -EIO; | |
1763 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1764 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1765 | return -EIO; | |
1766 | ||
ae18756b SW |
1767 | /* verify seq# */ |
1768 | seq = le64_to_cpu(con->in_hdr.seq); | |
1769 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1770 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1771 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1772 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1773 | seq, con->in_seq + 1); |
1774 | con->in_base_pos = -front_len - middle_len - data_len - | |
1775 | sizeof(m->footer); | |
1776 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1777 | return 0; |
1778 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1779 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1780 | seq, con->in_seq + 1); | |
1781 | con->error_msg = "bad message sequence # for incoming message"; | |
1782 | return -EBADMSG; | |
1783 | } | |
1784 | ||
31b8006e SW |
1785 | /* allocate message? */ |
1786 | if (!con->in_msg) { | |
1787 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1788 | con->in_hdr.front_len, con->in_hdr.data_len); | |
ae32be31 | 1789 | skip = 0; |
2450418c YS |
1790 | con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip); |
1791 | if (skip) { | |
31b8006e | 1792 | /* skip this message */ |
a79832f2 | 1793 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1794 | BUG_ON(con->in_msg); |
31b8006e SW |
1795 | con->in_base_pos = -front_len - middle_len - data_len - |
1796 | sizeof(m->footer); | |
1797 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1798 | con->in_seq++; |
31b8006e SW |
1799 | return 0; |
1800 | } | |
a79832f2 | 1801 | if (!con->in_msg) { |
5b3a4db3 SW |
1802 | con->error_msg = |
1803 | "error allocating memory for incoming message"; | |
a79832f2 | 1804 | return -ENOMEM; |
31b8006e SW |
1805 | } |
1806 | m = con->in_msg; | |
1807 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1808 | if (m->middle) |
1809 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1810 | |
1811 | con->in_msg_pos.page = 0; | |
68b4476b | 1812 | if (m->pages) |
c5c6b19d | 1813 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1814 | else |
1815 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1816 | con->in_msg_pos.data_pos = 0; |
31b8006e SW |
1817 | } |
1818 | ||
1819 | /* front */ | |
2450418c YS |
1820 | ret = read_partial_message_section(con, &m->front, front_len, |
1821 | &con->in_front_crc); | |
1822 | if (ret <= 0) | |
1823 | return ret; | |
31b8006e SW |
1824 | |
1825 | /* middle */ | |
2450418c | 1826 | if (m->middle) { |
213c99ee SW |
1827 | ret = read_partial_message_section(con, &m->middle->vec, |
1828 | middle_len, | |
2450418c | 1829 | &con->in_middle_crc); |
31b8006e SW |
1830 | if (ret <= 0) |
1831 | return ret; | |
31b8006e | 1832 | } |
68b4476b YS |
1833 | #ifdef CONFIG_BLOCK |
1834 | if (m->bio && !m->bio_iter) | |
1835 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1836 | #endif | |
31b8006e SW |
1837 | |
1838 | /* (page) data */ | |
31b8006e | 1839 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1840 | if (m->pages) { |
1841 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1842 | data_len, do_datacrc); |
68b4476b YS |
1843 | if (ret <= 0) |
1844 | return ret; | |
1845 | #ifdef CONFIG_BLOCK | |
1846 | } else if (m->bio) { | |
1847 | ||
1848 | ret = read_partial_message_bio(con, | |
1849 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1850 | data_len, do_datacrc); |
68b4476b YS |
1851 | if (ret <= 0) |
1852 | return ret; | |
1853 | #endif | |
1854 | } else { | |
1855 | BUG_ON(1); | |
31b8006e SW |
1856 | } |
1857 | } | |
1858 | ||
31b8006e | 1859 | /* footer */ |
fd51653f AE |
1860 | size = sizeof (m->footer); |
1861 | end += size; | |
1862 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
1863 | if (ret <= 0) |
1864 | return ret; | |
1865 | ||
31b8006e SW |
1866 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
1867 | m, front_len, m->footer.front_crc, middle_len, | |
1868 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1869 | ||
1870 | /* crc ok? */ | |
1871 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1872 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1873 | m, con->in_front_crc, m->footer.front_crc); | |
1874 | return -EBADMSG; | |
1875 | } | |
1876 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1877 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1878 | m, con->in_middle_crc, m->footer.middle_crc); | |
1879 | return -EBADMSG; | |
1880 | } | |
bca064d2 | 1881 | if (do_datacrc && |
31b8006e SW |
1882 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
1883 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1884 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1885 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1886 | return -EBADMSG; | |
1887 | } | |
1888 | ||
1889 | return 1; /* done! */ | |
1890 | } | |
1891 | ||
1892 | /* | |
1893 | * Process message. This happens in the worker thread. The callback should | |
1894 | * be careful not to do anything that waits on other incoming messages or it | |
1895 | * may deadlock. | |
1896 | */ | |
1897 | static void process_message(struct ceph_connection *con) | |
1898 | { | |
5e095e8b | 1899 | struct ceph_msg *msg; |
31b8006e | 1900 | |
5e095e8b | 1901 | msg = con->in_msg; |
31b8006e SW |
1902 | con->in_msg = NULL; |
1903 | ||
1904 | /* if first message, set peer_name */ | |
1905 | if (con->peer_name.type == 0) | |
dbad185d | 1906 | con->peer_name = msg->hdr.src; |
31b8006e | 1907 | |
31b8006e | 1908 | con->in_seq++; |
ec302645 | 1909 | mutex_unlock(&con->mutex); |
31b8006e SW |
1910 | |
1911 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1912 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1913 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1914 | le16_to_cpu(msg->hdr.type), |
1915 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1916 | le32_to_cpu(msg->hdr.front_len), | |
1917 | le32_to_cpu(msg->hdr.data_len), | |
1918 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1919 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1920 | |
1921 | mutex_lock(&con->mutex); | |
31b8006e SW |
1922 | prepare_read_tag(con); |
1923 | } | |
1924 | ||
1925 | ||
1926 | /* | |
1927 | * Write something to the socket. Called in a worker thread when the | |
1928 | * socket appears to be writeable and we have something ready to send. | |
1929 | */ | |
1930 | static int try_write(struct ceph_connection *con) | |
1931 | { | |
31b8006e SW |
1932 | int ret = 1; |
1933 | ||
1934 | dout("try_write start %p state %lu nref %d\n", con, con->state, | |
1935 | atomic_read(&con->nref)); | |
1936 | ||
31b8006e SW |
1937 | more: |
1938 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
1939 | ||
1940 | /* open the socket first? */ | |
1941 | if (con->sock == NULL) { | |
a5988c49 AE |
1942 | clear_bit(NEGOTIATING, &con->state); |
1943 | set_bit(CONNECTING, &con->state); | |
1944 | ||
e2200423 | 1945 | con_out_kvec_reset(con); |
e825a66d | 1946 | prepare_write_banner(con); |
5a0f8fdd AE |
1947 | ret = prepare_write_connect(con); |
1948 | if (ret < 0) | |
1949 | goto out; | |
eed0ef2c | 1950 | prepare_read_banner(con); |
31b8006e | 1951 | |
cf3e5c40 | 1952 | BUG_ON(con->in_msg); |
31b8006e SW |
1953 | con->in_tag = CEPH_MSGR_TAG_READY; |
1954 | dout("try_write initiating connect on %p new state %lu\n", | |
1955 | con, con->state); | |
41617d0c AE |
1956 | ret = ceph_tcp_connect(con); |
1957 | if (ret < 0) { | |
31b8006e | 1958 | con->error_msg = "connect error"; |
31b8006e SW |
1959 | goto out; |
1960 | } | |
1961 | } | |
1962 | ||
1963 | more_kvec: | |
1964 | /* kvec data queued? */ | |
1965 | if (con->out_skip) { | |
1966 | ret = write_partial_skip(con); | |
1967 | if (ret <= 0) | |
42961d23 | 1968 | goto out; |
31b8006e SW |
1969 | } |
1970 | if (con->out_kvec_left) { | |
1971 | ret = write_partial_kvec(con); | |
1972 | if (ret <= 0) | |
42961d23 | 1973 | goto out; |
31b8006e SW |
1974 | } |
1975 | ||
1976 | /* msg pages? */ | |
1977 | if (con->out_msg) { | |
c86a2930 SW |
1978 | if (con->out_msg_done) { |
1979 | ceph_msg_put(con->out_msg); | |
1980 | con->out_msg = NULL; /* we're done with this one */ | |
1981 | goto do_next; | |
1982 | } | |
1983 | ||
31b8006e SW |
1984 | ret = write_partial_msg_pages(con); |
1985 | if (ret == 1) | |
1986 | goto more_kvec; /* we need to send the footer, too! */ | |
1987 | if (ret == 0) | |
42961d23 | 1988 | goto out; |
31b8006e SW |
1989 | if (ret < 0) { |
1990 | dout("try_write write_partial_msg_pages err %d\n", | |
1991 | ret); | |
42961d23 | 1992 | goto out; |
31b8006e SW |
1993 | } |
1994 | } | |
1995 | ||
c86a2930 | 1996 | do_next: |
31b8006e SW |
1997 | if (!test_bit(CONNECTING, &con->state)) { |
1998 | /* is anything else pending? */ | |
1999 | if (!list_empty(&con->out_queue)) { | |
2000 | prepare_write_message(con); | |
2001 | goto more; | |
2002 | } | |
2003 | if (con->in_seq > con->in_seq_acked) { | |
2004 | prepare_write_ack(con); | |
2005 | goto more; | |
2006 | } | |
928443cd | 2007 | if (test_and_clear_bit(KEEPALIVE_PENDING, &con->flags)) { |
31b8006e SW |
2008 | prepare_write_keepalive(con); |
2009 | goto more; | |
2010 | } | |
2011 | } | |
2012 | ||
2013 | /* Nothing to do! */ | |
928443cd | 2014 | clear_bit(WRITE_PENDING, &con->flags); |
31b8006e | 2015 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2016 | ret = 0; |
2017 | out: | |
42961d23 | 2018 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2019 | return ret; |
2020 | } | |
2021 | ||
2022 | ||
2023 | ||
2024 | /* | |
2025 | * Read what we can from the socket. | |
2026 | */ | |
2027 | static int try_read(struct ceph_connection *con) | |
2028 | { | |
31b8006e SW |
2029 | int ret = -1; |
2030 | ||
2031 | if (!con->sock) | |
2032 | return 0; | |
2033 | ||
2034 | if (test_bit(STANDBY, &con->state)) | |
2035 | return 0; | |
2036 | ||
2037 | dout("try_read start on %p\n", con); | |
ec302645 | 2038 | |
31b8006e SW |
2039 | more: |
2040 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, | |
2041 | con->in_base_pos); | |
0da5d703 SW |
2042 | |
2043 | /* | |
2044 | * process_connect and process_message drop and re-take | |
2045 | * con->mutex. make sure we handle a racing close or reopen. | |
2046 | */ | |
2047 | if (test_bit(CLOSED, &con->state) || | |
2048 | test_bit(OPENING, &con->state)) { | |
2049 | ret = -EAGAIN; | |
2050 | goto out; | |
2051 | } | |
2052 | ||
31b8006e | 2053 | if (test_bit(CONNECTING, &con->state)) { |
eed0ef2c SW |
2054 | if (!test_bit(NEGOTIATING, &con->state)) { |
2055 | dout("try_read connecting\n"); | |
2056 | ret = read_partial_banner(con); | |
2057 | if (ret <= 0) | |
eed0ef2c | 2058 | goto out; |
98bdb0aa SW |
2059 | ret = process_banner(con); |
2060 | if (ret < 0) | |
2061 | goto out; | |
eed0ef2c | 2062 | } |
31b8006e SW |
2063 | ret = read_partial_connect(con); |
2064 | if (ret <= 0) | |
31b8006e | 2065 | goto out; |
98bdb0aa SW |
2066 | ret = process_connect(con); |
2067 | if (ret < 0) | |
2068 | goto out; | |
31b8006e SW |
2069 | goto more; |
2070 | } | |
2071 | ||
2072 | if (con->in_base_pos < 0) { | |
2073 | /* | |
2074 | * skipping + discarding content. | |
2075 | * | |
2076 | * FIXME: there must be a better way to do this! | |
2077 | */ | |
84495f49 AE |
2078 | static char buf[SKIP_BUF_SIZE]; |
2079 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2080 | ||
31b8006e SW |
2081 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2082 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2083 | if (ret <= 0) | |
98bdb0aa | 2084 | goto out; |
31b8006e SW |
2085 | con->in_base_pos += ret; |
2086 | if (con->in_base_pos) | |
2087 | goto more; | |
2088 | } | |
2089 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2090 | /* | |
2091 | * what's next? | |
2092 | */ | |
2093 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2094 | if (ret <= 0) | |
98bdb0aa | 2095 | goto out; |
31b8006e SW |
2096 | dout("try_read got tag %d\n", (int)con->in_tag); |
2097 | switch (con->in_tag) { | |
2098 | case CEPH_MSGR_TAG_MSG: | |
2099 | prepare_read_message(con); | |
2100 | break; | |
2101 | case CEPH_MSGR_TAG_ACK: | |
2102 | prepare_read_ack(con); | |
2103 | break; | |
2104 | case CEPH_MSGR_TAG_CLOSE: | |
2105 | set_bit(CLOSED, &con->state); /* fixme */ | |
98bdb0aa | 2106 | goto out; |
31b8006e SW |
2107 | default: |
2108 | goto bad_tag; | |
2109 | } | |
2110 | } | |
2111 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2112 | ret = read_partial_message(con); | |
2113 | if (ret <= 0) { | |
2114 | switch (ret) { | |
2115 | case -EBADMSG: | |
2116 | con->error_msg = "bad crc"; | |
2117 | ret = -EIO; | |
98bdb0aa | 2118 | break; |
31b8006e SW |
2119 | case -EIO: |
2120 | con->error_msg = "io error"; | |
98bdb0aa | 2121 | break; |
31b8006e | 2122 | } |
98bdb0aa | 2123 | goto out; |
31b8006e SW |
2124 | } |
2125 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2126 | goto more; | |
2127 | process_message(con); | |
2128 | goto more; | |
2129 | } | |
2130 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2131 | ret = read_partial_ack(con); | |
2132 | if (ret <= 0) | |
98bdb0aa | 2133 | goto out; |
31b8006e SW |
2134 | process_ack(con); |
2135 | goto more; | |
2136 | } | |
2137 | ||
31b8006e | 2138 | out: |
98bdb0aa | 2139 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2140 | return ret; |
2141 | ||
2142 | bad_tag: | |
2143 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2144 | con->error_msg = "protocol error, garbage tag"; | |
2145 | ret = -1; | |
2146 | goto out; | |
2147 | } | |
2148 | ||
2149 | ||
2150 | /* | |
2151 | * Atomically queue work on a connection. Bump @con reference to | |
2152 | * avoid races with connection teardown. | |
31b8006e SW |
2153 | */ |
2154 | static void queue_con(struct ceph_connection *con) | |
2155 | { | |
31b8006e SW |
2156 | if (!con->ops->get(con)) { |
2157 | dout("queue_con %p ref count 0\n", con); | |
2158 | return; | |
2159 | } | |
2160 | ||
f363e45f | 2161 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2162 | dout("queue_con %p - already queued\n", con); |
2163 | con->ops->put(con); | |
2164 | } else { | |
2165 | dout("queue_con %p\n", con); | |
2166 | } | |
2167 | } | |
2168 | ||
2169 | /* | |
2170 | * Do some work on a connection. Drop a connection ref when we're done. | |
2171 | */ | |
2172 | static void con_work(struct work_struct *work) | |
2173 | { | |
2174 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2175 | work.work); | |
0da5d703 | 2176 | int ret; |
31b8006e | 2177 | |
9dd4658d | 2178 | mutex_lock(&con->mutex); |
0da5d703 | 2179 | restart: |
928443cd | 2180 | if (test_and_clear_bit(BACKOFF, &con->flags)) { |
60bf8bf8 SW |
2181 | dout("con_work %p backing off\n", con); |
2182 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2183 | round_jiffies_relative(con->delay))) { | |
2184 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2185 | mutex_unlock(&con->mutex); | |
2186 | return; | |
2187 | } else { | |
2188 | con->ops->put(con); | |
2189 | dout("con_work %p FAILED to back off %lu\n", con, | |
2190 | con->delay); | |
2191 | } | |
2192 | } | |
9dd4658d | 2193 | |
e00de341 SW |
2194 | if (test_bit(STANDBY, &con->state)) { |
2195 | dout("con_work %p STANDBY\n", con); | |
2196 | goto done; | |
2197 | } | |
31b8006e SW |
2198 | if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */ |
2199 | dout("con_work CLOSED\n"); | |
2200 | con_close_socket(con); | |
2201 | goto done; | |
2202 | } | |
2203 | if (test_and_clear_bit(OPENING, &con->state)) { | |
2204 | /* reopen w/ new peer */ | |
2205 | dout("con_work OPENING\n"); | |
2206 | con_close_socket(con); | |
2207 | } | |
2208 | ||
928443cd | 2209 | if (test_and_clear_bit(SOCK_CLOSED, &con->flags)) |
0da5d703 SW |
2210 | goto fault; |
2211 | ||
2212 | ret = try_read(con); | |
2213 | if (ret == -EAGAIN) | |
2214 | goto restart; | |
2215 | if (ret < 0) | |
2216 | goto fault; | |
2217 | ||
2218 | ret = try_write(con); | |
2219 | if (ret == -EAGAIN) | |
2220 | goto restart; | |
2221 | if (ret < 0) | |
2222 | goto fault; | |
31b8006e SW |
2223 | |
2224 | done: | |
9dd4658d | 2225 | mutex_unlock(&con->mutex); |
9dd4658d | 2226 | done_unlocked: |
31b8006e | 2227 | con->ops->put(con); |
0da5d703 SW |
2228 | return; |
2229 | ||
2230 | fault: | |
2231 | mutex_unlock(&con->mutex); | |
2232 | ceph_fault(con); /* error/fault path */ | |
2233 | goto done_unlocked; | |
31b8006e SW |
2234 | } |
2235 | ||
2236 | ||
2237 | /* | |
2238 | * Generic error/fault handler. A retry mechanism is used with | |
2239 | * exponential backoff | |
2240 | */ | |
2241 | static void ceph_fault(struct ceph_connection *con) | |
2242 | { | |
2243 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2244 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2245 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2246 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2247 | |
928443cd | 2248 | if (test_bit(LOSSYTX, &con->flags)) { |
31b8006e SW |
2249 | dout("fault on LOSSYTX channel\n"); |
2250 | goto out; | |
2251 | } | |
2252 | ||
ec302645 | 2253 | mutex_lock(&con->mutex); |
91e45ce3 SW |
2254 | if (test_bit(CLOSED, &con->state)) |
2255 | goto out_unlock; | |
ec302645 | 2256 | |
31b8006e | 2257 | con_close_socket(con); |
5e095e8b SW |
2258 | |
2259 | if (con->in_msg) { | |
2260 | ceph_msg_put(con->in_msg); | |
2261 | con->in_msg = NULL; | |
2262 | } | |
31b8006e | 2263 | |
e80a52d1 SW |
2264 | /* Requeue anything that hasn't been acked */ |
2265 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2266 | |
e76661d0 SW |
2267 | /* If there are no messages queued or keepalive pending, place |
2268 | * the connection in a STANDBY state */ | |
2269 | if (list_empty(&con->out_queue) && | |
928443cd | 2270 | !test_bit(KEEPALIVE_PENDING, &con->flags)) { |
e00de341 | 2271 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
928443cd | 2272 | clear_bit(WRITE_PENDING, &con->flags); |
31b8006e | 2273 | set_bit(STANDBY, &con->state); |
e80a52d1 SW |
2274 | } else { |
2275 | /* retry after a delay. */ | |
2276 | if (con->delay == 0) | |
2277 | con->delay = BASE_DELAY_INTERVAL; | |
2278 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2279 | con->delay *= 2; | |
e80a52d1 SW |
2280 | con->ops->get(con); |
2281 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2282 | round_jiffies_relative(con->delay))) { |
2283 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2284 | } else { | |
e80a52d1 | 2285 | con->ops->put(con); |
60bf8bf8 SW |
2286 | dout("fault failed to queue %p delay %lu, backoff\n", |
2287 | con, con->delay); | |
2288 | /* | |
2289 | * In many cases we see a socket state change | |
2290 | * while con_work is running and end up | |
2291 | * queuing (non-delayed) work, such that we | |
2292 | * can't backoff with a delay. Set a flag so | |
2293 | * that when con_work restarts we schedule the | |
2294 | * delay then. | |
2295 | */ | |
928443cd | 2296 | set_bit(BACKOFF, &con->flags); |
60bf8bf8 | 2297 | } |
31b8006e SW |
2298 | } |
2299 | ||
91e45ce3 SW |
2300 | out_unlock: |
2301 | mutex_unlock(&con->mutex); | |
31b8006e | 2302 | out: |
161fd65a SW |
2303 | /* |
2304 | * in case we faulted due to authentication, invalidate our | |
2305 | * current tickets so that we can get new ones. | |
213c99ee | 2306 | */ |
161fd65a SW |
2307 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2308 | dout("calling invalidate_authorizer()\n"); | |
2309 | con->ops->invalidate_authorizer(con); | |
2310 | } | |
2311 | ||
31b8006e SW |
2312 | if (con->ops->fault) |
2313 | con->ops->fault(con); | |
2314 | } | |
2315 | ||
2316 | ||
2317 | ||
2318 | /* | |
15d9882c | 2319 | * initialize a new messenger instance |
31b8006e | 2320 | */ |
15d9882c AE |
2321 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2322 | struct ceph_entity_addr *myaddr, | |
2323 | u32 supported_features, | |
2324 | u32 required_features, | |
2325 | bool nocrc) | |
31b8006e | 2326 | { |
3d14c5d2 YS |
2327 | msgr->supported_features = supported_features; |
2328 | msgr->required_features = required_features; | |
2329 | ||
31b8006e SW |
2330 | spin_lock_init(&msgr->global_seq_lock); |
2331 | ||
31b8006e SW |
2332 | if (myaddr) |
2333 | msgr->inst.addr = *myaddr; | |
2334 | ||
2335 | /* select a random nonce */ | |
ac8839d7 | 2336 | msgr->inst.addr.type = 0; |
103e2d3a | 2337 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2338 | encode_my_addr(msgr); |
15d9882c | 2339 | msgr->nocrc = nocrc; |
31b8006e | 2340 | |
15d9882c | 2341 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2342 | } |
15d9882c | 2343 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2344 | |
e00de341 SW |
2345 | static void clear_standby(struct ceph_connection *con) |
2346 | { | |
2347 | /* come back from STANDBY? */ | |
2348 | if (test_and_clear_bit(STANDBY, &con->state)) { | |
2349 | mutex_lock(&con->mutex); | |
2350 | dout("clear_standby %p and ++connect_seq\n", con); | |
2351 | con->connect_seq++; | |
928443cd AE |
2352 | WARN_ON(test_bit(WRITE_PENDING, &con->flags)); |
2353 | WARN_ON(test_bit(KEEPALIVE_PENDING, &con->flags)); | |
e00de341 SW |
2354 | mutex_unlock(&con->mutex); |
2355 | } | |
2356 | } | |
2357 | ||
31b8006e SW |
2358 | /* |
2359 | * Queue up an outgoing message on the given connection. | |
2360 | */ | |
2361 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2362 | { | |
2363 | if (test_bit(CLOSED, &con->state)) { | |
2364 | dout("con_send %p closed, dropping %p\n", con, msg); | |
2365 | ceph_msg_put(msg); | |
2366 | return; | |
2367 | } | |
2368 | ||
2369 | /* set src+dst */ | |
dbad185d | 2370 | msg->hdr.src = con->msgr->inst.name; |
31b8006e | 2371 | |
3ca02ef9 SW |
2372 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
2373 | ||
e84346b7 SW |
2374 | msg->needs_out_seq = true; |
2375 | ||
31b8006e | 2376 | /* queue */ |
ec302645 | 2377 | mutex_lock(&con->mutex); |
31b8006e SW |
2378 | BUG_ON(!list_empty(&msg->list_head)); |
2379 | list_add_tail(&msg->list_head, &con->out_queue); | |
2380 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2381 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2382 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2383 | le32_to_cpu(msg->hdr.front_len), | |
2384 | le32_to_cpu(msg->hdr.middle_len), | |
2385 | le32_to_cpu(msg->hdr.data_len)); | |
ec302645 | 2386 | mutex_unlock(&con->mutex); |
31b8006e SW |
2387 | |
2388 | /* if there wasn't anything waiting to send before, queue | |
2389 | * new work */ | |
e00de341 | 2390 | clear_standby(con); |
928443cd | 2391 | if (test_and_set_bit(WRITE_PENDING, &con->flags) == 0) |
31b8006e SW |
2392 | queue_con(con); |
2393 | } | |
3d14c5d2 | 2394 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2395 | |
2396 | /* | |
2397 | * Revoke a message that was previously queued for send | |
2398 | */ | |
2399 | void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg) | |
2400 | { | |
ec302645 | 2401 | mutex_lock(&con->mutex); |
31b8006e | 2402 | if (!list_empty(&msg->list_head)) { |
ed98adad | 2403 | dout("con_revoke %p msg %p - was on queue\n", con, msg); |
31b8006e SW |
2404 | list_del_init(&msg->list_head); |
2405 | ceph_msg_put(msg); | |
2406 | msg->hdr.seq = 0; | |
ed98adad SW |
2407 | } |
2408 | if (con->out_msg == msg) { | |
2409 | dout("con_revoke %p msg %p - was sending\n", con, msg); | |
2410 | con->out_msg = NULL; | |
31b8006e SW |
2411 | if (con->out_kvec_is_msg) { |
2412 | con->out_skip = con->out_kvec_bytes; | |
2413 | con->out_kvec_is_msg = false; | |
2414 | } | |
ed98adad SW |
2415 | ceph_msg_put(msg); |
2416 | msg->hdr.seq = 0; | |
31b8006e | 2417 | } |
ec302645 | 2418 | mutex_unlock(&con->mutex); |
31b8006e SW |
2419 | } |
2420 | ||
350b1c32 | 2421 | /* |
0d59ab81 | 2422 | * Revoke a message that we may be reading data into |
350b1c32 | 2423 | */ |
0d59ab81 | 2424 | void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg) |
350b1c32 SW |
2425 | { |
2426 | mutex_lock(&con->mutex); | |
0d59ab81 YS |
2427 | if (con->in_msg && con->in_msg == msg) { |
2428 | unsigned front_len = le32_to_cpu(con->in_hdr.front_len); | |
2429 | unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
350b1c32 SW |
2430 | unsigned data_len = le32_to_cpu(con->in_hdr.data_len); |
2431 | ||
2432 | /* skip rest of message */ | |
0d59ab81 | 2433 | dout("con_revoke_pages %p msg %p revoked\n", con, msg); |
350b1c32 SW |
2434 | con->in_base_pos = con->in_base_pos - |
2435 | sizeof(struct ceph_msg_header) - | |
0d59ab81 YS |
2436 | front_len - |
2437 | middle_len - | |
2438 | data_len - | |
350b1c32 | 2439 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2440 | ceph_msg_put(con->in_msg); |
2441 | con->in_msg = NULL; | |
2442 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2443 | con->in_seq++; |
350b1c32 SW |
2444 | } else { |
2445 | dout("con_revoke_pages %p msg %p pages %p no-op\n", | |
0d59ab81 | 2446 | con, con->in_msg, msg); |
350b1c32 SW |
2447 | } |
2448 | mutex_unlock(&con->mutex); | |
2449 | } | |
2450 | ||
31b8006e SW |
2451 | /* |
2452 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2453 | */ | |
2454 | void ceph_con_keepalive(struct ceph_connection *con) | |
2455 | { | |
e00de341 SW |
2456 | dout("con_keepalive %p\n", con); |
2457 | clear_standby(con); | |
928443cd AE |
2458 | if (test_and_set_bit(KEEPALIVE_PENDING, &con->flags) == 0 && |
2459 | test_and_set_bit(WRITE_PENDING, &con->flags) == 0) | |
31b8006e SW |
2460 | queue_con(con); |
2461 | } | |
3d14c5d2 | 2462 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2463 | |
2464 | ||
2465 | /* | |
2466 | * construct a new message with given type, size | |
2467 | * the new msg has a ref count of 1. | |
2468 | */ | |
b61c2763 SW |
2469 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2470 | bool can_fail) | |
31b8006e SW |
2471 | { |
2472 | struct ceph_msg *m; | |
2473 | ||
34d23762 | 2474 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2475 | if (m == NULL) |
2476 | goto out; | |
c2e552e7 | 2477 | kref_init(&m->kref); |
31b8006e SW |
2478 | INIT_LIST_HEAD(&m->list_head); |
2479 | ||
45c6ceb5 | 2480 | m->hdr.tid = 0; |
31b8006e | 2481 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2482 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2483 | m->hdr.version = 0; | |
31b8006e SW |
2484 | m->hdr.front_len = cpu_to_le32(front_len); |
2485 | m->hdr.middle_len = 0; | |
bb257664 SW |
2486 | m->hdr.data_len = 0; |
2487 | m->hdr.data_off = 0; | |
45c6ceb5 | 2488 | m->hdr.reserved = 0; |
31b8006e SW |
2489 | m->footer.front_crc = 0; |
2490 | m->footer.middle_crc = 0; | |
2491 | m->footer.data_crc = 0; | |
45c6ceb5 | 2492 | m->footer.flags = 0; |
31b8006e SW |
2493 | m->front_max = front_len; |
2494 | m->front_is_vmalloc = false; | |
2495 | m->more_to_follow = false; | |
c0d5f9db | 2496 | m->ack_stamp = 0; |
31b8006e SW |
2497 | m->pool = NULL; |
2498 | ||
ca20892d HC |
2499 | /* middle */ |
2500 | m->middle = NULL; | |
2501 | ||
2502 | /* data */ | |
2503 | m->nr_pages = 0; | |
2504 | m->page_alignment = 0; | |
2505 | m->pages = NULL; | |
2506 | m->pagelist = NULL; | |
2507 | m->bio = NULL; | |
2508 | m->bio_iter = NULL; | |
2509 | m->bio_seg = 0; | |
2510 | m->trail = NULL; | |
2511 | ||
31b8006e SW |
2512 | /* front */ |
2513 | if (front_len) { | |
2514 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2515 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2516 | PAGE_KERNEL); |
2517 | m->front_is_vmalloc = true; | |
2518 | } else { | |
34d23762 | 2519 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2520 | } |
2521 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2522 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2523 | front_len); |
2524 | goto out2; | |
2525 | } | |
2526 | } else { | |
2527 | m->front.iov_base = NULL; | |
2528 | } | |
2529 | m->front.iov_len = front_len; | |
2530 | ||
bb257664 | 2531 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2532 | return m; |
2533 | ||
2534 | out2: | |
2535 | ceph_msg_put(m); | |
2536 | out: | |
b61c2763 SW |
2537 | if (!can_fail) { |
2538 | pr_err("msg_new can't create type %d front %d\n", type, | |
2539 | front_len); | |
f0ed1b7c | 2540 | WARN_ON(1); |
b61c2763 SW |
2541 | } else { |
2542 | dout("msg_new can't create type %d front %d\n", type, | |
2543 | front_len); | |
2544 | } | |
a79832f2 | 2545 | return NULL; |
31b8006e | 2546 | } |
3d14c5d2 | 2547 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2548 | |
31b8006e SW |
2549 | /* |
2550 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2551 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2552 | * per-type header in the front and then gracefully fail (i.e., | |
2553 | * propagate the error to the caller based on info in the front) when | |
2554 | * the middle is too large. | |
2555 | */ | |
2450418c | 2556 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2557 | { |
2558 | int type = le16_to_cpu(msg->hdr.type); | |
2559 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2560 | ||
2561 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2562 | ceph_msg_type_name(type), middle_len); | |
2563 | BUG_ON(!middle_len); | |
2564 | BUG_ON(msg->middle); | |
2565 | ||
b6c1d5b8 | 2566 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2567 | if (!msg->middle) |
2568 | return -ENOMEM; | |
2569 | return 0; | |
2570 | } | |
2571 | ||
2450418c YS |
2572 | /* |
2573 | * Generic message allocator, for incoming messages. | |
2574 | */ | |
2575 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, | |
2576 | struct ceph_msg_header *hdr, | |
2577 | int *skip) | |
2578 | { | |
2579 | int type = le16_to_cpu(hdr->type); | |
2580 | int front_len = le32_to_cpu(hdr->front_len); | |
2581 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2582 | struct ceph_msg *msg = NULL; | |
2583 | int ret; | |
2584 | ||
2585 | if (con->ops->alloc_msg) { | |
0547a9b3 | 2586 | mutex_unlock(&con->mutex); |
2450418c | 2587 | msg = con->ops->alloc_msg(con, hdr, skip); |
0547a9b3 | 2588 | mutex_lock(&con->mutex); |
a79832f2 | 2589 | if (!msg || *skip) |
2450418c YS |
2590 | return NULL; |
2591 | } | |
2592 | if (!msg) { | |
2593 | *skip = 0; | |
b61c2763 | 2594 | msg = ceph_msg_new(type, front_len, GFP_NOFS, false); |
2450418c YS |
2595 | if (!msg) { |
2596 | pr_err("unable to allocate msg type %d len %d\n", | |
2597 | type, front_len); | |
a79832f2 | 2598 | return NULL; |
2450418c | 2599 | } |
c5c6b19d | 2600 | msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2601 | } |
9d7f0f13 | 2602 | memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2603 | |
bb257664 | 2604 | if (middle_len && !msg->middle) { |
2450418c | 2605 | ret = ceph_alloc_middle(con, msg); |
2450418c YS |
2606 | if (ret < 0) { |
2607 | ceph_msg_put(msg); | |
a79832f2 | 2608 | return NULL; |
2450418c YS |
2609 | } |
2610 | } | |
9d7f0f13 | 2611 | |
2450418c YS |
2612 | return msg; |
2613 | } | |
2614 | ||
31b8006e SW |
2615 | |
2616 | /* | |
2617 | * Free a generically kmalloc'd message. | |
2618 | */ | |
2619 | void ceph_msg_kfree(struct ceph_msg *m) | |
2620 | { | |
2621 | dout("msg_kfree %p\n", m); | |
2622 | if (m->front_is_vmalloc) | |
2623 | vfree(m->front.iov_base); | |
2624 | else | |
2625 | kfree(m->front.iov_base); | |
2626 | kfree(m); | |
2627 | } | |
2628 | ||
2629 | /* | |
2630 | * Drop a msg ref. Destroy as needed. | |
2631 | */ | |
c2e552e7 SW |
2632 | void ceph_msg_last_put(struct kref *kref) |
2633 | { | |
2634 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2635 | |
c2e552e7 SW |
2636 | dout("ceph_msg_put last one on %p\n", m); |
2637 | WARN_ON(!list_empty(&m->list_head)); | |
2638 | ||
2639 | /* drop middle, data, if any */ | |
2640 | if (m->middle) { | |
2641 | ceph_buffer_put(m->middle); | |
2642 | m->middle = NULL; | |
31b8006e | 2643 | } |
c2e552e7 SW |
2644 | m->nr_pages = 0; |
2645 | m->pages = NULL; | |
2646 | ||
58bb3b37 SW |
2647 | if (m->pagelist) { |
2648 | ceph_pagelist_release(m->pagelist); | |
2649 | kfree(m->pagelist); | |
2650 | m->pagelist = NULL; | |
2651 | } | |
2652 | ||
68b4476b YS |
2653 | m->trail = NULL; |
2654 | ||
c2e552e7 SW |
2655 | if (m->pool) |
2656 | ceph_msgpool_put(m->pool, m); | |
2657 | else | |
2658 | ceph_msg_kfree(m); | |
31b8006e | 2659 | } |
3d14c5d2 | 2660 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2661 | |
2662 | void ceph_msg_dump(struct ceph_msg *msg) | |
2663 | { | |
2664 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2665 | msg->front_max, msg->nr_pages); | |
2666 | print_hex_dump(KERN_DEBUG, "header: ", | |
2667 | DUMP_PREFIX_OFFSET, 16, 1, | |
2668 | &msg->hdr, sizeof(msg->hdr), true); | |
2669 | print_hex_dump(KERN_DEBUG, " front: ", | |
2670 | DUMP_PREFIX_OFFSET, 16, 1, | |
2671 | msg->front.iov_base, msg->front.iov_len, true); | |
2672 | if (msg->middle) | |
2673 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2674 | DUMP_PREFIX_OFFSET, 16, 1, | |
2675 | msg->middle->vec.iov_base, | |
2676 | msg->middle->vec.iov_len, true); | |
2677 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2678 | DUMP_PREFIX_OFFSET, 16, 1, | |
2679 | &msg->footer, sizeof(msg->footer), true); | |
2680 | } | |
3d14c5d2 | 2681 | EXPORT_SYMBOL(ceph_msg_dump); |