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 | ||
bc18f4b1 AE |
32 | /* |
33 | * We track the state of the socket on a given connection using | |
34 | * values defined below. The transition to a new socket state is | |
35 | * handled by a function which verifies we aren't coming from an | |
36 | * unexpected state. | |
37 | * | |
38 | * -------- | |
39 | * | NEW* | transient initial state | |
40 | * -------- | |
41 | * | con_sock_state_init() | |
42 | * v | |
43 | * ---------- | |
44 | * | CLOSED | initialized, but no socket (and no | |
45 | * ---------- TCP connection) | |
46 | * ^ \ | |
47 | * | \ con_sock_state_connecting() | |
48 | * | ---------------------- | |
49 | * | \ | |
50 | * + con_sock_state_closed() \ | |
fbb85a47 SW |
51 | * |+--------------------------- \ |
52 | * | \ \ \ | |
53 | * | ----------- \ \ | |
54 | * | | CLOSING | socket event; \ \ | |
55 | * | ----------- await close \ \ | |
56 | * | ^ \ | | |
57 | * | | \ | | |
58 | * | + con_sock_state_closing() \ | | |
59 | * | / \ | | | |
60 | * | / --------------- | | | |
61 | * | / \ v v | |
bc18f4b1 AE |
62 | * | / -------------- |
63 | * | / -----------------| CONNECTING | socket created, TCP | |
64 | * | | / -------------- connect initiated | |
65 | * | | | con_sock_state_connected() | |
66 | * | | v | |
67 | * ------------- | |
68 | * | CONNECTED | TCP connection established | |
69 | * ------------- | |
70 | * | |
71 | * State values for ceph_connection->sock_state; NEW is assumed to be 0. | |
72 | */ | |
ce2c8903 AE |
73 | |
74 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ | |
75 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ | |
76 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ | |
77 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ | |
78 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ | |
79 | ||
31b8006e SW |
80 | /* static tag bytes (protocol control messages) */ |
81 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
82 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
83 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
84 | ||
a6a5349d SW |
85 | #ifdef CONFIG_LOCKDEP |
86 | static struct lock_class_key socket_class; | |
87 | #endif | |
88 | ||
84495f49 AE |
89 | /* |
90 | * When skipping (ignoring) a block of input we read it into a "skip | |
91 | * buffer," which is this many bytes in size. | |
92 | */ | |
93 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
94 | |
95 | static void queue_con(struct ceph_connection *con); | |
96 | static void con_work(struct work_struct *); | |
97 | static void ceph_fault(struct ceph_connection *con); | |
98 | ||
31b8006e | 99 | /* |
f64a9317 AE |
100 | * Nicely render a sockaddr as a string. An array of formatted |
101 | * strings is used, to approximate reentrancy. | |
31b8006e | 102 | */ |
f64a9317 AE |
103 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
104 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
105 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
106 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
107 | ||
108 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
109 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 110 | |
57666519 | 111 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 112 | |
3d14c5d2 | 113 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
114 | { |
115 | int i; | |
116 | char *s; | |
99f0f3b2 AE |
117 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
118 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 119 | |
f64a9317 | 120 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
121 | s = addr_str[i]; |
122 | ||
123 | switch (ss->ss_family) { | |
124 | case AF_INET: | |
bd406145 AE |
125 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
126 | ntohs(in4->sin_port)); | |
31b8006e SW |
127 | break; |
128 | ||
129 | case AF_INET6: | |
bd406145 AE |
130 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
131 | ntohs(in6->sin6_port)); | |
31b8006e SW |
132 | break; |
133 | ||
134 | default: | |
d3002b97 AE |
135 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
136 | ss->ss_family); | |
31b8006e SW |
137 | } |
138 | ||
139 | return s; | |
140 | } | |
3d14c5d2 | 141 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 142 | |
63f2d211 SW |
143 | static void encode_my_addr(struct ceph_messenger *msgr) |
144 | { | |
145 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
146 | ceph_encode_addr(&msgr->my_enc_addr); | |
147 | } | |
148 | ||
31b8006e SW |
149 | /* |
150 | * work queue for all reading and writing to/from the socket. | |
151 | */ | |
e0f43c94 | 152 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 153 | |
6173d1f0 AE |
154 | void _ceph_msgr_exit(void) |
155 | { | |
d3002b97 | 156 | if (ceph_msgr_wq) { |
6173d1f0 | 157 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
158 | ceph_msgr_wq = NULL; |
159 | } | |
6173d1f0 | 160 | |
6173d1f0 AE |
161 | BUG_ON(zero_page == NULL); |
162 | kunmap(zero_page); | |
163 | page_cache_release(zero_page); | |
164 | zero_page = NULL; | |
165 | } | |
166 | ||
3d14c5d2 | 167 | int ceph_msgr_init(void) |
31b8006e | 168 | { |
57666519 AE |
169 | BUG_ON(zero_page != NULL); |
170 | zero_page = ZERO_PAGE(0); | |
171 | page_cache_get(zero_page); | |
172 | ||
f363e45f | 173 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0); |
6173d1f0 AE |
174 | if (ceph_msgr_wq) |
175 | return 0; | |
57666519 | 176 | |
6173d1f0 AE |
177 | pr_err("msgr_init failed to create workqueue\n"); |
178 | _ceph_msgr_exit(); | |
57666519 | 179 | |
6173d1f0 | 180 | return -ENOMEM; |
31b8006e | 181 | } |
3d14c5d2 | 182 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
183 | |
184 | void ceph_msgr_exit(void) | |
185 | { | |
57666519 | 186 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 187 | |
6173d1f0 | 188 | _ceph_msgr_exit(); |
31b8006e | 189 | } |
3d14c5d2 | 190 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 191 | |
cd84db6e | 192 | void ceph_msgr_flush(void) |
a922d38f SW |
193 | { |
194 | flush_workqueue(ceph_msgr_wq); | |
195 | } | |
3d14c5d2 | 196 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 197 | |
ce2c8903 AE |
198 | /* Connection socket state transition functions */ |
199 | ||
200 | static void con_sock_state_init(struct ceph_connection *con) | |
201 | { | |
202 | int old_state; | |
203 | ||
204 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
205 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
206 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
207 | } | |
208 | ||
209 | static void con_sock_state_connecting(struct ceph_connection *con) | |
210 | { | |
211 | int old_state; | |
212 | ||
213 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
214 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
215 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
216 | } | |
217 | ||
218 | static void con_sock_state_connected(struct ceph_connection *con) | |
219 | { | |
220 | int old_state; | |
221 | ||
222 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
223 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
224 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
225 | } | |
226 | ||
227 | static void con_sock_state_closing(struct ceph_connection *con) | |
228 | { | |
229 | int old_state; | |
230 | ||
231 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
232 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
233 | old_state != CON_SOCK_STATE_CONNECTED && | |
234 | old_state != CON_SOCK_STATE_CLOSING)) | |
235 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
236 | } | |
237 | ||
238 | static void con_sock_state_closed(struct ceph_connection *con) | |
239 | { | |
240 | int old_state; | |
241 | ||
242 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
243 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 SW |
244 | old_state != CON_SOCK_STATE_CLOSING && |
245 | old_state != CON_SOCK_STATE_CONNECTING)) | |
ce2c8903 AE |
246 | printk("%s: unexpected old state %d\n", __func__, old_state); |
247 | } | |
a922d38f | 248 | |
31b8006e SW |
249 | /* |
250 | * socket callback functions | |
251 | */ | |
252 | ||
253 | /* data available on socket, or listen socket received a connect */ | |
327800bd | 254 | static void ceph_sock_data_ready(struct sock *sk, int count_unused) |
31b8006e | 255 | { |
bd406145 | 256 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
257 | if (atomic_read(&con->msgr->stopping)) { |
258 | return; | |
259 | } | |
bd406145 | 260 | |
31b8006e | 261 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 262 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
263 | con, con->state); |
264 | queue_con(con); | |
265 | } | |
266 | } | |
267 | ||
268 | /* socket has buffer space for writing */ | |
327800bd | 269 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 270 | { |
d3002b97 | 271 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 272 | |
182fac26 JS |
273 | /* only queue to workqueue if there is data we want to write, |
274 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 275 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
276 | * doesn't get called again until try_write() fills the socket |
277 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
278 | * and net/core/stream.c:sk_stream_write_space(). | |
279 | */ | |
928443cd | 280 | if (test_bit(WRITE_PENDING, &con->flags)) { |
182fac26 | 281 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
327800bd | 282 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
283 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
284 | queue_con(con); | |
285 | } | |
31b8006e | 286 | } else { |
327800bd | 287 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 288 | } |
31b8006e SW |
289 | } |
290 | ||
291 | /* socket's state has changed */ | |
327800bd | 292 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 293 | { |
bd406145 | 294 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 295 | |
327800bd | 296 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
297 | con, con->state, sk->sk_state); |
298 | ||
299 | if (test_bit(CLOSED, &con->state)) | |
300 | return; | |
301 | ||
302 | switch (sk->sk_state) { | |
303 | case TCP_CLOSE: | |
327800bd | 304 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 305 | case TCP_CLOSE_WAIT: |
327800bd | 306 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 307 | con_sock_state_closing(con); |
d65c9e0b AE |
308 | set_bit(SOCK_CLOSED, &con->flags); |
309 | queue_con(con); | |
31b8006e SW |
310 | break; |
311 | case TCP_ESTABLISHED: | |
327800bd | 312 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 313 | con_sock_state_connected(con); |
31b8006e SW |
314 | queue_con(con); |
315 | break; | |
d3002b97 AE |
316 | default: /* Everything else is uninteresting */ |
317 | break; | |
31b8006e SW |
318 | } |
319 | } | |
320 | ||
321 | /* | |
322 | * set up socket callbacks | |
323 | */ | |
324 | static void set_sock_callbacks(struct socket *sock, | |
325 | struct ceph_connection *con) | |
326 | { | |
327 | struct sock *sk = sock->sk; | |
bd406145 | 328 | sk->sk_user_data = con; |
327800bd AE |
329 | sk->sk_data_ready = ceph_sock_data_ready; |
330 | sk->sk_write_space = ceph_sock_write_space; | |
331 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
332 | } |
333 | ||
334 | ||
335 | /* | |
336 | * socket helpers | |
337 | */ | |
338 | ||
339 | /* | |
340 | * initiate connection to a remote socket. | |
341 | */ | |
41617d0c | 342 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 343 | { |
f91d3471 | 344 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
345 | struct socket *sock; |
346 | int ret; | |
347 | ||
348 | BUG_ON(con->sock); | |
f91d3471 SW |
349 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
350 | IPPROTO_TCP, &sock); | |
31b8006e | 351 | if (ret) |
41617d0c | 352 | return ret; |
31b8006e SW |
353 | sock->sk->sk_allocation = GFP_NOFS; |
354 | ||
a6a5349d SW |
355 | #ifdef CONFIG_LOCKDEP |
356 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
357 | #endif | |
358 | ||
31b8006e SW |
359 | set_sock_callbacks(sock, con); |
360 | ||
3d14c5d2 | 361 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 362 | |
89a86be0 | 363 | con_sock_state_connecting(con); |
f91d3471 SW |
364 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
365 | O_NONBLOCK); | |
31b8006e SW |
366 | if (ret == -EINPROGRESS) { |
367 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 368 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 369 | sock->sk->sk_state); |
a5bc3129 | 370 | } else if (ret < 0) { |
31b8006e | 371 | pr_err("connect %s error %d\n", |
3d14c5d2 | 372 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 373 | sock_release(sock); |
31b8006e | 374 | con->error_msg = "connect error"; |
31b8006e | 375 | |
41617d0c | 376 | return ret; |
a5bc3129 AE |
377 | } |
378 | con->sock = sock; | |
41617d0c | 379 | return 0; |
31b8006e SW |
380 | } |
381 | ||
382 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
383 | { | |
384 | struct kvec iov = {buf, len}; | |
385 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 386 | int r; |
31b8006e | 387 | |
98bdb0aa SW |
388 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
389 | if (r == -EAGAIN) | |
390 | r = 0; | |
391 | return r; | |
31b8006e SW |
392 | } |
393 | ||
394 | /* | |
395 | * write something. @more is true if caller will be sending more data | |
396 | * shortly. | |
397 | */ | |
398 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
399 | size_t kvlen, size_t len, int more) | |
400 | { | |
401 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 402 | int r; |
31b8006e SW |
403 | |
404 | if (more) | |
405 | msg.msg_flags |= MSG_MORE; | |
406 | else | |
407 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
408 | ||
42961d23 SW |
409 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
410 | if (r == -EAGAIN) | |
411 | r = 0; | |
412 | return r; | |
31b8006e SW |
413 | } |
414 | ||
31739139 AE |
415 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
416 | int offset, size_t size, int more) | |
417 | { | |
418 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
419 | int ret; | |
420 | ||
421 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
422 | if (ret == -EAGAIN) | |
423 | ret = 0; | |
424 | ||
425 | return ret; | |
426 | } | |
427 | ||
31b8006e SW |
428 | |
429 | /* | |
430 | * Shutdown/close the socket for the given connection. | |
431 | */ | |
432 | static int con_close_socket(struct ceph_connection *con) | |
433 | { | |
434 | int rc; | |
435 | ||
436 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
437 | if (!con->sock) | |
438 | return 0; | |
31b8006e SW |
439 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); |
440 | sock_release(con->sock); | |
441 | con->sock = NULL; | |
456ea468 AE |
442 | |
443 | /* | |
444 | * Forcibly clear the SOCK_CLOSE flag. It gets set | |
445 | * independent of the connection mutex, and we could have | |
446 | * received a socket close event before we had the chance to | |
447 | * shut the socket down. | |
448 | */ | |
a8d00e3c | 449 | clear_bit(SOCK_CLOSED, &con->flags); |
ce2c8903 | 450 | con_sock_state_closed(con); |
31b8006e SW |
451 | return rc; |
452 | } | |
453 | ||
454 | /* | |
455 | * Reset a connection. Discard all incoming and outgoing messages | |
456 | * and clear *_seq state. | |
457 | */ | |
458 | static void ceph_msg_remove(struct ceph_msg *msg) | |
459 | { | |
460 | list_del_init(&msg->list_head); | |
38941f80 | 461 | BUG_ON(msg->con == NULL); |
36eb71aa | 462 | msg->con->ops->put(msg->con); |
38941f80 AE |
463 | msg->con = NULL; |
464 | ||
31b8006e SW |
465 | ceph_msg_put(msg); |
466 | } | |
467 | static void ceph_msg_remove_list(struct list_head *head) | |
468 | { | |
469 | while (!list_empty(head)) { | |
470 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
471 | list_head); | |
472 | ceph_msg_remove(msg); | |
473 | } | |
474 | } | |
475 | ||
476 | static void reset_connection(struct ceph_connection *con) | |
477 | { | |
478 | /* reset connection, out_queue, msg_ and connect_seq */ | |
479 | /* discard existing out_queue and msg_seq */ | |
31b8006e SW |
480 | ceph_msg_remove_list(&con->out_queue); |
481 | ceph_msg_remove_list(&con->out_sent); | |
482 | ||
cf3e5c40 | 483 | if (con->in_msg) { |
38941f80 AE |
484 | BUG_ON(con->in_msg->con != con); |
485 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
486 | ceph_msg_put(con->in_msg); |
487 | con->in_msg = NULL; | |
36eb71aa | 488 | con->ops->put(con); |
cf3e5c40 SW |
489 | } |
490 | ||
31b8006e SW |
491 | con->connect_seq = 0; |
492 | con->out_seq = 0; | |
c86a2930 SW |
493 | if (con->out_msg) { |
494 | ceph_msg_put(con->out_msg); | |
495 | con->out_msg = NULL; | |
496 | } | |
31b8006e | 497 | con->in_seq = 0; |
0e0d5e0c | 498 | con->in_seq_acked = 0; |
31b8006e SW |
499 | } |
500 | ||
501 | /* | |
502 | * mark a peer down. drop any open connections. | |
503 | */ | |
504 | void ceph_con_close(struct ceph_connection *con) | |
505 | { | |
3d14c5d2 YS |
506 | dout("con_close %p peer %s\n", con, |
507 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
a5988c49 | 508 | clear_bit(NEGOTIATING, &con->state); |
bb9e6bba | 509 | clear_bit(CONNECTING, &con->state); |
e27947c7 | 510 | clear_bit(CONNECTED, &con->state); |
31b8006e | 511 | clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */ |
a5988c49 AE |
512 | set_bit(CLOSED, &con->state); |
513 | ||
928443cd AE |
514 | clear_bit(LOSSYTX, &con->flags); /* so we retry next connect */ |
515 | clear_bit(KEEPALIVE_PENDING, &con->flags); | |
516 | clear_bit(WRITE_PENDING, &con->flags); | |
a5988c49 | 517 | |
ec302645 | 518 | mutex_lock(&con->mutex); |
31b8006e | 519 | reset_connection(con); |
6f2bc3ff | 520 | con->peer_global_seq = 0; |
91e45ce3 | 521 | cancel_delayed_work(&con->work); |
ec302645 | 522 | mutex_unlock(&con->mutex); |
31b8006e SW |
523 | queue_con(con); |
524 | } | |
3d14c5d2 | 525 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 526 | |
31b8006e SW |
527 | /* |
528 | * Reopen a closed connection, with a new peer address. | |
529 | */ | |
b7a9e5dd SW |
530 | void ceph_con_open(struct ceph_connection *con, |
531 | __u8 entity_type, __u64 entity_num, | |
532 | struct ceph_entity_addr *addr) | |
31b8006e | 533 | { |
3d14c5d2 | 534 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
31b8006e | 535 | set_bit(OPENING, &con->state); |
a5988c49 AE |
536 | WARN_ON(!test_and_clear_bit(CLOSED, &con->state)); |
537 | ||
b7a9e5dd SW |
538 | con->peer_name.type = (__u8) entity_type; |
539 | con->peer_name.num = cpu_to_le64(entity_num); | |
540 | ||
31b8006e | 541 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 542 | con->delay = 0; /* reset backoff memory */ |
31b8006e SW |
543 | queue_con(con); |
544 | } | |
3d14c5d2 | 545 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 546 | |
87b315a5 SW |
547 | /* |
548 | * return true if this connection ever successfully opened | |
549 | */ | |
550 | bool ceph_con_opened(struct ceph_connection *con) | |
551 | { | |
552 | return con->connect_seq > 0; | |
553 | } | |
554 | ||
31b8006e SW |
555 | /* |
556 | * initialize a new connection. | |
557 | */ | |
1bfd89f4 AE |
558 | void ceph_con_init(struct ceph_connection *con, void *private, |
559 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 560 | struct ceph_messenger *msgr) |
31b8006e SW |
561 | { |
562 | dout("con_init %p\n", con); | |
563 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
564 | con->private = private; |
565 | con->ops = ops; | |
31b8006e | 566 | con->msgr = msgr; |
ce2c8903 AE |
567 | |
568 | con_sock_state_init(con); | |
569 | ||
ec302645 | 570 | mutex_init(&con->mutex); |
31b8006e SW |
571 | INIT_LIST_HEAD(&con->out_queue); |
572 | INIT_LIST_HEAD(&con->out_sent); | |
573 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 AE |
574 | |
575 | set_bit(CLOSED, &con->state); | |
31b8006e | 576 | } |
3d14c5d2 | 577 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
578 | |
579 | ||
580 | /* | |
581 | * We maintain a global counter to order connection attempts. Get | |
582 | * a unique seq greater than @gt. | |
583 | */ | |
584 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
585 | { | |
586 | u32 ret; | |
587 | ||
588 | spin_lock(&msgr->global_seq_lock); | |
589 | if (msgr->global_seq < gt) | |
590 | msgr->global_seq = gt; | |
591 | ret = ++msgr->global_seq; | |
592 | spin_unlock(&msgr->global_seq_lock); | |
593 | return ret; | |
594 | } | |
595 | ||
e2200423 | 596 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
597 | { |
598 | con->out_kvec_left = 0; | |
599 | con->out_kvec_bytes = 0; | |
600 | con->out_kvec_cur = &con->out_kvec[0]; | |
601 | } | |
602 | ||
e2200423 | 603 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
604 | size_t size, void *data) |
605 | { | |
606 | int index; | |
607 | ||
608 | index = con->out_kvec_left; | |
609 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
610 | ||
611 | con->out_kvec[index].iov_len = size; | |
612 | con->out_kvec[index].iov_base = data; | |
613 | con->out_kvec_left++; | |
614 | con->out_kvec_bytes += size; | |
615 | } | |
31b8006e | 616 | |
df6ad1f9 AE |
617 | #ifdef CONFIG_BLOCK |
618 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
619 | { | |
620 | if (!bio) { | |
621 | *iter = NULL; | |
622 | *seg = 0; | |
623 | return; | |
624 | } | |
625 | *iter = bio; | |
626 | *seg = bio->bi_idx; | |
627 | } | |
628 | ||
629 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
630 | { | |
631 | if (*bio_iter == NULL) | |
632 | return; | |
633 | ||
634 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
635 | ||
636 | (*seg)++; | |
637 | if (*seg == (*bio_iter)->bi_vcnt) | |
638 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
639 | } | |
640 | #endif | |
641 | ||
739c905b AE |
642 | static void prepare_write_message_data(struct ceph_connection *con) |
643 | { | |
644 | struct ceph_msg *msg = con->out_msg; | |
645 | ||
646 | BUG_ON(!msg); | |
647 | BUG_ON(!msg->hdr.data_len); | |
648 | ||
649 | /* initialize page iterator */ | |
650 | con->out_msg_pos.page = 0; | |
651 | if (msg->pages) | |
652 | con->out_msg_pos.page_pos = msg->page_alignment; | |
653 | else | |
654 | con->out_msg_pos.page_pos = 0; | |
572c588e | 655 | #ifdef CONFIG_BLOCK |
abdaa6a8 | 656 | if (msg->bio) |
572c588e AE |
657 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); |
658 | #endif | |
739c905b AE |
659 | con->out_msg_pos.data_pos = 0; |
660 | con->out_msg_pos.did_page_crc = false; | |
661 | con->out_more = 1; /* data + footer will follow */ | |
662 | } | |
663 | ||
31b8006e SW |
664 | /* |
665 | * Prepare footer for currently outgoing message, and finish things | |
666 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
667 | */ | |
859eb799 | 668 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
669 | { |
670 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 671 | int v = con->out_kvec_left; |
31b8006e | 672 | |
fd154f3c AE |
673 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
674 | ||
31b8006e SW |
675 | dout("prepare_write_message_footer %p\n", con); |
676 | con->out_kvec_is_msg = true; | |
677 | con->out_kvec[v].iov_base = &m->footer; | |
678 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
679 | con->out_kvec_bytes += sizeof(m->footer); | |
680 | con->out_kvec_left++; | |
681 | con->out_more = m->more_to_follow; | |
c86a2930 | 682 | con->out_msg_done = true; |
31b8006e SW |
683 | } |
684 | ||
685 | /* | |
686 | * Prepare headers for the next outgoing message. | |
687 | */ | |
688 | static void prepare_write_message(struct ceph_connection *con) | |
689 | { | |
690 | struct ceph_msg *m; | |
a9a0c51a | 691 | u32 crc; |
31b8006e | 692 | |
e2200423 | 693 | con_out_kvec_reset(con); |
31b8006e | 694 | con->out_kvec_is_msg = true; |
c86a2930 | 695 | con->out_msg_done = false; |
31b8006e SW |
696 | |
697 | /* Sneak an ack in there first? If we can get it into the same | |
698 | * TCP packet that's a good thing. */ | |
699 | if (con->in_seq > con->in_seq_acked) { | |
700 | con->in_seq_acked = con->in_seq; | |
e2200423 | 701 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 702 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 703 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 704 | &con->out_temp_ack); |
31b8006e SW |
705 | } |
706 | ||
38941f80 | 707 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 708 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 709 | con->out_msg = m; |
38941f80 | 710 | BUG_ON(m->con != con); |
4cf9d544 SW |
711 | |
712 | /* put message on sent list */ | |
713 | ceph_msg_get(m); | |
714 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 715 | |
e84346b7 SW |
716 | /* |
717 | * only assign outgoing seq # if we haven't sent this message | |
718 | * yet. if it is requeued, resend with it's original seq. | |
719 | */ | |
720 | if (m->needs_out_seq) { | |
721 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
722 | m->needs_out_seq = false; | |
723 | } | |
31b8006e SW |
724 | |
725 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
726 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
727 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
728 | le32_to_cpu(m->hdr.data_len), | |
729 | m->nr_pages); | |
730 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); | |
731 | ||
732 | /* tag + hdr + front + middle */ | |
e2200423 AE |
733 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
734 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
735 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 736 | |
31b8006e | 737 | if (m->middle) |
e2200423 | 738 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 739 | m->middle->vec.iov_base); |
31b8006e SW |
740 | |
741 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
742 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
743 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 744 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
745 | |
746 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
747 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
748 | if (m->middle) { | |
749 | crc = crc32c(0, m->middle->vec.iov_base, | |
750 | m->middle->vec.iov_len); | |
751 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
752 | } else | |
31b8006e | 753 | con->out_msg->footer.middle_crc = 0; |
739c905b | 754 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
755 | le32_to_cpu(con->out_msg->footer.front_crc), |
756 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
757 | ||
758 | /* is there a data payload? */ | |
739c905b AE |
759 | con->out_msg->footer.data_crc = 0; |
760 | if (m->hdr.data_len) | |
761 | prepare_write_message_data(con); | |
762 | else | |
31b8006e | 763 | /* no, queue up footer too and be done */ |
859eb799 | 764 | prepare_write_message_footer(con); |
31b8006e | 765 | |
928443cd | 766 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
767 | } |
768 | ||
769 | /* | |
770 | * Prepare an ack. | |
771 | */ | |
772 | static void prepare_write_ack(struct ceph_connection *con) | |
773 | { | |
774 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
775 | con->in_seq_acked, con->in_seq); | |
776 | con->in_seq_acked = con->in_seq; | |
777 | ||
e2200423 | 778 | con_out_kvec_reset(con); |
859eb799 | 779 | |
e2200423 | 780 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 781 | |
31b8006e | 782 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 783 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
784 | &con->out_temp_ack); |
785 | ||
31b8006e | 786 | con->out_more = 1; /* more will follow.. eventually.. */ |
928443cd | 787 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
788 | } |
789 | ||
790 | /* | |
791 | * Prepare to write keepalive byte. | |
792 | */ | |
793 | static void prepare_write_keepalive(struct ceph_connection *con) | |
794 | { | |
795 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
796 | con_out_kvec_reset(con); |
797 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
928443cd | 798 | set_bit(WRITE_PENDING, &con->flags); |
31b8006e SW |
799 | } |
800 | ||
801 | /* | |
802 | * Connection negotiation. | |
803 | */ | |
804 | ||
dac1e716 AE |
805 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
806 | int *auth_proto) | |
4e7a5dcd | 807 | { |
a3530df3 | 808 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
809 | |
810 | if (!con->ops->get_authorizer) { | |
811 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
812 | con->out_connect.authorizer_len = 0; | |
813 | ||
729796be | 814 | return NULL; |
b1c6b980 AE |
815 | } |
816 | ||
817 | /* Can't hold the mutex while getting authorizer */ | |
4e7a5dcd | 818 | |
ec302645 | 819 | mutex_unlock(&con->mutex); |
b1c6b980 | 820 | |
dac1e716 | 821 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
8f43fb53 | 822 | |
ec302645 | 823 | mutex_lock(&con->mutex); |
4e7a5dcd | 824 | |
a3530df3 | 825 | if (IS_ERR(auth)) |
729796be | 826 | return auth; |
928443cd | 827 | if (test_bit(CLOSED, &con->state) || test_bit(OPENING, &con->flags)) |
729796be | 828 | return ERR_PTR(-EAGAIN); |
0da5d703 | 829 | |
8f43fb53 AE |
830 | con->auth_reply_buf = auth->authorizer_reply_buf; |
831 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
832 | ||
859eb799 | 833 | |
729796be | 834 | return auth; |
4e7a5dcd SW |
835 | } |
836 | ||
31b8006e SW |
837 | /* |
838 | * We connected to a peer and are saying hello. | |
839 | */ | |
e825a66d | 840 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 841 | { |
e2200423 AE |
842 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
843 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 844 | &con->msgr->my_enc_addr); |
eed0ef2c | 845 | |
eed0ef2c | 846 | con->out_more = 0; |
928443cd | 847 | set_bit(WRITE_PENDING, &con->flags); |
eed0ef2c SW |
848 | } |
849 | ||
e825a66d | 850 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 851 | { |
95c96174 | 852 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 853 | int proto; |
dac1e716 | 854 | int auth_proto; |
729796be | 855 | struct ceph_auth_handshake *auth; |
31b8006e SW |
856 | |
857 | switch (con->peer_name.type) { | |
858 | case CEPH_ENTITY_TYPE_MON: | |
859 | proto = CEPH_MONC_PROTOCOL; | |
860 | break; | |
861 | case CEPH_ENTITY_TYPE_OSD: | |
862 | proto = CEPH_OSDC_PROTOCOL; | |
863 | break; | |
864 | case CEPH_ENTITY_TYPE_MDS: | |
865 | proto = CEPH_MDSC_PROTOCOL; | |
866 | break; | |
867 | default: | |
868 | BUG(); | |
869 | } | |
870 | ||
871 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
872 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 873 | |
e825a66d | 874 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
875 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
876 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
877 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
878 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
879 | con->out_connect.flags = 0; | |
31b8006e | 880 | |
dac1e716 AE |
881 | auth_proto = CEPH_AUTH_UNKNOWN; |
882 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
883 | if (IS_ERR(auth)) |
884 | return PTR_ERR(auth); | |
3da54776 | 885 | |
dac1e716 | 886 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
887 | con->out_connect.authorizer_len = auth ? |
888 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
889 | ||
ab166d5a | 890 | con_out_kvec_reset(con); |
e2200423 | 891 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
892 | &con->out_connect); |
893 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 894 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 895 | auth->authorizer_buf); |
859eb799 | 896 | |
31b8006e | 897 | con->out_more = 0; |
928443cd | 898 | set_bit(WRITE_PENDING, &con->flags); |
4e7a5dcd | 899 | |
e10c758e | 900 | return 0; |
31b8006e SW |
901 | } |
902 | ||
31b8006e SW |
903 | /* |
904 | * write as much of pending kvecs to the socket as we can. | |
905 | * 1 -> done | |
906 | * 0 -> socket full, but more to do | |
907 | * <0 -> error | |
908 | */ | |
909 | static int write_partial_kvec(struct ceph_connection *con) | |
910 | { | |
911 | int ret; | |
912 | ||
913 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
914 | while (con->out_kvec_bytes > 0) { | |
915 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
916 | con->out_kvec_left, con->out_kvec_bytes, | |
917 | con->out_more); | |
918 | if (ret <= 0) | |
919 | goto out; | |
920 | con->out_kvec_bytes -= ret; | |
921 | if (con->out_kvec_bytes == 0) | |
922 | break; /* done */ | |
f42299e6 AE |
923 | |
924 | /* account for full iov entries consumed */ | |
925 | while (ret >= con->out_kvec_cur->iov_len) { | |
926 | BUG_ON(!con->out_kvec_left); | |
927 | ret -= con->out_kvec_cur->iov_len; | |
928 | con->out_kvec_cur++; | |
929 | con->out_kvec_left--; | |
930 | } | |
931 | /* and for a partially-consumed entry */ | |
932 | if (ret) { | |
933 | con->out_kvec_cur->iov_len -= ret; | |
934 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
935 | } |
936 | } | |
937 | con->out_kvec_left = 0; | |
938 | con->out_kvec_is_msg = false; | |
939 | ret = 1; | |
940 | out: | |
941 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
942 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
943 | return ret; /* done! */ | |
944 | } | |
945 | ||
84ca8fc8 AE |
946 | static void out_msg_pos_next(struct ceph_connection *con, struct page *page, |
947 | size_t len, size_t sent, bool in_trail) | |
948 | { | |
949 | struct ceph_msg *msg = con->out_msg; | |
950 | ||
951 | BUG_ON(!msg); | |
952 | BUG_ON(!sent); | |
953 | ||
954 | con->out_msg_pos.data_pos += sent; | |
955 | con->out_msg_pos.page_pos += sent; | |
5821bd8c AE |
956 | if (sent < len) |
957 | return; | |
958 | ||
959 | BUG_ON(sent != len); | |
960 | con->out_msg_pos.page_pos = 0; | |
961 | con->out_msg_pos.page++; | |
962 | con->out_msg_pos.did_page_crc = false; | |
963 | if (in_trail) | |
964 | list_move_tail(&page->lru, | |
965 | &msg->trail->head); | |
966 | else if (msg->pagelist) | |
967 | list_move_tail(&page->lru, | |
968 | &msg->pagelist->head); | |
84ca8fc8 | 969 | #ifdef CONFIG_BLOCK |
5821bd8c AE |
970 | else if (msg->bio) |
971 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
84ca8fc8 | 972 | #endif |
84ca8fc8 AE |
973 | } |
974 | ||
31b8006e SW |
975 | /* |
976 | * Write as much message data payload as we can. If we finish, queue | |
977 | * up the footer. | |
978 | * 1 -> done, footer is now queued in out_kvec[]. | |
979 | * 0 -> socket full, but more to do | |
980 | * <0 -> error | |
981 | */ | |
982 | static int write_partial_msg_pages(struct ceph_connection *con) | |
983 | { | |
984 | struct ceph_msg *msg = con->out_msg; | |
95c96174 | 985 | unsigned int data_len = le32_to_cpu(msg->hdr.data_len); |
31b8006e | 986 | size_t len; |
37675b0f | 987 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 988 | int ret; |
68b4476b | 989 | int total_max_write; |
84ca8fc8 | 990 | bool in_trail = false; |
5821bd8c AE |
991 | const size_t trail_len = (msg->trail ? msg->trail->length : 0); |
992 | const size_t trail_off = data_len - trail_len; | |
31b8006e SW |
993 | |
994 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
84ca8fc8 | 995 | con, msg, con->out_msg_pos.page, msg->nr_pages, |
31b8006e SW |
996 | con->out_msg_pos.page_pos); |
997 | ||
5821bd8c AE |
998 | /* |
999 | * Iterate through each page that contains data to be | |
1000 | * written, and send as much as possible for each. | |
1001 | * | |
1002 | * If we are calculating the data crc (the default), we will | |
1003 | * need to map the page. If we have no pages, they have | |
1004 | * been revoked, so use the zero page. | |
1005 | */ | |
68b4476b | 1006 | while (data_len > con->out_msg_pos.data_pos) { |
31b8006e | 1007 | struct page *page = NULL; |
68b4476b | 1008 | int max_write = PAGE_SIZE; |
9bd19663 | 1009 | int bio_offset = 0; |
68b4476b | 1010 | |
5821bd8c AE |
1011 | in_trail = in_trail || con->out_msg_pos.data_pos >= trail_off; |
1012 | if (!in_trail) | |
1013 | total_max_write = trail_off - con->out_msg_pos.data_pos; | |
68b4476b | 1014 | |
5821bd8c | 1015 | if (in_trail) { |
68b4476b YS |
1016 | total_max_write = data_len - con->out_msg_pos.data_pos; |
1017 | ||
1018 | page = list_first_entry(&msg->trail->head, | |
1019 | struct page, lru); | |
68b4476b | 1020 | } else if (msg->pages) { |
31b8006e | 1021 | page = msg->pages[con->out_msg_pos.page]; |
58bb3b37 SW |
1022 | } else if (msg->pagelist) { |
1023 | page = list_first_entry(&msg->pagelist->head, | |
1024 | struct page, lru); | |
68b4476b YS |
1025 | #ifdef CONFIG_BLOCK |
1026 | } else if (msg->bio) { | |
1027 | struct bio_vec *bv; | |
1028 | ||
1029 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
1030 | page = bv->bv_page; | |
9bd19663 | 1031 | bio_offset = bv->bv_offset; |
68b4476b YS |
1032 | max_write = bv->bv_len; |
1033 | #endif | |
31b8006e | 1034 | } else { |
57666519 | 1035 | page = zero_page; |
31b8006e | 1036 | } |
68b4476b YS |
1037 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
1038 | total_max_write); | |
1039 | ||
37675b0f | 1040 | if (do_datacrc && !con->out_msg_pos.did_page_crc) { |
9bd19663 | 1041 | void *base; |
5821bd8c | 1042 | u32 crc = le32_to_cpu(msg->footer.data_crc); |
8d63e318 | 1043 | char *kaddr; |
31b8006e | 1044 | |
8d63e318 | 1045 | kaddr = kmap(page); |
31b8006e | 1046 | BUG_ON(kaddr == NULL); |
9bd19663 | 1047 | base = kaddr + con->out_msg_pos.page_pos + bio_offset; |
5821bd8c | 1048 | crc = crc32c(crc, base, len); |
84ca8fc8 | 1049 | msg->footer.data_crc = cpu_to_le32(crc); |
bca064d2 | 1050 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 1051 | } |
e36b13cc | 1052 | ret = ceph_tcp_sendpage(con->sock, page, |
9bd19663 | 1053 | con->out_msg_pos.page_pos + bio_offset, |
e36b13cc | 1054 | len, 1); |
31b8006e | 1055 | |
0cdf9e60 | 1056 | if (do_datacrc) |
31b8006e SW |
1057 | kunmap(page); |
1058 | ||
1059 | if (ret <= 0) | |
1060 | goto out; | |
1061 | ||
84ca8fc8 | 1062 | out_msg_pos_next(con, page, len, (size_t) ret, in_trail); |
31b8006e SW |
1063 | } |
1064 | ||
1065 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
1066 | ||
1067 | /* prepare and queue up footer, too */ | |
37675b0f | 1068 | if (!do_datacrc) |
84ca8fc8 | 1069 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1070 | con_out_kvec_reset(con); |
859eb799 | 1071 | prepare_write_message_footer(con); |
31b8006e SW |
1072 | ret = 1; |
1073 | out: | |
1074 | return ret; | |
1075 | } | |
1076 | ||
1077 | /* | |
1078 | * write some zeros | |
1079 | */ | |
1080 | static int write_partial_skip(struct ceph_connection *con) | |
1081 | { | |
1082 | int ret; | |
1083 | ||
1084 | while (con->out_skip > 0) { | |
31739139 | 1085 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1086 | |
31739139 | 1087 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1); |
31b8006e SW |
1088 | if (ret <= 0) |
1089 | goto out; | |
1090 | con->out_skip -= ret; | |
1091 | } | |
1092 | ret = 1; | |
1093 | out: | |
1094 | return ret; | |
1095 | } | |
1096 | ||
1097 | /* | |
1098 | * Prepare to read connection handshake, or an ack. | |
1099 | */ | |
eed0ef2c SW |
1100 | static void prepare_read_banner(struct ceph_connection *con) |
1101 | { | |
1102 | dout("prepare_read_banner %p\n", con); | |
1103 | con->in_base_pos = 0; | |
1104 | } | |
1105 | ||
31b8006e SW |
1106 | static void prepare_read_connect(struct ceph_connection *con) |
1107 | { | |
1108 | dout("prepare_read_connect %p\n", con); | |
1109 | con->in_base_pos = 0; | |
1110 | } | |
1111 | ||
1112 | static void prepare_read_ack(struct ceph_connection *con) | |
1113 | { | |
1114 | dout("prepare_read_ack %p\n", con); | |
1115 | con->in_base_pos = 0; | |
1116 | } | |
1117 | ||
1118 | static void prepare_read_tag(struct ceph_connection *con) | |
1119 | { | |
1120 | dout("prepare_read_tag %p\n", con); | |
1121 | con->in_base_pos = 0; | |
1122 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * Prepare to read a message. | |
1127 | */ | |
1128 | static int prepare_read_message(struct ceph_connection *con) | |
1129 | { | |
1130 | dout("prepare_read_message %p\n", con); | |
1131 | BUG_ON(con->in_msg != NULL); | |
1132 | con->in_base_pos = 0; | |
1133 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1134 | return 0; | |
1135 | } | |
1136 | ||
1137 | ||
1138 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1139 | int end, int size, void *object) |
31b8006e | 1140 | { |
e6cee71f AE |
1141 | while (con->in_base_pos < end) { |
1142 | int left = end - con->in_base_pos; | |
31b8006e SW |
1143 | int have = size - left; |
1144 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1145 | if (ret <= 0) | |
1146 | return ret; | |
1147 | con->in_base_pos += ret; | |
1148 | } | |
1149 | return 1; | |
1150 | } | |
1151 | ||
1152 | ||
1153 | /* | |
1154 | * Read all or part of the connect-side handshake on a new connection | |
1155 | */ | |
eed0ef2c | 1156 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1157 | { |
fd51653f AE |
1158 | int size; |
1159 | int end; | |
1160 | int ret; | |
31b8006e | 1161 | |
eed0ef2c | 1162 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1163 | |
1164 | /* peer's banner */ | |
fd51653f AE |
1165 | size = strlen(CEPH_BANNER); |
1166 | end = size; | |
1167 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1168 | if (ret <= 0) |
1169 | goto out; | |
fd51653f AE |
1170 | |
1171 | size = sizeof (con->actual_peer_addr); | |
1172 | end += size; | |
1173 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1174 | if (ret <= 0) |
1175 | goto out; | |
fd51653f AE |
1176 | |
1177 | size = sizeof (con->peer_addr_for_me); | |
1178 | end += size; | |
1179 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1180 | if (ret <= 0) |
1181 | goto out; | |
fd51653f | 1182 | |
eed0ef2c SW |
1183 | out: |
1184 | return ret; | |
1185 | } | |
1186 | ||
1187 | static int read_partial_connect(struct ceph_connection *con) | |
1188 | { | |
fd51653f AE |
1189 | int size; |
1190 | int end; | |
1191 | int ret; | |
eed0ef2c SW |
1192 | |
1193 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1194 | ||
fd51653f AE |
1195 | size = sizeof (con->in_reply); |
1196 | end = size; | |
1197 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1198 | if (ret <= 0) |
1199 | goto out; | |
fd51653f AE |
1200 | |
1201 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1202 | end += size; | |
1203 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1204 | if (ret <= 0) |
1205 | goto out; | |
31b8006e | 1206 | |
4e7a5dcd SW |
1207 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1208 | con, (int)con->in_reply.tag, | |
1209 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1210 | le32_to_cpu(con->in_reply.global_seq)); |
1211 | out: | |
1212 | return ret; | |
eed0ef2c | 1213 | |
31b8006e SW |
1214 | } |
1215 | ||
1216 | /* | |
1217 | * Verify the hello banner looks okay. | |
1218 | */ | |
1219 | static int verify_hello(struct ceph_connection *con) | |
1220 | { | |
1221 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1222 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1223 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1224 | con->error_msg = "protocol error, bad banner"; |
1225 | return -1; | |
1226 | } | |
1227 | return 0; | |
1228 | } | |
1229 | ||
1230 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1231 | { | |
1232 | switch (ss->ss_family) { | |
1233 | case AF_INET: | |
1234 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1235 | case AF_INET6: | |
1236 | return | |
1237 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1238 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1239 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1240 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1241 | } | |
1242 | return false; | |
1243 | } | |
1244 | ||
1245 | static int addr_port(struct sockaddr_storage *ss) | |
1246 | { | |
1247 | switch (ss->ss_family) { | |
1248 | case AF_INET: | |
f28bcfbe | 1249 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1250 | case AF_INET6: |
f28bcfbe | 1251 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1252 | } |
1253 | return 0; | |
1254 | } | |
1255 | ||
1256 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1257 | { | |
1258 | switch (ss->ss_family) { | |
1259 | case AF_INET: | |
1260 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1261 | break; |
31b8006e SW |
1262 | case AF_INET6: |
1263 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1264 | break; |
31b8006e SW |
1265 | } |
1266 | } | |
1267 | ||
ee3b56f2 NW |
1268 | /* |
1269 | * Unlike other *_pton function semantics, zero indicates success. | |
1270 | */ | |
1271 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1272 | char delim, const char **ipend) | |
1273 | { | |
99f0f3b2 AE |
1274 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1275 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1276 | |
1277 | memset(ss, 0, sizeof(*ss)); | |
1278 | ||
1279 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1280 | ss->ss_family = AF_INET; | |
1281 | return 0; | |
1282 | } | |
1283 | ||
1284 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1285 | ss->ss_family = AF_INET6; | |
1286 | return 0; | |
1287 | } | |
1288 | ||
1289 | return -EINVAL; | |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * Extract hostname string and resolve using kernel DNS facility. | |
1294 | */ | |
1295 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1296 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1297 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1298 | { | |
1299 | const char *end, *delim_p; | |
1300 | char *colon_p, *ip_addr = NULL; | |
1301 | int ip_len, ret; | |
1302 | ||
1303 | /* | |
1304 | * The end of the hostname occurs immediately preceding the delimiter or | |
1305 | * the port marker (':') where the delimiter takes precedence. | |
1306 | */ | |
1307 | delim_p = memchr(name, delim, namelen); | |
1308 | colon_p = memchr(name, ':', namelen); | |
1309 | ||
1310 | if (delim_p && colon_p) | |
1311 | end = delim_p < colon_p ? delim_p : colon_p; | |
1312 | else if (!delim_p && colon_p) | |
1313 | end = colon_p; | |
1314 | else { | |
1315 | end = delim_p; | |
1316 | if (!end) /* case: hostname:/ */ | |
1317 | end = name + namelen; | |
1318 | } | |
1319 | ||
1320 | if (end <= name) | |
1321 | return -EINVAL; | |
1322 | ||
1323 | /* do dns_resolve upcall */ | |
1324 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1325 | if (ip_len > 0) | |
1326 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1327 | else | |
1328 | ret = -ESRCH; | |
1329 | ||
1330 | kfree(ip_addr); | |
1331 | ||
1332 | *ipend = end; | |
1333 | ||
1334 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1335 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1336 | ||
1337 | return ret; | |
1338 | } | |
1339 | #else | |
1340 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1341 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1342 | { | |
1343 | return -EINVAL; | |
1344 | } | |
1345 | #endif | |
1346 | ||
1347 | /* | |
1348 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1349 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1350 | */ | |
1351 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1352 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1353 | { | |
1354 | int ret; | |
1355 | ||
1356 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1357 | if (ret) | |
1358 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1359 | ||
1360 | return ret; | |
1361 | } | |
1362 | ||
31b8006e SW |
1363 | /* |
1364 | * Parse an ip[:port] list into an addr array. Use the default | |
1365 | * monitor port if a port isn't specified. | |
1366 | */ | |
1367 | int ceph_parse_ips(const char *c, const char *end, | |
1368 | struct ceph_entity_addr *addr, | |
1369 | int max_count, int *count) | |
1370 | { | |
ee3b56f2 | 1371 | int i, ret = -EINVAL; |
31b8006e SW |
1372 | const char *p = c; |
1373 | ||
1374 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1375 | for (i = 0; i < max_count; i++) { | |
1376 | const char *ipend; | |
1377 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1378 | int port; |
39139f64 SW |
1379 | char delim = ','; |
1380 | ||
1381 | if (*p == '[') { | |
1382 | delim = ']'; | |
1383 | p++; | |
1384 | } | |
31b8006e | 1385 | |
ee3b56f2 NW |
1386 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1387 | if (ret) | |
31b8006e | 1388 | goto bad; |
ee3b56f2 NW |
1389 | ret = -EINVAL; |
1390 | ||
31b8006e SW |
1391 | p = ipend; |
1392 | ||
39139f64 SW |
1393 | if (delim == ']') { |
1394 | if (*p != ']') { | |
1395 | dout("missing matching ']'\n"); | |
1396 | goto bad; | |
1397 | } | |
1398 | p++; | |
1399 | } | |
1400 | ||
31b8006e SW |
1401 | /* port? */ |
1402 | if (p < end && *p == ':') { | |
1403 | port = 0; | |
1404 | p++; | |
1405 | while (p < end && *p >= '0' && *p <= '9') { | |
1406 | port = (port * 10) + (*p - '0'); | |
1407 | p++; | |
1408 | } | |
1409 | if (port > 65535 || port == 0) | |
1410 | goto bad; | |
1411 | } else { | |
1412 | port = CEPH_MON_PORT; | |
1413 | } | |
1414 | ||
1415 | addr_set_port(ss, port); | |
1416 | ||
3d14c5d2 | 1417 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1418 | |
1419 | if (p == end) | |
1420 | break; | |
1421 | if (*p != ',') | |
1422 | goto bad; | |
1423 | p++; | |
1424 | } | |
1425 | ||
1426 | if (p != end) | |
1427 | goto bad; | |
1428 | ||
1429 | if (count) | |
1430 | *count = i + 1; | |
1431 | return 0; | |
1432 | ||
1433 | bad: | |
39139f64 | 1434 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1435 | return ret; |
31b8006e | 1436 | } |
3d14c5d2 | 1437 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1438 | |
eed0ef2c | 1439 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1440 | { |
eed0ef2c | 1441 | dout("process_banner on %p\n", con); |
31b8006e SW |
1442 | |
1443 | if (verify_hello(con) < 0) | |
1444 | return -1; | |
1445 | ||
63f2d211 SW |
1446 | ceph_decode_addr(&con->actual_peer_addr); |
1447 | ceph_decode_addr(&con->peer_addr_for_me); | |
1448 | ||
31b8006e SW |
1449 | /* |
1450 | * Make sure the other end is who we wanted. note that the other | |
1451 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1452 | * them the benefit of the doubt. | |
1453 | */ | |
103e2d3a SW |
1454 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1455 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1456 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1457 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1458 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1459 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1460 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1461 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1462 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1463 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1464 | return -1; |
1465 | } | |
1466 | ||
1467 | /* | |
1468 | * did we learn our address? | |
1469 | */ | |
1470 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1471 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1472 | ||
1473 | memcpy(&con->msgr->inst.addr.in_addr, | |
1474 | &con->peer_addr_for_me.in_addr, | |
1475 | sizeof(con->peer_addr_for_me.in_addr)); | |
1476 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1477 | encode_my_addr(con->msgr); |
eed0ef2c | 1478 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1479 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1480 | } |
1481 | ||
eed0ef2c SW |
1482 | return 0; |
1483 | } | |
1484 | ||
04a419f9 SW |
1485 | static void fail_protocol(struct ceph_connection *con) |
1486 | { | |
1487 | reset_connection(con); | |
1488 | set_bit(CLOSED, &con->state); /* in case there's queued work */ | |
04a419f9 SW |
1489 | } |
1490 | ||
eed0ef2c SW |
1491 | static int process_connect(struct ceph_connection *con) |
1492 | { | |
3d14c5d2 YS |
1493 | u64 sup_feat = con->msgr->supported_features; |
1494 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1495 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1496 | int ret; |
04a419f9 | 1497 | |
eed0ef2c SW |
1498 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1499 | ||
31b8006e | 1500 | switch (con->in_reply.tag) { |
04a419f9 SW |
1501 | case CEPH_MSGR_TAG_FEATURES: |
1502 | pr_err("%s%lld %s feature set mismatch," | |
1503 | " my %llx < server's %llx, missing %llx\n", | |
1504 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1505 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1506 | sup_feat, server_feat, server_feat & ~sup_feat); |
1507 | con->error_msg = "missing required protocol features"; | |
1508 | fail_protocol(con); | |
1509 | return -1; | |
1510 | ||
31b8006e | 1511 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1512 | pr_err("%s%lld %s protocol version mismatch," |
1513 | " my %d != server's %d\n", | |
1514 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1515 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1516 | le32_to_cpu(con->out_connect.protocol_version), |
1517 | le32_to_cpu(con->in_reply.protocol_version)); | |
1518 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1519 | fail_protocol(con); |
31b8006e SW |
1520 | return -1; |
1521 | ||
4e7a5dcd SW |
1522 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1523 | con->auth_retry++; | |
1524 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1525 | con->auth_retry); | |
1526 | if (con->auth_retry == 2) { | |
1527 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1528 | return -1; |
1529 | } | |
1530 | con->auth_retry = 1; | |
e825a66d | 1531 | ret = prepare_write_connect(con); |
0da5d703 SW |
1532 | if (ret < 0) |
1533 | return ret; | |
63733a0f | 1534 | prepare_read_connect(con); |
4e7a5dcd | 1535 | break; |
31b8006e SW |
1536 | |
1537 | case CEPH_MSGR_TAG_RESETSESSION: | |
1538 | /* | |
1539 | * If we connected with a large connect_seq but the peer | |
1540 | * has no record of a session with us (no connection, or | |
1541 | * connect_seq == 0), they will send RESETSESION to indicate | |
1542 | * that they must have reset their session, and may have | |
1543 | * dropped messages. | |
1544 | */ | |
1545 | dout("process_connect got RESET peer seq %u\n", | |
a16cb1f7 | 1546 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
1547 | pr_err("%s%lld %s connection reset\n", |
1548 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1549 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1550 | reset_connection(con); |
5a0f8fdd AE |
1551 | ret = prepare_write_connect(con); |
1552 | if (ret < 0) | |
1553 | return ret; | |
31b8006e SW |
1554 | prepare_read_connect(con); |
1555 | ||
1556 | /* Tell ceph about it. */ | |
ec302645 | 1557 | mutex_unlock(&con->mutex); |
31b8006e SW |
1558 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1559 | if (con->ops->peer_reset) | |
1560 | con->ops->peer_reset(con); | |
ec302645 | 1561 | mutex_lock(&con->mutex); |
0da5d703 SW |
1562 | if (test_bit(CLOSED, &con->state) || |
1563 | test_bit(OPENING, &con->state)) | |
1564 | return -EAGAIN; | |
31b8006e SW |
1565 | break; |
1566 | ||
1567 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1568 | /* | |
1569 | * If we sent a smaller connect_seq than the peer has, try | |
1570 | * again with a larger value. | |
1571 | */ | |
a16cb1f7 | 1572 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 1573 | le32_to_cpu(con->out_connect.connect_seq), |
a16cb1f7 SW |
1574 | le32_to_cpu(con->in_reply.connect_seq)); |
1575 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
5a0f8fdd AE |
1576 | ret = prepare_write_connect(con); |
1577 | if (ret < 0) | |
1578 | return ret; | |
31b8006e SW |
1579 | prepare_read_connect(con); |
1580 | break; | |
1581 | ||
1582 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1583 | /* | |
1584 | * If we sent a smaller global_seq than the peer has, try | |
1585 | * again with a larger value. | |
1586 | */ | |
eed0ef2c | 1587 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 1588 | con->peer_global_seq, |
a16cb1f7 | 1589 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 1590 | get_global_seq(con->msgr, |
a16cb1f7 | 1591 | le32_to_cpu(con->in_reply.global_seq)); |
5a0f8fdd AE |
1592 | ret = prepare_write_connect(con); |
1593 | if (ret < 0) | |
1594 | return ret; | |
31b8006e SW |
1595 | prepare_read_connect(con); |
1596 | break; | |
1597 | ||
1598 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1599 | if (req_feat & ~server_feat) { |
1600 | pr_err("%s%lld %s protocol feature mismatch," | |
1601 | " my required %llx > server's %llx, need %llx\n", | |
1602 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1603 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1604 | req_feat, server_feat, req_feat & ~server_feat); |
1605 | con->error_msg = "missing required protocol features"; | |
1606 | fail_protocol(con); | |
1607 | return -1; | |
1608 | } | |
3ec50d18 | 1609 | clear_bit(NEGOTIATING, &con->state); |
e27947c7 | 1610 | set_bit(CONNECTED, &con->state); |
31b8006e SW |
1611 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1612 | con->connect_seq++; | |
aba558e2 | 1613 | con->peer_features = server_feat; |
31b8006e SW |
1614 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1615 | con->peer_global_seq, | |
1616 | le32_to_cpu(con->in_reply.connect_seq), | |
1617 | con->connect_seq); | |
1618 | WARN_ON(con->connect_seq != | |
1619 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1620 | |
1621 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
928443cd | 1622 | set_bit(LOSSYTX, &con->flags); |
92ac41d0 | 1623 | |
31b8006e SW |
1624 | prepare_read_tag(con); |
1625 | break; | |
1626 | ||
1627 | case CEPH_MSGR_TAG_WAIT: | |
1628 | /* | |
1629 | * If there is a connection race (we are opening | |
1630 | * connections to each other), one of us may just have | |
1631 | * to WAIT. This shouldn't happen if we are the | |
1632 | * client. | |
1633 | */ | |
04177882 SW |
1634 | pr_err("process_connect got WAIT as client\n"); |
1635 | con->error_msg = "protocol error, got WAIT as client"; | |
1636 | return -1; | |
31b8006e SW |
1637 | |
1638 | default: | |
1639 | pr_err("connect protocol error, will retry\n"); | |
1640 | con->error_msg = "protocol error, garbage tag during connect"; | |
1641 | return -1; | |
1642 | } | |
1643 | return 0; | |
1644 | } | |
1645 | ||
1646 | ||
1647 | /* | |
1648 | * read (part of) an ack | |
1649 | */ | |
1650 | static int read_partial_ack(struct ceph_connection *con) | |
1651 | { | |
fd51653f AE |
1652 | int size = sizeof (con->in_temp_ack); |
1653 | int end = size; | |
1654 | ||
1655 | return read_partial(con, end, size, &con->in_temp_ack); | |
31b8006e SW |
1656 | } |
1657 | ||
1658 | ||
1659 | /* | |
1660 | * We can finally discard anything that's been acked. | |
1661 | */ | |
1662 | static void process_ack(struct ceph_connection *con) | |
1663 | { | |
1664 | struct ceph_msg *m; | |
1665 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1666 | u64 seq; | |
1667 | ||
31b8006e SW |
1668 | while (!list_empty(&con->out_sent)) { |
1669 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1670 | list_head); | |
1671 | seq = le64_to_cpu(m->hdr.seq); | |
1672 | if (seq > ack) | |
1673 | break; | |
1674 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1675 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1676 | m->ack_stamp = jiffies; |
31b8006e SW |
1677 | ceph_msg_remove(m); |
1678 | } | |
31b8006e SW |
1679 | prepare_read_tag(con); |
1680 | } | |
1681 | ||
1682 | ||
1683 | ||
1684 | ||
2450418c | 1685 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1686 | struct kvec *section, |
1687 | unsigned int sec_len, u32 *crc) | |
2450418c | 1688 | { |
68b4476b | 1689 | int ret, left; |
2450418c YS |
1690 | |
1691 | BUG_ON(!section); | |
1692 | ||
1693 | while (section->iov_len < sec_len) { | |
1694 | BUG_ON(section->iov_base == NULL); | |
1695 | left = sec_len - section->iov_len; | |
1696 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1697 | section->iov_len, left); | |
1698 | if (ret <= 0) | |
1699 | return ret; | |
1700 | section->iov_len += ret; | |
2450418c | 1701 | } |
fe3ad593 AE |
1702 | if (section->iov_len == sec_len) |
1703 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1704 | |
2450418c YS |
1705 | return 1; |
1706 | } | |
31b8006e | 1707 | |
1c20f2d2 AE |
1708 | static bool ceph_con_in_msg_alloc(struct ceph_connection *con, |
1709 | struct ceph_msg_header *hdr); | |
68b4476b YS |
1710 | |
1711 | ||
1712 | static int read_partial_message_pages(struct ceph_connection *con, | |
1713 | struct page **pages, | |
95c96174 | 1714 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1715 | { |
1716 | void *p; | |
1717 | int ret; | |
1718 | int left; | |
1719 | ||
1720 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1721 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1722 | /* (page) data */ | |
1723 | BUG_ON(pages == NULL); | |
1724 | p = kmap(pages[con->in_msg_pos.page]); | |
1725 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1726 | left); | |
bca064d2 | 1727 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1728 | con->in_data_crc = |
1729 | crc32c(con->in_data_crc, | |
1730 | p + con->in_msg_pos.page_pos, ret); | |
1731 | kunmap(pages[con->in_msg_pos.page]); | |
1732 | if (ret <= 0) | |
1733 | return ret; | |
1734 | con->in_msg_pos.data_pos += ret; | |
1735 | con->in_msg_pos.page_pos += ret; | |
1736 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1737 | con->in_msg_pos.page_pos = 0; | |
1738 | con->in_msg_pos.page++; | |
1739 | } | |
1740 | ||
1741 | return ret; | |
1742 | } | |
1743 | ||
1744 | #ifdef CONFIG_BLOCK | |
1745 | static int read_partial_message_bio(struct ceph_connection *con, | |
1746 | struct bio **bio_iter, int *bio_seg, | |
95c96174 | 1747 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1748 | { |
1749 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1750 | void *p; | |
1751 | int ret, left; | |
1752 | ||
68b4476b YS |
1753 | left = min((int)(data_len - con->in_msg_pos.data_pos), |
1754 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1755 | ||
1756 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1757 | ||
1758 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1759 | left); | |
bca064d2 | 1760 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1761 | con->in_data_crc = |
1762 | crc32c(con->in_data_crc, | |
1763 | p + con->in_msg_pos.page_pos, ret); | |
1764 | kunmap(bv->bv_page); | |
1765 | if (ret <= 0) | |
1766 | return ret; | |
1767 | con->in_msg_pos.data_pos += ret; | |
1768 | con->in_msg_pos.page_pos += ret; | |
1769 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1770 | con->in_msg_pos.page_pos = 0; | |
1771 | iter_bio_next(bio_iter, bio_seg); | |
1772 | } | |
1773 | ||
1774 | return ret; | |
1775 | } | |
1776 | #endif | |
1777 | ||
31b8006e SW |
1778 | /* |
1779 | * read (part of) a message. | |
1780 | */ | |
1781 | static int read_partial_message(struct ceph_connection *con) | |
1782 | { | |
1783 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1784 | int size; |
1785 | int end; | |
31b8006e | 1786 | int ret; |
95c96174 | 1787 | unsigned int front_len, middle_len, data_len; |
37675b0f | 1788 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 1789 | u64 seq; |
fe3ad593 | 1790 | u32 crc; |
31b8006e SW |
1791 | |
1792 | dout("read_partial_message con %p msg %p\n", con, m); | |
1793 | ||
1794 | /* header */ | |
fd51653f AE |
1795 | size = sizeof (con->in_hdr); |
1796 | end = size; | |
1797 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1798 | if (ret <= 0) |
1799 | return ret; | |
fe3ad593 AE |
1800 | |
1801 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1802 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1803 | pr_err("read_partial_message bad hdr " | |
1804 | " crc %u != expected %u\n", | |
1805 | crc, con->in_hdr.crc); | |
1806 | return -EBADMSG; | |
1807 | } | |
1808 | ||
31b8006e SW |
1809 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1810 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1811 | return -EIO; | |
1812 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1813 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1814 | return -EIO; | |
1815 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1816 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1817 | return -EIO; | |
1818 | ||
ae18756b SW |
1819 | /* verify seq# */ |
1820 | seq = le64_to_cpu(con->in_hdr.seq); | |
1821 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1822 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1823 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1824 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1825 | seq, con->in_seq + 1); |
1826 | con->in_base_pos = -front_len - middle_len - data_len - | |
1827 | sizeof(m->footer); | |
1828 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1829 | return 0; |
1830 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1831 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1832 | seq, con->in_seq + 1); | |
1833 | con->error_msg = "bad message sequence # for incoming message"; | |
1834 | return -EBADMSG; | |
1835 | } | |
1836 | ||
31b8006e SW |
1837 | /* allocate message? */ |
1838 | if (!con->in_msg) { | |
1839 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1840 | con->in_hdr.front_len, con->in_hdr.data_len); | |
1c20f2d2 | 1841 | if (ceph_con_in_msg_alloc(con, &con->in_hdr)) { |
31b8006e | 1842 | /* skip this message */ |
a79832f2 | 1843 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1844 | BUG_ON(con->in_msg); |
31b8006e SW |
1845 | con->in_base_pos = -front_len - middle_len - data_len - |
1846 | sizeof(m->footer); | |
1847 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1848 | con->in_seq++; |
31b8006e SW |
1849 | return 0; |
1850 | } | |
a79832f2 | 1851 | if (!con->in_msg) { |
5b3a4db3 SW |
1852 | con->error_msg = |
1853 | "error allocating memory for incoming message"; | |
a79832f2 | 1854 | return -ENOMEM; |
31b8006e | 1855 | } |
38941f80 AE |
1856 | |
1857 | BUG_ON(con->in_msg->con != con); | |
31b8006e SW |
1858 | m = con->in_msg; |
1859 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1860 | if (m->middle) |
1861 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1862 | |
1863 | con->in_msg_pos.page = 0; | |
68b4476b | 1864 | if (m->pages) |
c5c6b19d | 1865 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1866 | else |
1867 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1868 | con->in_msg_pos.data_pos = 0; |
31b8006e SW |
1869 | } |
1870 | ||
1871 | /* front */ | |
2450418c YS |
1872 | ret = read_partial_message_section(con, &m->front, front_len, |
1873 | &con->in_front_crc); | |
1874 | if (ret <= 0) | |
1875 | return ret; | |
31b8006e SW |
1876 | |
1877 | /* middle */ | |
2450418c | 1878 | if (m->middle) { |
213c99ee SW |
1879 | ret = read_partial_message_section(con, &m->middle->vec, |
1880 | middle_len, | |
2450418c | 1881 | &con->in_middle_crc); |
31b8006e SW |
1882 | if (ret <= 0) |
1883 | return ret; | |
31b8006e | 1884 | } |
68b4476b YS |
1885 | #ifdef CONFIG_BLOCK |
1886 | if (m->bio && !m->bio_iter) | |
1887 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1888 | #endif | |
31b8006e SW |
1889 | |
1890 | /* (page) data */ | |
31b8006e | 1891 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1892 | if (m->pages) { |
1893 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1894 | data_len, do_datacrc); |
68b4476b YS |
1895 | if (ret <= 0) |
1896 | return ret; | |
1897 | #ifdef CONFIG_BLOCK | |
1898 | } else if (m->bio) { | |
1899 | ||
1900 | ret = read_partial_message_bio(con, | |
1901 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1902 | data_len, do_datacrc); |
68b4476b YS |
1903 | if (ret <= 0) |
1904 | return ret; | |
1905 | #endif | |
1906 | } else { | |
1907 | BUG_ON(1); | |
31b8006e SW |
1908 | } |
1909 | } | |
1910 | ||
31b8006e | 1911 | /* footer */ |
fd51653f AE |
1912 | size = sizeof (m->footer); |
1913 | end += size; | |
1914 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
1915 | if (ret <= 0) |
1916 | return ret; | |
1917 | ||
31b8006e SW |
1918 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
1919 | m, front_len, m->footer.front_crc, middle_len, | |
1920 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1921 | ||
1922 | /* crc ok? */ | |
1923 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1924 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1925 | m, con->in_front_crc, m->footer.front_crc); | |
1926 | return -EBADMSG; | |
1927 | } | |
1928 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1929 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1930 | m, con->in_middle_crc, m->footer.middle_crc); | |
1931 | return -EBADMSG; | |
1932 | } | |
bca064d2 | 1933 | if (do_datacrc && |
31b8006e SW |
1934 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
1935 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1936 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1937 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1938 | return -EBADMSG; | |
1939 | } | |
1940 | ||
1941 | return 1; /* done! */ | |
1942 | } | |
1943 | ||
1944 | /* | |
1945 | * Process message. This happens in the worker thread. The callback should | |
1946 | * be careful not to do anything that waits on other incoming messages or it | |
1947 | * may deadlock. | |
1948 | */ | |
1949 | static void process_message(struct ceph_connection *con) | |
1950 | { | |
5e095e8b | 1951 | struct ceph_msg *msg; |
31b8006e | 1952 | |
38941f80 AE |
1953 | BUG_ON(con->in_msg->con != con); |
1954 | con->in_msg->con = NULL; | |
5e095e8b | 1955 | msg = con->in_msg; |
31b8006e | 1956 | con->in_msg = NULL; |
36eb71aa | 1957 | con->ops->put(con); |
31b8006e SW |
1958 | |
1959 | /* if first message, set peer_name */ | |
1960 | if (con->peer_name.type == 0) | |
dbad185d | 1961 | con->peer_name = msg->hdr.src; |
31b8006e | 1962 | |
31b8006e | 1963 | con->in_seq++; |
ec302645 | 1964 | mutex_unlock(&con->mutex); |
31b8006e SW |
1965 | |
1966 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1967 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1968 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1969 | le16_to_cpu(msg->hdr.type), |
1970 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1971 | le32_to_cpu(msg->hdr.front_len), | |
1972 | le32_to_cpu(msg->hdr.data_len), | |
1973 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1974 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1975 | |
1976 | mutex_lock(&con->mutex); | |
31b8006e SW |
1977 | prepare_read_tag(con); |
1978 | } | |
1979 | ||
1980 | ||
1981 | /* | |
1982 | * Write something to the socket. Called in a worker thread when the | |
1983 | * socket appears to be writeable and we have something ready to send. | |
1984 | */ | |
1985 | static int try_write(struct ceph_connection *con) | |
1986 | { | |
31b8006e SW |
1987 | int ret = 1; |
1988 | ||
d59315ca | 1989 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 1990 | |
31b8006e SW |
1991 | more: |
1992 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
1993 | ||
1994 | /* open the socket first? */ | |
1995 | if (con->sock == NULL) { | |
a5988c49 AE |
1996 | set_bit(CONNECTING, &con->state); |
1997 | ||
e2200423 | 1998 | con_out_kvec_reset(con); |
e825a66d | 1999 | prepare_write_banner(con); |
eed0ef2c | 2000 | prepare_read_banner(con); |
31b8006e | 2001 | |
cf3e5c40 | 2002 | BUG_ON(con->in_msg); |
31b8006e SW |
2003 | con->in_tag = CEPH_MSGR_TAG_READY; |
2004 | dout("try_write initiating connect on %p new state %lu\n", | |
2005 | con, con->state); | |
41617d0c AE |
2006 | ret = ceph_tcp_connect(con); |
2007 | if (ret < 0) { | |
31b8006e | 2008 | con->error_msg = "connect error"; |
31b8006e SW |
2009 | goto out; |
2010 | } | |
2011 | } | |
2012 | ||
2013 | more_kvec: | |
2014 | /* kvec data queued? */ | |
2015 | if (con->out_skip) { | |
2016 | ret = write_partial_skip(con); | |
2017 | if (ret <= 0) | |
42961d23 | 2018 | goto out; |
31b8006e SW |
2019 | } |
2020 | if (con->out_kvec_left) { | |
2021 | ret = write_partial_kvec(con); | |
2022 | if (ret <= 0) | |
42961d23 | 2023 | goto out; |
31b8006e SW |
2024 | } |
2025 | ||
2026 | /* msg pages? */ | |
2027 | if (con->out_msg) { | |
c86a2930 SW |
2028 | if (con->out_msg_done) { |
2029 | ceph_msg_put(con->out_msg); | |
2030 | con->out_msg = NULL; /* we're done with this one */ | |
2031 | goto do_next; | |
2032 | } | |
2033 | ||
31b8006e SW |
2034 | ret = write_partial_msg_pages(con); |
2035 | if (ret == 1) | |
2036 | goto more_kvec; /* we need to send the footer, too! */ | |
2037 | if (ret == 0) | |
42961d23 | 2038 | goto out; |
31b8006e SW |
2039 | if (ret < 0) { |
2040 | dout("try_write write_partial_msg_pages err %d\n", | |
2041 | ret); | |
42961d23 | 2042 | goto out; |
31b8006e SW |
2043 | } |
2044 | } | |
2045 | ||
c86a2930 | 2046 | do_next: |
7593af92 AE |
2047 | if (!test_bit(CONNECTING, &con->state) && |
2048 | !test_bit(NEGOTIATING, &con->state)) { | |
31b8006e SW |
2049 | /* is anything else pending? */ |
2050 | if (!list_empty(&con->out_queue)) { | |
2051 | prepare_write_message(con); | |
2052 | goto more; | |
2053 | } | |
2054 | if (con->in_seq > con->in_seq_acked) { | |
2055 | prepare_write_ack(con); | |
2056 | goto more; | |
2057 | } | |
928443cd | 2058 | if (test_and_clear_bit(KEEPALIVE_PENDING, &con->flags)) { |
31b8006e SW |
2059 | prepare_write_keepalive(con); |
2060 | goto more; | |
2061 | } | |
2062 | } | |
2063 | ||
2064 | /* Nothing to do! */ | |
928443cd | 2065 | clear_bit(WRITE_PENDING, &con->flags); |
31b8006e | 2066 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2067 | ret = 0; |
2068 | out: | |
42961d23 | 2069 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2070 | return ret; |
2071 | } | |
2072 | ||
2073 | ||
2074 | ||
2075 | /* | |
2076 | * Read what we can from the socket. | |
2077 | */ | |
2078 | static int try_read(struct ceph_connection *con) | |
2079 | { | |
31b8006e SW |
2080 | int ret = -1; |
2081 | ||
2082 | if (!con->sock) | |
2083 | return 0; | |
2084 | ||
2085 | if (test_bit(STANDBY, &con->state)) | |
2086 | return 0; | |
2087 | ||
2088 | dout("try_read start on %p\n", con); | |
ec302645 | 2089 | |
31b8006e SW |
2090 | more: |
2091 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, | |
2092 | con->in_base_pos); | |
0da5d703 SW |
2093 | |
2094 | /* | |
2095 | * process_connect and process_message drop and re-take | |
2096 | * con->mutex. make sure we handle a racing close or reopen. | |
2097 | */ | |
2098 | if (test_bit(CLOSED, &con->state) || | |
2099 | test_bit(OPENING, &con->state)) { | |
2100 | ret = -EAGAIN; | |
2101 | goto out; | |
2102 | } | |
2103 | ||
31b8006e | 2104 | if (test_bit(CONNECTING, &con->state)) { |
7593af92 AE |
2105 | dout("try_read connecting\n"); |
2106 | ret = read_partial_banner(con); | |
2107 | if (ret <= 0) | |
ab166d5a | 2108 | goto out; |
7593af92 AE |
2109 | ret = process_banner(con); |
2110 | if (ret < 0) | |
2111 | goto out; | |
2112 | ||
2113 | clear_bit(CONNECTING, &con->state); | |
2114 | set_bit(NEGOTIATING, &con->state); | |
2115 | ||
2116 | /* Banner is good, exchange connection info */ | |
2117 | ret = prepare_write_connect(con); | |
2118 | if (ret < 0) | |
2119 | goto out; | |
2120 | prepare_read_connect(con); | |
2121 | ||
2122 | /* Send connection info before awaiting response */ | |
2123 | goto out; | |
2124 | } | |
2125 | ||
2126 | if (test_bit(NEGOTIATING, &con->state)) { | |
2127 | dout("try_read negotiating\n"); | |
31b8006e SW |
2128 | ret = read_partial_connect(con); |
2129 | if (ret <= 0) | |
31b8006e | 2130 | goto out; |
98bdb0aa SW |
2131 | ret = process_connect(con); |
2132 | if (ret < 0) | |
2133 | goto out; | |
31b8006e SW |
2134 | goto more; |
2135 | } | |
2136 | ||
2137 | if (con->in_base_pos < 0) { | |
2138 | /* | |
2139 | * skipping + discarding content. | |
2140 | * | |
2141 | * FIXME: there must be a better way to do this! | |
2142 | */ | |
84495f49 AE |
2143 | static char buf[SKIP_BUF_SIZE]; |
2144 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2145 | ||
31b8006e SW |
2146 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2147 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2148 | if (ret <= 0) | |
98bdb0aa | 2149 | goto out; |
31b8006e SW |
2150 | con->in_base_pos += ret; |
2151 | if (con->in_base_pos) | |
2152 | goto more; | |
2153 | } | |
2154 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2155 | /* | |
2156 | * what's next? | |
2157 | */ | |
2158 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2159 | if (ret <= 0) | |
98bdb0aa | 2160 | goto out; |
31b8006e SW |
2161 | dout("try_read got tag %d\n", (int)con->in_tag); |
2162 | switch (con->in_tag) { | |
2163 | case CEPH_MSGR_TAG_MSG: | |
2164 | prepare_read_message(con); | |
2165 | break; | |
2166 | case CEPH_MSGR_TAG_ACK: | |
2167 | prepare_read_ack(con); | |
2168 | break; | |
2169 | case CEPH_MSGR_TAG_CLOSE: | |
e27947c7 | 2170 | clear_bit(CONNECTED, &con->state); |
31b8006e | 2171 | set_bit(CLOSED, &con->state); /* fixme */ |
98bdb0aa | 2172 | goto out; |
31b8006e SW |
2173 | default: |
2174 | goto bad_tag; | |
2175 | } | |
2176 | } | |
2177 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2178 | ret = read_partial_message(con); | |
2179 | if (ret <= 0) { | |
2180 | switch (ret) { | |
2181 | case -EBADMSG: | |
2182 | con->error_msg = "bad crc"; | |
2183 | ret = -EIO; | |
98bdb0aa | 2184 | break; |
31b8006e SW |
2185 | case -EIO: |
2186 | con->error_msg = "io error"; | |
98bdb0aa | 2187 | break; |
31b8006e | 2188 | } |
98bdb0aa | 2189 | goto out; |
31b8006e SW |
2190 | } |
2191 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2192 | goto more; | |
2193 | process_message(con); | |
2194 | goto more; | |
2195 | } | |
2196 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2197 | ret = read_partial_ack(con); | |
2198 | if (ret <= 0) | |
98bdb0aa | 2199 | goto out; |
31b8006e SW |
2200 | process_ack(con); |
2201 | goto more; | |
2202 | } | |
2203 | ||
31b8006e | 2204 | out: |
98bdb0aa | 2205 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2206 | return ret; |
2207 | ||
2208 | bad_tag: | |
2209 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2210 | con->error_msg = "protocol error, garbage tag"; | |
2211 | ret = -1; | |
2212 | goto out; | |
2213 | } | |
2214 | ||
2215 | ||
2216 | /* | |
2217 | * Atomically queue work on a connection. Bump @con reference to | |
2218 | * avoid races with connection teardown. | |
31b8006e SW |
2219 | */ |
2220 | static void queue_con(struct ceph_connection *con) | |
2221 | { | |
31b8006e SW |
2222 | if (!con->ops->get(con)) { |
2223 | dout("queue_con %p ref count 0\n", con); | |
2224 | return; | |
2225 | } | |
2226 | ||
f363e45f | 2227 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2228 | dout("queue_con %p - already queued\n", con); |
2229 | con->ops->put(con); | |
2230 | } else { | |
2231 | dout("queue_con %p\n", con); | |
2232 | } | |
2233 | } | |
2234 | ||
2235 | /* | |
2236 | * Do some work on a connection. Drop a connection ref when we're done. | |
2237 | */ | |
2238 | static void con_work(struct work_struct *work) | |
2239 | { | |
2240 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2241 | work.work); | |
0da5d703 | 2242 | int ret; |
31b8006e | 2243 | |
9dd4658d | 2244 | mutex_lock(&con->mutex); |
0da5d703 | 2245 | restart: |
188048bc | 2246 | if (test_and_clear_bit(SOCK_CLOSED, &con->flags)) { |
e27947c7 AE |
2247 | if (test_and_clear_bit(CONNECTED, &con->state)) |
2248 | con->error_msg = "socket closed"; | |
7593af92 AE |
2249 | else if (test_and_clear_bit(NEGOTIATING, &con->state)) |
2250 | con->error_msg = "negotiation failed"; | |
2251 | else if (test_and_clear_bit(CONNECTING, &con->state)) | |
188048bc | 2252 | con->error_msg = "connection failed"; |
7593af92 | 2253 | else |
e27947c7 | 2254 | con->error_msg = "unrecognized con state"; |
188048bc AE |
2255 | goto fault; |
2256 | } | |
2257 | ||
928443cd | 2258 | if (test_and_clear_bit(BACKOFF, &con->flags)) { |
60bf8bf8 SW |
2259 | dout("con_work %p backing off\n", con); |
2260 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2261 | round_jiffies_relative(con->delay))) { | |
2262 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2263 | mutex_unlock(&con->mutex); | |
2264 | return; | |
2265 | } else { | |
2266 | con->ops->put(con); | |
2267 | dout("con_work %p FAILED to back off %lu\n", con, | |
2268 | con->delay); | |
2269 | } | |
2270 | } | |
9dd4658d | 2271 | |
e00de341 SW |
2272 | if (test_bit(STANDBY, &con->state)) { |
2273 | dout("con_work %p STANDBY\n", con); | |
2274 | goto done; | |
2275 | } | |
31b8006e SW |
2276 | if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */ |
2277 | dout("con_work CLOSED\n"); | |
2278 | con_close_socket(con); | |
2279 | goto done; | |
2280 | } | |
2281 | if (test_and_clear_bit(OPENING, &con->state)) { | |
2282 | /* reopen w/ new peer */ | |
2283 | dout("con_work OPENING\n"); | |
2284 | con_close_socket(con); | |
2285 | } | |
2286 | ||
0da5d703 SW |
2287 | ret = try_read(con); |
2288 | if (ret == -EAGAIN) | |
2289 | goto restart; | |
2290 | if (ret < 0) | |
2291 | goto fault; | |
2292 | ||
2293 | ret = try_write(con); | |
2294 | if (ret == -EAGAIN) | |
2295 | goto restart; | |
2296 | if (ret < 0) | |
2297 | goto fault; | |
31b8006e SW |
2298 | |
2299 | done: | |
9dd4658d | 2300 | mutex_unlock(&con->mutex); |
9dd4658d | 2301 | done_unlocked: |
31b8006e | 2302 | con->ops->put(con); |
0da5d703 SW |
2303 | return; |
2304 | ||
2305 | fault: | |
2306 | mutex_unlock(&con->mutex); | |
2307 | ceph_fault(con); /* error/fault path */ | |
2308 | goto done_unlocked; | |
31b8006e SW |
2309 | } |
2310 | ||
2311 | ||
2312 | /* | |
2313 | * Generic error/fault handler. A retry mechanism is used with | |
2314 | * exponential backoff | |
2315 | */ | |
2316 | static void ceph_fault(struct ceph_connection *con) | |
2317 | { | |
2318 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2319 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2320 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2321 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2322 | |
928443cd | 2323 | if (test_bit(LOSSYTX, &con->flags)) { |
31b8006e SW |
2324 | dout("fault on LOSSYTX channel\n"); |
2325 | goto out; | |
2326 | } | |
2327 | ||
ec302645 | 2328 | mutex_lock(&con->mutex); |
91e45ce3 SW |
2329 | if (test_bit(CLOSED, &con->state)) |
2330 | goto out_unlock; | |
ec302645 | 2331 | |
31b8006e | 2332 | con_close_socket(con); |
5e095e8b SW |
2333 | |
2334 | if (con->in_msg) { | |
38941f80 AE |
2335 | BUG_ON(con->in_msg->con != con); |
2336 | con->in_msg->con = NULL; | |
5e095e8b SW |
2337 | ceph_msg_put(con->in_msg); |
2338 | con->in_msg = NULL; | |
36eb71aa | 2339 | con->ops->put(con); |
5e095e8b | 2340 | } |
31b8006e | 2341 | |
e80a52d1 SW |
2342 | /* Requeue anything that hasn't been acked */ |
2343 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2344 | |
e76661d0 SW |
2345 | /* If there are no messages queued or keepalive pending, place |
2346 | * the connection in a STANDBY state */ | |
2347 | if (list_empty(&con->out_queue) && | |
928443cd | 2348 | !test_bit(KEEPALIVE_PENDING, &con->flags)) { |
e00de341 | 2349 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
928443cd | 2350 | clear_bit(WRITE_PENDING, &con->flags); |
31b8006e | 2351 | set_bit(STANDBY, &con->state); |
e80a52d1 SW |
2352 | } else { |
2353 | /* retry after a delay. */ | |
2354 | if (con->delay == 0) | |
2355 | con->delay = BASE_DELAY_INTERVAL; | |
2356 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2357 | con->delay *= 2; | |
e80a52d1 SW |
2358 | con->ops->get(con); |
2359 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2360 | round_jiffies_relative(con->delay))) { |
2361 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2362 | } else { | |
e80a52d1 | 2363 | con->ops->put(con); |
60bf8bf8 SW |
2364 | dout("fault failed to queue %p delay %lu, backoff\n", |
2365 | con, con->delay); | |
2366 | /* | |
2367 | * In many cases we see a socket state change | |
2368 | * while con_work is running and end up | |
2369 | * queuing (non-delayed) work, such that we | |
2370 | * can't backoff with a delay. Set a flag so | |
2371 | * that when con_work restarts we schedule the | |
2372 | * delay then. | |
2373 | */ | |
928443cd | 2374 | set_bit(BACKOFF, &con->flags); |
60bf8bf8 | 2375 | } |
31b8006e SW |
2376 | } |
2377 | ||
91e45ce3 SW |
2378 | out_unlock: |
2379 | mutex_unlock(&con->mutex); | |
31b8006e | 2380 | out: |
161fd65a SW |
2381 | /* |
2382 | * in case we faulted due to authentication, invalidate our | |
2383 | * current tickets so that we can get new ones. | |
213c99ee | 2384 | */ |
161fd65a SW |
2385 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2386 | dout("calling invalidate_authorizer()\n"); | |
2387 | con->ops->invalidate_authorizer(con); | |
2388 | } | |
2389 | ||
31b8006e SW |
2390 | if (con->ops->fault) |
2391 | con->ops->fault(con); | |
2392 | } | |
2393 | ||
2394 | ||
2395 | ||
2396 | /* | |
15d9882c | 2397 | * initialize a new messenger instance |
31b8006e | 2398 | */ |
15d9882c AE |
2399 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2400 | struct ceph_entity_addr *myaddr, | |
2401 | u32 supported_features, | |
2402 | u32 required_features, | |
2403 | bool nocrc) | |
31b8006e | 2404 | { |
3d14c5d2 YS |
2405 | msgr->supported_features = supported_features; |
2406 | msgr->required_features = required_features; | |
2407 | ||
31b8006e SW |
2408 | spin_lock_init(&msgr->global_seq_lock); |
2409 | ||
31b8006e SW |
2410 | if (myaddr) |
2411 | msgr->inst.addr = *myaddr; | |
2412 | ||
2413 | /* select a random nonce */ | |
ac8839d7 | 2414 | msgr->inst.addr.type = 0; |
103e2d3a | 2415 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2416 | encode_my_addr(msgr); |
15d9882c | 2417 | msgr->nocrc = nocrc; |
31b8006e | 2418 | |
a2a32584 GH |
2419 | atomic_set(&msgr->stopping, 0); |
2420 | ||
15d9882c | 2421 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2422 | } |
15d9882c | 2423 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2424 | |
e00de341 SW |
2425 | static void clear_standby(struct ceph_connection *con) |
2426 | { | |
2427 | /* come back from STANDBY? */ | |
2428 | if (test_and_clear_bit(STANDBY, &con->state)) { | |
2429 | mutex_lock(&con->mutex); | |
2430 | dout("clear_standby %p and ++connect_seq\n", con); | |
2431 | con->connect_seq++; | |
928443cd AE |
2432 | WARN_ON(test_bit(WRITE_PENDING, &con->flags)); |
2433 | WARN_ON(test_bit(KEEPALIVE_PENDING, &con->flags)); | |
e00de341 SW |
2434 | mutex_unlock(&con->mutex); |
2435 | } | |
2436 | } | |
2437 | ||
31b8006e SW |
2438 | /* |
2439 | * Queue up an outgoing message on the given connection. | |
2440 | */ | |
2441 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2442 | { | |
2443 | if (test_bit(CLOSED, &con->state)) { | |
2444 | dout("con_send %p closed, dropping %p\n", con, msg); | |
2445 | ceph_msg_put(msg); | |
2446 | return; | |
2447 | } | |
2448 | ||
2449 | /* set src+dst */ | |
dbad185d | 2450 | msg->hdr.src = con->msgr->inst.name; |
31b8006e | 2451 | |
3ca02ef9 SW |
2452 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
2453 | ||
e84346b7 SW |
2454 | msg->needs_out_seq = true; |
2455 | ||
31b8006e | 2456 | /* queue */ |
ec302645 | 2457 | mutex_lock(&con->mutex); |
92ce034b | 2458 | |
38941f80 | 2459 | BUG_ON(msg->con != NULL); |
36eb71aa | 2460 | msg->con = con->ops->get(con); |
92ce034b AE |
2461 | BUG_ON(msg->con == NULL); |
2462 | ||
31b8006e SW |
2463 | BUG_ON(!list_empty(&msg->list_head)); |
2464 | list_add_tail(&msg->list_head, &con->out_queue); | |
2465 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2466 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2467 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2468 | le32_to_cpu(msg->hdr.front_len), | |
2469 | le32_to_cpu(msg->hdr.middle_len), | |
2470 | le32_to_cpu(msg->hdr.data_len)); | |
ec302645 | 2471 | mutex_unlock(&con->mutex); |
31b8006e SW |
2472 | |
2473 | /* if there wasn't anything waiting to send before, queue | |
2474 | * new work */ | |
e00de341 | 2475 | clear_standby(con); |
928443cd | 2476 | if (test_and_set_bit(WRITE_PENDING, &con->flags) == 0) |
31b8006e SW |
2477 | queue_con(con); |
2478 | } | |
3d14c5d2 | 2479 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2480 | |
2481 | /* | |
2482 | * Revoke a message that was previously queued for send | |
2483 | */ | |
6740a845 | 2484 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2485 | { |
6740a845 AE |
2486 | struct ceph_connection *con = msg->con; |
2487 | ||
2488 | if (!con) | |
2489 | return; /* Message not in our possession */ | |
2490 | ||
ec302645 | 2491 | mutex_lock(&con->mutex); |
31b8006e | 2492 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2493 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2494 | list_del_init(&msg->list_head); |
38941f80 | 2495 | BUG_ON(msg->con == NULL); |
36eb71aa | 2496 | msg->con->ops->put(msg->con); |
38941f80 | 2497 | msg->con = NULL; |
92ce034b | 2498 | msg->hdr.seq = 0; |
38941f80 | 2499 | |
31b8006e | 2500 | ceph_msg_put(msg); |
ed98adad SW |
2501 | } |
2502 | if (con->out_msg == msg) { | |
38941f80 | 2503 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2504 | con->out_msg = NULL; |
31b8006e SW |
2505 | if (con->out_kvec_is_msg) { |
2506 | con->out_skip = con->out_kvec_bytes; | |
2507 | con->out_kvec_is_msg = false; | |
2508 | } | |
ed98adad | 2509 | msg->hdr.seq = 0; |
92ce034b AE |
2510 | |
2511 | ceph_msg_put(msg); | |
31b8006e | 2512 | } |
ec302645 | 2513 | mutex_unlock(&con->mutex); |
31b8006e SW |
2514 | } |
2515 | ||
350b1c32 | 2516 | /* |
0d59ab81 | 2517 | * Revoke a message that we may be reading data into |
350b1c32 | 2518 | */ |
8921d114 | 2519 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2520 | { |
8921d114 AE |
2521 | struct ceph_connection *con; |
2522 | ||
2523 | BUG_ON(msg == NULL); | |
2524 | if (!msg->con) { | |
2525 | dout("%s msg %p null con\n", __func__, msg); | |
2526 | ||
2527 | return; /* Message not in our possession */ | |
2528 | } | |
2529 | ||
2530 | con = msg->con; | |
350b1c32 | 2531 | mutex_lock(&con->mutex); |
8921d114 | 2532 | if (con->in_msg == msg) { |
95c96174 ED |
2533 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
2534 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
2535 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
2536 | |
2537 | /* skip rest of message */ | |
8921d114 AE |
2538 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
2539 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 2540 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
2541 | front_len - |
2542 | middle_len - | |
2543 | data_len - | |
350b1c32 | 2544 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2545 | ceph_msg_put(con->in_msg); |
2546 | con->in_msg = NULL; | |
2547 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2548 | con->in_seq++; |
350b1c32 | 2549 | } else { |
8921d114 AE |
2550 | dout("%s %p in_msg %p msg %p no-op\n", |
2551 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
2552 | } |
2553 | mutex_unlock(&con->mutex); | |
2554 | } | |
2555 | ||
31b8006e SW |
2556 | /* |
2557 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2558 | */ | |
2559 | void ceph_con_keepalive(struct ceph_connection *con) | |
2560 | { | |
e00de341 SW |
2561 | dout("con_keepalive %p\n", con); |
2562 | clear_standby(con); | |
928443cd AE |
2563 | if (test_and_set_bit(KEEPALIVE_PENDING, &con->flags) == 0 && |
2564 | test_and_set_bit(WRITE_PENDING, &con->flags) == 0) | |
31b8006e SW |
2565 | queue_con(con); |
2566 | } | |
3d14c5d2 | 2567 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2568 | |
2569 | ||
2570 | /* | |
2571 | * construct a new message with given type, size | |
2572 | * the new msg has a ref count of 1. | |
2573 | */ | |
b61c2763 SW |
2574 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2575 | bool can_fail) | |
31b8006e SW |
2576 | { |
2577 | struct ceph_msg *m; | |
2578 | ||
34d23762 | 2579 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2580 | if (m == NULL) |
2581 | goto out; | |
c2e552e7 | 2582 | kref_init(&m->kref); |
38941f80 AE |
2583 | |
2584 | m->con = NULL; | |
31b8006e SW |
2585 | INIT_LIST_HEAD(&m->list_head); |
2586 | ||
45c6ceb5 | 2587 | m->hdr.tid = 0; |
31b8006e | 2588 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2589 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2590 | m->hdr.version = 0; | |
31b8006e SW |
2591 | m->hdr.front_len = cpu_to_le32(front_len); |
2592 | m->hdr.middle_len = 0; | |
bb257664 SW |
2593 | m->hdr.data_len = 0; |
2594 | m->hdr.data_off = 0; | |
45c6ceb5 | 2595 | m->hdr.reserved = 0; |
31b8006e SW |
2596 | m->footer.front_crc = 0; |
2597 | m->footer.middle_crc = 0; | |
2598 | m->footer.data_crc = 0; | |
45c6ceb5 | 2599 | m->footer.flags = 0; |
31b8006e SW |
2600 | m->front_max = front_len; |
2601 | m->front_is_vmalloc = false; | |
2602 | m->more_to_follow = false; | |
c0d5f9db | 2603 | m->ack_stamp = 0; |
31b8006e SW |
2604 | m->pool = NULL; |
2605 | ||
ca20892d HC |
2606 | /* middle */ |
2607 | m->middle = NULL; | |
2608 | ||
2609 | /* data */ | |
2610 | m->nr_pages = 0; | |
2611 | m->page_alignment = 0; | |
2612 | m->pages = NULL; | |
2613 | m->pagelist = NULL; | |
2614 | m->bio = NULL; | |
2615 | m->bio_iter = NULL; | |
2616 | m->bio_seg = 0; | |
2617 | m->trail = NULL; | |
2618 | ||
31b8006e SW |
2619 | /* front */ |
2620 | if (front_len) { | |
2621 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2622 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2623 | PAGE_KERNEL); |
2624 | m->front_is_vmalloc = true; | |
2625 | } else { | |
34d23762 | 2626 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2627 | } |
2628 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2629 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2630 | front_len); |
2631 | goto out2; | |
2632 | } | |
2633 | } else { | |
2634 | m->front.iov_base = NULL; | |
2635 | } | |
2636 | m->front.iov_len = front_len; | |
2637 | ||
bb257664 | 2638 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2639 | return m; |
2640 | ||
2641 | out2: | |
2642 | ceph_msg_put(m); | |
2643 | out: | |
b61c2763 SW |
2644 | if (!can_fail) { |
2645 | pr_err("msg_new can't create type %d front %d\n", type, | |
2646 | front_len); | |
f0ed1b7c | 2647 | WARN_ON(1); |
b61c2763 SW |
2648 | } else { |
2649 | dout("msg_new can't create type %d front %d\n", type, | |
2650 | front_len); | |
2651 | } | |
a79832f2 | 2652 | return NULL; |
31b8006e | 2653 | } |
3d14c5d2 | 2654 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2655 | |
31b8006e SW |
2656 | /* |
2657 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2658 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2659 | * per-type header in the front and then gracefully fail (i.e., | |
2660 | * propagate the error to the caller based on info in the front) when | |
2661 | * the middle is too large. | |
2662 | */ | |
2450418c | 2663 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2664 | { |
2665 | int type = le16_to_cpu(msg->hdr.type); | |
2666 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2667 | ||
2668 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2669 | ceph_msg_type_name(type), middle_len); | |
2670 | BUG_ON(!middle_len); | |
2671 | BUG_ON(msg->middle); | |
2672 | ||
b6c1d5b8 | 2673 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2674 | if (!msg->middle) |
2675 | return -ENOMEM; | |
2676 | return 0; | |
2677 | } | |
2678 | ||
2450418c | 2679 | /* |
1c20f2d2 AE |
2680 | * Allocate a message for receiving an incoming message on a |
2681 | * connection, and save the result in con->in_msg. Uses the | |
2682 | * connection's private alloc_msg op if available. | |
2683 | * | |
2684 | * Returns true if the message should be skipped, false otherwise. | |
2685 | * If true is returned (skip message), con->in_msg will be NULL. | |
2686 | * If false is returned, con->in_msg will contain a pointer to the | |
2687 | * newly-allocated message, or NULL in case of memory exhaustion. | |
2450418c | 2688 | */ |
1c20f2d2 AE |
2689 | static bool ceph_con_in_msg_alloc(struct ceph_connection *con, |
2690 | struct ceph_msg_header *hdr) | |
2450418c YS |
2691 | { |
2692 | int type = le16_to_cpu(hdr->type); | |
2693 | int front_len = le32_to_cpu(hdr->front_len); | |
2694 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2450418c YS |
2695 | int ret; |
2696 | ||
1c20f2d2 AE |
2697 | BUG_ON(con->in_msg != NULL); |
2698 | ||
2450418c | 2699 | if (con->ops->alloc_msg) { |
1c20f2d2 AE |
2700 | int skip = 0; |
2701 | ||
0547a9b3 | 2702 | mutex_unlock(&con->mutex); |
1c20f2d2 | 2703 | con->in_msg = con->ops->alloc_msg(con, hdr, &skip); |
0547a9b3 | 2704 | mutex_lock(&con->mutex); |
92ce034b | 2705 | if (con->in_msg) { |
36eb71aa | 2706 | con->in_msg->con = con->ops->get(con); |
92ce034b AE |
2707 | BUG_ON(con->in_msg->con == NULL); |
2708 | } | |
1c20f2d2 AE |
2709 | if (skip) |
2710 | con->in_msg = NULL; | |
2711 | ||
2712 | if (!con->in_msg) | |
2713 | return skip != 0; | |
2450418c | 2714 | } |
1c20f2d2 AE |
2715 | if (!con->in_msg) { |
2716 | con->in_msg = ceph_msg_new(type, front_len, GFP_NOFS, false); | |
2717 | if (!con->in_msg) { | |
2450418c YS |
2718 | pr_err("unable to allocate msg type %d len %d\n", |
2719 | type, front_len); | |
1c20f2d2 | 2720 | return false; |
2450418c | 2721 | } |
36eb71aa | 2722 | con->in_msg->con = con->ops->get(con); |
92ce034b | 2723 | BUG_ON(con->in_msg->con == NULL); |
1c20f2d2 | 2724 | con->in_msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2725 | } |
1c20f2d2 | 2726 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2727 | |
1c20f2d2 AE |
2728 | if (middle_len && !con->in_msg->middle) { |
2729 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 2730 | if (ret < 0) { |
1c20f2d2 AE |
2731 | ceph_msg_put(con->in_msg); |
2732 | con->in_msg = NULL; | |
2450418c YS |
2733 | } |
2734 | } | |
9d7f0f13 | 2735 | |
1c20f2d2 | 2736 | return false; |
2450418c YS |
2737 | } |
2738 | ||
31b8006e SW |
2739 | |
2740 | /* | |
2741 | * Free a generically kmalloc'd message. | |
2742 | */ | |
2743 | void ceph_msg_kfree(struct ceph_msg *m) | |
2744 | { | |
2745 | dout("msg_kfree %p\n", m); | |
2746 | if (m->front_is_vmalloc) | |
2747 | vfree(m->front.iov_base); | |
2748 | else | |
2749 | kfree(m->front.iov_base); | |
2750 | kfree(m); | |
2751 | } | |
2752 | ||
2753 | /* | |
2754 | * Drop a msg ref. Destroy as needed. | |
2755 | */ | |
c2e552e7 SW |
2756 | void ceph_msg_last_put(struct kref *kref) |
2757 | { | |
2758 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2759 | |
c2e552e7 SW |
2760 | dout("ceph_msg_put last one on %p\n", m); |
2761 | WARN_ON(!list_empty(&m->list_head)); | |
2762 | ||
2763 | /* drop middle, data, if any */ | |
2764 | if (m->middle) { | |
2765 | ceph_buffer_put(m->middle); | |
2766 | m->middle = NULL; | |
31b8006e | 2767 | } |
c2e552e7 SW |
2768 | m->nr_pages = 0; |
2769 | m->pages = NULL; | |
2770 | ||
58bb3b37 SW |
2771 | if (m->pagelist) { |
2772 | ceph_pagelist_release(m->pagelist); | |
2773 | kfree(m->pagelist); | |
2774 | m->pagelist = NULL; | |
2775 | } | |
2776 | ||
68b4476b YS |
2777 | m->trail = NULL; |
2778 | ||
c2e552e7 SW |
2779 | if (m->pool) |
2780 | ceph_msgpool_put(m->pool, m); | |
2781 | else | |
2782 | ceph_msg_kfree(m); | |
31b8006e | 2783 | } |
3d14c5d2 | 2784 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2785 | |
2786 | void ceph_msg_dump(struct ceph_msg *msg) | |
2787 | { | |
2788 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2789 | msg->front_max, msg->nr_pages); | |
2790 | print_hex_dump(KERN_DEBUG, "header: ", | |
2791 | DUMP_PREFIX_OFFSET, 16, 1, | |
2792 | &msg->hdr, sizeof(msg->hdr), true); | |
2793 | print_hex_dump(KERN_DEBUG, " front: ", | |
2794 | DUMP_PREFIX_OFFSET, 16, 1, | |
2795 | msg->front.iov_base, msg->front.iov_len, true); | |
2796 | if (msg->middle) | |
2797 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2798 | DUMP_PREFIX_OFFSET, 16, 1, | |
2799 | msg->middle->vec.iov_base, | |
2800 | msg->middle->vec.iov_len, true); | |
2801 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2802 | DUMP_PREFIX_OFFSET, 16, 1, | |
2803 | &msg->footer, sizeof(msg->footer), true); | |
2804 | } | |
3d14c5d2 | 2805 | EXPORT_SYMBOL(ceph_msg_dump); |