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 | ||
8dacc7da SW |
80 | /* |
81 | * connection states | |
82 | */ | |
83 | #define CON_STATE_CLOSED 1 /* -> PREOPEN */ | |
84 | #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */ | |
85 | #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */ | |
86 | #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */ | |
87 | #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */ | |
88 | #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */ | |
89 | ||
4a861692 SW |
90 | /* |
91 | * ceph_connection flag bits | |
92 | */ | |
93 | #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop | |
94 | * messages on errors */ | |
95 | #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */ | |
96 | #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */ | |
97 | #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */ | |
98 | #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */ | |
8dacc7da | 99 | |
31b8006e SW |
100 | /* static tag bytes (protocol control messages) */ |
101 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
102 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
103 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
104 | ||
a6a5349d SW |
105 | #ifdef CONFIG_LOCKDEP |
106 | static struct lock_class_key socket_class; | |
107 | #endif | |
108 | ||
84495f49 AE |
109 | /* |
110 | * When skipping (ignoring) a block of input we read it into a "skip | |
111 | * buffer," which is this many bytes in size. | |
112 | */ | |
113 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
114 | |
115 | static void queue_con(struct ceph_connection *con); | |
116 | static void con_work(struct work_struct *); | |
117 | static void ceph_fault(struct ceph_connection *con); | |
118 | ||
31b8006e | 119 | /* |
f64a9317 AE |
120 | * Nicely render a sockaddr as a string. An array of formatted |
121 | * strings is used, to approximate reentrancy. | |
31b8006e | 122 | */ |
f64a9317 AE |
123 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
124 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
125 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
126 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
127 | ||
128 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
129 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 130 | |
57666519 | 131 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 132 | |
3d14c5d2 | 133 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
134 | { |
135 | int i; | |
136 | char *s; | |
99f0f3b2 AE |
137 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
138 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 139 | |
f64a9317 | 140 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
141 | s = addr_str[i]; |
142 | ||
143 | switch (ss->ss_family) { | |
144 | case AF_INET: | |
bd406145 AE |
145 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
146 | ntohs(in4->sin_port)); | |
31b8006e SW |
147 | break; |
148 | ||
149 | case AF_INET6: | |
bd406145 AE |
150 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
151 | ntohs(in6->sin6_port)); | |
31b8006e SW |
152 | break; |
153 | ||
154 | default: | |
d3002b97 AE |
155 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
156 | ss->ss_family); | |
31b8006e SW |
157 | } |
158 | ||
159 | return s; | |
160 | } | |
3d14c5d2 | 161 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 162 | |
63f2d211 SW |
163 | static void encode_my_addr(struct ceph_messenger *msgr) |
164 | { | |
165 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
166 | ceph_encode_addr(&msgr->my_enc_addr); | |
167 | } | |
168 | ||
31b8006e SW |
169 | /* |
170 | * work queue for all reading and writing to/from the socket. | |
171 | */ | |
e0f43c94 | 172 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 173 | |
6173d1f0 AE |
174 | void _ceph_msgr_exit(void) |
175 | { | |
d3002b97 | 176 | if (ceph_msgr_wq) { |
6173d1f0 | 177 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
178 | ceph_msgr_wq = NULL; |
179 | } | |
6173d1f0 | 180 | |
6173d1f0 AE |
181 | BUG_ON(zero_page == NULL); |
182 | kunmap(zero_page); | |
183 | page_cache_release(zero_page); | |
184 | zero_page = NULL; | |
185 | } | |
186 | ||
3d14c5d2 | 187 | int ceph_msgr_init(void) |
31b8006e | 188 | { |
57666519 AE |
189 | BUG_ON(zero_page != NULL); |
190 | zero_page = ZERO_PAGE(0); | |
191 | page_cache_get(zero_page); | |
192 | ||
f363e45f | 193 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0); |
6173d1f0 AE |
194 | if (ceph_msgr_wq) |
195 | return 0; | |
57666519 | 196 | |
6173d1f0 AE |
197 | pr_err("msgr_init failed to create workqueue\n"); |
198 | _ceph_msgr_exit(); | |
57666519 | 199 | |
6173d1f0 | 200 | return -ENOMEM; |
31b8006e | 201 | } |
3d14c5d2 | 202 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
203 | |
204 | void ceph_msgr_exit(void) | |
205 | { | |
57666519 | 206 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 207 | |
6173d1f0 | 208 | _ceph_msgr_exit(); |
31b8006e | 209 | } |
3d14c5d2 | 210 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 211 | |
cd84db6e | 212 | void ceph_msgr_flush(void) |
a922d38f SW |
213 | { |
214 | flush_workqueue(ceph_msgr_wq); | |
215 | } | |
3d14c5d2 | 216 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 217 | |
ce2c8903 AE |
218 | /* Connection socket state transition functions */ |
219 | ||
220 | static void con_sock_state_init(struct ceph_connection *con) | |
221 | { | |
222 | int old_state; | |
223 | ||
224 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
225 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
226 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
227 | } | |
228 | ||
229 | static void con_sock_state_connecting(struct ceph_connection *con) | |
230 | { | |
231 | int old_state; | |
232 | ||
233 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
234 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
235 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
236 | } | |
237 | ||
238 | static void con_sock_state_connected(struct ceph_connection *con) | |
239 | { | |
240 | int old_state; | |
241 | ||
242 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
243 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
244 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
245 | } | |
246 | ||
247 | static void con_sock_state_closing(struct ceph_connection *con) | |
248 | { | |
249 | int old_state; | |
250 | ||
251 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
252 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
253 | old_state != CON_SOCK_STATE_CONNECTED && | |
254 | old_state != CON_SOCK_STATE_CLOSING)) | |
255 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
256 | } | |
257 | ||
258 | static void con_sock_state_closed(struct ceph_connection *con) | |
259 | { | |
260 | int old_state; | |
261 | ||
262 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
263 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 SW |
264 | old_state != CON_SOCK_STATE_CLOSING && |
265 | old_state != CON_SOCK_STATE_CONNECTING)) | |
ce2c8903 AE |
266 | printk("%s: unexpected old state %d\n", __func__, old_state); |
267 | } | |
a922d38f | 268 | |
31b8006e SW |
269 | /* |
270 | * socket callback functions | |
271 | */ | |
272 | ||
273 | /* data available on socket, or listen socket received a connect */ | |
327800bd | 274 | static void ceph_sock_data_ready(struct sock *sk, int count_unused) |
31b8006e | 275 | { |
bd406145 | 276 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
277 | if (atomic_read(&con->msgr->stopping)) { |
278 | return; | |
279 | } | |
bd406145 | 280 | |
31b8006e | 281 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 282 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
283 | con, con->state); |
284 | queue_con(con); | |
285 | } | |
286 | } | |
287 | ||
288 | /* socket has buffer space for writing */ | |
327800bd | 289 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 290 | { |
d3002b97 | 291 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 292 | |
182fac26 JS |
293 | /* only queue to workqueue if there is data we want to write, |
294 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 295 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
296 | * doesn't get called again until try_write() fills the socket |
297 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
298 | * and net/core/stream.c:sk_stream_write_space(). | |
299 | */ | |
4a861692 | 300 | if (test_bit(CON_FLAG_WRITE_PENDING, &con->flags)) { |
182fac26 | 301 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
327800bd | 302 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
303 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
304 | queue_con(con); | |
305 | } | |
31b8006e | 306 | } else { |
327800bd | 307 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 308 | } |
31b8006e SW |
309 | } |
310 | ||
311 | /* socket's state has changed */ | |
327800bd | 312 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 313 | { |
bd406145 | 314 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 315 | |
327800bd | 316 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
317 | con, con->state, sk->sk_state); |
318 | ||
31b8006e SW |
319 | switch (sk->sk_state) { |
320 | case TCP_CLOSE: | |
327800bd | 321 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 322 | case TCP_CLOSE_WAIT: |
327800bd | 323 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 324 | con_sock_state_closing(con); |
4a861692 | 325 | set_bit(CON_FLAG_SOCK_CLOSED, &con->flags); |
d65c9e0b | 326 | queue_con(con); |
31b8006e SW |
327 | break; |
328 | case TCP_ESTABLISHED: | |
327800bd | 329 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 330 | con_sock_state_connected(con); |
31b8006e SW |
331 | queue_con(con); |
332 | break; | |
d3002b97 AE |
333 | default: /* Everything else is uninteresting */ |
334 | break; | |
31b8006e SW |
335 | } |
336 | } | |
337 | ||
338 | /* | |
339 | * set up socket callbacks | |
340 | */ | |
341 | static void set_sock_callbacks(struct socket *sock, | |
342 | struct ceph_connection *con) | |
343 | { | |
344 | struct sock *sk = sock->sk; | |
bd406145 | 345 | sk->sk_user_data = con; |
327800bd AE |
346 | sk->sk_data_ready = ceph_sock_data_ready; |
347 | sk->sk_write_space = ceph_sock_write_space; | |
348 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
349 | } |
350 | ||
351 | ||
352 | /* | |
353 | * socket helpers | |
354 | */ | |
355 | ||
356 | /* | |
357 | * initiate connection to a remote socket. | |
358 | */ | |
41617d0c | 359 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 360 | { |
f91d3471 | 361 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
362 | struct socket *sock; |
363 | int ret; | |
364 | ||
365 | BUG_ON(con->sock); | |
f91d3471 SW |
366 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
367 | IPPROTO_TCP, &sock); | |
31b8006e | 368 | if (ret) |
41617d0c | 369 | return ret; |
31b8006e SW |
370 | sock->sk->sk_allocation = GFP_NOFS; |
371 | ||
a6a5349d SW |
372 | #ifdef CONFIG_LOCKDEP |
373 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
374 | #endif | |
375 | ||
31b8006e SW |
376 | set_sock_callbacks(sock, con); |
377 | ||
3d14c5d2 | 378 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 379 | |
89a86be0 | 380 | con_sock_state_connecting(con); |
f91d3471 SW |
381 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
382 | O_NONBLOCK); | |
31b8006e SW |
383 | if (ret == -EINPROGRESS) { |
384 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 385 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 386 | sock->sk->sk_state); |
a5bc3129 | 387 | } else if (ret < 0) { |
31b8006e | 388 | pr_err("connect %s error %d\n", |
3d14c5d2 | 389 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 390 | sock_release(sock); |
31b8006e | 391 | con->error_msg = "connect error"; |
31b8006e | 392 | |
41617d0c | 393 | return ret; |
a5bc3129 AE |
394 | } |
395 | con->sock = sock; | |
41617d0c | 396 | return 0; |
31b8006e SW |
397 | } |
398 | ||
399 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
400 | { | |
401 | struct kvec iov = {buf, len}; | |
402 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 403 | int r; |
31b8006e | 404 | |
98bdb0aa SW |
405 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
406 | if (r == -EAGAIN) | |
407 | r = 0; | |
408 | return r; | |
31b8006e SW |
409 | } |
410 | ||
411 | /* | |
412 | * write something. @more is true if caller will be sending more data | |
413 | * shortly. | |
414 | */ | |
415 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
416 | size_t kvlen, size_t len, int more) | |
417 | { | |
418 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 419 | int r; |
31b8006e SW |
420 | |
421 | if (more) | |
422 | msg.msg_flags |= MSG_MORE; | |
423 | else | |
424 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
425 | ||
42961d23 SW |
426 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
427 | if (r == -EAGAIN) | |
428 | r = 0; | |
429 | return r; | |
31b8006e SW |
430 | } |
431 | ||
31739139 AE |
432 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
433 | int offset, size_t size, int more) | |
434 | { | |
435 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
436 | int ret; | |
437 | ||
438 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
439 | if (ret == -EAGAIN) | |
440 | ret = 0; | |
441 | ||
442 | return ret; | |
443 | } | |
444 | ||
31b8006e SW |
445 | |
446 | /* | |
447 | * Shutdown/close the socket for the given connection. | |
448 | */ | |
449 | static int con_close_socket(struct ceph_connection *con) | |
450 | { | |
451 | int rc; | |
452 | ||
453 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
454 | if (!con->sock) | |
455 | return 0; | |
31b8006e SW |
456 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); |
457 | sock_release(con->sock); | |
458 | con->sock = NULL; | |
456ea468 AE |
459 | |
460 | /* | |
4a861692 | 461 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
462 | * independent of the connection mutex, and we could have |
463 | * received a socket close event before we had the chance to | |
464 | * shut the socket down. | |
465 | */ | |
4a861692 | 466 | clear_bit(CON_FLAG_SOCK_CLOSED, &con->flags); |
ce2c8903 | 467 | con_sock_state_closed(con); |
31b8006e SW |
468 | return rc; |
469 | } | |
470 | ||
471 | /* | |
472 | * Reset a connection. Discard all incoming and outgoing messages | |
473 | * and clear *_seq state. | |
474 | */ | |
475 | static void ceph_msg_remove(struct ceph_msg *msg) | |
476 | { | |
477 | list_del_init(&msg->list_head); | |
38941f80 | 478 | BUG_ON(msg->con == NULL); |
36eb71aa | 479 | msg->con->ops->put(msg->con); |
38941f80 AE |
480 | msg->con = NULL; |
481 | ||
31b8006e SW |
482 | ceph_msg_put(msg); |
483 | } | |
484 | static void ceph_msg_remove_list(struct list_head *head) | |
485 | { | |
486 | while (!list_empty(head)) { | |
487 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
488 | list_head); | |
489 | ceph_msg_remove(msg); | |
490 | } | |
491 | } | |
492 | ||
493 | static void reset_connection(struct ceph_connection *con) | |
494 | { | |
495 | /* reset connection, out_queue, msg_ and connect_seq */ | |
496 | /* discard existing out_queue and msg_seq */ | |
31b8006e SW |
497 | ceph_msg_remove_list(&con->out_queue); |
498 | ceph_msg_remove_list(&con->out_sent); | |
499 | ||
cf3e5c40 | 500 | if (con->in_msg) { |
38941f80 AE |
501 | BUG_ON(con->in_msg->con != con); |
502 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
503 | ceph_msg_put(con->in_msg); |
504 | con->in_msg = NULL; | |
36eb71aa | 505 | con->ops->put(con); |
cf3e5c40 SW |
506 | } |
507 | ||
31b8006e SW |
508 | con->connect_seq = 0; |
509 | con->out_seq = 0; | |
c86a2930 SW |
510 | if (con->out_msg) { |
511 | ceph_msg_put(con->out_msg); | |
512 | con->out_msg = NULL; | |
513 | } | |
31b8006e | 514 | con->in_seq = 0; |
0e0d5e0c | 515 | con->in_seq_acked = 0; |
31b8006e SW |
516 | } |
517 | ||
518 | /* | |
519 | * mark a peer down. drop any open connections. | |
520 | */ | |
521 | void ceph_con_close(struct ceph_connection *con) | |
522 | { | |
8c50c817 | 523 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
524 | dout("con_close %p peer %s\n", con, |
525 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
8dacc7da | 526 | con->state = CON_STATE_CLOSED; |
a5988c49 | 527 | |
4a861692 SW |
528 | clear_bit(CON_FLAG_LOSSYTX, &con->flags); /* so we retry next connect */ |
529 | clear_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags); | |
530 | clear_bit(CON_FLAG_WRITE_PENDING, &con->flags); | |
43c7427d SW |
531 | clear_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags); |
532 | clear_bit(CON_FLAG_BACKOFF, &con->flags); | |
a5988c49 | 533 | |
31b8006e | 534 | reset_connection(con); |
6f2bc3ff | 535 | con->peer_global_seq = 0; |
91e45ce3 | 536 | cancel_delayed_work(&con->work); |
ee76e073 | 537 | con_close_socket(con); |
ec302645 | 538 | mutex_unlock(&con->mutex); |
31b8006e | 539 | } |
3d14c5d2 | 540 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 541 | |
31b8006e SW |
542 | /* |
543 | * Reopen a closed connection, with a new peer address. | |
544 | */ | |
b7a9e5dd SW |
545 | void ceph_con_open(struct ceph_connection *con, |
546 | __u8 entity_type, __u64 entity_num, | |
547 | struct ceph_entity_addr *addr) | |
31b8006e | 548 | { |
5469155f | 549 | mutex_lock(&con->mutex); |
3d14c5d2 | 550 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
8dacc7da SW |
551 | |
552 | BUG_ON(con->state != CON_STATE_CLOSED); | |
553 | con->state = CON_STATE_PREOPEN; | |
a5988c49 | 554 | |
b7a9e5dd SW |
555 | con->peer_name.type = (__u8) entity_type; |
556 | con->peer_name.num = cpu_to_le64(entity_num); | |
557 | ||
31b8006e | 558 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 559 | con->delay = 0; /* reset backoff memory */ |
5469155f | 560 | mutex_unlock(&con->mutex); |
31b8006e SW |
561 | queue_con(con); |
562 | } | |
3d14c5d2 | 563 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 564 | |
87b315a5 SW |
565 | /* |
566 | * return true if this connection ever successfully opened | |
567 | */ | |
568 | bool ceph_con_opened(struct ceph_connection *con) | |
569 | { | |
570 | return con->connect_seq > 0; | |
571 | } | |
572 | ||
31b8006e SW |
573 | /* |
574 | * initialize a new connection. | |
575 | */ | |
1bfd89f4 AE |
576 | void ceph_con_init(struct ceph_connection *con, void *private, |
577 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 578 | struct ceph_messenger *msgr) |
31b8006e SW |
579 | { |
580 | dout("con_init %p\n", con); | |
581 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
582 | con->private = private; |
583 | con->ops = ops; | |
31b8006e | 584 | con->msgr = msgr; |
ce2c8903 AE |
585 | |
586 | con_sock_state_init(con); | |
587 | ||
ec302645 | 588 | mutex_init(&con->mutex); |
31b8006e SW |
589 | INIT_LIST_HEAD(&con->out_queue); |
590 | INIT_LIST_HEAD(&con->out_sent); | |
591 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 | 592 | |
8dacc7da | 593 | con->state = CON_STATE_CLOSED; |
31b8006e | 594 | } |
3d14c5d2 | 595 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
596 | |
597 | ||
598 | /* | |
599 | * We maintain a global counter to order connection attempts. Get | |
600 | * a unique seq greater than @gt. | |
601 | */ | |
602 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
603 | { | |
604 | u32 ret; | |
605 | ||
606 | spin_lock(&msgr->global_seq_lock); | |
607 | if (msgr->global_seq < gt) | |
608 | msgr->global_seq = gt; | |
609 | ret = ++msgr->global_seq; | |
610 | spin_unlock(&msgr->global_seq_lock); | |
611 | return ret; | |
612 | } | |
613 | ||
e2200423 | 614 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
615 | { |
616 | con->out_kvec_left = 0; | |
617 | con->out_kvec_bytes = 0; | |
618 | con->out_kvec_cur = &con->out_kvec[0]; | |
619 | } | |
620 | ||
e2200423 | 621 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
622 | size_t size, void *data) |
623 | { | |
624 | int index; | |
625 | ||
626 | index = con->out_kvec_left; | |
627 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
628 | ||
629 | con->out_kvec[index].iov_len = size; | |
630 | con->out_kvec[index].iov_base = data; | |
631 | con->out_kvec_left++; | |
632 | con->out_kvec_bytes += size; | |
633 | } | |
31b8006e | 634 | |
df6ad1f9 AE |
635 | #ifdef CONFIG_BLOCK |
636 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
637 | { | |
638 | if (!bio) { | |
639 | *iter = NULL; | |
640 | *seg = 0; | |
641 | return; | |
642 | } | |
643 | *iter = bio; | |
644 | *seg = bio->bi_idx; | |
645 | } | |
646 | ||
647 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
648 | { | |
649 | if (*bio_iter == NULL) | |
650 | return; | |
651 | ||
652 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
653 | ||
654 | (*seg)++; | |
655 | if (*seg == (*bio_iter)->bi_vcnt) | |
656 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
657 | } | |
658 | #endif | |
659 | ||
739c905b AE |
660 | static void prepare_write_message_data(struct ceph_connection *con) |
661 | { | |
662 | struct ceph_msg *msg = con->out_msg; | |
663 | ||
664 | BUG_ON(!msg); | |
665 | BUG_ON(!msg->hdr.data_len); | |
666 | ||
667 | /* initialize page iterator */ | |
668 | con->out_msg_pos.page = 0; | |
669 | if (msg->pages) | |
670 | con->out_msg_pos.page_pos = msg->page_alignment; | |
671 | else | |
672 | con->out_msg_pos.page_pos = 0; | |
572c588e | 673 | #ifdef CONFIG_BLOCK |
abdaa6a8 | 674 | if (msg->bio) |
572c588e AE |
675 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); |
676 | #endif | |
739c905b AE |
677 | con->out_msg_pos.data_pos = 0; |
678 | con->out_msg_pos.did_page_crc = false; | |
679 | con->out_more = 1; /* data + footer will follow */ | |
680 | } | |
681 | ||
31b8006e SW |
682 | /* |
683 | * Prepare footer for currently outgoing message, and finish things | |
684 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
685 | */ | |
859eb799 | 686 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
687 | { |
688 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 689 | int v = con->out_kvec_left; |
31b8006e | 690 | |
fd154f3c AE |
691 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
692 | ||
31b8006e SW |
693 | dout("prepare_write_message_footer %p\n", con); |
694 | con->out_kvec_is_msg = true; | |
695 | con->out_kvec[v].iov_base = &m->footer; | |
696 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
697 | con->out_kvec_bytes += sizeof(m->footer); | |
698 | con->out_kvec_left++; | |
699 | con->out_more = m->more_to_follow; | |
c86a2930 | 700 | con->out_msg_done = true; |
31b8006e SW |
701 | } |
702 | ||
703 | /* | |
704 | * Prepare headers for the next outgoing message. | |
705 | */ | |
706 | static void prepare_write_message(struct ceph_connection *con) | |
707 | { | |
708 | struct ceph_msg *m; | |
a9a0c51a | 709 | u32 crc; |
31b8006e | 710 | |
e2200423 | 711 | con_out_kvec_reset(con); |
31b8006e | 712 | con->out_kvec_is_msg = true; |
c86a2930 | 713 | con->out_msg_done = false; |
31b8006e SW |
714 | |
715 | /* Sneak an ack in there first? If we can get it into the same | |
716 | * TCP packet that's a good thing. */ | |
717 | if (con->in_seq > con->in_seq_acked) { | |
718 | con->in_seq_acked = con->in_seq; | |
e2200423 | 719 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 720 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 721 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 722 | &con->out_temp_ack); |
31b8006e SW |
723 | } |
724 | ||
38941f80 | 725 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 726 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 727 | con->out_msg = m; |
38941f80 | 728 | BUG_ON(m->con != con); |
4cf9d544 SW |
729 | |
730 | /* put message on sent list */ | |
731 | ceph_msg_get(m); | |
732 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 733 | |
e84346b7 SW |
734 | /* |
735 | * only assign outgoing seq # if we haven't sent this message | |
736 | * yet. if it is requeued, resend with it's original seq. | |
737 | */ | |
738 | if (m->needs_out_seq) { | |
739 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
740 | m->needs_out_seq = false; | |
741 | } | |
31b8006e SW |
742 | |
743 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
744 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
745 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
746 | le32_to_cpu(m->hdr.data_len), | |
747 | m->nr_pages); | |
748 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); | |
749 | ||
750 | /* tag + hdr + front + middle */ | |
e2200423 AE |
751 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
752 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
753 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 754 | |
31b8006e | 755 | if (m->middle) |
e2200423 | 756 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 757 | m->middle->vec.iov_base); |
31b8006e SW |
758 | |
759 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
760 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
761 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 762 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
763 | |
764 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
765 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
766 | if (m->middle) { | |
767 | crc = crc32c(0, m->middle->vec.iov_base, | |
768 | m->middle->vec.iov_len); | |
769 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
770 | } else | |
31b8006e | 771 | con->out_msg->footer.middle_crc = 0; |
739c905b | 772 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
773 | le32_to_cpu(con->out_msg->footer.front_crc), |
774 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
775 | ||
776 | /* is there a data payload? */ | |
739c905b AE |
777 | con->out_msg->footer.data_crc = 0; |
778 | if (m->hdr.data_len) | |
779 | prepare_write_message_data(con); | |
780 | else | |
31b8006e | 781 | /* no, queue up footer too and be done */ |
859eb799 | 782 | prepare_write_message_footer(con); |
31b8006e | 783 | |
4a861692 | 784 | set_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
31b8006e SW |
785 | } |
786 | ||
787 | /* | |
788 | * Prepare an ack. | |
789 | */ | |
790 | static void prepare_write_ack(struct ceph_connection *con) | |
791 | { | |
792 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
793 | con->in_seq_acked, con->in_seq); | |
794 | con->in_seq_acked = con->in_seq; | |
795 | ||
e2200423 | 796 | con_out_kvec_reset(con); |
859eb799 | 797 | |
e2200423 | 798 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 799 | |
31b8006e | 800 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 801 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
802 | &con->out_temp_ack); |
803 | ||
31b8006e | 804 | con->out_more = 1; /* more will follow.. eventually.. */ |
4a861692 | 805 | set_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
31b8006e SW |
806 | } |
807 | ||
808 | /* | |
809 | * Prepare to write keepalive byte. | |
810 | */ | |
811 | static void prepare_write_keepalive(struct ceph_connection *con) | |
812 | { | |
813 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
814 | con_out_kvec_reset(con); |
815 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
4a861692 | 816 | set_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
31b8006e SW |
817 | } |
818 | ||
819 | /* | |
820 | * Connection negotiation. | |
821 | */ | |
822 | ||
dac1e716 AE |
823 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
824 | int *auth_proto) | |
4e7a5dcd | 825 | { |
a3530df3 | 826 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
827 | |
828 | if (!con->ops->get_authorizer) { | |
829 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
830 | con->out_connect.authorizer_len = 0; | |
729796be | 831 | return NULL; |
b1c6b980 AE |
832 | } |
833 | ||
834 | /* Can't hold the mutex while getting authorizer */ | |
ec302645 | 835 | mutex_unlock(&con->mutex); |
dac1e716 | 836 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
ec302645 | 837 | mutex_lock(&con->mutex); |
4e7a5dcd | 838 | |
a3530df3 | 839 | if (IS_ERR(auth)) |
729796be | 840 | return auth; |
8dacc7da | 841 | if (con->state != CON_STATE_NEGOTIATING) |
729796be | 842 | return ERR_PTR(-EAGAIN); |
0da5d703 | 843 | |
8f43fb53 AE |
844 | con->auth_reply_buf = auth->authorizer_reply_buf; |
845 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
729796be | 846 | return auth; |
4e7a5dcd SW |
847 | } |
848 | ||
31b8006e SW |
849 | /* |
850 | * We connected to a peer and are saying hello. | |
851 | */ | |
e825a66d | 852 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 853 | { |
e2200423 AE |
854 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
855 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 856 | &con->msgr->my_enc_addr); |
eed0ef2c | 857 | |
eed0ef2c | 858 | con->out_more = 0; |
4a861692 | 859 | set_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
eed0ef2c SW |
860 | } |
861 | ||
e825a66d | 862 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 863 | { |
95c96174 | 864 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 865 | int proto; |
dac1e716 | 866 | int auth_proto; |
729796be | 867 | struct ceph_auth_handshake *auth; |
31b8006e SW |
868 | |
869 | switch (con->peer_name.type) { | |
870 | case CEPH_ENTITY_TYPE_MON: | |
871 | proto = CEPH_MONC_PROTOCOL; | |
872 | break; | |
873 | case CEPH_ENTITY_TYPE_OSD: | |
874 | proto = CEPH_OSDC_PROTOCOL; | |
875 | break; | |
876 | case CEPH_ENTITY_TYPE_MDS: | |
877 | proto = CEPH_MDSC_PROTOCOL; | |
878 | break; | |
879 | default: | |
880 | BUG(); | |
881 | } | |
882 | ||
883 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
884 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 885 | |
e825a66d | 886 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
887 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
888 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
889 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
890 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
891 | con->out_connect.flags = 0; | |
31b8006e | 892 | |
dac1e716 AE |
893 | auth_proto = CEPH_AUTH_UNKNOWN; |
894 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
895 | if (IS_ERR(auth)) |
896 | return PTR_ERR(auth); | |
3da54776 | 897 | |
dac1e716 | 898 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
899 | con->out_connect.authorizer_len = auth ? |
900 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
901 | ||
ab166d5a | 902 | con_out_kvec_reset(con); |
e2200423 | 903 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
904 | &con->out_connect); |
905 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 906 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 907 | auth->authorizer_buf); |
859eb799 | 908 | |
31b8006e | 909 | con->out_more = 0; |
4a861692 | 910 | set_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
4e7a5dcd | 911 | |
e10c758e | 912 | return 0; |
31b8006e SW |
913 | } |
914 | ||
31b8006e SW |
915 | /* |
916 | * write as much of pending kvecs to the socket as we can. | |
917 | * 1 -> done | |
918 | * 0 -> socket full, but more to do | |
919 | * <0 -> error | |
920 | */ | |
921 | static int write_partial_kvec(struct ceph_connection *con) | |
922 | { | |
923 | int ret; | |
924 | ||
925 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
926 | while (con->out_kvec_bytes > 0) { | |
927 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
928 | con->out_kvec_left, con->out_kvec_bytes, | |
929 | con->out_more); | |
930 | if (ret <= 0) | |
931 | goto out; | |
932 | con->out_kvec_bytes -= ret; | |
933 | if (con->out_kvec_bytes == 0) | |
934 | break; /* done */ | |
f42299e6 AE |
935 | |
936 | /* account for full iov entries consumed */ | |
937 | while (ret >= con->out_kvec_cur->iov_len) { | |
938 | BUG_ON(!con->out_kvec_left); | |
939 | ret -= con->out_kvec_cur->iov_len; | |
940 | con->out_kvec_cur++; | |
941 | con->out_kvec_left--; | |
942 | } | |
943 | /* and for a partially-consumed entry */ | |
944 | if (ret) { | |
945 | con->out_kvec_cur->iov_len -= ret; | |
946 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
947 | } |
948 | } | |
949 | con->out_kvec_left = 0; | |
950 | con->out_kvec_is_msg = false; | |
951 | ret = 1; | |
952 | out: | |
953 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
954 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
955 | return ret; /* done! */ | |
956 | } | |
957 | ||
84ca8fc8 AE |
958 | static void out_msg_pos_next(struct ceph_connection *con, struct page *page, |
959 | size_t len, size_t sent, bool in_trail) | |
960 | { | |
961 | struct ceph_msg *msg = con->out_msg; | |
962 | ||
963 | BUG_ON(!msg); | |
964 | BUG_ON(!sent); | |
965 | ||
966 | con->out_msg_pos.data_pos += sent; | |
967 | con->out_msg_pos.page_pos += sent; | |
5821bd8c AE |
968 | if (sent < len) |
969 | return; | |
970 | ||
971 | BUG_ON(sent != len); | |
972 | con->out_msg_pos.page_pos = 0; | |
973 | con->out_msg_pos.page++; | |
974 | con->out_msg_pos.did_page_crc = false; | |
975 | if (in_trail) | |
976 | list_move_tail(&page->lru, | |
977 | &msg->trail->head); | |
978 | else if (msg->pagelist) | |
979 | list_move_tail(&page->lru, | |
980 | &msg->pagelist->head); | |
84ca8fc8 | 981 | #ifdef CONFIG_BLOCK |
5821bd8c AE |
982 | else if (msg->bio) |
983 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
84ca8fc8 | 984 | #endif |
84ca8fc8 AE |
985 | } |
986 | ||
31b8006e SW |
987 | /* |
988 | * Write as much message data payload as we can. If we finish, queue | |
989 | * up the footer. | |
990 | * 1 -> done, footer is now queued in out_kvec[]. | |
991 | * 0 -> socket full, but more to do | |
992 | * <0 -> error | |
993 | */ | |
994 | static int write_partial_msg_pages(struct ceph_connection *con) | |
995 | { | |
996 | struct ceph_msg *msg = con->out_msg; | |
95c96174 | 997 | unsigned int data_len = le32_to_cpu(msg->hdr.data_len); |
31b8006e | 998 | size_t len; |
37675b0f | 999 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 1000 | int ret; |
68b4476b | 1001 | int total_max_write; |
84ca8fc8 | 1002 | bool in_trail = false; |
5821bd8c AE |
1003 | const size_t trail_len = (msg->trail ? msg->trail->length : 0); |
1004 | const size_t trail_off = data_len - trail_len; | |
31b8006e SW |
1005 | |
1006 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
84ca8fc8 | 1007 | con, msg, con->out_msg_pos.page, msg->nr_pages, |
31b8006e SW |
1008 | con->out_msg_pos.page_pos); |
1009 | ||
5821bd8c AE |
1010 | /* |
1011 | * Iterate through each page that contains data to be | |
1012 | * written, and send as much as possible for each. | |
1013 | * | |
1014 | * If we are calculating the data crc (the default), we will | |
1015 | * need to map the page. If we have no pages, they have | |
1016 | * been revoked, so use the zero page. | |
1017 | */ | |
68b4476b | 1018 | while (data_len > con->out_msg_pos.data_pos) { |
31b8006e | 1019 | struct page *page = NULL; |
68b4476b | 1020 | int max_write = PAGE_SIZE; |
9bd19663 | 1021 | int bio_offset = 0; |
68b4476b | 1022 | |
5821bd8c AE |
1023 | in_trail = in_trail || con->out_msg_pos.data_pos >= trail_off; |
1024 | if (!in_trail) | |
1025 | total_max_write = trail_off - con->out_msg_pos.data_pos; | |
68b4476b | 1026 | |
5821bd8c | 1027 | if (in_trail) { |
68b4476b YS |
1028 | total_max_write = data_len - con->out_msg_pos.data_pos; |
1029 | ||
1030 | page = list_first_entry(&msg->trail->head, | |
1031 | struct page, lru); | |
68b4476b | 1032 | } else if (msg->pages) { |
31b8006e | 1033 | page = msg->pages[con->out_msg_pos.page]; |
58bb3b37 SW |
1034 | } else if (msg->pagelist) { |
1035 | page = list_first_entry(&msg->pagelist->head, | |
1036 | struct page, lru); | |
68b4476b YS |
1037 | #ifdef CONFIG_BLOCK |
1038 | } else if (msg->bio) { | |
1039 | struct bio_vec *bv; | |
1040 | ||
1041 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
1042 | page = bv->bv_page; | |
9bd19663 | 1043 | bio_offset = bv->bv_offset; |
68b4476b YS |
1044 | max_write = bv->bv_len; |
1045 | #endif | |
31b8006e | 1046 | } else { |
57666519 | 1047 | page = zero_page; |
31b8006e | 1048 | } |
68b4476b YS |
1049 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
1050 | total_max_write); | |
1051 | ||
37675b0f | 1052 | if (do_datacrc && !con->out_msg_pos.did_page_crc) { |
9bd19663 | 1053 | void *base; |
5821bd8c | 1054 | u32 crc = le32_to_cpu(msg->footer.data_crc); |
8d63e318 | 1055 | char *kaddr; |
31b8006e | 1056 | |
8d63e318 | 1057 | kaddr = kmap(page); |
31b8006e | 1058 | BUG_ON(kaddr == NULL); |
9bd19663 | 1059 | base = kaddr + con->out_msg_pos.page_pos + bio_offset; |
5821bd8c | 1060 | crc = crc32c(crc, base, len); |
84ca8fc8 | 1061 | msg->footer.data_crc = cpu_to_le32(crc); |
bca064d2 | 1062 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 1063 | } |
e36b13cc | 1064 | ret = ceph_tcp_sendpage(con->sock, page, |
9bd19663 | 1065 | con->out_msg_pos.page_pos + bio_offset, |
e36b13cc | 1066 | len, 1); |
31b8006e | 1067 | |
0cdf9e60 | 1068 | if (do_datacrc) |
31b8006e SW |
1069 | kunmap(page); |
1070 | ||
1071 | if (ret <= 0) | |
1072 | goto out; | |
1073 | ||
84ca8fc8 | 1074 | out_msg_pos_next(con, page, len, (size_t) ret, in_trail); |
31b8006e SW |
1075 | } |
1076 | ||
1077 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
1078 | ||
1079 | /* prepare and queue up footer, too */ | |
37675b0f | 1080 | if (!do_datacrc) |
84ca8fc8 | 1081 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1082 | con_out_kvec_reset(con); |
859eb799 | 1083 | prepare_write_message_footer(con); |
31b8006e SW |
1084 | ret = 1; |
1085 | out: | |
1086 | return ret; | |
1087 | } | |
1088 | ||
1089 | /* | |
1090 | * write some zeros | |
1091 | */ | |
1092 | static int write_partial_skip(struct ceph_connection *con) | |
1093 | { | |
1094 | int ret; | |
1095 | ||
1096 | while (con->out_skip > 0) { | |
31739139 | 1097 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1098 | |
31739139 | 1099 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1); |
31b8006e SW |
1100 | if (ret <= 0) |
1101 | goto out; | |
1102 | con->out_skip -= ret; | |
1103 | } | |
1104 | ret = 1; | |
1105 | out: | |
1106 | return ret; | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * Prepare to read connection handshake, or an ack. | |
1111 | */ | |
eed0ef2c SW |
1112 | static void prepare_read_banner(struct ceph_connection *con) |
1113 | { | |
1114 | dout("prepare_read_banner %p\n", con); | |
1115 | con->in_base_pos = 0; | |
1116 | } | |
1117 | ||
31b8006e SW |
1118 | static void prepare_read_connect(struct ceph_connection *con) |
1119 | { | |
1120 | dout("prepare_read_connect %p\n", con); | |
1121 | con->in_base_pos = 0; | |
1122 | } | |
1123 | ||
1124 | static void prepare_read_ack(struct ceph_connection *con) | |
1125 | { | |
1126 | dout("prepare_read_ack %p\n", con); | |
1127 | con->in_base_pos = 0; | |
1128 | } | |
1129 | ||
1130 | static void prepare_read_tag(struct ceph_connection *con) | |
1131 | { | |
1132 | dout("prepare_read_tag %p\n", con); | |
1133 | con->in_base_pos = 0; | |
1134 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1135 | } | |
1136 | ||
1137 | /* | |
1138 | * Prepare to read a message. | |
1139 | */ | |
1140 | static int prepare_read_message(struct ceph_connection *con) | |
1141 | { | |
1142 | dout("prepare_read_message %p\n", con); | |
1143 | BUG_ON(con->in_msg != NULL); | |
1144 | con->in_base_pos = 0; | |
1145 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1146 | return 0; | |
1147 | } | |
1148 | ||
1149 | ||
1150 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1151 | int end, int size, void *object) |
31b8006e | 1152 | { |
e6cee71f AE |
1153 | while (con->in_base_pos < end) { |
1154 | int left = end - con->in_base_pos; | |
31b8006e SW |
1155 | int have = size - left; |
1156 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1157 | if (ret <= 0) | |
1158 | return ret; | |
1159 | con->in_base_pos += ret; | |
1160 | } | |
1161 | return 1; | |
1162 | } | |
1163 | ||
1164 | ||
1165 | /* | |
1166 | * Read all or part of the connect-side handshake on a new connection | |
1167 | */ | |
eed0ef2c | 1168 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1169 | { |
fd51653f AE |
1170 | int size; |
1171 | int end; | |
1172 | int ret; | |
31b8006e | 1173 | |
eed0ef2c | 1174 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1175 | |
1176 | /* peer's banner */ | |
fd51653f AE |
1177 | size = strlen(CEPH_BANNER); |
1178 | end = size; | |
1179 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1180 | if (ret <= 0) |
1181 | goto out; | |
fd51653f AE |
1182 | |
1183 | size = sizeof (con->actual_peer_addr); | |
1184 | end += size; | |
1185 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1186 | if (ret <= 0) |
1187 | goto out; | |
fd51653f AE |
1188 | |
1189 | size = sizeof (con->peer_addr_for_me); | |
1190 | end += size; | |
1191 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1192 | if (ret <= 0) |
1193 | goto out; | |
fd51653f | 1194 | |
eed0ef2c SW |
1195 | out: |
1196 | return ret; | |
1197 | } | |
1198 | ||
1199 | static int read_partial_connect(struct ceph_connection *con) | |
1200 | { | |
fd51653f AE |
1201 | int size; |
1202 | int end; | |
1203 | int ret; | |
eed0ef2c SW |
1204 | |
1205 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1206 | ||
fd51653f AE |
1207 | size = sizeof (con->in_reply); |
1208 | end = size; | |
1209 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1210 | if (ret <= 0) |
1211 | goto out; | |
fd51653f AE |
1212 | |
1213 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1214 | end += size; | |
1215 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1216 | if (ret <= 0) |
1217 | goto out; | |
31b8006e | 1218 | |
4e7a5dcd SW |
1219 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1220 | con, (int)con->in_reply.tag, | |
1221 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1222 | le32_to_cpu(con->in_reply.global_seq)); |
1223 | out: | |
1224 | return ret; | |
eed0ef2c | 1225 | |
31b8006e SW |
1226 | } |
1227 | ||
1228 | /* | |
1229 | * Verify the hello banner looks okay. | |
1230 | */ | |
1231 | static int verify_hello(struct ceph_connection *con) | |
1232 | { | |
1233 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1234 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1235 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1236 | con->error_msg = "protocol error, bad banner"; |
1237 | return -1; | |
1238 | } | |
1239 | return 0; | |
1240 | } | |
1241 | ||
1242 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1243 | { | |
1244 | switch (ss->ss_family) { | |
1245 | case AF_INET: | |
1246 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1247 | case AF_INET6: | |
1248 | return | |
1249 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1250 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1251 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1252 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1253 | } | |
1254 | return false; | |
1255 | } | |
1256 | ||
1257 | static int addr_port(struct sockaddr_storage *ss) | |
1258 | { | |
1259 | switch (ss->ss_family) { | |
1260 | case AF_INET: | |
f28bcfbe | 1261 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1262 | case AF_INET6: |
f28bcfbe | 1263 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1264 | } |
1265 | return 0; | |
1266 | } | |
1267 | ||
1268 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1269 | { | |
1270 | switch (ss->ss_family) { | |
1271 | case AF_INET: | |
1272 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1273 | break; |
31b8006e SW |
1274 | case AF_INET6: |
1275 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1276 | break; |
31b8006e SW |
1277 | } |
1278 | } | |
1279 | ||
ee3b56f2 NW |
1280 | /* |
1281 | * Unlike other *_pton function semantics, zero indicates success. | |
1282 | */ | |
1283 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1284 | char delim, const char **ipend) | |
1285 | { | |
99f0f3b2 AE |
1286 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1287 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1288 | |
1289 | memset(ss, 0, sizeof(*ss)); | |
1290 | ||
1291 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1292 | ss->ss_family = AF_INET; | |
1293 | return 0; | |
1294 | } | |
1295 | ||
1296 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1297 | ss->ss_family = AF_INET6; | |
1298 | return 0; | |
1299 | } | |
1300 | ||
1301 | return -EINVAL; | |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * Extract hostname string and resolve using kernel DNS facility. | |
1306 | */ | |
1307 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1308 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1309 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1310 | { | |
1311 | const char *end, *delim_p; | |
1312 | char *colon_p, *ip_addr = NULL; | |
1313 | int ip_len, ret; | |
1314 | ||
1315 | /* | |
1316 | * The end of the hostname occurs immediately preceding the delimiter or | |
1317 | * the port marker (':') where the delimiter takes precedence. | |
1318 | */ | |
1319 | delim_p = memchr(name, delim, namelen); | |
1320 | colon_p = memchr(name, ':', namelen); | |
1321 | ||
1322 | if (delim_p && colon_p) | |
1323 | end = delim_p < colon_p ? delim_p : colon_p; | |
1324 | else if (!delim_p && colon_p) | |
1325 | end = colon_p; | |
1326 | else { | |
1327 | end = delim_p; | |
1328 | if (!end) /* case: hostname:/ */ | |
1329 | end = name + namelen; | |
1330 | } | |
1331 | ||
1332 | if (end <= name) | |
1333 | return -EINVAL; | |
1334 | ||
1335 | /* do dns_resolve upcall */ | |
1336 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1337 | if (ip_len > 0) | |
1338 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1339 | else | |
1340 | ret = -ESRCH; | |
1341 | ||
1342 | kfree(ip_addr); | |
1343 | ||
1344 | *ipend = end; | |
1345 | ||
1346 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1347 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1348 | ||
1349 | return ret; | |
1350 | } | |
1351 | #else | |
1352 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1353 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1354 | { | |
1355 | return -EINVAL; | |
1356 | } | |
1357 | #endif | |
1358 | ||
1359 | /* | |
1360 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1361 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1362 | */ | |
1363 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1364 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1365 | { | |
1366 | int ret; | |
1367 | ||
1368 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1369 | if (ret) | |
1370 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1371 | ||
1372 | return ret; | |
1373 | } | |
1374 | ||
31b8006e SW |
1375 | /* |
1376 | * Parse an ip[:port] list into an addr array. Use the default | |
1377 | * monitor port if a port isn't specified. | |
1378 | */ | |
1379 | int ceph_parse_ips(const char *c, const char *end, | |
1380 | struct ceph_entity_addr *addr, | |
1381 | int max_count, int *count) | |
1382 | { | |
ee3b56f2 | 1383 | int i, ret = -EINVAL; |
31b8006e SW |
1384 | const char *p = c; |
1385 | ||
1386 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1387 | for (i = 0; i < max_count; i++) { | |
1388 | const char *ipend; | |
1389 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1390 | int port; |
39139f64 SW |
1391 | char delim = ','; |
1392 | ||
1393 | if (*p == '[') { | |
1394 | delim = ']'; | |
1395 | p++; | |
1396 | } | |
31b8006e | 1397 | |
ee3b56f2 NW |
1398 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1399 | if (ret) | |
31b8006e | 1400 | goto bad; |
ee3b56f2 NW |
1401 | ret = -EINVAL; |
1402 | ||
31b8006e SW |
1403 | p = ipend; |
1404 | ||
39139f64 SW |
1405 | if (delim == ']') { |
1406 | if (*p != ']') { | |
1407 | dout("missing matching ']'\n"); | |
1408 | goto bad; | |
1409 | } | |
1410 | p++; | |
1411 | } | |
1412 | ||
31b8006e SW |
1413 | /* port? */ |
1414 | if (p < end && *p == ':') { | |
1415 | port = 0; | |
1416 | p++; | |
1417 | while (p < end && *p >= '0' && *p <= '9') { | |
1418 | port = (port * 10) + (*p - '0'); | |
1419 | p++; | |
1420 | } | |
1421 | if (port > 65535 || port == 0) | |
1422 | goto bad; | |
1423 | } else { | |
1424 | port = CEPH_MON_PORT; | |
1425 | } | |
1426 | ||
1427 | addr_set_port(ss, port); | |
1428 | ||
3d14c5d2 | 1429 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1430 | |
1431 | if (p == end) | |
1432 | break; | |
1433 | if (*p != ',') | |
1434 | goto bad; | |
1435 | p++; | |
1436 | } | |
1437 | ||
1438 | if (p != end) | |
1439 | goto bad; | |
1440 | ||
1441 | if (count) | |
1442 | *count = i + 1; | |
1443 | return 0; | |
1444 | ||
1445 | bad: | |
39139f64 | 1446 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1447 | return ret; |
31b8006e | 1448 | } |
3d14c5d2 | 1449 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1450 | |
eed0ef2c | 1451 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1452 | { |
eed0ef2c | 1453 | dout("process_banner on %p\n", con); |
31b8006e SW |
1454 | |
1455 | if (verify_hello(con) < 0) | |
1456 | return -1; | |
1457 | ||
63f2d211 SW |
1458 | ceph_decode_addr(&con->actual_peer_addr); |
1459 | ceph_decode_addr(&con->peer_addr_for_me); | |
1460 | ||
31b8006e SW |
1461 | /* |
1462 | * Make sure the other end is who we wanted. note that the other | |
1463 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1464 | * them the benefit of the doubt. | |
1465 | */ | |
103e2d3a SW |
1466 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1467 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1468 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1469 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1470 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1471 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1472 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1473 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1474 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1475 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1476 | return -1; |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * did we learn our address? | |
1481 | */ | |
1482 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1483 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1484 | ||
1485 | memcpy(&con->msgr->inst.addr.in_addr, | |
1486 | &con->peer_addr_for_me.in_addr, | |
1487 | sizeof(con->peer_addr_for_me.in_addr)); | |
1488 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1489 | encode_my_addr(con->msgr); |
eed0ef2c | 1490 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1491 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1492 | } |
1493 | ||
eed0ef2c SW |
1494 | return 0; |
1495 | } | |
1496 | ||
04a419f9 SW |
1497 | static void fail_protocol(struct ceph_connection *con) |
1498 | { | |
1499 | reset_connection(con); | |
8dacc7da SW |
1500 | BUG_ON(con->state != CON_STATE_NEGOTIATING); |
1501 | con->state = CON_STATE_CLOSED; | |
04a419f9 SW |
1502 | } |
1503 | ||
eed0ef2c SW |
1504 | static int process_connect(struct ceph_connection *con) |
1505 | { | |
3d14c5d2 YS |
1506 | u64 sup_feat = con->msgr->supported_features; |
1507 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1508 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1509 | int ret; |
04a419f9 | 1510 | |
eed0ef2c SW |
1511 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1512 | ||
31b8006e | 1513 | switch (con->in_reply.tag) { |
04a419f9 SW |
1514 | case CEPH_MSGR_TAG_FEATURES: |
1515 | pr_err("%s%lld %s feature set mismatch," | |
1516 | " my %llx < server's %llx, missing %llx\n", | |
1517 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1518 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1519 | sup_feat, server_feat, server_feat & ~sup_feat); |
1520 | con->error_msg = "missing required protocol features"; | |
1521 | fail_protocol(con); | |
1522 | return -1; | |
1523 | ||
31b8006e | 1524 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1525 | pr_err("%s%lld %s protocol version mismatch," |
1526 | " my %d != server's %d\n", | |
1527 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1528 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1529 | le32_to_cpu(con->out_connect.protocol_version), |
1530 | le32_to_cpu(con->in_reply.protocol_version)); | |
1531 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1532 | fail_protocol(con); |
31b8006e SW |
1533 | return -1; |
1534 | ||
4e7a5dcd SW |
1535 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1536 | con->auth_retry++; | |
1537 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1538 | con->auth_retry); | |
1539 | if (con->auth_retry == 2) { | |
1540 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1541 | return -1; |
1542 | } | |
1543 | con->auth_retry = 1; | |
e825a66d | 1544 | ret = prepare_write_connect(con); |
0da5d703 SW |
1545 | if (ret < 0) |
1546 | return ret; | |
63733a0f | 1547 | prepare_read_connect(con); |
4e7a5dcd | 1548 | break; |
31b8006e SW |
1549 | |
1550 | case CEPH_MSGR_TAG_RESETSESSION: | |
1551 | /* | |
1552 | * If we connected with a large connect_seq but the peer | |
1553 | * has no record of a session with us (no connection, or | |
1554 | * connect_seq == 0), they will send RESETSESION to indicate | |
1555 | * that they must have reset their session, and may have | |
1556 | * dropped messages. | |
1557 | */ | |
1558 | dout("process_connect got RESET peer seq %u\n", | |
a16cb1f7 | 1559 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
1560 | pr_err("%s%lld %s connection reset\n", |
1561 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1562 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1563 | reset_connection(con); |
5a0f8fdd AE |
1564 | ret = prepare_write_connect(con); |
1565 | if (ret < 0) | |
1566 | return ret; | |
31b8006e SW |
1567 | prepare_read_connect(con); |
1568 | ||
1569 | /* Tell ceph about it. */ | |
ec302645 | 1570 | mutex_unlock(&con->mutex); |
31b8006e SW |
1571 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1572 | if (con->ops->peer_reset) | |
1573 | con->ops->peer_reset(con); | |
ec302645 | 1574 | mutex_lock(&con->mutex); |
8dacc7da | 1575 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 1576 | return -EAGAIN; |
31b8006e SW |
1577 | break; |
1578 | ||
1579 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1580 | /* | |
1581 | * If we sent a smaller connect_seq than the peer has, try | |
1582 | * again with a larger value. | |
1583 | */ | |
a16cb1f7 | 1584 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 1585 | le32_to_cpu(con->out_connect.connect_seq), |
a16cb1f7 SW |
1586 | le32_to_cpu(con->in_reply.connect_seq)); |
1587 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
5a0f8fdd AE |
1588 | ret = prepare_write_connect(con); |
1589 | if (ret < 0) | |
1590 | return ret; | |
31b8006e SW |
1591 | prepare_read_connect(con); |
1592 | break; | |
1593 | ||
1594 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1595 | /* | |
1596 | * If we sent a smaller global_seq than the peer has, try | |
1597 | * again with a larger value. | |
1598 | */ | |
eed0ef2c | 1599 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 1600 | con->peer_global_seq, |
a16cb1f7 | 1601 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 1602 | get_global_seq(con->msgr, |
a16cb1f7 | 1603 | le32_to_cpu(con->in_reply.global_seq)); |
5a0f8fdd AE |
1604 | ret = prepare_write_connect(con); |
1605 | if (ret < 0) | |
1606 | return ret; | |
31b8006e SW |
1607 | prepare_read_connect(con); |
1608 | break; | |
1609 | ||
1610 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1611 | if (req_feat & ~server_feat) { |
1612 | pr_err("%s%lld %s protocol feature mismatch," | |
1613 | " my required %llx > server's %llx, need %llx\n", | |
1614 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1615 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1616 | req_feat, server_feat, req_feat & ~server_feat); |
1617 | con->error_msg = "missing required protocol features"; | |
1618 | fail_protocol(con); | |
1619 | return -1; | |
1620 | } | |
8dacc7da SW |
1621 | |
1622 | BUG_ON(con->state != CON_STATE_NEGOTIATING); | |
1623 | con->state = CON_STATE_OPEN; | |
1624 | ||
31b8006e SW |
1625 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1626 | con->connect_seq++; | |
aba558e2 | 1627 | con->peer_features = server_feat; |
31b8006e SW |
1628 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1629 | con->peer_global_seq, | |
1630 | le32_to_cpu(con->in_reply.connect_seq), | |
1631 | con->connect_seq); | |
1632 | WARN_ON(con->connect_seq != | |
1633 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1634 | |
1635 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
4a861692 | 1636 | set_bit(CON_FLAG_LOSSYTX, &con->flags); |
92ac41d0 | 1637 | |
85effe18 SW |
1638 | con->delay = 0; /* reset backoff memory */ |
1639 | ||
31b8006e SW |
1640 | prepare_read_tag(con); |
1641 | break; | |
1642 | ||
1643 | case CEPH_MSGR_TAG_WAIT: | |
1644 | /* | |
1645 | * If there is a connection race (we are opening | |
1646 | * connections to each other), one of us may just have | |
1647 | * to WAIT. This shouldn't happen if we are the | |
1648 | * client. | |
1649 | */ | |
04177882 SW |
1650 | pr_err("process_connect got WAIT as client\n"); |
1651 | con->error_msg = "protocol error, got WAIT as client"; | |
1652 | return -1; | |
31b8006e SW |
1653 | |
1654 | default: | |
1655 | pr_err("connect protocol error, will retry\n"); | |
1656 | con->error_msg = "protocol error, garbage tag during connect"; | |
1657 | return -1; | |
1658 | } | |
1659 | return 0; | |
1660 | } | |
1661 | ||
1662 | ||
1663 | /* | |
1664 | * read (part of) an ack | |
1665 | */ | |
1666 | static int read_partial_ack(struct ceph_connection *con) | |
1667 | { | |
fd51653f AE |
1668 | int size = sizeof (con->in_temp_ack); |
1669 | int end = size; | |
1670 | ||
1671 | return read_partial(con, end, size, &con->in_temp_ack); | |
31b8006e SW |
1672 | } |
1673 | ||
1674 | ||
1675 | /* | |
1676 | * We can finally discard anything that's been acked. | |
1677 | */ | |
1678 | static void process_ack(struct ceph_connection *con) | |
1679 | { | |
1680 | struct ceph_msg *m; | |
1681 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1682 | u64 seq; | |
1683 | ||
31b8006e SW |
1684 | while (!list_empty(&con->out_sent)) { |
1685 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1686 | list_head); | |
1687 | seq = le64_to_cpu(m->hdr.seq); | |
1688 | if (seq > ack) | |
1689 | break; | |
1690 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1691 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1692 | m->ack_stamp = jiffies; |
31b8006e SW |
1693 | ceph_msg_remove(m); |
1694 | } | |
31b8006e SW |
1695 | prepare_read_tag(con); |
1696 | } | |
1697 | ||
1698 | ||
1699 | ||
1700 | ||
2450418c | 1701 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1702 | struct kvec *section, |
1703 | unsigned int sec_len, u32 *crc) | |
2450418c | 1704 | { |
68b4476b | 1705 | int ret, left; |
2450418c YS |
1706 | |
1707 | BUG_ON(!section); | |
1708 | ||
1709 | while (section->iov_len < sec_len) { | |
1710 | BUG_ON(section->iov_base == NULL); | |
1711 | left = sec_len - section->iov_len; | |
1712 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1713 | section->iov_len, left); | |
1714 | if (ret <= 0) | |
1715 | return ret; | |
1716 | section->iov_len += ret; | |
2450418c | 1717 | } |
fe3ad593 AE |
1718 | if (section->iov_len == sec_len) |
1719 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1720 | |
2450418c YS |
1721 | return 1; |
1722 | } | |
31b8006e | 1723 | |
1c20f2d2 AE |
1724 | static bool ceph_con_in_msg_alloc(struct ceph_connection *con, |
1725 | struct ceph_msg_header *hdr); | |
68b4476b YS |
1726 | |
1727 | ||
1728 | static int read_partial_message_pages(struct ceph_connection *con, | |
1729 | struct page **pages, | |
95c96174 | 1730 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1731 | { |
1732 | void *p; | |
1733 | int ret; | |
1734 | int left; | |
1735 | ||
1736 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1737 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1738 | /* (page) data */ | |
1739 | BUG_ON(pages == NULL); | |
1740 | p = kmap(pages[con->in_msg_pos.page]); | |
1741 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1742 | left); | |
bca064d2 | 1743 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1744 | con->in_data_crc = |
1745 | crc32c(con->in_data_crc, | |
1746 | p + con->in_msg_pos.page_pos, ret); | |
1747 | kunmap(pages[con->in_msg_pos.page]); | |
1748 | if (ret <= 0) | |
1749 | return ret; | |
1750 | con->in_msg_pos.data_pos += ret; | |
1751 | con->in_msg_pos.page_pos += ret; | |
1752 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1753 | con->in_msg_pos.page_pos = 0; | |
1754 | con->in_msg_pos.page++; | |
1755 | } | |
1756 | ||
1757 | return ret; | |
1758 | } | |
1759 | ||
1760 | #ifdef CONFIG_BLOCK | |
1761 | static int read_partial_message_bio(struct ceph_connection *con, | |
1762 | struct bio **bio_iter, int *bio_seg, | |
95c96174 | 1763 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1764 | { |
1765 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1766 | void *p; | |
1767 | int ret, left; | |
1768 | ||
68b4476b YS |
1769 | left = min((int)(data_len - con->in_msg_pos.data_pos), |
1770 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1771 | ||
1772 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1773 | ||
1774 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1775 | left); | |
bca064d2 | 1776 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1777 | con->in_data_crc = |
1778 | crc32c(con->in_data_crc, | |
1779 | p + con->in_msg_pos.page_pos, ret); | |
1780 | kunmap(bv->bv_page); | |
1781 | if (ret <= 0) | |
1782 | return ret; | |
1783 | con->in_msg_pos.data_pos += ret; | |
1784 | con->in_msg_pos.page_pos += ret; | |
1785 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1786 | con->in_msg_pos.page_pos = 0; | |
1787 | iter_bio_next(bio_iter, bio_seg); | |
1788 | } | |
1789 | ||
1790 | return ret; | |
1791 | } | |
1792 | #endif | |
1793 | ||
31b8006e SW |
1794 | /* |
1795 | * read (part of) a message. | |
1796 | */ | |
1797 | static int read_partial_message(struct ceph_connection *con) | |
1798 | { | |
1799 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1800 | int size; |
1801 | int end; | |
31b8006e | 1802 | int ret; |
95c96174 | 1803 | unsigned int front_len, middle_len, data_len; |
37675b0f | 1804 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 1805 | u64 seq; |
fe3ad593 | 1806 | u32 crc; |
31b8006e SW |
1807 | |
1808 | dout("read_partial_message con %p msg %p\n", con, m); | |
1809 | ||
1810 | /* header */ | |
fd51653f AE |
1811 | size = sizeof (con->in_hdr); |
1812 | end = size; | |
1813 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1814 | if (ret <= 0) |
1815 | return ret; | |
fe3ad593 AE |
1816 | |
1817 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1818 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1819 | pr_err("read_partial_message bad hdr " | |
1820 | " crc %u != expected %u\n", | |
1821 | crc, con->in_hdr.crc); | |
1822 | return -EBADMSG; | |
1823 | } | |
1824 | ||
31b8006e SW |
1825 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1826 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1827 | return -EIO; | |
1828 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1829 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1830 | return -EIO; | |
1831 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1832 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1833 | return -EIO; | |
1834 | ||
ae18756b SW |
1835 | /* verify seq# */ |
1836 | seq = le64_to_cpu(con->in_hdr.seq); | |
1837 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1838 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1839 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1840 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1841 | seq, con->in_seq + 1); |
1842 | con->in_base_pos = -front_len - middle_len - data_len - | |
1843 | sizeof(m->footer); | |
1844 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1845 | return 0; |
1846 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1847 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1848 | seq, con->in_seq + 1); | |
1849 | con->error_msg = "bad message sequence # for incoming message"; | |
1850 | return -EBADMSG; | |
1851 | } | |
1852 | ||
31b8006e SW |
1853 | /* allocate message? */ |
1854 | if (!con->in_msg) { | |
1855 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1856 | con->in_hdr.front_len, con->in_hdr.data_len); | |
1c20f2d2 | 1857 | if (ceph_con_in_msg_alloc(con, &con->in_hdr)) { |
31b8006e | 1858 | /* skip this message */ |
a79832f2 | 1859 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1860 | BUG_ON(con->in_msg); |
31b8006e SW |
1861 | con->in_base_pos = -front_len - middle_len - data_len - |
1862 | sizeof(m->footer); | |
1863 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1864 | con->in_seq++; |
31b8006e SW |
1865 | return 0; |
1866 | } | |
a79832f2 | 1867 | if (!con->in_msg) { |
5b3a4db3 SW |
1868 | con->error_msg = |
1869 | "error allocating memory for incoming message"; | |
a79832f2 | 1870 | return -ENOMEM; |
31b8006e | 1871 | } |
38941f80 AE |
1872 | |
1873 | BUG_ON(con->in_msg->con != con); | |
31b8006e SW |
1874 | m = con->in_msg; |
1875 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1876 | if (m->middle) |
1877 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1878 | |
1879 | con->in_msg_pos.page = 0; | |
68b4476b | 1880 | if (m->pages) |
c5c6b19d | 1881 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1882 | else |
1883 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1884 | con->in_msg_pos.data_pos = 0; |
a4107026 SW |
1885 | |
1886 | #ifdef CONFIG_BLOCK | |
1887 | if (m->bio) | |
1888 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1889 | #endif | |
31b8006e SW |
1890 | } |
1891 | ||
1892 | /* front */ | |
2450418c YS |
1893 | ret = read_partial_message_section(con, &m->front, front_len, |
1894 | &con->in_front_crc); | |
1895 | if (ret <= 0) | |
1896 | return ret; | |
31b8006e SW |
1897 | |
1898 | /* middle */ | |
2450418c | 1899 | if (m->middle) { |
213c99ee SW |
1900 | ret = read_partial_message_section(con, &m->middle->vec, |
1901 | middle_len, | |
2450418c | 1902 | &con->in_middle_crc); |
31b8006e SW |
1903 | if (ret <= 0) |
1904 | return ret; | |
31b8006e SW |
1905 | } |
1906 | ||
1907 | /* (page) data */ | |
31b8006e | 1908 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1909 | if (m->pages) { |
1910 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1911 | data_len, do_datacrc); |
68b4476b YS |
1912 | if (ret <= 0) |
1913 | return ret; | |
1914 | #ifdef CONFIG_BLOCK | |
1915 | } else if (m->bio) { | |
a4107026 | 1916 | BUG_ON(!m->bio_iter); |
68b4476b YS |
1917 | ret = read_partial_message_bio(con, |
1918 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1919 | data_len, do_datacrc); |
68b4476b YS |
1920 | if (ret <= 0) |
1921 | return ret; | |
1922 | #endif | |
1923 | } else { | |
1924 | BUG_ON(1); | |
31b8006e SW |
1925 | } |
1926 | } | |
1927 | ||
31b8006e | 1928 | /* footer */ |
fd51653f AE |
1929 | size = sizeof (m->footer); |
1930 | end += size; | |
1931 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
1932 | if (ret <= 0) |
1933 | return ret; | |
1934 | ||
31b8006e SW |
1935 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
1936 | m, front_len, m->footer.front_crc, middle_len, | |
1937 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1938 | ||
1939 | /* crc ok? */ | |
1940 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1941 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1942 | m, con->in_front_crc, m->footer.front_crc); | |
1943 | return -EBADMSG; | |
1944 | } | |
1945 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1946 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1947 | m, con->in_middle_crc, m->footer.middle_crc); | |
1948 | return -EBADMSG; | |
1949 | } | |
bca064d2 | 1950 | if (do_datacrc && |
31b8006e SW |
1951 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
1952 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1953 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1954 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1955 | return -EBADMSG; | |
1956 | } | |
1957 | ||
1958 | return 1; /* done! */ | |
1959 | } | |
1960 | ||
1961 | /* | |
1962 | * Process message. This happens in the worker thread. The callback should | |
1963 | * be careful not to do anything that waits on other incoming messages or it | |
1964 | * may deadlock. | |
1965 | */ | |
1966 | static void process_message(struct ceph_connection *con) | |
1967 | { | |
5e095e8b | 1968 | struct ceph_msg *msg; |
31b8006e | 1969 | |
38941f80 AE |
1970 | BUG_ON(con->in_msg->con != con); |
1971 | con->in_msg->con = NULL; | |
5e095e8b | 1972 | msg = con->in_msg; |
31b8006e | 1973 | con->in_msg = NULL; |
36eb71aa | 1974 | con->ops->put(con); |
31b8006e SW |
1975 | |
1976 | /* if first message, set peer_name */ | |
1977 | if (con->peer_name.type == 0) | |
dbad185d | 1978 | con->peer_name = msg->hdr.src; |
31b8006e | 1979 | |
31b8006e | 1980 | con->in_seq++; |
ec302645 | 1981 | mutex_unlock(&con->mutex); |
31b8006e SW |
1982 | |
1983 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1984 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1985 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1986 | le16_to_cpu(msg->hdr.type), |
1987 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1988 | le32_to_cpu(msg->hdr.front_len), | |
1989 | le32_to_cpu(msg->hdr.data_len), | |
1990 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1991 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1992 | |
1993 | mutex_lock(&con->mutex); | |
31b8006e SW |
1994 | prepare_read_tag(con); |
1995 | } | |
1996 | ||
1997 | ||
1998 | /* | |
1999 | * Write something to the socket. Called in a worker thread when the | |
2000 | * socket appears to be writeable and we have something ready to send. | |
2001 | */ | |
2002 | static int try_write(struct ceph_connection *con) | |
2003 | { | |
31b8006e SW |
2004 | int ret = 1; |
2005 | ||
d59315ca | 2006 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 2007 | |
31b8006e SW |
2008 | more: |
2009 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
2010 | ||
2011 | /* open the socket first? */ | |
8dacc7da SW |
2012 | if (con->state == CON_STATE_PREOPEN) { |
2013 | BUG_ON(con->sock); | |
2014 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2015 | |
e2200423 | 2016 | con_out_kvec_reset(con); |
e825a66d | 2017 | prepare_write_banner(con); |
eed0ef2c | 2018 | prepare_read_banner(con); |
31b8006e | 2019 | |
cf3e5c40 | 2020 | BUG_ON(con->in_msg); |
31b8006e SW |
2021 | con->in_tag = CEPH_MSGR_TAG_READY; |
2022 | dout("try_write initiating connect on %p new state %lu\n", | |
2023 | con, con->state); | |
41617d0c AE |
2024 | ret = ceph_tcp_connect(con); |
2025 | if (ret < 0) { | |
31b8006e | 2026 | con->error_msg = "connect error"; |
31b8006e SW |
2027 | goto out; |
2028 | } | |
2029 | } | |
2030 | ||
2031 | more_kvec: | |
2032 | /* kvec data queued? */ | |
2033 | if (con->out_skip) { | |
2034 | ret = write_partial_skip(con); | |
2035 | if (ret <= 0) | |
42961d23 | 2036 | goto out; |
31b8006e SW |
2037 | } |
2038 | if (con->out_kvec_left) { | |
2039 | ret = write_partial_kvec(con); | |
2040 | if (ret <= 0) | |
42961d23 | 2041 | goto out; |
31b8006e SW |
2042 | } |
2043 | ||
2044 | /* msg pages? */ | |
2045 | if (con->out_msg) { | |
c86a2930 SW |
2046 | if (con->out_msg_done) { |
2047 | ceph_msg_put(con->out_msg); | |
2048 | con->out_msg = NULL; /* we're done with this one */ | |
2049 | goto do_next; | |
2050 | } | |
2051 | ||
31b8006e SW |
2052 | ret = write_partial_msg_pages(con); |
2053 | if (ret == 1) | |
2054 | goto more_kvec; /* we need to send the footer, too! */ | |
2055 | if (ret == 0) | |
42961d23 | 2056 | goto out; |
31b8006e SW |
2057 | if (ret < 0) { |
2058 | dout("try_write write_partial_msg_pages err %d\n", | |
2059 | ret); | |
42961d23 | 2060 | goto out; |
31b8006e SW |
2061 | } |
2062 | } | |
2063 | ||
c86a2930 | 2064 | do_next: |
8dacc7da | 2065 | if (con->state == CON_STATE_OPEN) { |
31b8006e SW |
2066 | /* is anything else pending? */ |
2067 | if (!list_empty(&con->out_queue)) { | |
2068 | prepare_write_message(con); | |
2069 | goto more; | |
2070 | } | |
2071 | if (con->in_seq > con->in_seq_acked) { | |
2072 | prepare_write_ack(con); | |
2073 | goto more; | |
2074 | } | |
4a861692 SW |
2075 | if (test_and_clear_bit(CON_FLAG_KEEPALIVE_PENDING, |
2076 | &con->flags)) { | |
31b8006e SW |
2077 | prepare_write_keepalive(con); |
2078 | goto more; | |
2079 | } | |
2080 | } | |
2081 | ||
2082 | /* Nothing to do! */ | |
4a861692 | 2083 | clear_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
31b8006e | 2084 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2085 | ret = 0; |
2086 | out: | |
42961d23 | 2087 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2088 | return ret; |
2089 | } | |
2090 | ||
2091 | ||
2092 | ||
2093 | /* | |
2094 | * Read what we can from the socket. | |
2095 | */ | |
2096 | static int try_read(struct ceph_connection *con) | |
2097 | { | |
31b8006e SW |
2098 | int ret = -1; |
2099 | ||
8dacc7da SW |
2100 | more: |
2101 | dout("try_read start on %p state %lu\n", con, con->state); | |
2102 | if (con->state != CON_STATE_CONNECTING && | |
2103 | con->state != CON_STATE_NEGOTIATING && | |
2104 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2105 | return 0; |
2106 | ||
8dacc7da | 2107 | BUG_ON(!con->sock); |
ec302645 | 2108 | |
31b8006e SW |
2109 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2110 | con->in_base_pos); | |
0da5d703 | 2111 | |
8dacc7da | 2112 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2113 | dout("try_read connecting\n"); |
2114 | ret = read_partial_banner(con); | |
2115 | if (ret <= 0) | |
ab166d5a | 2116 | goto out; |
7593af92 AE |
2117 | ret = process_banner(con); |
2118 | if (ret < 0) | |
2119 | goto out; | |
2120 | ||
8dacc7da SW |
2121 | BUG_ON(con->state != CON_STATE_CONNECTING); |
2122 | con->state = CON_STATE_NEGOTIATING; | |
7593af92 AE |
2123 | |
2124 | /* Banner is good, exchange connection info */ | |
2125 | ret = prepare_write_connect(con); | |
2126 | if (ret < 0) | |
2127 | goto out; | |
2128 | prepare_read_connect(con); | |
2129 | ||
2130 | /* Send connection info before awaiting response */ | |
2131 | goto out; | |
2132 | } | |
2133 | ||
8dacc7da | 2134 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2135 | dout("try_read negotiating\n"); |
31b8006e SW |
2136 | ret = read_partial_connect(con); |
2137 | if (ret <= 0) | |
31b8006e | 2138 | goto out; |
98bdb0aa SW |
2139 | ret = process_connect(con); |
2140 | if (ret < 0) | |
2141 | goto out; | |
31b8006e SW |
2142 | goto more; |
2143 | } | |
2144 | ||
8dacc7da SW |
2145 | BUG_ON(con->state != CON_STATE_OPEN); |
2146 | ||
31b8006e SW |
2147 | if (con->in_base_pos < 0) { |
2148 | /* | |
2149 | * skipping + discarding content. | |
2150 | * | |
2151 | * FIXME: there must be a better way to do this! | |
2152 | */ | |
84495f49 AE |
2153 | static char buf[SKIP_BUF_SIZE]; |
2154 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2155 | ||
31b8006e SW |
2156 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2157 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2158 | if (ret <= 0) | |
98bdb0aa | 2159 | goto out; |
31b8006e SW |
2160 | con->in_base_pos += ret; |
2161 | if (con->in_base_pos) | |
2162 | goto more; | |
2163 | } | |
2164 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2165 | /* | |
2166 | * what's next? | |
2167 | */ | |
2168 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2169 | if (ret <= 0) | |
98bdb0aa | 2170 | goto out; |
31b8006e SW |
2171 | dout("try_read got tag %d\n", (int)con->in_tag); |
2172 | switch (con->in_tag) { | |
2173 | case CEPH_MSGR_TAG_MSG: | |
2174 | prepare_read_message(con); | |
2175 | break; | |
2176 | case CEPH_MSGR_TAG_ACK: | |
2177 | prepare_read_ack(con); | |
2178 | break; | |
2179 | case CEPH_MSGR_TAG_CLOSE: | |
8dacc7da SW |
2180 | con_close_socket(con); |
2181 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2182 | goto out; |
31b8006e SW |
2183 | default: |
2184 | goto bad_tag; | |
2185 | } | |
2186 | } | |
2187 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2188 | ret = read_partial_message(con); | |
2189 | if (ret <= 0) { | |
2190 | switch (ret) { | |
2191 | case -EBADMSG: | |
2192 | con->error_msg = "bad crc"; | |
2193 | ret = -EIO; | |
98bdb0aa | 2194 | break; |
31b8006e SW |
2195 | case -EIO: |
2196 | con->error_msg = "io error"; | |
98bdb0aa | 2197 | break; |
31b8006e | 2198 | } |
98bdb0aa | 2199 | goto out; |
31b8006e SW |
2200 | } |
2201 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2202 | goto more; | |
2203 | process_message(con); | |
2204 | goto more; | |
2205 | } | |
2206 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2207 | ret = read_partial_ack(con); | |
2208 | if (ret <= 0) | |
98bdb0aa | 2209 | goto out; |
31b8006e SW |
2210 | process_ack(con); |
2211 | goto more; | |
2212 | } | |
2213 | ||
31b8006e | 2214 | out: |
98bdb0aa | 2215 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2216 | return ret; |
2217 | ||
2218 | bad_tag: | |
2219 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2220 | con->error_msg = "protocol error, garbage tag"; | |
2221 | ret = -1; | |
2222 | goto out; | |
2223 | } | |
2224 | ||
2225 | ||
2226 | /* | |
2227 | * Atomically queue work on a connection. Bump @con reference to | |
2228 | * avoid races with connection teardown. | |
31b8006e SW |
2229 | */ |
2230 | static void queue_con(struct ceph_connection *con) | |
2231 | { | |
31b8006e SW |
2232 | if (!con->ops->get(con)) { |
2233 | dout("queue_con %p ref count 0\n", con); | |
2234 | return; | |
2235 | } | |
2236 | ||
f363e45f | 2237 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2238 | dout("queue_con %p - already queued\n", con); |
2239 | con->ops->put(con); | |
2240 | } else { | |
2241 | dout("queue_con %p\n", con); | |
2242 | } | |
2243 | } | |
2244 | ||
2245 | /* | |
2246 | * Do some work on a connection. Drop a connection ref when we're done. | |
2247 | */ | |
2248 | static void con_work(struct work_struct *work) | |
2249 | { | |
2250 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2251 | work.work); | |
0da5d703 | 2252 | int ret; |
31b8006e | 2253 | |
9dd4658d | 2254 | mutex_lock(&con->mutex); |
0da5d703 | 2255 | restart: |
4a861692 | 2256 | if (test_and_clear_bit(CON_FLAG_SOCK_CLOSED, &con->flags)) { |
8dacc7da SW |
2257 | switch (con->state) { |
2258 | case CON_STATE_CONNECTING: | |
188048bc | 2259 | con->error_msg = "connection failed"; |
8dacc7da SW |
2260 | break; |
2261 | case CON_STATE_NEGOTIATING: | |
2262 | con->error_msg = "negotiation failed"; | |
2263 | break; | |
2264 | case CON_STATE_OPEN: | |
2265 | con->error_msg = "socket closed"; | |
2266 | break; | |
2267 | default: | |
2268 | dout("unrecognized con state %d\n", (int)con->state); | |
e27947c7 | 2269 | con->error_msg = "unrecognized con state"; |
8dacc7da SW |
2270 | BUG(); |
2271 | } | |
188048bc AE |
2272 | goto fault; |
2273 | } | |
2274 | ||
4a861692 | 2275 | if (test_and_clear_bit(CON_FLAG_BACKOFF, &con->flags)) { |
60bf8bf8 SW |
2276 | dout("con_work %p backing off\n", con); |
2277 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2278 | round_jiffies_relative(con->delay))) { | |
2279 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2280 | mutex_unlock(&con->mutex); | |
2281 | return; | |
2282 | } else { | |
2283 | con->ops->put(con); | |
2284 | dout("con_work %p FAILED to back off %lu\n", con, | |
2285 | con->delay); | |
2286 | } | |
2287 | } | |
9dd4658d | 2288 | |
8dacc7da | 2289 | if (con->state == CON_STATE_STANDBY) { |
e00de341 SW |
2290 | dout("con_work %p STANDBY\n", con); |
2291 | goto done; | |
2292 | } | |
8dacc7da | 2293 | if (con->state == CON_STATE_CLOSED) { |
2e8cb100 SW |
2294 | dout("con_work %p CLOSED\n", con); |
2295 | BUG_ON(con->sock); | |
31b8006e SW |
2296 | goto done; |
2297 | } | |
8dacc7da | 2298 | if (con->state == CON_STATE_PREOPEN) { |
31b8006e | 2299 | dout("con_work OPENING\n"); |
2e8cb100 | 2300 | BUG_ON(con->sock); |
31b8006e SW |
2301 | } |
2302 | ||
0da5d703 SW |
2303 | ret = try_read(con); |
2304 | if (ret == -EAGAIN) | |
2305 | goto restart; | |
3a140a0d SW |
2306 | if (ret < 0) { |
2307 | con->error_msg = "socket error on read"; | |
0da5d703 | 2308 | goto fault; |
3a140a0d | 2309 | } |
0da5d703 SW |
2310 | |
2311 | ret = try_write(con); | |
2312 | if (ret == -EAGAIN) | |
2313 | goto restart; | |
3a140a0d SW |
2314 | if (ret < 0) { |
2315 | con->error_msg = "socket error on write"; | |
0da5d703 | 2316 | goto fault; |
3a140a0d | 2317 | } |
31b8006e SW |
2318 | |
2319 | done: | |
9dd4658d | 2320 | mutex_unlock(&con->mutex); |
9dd4658d | 2321 | done_unlocked: |
31b8006e | 2322 | con->ops->put(con); |
0da5d703 SW |
2323 | return; |
2324 | ||
2325 | fault: | |
2326 | mutex_unlock(&con->mutex); | |
2327 | ceph_fault(con); /* error/fault path */ | |
2328 | goto done_unlocked; | |
31b8006e SW |
2329 | } |
2330 | ||
2331 | ||
2332 | /* | |
2333 | * Generic error/fault handler. A retry mechanism is used with | |
2334 | * exponential backoff | |
2335 | */ | |
2336 | static void ceph_fault(struct ceph_connection *con) | |
2337 | { | |
3b5ede07 SW |
2338 | mutex_lock(&con->mutex); |
2339 | ||
31b8006e | 2340 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2341 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2342 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2343 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2344 | |
8dacc7da SW |
2345 | BUG_ON(con->state != CON_STATE_CONNECTING && |
2346 | con->state != CON_STATE_NEGOTIATING && | |
2347 | con->state != CON_STATE_OPEN); | |
ec302645 | 2348 | |
31b8006e | 2349 | con_close_socket(con); |
5e095e8b | 2350 | |
4a861692 | 2351 | if (test_bit(CON_FLAG_LOSSYTX, &con->flags)) { |
8dacc7da SW |
2352 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
2353 | con->state = CON_STATE_CLOSED; | |
3b5ede07 SW |
2354 | goto out_unlock; |
2355 | } | |
2356 | ||
5e095e8b | 2357 | if (con->in_msg) { |
38941f80 AE |
2358 | BUG_ON(con->in_msg->con != con); |
2359 | con->in_msg->con = NULL; | |
5e095e8b SW |
2360 | ceph_msg_put(con->in_msg); |
2361 | con->in_msg = NULL; | |
36eb71aa | 2362 | con->ops->put(con); |
5e095e8b | 2363 | } |
31b8006e | 2364 | |
e80a52d1 SW |
2365 | /* Requeue anything that hasn't been acked */ |
2366 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2367 | |
e76661d0 SW |
2368 | /* If there are no messages queued or keepalive pending, place |
2369 | * the connection in a STANDBY state */ | |
2370 | if (list_empty(&con->out_queue) && | |
4a861692 | 2371 | !test_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags)) { |
e00de341 | 2372 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
4a861692 | 2373 | clear_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
8dacc7da | 2374 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
2375 | } else { |
2376 | /* retry after a delay. */ | |
8dacc7da | 2377 | con->state = CON_STATE_PREOPEN; |
e80a52d1 SW |
2378 | if (con->delay == 0) |
2379 | con->delay = BASE_DELAY_INTERVAL; | |
2380 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2381 | con->delay *= 2; | |
e80a52d1 SW |
2382 | con->ops->get(con); |
2383 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2384 | round_jiffies_relative(con->delay))) { |
2385 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2386 | } else { | |
e80a52d1 | 2387 | con->ops->put(con); |
60bf8bf8 SW |
2388 | dout("fault failed to queue %p delay %lu, backoff\n", |
2389 | con, con->delay); | |
2390 | /* | |
2391 | * In many cases we see a socket state change | |
2392 | * while con_work is running and end up | |
2393 | * queuing (non-delayed) work, such that we | |
2394 | * can't backoff with a delay. Set a flag so | |
2395 | * that when con_work restarts we schedule the | |
2396 | * delay then. | |
2397 | */ | |
4a861692 | 2398 | set_bit(CON_FLAG_BACKOFF, &con->flags); |
60bf8bf8 | 2399 | } |
31b8006e SW |
2400 | } |
2401 | ||
91e45ce3 SW |
2402 | out_unlock: |
2403 | mutex_unlock(&con->mutex); | |
161fd65a SW |
2404 | /* |
2405 | * in case we faulted due to authentication, invalidate our | |
2406 | * current tickets so that we can get new ones. | |
213c99ee | 2407 | */ |
161fd65a SW |
2408 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2409 | dout("calling invalidate_authorizer()\n"); | |
2410 | con->ops->invalidate_authorizer(con); | |
2411 | } | |
2412 | ||
31b8006e SW |
2413 | if (con->ops->fault) |
2414 | con->ops->fault(con); | |
2415 | } | |
2416 | ||
2417 | ||
2418 | ||
2419 | /* | |
15d9882c | 2420 | * initialize a new messenger instance |
31b8006e | 2421 | */ |
15d9882c AE |
2422 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2423 | struct ceph_entity_addr *myaddr, | |
2424 | u32 supported_features, | |
2425 | u32 required_features, | |
2426 | bool nocrc) | |
31b8006e | 2427 | { |
3d14c5d2 YS |
2428 | msgr->supported_features = supported_features; |
2429 | msgr->required_features = required_features; | |
2430 | ||
31b8006e SW |
2431 | spin_lock_init(&msgr->global_seq_lock); |
2432 | ||
31b8006e SW |
2433 | if (myaddr) |
2434 | msgr->inst.addr = *myaddr; | |
2435 | ||
2436 | /* select a random nonce */ | |
ac8839d7 | 2437 | msgr->inst.addr.type = 0; |
103e2d3a | 2438 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2439 | encode_my_addr(msgr); |
15d9882c | 2440 | msgr->nocrc = nocrc; |
31b8006e | 2441 | |
a2a32584 GH |
2442 | atomic_set(&msgr->stopping, 0); |
2443 | ||
15d9882c | 2444 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2445 | } |
15d9882c | 2446 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2447 | |
e00de341 SW |
2448 | static void clear_standby(struct ceph_connection *con) |
2449 | { | |
2450 | /* come back from STANDBY? */ | |
8dacc7da | 2451 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 2452 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 2453 | con->state = CON_STATE_PREOPEN; |
e00de341 | 2454 | con->connect_seq++; |
4a861692 SW |
2455 | WARN_ON(test_bit(CON_FLAG_WRITE_PENDING, &con->flags)); |
2456 | WARN_ON(test_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags)); | |
e00de341 SW |
2457 | } |
2458 | } | |
2459 | ||
31b8006e SW |
2460 | /* |
2461 | * Queue up an outgoing message on the given connection. | |
2462 | */ | |
2463 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2464 | { | |
31b8006e | 2465 | /* set src+dst */ |
dbad185d | 2466 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 2467 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
2468 | msg->needs_out_seq = true; |
2469 | ||
ec302645 | 2470 | mutex_lock(&con->mutex); |
92ce034b | 2471 | |
8dacc7da | 2472 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
2473 | dout("con_send %p closed, dropping %p\n", con, msg); |
2474 | ceph_msg_put(msg); | |
2475 | mutex_unlock(&con->mutex); | |
2476 | return; | |
2477 | } | |
2478 | ||
38941f80 | 2479 | BUG_ON(msg->con != NULL); |
36eb71aa | 2480 | msg->con = con->ops->get(con); |
92ce034b AE |
2481 | BUG_ON(msg->con == NULL); |
2482 | ||
31b8006e SW |
2483 | BUG_ON(!list_empty(&msg->list_head)); |
2484 | list_add_tail(&msg->list_head, &con->out_queue); | |
2485 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2486 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2487 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2488 | le32_to_cpu(msg->hdr.front_len), | |
2489 | le32_to_cpu(msg->hdr.middle_len), | |
2490 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
2491 | |
2492 | clear_standby(con); | |
ec302645 | 2493 | mutex_unlock(&con->mutex); |
31b8006e SW |
2494 | |
2495 | /* if there wasn't anything waiting to send before, queue | |
2496 | * new work */ | |
4a861692 | 2497 | if (test_and_set_bit(CON_FLAG_WRITE_PENDING, &con->flags) == 0) |
31b8006e SW |
2498 | queue_con(con); |
2499 | } | |
3d14c5d2 | 2500 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2501 | |
2502 | /* | |
2503 | * Revoke a message that was previously queued for send | |
2504 | */ | |
6740a845 | 2505 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2506 | { |
6740a845 AE |
2507 | struct ceph_connection *con = msg->con; |
2508 | ||
2509 | if (!con) | |
2510 | return; /* Message not in our possession */ | |
2511 | ||
ec302645 | 2512 | mutex_lock(&con->mutex); |
31b8006e | 2513 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2514 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2515 | list_del_init(&msg->list_head); |
38941f80 | 2516 | BUG_ON(msg->con == NULL); |
36eb71aa | 2517 | msg->con->ops->put(msg->con); |
38941f80 | 2518 | msg->con = NULL; |
92ce034b | 2519 | msg->hdr.seq = 0; |
38941f80 | 2520 | |
31b8006e | 2521 | ceph_msg_put(msg); |
ed98adad SW |
2522 | } |
2523 | if (con->out_msg == msg) { | |
38941f80 | 2524 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2525 | con->out_msg = NULL; |
31b8006e SW |
2526 | if (con->out_kvec_is_msg) { |
2527 | con->out_skip = con->out_kvec_bytes; | |
2528 | con->out_kvec_is_msg = false; | |
2529 | } | |
ed98adad | 2530 | msg->hdr.seq = 0; |
92ce034b AE |
2531 | |
2532 | ceph_msg_put(msg); | |
31b8006e | 2533 | } |
ec302645 | 2534 | mutex_unlock(&con->mutex); |
31b8006e SW |
2535 | } |
2536 | ||
350b1c32 | 2537 | /* |
0d59ab81 | 2538 | * Revoke a message that we may be reading data into |
350b1c32 | 2539 | */ |
8921d114 | 2540 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2541 | { |
8921d114 AE |
2542 | struct ceph_connection *con; |
2543 | ||
2544 | BUG_ON(msg == NULL); | |
2545 | if (!msg->con) { | |
2546 | dout("%s msg %p null con\n", __func__, msg); | |
2547 | ||
2548 | return; /* Message not in our possession */ | |
2549 | } | |
2550 | ||
2551 | con = msg->con; | |
350b1c32 | 2552 | mutex_lock(&con->mutex); |
8921d114 | 2553 | if (con->in_msg == msg) { |
95c96174 ED |
2554 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
2555 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
2556 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
2557 | |
2558 | /* skip rest of message */ | |
8921d114 AE |
2559 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
2560 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 2561 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
2562 | front_len - |
2563 | middle_len - | |
2564 | data_len - | |
350b1c32 | 2565 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2566 | ceph_msg_put(con->in_msg); |
2567 | con->in_msg = NULL; | |
2568 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2569 | con->in_seq++; |
350b1c32 | 2570 | } else { |
8921d114 AE |
2571 | dout("%s %p in_msg %p msg %p no-op\n", |
2572 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
2573 | } |
2574 | mutex_unlock(&con->mutex); | |
2575 | } | |
2576 | ||
31b8006e SW |
2577 | /* |
2578 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2579 | */ | |
2580 | void ceph_con_keepalive(struct ceph_connection *con) | |
2581 | { | |
e00de341 | 2582 | dout("con_keepalive %p\n", con); |
00650931 | 2583 | mutex_lock(&con->mutex); |
e00de341 | 2584 | clear_standby(con); |
00650931 | 2585 | mutex_unlock(&con->mutex); |
4a861692 SW |
2586 | if (test_and_set_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags) == 0 && |
2587 | test_and_set_bit(CON_FLAG_WRITE_PENDING, &con->flags) == 0) | |
31b8006e SW |
2588 | queue_con(con); |
2589 | } | |
3d14c5d2 | 2590 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2591 | |
2592 | ||
2593 | /* | |
2594 | * construct a new message with given type, size | |
2595 | * the new msg has a ref count of 1. | |
2596 | */ | |
b61c2763 SW |
2597 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2598 | bool can_fail) | |
31b8006e SW |
2599 | { |
2600 | struct ceph_msg *m; | |
2601 | ||
34d23762 | 2602 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2603 | if (m == NULL) |
2604 | goto out; | |
c2e552e7 | 2605 | kref_init(&m->kref); |
38941f80 AE |
2606 | |
2607 | m->con = NULL; | |
31b8006e SW |
2608 | INIT_LIST_HEAD(&m->list_head); |
2609 | ||
45c6ceb5 | 2610 | m->hdr.tid = 0; |
31b8006e | 2611 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2612 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2613 | m->hdr.version = 0; | |
31b8006e SW |
2614 | m->hdr.front_len = cpu_to_le32(front_len); |
2615 | m->hdr.middle_len = 0; | |
bb257664 SW |
2616 | m->hdr.data_len = 0; |
2617 | m->hdr.data_off = 0; | |
45c6ceb5 | 2618 | m->hdr.reserved = 0; |
31b8006e SW |
2619 | m->footer.front_crc = 0; |
2620 | m->footer.middle_crc = 0; | |
2621 | m->footer.data_crc = 0; | |
45c6ceb5 | 2622 | m->footer.flags = 0; |
31b8006e SW |
2623 | m->front_max = front_len; |
2624 | m->front_is_vmalloc = false; | |
2625 | m->more_to_follow = false; | |
c0d5f9db | 2626 | m->ack_stamp = 0; |
31b8006e SW |
2627 | m->pool = NULL; |
2628 | ||
ca20892d HC |
2629 | /* middle */ |
2630 | m->middle = NULL; | |
2631 | ||
2632 | /* data */ | |
2633 | m->nr_pages = 0; | |
2634 | m->page_alignment = 0; | |
2635 | m->pages = NULL; | |
2636 | m->pagelist = NULL; | |
2637 | m->bio = NULL; | |
2638 | m->bio_iter = NULL; | |
2639 | m->bio_seg = 0; | |
2640 | m->trail = NULL; | |
2641 | ||
31b8006e SW |
2642 | /* front */ |
2643 | if (front_len) { | |
2644 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2645 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2646 | PAGE_KERNEL); |
2647 | m->front_is_vmalloc = true; | |
2648 | } else { | |
34d23762 | 2649 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2650 | } |
2651 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2652 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2653 | front_len); |
2654 | goto out2; | |
2655 | } | |
2656 | } else { | |
2657 | m->front.iov_base = NULL; | |
2658 | } | |
2659 | m->front.iov_len = front_len; | |
2660 | ||
bb257664 | 2661 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2662 | return m; |
2663 | ||
2664 | out2: | |
2665 | ceph_msg_put(m); | |
2666 | out: | |
b61c2763 SW |
2667 | if (!can_fail) { |
2668 | pr_err("msg_new can't create type %d front %d\n", type, | |
2669 | front_len); | |
f0ed1b7c | 2670 | WARN_ON(1); |
b61c2763 SW |
2671 | } else { |
2672 | dout("msg_new can't create type %d front %d\n", type, | |
2673 | front_len); | |
2674 | } | |
a79832f2 | 2675 | return NULL; |
31b8006e | 2676 | } |
3d14c5d2 | 2677 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2678 | |
31b8006e SW |
2679 | /* |
2680 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2681 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2682 | * per-type header in the front and then gracefully fail (i.e., | |
2683 | * propagate the error to the caller based on info in the front) when | |
2684 | * the middle is too large. | |
2685 | */ | |
2450418c | 2686 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2687 | { |
2688 | int type = le16_to_cpu(msg->hdr.type); | |
2689 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2690 | ||
2691 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2692 | ceph_msg_type_name(type), middle_len); | |
2693 | BUG_ON(!middle_len); | |
2694 | BUG_ON(msg->middle); | |
2695 | ||
b6c1d5b8 | 2696 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2697 | if (!msg->middle) |
2698 | return -ENOMEM; | |
2699 | return 0; | |
2700 | } | |
2701 | ||
2450418c | 2702 | /* |
1c20f2d2 AE |
2703 | * Allocate a message for receiving an incoming message on a |
2704 | * connection, and save the result in con->in_msg. Uses the | |
2705 | * connection's private alloc_msg op if available. | |
2706 | * | |
2707 | * Returns true if the message should be skipped, false otherwise. | |
2708 | * If true is returned (skip message), con->in_msg will be NULL. | |
2709 | * If false is returned, con->in_msg will contain a pointer to the | |
2710 | * newly-allocated message, or NULL in case of memory exhaustion. | |
2450418c | 2711 | */ |
1c20f2d2 AE |
2712 | static bool ceph_con_in_msg_alloc(struct ceph_connection *con, |
2713 | struct ceph_msg_header *hdr) | |
2450418c YS |
2714 | { |
2715 | int type = le16_to_cpu(hdr->type); | |
2716 | int front_len = le32_to_cpu(hdr->front_len); | |
2717 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2450418c YS |
2718 | int ret; |
2719 | ||
1c20f2d2 AE |
2720 | BUG_ON(con->in_msg != NULL); |
2721 | ||
2450418c | 2722 | if (con->ops->alloc_msg) { |
1c20f2d2 AE |
2723 | int skip = 0; |
2724 | ||
0547a9b3 | 2725 | mutex_unlock(&con->mutex); |
1c20f2d2 | 2726 | con->in_msg = con->ops->alloc_msg(con, hdr, &skip); |
0547a9b3 | 2727 | mutex_lock(&con->mutex); |
92ce034b | 2728 | if (con->in_msg) { |
36eb71aa | 2729 | con->in_msg->con = con->ops->get(con); |
92ce034b AE |
2730 | BUG_ON(con->in_msg->con == NULL); |
2731 | } | |
1c20f2d2 AE |
2732 | if (skip) |
2733 | con->in_msg = NULL; | |
2734 | ||
2735 | if (!con->in_msg) | |
2736 | return skip != 0; | |
2450418c | 2737 | } |
1c20f2d2 AE |
2738 | if (!con->in_msg) { |
2739 | con->in_msg = ceph_msg_new(type, front_len, GFP_NOFS, false); | |
2740 | if (!con->in_msg) { | |
2450418c YS |
2741 | pr_err("unable to allocate msg type %d len %d\n", |
2742 | type, front_len); | |
1c20f2d2 | 2743 | return false; |
2450418c | 2744 | } |
36eb71aa | 2745 | con->in_msg->con = con->ops->get(con); |
92ce034b | 2746 | BUG_ON(con->in_msg->con == NULL); |
1c20f2d2 | 2747 | con->in_msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2748 | } |
1c20f2d2 | 2749 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2750 | |
1c20f2d2 AE |
2751 | if (middle_len && !con->in_msg->middle) { |
2752 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 2753 | if (ret < 0) { |
1c20f2d2 AE |
2754 | ceph_msg_put(con->in_msg); |
2755 | con->in_msg = NULL; | |
2450418c YS |
2756 | } |
2757 | } | |
9d7f0f13 | 2758 | |
1c20f2d2 | 2759 | return false; |
2450418c YS |
2760 | } |
2761 | ||
31b8006e SW |
2762 | |
2763 | /* | |
2764 | * Free a generically kmalloc'd message. | |
2765 | */ | |
2766 | void ceph_msg_kfree(struct ceph_msg *m) | |
2767 | { | |
2768 | dout("msg_kfree %p\n", m); | |
2769 | if (m->front_is_vmalloc) | |
2770 | vfree(m->front.iov_base); | |
2771 | else | |
2772 | kfree(m->front.iov_base); | |
2773 | kfree(m); | |
2774 | } | |
2775 | ||
2776 | /* | |
2777 | * Drop a msg ref. Destroy as needed. | |
2778 | */ | |
c2e552e7 SW |
2779 | void ceph_msg_last_put(struct kref *kref) |
2780 | { | |
2781 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2782 | |
c2e552e7 SW |
2783 | dout("ceph_msg_put last one on %p\n", m); |
2784 | WARN_ON(!list_empty(&m->list_head)); | |
2785 | ||
2786 | /* drop middle, data, if any */ | |
2787 | if (m->middle) { | |
2788 | ceph_buffer_put(m->middle); | |
2789 | m->middle = NULL; | |
31b8006e | 2790 | } |
c2e552e7 SW |
2791 | m->nr_pages = 0; |
2792 | m->pages = NULL; | |
2793 | ||
58bb3b37 SW |
2794 | if (m->pagelist) { |
2795 | ceph_pagelist_release(m->pagelist); | |
2796 | kfree(m->pagelist); | |
2797 | m->pagelist = NULL; | |
2798 | } | |
2799 | ||
68b4476b YS |
2800 | m->trail = NULL; |
2801 | ||
c2e552e7 SW |
2802 | if (m->pool) |
2803 | ceph_msgpool_put(m->pool, m); | |
2804 | else | |
2805 | ceph_msg_kfree(m); | |
31b8006e | 2806 | } |
3d14c5d2 | 2807 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2808 | |
2809 | void ceph_msg_dump(struct ceph_msg *msg) | |
2810 | { | |
2811 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2812 | msg->front_max, msg->nr_pages); | |
2813 | print_hex_dump(KERN_DEBUG, "header: ", | |
2814 | DUMP_PREFIX_OFFSET, 16, 1, | |
2815 | &msg->hdr, sizeof(msg->hdr), true); | |
2816 | print_hex_dump(KERN_DEBUG, " front: ", | |
2817 | DUMP_PREFIX_OFFSET, 16, 1, | |
2818 | msg->front.iov_base, msg->front.iov_len, true); | |
2819 | if (msg->middle) | |
2820 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2821 | DUMP_PREFIX_OFFSET, 16, 1, | |
2822 | msg->middle->vec.iov_base, | |
2823 | msg->middle->vec.iov_len, true); | |
2824 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2825 | DUMP_PREFIX_OFFSET, 16, 1, | |
2826 | &msg->footer, sizeof(msg->footer), true); | |
2827 | } | |
3d14c5d2 | 2828 | EXPORT_SYMBOL(ceph_msg_dump); |