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); | |
8007b8d6 SW |
227 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
228 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 AE |
229 | } |
230 | ||
231 | static void con_sock_state_connecting(struct ceph_connection *con) | |
232 | { | |
233 | int old_state; | |
234 | ||
235 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
236 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
237 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
238 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
239 | CON_SOCK_STATE_CONNECTING); | |
ce2c8903 AE |
240 | } |
241 | ||
242 | static void con_sock_state_connected(struct ceph_connection *con) | |
243 | { | |
244 | int old_state; | |
245 | ||
246 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
247 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
248 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
249 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
250 | CON_SOCK_STATE_CONNECTED); | |
ce2c8903 AE |
251 | } |
252 | ||
253 | static void con_sock_state_closing(struct ceph_connection *con) | |
254 | { | |
255 | int old_state; | |
256 | ||
257 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
258 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
259 | old_state != CON_SOCK_STATE_CONNECTED && | |
260 | old_state != CON_SOCK_STATE_CLOSING)) | |
261 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
262 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
263 | CON_SOCK_STATE_CLOSING); | |
ce2c8903 AE |
264 | } |
265 | ||
266 | static void con_sock_state_closed(struct ceph_connection *con) | |
267 | { | |
268 | int old_state; | |
269 | ||
270 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
271 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 | 272 | old_state != CON_SOCK_STATE_CLOSING && |
8007b8d6 SW |
273 | old_state != CON_SOCK_STATE_CONNECTING && |
274 | old_state != CON_SOCK_STATE_CLOSED)) | |
ce2c8903 | 275 | printk("%s: unexpected old state %d\n", __func__, old_state); |
8007b8d6 SW |
276 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
277 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 | 278 | } |
a922d38f | 279 | |
31b8006e SW |
280 | /* |
281 | * socket callback functions | |
282 | */ | |
283 | ||
284 | /* data available on socket, or listen socket received a connect */ | |
327800bd | 285 | static void ceph_sock_data_ready(struct sock *sk, int count_unused) |
31b8006e | 286 | { |
bd406145 | 287 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
288 | if (atomic_read(&con->msgr->stopping)) { |
289 | return; | |
290 | } | |
bd406145 | 291 | |
31b8006e | 292 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 293 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
294 | con, con->state); |
295 | queue_con(con); | |
296 | } | |
297 | } | |
298 | ||
299 | /* socket has buffer space for writing */ | |
327800bd | 300 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 301 | { |
d3002b97 | 302 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 303 | |
182fac26 JS |
304 | /* only queue to workqueue if there is data we want to write, |
305 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 306 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
307 | * doesn't get called again until try_write() fills the socket |
308 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
309 | * and net/core/stream.c:sk_stream_write_space(). | |
310 | */ | |
4a861692 | 311 | if (test_bit(CON_FLAG_WRITE_PENDING, &con->flags)) { |
182fac26 | 312 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
327800bd | 313 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
314 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
315 | queue_con(con); | |
316 | } | |
31b8006e | 317 | } else { |
327800bd | 318 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 319 | } |
31b8006e SW |
320 | } |
321 | ||
322 | /* socket's state has changed */ | |
327800bd | 323 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 324 | { |
bd406145 | 325 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 326 | |
327800bd | 327 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
328 | con, con->state, sk->sk_state); |
329 | ||
31b8006e SW |
330 | switch (sk->sk_state) { |
331 | case TCP_CLOSE: | |
327800bd | 332 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 333 | case TCP_CLOSE_WAIT: |
327800bd | 334 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 335 | con_sock_state_closing(con); |
4a861692 | 336 | set_bit(CON_FLAG_SOCK_CLOSED, &con->flags); |
d65c9e0b | 337 | queue_con(con); |
31b8006e SW |
338 | break; |
339 | case TCP_ESTABLISHED: | |
327800bd | 340 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 341 | con_sock_state_connected(con); |
31b8006e SW |
342 | queue_con(con); |
343 | break; | |
d3002b97 AE |
344 | default: /* Everything else is uninteresting */ |
345 | break; | |
31b8006e SW |
346 | } |
347 | } | |
348 | ||
349 | /* | |
350 | * set up socket callbacks | |
351 | */ | |
352 | static void set_sock_callbacks(struct socket *sock, | |
353 | struct ceph_connection *con) | |
354 | { | |
355 | struct sock *sk = sock->sk; | |
bd406145 | 356 | sk->sk_user_data = con; |
327800bd AE |
357 | sk->sk_data_ready = ceph_sock_data_ready; |
358 | sk->sk_write_space = ceph_sock_write_space; | |
359 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
360 | } |
361 | ||
362 | ||
363 | /* | |
364 | * socket helpers | |
365 | */ | |
366 | ||
367 | /* | |
368 | * initiate connection to a remote socket. | |
369 | */ | |
41617d0c | 370 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 371 | { |
f91d3471 | 372 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
373 | struct socket *sock; |
374 | int ret; | |
375 | ||
376 | BUG_ON(con->sock); | |
f91d3471 SW |
377 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
378 | IPPROTO_TCP, &sock); | |
31b8006e | 379 | if (ret) |
41617d0c | 380 | return ret; |
31b8006e SW |
381 | sock->sk->sk_allocation = GFP_NOFS; |
382 | ||
a6a5349d SW |
383 | #ifdef CONFIG_LOCKDEP |
384 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
385 | #endif | |
386 | ||
31b8006e SW |
387 | set_sock_callbacks(sock, con); |
388 | ||
3d14c5d2 | 389 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 390 | |
89a86be0 | 391 | con_sock_state_connecting(con); |
f91d3471 SW |
392 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
393 | O_NONBLOCK); | |
31b8006e SW |
394 | if (ret == -EINPROGRESS) { |
395 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 396 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 397 | sock->sk->sk_state); |
a5bc3129 | 398 | } else if (ret < 0) { |
31b8006e | 399 | pr_err("connect %s error %d\n", |
3d14c5d2 | 400 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 401 | sock_release(sock); |
31b8006e | 402 | con->error_msg = "connect error"; |
31b8006e | 403 | |
41617d0c | 404 | return ret; |
a5bc3129 AE |
405 | } |
406 | con->sock = sock; | |
41617d0c | 407 | return 0; |
31b8006e SW |
408 | } |
409 | ||
410 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
411 | { | |
412 | struct kvec iov = {buf, len}; | |
413 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 414 | int r; |
31b8006e | 415 | |
98bdb0aa SW |
416 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
417 | if (r == -EAGAIN) | |
418 | r = 0; | |
419 | return r; | |
31b8006e SW |
420 | } |
421 | ||
422 | /* | |
423 | * write something. @more is true if caller will be sending more data | |
424 | * shortly. | |
425 | */ | |
426 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
427 | size_t kvlen, size_t len, int more) | |
428 | { | |
429 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 430 | int r; |
31b8006e SW |
431 | |
432 | if (more) | |
433 | msg.msg_flags |= MSG_MORE; | |
434 | else | |
435 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
436 | ||
42961d23 SW |
437 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
438 | if (r == -EAGAIN) | |
439 | r = 0; | |
440 | return r; | |
31b8006e SW |
441 | } |
442 | ||
31739139 AE |
443 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
444 | int offset, size_t size, int more) | |
445 | { | |
446 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
447 | int ret; | |
448 | ||
449 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
450 | if (ret == -EAGAIN) | |
451 | ret = 0; | |
452 | ||
453 | return ret; | |
454 | } | |
455 | ||
31b8006e SW |
456 | |
457 | /* | |
458 | * Shutdown/close the socket for the given connection. | |
459 | */ | |
460 | static int con_close_socket(struct ceph_connection *con) | |
461 | { | |
8007b8d6 | 462 | int rc = 0; |
31b8006e SW |
463 | |
464 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
8007b8d6 SW |
465 | if (con->sock) { |
466 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
467 | sock_release(con->sock); | |
468 | con->sock = NULL; | |
469 | } | |
456ea468 AE |
470 | |
471 | /* | |
4a861692 | 472 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
473 | * independent of the connection mutex, and we could have |
474 | * received a socket close event before we had the chance to | |
475 | * shut the socket down. | |
476 | */ | |
4a861692 | 477 | clear_bit(CON_FLAG_SOCK_CLOSED, &con->flags); |
8007b8d6 | 478 | |
ce2c8903 | 479 | con_sock_state_closed(con); |
31b8006e SW |
480 | return rc; |
481 | } | |
482 | ||
483 | /* | |
484 | * Reset a connection. Discard all incoming and outgoing messages | |
485 | * and clear *_seq state. | |
486 | */ | |
487 | static void ceph_msg_remove(struct ceph_msg *msg) | |
488 | { | |
489 | list_del_init(&msg->list_head); | |
38941f80 | 490 | BUG_ON(msg->con == NULL); |
36eb71aa | 491 | msg->con->ops->put(msg->con); |
38941f80 AE |
492 | msg->con = NULL; |
493 | ||
31b8006e SW |
494 | ceph_msg_put(msg); |
495 | } | |
496 | static void ceph_msg_remove_list(struct list_head *head) | |
497 | { | |
498 | while (!list_empty(head)) { | |
499 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
500 | list_head); | |
501 | ceph_msg_remove(msg); | |
502 | } | |
503 | } | |
504 | ||
505 | static void reset_connection(struct ceph_connection *con) | |
506 | { | |
507 | /* reset connection, out_queue, msg_ and connect_seq */ | |
508 | /* discard existing out_queue and msg_seq */ | |
31b8006e SW |
509 | ceph_msg_remove_list(&con->out_queue); |
510 | ceph_msg_remove_list(&con->out_sent); | |
511 | ||
cf3e5c40 | 512 | if (con->in_msg) { |
38941f80 AE |
513 | BUG_ON(con->in_msg->con != con); |
514 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
515 | ceph_msg_put(con->in_msg); |
516 | con->in_msg = NULL; | |
36eb71aa | 517 | con->ops->put(con); |
cf3e5c40 SW |
518 | } |
519 | ||
31b8006e SW |
520 | con->connect_seq = 0; |
521 | con->out_seq = 0; | |
c86a2930 SW |
522 | if (con->out_msg) { |
523 | ceph_msg_put(con->out_msg); | |
524 | con->out_msg = NULL; | |
525 | } | |
31b8006e | 526 | con->in_seq = 0; |
0e0d5e0c | 527 | con->in_seq_acked = 0; |
31b8006e SW |
528 | } |
529 | ||
530 | /* | |
531 | * mark a peer down. drop any open connections. | |
532 | */ | |
533 | void ceph_con_close(struct ceph_connection *con) | |
534 | { | |
8c50c817 | 535 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
536 | dout("con_close %p peer %s\n", con, |
537 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
8dacc7da | 538 | con->state = CON_STATE_CLOSED; |
a5988c49 | 539 | |
4a861692 SW |
540 | clear_bit(CON_FLAG_LOSSYTX, &con->flags); /* so we retry next connect */ |
541 | clear_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags); | |
542 | clear_bit(CON_FLAG_WRITE_PENDING, &con->flags); | |
43c7427d SW |
543 | clear_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags); |
544 | clear_bit(CON_FLAG_BACKOFF, &con->flags); | |
a5988c49 | 545 | |
31b8006e | 546 | reset_connection(con); |
6f2bc3ff | 547 | con->peer_global_seq = 0; |
91e45ce3 | 548 | cancel_delayed_work(&con->work); |
ee76e073 | 549 | con_close_socket(con); |
ec302645 | 550 | mutex_unlock(&con->mutex); |
31b8006e | 551 | } |
3d14c5d2 | 552 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 553 | |
31b8006e SW |
554 | /* |
555 | * Reopen a closed connection, with a new peer address. | |
556 | */ | |
b7a9e5dd SW |
557 | void ceph_con_open(struct ceph_connection *con, |
558 | __u8 entity_type, __u64 entity_num, | |
559 | struct ceph_entity_addr *addr) | |
31b8006e | 560 | { |
5469155f | 561 | mutex_lock(&con->mutex); |
3d14c5d2 | 562 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
8dacc7da SW |
563 | |
564 | BUG_ON(con->state != CON_STATE_CLOSED); | |
565 | con->state = CON_STATE_PREOPEN; | |
a5988c49 | 566 | |
b7a9e5dd SW |
567 | con->peer_name.type = (__u8) entity_type; |
568 | con->peer_name.num = cpu_to_le64(entity_num); | |
569 | ||
31b8006e | 570 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 571 | con->delay = 0; /* reset backoff memory */ |
5469155f | 572 | mutex_unlock(&con->mutex); |
31b8006e SW |
573 | queue_con(con); |
574 | } | |
3d14c5d2 | 575 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 576 | |
87b315a5 SW |
577 | /* |
578 | * return true if this connection ever successfully opened | |
579 | */ | |
580 | bool ceph_con_opened(struct ceph_connection *con) | |
581 | { | |
582 | return con->connect_seq > 0; | |
583 | } | |
584 | ||
31b8006e SW |
585 | /* |
586 | * initialize a new connection. | |
587 | */ | |
1bfd89f4 AE |
588 | void ceph_con_init(struct ceph_connection *con, void *private, |
589 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 590 | struct ceph_messenger *msgr) |
31b8006e SW |
591 | { |
592 | dout("con_init %p\n", con); | |
593 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
594 | con->private = private; |
595 | con->ops = ops; | |
31b8006e | 596 | con->msgr = msgr; |
ce2c8903 AE |
597 | |
598 | con_sock_state_init(con); | |
599 | ||
ec302645 | 600 | mutex_init(&con->mutex); |
31b8006e SW |
601 | INIT_LIST_HEAD(&con->out_queue); |
602 | INIT_LIST_HEAD(&con->out_sent); | |
603 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 | 604 | |
8dacc7da | 605 | con->state = CON_STATE_CLOSED; |
31b8006e | 606 | } |
3d14c5d2 | 607 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
608 | |
609 | ||
610 | /* | |
611 | * We maintain a global counter to order connection attempts. Get | |
612 | * a unique seq greater than @gt. | |
613 | */ | |
614 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
615 | { | |
616 | u32 ret; | |
617 | ||
618 | spin_lock(&msgr->global_seq_lock); | |
619 | if (msgr->global_seq < gt) | |
620 | msgr->global_seq = gt; | |
621 | ret = ++msgr->global_seq; | |
622 | spin_unlock(&msgr->global_seq_lock); | |
623 | return ret; | |
624 | } | |
625 | ||
e2200423 | 626 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
627 | { |
628 | con->out_kvec_left = 0; | |
629 | con->out_kvec_bytes = 0; | |
630 | con->out_kvec_cur = &con->out_kvec[0]; | |
631 | } | |
632 | ||
e2200423 | 633 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
634 | size_t size, void *data) |
635 | { | |
636 | int index; | |
637 | ||
638 | index = con->out_kvec_left; | |
639 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
640 | ||
641 | con->out_kvec[index].iov_len = size; | |
642 | con->out_kvec[index].iov_base = data; | |
643 | con->out_kvec_left++; | |
644 | con->out_kvec_bytes += size; | |
645 | } | |
31b8006e | 646 | |
df6ad1f9 AE |
647 | #ifdef CONFIG_BLOCK |
648 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
649 | { | |
650 | if (!bio) { | |
651 | *iter = NULL; | |
652 | *seg = 0; | |
653 | return; | |
654 | } | |
655 | *iter = bio; | |
656 | *seg = bio->bi_idx; | |
657 | } | |
658 | ||
659 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
660 | { | |
661 | if (*bio_iter == NULL) | |
662 | return; | |
663 | ||
664 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
665 | ||
666 | (*seg)++; | |
667 | if (*seg == (*bio_iter)->bi_vcnt) | |
668 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
669 | } | |
670 | #endif | |
671 | ||
739c905b AE |
672 | static void prepare_write_message_data(struct ceph_connection *con) |
673 | { | |
674 | struct ceph_msg *msg = con->out_msg; | |
675 | ||
676 | BUG_ON(!msg); | |
677 | BUG_ON(!msg->hdr.data_len); | |
678 | ||
679 | /* initialize page iterator */ | |
680 | con->out_msg_pos.page = 0; | |
681 | if (msg->pages) | |
682 | con->out_msg_pos.page_pos = msg->page_alignment; | |
683 | else | |
684 | con->out_msg_pos.page_pos = 0; | |
572c588e | 685 | #ifdef CONFIG_BLOCK |
abdaa6a8 | 686 | if (msg->bio) |
572c588e AE |
687 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); |
688 | #endif | |
739c905b AE |
689 | con->out_msg_pos.data_pos = 0; |
690 | con->out_msg_pos.did_page_crc = false; | |
691 | con->out_more = 1; /* data + footer will follow */ | |
692 | } | |
693 | ||
31b8006e SW |
694 | /* |
695 | * Prepare footer for currently outgoing message, and finish things | |
696 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
697 | */ | |
859eb799 | 698 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
699 | { |
700 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 701 | int v = con->out_kvec_left; |
31b8006e | 702 | |
fd154f3c AE |
703 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
704 | ||
31b8006e SW |
705 | dout("prepare_write_message_footer %p\n", con); |
706 | con->out_kvec_is_msg = true; | |
707 | con->out_kvec[v].iov_base = &m->footer; | |
708 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
709 | con->out_kvec_bytes += sizeof(m->footer); | |
710 | con->out_kvec_left++; | |
711 | con->out_more = m->more_to_follow; | |
c86a2930 | 712 | con->out_msg_done = true; |
31b8006e SW |
713 | } |
714 | ||
715 | /* | |
716 | * Prepare headers for the next outgoing message. | |
717 | */ | |
718 | static void prepare_write_message(struct ceph_connection *con) | |
719 | { | |
720 | struct ceph_msg *m; | |
a9a0c51a | 721 | u32 crc; |
31b8006e | 722 | |
e2200423 | 723 | con_out_kvec_reset(con); |
31b8006e | 724 | con->out_kvec_is_msg = true; |
c86a2930 | 725 | con->out_msg_done = false; |
31b8006e SW |
726 | |
727 | /* Sneak an ack in there first? If we can get it into the same | |
728 | * TCP packet that's a good thing. */ | |
729 | if (con->in_seq > con->in_seq_acked) { | |
730 | con->in_seq_acked = con->in_seq; | |
e2200423 | 731 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 732 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 733 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 734 | &con->out_temp_ack); |
31b8006e SW |
735 | } |
736 | ||
38941f80 | 737 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 738 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 739 | con->out_msg = m; |
38941f80 | 740 | BUG_ON(m->con != con); |
4cf9d544 SW |
741 | |
742 | /* put message on sent list */ | |
743 | ceph_msg_get(m); | |
744 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 745 | |
e84346b7 SW |
746 | /* |
747 | * only assign outgoing seq # if we haven't sent this message | |
748 | * yet. if it is requeued, resend with it's original seq. | |
749 | */ | |
750 | if (m->needs_out_seq) { | |
751 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
752 | m->needs_out_seq = false; | |
753 | } | |
b132cf4c YZ |
754 | #ifdef CONFIG_BLOCK |
755 | else | |
756 | m->bio_iter = NULL; | |
757 | #endif | |
31b8006e SW |
758 | |
759 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
760 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
761 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
762 | le32_to_cpu(m->hdr.data_len), | |
763 | m->nr_pages); | |
764 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); | |
765 | ||
766 | /* tag + hdr + front + middle */ | |
e2200423 AE |
767 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
768 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
769 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 770 | |
31b8006e | 771 | if (m->middle) |
e2200423 | 772 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 773 | m->middle->vec.iov_base); |
31b8006e SW |
774 | |
775 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
776 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
777 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 778 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
779 | |
780 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
781 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
782 | if (m->middle) { | |
783 | crc = crc32c(0, m->middle->vec.iov_base, | |
784 | m->middle->vec.iov_len); | |
785 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
786 | } else | |
31b8006e | 787 | con->out_msg->footer.middle_crc = 0; |
739c905b | 788 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
789 | le32_to_cpu(con->out_msg->footer.front_crc), |
790 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
791 | ||
792 | /* is there a data payload? */ | |
739c905b AE |
793 | con->out_msg->footer.data_crc = 0; |
794 | if (m->hdr.data_len) | |
795 | prepare_write_message_data(con); | |
796 | else | |
31b8006e | 797 | /* no, queue up footer too and be done */ |
859eb799 | 798 | prepare_write_message_footer(con); |
31b8006e | 799 | |
4a861692 | 800 | set_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
31b8006e SW |
801 | } |
802 | ||
803 | /* | |
804 | * Prepare an ack. | |
805 | */ | |
806 | static void prepare_write_ack(struct ceph_connection *con) | |
807 | { | |
808 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
809 | con->in_seq_acked, con->in_seq); | |
810 | con->in_seq_acked = con->in_seq; | |
811 | ||
e2200423 | 812 | con_out_kvec_reset(con); |
859eb799 | 813 | |
e2200423 | 814 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 815 | |
31b8006e | 816 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 817 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
818 | &con->out_temp_ack); |
819 | ||
31b8006e | 820 | con->out_more = 1; /* more will follow.. eventually.. */ |
4a861692 | 821 | set_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
31b8006e SW |
822 | } |
823 | ||
824 | /* | |
825 | * Prepare to write keepalive byte. | |
826 | */ | |
827 | static void prepare_write_keepalive(struct ceph_connection *con) | |
828 | { | |
829 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
830 | con_out_kvec_reset(con); |
831 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
4a861692 | 832 | set_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
31b8006e SW |
833 | } |
834 | ||
835 | /* | |
836 | * Connection negotiation. | |
837 | */ | |
838 | ||
dac1e716 AE |
839 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
840 | int *auth_proto) | |
4e7a5dcd | 841 | { |
a3530df3 | 842 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
843 | |
844 | if (!con->ops->get_authorizer) { | |
845 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
846 | con->out_connect.authorizer_len = 0; | |
729796be | 847 | return NULL; |
b1c6b980 AE |
848 | } |
849 | ||
850 | /* Can't hold the mutex while getting authorizer */ | |
ec302645 | 851 | mutex_unlock(&con->mutex); |
dac1e716 | 852 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
ec302645 | 853 | mutex_lock(&con->mutex); |
4e7a5dcd | 854 | |
a3530df3 | 855 | if (IS_ERR(auth)) |
729796be | 856 | return auth; |
8dacc7da | 857 | if (con->state != CON_STATE_NEGOTIATING) |
729796be | 858 | return ERR_PTR(-EAGAIN); |
0da5d703 | 859 | |
8f43fb53 AE |
860 | con->auth_reply_buf = auth->authorizer_reply_buf; |
861 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
729796be | 862 | return auth; |
4e7a5dcd SW |
863 | } |
864 | ||
31b8006e SW |
865 | /* |
866 | * We connected to a peer and are saying hello. | |
867 | */ | |
e825a66d | 868 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 869 | { |
e2200423 AE |
870 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
871 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 872 | &con->msgr->my_enc_addr); |
eed0ef2c | 873 | |
eed0ef2c | 874 | con->out_more = 0; |
4a861692 | 875 | set_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
eed0ef2c SW |
876 | } |
877 | ||
e825a66d | 878 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 879 | { |
95c96174 | 880 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 881 | int proto; |
dac1e716 | 882 | int auth_proto; |
729796be | 883 | struct ceph_auth_handshake *auth; |
31b8006e SW |
884 | |
885 | switch (con->peer_name.type) { | |
886 | case CEPH_ENTITY_TYPE_MON: | |
887 | proto = CEPH_MONC_PROTOCOL; | |
888 | break; | |
889 | case CEPH_ENTITY_TYPE_OSD: | |
890 | proto = CEPH_OSDC_PROTOCOL; | |
891 | break; | |
892 | case CEPH_ENTITY_TYPE_MDS: | |
893 | proto = CEPH_MDSC_PROTOCOL; | |
894 | break; | |
895 | default: | |
896 | BUG(); | |
897 | } | |
898 | ||
899 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
900 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 901 | |
e825a66d | 902 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
903 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
904 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
905 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
906 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
907 | con->out_connect.flags = 0; | |
31b8006e | 908 | |
dac1e716 AE |
909 | auth_proto = CEPH_AUTH_UNKNOWN; |
910 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
911 | if (IS_ERR(auth)) |
912 | return PTR_ERR(auth); | |
3da54776 | 913 | |
dac1e716 | 914 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
915 | con->out_connect.authorizer_len = auth ? |
916 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
917 | ||
e2200423 | 918 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
919 | &con->out_connect); |
920 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 921 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 922 | auth->authorizer_buf); |
859eb799 | 923 | |
31b8006e | 924 | con->out_more = 0; |
4a861692 | 925 | set_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
4e7a5dcd | 926 | |
e10c758e | 927 | return 0; |
31b8006e SW |
928 | } |
929 | ||
31b8006e SW |
930 | /* |
931 | * write as much of pending kvecs to the socket as we can. | |
932 | * 1 -> done | |
933 | * 0 -> socket full, but more to do | |
934 | * <0 -> error | |
935 | */ | |
936 | static int write_partial_kvec(struct ceph_connection *con) | |
937 | { | |
938 | int ret; | |
939 | ||
940 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
941 | while (con->out_kvec_bytes > 0) { | |
942 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
943 | con->out_kvec_left, con->out_kvec_bytes, | |
944 | con->out_more); | |
945 | if (ret <= 0) | |
946 | goto out; | |
947 | con->out_kvec_bytes -= ret; | |
948 | if (con->out_kvec_bytes == 0) | |
949 | break; /* done */ | |
f42299e6 AE |
950 | |
951 | /* account for full iov entries consumed */ | |
952 | while (ret >= con->out_kvec_cur->iov_len) { | |
953 | BUG_ON(!con->out_kvec_left); | |
954 | ret -= con->out_kvec_cur->iov_len; | |
955 | con->out_kvec_cur++; | |
956 | con->out_kvec_left--; | |
957 | } | |
958 | /* and for a partially-consumed entry */ | |
959 | if (ret) { | |
960 | con->out_kvec_cur->iov_len -= ret; | |
961 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
962 | } |
963 | } | |
964 | con->out_kvec_left = 0; | |
965 | con->out_kvec_is_msg = false; | |
966 | ret = 1; | |
967 | out: | |
968 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
969 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
970 | return ret; /* done! */ | |
971 | } | |
972 | ||
84ca8fc8 AE |
973 | static void out_msg_pos_next(struct ceph_connection *con, struct page *page, |
974 | size_t len, size_t sent, bool in_trail) | |
68b4476b | 975 | { |
84ca8fc8 | 976 | struct ceph_msg *msg = con->out_msg; |
68b4476b | 977 | |
84ca8fc8 AE |
978 | BUG_ON(!msg); |
979 | BUG_ON(!sent); | |
68b4476b | 980 | |
84ca8fc8 AE |
981 | con->out_msg_pos.data_pos += sent; |
982 | con->out_msg_pos.page_pos += sent; | |
5821bd8c AE |
983 | if (sent < len) |
984 | return; | |
68b4476b | 985 | |
5821bd8c AE |
986 | BUG_ON(sent != len); |
987 | con->out_msg_pos.page_pos = 0; | |
988 | con->out_msg_pos.page++; | |
989 | con->out_msg_pos.did_page_crc = false; | |
990 | if (in_trail) | |
991 | list_move_tail(&page->lru, | |
992 | &msg->trail->head); | |
993 | else if (msg->pagelist) | |
994 | list_move_tail(&page->lru, | |
995 | &msg->pagelist->head); | |
84ca8fc8 | 996 | #ifdef CONFIG_BLOCK |
5821bd8c AE |
997 | else if (msg->bio) |
998 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
68b4476b | 999 | #endif |
84ca8fc8 | 1000 | } |
68b4476b | 1001 | |
31b8006e SW |
1002 | /* |
1003 | * Write as much message data payload as we can. If we finish, queue | |
1004 | * up the footer. | |
1005 | * 1 -> done, footer is now queued in out_kvec[]. | |
1006 | * 0 -> socket full, but more to do | |
1007 | * <0 -> error | |
1008 | */ | |
1009 | static int write_partial_msg_pages(struct ceph_connection *con) | |
1010 | { | |
1011 | struct ceph_msg *msg = con->out_msg; | |
95c96174 | 1012 | unsigned int data_len = le32_to_cpu(msg->hdr.data_len); |
31b8006e | 1013 | size_t len; |
37675b0f | 1014 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 1015 | int ret; |
68b4476b | 1016 | int total_max_write; |
84ca8fc8 | 1017 | bool in_trail = false; |
5821bd8c AE |
1018 | const size_t trail_len = (msg->trail ? msg->trail->length : 0); |
1019 | const size_t trail_off = data_len - trail_len; | |
31b8006e SW |
1020 | |
1021 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
84ca8fc8 | 1022 | con, msg, con->out_msg_pos.page, msg->nr_pages, |
31b8006e SW |
1023 | con->out_msg_pos.page_pos); |
1024 | ||
5821bd8c AE |
1025 | /* |
1026 | * Iterate through each page that contains data to be | |
1027 | * written, and send as much as possible for each. | |
1028 | * | |
1029 | * If we are calculating the data crc (the default), we will | |
1030 | * need to map the page. If we have no pages, they have | |
1031 | * been revoked, so use the zero page. | |
1032 | */ | |
68b4476b | 1033 | while (data_len > con->out_msg_pos.data_pos) { |
31b8006e | 1034 | struct page *page = NULL; |
68b4476b | 1035 | int max_write = PAGE_SIZE; |
9bd19663 | 1036 | int bio_offset = 0; |
68b4476b | 1037 | |
5821bd8c AE |
1038 | in_trail = in_trail || con->out_msg_pos.data_pos >= trail_off; |
1039 | if (!in_trail) | |
1040 | total_max_write = trail_off - con->out_msg_pos.data_pos; | |
68b4476b | 1041 | |
5821bd8c | 1042 | if (in_trail) { |
68b4476b YS |
1043 | total_max_write = data_len - con->out_msg_pos.data_pos; |
1044 | ||
1045 | page = list_first_entry(&msg->trail->head, | |
1046 | struct page, lru); | |
68b4476b | 1047 | } else if (msg->pages) { |
31b8006e | 1048 | page = msg->pages[con->out_msg_pos.page]; |
58bb3b37 SW |
1049 | } else if (msg->pagelist) { |
1050 | page = list_first_entry(&msg->pagelist->head, | |
1051 | struct page, lru); | |
68b4476b YS |
1052 | #ifdef CONFIG_BLOCK |
1053 | } else if (msg->bio) { | |
1054 | struct bio_vec *bv; | |
1055 | ||
1056 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
1057 | page = bv->bv_page; | |
9bd19663 | 1058 | bio_offset = bv->bv_offset; |
68b4476b YS |
1059 | max_write = bv->bv_len; |
1060 | #endif | |
31b8006e | 1061 | } else { |
57666519 | 1062 | page = zero_page; |
31b8006e | 1063 | } |
68b4476b YS |
1064 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
1065 | total_max_write); | |
1066 | ||
37675b0f | 1067 | if (do_datacrc && !con->out_msg_pos.did_page_crc) { |
9bd19663 | 1068 | void *base; |
5821bd8c | 1069 | u32 crc = le32_to_cpu(msg->footer.data_crc); |
8d63e318 | 1070 | char *kaddr; |
31b8006e | 1071 | |
8d63e318 | 1072 | kaddr = kmap(page); |
31b8006e | 1073 | BUG_ON(kaddr == NULL); |
9bd19663 | 1074 | base = kaddr + con->out_msg_pos.page_pos + bio_offset; |
5821bd8c | 1075 | crc = crc32c(crc, base, len); |
84ca8fc8 | 1076 | msg->footer.data_crc = cpu_to_le32(crc); |
bca064d2 | 1077 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 1078 | } |
e36b13cc | 1079 | ret = ceph_tcp_sendpage(con->sock, page, |
9bd19663 | 1080 | con->out_msg_pos.page_pos + bio_offset, |
e36b13cc | 1081 | len, 1); |
31b8006e | 1082 | |
0cdf9e60 | 1083 | if (do_datacrc) |
31b8006e SW |
1084 | kunmap(page); |
1085 | ||
1086 | if (ret <= 0) | |
1087 | goto out; | |
1088 | ||
84ca8fc8 | 1089 | out_msg_pos_next(con, page, len, (size_t) ret, in_trail); |
31b8006e SW |
1090 | } |
1091 | ||
1092 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
1093 | ||
1094 | /* prepare and queue up footer, too */ | |
37675b0f | 1095 | if (!do_datacrc) |
84ca8fc8 | 1096 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1097 | con_out_kvec_reset(con); |
859eb799 | 1098 | prepare_write_message_footer(con); |
31b8006e SW |
1099 | ret = 1; |
1100 | out: | |
1101 | return ret; | |
1102 | } | |
1103 | ||
1104 | /* | |
1105 | * write some zeros | |
1106 | */ | |
1107 | static int write_partial_skip(struct ceph_connection *con) | |
1108 | { | |
1109 | int ret; | |
1110 | ||
1111 | while (con->out_skip > 0) { | |
31739139 | 1112 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1113 | |
31739139 | 1114 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1); |
31b8006e SW |
1115 | if (ret <= 0) |
1116 | goto out; | |
1117 | con->out_skip -= ret; | |
1118 | } | |
1119 | ret = 1; | |
1120 | out: | |
1121 | return ret; | |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * Prepare to read connection handshake, or an ack. | |
1126 | */ | |
eed0ef2c SW |
1127 | static void prepare_read_banner(struct ceph_connection *con) |
1128 | { | |
1129 | dout("prepare_read_banner %p\n", con); | |
1130 | con->in_base_pos = 0; | |
1131 | } | |
1132 | ||
31b8006e SW |
1133 | static void prepare_read_connect(struct ceph_connection *con) |
1134 | { | |
1135 | dout("prepare_read_connect %p\n", con); | |
1136 | con->in_base_pos = 0; | |
1137 | } | |
1138 | ||
1139 | static void prepare_read_ack(struct ceph_connection *con) | |
1140 | { | |
1141 | dout("prepare_read_ack %p\n", con); | |
1142 | con->in_base_pos = 0; | |
1143 | } | |
1144 | ||
1145 | static void prepare_read_tag(struct ceph_connection *con) | |
1146 | { | |
1147 | dout("prepare_read_tag %p\n", con); | |
1148 | con->in_base_pos = 0; | |
1149 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1150 | } | |
1151 | ||
1152 | /* | |
1153 | * Prepare to read a message. | |
1154 | */ | |
1155 | static int prepare_read_message(struct ceph_connection *con) | |
1156 | { | |
1157 | dout("prepare_read_message %p\n", con); | |
1158 | BUG_ON(con->in_msg != NULL); | |
1159 | con->in_base_pos = 0; | |
1160 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1161 | return 0; | |
1162 | } | |
1163 | ||
1164 | ||
1165 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1166 | int end, int size, void *object) |
31b8006e | 1167 | { |
e6cee71f AE |
1168 | while (con->in_base_pos < end) { |
1169 | int left = end - con->in_base_pos; | |
31b8006e SW |
1170 | int have = size - left; |
1171 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1172 | if (ret <= 0) | |
1173 | return ret; | |
1174 | con->in_base_pos += ret; | |
1175 | } | |
1176 | return 1; | |
1177 | } | |
1178 | ||
1179 | ||
1180 | /* | |
1181 | * Read all or part of the connect-side handshake on a new connection | |
1182 | */ | |
eed0ef2c | 1183 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1184 | { |
fd51653f AE |
1185 | int size; |
1186 | int end; | |
1187 | int ret; | |
31b8006e | 1188 | |
eed0ef2c | 1189 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1190 | |
1191 | /* peer's banner */ | |
fd51653f AE |
1192 | size = strlen(CEPH_BANNER); |
1193 | end = size; | |
1194 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1195 | if (ret <= 0) |
1196 | goto out; | |
fd51653f AE |
1197 | |
1198 | size = sizeof (con->actual_peer_addr); | |
1199 | end += size; | |
1200 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1201 | if (ret <= 0) |
1202 | goto out; | |
fd51653f AE |
1203 | |
1204 | size = sizeof (con->peer_addr_for_me); | |
1205 | end += size; | |
1206 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1207 | if (ret <= 0) |
1208 | goto out; | |
fd51653f | 1209 | |
eed0ef2c SW |
1210 | out: |
1211 | return ret; | |
1212 | } | |
1213 | ||
1214 | static int read_partial_connect(struct ceph_connection *con) | |
1215 | { | |
fd51653f AE |
1216 | int size; |
1217 | int end; | |
1218 | int ret; | |
eed0ef2c SW |
1219 | |
1220 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1221 | ||
fd51653f AE |
1222 | size = sizeof (con->in_reply); |
1223 | end = size; | |
1224 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1225 | if (ret <= 0) |
1226 | goto out; | |
fd51653f AE |
1227 | |
1228 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1229 | end += size; | |
1230 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1231 | if (ret <= 0) |
1232 | goto out; | |
31b8006e | 1233 | |
4e7a5dcd SW |
1234 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1235 | con, (int)con->in_reply.tag, | |
1236 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1237 | le32_to_cpu(con->in_reply.global_seq)); |
1238 | out: | |
1239 | return ret; | |
eed0ef2c | 1240 | |
31b8006e SW |
1241 | } |
1242 | ||
1243 | /* | |
1244 | * Verify the hello banner looks okay. | |
1245 | */ | |
1246 | static int verify_hello(struct ceph_connection *con) | |
1247 | { | |
1248 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1249 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1250 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1251 | con->error_msg = "protocol error, bad banner"; |
1252 | return -1; | |
1253 | } | |
1254 | return 0; | |
1255 | } | |
1256 | ||
1257 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1258 | { | |
1259 | switch (ss->ss_family) { | |
1260 | case AF_INET: | |
1261 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1262 | case AF_INET6: | |
1263 | return | |
1264 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1265 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1266 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1267 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1268 | } | |
1269 | return false; | |
1270 | } | |
1271 | ||
1272 | static int addr_port(struct sockaddr_storage *ss) | |
1273 | { | |
1274 | switch (ss->ss_family) { | |
1275 | case AF_INET: | |
f28bcfbe | 1276 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1277 | case AF_INET6: |
f28bcfbe | 1278 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1279 | } |
1280 | return 0; | |
1281 | } | |
1282 | ||
1283 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1284 | { | |
1285 | switch (ss->ss_family) { | |
1286 | case AF_INET: | |
1287 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1288 | break; |
31b8006e SW |
1289 | case AF_INET6: |
1290 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1291 | break; |
31b8006e SW |
1292 | } |
1293 | } | |
1294 | ||
ee3b56f2 NW |
1295 | /* |
1296 | * Unlike other *_pton function semantics, zero indicates success. | |
1297 | */ | |
1298 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1299 | char delim, const char **ipend) | |
1300 | { | |
99f0f3b2 AE |
1301 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1302 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1303 | |
1304 | memset(ss, 0, sizeof(*ss)); | |
1305 | ||
1306 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1307 | ss->ss_family = AF_INET; | |
1308 | return 0; | |
1309 | } | |
1310 | ||
1311 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1312 | ss->ss_family = AF_INET6; | |
1313 | return 0; | |
1314 | } | |
1315 | ||
1316 | return -EINVAL; | |
1317 | } | |
1318 | ||
1319 | /* | |
1320 | * Extract hostname string and resolve using kernel DNS facility. | |
1321 | */ | |
1322 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1323 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1324 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1325 | { | |
1326 | const char *end, *delim_p; | |
1327 | char *colon_p, *ip_addr = NULL; | |
1328 | int ip_len, ret; | |
1329 | ||
1330 | /* | |
1331 | * The end of the hostname occurs immediately preceding the delimiter or | |
1332 | * the port marker (':') where the delimiter takes precedence. | |
1333 | */ | |
1334 | delim_p = memchr(name, delim, namelen); | |
1335 | colon_p = memchr(name, ':', namelen); | |
1336 | ||
1337 | if (delim_p && colon_p) | |
1338 | end = delim_p < colon_p ? delim_p : colon_p; | |
1339 | else if (!delim_p && colon_p) | |
1340 | end = colon_p; | |
1341 | else { | |
1342 | end = delim_p; | |
1343 | if (!end) /* case: hostname:/ */ | |
1344 | end = name + namelen; | |
1345 | } | |
1346 | ||
1347 | if (end <= name) | |
1348 | return -EINVAL; | |
1349 | ||
1350 | /* do dns_resolve upcall */ | |
1351 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1352 | if (ip_len > 0) | |
1353 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1354 | else | |
1355 | ret = -ESRCH; | |
1356 | ||
1357 | kfree(ip_addr); | |
1358 | ||
1359 | *ipend = end; | |
1360 | ||
1361 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1362 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1363 | ||
1364 | return ret; | |
1365 | } | |
1366 | #else | |
1367 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1368 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1369 | { | |
1370 | return -EINVAL; | |
1371 | } | |
1372 | #endif | |
1373 | ||
1374 | /* | |
1375 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1376 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1377 | */ | |
1378 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1379 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1380 | { | |
1381 | int ret; | |
1382 | ||
1383 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1384 | if (ret) | |
1385 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1386 | ||
1387 | return ret; | |
1388 | } | |
1389 | ||
31b8006e SW |
1390 | /* |
1391 | * Parse an ip[:port] list into an addr array. Use the default | |
1392 | * monitor port if a port isn't specified. | |
1393 | */ | |
1394 | int ceph_parse_ips(const char *c, const char *end, | |
1395 | struct ceph_entity_addr *addr, | |
1396 | int max_count, int *count) | |
1397 | { | |
ee3b56f2 | 1398 | int i, ret = -EINVAL; |
31b8006e SW |
1399 | const char *p = c; |
1400 | ||
1401 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1402 | for (i = 0; i < max_count; i++) { | |
1403 | const char *ipend; | |
1404 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1405 | int port; |
39139f64 SW |
1406 | char delim = ','; |
1407 | ||
1408 | if (*p == '[') { | |
1409 | delim = ']'; | |
1410 | p++; | |
1411 | } | |
31b8006e | 1412 | |
ee3b56f2 NW |
1413 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1414 | if (ret) | |
31b8006e | 1415 | goto bad; |
ee3b56f2 NW |
1416 | ret = -EINVAL; |
1417 | ||
31b8006e SW |
1418 | p = ipend; |
1419 | ||
39139f64 SW |
1420 | if (delim == ']') { |
1421 | if (*p != ']') { | |
1422 | dout("missing matching ']'\n"); | |
1423 | goto bad; | |
1424 | } | |
1425 | p++; | |
1426 | } | |
1427 | ||
31b8006e SW |
1428 | /* port? */ |
1429 | if (p < end && *p == ':') { | |
1430 | port = 0; | |
1431 | p++; | |
1432 | while (p < end && *p >= '0' && *p <= '9') { | |
1433 | port = (port * 10) + (*p - '0'); | |
1434 | p++; | |
1435 | } | |
1436 | if (port > 65535 || port == 0) | |
1437 | goto bad; | |
1438 | } else { | |
1439 | port = CEPH_MON_PORT; | |
1440 | } | |
1441 | ||
1442 | addr_set_port(ss, port); | |
1443 | ||
3d14c5d2 | 1444 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1445 | |
1446 | if (p == end) | |
1447 | break; | |
1448 | if (*p != ',') | |
1449 | goto bad; | |
1450 | p++; | |
1451 | } | |
1452 | ||
1453 | if (p != end) | |
1454 | goto bad; | |
1455 | ||
1456 | if (count) | |
1457 | *count = i + 1; | |
1458 | return 0; | |
1459 | ||
1460 | bad: | |
39139f64 | 1461 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1462 | return ret; |
31b8006e | 1463 | } |
3d14c5d2 | 1464 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1465 | |
eed0ef2c | 1466 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1467 | { |
eed0ef2c | 1468 | dout("process_banner on %p\n", con); |
31b8006e SW |
1469 | |
1470 | if (verify_hello(con) < 0) | |
1471 | return -1; | |
1472 | ||
63f2d211 SW |
1473 | ceph_decode_addr(&con->actual_peer_addr); |
1474 | ceph_decode_addr(&con->peer_addr_for_me); | |
1475 | ||
31b8006e SW |
1476 | /* |
1477 | * Make sure the other end is who we wanted. note that the other | |
1478 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1479 | * them the benefit of the doubt. | |
1480 | */ | |
103e2d3a SW |
1481 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1482 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1483 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1484 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1485 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1486 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1487 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1488 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1489 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1490 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1491 | return -1; |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * did we learn our address? | |
1496 | */ | |
1497 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1498 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1499 | ||
1500 | memcpy(&con->msgr->inst.addr.in_addr, | |
1501 | &con->peer_addr_for_me.in_addr, | |
1502 | sizeof(con->peer_addr_for_me.in_addr)); | |
1503 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1504 | encode_my_addr(con->msgr); |
eed0ef2c | 1505 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1506 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1507 | } |
1508 | ||
eed0ef2c SW |
1509 | return 0; |
1510 | } | |
1511 | ||
04a419f9 SW |
1512 | static void fail_protocol(struct ceph_connection *con) |
1513 | { | |
1514 | reset_connection(con); | |
8dacc7da SW |
1515 | BUG_ON(con->state != CON_STATE_NEGOTIATING); |
1516 | con->state = CON_STATE_CLOSED; | |
04a419f9 SW |
1517 | } |
1518 | ||
eed0ef2c SW |
1519 | static int process_connect(struct ceph_connection *con) |
1520 | { | |
3d14c5d2 YS |
1521 | u64 sup_feat = con->msgr->supported_features; |
1522 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1523 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1524 | int ret; |
04a419f9 | 1525 | |
eed0ef2c SW |
1526 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1527 | ||
31b8006e | 1528 | switch (con->in_reply.tag) { |
04a419f9 SW |
1529 | case CEPH_MSGR_TAG_FEATURES: |
1530 | pr_err("%s%lld %s feature set mismatch," | |
1531 | " my %llx < server's %llx, missing %llx\n", | |
1532 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1533 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1534 | sup_feat, server_feat, server_feat & ~sup_feat); |
1535 | con->error_msg = "missing required protocol features"; | |
1536 | fail_protocol(con); | |
1537 | return -1; | |
1538 | ||
31b8006e | 1539 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1540 | pr_err("%s%lld %s protocol version mismatch," |
1541 | " my %d != server's %d\n", | |
1542 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1543 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1544 | le32_to_cpu(con->out_connect.protocol_version), |
1545 | le32_to_cpu(con->in_reply.protocol_version)); | |
1546 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1547 | fail_protocol(con); |
31b8006e SW |
1548 | return -1; |
1549 | ||
4e7a5dcd SW |
1550 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1551 | con->auth_retry++; | |
1552 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1553 | con->auth_retry); | |
1554 | if (con->auth_retry == 2) { | |
1555 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1556 | return -1; |
1557 | } | |
1558 | con->auth_retry = 1; | |
6d4221b5 | 1559 | con_out_kvec_reset(con); |
e825a66d | 1560 | ret = prepare_write_connect(con); |
0da5d703 SW |
1561 | if (ret < 0) |
1562 | return ret; | |
63733a0f | 1563 | prepare_read_connect(con); |
4e7a5dcd | 1564 | break; |
31b8006e SW |
1565 | |
1566 | case CEPH_MSGR_TAG_RESETSESSION: | |
1567 | /* | |
1568 | * If we connected with a large connect_seq but the peer | |
1569 | * has no record of a session with us (no connection, or | |
1570 | * connect_seq == 0), they will send RESETSESION to indicate | |
1571 | * that they must have reset their session, and may have | |
1572 | * dropped messages. | |
1573 | */ | |
1574 | dout("process_connect got RESET peer seq %u\n", | |
5bdca4e0 | 1575 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
1576 | pr_err("%s%lld %s connection reset\n", |
1577 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1578 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1579 | reset_connection(con); |
6d4221b5 | 1580 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1581 | ret = prepare_write_connect(con); |
1582 | if (ret < 0) | |
1583 | return ret; | |
31b8006e SW |
1584 | prepare_read_connect(con); |
1585 | ||
1586 | /* Tell ceph about it. */ | |
ec302645 | 1587 | mutex_unlock(&con->mutex); |
31b8006e SW |
1588 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1589 | if (con->ops->peer_reset) | |
1590 | con->ops->peer_reset(con); | |
ec302645 | 1591 | mutex_lock(&con->mutex); |
8dacc7da | 1592 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 1593 | return -EAGAIN; |
31b8006e SW |
1594 | break; |
1595 | ||
1596 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1597 | /* | |
1598 | * If we sent a smaller connect_seq than the peer has, try | |
1599 | * again with a larger value. | |
1600 | */ | |
5bdca4e0 | 1601 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 1602 | le32_to_cpu(con->out_connect.connect_seq), |
5bdca4e0 SW |
1603 | le32_to_cpu(con->in_reply.connect_seq)); |
1604 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
6d4221b5 | 1605 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1606 | ret = prepare_write_connect(con); |
1607 | if (ret < 0) | |
1608 | return ret; | |
31b8006e SW |
1609 | prepare_read_connect(con); |
1610 | break; | |
1611 | ||
1612 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1613 | /* | |
1614 | * If we sent a smaller global_seq than the peer has, try | |
1615 | * again with a larger value. | |
1616 | */ | |
eed0ef2c | 1617 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 1618 | con->peer_global_seq, |
5bdca4e0 | 1619 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 1620 | get_global_seq(con->msgr, |
5bdca4e0 | 1621 | le32_to_cpu(con->in_reply.global_seq)); |
6d4221b5 | 1622 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1623 | ret = prepare_write_connect(con); |
1624 | if (ret < 0) | |
1625 | return ret; | |
31b8006e SW |
1626 | prepare_read_connect(con); |
1627 | break; | |
1628 | ||
1629 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1630 | if (req_feat & ~server_feat) { |
1631 | pr_err("%s%lld %s protocol feature mismatch," | |
1632 | " my required %llx > server's %llx, need %llx\n", | |
1633 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1634 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1635 | req_feat, server_feat, req_feat & ~server_feat); |
1636 | con->error_msg = "missing required protocol features"; | |
1637 | fail_protocol(con); | |
1638 | return -1; | |
1639 | } | |
8dacc7da SW |
1640 | |
1641 | BUG_ON(con->state != CON_STATE_NEGOTIATING); | |
1642 | con->state = CON_STATE_OPEN; | |
1643 | ||
31b8006e SW |
1644 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1645 | con->connect_seq++; | |
aba558e2 | 1646 | con->peer_features = server_feat; |
31b8006e SW |
1647 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1648 | con->peer_global_seq, | |
1649 | le32_to_cpu(con->in_reply.connect_seq), | |
1650 | con->connect_seq); | |
1651 | WARN_ON(con->connect_seq != | |
1652 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1653 | |
1654 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
4a861692 | 1655 | set_bit(CON_FLAG_LOSSYTX, &con->flags); |
92ac41d0 | 1656 | |
85effe18 | 1657 | con->delay = 0; /* reset backoff memory */ |
92ac41d0 | 1658 | |
31b8006e SW |
1659 | prepare_read_tag(con); |
1660 | break; | |
1661 | ||
1662 | case CEPH_MSGR_TAG_WAIT: | |
1663 | /* | |
1664 | * If there is a connection race (we are opening | |
1665 | * connections to each other), one of us may just have | |
1666 | * to WAIT. This shouldn't happen if we are the | |
1667 | * client. | |
1668 | */ | |
04177882 SW |
1669 | pr_err("process_connect got WAIT as client\n"); |
1670 | con->error_msg = "protocol error, got WAIT as client"; | |
1671 | return -1; | |
31b8006e SW |
1672 | |
1673 | default: | |
1674 | pr_err("connect protocol error, will retry\n"); | |
1675 | con->error_msg = "protocol error, garbage tag during connect"; | |
1676 | return -1; | |
1677 | } | |
1678 | return 0; | |
1679 | } | |
1680 | ||
1681 | ||
1682 | /* | |
1683 | * read (part of) an ack | |
1684 | */ | |
1685 | static int read_partial_ack(struct ceph_connection *con) | |
1686 | { | |
fd51653f AE |
1687 | int size = sizeof (con->in_temp_ack); |
1688 | int end = size; | |
31b8006e | 1689 | |
fd51653f | 1690 | return read_partial(con, end, size, &con->in_temp_ack); |
31b8006e SW |
1691 | } |
1692 | ||
1693 | ||
1694 | /* | |
1695 | * We can finally discard anything that's been acked. | |
1696 | */ | |
1697 | static void process_ack(struct ceph_connection *con) | |
1698 | { | |
1699 | struct ceph_msg *m; | |
1700 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1701 | u64 seq; | |
1702 | ||
31b8006e SW |
1703 | while (!list_empty(&con->out_sent)) { |
1704 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1705 | list_head); | |
1706 | seq = le64_to_cpu(m->hdr.seq); | |
1707 | if (seq > ack) | |
1708 | break; | |
1709 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1710 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1711 | m->ack_stamp = jiffies; |
31b8006e SW |
1712 | ceph_msg_remove(m); |
1713 | } | |
31b8006e SW |
1714 | prepare_read_tag(con); |
1715 | } | |
1716 | ||
1717 | ||
1718 | ||
1719 | ||
2450418c | 1720 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1721 | struct kvec *section, |
1722 | unsigned int sec_len, u32 *crc) | |
2450418c | 1723 | { |
68b4476b | 1724 | int ret, left; |
2450418c YS |
1725 | |
1726 | BUG_ON(!section); | |
1727 | ||
1728 | while (section->iov_len < sec_len) { | |
1729 | BUG_ON(section->iov_base == NULL); | |
1730 | left = sec_len - section->iov_len; | |
1731 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1732 | section->iov_len, left); | |
1733 | if (ret <= 0) | |
1734 | return ret; | |
1735 | section->iov_len += ret; | |
2450418c | 1736 | } |
fe3ad593 AE |
1737 | if (section->iov_len == sec_len) |
1738 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1739 | |
2450418c YS |
1740 | return 1; |
1741 | } | |
31b8006e | 1742 | |
4740a623 | 1743 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
68b4476b YS |
1744 | |
1745 | static int read_partial_message_pages(struct ceph_connection *con, | |
1746 | struct page **pages, | |
95c96174 | 1747 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1748 | { |
1749 | void *p; | |
1750 | int ret; | |
1751 | int left; | |
1752 | ||
1753 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1754 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1755 | /* (page) data */ | |
1756 | BUG_ON(pages == NULL); | |
1757 | p = kmap(pages[con->in_msg_pos.page]); | |
1758 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1759 | left); | |
bca064d2 | 1760 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1761 | con->in_data_crc = |
1762 | crc32c(con->in_data_crc, | |
1763 | p + con->in_msg_pos.page_pos, ret); | |
1764 | kunmap(pages[con->in_msg_pos.page]); | |
1765 | if (ret <= 0) | |
1766 | return ret; | |
1767 | con->in_msg_pos.data_pos += ret; | |
1768 | con->in_msg_pos.page_pos += ret; | |
1769 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1770 | con->in_msg_pos.page_pos = 0; | |
1771 | con->in_msg_pos.page++; | |
1772 | } | |
1773 | ||
1774 | return ret; | |
1775 | } | |
1776 | ||
1777 | #ifdef CONFIG_BLOCK | |
1778 | static int read_partial_message_bio(struct ceph_connection *con, | |
1779 | struct bio **bio_iter, int *bio_seg, | |
95c96174 | 1780 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1781 | { |
1782 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1783 | void *p; | |
1784 | int ret, left; | |
1785 | ||
68b4476b YS |
1786 | left = min((int)(data_len - con->in_msg_pos.data_pos), |
1787 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1788 | ||
1789 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1790 | ||
1791 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1792 | left); | |
bca064d2 | 1793 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1794 | con->in_data_crc = |
1795 | crc32c(con->in_data_crc, | |
1796 | p + con->in_msg_pos.page_pos, ret); | |
1797 | kunmap(bv->bv_page); | |
1798 | if (ret <= 0) | |
1799 | return ret; | |
1800 | con->in_msg_pos.data_pos += ret; | |
1801 | con->in_msg_pos.page_pos += ret; | |
1802 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1803 | con->in_msg_pos.page_pos = 0; | |
1804 | iter_bio_next(bio_iter, bio_seg); | |
1805 | } | |
1806 | ||
1807 | return ret; | |
1808 | } | |
1809 | #endif | |
1810 | ||
31b8006e SW |
1811 | /* |
1812 | * read (part of) a message. | |
1813 | */ | |
1814 | static int read_partial_message(struct ceph_connection *con) | |
1815 | { | |
1816 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1817 | int size; |
1818 | int end; | |
31b8006e | 1819 | int ret; |
95c96174 | 1820 | unsigned int front_len, middle_len, data_len; |
37675b0f | 1821 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 1822 | u64 seq; |
fe3ad593 | 1823 | u32 crc; |
31b8006e SW |
1824 | |
1825 | dout("read_partial_message con %p msg %p\n", con, m); | |
1826 | ||
1827 | /* header */ | |
fd51653f AE |
1828 | size = sizeof (con->in_hdr); |
1829 | end = size; | |
1830 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1831 | if (ret <= 0) |
1832 | return ret; | |
fe3ad593 AE |
1833 | |
1834 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1835 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1836 | pr_err("read_partial_message bad hdr " | |
1837 | " crc %u != expected %u\n", | |
1838 | crc, con->in_hdr.crc); | |
1839 | return -EBADMSG; | |
1840 | } | |
1841 | ||
31b8006e SW |
1842 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1843 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1844 | return -EIO; | |
1845 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1846 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1847 | return -EIO; | |
1848 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1849 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1850 | return -EIO; | |
1851 | ||
ae18756b SW |
1852 | /* verify seq# */ |
1853 | seq = le64_to_cpu(con->in_hdr.seq); | |
1854 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1855 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1856 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1857 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1858 | seq, con->in_seq + 1); |
1859 | con->in_base_pos = -front_len - middle_len - data_len - | |
1860 | sizeof(m->footer); | |
1861 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1862 | return 0; |
1863 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1864 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1865 | seq, con->in_seq + 1); | |
1866 | con->error_msg = "bad message sequence # for incoming message"; | |
1867 | return -EBADMSG; | |
1868 | } | |
1869 | ||
31b8006e SW |
1870 | /* allocate message? */ |
1871 | if (!con->in_msg) { | |
4740a623 SW |
1872 | int skip = 0; |
1873 | ||
31b8006e SW |
1874 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
1875 | con->in_hdr.front_len, con->in_hdr.data_len); | |
4740a623 SW |
1876 | ret = ceph_con_in_msg_alloc(con, &skip); |
1877 | if (ret < 0) | |
1878 | return ret; | |
2450418c | 1879 | if (skip) { |
31b8006e | 1880 | /* skip this message */ |
a79832f2 | 1881 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1882 | BUG_ON(con->in_msg); |
31b8006e SW |
1883 | con->in_base_pos = -front_len - middle_len - data_len - |
1884 | sizeof(m->footer); | |
1885 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1886 | con->in_seq++; |
31b8006e SW |
1887 | return 0; |
1888 | } | |
38941f80 | 1889 | |
4740a623 | 1890 | BUG_ON(!con->in_msg); |
38941f80 | 1891 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
1892 | m = con->in_msg; |
1893 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1894 | if (m->middle) |
1895 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1896 | |
1897 | con->in_msg_pos.page = 0; | |
68b4476b | 1898 | if (m->pages) |
c5c6b19d | 1899 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1900 | else |
1901 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1902 | con->in_msg_pos.data_pos = 0; |
a4107026 SW |
1903 | |
1904 | #ifdef CONFIG_BLOCK | |
1905 | if (m->bio) | |
1906 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1907 | #endif | |
31b8006e SW |
1908 | } |
1909 | ||
1910 | /* front */ | |
2450418c YS |
1911 | ret = read_partial_message_section(con, &m->front, front_len, |
1912 | &con->in_front_crc); | |
1913 | if (ret <= 0) | |
1914 | return ret; | |
31b8006e SW |
1915 | |
1916 | /* middle */ | |
2450418c | 1917 | if (m->middle) { |
213c99ee SW |
1918 | ret = read_partial_message_section(con, &m->middle->vec, |
1919 | middle_len, | |
2450418c | 1920 | &con->in_middle_crc); |
31b8006e SW |
1921 | if (ret <= 0) |
1922 | return ret; | |
31b8006e SW |
1923 | } |
1924 | ||
1925 | /* (page) data */ | |
31b8006e | 1926 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1927 | if (m->pages) { |
1928 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1929 | data_len, do_datacrc); |
68b4476b YS |
1930 | if (ret <= 0) |
1931 | return ret; | |
1932 | #ifdef CONFIG_BLOCK | |
1933 | } else if (m->bio) { | |
a4107026 | 1934 | BUG_ON(!m->bio_iter); |
68b4476b YS |
1935 | ret = read_partial_message_bio(con, |
1936 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1937 | data_len, do_datacrc); |
68b4476b YS |
1938 | if (ret <= 0) |
1939 | return ret; | |
1940 | #endif | |
1941 | } else { | |
1942 | BUG_ON(1); | |
31b8006e SW |
1943 | } |
1944 | } | |
1945 | ||
31b8006e | 1946 | /* footer */ |
fd51653f AE |
1947 | size = sizeof (m->footer); |
1948 | end += size; | |
1949 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
1950 | if (ret <= 0) |
1951 | return ret; | |
1952 | ||
31b8006e SW |
1953 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
1954 | m, front_len, m->footer.front_crc, middle_len, | |
1955 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1956 | ||
1957 | /* crc ok? */ | |
1958 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1959 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1960 | m, con->in_front_crc, m->footer.front_crc); | |
1961 | return -EBADMSG; | |
1962 | } | |
1963 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1964 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1965 | m, con->in_middle_crc, m->footer.middle_crc); | |
1966 | return -EBADMSG; | |
1967 | } | |
bca064d2 | 1968 | if (do_datacrc && |
31b8006e SW |
1969 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
1970 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1971 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1972 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1973 | return -EBADMSG; | |
1974 | } | |
1975 | ||
1976 | return 1; /* done! */ | |
1977 | } | |
1978 | ||
1979 | /* | |
1980 | * Process message. This happens in the worker thread. The callback should | |
1981 | * be careful not to do anything that waits on other incoming messages or it | |
1982 | * may deadlock. | |
1983 | */ | |
1984 | static void process_message(struct ceph_connection *con) | |
1985 | { | |
5e095e8b | 1986 | struct ceph_msg *msg; |
31b8006e | 1987 | |
38941f80 AE |
1988 | BUG_ON(con->in_msg->con != con); |
1989 | con->in_msg->con = NULL; | |
5e095e8b | 1990 | msg = con->in_msg; |
31b8006e | 1991 | con->in_msg = NULL; |
36eb71aa | 1992 | con->ops->put(con); |
31b8006e SW |
1993 | |
1994 | /* if first message, set peer_name */ | |
1995 | if (con->peer_name.type == 0) | |
dbad185d | 1996 | con->peer_name = msg->hdr.src; |
31b8006e | 1997 | |
31b8006e | 1998 | con->in_seq++; |
ec302645 | 1999 | mutex_unlock(&con->mutex); |
31b8006e SW |
2000 | |
2001 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
2002 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 2003 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
2004 | le16_to_cpu(msg->hdr.type), |
2005 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2006 | le32_to_cpu(msg->hdr.front_len), | |
2007 | le32_to_cpu(msg->hdr.data_len), | |
2008 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
2009 | con->ops->dispatch(con, msg); | |
ec302645 SW |
2010 | |
2011 | mutex_lock(&con->mutex); | |
31b8006e SW |
2012 | } |
2013 | ||
2014 | ||
2015 | /* | |
2016 | * Write something to the socket. Called in a worker thread when the | |
2017 | * socket appears to be writeable and we have something ready to send. | |
2018 | */ | |
2019 | static int try_write(struct ceph_connection *con) | |
2020 | { | |
31b8006e SW |
2021 | int ret = 1; |
2022 | ||
d59315ca | 2023 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 2024 | |
31b8006e SW |
2025 | more: |
2026 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
2027 | ||
2028 | /* open the socket first? */ | |
8dacc7da SW |
2029 | if (con->state == CON_STATE_PREOPEN) { |
2030 | BUG_ON(con->sock); | |
2031 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2032 | |
e2200423 | 2033 | con_out_kvec_reset(con); |
e825a66d | 2034 | prepare_write_banner(con); |
eed0ef2c | 2035 | prepare_read_banner(con); |
31b8006e | 2036 | |
cf3e5c40 | 2037 | BUG_ON(con->in_msg); |
31b8006e SW |
2038 | con->in_tag = CEPH_MSGR_TAG_READY; |
2039 | dout("try_write initiating connect on %p new state %lu\n", | |
2040 | con, con->state); | |
41617d0c AE |
2041 | ret = ceph_tcp_connect(con); |
2042 | if (ret < 0) { | |
31b8006e | 2043 | con->error_msg = "connect error"; |
31b8006e SW |
2044 | goto out; |
2045 | } | |
2046 | } | |
2047 | ||
2048 | more_kvec: | |
2049 | /* kvec data queued? */ | |
2050 | if (con->out_skip) { | |
2051 | ret = write_partial_skip(con); | |
2052 | if (ret <= 0) | |
42961d23 | 2053 | goto out; |
31b8006e SW |
2054 | } |
2055 | if (con->out_kvec_left) { | |
2056 | ret = write_partial_kvec(con); | |
2057 | if (ret <= 0) | |
42961d23 | 2058 | goto out; |
31b8006e SW |
2059 | } |
2060 | ||
2061 | /* msg pages? */ | |
2062 | if (con->out_msg) { | |
c86a2930 SW |
2063 | if (con->out_msg_done) { |
2064 | ceph_msg_put(con->out_msg); | |
2065 | con->out_msg = NULL; /* we're done with this one */ | |
2066 | goto do_next; | |
2067 | } | |
2068 | ||
31b8006e SW |
2069 | ret = write_partial_msg_pages(con); |
2070 | if (ret == 1) | |
2071 | goto more_kvec; /* we need to send the footer, too! */ | |
2072 | if (ret == 0) | |
42961d23 | 2073 | goto out; |
31b8006e SW |
2074 | if (ret < 0) { |
2075 | dout("try_write write_partial_msg_pages err %d\n", | |
2076 | ret); | |
42961d23 | 2077 | goto out; |
31b8006e SW |
2078 | } |
2079 | } | |
2080 | ||
c86a2930 | 2081 | do_next: |
8dacc7da | 2082 | if (con->state == CON_STATE_OPEN) { |
31b8006e SW |
2083 | /* is anything else pending? */ |
2084 | if (!list_empty(&con->out_queue)) { | |
2085 | prepare_write_message(con); | |
2086 | goto more; | |
2087 | } | |
2088 | if (con->in_seq > con->in_seq_acked) { | |
2089 | prepare_write_ack(con); | |
2090 | goto more; | |
2091 | } | |
4a861692 SW |
2092 | if (test_and_clear_bit(CON_FLAG_KEEPALIVE_PENDING, |
2093 | &con->flags)) { | |
31b8006e SW |
2094 | prepare_write_keepalive(con); |
2095 | goto more; | |
2096 | } | |
2097 | } | |
2098 | ||
2099 | /* Nothing to do! */ | |
4a861692 | 2100 | clear_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
31b8006e | 2101 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2102 | ret = 0; |
2103 | out: | |
42961d23 | 2104 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2105 | return ret; |
2106 | } | |
2107 | ||
2108 | ||
2109 | ||
2110 | /* | |
2111 | * Read what we can from the socket. | |
2112 | */ | |
2113 | static int try_read(struct ceph_connection *con) | |
2114 | { | |
31b8006e SW |
2115 | int ret = -1; |
2116 | ||
8dacc7da SW |
2117 | more: |
2118 | dout("try_read start on %p state %lu\n", con, con->state); | |
2119 | if (con->state != CON_STATE_CONNECTING && | |
2120 | con->state != CON_STATE_NEGOTIATING && | |
2121 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2122 | return 0; |
2123 | ||
8dacc7da | 2124 | BUG_ON(!con->sock); |
ec302645 | 2125 | |
31b8006e SW |
2126 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2127 | con->in_base_pos); | |
0da5d703 | 2128 | |
8dacc7da | 2129 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2130 | dout("try_read connecting\n"); |
2131 | ret = read_partial_banner(con); | |
2132 | if (ret <= 0) | |
ab166d5a | 2133 | goto out; |
7593af92 AE |
2134 | ret = process_banner(con); |
2135 | if (ret < 0) | |
2136 | goto out; | |
2137 | ||
8dacc7da SW |
2138 | BUG_ON(con->state != CON_STATE_CONNECTING); |
2139 | con->state = CON_STATE_NEGOTIATING; | |
7593af92 | 2140 | |
6d4221b5 JS |
2141 | /* |
2142 | * Received banner is good, exchange connection info. | |
2143 | * Do not reset out_kvec, as sending our banner raced | |
2144 | * with receiving peer banner after connect completed. | |
2145 | */ | |
7593af92 AE |
2146 | ret = prepare_write_connect(con); |
2147 | if (ret < 0) | |
2148 | goto out; | |
2149 | prepare_read_connect(con); | |
2150 | ||
2151 | /* Send connection info before awaiting response */ | |
0da5d703 SW |
2152 | goto out; |
2153 | } | |
2154 | ||
8dacc7da | 2155 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2156 | dout("try_read negotiating\n"); |
31b8006e SW |
2157 | ret = read_partial_connect(con); |
2158 | if (ret <= 0) | |
31b8006e | 2159 | goto out; |
98bdb0aa SW |
2160 | ret = process_connect(con); |
2161 | if (ret < 0) | |
2162 | goto out; | |
31b8006e SW |
2163 | goto more; |
2164 | } | |
2165 | ||
8dacc7da SW |
2166 | BUG_ON(con->state != CON_STATE_OPEN); |
2167 | ||
31b8006e SW |
2168 | if (con->in_base_pos < 0) { |
2169 | /* | |
2170 | * skipping + discarding content. | |
2171 | * | |
2172 | * FIXME: there must be a better way to do this! | |
2173 | */ | |
84495f49 AE |
2174 | static char buf[SKIP_BUF_SIZE]; |
2175 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2176 | ||
31b8006e SW |
2177 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2178 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2179 | if (ret <= 0) | |
98bdb0aa | 2180 | goto out; |
31b8006e SW |
2181 | con->in_base_pos += ret; |
2182 | if (con->in_base_pos) | |
2183 | goto more; | |
2184 | } | |
2185 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2186 | /* | |
2187 | * what's next? | |
2188 | */ | |
2189 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2190 | if (ret <= 0) | |
98bdb0aa | 2191 | goto out; |
31b8006e SW |
2192 | dout("try_read got tag %d\n", (int)con->in_tag); |
2193 | switch (con->in_tag) { | |
2194 | case CEPH_MSGR_TAG_MSG: | |
2195 | prepare_read_message(con); | |
2196 | break; | |
2197 | case CEPH_MSGR_TAG_ACK: | |
2198 | prepare_read_ack(con); | |
2199 | break; | |
2200 | case CEPH_MSGR_TAG_CLOSE: | |
8dacc7da SW |
2201 | con_close_socket(con); |
2202 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2203 | goto out; |
31b8006e SW |
2204 | default: |
2205 | goto bad_tag; | |
2206 | } | |
2207 | } | |
2208 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2209 | ret = read_partial_message(con); | |
2210 | if (ret <= 0) { | |
2211 | switch (ret) { | |
2212 | case -EBADMSG: | |
2213 | con->error_msg = "bad crc"; | |
2214 | ret = -EIO; | |
98bdb0aa | 2215 | break; |
31b8006e SW |
2216 | case -EIO: |
2217 | con->error_msg = "io error"; | |
98bdb0aa | 2218 | break; |
31b8006e | 2219 | } |
98bdb0aa | 2220 | goto out; |
31b8006e SW |
2221 | } |
2222 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2223 | goto more; | |
2224 | process_message(con); | |
7b862e07 SW |
2225 | if (con->state == CON_STATE_OPEN) |
2226 | prepare_read_tag(con); | |
31b8006e SW |
2227 | goto more; |
2228 | } | |
2229 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2230 | ret = read_partial_ack(con); | |
2231 | if (ret <= 0) | |
98bdb0aa | 2232 | goto out; |
31b8006e SW |
2233 | process_ack(con); |
2234 | goto more; | |
2235 | } | |
2236 | ||
31b8006e | 2237 | out: |
98bdb0aa | 2238 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2239 | return ret; |
2240 | ||
2241 | bad_tag: | |
2242 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2243 | con->error_msg = "protocol error, garbage tag"; | |
2244 | ret = -1; | |
2245 | goto out; | |
2246 | } | |
2247 | ||
2248 | ||
2249 | /* | |
2250 | * Atomically queue work on a connection. Bump @con reference to | |
2251 | * avoid races with connection teardown. | |
31b8006e SW |
2252 | */ |
2253 | static void queue_con(struct ceph_connection *con) | |
2254 | { | |
31b8006e SW |
2255 | if (!con->ops->get(con)) { |
2256 | dout("queue_con %p ref count 0\n", con); | |
2257 | return; | |
2258 | } | |
2259 | ||
f363e45f | 2260 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2261 | dout("queue_con %p - already queued\n", con); |
2262 | con->ops->put(con); | |
2263 | } else { | |
2264 | dout("queue_con %p\n", con); | |
2265 | } | |
2266 | } | |
2267 | ||
2268 | /* | |
2269 | * Do some work on a connection. Drop a connection ref when we're done. | |
2270 | */ | |
2271 | static void con_work(struct work_struct *work) | |
2272 | { | |
2273 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2274 | work.work); | |
0da5d703 | 2275 | int ret; |
31b8006e | 2276 | |
9dd4658d | 2277 | mutex_lock(&con->mutex); |
0da5d703 | 2278 | restart: |
4a861692 | 2279 | if (test_and_clear_bit(CON_FLAG_SOCK_CLOSED, &con->flags)) { |
8dacc7da SW |
2280 | switch (con->state) { |
2281 | case CON_STATE_CONNECTING: | |
188048bc | 2282 | con->error_msg = "connection failed"; |
8dacc7da SW |
2283 | break; |
2284 | case CON_STATE_NEGOTIATING: | |
2285 | con->error_msg = "negotiation failed"; | |
2286 | break; | |
2287 | case CON_STATE_OPEN: | |
2288 | con->error_msg = "socket closed"; | |
2289 | break; | |
2290 | default: | |
2291 | dout("unrecognized con state %d\n", (int)con->state); | |
e27947c7 | 2292 | con->error_msg = "unrecognized con state"; |
8dacc7da SW |
2293 | BUG(); |
2294 | } | |
188048bc AE |
2295 | goto fault; |
2296 | } | |
2297 | ||
4a861692 | 2298 | if (test_and_clear_bit(CON_FLAG_BACKOFF, &con->flags)) { |
60bf8bf8 SW |
2299 | dout("con_work %p backing off\n", con); |
2300 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2301 | round_jiffies_relative(con->delay))) { | |
2302 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2303 | mutex_unlock(&con->mutex); | |
2304 | return; | |
2305 | } else { | |
2306 | con->ops->put(con); | |
2307 | dout("con_work %p FAILED to back off %lu\n", con, | |
2308 | con->delay); | |
2309 | } | |
2310 | } | |
9dd4658d | 2311 | |
8dacc7da | 2312 | if (con->state == CON_STATE_STANDBY) { |
e00de341 SW |
2313 | dout("con_work %p STANDBY\n", con); |
2314 | goto done; | |
2315 | } | |
8dacc7da | 2316 | if (con->state == CON_STATE_CLOSED) { |
2e8cb100 SW |
2317 | dout("con_work %p CLOSED\n", con); |
2318 | BUG_ON(con->sock); | |
31b8006e SW |
2319 | goto done; |
2320 | } | |
8dacc7da | 2321 | if (con->state == CON_STATE_PREOPEN) { |
31b8006e | 2322 | dout("con_work OPENING\n"); |
2e8cb100 | 2323 | BUG_ON(con->sock); |
31b8006e SW |
2324 | } |
2325 | ||
0da5d703 SW |
2326 | ret = try_read(con); |
2327 | if (ret == -EAGAIN) | |
2328 | goto restart; | |
3a140a0d SW |
2329 | if (ret < 0) { |
2330 | con->error_msg = "socket error on read"; | |
0da5d703 | 2331 | goto fault; |
3a140a0d | 2332 | } |
0da5d703 SW |
2333 | |
2334 | ret = try_write(con); | |
2335 | if (ret == -EAGAIN) | |
2336 | goto restart; | |
3a140a0d SW |
2337 | if (ret < 0) { |
2338 | con->error_msg = "socket error on write"; | |
0da5d703 | 2339 | goto fault; |
3a140a0d | 2340 | } |
31b8006e SW |
2341 | |
2342 | done: | |
9dd4658d | 2343 | mutex_unlock(&con->mutex); |
9dd4658d | 2344 | done_unlocked: |
31b8006e | 2345 | con->ops->put(con); |
0da5d703 SW |
2346 | return; |
2347 | ||
2348 | fault: | |
0da5d703 SW |
2349 | ceph_fault(con); /* error/fault path */ |
2350 | goto done_unlocked; | |
31b8006e SW |
2351 | } |
2352 | ||
2353 | ||
2354 | /* | |
2355 | * Generic error/fault handler. A retry mechanism is used with | |
2356 | * exponential backoff | |
2357 | */ | |
2358 | static void ceph_fault(struct ceph_connection *con) | |
8636ea67 | 2359 | __releases(con->mutex) |
31b8006e SW |
2360 | { |
2361 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2362 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2363 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2364 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2365 | |
8dacc7da SW |
2366 | BUG_ON(con->state != CON_STATE_CONNECTING && |
2367 | con->state != CON_STATE_NEGOTIATING && | |
2368 | con->state != CON_STATE_OPEN); | |
ec302645 | 2369 | |
31b8006e | 2370 | con_close_socket(con); |
5e095e8b | 2371 | |
4a861692 | 2372 | if (test_bit(CON_FLAG_LOSSYTX, &con->flags)) { |
8dacc7da SW |
2373 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
2374 | con->state = CON_STATE_CLOSED; | |
3b5ede07 SW |
2375 | goto out_unlock; |
2376 | } | |
2377 | ||
5e095e8b | 2378 | if (con->in_msg) { |
38941f80 AE |
2379 | BUG_ON(con->in_msg->con != con); |
2380 | con->in_msg->con = NULL; | |
5e095e8b SW |
2381 | ceph_msg_put(con->in_msg); |
2382 | con->in_msg = NULL; | |
36eb71aa | 2383 | con->ops->put(con); |
5e095e8b | 2384 | } |
31b8006e | 2385 | |
e80a52d1 SW |
2386 | /* Requeue anything that hasn't been acked */ |
2387 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2388 | |
e76661d0 SW |
2389 | /* If there are no messages queued or keepalive pending, place |
2390 | * the connection in a STANDBY state */ | |
2391 | if (list_empty(&con->out_queue) && | |
4a861692 | 2392 | !test_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags)) { |
e00de341 | 2393 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
4a861692 | 2394 | clear_bit(CON_FLAG_WRITE_PENDING, &con->flags); |
8dacc7da | 2395 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
2396 | } else { |
2397 | /* retry after a delay. */ | |
8dacc7da | 2398 | con->state = CON_STATE_PREOPEN; |
e80a52d1 SW |
2399 | if (con->delay == 0) |
2400 | con->delay = BASE_DELAY_INTERVAL; | |
2401 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2402 | con->delay *= 2; | |
e80a52d1 SW |
2403 | con->ops->get(con); |
2404 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2405 | round_jiffies_relative(con->delay))) { |
2406 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2407 | } else { | |
e80a52d1 | 2408 | con->ops->put(con); |
60bf8bf8 SW |
2409 | dout("fault failed to queue %p delay %lu, backoff\n", |
2410 | con, con->delay); | |
2411 | /* | |
2412 | * In many cases we see a socket state change | |
2413 | * while con_work is running and end up | |
2414 | * queuing (non-delayed) work, such that we | |
2415 | * can't backoff with a delay. Set a flag so | |
2416 | * that when con_work restarts we schedule the | |
2417 | * delay then. | |
2418 | */ | |
4a861692 | 2419 | set_bit(CON_FLAG_BACKOFF, &con->flags); |
60bf8bf8 | 2420 | } |
31b8006e SW |
2421 | } |
2422 | ||
91e45ce3 SW |
2423 | out_unlock: |
2424 | mutex_unlock(&con->mutex); | |
161fd65a SW |
2425 | /* |
2426 | * in case we faulted due to authentication, invalidate our | |
2427 | * current tickets so that we can get new ones. | |
213c99ee | 2428 | */ |
161fd65a SW |
2429 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2430 | dout("calling invalidate_authorizer()\n"); | |
2431 | con->ops->invalidate_authorizer(con); | |
2432 | } | |
2433 | ||
31b8006e SW |
2434 | if (con->ops->fault) |
2435 | con->ops->fault(con); | |
2436 | } | |
2437 | ||
2438 | ||
2439 | ||
2440 | /* | |
15d9882c | 2441 | * initialize a new messenger instance |
31b8006e | 2442 | */ |
15d9882c AE |
2443 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2444 | struct ceph_entity_addr *myaddr, | |
2445 | u32 supported_features, | |
2446 | u32 required_features, | |
2447 | bool nocrc) | |
31b8006e | 2448 | { |
3d14c5d2 YS |
2449 | msgr->supported_features = supported_features; |
2450 | msgr->required_features = required_features; | |
2451 | ||
31b8006e SW |
2452 | spin_lock_init(&msgr->global_seq_lock); |
2453 | ||
31b8006e SW |
2454 | if (myaddr) |
2455 | msgr->inst.addr = *myaddr; | |
2456 | ||
2457 | /* select a random nonce */ | |
ac8839d7 | 2458 | msgr->inst.addr.type = 0; |
103e2d3a | 2459 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2460 | encode_my_addr(msgr); |
15d9882c | 2461 | msgr->nocrc = nocrc; |
31b8006e | 2462 | |
a2a32584 | 2463 | atomic_set(&msgr->stopping, 0); |
31b8006e | 2464 | |
15d9882c | 2465 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2466 | } |
15d9882c | 2467 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2468 | |
e00de341 SW |
2469 | static void clear_standby(struct ceph_connection *con) |
2470 | { | |
2471 | /* come back from STANDBY? */ | |
8dacc7da | 2472 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 2473 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 2474 | con->state = CON_STATE_PREOPEN; |
e00de341 | 2475 | con->connect_seq++; |
4a861692 SW |
2476 | WARN_ON(test_bit(CON_FLAG_WRITE_PENDING, &con->flags)); |
2477 | WARN_ON(test_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags)); | |
e00de341 SW |
2478 | } |
2479 | } | |
2480 | ||
31b8006e SW |
2481 | /* |
2482 | * Queue up an outgoing message on the given connection. | |
2483 | */ | |
2484 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2485 | { | |
31b8006e | 2486 | /* set src+dst */ |
dbad185d | 2487 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 2488 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
2489 | msg->needs_out_seq = true; |
2490 | ||
ec302645 | 2491 | mutex_lock(&con->mutex); |
92ce034b | 2492 | |
8dacc7da | 2493 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
2494 | dout("con_send %p closed, dropping %p\n", con, msg); |
2495 | ceph_msg_put(msg); | |
2496 | mutex_unlock(&con->mutex); | |
2497 | return; | |
2498 | } | |
2499 | ||
38941f80 | 2500 | BUG_ON(msg->con != NULL); |
36eb71aa | 2501 | msg->con = con->ops->get(con); |
92ce034b AE |
2502 | BUG_ON(msg->con == NULL); |
2503 | ||
31b8006e SW |
2504 | BUG_ON(!list_empty(&msg->list_head)); |
2505 | list_add_tail(&msg->list_head, &con->out_queue); | |
2506 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2507 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2508 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2509 | le32_to_cpu(msg->hdr.front_len), | |
2510 | le32_to_cpu(msg->hdr.middle_len), | |
2511 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
2512 | |
2513 | clear_standby(con); | |
ec302645 | 2514 | mutex_unlock(&con->mutex); |
31b8006e SW |
2515 | |
2516 | /* if there wasn't anything waiting to send before, queue | |
2517 | * new work */ | |
4a861692 | 2518 | if (test_and_set_bit(CON_FLAG_WRITE_PENDING, &con->flags) == 0) |
31b8006e SW |
2519 | queue_con(con); |
2520 | } | |
3d14c5d2 | 2521 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2522 | |
2523 | /* | |
2524 | * Revoke a message that was previously queued for send | |
2525 | */ | |
6740a845 | 2526 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2527 | { |
6740a845 AE |
2528 | struct ceph_connection *con = msg->con; |
2529 | ||
2530 | if (!con) | |
2531 | return; /* Message not in our possession */ | |
2532 | ||
ec302645 | 2533 | mutex_lock(&con->mutex); |
31b8006e | 2534 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2535 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2536 | list_del_init(&msg->list_head); |
38941f80 | 2537 | BUG_ON(msg->con == NULL); |
36eb71aa | 2538 | msg->con->ops->put(msg->con); |
38941f80 | 2539 | msg->con = NULL; |
31b8006e | 2540 | msg->hdr.seq = 0; |
38941f80 | 2541 | |
31b8006e | 2542 | ceph_msg_put(msg); |
ed98adad SW |
2543 | } |
2544 | if (con->out_msg == msg) { | |
38941f80 | 2545 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2546 | con->out_msg = NULL; |
31b8006e SW |
2547 | if (con->out_kvec_is_msg) { |
2548 | con->out_skip = con->out_kvec_bytes; | |
2549 | con->out_kvec_is_msg = false; | |
2550 | } | |
ed98adad | 2551 | msg->hdr.seq = 0; |
92ce034b AE |
2552 | |
2553 | ceph_msg_put(msg); | |
31b8006e | 2554 | } |
ec302645 | 2555 | mutex_unlock(&con->mutex); |
31b8006e SW |
2556 | } |
2557 | ||
350b1c32 | 2558 | /* |
0d59ab81 | 2559 | * Revoke a message that we may be reading data into |
350b1c32 | 2560 | */ |
8921d114 | 2561 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2562 | { |
8921d114 AE |
2563 | struct ceph_connection *con; |
2564 | ||
2565 | BUG_ON(msg == NULL); | |
2566 | if (!msg->con) { | |
2567 | dout("%s msg %p null con\n", __func__, msg); | |
2568 | ||
2569 | return; /* Message not in our possession */ | |
2570 | } | |
2571 | ||
2572 | con = msg->con; | |
350b1c32 | 2573 | mutex_lock(&con->mutex); |
8921d114 | 2574 | if (con->in_msg == msg) { |
95c96174 ED |
2575 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
2576 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
2577 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
2578 | |
2579 | /* skip rest of message */ | |
8921d114 AE |
2580 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
2581 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 2582 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
2583 | front_len - |
2584 | middle_len - | |
2585 | data_len - | |
350b1c32 | 2586 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2587 | ceph_msg_put(con->in_msg); |
2588 | con->in_msg = NULL; | |
2589 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2590 | con->in_seq++; |
350b1c32 | 2591 | } else { |
8921d114 AE |
2592 | dout("%s %p in_msg %p msg %p no-op\n", |
2593 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
2594 | } |
2595 | mutex_unlock(&con->mutex); | |
2596 | } | |
2597 | ||
31b8006e SW |
2598 | /* |
2599 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2600 | */ | |
2601 | void ceph_con_keepalive(struct ceph_connection *con) | |
2602 | { | |
e00de341 | 2603 | dout("con_keepalive %p\n", con); |
00650931 | 2604 | mutex_lock(&con->mutex); |
e00de341 | 2605 | clear_standby(con); |
00650931 | 2606 | mutex_unlock(&con->mutex); |
4a861692 SW |
2607 | if (test_and_set_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags) == 0 && |
2608 | test_and_set_bit(CON_FLAG_WRITE_PENDING, &con->flags) == 0) | |
31b8006e SW |
2609 | queue_con(con); |
2610 | } | |
3d14c5d2 | 2611 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2612 | |
2613 | ||
2614 | /* | |
2615 | * construct a new message with given type, size | |
2616 | * the new msg has a ref count of 1. | |
2617 | */ | |
b61c2763 SW |
2618 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2619 | bool can_fail) | |
31b8006e SW |
2620 | { |
2621 | struct ceph_msg *m; | |
2622 | ||
34d23762 | 2623 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2624 | if (m == NULL) |
2625 | goto out; | |
c2e552e7 | 2626 | kref_init(&m->kref); |
38941f80 AE |
2627 | |
2628 | m->con = NULL; | |
31b8006e SW |
2629 | INIT_LIST_HEAD(&m->list_head); |
2630 | ||
45c6ceb5 | 2631 | m->hdr.tid = 0; |
31b8006e | 2632 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2633 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2634 | m->hdr.version = 0; | |
31b8006e SW |
2635 | m->hdr.front_len = cpu_to_le32(front_len); |
2636 | m->hdr.middle_len = 0; | |
bb257664 SW |
2637 | m->hdr.data_len = 0; |
2638 | m->hdr.data_off = 0; | |
45c6ceb5 | 2639 | m->hdr.reserved = 0; |
31b8006e SW |
2640 | m->footer.front_crc = 0; |
2641 | m->footer.middle_crc = 0; | |
2642 | m->footer.data_crc = 0; | |
45c6ceb5 | 2643 | m->footer.flags = 0; |
31b8006e SW |
2644 | m->front_max = front_len; |
2645 | m->front_is_vmalloc = false; | |
2646 | m->more_to_follow = false; | |
c0d5f9db | 2647 | m->ack_stamp = 0; |
31b8006e SW |
2648 | m->pool = NULL; |
2649 | ||
ca20892d HC |
2650 | /* middle */ |
2651 | m->middle = NULL; | |
2652 | ||
2653 | /* data */ | |
2654 | m->nr_pages = 0; | |
2655 | m->page_alignment = 0; | |
2656 | m->pages = NULL; | |
2657 | m->pagelist = NULL; | |
2658 | m->bio = NULL; | |
2659 | m->bio_iter = NULL; | |
2660 | m->bio_seg = 0; | |
2661 | m->trail = NULL; | |
2662 | ||
31b8006e SW |
2663 | /* front */ |
2664 | if (front_len) { | |
2665 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2666 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2667 | PAGE_KERNEL); |
2668 | m->front_is_vmalloc = true; | |
2669 | } else { | |
34d23762 | 2670 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2671 | } |
2672 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2673 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2674 | front_len); |
2675 | goto out2; | |
2676 | } | |
2677 | } else { | |
2678 | m->front.iov_base = NULL; | |
2679 | } | |
2680 | m->front.iov_len = front_len; | |
2681 | ||
bb257664 | 2682 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2683 | return m; |
2684 | ||
2685 | out2: | |
2686 | ceph_msg_put(m); | |
2687 | out: | |
b61c2763 SW |
2688 | if (!can_fail) { |
2689 | pr_err("msg_new can't create type %d front %d\n", type, | |
2690 | front_len); | |
f0ed1b7c | 2691 | WARN_ON(1); |
b61c2763 SW |
2692 | } else { |
2693 | dout("msg_new can't create type %d front %d\n", type, | |
2694 | front_len); | |
2695 | } | |
a79832f2 | 2696 | return NULL; |
31b8006e | 2697 | } |
3d14c5d2 | 2698 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2699 | |
31b8006e SW |
2700 | /* |
2701 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2702 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2703 | * per-type header in the front and then gracefully fail (i.e., | |
2704 | * propagate the error to the caller based on info in the front) when | |
2705 | * the middle is too large. | |
2706 | */ | |
2450418c | 2707 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2708 | { |
2709 | int type = le16_to_cpu(msg->hdr.type); | |
2710 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2711 | ||
2712 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2713 | ceph_msg_type_name(type), middle_len); | |
2714 | BUG_ON(!middle_len); | |
2715 | BUG_ON(msg->middle); | |
2716 | ||
b6c1d5b8 | 2717 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2718 | if (!msg->middle) |
2719 | return -ENOMEM; | |
2720 | return 0; | |
2721 | } | |
2722 | ||
2450418c | 2723 | /* |
1c20f2d2 AE |
2724 | * Allocate a message for receiving an incoming message on a |
2725 | * connection, and save the result in con->in_msg. Uses the | |
2726 | * connection's private alloc_msg op if available. | |
2727 | * | |
4740a623 SW |
2728 | * Returns 0 on success, or a negative error code. |
2729 | * | |
2730 | * On success, if we set *skip = 1: | |
2731 | * - the next message should be skipped and ignored. | |
2732 | * - con->in_msg == NULL | |
2733 | * or if we set *skip = 0: | |
2734 | * - con->in_msg is non-null. | |
2735 | * On error (ENOMEM, EAGAIN, ...), | |
2736 | * - con->in_msg == NULL | |
2450418c | 2737 | */ |
4740a623 | 2738 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
2450418c | 2739 | { |
4740a623 | 2740 | struct ceph_msg_header *hdr = &con->in_hdr; |
2450418c YS |
2741 | int type = le16_to_cpu(hdr->type); |
2742 | int front_len = le32_to_cpu(hdr->front_len); | |
2743 | int middle_len = le32_to_cpu(hdr->middle_len); | |
4740a623 | 2744 | int ret = 0; |
2450418c | 2745 | |
1c20f2d2 | 2746 | BUG_ON(con->in_msg != NULL); |
2450418c YS |
2747 | |
2748 | if (con->ops->alloc_msg) { | |
61399191 SW |
2749 | struct ceph_msg *msg; |
2750 | ||
0547a9b3 | 2751 | mutex_unlock(&con->mutex); |
2450418c | 2752 | msg = con->ops->alloc_msg(con, hdr, skip); |
0547a9b3 | 2753 | mutex_lock(&con->mutex); |
61399191 SW |
2754 | if (con->state != CON_STATE_OPEN) { |
2755 | ceph_msg_put(msg); | |
2756 | return -EAGAIN; | |
2757 | } | |
2758 | con->in_msg = msg; | |
92ce034b | 2759 | if (con->in_msg) { |
36eb71aa | 2760 | con->in_msg->con = con->ops->get(con); |
92ce034b AE |
2761 | BUG_ON(con->in_msg->con == NULL); |
2762 | } | |
4740a623 | 2763 | if (*skip) { |
1c20f2d2 | 2764 | con->in_msg = NULL; |
4740a623 SW |
2765 | return 0; |
2766 | } | |
2767 | if (!con->in_msg) { | |
2768 | con->error_msg = | |
2769 | "error allocating memory for incoming message"; | |
2770 | return -ENOMEM; | |
2771 | } | |
2450418c | 2772 | } |
1c20f2d2 AE |
2773 | if (!con->in_msg) { |
2774 | con->in_msg = ceph_msg_new(type, front_len, GFP_NOFS, false); | |
2775 | if (!con->in_msg) { | |
2450418c YS |
2776 | pr_err("unable to allocate msg type %d len %d\n", |
2777 | type, front_len); | |
4740a623 | 2778 | return -ENOMEM; |
2450418c | 2779 | } |
36eb71aa | 2780 | con->in_msg->con = con->ops->get(con); |
92ce034b | 2781 | BUG_ON(con->in_msg->con == NULL); |
1c20f2d2 | 2782 | con->in_msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2783 | } |
1c20f2d2 | 2784 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2785 | |
1c20f2d2 AE |
2786 | if (middle_len && !con->in_msg->middle) { |
2787 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 2788 | if (ret < 0) { |
1c20f2d2 AE |
2789 | ceph_msg_put(con->in_msg); |
2790 | con->in_msg = NULL; | |
2450418c YS |
2791 | } |
2792 | } | |
9d7f0f13 | 2793 | |
4740a623 | 2794 | return ret; |
2450418c YS |
2795 | } |
2796 | ||
31b8006e SW |
2797 | |
2798 | /* | |
2799 | * Free a generically kmalloc'd message. | |
2800 | */ | |
2801 | void ceph_msg_kfree(struct ceph_msg *m) | |
2802 | { | |
2803 | dout("msg_kfree %p\n", m); | |
2804 | if (m->front_is_vmalloc) | |
2805 | vfree(m->front.iov_base); | |
2806 | else | |
2807 | kfree(m->front.iov_base); | |
2808 | kfree(m); | |
2809 | } | |
2810 | ||
2811 | /* | |
2812 | * Drop a msg ref. Destroy as needed. | |
2813 | */ | |
c2e552e7 SW |
2814 | void ceph_msg_last_put(struct kref *kref) |
2815 | { | |
2816 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2817 | |
c2e552e7 SW |
2818 | dout("ceph_msg_put last one on %p\n", m); |
2819 | WARN_ON(!list_empty(&m->list_head)); | |
2820 | ||
2821 | /* drop middle, data, if any */ | |
2822 | if (m->middle) { | |
2823 | ceph_buffer_put(m->middle); | |
2824 | m->middle = NULL; | |
31b8006e | 2825 | } |
c2e552e7 SW |
2826 | m->nr_pages = 0; |
2827 | m->pages = NULL; | |
2828 | ||
58bb3b37 SW |
2829 | if (m->pagelist) { |
2830 | ceph_pagelist_release(m->pagelist); | |
2831 | kfree(m->pagelist); | |
2832 | m->pagelist = NULL; | |
2833 | } | |
2834 | ||
68b4476b YS |
2835 | m->trail = NULL; |
2836 | ||
c2e552e7 SW |
2837 | if (m->pool) |
2838 | ceph_msgpool_put(m->pool, m); | |
2839 | else | |
2840 | ceph_msg_kfree(m); | |
31b8006e | 2841 | } |
3d14c5d2 | 2842 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2843 | |
2844 | void ceph_msg_dump(struct ceph_msg *msg) | |
2845 | { | |
2846 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2847 | msg->front_max, msg->nr_pages); | |
2848 | print_hex_dump(KERN_DEBUG, "header: ", | |
2849 | DUMP_PREFIX_OFFSET, 16, 1, | |
2850 | &msg->hdr, sizeof(msg->hdr), true); | |
2851 | print_hex_dump(KERN_DEBUG, " front: ", | |
2852 | DUMP_PREFIX_OFFSET, 16, 1, | |
2853 | msg->front.iov_base, msg->front.iov_len, true); | |
2854 | if (msg->middle) | |
2855 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2856 | DUMP_PREFIX_OFFSET, 16, 1, | |
2857 | msg->middle->vec.iov_base, | |
2858 | msg->middle->vec.iov_len, true); | |
2859 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2860 | DUMP_PREFIX_OFFSET, 16, 1, | |
2861 | &msg->footer, sizeof(msg->footer), true); | |
2862 | } | |
3d14c5d2 | 2863 | EXPORT_SYMBOL(ceph_msg_dump); |