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> | |
3ebc21f7 | 12 | #ifdef CONFIG_BLOCK |
68b4476b | 13 | #include <linux/bio.h> |
3ebc21f7 | 14 | #endif /* CONFIG_BLOCK */ |
ee3b56f2 | 15 | #include <linux/dns_resolver.h> |
31b8006e SW |
16 | #include <net/tcp.h> |
17 | ||
3d14c5d2 YS |
18 | #include <linux/ceph/libceph.h> |
19 | #include <linux/ceph/messenger.h> | |
20 | #include <linux/ceph/decode.h> | |
21 | #include <linux/ceph/pagelist.h> | |
bc3b2d7f | 22 | #include <linux/export.h> |
31b8006e SW |
23 | |
24 | /* | |
25 | * Ceph uses the messenger to exchange ceph_msg messages with other | |
26 | * hosts in the system. The messenger provides ordered and reliable | |
27 | * delivery. We tolerate TCP disconnects by reconnecting (with | |
28 | * exponential backoff) in the case of a fault (disconnection, bad | |
29 | * crc, protocol error). Acks allow sent messages to be discarded by | |
30 | * the sender. | |
31 | */ | |
32 | ||
bc18f4b1 AE |
33 | /* |
34 | * We track the state of the socket on a given connection using | |
35 | * values defined below. The transition to a new socket state is | |
36 | * handled by a function which verifies we aren't coming from an | |
37 | * unexpected state. | |
38 | * | |
39 | * -------- | |
40 | * | NEW* | transient initial state | |
41 | * -------- | |
42 | * | con_sock_state_init() | |
43 | * v | |
44 | * ---------- | |
45 | * | CLOSED | initialized, but no socket (and no | |
46 | * ---------- TCP connection) | |
47 | * ^ \ | |
48 | * | \ con_sock_state_connecting() | |
49 | * | ---------------------- | |
50 | * | \ | |
51 | * + con_sock_state_closed() \ | |
fbb85a47 SW |
52 | * |+--------------------------- \ |
53 | * | \ \ \ | |
54 | * | ----------- \ \ | |
55 | * | | CLOSING | socket event; \ \ | |
56 | * | ----------- await close \ \ | |
57 | * | ^ \ | | |
58 | * | | \ | | |
59 | * | + con_sock_state_closing() \ | | |
60 | * | / \ | | | |
61 | * | / --------------- | | | |
62 | * | / \ v v | |
bc18f4b1 AE |
63 | * | / -------------- |
64 | * | / -----------------| CONNECTING | socket created, TCP | |
65 | * | | / -------------- connect initiated | |
66 | * | | | con_sock_state_connected() | |
67 | * | | v | |
68 | * ------------- | |
69 | * | CONNECTED | TCP connection established | |
70 | * ------------- | |
71 | * | |
72 | * State values for ceph_connection->sock_state; NEW is assumed to be 0. | |
73 | */ | |
ce2c8903 AE |
74 | |
75 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ | |
76 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ | |
77 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ | |
78 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ | |
79 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ | |
80 | ||
8dacc7da SW |
81 | /* |
82 | * connection states | |
83 | */ | |
84 | #define CON_STATE_CLOSED 1 /* -> PREOPEN */ | |
85 | #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */ | |
86 | #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */ | |
87 | #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */ | |
88 | #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */ | |
89 | #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */ | |
90 | ||
4a861692 SW |
91 | /* |
92 | * ceph_connection flag bits | |
93 | */ | |
94 | #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop | |
95 | * messages on errors */ | |
96 | #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */ | |
97 | #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */ | |
98 | #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */ | |
99 | #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */ | |
8dacc7da | 100 | |
c9ffc77a AE |
101 | static bool con_flag_valid(unsigned long con_flag) |
102 | { | |
103 | switch (con_flag) { | |
104 | case CON_FLAG_LOSSYTX: | |
105 | case CON_FLAG_KEEPALIVE_PENDING: | |
106 | case CON_FLAG_WRITE_PENDING: | |
107 | case CON_FLAG_SOCK_CLOSED: | |
108 | case CON_FLAG_BACKOFF: | |
109 | return true; | |
110 | default: | |
111 | return false; | |
112 | } | |
113 | } | |
114 | ||
115 | static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag) | |
116 | { | |
117 | BUG_ON(!con_flag_valid(con_flag)); | |
118 | ||
119 | clear_bit(con_flag, &con->flags); | |
120 | } | |
121 | ||
122 | static void con_flag_set(struct ceph_connection *con, unsigned long con_flag) | |
123 | { | |
124 | BUG_ON(!con_flag_valid(con_flag)); | |
125 | ||
126 | set_bit(con_flag, &con->flags); | |
127 | } | |
128 | ||
129 | static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag) | |
130 | { | |
131 | BUG_ON(!con_flag_valid(con_flag)); | |
132 | ||
133 | return test_bit(con_flag, &con->flags); | |
134 | } | |
135 | ||
136 | static bool con_flag_test_and_clear(struct ceph_connection *con, | |
137 | unsigned long con_flag) | |
138 | { | |
139 | BUG_ON(!con_flag_valid(con_flag)); | |
140 | ||
141 | return test_and_clear_bit(con_flag, &con->flags); | |
142 | } | |
143 | ||
144 | static bool con_flag_test_and_set(struct ceph_connection *con, | |
145 | unsigned long con_flag) | |
146 | { | |
147 | BUG_ON(!con_flag_valid(con_flag)); | |
148 | ||
149 | return test_and_set_bit(con_flag, &con->flags); | |
150 | } | |
151 | ||
31b8006e SW |
152 | /* static tag bytes (protocol control messages) */ |
153 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
154 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
155 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
156 | ||
a6a5349d SW |
157 | #ifdef CONFIG_LOCKDEP |
158 | static struct lock_class_key socket_class; | |
159 | #endif | |
160 | ||
84495f49 AE |
161 | /* |
162 | * When skipping (ignoring) a block of input we read it into a "skip | |
163 | * buffer," which is this many bytes in size. | |
164 | */ | |
165 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
166 | |
167 | static void queue_con(struct ceph_connection *con); | |
168 | static void con_work(struct work_struct *); | |
93209264 | 169 | static void con_fault(struct ceph_connection *con); |
31b8006e | 170 | |
31b8006e | 171 | /* |
f64a9317 AE |
172 | * Nicely render a sockaddr as a string. An array of formatted |
173 | * strings is used, to approximate reentrancy. | |
31b8006e | 174 | */ |
f64a9317 AE |
175 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
176 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
177 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
178 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
179 | ||
180 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
181 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 182 | |
57666519 | 183 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 184 | |
3d14c5d2 | 185 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
186 | { |
187 | int i; | |
188 | char *s; | |
99f0f3b2 AE |
189 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
190 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 191 | |
f64a9317 | 192 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
193 | s = addr_str[i]; |
194 | ||
195 | switch (ss->ss_family) { | |
196 | case AF_INET: | |
bd406145 AE |
197 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
198 | ntohs(in4->sin_port)); | |
31b8006e SW |
199 | break; |
200 | ||
201 | case AF_INET6: | |
bd406145 AE |
202 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
203 | ntohs(in6->sin6_port)); | |
31b8006e SW |
204 | break; |
205 | ||
206 | default: | |
d3002b97 AE |
207 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
208 | ss->ss_family); | |
31b8006e SW |
209 | } |
210 | ||
211 | return s; | |
212 | } | |
3d14c5d2 | 213 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 214 | |
63f2d211 SW |
215 | static void encode_my_addr(struct ceph_messenger *msgr) |
216 | { | |
217 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
218 | ceph_encode_addr(&msgr->my_enc_addr); | |
219 | } | |
220 | ||
31b8006e SW |
221 | /* |
222 | * work queue for all reading and writing to/from the socket. | |
223 | */ | |
e0f43c94 | 224 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 225 | |
15417167 | 226 | static void _ceph_msgr_exit(void) |
6173d1f0 | 227 | { |
d3002b97 | 228 | if (ceph_msgr_wq) { |
6173d1f0 | 229 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
230 | ceph_msgr_wq = NULL; |
231 | } | |
6173d1f0 | 232 | |
6173d1f0 AE |
233 | BUG_ON(zero_page == NULL); |
234 | kunmap(zero_page); | |
235 | page_cache_release(zero_page); | |
236 | zero_page = NULL; | |
237 | } | |
238 | ||
3d14c5d2 | 239 | int ceph_msgr_init(void) |
31b8006e | 240 | { |
57666519 AE |
241 | BUG_ON(zero_page != NULL); |
242 | zero_page = ZERO_PAGE(0); | |
243 | page_cache_get(zero_page); | |
244 | ||
f363e45f | 245 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0); |
6173d1f0 AE |
246 | if (ceph_msgr_wq) |
247 | return 0; | |
57666519 | 248 | |
6173d1f0 AE |
249 | pr_err("msgr_init failed to create workqueue\n"); |
250 | _ceph_msgr_exit(); | |
57666519 | 251 | |
6173d1f0 | 252 | return -ENOMEM; |
31b8006e | 253 | } |
3d14c5d2 | 254 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
255 | |
256 | void ceph_msgr_exit(void) | |
257 | { | |
57666519 | 258 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 259 | |
6173d1f0 | 260 | _ceph_msgr_exit(); |
31b8006e | 261 | } |
3d14c5d2 | 262 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 263 | |
cd84db6e | 264 | void ceph_msgr_flush(void) |
a922d38f SW |
265 | { |
266 | flush_workqueue(ceph_msgr_wq); | |
267 | } | |
3d14c5d2 | 268 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 269 | |
ce2c8903 AE |
270 | /* Connection socket state transition functions */ |
271 | ||
272 | static void con_sock_state_init(struct ceph_connection *con) | |
273 | { | |
274 | int old_state; | |
275 | ||
276 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
277 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
278 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
279 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
280 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 AE |
281 | } |
282 | ||
283 | static void con_sock_state_connecting(struct ceph_connection *con) | |
284 | { | |
285 | int old_state; | |
286 | ||
287 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
288 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
289 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
290 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
291 | CON_SOCK_STATE_CONNECTING); | |
ce2c8903 AE |
292 | } |
293 | ||
294 | static void con_sock_state_connected(struct ceph_connection *con) | |
295 | { | |
296 | int old_state; | |
297 | ||
298 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
299 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
300 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
301 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
302 | CON_SOCK_STATE_CONNECTED); | |
ce2c8903 AE |
303 | } |
304 | ||
305 | static void con_sock_state_closing(struct ceph_connection *con) | |
306 | { | |
307 | int old_state; | |
308 | ||
309 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
310 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
311 | old_state != CON_SOCK_STATE_CONNECTED && | |
312 | old_state != CON_SOCK_STATE_CLOSING)) | |
313 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
314 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
315 | CON_SOCK_STATE_CLOSING); | |
ce2c8903 AE |
316 | } |
317 | ||
318 | static void con_sock_state_closed(struct ceph_connection *con) | |
319 | { | |
320 | int old_state; | |
321 | ||
322 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
323 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 | 324 | old_state != CON_SOCK_STATE_CLOSING && |
8007b8d6 SW |
325 | old_state != CON_SOCK_STATE_CONNECTING && |
326 | old_state != CON_SOCK_STATE_CLOSED)) | |
ce2c8903 | 327 | printk("%s: unexpected old state %d\n", __func__, old_state); |
8007b8d6 SW |
328 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
329 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 | 330 | } |
a922d38f | 331 | |
31b8006e SW |
332 | /* |
333 | * socket callback functions | |
334 | */ | |
335 | ||
336 | /* data available on socket, or listen socket received a connect */ | |
327800bd | 337 | static void ceph_sock_data_ready(struct sock *sk, int count_unused) |
31b8006e | 338 | { |
bd406145 | 339 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
340 | if (atomic_read(&con->msgr->stopping)) { |
341 | return; | |
342 | } | |
bd406145 | 343 | |
31b8006e | 344 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 345 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
346 | con, con->state); |
347 | queue_con(con); | |
348 | } | |
349 | } | |
350 | ||
351 | /* socket has buffer space for writing */ | |
327800bd | 352 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 353 | { |
d3002b97 | 354 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 355 | |
182fac26 JS |
356 | /* only queue to workqueue if there is data we want to write, |
357 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 358 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
359 | * doesn't get called again until try_write() fills the socket |
360 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
361 | * and net/core/stream.c:sk_stream_write_space(). | |
362 | */ | |
c9ffc77a | 363 | if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) { |
182fac26 | 364 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
327800bd | 365 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
366 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
367 | queue_con(con); | |
368 | } | |
31b8006e | 369 | } else { |
327800bd | 370 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 371 | } |
31b8006e SW |
372 | } |
373 | ||
374 | /* socket's state has changed */ | |
327800bd | 375 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 376 | { |
bd406145 | 377 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 378 | |
327800bd | 379 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
380 | con, con->state, sk->sk_state); |
381 | ||
31b8006e SW |
382 | switch (sk->sk_state) { |
383 | case TCP_CLOSE: | |
327800bd | 384 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 385 | case TCP_CLOSE_WAIT: |
327800bd | 386 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 387 | con_sock_state_closing(con); |
c9ffc77a | 388 | con_flag_set(con, CON_FLAG_SOCK_CLOSED); |
d65c9e0b | 389 | queue_con(con); |
31b8006e SW |
390 | break; |
391 | case TCP_ESTABLISHED: | |
327800bd | 392 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 393 | con_sock_state_connected(con); |
31b8006e SW |
394 | queue_con(con); |
395 | break; | |
d3002b97 AE |
396 | default: /* Everything else is uninteresting */ |
397 | break; | |
31b8006e SW |
398 | } |
399 | } | |
400 | ||
401 | /* | |
402 | * set up socket callbacks | |
403 | */ | |
404 | static void set_sock_callbacks(struct socket *sock, | |
405 | struct ceph_connection *con) | |
406 | { | |
407 | struct sock *sk = sock->sk; | |
bd406145 | 408 | sk->sk_user_data = con; |
327800bd AE |
409 | sk->sk_data_ready = ceph_sock_data_ready; |
410 | sk->sk_write_space = ceph_sock_write_space; | |
411 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
412 | } |
413 | ||
414 | ||
415 | /* | |
416 | * socket helpers | |
417 | */ | |
418 | ||
419 | /* | |
420 | * initiate connection to a remote socket. | |
421 | */ | |
41617d0c | 422 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 423 | { |
f91d3471 | 424 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
425 | struct socket *sock; |
426 | int ret; | |
427 | ||
428 | BUG_ON(con->sock); | |
f91d3471 SW |
429 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
430 | IPPROTO_TCP, &sock); | |
31b8006e | 431 | if (ret) |
41617d0c | 432 | return ret; |
31b8006e SW |
433 | sock->sk->sk_allocation = GFP_NOFS; |
434 | ||
a6a5349d SW |
435 | #ifdef CONFIG_LOCKDEP |
436 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
437 | #endif | |
438 | ||
31b8006e SW |
439 | set_sock_callbacks(sock, con); |
440 | ||
3d14c5d2 | 441 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 442 | |
89a86be0 | 443 | con_sock_state_connecting(con); |
f91d3471 SW |
444 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
445 | O_NONBLOCK); | |
31b8006e SW |
446 | if (ret == -EINPROGRESS) { |
447 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 448 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 449 | sock->sk->sk_state); |
a5bc3129 | 450 | } else if (ret < 0) { |
31b8006e | 451 | pr_err("connect %s error %d\n", |
3d14c5d2 | 452 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 453 | sock_release(sock); |
31b8006e | 454 | con->error_msg = "connect error"; |
31b8006e | 455 | |
41617d0c | 456 | return ret; |
a5bc3129 AE |
457 | } |
458 | con->sock = sock; | |
41617d0c | 459 | return 0; |
31b8006e SW |
460 | } |
461 | ||
462 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
463 | { | |
464 | struct kvec iov = {buf, len}; | |
465 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 466 | int r; |
31b8006e | 467 | |
98bdb0aa SW |
468 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
469 | if (r == -EAGAIN) | |
470 | r = 0; | |
471 | return r; | |
31b8006e SW |
472 | } |
473 | ||
474 | /* | |
475 | * write something. @more is true if caller will be sending more data | |
476 | * shortly. | |
477 | */ | |
478 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
479 | size_t kvlen, size_t len, int more) | |
480 | { | |
481 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 482 | int r; |
31b8006e SW |
483 | |
484 | if (more) | |
485 | msg.msg_flags |= MSG_MORE; | |
486 | else | |
487 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
488 | ||
42961d23 SW |
489 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
490 | if (r == -EAGAIN) | |
491 | r = 0; | |
492 | return r; | |
31b8006e SW |
493 | } |
494 | ||
31739139 | 495 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
e1dcb128 | 496 | int offset, size_t size, bool more) |
31739139 AE |
497 | { |
498 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
499 | int ret; | |
500 | ||
501 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
502 | if (ret == -EAGAIN) | |
503 | ret = 0; | |
504 | ||
505 | return ret; | |
506 | } | |
507 | ||
31b8006e SW |
508 | |
509 | /* | |
510 | * Shutdown/close the socket for the given connection. | |
511 | */ | |
512 | static int con_close_socket(struct ceph_connection *con) | |
513 | { | |
8007b8d6 | 514 | int rc = 0; |
31b8006e SW |
515 | |
516 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
8007b8d6 SW |
517 | if (con->sock) { |
518 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
519 | sock_release(con->sock); | |
520 | con->sock = NULL; | |
521 | } | |
456ea468 AE |
522 | |
523 | /* | |
4a861692 | 524 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
525 | * independent of the connection mutex, and we could have |
526 | * received a socket close event before we had the chance to | |
527 | * shut the socket down. | |
528 | */ | |
c9ffc77a | 529 | con_flag_clear(con, CON_FLAG_SOCK_CLOSED); |
8007b8d6 | 530 | |
ce2c8903 | 531 | con_sock_state_closed(con); |
31b8006e SW |
532 | return rc; |
533 | } | |
534 | ||
535 | /* | |
536 | * Reset a connection. Discard all incoming and outgoing messages | |
537 | * and clear *_seq state. | |
538 | */ | |
539 | static void ceph_msg_remove(struct ceph_msg *msg) | |
540 | { | |
541 | list_del_init(&msg->list_head); | |
38941f80 | 542 | BUG_ON(msg->con == NULL); |
36eb71aa | 543 | msg->con->ops->put(msg->con); |
38941f80 AE |
544 | msg->con = NULL; |
545 | ||
31b8006e SW |
546 | ceph_msg_put(msg); |
547 | } | |
548 | static void ceph_msg_remove_list(struct list_head *head) | |
549 | { | |
550 | while (!list_empty(head)) { | |
551 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
552 | list_head); | |
553 | ceph_msg_remove(msg); | |
554 | } | |
555 | } | |
556 | ||
557 | static void reset_connection(struct ceph_connection *con) | |
558 | { | |
559 | /* reset connection, out_queue, msg_ and connect_seq */ | |
560 | /* discard existing out_queue and msg_seq */ | |
0fa6ebc6 | 561 | dout("reset_connection %p\n", con); |
31b8006e SW |
562 | ceph_msg_remove_list(&con->out_queue); |
563 | ceph_msg_remove_list(&con->out_sent); | |
564 | ||
cf3e5c40 | 565 | if (con->in_msg) { |
38941f80 AE |
566 | BUG_ON(con->in_msg->con != con); |
567 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
568 | ceph_msg_put(con->in_msg); |
569 | con->in_msg = NULL; | |
36eb71aa | 570 | con->ops->put(con); |
cf3e5c40 SW |
571 | } |
572 | ||
31b8006e SW |
573 | con->connect_seq = 0; |
574 | con->out_seq = 0; | |
c86a2930 SW |
575 | if (con->out_msg) { |
576 | ceph_msg_put(con->out_msg); | |
577 | con->out_msg = NULL; | |
578 | } | |
31b8006e | 579 | con->in_seq = 0; |
0e0d5e0c | 580 | con->in_seq_acked = 0; |
31b8006e SW |
581 | } |
582 | ||
583 | /* | |
584 | * mark a peer down. drop any open connections. | |
585 | */ | |
586 | void ceph_con_close(struct ceph_connection *con) | |
587 | { | |
8c50c817 | 588 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
589 | dout("con_close %p peer %s\n", con, |
590 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
8dacc7da | 591 | con->state = CON_STATE_CLOSED; |
a5988c49 | 592 | |
c9ffc77a AE |
593 | con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */ |
594 | con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING); | |
595 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); | |
596 | con_flag_clear(con, CON_FLAG_BACKOFF); | |
a5988c49 | 597 | |
31b8006e | 598 | reset_connection(con); |
6f2bc3ff | 599 | con->peer_global_seq = 0; |
91e45ce3 | 600 | cancel_delayed_work(&con->work); |
ee76e073 | 601 | con_close_socket(con); |
ec302645 | 602 | mutex_unlock(&con->mutex); |
31b8006e | 603 | } |
3d14c5d2 | 604 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 605 | |
31b8006e SW |
606 | /* |
607 | * Reopen a closed connection, with a new peer address. | |
608 | */ | |
b7a9e5dd SW |
609 | void ceph_con_open(struct ceph_connection *con, |
610 | __u8 entity_type, __u64 entity_num, | |
611 | struct ceph_entity_addr *addr) | |
31b8006e | 612 | { |
5469155f | 613 | mutex_lock(&con->mutex); |
3d14c5d2 | 614 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
8dacc7da | 615 | |
122070a2 | 616 | WARN_ON(con->state != CON_STATE_CLOSED); |
8dacc7da | 617 | con->state = CON_STATE_PREOPEN; |
a5988c49 | 618 | |
b7a9e5dd SW |
619 | con->peer_name.type = (__u8) entity_type; |
620 | con->peer_name.num = cpu_to_le64(entity_num); | |
621 | ||
31b8006e | 622 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 623 | con->delay = 0; /* reset backoff memory */ |
5469155f | 624 | mutex_unlock(&con->mutex); |
31b8006e SW |
625 | queue_con(con); |
626 | } | |
3d14c5d2 | 627 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 628 | |
87b315a5 SW |
629 | /* |
630 | * return true if this connection ever successfully opened | |
631 | */ | |
632 | bool ceph_con_opened(struct ceph_connection *con) | |
633 | { | |
634 | return con->connect_seq > 0; | |
635 | } | |
636 | ||
31b8006e SW |
637 | /* |
638 | * initialize a new connection. | |
639 | */ | |
1bfd89f4 AE |
640 | void ceph_con_init(struct ceph_connection *con, void *private, |
641 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 642 | struct ceph_messenger *msgr) |
31b8006e SW |
643 | { |
644 | dout("con_init %p\n", con); | |
645 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
646 | con->private = private; |
647 | con->ops = ops; | |
31b8006e | 648 | con->msgr = msgr; |
ce2c8903 AE |
649 | |
650 | con_sock_state_init(con); | |
651 | ||
ec302645 | 652 | mutex_init(&con->mutex); |
31b8006e SW |
653 | INIT_LIST_HEAD(&con->out_queue); |
654 | INIT_LIST_HEAD(&con->out_sent); | |
655 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 | 656 | |
8dacc7da | 657 | con->state = CON_STATE_CLOSED; |
31b8006e | 658 | } |
3d14c5d2 | 659 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
660 | |
661 | ||
662 | /* | |
663 | * We maintain a global counter to order connection attempts. Get | |
664 | * a unique seq greater than @gt. | |
665 | */ | |
666 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
667 | { | |
668 | u32 ret; | |
669 | ||
670 | spin_lock(&msgr->global_seq_lock); | |
671 | if (msgr->global_seq < gt) | |
672 | msgr->global_seq = gt; | |
673 | ret = ++msgr->global_seq; | |
674 | spin_unlock(&msgr->global_seq_lock); | |
675 | return ret; | |
676 | } | |
677 | ||
e2200423 | 678 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
679 | { |
680 | con->out_kvec_left = 0; | |
681 | con->out_kvec_bytes = 0; | |
682 | con->out_kvec_cur = &con->out_kvec[0]; | |
683 | } | |
684 | ||
e2200423 | 685 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
686 | size_t size, void *data) |
687 | { | |
688 | int index; | |
689 | ||
690 | index = con->out_kvec_left; | |
691 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
692 | ||
693 | con->out_kvec[index].iov_len = size; | |
694 | con->out_kvec[index].iov_base = data; | |
695 | con->out_kvec_left++; | |
696 | con->out_kvec_bytes += size; | |
697 | } | |
31b8006e | 698 | |
df6ad1f9 | 699 | #ifdef CONFIG_BLOCK |
07c09b72 AE |
700 | static void init_bio_iter(struct bio *bio, struct bio **bio_iter, |
701 | unsigned int *bio_seg) | |
df6ad1f9 AE |
702 | { |
703 | if (!bio) { | |
07c09b72 AE |
704 | *bio_iter = NULL; |
705 | *bio_seg = 0; | |
df6ad1f9 AE |
706 | return; |
707 | } | |
07c09b72 AE |
708 | *bio_iter = bio; |
709 | *bio_seg = (unsigned int) bio->bi_idx; | |
df6ad1f9 AE |
710 | } |
711 | ||
07c09b72 | 712 | static void iter_bio_next(struct bio **bio_iter, unsigned int *seg) |
df6ad1f9 AE |
713 | { |
714 | if (*bio_iter == NULL) | |
715 | return; | |
716 | ||
717 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
718 | ||
719 | (*seg)++; | |
720 | if (*seg == (*bio_iter)->bi_vcnt) | |
721 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
722 | } | |
723 | #endif | |
724 | ||
739c905b AE |
725 | static void prepare_write_message_data(struct ceph_connection *con) |
726 | { | |
727 | struct ceph_msg *msg = con->out_msg; | |
728 | ||
729 | BUG_ON(!msg); | |
730 | BUG_ON(!msg->hdr.data_len); | |
731 | ||
732 | /* initialize page iterator */ | |
733 | con->out_msg_pos.page = 0; | |
734 | if (msg->pages) | |
735 | con->out_msg_pos.page_pos = msg->page_alignment; | |
736 | else | |
737 | con->out_msg_pos.page_pos = 0; | |
572c588e | 738 | #ifdef CONFIG_BLOCK |
abdaa6a8 | 739 | if (msg->bio) |
572c588e AE |
740 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); |
741 | #endif | |
739c905b AE |
742 | con->out_msg_pos.data_pos = 0; |
743 | con->out_msg_pos.did_page_crc = false; | |
744 | con->out_more = 1; /* data + footer will follow */ | |
745 | } | |
746 | ||
31b8006e SW |
747 | /* |
748 | * Prepare footer for currently outgoing message, and finish things | |
749 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
750 | */ | |
859eb799 | 751 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
752 | { |
753 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 754 | int v = con->out_kvec_left; |
31b8006e | 755 | |
fd154f3c AE |
756 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
757 | ||
31b8006e SW |
758 | dout("prepare_write_message_footer %p\n", con); |
759 | con->out_kvec_is_msg = true; | |
760 | con->out_kvec[v].iov_base = &m->footer; | |
761 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
762 | con->out_kvec_bytes += sizeof(m->footer); | |
763 | con->out_kvec_left++; | |
764 | con->out_more = m->more_to_follow; | |
c86a2930 | 765 | con->out_msg_done = true; |
31b8006e SW |
766 | } |
767 | ||
768 | /* | |
769 | * Prepare headers for the next outgoing message. | |
770 | */ | |
771 | static void prepare_write_message(struct ceph_connection *con) | |
772 | { | |
773 | struct ceph_msg *m; | |
a9a0c51a | 774 | u32 crc; |
31b8006e | 775 | |
e2200423 | 776 | con_out_kvec_reset(con); |
31b8006e | 777 | con->out_kvec_is_msg = true; |
c86a2930 | 778 | con->out_msg_done = false; |
31b8006e SW |
779 | |
780 | /* Sneak an ack in there first? If we can get it into the same | |
781 | * TCP packet that's a good thing. */ | |
782 | if (con->in_seq > con->in_seq_acked) { | |
783 | con->in_seq_acked = con->in_seq; | |
e2200423 | 784 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 785 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 786 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 787 | &con->out_temp_ack); |
31b8006e SW |
788 | } |
789 | ||
38941f80 | 790 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 791 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 792 | con->out_msg = m; |
38941f80 | 793 | BUG_ON(m->con != con); |
4cf9d544 SW |
794 | |
795 | /* put message on sent list */ | |
796 | ceph_msg_get(m); | |
797 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 798 | |
e84346b7 SW |
799 | /* |
800 | * only assign outgoing seq # if we haven't sent this message | |
801 | * yet. if it is requeued, resend with it's original seq. | |
802 | */ | |
803 | if (m->needs_out_seq) { | |
804 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
805 | m->needs_out_seq = false; | |
806 | } | |
b132cf4c YZ |
807 | #ifdef CONFIG_BLOCK |
808 | else | |
809 | m->bio_iter = NULL; | |
810 | #endif | |
31b8006e SW |
811 | |
812 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
813 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
814 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
815 | le32_to_cpu(m->hdr.data_len), | |
d4b515fa | 816 | m->page_count); |
31b8006e SW |
817 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); |
818 | ||
819 | /* tag + hdr + front + middle */ | |
e2200423 AE |
820 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
821 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
822 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 823 | |
31b8006e | 824 | if (m->middle) |
e2200423 | 825 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 826 | m->middle->vec.iov_base); |
31b8006e SW |
827 | |
828 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
829 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
830 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 831 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
832 | |
833 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
834 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
835 | if (m->middle) { | |
836 | crc = crc32c(0, m->middle->vec.iov_base, | |
837 | m->middle->vec.iov_len); | |
838 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
839 | } else | |
31b8006e | 840 | con->out_msg->footer.middle_crc = 0; |
739c905b | 841 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
842 | le32_to_cpu(con->out_msg->footer.front_crc), |
843 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
844 | ||
845 | /* is there a data payload? */ | |
739c905b AE |
846 | con->out_msg->footer.data_crc = 0; |
847 | if (m->hdr.data_len) | |
848 | prepare_write_message_data(con); | |
849 | else | |
31b8006e | 850 | /* no, queue up footer too and be done */ |
859eb799 | 851 | prepare_write_message_footer(con); |
31b8006e | 852 | |
c9ffc77a | 853 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
854 | } |
855 | ||
856 | /* | |
857 | * Prepare an ack. | |
858 | */ | |
859 | static void prepare_write_ack(struct ceph_connection *con) | |
860 | { | |
861 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
862 | con->in_seq_acked, con->in_seq); | |
863 | con->in_seq_acked = con->in_seq; | |
864 | ||
e2200423 | 865 | con_out_kvec_reset(con); |
859eb799 | 866 | |
e2200423 | 867 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 868 | |
31b8006e | 869 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 870 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
871 | &con->out_temp_ack); |
872 | ||
31b8006e | 873 | con->out_more = 1; /* more will follow.. eventually.. */ |
c9ffc77a | 874 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
875 | } |
876 | ||
877 | /* | |
878 | * Prepare to write keepalive byte. | |
879 | */ | |
880 | static void prepare_write_keepalive(struct ceph_connection *con) | |
881 | { | |
882 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
883 | con_out_kvec_reset(con); |
884 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
c9ffc77a | 885 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
886 | } |
887 | ||
888 | /* | |
889 | * Connection negotiation. | |
890 | */ | |
891 | ||
dac1e716 AE |
892 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
893 | int *auth_proto) | |
4e7a5dcd | 894 | { |
a3530df3 | 895 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
896 | |
897 | if (!con->ops->get_authorizer) { | |
898 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
899 | con->out_connect.authorizer_len = 0; | |
729796be | 900 | return NULL; |
b1c6b980 AE |
901 | } |
902 | ||
903 | /* Can't hold the mutex while getting authorizer */ | |
ec302645 | 904 | mutex_unlock(&con->mutex); |
dac1e716 | 905 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
ec302645 | 906 | mutex_lock(&con->mutex); |
4e7a5dcd | 907 | |
a3530df3 | 908 | if (IS_ERR(auth)) |
729796be | 909 | return auth; |
8dacc7da | 910 | if (con->state != CON_STATE_NEGOTIATING) |
729796be | 911 | return ERR_PTR(-EAGAIN); |
0da5d703 | 912 | |
8f43fb53 AE |
913 | con->auth_reply_buf = auth->authorizer_reply_buf; |
914 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
729796be | 915 | return auth; |
4e7a5dcd SW |
916 | } |
917 | ||
31b8006e SW |
918 | /* |
919 | * We connected to a peer and are saying hello. | |
920 | */ | |
e825a66d | 921 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 922 | { |
e2200423 AE |
923 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
924 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 925 | &con->msgr->my_enc_addr); |
eed0ef2c | 926 | |
eed0ef2c | 927 | con->out_more = 0; |
c9ffc77a | 928 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
eed0ef2c SW |
929 | } |
930 | ||
e825a66d | 931 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 932 | { |
95c96174 | 933 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 934 | int proto; |
dac1e716 | 935 | int auth_proto; |
729796be | 936 | struct ceph_auth_handshake *auth; |
31b8006e SW |
937 | |
938 | switch (con->peer_name.type) { | |
939 | case CEPH_ENTITY_TYPE_MON: | |
940 | proto = CEPH_MONC_PROTOCOL; | |
941 | break; | |
942 | case CEPH_ENTITY_TYPE_OSD: | |
943 | proto = CEPH_OSDC_PROTOCOL; | |
944 | break; | |
945 | case CEPH_ENTITY_TYPE_MDS: | |
946 | proto = CEPH_MDSC_PROTOCOL; | |
947 | break; | |
948 | default: | |
949 | BUG(); | |
950 | } | |
951 | ||
952 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
953 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 954 | |
e825a66d | 955 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
956 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
957 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
958 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
959 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
960 | con->out_connect.flags = 0; | |
31b8006e | 961 | |
dac1e716 AE |
962 | auth_proto = CEPH_AUTH_UNKNOWN; |
963 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
964 | if (IS_ERR(auth)) |
965 | return PTR_ERR(auth); | |
3da54776 | 966 | |
dac1e716 | 967 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
968 | con->out_connect.authorizer_len = auth ? |
969 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
970 | ||
e2200423 | 971 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
972 | &con->out_connect); |
973 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 974 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 975 | auth->authorizer_buf); |
859eb799 | 976 | |
31b8006e | 977 | con->out_more = 0; |
c9ffc77a | 978 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
4e7a5dcd | 979 | |
e10c758e | 980 | return 0; |
31b8006e SW |
981 | } |
982 | ||
31b8006e SW |
983 | /* |
984 | * write as much of pending kvecs to the socket as we can. | |
985 | * 1 -> done | |
986 | * 0 -> socket full, but more to do | |
987 | * <0 -> error | |
988 | */ | |
989 | static int write_partial_kvec(struct ceph_connection *con) | |
990 | { | |
991 | int ret; | |
992 | ||
993 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
994 | while (con->out_kvec_bytes > 0) { | |
995 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
996 | con->out_kvec_left, con->out_kvec_bytes, | |
997 | con->out_more); | |
998 | if (ret <= 0) | |
999 | goto out; | |
1000 | con->out_kvec_bytes -= ret; | |
1001 | if (con->out_kvec_bytes == 0) | |
1002 | break; /* done */ | |
f42299e6 AE |
1003 | |
1004 | /* account for full iov entries consumed */ | |
1005 | while (ret >= con->out_kvec_cur->iov_len) { | |
1006 | BUG_ON(!con->out_kvec_left); | |
1007 | ret -= con->out_kvec_cur->iov_len; | |
1008 | con->out_kvec_cur++; | |
1009 | con->out_kvec_left--; | |
1010 | } | |
1011 | /* and for a partially-consumed entry */ | |
1012 | if (ret) { | |
1013 | con->out_kvec_cur->iov_len -= ret; | |
1014 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
1015 | } |
1016 | } | |
1017 | con->out_kvec_left = 0; | |
1018 | con->out_kvec_is_msg = false; | |
1019 | ret = 1; | |
1020 | out: | |
1021 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
1022 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
1023 | return ret; /* done! */ | |
1024 | } | |
1025 | ||
84ca8fc8 AE |
1026 | static void out_msg_pos_next(struct ceph_connection *con, struct page *page, |
1027 | size_t len, size_t sent, bool in_trail) | |
68b4476b | 1028 | { |
84ca8fc8 | 1029 | struct ceph_msg *msg = con->out_msg; |
68b4476b | 1030 | |
84ca8fc8 AE |
1031 | BUG_ON(!msg); |
1032 | BUG_ON(!sent); | |
68b4476b | 1033 | |
84ca8fc8 AE |
1034 | con->out_msg_pos.data_pos += sent; |
1035 | con->out_msg_pos.page_pos += sent; | |
5821bd8c AE |
1036 | if (sent < len) |
1037 | return; | |
68b4476b | 1038 | |
5821bd8c AE |
1039 | BUG_ON(sent != len); |
1040 | con->out_msg_pos.page_pos = 0; | |
1041 | con->out_msg_pos.page++; | |
1042 | con->out_msg_pos.did_page_crc = false; | |
1043 | if (in_trail) | |
1044 | list_move_tail(&page->lru, | |
1045 | &msg->trail->head); | |
1046 | else if (msg->pagelist) | |
1047 | list_move_tail(&page->lru, | |
1048 | &msg->pagelist->head); | |
84ca8fc8 | 1049 | #ifdef CONFIG_BLOCK |
5821bd8c AE |
1050 | else if (msg->bio) |
1051 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
68b4476b | 1052 | #endif |
84ca8fc8 | 1053 | } |
68b4476b | 1054 | |
31b8006e SW |
1055 | /* |
1056 | * Write as much message data payload as we can. If we finish, queue | |
1057 | * up the footer. | |
1058 | * 1 -> done, footer is now queued in out_kvec[]. | |
1059 | * 0 -> socket full, but more to do | |
1060 | * <0 -> error | |
1061 | */ | |
1062 | static int write_partial_msg_pages(struct ceph_connection *con) | |
1063 | { | |
1064 | struct ceph_msg *msg = con->out_msg; | |
95c96174 | 1065 | unsigned int data_len = le32_to_cpu(msg->hdr.data_len); |
31b8006e | 1066 | size_t len; |
37675b0f | 1067 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 1068 | int ret; |
68b4476b | 1069 | int total_max_write; |
84ca8fc8 | 1070 | bool in_trail = false; |
5821bd8c AE |
1071 | const size_t trail_len = (msg->trail ? msg->trail->length : 0); |
1072 | const size_t trail_off = data_len - trail_len; | |
31b8006e SW |
1073 | |
1074 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
d4b515fa | 1075 | con, msg, con->out_msg_pos.page, msg->page_count, |
31b8006e SW |
1076 | con->out_msg_pos.page_pos); |
1077 | ||
5821bd8c AE |
1078 | /* |
1079 | * Iterate through each page that contains data to be | |
1080 | * written, and send as much as possible for each. | |
1081 | * | |
1082 | * If we are calculating the data crc (the default), we will | |
1083 | * need to map the page. If we have no pages, they have | |
1084 | * been revoked, so use the zero page. | |
1085 | */ | |
68b4476b | 1086 | while (data_len > con->out_msg_pos.data_pos) { |
31b8006e | 1087 | struct page *page = NULL; |
68b4476b | 1088 | int max_write = PAGE_SIZE; |
9bd19663 | 1089 | int bio_offset = 0; |
68b4476b | 1090 | |
5821bd8c AE |
1091 | in_trail = in_trail || con->out_msg_pos.data_pos >= trail_off; |
1092 | if (!in_trail) | |
1093 | total_max_write = trail_off - con->out_msg_pos.data_pos; | |
68b4476b | 1094 | |
5821bd8c | 1095 | if (in_trail) { |
68b4476b YS |
1096 | total_max_write = data_len - con->out_msg_pos.data_pos; |
1097 | ||
1098 | page = list_first_entry(&msg->trail->head, | |
1099 | struct page, lru); | |
68b4476b | 1100 | } else if (msg->pages) { |
31b8006e | 1101 | page = msg->pages[con->out_msg_pos.page]; |
58bb3b37 SW |
1102 | } else if (msg->pagelist) { |
1103 | page = list_first_entry(&msg->pagelist->head, | |
1104 | struct page, lru); | |
68b4476b YS |
1105 | #ifdef CONFIG_BLOCK |
1106 | } else if (msg->bio) { | |
1107 | struct bio_vec *bv; | |
1108 | ||
1109 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
1110 | page = bv->bv_page; | |
9bd19663 | 1111 | bio_offset = bv->bv_offset; |
68b4476b YS |
1112 | max_write = bv->bv_len; |
1113 | #endif | |
31b8006e | 1114 | } else { |
57666519 | 1115 | page = zero_page; |
31b8006e | 1116 | } |
68b4476b YS |
1117 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
1118 | total_max_write); | |
1119 | ||
37675b0f | 1120 | if (do_datacrc && !con->out_msg_pos.did_page_crc) { |
9bd19663 | 1121 | void *base; |
5821bd8c | 1122 | u32 crc = le32_to_cpu(msg->footer.data_crc); |
8d63e318 | 1123 | char *kaddr; |
31b8006e | 1124 | |
8d63e318 | 1125 | kaddr = kmap(page); |
31b8006e | 1126 | BUG_ON(kaddr == NULL); |
9bd19663 | 1127 | base = kaddr + con->out_msg_pos.page_pos + bio_offset; |
5821bd8c | 1128 | crc = crc32c(crc, base, len); |
5ce765a5 | 1129 | kunmap(page); |
84ca8fc8 | 1130 | msg->footer.data_crc = cpu_to_le32(crc); |
bca064d2 | 1131 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 1132 | } |
e36b13cc | 1133 | ret = ceph_tcp_sendpage(con->sock, page, |
9bd19663 | 1134 | con->out_msg_pos.page_pos + bio_offset, |
e1dcb128 | 1135 | len, true); |
31b8006e SW |
1136 | if (ret <= 0) |
1137 | goto out; | |
1138 | ||
84ca8fc8 | 1139 | out_msg_pos_next(con, page, len, (size_t) ret, in_trail); |
31b8006e SW |
1140 | } |
1141 | ||
1142 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
1143 | ||
1144 | /* prepare and queue up footer, too */ | |
37675b0f | 1145 | if (!do_datacrc) |
84ca8fc8 | 1146 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1147 | con_out_kvec_reset(con); |
859eb799 | 1148 | prepare_write_message_footer(con); |
31b8006e SW |
1149 | ret = 1; |
1150 | out: | |
1151 | return ret; | |
1152 | } | |
1153 | ||
1154 | /* | |
1155 | * write some zeros | |
1156 | */ | |
1157 | static int write_partial_skip(struct ceph_connection *con) | |
1158 | { | |
1159 | int ret; | |
1160 | ||
1161 | while (con->out_skip > 0) { | |
31739139 | 1162 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1163 | |
e1dcb128 | 1164 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true); |
31b8006e SW |
1165 | if (ret <= 0) |
1166 | goto out; | |
1167 | con->out_skip -= ret; | |
1168 | } | |
1169 | ret = 1; | |
1170 | out: | |
1171 | return ret; | |
1172 | } | |
1173 | ||
1174 | /* | |
1175 | * Prepare to read connection handshake, or an ack. | |
1176 | */ | |
eed0ef2c SW |
1177 | static void prepare_read_banner(struct ceph_connection *con) |
1178 | { | |
1179 | dout("prepare_read_banner %p\n", con); | |
1180 | con->in_base_pos = 0; | |
1181 | } | |
1182 | ||
31b8006e SW |
1183 | static void prepare_read_connect(struct ceph_connection *con) |
1184 | { | |
1185 | dout("prepare_read_connect %p\n", con); | |
1186 | con->in_base_pos = 0; | |
1187 | } | |
1188 | ||
1189 | static void prepare_read_ack(struct ceph_connection *con) | |
1190 | { | |
1191 | dout("prepare_read_ack %p\n", con); | |
1192 | con->in_base_pos = 0; | |
1193 | } | |
1194 | ||
1195 | static void prepare_read_tag(struct ceph_connection *con) | |
1196 | { | |
1197 | dout("prepare_read_tag %p\n", con); | |
1198 | con->in_base_pos = 0; | |
1199 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1200 | } | |
1201 | ||
1202 | /* | |
1203 | * Prepare to read a message. | |
1204 | */ | |
1205 | static int prepare_read_message(struct ceph_connection *con) | |
1206 | { | |
1207 | dout("prepare_read_message %p\n", con); | |
1208 | BUG_ON(con->in_msg != NULL); | |
1209 | con->in_base_pos = 0; | |
1210 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1211 | return 0; | |
1212 | } | |
1213 | ||
1214 | ||
1215 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1216 | int end, int size, void *object) |
31b8006e | 1217 | { |
e6cee71f AE |
1218 | while (con->in_base_pos < end) { |
1219 | int left = end - con->in_base_pos; | |
31b8006e SW |
1220 | int have = size - left; |
1221 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1222 | if (ret <= 0) | |
1223 | return ret; | |
1224 | con->in_base_pos += ret; | |
1225 | } | |
1226 | return 1; | |
1227 | } | |
1228 | ||
1229 | ||
1230 | /* | |
1231 | * Read all or part of the connect-side handshake on a new connection | |
1232 | */ | |
eed0ef2c | 1233 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1234 | { |
fd51653f AE |
1235 | int size; |
1236 | int end; | |
1237 | int ret; | |
31b8006e | 1238 | |
eed0ef2c | 1239 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1240 | |
1241 | /* peer's banner */ | |
fd51653f AE |
1242 | size = strlen(CEPH_BANNER); |
1243 | end = size; | |
1244 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1245 | if (ret <= 0) |
1246 | goto out; | |
fd51653f AE |
1247 | |
1248 | size = sizeof (con->actual_peer_addr); | |
1249 | end += size; | |
1250 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1251 | if (ret <= 0) |
1252 | goto out; | |
fd51653f AE |
1253 | |
1254 | size = sizeof (con->peer_addr_for_me); | |
1255 | end += size; | |
1256 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1257 | if (ret <= 0) |
1258 | goto out; | |
fd51653f | 1259 | |
eed0ef2c SW |
1260 | out: |
1261 | return ret; | |
1262 | } | |
1263 | ||
1264 | static int read_partial_connect(struct ceph_connection *con) | |
1265 | { | |
fd51653f AE |
1266 | int size; |
1267 | int end; | |
1268 | int ret; | |
eed0ef2c SW |
1269 | |
1270 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1271 | ||
fd51653f AE |
1272 | size = sizeof (con->in_reply); |
1273 | end = size; | |
1274 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1275 | if (ret <= 0) |
1276 | goto out; | |
fd51653f AE |
1277 | |
1278 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1279 | end += size; | |
1280 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1281 | if (ret <= 0) |
1282 | goto out; | |
31b8006e | 1283 | |
4e7a5dcd SW |
1284 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1285 | con, (int)con->in_reply.tag, | |
1286 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1287 | le32_to_cpu(con->in_reply.global_seq)); |
1288 | out: | |
1289 | return ret; | |
eed0ef2c | 1290 | |
31b8006e SW |
1291 | } |
1292 | ||
1293 | /* | |
1294 | * Verify the hello banner looks okay. | |
1295 | */ | |
1296 | static int verify_hello(struct ceph_connection *con) | |
1297 | { | |
1298 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1299 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1300 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1301 | con->error_msg = "protocol error, bad banner"; |
1302 | return -1; | |
1303 | } | |
1304 | return 0; | |
1305 | } | |
1306 | ||
1307 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1308 | { | |
1309 | switch (ss->ss_family) { | |
1310 | case AF_INET: | |
1311 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1312 | case AF_INET6: | |
1313 | return | |
1314 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1315 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1316 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1317 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1318 | } | |
1319 | return false; | |
1320 | } | |
1321 | ||
1322 | static int addr_port(struct sockaddr_storage *ss) | |
1323 | { | |
1324 | switch (ss->ss_family) { | |
1325 | case AF_INET: | |
f28bcfbe | 1326 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1327 | case AF_INET6: |
f28bcfbe | 1328 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1329 | } |
1330 | return 0; | |
1331 | } | |
1332 | ||
1333 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1334 | { | |
1335 | switch (ss->ss_family) { | |
1336 | case AF_INET: | |
1337 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1338 | break; |
31b8006e SW |
1339 | case AF_INET6: |
1340 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1341 | break; |
31b8006e SW |
1342 | } |
1343 | } | |
1344 | ||
ee3b56f2 NW |
1345 | /* |
1346 | * Unlike other *_pton function semantics, zero indicates success. | |
1347 | */ | |
1348 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1349 | char delim, const char **ipend) | |
1350 | { | |
99f0f3b2 AE |
1351 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1352 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1353 | |
1354 | memset(ss, 0, sizeof(*ss)); | |
1355 | ||
1356 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1357 | ss->ss_family = AF_INET; | |
1358 | return 0; | |
1359 | } | |
1360 | ||
1361 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1362 | ss->ss_family = AF_INET6; | |
1363 | return 0; | |
1364 | } | |
1365 | ||
1366 | return -EINVAL; | |
1367 | } | |
1368 | ||
1369 | /* | |
1370 | * Extract hostname string and resolve using kernel DNS facility. | |
1371 | */ | |
1372 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1373 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1374 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1375 | { | |
1376 | const char *end, *delim_p; | |
1377 | char *colon_p, *ip_addr = NULL; | |
1378 | int ip_len, ret; | |
1379 | ||
1380 | /* | |
1381 | * The end of the hostname occurs immediately preceding the delimiter or | |
1382 | * the port marker (':') where the delimiter takes precedence. | |
1383 | */ | |
1384 | delim_p = memchr(name, delim, namelen); | |
1385 | colon_p = memchr(name, ':', namelen); | |
1386 | ||
1387 | if (delim_p && colon_p) | |
1388 | end = delim_p < colon_p ? delim_p : colon_p; | |
1389 | else if (!delim_p && colon_p) | |
1390 | end = colon_p; | |
1391 | else { | |
1392 | end = delim_p; | |
1393 | if (!end) /* case: hostname:/ */ | |
1394 | end = name + namelen; | |
1395 | } | |
1396 | ||
1397 | if (end <= name) | |
1398 | return -EINVAL; | |
1399 | ||
1400 | /* do dns_resolve upcall */ | |
1401 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1402 | if (ip_len > 0) | |
1403 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1404 | else | |
1405 | ret = -ESRCH; | |
1406 | ||
1407 | kfree(ip_addr); | |
1408 | ||
1409 | *ipend = end; | |
1410 | ||
1411 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1412 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1413 | ||
1414 | return ret; | |
1415 | } | |
1416 | #else | |
1417 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1418 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1419 | { | |
1420 | return -EINVAL; | |
1421 | } | |
1422 | #endif | |
1423 | ||
1424 | /* | |
1425 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1426 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1427 | */ | |
1428 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1429 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1430 | { | |
1431 | int ret; | |
1432 | ||
1433 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1434 | if (ret) | |
1435 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1436 | ||
1437 | return ret; | |
1438 | } | |
1439 | ||
31b8006e SW |
1440 | /* |
1441 | * Parse an ip[:port] list into an addr array. Use the default | |
1442 | * monitor port if a port isn't specified. | |
1443 | */ | |
1444 | int ceph_parse_ips(const char *c, const char *end, | |
1445 | struct ceph_entity_addr *addr, | |
1446 | int max_count, int *count) | |
1447 | { | |
ee3b56f2 | 1448 | int i, ret = -EINVAL; |
31b8006e SW |
1449 | const char *p = c; |
1450 | ||
1451 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1452 | for (i = 0; i < max_count; i++) { | |
1453 | const char *ipend; | |
1454 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1455 | int port; |
39139f64 SW |
1456 | char delim = ','; |
1457 | ||
1458 | if (*p == '[') { | |
1459 | delim = ']'; | |
1460 | p++; | |
1461 | } | |
31b8006e | 1462 | |
ee3b56f2 NW |
1463 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1464 | if (ret) | |
31b8006e | 1465 | goto bad; |
ee3b56f2 NW |
1466 | ret = -EINVAL; |
1467 | ||
31b8006e SW |
1468 | p = ipend; |
1469 | ||
39139f64 SW |
1470 | if (delim == ']') { |
1471 | if (*p != ']') { | |
1472 | dout("missing matching ']'\n"); | |
1473 | goto bad; | |
1474 | } | |
1475 | p++; | |
1476 | } | |
1477 | ||
31b8006e SW |
1478 | /* port? */ |
1479 | if (p < end && *p == ':') { | |
1480 | port = 0; | |
1481 | p++; | |
1482 | while (p < end && *p >= '0' && *p <= '9') { | |
1483 | port = (port * 10) + (*p - '0'); | |
1484 | p++; | |
1485 | } | |
1486 | if (port > 65535 || port == 0) | |
1487 | goto bad; | |
1488 | } else { | |
1489 | port = CEPH_MON_PORT; | |
1490 | } | |
1491 | ||
1492 | addr_set_port(ss, port); | |
1493 | ||
3d14c5d2 | 1494 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1495 | |
1496 | if (p == end) | |
1497 | break; | |
1498 | if (*p != ',') | |
1499 | goto bad; | |
1500 | p++; | |
1501 | } | |
1502 | ||
1503 | if (p != end) | |
1504 | goto bad; | |
1505 | ||
1506 | if (count) | |
1507 | *count = i + 1; | |
1508 | return 0; | |
1509 | ||
1510 | bad: | |
39139f64 | 1511 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1512 | return ret; |
31b8006e | 1513 | } |
3d14c5d2 | 1514 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1515 | |
eed0ef2c | 1516 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1517 | { |
eed0ef2c | 1518 | dout("process_banner on %p\n", con); |
31b8006e SW |
1519 | |
1520 | if (verify_hello(con) < 0) | |
1521 | return -1; | |
1522 | ||
63f2d211 SW |
1523 | ceph_decode_addr(&con->actual_peer_addr); |
1524 | ceph_decode_addr(&con->peer_addr_for_me); | |
1525 | ||
31b8006e SW |
1526 | /* |
1527 | * Make sure the other end is who we wanted. note that the other | |
1528 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1529 | * them the benefit of the doubt. | |
1530 | */ | |
103e2d3a SW |
1531 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1532 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1533 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1534 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1535 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1536 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1537 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1538 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1539 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1540 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1541 | return -1; |
1542 | } | |
1543 | ||
1544 | /* | |
1545 | * did we learn our address? | |
1546 | */ | |
1547 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1548 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1549 | ||
1550 | memcpy(&con->msgr->inst.addr.in_addr, | |
1551 | &con->peer_addr_for_me.in_addr, | |
1552 | sizeof(con->peer_addr_for_me.in_addr)); | |
1553 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1554 | encode_my_addr(con->msgr); |
eed0ef2c | 1555 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1556 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1557 | } |
1558 | ||
eed0ef2c SW |
1559 | return 0; |
1560 | } | |
1561 | ||
1562 | static int process_connect(struct ceph_connection *con) | |
1563 | { | |
3d14c5d2 YS |
1564 | u64 sup_feat = con->msgr->supported_features; |
1565 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1566 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1567 | int ret; |
04a419f9 | 1568 | |
eed0ef2c SW |
1569 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1570 | ||
31b8006e | 1571 | switch (con->in_reply.tag) { |
04a419f9 SW |
1572 | case CEPH_MSGR_TAG_FEATURES: |
1573 | pr_err("%s%lld %s feature set mismatch," | |
1574 | " my %llx < server's %llx, missing %llx\n", | |
1575 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1576 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1577 | sup_feat, server_feat, server_feat & ~sup_feat); |
1578 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 1579 | reset_connection(con); |
04a419f9 SW |
1580 | return -1; |
1581 | ||
31b8006e | 1582 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1583 | pr_err("%s%lld %s protocol version mismatch," |
1584 | " my %d != server's %d\n", | |
1585 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1586 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1587 | le32_to_cpu(con->out_connect.protocol_version), |
1588 | le32_to_cpu(con->in_reply.protocol_version)); | |
1589 | con->error_msg = "protocol version mismatch"; | |
0fa6ebc6 | 1590 | reset_connection(con); |
31b8006e SW |
1591 | return -1; |
1592 | ||
4e7a5dcd SW |
1593 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1594 | con->auth_retry++; | |
1595 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1596 | con->auth_retry); | |
1597 | if (con->auth_retry == 2) { | |
1598 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1599 | return -1; |
1600 | } | |
1601 | con->auth_retry = 1; | |
6d4221b5 | 1602 | con_out_kvec_reset(con); |
e825a66d | 1603 | ret = prepare_write_connect(con); |
0da5d703 SW |
1604 | if (ret < 0) |
1605 | return ret; | |
63733a0f | 1606 | prepare_read_connect(con); |
4e7a5dcd | 1607 | break; |
31b8006e SW |
1608 | |
1609 | case CEPH_MSGR_TAG_RESETSESSION: | |
1610 | /* | |
1611 | * If we connected with a large connect_seq but the peer | |
1612 | * has no record of a session with us (no connection, or | |
1613 | * connect_seq == 0), they will send RESETSESION to indicate | |
1614 | * that they must have reset their session, and may have | |
1615 | * dropped messages. | |
1616 | */ | |
1617 | dout("process_connect got RESET peer seq %u\n", | |
5bdca4e0 | 1618 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
1619 | pr_err("%s%lld %s connection reset\n", |
1620 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1621 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1622 | reset_connection(con); |
6d4221b5 | 1623 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1624 | ret = prepare_write_connect(con); |
1625 | if (ret < 0) | |
1626 | return ret; | |
31b8006e SW |
1627 | prepare_read_connect(con); |
1628 | ||
1629 | /* Tell ceph about it. */ | |
ec302645 | 1630 | mutex_unlock(&con->mutex); |
31b8006e SW |
1631 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1632 | if (con->ops->peer_reset) | |
1633 | con->ops->peer_reset(con); | |
ec302645 | 1634 | mutex_lock(&con->mutex); |
8dacc7da | 1635 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 1636 | return -EAGAIN; |
31b8006e SW |
1637 | break; |
1638 | ||
1639 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1640 | /* | |
1641 | * If we sent a smaller connect_seq than the peer has, try | |
1642 | * again with a larger value. | |
1643 | */ | |
5bdca4e0 | 1644 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 1645 | le32_to_cpu(con->out_connect.connect_seq), |
5bdca4e0 SW |
1646 | le32_to_cpu(con->in_reply.connect_seq)); |
1647 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
6d4221b5 | 1648 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1649 | ret = prepare_write_connect(con); |
1650 | if (ret < 0) | |
1651 | return ret; | |
31b8006e SW |
1652 | prepare_read_connect(con); |
1653 | break; | |
1654 | ||
1655 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1656 | /* | |
1657 | * If we sent a smaller global_seq than the peer has, try | |
1658 | * again with a larger value. | |
1659 | */ | |
eed0ef2c | 1660 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 1661 | con->peer_global_seq, |
5bdca4e0 | 1662 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 1663 | get_global_seq(con->msgr, |
5bdca4e0 | 1664 | le32_to_cpu(con->in_reply.global_seq)); |
6d4221b5 | 1665 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1666 | ret = prepare_write_connect(con); |
1667 | if (ret < 0) | |
1668 | return ret; | |
31b8006e SW |
1669 | prepare_read_connect(con); |
1670 | break; | |
1671 | ||
1672 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1673 | if (req_feat & ~server_feat) { |
1674 | pr_err("%s%lld %s protocol feature mismatch," | |
1675 | " my required %llx > server's %llx, need %llx\n", | |
1676 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1677 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1678 | req_feat, server_feat, req_feat & ~server_feat); |
1679 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 1680 | reset_connection(con); |
04a419f9 SW |
1681 | return -1; |
1682 | } | |
8dacc7da | 1683 | |
122070a2 | 1684 | WARN_ON(con->state != CON_STATE_NEGOTIATING); |
8dacc7da SW |
1685 | con->state = CON_STATE_OPEN; |
1686 | ||
31b8006e SW |
1687 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1688 | con->connect_seq++; | |
aba558e2 | 1689 | con->peer_features = server_feat; |
31b8006e SW |
1690 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1691 | con->peer_global_seq, | |
1692 | le32_to_cpu(con->in_reply.connect_seq), | |
1693 | con->connect_seq); | |
1694 | WARN_ON(con->connect_seq != | |
1695 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1696 | |
1697 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
c9ffc77a | 1698 | con_flag_set(con, CON_FLAG_LOSSYTX); |
92ac41d0 | 1699 | |
85effe18 | 1700 | con->delay = 0; /* reset backoff memory */ |
92ac41d0 | 1701 | |
31b8006e SW |
1702 | prepare_read_tag(con); |
1703 | break; | |
1704 | ||
1705 | case CEPH_MSGR_TAG_WAIT: | |
1706 | /* | |
1707 | * If there is a connection race (we are opening | |
1708 | * connections to each other), one of us may just have | |
1709 | * to WAIT. This shouldn't happen if we are the | |
1710 | * client. | |
1711 | */ | |
04177882 SW |
1712 | pr_err("process_connect got WAIT as client\n"); |
1713 | con->error_msg = "protocol error, got WAIT as client"; | |
1714 | return -1; | |
31b8006e SW |
1715 | |
1716 | default: | |
1717 | pr_err("connect protocol error, will retry\n"); | |
1718 | con->error_msg = "protocol error, garbage tag during connect"; | |
1719 | return -1; | |
1720 | } | |
1721 | return 0; | |
1722 | } | |
1723 | ||
1724 | ||
1725 | /* | |
1726 | * read (part of) an ack | |
1727 | */ | |
1728 | static int read_partial_ack(struct ceph_connection *con) | |
1729 | { | |
fd51653f AE |
1730 | int size = sizeof (con->in_temp_ack); |
1731 | int end = size; | |
31b8006e | 1732 | |
fd51653f | 1733 | return read_partial(con, end, size, &con->in_temp_ack); |
31b8006e SW |
1734 | } |
1735 | ||
1736 | ||
1737 | /* | |
1738 | * We can finally discard anything that's been acked. | |
1739 | */ | |
1740 | static void process_ack(struct ceph_connection *con) | |
1741 | { | |
1742 | struct ceph_msg *m; | |
1743 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1744 | u64 seq; | |
1745 | ||
31b8006e SW |
1746 | while (!list_empty(&con->out_sent)) { |
1747 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1748 | list_head); | |
1749 | seq = le64_to_cpu(m->hdr.seq); | |
1750 | if (seq > ack) | |
1751 | break; | |
1752 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1753 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1754 | m->ack_stamp = jiffies; |
31b8006e SW |
1755 | ceph_msg_remove(m); |
1756 | } | |
31b8006e SW |
1757 | prepare_read_tag(con); |
1758 | } | |
1759 | ||
1760 | ||
1761 | ||
1762 | ||
2450418c | 1763 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1764 | struct kvec *section, |
1765 | unsigned int sec_len, u32 *crc) | |
2450418c | 1766 | { |
68b4476b | 1767 | int ret, left; |
2450418c YS |
1768 | |
1769 | BUG_ON(!section); | |
1770 | ||
1771 | while (section->iov_len < sec_len) { | |
1772 | BUG_ON(section->iov_base == NULL); | |
1773 | left = sec_len - section->iov_len; | |
1774 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1775 | section->iov_len, left); | |
1776 | if (ret <= 0) | |
1777 | return ret; | |
1778 | section->iov_len += ret; | |
2450418c | 1779 | } |
fe3ad593 AE |
1780 | if (section->iov_len == sec_len) |
1781 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1782 | |
2450418c YS |
1783 | return 1; |
1784 | } | |
31b8006e | 1785 | |
4740a623 | 1786 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
68b4476b YS |
1787 | |
1788 | static int read_partial_message_pages(struct ceph_connection *con, | |
1789 | struct page **pages, | |
95c96174 | 1790 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1791 | { |
1792 | void *p; | |
1793 | int ret; | |
1794 | int left; | |
1795 | ||
1796 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1797 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1798 | /* (page) data */ | |
1799 | BUG_ON(pages == NULL); | |
1800 | p = kmap(pages[con->in_msg_pos.page]); | |
1801 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1802 | left); | |
bca064d2 | 1803 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1804 | con->in_data_crc = |
1805 | crc32c(con->in_data_crc, | |
1806 | p + con->in_msg_pos.page_pos, ret); | |
1807 | kunmap(pages[con->in_msg_pos.page]); | |
1808 | if (ret <= 0) | |
1809 | return ret; | |
1810 | con->in_msg_pos.data_pos += ret; | |
1811 | con->in_msg_pos.page_pos += ret; | |
1812 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1813 | con->in_msg_pos.page_pos = 0; | |
1814 | con->in_msg_pos.page++; | |
1815 | } | |
1816 | ||
1817 | return ret; | |
1818 | } | |
1819 | ||
1820 | #ifdef CONFIG_BLOCK | |
1821 | static int read_partial_message_bio(struct ceph_connection *con, | |
07c09b72 AE |
1822 | struct bio **bio_iter, |
1823 | unsigned int *bio_seg, | |
95c96174 | 1824 | unsigned int data_len, bool do_datacrc) |
68b4476b YS |
1825 | { |
1826 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1827 | void *p; | |
1828 | int ret, left; | |
1829 | ||
68b4476b YS |
1830 | left = min((int)(data_len - con->in_msg_pos.data_pos), |
1831 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1832 | ||
1833 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1834 | ||
1835 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1836 | left); | |
bca064d2 | 1837 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1838 | con->in_data_crc = |
1839 | crc32c(con->in_data_crc, | |
1840 | p + con->in_msg_pos.page_pos, ret); | |
1841 | kunmap(bv->bv_page); | |
1842 | if (ret <= 0) | |
1843 | return ret; | |
1844 | con->in_msg_pos.data_pos += ret; | |
1845 | con->in_msg_pos.page_pos += ret; | |
1846 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1847 | con->in_msg_pos.page_pos = 0; | |
1848 | iter_bio_next(bio_iter, bio_seg); | |
1849 | } | |
1850 | ||
1851 | return ret; | |
1852 | } | |
1853 | #endif | |
1854 | ||
31b8006e SW |
1855 | /* |
1856 | * read (part of) a message. | |
1857 | */ | |
1858 | static int read_partial_message(struct ceph_connection *con) | |
1859 | { | |
1860 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1861 | int size; |
1862 | int end; | |
31b8006e | 1863 | int ret; |
95c96174 | 1864 | unsigned int front_len, middle_len, data_len; |
37675b0f | 1865 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 1866 | u64 seq; |
fe3ad593 | 1867 | u32 crc; |
31b8006e SW |
1868 | |
1869 | dout("read_partial_message con %p msg %p\n", con, m); | |
1870 | ||
1871 | /* header */ | |
fd51653f AE |
1872 | size = sizeof (con->in_hdr); |
1873 | end = size; | |
1874 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1875 | if (ret <= 0) |
1876 | return ret; | |
fe3ad593 AE |
1877 | |
1878 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1879 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1880 | pr_err("read_partial_message bad hdr " | |
1881 | " crc %u != expected %u\n", | |
1882 | crc, con->in_hdr.crc); | |
1883 | return -EBADMSG; | |
1884 | } | |
1885 | ||
31b8006e SW |
1886 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1887 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1888 | return -EIO; | |
1889 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
7b11ba37 | 1890 | if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN) |
31b8006e SW |
1891 | return -EIO; |
1892 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1893 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1894 | return -EIO; | |
1895 | ||
ae18756b SW |
1896 | /* verify seq# */ |
1897 | seq = le64_to_cpu(con->in_hdr.seq); | |
1898 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1899 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1900 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1901 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1902 | seq, con->in_seq + 1); |
1903 | con->in_base_pos = -front_len - middle_len - data_len - | |
1904 | sizeof(m->footer); | |
1905 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1906 | return 0; |
1907 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1908 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1909 | seq, con->in_seq + 1); | |
1910 | con->error_msg = "bad message sequence # for incoming message"; | |
1911 | return -EBADMSG; | |
1912 | } | |
1913 | ||
31b8006e SW |
1914 | /* allocate message? */ |
1915 | if (!con->in_msg) { | |
4740a623 SW |
1916 | int skip = 0; |
1917 | ||
31b8006e | 1918 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
6ebc8b32 | 1919 | front_len, data_len); |
4740a623 SW |
1920 | ret = ceph_con_in_msg_alloc(con, &skip); |
1921 | if (ret < 0) | |
1922 | return ret; | |
2450418c | 1923 | if (skip) { |
31b8006e | 1924 | /* skip this message */ |
a79832f2 | 1925 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1926 | BUG_ON(con->in_msg); |
31b8006e SW |
1927 | con->in_base_pos = -front_len - middle_len - data_len - |
1928 | sizeof(m->footer); | |
1929 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1930 | con->in_seq++; |
31b8006e SW |
1931 | return 0; |
1932 | } | |
38941f80 | 1933 | |
4740a623 | 1934 | BUG_ON(!con->in_msg); |
38941f80 | 1935 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
1936 | m = con->in_msg; |
1937 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1938 | if (m->middle) |
1939 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1940 | |
1941 | con->in_msg_pos.page = 0; | |
68b4476b | 1942 | if (m->pages) |
c5c6b19d | 1943 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1944 | else |
1945 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1946 | con->in_msg_pos.data_pos = 0; |
a4107026 SW |
1947 | |
1948 | #ifdef CONFIG_BLOCK | |
1949 | if (m->bio) | |
1950 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1951 | #endif | |
31b8006e SW |
1952 | } |
1953 | ||
1954 | /* front */ | |
2450418c YS |
1955 | ret = read_partial_message_section(con, &m->front, front_len, |
1956 | &con->in_front_crc); | |
1957 | if (ret <= 0) | |
1958 | return ret; | |
31b8006e SW |
1959 | |
1960 | /* middle */ | |
2450418c | 1961 | if (m->middle) { |
213c99ee SW |
1962 | ret = read_partial_message_section(con, &m->middle->vec, |
1963 | middle_len, | |
2450418c | 1964 | &con->in_middle_crc); |
31b8006e SW |
1965 | if (ret <= 0) |
1966 | return ret; | |
31b8006e SW |
1967 | } |
1968 | ||
1969 | /* (page) data */ | |
31b8006e | 1970 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1971 | if (m->pages) { |
1972 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1973 | data_len, do_datacrc); |
68b4476b YS |
1974 | if (ret <= 0) |
1975 | return ret; | |
1976 | #ifdef CONFIG_BLOCK | |
1977 | } else if (m->bio) { | |
a4107026 | 1978 | BUG_ON(!m->bio_iter); |
68b4476b YS |
1979 | ret = read_partial_message_bio(con, |
1980 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1981 | data_len, do_datacrc); |
68b4476b YS |
1982 | if (ret <= 0) |
1983 | return ret; | |
1984 | #endif | |
1985 | } else { | |
1986 | BUG_ON(1); | |
31b8006e SW |
1987 | } |
1988 | } | |
1989 | ||
31b8006e | 1990 | /* footer */ |
fd51653f AE |
1991 | size = sizeof (m->footer); |
1992 | end += size; | |
1993 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
1994 | if (ret <= 0) |
1995 | return ret; | |
1996 | ||
31b8006e SW |
1997 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
1998 | m, front_len, m->footer.front_crc, middle_len, | |
1999 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
2000 | ||
2001 | /* crc ok? */ | |
2002 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
2003 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
2004 | m, con->in_front_crc, m->footer.front_crc); | |
2005 | return -EBADMSG; | |
2006 | } | |
2007 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
2008 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
2009 | m, con->in_middle_crc, m->footer.middle_crc); | |
2010 | return -EBADMSG; | |
2011 | } | |
bca064d2 | 2012 | if (do_datacrc && |
31b8006e SW |
2013 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
2014 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
2015 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
2016 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
2017 | return -EBADMSG; | |
2018 | } | |
2019 | ||
2020 | return 1; /* done! */ | |
2021 | } | |
2022 | ||
2023 | /* | |
2024 | * Process message. This happens in the worker thread. The callback should | |
2025 | * be careful not to do anything that waits on other incoming messages or it | |
2026 | * may deadlock. | |
2027 | */ | |
2028 | static void process_message(struct ceph_connection *con) | |
2029 | { | |
5e095e8b | 2030 | struct ceph_msg *msg; |
31b8006e | 2031 | |
38941f80 AE |
2032 | BUG_ON(con->in_msg->con != con); |
2033 | con->in_msg->con = NULL; | |
5e095e8b | 2034 | msg = con->in_msg; |
31b8006e | 2035 | con->in_msg = NULL; |
36eb71aa | 2036 | con->ops->put(con); |
31b8006e SW |
2037 | |
2038 | /* if first message, set peer_name */ | |
2039 | if (con->peer_name.type == 0) | |
dbad185d | 2040 | con->peer_name = msg->hdr.src; |
31b8006e | 2041 | |
31b8006e | 2042 | con->in_seq++; |
ec302645 | 2043 | mutex_unlock(&con->mutex); |
31b8006e SW |
2044 | |
2045 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
2046 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 2047 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
2048 | le16_to_cpu(msg->hdr.type), |
2049 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2050 | le32_to_cpu(msg->hdr.front_len), | |
2051 | le32_to_cpu(msg->hdr.data_len), | |
2052 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
2053 | con->ops->dispatch(con, msg); | |
ec302645 SW |
2054 | |
2055 | mutex_lock(&con->mutex); | |
31b8006e SW |
2056 | } |
2057 | ||
2058 | ||
2059 | /* | |
2060 | * Write something to the socket. Called in a worker thread when the | |
2061 | * socket appears to be writeable and we have something ready to send. | |
2062 | */ | |
2063 | static int try_write(struct ceph_connection *con) | |
2064 | { | |
31b8006e SW |
2065 | int ret = 1; |
2066 | ||
d59315ca | 2067 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 2068 | |
31b8006e SW |
2069 | more: |
2070 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
2071 | ||
2072 | /* open the socket first? */ | |
8dacc7da SW |
2073 | if (con->state == CON_STATE_PREOPEN) { |
2074 | BUG_ON(con->sock); | |
2075 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2076 | |
e2200423 | 2077 | con_out_kvec_reset(con); |
e825a66d | 2078 | prepare_write_banner(con); |
eed0ef2c | 2079 | prepare_read_banner(con); |
31b8006e | 2080 | |
cf3e5c40 | 2081 | BUG_ON(con->in_msg); |
31b8006e SW |
2082 | con->in_tag = CEPH_MSGR_TAG_READY; |
2083 | dout("try_write initiating connect on %p new state %lu\n", | |
2084 | con, con->state); | |
41617d0c AE |
2085 | ret = ceph_tcp_connect(con); |
2086 | if (ret < 0) { | |
31b8006e | 2087 | con->error_msg = "connect error"; |
31b8006e SW |
2088 | goto out; |
2089 | } | |
2090 | } | |
2091 | ||
2092 | more_kvec: | |
2093 | /* kvec data queued? */ | |
2094 | if (con->out_skip) { | |
2095 | ret = write_partial_skip(con); | |
2096 | if (ret <= 0) | |
42961d23 | 2097 | goto out; |
31b8006e SW |
2098 | } |
2099 | if (con->out_kvec_left) { | |
2100 | ret = write_partial_kvec(con); | |
2101 | if (ret <= 0) | |
42961d23 | 2102 | goto out; |
31b8006e SW |
2103 | } |
2104 | ||
2105 | /* msg pages? */ | |
2106 | if (con->out_msg) { | |
c86a2930 SW |
2107 | if (con->out_msg_done) { |
2108 | ceph_msg_put(con->out_msg); | |
2109 | con->out_msg = NULL; /* we're done with this one */ | |
2110 | goto do_next; | |
2111 | } | |
2112 | ||
31b8006e SW |
2113 | ret = write_partial_msg_pages(con); |
2114 | if (ret == 1) | |
2115 | goto more_kvec; /* we need to send the footer, too! */ | |
2116 | if (ret == 0) | |
42961d23 | 2117 | goto out; |
31b8006e SW |
2118 | if (ret < 0) { |
2119 | dout("try_write write_partial_msg_pages err %d\n", | |
2120 | ret); | |
42961d23 | 2121 | goto out; |
31b8006e SW |
2122 | } |
2123 | } | |
2124 | ||
c86a2930 | 2125 | do_next: |
8dacc7da | 2126 | if (con->state == CON_STATE_OPEN) { |
31b8006e SW |
2127 | /* is anything else pending? */ |
2128 | if (!list_empty(&con->out_queue)) { | |
2129 | prepare_write_message(con); | |
2130 | goto more; | |
2131 | } | |
2132 | if (con->in_seq > con->in_seq_acked) { | |
2133 | prepare_write_ack(con); | |
2134 | goto more; | |
2135 | } | |
c9ffc77a | 2136 | if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) { |
31b8006e SW |
2137 | prepare_write_keepalive(con); |
2138 | goto more; | |
2139 | } | |
2140 | } | |
2141 | ||
2142 | /* Nothing to do! */ | |
c9ffc77a | 2143 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
31b8006e | 2144 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2145 | ret = 0; |
2146 | out: | |
42961d23 | 2147 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2148 | return ret; |
2149 | } | |
2150 | ||
2151 | ||
2152 | ||
2153 | /* | |
2154 | * Read what we can from the socket. | |
2155 | */ | |
2156 | static int try_read(struct ceph_connection *con) | |
2157 | { | |
31b8006e SW |
2158 | int ret = -1; |
2159 | ||
8dacc7da SW |
2160 | more: |
2161 | dout("try_read start on %p state %lu\n", con, con->state); | |
2162 | if (con->state != CON_STATE_CONNECTING && | |
2163 | con->state != CON_STATE_NEGOTIATING && | |
2164 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2165 | return 0; |
2166 | ||
8dacc7da | 2167 | BUG_ON(!con->sock); |
ec302645 | 2168 | |
31b8006e SW |
2169 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2170 | con->in_base_pos); | |
0da5d703 | 2171 | |
8dacc7da | 2172 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2173 | dout("try_read connecting\n"); |
2174 | ret = read_partial_banner(con); | |
2175 | if (ret <= 0) | |
ab166d5a | 2176 | goto out; |
7593af92 AE |
2177 | ret = process_banner(con); |
2178 | if (ret < 0) | |
2179 | goto out; | |
2180 | ||
8dacc7da | 2181 | con->state = CON_STATE_NEGOTIATING; |
7593af92 | 2182 | |
6d4221b5 JS |
2183 | /* |
2184 | * Received banner is good, exchange connection info. | |
2185 | * Do not reset out_kvec, as sending our banner raced | |
2186 | * with receiving peer banner after connect completed. | |
2187 | */ | |
7593af92 AE |
2188 | ret = prepare_write_connect(con); |
2189 | if (ret < 0) | |
2190 | goto out; | |
2191 | prepare_read_connect(con); | |
2192 | ||
2193 | /* Send connection info before awaiting response */ | |
0da5d703 SW |
2194 | goto out; |
2195 | } | |
2196 | ||
8dacc7da | 2197 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2198 | dout("try_read negotiating\n"); |
31b8006e SW |
2199 | ret = read_partial_connect(con); |
2200 | if (ret <= 0) | |
31b8006e | 2201 | goto out; |
98bdb0aa SW |
2202 | ret = process_connect(con); |
2203 | if (ret < 0) | |
2204 | goto out; | |
31b8006e SW |
2205 | goto more; |
2206 | } | |
2207 | ||
122070a2 | 2208 | WARN_ON(con->state != CON_STATE_OPEN); |
8dacc7da | 2209 | |
31b8006e SW |
2210 | if (con->in_base_pos < 0) { |
2211 | /* | |
2212 | * skipping + discarding content. | |
2213 | * | |
2214 | * FIXME: there must be a better way to do this! | |
2215 | */ | |
84495f49 AE |
2216 | static char buf[SKIP_BUF_SIZE]; |
2217 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2218 | ||
31b8006e SW |
2219 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2220 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2221 | if (ret <= 0) | |
98bdb0aa | 2222 | goto out; |
31b8006e SW |
2223 | con->in_base_pos += ret; |
2224 | if (con->in_base_pos) | |
2225 | goto more; | |
2226 | } | |
2227 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2228 | /* | |
2229 | * what's next? | |
2230 | */ | |
2231 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2232 | if (ret <= 0) | |
98bdb0aa | 2233 | goto out; |
31b8006e SW |
2234 | dout("try_read got tag %d\n", (int)con->in_tag); |
2235 | switch (con->in_tag) { | |
2236 | case CEPH_MSGR_TAG_MSG: | |
2237 | prepare_read_message(con); | |
2238 | break; | |
2239 | case CEPH_MSGR_TAG_ACK: | |
2240 | prepare_read_ack(con); | |
2241 | break; | |
2242 | case CEPH_MSGR_TAG_CLOSE: | |
8dacc7da SW |
2243 | con_close_socket(con); |
2244 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2245 | goto out; |
31b8006e SW |
2246 | default: |
2247 | goto bad_tag; | |
2248 | } | |
2249 | } | |
2250 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2251 | ret = read_partial_message(con); | |
2252 | if (ret <= 0) { | |
2253 | switch (ret) { | |
2254 | case -EBADMSG: | |
2255 | con->error_msg = "bad crc"; | |
2256 | ret = -EIO; | |
98bdb0aa | 2257 | break; |
31b8006e SW |
2258 | case -EIO: |
2259 | con->error_msg = "io error"; | |
98bdb0aa | 2260 | break; |
31b8006e | 2261 | } |
98bdb0aa | 2262 | goto out; |
31b8006e SW |
2263 | } |
2264 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2265 | goto more; | |
2266 | process_message(con); | |
7b862e07 SW |
2267 | if (con->state == CON_STATE_OPEN) |
2268 | prepare_read_tag(con); | |
31b8006e SW |
2269 | goto more; |
2270 | } | |
2271 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2272 | ret = read_partial_ack(con); | |
2273 | if (ret <= 0) | |
98bdb0aa | 2274 | goto out; |
31b8006e SW |
2275 | process_ack(con); |
2276 | goto more; | |
2277 | } | |
2278 | ||
31b8006e | 2279 | out: |
98bdb0aa | 2280 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2281 | return ret; |
2282 | ||
2283 | bad_tag: | |
2284 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2285 | con->error_msg = "protocol error, garbage tag"; | |
2286 | ret = -1; | |
2287 | goto out; | |
2288 | } | |
2289 | ||
2290 | ||
2291 | /* | |
802c6d96 AE |
2292 | * Atomically queue work on a connection after the specified delay. |
2293 | * Bump @con reference to avoid races with connection teardown. | |
2294 | * Returns 0 if work was queued, or an error code otherwise. | |
31b8006e | 2295 | */ |
802c6d96 | 2296 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
31b8006e | 2297 | { |
31b8006e | 2298 | if (!con->ops->get(con)) { |
802c6d96 AE |
2299 | dout("%s %p ref count 0\n", __func__, con); |
2300 | ||
2301 | return -ENOENT; | |
31b8006e SW |
2302 | } |
2303 | ||
802c6d96 AE |
2304 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
2305 | dout("%s %p - already queued\n", __func__, con); | |
31b8006e | 2306 | con->ops->put(con); |
802c6d96 AE |
2307 | |
2308 | return -EBUSY; | |
31b8006e | 2309 | } |
802c6d96 AE |
2310 | |
2311 | dout("%s %p %lu\n", __func__, con, delay); | |
2312 | ||
2313 | return 0; | |
2314 | } | |
2315 | ||
2316 | static void queue_con(struct ceph_connection *con) | |
2317 | { | |
2318 | (void) queue_con_delay(con, 0); | |
31b8006e SW |
2319 | } |
2320 | ||
7bb21d68 AE |
2321 | static bool con_sock_closed(struct ceph_connection *con) |
2322 | { | |
c9ffc77a | 2323 | if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED)) |
7bb21d68 AE |
2324 | return false; |
2325 | ||
2326 | #define CASE(x) \ | |
2327 | case CON_STATE_ ## x: \ | |
2328 | con->error_msg = "socket closed (con state " #x ")"; \ | |
2329 | break; | |
2330 | ||
2331 | switch (con->state) { | |
2332 | CASE(CLOSED); | |
2333 | CASE(PREOPEN); | |
2334 | CASE(CONNECTING); | |
2335 | CASE(NEGOTIATING); | |
2336 | CASE(OPEN); | |
2337 | CASE(STANDBY); | |
2338 | default: | |
2339 | pr_warning("%s con %p unrecognized state %lu\n", | |
2340 | __func__, con, con->state); | |
2341 | con->error_msg = "unrecognized con state"; | |
2342 | BUG(); | |
2343 | break; | |
2344 | } | |
2345 | #undef CASE | |
2346 | ||
2347 | return true; | |
2348 | } | |
2349 | ||
f20a39fd AE |
2350 | static bool con_backoff(struct ceph_connection *con) |
2351 | { | |
2352 | int ret; | |
2353 | ||
2354 | if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF)) | |
2355 | return false; | |
2356 | ||
2357 | ret = queue_con_delay(con, round_jiffies_relative(con->delay)); | |
2358 | if (ret) { | |
2359 | dout("%s: con %p FAILED to back off %lu\n", __func__, | |
2360 | con, con->delay); | |
2361 | BUG_ON(ret == -ENOENT); | |
2362 | con_flag_set(con, CON_FLAG_BACKOFF); | |
2363 | } | |
2364 | ||
2365 | return true; | |
2366 | } | |
2367 | ||
93209264 AE |
2368 | /* Finish fault handling; con->mutex must *not* be held here */ |
2369 | ||
2370 | static void con_fault_finish(struct ceph_connection *con) | |
2371 | { | |
2372 | /* | |
2373 | * in case we faulted due to authentication, invalidate our | |
2374 | * current tickets so that we can get new ones. | |
2375 | */ | |
2376 | if (con->auth_retry && con->ops->invalidate_authorizer) { | |
2377 | dout("calling invalidate_authorizer()\n"); | |
2378 | con->ops->invalidate_authorizer(con); | |
2379 | } | |
2380 | ||
2381 | if (con->ops->fault) | |
2382 | con->ops->fault(con); | |
2383 | } | |
2384 | ||
31b8006e SW |
2385 | /* |
2386 | * Do some work on a connection. Drop a connection ref when we're done. | |
2387 | */ | |
2388 | static void con_work(struct work_struct *work) | |
2389 | { | |
2390 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2391 | work.work); | |
49659416 | 2392 | bool fault; |
31b8006e | 2393 | |
9dd4658d | 2394 | mutex_lock(&con->mutex); |
49659416 AE |
2395 | while (true) { |
2396 | int ret; | |
31b8006e | 2397 | |
49659416 AE |
2398 | if ((fault = con_sock_closed(con))) { |
2399 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); | |
2400 | break; | |
2401 | } | |
2402 | if (con_backoff(con)) { | |
2403 | dout("%s: con %p BACKOFF\n", __func__, con); | |
2404 | break; | |
2405 | } | |
2406 | if (con->state == CON_STATE_STANDBY) { | |
2407 | dout("%s: con %p STANDBY\n", __func__, con); | |
2408 | break; | |
2409 | } | |
2410 | if (con->state == CON_STATE_CLOSED) { | |
2411 | dout("%s: con %p CLOSED\n", __func__, con); | |
2412 | BUG_ON(con->sock); | |
2413 | break; | |
2414 | } | |
2415 | if (con->state == CON_STATE_PREOPEN) { | |
2416 | dout("%s: con %p PREOPEN\n", __func__, con); | |
2417 | BUG_ON(con->sock); | |
2418 | } | |
0da5d703 | 2419 | |
49659416 AE |
2420 | ret = try_read(con); |
2421 | if (ret < 0) { | |
2422 | if (ret == -EAGAIN) | |
2423 | continue; | |
2424 | con->error_msg = "socket error on read"; | |
2425 | fault = true; | |
2426 | break; | |
2427 | } | |
2428 | ||
2429 | ret = try_write(con); | |
2430 | if (ret < 0) { | |
2431 | if (ret == -EAGAIN) | |
2432 | continue; | |
2433 | con->error_msg = "socket error on write"; | |
2434 | fault = true; | |
2435 | } | |
2436 | ||
2437 | break; /* If we make it to here, we're done */ | |
3a140a0d | 2438 | } |
b6e7b6a1 AE |
2439 | if (fault) |
2440 | con_fault(con); | |
9dd4658d | 2441 | mutex_unlock(&con->mutex); |
0da5d703 | 2442 | |
b6e7b6a1 AE |
2443 | if (fault) |
2444 | con_fault_finish(con); | |
2445 | ||
2446 | con->ops->put(con); | |
31b8006e SW |
2447 | } |
2448 | ||
31b8006e SW |
2449 | /* |
2450 | * Generic error/fault handler. A retry mechanism is used with | |
2451 | * exponential backoff | |
2452 | */ | |
93209264 | 2453 | static void con_fault(struct ceph_connection *con) |
31b8006e | 2454 | { |
28362986 | 2455 | pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2456 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2457 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2458 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2459 | |
122070a2 | 2460 | WARN_ON(con->state != CON_STATE_CONNECTING && |
8dacc7da SW |
2461 | con->state != CON_STATE_NEGOTIATING && |
2462 | con->state != CON_STATE_OPEN); | |
ec302645 | 2463 | |
31b8006e | 2464 | con_close_socket(con); |
5e095e8b | 2465 | |
c9ffc77a | 2466 | if (con_flag_test(con, CON_FLAG_LOSSYTX)) { |
8dacc7da SW |
2467 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
2468 | con->state = CON_STATE_CLOSED; | |
93209264 | 2469 | return; |
3b5ede07 SW |
2470 | } |
2471 | ||
5e095e8b | 2472 | if (con->in_msg) { |
38941f80 AE |
2473 | BUG_ON(con->in_msg->con != con); |
2474 | con->in_msg->con = NULL; | |
5e095e8b SW |
2475 | ceph_msg_put(con->in_msg); |
2476 | con->in_msg = NULL; | |
36eb71aa | 2477 | con->ops->put(con); |
5e095e8b | 2478 | } |
31b8006e | 2479 | |
e80a52d1 SW |
2480 | /* Requeue anything that hasn't been acked */ |
2481 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2482 | |
e76661d0 SW |
2483 | /* If there are no messages queued or keepalive pending, place |
2484 | * the connection in a STANDBY state */ | |
2485 | if (list_empty(&con->out_queue) && | |
c9ffc77a | 2486 | !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) { |
e00de341 | 2487 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
c9ffc77a | 2488 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
8dacc7da | 2489 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
2490 | } else { |
2491 | /* retry after a delay. */ | |
8dacc7da | 2492 | con->state = CON_STATE_PREOPEN; |
e80a52d1 SW |
2493 | if (con->delay == 0) |
2494 | con->delay = BASE_DELAY_INTERVAL; | |
2495 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2496 | con->delay *= 2; | |
c9ffc77a | 2497 | con_flag_set(con, CON_FLAG_BACKOFF); |
8618e30b | 2498 | queue_con(con); |
31b8006e | 2499 | } |
31b8006e SW |
2500 | } |
2501 | ||
2502 | ||
2503 | ||
2504 | /* | |
15d9882c | 2505 | * initialize a new messenger instance |
31b8006e | 2506 | */ |
15d9882c AE |
2507 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2508 | struct ceph_entity_addr *myaddr, | |
2509 | u32 supported_features, | |
2510 | u32 required_features, | |
2511 | bool nocrc) | |
31b8006e | 2512 | { |
3d14c5d2 YS |
2513 | msgr->supported_features = supported_features; |
2514 | msgr->required_features = required_features; | |
2515 | ||
31b8006e SW |
2516 | spin_lock_init(&msgr->global_seq_lock); |
2517 | ||
31b8006e SW |
2518 | if (myaddr) |
2519 | msgr->inst.addr = *myaddr; | |
2520 | ||
2521 | /* select a random nonce */ | |
ac8839d7 | 2522 | msgr->inst.addr.type = 0; |
103e2d3a | 2523 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2524 | encode_my_addr(msgr); |
15d9882c | 2525 | msgr->nocrc = nocrc; |
31b8006e | 2526 | |
a2a32584 | 2527 | atomic_set(&msgr->stopping, 0); |
31b8006e | 2528 | |
15d9882c | 2529 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2530 | } |
15d9882c | 2531 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2532 | |
e00de341 SW |
2533 | static void clear_standby(struct ceph_connection *con) |
2534 | { | |
2535 | /* come back from STANDBY? */ | |
8dacc7da | 2536 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 2537 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 2538 | con->state = CON_STATE_PREOPEN; |
e00de341 | 2539 | con->connect_seq++; |
c9ffc77a AE |
2540 | WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING)); |
2541 | WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)); | |
e00de341 SW |
2542 | } |
2543 | } | |
2544 | ||
31b8006e SW |
2545 | /* |
2546 | * Queue up an outgoing message on the given connection. | |
2547 | */ | |
2548 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2549 | { | |
31b8006e | 2550 | /* set src+dst */ |
dbad185d | 2551 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 2552 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
2553 | msg->needs_out_seq = true; |
2554 | ||
ec302645 | 2555 | mutex_lock(&con->mutex); |
92ce034b | 2556 | |
8dacc7da | 2557 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
2558 | dout("con_send %p closed, dropping %p\n", con, msg); |
2559 | ceph_msg_put(msg); | |
2560 | mutex_unlock(&con->mutex); | |
2561 | return; | |
2562 | } | |
2563 | ||
38941f80 | 2564 | BUG_ON(msg->con != NULL); |
36eb71aa | 2565 | msg->con = con->ops->get(con); |
92ce034b AE |
2566 | BUG_ON(msg->con == NULL); |
2567 | ||
31b8006e SW |
2568 | BUG_ON(!list_empty(&msg->list_head)); |
2569 | list_add_tail(&msg->list_head, &con->out_queue); | |
2570 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2571 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2572 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2573 | le32_to_cpu(msg->hdr.front_len), | |
2574 | le32_to_cpu(msg->hdr.middle_len), | |
2575 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
2576 | |
2577 | clear_standby(con); | |
ec302645 | 2578 | mutex_unlock(&con->mutex); |
31b8006e SW |
2579 | |
2580 | /* if there wasn't anything waiting to send before, queue | |
2581 | * new work */ | |
c9ffc77a | 2582 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
31b8006e SW |
2583 | queue_con(con); |
2584 | } | |
3d14c5d2 | 2585 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2586 | |
2587 | /* | |
2588 | * Revoke a message that was previously queued for send | |
2589 | */ | |
6740a845 | 2590 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2591 | { |
6740a845 AE |
2592 | struct ceph_connection *con = msg->con; |
2593 | ||
2594 | if (!con) | |
2595 | return; /* Message not in our possession */ | |
2596 | ||
ec302645 | 2597 | mutex_lock(&con->mutex); |
31b8006e | 2598 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2599 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2600 | list_del_init(&msg->list_head); |
38941f80 | 2601 | BUG_ON(msg->con == NULL); |
36eb71aa | 2602 | msg->con->ops->put(msg->con); |
38941f80 | 2603 | msg->con = NULL; |
31b8006e | 2604 | msg->hdr.seq = 0; |
38941f80 | 2605 | |
31b8006e | 2606 | ceph_msg_put(msg); |
ed98adad SW |
2607 | } |
2608 | if (con->out_msg == msg) { | |
38941f80 | 2609 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2610 | con->out_msg = NULL; |
31b8006e SW |
2611 | if (con->out_kvec_is_msg) { |
2612 | con->out_skip = con->out_kvec_bytes; | |
2613 | con->out_kvec_is_msg = false; | |
2614 | } | |
ed98adad | 2615 | msg->hdr.seq = 0; |
92ce034b AE |
2616 | |
2617 | ceph_msg_put(msg); | |
31b8006e | 2618 | } |
ec302645 | 2619 | mutex_unlock(&con->mutex); |
31b8006e SW |
2620 | } |
2621 | ||
350b1c32 | 2622 | /* |
0d59ab81 | 2623 | * Revoke a message that we may be reading data into |
350b1c32 | 2624 | */ |
8921d114 | 2625 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2626 | { |
8921d114 AE |
2627 | struct ceph_connection *con; |
2628 | ||
2629 | BUG_ON(msg == NULL); | |
2630 | if (!msg->con) { | |
2631 | dout("%s msg %p null con\n", __func__, msg); | |
2632 | ||
2633 | return; /* Message not in our possession */ | |
2634 | } | |
2635 | ||
2636 | con = msg->con; | |
350b1c32 | 2637 | mutex_lock(&con->mutex); |
8921d114 | 2638 | if (con->in_msg == msg) { |
95c96174 ED |
2639 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
2640 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
2641 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
2642 | |
2643 | /* skip rest of message */ | |
8921d114 AE |
2644 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
2645 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 2646 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
2647 | front_len - |
2648 | middle_len - | |
2649 | data_len - | |
350b1c32 | 2650 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2651 | ceph_msg_put(con->in_msg); |
2652 | con->in_msg = NULL; | |
2653 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2654 | con->in_seq++; |
350b1c32 | 2655 | } else { |
8921d114 AE |
2656 | dout("%s %p in_msg %p msg %p no-op\n", |
2657 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
2658 | } |
2659 | mutex_unlock(&con->mutex); | |
2660 | } | |
2661 | ||
31b8006e SW |
2662 | /* |
2663 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2664 | */ | |
2665 | void ceph_con_keepalive(struct ceph_connection *con) | |
2666 | { | |
e00de341 | 2667 | dout("con_keepalive %p\n", con); |
00650931 | 2668 | mutex_lock(&con->mutex); |
e00de341 | 2669 | clear_standby(con); |
00650931 | 2670 | mutex_unlock(&con->mutex); |
c9ffc77a AE |
2671 | if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 && |
2672 | con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) | |
31b8006e SW |
2673 | queue_con(con); |
2674 | } | |
3d14c5d2 | 2675 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2676 | |
2677 | ||
2678 | /* | |
2679 | * construct a new message with given type, size | |
2680 | * the new msg has a ref count of 1. | |
2681 | */ | |
b61c2763 SW |
2682 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2683 | bool can_fail) | |
31b8006e SW |
2684 | { |
2685 | struct ceph_msg *m; | |
2686 | ||
34d23762 | 2687 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2688 | if (m == NULL) |
2689 | goto out; | |
c2e552e7 | 2690 | kref_init(&m->kref); |
38941f80 AE |
2691 | |
2692 | m->con = NULL; | |
31b8006e SW |
2693 | INIT_LIST_HEAD(&m->list_head); |
2694 | ||
45c6ceb5 | 2695 | m->hdr.tid = 0; |
31b8006e | 2696 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2697 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2698 | m->hdr.version = 0; | |
31b8006e SW |
2699 | m->hdr.front_len = cpu_to_le32(front_len); |
2700 | m->hdr.middle_len = 0; | |
bb257664 SW |
2701 | m->hdr.data_len = 0; |
2702 | m->hdr.data_off = 0; | |
45c6ceb5 | 2703 | m->hdr.reserved = 0; |
31b8006e SW |
2704 | m->footer.front_crc = 0; |
2705 | m->footer.middle_crc = 0; | |
2706 | m->footer.data_crc = 0; | |
45c6ceb5 | 2707 | m->footer.flags = 0; |
31b8006e SW |
2708 | m->front_max = front_len; |
2709 | m->front_is_vmalloc = false; | |
2710 | m->more_to_follow = false; | |
c0d5f9db | 2711 | m->ack_stamp = 0; |
31b8006e SW |
2712 | m->pool = NULL; |
2713 | ||
ca20892d HC |
2714 | /* middle */ |
2715 | m->middle = NULL; | |
2716 | ||
2717 | /* data */ | |
d4b515fa | 2718 | m->page_count = 0; |
ca20892d HC |
2719 | m->page_alignment = 0; |
2720 | m->pages = NULL; | |
2721 | m->pagelist = NULL; | |
3ebc21f7 | 2722 | #ifdef CONFIG_BLOCK |
ca20892d HC |
2723 | m->bio = NULL; |
2724 | m->bio_iter = NULL; | |
2725 | m->bio_seg = 0; | |
3ebc21f7 | 2726 | #endif /* CONFIG_BLOCK */ |
ca20892d HC |
2727 | m->trail = NULL; |
2728 | ||
31b8006e SW |
2729 | /* front */ |
2730 | if (front_len) { | |
2731 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2732 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2733 | PAGE_KERNEL); |
2734 | m->front_is_vmalloc = true; | |
2735 | } else { | |
34d23762 | 2736 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2737 | } |
2738 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2739 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2740 | front_len); |
2741 | goto out2; | |
2742 | } | |
2743 | } else { | |
2744 | m->front.iov_base = NULL; | |
2745 | } | |
2746 | m->front.iov_len = front_len; | |
2747 | ||
bb257664 | 2748 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2749 | return m; |
2750 | ||
2751 | out2: | |
2752 | ceph_msg_put(m); | |
2753 | out: | |
b61c2763 SW |
2754 | if (!can_fail) { |
2755 | pr_err("msg_new can't create type %d front %d\n", type, | |
2756 | front_len); | |
f0ed1b7c | 2757 | WARN_ON(1); |
b61c2763 SW |
2758 | } else { |
2759 | dout("msg_new can't create type %d front %d\n", type, | |
2760 | front_len); | |
2761 | } | |
a79832f2 | 2762 | return NULL; |
31b8006e | 2763 | } |
3d14c5d2 | 2764 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2765 | |
31b8006e SW |
2766 | /* |
2767 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2768 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2769 | * per-type header in the front and then gracefully fail (i.e., | |
2770 | * propagate the error to the caller based on info in the front) when | |
2771 | * the middle is too large. | |
2772 | */ | |
2450418c | 2773 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2774 | { |
2775 | int type = le16_to_cpu(msg->hdr.type); | |
2776 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2777 | ||
2778 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2779 | ceph_msg_type_name(type), middle_len); | |
2780 | BUG_ON(!middle_len); | |
2781 | BUG_ON(msg->middle); | |
2782 | ||
b6c1d5b8 | 2783 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2784 | if (!msg->middle) |
2785 | return -ENOMEM; | |
2786 | return 0; | |
2787 | } | |
2788 | ||
2450418c | 2789 | /* |
1c20f2d2 AE |
2790 | * Allocate a message for receiving an incoming message on a |
2791 | * connection, and save the result in con->in_msg. Uses the | |
2792 | * connection's private alloc_msg op if available. | |
2793 | * | |
4740a623 SW |
2794 | * Returns 0 on success, or a negative error code. |
2795 | * | |
2796 | * On success, if we set *skip = 1: | |
2797 | * - the next message should be skipped and ignored. | |
2798 | * - con->in_msg == NULL | |
2799 | * or if we set *skip = 0: | |
2800 | * - con->in_msg is non-null. | |
2801 | * On error (ENOMEM, EAGAIN, ...), | |
2802 | * - con->in_msg == NULL | |
2450418c | 2803 | */ |
4740a623 | 2804 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
2450418c | 2805 | { |
4740a623 | 2806 | struct ceph_msg_header *hdr = &con->in_hdr; |
2450418c | 2807 | int middle_len = le32_to_cpu(hdr->middle_len); |
1d866d1c | 2808 | struct ceph_msg *msg; |
4740a623 | 2809 | int ret = 0; |
2450418c | 2810 | |
1c20f2d2 | 2811 | BUG_ON(con->in_msg != NULL); |
53ded495 | 2812 | BUG_ON(!con->ops->alloc_msg); |
2450418c | 2813 | |
53ded495 AE |
2814 | mutex_unlock(&con->mutex); |
2815 | msg = con->ops->alloc_msg(con, hdr, skip); | |
2816 | mutex_lock(&con->mutex); | |
2817 | if (con->state != CON_STATE_OPEN) { | |
2818 | if (msg) | |
1d866d1c | 2819 | ceph_msg_put(msg); |
53ded495 AE |
2820 | return -EAGAIN; |
2821 | } | |
4137577a AE |
2822 | if (msg) { |
2823 | BUG_ON(*skip); | |
2824 | con->in_msg = msg; | |
36eb71aa | 2825 | con->in_msg->con = con->ops->get(con); |
92ce034b | 2826 | BUG_ON(con->in_msg->con == NULL); |
4137577a AE |
2827 | } else { |
2828 | /* | |
2829 | * Null message pointer means either we should skip | |
2830 | * this message or we couldn't allocate memory. The | |
2831 | * former is not an error. | |
2832 | */ | |
2833 | if (*skip) | |
2834 | return 0; | |
2835 | con->error_msg = "error allocating memory for incoming message"; | |
2836 | ||
53ded495 | 2837 | return -ENOMEM; |
2450418c | 2838 | } |
1c20f2d2 | 2839 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2840 | |
1c20f2d2 AE |
2841 | if (middle_len && !con->in_msg->middle) { |
2842 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 2843 | if (ret < 0) { |
1c20f2d2 AE |
2844 | ceph_msg_put(con->in_msg); |
2845 | con->in_msg = NULL; | |
2450418c YS |
2846 | } |
2847 | } | |
9d7f0f13 | 2848 | |
4740a623 | 2849 | return ret; |
2450418c YS |
2850 | } |
2851 | ||
31b8006e SW |
2852 | |
2853 | /* | |
2854 | * Free a generically kmalloc'd message. | |
2855 | */ | |
2856 | void ceph_msg_kfree(struct ceph_msg *m) | |
2857 | { | |
2858 | dout("msg_kfree %p\n", m); | |
2859 | if (m->front_is_vmalloc) | |
2860 | vfree(m->front.iov_base); | |
2861 | else | |
2862 | kfree(m->front.iov_base); | |
2863 | kfree(m); | |
2864 | } | |
2865 | ||
2866 | /* | |
2867 | * Drop a msg ref. Destroy as needed. | |
2868 | */ | |
c2e552e7 SW |
2869 | void ceph_msg_last_put(struct kref *kref) |
2870 | { | |
2871 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2872 | |
c2e552e7 SW |
2873 | dout("ceph_msg_put last one on %p\n", m); |
2874 | WARN_ON(!list_empty(&m->list_head)); | |
2875 | ||
2876 | /* drop middle, data, if any */ | |
2877 | if (m->middle) { | |
2878 | ceph_buffer_put(m->middle); | |
2879 | m->middle = NULL; | |
31b8006e | 2880 | } |
d4b515fa | 2881 | m->page_count = 0; |
c2e552e7 SW |
2882 | m->pages = NULL; |
2883 | ||
58bb3b37 SW |
2884 | if (m->pagelist) { |
2885 | ceph_pagelist_release(m->pagelist); | |
2886 | kfree(m->pagelist); | |
2887 | m->pagelist = NULL; | |
2888 | } | |
2889 | ||
68b4476b YS |
2890 | m->trail = NULL; |
2891 | ||
c2e552e7 SW |
2892 | if (m->pool) |
2893 | ceph_msgpool_put(m->pool, m); | |
2894 | else | |
2895 | ceph_msg_kfree(m); | |
31b8006e | 2896 | } |
3d14c5d2 | 2897 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2898 | |
2899 | void ceph_msg_dump(struct ceph_msg *msg) | |
2900 | { | |
d4b515fa AE |
2901 | pr_debug("msg_dump %p (front_max %d page_count %d)\n", msg, |
2902 | msg->front_max, msg->page_count); | |
9ec7cab1 SW |
2903 | print_hex_dump(KERN_DEBUG, "header: ", |
2904 | DUMP_PREFIX_OFFSET, 16, 1, | |
2905 | &msg->hdr, sizeof(msg->hdr), true); | |
2906 | print_hex_dump(KERN_DEBUG, " front: ", | |
2907 | DUMP_PREFIX_OFFSET, 16, 1, | |
2908 | msg->front.iov_base, msg->front.iov_len, true); | |
2909 | if (msg->middle) | |
2910 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2911 | DUMP_PREFIX_OFFSET, 16, 1, | |
2912 | msg->middle->vec.iov_base, | |
2913 | msg->middle->vec.iov_len, true); | |
2914 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2915 | DUMP_PREFIX_OFFSET, 16, 1, | |
2916 | &msg->footer, sizeof(msg->footer), true); | |
2917 | } | |
3d14c5d2 | 2918 | EXPORT_SYMBOL(ceph_msg_dump); |