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 | ||
afb3d90e AE |
474 | static int ceph_tcp_recvpage(struct socket *sock, struct page *page, |
475 | int page_offset, size_t length) | |
476 | { | |
477 | void *kaddr; | |
478 | int ret; | |
479 | ||
480 | BUG_ON(page_offset + length > PAGE_SIZE); | |
481 | ||
482 | kaddr = kmap(page); | |
483 | BUG_ON(!kaddr); | |
484 | ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length); | |
485 | kunmap(page); | |
486 | ||
487 | return ret; | |
488 | } | |
489 | ||
31b8006e SW |
490 | /* |
491 | * write something. @more is true if caller will be sending more data | |
492 | * shortly. | |
493 | */ | |
494 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
495 | size_t kvlen, size_t len, int more) | |
496 | { | |
497 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 498 | int r; |
31b8006e SW |
499 | |
500 | if (more) | |
501 | msg.msg_flags |= MSG_MORE; | |
502 | else | |
503 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
504 | ||
42961d23 SW |
505 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
506 | if (r == -EAGAIN) | |
507 | r = 0; | |
508 | return r; | |
31b8006e SW |
509 | } |
510 | ||
31739139 | 511 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
e1dcb128 | 512 | int offset, size_t size, bool more) |
31739139 AE |
513 | { |
514 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
515 | int ret; | |
516 | ||
517 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
518 | if (ret == -EAGAIN) | |
519 | ret = 0; | |
520 | ||
521 | return ret; | |
522 | } | |
523 | ||
31b8006e SW |
524 | |
525 | /* | |
526 | * Shutdown/close the socket for the given connection. | |
527 | */ | |
528 | static int con_close_socket(struct ceph_connection *con) | |
529 | { | |
8007b8d6 | 530 | int rc = 0; |
31b8006e SW |
531 | |
532 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
8007b8d6 SW |
533 | if (con->sock) { |
534 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
535 | sock_release(con->sock); | |
536 | con->sock = NULL; | |
537 | } | |
456ea468 AE |
538 | |
539 | /* | |
4a861692 | 540 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
541 | * independent of the connection mutex, and we could have |
542 | * received a socket close event before we had the chance to | |
543 | * shut the socket down. | |
544 | */ | |
c9ffc77a | 545 | con_flag_clear(con, CON_FLAG_SOCK_CLOSED); |
8007b8d6 | 546 | |
ce2c8903 | 547 | con_sock_state_closed(con); |
31b8006e SW |
548 | return rc; |
549 | } | |
550 | ||
551 | /* | |
552 | * Reset a connection. Discard all incoming and outgoing messages | |
553 | * and clear *_seq state. | |
554 | */ | |
555 | static void ceph_msg_remove(struct ceph_msg *msg) | |
556 | { | |
557 | list_del_init(&msg->list_head); | |
38941f80 | 558 | BUG_ON(msg->con == NULL); |
36eb71aa | 559 | msg->con->ops->put(msg->con); |
38941f80 AE |
560 | msg->con = NULL; |
561 | ||
31b8006e SW |
562 | ceph_msg_put(msg); |
563 | } | |
564 | static void ceph_msg_remove_list(struct list_head *head) | |
565 | { | |
566 | while (!list_empty(head)) { | |
567 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
568 | list_head); | |
569 | ceph_msg_remove(msg); | |
570 | } | |
571 | } | |
572 | ||
573 | static void reset_connection(struct ceph_connection *con) | |
574 | { | |
575 | /* reset connection, out_queue, msg_ and connect_seq */ | |
576 | /* discard existing out_queue and msg_seq */ | |
0fa6ebc6 | 577 | dout("reset_connection %p\n", con); |
31b8006e SW |
578 | ceph_msg_remove_list(&con->out_queue); |
579 | ceph_msg_remove_list(&con->out_sent); | |
580 | ||
cf3e5c40 | 581 | if (con->in_msg) { |
38941f80 AE |
582 | BUG_ON(con->in_msg->con != con); |
583 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
584 | ceph_msg_put(con->in_msg); |
585 | con->in_msg = NULL; | |
36eb71aa | 586 | con->ops->put(con); |
cf3e5c40 SW |
587 | } |
588 | ||
31b8006e SW |
589 | con->connect_seq = 0; |
590 | con->out_seq = 0; | |
c86a2930 SW |
591 | if (con->out_msg) { |
592 | ceph_msg_put(con->out_msg); | |
593 | con->out_msg = NULL; | |
594 | } | |
31b8006e | 595 | con->in_seq = 0; |
0e0d5e0c | 596 | con->in_seq_acked = 0; |
31b8006e SW |
597 | } |
598 | ||
599 | /* | |
600 | * mark a peer down. drop any open connections. | |
601 | */ | |
602 | void ceph_con_close(struct ceph_connection *con) | |
603 | { | |
8c50c817 | 604 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
605 | dout("con_close %p peer %s\n", con, |
606 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
8dacc7da | 607 | con->state = CON_STATE_CLOSED; |
a5988c49 | 608 | |
c9ffc77a AE |
609 | con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */ |
610 | con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING); | |
611 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); | |
612 | con_flag_clear(con, CON_FLAG_BACKOFF); | |
a5988c49 | 613 | |
31b8006e | 614 | reset_connection(con); |
6f2bc3ff | 615 | con->peer_global_seq = 0; |
91e45ce3 | 616 | cancel_delayed_work(&con->work); |
ee76e073 | 617 | con_close_socket(con); |
ec302645 | 618 | mutex_unlock(&con->mutex); |
31b8006e | 619 | } |
3d14c5d2 | 620 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 621 | |
31b8006e SW |
622 | /* |
623 | * Reopen a closed connection, with a new peer address. | |
624 | */ | |
b7a9e5dd SW |
625 | void ceph_con_open(struct ceph_connection *con, |
626 | __u8 entity_type, __u64 entity_num, | |
627 | struct ceph_entity_addr *addr) | |
31b8006e | 628 | { |
5469155f | 629 | mutex_lock(&con->mutex); |
3d14c5d2 | 630 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
8dacc7da | 631 | |
122070a2 | 632 | WARN_ON(con->state != CON_STATE_CLOSED); |
8dacc7da | 633 | con->state = CON_STATE_PREOPEN; |
a5988c49 | 634 | |
b7a9e5dd SW |
635 | con->peer_name.type = (__u8) entity_type; |
636 | con->peer_name.num = cpu_to_le64(entity_num); | |
637 | ||
31b8006e | 638 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 639 | con->delay = 0; /* reset backoff memory */ |
5469155f | 640 | mutex_unlock(&con->mutex); |
31b8006e SW |
641 | queue_con(con); |
642 | } | |
3d14c5d2 | 643 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 644 | |
87b315a5 SW |
645 | /* |
646 | * return true if this connection ever successfully opened | |
647 | */ | |
648 | bool ceph_con_opened(struct ceph_connection *con) | |
649 | { | |
650 | return con->connect_seq > 0; | |
651 | } | |
652 | ||
31b8006e SW |
653 | /* |
654 | * initialize a new connection. | |
655 | */ | |
1bfd89f4 AE |
656 | void ceph_con_init(struct ceph_connection *con, void *private, |
657 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 658 | struct ceph_messenger *msgr) |
31b8006e SW |
659 | { |
660 | dout("con_init %p\n", con); | |
661 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
662 | con->private = private; |
663 | con->ops = ops; | |
31b8006e | 664 | con->msgr = msgr; |
ce2c8903 AE |
665 | |
666 | con_sock_state_init(con); | |
667 | ||
ec302645 | 668 | mutex_init(&con->mutex); |
31b8006e SW |
669 | INIT_LIST_HEAD(&con->out_queue); |
670 | INIT_LIST_HEAD(&con->out_sent); | |
671 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 | 672 | |
8dacc7da | 673 | con->state = CON_STATE_CLOSED; |
31b8006e | 674 | } |
3d14c5d2 | 675 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
676 | |
677 | ||
678 | /* | |
679 | * We maintain a global counter to order connection attempts. Get | |
680 | * a unique seq greater than @gt. | |
681 | */ | |
682 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
683 | { | |
684 | u32 ret; | |
685 | ||
686 | spin_lock(&msgr->global_seq_lock); | |
687 | if (msgr->global_seq < gt) | |
688 | msgr->global_seq = gt; | |
689 | ret = ++msgr->global_seq; | |
690 | spin_unlock(&msgr->global_seq_lock); | |
691 | return ret; | |
692 | } | |
693 | ||
e2200423 | 694 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
695 | { |
696 | con->out_kvec_left = 0; | |
697 | con->out_kvec_bytes = 0; | |
698 | con->out_kvec_cur = &con->out_kvec[0]; | |
699 | } | |
700 | ||
e2200423 | 701 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
702 | size_t size, void *data) |
703 | { | |
704 | int index; | |
705 | ||
706 | index = con->out_kvec_left; | |
707 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
708 | ||
709 | con->out_kvec[index].iov_len = size; | |
710 | con->out_kvec[index].iov_base = data; | |
711 | con->out_kvec_left++; | |
712 | con->out_kvec_bytes += size; | |
713 | } | |
31b8006e | 714 | |
df6ad1f9 | 715 | #ifdef CONFIG_BLOCK |
07c09b72 AE |
716 | static void init_bio_iter(struct bio *bio, struct bio **bio_iter, |
717 | unsigned int *bio_seg) | |
df6ad1f9 AE |
718 | { |
719 | if (!bio) { | |
07c09b72 AE |
720 | *bio_iter = NULL; |
721 | *bio_seg = 0; | |
df6ad1f9 AE |
722 | return; |
723 | } | |
07c09b72 AE |
724 | *bio_iter = bio; |
725 | *bio_seg = (unsigned int) bio->bi_idx; | |
df6ad1f9 AE |
726 | } |
727 | ||
07c09b72 | 728 | static void iter_bio_next(struct bio **bio_iter, unsigned int *seg) |
df6ad1f9 AE |
729 | { |
730 | if (*bio_iter == NULL) | |
731 | return; | |
732 | ||
733 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
734 | ||
735 | (*seg)++; | |
736 | if (*seg == (*bio_iter)->bi_vcnt) | |
737 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
738 | } | |
739 | #endif | |
740 | ||
78625051 AE |
741 | static void prepare_message_data(struct ceph_msg *msg, |
742 | struct ceph_msg_pos *msg_pos) | |
739c905b | 743 | { |
739c905b AE |
744 | BUG_ON(!msg); |
745 | BUG_ON(!msg->hdr.data_len); | |
746 | ||
747 | /* initialize page iterator */ | |
bae6acd9 | 748 | msg_pos->page = 0; |
739c905b | 749 | if (msg->pages) |
bae6acd9 | 750 | msg_pos->page_pos = msg->page_alignment; |
739c905b | 751 | else |
bae6acd9 | 752 | msg_pos->page_pos = 0; |
572c588e | 753 | #ifdef CONFIG_BLOCK |
abdaa6a8 | 754 | if (msg->bio) |
572c588e AE |
755 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); |
756 | #endif | |
bae6acd9 AE |
757 | msg_pos->data_pos = 0; |
758 | msg_pos->did_page_crc = false; | |
739c905b AE |
759 | } |
760 | ||
31b8006e SW |
761 | /* |
762 | * Prepare footer for currently outgoing message, and finish things | |
763 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
764 | */ | |
859eb799 | 765 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
766 | { |
767 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 768 | int v = con->out_kvec_left; |
31b8006e | 769 | |
fd154f3c AE |
770 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
771 | ||
31b8006e SW |
772 | dout("prepare_write_message_footer %p\n", con); |
773 | con->out_kvec_is_msg = true; | |
774 | con->out_kvec[v].iov_base = &m->footer; | |
775 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
776 | con->out_kvec_bytes += sizeof(m->footer); | |
777 | con->out_kvec_left++; | |
778 | con->out_more = m->more_to_follow; | |
c86a2930 | 779 | con->out_msg_done = true; |
31b8006e SW |
780 | } |
781 | ||
782 | /* | |
783 | * Prepare headers for the next outgoing message. | |
784 | */ | |
785 | static void prepare_write_message(struct ceph_connection *con) | |
786 | { | |
787 | struct ceph_msg *m; | |
a9a0c51a | 788 | u32 crc; |
31b8006e | 789 | |
e2200423 | 790 | con_out_kvec_reset(con); |
31b8006e | 791 | con->out_kvec_is_msg = true; |
c86a2930 | 792 | con->out_msg_done = false; |
31b8006e SW |
793 | |
794 | /* Sneak an ack in there first? If we can get it into the same | |
795 | * TCP packet that's a good thing. */ | |
796 | if (con->in_seq > con->in_seq_acked) { | |
797 | con->in_seq_acked = con->in_seq; | |
e2200423 | 798 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 799 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 800 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 801 | &con->out_temp_ack); |
31b8006e SW |
802 | } |
803 | ||
38941f80 | 804 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 805 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 806 | con->out_msg = m; |
38941f80 | 807 | BUG_ON(m->con != con); |
4cf9d544 SW |
808 | |
809 | /* put message on sent list */ | |
810 | ceph_msg_get(m); | |
811 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 812 | |
e84346b7 SW |
813 | /* |
814 | * only assign outgoing seq # if we haven't sent this message | |
815 | * yet. if it is requeued, resend with it's original seq. | |
816 | */ | |
817 | if (m->needs_out_seq) { | |
818 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
819 | m->needs_out_seq = false; | |
820 | } | |
31b8006e | 821 | |
4a73ef27 | 822 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n", |
31b8006e SW |
823 | m, con->out_seq, le16_to_cpu(m->hdr.type), |
824 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
4a73ef27 | 825 | le32_to_cpu(m->hdr.data_len), m->length); |
31b8006e SW |
826 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); |
827 | ||
828 | /* tag + hdr + front + middle */ | |
e2200423 AE |
829 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
830 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
831 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 832 | |
31b8006e | 833 | if (m->middle) |
e2200423 | 834 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 835 | m->middle->vec.iov_base); |
31b8006e SW |
836 | |
837 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
838 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
839 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 840 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
841 | |
842 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
843 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
844 | if (m->middle) { | |
845 | crc = crc32c(0, m->middle->vec.iov_base, | |
846 | m->middle->vec.iov_len); | |
847 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
848 | } else | |
31b8006e | 849 | con->out_msg->footer.middle_crc = 0; |
739c905b | 850 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
851 | le32_to_cpu(con->out_msg->footer.front_crc), |
852 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
853 | ||
854 | /* is there a data payload? */ | |
739c905b | 855 | con->out_msg->footer.data_crc = 0; |
78625051 AE |
856 | if (m->hdr.data_len) { |
857 | prepare_message_data(con->out_msg, &con->out_msg_pos); | |
858 | con->out_more = 1; /* data + footer will follow */ | |
859 | } else { | |
31b8006e | 860 | /* no, queue up footer too and be done */ |
859eb799 | 861 | prepare_write_message_footer(con); |
78625051 | 862 | } |
31b8006e | 863 | |
c9ffc77a | 864 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
865 | } |
866 | ||
867 | /* | |
868 | * Prepare an ack. | |
869 | */ | |
870 | static void prepare_write_ack(struct ceph_connection *con) | |
871 | { | |
872 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
873 | con->in_seq_acked, con->in_seq); | |
874 | con->in_seq_acked = con->in_seq; | |
875 | ||
e2200423 | 876 | con_out_kvec_reset(con); |
859eb799 | 877 | |
e2200423 | 878 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 879 | |
31b8006e | 880 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 881 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
882 | &con->out_temp_ack); |
883 | ||
31b8006e | 884 | con->out_more = 1; /* more will follow.. eventually.. */ |
c9ffc77a | 885 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
886 | } |
887 | ||
888 | /* | |
889 | * Prepare to write keepalive byte. | |
890 | */ | |
891 | static void prepare_write_keepalive(struct ceph_connection *con) | |
892 | { | |
893 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
894 | con_out_kvec_reset(con); |
895 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
c9ffc77a | 896 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
897 | } |
898 | ||
899 | /* | |
900 | * Connection negotiation. | |
901 | */ | |
902 | ||
dac1e716 AE |
903 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
904 | int *auth_proto) | |
4e7a5dcd | 905 | { |
a3530df3 | 906 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
907 | |
908 | if (!con->ops->get_authorizer) { | |
909 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
910 | con->out_connect.authorizer_len = 0; | |
729796be | 911 | return NULL; |
b1c6b980 AE |
912 | } |
913 | ||
914 | /* Can't hold the mutex while getting authorizer */ | |
ec302645 | 915 | mutex_unlock(&con->mutex); |
dac1e716 | 916 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
ec302645 | 917 | mutex_lock(&con->mutex); |
4e7a5dcd | 918 | |
a3530df3 | 919 | if (IS_ERR(auth)) |
729796be | 920 | return auth; |
8dacc7da | 921 | if (con->state != CON_STATE_NEGOTIATING) |
729796be | 922 | return ERR_PTR(-EAGAIN); |
0da5d703 | 923 | |
8f43fb53 AE |
924 | con->auth_reply_buf = auth->authorizer_reply_buf; |
925 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
729796be | 926 | return auth; |
4e7a5dcd SW |
927 | } |
928 | ||
31b8006e SW |
929 | /* |
930 | * We connected to a peer and are saying hello. | |
931 | */ | |
e825a66d | 932 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 933 | { |
e2200423 AE |
934 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
935 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 936 | &con->msgr->my_enc_addr); |
eed0ef2c | 937 | |
eed0ef2c | 938 | con->out_more = 0; |
c9ffc77a | 939 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
eed0ef2c SW |
940 | } |
941 | ||
e825a66d | 942 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 943 | { |
95c96174 | 944 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 945 | int proto; |
dac1e716 | 946 | int auth_proto; |
729796be | 947 | struct ceph_auth_handshake *auth; |
31b8006e SW |
948 | |
949 | switch (con->peer_name.type) { | |
950 | case CEPH_ENTITY_TYPE_MON: | |
951 | proto = CEPH_MONC_PROTOCOL; | |
952 | break; | |
953 | case CEPH_ENTITY_TYPE_OSD: | |
954 | proto = CEPH_OSDC_PROTOCOL; | |
955 | break; | |
956 | case CEPH_ENTITY_TYPE_MDS: | |
957 | proto = CEPH_MDSC_PROTOCOL; | |
958 | break; | |
959 | default: | |
960 | BUG(); | |
961 | } | |
962 | ||
963 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
964 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 965 | |
e825a66d | 966 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
967 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
968 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
969 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
970 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
971 | con->out_connect.flags = 0; | |
31b8006e | 972 | |
dac1e716 AE |
973 | auth_proto = CEPH_AUTH_UNKNOWN; |
974 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
975 | if (IS_ERR(auth)) |
976 | return PTR_ERR(auth); | |
3da54776 | 977 | |
dac1e716 | 978 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
979 | con->out_connect.authorizer_len = auth ? |
980 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
981 | ||
e2200423 | 982 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
983 | &con->out_connect); |
984 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 985 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 986 | auth->authorizer_buf); |
859eb799 | 987 | |
31b8006e | 988 | con->out_more = 0; |
c9ffc77a | 989 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
4e7a5dcd | 990 | |
e10c758e | 991 | return 0; |
31b8006e SW |
992 | } |
993 | ||
31b8006e SW |
994 | /* |
995 | * write as much of pending kvecs to the socket as we can. | |
996 | * 1 -> done | |
997 | * 0 -> socket full, but more to do | |
998 | * <0 -> error | |
999 | */ | |
1000 | static int write_partial_kvec(struct ceph_connection *con) | |
1001 | { | |
1002 | int ret; | |
1003 | ||
1004 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
1005 | while (con->out_kvec_bytes > 0) { | |
1006 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
1007 | con->out_kvec_left, con->out_kvec_bytes, | |
1008 | con->out_more); | |
1009 | if (ret <= 0) | |
1010 | goto out; | |
1011 | con->out_kvec_bytes -= ret; | |
1012 | if (con->out_kvec_bytes == 0) | |
1013 | break; /* done */ | |
f42299e6 AE |
1014 | |
1015 | /* account for full iov entries consumed */ | |
1016 | while (ret >= con->out_kvec_cur->iov_len) { | |
1017 | BUG_ON(!con->out_kvec_left); | |
1018 | ret -= con->out_kvec_cur->iov_len; | |
1019 | con->out_kvec_cur++; | |
1020 | con->out_kvec_left--; | |
1021 | } | |
1022 | /* and for a partially-consumed entry */ | |
1023 | if (ret) { | |
1024 | con->out_kvec_cur->iov_len -= ret; | |
1025 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
1026 | } |
1027 | } | |
1028 | con->out_kvec_left = 0; | |
1029 | con->out_kvec_is_msg = false; | |
1030 | ret = 1; | |
1031 | out: | |
1032 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
1033 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
1034 | return ret; /* done! */ | |
1035 | } | |
1036 | ||
84ca8fc8 AE |
1037 | static void out_msg_pos_next(struct ceph_connection *con, struct page *page, |
1038 | size_t len, size_t sent, bool in_trail) | |
68b4476b | 1039 | { |
84ca8fc8 | 1040 | struct ceph_msg *msg = con->out_msg; |
bae6acd9 | 1041 | struct ceph_msg_pos *msg_pos = &con->out_msg_pos; |
68b4476b | 1042 | |
84ca8fc8 AE |
1043 | BUG_ON(!msg); |
1044 | BUG_ON(!sent); | |
68b4476b | 1045 | |
bae6acd9 AE |
1046 | msg_pos->data_pos += sent; |
1047 | msg_pos->page_pos += sent; | |
5821bd8c AE |
1048 | if (sent < len) |
1049 | return; | |
68b4476b | 1050 | |
5821bd8c | 1051 | BUG_ON(sent != len); |
bae6acd9 AE |
1052 | msg_pos->page_pos = 0; |
1053 | msg_pos->page++; | |
1054 | msg_pos->did_page_crc = false; | |
5821bd8c | 1055 | if (in_trail) |
35c7bfbc | 1056 | list_rotate_left(&msg->trail->head); |
5821bd8c | 1057 | else if (msg->pagelist) |
35c7bfbc | 1058 | list_rotate_left(&msg->pagelist->head); |
84ca8fc8 | 1059 | #ifdef CONFIG_BLOCK |
5821bd8c AE |
1060 | else if (msg->bio) |
1061 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
68b4476b | 1062 | #endif |
84ca8fc8 | 1063 | } |
68b4476b | 1064 | |
e788182f AE |
1065 | static void in_msg_pos_next(struct ceph_connection *con, size_t len, |
1066 | size_t received) | |
1067 | { | |
1068 | struct ceph_msg *msg = con->in_msg; | |
bae6acd9 | 1069 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; |
e788182f AE |
1070 | |
1071 | BUG_ON(!msg); | |
1072 | BUG_ON(!received); | |
1073 | ||
bae6acd9 AE |
1074 | msg_pos->data_pos += received; |
1075 | msg_pos->page_pos += received; | |
e788182f AE |
1076 | if (received < len) |
1077 | return; | |
1078 | ||
1079 | BUG_ON(received != len); | |
bae6acd9 AE |
1080 | msg_pos->page_pos = 0; |
1081 | msg_pos->page++; | |
e788182f AE |
1082 | #ifdef CONFIG_BLOCK |
1083 | if (msg->bio) | |
1084 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
1085 | #endif /* CONFIG_BLOCK */ | |
1086 | } | |
1087 | ||
31b8006e SW |
1088 | /* |
1089 | * Write as much message data payload as we can. If we finish, queue | |
1090 | * up the footer. | |
1091 | * 1 -> done, footer is now queued in out_kvec[]. | |
1092 | * 0 -> socket full, but more to do | |
1093 | * <0 -> error | |
1094 | */ | |
34d2d200 | 1095 | static int write_partial_message_data(struct ceph_connection *con) |
31b8006e SW |
1096 | { |
1097 | struct ceph_msg *msg = con->out_msg; | |
bae6acd9 | 1098 | struct ceph_msg_pos *msg_pos = &con->out_msg_pos; |
95c96174 | 1099 | unsigned int data_len = le32_to_cpu(msg->hdr.data_len); |
37675b0f | 1100 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 1101 | int ret; |
68b4476b | 1102 | int total_max_write; |
84ca8fc8 | 1103 | bool in_trail = false; |
5821bd8c AE |
1104 | const size_t trail_len = (msg->trail ? msg->trail->length : 0); |
1105 | const size_t trail_off = data_len - trail_len; | |
31b8006e | 1106 | |
34d2d200 | 1107 | dout("%s %p msg %p page %d offset %d\n", __func__, |
bae6acd9 | 1108 | con, msg, msg_pos->page, msg_pos->page_pos); |
31b8006e | 1109 | |
5821bd8c AE |
1110 | /* |
1111 | * Iterate through each page that contains data to be | |
1112 | * written, and send as much as possible for each. | |
1113 | * | |
1114 | * If we are calculating the data crc (the default), we will | |
1115 | * need to map the page. If we have no pages, they have | |
1116 | * been revoked, so use the zero page. | |
1117 | */ | |
bae6acd9 | 1118 | while (data_len > msg_pos->data_pos) { |
31b8006e | 1119 | struct page *page = NULL; |
e387d525 AE |
1120 | size_t page_offset; |
1121 | size_t length; | |
68b4476b | 1122 | int max_write = PAGE_SIZE; |
9bd19663 | 1123 | int bio_offset = 0; |
68b4476b | 1124 | |
bae6acd9 | 1125 | in_trail = in_trail || msg_pos->data_pos >= trail_off; |
5821bd8c | 1126 | if (!in_trail) |
bae6acd9 | 1127 | total_max_write = trail_off - msg_pos->data_pos; |
68b4476b | 1128 | |
5821bd8c | 1129 | if (in_trail) { |
bae6acd9 | 1130 | total_max_write = data_len - msg_pos->data_pos; |
68b4476b YS |
1131 | |
1132 | page = list_first_entry(&msg->trail->head, | |
1133 | struct page, lru); | |
68b4476b | 1134 | } else if (msg->pages) { |
bae6acd9 | 1135 | page = msg->pages[msg_pos->page]; |
58bb3b37 SW |
1136 | } else if (msg->pagelist) { |
1137 | page = list_first_entry(&msg->pagelist->head, | |
1138 | struct page, lru); | |
68b4476b YS |
1139 | #ifdef CONFIG_BLOCK |
1140 | } else if (msg->bio) { | |
1141 | struct bio_vec *bv; | |
1142 | ||
1143 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
1144 | page = bv->bv_page; | |
9bd19663 | 1145 | bio_offset = bv->bv_offset; |
68b4476b YS |
1146 | max_write = bv->bv_len; |
1147 | #endif | |
31b8006e | 1148 | } else { |
57666519 | 1149 | page = zero_page; |
31b8006e | 1150 | } |
e387d525 | 1151 | length = min_t(int, max_write - msg_pos->page_pos, |
68b4476b YS |
1152 | total_max_write); |
1153 | ||
e387d525 | 1154 | page_offset = msg_pos->page_pos + bio_offset; |
bae6acd9 | 1155 | if (do_datacrc && !msg_pos->did_page_crc) { |
9bd19663 | 1156 | void *base; |
5821bd8c | 1157 | u32 crc = le32_to_cpu(msg->footer.data_crc); |
8d63e318 | 1158 | char *kaddr; |
31b8006e | 1159 | |
8d63e318 | 1160 | kaddr = kmap(page); |
31b8006e | 1161 | BUG_ON(kaddr == NULL); |
e387d525 AE |
1162 | base = kaddr + page_offset; |
1163 | crc = crc32c(crc, base, length); | |
5ce765a5 | 1164 | kunmap(page); |
84ca8fc8 | 1165 | msg->footer.data_crc = cpu_to_le32(crc); |
bae6acd9 | 1166 | msg_pos->did_page_crc = true; |
31b8006e | 1167 | } |
e387d525 AE |
1168 | ret = ceph_tcp_sendpage(con->sock, page, page_offset, |
1169 | length, true); | |
31b8006e SW |
1170 | if (ret <= 0) |
1171 | goto out; | |
1172 | ||
e387d525 | 1173 | out_msg_pos_next(con, page, length, (size_t) ret, in_trail); |
31b8006e SW |
1174 | } |
1175 | ||
34d2d200 | 1176 | dout("%s %p msg %p done\n", __func__, con, msg); |
31b8006e SW |
1177 | |
1178 | /* prepare and queue up footer, too */ | |
37675b0f | 1179 | if (!do_datacrc) |
84ca8fc8 | 1180 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1181 | con_out_kvec_reset(con); |
859eb799 | 1182 | prepare_write_message_footer(con); |
31b8006e SW |
1183 | ret = 1; |
1184 | out: | |
1185 | return ret; | |
1186 | } | |
1187 | ||
1188 | /* | |
1189 | * write some zeros | |
1190 | */ | |
1191 | static int write_partial_skip(struct ceph_connection *con) | |
1192 | { | |
1193 | int ret; | |
1194 | ||
1195 | while (con->out_skip > 0) { | |
31739139 | 1196 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1197 | |
e1dcb128 | 1198 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true); |
31b8006e SW |
1199 | if (ret <= 0) |
1200 | goto out; | |
1201 | con->out_skip -= ret; | |
1202 | } | |
1203 | ret = 1; | |
1204 | out: | |
1205 | return ret; | |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * Prepare to read connection handshake, or an ack. | |
1210 | */ | |
eed0ef2c SW |
1211 | static void prepare_read_banner(struct ceph_connection *con) |
1212 | { | |
1213 | dout("prepare_read_banner %p\n", con); | |
1214 | con->in_base_pos = 0; | |
1215 | } | |
1216 | ||
31b8006e SW |
1217 | static void prepare_read_connect(struct ceph_connection *con) |
1218 | { | |
1219 | dout("prepare_read_connect %p\n", con); | |
1220 | con->in_base_pos = 0; | |
1221 | } | |
1222 | ||
1223 | static void prepare_read_ack(struct ceph_connection *con) | |
1224 | { | |
1225 | dout("prepare_read_ack %p\n", con); | |
1226 | con->in_base_pos = 0; | |
1227 | } | |
1228 | ||
1229 | static void prepare_read_tag(struct ceph_connection *con) | |
1230 | { | |
1231 | dout("prepare_read_tag %p\n", con); | |
1232 | con->in_base_pos = 0; | |
1233 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1234 | } | |
1235 | ||
1236 | /* | |
1237 | * Prepare to read a message. | |
1238 | */ | |
1239 | static int prepare_read_message(struct ceph_connection *con) | |
1240 | { | |
1241 | dout("prepare_read_message %p\n", con); | |
1242 | BUG_ON(con->in_msg != NULL); | |
1243 | con->in_base_pos = 0; | |
1244 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1245 | return 0; | |
1246 | } | |
1247 | ||
1248 | ||
1249 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1250 | int end, int size, void *object) |
31b8006e | 1251 | { |
e6cee71f AE |
1252 | while (con->in_base_pos < end) { |
1253 | int left = end - con->in_base_pos; | |
31b8006e SW |
1254 | int have = size - left; |
1255 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1256 | if (ret <= 0) | |
1257 | return ret; | |
1258 | con->in_base_pos += ret; | |
1259 | } | |
1260 | return 1; | |
1261 | } | |
1262 | ||
1263 | ||
1264 | /* | |
1265 | * Read all or part of the connect-side handshake on a new connection | |
1266 | */ | |
eed0ef2c | 1267 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1268 | { |
fd51653f AE |
1269 | int size; |
1270 | int end; | |
1271 | int ret; | |
31b8006e | 1272 | |
eed0ef2c | 1273 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1274 | |
1275 | /* peer's banner */ | |
fd51653f AE |
1276 | size = strlen(CEPH_BANNER); |
1277 | end = size; | |
1278 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1279 | if (ret <= 0) |
1280 | goto out; | |
fd51653f AE |
1281 | |
1282 | size = sizeof (con->actual_peer_addr); | |
1283 | end += size; | |
1284 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1285 | if (ret <= 0) |
1286 | goto out; | |
fd51653f AE |
1287 | |
1288 | size = sizeof (con->peer_addr_for_me); | |
1289 | end += size; | |
1290 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1291 | if (ret <= 0) |
1292 | goto out; | |
fd51653f | 1293 | |
eed0ef2c SW |
1294 | out: |
1295 | return ret; | |
1296 | } | |
1297 | ||
1298 | static int read_partial_connect(struct ceph_connection *con) | |
1299 | { | |
fd51653f AE |
1300 | int size; |
1301 | int end; | |
1302 | int ret; | |
eed0ef2c SW |
1303 | |
1304 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1305 | ||
fd51653f AE |
1306 | size = sizeof (con->in_reply); |
1307 | end = size; | |
1308 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1309 | if (ret <= 0) |
1310 | goto out; | |
fd51653f AE |
1311 | |
1312 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1313 | end += size; | |
1314 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1315 | if (ret <= 0) |
1316 | goto out; | |
31b8006e | 1317 | |
4e7a5dcd SW |
1318 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1319 | con, (int)con->in_reply.tag, | |
1320 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1321 | le32_to_cpu(con->in_reply.global_seq)); |
1322 | out: | |
1323 | return ret; | |
eed0ef2c | 1324 | |
31b8006e SW |
1325 | } |
1326 | ||
1327 | /* | |
1328 | * Verify the hello banner looks okay. | |
1329 | */ | |
1330 | static int verify_hello(struct ceph_connection *con) | |
1331 | { | |
1332 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1333 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1334 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1335 | con->error_msg = "protocol error, bad banner"; |
1336 | return -1; | |
1337 | } | |
1338 | return 0; | |
1339 | } | |
1340 | ||
1341 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1342 | { | |
1343 | switch (ss->ss_family) { | |
1344 | case AF_INET: | |
1345 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1346 | case AF_INET6: | |
1347 | return | |
1348 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1349 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1350 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1351 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1352 | } | |
1353 | return false; | |
1354 | } | |
1355 | ||
1356 | static int addr_port(struct sockaddr_storage *ss) | |
1357 | { | |
1358 | switch (ss->ss_family) { | |
1359 | case AF_INET: | |
f28bcfbe | 1360 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1361 | case AF_INET6: |
f28bcfbe | 1362 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1363 | } |
1364 | return 0; | |
1365 | } | |
1366 | ||
1367 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1368 | { | |
1369 | switch (ss->ss_family) { | |
1370 | case AF_INET: | |
1371 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1372 | break; |
31b8006e SW |
1373 | case AF_INET6: |
1374 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1375 | break; |
31b8006e SW |
1376 | } |
1377 | } | |
1378 | ||
ee3b56f2 NW |
1379 | /* |
1380 | * Unlike other *_pton function semantics, zero indicates success. | |
1381 | */ | |
1382 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1383 | char delim, const char **ipend) | |
1384 | { | |
99f0f3b2 AE |
1385 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1386 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1387 | |
1388 | memset(ss, 0, sizeof(*ss)); | |
1389 | ||
1390 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1391 | ss->ss_family = AF_INET; | |
1392 | return 0; | |
1393 | } | |
1394 | ||
1395 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1396 | ss->ss_family = AF_INET6; | |
1397 | return 0; | |
1398 | } | |
1399 | ||
1400 | return -EINVAL; | |
1401 | } | |
1402 | ||
1403 | /* | |
1404 | * Extract hostname string and resolve using kernel DNS facility. | |
1405 | */ | |
1406 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1407 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1408 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1409 | { | |
1410 | const char *end, *delim_p; | |
1411 | char *colon_p, *ip_addr = NULL; | |
1412 | int ip_len, ret; | |
1413 | ||
1414 | /* | |
1415 | * The end of the hostname occurs immediately preceding the delimiter or | |
1416 | * the port marker (':') where the delimiter takes precedence. | |
1417 | */ | |
1418 | delim_p = memchr(name, delim, namelen); | |
1419 | colon_p = memchr(name, ':', namelen); | |
1420 | ||
1421 | if (delim_p && colon_p) | |
1422 | end = delim_p < colon_p ? delim_p : colon_p; | |
1423 | else if (!delim_p && colon_p) | |
1424 | end = colon_p; | |
1425 | else { | |
1426 | end = delim_p; | |
1427 | if (!end) /* case: hostname:/ */ | |
1428 | end = name + namelen; | |
1429 | } | |
1430 | ||
1431 | if (end <= name) | |
1432 | return -EINVAL; | |
1433 | ||
1434 | /* do dns_resolve upcall */ | |
1435 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1436 | if (ip_len > 0) | |
1437 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1438 | else | |
1439 | ret = -ESRCH; | |
1440 | ||
1441 | kfree(ip_addr); | |
1442 | ||
1443 | *ipend = end; | |
1444 | ||
1445 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1446 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1447 | ||
1448 | return ret; | |
1449 | } | |
1450 | #else | |
1451 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1452 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1453 | { | |
1454 | return -EINVAL; | |
1455 | } | |
1456 | #endif | |
1457 | ||
1458 | /* | |
1459 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1460 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1461 | */ | |
1462 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1463 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1464 | { | |
1465 | int ret; | |
1466 | ||
1467 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1468 | if (ret) | |
1469 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1470 | ||
1471 | return ret; | |
1472 | } | |
1473 | ||
31b8006e SW |
1474 | /* |
1475 | * Parse an ip[:port] list into an addr array. Use the default | |
1476 | * monitor port if a port isn't specified. | |
1477 | */ | |
1478 | int ceph_parse_ips(const char *c, const char *end, | |
1479 | struct ceph_entity_addr *addr, | |
1480 | int max_count, int *count) | |
1481 | { | |
ee3b56f2 | 1482 | int i, ret = -EINVAL; |
31b8006e SW |
1483 | const char *p = c; |
1484 | ||
1485 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1486 | for (i = 0; i < max_count; i++) { | |
1487 | const char *ipend; | |
1488 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1489 | int port; |
39139f64 SW |
1490 | char delim = ','; |
1491 | ||
1492 | if (*p == '[') { | |
1493 | delim = ']'; | |
1494 | p++; | |
1495 | } | |
31b8006e | 1496 | |
ee3b56f2 NW |
1497 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1498 | if (ret) | |
31b8006e | 1499 | goto bad; |
ee3b56f2 NW |
1500 | ret = -EINVAL; |
1501 | ||
31b8006e SW |
1502 | p = ipend; |
1503 | ||
39139f64 SW |
1504 | if (delim == ']') { |
1505 | if (*p != ']') { | |
1506 | dout("missing matching ']'\n"); | |
1507 | goto bad; | |
1508 | } | |
1509 | p++; | |
1510 | } | |
1511 | ||
31b8006e SW |
1512 | /* port? */ |
1513 | if (p < end && *p == ':') { | |
1514 | port = 0; | |
1515 | p++; | |
1516 | while (p < end && *p >= '0' && *p <= '9') { | |
1517 | port = (port * 10) + (*p - '0'); | |
1518 | p++; | |
1519 | } | |
1520 | if (port > 65535 || port == 0) | |
1521 | goto bad; | |
1522 | } else { | |
1523 | port = CEPH_MON_PORT; | |
1524 | } | |
1525 | ||
1526 | addr_set_port(ss, port); | |
1527 | ||
3d14c5d2 | 1528 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1529 | |
1530 | if (p == end) | |
1531 | break; | |
1532 | if (*p != ',') | |
1533 | goto bad; | |
1534 | p++; | |
1535 | } | |
1536 | ||
1537 | if (p != end) | |
1538 | goto bad; | |
1539 | ||
1540 | if (count) | |
1541 | *count = i + 1; | |
1542 | return 0; | |
1543 | ||
1544 | bad: | |
39139f64 | 1545 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1546 | return ret; |
31b8006e | 1547 | } |
3d14c5d2 | 1548 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1549 | |
eed0ef2c | 1550 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1551 | { |
eed0ef2c | 1552 | dout("process_banner on %p\n", con); |
31b8006e SW |
1553 | |
1554 | if (verify_hello(con) < 0) | |
1555 | return -1; | |
1556 | ||
63f2d211 SW |
1557 | ceph_decode_addr(&con->actual_peer_addr); |
1558 | ceph_decode_addr(&con->peer_addr_for_me); | |
1559 | ||
31b8006e SW |
1560 | /* |
1561 | * Make sure the other end is who we wanted. note that the other | |
1562 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1563 | * them the benefit of the doubt. | |
1564 | */ | |
103e2d3a SW |
1565 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1566 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1567 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1568 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1569 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1570 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1571 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1572 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1573 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1574 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1575 | return -1; |
1576 | } | |
1577 | ||
1578 | /* | |
1579 | * did we learn our address? | |
1580 | */ | |
1581 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1582 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1583 | ||
1584 | memcpy(&con->msgr->inst.addr.in_addr, | |
1585 | &con->peer_addr_for_me.in_addr, | |
1586 | sizeof(con->peer_addr_for_me.in_addr)); | |
1587 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1588 | encode_my_addr(con->msgr); |
eed0ef2c | 1589 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1590 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1591 | } |
1592 | ||
eed0ef2c SW |
1593 | return 0; |
1594 | } | |
1595 | ||
1596 | static int process_connect(struct ceph_connection *con) | |
1597 | { | |
3d14c5d2 YS |
1598 | u64 sup_feat = con->msgr->supported_features; |
1599 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1600 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1601 | int ret; |
04a419f9 | 1602 | |
eed0ef2c SW |
1603 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1604 | ||
31b8006e | 1605 | switch (con->in_reply.tag) { |
04a419f9 SW |
1606 | case CEPH_MSGR_TAG_FEATURES: |
1607 | pr_err("%s%lld %s feature set mismatch," | |
1608 | " my %llx < server's %llx, missing %llx\n", | |
1609 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1610 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1611 | sup_feat, server_feat, server_feat & ~sup_feat); |
1612 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 1613 | reset_connection(con); |
04a419f9 SW |
1614 | return -1; |
1615 | ||
31b8006e | 1616 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1617 | pr_err("%s%lld %s protocol version mismatch," |
1618 | " my %d != server's %d\n", | |
1619 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1620 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1621 | le32_to_cpu(con->out_connect.protocol_version), |
1622 | le32_to_cpu(con->in_reply.protocol_version)); | |
1623 | con->error_msg = "protocol version mismatch"; | |
0fa6ebc6 | 1624 | reset_connection(con); |
31b8006e SW |
1625 | return -1; |
1626 | ||
4e7a5dcd SW |
1627 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1628 | con->auth_retry++; | |
1629 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1630 | con->auth_retry); | |
1631 | if (con->auth_retry == 2) { | |
1632 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1633 | return -1; |
1634 | } | |
1635 | con->auth_retry = 1; | |
6d4221b5 | 1636 | con_out_kvec_reset(con); |
e825a66d | 1637 | ret = prepare_write_connect(con); |
0da5d703 SW |
1638 | if (ret < 0) |
1639 | return ret; | |
63733a0f | 1640 | prepare_read_connect(con); |
4e7a5dcd | 1641 | break; |
31b8006e SW |
1642 | |
1643 | case CEPH_MSGR_TAG_RESETSESSION: | |
1644 | /* | |
1645 | * If we connected with a large connect_seq but the peer | |
1646 | * has no record of a session with us (no connection, or | |
1647 | * connect_seq == 0), they will send RESETSESION to indicate | |
1648 | * that they must have reset their session, and may have | |
1649 | * dropped messages. | |
1650 | */ | |
1651 | dout("process_connect got RESET peer seq %u\n", | |
5bdca4e0 | 1652 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
1653 | pr_err("%s%lld %s connection reset\n", |
1654 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1655 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1656 | reset_connection(con); |
6d4221b5 | 1657 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1658 | ret = prepare_write_connect(con); |
1659 | if (ret < 0) | |
1660 | return ret; | |
31b8006e SW |
1661 | prepare_read_connect(con); |
1662 | ||
1663 | /* Tell ceph about it. */ | |
ec302645 | 1664 | mutex_unlock(&con->mutex); |
31b8006e SW |
1665 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1666 | if (con->ops->peer_reset) | |
1667 | con->ops->peer_reset(con); | |
ec302645 | 1668 | mutex_lock(&con->mutex); |
8dacc7da | 1669 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 1670 | return -EAGAIN; |
31b8006e SW |
1671 | break; |
1672 | ||
1673 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1674 | /* | |
1675 | * If we sent a smaller connect_seq than the peer has, try | |
1676 | * again with a larger value. | |
1677 | */ | |
5bdca4e0 | 1678 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 1679 | le32_to_cpu(con->out_connect.connect_seq), |
5bdca4e0 SW |
1680 | le32_to_cpu(con->in_reply.connect_seq)); |
1681 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
6d4221b5 | 1682 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1683 | ret = prepare_write_connect(con); |
1684 | if (ret < 0) | |
1685 | return ret; | |
31b8006e SW |
1686 | prepare_read_connect(con); |
1687 | break; | |
1688 | ||
1689 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1690 | /* | |
1691 | * If we sent a smaller global_seq than the peer has, try | |
1692 | * again with a larger value. | |
1693 | */ | |
eed0ef2c | 1694 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 1695 | con->peer_global_seq, |
5bdca4e0 | 1696 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 1697 | get_global_seq(con->msgr, |
5bdca4e0 | 1698 | le32_to_cpu(con->in_reply.global_seq)); |
6d4221b5 | 1699 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1700 | ret = prepare_write_connect(con); |
1701 | if (ret < 0) | |
1702 | return ret; | |
31b8006e SW |
1703 | prepare_read_connect(con); |
1704 | break; | |
1705 | ||
1706 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1707 | if (req_feat & ~server_feat) { |
1708 | pr_err("%s%lld %s protocol feature mismatch," | |
1709 | " my required %llx > server's %llx, need %llx\n", | |
1710 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1711 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1712 | req_feat, server_feat, req_feat & ~server_feat); |
1713 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 1714 | reset_connection(con); |
04a419f9 SW |
1715 | return -1; |
1716 | } | |
8dacc7da | 1717 | |
122070a2 | 1718 | WARN_ON(con->state != CON_STATE_NEGOTIATING); |
8dacc7da SW |
1719 | con->state = CON_STATE_OPEN; |
1720 | ||
31b8006e SW |
1721 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1722 | con->connect_seq++; | |
aba558e2 | 1723 | con->peer_features = server_feat; |
31b8006e SW |
1724 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1725 | con->peer_global_seq, | |
1726 | le32_to_cpu(con->in_reply.connect_seq), | |
1727 | con->connect_seq); | |
1728 | WARN_ON(con->connect_seq != | |
1729 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1730 | |
1731 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
c9ffc77a | 1732 | con_flag_set(con, CON_FLAG_LOSSYTX); |
92ac41d0 | 1733 | |
85effe18 | 1734 | con->delay = 0; /* reset backoff memory */ |
92ac41d0 | 1735 | |
31b8006e SW |
1736 | prepare_read_tag(con); |
1737 | break; | |
1738 | ||
1739 | case CEPH_MSGR_TAG_WAIT: | |
1740 | /* | |
1741 | * If there is a connection race (we are opening | |
1742 | * connections to each other), one of us may just have | |
1743 | * to WAIT. This shouldn't happen if we are the | |
1744 | * client. | |
1745 | */ | |
04177882 SW |
1746 | pr_err("process_connect got WAIT as client\n"); |
1747 | con->error_msg = "protocol error, got WAIT as client"; | |
1748 | return -1; | |
31b8006e SW |
1749 | |
1750 | default: | |
1751 | pr_err("connect protocol error, will retry\n"); | |
1752 | con->error_msg = "protocol error, garbage tag during connect"; | |
1753 | return -1; | |
1754 | } | |
1755 | return 0; | |
1756 | } | |
1757 | ||
1758 | ||
1759 | /* | |
1760 | * read (part of) an ack | |
1761 | */ | |
1762 | static int read_partial_ack(struct ceph_connection *con) | |
1763 | { | |
fd51653f AE |
1764 | int size = sizeof (con->in_temp_ack); |
1765 | int end = size; | |
31b8006e | 1766 | |
fd51653f | 1767 | return read_partial(con, end, size, &con->in_temp_ack); |
31b8006e SW |
1768 | } |
1769 | ||
1770 | ||
1771 | /* | |
1772 | * We can finally discard anything that's been acked. | |
1773 | */ | |
1774 | static void process_ack(struct ceph_connection *con) | |
1775 | { | |
1776 | struct ceph_msg *m; | |
1777 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1778 | u64 seq; | |
1779 | ||
31b8006e SW |
1780 | while (!list_empty(&con->out_sent)) { |
1781 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1782 | list_head); | |
1783 | seq = le64_to_cpu(m->hdr.seq); | |
1784 | if (seq > ack) | |
1785 | break; | |
1786 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1787 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1788 | m->ack_stamp = jiffies; |
31b8006e SW |
1789 | ceph_msg_remove(m); |
1790 | } | |
31b8006e SW |
1791 | prepare_read_tag(con); |
1792 | } | |
1793 | ||
1794 | ||
1795 | ||
1796 | ||
2450418c | 1797 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1798 | struct kvec *section, |
1799 | unsigned int sec_len, u32 *crc) | |
2450418c | 1800 | { |
68b4476b | 1801 | int ret, left; |
2450418c YS |
1802 | |
1803 | BUG_ON(!section); | |
1804 | ||
1805 | while (section->iov_len < sec_len) { | |
1806 | BUG_ON(section->iov_base == NULL); | |
1807 | left = sec_len - section->iov_len; | |
1808 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1809 | section->iov_len, left); | |
1810 | if (ret <= 0) | |
1811 | return ret; | |
1812 | section->iov_len += ret; | |
2450418c | 1813 | } |
fe3ad593 AE |
1814 | if (section->iov_len == sec_len) |
1815 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1816 | |
2450418c YS |
1817 | return 1; |
1818 | } | |
31b8006e | 1819 | |
4740a623 | 1820 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
68b4476b YS |
1821 | |
1822 | static int read_partial_message_pages(struct ceph_connection *con, | |
1823 | struct page **pages, | |
95c96174 | 1824 | unsigned int data_len, bool do_datacrc) |
68b4476b | 1825 | { |
bae6acd9 | 1826 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; |
e788182f | 1827 | struct page *page; |
afb3d90e AE |
1828 | size_t page_offset; |
1829 | size_t length; | |
1830 | unsigned int left; | |
68b4476b | 1831 | int ret; |
68b4476b | 1832 | |
68b4476b YS |
1833 | /* (page) data */ |
1834 | BUG_ON(pages == NULL); | |
bae6acd9 | 1835 | page = pages[msg_pos->page]; |
afb3d90e AE |
1836 | page_offset = msg_pos->page_pos; |
1837 | BUG_ON(msg_pos->data_pos >= data_len); | |
1838 | left = data_len - msg_pos->data_pos; | |
1839 | BUG_ON(page_offset >= PAGE_SIZE); | |
1840 | length = min_t(unsigned int, PAGE_SIZE - page_offset, left); | |
1841 | ||
1842 | ret = ceph_tcp_recvpage(con->sock, page, page_offset, length); | |
68b4476b YS |
1843 | if (ret <= 0) |
1844 | return ret; | |
e788182f | 1845 | |
afb3d90e AE |
1846 | if (do_datacrc) { |
1847 | void *kaddr; | |
1848 | void *base; | |
1849 | ||
1850 | kaddr = kmap(page); | |
1851 | BUG_ON(!kaddr); | |
1852 | base = kaddr + page_offset; | |
1853 | con->in_data_crc = crc32c(con->in_data_crc, base, ret); | |
1854 | kunmap(page); | |
1855 | } | |
1856 | ||
1857 | in_msg_pos_next(con, length, ret); | |
68b4476b YS |
1858 | |
1859 | return ret; | |
1860 | } | |
1861 | ||
1862 | #ifdef CONFIG_BLOCK | |
1863 | static int read_partial_message_bio(struct ceph_connection *con, | |
95c96174 | 1864 | unsigned int data_len, bool do_datacrc) |
68b4476b | 1865 | { |
b3d56fab | 1866 | struct ceph_msg *msg = con->in_msg; |
bae6acd9 | 1867 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; |
b3d56fab | 1868 | struct bio_vec *bv; |
e788182f | 1869 | struct page *page; |
afb3d90e AE |
1870 | size_t page_offset; |
1871 | size_t length; | |
1872 | unsigned int left; | |
1873 | int ret; | |
68b4476b | 1874 | |
b3d56fab AE |
1875 | BUG_ON(!msg); |
1876 | BUG_ON(!msg->bio_iter); | |
1877 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
e788182f | 1878 | page = bv->bv_page; |
afb3d90e AE |
1879 | page_offset = bv->bv_offset + msg_pos->page_pos; |
1880 | BUG_ON(msg_pos->data_pos >= data_len); | |
1881 | left = data_len - msg_pos->data_pos; | |
1882 | BUG_ON(msg_pos->page_pos >= bv->bv_len); | |
1883 | length = min_t(unsigned int, bv->bv_len - msg_pos->page_pos, left); | |
68b4476b | 1884 | |
afb3d90e | 1885 | ret = ceph_tcp_recvpage(con->sock, page, page_offset, length); |
68b4476b YS |
1886 | if (ret <= 0) |
1887 | return ret; | |
e788182f | 1888 | |
afb3d90e AE |
1889 | if (do_datacrc) { |
1890 | void *kaddr; | |
1891 | void *base; | |
1892 | ||
1893 | kaddr = kmap(page); | |
1894 | BUG_ON(!kaddr); | |
1895 | base = kaddr + page_offset; | |
1896 | con->in_data_crc = crc32c(con->in_data_crc, base, ret); | |
1897 | kunmap(page); | |
1898 | } | |
1899 | ||
1900 | in_msg_pos_next(con, length, ret); | |
68b4476b YS |
1901 | |
1902 | return ret; | |
1903 | } | |
1904 | #endif | |
1905 | ||
34d2d200 AE |
1906 | static int read_partial_msg_data(struct ceph_connection *con) |
1907 | { | |
1908 | struct ceph_msg *msg = con->in_msg; | |
1909 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; | |
1910 | const bool do_datacrc = !con->msgr->nocrc; | |
1911 | unsigned int data_len; | |
1912 | int ret; | |
1913 | ||
1914 | BUG_ON(!msg); | |
1915 | ||
1916 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1917 | while (msg_pos->data_pos < data_len) { | |
1918 | if (msg->pages) { | |
1919 | ret = read_partial_message_pages(con, msg->pages, | |
1920 | data_len, do_datacrc); | |
1921 | if (ret <= 0) | |
1922 | return ret; | |
1923 | #ifdef CONFIG_BLOCK | |
1924 | } else if (msg->bio) { | |
1925 | ret = read_partial_message_bio(con, | |
1926 | data_len, do_datacrc); | |
1927 | if (ret <= 0) | |
1928 | return ret; | |
1929 | #endif | |
1930 | } else { | |
1931 | BUG_ON(1); | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | return 1; /* must return > 0 to indicate success */ | |
1936 | } | |
1937 | ||
31b8006e SW |
1938 | /* |
1939 | * read (part of) a message. | |
1940 | */ | |
1941 | static int read_partial_message(struct ceph_connection *con) | |
1942 | { | |
1943 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
1944 | int size; |
1945 | int end; | |
31b8006e | 1946 | int ret; |
95c96174 | 1947 | unsigned int front_len, middle_len, data_len; |
37675b0f | 1948 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 1949 | u64 seq; |
fe3ad593 | 1950 | u32 crc; |
31b8006e SW |
1951 | |
1952 | dout("read_partial_message con %p msg %p\n", con, m); | |
1953 | ||
1954 | /* header */ | |
fd51653f AE |
1955 | size = sizeof (con->in_hdr); |
1956 | end = size; | |
1957 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
1958 | if (ret <= 0) |
1959 | return ret; | |
fe3ad593 AE |
1960 | |
1961 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1962 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1963 | pr_err("read_partial_message bad hdr " | |
1964 | " crc %u != expected %u\n", | |
1965 | crc, con->in_hdr.crc); | |
1966 | return -EBADMSG; | |
1967 | } | |
1968 | ||
31b8006e SW |
1969 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1970 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1971 | return -EIO; | |
1972 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
7b11ba37 | 1973 | if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN) |
31b8006e SW |
1974 | return -EIO; |
1975 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1976 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1977 | return -EIO; | |
1978 | ||
ae18756b SW |
1979 | /* verify seq# */ |
1980 | seq = le64_to_cpu(con->in_hdr.seq); | |
1981 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1982 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1983 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1984 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1985 | seq, con->in_seq + 1); |
1986 | con->in_base_pos = -front_len - middle_len - data_len - | |
1987 | sizeof(m->footer); | |
1988 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1989 | return 0; |
1990 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1991 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1992 | seq, con->in_seq + 1); | |
1993 | con->error_msg = "bad message sequence # for incoming message"; | |
1994 | return -EBADMSG; | |
1995 | } | |
1996 | ||
31b8006e SW |
1997 | /* allocate message? */ |
1998 | if (!con->in_msg) { | |
4740a623 SW |
1999 | int skip = 0; |
2000 | ||
31b8006e | 2001 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
6ebc8b32 | 2002 | front_len, data_len); |
4740a623 SW |
2003 | ret = ceph_con_in_msg_alloc(con, &skip); |
2004 | if (ret < 0) | |
2005 | return ret; | |
2450418c | 2006 | if (skip) { |
31b8006e | 2007 | /* skip this message */ |
a79832f2 | 2008 | dout("alloc_msg said skip message\n"); |
ae32be31 | 2009 | BUG_ON(con->in_msg); |
31b8006e SW |
2010 | con->in_base_pos = -front_len - middle_len - data_len - |
2011 | sizeof(m->footer); | |
2012 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2013 | con->in_seq++; |
31b8006e SW |
2014 | return 0; |
2015 | } | |
38941f80 | 2016 | |
4740a623 | 2017 | BUG_ON(!con->in_msg); |
38941f80 | 2018 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
2019 | m = con->in_msg; |
2020 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
2021 | if (m->middle) |
2022 | m->middle->vec.iov_len = 0; | |
9d7f0f13 | 2023 | |
78625051 | 2024 | /* prepare for data payload, if any */ |
a4107026 | 2025 | |
78625051 AE |
2026 | if (data_len) |
2027 | prepare_message_data(con->in_msg, &con->in_msg_pos); | |
31b8006e SW |
2028 | } |
2029 | ||
2030 | /* front */ | |
2450418c YS |
2031 | ret = read_partial_message_section(con, &m->front, front_len, |
2032 | &con->in_front_crc); | |
2033 | if (ret <= 0) | |
2034 | return ret; | |
31b8006e SW |
2035 | |
2036 | /* middle */ | |
2450418c | 2037 | if (m->middle) { |
213c99ee SW |
2038 | ret = read_partial_message_section(con, &m->middle->vec, |
2039 | middle_len, | |
2450418c | 2040 | &con->in_middle_crc); |
31b8006e SW |
2041 | if (ret <= 0) |
2042 | return ret; | |
31b8006e SW |
2043 | } |
2044 | ||
2045 | /* (page) data */ | |
34d2d200 AE |
2046 | if (data_len) { |
2047 | ret = read_partial_msg_data(con); | |
2048 | if (ret <= 0) | |
2049 | return ret; | |
31b8006e SW |
2050 | } |
2051 | ||
31b8006e | 2052 | /* footer */ |
fd51653f AE |
2053 | size = sizeof (m->footer); |
2054 | end += size; | |
2055 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
2056 | if (ret <= 0) |
2057 | return ret; | |
2058 | ||
31b8006e SW |
2059 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
2060 | m, front_len, m->footer.front_crc, middle_len, | |
2061 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
2062 | ||
2063 | /* crc ok? */ | |
2064 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
2065 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
2066 | m, con->in_front_crc, m->footer.front_crc); | |
2067 | return -EBADMSG; | |
2068 | } | |
2069 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
2070 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
2071 | m, con->in_middle_crc, m->footer.middle_crc); | |
2072 | return -EBADMSG; | |
2073 | } | |
bca064d2 | 2074 | if (do_datacrc && |
31b8006e SW |
2075 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
2076 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
2077 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
2078 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
2079 | return -EBADMSG; | |
2080 | } | |
2081 | ||
2082 | return 1; /* done! */ | |
2083 | } | |
2084 | ||
2085 | /* | |
2086 | * Process message. This happens in the worker thread. The callback should | |
2087 | * be careful not to do anything that waits on other incoming messages or it | |
2088 | * may deadlock. | |
2089 | */ | |
2090 | static void process_message(struct ceph_connection *con) | |
2091 | { | |
5e095e8b | 2092 | struct ceph_msg *msg; |
31b8006e | 2093 | |
38941f80 AE |
2094 | BUG_ON(con->in_msg->con != con); |
2095 | con->in_msg->con = NULL; | |
5e095e8b | 2096 | msg = con->in_msg; |
31b8006e | 2097 | con->in_msg = NULL; |
36eb71aa | 2098 | con->ops->put(con); |
31b8006e SW |
2099 | |
2100 | /* if first message, set peer_name */ | |
2101 | if (con->peer_name.type == 0) | |
dbad185d | 2102 | con->peer_name = msg->hdr.src; |
31b8006e | 2103 | |
31b8006e | 2104 | con->in_seq++; |
ec302645 | 2105 | mutex_unlock(&con->mutex); |
31b8006e SW |
2106 | |
2107 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
2108 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 2109 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
2110 | le16_to_cpu(msg->hdr.type), |
2111 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2112 | le32_to_cpu(msg->hdr.front_len), | |
2113 | le32_to_cpu(msg->hdr.data_len), | |
2114 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
2115 | con->ops->dispatch(con, msg); | |
ec302645 SW |
2116 | |
2117 | mutex_lock(&con->mutex); | |
31b8006e SW |
2118 | } |
2119 | ||
2120 | ||
2121 | /* | |
2122 | * Write something to the socket. Called in a worker thread when the | |
2123 | * socket appears to be writeable and we have something ready to send. | |
2124 | */ | |
2125 | static int try_write(struct ceph_connection *con) | |
2126 | { | |
31b8006e SW |
2127 | int ret = 1; |
2128 | ||
d59315ca | 2129 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 2130 | |
31b8006e SW |
2131 | more: |
2132 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
2133 | ||
2134 | /* open the socket first? */ | |
8dacc7da SW |
2135 | if (con->state == CON_STATE_PREOPEN) { |
2136 | BUG_ON(con->sock); | |
2137 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2138 | |
e2200423 | 2139 | con_out_kvec_reset(con); |
e825a66d | 2140 | prepare_write_banner(con); |
eed0ef2c | 2141 | prepare_read_banner(con); |
31b8006e | 2142 | |
cf3e5c40 | 2143 | BUG_ON(con->in_msg); |
31b8006e SW |
2144 | con->in_tag = CEPH_MSGR_TAG_READY; |
2145 | dout("try_write initiating connect on %p new state %lu\n", | |
2146 | con, con->state); | |
41617d0c AE |
2147 | ret = ceph_tcp_connect(con); |
2148 | if (ret < 0) { | |
31b8006e | 2149 | con->error_msg = "connect error"; |
31b8006e SW |
2150 | goto out; |
2151 | } | |
2152 | } | |
2153 | ||
2154 | more_kvec: | |
2155 | /* kvec data queued? */ | |
2156 | if (con->out_skip) { | |
2157 | ret = write_partial_skip(con); | |
2158 | if (ret <= 0) | |
42961d23 | 2159 | goto out; |
31b8006e SW |
2160 | } |
2161 | if (con->out_kvec_left) { | |
2162 | ret = write_partial_kvec(con); | |
2163 | if (ret <= 0) | |
42961d23 | 2164 | goto out; |
31b8006e SW |
2165 | } |
2166 | ||
2167 | /* msg pages? */ | |
2168 | if (con->out_msg) { | |
c86a2930 SW |
2169 | if (con->out_msg_done) { |
2170 | ceph_msg_put(con->out_msg); | |
2171 | con->out_msg = NULL; /* we're done with this one */ | |
2172 | goto do_next; | |
2173 | } | |
2174 | ||
34d2d200 | 2175 | ret = write_partial_message_data(con); |
31b8006e SW |
2176 | if (ret == 1) |
2177 | goto more_kvec; /* we need to send the footer, too! */ | |
2178 | if (ret == 0) | |
42961d23 | 2179 | goto out; |
31b8006e | 2180 | if (ret < 0) { |
34d2d200 | 2181 | dout("try_write write_partial_message_data err %d\n", |
31b8006e | 2182 | ret); |
42961d23 | 2183 | goto out; |
31b8006e SW |
2184 | } |
2185 | } | |
2186 | ||
c86a2930 | 2187 | do_next: |
8dacc7da | 2188 | if (con->state == CON_STATE_OPEN) { |
31b8006e SW |
2189 | /* is anything else pending? */ |
2190 | if (!list_empty(&con->out_queue)) { | |
2191 | prepare_write_message(con); | |
2192 | goto more; | |
2193 | } | |
2194 | if (con->in_seq > con->in_seq_acked) { | |
2195 | prepare_write_ack(con); | |
2196 | goto more; | |
2197 | } | |
c9ffc77a | 2198 | if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) { |
31b8006e SW |
2199 | prepare_write_keepalive(con); |
2200 | goto more; | |
2201 | } | |
2202 | } | |
2203 | ||
2204 | /* Nothing to do! */ | |
c9ffc77a | 2205 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
31b8006e | 2206 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2207 | ret = 0; |
2208 | out: | |
42961d23 | 2209 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2210 | return ret; |
2211 | } | |
2212 | ||
2213 | ||
2214 | ||
2215 | /* | |
2216 | * Read what we can from the socket. | |
2217 | */ | |
2218 | static int try_read(struct ceph_connection *con) | |
2219 | { | |
31b8006e SW |
2220 | int ret = -1; |
2221 | ||
8dacc7da SW |
2222 | more: |
2223 | dout("try_read start on %p state %lu\n", con, con->state); | |
2224 | if (con->state != CON_STATE_CONNECTING && | |
2225 | con->state != CON_STATE_NEGOTIATING && | |
2226 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2227 | return 0; |
2228 | ||
8dacc7da | 2229 | BUG_ON(!con->sock); |
ec302645 | 2230 | |
31b8006e SW |
2231 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2232 | con->in_base_pos); | |
0da5d703 | 2233 | |
8dacc7da | 2234 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2235 | dout("try_read connecting\n"); |
2236 | ret = read_partial_banner(con); | |
2237 | if (ret <= 0) | |
ab166d5a | 2238 | goto out; |
7593af92 AE |
2239 | ret = process_banner(con); |
2240 | if (ret < 0) | |
2241 | goto out; | |
2242 | ||
8dacc7da | 2243 | con->state = CON_STATE_NEGOTIATING; |
7593af92 | 2244 | |
6d4221b5 JS |
2245 | /* |
2246 | * Received banner is good, exchange connection info. | |
2247 | * Do not reset out_kvec, as sending our banner raced | |
2248 | * with receiving peer banner after connect completed. | |
2249 | */ | |
7593af92 AE |
2250 | ret = prepare_write_connect(con); |
2251 | if (ret < 0) | |
2252 | goto out; | |
2253 | prepare_read_connect(con); | |
2254 | ||
2255 | /* Send connection info before awaiting response */ | |
0da5d703 SW |
2256 | goto out; |
2257 | } | |
2258 | ||
8dacc7da | 2259 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2260 | dout("try_read negotiating\n"); |
31b8006e SW |
2261 | ret = read_partial_connect(con); |
2262 | if (ret <= 0) | |
31b8006e | 2263 | goto out; |
98bdb0aa SW |
2264 | ret = process_connect(con); |
2265 | if (ret < 0) | |
2266 | goto out; | |
31b8006e SW |
2267 | goto more; |
2268 | } | |
2269 | ||
122070a2 | 2270 | WARN_ON(con->state != CON_STATE_OPEN); |
8dacc7da | 2271 | |
31b8006e SW |
2272 | if (con->in_base_pos < 0) { |
2273 | /* | |
2274 | * skipping + discarding content. | |
2275 | * | |
2276 | * FIXME: there must be a better way to do this! | |
2277 | */ | |
84495f49 AE |
2278 | static char buf[SKIP_BUF_SIZE]; |
2279 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2280 | ||
31b8006e SW |
2281 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2282 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2283 | if (ret <= 0) | |
98bdb0aa | 2284 | goto out; |
31b8006e SW |
2285 | con->in_base_pos += ret; |
2286 | if (con->in_base_pos) | |
2287 | goto more; | |
2288 | } | |
2289 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2290 | /* | |
2291 | * what's next? | |
2292 | */ | |
2293 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2294 | if (ret <= 0) | |
98bdb0aa | 2295 | goto out; |
31b8006e SW |
2296 | dout("try_read got tag %d\n", (int)con->in_tag); |
2297 | switch (con->in_tag) { | |
2298 | case CEPH_MSGR_TAG_MSG: | |
2299 | prepare_read_message(con); | |
2300 | break; | |
2301 | case CEPH_MSGR_TAG_ACK: | |
2302 | prepare_read_ack(con); | |
2303 | break; | |
2304 | case CEPH_MSGR_TAG_CLOSE: | |
8dacc7da SW |
2305 | con_close_socket(con); |
2306 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2307 | goto out; |
31b8006e SW |
2308 | default: |
2309 | goto bad_tag; | |
2310 | } | |
2311 | } | |
2312 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2313 | ret = read_partial_message(con); | |
2314 | if (ret <= 0) { | |
2315 | switch (ret) { | |
2316 | case -EBADMSG: | |
2317 | con->error_msg = "bad crc"; | |
2318 | ret = -EIO; | |
98bdb0aa | 2319 | break; |
31b8006e SW |
2320 | case -EIO: |
2321 | con->error_msg = "io error"; | |
98bdb0aa | 2322 | break; |
31b8006e | 2323 | } |
98bdb0aa | 2324 | goto out; |
31b8006e SW |
2325 | } |
2326 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2327 | goto more; | |
2328 | process_message(con); | |
7b862e07 SW |
2329 | if (con->state == CON_STATE_OPEN) |
2330 | prepare_read_tag(con); | |
31b8006e SW |
2331 | goto more; |
2332 | } | |
2333 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2334 | ret = read_partial_ack(con); | |
2335 | if (ret <= 0) | |
98bdb0aa | 2336 | goto out; |
31b8006e SW |
2337 | process_ack(con); |
2338 | goto more; | |
2339 | } | |
2340 | ||
31b8006e | 2341 | out: |
98bdb0aa | 2342 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2343 | return ret; |
2344 | ||
2345 | bad_tag: | |
2346 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2347 | con->error_msg = "protocol error, garbage tag"; | |
2348 | ret = -1; | |
2349 | goto out; | |
2350 | } | |
2351 | ||
2352 | ||
2353 | /* | |
802c6d96 AE |
2354 | * Atomically queue work on a connection after the specified delay. |
2355 | * Bump @con reference to avoid races with connection teardown. | |
2356 | * Returns 0 if work was queued, or an error code otherwise. | |
31b8006e | 2357 | */ |
802c6d96 | 2358 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
31b8006e | 2359 | { |
31b8006e | 2360 | if (!con->ops->get(con)) { |
802c6d96 AE |
2361 | dout("%s %p ref count 0\n", __func__, con); |
2362 | ||
2363 | return -ENOENT; | |
31b8006e SW |
2364 | } |
2365 | ||
802c6d96 AE |
2366 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
2367 | dout("%s %p - already queued\n", __func__, con); | |
31b8006e | 2368 | con->ops->put(con); |
802c6d96 AE |
2369 | |
2370 | return -EBUSY; | |
31b8006e | 2371 | } |
802c6d96 AE |
2372 | |
2373 | dout("%s %p %lu\n", __func__, con, delay); | |
2374 | ||
2375 | return 0; | |
2376 | } | |
2377 | ||
2378 | static void queue_con(struct ceph_connection *con) | |
2379 | { | |
2380 | (void) queue_con_delay(con, 0); | |
31b8006e SW |
2381 | } |
2382 | ||
7bb21d68 AE |
2383 | static bool con_sock_closed(struct ceph_connection *con) |
2384 | { | |
c9ffc77a | 2385 | if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED)) |
7bb21d68 AE |
2386 | return false; |
2387 | ||
2388 | #define CASE(x) \ | |
2389 | case CON_STATE_ ## x: \ | |
2390 | con->error_msg = "socket closed (con state " #x ")"; \ | |
2391 | break; | |
2392 | ||
2393 | switch (con->state) { | |
2394 | CASE(CLOSED); | |
2395 | CASE(PREOPEN); | |
2396 | CASE(CONNECTING); | |
2397 | CASE(NEGOTIATING); | |
2398 | CASE(OPEN); | |
2399 | CASE(STANDBY); | |
2400 | default: | |
2401 | pr_warning("%s con %p unrecognized state %lu\n", | |
2402 | __func__, con, con->state); | |
2403 | con->error_msg = "unrecognized con state"; | |
2404 | BUG(); | |
2405 | break; | |
2406 | } | |
2407 | #undef CASE | |
2408 | ||
2409 | return true; | |
2410 | } | |
2411 | ||
f20a39fd AE |
2412 | static bool con_backoff(struct ceph_connection *con) |
2413 | { | |
2414 | int ret; | |
2415 | ||
2416 | if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF)) | |
2417 | return false; | |
2418 | ||
2419 | ret = queue_con_delay(con, round_jiffies_relative(con->delay)); | |
2420 | if (ret) { | |
2421 | dout("%s: con %p FAILED to back off %lu\n", __func__, | |
2422 | con, con->delay); | |
2423 | BUG_ON(ret == -ENOENT); | |
2424 | con_flag_set(con, CON_FLAG_BACKOFF); | |
2425 | } | |
2426 | ||
2427 | return true; | |
2428 | } | |
2429 | ||
93209264 AE |
2430 | /* Finish fault handling; con->mutex must *not* be held here */ |
2431 | ||
2432 | static void con_fault_finish(struct ceph_connection *con) | |
2433 | { | |
2434 | /* | |
2435 | * in case we faulted due to authentication, invalidate our | |
2436 | * current tickets so that we can get new ones. | |
2437 | */ | |
2438 | if (con->auth_retry && con->ops->invalidate_authorizer) { | |
2439 | dout("calling invalidate_authorizer()\n"); | |
2440 | con->ops->invalidate_authorizer(con); | |
2441 | } | |
2442 | ||
2443 | if (con->ops->fault) | |
2444 | con->ops->fault(con); | |
2445 | } | |
2446 | ||
31b8006e SW |
2447 | /* |
2448 | * Do some work on a connection. Drop a connection ref when we're done. | |
2449 | */ | |
2450 | static void con_work(struct work_struct *work) | |
2451 | { | |
2452 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2453 | work.work); | |
49659416 | 2454 | bool fault; |
31b8006e | 2455 | |
9dd4658d | 2456 | mutex_lock(&con->mutex); |
49659416 AE |
2457 | while (true) { |
2458 | int ret; | |
31b8006e | 2459 | |
49659416 AE |
2460 | if ((fault = con_sock_closed(con))) { |
2461 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); | |
2462 | break; | |
2463 | } | |
2464 | if (con_backoff(con)) { | |
2465 | dout("%s: con %p BACKOFF\n", __func__, con); | |
2466 | break; | |
2467 | } | |
2468 | if (con->state == CON_STATE_STANDBY) { | |
2469 | dout("%s: con %p STANDBY\n", __func__, con); | |
2470 | break; | |
2471 | } | |
2472 | if (con->state == CON_STATE_CLOSED) { | |
2473 | dout("%s: con %p CLOSED\n", __func__, con); | |
2474 | BUG_ON(con->sock); | |
2475 | break; | |
2476 | } | |
2477 | if (con->state == CON_STATE_PREOPEN) { | |
2478 | dout("%s: con %p PREOPEN\n", __func__, con); | |
2479 | BUG_ON(con->sock); | |
2480 | } | |
0da5d703 | 2481 | |
49659416 AE |
2482 | ret = try_read(con); |
2483 | if (ret < 0) { | |
2484 | if (ret == -EAGAIN) | |
2485 | continue; | |
2486 | con->error_msg = "socket error on read"; | |
2487 | fault = true; | |
2488 | break; | |
2489 | } | |
2490 | ||
2491 | ret = try_write(con); | |
2492 | if (ret < 0) { | |
2493 | if (ret == -EAGAIN) | |
2494 | continue; | |
2495 | con->error_msg = "socket error on write"; | |
2496 | fault = true; | |
2497 | } | |
2498 | ||
2499 | break; /* If we make it to here, we're done */ | |
3a140a0d | 2500 | } |
b6e7b6a1 AE |
2501 | if (fault) |
2502 | con_fault(con); | |
9dd4658d | 2503 | mutex_unlock(&con->mutex); |
0da5d703 | 2504 | |
b6e7b6a1 AE |
2505 | if (fault) |
2506 | con_fault_finish(con); | |
2507 | ||
2508 | con->ops->put(con); | |
31b8006e SW |
2509 | } |
2510 | ||
31b8006e SW |
2511 | /* |
2512 | * Generic error/fault handler. A retry mechanism is used with | |
2513 | * exponential backoff | |
2514 | */ | |
93209264 | 2515 | static void con_fault(struct ceph_connection *con) |
31b8006e | 2516 | { |
28362986 | 2517 | pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2518 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2519 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2520 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2521 | |
122070a2 | 2522 | WARN_ON(con->state != CON_STATE_CONNECTING && |
8dacc7da SW |
2523 | con->state != CON_STATE_NEGOTIATING && |
2524 | con->state != CON_STATE_OPEN); | |
ec302645 | 2525 | |
31b8006e | 2526 | con_close_socket(con); |
5e095e8b | 2527 | |
c9ffc77a | 2528 | if (con_flag_test(con, CON_FLAG_LOSSYTX)) { |
8dacc7da SW |
2529 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
2530 | con->state = CON_STATE_CLOSED; | |
93209264 | 2531 | return; |
3b5ede07 SW |
2532 | } |
2533 | ||
5e095e8b | 2534 | if (con->in_msg) { |
38941f80 AE |
2535 | BUG_ON(con->in_msg->con != con); |
2536 | con->in_msg->con = NULL; | |
5e095e8b SW |
2537 | ceph_msg_put(con->in_msg); |
2538 | con->in_msg = NULL; | |
36eb71aa | 2539 | con->ops->put(con); |
5e095e8b | 2540 | } |
31b8006e | 2541 | |
e80a52d1 SW |
2542 | /* Requeue anything that hasn't been acked */ |
2543 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2544 | |
e76661d0 SW |
2545 | /* If there are no messages queued or keepalive pending, place |
2546 | * the connection in a STANDBY state */ | |
2547 | if (list_empty(&con->out_queue) && | |
c9ffc77a | 2548 | !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) { |
e00de341 | 2549 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
c9ffc77a | 2550 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
8dacc7da | 2551 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
2552 | } else { |
2553 | /* retry after a delay. */ | |
8dacc7da | 2554 | con->state = CON_STATE_PREOPEN; |
e80a52d1 SW |
2555 | if (con->delay == 0) |
2556 | con->delay = BASE_DELAY_INTERVAL; | |
2557 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2558 | con->delay *= 2; | |
c9ffc77a | 2559 | con_flag_set(con, CON_FLAG_BACKOFF); |
8618e30b | 2560 | queue_con(con); |
31b8006e | 2561 | } |
31b8006e SW |
2562 | } |
2563 | ||
2564 | ||
2565 | ||
2566 | /* | |
15d9882c | 2567 | * initialize a new messenger instance |
31b8006e | 2568 | */ |
15d9882c AE |
2569 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2570 | struct ceph_entity_addr *myaddr, | |
2571 | u32 supported_features, | |
2572 | u32 required_features, | |
2573 | bool nocrc) | |
31b8006e | 2574 | { |
3d14c5d2 YS |
2575 | msgr->supported_features = supported_features; |
2576 | msgr->required_features = required_features; | |
2577 | ||
31b8006e SW |
2578 | spin_lock_init(&msgr->global_seq_lock); |
2579 | ||
31b8006e SW |
2580 | if (myaddr) |
2581 | msgr->inst.addr = *myaddr; | |
2582 | ||
2583 | /* select a random nonce */ | |
ac8839d7 | 2584 | msgr->inst.addr.type = 0; |
103e2d3a | 2585 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2586 | encode_my_addr(msgr); |
15d9882c | 2587 | msgr->nocrc = nocrc; |
31b8006e | 2588 | |
a2a32584 | 2589 | atomic_set(&msgr->stopping, 0); |
31b8006e | 2590 | |
15d9882c | 2591 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2592 | } |
15d9882c | 2593 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2594 | |
e00de341 SW |
2595 | static void clear_standby(struct ceph_connection *con) |
2596 | { | |
2597 | /* come back from STANDBY? */ | |
8dacc7da | 2598 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 2599 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 2600 | con->state = CON_STATE_PREOPEN; |
e00de341 | 2601 | con->connect_seq++; |
c9ffc77a AE |
2602 | WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING)); |
2603 | WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)); | |
e00de341 SW |
2604 | } |
2605 | } | |
2606 | ||
31b8006e SW |
2607 | /* |
2608 | * Queue up an outgoing message on the given connection. | |
2609 | */ | |
2610 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2611 | { | |
31b8006e | 2612 | /* set src+dst */ |
dbad185d | 2613 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 2614 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
2615 | msg->needs_out_seq = true; |
2616 | ||
ec302645 | 2617 | mutex_lock(&con->mutex); |
92ce034b | 2618 | |
8dacc7da | 2619 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
2620 | dout("con_send %p closed, dropping %p\n", con, msg); |
2621 | ceph_msg_put(msg); | |
2622 | mutex_unlock(&con->mutex); | |
2623 | return; | |
2624 | } | |
2625 | ||
38941f80 | 2626 | BUG_ON(msg->con != NULL); |
36eb71aa | 2627 | msg->con = con->ops->get(con); |
92ce034b AE |
2628 | BUG_ON(msg->con == NULL); |
2629 | ||
31b8006e SW |
2630 | BUG_ON(!list_empty(&msg->list_head)); |
2631 | list_add_tail(&msg->list_head, &con->out_queue); | |
2632 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2633 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2634 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2635 | le32_to_cpu(msg->hdr.front_len), | |
2636 | le32_to_cpu(msg->hdr.middle_len), | |
2637 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
2638 | |
2639 | clear_standby(con); | |
ec302645 | 2640 | mutex_unlock(&con->mutex); |
31b8006e SW |
2641 | |
2642 | /* if there wasn't anything waiting to send before, queue | |
2643 | * new work */ | |
c9ffc77a | 2644 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
31b8006e SW |
2645 | queue_con(con); |
2646 | } | |
3d14c5d2 | 2647 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2648 | |
2649 | /* | |
2650 | * Revoke a message that was previously queued for send | |
2651 | */ | |
6740a845 | 2652 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2653 | { |
6740a845 AE |
2654 | struct ceph_connection *con = msg->con; |
2655 | ||
2656 | if (!con) | |
2657 | return; /* Message not in our possession */ | |
2658 | ||
ec302645 | 2659 | mutex_lock(&con->mutex); |
31b8006e | 2660 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2661 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2662 | list_del_init(&msg->list_head); |
38941f80 | 2663 | BUG_ON(msg->con == NULL); |
36eb71aa | 2664 | msg->con->ops->put(msg->con); |
38941f80 | 2665 | msg->con = NULL; |
31b8006e | 2666 | msg->hdr.seq = 0; |
38941f80 | 2667 | |
31b8006e | 2668 | ceph_msg_put(msg); |
ed98adad SW |
2669 | } |
2670 | if (con->out_msg == msg) { | |
38941f80 | 2671 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2672 | con->out_msg = NULL; |
31b8006e SW |
2673 | if (con->out_kvec_is_msg) { |
2674 | con->out_skip = con->out_kvec_bytes; | |
2675 | con->out_kvec_is_msg = false; | |
2676 | } | |
ed98adad | 2677 | msg->hdr.seq = 0; |
92ce034b AE |
2678 | |
2679 | ceph_msg_put(msg); | |
31b8006e | 2680 | } |
ec302645 | 2681 | mutex_unlock(&con->mutex); |
31b8006e SW |
2682 | } |
2683 | ||
350b1c32 | 2684 | /* |
0d59ab81 | 2685 | * Revoke a message that we may be reading data into |
350b1c32 | 2686 | */ |
8921d114 | 2687 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2688 | { |
8921d114 AE |
2689 | struct ceph_connection *con; |
2690 | ||
2691 | BUG_ON(msg == NULL); | |
2692 | if (!msg->con) { | |
2693 | dout("%s msg %p null con\n", __func__, msg); | |
2694 | ||
2695 | return; /* Message not in our possession */ | |
2696 | } | |
2697 | ||
2698 | con = msg->con; | |
350b1c32 | 2699 | mutex_lock(&con->mutex); |
8921d114 | 2700 | if (con->in_msg == msg) { |
95c96174 ED |
2701 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
2702 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
2703 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
2704 | |
2705 | /* skip rest of message */ | |
8921d114 AE |
2706 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
2707 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 2708 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
2709 | front_len - |
2710 | middle_len - | |
2711 | data_len - | |
350b1c32 | 2712 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2713 | ceph_msg_put(con->in_msg); |
2714 | con->in_msg = NULL; | |
2715 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2716 | con->in_seq++; |
350b1c32 | 2717 | } else { |
8921d114 AE |
2718 | dout("%s %p in_msg %p msg %p no-op\n", |
2719 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
2720 | } |
2721 | mutex_unlock(&con->mutex); | |
2722 | } | |
2723 | ||
31b8006e SW |
2724 | /* |
2725 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2726 | */ | |
2727 | void ceph_con_keepalive(struct ceph_connection *con) | |
2728 | { | |
e00de341 | 2729 | dout("con_keepalive %p\n", con); |
00650931 | 2730 | mutex_lock(&con->mutex); |
e00de341 | 2731 | clear_standby(con); |
00650931 | 2732 | mutex_unlock(&con->mutex); |
c9ffc77a AE |
2733 | if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 && |
2734 | con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) | |
31b8006e SW |
2735 | queue_con(con); |
2736 | } | |
3d14c5d2 | 2737 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e | 2738 | |
02afca6c | 2739 | void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages, |
f1baeb2b | 2740 | size_t length, size_t alignment) |
02afca6c | 2741 | { |
07aa1558 AE |
2742 | BUG_ON(!pages); |
2743 | BUG_ON(!length); | |
2744 | BUG_ON(msg->pages); | |
2745 | BUG_ON(msg->length); | |
02afca6c AE |
2746 | |
2747 | msg->pages = pages; | |
4a73ef27 | 2748 | msg->length = length; |
02afca6c AE |
2749 | msg->page_alignment = alignment & ~PAGE_MASK; |
2750 | } | |
2751 | EXPORT_SYMBOL(ceph_msg_data_set_pages); | |
31b8006e | 2752 | |
27fa8385 AE |
2753 | void ceph_msg_data_set_pagelist(struct ceph_msg *msg, |
2754 | struct ceph_pagelist *pagelist) | |
2755 | { | |
07aa1558 AE |
2756 | BUG_ON(!pagelist); |
2757 | BUG_ON(!pagelist->length); | |
2758 | BUG_ON(msg->pagelist); | |
27fa8385 AE |
2759 | |
2760 | msg->pagelist = pagelist; | |
2761 | } | |
2762 | EXPORT_SYMBOL(ceph_msg_data_set_pagelist); | |
2763 | ||
2764 | void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio) | |
2765 | { | |
07aa1558 AE |
2766 | BUG_ON(!bio); |
2767 | BUG_ON(msg->bio); | |
27fa8385 AE |
2768 | |
2769 | msg->bio = bio; | |
2770 | } | |
2771 | EXPORT_SYMBOL(ceph_msg_data_set_bio); | |
2772 | ||
2773 | void ceph_msg_data_set_trail(struct ceph_msg *msg, struct ceph_pagelist *trail) | |
2774 | { | |
07aa1558 AE |
2775 | BUG_ON(!trail); |
2776 | BUG_ON(!trail->length); | |
2777 | BUG_ON(msg->trail); | |
27fa8385 AE |
2778 | |
2779 | msg->trail = trail; | |
2780 | } | |
2781 | EXPORT_SYMBOL(ceph_msg_data_set_trail); | |
2782 | ||
31b8006e SW |
2783 | /* |
2784 | * construct a new message with given type, size | |
2785 | * the new msg has a ref count of 1. | |
2786 | */ | |
b61c2763 SW |
2787 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2788 | bool can_fail) | |
31b8006e SW |
2789 | { |
2790 | struct ceph_msg *m; | |
2791 | ||
9516e45b | 2792 | m = kzalloc(sizeof(*m), flags); |
31b8006e SW |
2793 | if (m == NULL) |
2794 | goto out; | |
31b8006e SW |
2795 | |
2796 | m->hdr.type = cpu_to_le16(type); | |
45c6ceb5 | 2797 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
31b8006e | 2798 | m->hdr.front_len = cpu_to_le32(front_len); |
ca20892d | 2799 | |
9516e45b AE |
2800 | INIT_LIST_HEAD(&m->list_head); |
2801 | kref_init(&m->kref); | |
ca20892d | 2802 | |
31b8006e | 2803 | /* front */ |
9516e45b | 2804 | m->front_max = front_len; |
31b8006e SW |
2805 | if (front_len) { |
2806 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2807 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2808 | PAGE_KERNEL); |
2809 | m->front_is_vmalloc = true; | |
2810 | } else { | |
34d23762 | 2811 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2812 | } |
2813 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2814 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2815 | front_len); |
2816 | goto out2; | |
2817 | } | |
2818 | } else { | |
2819 | m->front.iov_base = NULL; | |
2820 | } | |
2821 | m->front.iov_len = front_len; | |
2822 | ||
bb257664 | 2823 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2824 | return m; |
2825 | ||
2826 | out2: | |
2827 | ceph_msg_put(m); | |
2828 | out: | |
b61c2763 SW |
2829 | if (!can_fail) { |
2830 | pr_err("msg_new can't create type %d front %d\n", type, | |
2831 | front_len); | |
f0ed1b7c | 2832 | WARN_ON(1); |
b61c2763 SW |
2833 | } else { |
2834 | dout("msg_new can't create type %d front %d\n", type, | |
2835 | front_len); | |
2836 | } | |
a79832f2 | 2837 | return NULL; |
31b8006e | 2838 | } |
3d14c5d2 | 2839 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2840 | |
31b8006e SW |
2841 | /* |
2842 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2843 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2844 | * per-type header in the front and then gracefully fail (i.e., | |
2845 | * propagate the error to the caller based on info in the front) when | |
2846 | * the middle is too large. | |
2847 | */ | |
2450418c | 2848 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2849 | { |
2850 | int type = le16_to_cpu(msg->hdr.type); | |
2851 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2852 | ||
2853 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2854 | ceph_msg_type_name(type), middle_len); | |
2855 | BUG_ON(!middle_len); | |
2856 | BUG_ON(msg->middle); | |
2857 | ||
b6c1d5b8 | 2858 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2859 | if (!msg->middle) |
2860 | return -ENOMEM; | |
2861 | return 0; | |
2862 | } | |
2863 | ||
2450418c | 2864 | /* |
1c20f2d2 AE |
2865 | * Allocate a message for receiving an incoming message on a |
2866 | * connection, and save the result in con->in_msg. Uses the | |
2867 | * connection's private alloc_msg op if available. | |
2868 | * | |
4740a623 SW |
2869 | * Returns 0 on success, or a negative error code. |
2870 | * | |
2871 | * On success, if we set *skip = 1: | |
2872 | * - the next message should be skipped and ignored. | |
2873 | * - con->in_msg == NULL | |
2874 | * or if we set *skip = 0: | |
2875 | * - con->in_msg is non-null. | |
2876 | * On error (ENOMEM, EAGAIN, ...), | |
2877 | * - con->in_msg == NULL | |
2450418c | 2878 | */ |
4740a623 | 2879 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
2450418c | 2880 | { |
4740a623 | 2881 | struct ceph_msg_header *hdr = &con->in_hdr; |
2450418c | 2882 | int middle_len = le32_to_cpu(hdr->middle_len); |
1d866d1c | 2883 | struct ceph_msg *msg; |
4740a623 | 2884 | int ret = 0; |
2450418c | 2885 | |
1c20f2d2 | 2886 | BUG_ON(con->in_msg != NULL); |
53ded495 | 2887 | BUG_ON(!con->ops->alloc_msg); |
2450418c | 2888 | |
53ded495 AE |
2889 | mutex_unlock(&con->mutex); |
2890 | msg = con->ops->alloc_msg(con, hdr, skip); | |
2891 | mutex_lock(&con->mutex); | |
2892 | if (con->state != CON_STATE_OPEN) { | |
2893 | if (msg) | |
1d866d1c | 2894 | ceph_msg_put(msg); |
53ded495 AE |
2895 | return -EAGAIN; |
2896 | } | |
4137577a AE |
2897 | if (msg) { |
2898 | BUG_ON(*skip); | |
2899 | con->in_msg = msg; | |
36eb71aa | 2900 | con->in_msg->con = con->ops->get(con); |
92ce034b | 2901 | BUG_ON(con->in_msg->con == NULL); |
4137577a AE |
2902 | } else { |
2903 | /* | |
2904 | * Null message pointer means either we should skip | |
2905 | * this message or we couldn't allocate memory. The | |
2906 | * former is not an error. | |
2907 | */ | |
2908 | if (*skip) | |
2909 | return 0; | |
2910 | con->error_msg = "error allocating memory for incoming message"; | |
2911 | ||
53ded495 | 2912 | return -ENOMEM; |
2450418c | 2913 | } |
1c20f2d2 | 2914 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2915 | |
1c20f2d2 AE |
2916 | if (middle_len && !con->in_msg->middle) { |
2917 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 2918 | if (ret < 0) { |
1c20f2d2 AE |
2919 | ceph_msg_put(con->in_msg); |
2920 | con->in_msg = NULL; | |
2450418c YS |
2921 | } |
2922 | } | |
9d7f0f13 | 2923 | |
4740a623 | 2924 | return ret; |
2450418c YS |
2925 | } |
2926 | ||
31b8006e SW |
2927 | |
2928 | /* | |
2929 | * Free a generically kmalloc'd message. | |
2930 | */ | |
2931 | void ceph_msg_kfree(struct ceph_msg *m) | |
2932 | { | |
2933 | dout("msg_kfree %p\n", m); | |
2934 | if (m->front_is_vmalloc) | |
2935 | vfree(m->front.iov_base); | |
2936 | else | |
2937 | kfree(m->front.iov_base); | |
2938 | kfree(m); | |
2939 | } | |
2940 | ||
2941 | /* | |
2942 | * Drop a msg ref. Destroy as needed. | |
2943 | */ | |
c2e552e7 SW |
2944 | void ceph_msg_last_put(struct kref *kref) |
2945 | { | |
2946 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2947 | |
c2e552e7 SW |
2948 | dout("ceph_msg_put last one on %p\n", m); |
2949 | WARN_ON(!list_empty(&m->list_head)); | |
2950 | ||
2951 | /* drop middle, data, if any */ | |
2952 | if (m->middle) { | |
2953 | ceph_buffer_put(m->middle); | |
2954 | m->middle = NULL; | |
31b8006e | 2955 | } |
4a73ef27 | 2956 | m->length = 0; |
c2e552e7 SW |
2957 | m->pages = NULL; |
2958 | ||
58bb3b37 SW |
2959 | if (m->pagelist) { |
2960 | ceph_pagelist_release(m->pagelist); | |
2961 | kfree(m->pagelist); | |
2962 | m->pagelist = NULL; | |
2963 | } | |
2964 | ||
68b4476b YS |
2965 | m->trail = NULL; |
2966 | ||
c2e552e7 SW |
2967 | if (m->pool) |
2968 | ceph_msgpool_put(m->pool, m); | |
2969 | else | |
2970 | ceph_msg_kfree(m); | |
31b8006e | 2971 | } |
3d14c5d2 | 2972 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2973 | |
2974 | void ceph_msg_dump(struct ceph_msg *msg) | |
2975 | { | |
4a73ef27 AE |
2976 | pr_debug("msg_dump %p (front_max %d length %zd)\n", msg, |
2977 | msg->front_max, msg->length); | |
9ec7cab1 SW |
2978 | print_hex_dump(KERN_DEBUG, "header: ", |
2979 | DUMP_PREFIX_OFFSET, 16, 1, | |
2980 | &msg->hdr, sizeof(msg->hdr), true); | |
2981 | print_hex_dump(KERN_DEBUG, " front: ", | |
2982 | DUMP_PREFIX_OFFSET, 16, 1, | |
2983 | msg->front.iov_base, msg->front.iov_len, true); | |
2984 | if (msg->middle) | |
2985 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2986 | DUMP_PREFIX_OFFSET, 16, 1, | |
2987 | msg->middle->vec.iov_base, | |
2988 | msg->middle->vec.iov_len, true); | |
2989 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2990 | DUMP_PREFIX_OFFSET, 16, 1, | |
2991 | &msg->footer, sizeof(msg->footer), true); | |
2992 | } | |
3d14c5d2 | 2993 | EXPORT_SYMBOL(ceph_msg_dump); |