]>
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 | 23 | |
fe38a2b6 AE |
24 | #define list_entry_next(pos, member) \ |
25 | list_entry(pos->member.next, typeof(*pos), member) | |
26 | ||
31b8006e SW |
27 | /* |
28 | * Ceph uses the messenger to exchange ceph_msg messages with other | |
29 | * hosts in the system. The messenger provides ordered and reliable | |
30 | * delivery. We tolerate TCP disconnects by reconnecting (with | |
31 | * exponential backoff) in the case of a fault (disconnection, bad | |
32 | * crc, protocol error). Acks allow sent messages to be discarded by | |
33 | * the sender. | |
34 | */ | |
35 | ||
bc18f4b1 AE |
36 | /* |
37 | * We track the state of the socket on a given connection using | |
38 | * values defined below. The transition to a new socket state is | |
39 | * handled by a function which verifies we aren't coming from an | |
40 | * unexpected state. | |
41 | * | |
42 | * -------- | |
43 | * | NEW* | transient initial state | |
44 | * -------- | |
45 | * | con_sock_state_init() | |
46 | * v | |
47 | * ---------- | |
48 | * | CLOSED | initialized, but no socket (and no | |
49 | * ---------- TCP connection) | |
50 | * ^ \ | |
51 | * | \ con_sock_state_connecting() | |
52 | * | ---------------------- | |
53 | * | \ | |
54 | * + con_sock_state_closed() \ | |
fbb85a47 SW |
55 | * |+--------------------------- \ |
56 | * | \ \ \ | |
57 | * | ----------- \ \ | |
58 | * | | CLOSING | socket event; \ \ | |
59 | * | ----------- await close \ \ | |
60 | * | ^ \ | | |
61 | * | | \ | | |
62 | * | + con_sock_state_closing() \ | | |
63 | * | / \ | | | |
64 | * | / --------------- | | | |
65 | * | / \ v v | |
bc18f4b1 AE |
66 | * | / -------------- |
67 | * | / -----------------| CONNECTING | socket created, TCP | |
68 | * | | / -------------- connect initiated | |
69 | * | | | con_sock_state_connected() | |
70 | * | | v | |
71 | * ------------- | |
72 | * | CONNECTED | TCP connection established | |
73 | * ------------- | |
74 | * | |
75 | * State values for ceph_connection->sock_state; NEW is assumed to be 0. | |
76 | */ | |
ce2c8903 AE |
77 | |
78 | #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */ | |
79 | #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */ | |
80 | #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */ | |
81 | #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */ | |
82 | #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */ | |
83 | ||
8dacc7da SW |
84 | /* |
85 | * connection states | |
86 | */ | |
87 | #define CON_STATE_CLOSED 1 /* -> PREOPEN */ | |
88 | #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */ | |
89 | #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */ | |
90 | #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */ | |
91 | #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */ | |
92 | #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */ | |
93 | ||
4a861692 SW |
94 | /* |
95 | * ceph_connection flag bits | |
96 | */ | |
97 | #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop | |
98 | * messages on errors */ | |
99 | #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */ | |
100 | #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */ | |
101 | #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */ | |
102 | #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */ | |
8dacc7da | 103 | |
c9ffc77a AE |
104 | static bool con_flag_valid(unsigned long con_flag) |
105 | { | |
106 | switch (con_flag) { | |
107 | case CON_FLAG_LOSSYTX: | |
108 | case CON_FLAG_KEEPALIVE_PENDING: | |
109 | case CON_FLAG_WRITE_PENDING: | |
110 | case CON_FLAG_SOCK_CLOSED: | |
111 | case CON_FLAG_BACKOFF: | |
112 | return true; | |
113 | default: | |
114 | return false; | |
115 | } | |
116 | } | |
117 | ||
118 | static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag) | |
119 | { | |
120 | BUG_ON(!con_flag_valid(con_flag)); | |
121 | ||
122 | clear_bit(con_flag, &con->flags); | |
123 | } | |
124 | ||
125 | static void con_flag_set(struct ceph_connection *con, unsigned long con_flag) | |
126 | { | |
127 | BUG_ON(!con_flag_valid(con_flag)); | |
128 | ||
129 | set_bit(con_flag, &con->flags); | |
130 | } | |
131 | ||
132 | static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag) | |
133 | { | |
134 | BUG_ON(!con_flag_valid(con_flag)); | |
135 | ||
136 | return test_bit(con_flag, &con->flags); | |
137 | } | |
138 | ||
139 | static bool con_flag_test_and_clear(struct ceph_connection *con, | |
140 | unsigned long con_flag) | |
141 | { | |
142 | BUG_ON(!con_flag_valid(con_flag)); | |
143 | ||
144 | return test_and_clear_bit(con_flag, &con->flags); | |
145 | } | |
146 | ||
147 | static bool con_flag_test_and_set(struct ceph_connection *con, | |
148 | unsigned long con_flag) | |
149 | { | |
150 | BUG_ON(!con_flag_valid(con_flag)); | |
151 | ||
152 | return test_and_set_bit(con_flag, &con->flags); | |
153 | } | |
154 | ||
31b8006e SW |
155 | /* static tag bytes (protocol control messages) */ |
156 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
157 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
158 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
159 | ||
a6a5349d SW |
160 | #ifdef CONFIG_LOCKDEP |
161 | static struct lock_class_key socket_class; | |
162 | #endif | |
163 | ||
84495f49 AE |
164 | /* |
165 | * When skipping (ignoring) a block of input we read it into a "skip | |
166 | * buffer," which is this many bytes in size. | |
167 | */ | |
168 | #define SKIP_BUF_SIZE 1024 | |
31b8006e SW |
169 | |
170 | static void queue_con(struct ceph_connection *con); | |
171 | static void con_work(struct work_struct *); | |
93209264 | 172 | static void con_fault(struct ceph_connection *con); |
31b8006e | 173 | |
31b8006e | 174 | /* |
f64a9317 AE |
175 | * Nicely render a sockaddr as a string. An array of formatted |
176 | * strings is used, to approximate reentrancy. | |
31b8006e | 177 | */ |
f64a9317 AE |
178 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
179 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
180 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
181 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
182 | ||
183 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
184 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 185 | |
57666519 | 186 | static struct page *zero_page; /* used in certain error cases */ |
57666519 | 187 | |
3d14c5d2 | 188 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
189 | { |
190 | int i; | |
191 | char *s; | |
99f0f3b2 AE |
192 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
193 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 194 | |
f64a9317 | 195 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
196 | s = addr_str[i]; |
197 | ||
198 | switch (ss->ss_family) { | |
199 | case AF_INET: | |
bd406145 AE |
200 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
201 | ntohs(in4->sin_port)); | |
31b8006e SW |
202 | break; |
203 | ||
204 | case AF_INET6: | |
bd406145 AE |
205 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
206 | ntohs(in6->sin6_port)); | |
31b8006e SW |
207 | break; |
208 | ||
209 | default: | |
d3002b97 AE |
210 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
211 | ss->ss_family); | |
31b8006e SW |
212 | } |
213 | ||
214 | return s; | |
215 | } | |
3d14c5d2 | 216 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 217 | |
63f2d211 SW |
218 | static void encode_my_addr(struct ceph_messenger *msgr) |
219 | { | |
220 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
221 | ceph_encode_addr(&msgr->my_enc_addr); | |
222 | } | |
223 | ||
31b8006e SW |
224 | /* |
225 | * work queue for all reading and writing to/from the socket. | |
226 | */ | |
e0f43c94 | 227 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 228 | |
15417167 | 229 | static void _ceph_msgr_exit(void) |
6173d1f0 | 230 | { |
d3002b97 | 231 | if (ceph_msgr_wq) { |
6173d1f0 | 232 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
233 | ceph_msgr_wq = NULL; |
234 | } | |
6173d1f0 | 235 | |
6173d1f0 AE |
236 | BUG_ON(zero_page == NULL); |
237 | kunmap(zero_page); | |
238 | page_cache_release(zero_page); | |
239 | zero_page = NULL; | |
240 | } | |
241 | ||
3d14c5d2 | 242 | int ceph_msgr_init(void) |
31b8006e | 243 | { |
57666519 AE |
244 | BUG_ON(zero_page != NULL); |
245 | zero_page = ZERO_PAGE(0); | |
246 | page_cache_get(zero_page); | |
247 | ||
f363e45f | 248 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0); |
6173d1f0 AE |
249 | if (ceph_msgr_wq) |
250 | return 0; | |
57666519 | 251 | |
6173d1f0 AE |
252 | pr_err("msgr_init failed to create workqueue\n"); |
253 | _ceph_msgr_exit(); | |
57666519 | 254 | |
6173d1f0 | 255 | return -ENOMEM; |
31b8006e | 256 | } |
3d14c5d2 | 257 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
258 | |
259 | void ceph_msgr_exit(void) | |
260 | { | |
57666519 | 261 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 262 | |
6173d1f0 | 263 | _ceph_msgr_exit(); |
31b8006e | 264 | } |
3d14c5d2 | 265 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 266 | |
cd84db6e | 267 | void ceph_msgr_flush(void) |
a922d38f SW |
268 | { |
269 | flush_workqueue(ceph_msgr_wq); | |
270 | } | |
3d14c5d2 | 271 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f | 272 | |
ce2c8903 AE |
273 | /* Connection socket state transition functions */ |
274 | ||
275 | static void con_sock_state_init(struct ceph_connection *con) | |
276 | { | |
277 | int old_state; | |
278 | ||
279 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
280 | if (WARN_ON(old_state != CON_SOCK_STATE_NEW)) | |
281 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
282 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
283 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 AE |
284 | } |
285 | ||
286 | static void con_sock_state_connecting(struct ceph_connection *con) | |
287 | { | |
288 | int old_state; | |
289 | ||
290 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING); | |
291 | if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED)) | |
292 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
293 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
294 | CON_SOCK_STATE_CONNECTING); | |
ce2c8903 AE |
295 | } |
296 | ||
297 | static void con_sock_state_connected(struct ceph_connection *con) | |
298 | { | |
299 | int old_state; | |
300 | ||
301 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED); | |
302 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING)) | |
303 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
304 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
305 | CON_SOCK_STATE_CONNECTED); | |
ce2c8903 AE |
306 | } |
307 | ||
308 | static void con_sock_state_closing(struct ceph_connection *con) | |
309 | { | |
310 | int old_state; | |
311 | ||
312 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING); | |
313 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING && | |
314 | old_state != CON_SOCK_STATE_CONNECTED && | |
315 | old_state != CON_SOCK_STATE_CLOSING)) | |
316 | printk("%s: unexpected old state %d\n", __func__, old_state); | |
8007b8d6 SW |
317 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
318 | CON_SOCK_STATE_CLOSING); | |
ce2c8903 AE |
319 | } |
320 | ||
321 | static void con_sock_state_closed(struct ceph_connection *con) | |
322 | { | |
323 | int old_state; | |
324 | ||
325 | old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED); | |
326 | if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED && | |
fbb85a47 | 327 | old_state != CON_SOCK_STATE_CLOSING && |
8007b8d6 SW |
328 | old_state != CON_SOCK_STATE_CONNECTING && |
329 | old_state != CON_SOCK_STATE_CLOSED)) | |
ce2c8903 | 330 | printk("%s: unexpected old state %d\n", __func__, old_state); |
8007b8d6 SW |
331 | dout("%s con %p sock %d -> %d\n", __func__, con, old_state, |
332 | CON_SOCK_STATE_CLOSED); | |
ce2c8903 | 333 | } |
a922d38f | 334 | |
31b8006e SW |
335 | /* |
336 | * socket callback functions | |
337 | */ | |
338 | ||
339 | /* data available on socket, or listen socket received a connect */ | |
327800bd | 340 | static void ceph_sock_data_ready(struct sock *sk, int count_unused) |
31b8006e | 341 | { |
bd406145 | 342 | struct ceph_connection *con = sk->sk_user_data; |
a2a32584 GH |
343 | if (atomic_read(&con->msgr->stopping)) { |
344 | return; | |
345 | } | |
bd406145 | 346 | |
31b8006e | 347 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
327800bd | 348 | dout("%s on %p state = %lu, queueing work\n", __func__, |
31b8006e SW |
349 | con, con->state); |
350 | queue_con(con); | |
351 | } | |
352 | } | |
353 | ||
354 | /* socket has buffer space for writing */ | |
327800bd | 355 | static void ceph_sock_write_space(struct sock *sk) |
31b8006e | 356 | { |
d3002b97 | 357 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 358 | |
182fac26 JS |
359 | /* only queue to workqueue if there is data we want to write, |
360 | * and there is sufficient space in the socket buffer to accept | |
327800bd | 361 | * more data. clear SOCK_NOSPACE so that ceph_sock_write_space() |
182fac26 JS |
362 | * doesn't get called again until try_write() fills the socket |
363 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
364 | * and net/core/stream.c:sk_stream_write_space(). | |
365 | */ | |
c9ffc77a | 366 | if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) { |
182fac26 | 367 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
327800bd | 368 | dout("%s %p queueing write work\n", __func__, con); |
182fac26 JS |
369 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); |
370 | queue_con(con); | |
371 | } | |
31b8006e | 372 | } else { |
327800bd | 373 | dout("%s %p nothing to write\n", __func__, con); |
31b8006e | 374 | } |
31b8006e SW |
375 | } |
376 | ||
377 | /* socket's state has changed */ | |
327800bd | 378 | static void ceph_sock_state_change(struct sock *sk) |
31b8006e | 379 | { |
bd406145 | 380 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 381 | |
327800bd | 382 | dout("%s %p state = %lu sk_state = %u\n", __func__, |
31b8006e SW |
383 | con, con->state, sk->sk_state); |
384 | ||
31b8006e SW |
385 | switch (sk->sk_state) { |
386 | case TCP_CLOSE: | |
327800bd | 387 | dout("%s TCP_CLOSE\n", __func__); |
31b8006e | 388 | case TCP_CLOSE_WAIT: |
327800bd | 389 | dout("%s TCP_CLOSE_WAIT\n", __func__); |
ce2c8903 | 390 | con_sock_state_closing(con); |
c9ffc77a | 391 | con_flag_set(con, CON_FLAG_SOCK_CLOSED); |
d65c9e0b | 392 | queue_con(con); |
31b8006e SW |
393 | break; |
394 | case TCP_ESTABLISHED: | |
327800bd | 395 | dout("%s TCP_ESTABLISHED\n", __func__); |
ce2c8903 | 396 | con_sock_state_connected(con); |
31b8006e SW |
397 | queue_con(con); |
398 | break; | |
d3002b97 AE |
399 | default: /* Everything else is uninteresting */ |
400 | break; | |
31b8006e SW |
401 | } |
402 | } | |
403 | ||
404 | /* | |
405 | * set up socket callbacks | |
406 | */ | |
407 | static void set_sock_callbacks(struct socket *sock, | |
408 | struct ceph_connection *con) | |
409 | { | |
410 | struct sock *sk = sock->sk; | |
bd406145 | 411 | sk->sk_user_data = con; |
327800bd AE |
412 | sk->sk_data_ready = ceph_sock_data_ready; |
413 | sk->sk_write_space = ceph_sock_write_space; | |
414 | sk->sk_state_change = ceph_sock_state_change; | |
31b8006e SW |
415 | } |
416 | ||
417 | ||
418 | /* | |
419 | * socket helpers | |
420 | */ | |
421 | ||
422 | /* | |
423 | * initiate connection to a remote socket. | |
424 | */ | |
41617d0c | 425 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 426 | { |
f91d3471 | 427 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
428 | struct socket *sock; |
429 | int ret; | |
430 | ||
431 | BUG_ON(con->sock); | |
f91d3471 SW |
432 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
433 | IPPROTO_TCP, &sock); | |
31b8006e | 434 | if (ret) |
41617d0c | 435 | return ret; |
31b8006e SW |
436 | sock->sk->sk_allocation = GFP_NOFS; |
437 | ||
a6a5349d SW |
438 | #ifdef CONFIG_LOCKDEP |
439 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
440 | #endif | |
441 | ||
31b8006e SW |
442 | set_sock_callbacks(sock, con); |
443 | ||
3d14c5d2 | 444 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 445 | |
89a86be0 | 446 | con_sock_state_connecting(con); |
f91d3471 SW |
447 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
448 | O_NONBLOCK); | |
31b8006e SW |
449 | if (ret == -EINPROGRESS) { |
450 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 451 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 452 | sock->sk->sk_state); |
a5bc3129 | 453 | } else if (ret < 0) { |
31b8006e | 454 | pr_err("connect %s error %d\n", |
3d14c5d2 | 455 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 456 | sock_release(sock); |
31b8006e | 457 | con->error_msg = "connect error"; |
31b8006e | 458 | |
41617d0c | 459 | return ret; |
a5bc3129 AE |
460 | } |
461 | con->sock = sock; | |
41617d0c | 462 | return 0; |
31b8006e SW |
463 | } |
464 | ||
465 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
466 | { | |
467 | struct kvec iov = {buf, len}; | |
468 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 469 | int r; |
31b8006e | 470 | |
98bdb0aa SW |
471 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
472 | if (r == -EAGAIN) | |
473 | r = 0; | |
474 | return r; | |
31b8006e SW |
475 | } |
476 | ||
afb3d90e AE |
477 | static int ceph_tcp_recvpage(struct socket *sock, struct page *page, |
478 | int page_offset, size_t length) | |
479 | { | |
480 | void *kaddr; | |
481 | int ret; | |
482 | ||
483 | BUG_ON(page_offset + length > PAGE_SIZE); | |
484 | ||
485 | kaddr = kmap(page); | |
486 | BUG_ON(!kaddr); | |
487 | ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length); | |
488 | kunmap(page); | |
489 | ||
490 | return ret; | |
491 | } | |
492 | ||
31b8006e SW |
493 | /* |
494 | * write something. @more is true if caller will be sending more data | |
495 | * shortly. | |
496 | */ | |
497 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
498 | size_t kvlen, size_t len, int more) | |
499 | { | |
500 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 501 | int r; |
31b8006e SW |
502 | |
503 | if (more) | |
504 | msg.msg_flags |= MSG_MORE; | |
505 | else | |
506 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
507 | ||
42961d23 SW |
508 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
509 | if (r == -EAGAIN) | |
510 | r = 0; | |
511 | return r; | |
31b8006e SW |
512 | } |
513 | ||
31739139 | 514 | static int ceph_tcp_sendpage(struct socket *sock, struct page *page, |
e1dcb128 | 515 | int offset, size_t size, bool more) |
31739139 AE |
516 | { |
517 | int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR); | |
518 | int ret; | |
519 | ||
520 | ret = kernel_sendpage(sock, page, offset, size, flags); | |
521 | if (ret == -EAGAIN) | |
522 | ret = 0; | |
523 | ||
524 | return ret; | |
525 | } | |
526 | ||
31b8006e SW |
527 | |
528 | /* | |
529 | * Shutdown/close the socket for the given connection. | |
530 | */ | |
531 | static int con_close_socket(struct ceph_connection *con) | |
532 | { | |
8007b8d6 | 533 | int rc = 0; |
31b8006e SW |
534 | |
535 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
8007b8d6 SW |
536 | if (con->sock) { |
537 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
538 | sock_release(con->sock); | |
539 | con->sock = NULL; | |
540 | } | |
456ea468 AE |
541 | |
542 | /* | |
4a861692 | 543 | * Forcibly clear the SOCK_CLOSED flag. It gets set |
456ea468 AE |
544 | * independent of the connection mutex, and we could have |
545 | * received a socket close event before we had the chance to | |
546 | * shut the socket down. | |
547 | */ | |
c9ffc77a | 548 | con_flag_clear(con, CON_FLAG_SOCK_CLOSED); |
8007b8d6 | 549 | |
ce2c8903 | 550 | con_sock_state_closed(con); |
31b8006e SW |
551 | return rc; |
552 | } | |
553 | ||
554 | /* | |
555 | * Reset a connection. Discard all incoming and outgoing messages | |
556 | * and clear *_seq state. | |
557 | */ | |
558 | static void ceph_msg_remove(struct ceph_msg *msg) | |
559 | { | |
560 | list_del_init(&msg->list_head); | |
38941f80 | 561 | BUG_ON(msg->con == NULL); |
36eb71aa | 562 | msg->con->ops->put(msg->con); |
38941f80 AE |
563 | msg->con = NULL; |
564 | ||
31b8006e SW |
565 | ceph_msg_put(msg); |
566 | } | |
567 | static void ceph_msg_remove_list(struct list_head *head) | |
568 | { | |
569 | while (!list_empty(head)) { | |
570 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
571 | list_head); | |
572 | ceph_msg_remove(msg); | |
573 | } | |
574 | } | |
575 | ||
576 | static void reset_connection(struct ceph_connection *con) | |
577 | { | |
578 | /* reset connection, out_queue, msg_ and connect_seq */ | |
579 | /* discard existing out_queue and msg_seq */ | |
0fa6ebc6 | 580 | dout("reset_connection %p\n", con); |
31b8006e SW |
581 | ceph_msg_remove_list(&con->out_queue); |
582 | ceph_msg_remove_list(&con->out_sent); | |
583 | ||
cf3e5c40 | 584 | if (con->in_msg) { |
38941f80 AE |
585 | BUG_ON(con->in_msg->con != con); |
586 | con->in_msg->con = NULL; | |
cf3e5c40 SW |
587 | ceph_msg_put(con->in_msg); |
588 | con->in_msg = NULL; | |
36eb71aa | 589 | con->ops->put(con); |
cf3e5c40 SW |
590 | } |
591 | ||
31b8006e SW |
592 | con->connect_seq = 0; |
593 | con->out_seq = 0; | |
c86a2930 SW |
594 | if (con->out_msg) { |
595 | ceph_msg_put(con->out_msg); | |
596 | con->out_msg = NULL; | |
597 | } | |
31b8006e | 598 | con->in_seq = 0; |
0e0d5e0c | 599 | con->in_seq_acked = 0; |
31b8006e SW |
600 | } |
601 | ||
602 | /* | |
603 | * mark a peer down. drop any open connections. | |
604 | */ | |
605 | void ceph_con_close(struct ceph_connection *con) | |
606 | { | |
8c50c817 | 607 | mutex_lock(&con->mutex); |
3d14c5d2 YS |
608 | dout("con_close %p peer %s\n", con, |
609 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
8dacc7da | 610 | con->state = CON_STATE_CLOSED; |
a5988c49 | 611 | |
c9ffc77a AE |
612 | con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */ |
613 | con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING); | |
614 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); | |
615 | con_flag_clear(con, CON_FLAG_BACKOFF); | |
a5988c49 | 616 | |
31b8006e | 617 | reset_connection(con); |
6f2bc3ff | 618 | con->peer_global_seq = 0; |
91e45ce3 | 619 | cancel_delayed_work(&con->work); |
ee76e073 | 620 | con_close_socket(con); |
ec302645 | 621 | mutex_unlock(&con->mutex); |
31b8006e | 622 | } |
3d14c5d2 | 623 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 624 | |
31b8006e SW |
625 | /* |
626 | * Reopen a closed connection, with a new peer address. | |
627 | */ | |
b7a9e5dd SW |
628 | void ceph_con_open(struct ceph_connection *con, |
629 | __u8 entity_type, __u64 entity_num, | |
630 | struct ceph_entity_addr *addr) | |
31b8006e | 631 | { |
5469155f | 632 | mutex_lock(&con->mutex); |
3d14c5d2 | 633 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
8dacc7da | 634 | |
122070a2 | 635 | WARN_ON(con->state != CON_STATE_CLOSED); |
8dacc7da | 636 | con->state = CON_STATE_PREOPEN; |
a5988c49 | 637 | |
b7a9e5dd SW |
638 | con->peer_name.type = (__u8) entity_type; |
639 | con->peer_name.num = cpu_to_le64(entity_num); | |
640 | ||
31b8006e | 641 | memcpy(&con->peer_addr, addr, sizeof(*addr)); |
03c677e1 | 642 | con->delay = 0; /* reset backoff memory */ |
5469155f | 643 | mutex_unlock(&con->mutex); |
31b8006e SW |
644 | queue_con(con); |
645 | } | |
3d14c5d2 | 646 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 647 | |
87b315a5 SW |
648 | /* |
649 | * return true if this connection ever successfully opened | |
650 | */ | |
651 | bool ceph_con_opened(struct ceph_connection *con) | |
652 | { | |
653 | return con->connect_seq > 0; | |
654 | } | |
655 | ||
31b8006e SW |
656 | /* |
657 | * initialize a new connection. | |
658 | */ | |
1bfd89f4 AE |
659 | void ceph_con_init(struct ceph_connection *con, void *private, |
660 | const struct ceph_connection_operations *ops, | |
b7a9e5dd | 661 | struct ceph_messenger *msgr) |
31b8006e SW |
662 | { |
663 | dout("con_init %p\n", con); | |
664 | memset(con, 0, sizeof(*con)); | |
1bfd89f4 AE |
665 | con->private = private; |
666 | con->ops = ops; | |
31b8006e | 667 | con->msgr = msgr; |
ce2c8903 AE |
668 | |
669 | con_sock_state_init(con); | |
670 | ||
ec302645 | 671 | mutex_init(&con->mutex); |
31b8006e SW |
672 | INIT_LIST_HEAD(&con->out_queue); |
673 | INIT_LIST_HEAD(&con->out_sent); | |
674 | INIT_DELAYED_WORK(&con->work, con_work); | |
a5988c49 | 675 | |
8dacc7da | 676 | con->state = CON_STATE_CLOSED; |
31b8006e | 677 | } |
3d14c5d2 | 678 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
679 | |
680 | ||
681 | /* | |
682 | * We maintain a global counter to order connection attempts. Get | |
683 | * a unique seq greater than @gt. | |
684 | */ | |
685 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
686 | { | |
687 | u32 ret; | |
688 | ||
689 | spin_lock(&msgr->global_seq_lock); | |
690 | if (msgr->global_seq < gt) | |
691 | msgr->global_seq = gt; | |
692 | ret = ++msgr->global_seq; | |
693 | spin_unlock(&msgr->global_seq_lock); | |
694 | return ret; | |
695 | } | |
696 | ||
e2200423 | 697 | static void con_out_kvec_reset(struct ceph_connection *con) |
859eb799 AE |
698 | { |
699 | con->out_kvec_left = 0; | |
700 | con->out_kvec_bytes = 0; | |
701 | con->out_kvec_cur = &con->out_kvec[0]; | |
702 | } | |
703 | ||
e2200423 | 704 | static void con_out_kvec_add(struct ceph_connection *con, |
859eb799 AE |
705 | size_t size, void *data) |
706 | { | |
707 | int index; | |
708 | ||
709 | index = con->out_kvec_left; | |
710 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
711 | ||
712 | con->out_kvec[index].iov_len = size; | |
713 | con->out_kvec[index].iov_base = data; | |
714 | con->out_kvec_left++; | |
715 | con->out_kvec_bytes += size; | |
716 | } | |
31b8006e | 717 | |
df6ad1f9 | 718 | #ifdef CONFIG_BLOCK |
07c09b72 AE |
719 | static void init_bio_iter(struct bio *bio, struct bio **bio_iter, |
720 | unsigned int *bio_seg) | |
df6ad1f9 AE |
721 | { |
722 | if (!bio) { | |
07c09b72 AE |
723 | *bio_iter = NULL; |
724 | *bio_seg = 0; | |
df6ad1f9 AE |
725 | return; |
726 | } | |
07c09b72 AE |
727 | *bio_iter = bio; |
728 | *bio_seg = (unsigned int) bio->bi_idx; | |
df6ad1f9 AE |
729 | } |
730 | ||
07c09b72 | 731 | static void iter_bio_next(struct bio **bio_iter, unsigned int *seg) |
df6ad1f9 AE |
732 | { |
733 | if (*bio_iter == NULL) | |
734 | return; | |
735 | ||
736 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
737 | ||
738 | (*seg)++; | |
739 | if (*seg == (*bio_iter)->bi_vcnt) | |
740 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
741 | } | |
742 | #endif | |
743 | ||
fe38a2b6 AE |
744 | /* |
745 | * Message data is handled (sent or received) in pieces, where each | |
746 | * piece resides on a single page. The network layer might not | |
747 | * consume an entire piece at once. A data item's cursor keeps | |
748 | * track of which piece is next to process and how much remains to | |
749 | * be processed in that piece. It also tracks whether the current | |
750 | * piece is the last one in the data item. | |
751 | */ | |
752 | static void ceph_msg_data_cursor_init(struct ceph_msg_data *data) | |
753 | { | |
754 | struct ceph_msg_data_cursor *cursor = &data->cursor; | |
755 | struct ceph_pagelist *pagelist; | |
756 | struct page *page; | |
757 | ||
758 | if (data->type != CEPH_MSG_DATA_PAGELIST) | |
759 | return; | |
760 | ||
761 | pagelist = data->pagelist; | |
762 | BUG_ON(!pagelist); | |
763 | if (!pagelist->length) | |
764 | return; /* pagelist can be assigned but empty */ | |
765 | ||
766 | BUG_ON(list_empty(&pagelist->head)); | |
767 | page = list_first_entry(&pagelist->head, struct page, lru); | |
768 | ||
769 | cursor->page = page; | |
770 | cursor->offset = 0; | |
771 | cursor->last_piece = pagelist->length <= PAGE_SIZE; | |
772 | } | |
773 | ||
774 | /* | |
775 | * Return the page containing the next piece to process for a given | |
776 | * data item, and supply the page offset and length of that piece. | |
777 | * Indicate whether this is the last piece in this data item. | |
778 | */ | |
779 | static struct page *ceph_msg_data_next(struct ceph_msg_data *data, | |
780 | size_t *page_offset, | |
781 | size_t *length, | |
782 | bool *last_piece) | |
783 | { | |
784 | struct ceph_msg_data_cursor *cursor = &data->cursor; | |
785 | struct ceph_pagelist *pagelist; | |
786 | size_t piece_end; | |
787 | ||
788 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
789 | ||
790 | pagelist = data->pagelist; | |
791 | BUG_ON(!pagelist); | |
792 | ||
793 | BUG_ON(!cursor->page); | |
794 | BUG_ON(cursor->offset >= pagelist->length); | |
795 | ||
796 | *last_piece = cursor->last_piece; | |
797 | if (*last_piece) { | |
798 | /* pagelist offset is always 0 */ | |
799 | piece_end = pagelist->length & ~PAGE_MASK; | |
800 | if (!piece_end) | |
801 | piece_end = PAGE_SIZE; | |
802 | } else { | |
803 | piece_end = PAGE_SIZE; | |
804 | } | |
805 | *page_offset = cursor->offset & ~PAGE_MASK; | |
806 | *length = piece_end - *page_offset; | |
807 | ||
808 | return data->cursor.page; | |
809 | } | |
810 | ||
811 | /* | |
812 | * Returns true if the result moves the cursor on to the next piece | |
813 | * (the next page) of the pagelist. | |
814 | */ | |
815 | static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes) | |
816 | { | |
817 | struct ceph_msg_data_cursor *cursor = &data->cursor; | |
818 | struct ceph_pagelist *pagelist; | |
819 | ||
820 | BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST); | |
821 | ||
822 | pagelist = data->pagelist; | |
823 | BUG_ON(!pagelist); | |
824 | BUG_ON(!cursor->page); | |
825 | BUG_ON(cursor->offset + bytes > pagelist->length); | |
826 | BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE); | |
827 | ||
828 | /* Advance the cursor offset */ | |
829 | ||
830 | cursor->offset += bytes; | |
831 | /* pagelist offset is always 0 */ | |
832 | if (!bytes || cursor->offset & ~PAGE_MASK) | |
833 | return false; /* more bytes to process in the current page */ | |
834 | ||
835 | /* Move on to the next page */ | |
836 | ||
837 | BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head)); | |
838 | cursor->page = list_entry_next(cursor->page, lru); | |
839 | ||
840 | /* cursor offset is at page boundary; pagelist offset is always 0 */ | |
841 | if (pagelist->length - cursor->offset <= PAGE_SIZE) | |
842 | cursor->last_piece = true; | |
843 | ||
844 | return true; | |
845 | } | |
846 | ||
78625051 AE |
847 | static void prepare_message_data(struct ceph_msg *msg, |
848 | struct ceph_msg_pos *msg_pos) | |
739c905b | 849 | { |
739c905b AE |
850 | BUG_ON(!msg); |
851 | BUG_ON(!msg->hdr.data_len); | |
852 | ||
853 | /* initialize page iterator */ | |
bae6acd9 | 854 | msg_pos->page = 0; |
97fb1c7f | 855 | if (ceph_msg_has_pages(msg)) |
f9e15777 | 856 | msg_pos->page_pos = msg->p.alignment; |
739c905b | 857 | else |
bae6acd9 | 858 | msg_pos->page_pos = 0; |
572c588e | 859 | #ifdef CONFIG_BLOCK |
97fb1c7f | 860 | if (ceph_msg_has_bio(msg)) |
f9e15777 | 861 | init_bio_iter(msg->b.bio, &msg->b.bio_iter, &msg->b.bio_seg); |
572c588e | 862 | #endif |
bae6acd9 | 863 | msg_pos->data_pos = 0; |
fe38a2b6 AE |
864 | |
865 | /* If there's a trail, initialize its cursor */ | |
866 | ||
867 | if (ceph_msg_has_trail(msg)) | |
868 | ceph_msg_data_cursor_init(&msg->t); | |
869 | ||
bae6acd9 | 870 | msg_pos->did_page_crc = false; |
739c905b AE |
871 | } |
872 | ||
31b8006e SW |
873 | /* |
874 | * Prepare footer for currently outgoing message, and finish things | |
875 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
876 | */ | |
859eb799 | 877 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
878 | { |
879 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 880 | int v = con->out_kvec_left; |
31b8006e | 881 | |
fd154f3c AE |
882 | m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE; |
883 | ||
31b8006e SW |
884 | dout("prepare_write_message_footer %p\n", con); |
885 | con->out_kvec_is_msg = true; | |
886 | con->out_kvec[v].iov_base = &m->footer; | |
887 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
888 | con->out_kvec_bytes += sizeof(m->footer); | |
889 | con->out_kvec_left++; | |
890 | con->out_more = m->more_to_follow; | |
c86a2930 | 891 | con->out_msg_done = true; |
31b8006e SW |
892 | } |
893 | ||
894 | /* | |
895 | * Prepare headers for the next outgoing message. | |
896 | */ | |
897 | static void prepare_write_message(struct ceph_connection *con) | |
898 | { | |
899 | struct ceph_msg *m; | |
a9a0c51a | 900 | u32 crc; |
31b8006e | 901 | |
e2200423 | 902 | con_out_kvec_reset(con); |
31b8006e | 903 | con->out_kvec_is_msg = true; |
c86a2930 | 904 | con->out_msg_done = false; |
31b8006e SW |
905 | |
906 | /* Sneak an ack in there first? If we can get it into the same | |
907 | * TCP packet that's a good thing. */ | |
908 | if (con->in_seq > con->in_seq_acked) { | |
909 | con->in_seq_acked = con->in_seq; | |
e2200423 | 910 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 911 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 912 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 | 913 | &con->out_temp_ack); |
31b8006e SW |
914 | } |
915 | ||
38941f80 | 916 | BUG_ON(list_empty(&con->out_queue)); |
859eb799 | 917 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 918 | con->out_msg = m; |
38941f80 | 919 | BUG_ON(m->con != con); |
4cf9d544 SW |
920 | |
921 | /* put message on sent list */ | |
922 | ceph_msg_get(m); | |
923 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 924 | |
e84346b7 SW |
925 | /* |
926 | * only assign outgoing seq # if we haven't sent this message | |
927 | * yet. if it is requeued, resend with it's original seq. | |
928 | */ | |
929 | if (m->needs_out_seq) { | |
930 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
931 | m->needs_out_seq = false; | |
932 | } | |
31b8006e | 933 | |
4a73ef27 | 934 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n", |
31b8006e SW |
935 | m, con->out_seq, le16_to_cpu(m->hdr.type), |
936 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
f9e15777 | 937 | le32_to_cpu(m->hdr.data_len), m->p.length); |
31b8006e SW |
938 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); |
939 | ||
940 | /* tag + hdr + front + middle */ | |
e2200423 AE |
941 | con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
942 | con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
943 | con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
859eb799 | 944 | |
31b8006e | 945 | if (m->middle) |
e2200423 | 946 | con_out_kvec_add(con, m->middle->vec.iov_len, |
859eb799 | 947 | m->middle->vec.iov_base); |
31b8006e SW |
948 | |
949 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
950 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
951 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
fd154f3c | 952 | con->out_msg->footer.flags = 0; |
a9a0c51a AE |
953 | |
954 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
955 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
956 | if (m->middle) { | |
957 | crc = crc32c(0, m->middle->vec.iov_base, | |
958 | m->middle->vec.iov_len); | |
959 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
960 | } else | |
31b8006e | 961 | con->out_msg->footer.middle_crc = 0; |
739c905b | 962 | dout("%s front_crc %u middle_crc %u\n", __func__, |
31b8006e SW |
963 | le32_to_cpu(con->out_msg->footer.front_crc), |
964 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
965 | ||
966 | /* is there a data payload? */ | |
739c905b | 967 | con->out_msg->footer.data_crc = 0; |
78625051 AE |
968 | if (m->hdr.data_len) { |
969 | prepare_message_data(con->out_msg, &con->out_msg_pos); | |
970 | con->out_more = 1; /* data + footer will follow */ | |
971 | } else { | |
31b8006e | 972 | /* no, queue up footer too and be done */ |
859eb799 | 973 | prepare_write_message_footer(con); |
78625051 | 974 | } |
31b8006e | 975 | |
c9ffc77a | 976 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
977 | } |
978 | ||
979 | /* | |
980 | * Prepare an ack. | |
981 | */ | |
982 | static void prepare_write_ack(struct ceph_connection *con) | |
983 | { | |
984 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
985 | con->in_seq_acked, con->in_seq); | |
986 | con->in_seq_acked = con->in_seq; | |
987 | ||
e2200423 | 988 | con_out_kvec_reset(con); |
859eb799 | 989 | |
e2200423 | 990 | con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
859eb799 | 991 | |
31b8006e | 992 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
e2200423 | 993 | con_out_kvec_add(con, sizeof (con->out_temp_ack), |
859eb799 AE |
994 | &con->out_temp_ack); |
995 | ||
31b8006e | 996 | con->out_more = 1; /* more will follow.. eventually.. */ |
c9ffc77a | 997 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
998 | } |
999 | ||
1000 | /* | |
1001 | * Prepare to write keepalive byte. | |
1002 | */ | |
1003 | static void prepare_write_keepalive(struct ceph_connection *con) | |
1004 | { | |
1005 | dout("prepare_write_keepalive %p\n", con); | |
e2200423 AE |
1006 | con_out_kvec_reset(con); |
1007 | con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
c9ffc77a | 1008 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
31b8006e SW |
1009 | } |
1010 | ||
1011 | /* | |
1012 | * Connection negotiation. | |
1013 | */ | |
1014 | ||
dac1e716 AE |
1015 | static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con, |
1016 | int *auth_proto) | |
4e7a5dcd | 1017 | { |
a3530df3 | 1018 | struct ceph_auth_handshake *auth; |
b1c6b980 AE |
1019 | |
1020 | if (!con->ops->get_authorizer) { | |
1021 | con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN; | |
1022 | con->out_connect.authorizer_len = 0; | |
729796be | 1023 | return NULL; |
b1c6b980 AE |
1024 | } |
1025 | ||
1026 | /* Can't hold the mutex while getting authorizer */ | |
ec302645 | 1027 | mutex_unlock(&con->mutex); |
dac1e716 | 1028 | auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry); |
ec302645 | 1029 | mutex_lock(&con->mutex); |
4e7a5dcd | 1030 | |
a3530df3 | 1031 | if (IS_ERR(auth)) |
729796be | 1032 | return auth; |
8dacc7da | 1033 | if (con->state != CON_STATE_NEGOTIATING) |
729796be | 1034 | return ERR_PTR(-EAGAIN); |
0da5d703 | 1035 | |
8f43fb53 AE |
1036 | con->auth_reply_buf = auth->authorizer_reply_buf; |
1037 | con->auth_reply_buf_len = auth->authorizer_reply_buf_len; | |
729796be | 1038 | return auth; |
4e7a5dcd SW |
1039 | } |
1040 | ||
31b8006e SW |
1041 | /* |
1042 | * We connected to a peer and are saying hello. | |
1043 | */ | |
e825a66d | 1044 | static void prepare_write_banner(struct ceph_connection *con) |
31b8006e | 1045 | { |
e2200423 AE |
1046 | con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); |
1047 | con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr), | |
e825a66d | 1048 | &con->msgr->my_enc_addr); |
eed0ef2c | 1049 | |
eed0ef2c | 1050 | con->out_more = 0; |
c9ffc77a | 1051 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
eed0ef2c SW |
1052 | } |
1053 | ||
e825a66d | 1054 | static int prepare_write_connect(struct ceph_connection *con) |
eed0ef2c | 1055 | { |
95c96174 | 1056 | unsigned int global_seq = get_global_seq(con->msgr, 0); |
31b8006e | 1057 | int proto; |
dac1e716 | 1058 | int auth_proto; |
729796be | 1059 | struct ceph_auth_handshake *auth; |
31b8006e SW |
1060 | |
1061 | switch (con->peer_name.type) { | |
1062 | case CEPH_ENTITY_TYPE_MON: | |
1063 | proto = CEPH_MONC_PROTOCOL; | |
1064 | break; | |
1065 | case CEPH_ENTITY_TYPE_OSD: | |
1066 | proto = CEPH_OSDC_PROTOCOL; | |
1067 | break; | |
1068 | case CEPH_ENTITY_TYPE_MDS: | |
1069 | proto = CEPH_MDSC_PROTOCOL; | |
1070 | break; | |
1071 | default: | |
1072 | BUG(); | |
1073 | } | |
1074 | ||
1075 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
1076 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 1077 | |
e825a66d | 1078 | con->out_connect.features = cpu_to_le64(con->msgr->supported_features); |
31b8006e SW |
1079 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
1080 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
1081 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
1082 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
1083 | con->out_connect.flags = 0; | |
31b8006e | 1084 | |
dac1e716 AE |
1085 | auth_proto = CEPH_AUTH_UNKNOWN; |
1086 | auth = get_connect_authorizer(con, &auth_proto); | |
729796be AE |
1087 | if (IS_ERR(auth)) |
1088 | return PTR_ERR(auth); | |
3da54776 | 1089 | |
dac1e716 | 1090 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto); |
3da54776 AE |
1091 | con->out_connect.authorizer_len = auth ? |
1092 | cpu_to_le32(auth->authorizer_buf_len) : 0; | |
1093 | ||
e2200423 | 1094 | con_out_kvec_add(con, sizeof (con->out_connect), |
3da54776 AE |
1095 | &con->out_connect); |
1096 | if (auth && auth->authorizer_buf_len) | |
e2200423 | 1097 | con_out_kvec_add(con, auth->authorizer_buf_len, |
3da54776 | 1098 | auth->authorizer_buf); |
859eb799 | 1099 | |
31b8006e | 1100 | con->out_more = 0; |
c9ffc77a | 1101 | con_flag_set(con, CON_FLAG_WRITE_PENDING); |
4e7a5dcd | 1102 | |
e10c758e | 1103 | return 0; |
31b8006e SW |
1104 | } |
1105 | ||
31b8006e SW |
1106 | /* |
1107 | * write as much of pending kvecs to the socket as we can. | |
1108 | * 1 -> done | |
1109 | * 0 -> socket full, but more to do | |
1110 | * <0 -> error | |
1111 | */ | |
1112 | static int write_partial_kvec(struct ceph_connection *con) | |
1113 | { | |
1114 | int ret; | |
1115 | ||
1116 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
1117 | while (con->out_kvec_bytes > 0) { | |
1118 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
1119 | con->out_kvec_left, con->out_kvec_bytes, | |
1120 | con->out_more); | |
1121 | if (ret <= 0) | |
1122 | goto out; | |
1123 | con->out_kvec_bytes -= ret; | |
1124 | if (con->out_kvec_bytes == 0) | |
1125 | break; /* done */ | |
f42299e6 AE |
1126 | |
1127 | /* account for full iov entries consumed */ | |
1128 | while (ret >= con->out_kvec_cur->iov_len) { | |
1129 | BUG_ON(!con->out_kvec_left); | |
1130 | ret -= con->out_kvec_cur->iov_len; | |
1131 | con->out_kvec_cur++; | |
1132 | con->out_kvec_left--; | |
1133 | } | |
1134 | /* and for a partially-consumed entry */ | |
1135 | if (ret) { | |
1136 | con->out_kvec_cur->iov_len -= ret; | |
1137 | con->out_kvec_cur->iov_base += ret; | |
31b8006e SW |
1138 | } |
1139 | } | |
1140 | con->out_kvec_left = 0; | |
1141 | con->out_kvec_is_msg = false; | |
1142 | ret = 1; | |
1143 | out: | |
1144 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
1145 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
1146 | return ret; /* done! */ | |
1147 | } | |
1148 | ||
84ca8fc8 AE |
1149 | static void out_msg_pos_next(struct ceph_connection *con, struct page *page, |
1150 | size_t len, size_t sent, bool in_trail) | |
68b4476b | 1151 | { |
84ca8fc8 | 1152 | struct ceph_msg *msg = con->out_msg; |
bae6acd9 | 1153 | struct ceph_msg_pos *msg_pos = &con->out_msg_pos; |
68b4476b | 1154 | |
84ca8fc8 AE |
1155 | BUG_ON(!msg); |
1156 | BUG_ON(!sent); | |
68b4476b | 1157 | |
bae6acd9 AE |
1158 | msg_pos->data_pos += sent; |
1159 | msg_pos->page_pos += sent; | |
fe38a2b6 AE |
1160 | if (in_trail) { |
1161 | bool need_crc; | |
1162 | ||
1163 | need_crc = ceph_msg_data_advance(&msg->t, sent); | |
1164 | BUG_ON(need_crc && sent != len); | |
1165 | } | |
5821bd8c AE |
1166 | if (sent < len) |
1167 | return; | |
68b4476b | 1168 | |
5821bd8c | 1169 | BUG_ON(sent != len); |
bae6acd9 AE |
1170 | msg_pos->page_pos = 0; |
1171 | msg_pos->page++; | |
1172 | msg_pos->did_page_crc = false; | |
fe38a2b6 | 1173 | if (ceph_msg_has_pagelist(msg)) { |
f9e15777 | 1174 | list_rotate_left(&msg->l.pagelist->head); |
84ca8fc8 | 1175 | #ifdef CONFIG_BLOCK |
97fb1c7f | 1176 | } else if (ceph_msg_has_bio(msg)) { |
f9e15777 | 1177 | iter_bio_next(&msg->b.bio_iter, &msg->b.bio_seg); |
68b4476b | 1178 | #endif |
97fb1c7f | 1179 | } |
84ca8fc8 | 1180 | } |
68b4476b | 1181 | |
e788182f AE |
1182 | static void in_msg_pos_next(struct ceph_connection *con, size_t len, |
1183 | size_t received) | |
1184 | { | |
1185 | struct ceph_msg *msg = con->in_msg; | |
bae6acd9 | 1186 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; |
e788182f AE |
1187 | |
1188 | BUG_ON(!msg); | |
1189 | BUG_ON(!received); | |
1190 | ||
bae6acd9 AE |
1191 | msg_pos->data_pos += received; |
1192 | msg_pos->page_pos += received; | |
e788182f AE |
1193 | if (received < len) |
1194 | return; | |
1195 | ||
1196 | BUG_ON(received != len); | |
bae6acd9 AE |
1197 | msg_pos->page_pos = 0; |
1198 | msg_pos->page++; | |
e788182f | 1199 | #ifdef CONFIG_BLOCK |
f9e15777 AE |
1200 | if (msg->b.bio) |
1201 | iter_bio_next(&msg->b.bio_iter, &msg->b.bio_seg); | |
e788182f AE |
1202 | #endif /* CONFIG_BLOCK */ |
1203 | } | |
1204 | ||
35b62808 AE |
1205 | static u32 ceph_crc32c_page(u32 crc, struct page *page, |
1206 | unsigned int page_offset, | |
1207 | unsigned int length) | |
1208 | { | |
1209 | char *kaddr; | |
1210 | ||
1211 | kaddr = kmap(page); | |
1212 | BUG_ON(kaddr == NULL); | |
1213 | crc = crc32c(crc, kaddr + page_offset, length); | |
1214 | kunmap(page); | |
1215 | ||
1216 | return crc; | |
1217 | } | |
31b8006e SW |
1218 | /* |
1219 | * Write as much message data payload as we can. If we finish, queue | |
1220 | * up the footer. | |
1221 | * 1 -> done, footer is now queued in out_kvec[]. | |
1222 | * 0 -> socket full, but more to do | |
1223 | * <0 -> error | |
1224 | */ | |
34d2d200 | 1225 | static int write_partial_message_data(struct ceph_connection *con) |
31b8006e SW |
1226 | { |
1227 | struct ceph_msg *msg = con->out_msg; | |
bae6acd9 | 1228 | struct ceph_msg_pos *msg_pos = &con->out_msg_pos; |
95c96174 | 1229 | unsigned int data_len = le32_to_cpu(msg->hdr.data_len); |
37675b0f | 1230 | bool do_datacrc = !con->msgr->nocrc; |
31b8006e | 1231 | int ret; |
68b4476b | 1232 | int total_max_write; |
84ca8fc8 | 1233 | bool in_trail = false; |
97fb1c7f AE |
1234 | size_t trail_len = 0; |
1235 | size_t trail_off = data_len; | |
1236 | ||
1237 | if (ceph_msg_has_trail(msg)) { | |
43794509 | 1238 | trail_len = msg->t.pagelist->length; |
97fb1c7f AE |
1239 | trail_off -= trail_len; |
1240 | } | |
31b8006e | 1241 | |
34d2d200 | 1242 | dout("%s %p msg %p page %d offset %d\n", __func__, |
bae6acd9 | 1243 | con, msg, msg_pos->page, msg_pos->page_pos); |
31b8006e | 1244 | |
5821bd8c AE |
1245 | /* |
1246 | * Iterate through each page that contains data to be | |
1247 | * written, and send as much as possible for each. | |
1248 | * | |
1249 | * If we are calculating the data crc (the default), we will | |
1250 | * need to map the page. If we have no pages, they have | |
1251 | * been revoked, so use the zero page. | |
1252 | */ | |
bae6acd9 | 1253 | while (data_len > msg_pos->data_pos) { |
31b8006e | 1254 | struct page *page = NULL; |
e387d525 AE |
1255 | size_t page_offset; |
1256 | size_t length; | |
68b4476b | 1257 | int max_write = PAGE_SIZE; |
9bd19663 | 1258 | int bio_offset = 0; |
fe38a2b6 AE |
1259 | bool use_cursor = false; |
1260 | bool last_piece = true; /* preserve existing behavior */ | |
68b4476b | 1261 | |
bae6acd9 | 1262 | in_trail = in_trail || msg_pos->data_pos >= trail_off; |
5821bd8c | 1263 | if (!in_trail) |
bae6acd9 | 1264 | total_max_write = trail_off - msg_pos->data_pos; |
68b4476b | 1265 | |
5821bd8c | 1266 | if (in_trail) { |
97fb1c7f | 1267 | BUG_ON(!ceph_msg_has_trail(msg)); |
fe38a2b6 AE |
1268 | use_cursor = true; |
1269 | page = ceph_msg_data_next(&msg->t, &page_offset, | |
1270 | &length, &last_piece); | |
97fb1c7f | 1271 | } else if (ceph_msg_has_pages(msg)) { |
f9e15777 | 1272 | page = msg->p.pages[msg_pos->page]; |
97fb1c7f | 1273 | } else if (ceph_msg_has_pagelist(msg)) { |
f9e15777 | 1274 | page = list_first_entry(&msg->l.pagelist->head, |
58bb3b37 | 1275 | struct page, lru); |
68b4476b | 1276 | #ifdef CONFIG_BLOCK |
97fb1c7f | 1277 | } else if (ceph_msg_has_bio(msg)) { |
68b4476b YS |
1278 | struct bio_vec *bv; |
1279 | ||
f9e15777 | 1280 | bv = bio_iovec_idx(msg->b.bio_iter, msg->b.bio_seg); |
68b4476b | 1281 | page = bv->bv_page; |
9bd19663 | 1282 | bio_offset = bv->bv_offset; |
68b4476b YS |
1283 | max_write = bv->bv_len; |
1284 | #endif | |
31b8006e | 1285 | } else { |
57666519 | 1286 | page = zero_page; |
31b8006e | 1287 | } |
fe38a2b6 AE |
1288 | if (!use_cursor) |
1289 | length = min_t(int, max_write - msg_pos->page_pos, | |
1290 | total_max_write); | |
68b4476b | 1291 | |
e387d525 | 1292 | page_offset = msg_pos->page_pos + bio_offset; |
bae6acd9 | 1293 | if (do_datacrc && !msg_pos->did_page_crc) { |
5821bd8c | 1294 | u32 crc = le32_to_cpu(msg->footer.data_crc); |
31b8006e | 1295 | |
35b62808 | 1296 | crc = ceph_crc32c_page(crc, page, page_offset, length); |
84ca8fc8 | 1297 | msg->footer.data_crc = cpu_to_le32(crc); |
bae6acd9 | 1298 | msg_pos->did_page_crc = true; |
31b8006e | 1299 | } |
e387d525 | 1300 | ret = ceph_tcp_sendpage(con->sock, page, page_offset, |
fe38a2b6 | 1301 | length, last_piece); |
31b8006e SW |
1302 | if (ret <= 0) |
1303 | goto out; | |
1304 | ||
e387d525 | 1305 | out_msg_pos_next(con, page, length, (size_t) ret, in_trail); |
31b8006e SW |
1306 | } |
1307 | ||
34d2d200 | 1308 | dout("%s %p msg %p done\n", __func__, con, msg); |
31b8006e SW |
1309 | |
1310 | /* prepare and queue up footer, too */ | |
37675b0f | 1311 | if (!do_datacrc) |
84ca8fc8 | 1312 | msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
e2200423 | 1313 | con_out_kvec_reset(con); |
859eb799 | 1314 | prepare_write_message_footer(con); |
31b8006e SW |
1315 | ret = 1; |
1316 | out: | |
1317 | return ret; | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * write some zeros | |
1322 | */ | |
1323 | static int write_partial_skip(struct ceph_connection *con) | |
1324 | { | |
1325 | int ret; | |
1326 | ||
1327 | while (con->out_skip > 0) { | |
31739139 | 1328 | size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE); |
31b8006e | 1329 | |
e1dcb128 | 1330 | ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true); |
31b8006e SW |
1331 | if (ret <= 0) |
1332 | goto out; | |
1333 | con->out_skip -= ret; | |
1334 | } | |
1335 | ret = 1; | |
1336 | out: | |
1337 | return ret; | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * Prepare to read connection handshake, or an ack. | |
1342 | */ | |
eed0ef2c SW |
1343 | static void prepare_read_banner(struct ceph_connection *con) |
1344 | { | |
1345 | dout("prepare_read_banner %p\n", con); | |
1346 | con->in_base_pos = 0; | |
1347 | } | |
1348 | ||
31b8006e SW |
1349 | static void prepare_read_connect(struct ceph_connection *con) |
1350 | { | |
1351 | dout("prepare_read_connect %p\n", con); | |
1352 | con->in_base_pos = 0; | |
1353 | } | |
1354 | ||
1355 | static void prepare_read_ack(struct ceph_connection *con) | |
1356 | { | |
1357 | dout("prepare_read_ack %p\n", con); | |
1358 | con->in_base_pos = 0; | |
1359 | } | |
1360 | ||
1361 | static void prepare_read_tag(struct ceph_connection *con) | |
1362 | { | |
1363 | dout("prepare_read_tag %p\n", con); | |
1364 | con->in_base_pos = 0; | |
1365 | con->in_tag = CEPH_MSGR_TAG_READY; | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * Prepare to read a message. | |
1370 | */ | |
1371 | static int prepare_read_message(struct ceph_connection *con) | |
1372 | { | |
1373 | dout("prepare_read_message %p\n", con); | |
1374 | BUG_ON(con->in_msg != NULL); | |
1375 | con->in_base_pos = 0; | |
1376 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
1377 | return 0; | |
1378 | } | |
1379 | ||
1380 | ||
1381 | static int read_partial(struct ceph_connection *con, | |
fd51653f | 1382 | int end, int size, void *object) |
31b8006e | 1383 | { |
e6cee71f AE |
1384 | while (con->in_base_pos < end) { |
1385 | int left = end - con->in_base_pos; | |
31b8006e SW |
1386 | int have = size - left; |
1387 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1388 | if (ret <= 0) | |
1389 | return ret; | |
1390 | con->in_base_pos += ret; | |
1391 | } | |
1392 | return 1; | |
1393 | } | |
1394 | ||
1395 | ||
1396 | /* | |
1397 | * Read all or part of the connect-side handshake on a new connection | |
1398 | */ | |
eed0ef2c | 1399 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e | 1400 | { |
fd51653f AE |
1401 | int size; |
1402 | int end; | |
1403 | int ret; | |
31b8006e | 1404 | |
eed0ef2c | 1405 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1406 | |
1407 | /* peer's banner */ | |
fd51653f AE |
1408 | size = strlen(CEPH_BANNER); |
1409 | end = size; | |
1410 | ret = read_partial(con, end, size, con->in_banner); | |
31b8006e SW |
1411 | if (ret <= 0) |
1412 | goto out; | |
fd51653f AE |
1413 | |
1414 | size = sizeof (con->actual_peer_addr); | |
1415 | end += size; | |
1416 | ret = read_partial(con, end, size, &con->actual_peer_addr); | |
31b8006e SW |
1417 | if (ret <= 0) |
1418 | goto out; | |
fd51653f AE |
1419 | |
1420 | size = sizeof (con->peer_addr_for_me); | |
1421 | end += size; | |
1422 | ret = read_partial(con, end, size, &con->peer_addr_for_me); | |
31b8006e SW |
1423 | if (ret <= 0) |
1424 | goto out; | |
fd51653f | 1425 | |
eed0ef2c SW |
1426 | out: |
1427 | return ret; | |
1428 | } | |
1429 | ||
1430 | static int read_partial_connect(struct ceph_connection *con) | |
1431 | { | |
fd51653f AE |
1432 | int size; |
1433 | int end; | |
1434 | int ret; | |
eed0ef2c SW |
1435 | |
1436 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1437 | ||
fd51653f AE |
1438 | size = sizeof (con->in_reply); |
1439 | end = size; | |
1440 | ret = read_partial(con, end, size, &con->in_reply); | |
31b8006e SW |
1441 | if (ret <= 0) |
1442 | goto out; | |
fd51653f AE |
1443 | |
1444 | size = le32_to_cpu(con->in_reply.authorizer_len); | |
1445 | end += size; | |
1446 | ret = read_partial(con, end, size, con->auth_reply_buf); | |
4e7a5dcd SW |
1447 | if (ret <= 0) |
1448 | goto out; | |
31b8006e | 1449 | |
4e7a5dcd SW |
1450 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1451 | con, (int)con->in_reply.tag, | |
1452 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1453 | le32_to_cpu(con->in_reply.global_seq)); |
1454 | out: | |
1455 | return ret; | |
eed0ef2c | 1456 | |
31b8006e SW |
1457 | } |
1458 | ||
1459 | /* | |
1460 | * Verify the hello banner looks okay. | |
1461 | */ | |
1462 | static int verify_hello(struct ceph_connection *con) | |
1463 | { | |
1464 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1465 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1466 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1467 | con->error_msg = "protocol error, bad banner"; |
1468 | return -1; | |
1469 | } | |
1470 | return 0; | |
1471 | } | |
1472 | ||
1473 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1474 | { | |
1475 | switch (ss->ss_family) { | |
1476 | case AF_INET: | |
1477 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1478 | case AF_INET6: | |
1479 | return | |
1480 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1481 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1482 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1483 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1484 | } | |
1485 | return false; | |
1486 | } | |
1487 | ||
1488 | static int addr_port(struct sockaddr_storage *ss) | |
1489 | { | |
1490 | switch (ss->ss_family) { | |
1491 | case AF_INET: | |
f28bcfbe | 1492 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1493 | case AF_INET6: |
f28bcfbe | 1494 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1495 | } |
1496 | return 0; | |
1497 | } | |
1498 | ||
1499 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1500 | { | |
1501 | switch (ss->ss_family) { | |
1502 | case AF_INET: | |
1503 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1504 | break; |
31b8006e SW |
1505 | case AF_INET6: |
1506 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1507 | break; |
31b8006e SW |
1508 | } |
1509 | } | |
1510 | ||
ee3b56f2 NW |
1511 | /* |
1512 | * Unlike other *_pton function semantics, zero indicates success. | |
1513 | */ | |
1514 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1515 | char delim, const char **ipend) | |
1516 | { | |
99f0f3b2 AE |
1517 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1518 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1519 | |
1520 | memset(ss, 0, sizeof(*ss)); | |
1521 | ||
1522 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1523 | ss->ss_family = AF_INET; | |
1524 | return 0; | |
1525 | } | |
1526 | ||
1527 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1528 | ss->ss_family = AF_INET6; | |
1529 | return 0; | |
1530 | } | |
1531 | ||
1532 | return -EINVAL; | |
1533 | } | |
1534 | ||
1535 | /* | |
1536 | * Extract hostname string and resolve using kernel DNS facility. | |
1537 | */ | |
1538 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1539 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1540 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1541 | { | |
1542 | const char *end, *delim_p; | |
1543 | char *colon_p, *ip_addr = NULL; | |
1544 | int ip_len, ret; | |
1545 | ||
1546 | /* | |
1547 | * The end of the hostname occurs immediately preceding the delimiter or | |
1548 | * the port marker (':') where the delimiter takes precedence. | |
1549 | */ | |
1550 | delim_p = memchr(name, delim, namelen); | |
1551 | colon_p = memchr(name, ':', namelen); | |
1552 | ||
1553 | if (delim_p && colon_p) | |
1554 | end = delim_p < colon_p ? delim_p : colon_p; | |
1555 | else if (!delim_p && colon_p) | |
1556 | end = colon_p; | |
1557 | else { | |
1558 | end = delim_p; | |
1559 | if (!end) /* case: hostname:/ */ | |
1560 | end = name + namelen; | |
1561 | } | |
1562 | ||
1563 | if (end <= name) | |
1564 | return -EINVAL; | |
1565 | ||
1566 | /* do dns_resolve upcall */ | |
1567 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1568 | if (ip_len > 0) | |
1569 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1570 | else | |
1571 | ret = -ESRCH; | |
1572 | ||
1573 | kfree(ip_addr); | |
1574 | ||
1575 | *ipend = end; | |
1576 | ||
1577 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1578 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1579 | ||
1580 | return ret; | |
1581 | } | |
1582 | #else | |
1583 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1584 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1585 | { | |
1586 | return -EINVAL; | |
1587 | } | |
1588 | #endif | |
1589 | ||
1590 | /* | |
1591 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1592 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1593 | */ | |
1594 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1595 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1596 | { | |
1597 | int ret; | |
1598 | ||
1599 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1600 | if (ret) | |
1601 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1602 | ||
1603 | return ret; | |
1604 | } | |
1605 | ||
31b8006e SW |
1606 | /* |
1607 | * Parse an ip[:port] list into an addr array. Use the default | |
1608 | * monitor port if a port isn't specified. | |
1609 | */ | |
1610 | int ceph_parse_ips(const char *c, const char *end, | |
1611 | struct ceph_entity_addr *addr, | |
1612 | int max_count, int *count) | |
1613 | { | |
ee3b56f2 | 1614 | int i, ret = -EINVAL; |
31b8006e SW |
1615 | const char *p = c; |
1616 | ||
1617 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1618 | for (i = 0; i < max_count; i++) { | |
1619 | const char *ipend; | |
1620 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1621 | int port; |
39139f64 SW |
1622 | char delim = ','; |
1623 | ||
1624 | if (*p == '[') { | |
1625 | delim = ']'; | |
1626 | p++; | |
1627 | } | |
31b8006e | 1628 | |
ee3b56f2 NW |
1629 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1630 | if (ret) | |
31b8006e | 1631 | goto bad; |
ee3b56f2 NW |
1632 | ret = -EINVAL; |
1633 | ||
31b8006e SW |
1634 | p = ipend; |
1635 | ||
39139f64 SW |
1636 | if (delim == ']') { |
1637 | if (*p != ']') { | |
1638 | dout("missing matching ']'\n"); | |
1639 | goto bad; | |
1640 | } | |
1641 | p++; | |
1642 | } | |
1643 | ||
31b8006e SW |
1644 | /* port? */ |
1645 | if (p < end && *p == ':') { | |
1646 | port = 0; | |
1647 | p++; | |
1648 | while (p < end && *p >= '0' && *p <= '9') { | |
1649 | port = (port * 10) + (*p - '0'); | |
1650 | p++; | |
1651 | } | |
1652 | if (port > 65535 || port == 0) | |
1653 | goto bad; | |
1654 | } else { | |
1655 | port = CEPH_MON_PORT; | |
1656 | } | |
1657 | ||
1658 | addr_set_port(ss, port); | |
1659 | ||
3d14c5d2 | 1660 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1661 | |
1662 | if (p == end) | |
1663 | break; | |
1664 | if (*p != ',') | |
1665 | goto bad; | |
1666 | p++; | |
1667 | } | |
1668 | ||
1669 | if (p != end) | |
1670 | goto bad; | |
1671 | ||
1672 | if (count) | |
1673 | *count = i + 1; | |
1674 | return 0; | |
1675 | ||
1676 | bad: | |
39139f64 | 1677 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1678 | return ret; |
31b8006e | 1679 | } |
3d14c5d2 | 1680 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1681 | |
eed0ef2c | 1682 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1683 | { |
eed0ef2c | 1684 | dout("process_banner on %p\n", con); |
31b8006e SW |
1685 | |
1686 | if (verify_hello(con) < 0) | |
1687 | return -1; | |
1688 | ||
63f2d211 SW |
1689 | ceph_decode_addr(&con->actual_peer_addr); |
1690 | ceph_decode_addr(&con->peer_addr_for_me); | |
1691 | ||
31b8006e SW |
1692 | /* |
1693 | * Make sure the other end is who we wanted. note that the other | |
1694 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1695 | * them the benefit of the doubt. | |
1696 | */ | |
103e2d3a SW |
1697 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1698 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1699 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1700 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1701 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1702 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1703 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1704 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1705 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1706 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1707 | return -1; |
1708 | } | |
1709 | ||
1710 | /* | |
1711 | * did we learn our address? | |
1712 | */ | |
1713 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1714 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1715 | ||
1716 | memcpy(&con->msgr->inst.addr.in_addr, | |
1717 | &con->peer_addr_for_me.in_addr, | |
1718 | sizeof(con->peer_addr_for_me.in_addr)); | |
1719 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1720 | encode_my_addr(con->msgr); |
eed0ef2c | 1721 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1722 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1723 | } |
1724 | ||
eed0ef2c SW |
1725 | return 0; |
1726 | } | |
1727 | ||
1728 | static int process_connect(struct ceph_connection *con) | |
1729 | { | |
3d14c5d2 YS |
1730 | u64 sup_feat = con->msgr->supported_features; |
1731 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1732 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1733 | int ret; |
04a419f9 | 1734 | |
eed0ef2c SW |
1735 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1736 | ||
31b8006e | 1737 | switch (con->in_reply.tag) { |
04a419f9 SW |
1738 | case CEPH_MSGR_TAG_FEATURES: |
1739 | pr_err("%s%lld %s feature set mismatch," | |
1740 | " my %llx < server's %llx, missing %llx\n", | |
1741 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1742 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1743 | sup_feat, server_feat, server_feat & ~sup_feat); |
1744 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 1745 | reset_connection(con); |
04a419f9 SW |
1746 | return -1; |
1747 | ||
31b8006e | 1748 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1749 | pr_err("%s%lld %s protocol version mismatch," |
1750 | " my %d != server's %d\n", | |
1751 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1752 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1753 | le32_to_cpu(con->out_connect.protocol_version), |
1754 | le32_to_cpu(con->in_reply.protocol_version)); | |
1755 | con->error_msg = "protocol version mismatch"; | |
0fa6ebc6 | 1756 | reset_connection(con); |
31b8006e SW |
1757 | return -1; |
1758 | ||
4e7a5dcd SW |
1759 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1760 | con->auth_retry++; | |
1761 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1762 | con->auth_retry); | |
1763 | if (con->auth_retry == 2) { | |
1764 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1765 | return -1; |
1766 | } | |
1767 | con->auth_retry = 1; | |
6d4221b5 | 1768 | con_out_kvec_reset(con); |
e825a66d | 1769 | ret = prepare_write_connect(con); |
0da5d703 SW |
1770 | if (ret < 0) |
1771 | return ret; | |
63733a0f | 1772 | prepare_read_connect(con); |
4e7a5dcd | 1773 | break; |
31b8006e SW |
1774 | |
1775 | case CEPH_MSGR_TAG_RESETSESSION: | |
1776 | /* | |
1777 | * If we connected with a large connect_seq but the peer | |
1778 | * has no record of a session with us (no connection, or | |
1779 | * connect_seq == 0), they will send RESETSESION to indicate | |
1780 | * that they must have reset their session, and may have | |
1781 | * dropped messages. | |
1782 | */ | |
1783 | dout("process_connect got RESET peer seq %u\n", | |
5bdca4e0 | 1784 | le32_to_cpu(con->in_reply.connect_seq)); |
31b8006e SW |
1785 | pr_err("%s%lld %s connection reset\n", |
1786 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1787 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1788 | reset_connection(con); |
6d4221b5 | 1789 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1790 | ret = prepare_write_connect(con); |
1791 | if (ret < 0) | |
1792 | return ret; | |
31b8006e SW |
1793 | prepare_read_connect(con); |
1794 | ||
1795 | /* Tell ceph about it. */ | |
ec302645 | 1796 | mutex_unlock(&con->mutex); |
31b8006e SW |
1797 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1798 | if (con->ops->peer_reset) | |
1799 | con->ops->peer_reset(con); | |
ec302645 | 1800 | mutex_lock(&con->mutex); |
8dacc7da | 1801 | if (con->state != CON_STATE_NEGOTIATING) |
0da5d703 | 1802 | return -EAGAIN; |
31b8006e SW |
1803 | break; |
1804 | ||
1805 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1806 | /* | |
1807 | * If we sent a smaller connect_seq than the peer has, try | |
1808 | * again with a larger value. | |
1809 | */ | |
5bdca4e0 | 1810 | dout("process_connect got RETRY_SESSION my seq %u, peer %u\n", |
31b8006e | 1811 | le32_to_cpu(con->out_connect.connect_seq), |
5bdca4e0 SW |
1812 | le32_to_cpu(con->in_reply.connect_seq)); |
1813 | con->connect_seq = le32_to_cpu(con->in_reply.connect_seq); | |
6d4221b5 | 1814 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1815 | ret = prepare_write_connect(con); |
1816 | if (ret < 0) | |
1817 | return ret; | |
31b8006e SW |
1818 | prepare_read_connect(con); |
1819 | break; | |
1820 | ||
1821 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1822 | /* | |
1823 | * If we sent a smaller global_seq than the peer has, try | |
1824 | * again with a larger value. | |
1825 | */ | |
eed0ef2c | 1826 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e | 1827 | con->peer_global_seq, |
5bdca4e0 | 1828 | le32_to_cpu(con->in_reply.global_seq)); |
31b8006e | 1829 | get_global_seq(con->msgr, |
5bdca4e0 | 1830 | le32_to_cpu(con->in_reply.global_seq)); |
6d4221b5 | 1831 | con_out_kvec_reset(con); |
5a0f8fdd AE |
1832 | ret = prepare_write_connect(con); |
1833 | if (ret < 0) | |
1834 | return ret; | |
31b8006e SW |
1835 | prepare_read_connect(con); |
1836 | break; | |
1837 | ||
1838 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1839 | if (req_feat & ~server_feat) { |
1840 | pr_err("%s%lld %s protocol feature mismatch," | |
1841 | " my required %llx > server's %llx, need %llx\n", | |
1842 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1843 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1844 | req_feat, server_feat, req_feat & ~server_feat); |
1845 | con->error_msg = "missing required protocol features"; | |
0fa6ebc6 | 1846 | reset_connection(con); |
04a419f9 SW |
1847 | return -1; |
1848 | } | |
8dacc7da | 1849 | |
122070a2 | 1850 | WARN_ON(con->state != CON_STATE_NEGOTIATING); |
8dacc7da SW |
1851 | con->state = CON_STATE_OPEN; |
1852 | ||
31b8006e SW |
1853 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1854 | con->connect_seq++; | |
aba558e2 | 1855 | con->peer_features = server_feat; |
31b8006e SW |
1856 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1857 | con->peer_global_seq, | |
1858 | le32_to_cpu(con->in_reply.connect_seq), | |
1859 | con->connect_seq); | |
1860 | WARN_ON(con->connect_seq != | |
1861 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1862 | |
1863 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
c9ffc77a | 1864 | con_flag_set(con, CON_FLAG_LOSSYTX); |
92ac41d0 | 1865 | |
85effe18 | 1866 | con->delay = 0; /* reset backoff memory */ |
92ac41d0 | 1867 | |
31b8006e SW |
1868 | prepare_read_tag(con); |
1869 | break; | |
1870 | ||
1871 | case CEPH_MSGR_TAG_WAIT: | |
1872 | /* | |
1873 | * If there is a connection race (we are opening | |
1874 | * connections to each other), one of us may just have | |
1875 | * to WAIT. This shouldn't happen if we are the | |
1876 | * client. | |
1877 | */ | |
04177882 SW |
1878 | pr_err("process_connect got WAIT as client\n"); |
1879 | con->error_msg = "protocol error, got WAIT as client"; | |
1880 | return -1; | |
31b8006e SW |
1881 | |
1882 | default: | |
1883 | pr_err("connect protocol error, will retry\n"); | |
1884 | con->error_msg = "protocol error, garbage tag during connect"; | |
1885 | return -1; | |
1886 | } | |
1887 | return 0; | |
1888 | } | |
1889 | ||
1890 | ||
1891 | /* | |
1892 | * read (part of) an ack | |
1893 | */ | |
1894 | static int read_partial_ack(struct ceph_connection *con) | |
1895 | { | |
fd51653f AE |
1896 | int size = sizeof (con->in_temp_ack); |
1897 | int end = size; | |
31b8006e | 1898 | |
fd51653f | 1899 | return read_partial(con, end, size, &con->in_temp_ack); |
31b8006e SW |
1900 | } |
1901 | ||
1902 | ||
1903 | /* | |
1904 | * We can finally discard anything that's been acked. | |
1905 | */ | |
1906 | static void process_ack(struct ceph_connection *con) | |
1907 | { | |
1908 | struct ceph_msg *m; | |
1909 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1910 | u64 seq; | |
1911 | ||
31b8006e SW |
1912 | while (!list_empty(&con->out_sent)) { |
1913 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1914 | list_head); | |
1915 | seq = le64_to_cpu(m->hdr.seq); | |
1916 | if (seq > ack) | |
1917 | break; | |
1918 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1919 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1920 | m->ack_stamp = jiffies; |
31b8006e SW |
1921 | ceph_msg_remove(m); |
1922 | } | |
31b8006e SW |
1923 | prepare_read_tag(con); |
1924 | } | |
1925 | ||
1926 | ||
1927 | ||
1928 | ||
2450418c | 1929 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1930 | struct kvec *section, |
1931 | unsigned int sec_len, u32 *crc) | |
2450418c | 1932 | { |
68b4476b | 1933 | int ret, left; |
2450418c YS |
1934 | |
1935 | BUG_ON(!section); | |
1936 | ||
1937 | while (section->iov_len < sec_len) { | |
1938 | BUG_ON(section->iov_base == NULL); | |
1939 | left = sec_len - section->iov_len; | |
1940 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1941 | section->iov_len, left); | |
1942 | if (ret <= 0) | |
1943 | return ret; | |
1944 | section->iov_len += ret; | |
2450418c | 1945 | } |
fe3ad593 AE |
1946 | if (section->iov_len == sec_len) |
1947 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1948 | |
2450418c YS |
1949 | return 1; |
1950 | } | |
31b8006e | 1951 | |
4740a623 | 1952 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip); |
68b4476b YS |
1953 | |
1954 | static int read_partial_message_pages(struct ceph_connection *con, | |
1955 | struct page **pages, | |
95c96174 | 1956 | unsigned int data_len, bool do_datacrc) |
68b4476b | 1957 | { |
bae6acd9 | 1958 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; |
e788182f | 1959 | struct page *page; |
afb3d90e AE |
1960 | size_t page_offset; |
1961 | size_t length; | |
1962 | unsigned int left; | |
68b4476b | 1963 | int ret; |
68b4476b | 1964 | |
68b4476b YS |
1965 | /* (page) data */ |
1966 | BUG_ON(pages == NULL); | |
bae6acd9 | 1967 | page = pages[msg_pos->page]; |
afb3d90e AE |
1968 | page_offset = msg_pos->page_pos; |
1969 | BUG_ON(msg_pos->data_pos >= data_len); | |
1970 | left = data_len - msg_pos->data_pos; | |
1971 | BUG_ON(page_offset >= PAGE_SIZE); | |
1972 | length = min_t(unsigned int, PAGE_SIZE - page_offset, left); | |
1973 | ||
1974 | ret = ceph_tcp_recvpage(con->sock, page, page_offset, length); | |
68b4476b YS |
1975 | if (ret <= 0) |
1976 | return ret; | |
e788182f | 1977 | |
35b62808 AE |
1978 | if (do_datacrc) |
1979 | con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page, | |
1980 | page_offset, ret); | |
afb3d90e AE |
1981 | |
1982 | in_msg_pos_next(con, length, ret); | |
68b4476b YS |
1983 | |
1984 | return ret; | |
1985 | } | |
1986 | ||
1987 | #ifdef CONFIG_BLOCK | |
1988 | static int read_partial_message_bio(struct ceph_connection *con, | |
95c96174 | 1989 | unsigned int data_len, bool do_datacrc) |
68b4476b | 1990 | { |
b3d56fab | 1991 | struct ceph_msg *msg = con->in_msg; |
bae6acd9 | 1992 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; |
b3d56fab | 1993 | struct bio_vec *bv; |
e788182f | 1994 | struct page *page; |
afb3d90e AE |
1995 | size_t page_offset; |
1996 | size_t length; | |
1997 | unsigned int left; | |
1998 | int ret; | |
68b4476b | 1999 | |
b3d56fab | 2000 | BUG_ON(!msg); |
f9e15777 AE |
2001 | BUG_ON(!msg->b.bio_iter); |
2002 | bv = bio_iovec_idx(msg->b.bio_iter, msg->b.bio_seg); | |
e788182f | 2003 | page = bv->bv_page; |
afb3d90e AE |
2004 | page_offset = bv->bv_offset + msg_pos->page_pos; |
2005 | BUG_ON(msg_pos->data_pos >= data_len); | |
2006 | left = data_len - msg_pos->data_pos; | |
2007 | BUG_ON(msg_pos->page_pos >= bv->bv_len); | |
2008 | length = min_t(unsigned int, bv->bv_len - msg_pos->page_pos, left); | |
68b4476b | 2009 | |
afb3d90e | 2010 | ret = ceph_tcp_recvpage(con->sock, page, page_offset, length); |
68b4476b YS |
2011 | if (ret <= 0) |
2012 | return ret; | |
e788182f | 2013 | |
35b62808 AE |
2014 | if (do_datacrc) |
2015 | con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page, | |
2016 | page_offset, ret); | |
afb3d90e AE |
2017 | |
2018 | in_msg_pos_next(con, length, ret); | |
68b4476b YS |
2019 | |
2020 | return ret; | |
2021 | } | |
2022 | #endif | |
2023 | ||
34d2d200 AE |
2024 | static int read_partial_msg_data(struct ceph_connection *con) |
2025 | { | |
2026 | struct ceph_msg *msg = con->in_msg; | |
2027 | struct ceph_msg_pos *msg_pos = &con->in_msg_pos; | |
2028 | const bool do_datacrc = !con->msgr->nocrc; | |
2029 | unsigned int data_len; | |
2030 | int ret; | |
2031 | ||
2032 | BUG_ON(!msg); | |
2033 | ||
2034 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
2035 | while (msg_pos->data_pos < data_len) { | |
97fb1c7f | 2036 | if (ceph_msg_has_pages(msg)) { |
f9e15777 | 2037 | ret = read_partial_message_pages(con, msg->p.pages, |
34d2d200 AE |
2038 | data_len, do_datacrc); |
2039 | if (ret <= 0) | |
2040 | return ret; | |
2041 | #ifdef CONFIG_BLOCK | |
97fb1c7f | 2042 | } else if (ceph_msg_has_bio(msg)) { |
34d2d200 AE |
2043 | ret = read_partial_message_bio(con, |
2044 | data_len, do_datacrc); | |
2045 | if (ret <= 0) | |
2046 | return ret; | |
2047 | #endif | |
2048 | } else { | |
2049 | BUG_ON(1); | |
2050 | } | |
2051 | } | |
2052 | ||
2053 | return 1; /* must return > 0 to indicate success */ | |
2054 | } | |
2055 | ||
31b8006e SW |
2056 | /* |
2057 | * read (part of) a message. | |
2058 | */ | |
2059 | static int read_partial_message(struct ceph_connection *con) | |
2060 | { | |
2061 | struct ceph_msg *m = con->in_msg; | |
fd51653f AE |
2062 | int size; |
2063 | int end; | |
31b8006e | 2064 | int ret; |
95c96174 | 2065 | unsigned int front_len, middle_len, data_len; |
37675b0f | 2066 | bool do_datacrc = !con->msgr->nocrc; |
ae18756b | 2067 | u64 seq; |
fe3ad593 | 2068 | u32 crc; |
31b8006e SW |
2069 | |
2070 | dout("read_partial_message con %p msg %p\n", con, m); | |
2071 | ||
2072 | /* header */ | |
fd51653f AE |
2073 | size = sizeof (con->in_hdr); |
2074 | end = size; | |
2075 | ret = read_partial(con, end, size, &con->in_hdr); | |
57dac9d1 AE |
2076 | if (ret <= 0) |
2077 | return ret; | |
fe3ad593 AE |
2078 | |
2079 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
2080 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
2081 | pr_err("read_partial_message bad hdr " | |
2082 | " crc %u != expected %u\n", | |
2083 | crc, con->in_hdr.crc); | |
2084 | return -EBADMSG; | |
2085 | } | |
2086 | ||
31b8006e SW |
2087 | front_len = le32_to_cpu(con->in_hdr.front_len); |
2088 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
2089 | return -EIO; | |
2090 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
7b11ba37 | 2091 | if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN) |
31b8006e SW |
2092 | return -EIO; |
2093 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
2094 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
2095 | return -EIO; | |
2096 | ||
ae18756b SW |
2097 | /* verify seq# */ |
2098 | seq = le64_to_cpu(con->in_hdr.seq); | |
2099 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 2100 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 2101 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2102 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
2103 | seq, con->in_seq + 1); |
2104 | con->in_base_pos = -front_len - middle_len - data_len - | |
2105 | sizeof(m->footer); | |
2106 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
2107 | return 0; |
2108 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
2109 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
2110 | seq, con->in_seq + 1); | |
2111 | con->error_msg = "bad message sequence # for incoming message"; | |
2112 | return -EBADMSG; | |
2113 | } | |
2114 | ||
31b8006e SW |
2115 | /* allocate message? */ |
2116 | if (!con->in_msg) { | |
4740a623 SW |
2117 | int skip = 0; |
2118 | ||
31b8006e | 2119 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, |
6ebc8b32 | 2120 | front_len, data_len); |
4740a623 SW |
2121 | ret = ceph_con_in_msg_alloc(con, &skip); |
2122 | if (ret < 0) | |
2123 | return ret; | |
2450418c | 2124 | if (skip) { |
31b8006e | 2125 | /* skip this message */ |
a79832f2 | 2126 | dout("alloc_msg said skip message\n"); |
ae32be31 | 2127 | BUG_ON(con->in_msg); |
31b8006e SW |
2128 | con->in_base_pos = -front_len - middle_len - data_len - |
2129 | sizeof(m->footer); | |
2130 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2131 | con->in_seq++; |
31b8006e SW |
2132 | return 0; |
2133 | } | |
38941f80 | 2134 | |
4740a623 | 2135 | BUG_ON(!con->in_msg); |
38941f80 | 2136 | BUG_ON(con->in_msg->con != con); |
31b8006e SW |
2137 | m = con->in_msg; |
2138 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
2139 | if (m->middle) |
2140 | m->middle->vec.iov_len = 0; | |
9d7f0f13 | 2141 | |
78625051 | 2142 | /* prepare for data payload, if any */ |
a4107026 | 2143 | |
78625051 AE |
2144 | if (data_len) |
2145 | prepare_message_data(con->in_msg, &con->in_msg_pos); | |
31b8006e SW |
2146 | } |
2147 | ||
2148 | /* front */ | |
2450418c YS |
2149 | ret = read_partial_message_section(con, &m->front, front_len, |
2150 | &con->in_front_crc); | |
2151 | if (ret <= 0) | |
2152 | return ret; | |
31b8006e SW |
2153 | |
2154 | /* middle */ | |
2450418c | 2155 | if (m->middle) { |
213c99ee SW |
2156 | ret = read_partial_message_section(con, &m->middle->vec, |
2157 | middle_len, | |
2450418c | 2158 | &con->in_middle_crc); |
31b8006e SW |
2159 | if (ret <= 0) |
2160 | return ret; | |
31b8006e SW |
2161 | } |
2162 | ||
2163 | /* (page) data */ | |
34d2d200 AE |
2164 | if (data_len) { |
2165 | ret = read_partial_msg_data(con); | |
2166 | if (ret <= 0) | |
2167 | return ret; | |
31b8006e SW |
2168 | } |
2169 | ||
31b8006e | 2170 | /* footer */ |
fd51653f AE |
2171 | size = sizeof (m->footer); |
2172 | end += size; | |
2173 | ret = read_partial(con, end, size, &m->footer); | |
57dac9d1 AE |
2174 | if (ret <= 0) |
2175 | return ret; | |
2176 | ||
31b8006e SW |
2177 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", |
2178 | m, front_len, m->footer.front_crc, middle_len, | |
2179 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
2180 | ||
2181 | /* crc ok? */ | |
2182 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
2183 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
2184 | m, con->in_front_crc, m->footer.front_crc); | |
2185 | return -EBADMSG; | |
2186 | } | |
2187 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
2188 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
2189 | m, con->in_middle_crc, m->footer.middle_crc); | |
2190 | return -EBADMSG; | |
2191 | } | |
bca064d2 | 2192 | if (do_datacrc && |
31b8006e SW |
2193 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
2194 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
2195 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
2196 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
2197 | return -EBADMSG; | |
2198 | } | |
2199 | ||
2200 | return 1; /* done! */ | |
2201 | } | |
2202 | ||
2203 | /* | |
2204 | * Process message. This happens in the worker thread. The callback should | |
2205 | * be careful not to do anything that waits on other incoming messages or it | |
2206 | * may deadlock. | |
2207 | */ | |
2208 | static void process_message(struct ceph_connection *con) | |
2209 | { | |
5e095e8b | 2210 | struct ceph_msg *msg; |
31b8006e | 2211 | |
38941f80 AE |
2212 | BUG_ON(con->in_msg->con != con); |
2213 | con->in_msg->con = NULL; | |
5e095e8b | 2214 | msg = con->in_msg; |
31b8006e | 2215 | con->in_msg = NULL; |
36eb71aa | 2216 | con->ops->put(con); |
31b8006e SW |
2217 | |
2218 | /* if first message, set peer_name */ | |
2219 | if (con->peer_name.type == 0) | |
dbad185d | 2220 | con->peer_name = msg->hdr.src; |
31b8006e | 2221 | |
31b8006e | 2222 | con->in_seq++; |
ec302645 | 2223 | mutex_unlock(&con->mutex); |
31b8006e SW |
2224 | |
2225 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
2226 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 2227 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
2228 | le16_to_cpu(msg->hdr.type), |
2229 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2230 | le32_to_cpu(msg->hdr.front_len), | |
2231 | le32_to_cpu(msg->hdr.data_len), | |
2232 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
2233 | con->ops->dispatch(con, msg); | |
ec302645 SW |
2234 | |
2235 | mutex_lock(&con->mutex); | |
31b8006e SW |
2236 | } |
2237 | ||
2238 | ||
2239 | /* | |
2240 | * Write something to the socket. Called in a worker thread when the | |
2241 | * socket appears to be writeable and we have something ready to send. | |
2242 | */ | |
2243 | static int try_write(struct ceph_connection *con) | |
2244 | { | |
31b8006e SW |
2245 | int ret = 1; |
2246 | ||
d59315ca | 2247 | dout("try_write start %p state %lu\n", con, con->state); |
31b8006e | 2248 | |
31b8006e SW |
2249 | more: |
2250 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
2251 | ||
2252 | /* open the socket first? */ | |
8dacc7da SW |
2253 | if (con->state == CON_STATE_PREOPEN) { |
2254 | BUG_ON(con->sock); | |
2255 | con->state = CON_STATE_CONNECTING; | |
a5988c49 | 2256 | |
e2200423 | 2257 | con_out_kvec_reset(con); |
e825a66d | 2258 | prepare_write_banner(con); |
eed0ef2c | 2259 | prepare_read_banner(con); |
31b8006e | 2260 | |
cf3e5c40 | 2261 | BUG_ON(con->in_msg); |
31b8006e SW |
2262 | con->in_tag = CEPH_MSGR_TAG_READY; |
2263 | dout("try_write initiating connect on %p new state %lu\n", | |
2264 | con, con->state); | |
41617d0c AE |
2265 | ret = ceph_tcp_connect(con); |
2266 | if (ret < 0) { | |
31b8006e | 2267 | con->error_msg = "connect error"; |
31b8006e SW |
2268 | goto out; |
2269 | } | |
2270 | } | |
2271 | ||
2272 | more_kvec: | |
2273 | /* kvec data queued? */ | |
2274 | if (con->out_skip) { | |
2275 | ret = write_partial_skip(con); | |
2276 | if (ret <= 0) | |
42961d23 | 2277 | goto out; |
31b8006e SW |
2278 | } |
2279 | if (con->out_kvec_left) { | |
2280 | ret = write_partial_kvec(con); | |
2281 | if (ret <= 0) | |
42961d23 | 2282 | goto out; |
31b8006e SW |
2283 | } |
2284 | ||
2285 | /* msg pages? */ | |
2286 | if (con->out_msg) { | |
c86a2930 SW |
2287 | if (con->out_msg_done) { |
2288 | ceph_msg_put(con->out_msg); | |
2289 | con->out_msg = NULL; /* we're done with this one */ | |
2290 | goto do_next; | |
2291 | } | |
2292 | ||
34d2d200 | 2293 | ret = write_partial_message_data(con); |
31b8006e SW |
2294 | if (ret == 1) |
2295 | goto more_kvec; /* we need to send the footer, too! */ | |
2296 | if (ret == 0) | |
42961d23 | 2297 | goto out; |
31b8006e | 2298 | if (ret < 0) { |
34d2d200 | 2299 | dout("try_write write_partial_message_data err %d\n", |
31b8006e | 2300 | ret); |
42961d23 | 2301 | goto out; |
31b8006e SW |
2302 | } |
2303 | } | |
2304 | ||
c86a2930 | 2305 | do_next: |
8dacc7da | 2306 | if (con->state == CON_STATE_OPEN) { |
31b8006e SW |
2307 | /* is anything else pending? */ |
2308 | if (!list_empty(&con->out_queue)) { | |
2309 | prepare_write_message(con); | |
2310 | goto more; | |
2311 | } | |
2312 | if (con->in_seq > con->in_seq_acked) { | |
2313 | prepare_write_ack(con); | |
2314 | goto more; | |
2315 | } | |
c9ffc77a | 2316 | if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) { |
31b8006e SW |
2317 | prepare_write_keepalive(con); |
2318 | goto more; | |
2319 | } | |
2320 | } | |
2321 | ||
2322 | /* Nothing to do! */ | |
c9ffc77a | 2323 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
31b8006e | 2324 | dout("try_write nothing else to write.\n"); |
31b8006e SW |
2325 | ret = 0; |
2326 | out: | |
42961d23 | 2327 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
2328 | return ret; |
2329 | } | |
2330 | ||
2331 | ||
2332 | ||
2333 | /* | |
2334 | * Read what we can from the socket. | |
2335 | */ | |
2336 | static int try_read(struct ceph_connection *con) | |
2337 | { | |
31b8006e SW |
2338 | int ret = -1; |
2339 | ||
8dacc7da SW |
2340 | more: |
2341 | dout("try_read start on %p state %lu\n", con, con->state); | |
2342 | if (con->state != CON_STATE_CONNECTING && | |
2343 | con->state != CON_STATE_NEGOTIATING && | |
2344 | con->state != CON_STATE_OPEN) | |
31b8006e SW |
2345 | return 0; |
2346 | ||
8dacc7da | 2347 | BUG_ON(!con->sock); |
ec302645 | 2348 | |
31b8006e SW |
2349 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, |
2350 | con->in_base_pos); | |
0da5d703 | 2351 | |
8dacc7da | 2352 | if (con->state == CON_STATE_CONNECTING) { |
7593af92 AE |
2353 | dout("try_read connecting\n"); |
2354 | ret = read_partial_banner(con); | |
2355 | if (ret <= 0) | |
ab166d5a | 2356 | goto out; |
7593af92 AE |
2357 | ret = process_banner(con); |
2358 | if (ret < 0) | |
2359 | goto out; | |
2360 | ||
8dacc7da | 2361 | con->state = CON_STATE_NEGOTIATING; |
7593af92 | 2362 | |
6d4221b5 JS |
2363 | /* |
2364 | * Received banner is good, exchange connection info. | |
2365 | * Do not reset out_kvec, as sending our banner raced | |
2366 | * with receiving peer banner after connect completed. | |
2367 | */ | |
7593af92 AE |
2368 | ret = prepare_write_connect(con); |
2369 | if (ret < 0) | |
2370 | goto out; | |
2371 | prepare_read_connect(con); | |
2372 | ||
2373 | /* Send connection info before awaiting response */ | |
0da5d703 SW |
2374 | goto out; |
2375 | } | |
2376 | ||
8dacc7da | 2377 | if (con->state == CON_STATE_NEGOTIATING) { |
7593af92 | 2378 | dout("try_read negotiating\n"); |
31b8006e SW |
2379 | ret = read_partial_connect(con); |
2380 | if (ret <= 0) | |
31b8006e | 2381 | goto out; |
98bdb0aa SW |
2382 | ret = process_connect(con); |
2383 | if (ret < 0) | |
2384 | goto out; | |
31b8006e SW |
2385 | goto more; |
2386 | } | |
2387 | ||
122070a2 | 2388 | WARN_ON(con->state != CON_STATE_OPEN); |
8dacc7da | 2389 | |
31b8006e SW |
2390 | if (con->in_base_pos < 0) { |
2391 | /* | |
2392 | * skipping + discarding content. | |
2393 | * | |
2394 | * FIXME: there must be a better way to do this! | |
2395 | */ | |
84495f49 AE |
2396 | static char buf[SKIP_BUF_SIZE]; |
2397 | int skip = min((int) sizeof (buf), -con->in_base_pos); | |
2398 | ||
31b8006e SW |
2399 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); |
2400 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
2401 | if (ret <= 0) | |
98bdb0aa | 2402 | goto out; |
31b8006e SW |
2403 | con->in_base_pos += ret; |
2404 | if (con->in_base_pos) | |
2405 | goto more; | |
2406 | } | |
2407 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
2408 | /* | |
2409 | * what's next? | |
2410 | */ | |
2411 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
2412 | if (ret <= 0) | |
98bdb0aa | 2413 | goto out; |
31b8006e SW |
2414 | dout("try_read got tag %d\n", (int)con->in_tag); |
2415 | switch (con->in_tag) { | |
2416 | case CEPH_MSGR_TAG_MSG: | |
2417 | prepare_read_message(con); | |
2418 | break; | |
2419 | case CEPH_MSGR_TAG_ACK: | |
2420 | prepare_read_ack(con); | |
2421 | break; | |
2422 | case CEPH_MSGR_TAG_CLOSE: | |
8dacc7da SW |
2423 | con_close_socket(con); |
2424 | con->state = CON_STATE_CLOSED; | |
98bdb0aa | 2425 | goto out; |
31b8006e SW |
2426 | default: |
2427 | goto bad_tag; | |
2428 | } | |
2429 | } | |
2430 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2431 | ret = read_partial_message(con); | |
2432 | if (ret <= 0) { | |
2433 | switch (ret) { | |
2434 | case -EBADMSG: | |
2435 | con->error_msg = "bad crc"; | |
2436 | ret = -EIO; | |
98bdb0aa | 2437 | break; |
31b8006e SW |
2438 | case -EIO: |
2439 | con->error_msg = "io error"; | |
98bdb0aa | 2440 | break; |
31b8006e | 2441 | } |
98bdb0aa | 2442 | goto out; |
31b8006e SW |
2443 | } |
2444 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2445 | goto more; | |
2446 | process_message(con); | |
7b862e07 SW |
2447 | if (con->state == CON_STATE_OPEN) |
2448 | prepare_read_tag(con); | |
31b8006e SW |
2449 | goto more; |
2450 | } | |
2451 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2452 | ret = read_partial_ack(con); | |
2453 | if (ret <= 0) | |
98bdb0aa | 2454 | goto out; |
31b8006e SW |
2455 | process_ack(con); |
2456 | goto more; | |
2457 | } | |
2458 | ||
31b8006e | 2459 | out: |
98bdb0aa | 2460 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2461 | return ret; |
2462 | ||
2463 | bad_tag: | |
2464 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2465 | con->error_msg = "protocol error, garbage tag"; | |
2466 | ret = -1; | |
2467 | goto out; | |
2468 | } | |
2469 | ||
2470 | ||
2471 | /* | |
802c6d96 AE |
2472 | * Atomically queue work on a connection after the specified delay. |
2473 | * Bump @con reference to avoid races with connection teardown. | |
2474 | * Returns 0 if work was queued, or an error code otherwise. | |
31b8006e | 2475 | */ |
802c6d96 | 2476 | static int queue_con_delay(struct ceph_connection *con, unsigned long delay) |
31b8006e | 2477 | { |
31b8006e | 2478 | if (!con->ops->get(con)) { |
802c6d96 AE |
2479 | dout("%s %p ref count 0\n", __func__, con); |
2480 | ||
2481 | return -ENOENT; | |
31b8006e SW |
2482 | } |
2483 | ||
802c6d96 AE |
2484 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) { |
2485 | dout("%s %p - already queued\n", __func__, con); | |
31b8006e | 2486 | con->ops->put(con); |
802c6d96 AE |
2487 | |
2488 | return -EBUSY; | |
31b8006e | 2489 | } |
802c6d96 AE |
2490 | |
2491 | dout("%s %p %lu\n", __func__, con, delay); | |
2492 | ||
2493 | return 0; | |
2494 | } | |
2495 | ||
2496 | static void queue_con(struct ceph_connection *con) | |
2497 | { | |
2498 | (void) queue_con_delay(con, 0); | |
31b8006e SW |
2499 | } |
2500 | ||
7bb21d68 AE |
2501 | static bool con_sock_closed(struct ceph_connection *con) |
2502 | { | |
c9ffc77a | 2503 | if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED)) |
7bb21d68 AE |
2504 | return false; |
2505 | ||
2506 | #define CASE(x) \ | |
2507 | case CON_STATE_ ## x: \ | |
2508 | con->error_msg = "socket closed (con state " #x ")"; \ | |
2509 | break; | |
2510 | ||
2511 | switch (con->state) { | |
2512 | CASE(CLOSED); | |
2513 | CASE(PREOPEN); | |
2514 | CASE(CONNECTING); | |
2515 | CASE(NEGOTIATING); | |
2516 | CASE(OPEN); | |
2517 | CASE(STANDBY); | |
2518 | default: | |
2519 | pr_warning("%s con %p unrecognized state %lu\n", | |
2520 | __func__, con, con->state); | |
2521 | con->error_msg = "unrecognized con state"; | |
2522 | BUG(); | |
2523 | break; | |
2524 | } | |
2525 | #undef CASE | |
2526 | ||
2527 | return true; | |
2528 | } | |
2529 | ||
f20a39fd AE |
2530 | static bool con_backoff(struct ceph_connection *con) |
2531 | { | |
2532 | int ret; | |
2533 | ||
2534 | if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF)) | |
2535 | return false; | |
2536 | ||
2537 | ret = queue_con_delay(con, round_jiffies_relative(con->delay)); | |
2538 | if (ret) { | |
2539 | dout("%s: con %p FAILED to back off %lu\n", __func__, | |
2540 | con, con->delay); | |
2541 | BUG_ON(ret == -ENOENT); | |
2542 | con_flag_set(con, CON_FLAG_BACKOFF); | |
2543 | } | |
2544 | ||
2545 | return true; | |
2546 | } | |
2547 | ||
93209264 AE |
2548 | /* Finish fault handling; con->mutex must *not* be held here */ |
2549 | ||
2550 | static void con_fault_finish(struct ceph_connection *con) | |
2551 | { | |
2552 | /* | |
2553 | * in case we faulted due to authentication, invalidate our | |
2554 | * current tickets so that we can get new ones. | |
2555 | */ | |
2556 | if (con->auth_retry && con->ops->invalidate_authorizer) { | |
2557 | dout("calling invalidate_authorizer()\n"); | |
2558 | con->ops->invalidate_authorizer(con); | |
2559 | } | |
2560 | ||
2561 | if (con->ops->fault) | |
2562 | con->ops->fault(con); | |
2563 | } | |
2564 | ||
31b8006e SW |
2565 | /* |
2566 | * Do some work on a connection. Drop a connection ref when we're done. | |
2567 | */ | |
2568 | static void con_work(struct work_struct *work) | |
2569 | { | |
2570 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2571 | work.work); | |
49659416 | 2572 | bool fault; |
31b8006e | 2573 | |
9dd4658d | 2574 | mutex_lock(&con->mutex); |
49659416 AE |
2575 | while (true) { |
2576 | int ret; | |
31b8006e | 2577 | |
49659416 AE |
2578 | if ((fault = con_sock_closed(con))) { |
2579 | dout("%s: con %p SOCK_CLOSED\n", __func__, con); | |
2580 | break; | |
2581 | } | |
2582 | if (con_backoff(con)) { | |
2583 | dout("%s: con %p BACKOFF\n", __func__, con); | |
2584 | break; | |
2585 | } | |
2586 | if (con->state == CON_STATE_STANDBY) { | |
2587 | dout("%s: con %p STANDBY\n", __func__, con); | |
2588 | break; | |
2589 | } | |
2590 | if (con->state == CON_STATE_CLOSED) { | |
2591 | dout("%s: con %p CLOSED\n", __func__, con); | |
2592 | BUG_ON(con->sock); | |
2593 | break; | |
2594 | } | |
2595 | if (con->state == CON_STATE_PREOPEN) { | |
2596 | dout("%s: con %p PREOPEN\n", __func__, con); | |
2597 | BUG_ON(con->sock); | |
2598 | } | |
0da5d703 | 2599 | |
49659416 AE |
2600 | ret = try_read(con); |
2601 | if (ret < 0) { | |
2602 | if (ret == -EAGAIN) | |
2603 | continue; | |
2604 | con->error_msg = "socket error on read"; | |
2605 | fault = true; | |
2606 | break; | |
2607 | } | |
2608 | ||
2609 | ret = try_write(con); | |
2610 | if (ret < 0) { | |
2611 | if (ret == -EAGAIN) | |
2612 | continue; | |
2613 | con->error_msg = "socket error on write"; | |
2614 | fault = true; | |
2615 | } | |
2616 | ||
2617 | break; /* If we make it to here, we're done */ | |
3a140a0d | 2618 | } |
b6e7b6a1 AE |
2619 | if (fault) |
2620 | con_fault(con); | |
9dd4658d | 2621 | mutex_unlock(&con->mutex); |
0da5d703 | 2622 | |
b6e7b6a1 AE |
2623 | if (fault) |
2624 | con_fault_finish(con); | |
2625 | ||
2626 | con->ops->put(con); | |
31b8006e SW |
2627 | } |
2628 | ||
31b8006e SW |
2629 | /* |
2630 | * Generic error/fault handler. A retry mechanism is used with | |
2631 | * exponential backoff | |
2632 | */ | |
93209264 | 2633 | static void con_fault(struct ceph_connection *con) |
31b8006e | 2634 | { |
28362986 | 2635 | pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), |
3d14c5d2 | 2636 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2637 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2638 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 2639 | |
122070a2 | 2640 | WARN_ON(con->state != CON_STATE_CONNECTING && |
8dacc7da SW |
2641 | con->state != CON_STATE_NEGOTIATING && |
2642 | con->state != CON_STATE_OPEN); | |
ec302645 | 2643 | |
31b8006e | 2644 | con_close_socket(con); |
5e095e8b | 2645 | |
c9ffc77a | 2646 | if (con_flag_test(con, CON_FLAG_LOSSYTX)) { |
8dacc7da SW |
2647 | dout("fault on LOSSYTX channel, marking CLOSED\n"); |
2648 | con->state = CON_STATE_CLOSED; | |
93209264 | 2649 | return; |
3b5ede07 SW |
2650 | } |
2651 | ||
5e095e8b | 2652 | if (con->in_msg) { |
38941f80 AE |
2653 | BUG_ON(con->in_msg->con != con); |
2654 | con->in_msg->con = NULL; | |
5e095e8b SW |
2655 | ceph_msg_put(con->in_msg); |
2656 | con->in_msg = NULL; | |
36eb71aa | 2657 | con->ops->put(con); |
5e095e8b | 2658 | } |
31b8006e | 2659 | |
e80a52d1 SW |
2660 | /* Requeue anything that hasn't been acked */ |
2661 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2662 | |
e76661d0 SW |
2663 | /* If there are no messages queued or keepalive pending, place |
2664 | * the connection in a STANDBY state */ | |
2665 | if (list_empty(&con->out_queue) && | |
c9ffc77a | 2666 | !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) { |
e00de341 | 2667 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
c9ffc77a | 2668 | con_flag_clear(con, CON_FLAG_WRITE_PENDING); |
8dacc7da | 2669 | con->state = CON_STATE_STANDBY; |
e80a52d1 SW |
2670 | } else { |
2671 | /* retry after a delay. */ | |
8dacc7da | 2672 | con->state = CON_STATE_PREOPEN; |
e80a52d1 SW |
2673 | if (con->delay == 0) |
2674 | con->delay = BASE_DELAY_INTERVAL; | |
2675 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2676 | con->delay *= 2; | |
c9ffc77a | 2677 | con_flag_set(con, CON_FLAG_BACKOFF); |
8618e30b | 2678 | queue_con(con); |
31b8006e | 2679 | } |
31b8006e SW |
2680 | } |
2681 | ||
2682 | ||
2683 | ||
2684 | /* | |
15d9882c | 2685 | * initialize a new messenger instance |
31b8006e | 2686 | */ |
15d9882c AE |
2687 | void ceph_messenger_init(struct ceph_messenger *msgr, |
2688 | struct ceph_entity_addr *myaddr, | |
2689 | u32 supported_features, | |
2690 | u32 required_features, | |
2691 | bool nocrc) | |
31b8006e | 2692 | { |
3d14c5d2 YS |
2693 | msgr->supported_features = supported_features; |
2694 | msgr->required_features = required_features; | |
2695 | ||
31b8006e SW |
2696 | spin_lock_init(&msgr->global_seq_lock); |
2697 | ||
31b8006e SW |
2698 | if (myaddr) |
2699 | msgr->inst.addr = *myaddr; | |
2700 | ||
2701 | /* select a random nonce */ | |
ac8839d7 | 2702 | msgr->inst.addr.type = 0; |
103e2d3a | 2703 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2704 | encode_my_addr(msgr); |
15d9882c | 2705 | msgr->nocrc = nocrc; |
31b8006e | 2706 | |
a2a32584 | 2707 | atomic_set(&msgr->stopping, 0); |
31b8006e | 2708 | |
15d9882c | 2709 | dout("%s %p\n", __func__, msgr); |
31b8006e | 2710 | } |
15d9882c | 2711 | EXPORT_SYMBOL(ceph_messenger_init); |
31b8006e | 2712 | |
e00de341 SW |
2713 | static void clear_standby(struct ceph_connection *con) |
2714 | { | |
2715 | /* come back from STANDBY? */ | |
8dacc7da | 2716 | if (con->state == CON_STATE_STANDBY) { |
e00de341 | 2717 | dout("clear_standby %p and ++connect_seq\n", con); |
8dacc7da | 2718 | con->state = CON_STATE_PREOPEN; |
e00de341 | 2719 | con->connect_seq++; |
c9ffc77a AE |
2720 | WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING)); |
2721 | WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)); | |
e00de341 SW |
2722 | } |
2723 | } | |
2724 | ||
31b8006e SW |
2725 | /* |
2726 | * Queue up an outgoing message on the given connection. | |
2727 | */ | |
2728 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2729 | { | |
31b8006e | 2730 | /* set src+dst */ |
dbad185d | 2731 | msg->hdr.src = con->msgr->inst.name; |
3ca02ef9 | 2732 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
e84346b7 SW |
2733 | msg->needs_out_seq = true; |
2734 | ||
ec302645 | 2735 | mutex_lock(&con->mutex); |
92ce034b | 2736 | |
8dacc7da | 2737 | if (con->state == CON_STATE_CLOSED) { |
a59b55a6 SW |
2738 | dout("con_send %p closed, dropping %p\n", con, msg); |
2739 | ceph_msg_put(msg); | |
2740 | mutex_unlock(&con->mutex); | |
2741 | return; | |
2742 | } | |
2743 | ||
38941f80 | 2744 | BUG_ON(msg->con != NULL); |
36eb71aa | 2745 | msg->con = con->ops->get(con); |
92ce034b AE |
2746 | BUG_ON(msg->con == NULL); |
2747 | ||
31b8006e SW |
2748 | BUG_ON(!list_empty(&msg->list_head)); |
2749 | list_add_tail(&msg->list_head, &con->out_queue); | |
2750 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2751 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2752 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2753 | le32_to_cpu(msg->hdr.front_len), | |
2754 | le32_to_cpu(msg->hdr.middle_len), | |
2755 | le32_to_cpu(msg->hdr.data_len)); | |
00650931 SW |
2756 | |
2757 | clear_standby(con); | |
ec302645 | 2758 | mutex_unlock(&con->mutex); |
31b8006e SW |
2759 | |
2760 | /* if there wasn't anything waiting to send before, queue | |
2761 | * new work */ | |
c9ffc77a | 2762 | if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) |
31b8006e SW |
2763 | queue_con(con); |
2764 | } | |
3d14c5d2 | 2765 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2766 | |
2767 | /* | |
2768 | * Revoke a message that was previously queued for send | |
2769 | */ | |
6740a845 | 2770 | void ceph_msg_revoke(struct ceph_msg *msg) |
31b8006e | 2771 | { |
6740a845 AE |
2772 | struct ceph_connection *con = msg->con; |
2773 | ||
2774 | if (!con) | |
2775 | return; /* Message not in our possession */ | |
2776 | ||
ec302645 | 2777 | mutex_lock(&con->mutex); |
31b8006e | 2778 | if (!list_empty(&msg->list_head)) { |
38941f80 | 2779 | dout("%s %p msg %p - was on queue\n", __func__, con, msg); |
31b8006e | 2780 | list_del_init(&msg->list_head); |
38941f80 | 2781 | BUG_ON(msg->con == NULL); |
36eb71aa | 2782 | msg->con->ops->put(msg->con); |
38941f80 | 2783 | msg->con = NULL; |
31b8006e | 2784 | msg->hdr.seq = 0; |
38941f80 | 2785 | |
31b8006e | 2786 | ceph_msg_put(msg); |
ed98adad SW |
2787 | } |
2788 | if (con->out_msg == msg) { | |
38941f80 | 2789 | dout("%s %p msg %p - was sending\n", __func__, con, msg); |
ed98adad | 2790 | con->out_msg = NULL; |
31b8006e SW |
2791 | if (con->out_kvec_is_msg) { |
2792 | con->out_skip = con->out_kvec_bytes; | |
2793 | con->out_kvec_is_msg = false; | |
2794 | } | |
ed98adad | 2795 | msg->hdr.seq = 0; |
92ce034b AE |
2796 | |
2797 | ceph_msg_put(msg); | |
31b8006e | 2798 | } |
ec302645 | 2799 | mutex_unlock(&con->mutex); |
31b8006e SW |
2800 | } |
2801 | ||
350b1c32 | 2802 | /* |
0d59ab81 | 2803 | * Revoke a message that we may be reading data into |
350b1c32 | 2804 | */ |
8921d114 | 2805 | void ceph_msg_revoke_incoming(struct ceph_msg *msg) |
350b1c32 | 2806 | { |
8921d114 AE |
2807 | struct ceph_connection *con; |
2808 | ||
2809 | BUG_ON(msg == NULL); | |
2810 | if (!msg->con) { | |
2811 | dout("%s msg %p null con\n", __func__, msg); | |
2812 | ||
2813 | return; /* Message not in our possession */ | |
2814 | } | |
2815 | ||
2816 | con = msg->con; | |
350b1c32 | 2817 | mutex_lock(&con->mutex); |
8921d114 | 2818 | if (con->in_msg == msg) { |
95c96174 ED |
2819 | unsigned int front_len = le32_to_cpu(con->in_hdr.front_len); |
2820 | unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
2821 | unsigned int data_len = le32_to_cpu(con->in_hdr.data_len); | |
350b1c32 SW |
2822 | |
2823 | /* skip rest of message */ | |
8921d114 AE |
2824 | dout("%s %p msg %p revoked\n", __func__, con, msg); |
2825 | con->in_base_pos = con->in_base_pos - | |
350b1c32 | 2826 | sizeof(struct ceph_msg_header) - |
0d59ab81 YS |
2827 | front_len - |
2828 | middle_len - | |
2829 | data_len - | |
350b1c32 | 2830 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2831 | ceph_msg_put(con->in_msg); |
2832 | con->in_msg = NULL; | |
2833 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2834 | con->in_seq++; |
350b1c32 | 2835 | } else { |
8921d114 AE |
2836 | dout("%s %p in_msg %p msg %p no-op\n", |
2837 | __func__, con, con->in_msg, msg); | |
350b1c32 SW |
2838 | } |
2839 | mutex_unlock(&con->mutex); | |
2840 | } | |
2841 | ||
31b8006e SW |
2842 | /* |
2843 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2844 | */ | |
2845 | void ceph_con_keepalive(struct ceph_connection *con) | |
2846 | { | |
e00de341 | 2847 | dout("con_keepalive %p\n", con); |
00650931 | 2848 | mutex_lock(&con->mutex); |
e00de341 | 2849 | clear_standby(con); |
00650931 | 2850 | mutex_unlock(&con->mutex); |
c9ffc77a AE |
2851 | if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 && |
2852 | con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0) | |
31b8006e SW |
2853 | queue_con(con); |
2854 | } | |
3d14c5d2 | 2855 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e | 2856 | |
43794509 AE |
2857 | static void ceph_msg_data_init(struct ceph_msg_data *data) |
2858 | { | |
2859 | data->type = CEPH_MSG_DATA_NONE; | |
2860 | } | |
2861 | ||
02afca6c | 2862 | void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages, |
f1baeb2b | 2863 | size_t length, size_t alignment) |
02afca6c | 2864 | { |
07aa1558 AE |
2865 | BUG_ON(!pages); |
2866 | BUG_ON(!length); | |
43794509 | 2867 | BUG_ON(msg->p.type != CEPH_MSG_DATA_NONE); |
02afca6c | 2868 | |
43794509 | 2869 | msg->p.type = CEPH_MSG_DATA_PAGES; |
f9e15777 AE |
2870 | msg->p.pages = pages; |
2871 | msg->p.length = length; | |
2872 | msg->p.alignment = alignment & ~PAGE_MASK; | |
02afca6c AE |
2873 | } |
2874 | EXPORT_SYMBOL(ceph_msg_data_set_pages); | |
31b8006e | 2875 | |
27fa8385 AE |
2876 | void ceph_msg_data_set_pagelist(struct ceph_msg *msg, |
2877 | struct ceph_pagelist *pagelist) | |
2878 | { | |
07aa1558 AE |
2879 | BUG_ON(!pagelist); |
2880 | BUG_ON(!pagelist->length); | |
43794509 | 2881 | BUG_ON(msg->l.type != CEPH_MSG_DATA_NONE); |
27fa8385 | 2882 | |
43794509 | 2883 | msg->l.type = CEPH_MSG_DATA_PAGELIST; |
f9e15777 | 2884 | msg->l.pagelist = pagelist; |
27fa8385 AE |
2885 | } |
2886 | EXPORT_SYMBOL(ceph_msg_data_set_pagelist); | |
2887 | ||
2888 | void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio) | |
2889 | { | |
07aa1558 | 2890 | BUG_ON(!bio); |
43794509 | 2891 | BUG_ON(msg->b.type != CEPH_MSG_DATA_NONE); |
27fa8385 | 2892 | |
43794509 | 2893 | msg->b.type = CEPH_MSG_DATA_BIO; |
f9e15777 | 2894 | msg->b.bio = bio; |
27fa8385 AE |
2895 | } |
2896 | EXPORT_SYMBOL(ceph_msg_data_set_bio); | |
2897 | ||
2898 | void ceph_msg_data_set_trail(struct ceph_msg *msg, struct ceph_pagelist *trail) | |
2899 | { | |
07aa1558 AE |
2900 | BUG_ON(!trail); |
2901 | BUG_ON(!trail->length); | |
43794509 | 2902 | BUG_ON(msg->b.type != CEPH_MSG_DATA_NONE); |
27fa8385 | 2903 | |
43794509 AE |
2904 | msg->t.type = CEPH_MSG_DATA_PAGELIST; |
2905 | msg->t.pagelist = trail; | |
27fa8385 AE |
2906 | } |
2907 | EXPORT_SYMBOL(ceph_msg_data_set_trail); | |
2908 | ||
31b8006e SW |
2909 | /* |
2910 | * construct a new message with given type, size | |
2911 | * the new msg has a ref count of 1. | |
2912 | */ | |
b61c2763 SW |
2913 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2914 | bool can_fail) | |
31b8006e SW |
2915 | { |
2916 | struct ceph_msg *m; | |
2917 | ||
9516e45b | 2918 | m = kzalloc(sizeof(*m), flags); |
31b8006e SW |
2919 | if (m == NULL) |
2920 | goto out; | |
31b8006e SW |
2921 | |
2922 | m->hdr.type = cpu_to_le16(type); | |
45c6ceb5 | 2923 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
31b8006e | 2924 | m->hdr.front_len = cpu_to_le32(front_len); |
ca20892d | 2925 | |
9516e45b AE |
2926 | INIT_LIST_HEAD(&m->list_head); |
2927 | kref_init(&m->kref); | |
ca20892d | 2928 | |
43794509 AE |
2929 | ceph_msg_data_init(&m->p); |
2930 | ceph_msg_data_init(&m->l); | |
2931 | ceph_msg_data_init(&m->b); | |
2932 | ceph_msg_data_init(&m->t); | |
2933 | ||
31b8006e | 2934 | /* front */ |
9516e45b | 2935 | m->front_max = front_len; |
31b8006e SW |
2936 | if (front_len) { |
2937 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2938 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2939 | PAGE_KERNEL); |
2940 | m->front_is_vmalloc = true; | |
2941 | } else { | |
34d23762 | 2942 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2943 | } |
2944 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2945 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2946 | front_len); |
2947 | goto out2; | |
2948 | } | |
2949 | } else { | |
2950 | m->front.iov_base = NULL; | |
2951 | } | |
2952 | m->front.iov_len = front_len; | |
2953 | ||
bb257664 | 2954 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2955 | return m; |
2956 | ||
2957 | out2: | |
2958 | ceph_msg_put(m); | |
2959 | out: | |
b61c2763 SW |
2960 | if (!can_fail) { |
2961 | pr_err("msg_new can't create type %d front %d\n", type, | |
2962 | front_len); | |
f0ed1b7c | 2963 | WARN_ON(1); |
b61c2763 SW |
2964 | } else { |
2965 | dout("msg_new can't create type %d front %d\n", type, | |
2966 | front_len); | |
2967 | } | |
a79832f2 | 2968 | return NULL; |
31b8006e | 2969 | } |
3d14c5d2 | 2970 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2971 | |
31b8006e SW |
2972 | /* |
2973 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2974 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2975 | * per-type header in the front and then gracefully fail (i.e., | |
2976 | * propagate the error to the caller based on info in the front) when | |
2977 | * the middle is too large. | |
2978 | */ | |
2450418c | 2979 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2980 | { |
2981 | int type = le16_to_cpu(msg->hdr.type); | |
2982 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2983 | ||
2984 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2985 | ceph_msg_type_name(type), middle_len); | |
2986 | BUG_ON(!middle_len); | |
2987 | BUG_ON(msg->middle); | |
2988 | ||
b6c1d5b8 | 2989 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2990 | if (!msg->middle) |
2991 | return -ENOMEM; | |
2992 | return 0; | |
2993 | } | |
2994 | ||
2450418c | 2995 | /* |
1c20f2d2 AE |
2996 | * Allocate a message for receiving an incoming message on a |
2997 | * connection, and save the result in con->in_msg. Uses the | |
2998 | * connection's private alloc_msg op if available. | |
2999 | * | |
4740a623 SW |
3000 | * Returns 0 on success, or a negative error code. |
3001 | * | |
3002 | * On success, if we set *skip = 1: | |
3003 | * - the next message should be skipped and ignored. | |
3004 | * - con->in_msg == NULL | |
3005 | * or if we set *skip = 0: | |
3006 | * - con->in_msg is non-null. | |
3007 | * On error (ENOMEM, EAGAIN, ...), | |
3008 | * - con->in_msg == NULL | |
2450418c | 3009 | */ |
4740a623 | 3010 | static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip) |
2450418c | 3011 | { |
4740a623 | 3012 | struct ceph_msg_header *hdr = &con->in_hdr; |
2450418c | 3013 | int middle_len = le32_to_cpu(hdr->middle_len); |
1d866d1c | 3014 | struct ceph_msg *msg; |
4740a623 | 3015 | int ret = 0; |
2450418c | 3016 | |
1c20f2d2 | 3017 | BUG_ON(con->in_msg != NULL); |
53ded495 | 3018 | BUG_ON(!con->ops->alloc_msg); |
2450418c | 3019 | |
53ded495 AE |
3020 | mutex_unlock(&con->mutex); |
3021 | msg = con->ops->alloc_msg(con, hdr, skip); | |
3022 | mutex_lock(&con->mutex); | |
3023 | if (con->state != CON_STATE_OPEN) { | |
3024 | if (msg) | |
1d866d1c | 3025 | ceph_msg_put(msg); |
53ded495 AE |
3026 | return -EAGAIN; |
3027 | } | |
4137577a AE |
3028 | if (msg) { |
3029 | BUG_ON(*skip); | |
3030 | con->in_msg = msg; | |
36eb71aa | 3031 | con->in_msg->con = con->ops->get(con); |
92ce034b | 3032 | BUG_ON(con->in_msg->con == NULL); |
4137577a AE |
3033 | } else { |
3034 | /* | |
3035 | * Null message pointer means either we should skip | |
3036 | * this message or we couldn't allocate memory. The | |
3037 | * former is not an error. | |
3038 | */ | |
3039 | if (*skip) | |
3040 | return 0; | |
3041 | con->error_msg = "error allocating memory for incoming message"; | |
3042 | ||
53ded495 | 3043 | return -ENOMEM; |
2450418c | 3044 | } |
1c20f2d2 | 3045 | memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 3046 | |
1c20f2d2 AE |
3047 | if (middle_len && !con->in_msg->middle) { |
3048 | ret = ceph_alloc_middle(con, con->in_msg); | |
2450418c | 3049 | if (ret < 0) { |
1c20f2d2 AE |
3050 | ceph_msg_put(con->in_msg); |
3051 | con->in_msg = NULL; | |
2450418c YS |
3052 | } |
3053 | } | |
9d7f0f13 | 3054 | |
4740a623 | 3055 | return ret; |
2450418c YS |
3056 | } |
3057 | ||
31b8006e SW |
3058 | |
3059 | /* | |
3060 | * Free a generically kmalloc'd message. | |
3061 | */ | |
3062 | void ceph_msg_kfree(struct ceph_msg *m) | |
3063 | { | |
3064 | dout("msg_kfree %p\n", m); | |
3065 | if (m->front_is_vmalloc) | |
3066 | vfree(m->front.iov_base); | |
3067 | else | |
3068 | kfree(m->front.iov_base); | |
3069 | kfree(m); | |
3070 | } | |
3071 | ||
3072 | /* | |
3073 | * Drop a msg ref. Destroy as needed. | |
3074 | */ | |
c2e552e7 SW |
3075 | void ceph_msg_last_put(struct kref *kref) |
3076 | { | |
3077 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 3078 | |
c2e552e7 SW |
3079 | dout("ceph_msg_put last one on %p\n", m); |
3080 | WARN_ON(!list_empty(&m->list_head)); | |
3081 | ||
3082 | /* drop middle, data, if any */ | |
3083 | if (m->middle) { | |
3084 | ceph_buffer_put(m->middle); | |
3085 | m->middle = NULL; | |
31b8006e | 3086 | } |
97fb1c7f | 3087 | if (ceph_msg_has_pages(m)) { |
f9e15777 AE |
3088 | m->p.length = 0; |
3089 | m->p.pages = NULL; | |
97fb1c7f | 3090 | } |
c2e552e7 | 3091 | |
97fb1c7f | 3092 | if (ceph_msg_has_pagelist(m)) { |
f9e15777 AE |
3093 | ceph_pagelist_release(m->l.pagelist); |
3094 | kfree(m->l.pagelist); | |
3095 | m->l.pagelist = NULL; | |
58bb3b37 SW |
3096 | } |
3097 | ||
97fb1c7f | 3098 | if (ceph_msg_has_trail(m)) |
43794509 | 3099 | m->t.pagelist = NULL; |
68b4476b | 3100 | |
c2e552e7 SW |
3101 | if (m->pool) |
3102 | ceph_msgpool_put(m->pool, m); | |
3103 | else | |
3104 | ceph_msg_kfree(m); | |
31b8006e | 3105 | } |
3d14c5d2 | 3106 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
3107 | |
3108 | void ceph_msg_dump(struct ceph_msg *msg) | |
3109 | { | |
4a73ef27 | 3110 | pr_debug("msg_dump %p (front_max %d length %zd)\n", msg, |
f9e15777 | 3111 | msg->front_max, msg->p.length); |
9ec7cab1 SW |
3112 | print_hex_dump(KERN_DEBUG, "header: ", |
3113 | DUMP_PREFIX_OFFSET, 16, 1, | |
3114 | &msg->hdr, sizeof(msg->hdr), true); | |
3115 | print_hex_dump(KERN_DEBUG, " front: ", | |
3116 | DUMP_PREFIX_OFFSET, 16, 1, | |
3117 | msg->front.iov_base, msg->front.iov_len, true); | |
3118 | if (msg->middle) | |
3119 | print_hex_dump(KERN_DEBUG, "middle: ", | |
3120 | DUMP_PREFIX_OFFSET, 16, 1, | |
3121 | msg->middle->vec.iov_base, | |
3122 | msg->middle->vec.iov_len, true); | |
3123 | print_hex_dump(KERN_DEBUG, "footer: ", | |
3124 | DUMP_PREFIX_OFFSET, 16, 1, | |
3125 | &msg->footer, sizeof(msg->footer), true); | |
3126 | } | |
3d14c5d2 | 3127 | EXPORT_SYMBOL(ceph_msg_dump); |