]>
Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
31b8006e SW |
2 | |
3 | #include <linux/crc32c.h> | |
4 | #include <linux/ctype.h> | |
5 | #include <linux/highmem.h> | |
6 | #include <linux/inet.h> | |
7 | #include <linux/kthread.h> | |
8 | #include <linux/net.h> | |
5a0e3ad6 | 9 | #include <linux/slab.h> |
31b8006e SW |
10 | #include <linux/socket.h> |
11 | #include <linux/string.h> | |
68b4476b YS |
12 | #include <linux/bio.h> |
13 | #include <linux/blkdev.h> | |
ee3b56f2 | 14 | #include <linux/dns_resolver.h> |
31b8006e SW |
15 | #include <net/tcp.h> |
16 | ||
3d14c5d2 YS |
17 | #include <linux/ceph/libceph.h> |
18 | #include <linux/ceph/messenger.h> | |
19 | #include <linux/ceph/decode.h> | |
20 | #include <linux/ceph/pagelist.h> | |
bc3b2d7f | 21 | #include <linux/export.h> |
31b8006e SW |
22 | |
23 | /* | |
24 | * Ceph uses the messenger to exchange ceph_msg messages with other | |
25 | * hosts in the system. The messenger provides ordered and reliable | |
26 | * delivery. We tolerate TCP disconnects by reconnecting (with | |
27 | * exponential backoff) in the case of a fault (disconnection, bad | |
28 | * crc, protocol error). Acks allow sent messages to be discarded by | |
29 | * the sender. | |
30 | */ | |
31 | ||
32 | /* static tag bytes (protocol control messages) */ | |
33 | static char tag_msg = CEPH_MSGR_TAG_MSG; | |
34 | static char tag_ack = CEPH_MSGR_TAG_ACK; | |
35 | static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE; | |
36 | ||
a6a5349d SW |
37 | #ifdef CONFIG_LOCKDEP |
38 | static struct lock_class_key socket_class; | |
39 | #endif | |
40 | ||
31b8006e SW |
41 | |
42 | static void queue_con(struct ceph_connection *con); | |
43 | static void con_work(struct work_struct *); | |
44 | static void ceph_fault(struct ceph_connection *con); | |
45 | ||
31b8006e | 46 | /* |
f64a9317 AE |
47 | * Nicely render a sockaddr as a string. An array of formatted |
48 | * strings is used, to approximate reentrancy. | |
31b8006e | 49 | */ |
f64a9317 AE |
50 | #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */ |
51 | #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG) | |
52 | #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1) | |
53 | #define MAX_ADDR_STR_LEN 64 /* 54 is enough */ | |
54 | ||
55 | static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN]; | |
56 | static atomic_t addr_str_seq = ATOMIC_INIT(0); | |
31b8006e | 57 | |
57666519 AE |
58 | static struct page *zero_page; /* used in certain error cases */ |
59 | static void *zero_page_address; /* kernel virtual addr of zero_page */ | |
60 | ||
3d14c5d2 | 61 | const char *ceph_pr_addr(const struct sockaddr_storage *ss) |
31b8006e SW |
62 | { |
63 | int i; | |
64 | char *s; | |
99f0f3b2 AE |
65 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
66 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
31b8006e | 67 | |
f64a9317 | 68 | i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK; |
31b8006e SW |
69 | s = addr_str[i]; |
70 | ||
71 | switch (ss->ss_family) { | |
72 | case AF_INET: | |
bd406145 AE |
73 | snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr, |
74 | ntohs(in4->sin_port)); | |
31b8006e SW |
75 | break; |
76 | ||
77 | case AF_INET6: | |
bd406145 AE |
78 | snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr, |
79 | ntohs(in6->sin6_port)); | |
31b8006e SW |
80 | break; |
81 | ||
82 | default: | |
d3002b97 AE |
83 | snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)", |
84 | ss->ss_family); | |
31b8006e SW |
85 | } |
86 | ||
87 | return s; | |
88 | } | |
3d14c5d2 | 89 | EXPORT_SYMBOL(ceph_pr_addr); |
31b8006e | 90 | |
63f2d211 SW |
91 | static void encode_my_addr(struct ceph_messenger *msgr) |
92 | { | |
93 | memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr)); | |
94 | ceph_encode_addr(&msgr->my_enc_addr); | |
95 | } | |
96 | ||
31b8006e SW |
97 | /* |
98 | * work queue for all reading and writing to/from the socket. | |
99 | */ | |
e0f43c94 | 100 | static struct workqueue_struct *ceph_msgr_wq; |
31b8006e | 101 | |
6173d1f0 AE |
102 | void _ceph_msgr_exit(void) |
103 | { | |
d3002b97 | 104 | if (ceph_msgr_wq) { |
6173d1f0 | 105 | destroy_workqueue(ceph_msgr_wq); |
d3002b97 AE |
106 | ceph_msgr_wq = NULL; |
107 | } | |
6173d1f0 AE |
108 | |
109 | BUG_ON(zero_page_address == NULL); | |
110 | zero_page_address = NULL; | |
111 | ||
112 | BUG_ON(zero_page == NULL); | |
113 | kunmap(zero_page); | |
114 | page_cache_release(zero_page); | |
115 | zero_page = NULL; | |
116 | } | |
117 | ||
3d14c5d2 | 118 | int ceph_msgr_init(void) |
31b8006e | 119 | { |
57666519 AE |
120 | BUG_ON(zero_page != NULL); |
121 | zero_page = ZERO_PAGE(0); | |
122 | page_cache_get(zero_page); | |
123 | ||
124 | BUG_ON(zero_page_address != NULL); | |
125 | zero_page_address = kmap(zero_page); | |
126 | ||
f363e45f | 127 | ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0); |
6173d1f0 AE |
128 | if (ceph_msgr_wq) |
129 | return 0; | |
57666519 | 130 | |
6173d1f0 AE |
131 | pr_err("msgr_init failed to create workqueue\n"); |
132 | _ceph_msgr_exit(); | |
57666519 | 133 | |
6173d1f0 | 134 | return -ENOMEM; |
31b8006e | 135 | } |
3d14c5d2 | 136 | EXPORT_SYMBOL(ceph_msgr_init); |
31b8006e SW |
137 | |
138 | void ceph_msgr_exit(void) | |
139 | { | |
57666519 | 140 | BUG_ON(ceph_msgr_wq == NULL); |
57666519 | 141 | |
6173d1f0 | 142 | _ceph_msgr_exit(); |
31b8006e | 143 | } |
3d14c5d2 | 144 | EXPORT_SYMBOL(ceph_msgr_exit); |
31b8006e | 145 | |
cd84db6e | 146 | void ceph_msgr_flush(void) |
a922d38f SW |
147 | { |
148 | flush_workqueue(ceph_msgr_wq); | |
149 | } | |
3d14c5d2 | 150 | EXPORT_SYMBOL(ceph_msgr_flush); |
a922d38f SW |
151 | |
152 | ||
31b8006e SW |
153 | /* |
154 | * socket callback functions | |
155 | */ | |
156 | ||
157 | /* data available on socket, or listen socket received a connect */ | |
158 | static void ceph_data_ready(struct sock *sk, int count_unused) | |
159 | { | |
bd406145 AE |
160 | struct ceph_connection *con = sk->sk_user_data; |
161 | ||
31b8006e SW |
162 | if (sk->sk_state != TCP_CLOSE_WAIT) { |
163 | dout("ceph_data_ready on %p state = %lu, queueing work\n", | |
164 | con, con->state); | |
165 | queue_con(con); | |
166 | } | |
167 | } | |
168 | ||
169 | /* socket has buffer space for writing */ | |
170 | static void ceph_write_space(struct sock *sk) | |
171 | { | |
d3002b97 | 172 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e | 173 | |
182fac26 JS |
174 | /* only queue to workqueue if there is data we want to write, |
175 | * and there is sufficient space in the socket buffer to accept | |
176 | * more data. clear SOCK_NOSPACE so that ceph_write_space() | |
177 | * doesn't get called again until try_write() fills the socket | |
178 | * buffer. See net/ipv4/tcp_input.c:tcp_check_space() | |
179 | * and net/core/stream.c:sk_stream_write_space(). | |
180 | */ | |
31b8006e | 181 | if (test_bit(WRITE_PENDING, &con->state)) { |
182fac26 JS |
182 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { |
183 | dout("ceph_write_space %p queueing write work\n", con); | |
184 | clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); | |
185 | queue_con(con); | |
186 | } | |
31b8006e SW |
187 | } else { |
188 | dout("ceph_write_space %p nothing to write\n", con); | |
189 | } | |
31b8006e SW |
190 | } |
191 | ||
192 | /* socket's state has changed */ | |
193 | static void ceph_state_change(struct sock *sk) | |
194 | { | |
bd406145 | 195 | struct ceph_connection *con = sk->sk_user_data; |
31b8006e SW |
196 | |
197 | dout("ceph_state_change %p state = %lu sk_state = %u\n", | |
198 | con, con->state, sk->sk_state); | |
199 | ||
200 | if (test_bit(CLOSED, &con->state)) | |
201 | return; | |
202 | ||
203 | switch (sk->sk_state) { | |
204 | case TCP_CLOSE: | |
205 | dout("ceph_state_change TCP_CLOSE\n"); | |
206 | case TCP_CLOSE_WAIT: | |
207 | dout("ceph_state_change TCP_CLOSE_WAIT\n"); | |
208 | if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) { | |
209 | if (test_bit(CONNECTING, &con->state)) | |
210 | con->error_msg = "connection failed"; | |
211 | else | |
212 | con->error_msg = "socket closed"; | |
213 | queue_con(con); | |
214 | } | |
215 | break; | |
216 | case TCP_ESTABLISHED: | |
217 | dout("ceph_state_change TCP_ESTABLISHED\n"); | |
218 | queue_con(con); | |
219 | break; | |
d3002b97 AE |
220 | default: /* Everything else is uninteresting */ |
221 | break; | |
31b8006e SW |
222 | } |
223 | } | |
224 | ||
225 | /* | |
226 | * set up socket callbacks | |
227 | */ | |
228 | static void set_sock_callbacks(struct socket *sock, | |
229 | struct ceph_connection *con) | |
230 | { | |
231 | struct sock *sk = sock->sk; | |
bd406145 | 232 | sk->sk_user_data = con; |
31b8006e SW |
233 | sk->sk_data_ready = ceph_data_ready; |
234 | sk->sk_write_space = ceph_write_space; | |
235 | sk->sk_state_change = ceph_state_change; | |
236 | } | |
237 | ||
238 | ||
239 | /* | |
240 | * socket helpers | |
241 | */ | |
242 | ||
243 | /* | |
244 | * initiate connection to a remote socket. | |
245 | */ | |
41617d0c | 246 | static int ceph_tcp_connect(struct ceph_connection *con) |
31b8006e | 247 | { |
f91d3471 | 248 | struct sockaddr_storage *paddr = &con->peer_addr.in_addr; |
31b8006e SW |
249 | struct socket *sock; |
250 | int ret; | |
251 | ||
252 | BUG_ON(con->sock); | |
f91d3471 SW |
253 | ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, |
254 | IPPROTO_TCP, &sock); | |
31b8006e | 255 | if (ret) |
41617d0c | 256 | return ret; |
31b8006e SW |
257 | sock->sk->sk_allocation = GFP_NOFS; |
258 | ||
a6a5349d SW |
259 | #ifdef CONFIG_LOCKDEP |
260 | lockdep_set_class(&sock->sk->sk_lock, &socket_class); | |
261 | #endif | |
262 | ||
31b8006e SW |
263 | set_sock_callbacks(sock, con); |
264 | ||
3d14c5d2 | 265 | dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 266 | |
f91d3471 SW |
267 | ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr), |
268 | O_NONBLOCK); | |
31b8006e SW |
269 | if (ret == -EINPROGRESS) { |
270 | dout("connect %s EINPROGRESS sk_state = %u\n", | |
3d14c5d2 | 271 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e | 272 | sock->sk->sk_state); |
a5bc3129 | 273 | } else if (ret < 0) { |
31b8006e | 274 | pr_err("connect %s error %d\n", |
3d14c5d2 | 275 | ceph_pr_addr(&con->peer_addr.in_addr), ret); |
31b8006e | 276 | sock_release(sock); |
31b8006e | 277 | con->error_msg = "connect error"; |
31b8006e | 278 | |
41617d0c | 279 | return ret; |
a5bc3129 AE |
280 | } |
281 | con->sock = sock; | |
282 | ||
41617d0c | 283 | return 0; |
31b8006e SW |
284 | } |
285 | ||
286 | static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len) | |
287 | { | |
288 | struct kvec iov = {buf, len}; | |
289 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
98bdb0aa | 290 | int r; |
31b8006e | 291 | |
98bdb0aa SW |
292 | r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags); |
293 | if (r == -EAGAIN) | |
294 | r = 0; | |
295 | return r; | |
31b8006e SW |
296 | } |
297 | ||
298 | /* | |
299 | * write something. @more is true if caller will be sending more data | |
300 | * shortly. | |
301 | */ | |
302 | static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov, | |
303 | size_t kvlen, size_t len, int more) | |
304 | { | |
305 | struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL }; | |
42961d23 | 306 | int r; |
31b8006e SW |
307 | |
308 | if (more) | |
309 | msg.msg_flags |= MSG_MORE; | |
310 | else | |
311 | msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */ | |
312 | ||
42961d23 SW |
313 | r = kernel_sendmsg(sock, &msg, iov, kvlen, len); |
314 | if (r == -EAGAIN) | |
315 | r = 0; | |
316 | return r; | |
31b8006e SW |
317 | } |
318 | ||
319 | ||
320 | /* | |
321 | * Shutdown/close the socket for the given connection. | |
322 | */ | |
323 | static int con_close_socket(struct ceph_connection *con) | |
324 | { | |
325 | int rc; | |
326 | ||
327 | dout("con_close_socket on %p sock %p\n", con, con->sock); | |
328 | if (!con->sock) | |
329 | return 0; | |
330 | set_bit(SOCK_CLOSED, &con->state); | |
331 | rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR); | |
332 | sock_release(con->sock); | |
333 | con->sock = NULL; | |
334 | clear_bit(SOCK_CLOSED, &con->state); | |
335 | return rc; | |
336 | } | |
337 | ||
338 | /* | |
339 | * Reset a connection. Discard all incoming and outgoing messages | |
340 | * and clear *_seq state. | |
341 | */ | |
342 | static void ceph_msg_remove(struct ceph_msg *msg) | |
343 | { | |
344 | list_del_init(&msg->list_head); | |
345 | ceph_msg_put(msg); | |
346 | } | |
347 | static void ceph_msg_remove_list(struct list_head *head) | |
348 | { | |
349 | while (!list_empty(head)) { | |
350 | struct ceph_msg *msg = list_first_entry(head, struct ceph_msg, | |
351 | list_head); | |
352 | ceph_msg_remove(msg); | |
353 | } | |
354 | } | |
355 | ||
356 | static void reset_connection(struct ceph_connection *con) | |
357 | { | |
358 | /* reset connection, out_queue, msg_ and connect_seq */ | |
359 | /* discard existing out_queue and msg_seq */ | |
31b8006e SW |
360 | ceph_msg_remove_list(&con->out_queue); |
361 | ceph_msg_remove_list(&con->out_sent); | |
362 | ||
cf3e5c40 SW |
363 | if (con->in_msg) { |
364 | ceph_msg_put(con->in_msg); | |
365 | con->in_msg = NULL; | |
366 | } | |
367 | ||
31b8006e SW |
368 | con->connect_seq = 0; |
369 | con->out_seq = 0; | |
c86a2930 SW |
370 | if (con->out_msg) { |
371 | ceph_msg_put(con->out_msg); | |
372 | con->out_msg = NULL; | |
373 | } | |
31b8006e | 374 | con->in_seq = 0; |
0e0d5e0c | 375 | con->in_seq_acked = 0; |
31b8006e SW |
376 | } |
377 | ||
378 | /* | |
379 | * mark a peer down. drop any open connections. | |
380 | */ | |
381 | void ceph_con_close(struct ceph_connection *con) | |
382 | { | |
3d14c5d2 YS |
383 | dout("con_close %p peer %s\n", con, |
384 | ceph_pr_addr(&con->peer_addr.in_addr)); | |
31b8006e SW |
385 | set_bit(CLOSED, &con->state); /* in case there's queued work */ |
386 | clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */ | |
1679f876 SW |
387 | clear_bit(LOSSYTX, &con->state); /* so we retry next connect */ |
388 | clear_bit(KEEPALIVE_PENDING, &con->state); | |
389 | clear_bit(WRITE_PENDING, &con->state); | |
ec302645 | 390 | mutex_lock(&con->mutex); |
31b8006e | 391 | reset_connection(con); |
6f2bc3ff | 392 | con->peer_global_seq = 0; |
91e45ce3 | 393 | cancel_delayed_work(&con->work); |
ec302645 | 394 | mutex_unlock(&con->mutex); |
31b8006e SW |
395 | queue_con(con); |
396 | } | |
3d14c5d2 | 397 | EXPORT_SYMBOL(ceph_con_close); |
31b8006e | 398 | |
31b8006e SW |
399 | /* |
400 | * Reopen a closed connection, with a new peer address. | |
401 | */ | |
402 | void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr) | |
403 | { | |
3d14c5d2 | 404 | dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr)); |
31b8006e SW |
405 | set_bit(OPENING, &con->state); |
406 | clear_bit(CLOSED, &con->state); | |
407 | memcpy(&con->peer_addr, addr, sizeof(*addr)); | |
03c677e1 | 408 | con->delay = 0; /* reset backoff memory */ |
31b8006e SW |
409 | queue_con(con); |
410 | } | |
3d14c5d2 | 411 | EXPORT_SYMBOL(ceph_con_open); |
31b8006e | 412 | |
87b315a5 SW |
413 | /* |
414 | * return true if this connection ever successfully opened | |
415 | */ | |
416 | bool ceph_con_opened(struct ceph_connection *con) | |
417 | { | |
418 | return con->connect_seq > 0; | |
419 | } | |
420 | ||
31b8006e SW |
421 | /* |
422 | * generic get/put | |
423 | */ | |
424 | struct ceph_connection *ceph_con_get(struct ceph_connection *con) | |
425 | { | |
d3002b97 AE |
426 | int nref = __atomic_add_unless(&con->nref, 1, 0); |
427 | ||
428 | dout("con_get %p nref = %d -> %d\n", con, nref, nref + 1); | |
429 | ||
430 | return nref ? con : NULL; | |
31b8006e SW |
431 | } |
432 | ||
433 | void ceph_con_put(struct ceph_connection *con) | |
434 | { | |
d3002b97 AE |
435 | int nref = atomic_dec_return(&con->nref); |
436 | ||
437 | BUG_ON(nref < 0); | |
438 | if (nref == 0) { | |
71ececda | 439 | BUG_ON(con->sock); |
31b8006e SW |
440 | kfree(con); |
441 | } | |
d3002b97 | 442 | dout("con_put %p nref = %d -> %d\n", con, nref + 1, nref); |
31b8006e SW |
443 | } |
444 | ||
445 | /* | |
446 | * initialize a new connection. | |
447 | */ | |
448 | void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con) | |
449 | { | |
450 | dout("con_init %p\n", con); | |
451 | memset(con, 0, sizeof(*con)); | |
452 | atomic_set(&con->nref, 1); | |
453 | con->msgr = msgr; | |
ec302645 | 454 | mutex_init(&con->mutex); |
31b8006e SW |
455 | INIT_LIST_HEAD(&con->out_queue); |
456 | INIT_LIST_HEAD(&con->out_sent); | |
457 | INIT_DELAYED_WORK(&con->work, con_work); | |
458 | } | |
3d14c5d2 | 459 | EXPORT_SYMBOL(ceph_con_init); |
31b8006e SW |
460 | |
461 | ||
462 | /* | |
463 | * We maintain a global counter to order connection attempts. Get | |
464 | * a unique seq greater than @gt. | |
465 | */ | |
466 | static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt) | |
467 | { | |
468 | u32 ret; | |
469 | ||
470 | spin_lock(&msgr->global_seq_lock); | |
471 | if (msgr->global_seq < gt) | |
472 | msgr->global_seq = gt; | |
473 | ret = ++msgr->global_seq; | |
474 | spin_unlock(&msgr->global_seq_lock); | |
475 | return ret; | |
476 | } | |
477 | ||
859eb799 AE |
478 | static void ceph_con_out_kvec_reset(struct ceph_connection *con) |
479 | { | |
480 | con->out_kvec_left = 0; | |
481 | con->out_kvec_bytes = 0; | |
482 | con->out_kvec_cur = &con->out_kvec[0]; | |
483 | } | |
484 | ||
485 | static void ceph_con_out_kvec_add(struct ceph_connection *con, | |
486 | size_t size, void *data) | |
487 | { | |
488 | int index; | |
489 | ||
490 | index = con->out_kvec_left; | |
491 | BUG_ON(index >= ARRAY_SIZE(con->out_kvec)); | |
492 | ||
493 | con->out_kvec[index].iov_len = size; | |
494 | con->out_kvec[index].iov_base = data; | |
495 | con->out_kvec_left++; | |
496 | con->out_kvec_bytes += size; | |
497 | } | |
31b8006e SW |
498 | |
499 | /* | |
500 | * Prepare footer for currently outgoing message, and finish things | |
501 | * off. Assumes out_kvec* are already valid.. we just add on to the end. | |
502 | */ | |
859eb799 | 503 | static void prepare_write_message_footer(struct ceph_connection *con) |
31b8006e SW |
504 | { |
505 | struct ceph_msg *m = con->out_msg; | |
859eb799 | 506 | int v = con->out_kvec_left; |
31b8006e SW |
507 | |
508 | dout("prepare_write_message_footer %p\n", con); | |
509 | con->out_kvec_is_msg = true; | |
510 | con->out_kvec[v].iov_base = &m->footer; | |
511 | con->out_kvec[v].iov_len = sizeof(m->footer); | |
512 | con->out_kvec_bytes += sizeof(m->footer); | |
513 | con->out_kvec_left++; | |
514 | con->out_more = m->more_to_follow; | |
c86a2930 | 515 | con->out_msg_done = true; |
31b8006e SW |
516 | } |
517 | ||
518 | /* | |
519 | * Prepare headers for the next outgoing message. | |
520 | */ | |
521 | static void prepare_write_message(struct ceph_connection *con) | |
522 | { | |
523 | struct ceph_msg *m; | |
a9a0c51a | 524 | u32 crc; |
31b8006e | 525 | |
859eb799 | 526 | ceph_con_out_kvec_reset(con); |
31b8006e | 527 | con->out_kvec_is_msg = true; |
c86a2930 | 528 | con->out_msg_done = false; |
31b8006e SW |
529 | |
530 | /* Sneak an ack in there first? If we can get it into the same | |
531 | * TCP packet that's a good thing. */ | |
532 | if (con->in_seq > con->in_seq_acked) { | |
533 | con->in_seq_acked = con->in_seq; | |
859eb799 | 534 | ceph_con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); |
31b8006e | 535 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
859eb799 AE |
536 | ceph_con_out_kvec_add(con, sizeof (con->out_temp_ack), |
537 | &con->out_temp_ack); | |
31b8006e SW |
538 | } |
539 | ||
859eb799 | 540 | m = list_first_entry(&con->out_queue, struct ceph_msg, list_head); |
c86a2930 | 541 | con->out_msg = m; |
4cf9d544 SW |
542 | |
543 | /* put message on sent list */ | |
544 | ceph_msg_get(m); | |
545 | list_move_tail(&m->list_head, &con->out_sent); | |
31b8006e | 546 | |
e84346b7 SW |
547 | /* |
548 | * only assign outgoing seq # if we haven't sent this message | |
549 | * yet. if it is requeued, resend with it's original seq. | |
550 | */ | |
551 | if (m->needs_out_seq) { | |
552 | m->hdr.seq = cpu_to_le64(++con->out_seq); | |
553 | m->needs_out_seq = false; | |
554 | } | |
31b8006e SW |
555 | |
556 | dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n", | |
557 | m, con->out_seq, le16_to_cpu(m->hdr.type), | |
558 | le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len), | |
559 | le32_to_cpu(m->hdr.data_len), | |
560 | m->nr_pages); | |
561 | BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len); | |
562 | ||
563 | /* tag + hdr + front + middle */ | |
859eb799 AE |
564 | ceph_con_out_kvec_add(con, sizeof (tag_msg), &tag_msg); |
565 | ceph_con_out_kvec_add(con, sizeof (m->hdr), &m->hdr); | |
566 | ceph_con_out_kvec_add(con, m->front.iov_len, m->front.iov_base); | |
567 | ||
31b8006e | 568 | if (m->middle) |
859eb799 AE |
569 | ceph_con_out_kvec_add(con, m->middle->vec.iov_len, |
570 | m->middle->vec.iov_base); | |
31b8006e SW |
571 | |
572 | /* fill in crc (except data pages), footer */ | |
a9a0c51a AE |
573 | crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc)); |
574 | con->out_msg->hdr.crc = cpu_to_le32(crc); | |
31b8006e | 575 | con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE; |
a9a0c51a AE |
576 | |
577 | crc = crc32c(0, m->front.iov_base, m->front.iov_len); | |
578 | con->out_msg->footer.front_crc = cpu_to_le32(crc); | |
579 | if (m->middle) { | |
580 | crc = crc32c(0, m->middle->vec.iov_base, | |
581 | m->middle->vec.iov_len); | |
582 | con->out_msg->footer.middle_crc = cpu_to_le32(crc); | |
583 | } else | |
31b8006e SW |
584 | con->out_msg->footer.middle_crc = 0; |
585 | con->out_msg->footer.data_crc = 0; | |
586 | dout("prepare_write_message front_crc %u data_crc %u\n", | |
587 | le32_to_cpu(con->out_msg->footer.front_crc), | |
588 | le32_to_cpu(con->out_msg->footer.middle_crc)); | |
589 | ||
590 | /* is there a data payload? */ | |
591 | if (le32_to_cpu(m->hdr.data_len) > 0) { | |
592 | /* initialize page iterator */ | |
593 | con->out_msg_pos.page = 0; | |
68b4476b | 594 | if (m->pages) |
c5c6b19d | 595 | con->out_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
596 | else |
597 | con->out_msg_pos.page_pos = 0; | |
31b8006e | 598 | con->out_msg_pos.data_pos = 0; |
bca064d2 | 599 | con->out_msg_pos.did_page_crc = false; |
31b8006e SW |
600 | con->out_more = 1; /* data + footer will follow */ |
601 | } else { | |
602 | /* no, queue up footer too and be done */ | |
859eb799 | 603 | prepare_write_message_footer(con); |
31b8006e SW |
604 | } |
605 | ||
606 | set_bit(WRITE_PENDING, &con->state); | |
607 | } | |
608 | ||
609 | /* | |
610 | * Prepare an ack. | |
611 | */ | |
612 | static void prepare_write_ack(struct ceph_connection *con) | |
613 | { | |
614 | dout("prepare_write_ack %p %llu -> %llu\n", con, | |
615 | con->in_seq_acked, con->in_seq); | |
616 | con->in_seq_acked = con->in_seq; | |
617 | ||
859eb799 AE |
618 | ceph_con_out_kvec_reset(con); |
619 | ||
620 | ceph_con_out_kvec_add(con, sizeof (tag_ack), &tag_ack); | |
621 | ||
31b8006e | 622 | con->out_temp_ack = cpu_to_le64(con->in_seq_acked); |
859eb799 AE |
623 | ceph_con_out_kvec_add(con, sizeof (con->out_temp_ack), |
624 | &con->out_temp_ack); | |
625 | ||
31b8006e SW |
626 | con->out_more = 1; /* more will follow.. eventually.. */ |
627 | set_bit(WRITE_PENDING, &con->state); | |
628 | } | |
629 | ||
630 | /* | |
631 | * Prepare to write keepalive byte. | |
632 | */ | |
633 | static void prepare_write_keepalive(struct ceph_connection *con) | |
634 | { | |
635 | dout("prepare_write_keepalive %p\n", con); | |
859eb799 AE |
636 | ceph_con_out_kvec_reset(con); |
637 | ceph_con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive); | |
31b8006e SW |
638 | set_bit(WRITE_PENDING, &con->state); |
639 | } | |
640 | ||
641 | /* | |
642 | * Connection negotiation. | |
643 | */ | |
644 | ||
0da5d703 | 645 | static int prepare_connect_authorizer(struct ceph_connection *con) |
4e7a5dcd SW |
646 | { |
647 | void *auth_buf; | |
648 | int auth_len = 0; | |
649 | int auth_protocol = 0; | |
650 | ||
ec302645 | 651 | mutex_unlock(&con->mutex); |
4e7a5dcd SW |
652 | if (con->ops->get_authorizer) |
653 | con->ops->get_authorizer(con, &auth_buf, &auth_len, | |
654 | &auth_protocol, &con->auth_reply_buf, | |
655 | &con->auth_reply_buf_len, | |
656 | con->auth_retry); | |
ec302645 | 657 | mutex_lock(&con->mutex); |
4e7a5dcd | 658 | |
0da5d703 SW |
659 | if (test_bit(CLOSED, &con->state) || |
660 | test_bit(OPENING, &con->state)) | |
661 | return -EAGAIN; | |
662 | ||
4e7a5dcd SW |
663 | con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol); |
664 | con->out_connect.authorizer_len = cpu_to_le32(auth_len); | |
665 | ||
859eb799 AE |
666 | if (auth_len) |
667 | ceph_con_out_kvec_add(con, auth_len, auth_buf); | |
668 | ||
0da5d703 | 669 | return 0; |
4e7a5dcd SW |
670 | } |
671 | ||
31b8006e SW |
672 | /* |
673 | * We connected to a peer and are saying hello. | |
674 | */ | |
eed0ef2c SW |
675 | static void prepare_write_banner(struct ceph_messenger *msgr, |
676 | struct ceph_connection *con) | |
31b8006e | 677 | { |
859eb799 AE |
678 | ceph_con_out_kvec_reset(con); |
679 | ceph_con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER); | |
680 | ceph_con_out_kvec_add(con, sizeof (msgr->my_enc_addr), | |
681 | &msgr->my_enc_addr); | |
eed0ef2c | 682 | |
eed0ef2c SW |
683 | con->out_more = 0; |
684 | set_bit(WRITE_PENDING, &con->state); | |
685 | } | |
686 | ||
0da5d703 SW |
687 | static int prepare_write_connect(struct ceph_messenger *msgr, |
688 | struct ceph_connection *con, | |
963be4d7 | 689 | int include_banner) |
eed0ef2c | 690 | { |
31b8006e SW |
691 | unsigned global_seq = get_global_seq(con->msgr, 0); |
692 | int proto; | |
693 | ||
694 | switch (con->peer_name.type) { | |
695 | case CEPH_ENTITY_TYPE_MON: | |
696 | proto = CEPH_MONC_PROTOCOL; | |
697 | break; | |
698 | case CEPH_ENTITY_TYPE_OSD: | |
699 | proto = CEPH_OSDC_PROTOCOL; | |
700 | break; | |
701 | case CEPH_ENTITY_TYPE_MDS: | |
702 | proto = CEPH_MDSC_PROTOCOL; | |
703 | break; | |
704 | default: | |
705 | BUG(); | |
706 | } | |
707 | ||
708 | dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con, | |
709 | con->connect_seq, global_seq, proto); | |
4e7a5dcd | 710 | |
3d14c5d2 | 711 | con->out_connect.features = cpu_to_le64(msgr->supported_features); |
31b8006e SW |
712 | con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT); |
713 | con->out_connect.connect_seq = cpu_to_le32(con->connect_seq); | |
714 | con->out_connect.global_seq = cpu_to_le32(global_seq); | |
715 | con->out_connect.protocol_version = cpu_to_le32(proto); | |
716 | con->out_connect.flags = 0; | |
31b8006e | 717 | |
963be4d7 AE |
718 | if (include_banner) |
719 | prepare_write_banner(msgr, con); | |
859eb799 AE |
720 | else |
721 | ceph_con_out_kvec_reset(con); | |
722 | ceph_con_out_kvec_add(con, sizeof (con->out_connect), &con->out_connect); | |
723 | ||
31b8006e SW |
724 | con->out_more = 0; |
725 | set_bit(WRITE_PENDING, &con->state); | |
4e7a5dcd | 726 | |
0da5d703 | 727 | return prepare_connect_authorizer(con); |
31b8006e SW |
728 | } |
729 | ||
31b8006e SW |
730 | /* |
731 | * write as much of pending kvecs to the socket as we can. | |
732 | * 1 -> done | |
733 | * 0 -> socket full, but more to do | |
734 | * <0 -> error | |
735 | */ | |
736 | static int write_partial_kvec(struct ceph_connection *con) | |
737 | { | |
738 | int ret; | |
739 | ||
740 | dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes); | |
741 | while (con->out_kvec_bytes > 0) { | |
742 | ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur, | |
743 | con->out_kvec_left, con->out_kvec_bytes, | |
744 | con->out_more); | |
745 | if (ret <= 0) | |
746 | goto out; | |
747 | con->out_kvec_bytes -= ret; | |
748 | if (con->out_kvec_bytes == 0) | |
749 | break; /* done */ | |
750 | while (ret > 0) { | |
751 | if (ret >= con->out_kvec_cur->iov_len) { | |
752 | ret -= con->out_kvec_cur->iov_len; | |
753 | con->out_kvec_cur++; | |
754 | con->out_kvec_left--; | |
755 | } else { | |
756 | con->out_kvec_cur->iov_len -= ret; | |
757 | con->out_kvec_cur->iov_base += ret; | |
758 | ret = 0; | |
759 | break; | |
760 | } | |
761 | } | |
762 | } | |
763 | con->out_kvec_left = 0; | |
764 | con->out_kvec_is_msg = false; | |
765 | ret = 1; | |
766 | out: | |
767 | dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con, | |
768 | con->out_kvec_bytes, con->out_kvec_left, ret); | |
769 | return ret; /* done! */ | |
770 | } | |
771 | ||
68b4476b YS |
772 | #ifdef CONFIG_BLOCK |
773 | static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg) | |
774 | { | |
775 | if (!bio) { | |
776 | *iter = NULL; | |
777 | *seg = 0; | |
778 | return; | |
779 | } | |
780 | *iter = bio; | |
781 | *seg = bio->bi_idx; | |
782 | } | |
783 | ||
784 | static void iter_bio_next(struct bio **bio_iter, int *seg) | |
785 | { | |
786 | if (*bio_iter == NULL) | |
787 | return; | |
788 | ||
789 | BUG_ON(*seg >= (*bio_iter)->bi_vcnt); | |
790 | ||
791 | (*seg)++; | |
792 | if (*seg == (*bio_iter)->bi_vcnt) | |
793 | init_bio_iter((*bio_iter)->bi_next, bio_iter, seg); | |
794 | } | |
795 | #endif | |
796 | ||
31b8006e SW |
797 | /* |
798 | * Write as much message data payload as we can. If we finish, queue | |
799 | * up the footer. | |
800 | * 1 -> done, footer is now queued in out_kvec[]. | |
801 | * 0 -> socket full, but more to do | |
802 | * <0 -> error | |
803 | */ | |
804 | static int write_partial_msg_pages(struct ceph_connection *con) | |
805 | { | |
806 | struct ceph_msg *msg = con->out_msg; | |
807 | unsigned data_len = le32_to_cpu(msg->hdr.data_len); | |
808 | size_t len; | |
bca064d2 | 809 | bool do_crc = con->msgr->nocrc; |
31b8006e | 810 | int ret; |
68b4476b YS |
811 | int total_max_write; |
812 | int in_trail = 0; | |
813 | size_t trail_len = (msg->trail ? msg->trail->length : 0); | |
31b8006e SW |
814 | |
815 | dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n", | |
816 | con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages, | |
817 | con->out_msg_pos.page_pos); | |
818 | ||
68b4476b YS |
819 | #ifdef CONFIG_BLOCK |
820 | if (msg->bio && !msg->bio_iter) | |
821 | init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg); | |
822 | #endif | |
823 | ||
824 | while (data_len > con->out_msg_pos.data_pos) { | |
31b8006e SW |
825 | struct page *page = NULL; |
826 | void *kaddr = NULL; | |
68b4476b YS |
827 | int max_write = PAGE_SIZE; |
828 | int page_shift = 0; | |
829 | ||
830 | total_max_write = data_len - trail_len - | |
831 | con->out_msg_pos.data_pos; | |
31b8006e SW |
832 | |
833 | /* | |
834 | * if we are calculating the data crc (the default), we need | |
835 | * to map the page. if our pages[] has been revoked, use the | |
836 | * zero page. | |
837 | */ | |
68b4476b YS |
838 | |
839 | /* have we reached the trail part of the data? */ | |
840 | if (con->out_msg_pos.data_pos >= data_len - trail_len) { | |
841 | in_trail = 1; | |
842 | ||
843 | total_max_write = data_len - con->out_msg_pos.data_pos; | |
844 | ||
845 | page = list_first_entry(&msg->trail->head, | |
846 | struct page, lru); | |
bca064d2 | 847 | if (do_crc) |
68b4476b YS |
848 | kaddr = kmap(page); |
849 | max_write = PAGE_SIZE; | |
850 | } else if (msg->pages) { | |
31b8006e | 851 | page = msg->pages[con->out_msg_pos.page]; |
bca064d2 | 852 | if (do_crc) |
31b8006e | 853 | kaddr = kmap(page); |
58bb3b37 SW |
854 | } else if (msg->pagelist) { |
855 | page = list_first_entry(&msg->pagelist->head, | |
856 | struct page, lru); | |
bca064d2 | 857 | if (do_crc) |
58bb3b37 | 858 | kaddr = kmap(page); |
68b4476b YS |
859 | #ifdef CONFIG_BLOCK |
860 | } else if (msg->bio) { | |
861 | struct bio_vec *bv; | |
862 | ||
863 | bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg); | |
864 | page = bv->bv_page; | |
865 | page_shift = bv->bv_offset; | |
bca064d2 | 866 | if (do_crc) |
68b4476b YS |
867 | kaddr = kmap(page) + page_shift; |
868 | max_write = bv->bv_len; | |
869 | #endif | |
31b8006e | 870 | } else { |
57666519 | 871 | page = zero_page; |
bca064d2 | 872 | if (do_crc) |
57666519 | 873 | kaddr = zero_page_address; |
31b8006e | 874 | } |
68b4476b YS |
875 | len = min_t(int, max_write - con->out_msg_pos.page_pos, |
876 | total_max_write); | |
877 | ||
bca064d2 | 878 | if (do_crc && !con->out_msg_pos.did_page_crc) { |
a9a0c51a | 879 | u32 crc; |
31b8006e SW |
880 | void *base = kaddr + con->out_msg_pos.page_pos; |
881 | u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc); | |
882 | ||
883 | BUG_ON(kaddr == NULL); | |
a9a0c51a AE |
884 | crc = crc32c(tmpcrc, base, len); |
885 | con->out_msg->footer.data_crc = cpu_to_le32(crc); | |
bca064d2 | 886 | con->out_msg_pos.did_page_crc = true; |
31b8006e | 887 | } |
31b8006e | 888 | ret = kernel_sendpage(con->sock, page, |
68b4476b YS |
889 | con->out_msg_pos.page_pos + page_shift, |
890 | len, | |
31b8006e SW |
891 | MSG_DONTWAIT | MSG_NOSIGNAL | |
892 | MSG_MORE); | |
893 | ||
bca064d2 | 894 | if (do_crc && |
68b4476b | 895 | (msg->pages || msg->pagelist || msg->bio || in_trail)) |
31b8006e SW |
896 | kunmap(page); |
897 | ||
42961d23 SW |
898 | if (ret == -EAGAIN) |
899 | ret = 0; | |
31b8006e SW |
900 | if (ret <= 0) |
901 | goto out; | |
902 | ||
903 | con->out_msg_pos.data_pos += ret; | |
904 | con->out_msg_pos.page_pos += ret; | |
905 | if (ret == len) { | |
906 | con->out_msg_pos.page_pos = 0; | |
907 | con->out_msg_pos.page++; | |
bca064d2 | 908 | con->out_msg_pos.did_page_crc = false; |
68b4476b YS |
909 | if (in_trail) |
910 | list_move_tail(&page->lru, | |
911 | &msg->trail->head); | |
912 | else if (msg->pagelist) | |
58bb3b37 SW |
913 | list_move_tail(&page->lru, |
914 | &msg->pagelist->head); | |
68b4476b YS |
915 | #ifdef CONFIG_BLOCK |
916 | else if (msg->bio) | |
917 | iter_bio_next(&msg->bio_iter, &msg->bio_seg); | |
918 | #endif | |
31b8006e SW |
919 | } |
920 | } | |
921 | ||
922 | dout("write_partial_msg_pages %p msg %p done\n", con, msg); | |
923 | ||
924 | /* prepare and queue up footer, too */ | |
bca064d2 | 925 | if (!do_crc) |
31b8006e | 926 | con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC; |
859eb799 AE |
927 | ceph_con_out_kvec_reset(con); |
928 | prepare_write_message_footer(con); | |
31b8006e SW |
929 | ret = 1; |
930 | out: | |
931 | return ret; | |
932 | } | |
933 | ||
934 | /* | |
935 | * write some zeros | |
936 | */ | |
937 | static int write_partial_skip(struct ceph_connection *con) | |
938 | { | |
939 | int ret; | |
940 | ||
941 | while (con->out_skip > 0) { | |
942 | struct kvec iov = { | |
57666519 | 943 | .iov_base = zero_page_address, |
31b8006e SW |
944 | .iov_len = min(con->out_skip, (int)PAGE_CACHE_SIZE) |
945 | }; | |
946 | ||
947 | ret = ceph_tcp_sendmsg(con->sock, &iov, 1, iov.iov_len, 1); | |
948 | if (ret <= 0) | |
949 | goto out; | |
950 | con->out_skip -= ret; | |
951 | } | |
952 | ret = 1; | |
953 | out: | |
954 | return ret; | |
955 | } | |
956 | ||
957 | /* | |
958 | * Prepare to read connection handshake, or an ack. | |
959 | */ | |
eed0ef2c SW |
960 | static void prepare_read_banner(struct ceph_connection *con) |
961 | { | |
962 | dout("prepare_read_banner %p\n", con); | |
963 | con->in_base_pos = 0; | |
964 | } | |
965 | ||
31b8006e SW |
966 | static void prepare_read_connect(struct ceph_connection *con) |
967 | { | |
968 | dout("prepare_read_connect %p\n", con); | |
969 | con->in_base_pos = 0; | |
970 | } | |
971 | ||
972 | static void prepare_read_ack(struct ceph_connection *con) | |
973 | { | |
974 | dout("prepare_read_ack %p\n", con); | |
975 | con->in_base_pos = 0; | |
976 | } | |
977 | ||
978 | static void prepare_read_tag(struct ceph_connection *con) | |
979 | { | |
980 | dout("prepare_read_tag %p\n", con); | |
981 | con->in_base_pos = 0; | |
982 | con->in_tag = CEPH_MSGR_TAG_READY; | |
983 | } | |
984 | ||
985 | /* | |
986 | * Prepare to read a message. | |
987 | */ | |
988 | static int prepare_read_message(struct ceph_connection *con) | |
989 | { | |
990 | dout("prepare_read_message %p\n", con); | |
991 | BUG_ON(con->in_msg != NULL); | |
992 | con->in_base_pos = 0; | |
993 | con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0; | |
994 | return 0; | |
995 | } | |
996 | ||
997 | ||
998 | static int read_partial(struct ceph_connection *con, | |
999 | int *to, int size, void *object) | |
1000 | { | |
1001 | *to += size; | |
1002 | while (con->in_base_pos < *to) { | |
1003 | int left = *to - con->in_base_pos; | |
1004 | int have = size - left; | |
1005 | int ret = ceph_tcp_recvmsg(con->sock, object + have, left); | |
1006 | if (ret <= 0) | |
1007 | return ret; | |
1008 | con->in_base_pos += ret; | |
1009 | } | |
1010 | return 1; | |
1011 | } | |
1012 | ||
1013 | ||
1014 | /* | |
1015 | * Read all or part of the connect-side handshake on a new connection | |
1016 | */ | |
eed0ef2c | 1017 | static int read_partial_banner(struct ceph_connection *con) |
31b8006e SW |
1018 | { |
1019 | int ret, to = 0; | |
1020 | ||
eed0ef2c | 1021 | dout("read_partial_banner %p at %d\n", con, con->in_base_pos); |
31b8006e SW |
1022 | |
1023 | /* peer's banner */ | |
1024 | ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner); | |
1025 | if (ret <= 0) | |
1026 | goto out; | |
1027 | ret = read_partial(con, &to, sizeof(con->actual_peer_addr), | |
1028 | &con->actual_peer_addr); | |
1029 | if (ret <= 0) | |
1030 | goto out; | |
1031 | ret = read_partial(con, &to, sizeof(con->peer_addr_for_me), | |
1032 | &con->peer_addr_for_me); | |
1033 | if (ret <= 0) | |
1034 | goto out; | |
eed0ef2c SW |
1035 | out: |
1036 | return ret; | |
1037 | } | |
1038 | ||
1039 | static int read_partial_connect(struct ceph_connection *con) | |
1040 | { | |
1041 | int ret, to = 0; | |
1042 | ||
1043 | dout("read_partial_connect %p at %d\n", con, con->in_base_pos); | |
1044 | ||
31b8006e SW |
1045 | ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply); |
1046 | if (ret <= 0) | |
1047 | goto out; | |
4e7a5dcd SW |
1048 | ret = read_partial(con, &to, le32_to_cpu(con->in_reply.authorizer_len), |
1049 | con->auth_reply_buf); | |
1050 | if (ret <= 0) | |
1051 | goto out; | |
31b8006e | 1052 | |
4e7a5dcd SW |
1053 | dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n", |
1054 | con, (int)con->in_reply.tag, | |
1055 | le32_to_cpu(con->in_reply.connect_seq), | |
31b8006e SW |
1056 | le32_to_cpu(con->in_reply.global_seq)); |
1057 | out: | |
1058 | return ret; | |
eed0ef2c | 1059 | |
31b8006e SW |
1060 | } |
1061 | ||
1062 | /* | |
1063 | * Verify the hello banner looks okay. | |
1064 | */ | |
1065 | static int verify_hello(struct ceph_connection *con) | |
1066 | { | |
1067 | if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) { | |
13e38c8a | 1068 | pr_err("connect to %s got bad banner\n", |
3d14c5d2 | 1069 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
1070 | con->error_msg = "protocol error, bad banner"; |
1071 | return -1; | |
1072 | } | |
1073 | return 0; | |
1074 | } | |
1075 | ||
1076 | static bool addr_is_blank(struct sockaddr_storage *ss) | |
1077 | { | |
1078 | switch (ss->ss_family) { | |
1079 | case AF_INET: | |
1080 | return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0; | |
1081 | case AF_INET6: | |
1082 | return | |
1083 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 && | |
1084 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 && | |
1085 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 && | |
1086 | ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0; | |
1087 | } | |
1088 | return false; | |
1089 | } | |
1090 | ||
1091 | static int addr_port(struct sockaddr_storage *ss) | |
1092 | { | |
1093 | switch (ss->ss_family) { | |
1094 | case AF_INET: | |
f28bcfbe | 1095 | return ntohs(((struct sockaddr_in *)ss)->sin_port); |
31b8006e | 1096 | case AF_INET6: |
f28bcfbe | 1097 | return ntohs(((struct sockaddr_in6 *)ss)->sin6_port); |
31b8006e SW |
1098 | } |
1099 | return 0; | |
1100 | } | |
1101 | ||
1102 | static void addr_set_port(struct sockaddr_storage *ss, int p) | |
1103 | { | |
1104 | switch (ss->ss_family) { | |
1105 | case AF_INET: | |
1106 | ((struct sockaddr_in *)ss)->sin_port = htons(p); | |
a2a79609 | 1107 | break; |
31b8006e SW |
1108 | case AF_INET6: |
1109 | ((struct sockaddr_in6 *)ss)->sin6_port = htons(p); | |
a2a79609 | 1110 | break; |
31b8006e SW |
1111 | } |
1112 | } | |
1113 | ||
ee3b56f2 NW |
1114 | /* |
1115 | * Unlike other *_pton function semantics, zero indicates success. | |
1116 | */ | |
1117 | static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss, | |
1118 | char delim, const char **ipend) | |
1119 | { | |
99f0f3b2 AE |
1120 | struct sockaddr_in *in4 = (struct sockaddr_in *) ss; |
1121 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss; | |
ee3b56f2 NW |
1122 | |
1123 | memset(ss, 0, sizeof(*ss)); | |
1124 | ||
1125 | if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) { | |
1126 | ss->ss_family = AF_INET; | |
1127 | return 0; | |
1128 | } | |
1129 | ||
1130 | if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) { | |
1131 | ss->ss_family = AF_INET6; | |
1132 | return 0; | |
1133 | } | |
1134 | ||
1135 | return -EINVAL; | |
1136 | } | |
1137 | ||
1138 | /* | |
1139 | * Extract hostname string and resolve using kernel DNS facility. | |
1140 | */ | |
1141 | #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER | |
1142 | static int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1143 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1144 | { | |
1145 | const char *end, *delim_p; | |
1146 | char *colon_p, *ip_addr = NULL; | |
1147 | int ip_len, ret; | |
1148 | ||
1149 | /* | |
1150 | * The end of the hostname occurs immediately preceding the delimiter or | |
1151 | * the port marker (':') where the delimiter takes precedence. | |
1152 | */ | |
1153 | delim_p = memchr(name, delim, namelen); | |
1154 | colon_p = memchr(name, ':', namelen); | |
1155 | ||
1156 | if (delim_p && colon_p) | |
1157 | end = delim_p < colon_p ? delim_p : colon_p; | |
1158 | else if (!delim_p && colon_p) | |
1159 | end = colon_p; | |
1160 | else { | |
1161 | end = delim_p; | |
1162 | if (!end) /* case: hostname:/ */ | |
1163 | end = name + namelen; | |
1164 | } | |
1165 | ||
1166 | if (end <= name) | |
1167 | return -EINVAL; | |
1168 | ||
1169 | /* do dns_resolve upcall */ | |
1170 | ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL); | |
1171 | if (ip_len > 0) | |
1172 | ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL); | |
1173 | else | |
1174 | ret = -ESRCH; | |
1175 | ||
1176 | kfree(ip_addr); | |
1177 | ||
1178 | *ipend = end; | |
1179 | ||
1180 | pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name, | |
1181 | ret, ret ? "failed" : ceph_pr_addr(ss)); | |
1182 | ||
1183 | return ret; | |
1184 | } | |
1185 | #else | |
1186 | static inline int ceph_dns_resolve_name(const char *name, size_t namelen, | |
1187 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1188 | { | |
1189 | return -EINVAL; | |
1190 | } | |
1191 | #endif | |
1192 | ||
1193 | /* | |
1194 | * Parse a server name (IP or hostname). If a valid IP address is not found | |
1195 | * then try to extract a hostname to resolve using userspace DNS upcall. | |
1196 | */ | |
1197 | static int ceph_parse_server_name(const char *name, size_t namelen, | |
1198 | struct sockaddr_storage *ss, char delim, const char **ipend) | |
1199 | { | |
1200 | int ret; | |
1201 | ||
1202 | ret = ceph_pton(name, namelen, ss, delim, ipend); | |
1203 | if (ret) | |
1204 | ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend); | |
1205 | ||
1206 | return ret; | |
1207 | } | |
1208 | ||
31b8006e SW |
1209 | /* |
1210 | * Parse an ip[:port] list into an addr array. Use the default | |
1211 | * monitor port if a port isn't specified. | |
1212 | */ | |
1213 | int ceph_parse_ips(const char *c, const char *end, | |
1214 | struct ceph_entity_addr *addr, | |
1215 | int max_count, int *count) | |
1216 | { | |
ee3b56f2 | 1217 | int i, ret = -EINVAL; |
31b8006e SW |
1218 | const char *p = c; |
1219 | ||
1220 | dout("parse_ips on '%.*s'\n", (int)(end-c), c); | |
1221 | for (i = 0; i < max_count; i++) { | |
1222 | const char *ipend; | |
1223 | struct sockaddr_storage *ss = &addr[i].in_addr; | |
31b8006e | 1224 | int port; |
39139f64 SW |
1225 | char delim = ','; |
1226 | ||
1227 | if (*p == '[') { | |
1228 | delim = ']'; | |
1229 | p++; | |
1230 | } | |
31b8006e | 1231 | |
ee3b56f2 NW |
1232 | ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend); |
1233 | if (ret) | |
31b8006e | 1234 | goto bad; |
ee3b56f2 NW |
1235 | ret = -EINVAL; |
1236 | ||
31b8006e SW |
1237 | p = ipend; |
1238 | ||
39139f64 SW |
1239 | if (delim == ']') { |
1240 | if (*p != ']') { | |
1241 | dout("missing matching ']'\n"); | |
1242 | goto bad; | |
1243 | } | |
1244 | p++; | |
1245 | } | |
1246 | ||
31b8006e SW |
1247 | /* port? */ |
1248 | if (p < end && *p == ':') { | |
1249 | port = 0; | |
1250 | p++; | |
1251 | while (p < end && *p >= '0' && *p <= '9') { | |
1252 | port = (port * 10) + (*p - '0'); | |
1253 | p++; | |
1254 | } | |
1255 | if (port > 65535 || port == 0) | |
1256 | goto bad; | |
1257 | } else { | |
1258 | port = CEPH_MON_PORT; | |
1259 | } | |
1260 | ||
1261 | addr_set_port(ss, port); | |
1262 | ||
3d14c5d2 | 1263 | dout("parse_ips got %s\n", ceph_pr_addr(ss)); |
31b8006e SW |
1264 | |
1265 | if (p == end) | |
1266 | break; | |
1267 | if (*p != ',') | |
1268 | goto bad; | |
1269 | p++; | |
1270 | } | |
1271 | ||
1272 | if (p != end) | |
1273 | goto bad; | |
1274 | ||
1275 | if (count) | |
1276 | *count = i + 1; | |
1277 | return 0; | |
1278 | ||
1279 | bad: | |
39139f64 | 1280 | pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c); |
ee3b56f2 | 1281 | return ret; |
31b8006e | 1282 | } |
3d14c5d2 | 1283 | EXPORT_SYMBOL(ceph_parse_ips); |
31b8006e | 1284 | |
eed0ef2c | 1285 | static int process_banner(struct ceph_connection *con) |
31b8006e | 1286 | { |
eed0ef2c | 1287 | dout("process_banner on %p\n", con); |
31b8006e SW |
1288 | |
1289 | if (verify_hello(con) < 0) | |
1290 | return -1; | |
1291 | ||
63f2d211 SW |
1292 | ceph_decode_addr(&con->actual_peer_addr); |
1293 | ceph_decode_addr(&con->peer_addr_for_me); | |
1294 | ||
31b8006e SW |
1295 | /* |
1296 | * Make sure the other end is who we wanted. note that the other | |
1297 | * end may not yet know their ip address, so if it's 0.0.0.0, give | |
1298 | * them the benefit of the doubt. | |
1299 | */ | |
103e2d3a SW |
1300 | if (memcmp(&con->peer_addr, &con->actual_peer_addr, |
1301 | sizeof(con->peer_addr)) != 0 && | |
31b8006e SW |
1302 | !(addr_is_blank(&con->actual_peer_addr.in_addr) && |
1303 | con->actual_peer_addr.nonce == con->peer_addr.nonce)) { | |
cd84db6e | 1304 | pr_warning("wrong peer, want %s/%d, got %s/%d\n", |
3d14c5d2 | 1305 | ceph_pr_addr(&con->peer_addr.in_addr), |
cd84db6e | 1306 | (int)le32_to_cpu(con->peer_addr.nonce), |
3d14c5d2 | 1307 | ceph_pr_addr(&con->actual_peer_addr.in_addr), |
cd84db6e | 1308 | (int)le32_to_cpu(con->actual_peer_addr.nonce)); |
58bb3b37 | 1309 | con->error_msg = "wrong peer at address"; |
31b8006e SW |
1310 | return -1; |
1311 | } | |
1312 | ||
1313 | /* | |
1314 | * did we learn our address? | |
1315 | */ | |
1316 | if (addr_is_blank(&con->msgr->inst.addr.in_addr)) { | |
1317 | int port = addr_port(&con->msgr->inst.addr.in_addr); | |
1318 | ||
1319 | memcpy(&con->msgr->inst.addr.in_addr, | |
1320 | &con->peer_addr_for_me.in_addr, | |
1321 | sizeof(con->peer_addr_for_me.in_addr)); | |
1322 | addr_set_port(&con->msgr->inst.addr.in_addr, port); | |
63f2d211 | 1323 | encode_my_addr(con->msgr); |
eed0ef2c | 1324 | dout("process_banner learned my addr is %s\n", |
3d14c5d2 | 1325 | ceph_pr_addr(&con->msgr->inst.addr.in_addr)); |
31b8006e SW |
1326 | } |
1327 | ||
eed0ef2c SW |
1328 | set_bit(NEGOTIATING, &con->state); |
1329 | prepare_read_connect(con); | |
1330 | return 0; | |
1331 | } | |
1332 | ||
04a419f9 SW |
1333 | static void fail_protocol(struct ceph_connection *con) |
1334 | { | |
1335 | reset_connection(con); | |
1336 | set_bit(CLOSED, &con->state); /* in case there's queued work */ | |
1337 | ||
1338 | mutex_unlock(&con->mutex); | |
1339 | if (con->ops->bad_proto) | |
1340 | con->ops->bad_proto(con); | |
1341 | mutex_lock(&con->mutex); | |
1342 | } | |
1343 | ||
eed0ef2c SW |
1344 | static int process_connect(struct ceph_connection *con) |
1345 | { | |
3d14c5d2 YS |
1346 | u64 sup_feat = con->msgr->supported_features; |
1347 | u64 req_feat = con->msgr->required_features; | |
04a419f9 | 1348 | u64 server_feat = le64_to_cpu(con->in_reply.features); |
0da5d703 | 1349 | int ret; |
04a419f9 | 1350 | |
eed0ef2c SW |
1351 | dout("process_connect on %p tag %d\n", con, (int)con->in_tag); |
1352 | ||
31b8006e | 1353 | switch (con->in_reply.tag) { |
04a419f9 SW |
1354 | case CEPH_MSGR_TAG_FEATURES: |
1355 | pr_err("%s%lld %s feature set mismatch," | |
1356 | " my %llx < server's %llx, missing %llx\n", | |
1357 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1358 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1359 | sup_feat, server_feat, server_feat & ~sup_feat); |
1360 | con->error_msg = "missing required protocol features"; | |
1361 | fail_protocol(con); | |
1362 | return -1; | |
1363 | ||
31b8006e | 1364 | case CEPH_MSGR_TAG_BADPROTOVER: |
31b8006e SW |
1365 | pr_err("%s%lld %s protocol version mismatch," |
1366 | " my %d != server's %d\n", | |
1367 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1368 | ceph_pr_addr(&con->peer_addr.in_addr), |
31b8006e SW |
1369 | le32_to_cpu(con->out_connect.protocol_version), |
1370 | le32_to_cpu(con->in_reply.protocol_version)); | |
1371 | con->error_msg = "protocol version mismatch"; | |
04a419f9 | 1372 | fail_protocol(con); |
31b8006e SW |
1373 | return -1; |
1374 | ||
4e7a5dcd SW |
1375 | case CEPH_MSGR_TAG_BADAUTHORIZER: |
1376 | con->auth_retry++; | |
1377 | dout("process_connect %p got BADAUTHORIZER attempt %d\n", con, | |
1378 | con->auth_retry); | |
1379 | if (con->auth_retry == 2) { | |
1380 | con->error_msg = "connect authorization failure"; | |
4e7a5dcd SW |
1381 | return -1; |
1382 | } | |
1383 | con->auth_retry = 1; | |
0da5d703 SW |
1384 | ret = prepare_write_connect(con->msgr, con, 0); |
1385 | if (ret < 0) | |
1386 | return ret; | |
63733a0f | 1387 | prepare_read_connect(con); |
4e7a5dcd | 1388 | break; |
31b8006e SW |
1389 | |
1390 | case CEPH_MSGR_TAG_RESETSESSION: | |
1391 | /* | |
1392 | * If we connected with a large connect_seq but the peer | |
1393 | * has no record of a session with us (no connection, or | |
1394 | * connect_seq == 0), they will send RESETSESION to indicate | |
1395 | * that they must have reset their session, and may have | |
1396 | * dropped messages. | |
1397 | */ | |
1398 | dout("process_connect got RESET peer seq %u\n", | |
1399 | le32_to_cpu(con->in_connect.connect_seq)); | |
1400 | pr_err("%s%lld %s connection reset\n", | |
1401 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1402 | ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e | 1403 | reset_connection(con); |
eed0ef2c | 1404 | prepare_write_connect(con->msgr, con, 0); |
31b8006e SW |
1405 | prepare_read_connect(con); |
1406 | ||
1407 | /* Tell ceph about it. */ | |
ec302645 | 1408 | mutex_unlock(&con->mutex); |
31b8006e SW |
1409 | pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name)); |
1410 | if (con->ops->peer_reset) | |
1411 | con->ops->peer_reset(con); | |
ec302645 | 1412 | mutex_lock(&con->mutex); |
0da5d703 SW |
1413 | if (test_bit(CLOSED, &con->state) || |
1414 | test_bit(OPENING, &con->state)) | |
1415 | return -EAGAIN; | |
31b8006e SW |
1416 | break; |
1417 | ||
1418 | case CEPH_MSGR_TAG_RETRY_SESSION: | |
1419 | /* | |
1420 | * If we sent a smaller connect_seq than the peer has, try | |
1421 | * again with a larger value. | |
1422 | */ | |
1423 | dout("process_connect got RETRY my seq = %u, peer_seq = %u\n", | |
1424 | le32_to_cpu(con->out_connect.connect_seq), | |
1425 | le32_to_cpu(con->in_connect.connect_seq)); | |
1426 | con->connect_seq = le32_to_cpu(con->in_connect.connect_seq); | |
eed0ef2c | 1427 | prepare_write_connect(con->msgr, con, 0); |
31b8006e SW |
1428 | prepare_read_connect(con); |
1429 | break; | |
1430 | ||
1431 | case CEPH_MSGR_TAG_RETRY_GLOBAL: | |
1432 | /* | |
1433 | * If we sent a smaller global_seq than the peer has, try | |
1434 | * again with a larger value. | |
1435 | */ | |
eed0ef2c | 1436 | dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n", |
31b8006e SW |
1437 | con->peer_global_seq, |
1438 | le32_to_cpu(con->in_connect.global_seq)); | |
1439 | get_global_seq(con->msgr, | |
1440 | le32_to_cpu(con->in_connect.global_seq)); | |
eed0ef2c | 1441 | prepare_write_connect(con->msgr, con, 0); |
31b8006e SW |
1442 | prepare_read_connect(con); |
1443 | break; | |
1444 | ||
1445 | case CEPH_MSGR_TAG_READY: | |
04a419f9 SW |
1446 | if (req_feat & ~server_feat) { |
1447 | pr_err("%s%lld %s protocol feature mismatch," | |
1448 | " my required %llx > server's %llx, need %llx\n", | |
1449 | ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 1450 | ceph_pr_addr(&con->peer_addr.in_addr), |
04a419f9 SW |
1451 | req_feat, server_feat, req_feat & ~server_feat); |
1452 | con->error_msg = "missing required protocol features"; | |
1453 | fail_protocol(con); | |
1454 | return -1; | |
1455 | } | |
31b8006e | 1456 | clear_bit(CONNECTING, &con->state); |
31b8006e SW |
1457 | con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq); |
1458 | con->connect_seq++; | |
aba558e2 | 1459 | con->peer_features = server_feat; |
31b8006e SW |
1460 | dout("process_connect got READY gseq %d cseq %d (%d)\n", |
1461 | con->peer_global_seq, | |
1462 | le32_to_cpu(con->in_reply.connect_seq), | |
1463 | con->connect_seq); | |
1464 | WARN_ON(con->connect_seq != | |
1465 | le32_to_cpu(con->in_reply.connect_seq)); | |
92ac41d0 SW |
1466 | |
1467 | if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY) | |
1468 | set_bit(LOSSYTX, &con->state); | |
1469 | ||
31b8006e SW |
1470 | prepare_read_tag(con); |
1471 | break; | |
1472 | ||
1473 | case CEPH_MSGR_TAG_WAIT: | |
1474 | /* | |
1475 | * If there is a connection race (we are opening | |
1476 | * connections to each other), one of us may just have | |
1477 | * to WAIT. This shouldn't happen if we are the | |
1478 | * client. | |
1479 | */ | |
04177882 SW |
1480 | pr_err("process_connect got WAIT as client\n"); |
1481 | con->error_msg = "protocol error, got WAIT as client"; | |
1482 | return -1; | |
31b8006e SW |
1483 | |
1484 | default: | |
1485 | pr_err("connect protocol error, will retry\n"); | |
1486 | con->error_msg = "protocol error, garbage tag during connect"; | |
1487 | return -1; | |
1488 | } | |
1489 | return 0; | |
1490 | } | |
1491 | ||
1492 | ||
1493 | /* | |
1494 | * read (part of) an ack | |
1495 | */ | |
1496 | static int read_partial_ack(struct ceph_connection *con) | |
1497 | { | |
1498 | int to = 0; | |
1499 | ||
1500 | return read_partial(con, &to, sizeof(con->in_temp_ack), | |
1501 | &con->in_temp_ack); | |
1502 | } | |
1503 | ||
1504 | ||
1505 | /* | |
1506 | * We can finally discard anything that's been acked. | |
1507 | */ | |
1508 | static void process_ack(struct ceph_connection *con) | |
1509 | { | |
1510 | struct ceph_msg *m; | |
1511 | u64 ack = le64_to_cpu(con->in_temp_ack); | |
1512 | u64 seq; | |
1513 | ||
31b8006e SW |
1514 | while (!list_empty(&con->out_sent)) { |
1515 | m = list_first_entry(&con->out_sent, struct ceph_msg, | |
1516 | list_head); | |
1517 | seq = le64_to_cpu(m->hdr.seq); | |
1518 | if (seq > ack) | |
1519 | break; | |
1520 | dout("got ack for seq %llu type %d at %p\n", seq, | |
1521 | le16_to_cpu(m->hdr.type), m); | |
4cf9d544 | 1522 | m->ack_stamp = jiffies; |
31b8006e SW |
1523 | ceph_msg_remove(m); |
1524 | } | |
31b8006e SW |
1525 | prepare_read_tag(con); |
1526 | } | |
1527 | ||
1528 | ||
1529 | ||
1530 | ||
2450418c | 1531 | static int read_partial_message_section(struct ceph_connection *con, |
213c99ee SW |
1532 | struct kvec *section, |
1533 | unsigned int sec_len, u32 *crc) | |
2450418c | 1534 | { |
68b4476b | 1535 | int ret, left; |
2450418c YS |
1536 | |
1537 | BUG_ON(!section); | |
1538 | ||
1539 | while (section->iov_len < sec_len) { | |
1540 | BUG_ON(section->iov_base == NULL); | |
1541 | left = sec_len - section->iov_len; | |
1542 | ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base + | |
1543 | section->iov_len, left); | |
1544 | if (ret <= 0) | |
1545 | return ret; | |
1546 | section->iov_len += ret; | |
2450418c | 1547 | } |
fe3ad593 AE |
1548 | if (section->iov_len == sec_len) |
1549 | *crc = crc32c(0, section->iov_base, section->iov_len); | |
31b8006e | 1550 | |
2450418c YS |
1551 | return 1; |
1552 | } | |
31b8006e | 1553 | |
2450418c YS |
1554 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, |
1555 | struct ceph_msg_header *hdr, | |
1556 | int *skip); | |
68b4476b YS |
1557 | |
1558 | ||
1559 | static int read_partial_message_pages(struct ceph_connection *con, | |
1560 | struct page **pages, | |
bca064d2 | 1561 | unsigned data_len, bool do_datacrc) |
68b4476b YS |
1562 | { |
1563 | void *p; | |
1564 | int ret; | |
1565 | int left; | |
1566 | ||
1567 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1568 | (int)(PAGE_SIZE - con->in_msg_pos.page_pos)); | |
1569 | /* (page) data */ | |
1570 | BUG_ON(pages == NULL); | |
1571 | p = kmap(pages[con->in_msg_pos.page]); | |
1572 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1573 | left); | |
bca064d2 | 1574 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1575 | con->in_data_crc = |
1576 | crc32c(con->in_data_crc, | |
1577 | p + con->in_msg_pos.page_pos, ret); | |
1578 | kunmap(pages[con->in_msg_pos.page]); | |
1579 | if (ret <= 0) | |
1580 | return ret; | |
1581 | con->in_msg_pos.data_pos += ret; | |
1582 | con->in_msg_pos.page_pos += ret; | |
1583 | if (con->in_msg_pos.page_pos == PAGE_SIZE) { | |
1584 | con->in_msg_pos.page_pos = 0; | |
1585 | con->in_msg_pos.page++; | |
1586 | } | |
1587 | ||
1588 | return ret; | |
1589 | } | |
1590 | ||
1591 | #ifdef CONFIG_BLOCK | |
1592 | static int read_partial_message_bio(struct ceph_connection *con, | |
1593 | struct bio **bio_iter, int *bio_seg, | |
bca064d2 | 1594 | unsigned data_len, bool do_datacrc) |
68b4476b YS |
1595 | { |
1596 | struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg); | |
1597 | void *p; | |
1598 | int ret, left; | |
1599 | ||
1600 | if (IS_ERR(bv)) | |
1601 | return PTR_ERR(bv); | |
1602 | ||
1603 | left = min((int)(data_len - con->in_msg_pos.data_pos), | |
1604 | (int)(bv->bv_len - con->in_msg_pos.page_pos)); | |
1605 | ||
1606 | p = kmap(bv->bv_page) + bv->bv_offset; | |
1607 | ||
1608 | ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, | |
1609 | left); | |
bca064d2 | 1610 | if (ret > 0 && do_datacrc) |
68b4476b YS |
1611 | con->in_data_crc = |
1612 | crc32c(con->in_data_crc, | |
1613 | p + con->in_msg_pos.page_pos, ret); | |
1614 | kunmap(bv->bv_page); | |
1615 | if (ret <= 0) | |
1616 | return ret; | |
1617 | con->in_msg_pos.data_pos += ret; | |
1618 | con->in_msg_pos.page_pos += ret; | |
1619 | if (con->in_msg_pos.page_pos == bv->bv_len) { | |
1620 | con->in_msg_pos.page_pos = 0; | |
1621 | iter_bio_next(bio_iter, bio_seg); | |
1622 | } | |
1623 | ||
1624 | return ret; | |
1625 | } | |
1626 | #endif | |
1627 | ||
31b8006e SW |
1628 | /* |
1629 | * read (part of) a message. | |
1630 | */ | |
1631 | static int read_partial_message(struct ceph_connection *con) | |
1632 | { | |
1633 | struct ceph_msg *m = con->in_msg; | |
31b8006e | 1634 | int ret; |
9d7f0f13 | 1635 | int to, left; |
c5c6b19d | 1636 | unsigned front_len, middle_len, data_len; |
bca064d2 | 1637 | bool do_datacrc = con->msgr->nocrc; |
2450418c | 1638 | int skip; |
ae18756b | 1639 | u64 seq; |
fe3ad593 | 1640 | u32 crc; |
31b8006e SW |
1641 | |
1642 | dout("read_partial_message con %p msg %p\n", con, m); | |
1643 | ||
1644 | /* header */ | |
1645 | while (con->in_base_pos < sizeof(con->in_hdr)) { | |
1646 | left = sizeof(con->in_hdr) - con->in_base_pos; | |
1647 | ret = ceph_tcp_recvmsg(con->sock, | |
1648 | (char *)&con->in_hdr + con->in_base_pos, | |
1649 | left); | |
1650 | if (ret <= 0) | |
1651 | return ret; | |
1652 | con->in_base_pos += ret; | |
31b8006e | 1653 | } |
fe3ad593 AE |
1654 | |
1655 | crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc)); | |
1656 | if (cpu_to_le32(crc) != con->in_hdr.crc) { | |
1657 | pr_err("read_partial_message bad hdr " | |
1658 | " crc %u != expected %u\n", | |
1659 | crc, con->in_hdr.crc); | |
1660 | return -EBADMSG; | |
1661 | } | |
1662 | ||
31b8006e SW |
1663 | front_len = le32_to_cpu(con->in_hdr.front_len); |
1664 | if (front_len > CEPH_MSG_MAX_FRONT_LEN) | |
1665 | return -EIO; | |
1666 | middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
1667 | if (middle_len > CEPH_MSG_MAX_DATA_LEN) | |
1668 | return -EIO; | |
1669 | data_len = le32_to_cpu(con->in_hdr.data_len); | |
1670 | if (data_len > CEPH_MSG_MAX_DATA_LEN) | |
1671 | return -EIO; | |
1672 | ||
ae18756b SW |
1673 | /* verify seq# */ |
1674 | seq = le64_to_cpu(con->in_hdr.seq); | |
1675 | if ((s64)seq - (s64)con->in_seq < 1) { | |
df9f86fa | 1676 | pr_info("skipping %s%lld %s seq %lld expected %lld\n", |
ae18756b | 1677 | ENTITY_NAME(con->peer_name), |
3d14c5d2 | 1678 | ceph_pr_addr(&con->peer_addr.in_addr), |
ae18756b SW |
1679 | seq, con->in_seq + 1); |
1680 | con->in_base_pos = -front_len - middle_len - data_len - | |
1681 | sizeof(m->footer); | |
1682 | con->in_tag = CEPH_MSGR_TAG_READY; | |
ae18756b SW |
1683 | return 0; |
1684 | } else if ((s64)seq - (s64)con->in_seq > 1) { | |
1685 | pr_err("read_partial_message bad seq %lld expected %lld\n", | |
1686 | seq, con->in_seq + 1); | |
1687 | con->error_msg = "bad message sequence # for incoming message"; | |
1688 | return -EBADMSG; | |
1689 | } | |
1690 | ||
31b8006e SW |
1691 | /* allocate message? */ |
1692 | if (!con->in_msg) { | |
1693 | dout("got hdr type %d front %d data %d\n", con->in_hdr.type, | |
1694 | con->in_hdr.front_len, con->in_hdr.data_len); | |
ae32be31 | 1695 | skip = 0; |
2450418c YS |
1696 | con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip); |
1697 | if (skip) { | |
31b8006e | 1698 | /* skip this message */ |
a79832f2 | 1699 | dout("alloc_msg said skip message\n"); |
ae32be31 | 1700 | BUG_ON(con->in_msg); |
31b8006e SW |
1701 | con->in_base_pos = -front_len - middle_len - data_len - |
1702 | sizeof(m->footer); | |
1703 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 1704 | con->in_seq++; |
31b8006e SW |
1705 | return 0; |
1706 | } | |
a79832f2 | 1707 | if (!con->in_msg) { |
5b3a4db3 SW |
1708 | con->error_msg = |
1709 | "error allocating memory for incoming message"; | |
a79832f2 | 1710 | return -ENOMEM; |
31b8006e SW |
1711 | } |
1712 | m = con->in_msg; | |
1713 | m->front.iov_len = 0; /* haven't read it yet */ | |
2450418c YS |
1714 | if (m->middle) |
1715 | m->middle->vec.iov_len = 0; | |
9d7f0f13 YS |
1716 | |
1717 | con->in_msg_pos.page = 0; | |
68b4476b | 1718 | if (m->pages) |
c5c6b19d | 1719 | con->in_msg_pos.page_pos = m->page_alignment; |
68b4476b YS |
1720 | else |
1721 | con->in_msg_pos.page_pos = 0; | |
9d7f0f13 | 1722 | con->in_msg_pos.data_pos = 0; |
31b8006e SW |
1723 | } |
1724 | ||
1725 | /* front */ | |
2450418c YS |
1726 | ret = read_partial_message_section(con, &m->front, front_len, |
1727 | &con->in_front_crc); | |
1728 | if (ret <= 0) | |
1729 | return ret; | |
31b8006e SW |
1730 | |
1731 | /* middle */ | |
2450418c | 1732 | if (m->middle) { |
213c99ee SW |
1733 | ret = read_partial_message_section(con, &m->middle->vec, |
1734 | middle_len, | |
2450418c | 1735 | &con->in_middle_crc); |
31b8006e SW |
1736 | if (ret <= 0) |
1737 | return ret; | |
31b8006e | 1738 | } |
68b4476b YS |
1739 | #ifdef CONFIG_BLOCK |
1740 | if (m->bio && !m->bio_iter) | |
1741 | init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg); | |
1742 | #endif | |
31b8006e SW |
1743 | |
1744 | /* (page) data */ | |
31b8006e | 1745 | while (con->in_msg_pos.data_pos < data_len) { |
68b4476b YS |
1746 | if (m->pages) { |
1747 | ret = read_partial_message_pages(con, m->pages, | |
bca064d2 | 1748 | data_len, do_datacrc); |
68b4476b YS |
1749 | if (ret <= 0) |
1750 | return ret; | |
1751 | #ifdef CONFIG_BLOCK | |
1752 | } else if (m->bio) { | |
1753 | ||
1754 | ret = read_partial_message_bio(con, | |
1755 | &m->bio_iter, &m->bio_seg, | |
bca064d2 | 1756 | data_len, do_datacrc); |
68b4476b YS |
1757 | if (ret <= 0) |
1758 | return ret; | |
1759 | #endif | |
1760 | } else { | |
1761 | BUG_ON(1); | |
31b8006e SW |
1762 | } |
1763 | } | |
1764 | ||
31b8006e SW |
1765 | /* footer */ |
1766 | to = sizeof(m->hdr) + sizeof(m->footer); | |
1767 | while (con->in_base_pos < to) { | |
1768 | left = to - con->in_base_pos; | |
1769 | ret = ceph_tcp_recvmsg(con->sock, (char *)&m->footer + | |
1770 | (con->in_base_pos - sizeof(m->hdr)), | |
1771 | left); | |
1772 | if (ret <= 0) | |
1773 | return ret; | |
1774 | con->in_base_pos += ret; | |
1775 | } | |
1776 | dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n", | |
1777 | m, front_len, m->footer.front_crc, middle_len, | |
1778 | m->footer.middle_crc, data_len, m->footer.data_crc); | |
1779 | ||
1780 | /* crc ok? */ | |
1781 | if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) { | |
1782 | pr_err("read_partial_message %p front crc %u != exp. %u\n", | |
1783 | m, con->in_front_crc, m->footer.front_crc); | |
1784 | return -EBADMSG; | |
1785 | } | |
1786 | if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) { | |
1787 | pr_err("read_partial_message %p middle crc %u != exp %u\n", | |
1788 | m, con->in_middle_crc, m->footer.middle_crc); | |
1789 | return -EBADMSG; | |
1790 | } | |
bca064d2 | 1791 | if (do_datacrc && |
31b8006e SW |
1792 | (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 && |
1793 | con->in_data_crc != le32_to_cpu(m->footer.data_crc)) { | |
1794 | pr_err("read_partial_message %p data crc %u != exp. %u\n", m, | |
1795 | con->in_data_crc, le32_to_cpu(m->footer.data_crc)); | |
1796 | return -EBADMSG; | |
1797 | } | |
1798 | ||
1799 | return 1; /* done! */ | |
1800 | } | |
1801 | ||
1802 | /* | |
1803 | * Process message. This happens in the worker thread. The callback should | |
1804 | * be careful not to do anything that waits on other incoming messages or it | |
1805 | * may deadlock. | |
1806 | */ | |
1807 | static void process_message(struct ceph_connection *con) | |
1808 | { | |
5e095e8b | 1809 | struct ceph_msg *msg; |
31b8006e | 1810 | |
5e095e8b | 1811 | msg = con->in_msg; |
31b8006e SW |
1812 | con->in_msg = NULL; |
1813 | ||
1814 | /* if first message, set peer_name */ | |
1815 | if (con->peer_name.type == 0) | |
dbad185d | 1816 | con->peer_name = msg->hdr.src; |
31b8006e | 1817 | |
31b8006e | 1818 | con->in_seq++; |
ec302645 | 1819 | mutex_unlock(&con->mutex); |
31b8006e SW |
1820 | |
1821 | dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n", | |
1822 | msg, le64_to_cpu(msg->hdr.seq), | |
dbad185d | 1823 | ENTITY_NAME(msg->hdr.src), |
31b8006e SW |
1824 | le16_to_cpu(msg->hdr.type), |
1825 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
1826 | le32_to_cpu(msg->hdr.front_len), | |
1827 | le32_to_cpu(msg->hdr.data_len), | |
1828 | con->in_front_crc, con->in_middle_crc, con->in_data_crc); | |
1829 | con->ops->dispatch(con, msg); | |
ec302645 SW |
1830 | |
1831 | mutex_lock(&con->mutex); | |
31b8006e SW |
1832 | prepare_read_tag(con); |
1833 | } | |
1834 | ||
1835 | ||
1836 | /* | |
1837 | * Write something to the socket. Called in a worker thread when the | |
1838 | * socket appears to be writeable and we have something ready to send. | |
1839 | */ | |
1840 | static int try_write(struct ceph_connection *con) | |
1841 | { | |
1842 | struct ceph_messenger *msgr = con->msgr; | |
1843 | int ret = 1; | |
1844 | ||
1845 | dout("try_write start %p state %lu nref %d\n", con, con->state, | |
1846 | atomic_read(&con->nref)); | |
1847 | ||
31b8006e SW |
1848 | more: |
1849 | dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes); | |
1850 | ||
1851 | /* open the socket first? */ | |
1852 | if (con->sock == NULL) { | |
eed0ef2c SW |
1853 | prepare_write_connect(msgr, con, 1); |
1854 | prepare_read_banner(con); | |
31b8006e | 1855 | set_bit(CONNECTING, &con->state); |
eed0ef2c | 1856 | clear_bit(NEGOTIATING, &con->state); |
31b8006e | 1857 | |
cf3e5c40 | 1858 | BUG_ON(con->in_msg); |
31b8006e SW |
1859 | con->in_tag = CEPH_MSGR_TAG_READY; |
1860 | dout("try_write initiating connect on %p new state %lu\n", | |
1861 | con, con->state); | |
41617d0c AE |
1862 | ret = ceph_tcp_connect(con); |
1863 | if (ret < 0) { | |
31b8006e | 1864 | con->error_msg = "connect error"; |
31b8006e SW |
1865 | goto out; |
1866 | } | |
1867 | } | |
1868 | ||
1869 | more_kvec: | |
1870 | /* kvec data queued? */ | |
1871 | if (con->out_skip) { | |
1872 | ret = write_partial_skip(con); | |
1873 | if (ret <= 0) | |
42961d23 | 1874 | goto out; |
31b8006e SW |
1875 | } |
1876 | if (con->out_kvec_left) { | |
1877 | ret = write_partial_kvec(con); | |
1878 | if (ret <= 0) | |
42961d23 | 1879 | goto out; |
31b8006e SW |
1880 | } |
1881 | ||
1882 | /* msg pages? */ | |
1883 | if (con->out_msg) { | |
c86a2930 SW |
1884 | if (con->out_msg_done) { |
1885 | ceph_msg_put(con->out_msg); | |
1886 | con->out_msg = NULL; /* we're done with this one */ | |
1887 | goto do_next; | |
1888 | } | |
1889 | ||
31b8006e SW |
1890 | ret = write_partial_msg_pages(con); |
1891 | if (ret == 1) | |
1892 | goto more_kvec; /* we need to send the footer, too! */ | |
1893 | if (ret == 0) | |
42961d23 | 1894 | goto out; |
31b8006e SW |
1895 | if (ret < 0) { |
1896 | dout("try_write write_partial_msg_pages err %d\n", | |
1897 | ret); | |
42961d23 | 1898 | goto out; |
31b8006e SW |
1899 | } |
1900 | } | |
1901 | ||
c86a2930 | 1902 | do_next: |
31b8006e SW |
1903 | if (!test_bit(CONNECTING, &con->state)) { |
1904 | /* is anything else pending? */ | |
1905 | if (!list_empty(&con->out_queue)) { | |
1906 | prepare_write_message(con); | |
1907 | goto more; | |
1908 | } | |
1909 | if (con->in_seq > con->in_seq_acked) { | |
1910 | prepare_write_ack(con); | |
1911 | goto more; | |
1912 | } | |
1913 | if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) { | |
1914 | prepare_write_keepalive(con); | |
1915 | goto more; | |
1916 | } | |
1917 | } | |
1918 | ||
1919 | /* Nothing to do! */ | |
1920 | clear_bit(WRITE_PENDING, &con->state); | |
1921 | dout("try_write nothing else to write.\n"); | |
31b8006e SW |
1922 | ret = 0; |
1923 | out: | |
42961d23 | 1924 | dout("try_write done on %p ret %d\n", con, ret); |
31b8006e SW |
1925 | return ret; |
1926 | } | |
1927 | ||
1928 | ||
1929 | ||
1930 | /* | |
1931 | * Read what we can from the socket. | |
1932 | */ | |
1933 | static int try_read(struct ceph_connection *con) | |
1934 | { | |
31b8006e SW |
1935 | int ret = -1; |
1936 | ||
1937 | if (!con->sock) | |
1938 | return 0; | |
1939 | ||
1940 | if (test_bit(STANDBY, &con->state)) | |
1941 | return 0; | |
1942 | ||
1943 | dout("try_read start on %p\n", con); | |
ec302645 | 1944 | |
31b8006e SW |
1945 | more: |
1946 | dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag, | |
1947 | con->in_base_pos); | |
0da5d703 SW |
1948 | |
1949 | /* | |
1950 | * process_connect and process_message drop and re-take | |
1951 | * con->mutex. make sure we handle a racing close or reopen. | |
1952 | */ | |
1953 | if (test_bit(CLOSED, &con->state) || | |
1954 | test_bit(OPENING, &con->state)) { | |
1955 | ret = -EAGAIN; | |
1956 | goto out; | |
1957 | } | |
1958 | ||
31b8006e | 1959 | if (test_bit(CONNECTING, &con->state)) { |
eed0ef2c SW |
1960 | if (!test_bit(NEGOTIATING, &con->state)) { |
1961 | dout("try_read connecting\n"); | |
1962 | ret = read_partial_banner(con); | |
1963 | if (ret <= 0) | |
eed0ef2c | 1964 | goto out; |
98bdb0aa SW |
1965 | ret = process_banner(con); |
1966 | if (ret < 0) | |
1967 | goto out; | |
eed0ef2c | 1968 | } |
31b8006e SW |
1969 | ret = read_partial_connect(con); |
1970 | if (ret <= 0) | |
31b8006e | 1971 | goto out; |
98bdb0aa SW |
1972 | ret = process_connect(con); |
1973 | if (ret < 0) | |
1974 | goto out; | |
31b8006e SW |
1975 | goto more; |
1976 | } | |
1977 | ||
1978 | if (con->in_base_pos < 0) { | |
1979 | /* | |
1980 | * skipping + discarding content. | |
1981 | * | |
1982 | * FIXME: there must be a better way to do this! | |
1983 | */ | |
1984 | static char buf[1024]; | |
1985 | int skip = min(1024, -con->in_base_pos); | |
1986 | dout("skipping %d / %d bytes\n", skip, -con->in_base_pos); | |
1987 | ret = ceph_tcp_recvmsg(con->sock, buf, skip); | |
1988 | if (ret <= 0) | |
98bdb0aa | 1989 | goto out; |
31b8006e SW |
1990 | con->in_base_pos += ret; |
1991 | if (con->in_base_pos) | |
1992 | goto more; | |
1993 | } | |
1994 | if (con->in_tag == CEPH_MSGR_TAG_READY) { | |
1995 | /* | |
1996 | * what's next? | |
1997 | */ | |
1998 | ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1); | |
1999 | if (ret <= 0) | |
98bdb0aa | 2000 | goto out; |
31b8006e SW |
2001 | dout("try_read got tag %d\n", (int)con->in_tag); |
2002 | switch (con->in_tag) { | |
2003 | case CEPH_MSGR_TAG_MSG: | |
2004 | prepare_read_message(con); | |
2005 | break; | |
2006 | case CEPH_MSGR_TAG_ACK: | |
2007 | prepare_read_ack(con); | |
2008 | break; | |
2009 | case CEPH_MSGR_TAG_CLOSE: | |
2010 | set_bit(CLOSED, &con->state); /* fixme */ | |
98bdb0aa | 2011 | goto out; |
31b8006e SW |
2012 | default: |
2013 | goto bad_tag; | |
2014 | } | |
2015 | } | |
2016 | if (con->in_tag == CEPH_MSGR_TAG_MSG) { | |
2017 | ret = read_partial_message(con); | |
2018 | if (ret <= 0) { | |
2019 | switch (ret) { | |
2020 | case -EBADMSG: | |
2021 | con->error_msg = "bad crc"; | |
2022 | ret = -EIO; | |
98bdb0aa | 2023 | break; |
31b8006e SW |
2024 | case -EIO: |
2025 | con->error_msg = "io error"; | |
98bdb0aa | 2026 | break; |
31b8006e | 2027 | } |
98bdb0aa | 2028 | goto out; |
31b8006e SW |
2029 | } |
2030 | if (con->in_tag == CEPH_MSGR_TAG_READY) | |
2031 | goto more; | |
2032 | process_message(con); | |
2033 | goto more; | |
2034 | } | |
2035 | if (con->in_tag == CEPH_MSGR_TAG_ACK) { | |
2036 | ret = read_partial_ack(con); | |
2037 | if (ret <= 0) | |
98bdb0aa | 2038 | goto out; |
31b8006e SW |
2039 | process_ack(con); |
2040 | goto more; | |
2041 | } | |
2042 | ||
31b8006e | 2043 | out: |
98bdb0aa | 2044 | dout("try_read done on %p ret %d\n", con, ret); |
31b8006e SW |
2045 | return ret; |
2046 | ||
2047 | bad_tag: | |
2048 | pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag); | |
2049 | con->error_msg = "protocol error, garbage tag"; | |
2050 | ret = -1; | |
2051 | goto out; | |
2052 | } | |
2053 | ||
2054 | ||
2055 | /* | |
2056 | * Atomically queue work on a connection. Bump @con reference to | |
2057 | * avoid races with connection teardown. | |
31b8006e SW |
2058 | */ |
2059 | static void queue_con(struct ceph_connection *con) | |
2060 | { | |
2061 | if (test_bit(DEAD, &con->state)) { | |
2062 | dout("queue_con %p ignoring: DEAD\n", | |
2063 | con); | |
2064 | return; | |
2065 | } | |
2066 | ||
2067 | if (!con->ops->get(con)) { | |
2068 | dout("queue_con %p ref count 0\n", con); | |
2069 | return; | |
2070 | } | |
2071 | ||
f363e45f | 2072 | if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) { |
31b8006e SW |
2073 | dout("queue_con %p - already queued\n", con); |
2074 | con->ops->put(con); | |
2075 | } else { | |
2076 | dout("queue_con %p\n", con); | |
2077 | } | |
2078 | } | |
2079 | ||
2080 | /* | |
2081 | * Do some work on a connection. Drop a connection ref when we're done. | |
2082 | */ | |
2083 | static void con_work(struct work_struct *work) | |
2084 | { | |
2085 | struct ceph_connection *con = container_of(work, struct ceph_connection, | |
2086 | work.work); | |
0da5d703 | 2087 | int ret; |
31b8006e | 2088 | |
9dd4658d | 2089 | mutex_lock(&con->mutex); |
0da5d703 | 2090 | restart: |
60bf8bf8 SW |
2091 | if (test_and_clear_bit(BACKOFF, &con->state)) { |
2092 | dout("con_work %p backing off\n", con); | |
2093 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
2094 | round_jiffies_relative(con->delay))) { | |
2095 | dout("con_work %p backoff %lu\n", con, con->delay); | |
2096 | mutex_unlock(&con->mutex); | |
2097 | return; | |
2098 | } else { | |
2099 | con->ops->put(con); | |
2100 | dout("con_work %p FAILED to back off %lu\n", con, | |
2101 | con->delay); | |
2102 | } | |
2103 | } | |
9dd4658d | 2104 | |
e00de341 SW |
2105 | if (test_bit(STANDBY, &con->state)) { |
2106 | dout("con_work %p STANDBY\n", con); | |
2107 | goto done; | |
2108 | } | |
31b8006e SW |
2109 | if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */ |
2110 | dout("con_work CLOSED\n"); | |
2111 | con_close_socket(con); | |
2112 | goto done; | |
2113 | } | |
2114 | if (test_and_clear_bit(OPENING, &con->state)) { | |
2115 | /* reopen w/ new peer */ | |
2116 | dout("con_work OPENING\n"); | |
2117 | con_close_socket(con); | |
2118 | } | |
2119 | ||
0da5d703 SW |
2120 | if (test_and_clear_bit(SOCK_CLOSED, &con->state)) |
2121 | goto fault; | |
2122 | ||
2123 | ret = try_read(con); | |
2124 | if (ret == -EAGAIN) | |
2125 | goto restart; | |
2126 | if (ret < 0) | |
2127 | goto fault; | |
2128 | ||
2129 | ret = try_write(con); | |
2130 | if (ret == -EAGAIN) | |
2131 | goto restart; | |
2132 | if (ret < 0) | |
2133 | goto fault; | |
31b8006e SW |
2134 | |
2135 | done: | |
9dd4658d | 2136 | mutex_unlock(&con->mutex); |
9dd4658d | 2137 | done_unlocked: |
31b8006e | 2138 | con->ops->put(con); |
0da5d703 SW |
2139 | return; |
2140 | ||
2141 | fault: | |
2142 | mutex_unlock(&con->mutex); | |
2143 | ceph_fault(con); /* error/fault path */ | |
2144 | goto done_unlocked; | |
31b8006e SW |
2145 | } |
2146 | ||
2147 | ||
2148 | /* | |
2149 | * Generic error/fault handler. A retry mechanism is used with | |
2150 | * exponential backoff | |
2151 | */ | |
2152 | static void ceph_fault(struct ceph_connection *con) | |
2153 | { | |
2154 | pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name), | |
3d14c5d2 | 2155 | ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg); |
31b8006e | 2156 | dout("fault %p state %lu to peer %s\n", |
3d14c5d2 | 2157 | con, con->state, ceph_pr_addr(&con->peer_addr.in_addr)); |
31b8006e SW |
2158 | |
2159 | if (test_bit(LOSSYTX, &con->state)) { | |
2160 | dout("fault on LOSSYTX channel\n"); | |
2161 | goto out; | |
2162 | } | |
2163 | ||
ec302645 | 2164 | mutex_lock(&con->mutex); |
91e45ce3 SW |
2165 | if (test_bit(CLOSED, &con->state)) |
2166 | goto out_unlock; | |
ec302645 | 2167 | |
31b8006e | 2168 | con_close_socket(con); |
5e095e8b SW |
2169 | |
2170 | if (con->in_msg) { | |
2171 | ceph_msg_put(con->in_msg); | |
2172 | con->in_msg = NULL; | |
2173 | } | |
31b8006e | 2174 | |
e80a52d1 SW |
2175 | /* Requeue anything that hasn't been acked */ |
2176 | list_splice_init(&con->out_sent, &con->out_queue); | |
9bd2e6f8 | 2177 | |
e76661d0 SW |
2178 | /* If there are no messages queued or keepalive pending, place |
2179 | * the connection in a STANDBY state */ | |
2180 | if (list_empty(&con->out_queue) && | |
2181 | !test_bit(KEEPALIVE_PENDING, &con->state)) { | |
e00de341 SW |
2182 | dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con); |
2183 | clear_bit(WRITE_PENDING, &con->state); | |
31b8006e | 2184 | set_bit(STANDBY, &con->state); |
e80a52d1 SW |
2185 | } else { |
2186 | /* retry after a delay. */ | |
2187 | if (con->delay == 0) | |
2188 | con->delay = BASE_DELAY_INTERVAL; | |
2189 | else if (con->delay < MAX_DELAY_INTERVAL) | |
2190 | con->delay *= 2; | |
e80a52d1 SW |
2191 | con->ops->get(con); |
2192 | if (queue_delayed_work(ceph_msgr_wq, &con->work, | |
60bf8bf8 SW |
2193 | round_jiffies_relative(con->delay))) { |
2194 | dout("fault queued %p delay %lu\n", con, con->delay); | |
2195 | } else { | |
e80a52d1 | 2196 | con->ops->put(con); |
60bf8bf8 SW |
2197 | dout("fault failed to queue %p delay %lu, backoff\n", |
2198 | con, con->delay); | |
2199 | /* | |
2200 | * In many cases we see a socket state change | |
2201 | * while con_work is running and end up | |
2202 | * queuing (non-delayed) work, such that we | |
2203 | * can't backoff with a delay. Set a flag so | |
2204 | * that when con_work restarts we schedule the | |
2205 | * delay then. | |
2206 | */ | |
2207 | set_bit(BACKOFF, &con->state); | |
2208 | } | |
31b8006e SW |
2209 | } |
2210 | ||
91e45ce3 SW |
2211 | out_unlock: |
2212 | mutex_unlock(&con->mutex); | |
31b8006e | 2213 | out: |
161fd65a SW |
2214 | /* |
2215 | * in case we faulted due to authentication, invalidate our | |
2216 | * current tickets so that we can get new ones. | |
213c99ee | 2217 | */ |
161fd65a SW |
2218 | if (con->auth_retry && con->ops->invalidate_authorizer) { |
2219 | dout("calling invalidate_authorizer()\n"); | |
2220 | con->ops->invalidate_authorizer(con); | |
2221 | } | |
2222 | ||
31b8006e SW |
2223 | if (con->ops->fault) |
2224 | con->ops->fault(con); | |
2225 | } | |
2226 | ||
2227 | ||
2228 | ||
2229 | /* | |
2230 | * create a new messenger instance | |
2231 | */ | |
3d14c5d2 YS |
2232 | struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr, |
2233 | u32 supported_features, | |
2234 | u32 required_features) | |
31b8006e SW |
2235 | { |
2236 | struct ceph_messenger *msgr; | |
2237 | ||
2238 | msgr = kzalloc(sizeof(*msgr), GFP_KERNEL); | |
2239 | if (msgr == NULL) | |
2240 | return ERR_PTR(-ENOMEM); | |
2241 | ||
3d14c5d2 YS |
2242 | msgr->supported_features = supported_features; |
2243 | msgr->required_features = required_features; | |
2244 | ||
31b8006e SW |
2245 | spin_lock_init(&msgr->global_seq_lock); |
2246 | ||
31b8006e SW |
2247 | if (myaddr) |
2248 | msgr->inst.addr = *myaddr; | |
2249 | ||
2250 | /* select a random nonce */ | |
ac8839d7 | 2251 | msgr->inst.addr.type = 0; |
103e2d3a | 2252 | get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce)); |
63f2d211 | 2253 | encode_my_addr(msgr); |
31b8006e SW |
2254 | |
2255 | dout("messenger_create %p\n", msgr); | |
2256 | return msgr; | |
2257 | } | |
3d14c5d2 | 2258 | EXPORT_SYMBOL(ceph_messenger_create); |
31b8006e SW |
2259 | |
2260 | void ceph_messenger_destroy(struct ceph_messenger *msgr) | |
2261 | { | |
2262 | dout("destroy %p\n", msgr); | |
31b8006e SW |
2263 | kfree(msgr); |
2264 | dout("destroyed messenger %p\n", msgr); | |
2265 | } | |
3d14c5d2 | 2266 | EXPORT_SYMBOL(ceph_messenger_destroy); |
31b8006e | 2267 | |
e00de341 SW |
2268 | static void clear_standby(struct ceph_connection *con) |
2269 | { | |
2270 | /* come back from STANDBY? */ | |
2271 | if (test_and_clear_bit(STANDBY, &con->state)) { | |
2272 | mutex_lock(&con->mutex); | |
2273 | dout("clear_standby %p and ++connect_seq\n", con); | |
2274 | con->connect_seq++; | |
2275 | WARN_ON(test_bit(WRITE_PENDING, &con->state)); | |
2276 | WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state)); | |
2277 | mutex_unlock(&con->mutex); | |
2278 | } | |
2279 | } | |
2280 | ||
31b8006e SW |
2281 | /* |
2282 | * Queue up an outgoing message on the given connection. | |
2283 | */ | |
2284 | void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg) | |
2285 | { | |
2286 | if (test_bit(CLOSED, &con->state)) { | |
2287 | dout("con_send %p closed, dropping %p\n", con, msg); | |
2288 | ceph_msg_put(msg); | |
2289 | return; | |
2290 | } | |
2291 | ||
2292 | /* set src+dst */ | |
dbad185d | 2293 | msg->hdr.src = con->msgr->inst.name; |
31b8006e | 2294 | |
3ca02ef9 SW |
2295 | BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len)); |
2296 | ||
e84346b7 SW |
2297 | msg->needs_out_seq = true; |
2298 | ||
31b8006e | 2299 | /* queue */ |
ec302645 | 2300 | mutex_lock(&con->mutex); |
31b8006e SW |
2301 | BUG_ON(!list_empty(&msg->list_head)); |
2302 | list_add_tail(&msg->list_head, &con->out_queue); | |
2303 | dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg, | |
2304 | ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type), | |
2305 | ceph_msg_type_name(le16_to_cpu(msg->hdr.type)), | |
2306 | le32_to_cpu(msg->hdr.front_len), | |
2307 | le32_to_cpu(msg->hdr.middle_len), | |
2308 | le32_to_cpu(msg->hdr.data_len)); | |
ec302645 | 2309 | mutex_unlock(&con->mutex); |
31b8006e SW |
2310 | |
2311 | /* if there wasn't anything waiting to send before, queue | |
2312 | * new work */ | |
e00de341 | 2313 | clear_standby(con); |
31b8006e SW |
2314 | if (test_and_set_bit(WRITE_PENDING, &con->state) == 0) |
2315 | queue_con(con); | |
2316 | } | |
3d14c5d2 | 2317 | EXPORT_SYMBOL(ceph_con_send); |
31b8006e SW |
2318 | |
2319 | /* | |
2320 | * Revoke a message that was previously queued for send | |
2321 | */ | |
2322 | void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg) | |
2323 | { | |
ec302645 | 2324 | mutex_lock(&con->mutex); |
31b8006e | 2325 | if (!list_empty(&msg->list_head)) { |
ed98adad | 2326 | dout("con_revoke %p msg %p - was on queue\n", con, msg); |
31b8006e SW |
2327 | list_del_init(&msg->list_head); |
2328 | ceph_msg_put(msg); | |
2329 | msg->hdr.seq = 0; | |
ed98adad SW |
2330 | } |
2331 | if (con->out_msg == msg) { | |
2332 | dout("con_revoke %p msg %p - was sending\n", con, msg); | |
2333 | con->out_msg = NULL; | |
31b8006e SW |
2334 | if (con->out_kvec_is_msg) { |
2335 | con->out_skip = con->out_kvec_bytes; | |
2336 | con->out_kvec_is_msg = false; | |
2337 | } | |
ed98adad SW |
2338 | ceph_msg_put(msg); |
2339 | msg->hdr.seq = 0; | |
31b8006e | 2340 | } |
ec302645 | 2341 | mutex_unlock(&con->mutex); |
31b8006e SW |
2342 | } |
2343 | ||
350b1c32 | 2344 | /* |
0d59ab81 | 2345 | * Revoke a message that we may be reading data into |
350b1c32 | 2346 | */ |
0d59ab81 | 2347 | void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg) |
350b1c32 SW |
2348 | { |
2349 | mutex_lock(&con->mutex); | |
0d59ab81 YS |
2350 | if (con->in_msg && con->in_msg == msg) { |
2351 | unsigned front_len = le32_to_cpu(con->in_hdr.front_len); | |
2352 | unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len); | |
350b1c32 SW |
2353 | unsigned data_len = le32_to_cpu(con->in_hdr.data_len); |
2354 | ||
2355 | /* skip rest of message */ | |
0d59ab81 | 2356 | dout("con_revoke_pages %p msg %p revoked\n", con, msg); |
350b1c32 SW |
2357 | con->in_base_pos = con->in_base_pos - |
2358 | sizeof(struct ceph_msg_header) - | |
0d59ab81 YS |
2359 | front_len - |
2360 | middle_len - | |
2361 | data_len - | |
350b1c32 | 2362 | sizeof(struct ceph_msg_footer); |
350b1c32 SW |
2363 | ceph_msg_put(con->in_msg); |
2364 | con->in_msg = NULL; | |
2365 | con->in_tag = CEPH_MSGR_TAG_READY; | |
684be25c | 2366 | con->in_seq++; |
350b1c32 SW |
2367 | } else { |
2368 | dout("con_revoke_pages %p msg %p pages %p no-op\n", | |
0d59ab81 | 2369 | con, con->in_msg, msg); |
350b1c32 SW |
2370 | } |
2371 | mutex_unlock(&con->mutex); | |
2372 | } | |
2373 | ||
31b8006e SW |
2374 | /* |
2375 | * Queue a keepalive byte to ensure the tcp connection is alive. | |
2376 | */ | |
2377 | void ceph_con_keepalive(struct ceph_connection *con) | |
2378 | { | |
e00de341 SW |
2379 | dout("con_keepalive %p\n", con); |
2380 | clear_standby(con); | |
31b8006e SW |
2381 | if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 && |
2382 | test_and_set_bit(WRITE_PENDING, &con->state) == 0) | |
2383 | queue_con(con); | |
2384 | } | |
3d14c5d2 | 2385 | EXPORT_SYMBOL(ceph_con_keepalive); |
31b8006e SW |
2386 | |
2387 | ||
2388 | /* | |
2389 | * construct a new message with given type, size | |
2390 | * the new msg has a ref count of 1. | |
2391 | */ | |
b61c2763 SW |
2392 | struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
2393 | bool can_fail) | |
31b8006e SW |
2394 | { |
2395 | struct ceph_msg *m; | |
2396 | ||
34d23762 | 2397 | m = kmalloc(sizeof(*m), flags); |
31b8006e SW |
2398 | if (m == NULL) |
2399 | goto out; | |
c2e552e7 | 2400 | kref_init(&m->kref); |
31b8006e SW |
2401 | INIT_LIST_HEAD(&m->list_head); |
2402 | ||
45c6ceb5 | 2403 | m->hdr.tid = 0; |
31b8006e | 2404 | m->hdr.type = cpu_to_le16(type); |
45c6ceb5 SW |
2405 | m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT); |
2406 | m->hdr.version = 0; | |
31b8006e SW |
2407 | m->hdr.front_len = cpu_to_le32(front_len); |
2408 | m->hdr.middle_len = 0; | |
bb257664 SW |
2409 | m->hdr.data_len = 0; |
2410 | m->hdr.data_off = 0; | |
45c6ceb5 | 2411 | m->hdr.reserved = 0; |
31b8006e SW |
2412 | m->footer.front_crc = 0; |
2413 | m->footer.middle_crc = 0; | |
2414 | m->footer.data_crc = 0; | |
45c6ceb5 | 2415 | m->footer.flags = 0; |
31b8006e SW |
2416 | m->front_max = front_len; |
2417 | m->front_is_vmalloc = false; | |
2418 | m->more_to_follow = false; | |
c0d5f9db | 2419 | m->ack_stamp = 0; |
31b8006e SW |
2420 | m->pool = NULL; |
2421 | ||
ca20892d HC |
2422 | /* middle */ |
2423 | m->middle = NULL; | |
2424 | ||
2425 | /* data */ | |
2426 | m->nr_pages = 0; | |
2427 | m->page_alignment = 0; | |
2428 | m->pages = NULL; | |
2429 | m->pagelist = NULL; | |
2430 | m->bio = NULL; | |
2431 | m->bio_iter = NULL; | |
2432 | m->bio_seg = 0; | |
2433 | m->trail = NULL; | |
2434 | ||
31b8006e SW |
2435 | /* front */ |
2436 | if (front_len) { | |
2437 | if (front_len > PAGE_CACHE_SIZE) { | |
34d23762 | 2438 | m->front.iov_base = __vmalloc(front_len, flags, |
31b8006e SW |
2439 | PAGE_KERNEL); |
2440 | m->front_is_vmalloc = true; | |
2441 | } else { | |
34d23762 | 2442 | m->front.iov_base = kmalloc(front_len, flags); |
31b8006e SW |
2443 | } |
2444 | if (m->front.iov_base == NULL) { | |
b61c2763 | 2445 | dout("ceph_msg_new can't allocate %d bytes\n", |
31b8006e SW |
2446 | front_len); |
2447 | goto out2; | |
2448 | } | |
2449 | } else { | |
2450 | m->front.iov_base = NULL; | |
2451 | } | |
2452 | m->front.iov_len = front_len; | |
2453 | ||
bb257664 | 2454 | dout("ceph_msg_new %p front %d\n", m, front_len); |
31b8006e SW |
2455 | return m; |
2456 | ||
2457 | out2: | |
2458 | ceph_msg_put(m); | |
2459 | out: | |
b61c2763 SW |
2460 | if (!can_fail) { |
2461 | pr_err("msg_new can't create type %d front %d\n", type, | |
2462 | front_len); | |
f0ed1b7c | 2463 | WARN_ON(1); |
b61c2763 SW |
2464 | } else { |
2465 | dout("msg_new can't create type %d front %d\n", type, | |
2466 | front_len); | |
2467 | } | |
a79832f2 | 2468 | return NULL; |
31b8006e | 2469 | } |
3d14c5d2 | 2470 | EXPORT_SYMBOL(ceph_msg_new); |
31b8006e | 2471 | |
31b8006e SW |
2472 | /* |
2473 | * Allocate "middle" portion of a message, if it is needed and wasn't | |
2474 | * allocated by alloc_msg. This allows us to read a small fixed-size | |
2475 | * per-type header in the front and then gracefully fail (i.e., | |
2476 | * propagate the error to the caller based on info in the front) when | |
2477 | * the middle is too large. | |
2478 | */ | |
2450418c | 2479 | static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg) |
31b8006e SW |
2480 | { |
2481 | int type = le16_to_cpu(msg->hdr.type); | |
2482 | int middle_len = le32_to_cpu(msg->hdr.middle_len); | |
2483 | ||
2484 | dout("alloc_middle %p type %d %s middle_len %d\n", msg, type, | |
2485 | ceph_msg_type_name(type), middle_len); | |
2486 | BUG_ON(!middle_len); | |
2487 | BUG_ON(msg->middle); | |
2488 | ||
b6c1d5b8 | 2489 | msg->middle = ceph_buffer_new(middle_len, GFP_NOFS); |
31b8006e SW |
2490 | if (!msg->middle) |
2491 | return -ENOMEM; | |
2492 | return 0; | |
2493 | } | |
2494 | ||
2450418c YS |
2495 | /* |
2496 | * Generic message allocator, for incoming messages. | |
2497 | */ | |
2498 | static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, | |
2499 | struct ceph_msg_header *hdr, | |
2500 | int *skip) | |
2501 | { | |
2502 | int type = le16_to_cpu(hdr->type); | |
2503 | int front_len = le32_to_cpu(hdr->front_len); | |
2504 | int middle_len = le32_to_cpu(hdr->middle_len); | |
2505 | struct ceph_msg *msg = NULL; | |
2506 | int ret; | |
2507 | ||
2508 | if (con->ops->alloc_msg) { | |
0547a9b3 | 2509 | mutex_unlock(&con->mutex); |
2450418c | 2510 | msg = con->ops->alloc_msg(con, hdr, skip); |
0547a9b3 | 2511 | mutex_lock(&con->mutex); |
a79832f2 | 2512 | if (!msg || *skip) |
2450418c YS |
2513 | return NULL; |
2514 | } | |
2515 | if (!msg) { | |
2516 | *skip = 0; | |
b61c2763 | 2517 | msg = ceph_msg_new(type, front_len, GFP_NOFS, false); |
2450418c YS |
2518 | if (!msg) { |
2519 | pr_err("unable to allocate msg type %d len %d\n", | |
2520 | type, front_len); | |
a79832f2 | 2521 | return NULL; |
2450418c | 2522 | } |
c5c6b19d | 2523 | msg->page_alignment = le16_to_cpu(hdr->data_off); |
2450418c | 2524 | } |
9d7f0f13 | 2525 | memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); |
2450418c | 2526 | |
bb257664 | 2527 | if (middle_len && !msg->middle) { |
2450418c | 2528 | ret = ceph_alloc_middle(con, msg); |
2450418c YS |
2529 | if (ret < 0) { |
2530 | ceph_msg_put(msg); | |
a79832f2 | 2531 | return NULL; |
2450418c YS |
2532 | } |
2533 | } | |
9d7f0f13 | 2534 | |
2450418c YS |
2535 | return msg; |
2536 | } | |
2537 | ||
31b8006e SW |
2538 | |
2539 | /* | |
2540 | * Free a generically kmalloc'd message. | |
2541 | */ | |
2542 | void ceph_msg_kfree(struct ceph_msg *m) | |
2543 | { | |
2544 | dout("msg_kfree %p\n", m); | |
2545 | if (m->front_is_vmalloc) | |
2546 | vfree(m->front.iov_base); | |
2547 | else | |
2548 | kfree(m->front.iov_base); | |
2549 | kfree(m); | |
2550 | } | |
2551 | ||
2552 | /* | |
2553 | * Drop a msg ref. Destroy as needed. | |
2554 | */ | |
c2e552e7 SW |
2555 | void ceph_msg_last_put(struct kref *kref) |
2556 | { | |
2557 | struct ceph_msg *m = container_of(kref, struct ceph_msg, kref); | |
31b8006e | 2558 | |
c2e552e7 SW |
2559 | dout("ceph_msg_put last one on %p\n", m); |
2560 | WARN_ON(!list_empty(&m->list_head)); | |
2561 | ||
2562 | /* drop middle, data, if any */ | |
2563 | if (m->middle) { | |
2564 | ceph_buffer_put(m->middle); | |
2565 | m->middle = NULL; | |
31b8006e | 2566 | } |
c2e552e7 SW |
2567 | m->nr_pages = 0; |
2568 | m->pages = NULL; | |
2569 | ||
58bb3b37 SW |
2570 | if (m->pagelist) { |
2571 | ceph_pagelist_release(m->pagelist); | |
2572 | kfree(m->pagelist); | |
2573 | m->pagelist = NULL; | |
2574 | } | |
2575 | ||
68b4476b YS |
2576 | m->trail = NULL; |
2577 | ||
c2e552e7 SW |
2578 | if (m->pool) |
2579 | ceph_msgpool_put(m->pool, m); | |
2580 | else | |
2581 | ceph_msg_kfree(m); | |
31b8006e | 2582 | } |
3d14c5d2 | 2583 | EXPORT_SYMBOL(ceph_msg_last_put); |
9ec7cab1 SW |
2584 | |
2585 | void ceph_msg_dump(struct ceph_msg *msg) | |
2586 | { | |
2587 | pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg, | |
2588 | msg->front_max, msg->nr_pages); | |
2589 | print_hex_dump(KERN_DEBUG, "header: ", | |
2590 | DUMP_PREFIX_OFFSET, 16, 1, | |
2591 | &msg->hdr, sizeof(msg->hdr), true); | |
2592 | print_hex_dump(KERN_DEBUG, " front: ", | |
2593 | DUMP_PREFIX_OFFSET, 16, 1, | |
2594 | msg->front.iov_base, msg->front.iov_len, true); | |
2595 | if (msg->middle) | |
2596 | print_hex_dump(KERN_DEBUG, "middle: ", | |
2597 | DUMP_PREFIX_OFFSET, 16, 1, | |
2598 | msg->middle->vec.iov_base, | |
2599 | msg->middle->vec.iov_len, true); | |
2600 | print_hex_dump(KERN_DEBUG, "footer: ", | |
2601 | DUMP_PREFIX_OFFSET, 16, 1, | |
2602 | &msg->footer, sizeof(msg->footer), true); | |
2603 | } | |
3d14c5d2 | 2604 | EXPORT_SYMBOL(ceph_msg_dump); |