blob: 71263a2a78d71c06ca1c647327c42d8e8b4aa9f5 [file] [log] [blame]
Sage Weil31b80062009-10-06 11:31:13 -07001#include "ceph_debug.h"
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09009#include <linux/slab.h>
Sage Weil31b80062009-10-06 11:31:13 -070010#include <linux/socket.h>
11#include <linux/string.h>
12#include <net/tcp.h>
13
14#include "super.h"
15#include "messenger.h"
Sage Weil63f2d212009-11-03 15:17:56 -080016#include "decode.h"
Sage Weil58bb3b32009-12-23 12:12:31 -080017#include "pagelist.h"
Sage Weil31b80062009-10-06 11:31:13 -070018
19/*
20 * Ceph uses the messenger to exchange ceph_msg messages with other
21 * hosts in the system. The messenger provides ordered and reliable
22 * delivery. We tolerate TCP disconnects by reconnecting (with
23 * exponential backoff) in the case of a fault (disconnection, bad
24 * crc, protocol error). Acks allow sent messages to be discarded by
25 * the sender.
26 */
27
28/* static tag bytes (protocol control messages) */
29static char tag_msg = CEPH_MSGR_TAG_MSG;
30static char tag_ack = CEPH_MSGR_TAG_ACK;
31static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
32
Sage Weila6a53492010-04-13 14:07:07 -070033#ifdef CONFIG_LOCKDEP
34static struct lock_class_key socket_class;
35#endif
36
Sage Weil31b80062009-10-06 11:31:13 -070037
38static void queue_con(struct ceph_connection *con);
39static void con_work(struct work_struct *);
40static void ceph_fault(struct ceph_connection *con);
41
Sage Weil31b80062009-10-06 11:31:13 -070042/*
43 * nicely render a sockaddr as a string.
44 */
45#define MAX_ADDR_STR 20
Sage Weild06dbaf2010-07-08 10:47:16 -070046#define MAX_ADDR_STR_LEN 60
47static char addr_str[MAX_ADDR_STR][MAX_ADDR_STR_LEN];
Sage Weil31b80062009-10-06 11:31:13 -070048static DEFINE_SPINLOCK(addr_str_lock);
49static int last_addr_str;
50
51const char *pr_addr(const struct sockaddr_storage *ss)
52{
53 int i;
54 char *s;
55 struct sockaddr_in *in4 = (void *)ss;
Sage Weil31b80062009-10-06 11:31:13 -070056 struct sockaddr_in6 *in6 = (void *)ss;
57
58 spin_lock(&addr_str_lock);
59 i = last_addr_str++;
60 if (last_addr_str == MAX_ADDR_STR)
61 last_addr_str = 0;
62 spin_unlock(&addr_str_lock);
63 s = addr_str[i];
64
65 switch (ss->ss_family) {
66 case AF_INET:
Sage Weild06dbaf2010-07-08 10:47:16 -070067 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%u", &in4->sin_addr,
68 (unsigned int)ntohs(in4->sin_port));
Sage Weil31b80062009-10-06 11:31:13 -070069 break;
70
71 case AF_INET6:
Sage Weild06dbaf2010-07-08 10:47:16 -070072 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%u", &in6->sin6_addr,
73 (unsigned int)ntohs(in6->sin6_port));
Sage Weil31b80062009-10-06 11:31:13 -070074 break;
75
76 default:
77 sprintf(s, "(unknown sockaddr family %d)", (int)ss->ss_family);
78 }
79
80 return s;
81}
82
Sage Weil63f2d212009-11-03 15:17:56 -080083static void encode_my_addr(struct ceph_messenger *msgr)
84{
85 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
86 ceph_encode_addr(&msgr->my_enc_addr);
87}
88
Sage Weil31b80062009-10-06 11:31:13 -070089/*
90 * work queue for all reading and writing to/from the socket.
91 */
92struct workqueue_struct *ceph_msgr_wq;
93
94int __init ceph_msgr_init(void)
95{
96 ceph_msgr_wq = create_workqueue("ceph-msgr");
97 if (IS_ERR(ceph_msgr_wq)) {
98 int ret = PTR_ERR(ceph_msgr_wq);
99 pr_err("msgr_init failed to create workqueue: %d\n", ret);
100 ceph_msgr_wq = NULL;
101 return ret;
102 }
103 return 0;
104}
105
106void ceph_msgr_exit(void)
107{
108 destroy_workqueue(ceph_msgr_wq);
109}
110
Yehuda Sadehcd84db62010-06-11 16:58:48 -0700111void ceph_msgr_flush(void)
Sage Weila922d382010-05-29 09:41:23 -0700112{
113 flush_workqueue(ceph_msgr_wq);
114}
115
116
Sage Weil31b80062009-10-06 11:31:13 -0700117/*
118 * socket callback functions
119 */
120
121/* data available on socket, or listen socket received a connect */
122static void ceph_data_ready(struct sock *sk, int count_unused)
123{
124 struct ceph_connection *con =
125 (struct ceph_connection *)sk->sk_user_data;
126 if (sk->sk_state != TCP_CLOSE_WAIT) {
127 dout("ceph_data_ready on %p state = %lu, queueing work\n",
128 con, con->state);
129 queue_con(con);
130 }
131}
132
133/* socket has buffer space for writing */
134static void ceph_write_space(struct sock *sk)
135{
136 struct ceph_connection *con =
137 (struct ceph_connection *)sk->sk_user_data;
138
139 /* only queue to workqueue if there is data we want to write. */
140 if (test_bit(WRITE_PENDING, &con->state)) {
141 dout("ceph_write_space %p queueing write work\n", con);
142 queue_con(con);
143 } else {
144 dout("ceph_write_space %p nothing to write\n", con);
145 }
146
147 /* since we have our own write_space, clear the SOCK_NOSPACE flag */
148 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
149}
150
151/* socket's state has changed */
152static void ceph_state_change(struct sock *sk)
153{
154 struct ceph_connection *con =
155 (struct ceph_connection *)sk->sk_user_data;
156
157 dout("ceph_state_change %p state = %lu sk_state = %u\n",
158 con, con->state, sk->sk_state);
159
160 if (test_bit(CLOSED, &con->state))
161 return;
162
163 switch (sk->sk_state) {
164 case TCP_CLOSE:
165 dout("ceph_state_change TCP_CLOSE\n");
166 case TCP_CLOSE_WAIT:
167 dout("ceph_state_change TCP_CLOSE_WAIT\n");
168 if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) {
169 if (test_bit(CONNECTING, &con->state))
170 con->error_msg = "connection failed";
171 else
172 con->error_msg = "socket closed";
173 queue_con(con);
174 }
175 break;
176 case TCP_ESTABLISHED:
177 dout("ceph_state_change TCP_ESTABLISHED\n");
178 queue_con(con);
179 break;
180 }
181}
182
183/*
184 * set up socket callbacks
185 */
186static void set_sock_callbacks(struct socket *sock,
187 struct ceph_connection *con)
188{
189 struct sock *sk = sock->sk;
190 sk->sk_user_data = (void *)con;
191 sk->sk_data_ready = ceph_data_ready;
192 sk->sk_write_space = ceph_write_space;
193 sk->sk_state_change = ceph_state_change;
194}
195
196
197/*
198 * socket helpers
199 */
200
201/*
202 * initiate connection to a remote socket.
203 */
204static struct socket *ceph_tcp_connect(struct ceph_connection *con)
205{
Sage Weilf91d3472010-07-01 15:18:31 -0700206 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
Sage Weil31b80062009-10-06 11:31:13 -0700207 struct socket *sock;
208 int ret;
209
210 BUG_ON(con->sock);
Sage Weilf91d3472010-07-01 15:18:31 -0700211 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
212 IPPROTO_TCP, &sock);
Sage Weil31b80062009-10-06 11:31:13 -0700213 if (ret)
214 return ERR_PTR(ret);
215 con->sock = sock;
216 sock->sk->sk_allocation = GFP_NOFS;
217
Sage Weila6a53492010-04-13 14:07:07 -0700218#ifdef CONFIG_LOCKDEP
219 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
220#endif
221
Sage Weil31b80062009-10-06 11:31:13 -0700222 set_sock_callbacks(sock, con);
223
224 dout("connect %s\n", pr_addr(&con->peer_addr.in_addr));
225
Sage Weilf91d3472010-07-01 15:18:31 -0700226 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
227 O_NONBLOCK);
Sage Weil31b80062009-10-06 11:31:13 -0700228 if (ret == -EINPROGRESS) {
229 dout("connect %s EINPROGRESS sk_state = %u\n",
230 pr_addr(&con->peer_addr.in_addr),
231 sock->sk->sk_state);
232 ret = 0;
233 }
234 if (ret < 0) {
235 pr_err("connect %s error %d\n",
236 pr_addr(&con->peer_addr.in_addr), ret);
237 sock_release(sock);
238 con->sock = NULL;
239 con->error_msg = "connect error";
240 }
241
242 if (ret < 0)
243 return ERR_PTR(ret);
244 return sock;
245}
246
247static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
248{
249 struct kvec iov = {buf, len};
250 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
251
252 return kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
253}
254
255/*
256 * write something. @more is true if caller will be sending more data
257 * shortly.
258 */
259static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
260 size_t kvlen, size_t len, int more)
261{
262 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
263
264 if (more)
265 msg.msg_flags |= MSG_MORE;
266 else
267 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
268
269 return kernel_sendmsg(sock, &msg, iov, kvlen, len);
270}
271
272
273/*
274 * Shutdown/close the socket for the given connection.
275 */
276static int con_close_socket(struct ceph_connection *con)
277{
278 int rc;
279
280 dout("con_close_socket on %p sock %p\n", con, con->sock);
281 if (!con->sock)
282 return 0;
283 set_bit(SOCK_CLOSED, &con->state);
284 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
285 sock_release(con->sock);
286 con->sock = NULL;
287 clear_bit(SOCK_CLOSED, &con->state);
288 return rc;
289}
290
291/*
292 * Reset a connection. Discard all incoming and outgoing messages
293 * and clear *_seq state.
294 */
295static void ceph_msg_remove(struct ceph_msg *msg)
296{
297 list_del_init(&msg->list_head);
298 ceph_msg_put(msg);
299}
300static void ceph_msg_remove_list(struct list_head *head)
301{
302 while (!list_empty(head)) {
303 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
304 list_head);
305 ceph_msg_remove(msg);
306 }
307}
308
309static void reset_connection(struct ceph_connection *con)
310{
311 /* reset connection, out_queue, msg_ and connect_seq */
312 /* discard existing out_queue and msg_seq */
Sage Weil31b80062009-10-06 11:31:13 -0700313 ceph_msg_remove_list(&con->out_queue);
314 ceph_msg_remove_list(&con->out_sent);
315
Sage Weilcf3e5c42009-12-11 09:48:05 -0800316 if (con->in_msg) {
317 ceph_msg_put(con->in_msg);
318 con->in_msg = NULL;
319 }
320
Sage Weil31b80062009-10-06 11:31:13 -0700321 con->connect_seq = 0;
322 con->out_seq = 0;
Sage Weilc86a2932009-12-14 14:04:30 -0800323 if (con->out_msg) {
324 ceph_msg_put(con->out_msg);
325 con->out_msg = NULL;
326 }
Sage Weil6f2bc3f2010-04-02 16:16:34 -0700327 con->out_keepalive_pending = false;
Sage Weil31b80062009-10-06 11:31:13 -0700328 con->in_seq = 0;
Sage Weil0e0d5e02010-04-02 16:07:19 -0700329 con->in_seq_acked = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700330}
331
332/*
333 * mark a peer down. drop any open connections.
334 */
335void ceph_con_close(struct ceph_connection *con)
336{
337 dout("con_close %p peer %s\n", con, pr_addr(&con->peer_addr.in_addr));
338 set_bit(CLOSED, &con->state); /* in case there's queued work */
339 clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */
Sage Weil1679f872010-02-26 13:55:51 -0800340 clear_bit(LOSSYTX, &con->state); /* so we retry next connect */
341 clear_bit(KEEPALIVE_PENDING, &con->state);
342 clear_bit(WRITE_PENDING, &con->state);
Sage Weilec302642009-12-22 10:43:42 -0800343 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700344 reset_connection(con);
Sage Weil6f2bc3f2010-04-02 16:16:34 -0700345 con->peer_global_seq = 0;
Sage Weil91e45ce32010-02-15 12:05:09 -0800346 cancel_delayed_work(&con->work);
Sage Weilec302642009-12-22 10:43:42 -0800347 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700348 queue_con(con);
349}
350
351/*
Sage Weil31b80062009-10-06 11:31:13 -0700352 * Reopen a closed connection, with a new peer address.
353 */
354void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr)
355{
356 dout("con_open %p %s\n", con, pr_addr(&addr->in_addr));
357 set_bit(OPENING, &con->state);
358 clear_bit(CLOSED, &con->state);
359 memcpy(&con->peer_addr, addr, sizeof(*addr));
Sage Weil03c677e2009-11-20 15:14:15 -0800360 con->delay = 0; /* reset backoff memory */
Sage Weil31b80062009-10-06 11:31:13 -0700361 queue_con(con);
362}
363
364/*
Sage Weil87b315a2010-03-22 14:51:18 -0700365 * return true if this connection ever successfully opened
366 */
367bool ceph_con_opened(struct ceph_connection *con)
368{
369 return con->connect_seq > 0;
370}
371
372/*
Sage Weil31b80062009-10-06 11:31:13 -0700373 * generic get/put
374 */
375struct ceph_connection *ceph_con_get(struct ceph_connection *con)
376{
377 dout("con_get %p nref = %d -> %d\n", con,
378 atomic_read(&con->nref), atomic_read(&con->nref) + 1);
379 if (atomic_inc_not_zero(&con->nref))
380 return con;
381 return NULL;
382}
383
384void ceph_con_put(struct ceph_connection *con)
385{
386 dout("con_put %p nref = %d -> %d\n", con,
387 atomic_read(&con->nref), atomic_read(&con->nref) - 1);
388 BUG_ON(atomic_read(&con->nref) == 0);
389 if (atomic_dec_and_test(&con->nref)) {
Sage Weil71ececd2009-11-18 11:27:06 -0800390 BUG_ON(con->sock);
Sage Weil31b80062009-10-06 11:31:13 -0700391 kfree(con);
392 }
393}
394
395/*
396 * initialize a new connection.
397 */
398void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con)
399{
400 dout("con_init %p\n", con);
401 memset(con, 0, sizeof(*con));
402 atomic_set(&con->nref, 1);
403 con->msgr = msgr;
Sage Weilec302642009-12-22 10:43:42 -0800404 mutex_init(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700405 INIT_LIST_HEAD(&con->out_queue);
406 INIT_LIST_HEAD(&con->out_sent);
407 INIT_DELAYED_WORK(&con->work, con_work);
408}
409
410
411/*
412 * We maintain a global counter to order connection attempts. Get
413 * a unique seq greater than @gt.
414 */
415static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
416{
417 u32 ret;
418
419 spin_lock(&msgr->global_seq_lock);
420 if (msgr->global_seq < gt)
421 msgr->global_seq = gt;
422 ret = ++msgr->global_seq;
423 spin_unlock(&msgr->global_seq_lock);
424 return ret;
425}
426
427
428/*
429 * Prepare footer for currently outgoing message, and finish things
430 * off. Assumes out_kvec* are already valid.. we just add on to the end.
431 */
432static void prepare_write_message_footer(struct ceph_connection *con, int v)
433{
434 struct ceph_msg *m = con->out_msg;
435
436 dout("prepare_write_message_footer %p\n", con);
437 con->out_kvec_is_msg = true;
438 con->out_kvec[v].iov_base = &m->footer;
439 con->out_kvec[v].iov_len = sizeof(m->footer);
440 con->out_kvec_bytes += sizeof(m->footer);
441 con->out_kvec_left++;
442 con->out_more = m->more_to_follow;
Sage Weilc86a2932009-12-14 14:04:30 -0800443 con->out_msg_done = true;
Sage Weil31b80062009-10-06 11:31:13 -0700444}
445
446/*
447 * Prepare headers for the next outgoing message.
448 */
449static void prepare_write_message(struct ceph_connection *con)
450{
451 struct ceph_msg *m;
452 int v = 0;
453
454 con->out_kvec_bytes = 0;
455 con->out_kvec_is_msg = true;
Sage Weilc86a2932009-12-14 14:04:30 -0800456 con->out_msg_done = false;
Sage Weil31b80062009-10-06 11:31:13 -0700457
458 /* Sneak an ack in there first? If we can get it into the same
459 * TCP packet that's a good thing. */
460 if (con->in_seq > con->in_seq_acked) {
461 con->in_seq_acked = con->in_seq;
462 con->out_kvec[v].iov_base = &tag_ack;
463 con->out_kvec[v++].iov_len = 1;
464 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
465 con->out_kvec[v].iov_base = &con->out_temp_ack;
466 con->out_kvec[v++].iov_len = sizeof(con->out_temp_ack);
467 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
468 }
469
Sage Weil31b80062009-10-06 11:31:13 -0700470 m = list_first_entry(&con->out_queue,
471 struct ceph_msg, list_head);
Sage Weilc86a2932009-12-14 14:04:30 -0800472 con->out_msg = m;
Sage Weilb3d1dbb2009-12-14 14:58:11 -0800473 if (test_bit(LOSSYTX, &con->state)) {
Sage Weil6c5d1a42010-02-13 20:29:31 -0800474 list_del_init(&m->list_head);
475 } else {
Sage Weilb3d1dbb2009-12-14 14:58:11 -0800476 /* put message on sent list */
477 ceph_msg_get(m);
478 list_move_tail(&m->list_head, &con->out_sent);
Sage Weilb3d1dbb2009-12-14 14:58:11 -0800479 }
Sage Weil31b80062009-10-06 11:31:13 -0700480
Sage Weile84346b2010-05-11 21:20:38 -0700481 /*
482 * only assign outgoing seq # if we haven't sent this message
483 * yet. if it is requeued, resend with it's original seq.
484 */
485 if (m->needs_out_seq) {
486 m->hdr.seq = cpu_to_le64(++con->out_seq);
487 m->needs_out_seq = false;
488 }
Sage Weil31b80062009-10-06 11:31:13 -0700489
490 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
491 m, con->out_seq, le16_to_cpu(m->hdr.type),
492 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
493 le32_to_cpu(m->hdr.data_len),
494 m->nr_pages);
495 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
496
497 /* tag + hdr + front + middle */
498 con->out_kvec[v].iov_base = &tag_msg;
499 con->out_kvec[v++].iov_len = 1;
500 con->out_kvec[v].iov_base = &m->hdr;
501 con->out_kvec[v++].iov_len = sizeof(m->hdr);
502 con->out_kvec[v++] = m->front;
503 if (m->middle)
504 con->out_kvec[v++] = m->middle->vec;
505 con->out_kvec_left = v;
506 con->out_kvec_bytes += 1 + sizeof(m->hdr) + m->front.iov_len +
507 (m->middle ? m->middle->vec.iov_len : 0);
508 con->out_kvec_cur = con->out_kvec;
509
510 /* fill in crc (except data pages), footer */
511 con->out_msg->hdr.crc =
512 cpu_to_le32(crc32c(0, (void *)&m->hdr,
513 sizeof(m->hdr) - sizeof(m->hdr.crc)));
514 con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE;
515 con->out_msg->footer.front_crc =
516 cpu_to_le32(crc32c(0, m->front.iov_base, m->front.iov_len));
517 if (m->middle)
518 con->out_msg->footer.middle_crc =
519 cpu_to_le32(crc32c(0, m->middle->vec.iov_base,
520 m->middle->vec.iov_len));
521 else
522 con->out_msg->footer.middle_crc = 0;
523 con->out_msg->footer.data_crc = 0;
524 dout("prepare_write_message front_crc %u data_crc %u\n",
525 le32_to_cpu(con->out_msg->footer.front_crc),
526 le32_to_cpu(con->out_msg->footer.middle_crc));
527
528 /* is there a data payload? */
529 if (le32_to_cpu(m->hdr.data_len) > 0) {
530 /* initialize page iterator */
531 con->out_msg_pos.page = 0;
532 con->out_msg_pos.page_pos =
533 le16_to_cpu(m->hdr.data_off) & ~PAGE_MASK;
534 con->out_msg_pos.data_pos = 0;
535 con->out_msg_pos.did_page_crc = 0;
536 con->out_more = 1; /* data + footer will follow */
537 } else {
538 /* no, queue up footer too and be done */
539 prepare_write_message_footer(con, v);
540 }
541
542 set_bit(WRITE_PENDING, &con->state);
543}
544
545/*
546 * Prepare an ack.
547 */
548static void prepare_write_ack(struct ceph_connection *con)
549{
550 dout("prepare_write_ack %p %llu -> %llu\n", con,
551 con->in_seq_acked, con->in_seq);
552 con->in_seq_acked = con->in_seq;
553
554 con->out_kvec[0].iov_base = &tag_ack;
555 con->out_kvec[0].iov_len = 1;
556 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
557 con->out_kvec[1].iov_base = &con->out_temp_ack;
558 con->out_kvec[1].iov_len = sizeof(con->out_temp_ack);
559 con->out_kvec_left = 2;
560 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
561 con->out_kvec_cur = con->out_kvec;
562 con->out_more = 1; /* more will follow.. eventually.. */
563 set_bit(WRITE_PENDING, &con->state);
564}
565
566/*
567 * Prepare to write keepalive byte.
568 */
569static void prepare_write_keepalive(struct ceph_connection *con)
570{
571 dout("prepare_write_keepalive %p\n", con);
572 con->out_kvec[0].iov_base = &tag_keepalive;
573 con->out_kvec[0].iov_len = 1;
574 con->out_kvec_left = 1;
575 con->out_kvec_bytes = 1;
576 con->out_kvec_cur = con->out_kvec;
577 set_bit(WRITE_PENDING, &con->state);
578}
579
580/*
581 * Connection negotiation.
582 */
583
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800584static void prepare_connect_authorizer(struct ceph_connection *con)
585{
586 void *auth_buf;
587 int auth_len = 0;
588 int auth_protocol = 0;
589
Sage Weilec302642009-12-22 10:43:42 -0800590 mutex_unlock(&con->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800591 if (con->ops->get_authorizer)
592 con->ops->get_authorizer(con, &auth_buf, &auth_len,
593 &auth_protocol, &con->auth_reply_buf,
594 &con->auth_reply_buf_len,
595 con->auth_retry);
Sage Weilec302642009-12-22 10:43:42 -0800596 mutex_lock(&con->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800597
598 con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol);
599 con->out_connect.authorizer_len = cpu_to_le32(auth_len);
600
601 con->out_kvec[con->out_kvec_left].iov_base = auth_buf;
602 con->out_kvec[con->out_kvec_left].iov_len = auth_len;
603 con->out_kvec_left++;
604 con->out_kvec_bytes += auth_len;
605}
606
Sage Weil31b80062009-10-06 11:31:13 -0700607/*
608 * We connected to a peer and are saying hello.
609 */
Sage Weileed0ef22009-11-10 14:34:36 -0800610static void prepare_write_banner(struct ceph_messenger *msgr,
611 struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -0700612{
613 int len = strlen(CEPH_BANNER);
Sage Weileed0ef22009-11-10 14:34:36 -0800614
615 con->out_kvec[0].iov_base = CEPH_BANNER;
616 con->out_kvec[0].iov_len = len;
617 con->out_kvec[1].iov_base = &msgr->my_enc_addr;
618 con->out_kvec[1].iov_len = sizeof(msgr->my_enc_addr);
619 con->out_kvec_left = 2;
620 con->out_kvec_bytes = len + sizeof(msgr->my_enc_addr);
621 con->out_kvec_cur = con->out_kvec;
622 con->out_more = 0;
623 set_bit(WRITE_PENDING, &con->state);
624}
625
626static void prepare_write_connect(struct ceph_messenger *msgr,
627 struct ceph_connection *con,
628 int after_banner)
629{
Sage Weil31b80062009-10-06 11:31:13 -0700630 unsigned global_seq = get_global_seq(con->msgr, 0);
631 int proto;
632
633 switch (con->peer_name.type) {
634 case CEPH_ENTITY_TYPE_MON:
635 proto = CEPH_MONC_PROTOCOL;
636 break;
637 case CEPH_ENTITY_TYPE_OSD:
638 proto = CEPH_OSDC_PROTOCOL;
639 break;
640 case CEPH_ENTITY_TYPE_MDS:
641 proto = CEPH_MDSC_PROTOCOL;
642 break;
643 default:
644 BUG();
645 }
646
647 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
648 con->connect_seq, global_seq, proto);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800649
Yehuda Sadeh0cf55372010-06-11 15:57:06 -0700650 con->out_connect.features = cpu_to_le64(CEPH_FEATURE_SUPPORTED_CLIENT);
Sage Weil31b80062009-10-06 11:31:13 -0700651 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
652 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
653 con->out_connect.global_seq = cpu_to_le32(global_seq);
654 con->out_connect.protocol_version = cpu_to_le32(proto);
655 con->out_connect.flags = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700656
Sage Weileed0ef22009-11-10 14:34:36 -0800657 if (!after_banner) {
658 con->out_kvec_left = 0;
659 con->out_kvec_bytes = 0;
660 }
661 con->out_kvec[con->out_kvec_left].iov_base = &con->out_connect;
662 con->out_kvec[con->out_kvec_left].iov_len = sizeof(con->out_connect);
663 con->out_kvec_left++;
664 con->out_kvec_bytes += sizeof(con->out_connect);
Sage Weil31b80062009-10-06 11:31:13 -0700665 con->out_kvec_cur = con->out_kvec;
666 con->out_more = 0;
667 set_bit(WRITE_PENDING, &con->state);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800668
669 prepare_connect_authorizer(con);
Sage Weil31b80062009-10-06 11:31:13 -0700670}
671
672
673/*
674 * write as much of pending kvecs to the socket as we can.
675 * 1 -> done
676 * 0 -> socket full, but more to do
677 * <0 -> error
678 */
679static int write_partial_kvec(struct ceph_connection *con)
680{
681 int ret;
682
683 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
684 while (con->out_kvec_bytes > 0) {
685 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
686 con->out_kvec_left, con->out_kvec_bytes,
687 con->out_more);
688 if (ret <= 0)
689 goto out;
690 con->out_kvec_bytes -= ret;
691 if (con->out_kvec_bytes == 0)
692 break; /* done */
693 while (ret > 0) {
694 if (ret >= con->out_kvec_cur->iov_len) {
695 ret -= con->out_kvec_cur->iov_len;
696 con->out_kvec_cur++;
697 con->out_kvec_left--;
698 } else {
699 con->out_kvec_cur->iov_len -= ret;
700 con->out_kvec_cur->iov_base += ret;
701 ret = 0;
702 break;
703 }
704 }
705 }
706 con->out_kvec_left = 0;
707 con->out_kvec_is_msg = false;
708 ret = 1;
709out:
710 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
711 con->out_kvec_bytes, con->out_kvec_left, ret);
712 return ret; /* done! */
713}
714
715/*
716 * Write as much message data payload as we can. If we finish, queue
717 * up the footer.
718 * 1 -> done, footer is now queued in out_kvec[].
719 * 0 -> socket full, but more to do
720 * <0 -> error
721 */
722static int write_partial_msg_pages(struct ceph_connection *con)
723{
724 struct ceph_msg *msg = con->out_msg;
725 unsigned data_len = le32_to_cpu(msg->hdr.data_len);
726 size_t len;
727 int crc = con->msgr->nocrc;
728 int ret;
729
730 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
731 con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages,
732 con->out_msg_pos.page_pos);
733
734 while (con->out_msg_pos.page < con->out_msg->nr_pages) {
735 struct page *page = NULL;
736 void *kaddr = NULL;
737
738 /*
739 * if we are calculating the data crc (the default), we need
740 * to map the page. if our pages[] has been revoked, use the
741 * zero page.
742 */
743 if (msg->pages) {
744 page = msg->pages[con->out_msg_pos.page];
745 if (crc)
746 kaddr = kmap(page);
Sage Weil58bb3b32009-12-23 12:12:31 -0800747 } else if (msg->pagelist) {
748 page = list_first_entry(&msg->pagelist->head,
749 struct page, lru);
750 if (crc)
751 kaddr = kmap(page);
Sage Weil31b80062009-10-06 11:31:13 -0700752 } else {
753 page = con->msgr->zero_page;
754 if (crc)
755 kaddr = page_address(con->msgr->zero_page);
756 }
757 len = min((int)(PAGE_SIZE - con->out_msg_pos.page_pos),
758 (int)(data_len - con->out_msg_pos.data_pos));
759 if (crc && !con->out_msg_pos.did_page_crc) {
760 void *base = kaddr + con->out_msg_pos.page_pos;
761 u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc);
762
763 BUG_ON(kaddr == NULL);
764 con->out_msg->footer.data_crc =
765 cpu_to_le32(crc32c(tmpcrc, base, len));
766 con->out_msg_pos.did_page_crc = 1;
767 }
768
769 ret = kernel_sendpage(con->sock, page,
770 con->out_msg_pos.page_pos, len,
771 MSG_DONTWAIT | MSG_NOSIGNAL |
772 MSG_MORE);
773
Sage Weil58bb3b32009-12-23 12:12:31 -0800774 if (crc && (msg->pages || msg->pagelist))
Sage Weil31b80062009-10-06 11:31:13 -0700775 kunmap(page);
776
777 if (ret <= 0)
778 goto out;
779
780 con->out_msg_pos.data_pos += ret;
781 con->out_msg_pos.page_pos += ret;
782 if (ret == len) {
783 con->out_msg_pos.page_pos = 0;
784 con->out_msg_pos.page++;
785 con->out_msg_pos.did_page_crc = 0;
Sage Weil58bb3b32009-12-23 12:12:31 -0800786 if (msg->pagelist)
787 list_move_tail(&page->lru,
788 &msg->pagelist->head);
Sage Weil31b80062009-10-06 11:31:13 -0700789 }
790 }
791
792 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
793
794 /* prepare and queue up footer, too */
795 if (!crc)
796 con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
797 con->out_kvec_bytes = 0;
798 con->out_kvec_left = 0;
799 con->out_kvec_cur = con->out_kvec;
800 prepare_write_message_footer(con, 0);
801 ret = 1;
802out:
803 return ret;
804}
805
806/*
807 * write some zeros
808 */
809static int write_partial_skip(struct ceph_connection *con)
810{
811 int ret;
812
813 while (con->out_skip > 0) {
814 struct kvec iov = {
815 .iov_base = page_address(con->msgr->zero_page),
816 .iov_len = min(con->out_skip, (int)PAGE_CACHE_SIZE)
817 };
818
819 ret = ceph_tcp_sendmsg(con->sock, &iov, 1, iov.iov_len, 1);
820 if (ret <= 0)
821 goto out;
822 con->out_skip -= ret;
823 }
824 ret = 1;
825out:
826 return ret;
827}
828
829/*
830 * Prepare to read connection handshake, or an ack.
831 */
Sage Weileed0ef22009-11-10 14:34:36 -0800832static void prepare_read_banner(struct ceph_connection *con)
833{
834 dout("prepare_read_banner %p\n", con);
835 con->in_base_pos = 0;
836}
837
Sage Weil31b80062009-10-06 11:31:13 -0700838static void prepare_read_connect(struct ceph_connection *con)
839{
840 dout("prepare_read_connect %p\n", con);
841 con->in_base_pos = 0;
842}
843
844static void prepare_read_ack(struct ceph_connection *con)
845{
846 dout("prepare_read_ack %p\n", con);
847 con->in_base_pos = 0;
848}
849
850static void prepare_read_tag(struct ceph_connection *con)
851{
852 dout("prepare_read_tag %p\n", con);
853 con->in_base_pos = 0;
854 con->in_tag = CEPH_MSGR_TAG_READY;
855}
856
857/*
858 * Prepare to read a message.
859 */
860static int prepare_read_message(struct ceph_connection *con)
861{
862 dout("prepare_read_message %p\n", con);
863 BUG_ON(con->in_msg != NULL);
864 con->in_base_pos = 0;
865 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
866 return 0;
867}
868
869
870static int read_partial(struct ceph_connection *con,
871 int *to, int size, void *object)
872{
873 *to += size;
874 while (con->in_base_pos < *to) {
875 int left = *to - con->in_base_pos;
876 int have = size - left;
877 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
878 if (ret <= 0)
879 return ret;
880 con->in_base_pos += ret;
881 }
882 return 1;
883}
884
885
886/*
887 * Read all or part of the connect-side handshake on a new connection
888 */
Sage Weileed0ef22009-11-10 14:34:36 -0800889static int read_partial_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -0700890{
891 int ret, to = 0;
892
Sage Weileed0ef22009-11-10 14:34:36 -0800893 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
Sage Weil31b80062009-10-06 11:31:13 -0700894
895 /* peer's banner */
896 ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner);
897 if (ret <= 0)
898 goto out;
899 ret = read_partial(con, &to, sizeof(con->actual_peer_addr),
900 &con->actual_peer_addr);
901 if (ret <= 0)
902 goto out;
903 ret = read_partial(con, &to, sizeof(con->peer_addr_for_me),
904 &con->peer_addr_for_me);
905 if (ret <= 0)
906 goto out;
Sage Weileed0ef22009-11-10 14:34:36 -0800907out:
908 return ret;
909}
910
911static int read_partial_connect(struct ceph_connection *con)
912{
913 int ret, to = 0;
914
915 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
916
Sage Weil31b80062009-10-06 11:31:13 -0700917 ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply);
918 if (ret <= 0)
919 goto out;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800920 ret = read_partial(con, &to, le32_to_cpu(con->in_reply.authorizer_len),
921 con->auth_reply_buf);
922 if (ret <= 0)
923 goto out;
Sage Weil31b80062009-10-06 11:31:13 -0700924
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800925 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
926 con, (int)con->in_reply.tag,
927 le32_to_cpu(con->in_reply.connect_seq),
Sage Weil31b80062009-10-06 11:31:13 -0700928 le32_to_cpu(con->in_reply.global_seq));
929out:
930 return ret;
Sage Weileed0ef22009-11-10 14:34:36 -0800931
Sage Weil31b80062009-10-06 11:31:13 -0700932}
933
934/*
935 * Verify the hello banner looks okay.
936 */
937static int verify_hello(struct ceph_connection *con)
938{
939 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
Sage Weil13e38c82009-10-09 16:36:34 -0700940 pr_err("connect to %s got bad banner\n",
Sage Weil31b80062009-10-06 11:31:13 -0700941 pr_addr(&con->peer_addr.in_addr));
942 con->error_msg = "protocol error, bad banner";
943 return -1;
944 }
945 return 0;
946}
947
948static bool addr_is_blank(struct sockaddr_storage *ss)
949{
950 switch (ss->ss_family) {
951 case AF_INET:
952 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
953 case AF_INET6:
954 return
955 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
956 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
957 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
958 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
959 }
960 return false;
961}
962
963static int addr_port(struct sockaddr_storage *ss)
964{
965 switch (ss->ss_family) {
966 case AF_INET:
Sage Weilf28bcfb2009-11-04 11:46:35 -0800967 return ntohs(((struct sockaddr_in *)ss)->sin_port);
Sage Weil31b80062009-10-06 11:31:13 -0700968 case AF_INET6:
Sage Weilf28bcfb2009-11-04 11:46:35 -0800969 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
Sage Weil31b80062009-10-06 11:31:13 -0700970 }
971 return 0;
972}
973
974static void addr_set_port(struct sockaddr_storage *ss, int p)
975{
976 switch (ss->ss_family) {
977 case AF_INET:
978 ((struct sockaddr_in *)ss)->sin_port = htons(p);
979 case AF_INET6:
980 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
981 }
982}
983
984/*
985 * Parse an ip[:port] list into an addr array. Use the default
986 * monitor port if a port isn't specified.
987 */
988int ceph_parse_ips(const char *c, const char *end,
989 struct ceph_entity_addr *addr,
990 int max_count, int *count)
991{
992 int i;
993 const char *p = c;
994
995 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
996 for (i = 0; i < max_count; i++) {
997 const char *ipend;
998 struct sockaddr_storage *ss = &addr[i].in_addr;
999 struct sockaddr_in *in4 = (void *)ss;
1000 struct sockaddr_in6 *in6 = (void *)ss;
1001 int port;
Sage Weil39139f62010-07-08 09:54:52 -07001002 char delim = ',';
1003
1004 if (*p == '[') {
1005 delim = ']';
1006 p++;
1007 }
Sage Weil31b80062009-10-06 11:31:13 -07001008
1009 memset(ss, 0, sizeof(*ss));
1010 if (in4_pton(p, end - p, (u8 *)&in4->sin_addr.s_addr,
Sage Weil39139f62010-07-08 09:54:52 -07001011 delim, &ipend))
Sage Weil31b80062009-10-06 11:31:13 -07001012 ss->ss_family = AF_INET;
Sage Weil39139f62010-07-08 09:54:52 -07001013 else if (in6_pton(p, end - p, (u8 *)&in6->sin6_addr.s6_addr,
1014 delim, &ipend))
Sage Weil31b80062009-10-06 11:31:13 -07001015 ss->ss_family = AF_INET6;
Sage Weil39139f62010-07-08 09:54:52 -07001016 else
Sage Weil31b80062009-10-06 11:31:13 -07001017 goto bad;
Sage Weil31b80062009-10-06 11:31:13 -07001018 p = ipend;
1019
Sage Weil39139f62010-07-08 09:54:52 -07001020 if (delim == ']') {
1021 if (*p != ']') {
1022 dout("missing matching ']'\n");
1023 goto bad;
1024 }
1025 p++;
1026 }
1027
Sage Weil31b80062009-10-06 11:31:13 -07001028 /* port? */
1029 if (p < end && *p == ':') {
1030 port = 0;
1031 p++;
1032 while (p < end && *p >= '0' && *p <= '9') {
1033 port = (port * 10) + (*p - '0');
1034 p++;
1035 }
1036 if (port > 65535 || port == 0)
1037 goto bad;
1038 } else {
1039 port = CEPH_MON_PORT;
1040 }
1041
1042 addr_set_port(ss, port);
1043
1044 dout("parse_ips got %s\n", pr_addr(ss));
1045
1046 if (p == end)
1047 break;
1048 if (*p != ',')
1049 goto bad;
1050 p++;
1051 }
1052
1053 if (p != end)
1054 goto bad;
1055
1056 if (count)
1057 *count = i + 1;
1058 return 0;
1059
1060bad:
Sage Weil39139f62010-07-08 09:54:52 -07001061 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
Sage Weil31b80062009-10-06 11:31:13 -07001062 return -EINVAL;
1063}
1064
Sage Weileed0ef22009-11-10 14:34:36 -08001065static int process_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001066{
Sage Weileed0ef22009-11-10 14:34:36 -08001067 dout("process_banner on %p\n", con);
Sage Weil31b80062009-10-06 11:31:13 -07001068
1069 if (verify_hello(con) < 0)
1070 return -1;
1071
Sage Weil63f2d212009-11-03 15:17:56 -08001072 ceph_decode_addr(&con->actual_peer_addr);
1073 ceph_decode_addr(&con->peer_addr_for_me);
1074
Sage Weil31b80062009-10-06 11:31:13 -07001075 /*
1076 * Make sure the other end is who we wanted. note that the other
1077 * end may not yet know their ip address, so if it's 0.0.0.0, give
1078 * them the benefit of the doubt.
1079 */
Sage Weil103e2d32010-01-07 16:12:36 -08001080 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1081 sizeof(con->peer_addr)) != 0 &&
Sage Weil31b80062009-10-06 11:31:13 -07001082 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1083 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001084 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
Sage Weil103e2d32010-01-07 16:12:36 -08001085 pr_addr(&con->peer_addr.in_addr),
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001086 (int)le32_to_cpu(con->peer_addr.nonce),
Sage Weil103e2d32010-01-07 16:12:36 -08001087 pr_addr(&con->actual_peer_addr.in_addr),
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001088 (int)le32_to_cpu(con->actual_peer_addr.nonce));
Sage Weil58bb3b32009-12-23 12:12:31 -08001089 con->error_msg = "wrong peer at address";
Sage Weil31b80062009-10-06 11:31:13 -07001090 return -1;
1091 }
1092
1093 /*
1094 * did we learn our address?
1095 */
1096 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1097 int port = addr_port(&con->msgr->inst.addr.in_addr);
1098
1099 memcpy(&con->msgr->inst.addr.in_addr,
1100 &con->peer_addr_for_me.in_addr,
1101 sizeof(con->peer_addr_for_me.in_addr));
1102 addr_set_port(&con->msgr->inst.addr.in_addr, port);
Sage Weil63f2d212009-11-03 15:17:56 -08001103 encode_my_addr(con->msgr);
Sage Weileed0ef22009-11-10 14:34:36 -08001104 dout("process_banner learned my addr is %s\n",
Sage Weil31b80062009-10-06 11:31:13 -07001105 pr_addr(&con->msgr->inst.addr.in_addr));
1106 }
1107
Sage Weileed0ef22009-11-10 14:34:36 -08001108 set_bit(NEGOTIATING, &con->state);
1109 prepare_read_connect(con);
1110 return 0;
1111}
1112
Sage Weil04a419f2009-12-23 09:30:21 -08001113static void fail_protocol(struct ceph_connection *con)
1114{
1115 reset_connection(con);
1116 set_bit(CLOSED, &con->state); /* in case there's queued work */
1117
1118 mutex_unlock(&con->mutex);
1119 if (con->ops->bad_proto)
1120 con->ops->bad_proto(con);
1121 mutex_lock(&con->mutex);
1122}
1123
Sage Weileed0ef22009-11-10 14:34:36 -08001124static int process_connect(struct ceph_connection *con)
1125{
Sage Weildbad1852010-03-25 15:45:38 -07001126 u64 sup_feat = CEPH_FEATURE_SUPPORTED_CLIENT;
1127 u64 req_feat = CEPH_FEATURE_REQUIRED_CLIENT;
Sage Weil04a419f2009-12-23 09:30:21 -08001128 u64 server_feat = le64_to_cpu(con->in_reply.features);
1129
Sage Weileed0ef22009-11-10 14:34:36 -08001130 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1131
Sage Weil31b80062009-10-06 11:31:13 -07001132 switch (con->in_reply.tag) {
Sage Weil04a419f2009-12-23 09:30:21 -08001133 case CEPH_MSGR_TAG_FEATURES:
1134 pr_err("%s%lld %s feature set mismatch,"
1135 " my %llx < server's %llx, missing %llx\n",
1136 ENTITY_NAME(con->peer_name),
1137 pr_addr(&con->peer_addr.in_addr),
1138 sup_feat, server_feat, server_feat & ~sup_feat);
1139 con->error_msg = "missing required protocol features";
1140 fail_protocol(con);
1141 return -1;
1142
Sage Weil31b80062009-10-06 11:31:13 -07001143 case CEPH_MSGR_TAG_BADPROTOVER:
Sage Weil31b80062009-10-06 11:31:13 -07001144 pr_err("%s%lld %s protocol version mismatch,"
1145 " my %d != server's %d\n",
1146 ENTITY_NAME(con->peer_name),
1147 pr_addr(&con->peer_addr.in_addr),
1148 le32_to_cpu(con->out_connect.protocol_version),
1149 le32_to_cpu(con->in_reply.protocol_version));
1150 con->error_msg = "protocol version mismatch";
Sage Weil04a419f2009-12-23 09:30:21 -08001151 fail_protocol(con);
Sage Weil31b80062009-10-06 11:31:13 -07001152 return -1;
1153
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001154 case CEPH_MSGR_TAG_BADAUTHORIZER:
1155 con->auth_retry++;
1156 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1157 con->auth_retry);
1158 if (con->auth_retry == 2) {
1159 con->error_msg = "connect authorization failure";
1160 reset_connection(con);
1161 set_bit(CLOSED, &con->state);
1162 return -1;
1163 }
1164 con->auth_retry = 1;
1165 prepare_write_connect(con->msgr, con, 0);
Sage Weil63733a02010-03-15 15:47:22 -07001166 prepare_read_connect(con);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001167 break;
Sage Weil31b80062009-10-06 11:31:13 -07001168
1169 case CEPH_MSGR_TAG_RESETSESSION:
1170 /*
1171 * If we connected with a large connect_seq but the peer
1172 * has no record of a session with us (no connection, or
1173 * connect_seq == 0), they will send RESETSESION to indicate
1174 * that they must have reset their session, and may have
1175 * dropped messages.
1176 */
1177 dout("process_connect got RESET peer seq %u\n",
1178 le32_to_cpu(con->in_connect.connect_seq));
1179 pr_err("%s%lld %s connection reset\n",
1180 ENTITY_NAME(con->peer_name),
1181 pr_addr(&con->peer_addr.in_addr));
1182 reset_connection(con);
Sage Weileed0ef22009-11-10 14:34:36 -08001183 prepare_write_connect(con->msgr, con, 0);
Sage Weil31b80062009-10-06 11:31:13 -07001184 prepare_read_connect(con);
1185
1186 /* Tell ceph about it. */
Sage Weilec302642009-12-22 10:43:42 -08001187 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001188 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1189 if (con->ops->peer_reset)
1190 con->ops->peer_reset(con);
Sage Weilec302642009-12-22 10:43:42 -08001191 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001192 break;
1193
1194 case CEPH_MSGR_TAG_RETRY_SESSION:
1195 /*
1196 * If we sent a smaller connect_seq than the peer has, try
1197 * again with a larger value.
1198 */
1199 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1200 le32_to_cpu(con->out_connect.connect_seq),
1201 le32_to_cpu(con->in_connect.connect_seq));
1202 con->connect_seq = le32_to_cpu(con->in_connect.connect_seq);
Sage Weileed0ef22009-11-10 14:34:36 -08001203 prepare_write_connect(con->msgr, con, 0);
Sage Weil31b80062009-10-06 11:31:13 -07001204 prepare_read_connect(con);
1205 break;
1206
1207 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1208 /*
1209 * If we sent a smaller global_seq than the peer has, try
1210 * again with a larger value.
1211 */
Sage Weileed0ef22009-11-10 14:34:36 -08001212 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
Sage Weil31b80062009-10-06 11:31:13 -07001213 con->peer_global_seq,
1214 le32_to_cpu(con->in_connect.global_seq));
1215 get_global_seq(con->msgr,
1216 le32_to_cpu(con->in_connect.global_seq));
Sage Weileed0ef22009-11-10 14:34:36 -08001217 prepare_write_connect(con->msgr, con, 0);
Sage Weil31b80062009-10-06 11:31:13 -07001218 prepare_read_connect(con);
1219 break;
1220
1221 case CEPH_MSGR_TAG_READY:
Sage Weil04a419f2009-12-23 09:30:21 -08001222 if (req_feat & ~server_feat) {
1223 pr_err("%s%lld %s protocol feature mismatch,"
1224 " my required %llx > server's %llx, need %llx\n",
1225 ENTITY_NAME(con->peer_name),
1226 pr_addr(&con->peer_addr.in_addr),
1227 req_feat, server_feat, req_feat & ~server_feat);
1228 con->error_msg = "missing required protocol features";
1229 fail_protocol(con);
1230 return -1;
1231 }
Sage Weil31b80062009-10-06 11:31:13 -07001232 clear_bit(CONNECTING, &con->state);
Sage Weil31b80062009-10-06 11:31:13 -07001233 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1234 con->connect_seq++;
Sage Weilaba558e2010-05-12 15:23:30 -07001235 con->peer_features = server_feat;
Sage Weil31b80062009-10-06 11:31:13 -07001236 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1237 con->peer_global_seq,
1238 le32_to_cpu(con->in_reply.connect_seq),
1239 con->connect_seq);
1240 WARN_ON(con->connect_seq !=
1241 le32_to_cpu(con->in_reply.connect_seq));
Sage Weil92ac41d2009-12-14 14:56:56 -08001242
1243 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1244 set_bit(LOSSYTX, &con->state);
1245
Sage Weil31b80062009-10-06 11:31:13 -07001246 prepare_read_tag(con);
1247 break;
1248
1249 case CEPH_MSGR_TAG_WAIT:
1250 /*
1251 * If there is a connection race (we are opening
1252 * connections to each other), one of us may just have
1253 * to WAIT. This shouldn't happen if we are the
1254 * client.
1255 */
1256 pr_err("process_connect peer connecting WAIT\n");
1257
1258 default:
1259 pr_err("connect protocol error, will retry\n");
1260 con->error_msg = "protocol error, garbage tag during connect";
1261 return -1;
1262 }
1263 return 0;
1264}
1265
1266
1267/*
1268 * read (part of) an ack
1269 */
1270static int read_partial_ack(struct ceph_connection *con)
1271{
1272 int to = 0;
1273
1274 return read_partial(con, &to, sizeof(con->in_temp_ack),
1275 &con->in_temp_ack);
1276}
1277
1278
1279/*
1280 * We can finally discard anything that's been acked.
1281 */
1282static void process_ack(struct ceph_connection *con)
1283{
1284 struct ceph_msg *m;
1285 u64 ack = le64_to_cpu(con->in_temp_ack);
1286 u64 seq;
1287
Sage Weil31b80062009-10-06 11:31:13 -07001288 while (!list_empty(&con->out_sent)) {
1289 m = list_first_entry(&con->out_sent, struct ceph_msg,
1290 list_head);
1291 seq = le64_to_cpu(m->hdr.seq);
1292 if (seq > ack)
1293 break;
1294 dout("got ack for seq %llu type %d at %p\n", seq,
1295 le16_to_cpu(m->hdr.type), m);
1296 ceph_msg_remove(m);
1297 }
Sage Weil31b80062009-10-06 11:31:13 -07001298 prepare_read_tag(con);
1299}
1300
1301
1302
1303
Yehuda Sadeh24504182010-01-08 13:58:34 -08001304static int read_partial_message_section(struct ceph_connection *con,
1305 struct kvec *section, unsigned int sec_len,
1306 u32 *crc)
1307{
1308 int left;
1309 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07001310
Yehuda Sadeh24504182010-01-08 13:58:34 -08001311 BUG_ON(!section);
Sage Weil31b80062009-10-06 11:31:13 -07001312
Yehuda Sadeh24504182010-01-08 13:58:34 -08001313 while (section->iov_len < sec_len) {
1314 BUG_ON(section->iov_base == NULL);
1315 left = sec_len - section->iov_len;
1316 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1317 section->iov_len, left);
1318 if (ret <= 0)
1319 return ret;
1320 section->iov_len += ret;
1321 if (section->iov_len == sec_len)
1322 *crc = crc32c(0, section->iov_base,
1323 section->iov_len);
1324 }
1325
1326 return 1;
1327}
1328
1329static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
1330 struct ceph_msg_header *hdr,
1331 int *skip);
Sage Weil31b80062009-10-06 11:31:13 -07001332/*
1333 * read (part of) a message.
1334 */
1335static int read_partial_message(struct ceph_connection *con)
1336{
1337 struct ceph_msg *m = con->in_msg;
1338 void *p;
1339 int ret;
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08001340 int to, left;
Sage Weil31b80062009-10-06 11:31:13 -07001341 unsigned front_len, middle_len, data_len, data_off;
1342 int datacrc = con->msgr->nocrc;
Yehuda Sadeh24504182010-01-08 13:58:34 -08001343 int skip;
Sage Weilae187562010-04-22 07:47:01 -07001344 u64 seq;
Sage Weil31b80062009-10-06 11:31:13 -07001345
1346 dout("read_partial_message con %p msg %p\n", con, m);
1347
1348 /* header */
1349 while (con->in_base_pos < sizeof(con->in_hdr)) {
1350 left = sizeof(con->in_hdr) - con->in_base_pos;
1351 ret = ceph_tcp_recvmsg(con->sock,
1352 (char *)&con->in_hdr + con->in_base_pos,
1353 left);
1354 if (ret <= 0)
1355 return ret;
1356 con->in_base_pos += ret;
1357 if (con->in_base_pos == sizeof(con->in_hdr)) {
1358 u32 crc = crc32c(0, (void *)&con->in_hdr,
1359 sizeof(con->in_hdr) - sizeof(con->in_hdr.crc));
1360 if (crc != le32_to_cpu(con->in_hdr.crc)) {
1361 pr_err("read_partial_message bad hdr "
1362 " crc %u != expected %u\n",
1363 crc, con->in_hdr.crc);
1364 return -EBADMSG;
1365 }
1366 }
1367 }
Sage Weil31b80062009-10-06 11:31:13 -07001368 front_len = le32_to_cpu(con->in_hdr.front_len);
1369 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1370 return -EIO;
1371 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1372 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1373 return -EIO;
1374 data_len = le32_to_cpu(con->in_hdr.data_len);
1375 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1376 return -EIO;
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08001377 data_off = le16_to_cpu(con->in_hdr.data_off);
Sage Weil31b80062009-10-06 11:31:13 -07001378
Sage Weilae187562010-04-22 07:47:01 -07001379 /* verify seq# */
1380 seq = le64_to_cpu(con->in_hdr.seq);
1381 if ((s64)seq - (s64)con->in_seq < 1) {
1382 pr_info("skipping %s%lld %s seq %lld, expected %lld\n",
1383 ENTITY_NAME(con->peer_name),
1384 pr_addr(&con->peer_addr.in_addr),
1385 seq, con->in_seq + 1);
1386 con->in_base_pos = -front_len - middle_len - data_len -
1387 sizeof(m->footer);
1388 con->in_tag = CEPH_MSGR_TAG_READY;
1389 con->in_seq++;
1390 return 0;
1391 } else if ((s64)seq - (s64)con->in_seq > 1) {
1392 pr_err("read_partial_message bad seq %lld expected %lld\n",
1393 seq, con->in_seq + 1);
1394 con->error_msg = "bad message sequence # for incoming message";
1395 return -EBADMSG;
1396 }
1397
Sage Weil31b80062009-10-06 11:31:13 -07001398 /* allocate message? */
1399 if (!con->in_msg) {
1400 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1401 con->in_hdr.front_len, con->in_hdr.data_len);
Sage Weilae32be32010-06-13 10:30:19 -07001402 skip = 0;
Yehuda Sadeh24504182010-01-08 13:58:34 -08001403 con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip);
1404 if (skip) {
Sage Weil31b80062009-10-06 11:31:13 -07001405 /* skip this message */
Sage Weila79832f2010-04-01 16:06:19 -07001406 dout("alloc_msg said skip message\n");
Sage Weilae32be32010-06-13 10:30:19 -07001407 BUG_ON(con->in_msg);
Sage Weil31b80062009-10-06 11:31:13 -07001408 con->in_base_pos = -front_len - middle_len - data_len -
1409 sizeof(m->footer);
1410 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weil684be252010-04-21 20:45:59 -07001411 con->in_seq++;
Sage Weil31b80062009-10-06 11:31:13 -07001412 return 0;
1413 }
Sage Weila79832f2010-04-01 16:06:19 -07001414 if (!con->in_msg) {
Sage Weil5b3a4db2010-02-19 21:43:23 -08001415 con->error_msg =
1416 "error allocating memory for incoming message";
Sage Weila79832f2010-04-01 16:06:19 -07001417 return -ENOMEM;
Sage Weil31b80062009-10-06 11:31:13 -07001418 }
1419 m = con->in_msg;
1420 m->front.iov_len = 0; /* haven't read it yet */
Yehuda Sadeh24504182010-01-08 13:58:34 -08001421 if (m->middle)
1422 m->middle->vec.iov_len = 0;
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08001423
1424 con->in_msg_pos.page = 0;
1425 con->in_msg_pos.page_pos = data_off & ~PAGE_MASK;
1426 con->in_msg_pos.data_pos = 0;
Sage Weil31b80062009-10-06 11:31:13 -07001427 }
1428
1429 /* front */
Yehuda Sadeh24504182010-01-08 13:58:34 -08001430 ret = read_partial_message_section(con, &m->front, front_len,
1431 &con->in_front_crc);
1432 if (ret <= 0)
1433 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07001434
1435 /* middle */
Yehuda Sadeh24504182010-01-08 13:58:34 -08001436 if (m->middle) {
1437 ret = read_partial_message_section(con, &m->middle->vec, middle_len,
1438 &con->in_middle_crc);
Sage Weil31b80062009-10-06 11:31:13 -07001439 if (ret <= 0)
1440 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07001441 }
1442
1443 /* (page) data */
Sage Weil31b80062009-10-06 11:31:13 -07001444 while (con->in_msg_pos.data_pos < data_len) {
1445 left = min((int)(data_len - con->in_msg_pos.data_pos),
1446 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1447 BUG_ON(m->pages == NULL);
1448 p = kmap(m->pages[con->in_msg_pos.page]);
1449 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1450 left);
1451 if (ret > 0 && datacrc)
1452 con->in_data_crc =
1453 crc32c(con->in_data_crc,
1454 p + con->in_msg_pos.page_pos, ret);
1455 kunmap(m->pages[con->in_msg_pos.page]);
1456 if (ret <= 0)
1457 return ret;
1458 con->in_msg_pos.data_pos += ret;
1459 con->in_msg_pos.page_pos += ret;
1460 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1461 con->in_msg_pos.page_pos = 0;
1462 con->in_msg_pos.page++;
1463 }
1464 }
1465
Sage Weil31b80062009-10-06 11:31:13 -07001466 /* footer */
1467 to = sizeof(m->hdr) + sizeof(m->footer);
1468 while (con->in_base_pos < to) {
1469 left = to - con->in_base_pos;
1470 ret = ceph_tcp_recvmsg(con->sock, (char *)&m->footer +
1471 (con->in_base_pos - sizeof(m->hdr)),
1472 left);
1473 if (ret <= 0)
1474 return ret;
1475 con->in_base_pos += ret;
1476 }
1477 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1478 m, front_len, m->footer.front_crc, middle_len,
1479 m->footer.middle_crc, data_len, m->footer.data_crc);
1480
1481 /* crc ok? */
1482 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1483 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1484 m, con->in_front_crc, m->footer.front_crc);
1485 return -EBADMSG;
1486 }
1487 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1488 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1489 m, con->in_middle_crc, m->footer.middle_crc);
1490 return -EBADMSG;
1491 }
1492 if (datacrc &&
1493 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1494 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1495 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1496 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1497 return -EBADMSG;
1498 }
1499
1500 return 1; /* done! */
1501}
1502
1503/*
1504 * Process message. This happens in the worker thread. The callback should
1505 * be careful not to do anything that waits on other incoming messages or it
1506 * may deadlock.
1507 */
1508static void process_message(struct ceph_connection *con)
1509{
Sage Weil5e095e82009-12-14 14:30:34 -08001510 struct ceph_msg *msg;
Sage Weil31b80062009-10-06 11:31:13 -07001511
Sage Weil5e095e82009-12-14 14:30:34 -08001512 msg = con->in_msg;
Sage Weil31b80062009-10-06 11:31:13 -07001513 con->in_msg = NULL;
1514
1515 /* if first message, set peer_name */
1516 if (con->peer_name.type == 0)
Sage Weildbad1852010-03-25 15:45:38 -07001517 con->peer_name = msg->hdr.src;
Sage Weil31b80062009-10-06 11:31:13 -07001518
Sage Weil31b80062009-10-06 11:31:13 -07001519 con->in_seq++;
Sage Weilec302642009-12-22 10:43:42 -08001520 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001521
1522 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1523 msg, le64_to_cpu(msg->hdr.seq),
Sage Weildbad1852010-03-25 15:45:38 -07001524 ENTITY_NAME(msg->hdr.src),
Sage Weil31b80062009-10-06 11:31:13 -07001525 le16_to_cpu(msg->hdr.type),
1526 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1527 le32_to_cpu(msg->hdr.front_len),
1528 le32_to_cpu(msg->hdr.data_len),
1529 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1530 con->ops->dispatch(con, msg);
Sage Weilec302642009-12-22 10:43:42 -08001531
1532 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001533 prepare_read_tag(con);
1534}
1535
1536
1537/*
1538 * Write something to the socket. Called in a worker thread when the
1539 * socket appears to be writeable and we have something ready to send.
1540 */
1541static int try_write(struct ceph_connection *con)
1542{
1543 struct ceph_messenger *msgr = con->msgr;
1544 int ret = 1;
1545
1546 dout("try_write start %p state %lu nref %d\n", con, con->state,
1547 atomic_read(&con->nref));
1548
Sage Weil31b80062009-10-06 11:31:13 -07001549more:
1550 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1551
1552 /* open the socket first? */
1553 if (con->sock == NULL) {
1554 /*
1555 * if we were STANDBY and are reconnecting _this_
1556 * connection, bump connect_seq now. Always bump
1557 * global_seq.
1558 */
1559 if (test_and_clear_bit(STANDBY, &con->state))
1560 con->connect_seq++;
1561
Sage Weileed0ef22009-11-10 14:34:36 -08001562 prepare_write_banner(msgr, con);
1563 prepare_write_connect(msgr, con, 1);
1564 prepare_read_banner(con);
Sage Weil31b80062009-10-06 11:31:13 -07001565 set_bit(CONNECTING, &con->state);
Sage Weileed0ef22009-11-10 14:34:36 -08001566 clear_bit(NEGOTIATING, &con->state);
Sage Weil31b80062009-10-06 11:31:13 -07001567
Sage Weilcf3e5c42009-12-11 09:48:05 -08001568 BUG_ON(con->in_msg);
Sage Weil31b80062009-10-06 11:31:13 -07001569 con->in_tag = CEPH_MSGR_TAG_READY;
1570 dout("try_write initiating connect on %p new state %lu\n",
1571 con, con->state);
1572 con->sock = ceph_tcp_connect(con);
1573 if (IS_ERR(con->sock)) {
1574 con->sock = NULL;
1575 con->error_msg = "connect error";
1576 ret = -1;
1577 goto out;
1578 }
1579 }
1580
1581more_kvec:
1582 /* kvec data queued? */
1583 if (con->out_skip) {
1584 ret = write_partial_skip(con);
1585 if (ret <= 0)
1586 goto done;
1587 if (ret < 0) {
1588 dout("try_write write_partial_skip err %d\n", ret);
1589 goto done;
1590 }
1591 }
1592 if (con->out_kvec_left) {
1593 ret = write_partial_kvec(con);
1594 if (ret <= 0)
1595 goto done;
Sage Weil31b80062009-10-06 11:31:13 -07001596 }
1597
1598 /* msg pages? */
1599 if (con->out_msg) {
Sage Weilc86a2932009-12-14 14:04:30 -08001600 if (con->out_msg_done) {
1601 ceph_msg_put(con->out_msg);
1602 con->out_msg = NULL; /* we're done with this one */
1603 goto do_next;
1604 }
1605
Sage Weil31b80062009-10-06 11:31:13 -07001606 ret = write_partial_msg_pages(con);
1607 if (ret == 1)
1608 goto more_kvec; /* we need to send the footer, too! */
1609 if (ret == 0)
1610 goto done;
1611 if (ret < 0) {
1612 dout("try_write write_partial_msg_pages err %d\n",
1613 ret);
1614 goto done;
1615 }
1616 }
1617
Sage Weilc86a2932009-12-14 14:04:30 -08001618do_next:
Sage Weil31b80062009-10-06 11:31:13 -07001619 if (!test_bit(CONNECTING, &con->state)) {
1620 /* is anything else pending? */
1621 if (!list_empty(&con->out_queue)) {
1622 prepare_write_message(con);
1623 goto more;
1624 }
1625 if (con->in_seq > con->in_seq_acked) {
1626 prepare_write_ack(con);
1627 goto more;
1628 }
1629 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) {
1630 prepare_write_keepalive(con);
1631 goto more;
1632 }
1633 }
1634
1635 /* Nothing to do! */
1636 clear_bit(WRITE_PENDING, &con->state);
1637 dout("try_write nothing else to write.\n");
1638done:
1639 ret = 0;
1640out:
Sage Weil31b80062009-10-06 11:31:13 -07001641 dout("try_write done on %p\n", con);
1642 return ret;
1643}
1644
1645
1646
1647/*
1648 * Read what we can from the socket.
1649 */
1650static int try_read(struct ceph_connection *con)
1651{
Sage Weil31b80062009-10-06 11:31:13 -07001652 int ret = -1;
1653
1654 if (!con->sock)
1655 return 0;
1656
1657 if (test_bit(STANDBY, &con->state))
1658 return 0;
1659
1660 dout("try_read start on %p\n", con);
Sage Weilec302642009-12-22 10:43:42 -08001661
Sage Weil31b80062009-10-06 11:31:13 -07001662more:
1663 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
1664 con->in_base_pos);
1665 if (test_bit(CONNECTING, &con->state)) {
Sage Weileed0ef22009-11-10 14:34:36 -08001666 if (!test_bit(NEGOTIATING, &con->state)) {
1667 dout("try_read connecting\n");
1668 ret = read_partial_banner(con);
1669 if (ret <= 0)
1670 goto done;
1671 if (process_banner(con) < 0) {
1672 ret = -1;
1673 goto out;
1674 }
1675 }
Sage Weil31b80062009-10-06 11:31:13 -07001676 ret = read_partial_connect(con);
1677 if (ret <= 0)
1678 goto done;
1679 if (process_connect(con) < 0) {
1680 ret = -1;
1681 goto out;
1682 }
1683 goto more;
1684 }
1685
1686 if (con->in_base_pos < 0) {
1687 /*
1688 * skipping + discarding content.
1689 *
1690 * FIXME: there must be a better way to do this!
1691 */
1692 static char buf[1024];
1693 int skip = min(1024, -con->in_base_pos);
1694 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
1695 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
1696 if (ret <= 0)
1697 goto done;
1698 con->in_base_pos += ret;
1699 if (con->in_base_pos)
1700 goto more;
1701 }
1702 if (con->in_tag == CEPH_MSGR_TAG_READY) {
1703 /*
1704 * what's next?
1705 */
1706 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
1707 if (ret <= 0)
1708 goto done;
1709 dout("try_read got tag %d\n", (int)con->in_tag);
1710 switch (con->in_tag) {
1711 case CEPH_MSGR_TAG_MSG:
1712 prepare_read_message(con);
1713 break;
1714 case CEPH_MSGR_TAG_ACK:
1715 prepare_read_ack(con);
1716 break;
1717 case CEPH_MSGR_TAG_CLOSE:
1718 set_bit(CLOSED, &con->state); /* fixme */
1719 goto done;
1720 default:
1721 goto bad_tag;
1722 }
1723 }
1724 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
1725 ret = read_partial_message(con);
1726 if (ret <= 0) {
1727 switch (ret) {
1728 case -EBADMSG:
1729 con->error_msg = "bad crc";
1730 ret = -EIO;
1731 goto out;
1732 case -EIO:
1733 con->error_msg = "io error";
1734 goto out;
1735 default:
1736 goto done;
1737 }
1738 }
1739 if (con->in_tag == CEPH_MSGR_TAG_READY)
1740 goto more;
1741 process_message(con);
1742 goto more;
1743 }
1744 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
1745 ret = read_partial_ack(con);
1746 if (ret <= 0)
1747 goto done;
1748 process_ack(con);
1749 goto more;
1750 }
1751
1752done:
1753 ret = 0;
1754out:
1755 dout("try_read done on %p\n", con);
1756 return ret;
1757
1758bad_tag:
1759 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
1760 con->error_msg = "protocol error, garbage tag";
1761 ret = -1;
1762 goto out;
1763}
1764
1765
1766/*
1767 * Atomically queue work on a connection. Bump @con reference to
1768 * avoid races with connection teardown.
1769 *
1770 * There is some trickery going on with QUEUED and BUSY because we
1771 * only want a _single_ thread operating on each connection at any
1772 * point in time, but we want to use all available CPUs.
1773 *
1774 * The worker thread only proceeds if it can atomically set BUSY. It
1775 * clears QUEUED and does it's thing. When it thinks it's done, it
1776 * clears BUSY, then rechecks QUEUED.. if it's set again, it loops
1777 * (tries again to set BUSY).
1778 *
1779 * To queue work, we first set QUEUED, _then_ if BUSY isn't set, we
1780 * try to queue work. If that fails (work is already queued, or BUSY)
1781 * we give up (work also already being done or is queued) but leave QUEUED
1782 * set so that the worker thread will loop if necessary.
1783 */
1784static void queue_con(struct ceph_connection *con)
1785{
1786 if (test_bit(DEAD, &con->state)) {
1787 dout("queue_con %p ignoring: DEAD\n",
1788 con);
1789 return;
1790 }
1791
1792 if (!con->ops->get(con)) {
1793 dout("queue_con %p ref count 0\n", con);
1794 return;
1795 }
1796
1797 set_bit(QUEUED, &con->state);
1798 if (test_bit(BUSY, &con->state)) {
1799 dout("queue_con %p - already BUSY\n", con);
1800 con->ops->put(con);
1801 } else if (!queue_work(ceph_msgr_wq, &con->work.work)) {
1802 dout("queue_con %p - already queued\n", con);
1803 con->ops->put(con);
1804 } else {
1805 dout("queue_con %p\n", con);
1806 }
1807}
1808
1809/*
1810 * Do some work on a connection. Drop a connection ref when we're done.
1811 */
1812static void con_work(struct work_struct *work)
1813{
1814 struct ceph_connection *con = container_of(work, struct ceph_connection,
1815 work.work);
1816 int backoff = 0;
1817
1818more:
1819 if (test_and_set_bit(BUSY, &con->state) != 0) {
1820 dout("con_work %p BUSY already set\n", con);
1821 goto out;
1822 }
1823 dout("con_work %p start, clearing QUEUED\n", con);
1824 clear_bit(QUEUED, &con->state);
1825
Sage Weil9dd46582010-04-28 13:51:50 -07001826 mutex_lock(&con->mutex);
1827
Sage Weil31b80062009-10-06 11:31:13 -07001828 if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
1829 dout("con_work CLOSED\n");
1830 con_close_socket(con);
1831 goto done;
1832 }
1833 if (test_and_clear_bit(OPENING, &con->state)) {
1834 /* reopen w/ new peer */
1835 dout("con_work OPENING\n");
1836 con_close_socket(con);
1837 }
1838
1839 if (test_and_clear_bit(SOCK_CLOSED, &con->state) ||
1840 try_read(con) < 0 ||
1841 try_write(con) < 0) {
Sage Weil9dd46582010-04-28 13:51:50 -07001842 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001843 backoff = 1;
1844 ceph_fault(con); /* error/fault path */
Sage Weil9dd46582010-04-28 13:51:50 -07001845 goto done_unlocked;
Sage Weil31b80062009-10-06 11:31:13 -07001846 }
1847
1848done:
Sage Weil9dd46582010-04-28 13:51:50 -07001849 mutex_unlock(&con->mutex);
1850
1851done_unlocked:
Sage Weil31b80062009-10-06 11:31:13 -07001852 clear_bit(BUSY, &con->state);
1853 dout("con->state=%lu\n", con->state);
1854 if (test_bit(QUEUED, &con->state)) {
Sage Weile2663ab2010-02-16 22:01:03 -08001855 if (!backoff || test_bit(OPENING, &con->state)) {
Sage Weil31b80062009-10-06 11:31:13 -07001856 dout("con_work %p QUEUED reset, looping\n", con);
1857 goto more;
1858 }
1859 dout("con_work %p QUEUED reset, but just faulted\n", con);
1860 clear_bit(QUEUED, &con->state);
1861 }
1862 dout("con_work %p done\n", con);
1863
1864out:
1865 con->ops->put(con);
1866}
1867
1868
1869/*
1870 * Generic error/fault handler. A retry mechanism is used with
1871 * exponential backoff
1872 */
1873static void ceph_fault(struct ceph_connection *con)
1874{
1875 pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
1876 pr_addr(&con->peer_addr.in_addr), con->error_msg);
1877 dout("fault %p state %lu to peer %s\n",
1878 con, con->state, pr_addr(&con->peer_addr.in_addr));
1879
1880 if (test_bit(LOSSYTX, &con->state)) {
1881 dout("fault on LOSSYTX channel\n");
1882 goto out;
1883 }
1884
Sage Weilec302642009-12-22 10:43:42 -08001885 mutex_lock(&con->mutex);
Sage Weil91e45ce32010-02-15 12:05:09 -08001886 if (test_bit(CLOSED, &con->state))
1887 goto out_unlock;
Sage Weilec302642009-12-22 10:43:42 -08001888
Sage Weil31b80062009-10-06 11:31:13 -07001889 con_close_socket(con);
Sage Weil5e095e82009-12-14 14:30:34 -08001890
1891 if (con->in_msg) {
1892 ceph_msg_put(con->in_msg);
1893 con->in_msg = NULL;
1894 }
Sage Weil31b80062009-10-06 11:31:13 -07001895
Sage Weile80a52d2010-02-25 12:40:45 -08001896 /* Requeue anything that hasn't been acked */
1897 list_splice_init(&con->out_sent, &con->out_queue);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001898
Sage Weil31b80062009-10-06 11:31:13 -07001899 /* If there are no messages in the queue, place the connection
1900 * in a STANDBY state (i.e., don't try to reconnect just yet). */
Sage Weil31b80062009-10-06 11:31:13 -07001901 if (list_empty(&con->out_queue) && !con->out_keepalive_pending) {
1902 dout("fault setting STANDBY\n");
1903 set_bit(STANDBY, &con->state);
Sage Weile80a52d2010-02-25 12:40:45 -08001904 } else {
1905 /* retry after a delay. */
1906 if (con->delay == 0)
1907 con->delay = BASE_DELAY_INTERVAL;
1908 else if (con->delay < MAX_DELAY_INTERVAL)
1909 con->delay *= 2;
1910 dout("fault queueing %p delay %lu\n", con, con->delay);
1911 con->ops->get(con);
1912 if (queue_delayed_work(ceph_msgr_wq, &con->work,
1913 round_jiffies_relative(con->delay)) == 0)
1914 con->ops->put(con);
Sage Weil31b80062009-10-06 11:31:13 -07001915 }
1916
Sage Weil91e45ce32010-02-15 12:05:09 -08001917out_unlock:
1918 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001919out:
Sage Weil161fd652010-02-25 12:38:57 -08001920 /*
1921 * in case we faulted due to authentication, invalidate our
1922 * current tickets so that we can get new ones.
1923 */
1924 if (con->auth_retry && con->ops->invalidate_authorizer) {
1925 dout("calling invalidate_authorizer()\n");
1926 con->ops->invalidate_authorizer(con);
1927 }
1928
Sage Weil31b80062009-10-06 11:31:13 -07001929 if (con->ops->fault)
1930 con->ops->fault(con);
1931}
1932
1933
1934
1935/*
1936 * create a new messenger instance
1937 */
1938struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr)
1939{
1940 struct ceph_messenger *msgr;
1941
1942 msgr = kzalloc(sizeof(*msgr), GFP_KERNEL);
1943 if (msgr == NULL)
1944 return ERR_PTR(-ENOMEM);
1945
1946 spin_lock_init(&msgr->global_seq_lock);
1947
1948 /* the zero page is needed if a request is "canceled" while the message
1949 * is being written over the socket */
Yehuda Sadeh31459fe2010-03-17 13:54:02 -07001950 msgr->zero_page = __page_cache_alloc(GFP_KERNEL | __GFP_ZERO);
Sage Weil31b80062009-10-06 11:31:13 -07001951 if (!msgr->zero_page) {
1952 kfree(msgr);
1953 return ERR_PTR(-ENOMEM);
1954 }
1955 kmap(msgr->zero_page);
1956
1957 if (myaddr)
1958 msgr->inst.addr = *myaddr;
1959
1960 /* select a random nonce */
Sage Weilac8839d2010-01-27 14:28:10 -08001961 msgr->inst.addr.type = 0;
Sage Weil103e2d32010-01-07 16:12:36 -08001962 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
Sage Weil63f2d212009-11-03 15:17:56 -08001963 encode_my_addr(msgr);
Sage Weil31b80062009-10-06 11:31:13 -07001964
1965 dout("messenger_create %p\n", msgr);
1966 return msgr;
1967}
1968
1969void ceph_messenger_destroy(struct ceph_messenger *msgr)
1970{
1971 dout("destroy %p\n", msgr);
1972 kunmap(msgr->zero_page);
1973 __free_page(msgr->zero_page);
1974 kfree(msgr);
1975 dout("destroyed messenger %p\n", msgr);
1976}
1977
1978/*
1979 * Queue up an outgoing message on the given connection.
1980 */
1981void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
1982{
1983 if (test_bit(CLOSED, &con->state)) {
1984 dout("con_send %p closed, dropping %p\n", con, msg);
1985 ceph_msg_put(msg);
1986 return;
1987 }
1988
1989 /* set src+dst */
Sage Weildbad1852010-03-25 15:45:38 -07001990 msg->hdr.src = con->msgr->inst.name;
Sage Weil31b80062009-10-06 11:31:13 -07001991
Sage Weil3ca02ef2010-03-01 15:25:00 -08001992 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
1993
Sage Weile84346b2010-05-11 21:20:38 -07001994 msg->needs_out_seq = true;
1995
Sage Weil31b80062009-10-06 11:31:13 -07001996 /* queue */
Sage Weilec302642009-12-22 10:43:42 -08001997 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001998 BUG_ON(!list_empty(&msg->list_head));
1999 list_add_tail(&msg->list_head, &con->out_queue);
2000 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2001 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2002 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2003 le32_to_cpu(msg->hdr.front_len),
2004 le32_to_cpu(msg->hdr.middle_len),
2005 le32_to_cpu(msg->hdr.data_len));
Sage Weilec302642009-12-22 10:43:42 -08002006 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002007
2008 /* if there wasn't anything waiting to send before, queue
2009 * new work */
2010 if (test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2011 queue_con(con);
2012}
2013
2014/*
2015 * Revoke a message that was previously queued for send
2016 */
2017void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg)
2018{
Sage Weilec302642009-12-22 10:43:42 -08002019 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002020 if (!list_empty(&msg->list_head)) {
Sage Weiled98ada2010-07-05 12:15:14 -07002021 dout("con_revoke %p msg %p - was on queue\n", con, msg);
Sage Weil31b80062009-10-06 11:31:13 -07002022 list_del_init(&msg->list_head);
2023 ceph_msg_put(msg);
2024 msg->hdr.seq = 0;
Sage Weiled98ada2010-07-05 12:15:14 -07002025 }
2026 if (con->out_msg == msg) {
2027 dout("con_revoke %p msg %p - was sending\n", con, msg);
2028 con->out_msg = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002029 if (con->out_kvec_is_msg) {
2030 con->out_skip = con->out_kvec_bytes;
2031 con->out_kvec_is_msg = false;
2032 }
Sage Weiled98ada2010-07-05 12:15:14 -07002033 ceph_msg_put(msg);
2034 msg->hdr.seq = 0;
Sage Weil31b80062009-10-06 11:31:13 -07002035 }
Sage Weilec302642009-12-22 10:43:42 -08002036 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002037}
2038
2039/*
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002040 * Revoke a message that we may be reading data into
Sage Weil350b1c32009-12-22 10:45:45 -08002041 */
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002042void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg)
Sage Weil350b1c32009-12-22 10:45:45 -08002043{
2044 mutex_lock(&con->mutex);
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002045 if (con->in_msg && con->in_msg == msg) {
2046 unsigned front_len = le32_to_cpu(con->in_hdr.front_len);
2047 unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len);
Sage Weil350b1c32009-12-22 10:45:45 -08002048 unsigned data_len = le32_to_cpu(con->in_hdr.data_len);
2049
2050 /* skip rest of message */
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002051 dout("con_revoke_pages %p msg %p revoked\n", con, msg);
Sage Weil350b1c32009-12-22 10:45:45 -08002052 con->in_base_pos = con->in_base_pos -
2053 sizeof(struct ceph_msg_header) -
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002054 front_len -
2055 middle_len -
2056 data_len -
Sage Weil350b1c32009-12-22 10:45:45 -08002057 sizeof(struct ceph_msg_footer);
Sage Weil350b1c32009-12-22 10:45:45 -08002058 ceph_msg_put(con->in_msg);
2059 con->in_msg = NULL;
2060 con->in_tag = CEPH_MSGR_TAG_READY;
Sage Weil684be252010-04-21 20:45:59 -07002061 con->in_seq++;
Sage Weil350b1c32009-12-22 10:45:45 -08002062 } else {
2063 dout("con_revoke_pages %p msg %p pages %p no-op\n",
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002064 con, con->in_msg, msg);
Sage Weil350b1c32009-12-22 10:45:45 -08002065 }
2066 mutex_unlock(&con->mutex);
2067}
2068
2069/*
Sage Weil31b80062009-10-06 11:31:13 -07002070 * Queue a keepalive byte to ensure the tcp connection is alive.
2071 */
2072void ceph_con_keepalive(struct ceph_connection *con)
2073{
2074 if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 &&
2075 test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2076 queue_con(con);
2077}
2078
2079
2080/*
2081 * construct a new message with given type, size
2082 * the new msg has a ref count of 1.
2083 */
Yehuda Sadeh34d23762010-04-06 14:33:58 -07002084struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags)
Sage Weil31b80062009-10-06 11:31:13 -07002085{
2086 struct ceph_msg *m;
2087
Yehuda Sadeh34d23762010-04-06 14:33:58 -07002088 m = kmalloc(sizeof(*m), flags);
Sage Weil31b80062009-10-06 11:31:13 -07002089 if (m == NULL)
2090 goto out;
Sage Weilc2e552e2009-12-07 15:55:05 -08002091 kref_init(&m->kref);
Sage Weil31b80062009-10-06 11:31:13 -07002092 INIT_LIST_HEAD(&m->list_head);
2093
Sage Weil45c6ceb2010-05-11 15:01:51 -07002094 m->hdr.tid = 0;
Sage Weil31b80062009-10-06 11:31:13 -07002095 m->hdr.type = cpu_to_le16(type);
Sage Weil45c6ceb2010-05-11 15:01:51 -07002096 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2097 m->hdr.version = 0;
Sage Weil31b80062009-10-06 11:31:13 -07002098 m->hdr.front_len = cpu_to_le32(front_len);
2099 m->hdr.middle_len = 0;
Sage Weilbb257662010-04-01 16:07:23 -07002100 m->hdr.data_len = 0;
2101 m->hdr.data_off = 0;
Sage Weil45c6ceb2010-05-11 15:01:51 -07002102 m->hdr.reserved = 0;
Sage Weil31b80062009-10-06 11:31:13 -07002103 m->footer.front_crc = 0;
2104 m->footer.middle_crc = 0;
2105 m->footer.data_crc = 0;
Sage Weil45c6ceb2010-05-11 15:01:51 -07002106 m->footer.flags = 0;
Sage Weil31b80062009-10-06 11:31:13 -07002107 m->front_max = front_len;
2108 m->front_is_vmalloc = false;
2109 m->more_to_follow = false;
2110 m->pool = NULL;
2111
2112 /* front */
2113 if (front_len) {
2114 if (front_len > PAGE_CACHE_SIZE) {
Yehuda Sadeh34d23762010-04-06 14:33:58 -07002115 m->front.iov_base = __vmalloc(front_len, flags,
Sage Weil31b80062009-10-06 11:31:13 -07002116 PAGE_KERNEL);
2117 m->front_is_vmalloc = true;
2118 } else {
Yehuda Sadeh34d23762010-04-06 14:33:58 -07002119 m->front.iov_base = kmalloc(front_len, flags);
Sage Weil31b80062009-10-06 11:31:13 -07002120 }
2121 if (m->front.iov_base == NULL) {
2122 pr_err("msg_new can't allocate %d bytes\n",
2123 front_len);
2124 goto out2;
2125 }
2126 } else {
2127 m->front.iov_base = NULL;
2128 }
2129 m->front.iov_len = front_len;
2130
2131 /* middle */
2132 m->middle = NULL;
2133
2134 /* data */
Sage Weilbb257662010-04-01 16:07:23 -07002135 m->nr_pages = 0;
2136 m->pages = NULL;
Sage Weil58bb3b32009-12-23 12:12:31 -08002137 m->pagelist = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002138
Sage Weilbb257662010-04-01 16:07:23 -07002139 dout("ceph_msg_new %p front %d\n", m, front_len);
Sage Weil31b80062009-10-06 11:31:13 -07002140 return m;
2141
2142out2:
2143 ceph_msg_put(m);
2144out:
Sage Weilbb257662010-04-01 16:07:23 -07002145 pr_err("msg_new can't create type %d front %d\n", type, front_len);
Sage Weila79832f2010-04-01 16:06:19 -07002146 return NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002147}
2148
2149/*
Sage Weil31b80062009-10-06 11:31:13 -07002150 * Allocate "middle" portion of a message, if it is needed and wasn't
2151 * allocated by alloc_msg. This allows us to read a small fixed-size
2152 * per-type header in the front and then gracefully fail (i.e.,
2153 * propagate the error to the caller based on info in the front) when
2154 * the middle is too large.
2155 */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002156static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
Sage Weil31b80062009-10-06 11:31:13 -07002157{
2158 int type = le16_to_cpu(msg->hdr.type);
2159 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2160
2161 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2162 ceph_msg_type_name(type), middle_len);
2163 BUG_ON(!middle_len);
2164 BUG_ON(msg->middle);
2165
Sage Weilb6c1d5b2009-12-07 12:17:17 -08002166 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
Sage Weil31b80062009-10-06 11:31:13 -07002167 if (!msg->middle)
2168 return -ENOMEM;
2169 return 0;
2170}
2171
Yehuda Sadeh24504182010-01-08 13:58:34 -08002172/*
2173 * Generic message allocator, for incoming messages.
2174 */
2175static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
2176 struct ceph_msg_header *hdr,
2177 int *skip)
2178{
2179 int type = le16_to_cpu(hdr->type);
2180 int front_len = le32_to_cpu(hdr->front_len);
2181 int middle_len = le32_to_cpu(hdr->middle_len);
2182 struct ceph_msg *msg = NULL;
2183 int ret;
2184
2185 if (con->ops->alloc_msg) {
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002186 mutex_unlock(&con->mutex);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002187 msg = con->ops->alloc_msg(con, hdr, skip);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002188 mutex_lock(&con->mutex);
Sage Weila79832f2010-04-01 16:06:19 -07002189 if (!msg || *skip)
Yehuda Sadeh24504182010-01-08 13:58:34 -08002190 return NULL;
2191 }
2192 if (!msg) {
2193 *skip = 0;
Yehuda Sadeh34d23762010-04-06 14:33:58 -07002194 msg = ceph_msg_new(type, front_len, GFP_NOFS);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002195 if (!msg) {
2196 pr_err("unable to allocate msg type %d len %d\n",
2197 type, front_len);
Sage Weila79832f2010-04-01 16:06:19 -07002198 return NULL;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002199 }
2200 }
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08002201 memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
Yehuda Sadeh24504182010-01-08 13:58:34 -08002202
Sage Weilbb257662010-04-01 16:07:23 -07002203 if (middle_len && !msg->middle) {
Yehuda Sadeh24504182010-01-08 13:58:34 -08002204 ret = ceph_alloc_middle(con, msg);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002205 if (ret < 0) {
2206 ceph_msg_put(msg);
Sage Weila79832f2010-04-01 16:06:19 -07002207 return NULL;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002208 }
2209 }
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08002210
Yehuda Sadeh24504182010-01-08 13:58:34 -08002211 return msg;
2212}
2213
Sage Weil31b80062009-10-06 11:31:13 -07002214
2215/*
2216 * Free a generically kmalloc'd message.
2217 */
2218void ceph_msg_kfree(struct ceph_msg *m)
2219{
2220 dout("msg_kfree %p\n", m);
2221 if (m->front_is_vmalloc)
2222 vfree(m->front.iov_base);
2223 else
2224 kfree(m->front.iov_base);
2225 kfree(m);
2226}
2227
2228/*
2229 * Drop a msg ref. Destroy as needed.
2230 */
Sage Weilc2e552e2009-12-07 15:55:05 -08002231void ceph_msg_last_put(struct kref *kref)
Sage Weil31b80062009-10-06 11:31:13 -07002232{
Sage Weilc2e552e2009-12-07 15:55:05 -08002233 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
Sage Weil31b80062009-10-06 11:31:13 -07002234
Sage Weilc2e552e2009-12-07 15:55:05 -08002235 dout("ceph_msg_put last one on %p\n", m);
2236 WARN_ON(!list_empty(&m->list_head));
Sage Weil31b80062009-10-06 11:31:13 -07002237
Sage Weilc2e552e2009-12-07 15:55:05 -08002238 /* drop middle, data, if any */
2239 if (m->middle) {
2240 ceph_buffer_put(m->middle);
2241 m->middle = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002242 }
Sage Weilc2e552e2009-12-07 15:55:05 -08002243 m->nr_pages = 0;
2244 m->pages = NULL;
2245
Sage Weil58bb3b32009-12-23 12:12:31 -08002246 if (m->pagelist) {
2247 ceph_pagelist_release(m->pagelist);
2248 kfree(m->pagelist);
2249 m->pagelist = NULL;
2250 }
2251
Sage Weilc2e552e2009-12-07 15:55:05 -08002252 if (m->pool)
2253 ceph_msgpool_put(m->pool, m);
2254 else
2255 ceph_msg_kfree(m);
Sage Weil31b80062009-10-06 11:31:13 -07002256}
Sage Weil9ec7cab2009-12-14 15:13:47 -08002257
2258void ceph_msg_dump(struct ceph_msg *msg)
2259{
2260 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2261 msg->front_max, msg->nr_pages);
2262 print_hex_dump(KERN_DEBUG, "header: ",
2263 DUMP_PREFIX_OFFSET, 16, 1,
2264 &msg->hdr, sizeof(msg->hdr), true);
2265 print_hex_dump(KERN_DEBUG, " front: ",
2266 DUMP_PREFIX_OFFSET, 16, 1,
2267 msg->front.iov_base, msg->front.iov_len, true);
2268 if (msg->middle)
2269 print_hex_dump(KERN_DEBUG, "middle: ",
2270 DUMP_PREFIX_OFFSET, 16, 1,
2271 msg->middle->vec.iov_base,
2272 msg->middle->vec.iov_len, true);
2273 print_hex_dump(KERN_DEBUG, "footer: ",
2274 DUMP_PREFIX_OFFSET, 16, 1,
2275 &msg->footer, sizeof(msg->footer), true);
2276}