projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: use kref for ceph_osd_request
[deliverable/linux.git] / fs / ceph / messenger.c
CommitLineData
31b8006e
SW		 1#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>
9#include <linux/socket.h>
10#include <linux/string.h>
11#include <net/tcp.h>
12
13#include "super.h"
14#include "messenger.h"
63f2d211 15#include "decode.h"
31b8006e
SW		 16
17/*
18 * Ceph uses the messenger to exchange ceph_msg messages with other
19 * hosts in the system. The messenger provides ordered and reliable
20 * delivery. We tolerate TCP disconnects by reconnecting (with
21 * exponential backoff) in the case of a fault (disconnection, bad
22 * crc, protocol error). Acks allow sent messages to be discarded by
23 * the sender.
24 */
25
26/* static tag bytes (protocol control messages) */
27static char tag_msg = CEPH_MSGR_TAG_MSG;
28static char tag_ack = CEPH_MSGR_TAG_ACK;
29static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
30
31
32static void queue_con(struct ceph_connection *con);
33static void con_work(struct work_struct *);
34static void ceph_fault(struct ceph_connection *con);
35
36const char *ceph_name_type_str(int t)
37{
38 switch (t) {
39 case CEPH_ENTITY_TYPE_MON: return "mon";
40 case CEPH_ENTITY_TYPE_MDS: return "mds";
41 case CEPH_ENTITY_TYPE_OSD: return "osd";
42 case CEPH_ENTITY_TYPE_CLIENT: return "client";
43 case CEPH_ENTITY_TYPE_ADMIN: return "admin";
44 default: return "???";
45 }
46}
47
48/*
49 * nicely render a sockaddr as a string.
50 */
51#define MAX_ADDR_STR 20
52static char addr_str[MAX_ADDR_STR][40];
53static DEFINE_SPINLOCK(addr_str_lock);
54static int last_addr_str;
55
56const char *pr_addr(const struct sockaddr_storage *ss)
57{
58 int i;
59 char *s;
60 struct sockaddr_in *in4 = (void *)ss;
61 unsigned char *quad = (void *)&in4->sin_addr.s_addr;
62 struct sockaddr_in6 *in6 = (void *)ss;
63
64 spin_lock(&addr_str_lock);
65 i = last_addr_str++;
66 if (last_addr_str == MAX_ADDR_STR)
67 last_addr_str = 0;
68 spin_unlock(&addr_str_lock);
69 s = addr_str[i];
70
71 switch (ss->ss_family) {
72 case AF_INET:
73 sprintf(s, "%u.%u.%u.%u:%u",
74 (unsigned int)quad[0],
75 (unsigned int)quad[1],
76 (unsigned int)quad[2],
77 (unsigned int)quad[3],
78 (unsigned int)ntohs(in4->sin_port));
79 break;
80
81 case AF_INET6:
82 sprintf(s, "%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x:%u",
83 in6->sin6_addr.s6_addr16[0],
84 in6->sin6_addr.s6_addr16[1],
85 in6->sin6_addr.s6_addr16[2],
86 in6->sin6_addr.s6_addr16[3],
87 in6->sin6_addr.s6_addr16[4],
88 in6->sin6_addr.s6_addr16[5],
89 in6->sin6_addr.s6_addr16[6],
90 in6->sin6_addr.s6_addr16[7],
91 (unsigned int)ntohs(in6->sin6_port));
92 break;
93
94 default:
95 sprintf(s, "(unknown sockaddr family %d)", (int)ss->ss_family);
96 }
97
98 return s;
99}
100
63f2d211
SW		 101static void encode_my_addr(struct ceph_messenger *msgr)
102{
103 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
104 ceph_encode_addr(&msgr->my_enc_addr);
105}
106
31b8006e
SW		 107/*
108 * work queue for all reading and writing to/from the socket.
109 */
110struct workqueue_struct *ceph_msgr_wq;
111
112int __init ceph_msgr_init(void)
113{
114 ceph_msgr_wq = create_workqueue("ceph-msgr");
115 if (IS_ERR(ceph_msgr_wq)) {
116 int ret = PTR_ERR(ceph_msgr_wq);
117 pr_err("msgr_init failed to create workqueue: %d\n", ret);
118 ceph_msgr_wq = NULL;
119 return ret;
120 }
121 return 0;
122}
123
124void ceph_msgr_exit(void)
125{
126 destroy_workqueue(ceph_msgr_wq);
127}
128
129/*
130 * socket callback functions
131 */
132
133/* data available on socket, or listen socket received a connect */
134static void ceph_data_ready(struct sock *sk, int count_unused)
135{
136 struct ceph_connection *con =
137 (struct ceph_connection *)sk->sk_user_data;
138 if (sk->sk_state != TCP_CLOSE_WAIT) {
139 dout("ceph_data_ready on %p state = %lu, queueing work\n",
140 con, con->state);
141 queue_con(con);
142 }
143}
144
145/* socket has buffer space for writing */
146static void ceph_write_space(struct sock *sk)
147{
148 struct ceph_connection *con =
149 (struct ceph_connection *)sk->sk_user_data;
150
151 /* only queue to workqueue if there is data we want to write. */
152 if (test_bit(WRITE_PENDING, &con->state)) {
153 dout("ceph_write_space %p queueing write work\n", con);
154 queue_con(con);
155 } else {
156 dout("ceph_write_space %p nothing to write\n", con);
157 }
158
159 /* since we have our own write_space, clear the SOCK_NOSPACE flag */
160 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
161}
162
163/* socket's state has changed */
164static void ceph_state_change(struct sock *sk)
165{
166 struct ceph_connection *con =
167 (struct ceph_connection *)sk->sk_user_data;
168
169 dout("ceph_state_change %p state = %lu sk_state = %u\n",
170 con, con->state, sk->sk_state);
171
172 if (test_bit(CLOSED, &con->state))
173 return;
174
175 switch (sk->sk_state) {
176 case TCP_CLOSE:
177 dout("ceph_state_change TCP_CLOSE\n");
178 case TCP_CLOSE_WAIT:
179 dout("ceph_state_change TCP_CLOSE_WAIT\n");
180 if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) {
181 if (test_bit(CONNECTING, &con->state))
182 con->error_msg = "connection failed";
183 else
184 con->error_msg = "socket closed";
185 queue_con(con);
186 }
187 break;
188 case TCP_ESTABLISHED:
189 dout("ceph_state_change TCP_ESTABLISHED\n");
190 queue_con(con);
191 break;
192 }
193}
194
195/*
196 * set up socket callbacks
197 */
198static void set_sock_callbacks(struct socket *sock,
199 struct ceph_connection *con)
200{
201 struct sock *sk = sock->sk;
202 sk->sk_user_data = (void *)con;
203 sk->sk_data_ready = ceph_data_ready;
204 sk->sk_write_space = ceph_write_space;
205 sk->sk_state_change = ceph_state_change;
206}
207
208
209/*
210 * socket helpers
211 */
212
213/*
214 * initiate connection to a remote socket.
215 */
216static struct socket *ceph_tcp_connect(struct ceph_connection *con)
217{
218 struct sockaddr *paddr = (struct sockaddr *)&con->peer_addr.in_addr;
219 struct socket *sock;
220 int ret;
221
222 BUG_ON(con->sock);
223 ret = sock_create_kern(AF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
224 if (ret)
225 return ERR_PTR(ret);
226 con->sock = sock;
227 sock->sk->sk_allocation = GFP_NOFS;
228
229 set_sock_callbacks(sock, con);
230
231 dout("connect %s\n", pr_addr(&con->peer_addr.in_addr));
232
233 ret = sock->ops->connect(sock, paddr, sizeof(*paddr), O_NONBLOCK);
234 if (ret == -EINPROGRESS) {
235 dout("connect %s EINPROGRESS sk_state = %u\n",
236 pr_addr(&con->peer_addr.in_addr),
237 sock->sk->sk_state);
238 ret = 0;
239 }
240 if (ret < 0) {
241 pr_err("connect %s error %d\n",
242 pr_addr(&con->peer_addr.in_addr), ret);
243 sock_release(sock);
244 con->sock = NULL;
245 con->error_msg = "connect error";
246 }
247
248 if (ret < 0)
249 return ERR_PTR(ret);
250 return sock;
251}
252
253static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
254{
255 struct kvec iov = {buf, len};
256 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
257
258 return kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
259}
260
261/*
262 * write something. @more is true if caller will be sending more data
263 * shortly.
264 */
265static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
266 size_t kvlen, size_t len, int more)
267{
268 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
269
270 if (more)
271 msg.msg_flags |= MSG_MORE;
272 else
273 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
274
275 return kernel_sendmsg(sock, &msg, iov, kvlen, len);
276}
277
278
279/*
280 * Shutdown/close the socket for the given connection.
281 */
282static int con_close_socket(struct ceph_connection *con)
283{
284 int rc;
285
286 dout("con_close_socket on %p sock %p\n", con, con->sock);
287 if (!con->sock)
288 return 0;
289 set_bit(SOCK_CLOSED, &con->state);
290 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
291 sock_release(con->sock);
292 con->sock = NULL;
293 clear_bit(SOCK_CLOSED, &con->state);
294 return rc;
295}
296
297/*
298 * Reset a connection. Discard all incoming and outgoing messages
299 * and clear *_seq state.
300 */
301static void ceph_msg_remove(struct ceph_msg *msg)
302{
303 list_del_init(&msg->list_head);
304 ceph_msg_put(msg);
305}
306static void ceph_msg_remove_list(struct list_head *head)
307{
308 while (!list_empty(head)) {
309 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
310 list_head);
311 ceph_msg_remove(msg);
312 }
313}
314
315static void reset_connection(struct ceph_connection *con)
316{
317 /* reset connection, out_queue, msg_ and connect_seq */
318 /* discard existing out_queue and msg_seq */
319 mutex_lock(&con->out_mutex);
320 ceph_msg_remove_list(&con->out_queue);
321 ceph_msg_remove_list(&con->out_sent);
322
323 con->connect_seq = 0;
324 con->out_seq = 0;
325 con->out_msg = NULL;
326 con->in_seq = 0;
327 mutex_unlock(&con->out_mutex);
328}
329
330/*
331 * mark a peer down. drop any open connections.
332 */
333void ceph_con_close(struct ceph_connection *con)
334{
335 dout("con_close %p peer %s\n", con, pr_addr(&con->peer_addr.in_addr));
336 set_bit(CLOSED, &con->state); /* in case there's queued work */
337 clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */
338 reset_connection(con);
339 queue_con(con);
340}
341
31b8006e
SW		 342/*
343 * Reopen a closed connection, with a new peer address.
344 */
345void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr)
346{
347 dout("con_open %p %s\n", con, pr_addr(&addr->in_addr));
348 set_bit(OPENING, &con->state);
349 clear_bit(CLOSED, &con->state);
350 memcpy(&con->peer_addr, addr, sizeof(*addr));
03c677e1 351 con->delay = 0; /* reset backoff memory */
31b8006e
SW		 352 queue_con(con);
353}
354
355/*
356 * generic get/put
357 */
358struct ceph_connection *ceph_con_get(struct ceph_connection *con)
359{
360 dout("con_get %p nref = %d -> %d\n", con,
361 atomic_read(&con->nref), atomic_read(&con->nref) + 1);
362 if (atomic_inc_not_zero(&con->nref))
363 return con;
364 return NULL;
365}
366
367void ceph_con_put(struct ceph_connection *con)
368{
369 dout("con_put %p nref = %d -> %d\n", con,
370 atomic_read(&con->nref), atomic_read(&con->nref) - 1);
371 BUG_ON(atomic_read(&con->nref) == 0);
372 if (atomic_dec_and_test(&con->nref)) {
71ececda 373 BUG_ON(con->sock);
31b8006e
SW		 374 kfree(con);
375 }
376}
377
378/*
379 * initialize a new connection.
380 */
381void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con)
382{
383 dout("con_init %p\n", con);
384 memset(con, 0, sizeof(*con));
385 atomic_set(&con->nref, 1);
386 con->msgr = msgr;
387 mutex_init(&con->out_mutex);
388 INIT_LIST_HEAD(&con->out_queue);
389 INIT_LIST_HEAD(&con->out_sent);
390 INIT_DELAYED_WORK(&con->work, con_work);
391}
392
393
394/*
395 * We maintain a global counter to order connection attempts. Get
396 * a unique seq greater than @gt.
397 */
398static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
399{
400 u32 ret;
401
402 spin_lock(&msgr->global_seq_lock);
403 if (msgr->global_seq < gt)
404 msgr->global_seq = gt;
405 ret = ++msgr->global_seq;
406 spin_unlock(&msgr->global_seq_lock);
407 return ret;
408}
409
410
411/*
412 * Prepare footer for currently outgoing message, and finish things
413 * off. Assumes out_kvec* are already valid.. we just add on to the end.
414 */
415static void prepare_write_message_footer(struct ceph_connection *con, int v)
416{
417 struct ceph_msg *m = con->out_msg;
418
419 dout("prepare_write_message_footer %p\n", con);
420 con->out_kvec_is_msg = true;
421 con->out_kvec[v].iov_base = &m->footer;
422 con->out_kvec[v].iov_len = sizeof(m->footer);
423 con->out_kvec_bytes += sizeof(m->footer);
424 con->out_kvec_left++;
425 con->out_more = m->more_to_follow;
426 con->out_msg = NULL; /* we're done with this one */
427}
428
429/*
430 * Prepare headers for the next outgoing message.
431 */
432static void prepare_write_message(struct ceph_connection *con)
433{
434 struct ceph_msg *m;
435 int v = 0;
436
437 con->out_kvec_bytes = 0;
438 con->out_kvec_is_msg = true;
439
440 /* Sneak an ack in there first? If we can get it into the same
441 * TCP packet that's a good thing. */
442 if (con->in_seq > con->in_seq_acked) {
443 con->in_seq_acked = con->in_seq;
444 con->out_kvec[v].iov_base = &tag_ack;
445 con->out_kvec[v++].iov_len = 1;
446 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
447 con->out_kvec[v].iov_base = &con->out_temp_ack;
448 con->out_kvec[v++].iov_len = sizeof(con->out_temp_ack);
449 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
450 }
451
452 /* move message to sending/sent list */
453 m = list_first_entry(&con->out_queue,
454 struct ceph_msg, list_head);
455 list_move_tail(&m->list_head, &con->out_sent);
456 con->out_msg = m; /* we don't bother taking a reference here. */
457
458 m->hdr.seq = cpu_to_le64(++con->out_seq);
459
460 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
461 m, con->out_seq, le16_to_cpu(m->hdr.type),
462 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
463 le32_to_cpu(m->hdr.data_len),
464 m->nr_pages);
465 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
466
467 /* tag + hdr + front + middle */
468 con->out_kvec[v].iov_base = &tag_msg;
469 con->out_kvec[v++].iov_len = 1;
470 con->out_kvec[v].iov_base = &m->hdr;
471 con->out_kvec[v++].iov_len = sizeof(m->hdr);
472 con->out_kvec[v++] = m->front;
473 if (m->middle)
474 con->out_kvec[v++] = m->middle->vec;
475 con->out_kvec_left = v;
476 con->out_kvec_bytes += 1 + sizeof(m->hdr) + m->front.iov_len +
477 (m->middle ? m->middle->vec.iov_len : 0);
478 con->out_kvec_cur = con->out_kvec;
479
480 /* fill in crc (except data pages), footer */
481 con->out_msg->hdr.crc =
482 cpu_to_le32(crc32c(0, (void *)&m->hdr,
483 sizeof(m->hdr) - sizeof(m->hdr.crc)));
484 con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE;
485 con->out_msg->footer.front_crc =
486 cpu_to_le32(crc32c(0, m->front.iov_base, m->front.iov_len));
487 if (m->middle)
488 con->out_msg->footer.middle_crc =
489 cpu_to_le32(crc32c(0, m->middle->vec.iov_base,
490 m->middle->vec.iov_len));
491 else
492 con->out_msg->footer.middle_crc = 0;
493 con->out_msg->footer.data_crc = 0;
494 dout("prepare_write_message front_crc %u data_crc %u\n",
495 le32_to_cpu(con->out_msg->footer.front_crc),
496 le32_to_cpu(con->out_msg->footer.middle_crc));
497
498 /* is there a data payload? */
499 if (le32_to_cpu(m->hdr.data_len) > 0) {
500 /* initialize page iterator */
501 con->out_msg_pos.page = 0;
502 con->out_msg_pos.page_pos =
503 le16_to_cpu(m->hdr.data_off) & ~PAGE_MASK;
504 con->out_msg_pos.data_pos = 0;
505 con->out_msg_pos.did_page_crc = 0;
506 con->out_more = 1; /* data + footer will follow */
507 } else {
508 /* no, queue up footer too and be done */
509 prepare_write_message_footer(con, v);
510 }
511
512 set_bit(WRITE_PENDING, &con->state);
513}
514
515/*
516 * Prepare an ack.
517 */
518static void prepare_write_ack(struct ceph_connection *con)
519{
520 dout("prepare_write_ack %p %llu -> %llu\n", con,
521 con->in_seq_acked, con->in_seq);
522 con->in_seq_acked = con->in_seq;
523
524 con->out_kvec[0].iov_base = &tag_ack;
525 con->out_kvec[0].iov_len = 1;
526 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
527 con->out_kvec[1].iov_base = &con->out_temp_ack;
528 con->out_kvec[1].iov_len = sizeof(con->out_temp_ack);
529 con->out_kvec_left = 2;
530 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
531 con->out_kvec_cur = con->out_kvec;
532 con->out_more = 1; /* more will follow.. eventually.. */
533 set_bit(WRITE_PENDING, &con->state);
534}
535
536/*
537 * Prepare to write keepalive byte.
538 */
539static void prepare_write_keepalive(struct ceph_connection *con)
540{
541 dout("prepare_write_keepalive %p\n", con);
542 con->out_kvec[0].iov_base = &tag_keepalive;
543 con->out_kvec[0].iov_len = 1;
544 con->out_kvec_left = 1;
545 con->out_kvec_bytes = 1;
546 con->out_kvec_cur = con->out_kvec;
547 set_bit(WRITE_PENDING, &con->state);
548}
549
550/*
551 * Connection negotiation.
552 */
553
4e7a5dcd
SW		 554static void prepare_connect_authorizer(struct ceph_connection *con)
555{
556 void *auth_buf;
557 int auth_len = 0;
558 int auth_protocol = 0;
559
560 if (con->ops->get_authorizer)
561 con->ops->get_authorizer(con, &auth_buf, &auth_len,
562 &auth_protocol, &con->auth_reply_buf,
563 &con->auth_reply_buf_len,
564 con->auth_retry);
565
566 con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol);
567 con->out_connect.authorizer_len = cpu_to_le32(auth_len);
568
569 con->out_kvec[con->out_kvec_left].iov_base = auth_buf;
570 con->out_kvec[con->out_kvec_left].iov_len = auth_len;
571 con->out_kvec_left++;
572 con->out_kvec_bytes += auth_len;
573}
574
31b8006e
SW		 575/*
576 * We connected to a peer and are saying hello.
577 */
eed0ef2c
SW		 578static void prepare_write_banner(struct ceph_messenger *msgr,
579 struct ceph_connection *con)
31b8006e
SW		 580{
581 int len = strlen(CEPH_BANNER);
eed0ef2c
SW		 582
583 con->out_kvec[0].iov_base = CEPH_BANNER;
584 con->out_kvec[0].iov_len = len;
585 con->out_kvec[1].iov_base = &msgr->my_enc_addr;
586 con->out_kvec[1].iov_len = sizeof(msgr->my_enc_addr);
587 con->out_kvec_left = 2;
588 con->out_kvec_bytes = len + sizeof(msgr->my_enc_addr);
589 con->out_kvec_cur = con->out_kvec;
590 con->out_more = 0;
591 set_bit(WRITE_PENDING, &con->state);
592}
593
594static void prepare_write_connect(struct ceph_messenger *msgr,
595 struct ceph_connection *con,
596 int after_banner)
597{
31b8006e
SW		 598 unsigned global_seq = get_global_seq(con->msgr, 0);
599 int proto;
600
601 switch (con->peer_name.type) {
602 case CEPH_ENTITY_TYPE_MON:
603 proto = CEPH_MONC_PROTOCOL;
604 break;
605 case CEPH_ENTITY_TYPE_OSD:
606 proto = CEPH_OSDC_PROTOCOL;
607 break;
608 case CEPH_ENTITY_TYPE_MDS:
609 proto = CEPH_MDSC_PROTOCOL;
610 break;
611 default:
612 BUG();
613 }
614
615 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
616 con->connect_seq, global_seq, proto);
4e7a5dcd 617
31b8006e
SW		 618 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
619 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
620 con->out_connect.global_seq = cpu_to_le32(global_seq);
621 con->out_connect.protocol_version = cpu_to_le32(proto);
622 con->out_connect.flags = 0;
623 if (test_bit(LOSSYTX, &con->state))
624 con->out_connect.flags = CEPH_MSG_CONNECT_LOSSY;
625
eed0ef2c
SW		 626 if (!after_banner) {
627 con->out_kvec_left = 0;
628 con->out_kvec_bytes = 0;
629 }
630 con->out_kvec[con->out_kvec_left].iov_base = &con->out_connect;
631 con->out_kvec[con->out_kvec_left].iov_len = sizeof(con->out_connect);
632 con->out_kvec_left++;
633 con->out_kvec_bytes += sizeof(con->out_connect);
31b8006e
SW		 634 con->out_kvec_cur = con->out_kvec;
635 con->out_more = 0;
636 set_bit(WRITE_PENDING, &con->state);
4e7a5dcd
SW		 637
638 prepare_connect_authorizer(con);
31b8006e
SW		 639}
640
641
642/*
643 * write as much of pending kvecs to the socket as we can.
644 * 1 -> done
645 * 0 -> socket full, but more to do
646 * <0 -> error
647 */
648static int write_partial_kvec(struct ceph_connection *con)
649{
650 int ret;
651
652 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
653 while (con->out_kvec_bytes > 0) {
654 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
655 con->out_kvec_left, con->out_kvec_bytes,
656 con->out_more);
657 if (ret <= 0)
658 goto out;
659 con->out_kvec_bytes -= ret;
660 if (con->out_kvec_bytes == 0)
661 break; /* done */
662 while (ret > 0) {
663 if (ret >= con->out_kvec_cur->iov_len) {
664 ret -= con->out_kvec_cur->iov_len;
665 con->out_kvec_cur++;
666 con->out_kvec_left--;
667 } else {
668 con->out_kvec_cur->iov_len -= ret;
669 con->out_kvec_cur->iov_base += ret;
670 ret = 0;
671 break;
672 }
673 }
674 }
675 con->out_kvec_left = 0;
676 con->out_kvec_is_msg = false;
677 ret = 1;
678out:
679 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
680 con->out_kvec_bytes, con->out_kvec_left, ret);
681 return ret; /* done! */
682}
683
684/*
685 * Write as much message data payload as we can. If we finish, queue
686 * up the footer.
687 * 1 -> done, footer is now queued in out_kvec[].
688 * 0 -> socket full, but more to do
689 * <0 -> error
690 */
691static int write_partial_msg_pages(struct ceph_connection *con)
692{
693 struct ceph_msg *msg = con->out_msg;
694 unsigned data_len = le32_to_cpu(msg->hdr.data_len);
695 size_t len;
696 int crc = con->msgr->nocrc;
697 int ret;
698
699 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
700 con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages,
701 con->out_msg_pos.page_pos);
702
703 while (con->out_msg_pos.page < con->out_msg->nr_pages) {
704 struct page *page = NULL;
705 void *kaddr = NULL;
706
707 /*
708 * if we are calculating the data crc (the default), we need
709 * to map the page. if our pages[] has been revoked, use the
710 * zero page.
711 */
712 if (msg->pages) {
713 page = msg->pages[con->out_msg_pos.page];
714 if (crc)
715 kaddr = kmap(page);
716 } else {
717 page = con->msgr->zero_page;
718 if (crc)
719 kaddr = page_address(con->msgr->zero_page);
720 }
721 len = min((int)(PAGE_SIZE - con->out_msg_pos.page_pos),
722 (int)(data_len - con->out_msg_pos.data_pos));
723 if (crc && !con->out_msg_pos.did_page_crc) {
724 void *base = kaddr + con->out_msg_pos.page_pos;
725 u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc);
726
727 BUG_ON(kaddr == NULL);
728 con->out_msg->footer.data_crc =
729 cpu_to_le32(crc32c(tmpcrc, base, len));
730 con->out_msg_pos.did_page_crc = 1;
731 }
732
733 ret = kernel_sendpage(con->sock, page,
734 con->out_msg_pos.page_pos, len,
735 MSG_DONTWAIT | MSG_NOSIGNAL |
736 MSG_MORE);
737
738 if (crc && msg->pages)
739 kunmap(page);
740
741 if (ret <= 0)
742 goto out;
743
744 con->out_msg_pos.data_pos += ret;
745 con->out_msg_pos.page_pos += ret;
746 if (ret == len) {
747 con->out_msg_pos.page_pos = 0;
748 con->out_msg_pos.page++;
749 con->out_msg_pos.did_page_crc = 0;
750 }
751 }
752
753 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
754
755 /* prepare and queue up footer, too */
756 if (!crc)
757 con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
758 con->out_kvec_bytes = 0;
759 con->out_kvec_left = 0;
760 con->out_kvec_cur = con->out_kvec;
761 prepare_write_message_footer(con, 0);
762 ret = 1;
763out:
764 return ret;
765}
766
767/*
768 * write some zeros
769 */
770static int write_partial_skip(struct ceph_connection *con)
771{
772 int ret;
773
774 while (con->out_skip > 0) {
775 struct kvec iov = {
776 .iov_base = page_address(con->msgr->zero_page),
777 .iov_len = min(con->out_skip, (int)PAGE_CACHE_SIZE)
778 };
779
780 ret = ceph_tcp_sendmsg(con->sock, &iov, 1, iov.iov_len, 1);
781 if (ret <= 0)
782 goto out;
783 con->out_skip -= ret;
784 }
785 ret = 1;
786out:
787 return ret;
788}
789
790/*
791 * Prepare to read connection handshake, or an ack.
792 */
eed0ef2c
SW		 793static void prepare_read_banner(struct ceph_connection *con)
794{
795 dout("prepare_read_banner %p\n", con);
796 con->in_base_pos = 0;
797}
798
31b8006e
SW		 799static void prepare_read_connect(struct ceph_connection *con)
800{
801 dout("prepare_read_connect %p\n", con);
802 con->in_base_pos = 0;
803}
804
4e7a5dcd
SW		 805static void prepare_read_connect_retry(struct ceph_connection *con)
806{
807 dout("prepare_read_connect_retry %p\n", con);
808 con->in_base_pos = strlen(CEPH_BANNER) + sizeof(con->actual_peer_addr)
809 + sizeof(con->peer_addr_for_me);
810}
811
31b8006e
SW		 812static void prepare_read_ack(struct ceph_connection *con)
813{
814 dout("prepare_read_ack %p\n", con);
815 con->in_base_pos = 0;
816}
817
818static void prepare_read_tag(struct ceph_connection *con)
819{
820 dout("prepare_read_tag %p\n", con);
821 con->in_base_pos = 0;
822 con->in_tag = CEPH_MSGR_TAG_READY;
823}
824
825/*
826 * Prepare to read a message.
827 */
828static int prepare_read_message(struct ceph_connection *con)
829{
830 dout("prepare_read_message %p\n", con);
831 BUG_ON(con->in_msg != NULL);
832 con->in_base_pos = 0;
833 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
834 return 0;
835}
836
837
838static int read_partial(struct ceph_connection *con,
839 int *to, int size, void *object)
840{
841 *to += size;
842 while (con->in_base_pos < *to) {
843 int left = *to - con->in_base_pos;
844 int have = size - left;
845 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
846 if (ret <= 0)
847 return ret;
848 con->in_base_pos += ret;
849 }
850 return 1;
851}
852
853
854/*
855 * Read all or part of the connect-side handshake on a new connection
856 */
eed0ef2c 857static int read_partial_banner(struct ceph_connection *con)
31b8006e
SW		 858{
859 int ret, to = 0;
860
eed0ef2c 861 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
31b8006e
SW		 862
863 /* peer's banner */
864 ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner);
865 if (ret <= 0)
866 goto out;
867 ret = read_partial(con, &to, sizeof(con->actual_peer_addr),
868 &con->actual_peer_addr);
869 if (ret <= 0)
870 goto out;
871 ret = read_partial(con, &to, sizeof(con->peer_addr_for_me),
872 &con->peer_addr_for_me);
873 if (ret <= 0)
874 goto out;
eed0ef2c
SW		 875out:
876 return ret;
877}
878
879static int read_partial_connect(struct ceph_connection *con)
880{
881 int ret, to = 0;
882
883 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
884
31b8006e
SW		 885 ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply);
886 if (ret <= 0)
887 goto out;
4e7a5dcd
SW		 888 ret = read_partial(con, &to, le32_to_cpu(con->in_reply.authorizer_len),
889 con->auth_reply_buf);
890 if (ret <= 0)
891 goto out;
31b8006e 892
4e7a5dcd
SW		 893 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
894 con, (int)con->in_reply.tag,
895 le32_to_cpu(con->in_reply.connect_seq),
31b8006e
SW		 896 le32_to_cpu(con->in_reply.global_seq));
897out:
898 return ret;
eed0ef2c 899
31b8006e
SW		 900}
901
902/*
903 * Verify the hello banner looks okay.
904 */
905static int verify_hello(struct ceph_connection *con)
906{
907 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
13e38c8a 908 pr_err("connect to %s got bad banner\n",
31b8006e
SW		 909 pr_addr(&con->peer_addr.in_addr));
910 con->error_msg = "protocol error, bad banner";
911 return -1;
912 }
913 return 0;
914}
915
916static bool addr_is_blank(struct sockaddr_storage *ss)
917{
918 switch (ss->ss_family) {
919 case AF_INET:
920 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
921 case AF_INET6:
922 return
923 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
924 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
925 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
926 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
927 }
928 return false;
929}
930
931static int addr_port(struct sockaddr_storage *ss)
932{
933 switch (ss->ss_family) {
934 case AF_INET:
f28bcfbe 935 return ntohs(((struct sockaddr_in *)ss)->sin_port);
31b8006e 936 case AF_INET6:
f28bcfbe 937 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
31b8006e
SW		 938 }
939 return 0;
940}
941
942static void addr_set_port(struct sockaddr_storage *ss, int p)
943{
944 switch (ss->ss_family) {
945 case AF_INET:
946 ((struct sockaddr_in *)ss)->sin_port = htons(p);
947 case AF_INET6:
948 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
949 }
950}
951
952/*
953 * Parse an ip[:port] list into an addr array. Use the default
954 * monitor port if a port isn't specified.
955 */
956int ceph_parse_ips(const char *c, const char *end,
957 struct ceph_entity_addr *addr,
958 int max_count, int *count)
959{
960 int i;
961 const char *p = c;
962
963 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
964 for (i = 0; i < max_count; i++) {
965 const char *ipend;
966 struct sockaddr_storage *ss = &addr[i].in_addr;
967 struct sockaddr_in *in4 = (void *)ss;
968 struct sockaddr_in6 *in6 = (void *)ss;
969 int port;
970
971 memset(ss, 0, sizeof(*ss));
972 if (in4_pton(p, end - p, (u8 *)&in4->sin_addr.s_addr,
973 ',', &ipend)) {
974 ss->ss_family = AF_INET;
975 } else if (in6_pton(p, end - p, (u8 *)&in6->sin6_addr.s6_addr,
976 ',', &ipend)) {
977 ss->ss_family = AF_INET6;
978 } else {
979 goto bad;
980 }
981 p = ipend;
982
983 /* port? */
984 if (p < end && *p == ':') {
985 port = 0;
986 p++;
987 while (p < end && *p >= '0' && *p <= '9') {
988 port = (port * 10) + (*p - '0');
989 p++;
990 }
991 if (port > 65535 || port == 0)
992 goto bad;
993 } else {
994 port = CEPH_MON_PORT;
995 }
996
997 addr_set_port(ss, port);
998
999 dout("parse_ips got %s\n", pr_addr(ss));
1000
1001 if (p == end)
1002 break;
1003 if (*p != ',')
1004 goto bad;
1005 p++;
1006 }
1007
1008 if (p != end)
1009 goto bad;
1010
1011 if (count)
1012 *count = i + 1;
1013 return 0;
1014
1015bad:
1016 pr_err("parse_ips bad ip '%s'\n", c);
1017 return -EINVAL;
1018}
1019
eed0ef2c 1020static int process_banner(struct ceph_connection *con)
31b8006e 1021{
eed0ef2c 1022 dout("process_banner on %p\n", con);
31b8006e
SW		 1023
1024 if (verify_hello(con) < 0)
1025 return -1;
1026
63f2d211
SW		 1027 ceph_decode_addr(&con->actual_peer_addr);
1028 ceph_decode_addr(&con->peer_addr_for_me);
1029
31b8006e
SW		 1030 /*
1031 * Make sure the other end is who we wanted. note that the other
1032 * end may not yet know their ip address, so if it's 0.0.0.0, give
1033 * them the benefit of the doubt.
1034 */
1035 if (!ceph_entity_addr_is_local(&con->peer_addr,
1036 &con->actual_peer_addr) &&
1037 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1038 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1039 pr_err("wrong peer, want %s/%d, "
1040 "got %s/%d, wtf\n",
1041 pr_addr(&con->peer_addr.in_addr),
1042 con->peer_addr.nonce,
1043 pr_addr(&con->actual_peer_addr.in_addr),
1044 con->actual_peer_addr.nonce);
1045 con->error_msg = "protocol error, wrong peer";
1046 return -1;
1047 }
1048
1049 /*
1050 * did we learn our address?
1051 */
1052 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1053 int port = addr_port(&con->msgr->inst.addr.in_addr);
1054
1055 memcpy(&con->msgr->inst.addr.in_addr,
1056 &con->peer_addr_for_me.in_addr,
1057 sizeof(con->peer_addr_for_me.in_addr));
1058 addr_set_port(&con->msgr->inst.addr.in_addr, port);
63f2d211 1059 encode_my_addr(con->msgr);
eed0ef2c 1060 dout("process_banner learned my addr is %s\n",
31b8006e
SW		 1061 pr_addr(&con->msgr->inst.addr.in_addr));
1062 }
1063
eed0ef2c
SW		 1064 set_bit(NEGOTIATING, &con->state);
1065 prepare_read_connect(con);
1066 return 0;
1067}
1068
1069static int process_connect(struct ceph_connection *con)
1070{
1071 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1072
31b8006e
SW		 1073 switch (con->in_reply.tag) {
1074 case CEPH_MSGR_TAG_BADPROTOVER:
1075 dout("process_connect got BADPROTOVER my %d != their %d\n",
1076 le32_to_cpu(con->out_connect.protocol_version),
1077 le32_to_cpu(con->in_reply.protocol_version));
1078 pr_err("%s%lld %s protocol version mismatch,"
1079 " my %d != server's %d\n",
1080 ENTITY_NAME(con->peer_name),
1081 pr_addr(&con->peer_addr.in_addr),
1082 le32_to_cpu(con->out_connect.protocol_version),
1083 le32_to_cpu(con->in_reply.protocol_version));
1084 con->error_msg = "protocol version mismatch";
1085 if (con->ops->bad_proto)
1086 con->ops->bad_proto(con);
1087 reset_connection(con);
1088 set_bit(CLOSED, &con->state); /* in case there's queued work */
1089 return -1;
1090
4e7a5dcd
SW		 1091 case CEPH_MSGR_TAG_BADAUTHORIZER:
1092 con->auth_retry++;
1093 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1094 con->auth_retry);
1095 if (con->auth_retry == 2) {
1096 con->error_msg = "connect authorization failure";
1097 reset_connection(con);
1098 set_bit(CLOSED, &con->state);
1099 return -1;
1100 }
1101 con->auth_retry = 1;
1102 prepare_write_connect(con->msgr, con, 0);
1103 prepare_read_connect_retry(con);
1104 break;
31b8006e
SW		 1105
1106 case CEPH_MSGR_TAG_RESETSESSION:
1107 /*
1108 * If we connected with a large connect_seq but the peer
1109 * has no record of a session with us (no connection, or
1110 * connect_seq == 0), they will send RESETSESION to indicate
1111 * that they must have reset their session, and may have
1112 * dropped messages.
1113 */
1114 dout("process_connect got RESET peer seq %u\n",
1115 le32_to_cpu(con->in_connect.connect_seq));
1116 pr_err("%s%lld %s connection reset\n",
1117 ENTITY_NAME(con->peer_name),
1118 pr_addr(&con->peer_addr.in_addr));
1119 reset_connection(con);
eed0ef2c 1120 prepare_write_connect(con->msgr, con, 0);
31b8006e
SW		 1121 prepare_read_connect(con);
1122
1123 /* Tell ceph about it. */
1124 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1125 if (con->ops->peer_reset)
1126 con->ops->peer_reset(con);
1127 break;
1128
1129 case CEPH_MSGR_TAG_RETRY_SESSION:
1130 /*
1131 * If we sent a smaller connect_seq than the peer has, try
1132 * again with a larger value.
1133 */
1134 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1135 le32_to_cpu(con->out_connect.connect_seq),
1136 le32_to_cpu(con->in_connect.connect_seq));
1137 con->connect_seq = le32_to_cpu(con->in_connect.connect_seq);
eed0ef2c 1138 prepare_write_connect(con->msgr, con, 0);
31b8006e
SW		 1139 prepare_read_connect(con);
1140 break;
1141
1142 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1143 /*
1144 * If we sent a smaller global_seq than the peer has, try
1145 * again with a larger value.
1146 */
eed0ef2c 1147 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
31b8006e
SW		 1148 con->peer_global_seq,
1149 le32_to_cpu(con->in_connect.global_seq));
1150 get_global_seq(con->msgr,
1151 le32_to_cpu(con->in_connect.global_seq));
eed0ef2c 1152 prepare_write_connect(con->msgr, con, 0);
31b8006e
SW		 1153 prepare_read_connect(con);
1154 break;
1155
1156 case CEPH_MSGR_TAG_READY:
1157 clear_bit(CONNECTING, &con->state);
31b8006e
SW		 1158 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1159 con->connect_seq++;
1160 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1161 con->peer_global_seq,
1162 le32_to_cpu(con->in_reply.connect_seq),
1163 con->connect_seq);
1164 WARN_ON(con->connect_seq !=
1165 le32_to_cpu(con->in_reply.connect_seq));
31b8006e
SW		 1166 prepare_read_tag(con);
1167 break;
1168
1169 case CEPH_MSGR_TAG_WAIT:
1170 /*
1171 * If there is a connection race (we are opening
1172 * connections to each other), one of us may just have
1173 * to WAIT. This shouldn't happen if we are the
1174 * client.
1175 */
1176 pr_err("process_connect peer connecting WAIT\n");
1177
1178 default:
1179 pr_err("connect protocol error, will retry\n");
1180 con->error_msg = "protocol error, garbage tag during connect";
1181 return -1;
1182 }
1183 return 0;
1184}
1185
1186
1187/*
1188 * read (part of) an ack
1189 */
1190static int read_partial_ack(struct ceph_connection *con)
1191{
1192 int to = 0;
1193
1194 return read_partial(con, &to, sizeof(con->in_temp_ack),
1195 &con->in_temp_ack);
1196}
1197
1198
1199/*
1200 * We can finally discard anything that's been acked.
1201 */
1202static void process_ack(struct ceph_connection *con)
1203{
1204 struct ceph_msg *m;
1205 u64 ack = le64_to_cpu(con->in_temp_ack);
1206 u64 seq;
1207
1208 mutex_lock(&con->out_mutex);
1209 while (!list_empty(&con->out_sent)) {
1210 m = list_first_entry(&con->out_sent, struct ceph_msg,
1211 list_head);
1212 seq = le64_to_cpu(m->hdr.seq);
1213 if (seq > ack)
1214 break;
1215 dout("got ack for seq %llu type %d at %p\n", seq,
1216 le16_to_cpu(m->hdr.type), m);
1217 ceph_msg_remove(m);
1218 }
1219 mutex_unlock(&con->out_mutex);
1220 prepare_read_tag(con);
1221}
1222
1223
1224
1225
1226
1227
1228/*
1229 * read (part of) a message.
1230 */
1231static int read_partial_message(struct ceph_connection *con)
1232{
1233 struct ceph_msg *m = con->in_msg;
1234 void *p;
1235 int ret;
1236 int to, want, left;
1237 unsigned front_len, middle_len, data_len, data_off;
1238 int datacrc = con->msgr->nocrc;
1239
1240 dout("read_partial_message con %p msg %p\n", con, m);
1241
1242 /* header */
1243 while (con->in_base_pos < sizeof(con->in_hdr)) {
1244 left = sizeof(con->in_hdr) - con->in_base_pos;
1245 ret = ceph_tcp_recvmsg(con->sock,
1246 (char *)&con->in_hdr + con->in_base_pos,
1247 left);
1248 if (ret <= 0)
1249 return ret;
1250 con->in_base_pos += ret;
1251 if (con->in_base_pos == sizeof(con->in_hdr)) {
1252 u32 crc = crc32c(0, (void *)&con->in_hdr,
1253 sizeof(con->in_hdr) - sizeof(con->in_hdr.crc));
1254 if (crc != le32_to_cpu(con->in_hdr.crc)) {
1255 pr_err("read_partial_message bad hdr "
1256 " crc %u != expected %u\n",
1257 crc, con->in_hdr.crc);
1258 return -EBADMSG;
1259 }
1260 }
1261 }
1262
1263 front_len = le32_to_cpu(con->in_hdr.front_len);
1264 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1265 return -EIO;
1266 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1267 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1268 return -EIO;
1269 data_len = le32_to_cpu(con->in_hdr.data_len);
1270 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1271 return -EIO;
1272
1273 /* allocate message? */
1274 if (!con->in_msg) {
1275 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1276 con->in_hdr.front_len, con->in_hdr.data_len);
1277 con->in_msg = con->ops->alloc_msg(con, &con->in_hdr);
1278 if (!con->in_msg) {
1279 /* skip this message */
1280 dout("alloc_msg returned NULL, skipping message\n");
1281 con->in_base_pos = -front_len - middle_len - data_len -
1282 sizeof(m->footer);
1283 con->in_tag = CEPH_MSGR_TAG_READY;
1284 return 0;
1285 }
1286 if (IS_ERR(con->in_msg)) {
1287 ret = PTR_ERR(con->in_msg);
1288 con->in_msg = NULL;
1289 con->error_msg = "out of memory for incoming message";
1290 return ret;
1291 }
1292 m = con->in_msg;
1293 m->front.iov_len = 0; /* haven't read it yet */
1294 memcpy(&m->hdr, &con->in_hdr, sizeof(con->in_hdr));
1295 }
1296
1297 /* front */
1298 while (m->front.iov_len < front_len) {
1299 BUG_ON(m->front.iov_base == NULL);
1300 left = front_len - m->front.iov_len;
1301 ret = ceph_tcp_recvmsg(con->sock, (char *)m->front.iov_base +
1302 m->front.iov_len, left);
1303 if (ret <= 0)
1304 return ret;
1305 m->front.iov_len += ret;
1306 if (m->front.iov_len == front_len)
1307 con->in_front_crc = crc32c(0, m->front.iov_base,
1308 m->front.iov_len);
1309 }
1310
1311 /* middle */
1312 while (middle_len > 0 && (!m->middle ||
1313 m->middle->vec.iov_len < middle_len)) {
1314 if (m->middle == NULL) {
1315 ret = -EOPNOTSUPP;
1316 if (con->ops->alloc_middle)
1317 ret = con->ops->alloc_middle(con, m);
1318 if (ret < 0) {
1319 dout("alloc_middle failed, skipping payload\n");
1320 con->in_base_pos = -middle_len - data_len
1321 - sizeof(m->footer);
1322 ceph_msg_put(con->in_msg);
1323 con->in_msg = NULL;
1324 con->in_tag = CEPH_MSGR_TAG_READY;
1325 return 0;
1326 }
1327 m->middle->vec.iov_len = 0;
1328 }
1329 left = middle_len - m->middle->vec.iov_len;
1330 ret = ceph_tcp_recvmsg(con->sock,
1331 (char *)m->middle->vec.iov_base +
1332 m->middle->vec.iov_len, left);
1333 if (ret <= 0)
1334 return ret;
1335 m->middle->vec.iov_len += ret;
1336 if (m->middle->vec.iov_len == middle_len)
1337 con->in_middle_crc = crc32c(0, m->middle->vec.iov_base,
1338 m->middle->vec.iov_len);
1339 }
1340
1341 /* (page) data */
1342 data_off = le16_to_cpu(m->hdr.data_off);
1343 if (data_len == 0)
1344 goto no_data;
1345
1346 if (m->nr_pages == 0) {
1347 con->in_msg_pos.page = 0;
1348 con->in_msg_pos.page_pos = data_off & ~PAGE_MASK;
1349 con->in_msg_pos.data_pos = 0;
1350 /* find pages for data payload */
1351 want = calc_pages_for(data_off & ~PAGE_MASK, data_len);
1352 ret = -1;
1353 if (con->ops->prepare_pages)
1354 ret = con->ops->prepare_pages(con, m, want);
1355 if (ret < 0) {
1356 dout("%p prepare_pages failed, skipping payload\n", m);
1357 con->in_base_pos = -data_len - sizeof(m->footer);
1358 ceph_msg_put(con->in_msg);
1359 con->in_msg = NULL;
1360 con->in_tag = CEPH_MSGR_TAG_READY;
1361 return 0;
1362 }
1363 BUG_ON(m->nr_pages < want);
1364 }
1365 while (con->in_msg_pos.data_pos < data_len) {
1366 left = min((int)(data_len - con->in_msg_pos.data_pos),
1367 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1368 BUG_ON(m->pages == NULL);
1369 p = kmap(m->pages[con->in_msg_pos.page]);
1370 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1371 left);
1372 if (ret > 0 && datacrc)
1373 con->in_data_crc =
1374 crc32c(con->in_data_crc,
1375 p + con->in_msg_pos.page_pos, ret);
1376 kunmap(m->pages[con->in_msg_pos.page]);
1377 if (ret <= 0)
1378 return ret;
1379 con->in_msg_pos.data_pos += ret;
1380 con->in_msg_pos.page_pos += ret;
1381 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1382 con->in_msg_pos.page_pos = 0;
1383 con->in_msg_pos.page++;
1384 }
1385 }
1386
1387no_data:
1388 /* footer */
1389 to = sizeof(m->hdr) + sizeof(m->footer);
1390 while (con->in_base_pos < to) {
1391 left = to - con->in_base_pos;
1392 ret = ceph_tcp_recvmsg(con->sock, (char *)&m->footer +
1393 (con->in_base_pos - sizeof(m->hdr)),
1394 left);
1395 if (ret <= 0)
1396 return ret;
1397 con->in_base_pos += ret;
1398 }
1399 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1400 m, front_len, m->footer.front_crc, middle_len,
1401 m->footer.middle_crc, data_len, m->footer.data_crc);
1402
1403 /* crc ok? */
1404 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1405 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1406 m, con->in_front_crc, m->footer.front_crc);
1407 return -EBADMSG;
1408 }
1409 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1410 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1411 m, con->in_middle_crc, m->footer.middle_crc);
1412 return -EBADMSG;
1413 }
1414 if (datacrc &&
1415 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1416 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1417 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1418 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1419 return -EBADMSG;
1420 }
1421
1422 return 1; /* done! */
1423}
1424
1425/*
1426 * Process message. This happens in the worker thread. The callback should
1427 * be careful not to do anything that waits on other incoming messages or it
1428 * may deadlock.
1429 */
1430static void process_message(struct ceph_connection *con)
1431{
1432 struct ceph_msg *msg = con->in_msg;
1433
1434 con->in_msg = NULL;
1435
1436 /* if first message, set peer_name */
1437 if (con->peer_name.type == 0)
1438 con->peer_name = msg->hdr.src.name;
1439
1440 mutex_lock(&con->out_mutex);
1441 con->in_seq++;
1442 mutex_unlock(&con->out_mutex);
1443
1444 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1445 msg, le64_to_cpu(msg->hdr.seq),
1446 ENTITY_NAME(msg->hdr.src.name),
1447 le16_to_cpu(msg->hdr.type),
1448 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1449 le32_to_cpu(msg->hdr.front_len),
1450 le32_to_cpu(msg->hdr.data_len),
1451 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1452 con->ops->dispatch(con, msg);
1453 prepare_read_tag(con);
1454}
1455
1456
1457/*
1458 * Write something to the socket. Called in a worker thread when the
1459 * socket appears to be writeable and we have something ready to send.
1460 */
1461static int try_write(struct ceph_connection *con)
1462{
1463 struct ceph_messenger *msgr = con->msgr;
1464 int ret = 1;
1465
1466 dout("try_write start %p state %lu nref %d\n", con, con->state,
1467 atomic_read(&con->nref));
1468
1469 mutex_lock(&con->out_mutex);
1470more:
1471 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1472
1473 /* open the socket first? */
1474 if (con->sock == NULL) {
1475 /*
1476 * if we were STANDBY and are reconnecting _this_
1477 * connection, bump connect_seq now. Always bump
1478 * global_seq.
1479 */
1480 if (test_and_clear_bit(STANDBY, &con->state))
1481 con->connect_seq++;
1482
eed0ef2c
SW		 1483 prepare_write_banner(msgr, con);
1484 prepare_write_connect(msgr, con, 1);
1485 prepare_read_banner(con);
31b8006e 1486 set_bit(CONNECTING, &con->state);
eed0ef2c 1487 clear_bit(NEGOTIATING, &con->state);
31b8006e
SW		 1488
1489 con->in_tag = CEPH_MSGR_TAG_READY;
1490 dout("try_write initiating connect on %p new state %lu\n",
1491 con, con->state);
1492 con->sock = ceph_tcp_connect(con);
1493 if (IS_ERR(con->sock)) {
1494 con->sock = NULL;
1495 con->error_msg = "connect error";
1496 ret = -1;
1497 goto out;
1498 }
1499 }
1500
1501more_kvec:
1502 /* kvec data queued? */
1503 if (con->out_skip) {
1504 ret = write_partial_skip(con);
1505 if (ret <= 0)
1506 goto done;
1507 if (ret < 0) {
1508 dout("try_write write_partial_skip err %d\n", ret);
1509 goto done;
1510 }
1511 }
1512 if (con->out_kvec_left) {
1513 ret = write_partial_kvec(con);
1514 if (ret <= 0)
1515 goto done;
1516 if (ret < 0) {
1517 dout("try_write write_partial_kvec err %d\n", ret);
1518 goto done;
1519 }
1520 }
1521
1522 /* msg pages? */
1523 if (con->out_msg) {
1524 ret = write_partial_msg_pages(con);
1525 if (ret == 1)
1526 goto more_kvec; /* we need to send the footer, too! */
1527 if (ret == 0)
1528 goto done;
1529 if (ret < 0) {
1530 dout("try_write write_partial_msg_pages err %d\n",
1531 ret);
1532 goto done;
1533 }
1534 }
1535
1536 if (!test_bit(CONNECTING, &con->state)) {
1537 /* is anything else pending? */
1538 if (!list_empty(&con->out_queue)) {
1539 prepare_write_message(con);
1540 goto more;
1541 }
1542 if (con->in_seq > con->in_seq_acked) {
1543 prepare_write_ack(con);
1544 goto more;
1545 }
1546 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) {
1547 prepare_write_keepalive(con);
1548 goto more;
1549 }
1550 }
1551
1552 /* Nothing to do! */
1553 clear_bit(WRITE_PENDING, &con->state);
1554 dout("try_write nothing else to write.\n");
1555done:
1556 ret = 0;
1557out:
1558 mutex_unlock(&con->out_mutex);
1559 dout("try_write done on %p\n", con);
1560 return ret;
1561}
1562
1563
1564
1565/*
1566 * Read what we can from the socket.
1567 */
1568static int try_read(struct ceph_connection *con)
1569{
1570 struct ceph_messenger *msgr;
1571 int ret = -1;
1572
1573 if (!con->sock)
1574 return 0;
1575
1576 if (test_bit(STANDBY, &con->state))
1577 return 0;
1578
1579 dout("try_read start on %p\n", con);
1580 msgr = con->msgr;
1581
1582more:
1583 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
1584 con->in_base_pos);
1585 if (test_bit(CONNECTING, &con->state)) {
eed0ef2c
SW		 1586 if (!test_bit(NEGOTIATING, &con->state)) {
1587 dout("try_read connecting\n");
1588 ret = read_partial_banner(con);
1589 if (ret <= 0)
1590 goto done;
1591 if (process_banner(con) < 0) {
1592 ret = -1;
1593 goto out;
1594 }
1595 }
31b8006e
SW		 1596 ret = read_partial_connect(con);
1597 if (ret <= 0)
1598 goto done;
1599 if (process_connect(con) < 0) {
1600 ret = -1;
1601 goto out;
1602 }
1603 goto more;
1604 }
1605
1606 if (con->in_base_pos < 0) {
1607 /*
1608 * skipping + discarding content.
1609 *
1610 * FIXME: there must be a better way to do this!
1611 */
1612 static char buf[1024];
1613 int skip = min(1024, -con->in_base_pos);
1614 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
1615 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
1616 if (ret <= 0)
1617 goto done;
1618 con->in_base_pos += ret;
1619 if (con->in_base_pos)
1620 goto more;
1621 }
1622 if (con->in_tag == CEPH_MSGR_TAG_READY) {
1623 /*
1624 * what's next?
1625 */
1626 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
1627 if (ret <= 0)
1628 goto done;
1629 dout("try_read got tag %d\n", (int)con->in_tag);
1630 switch (con->in_tag) {
1631 case CEPH_MSGR_TAG_MSG:
1632 prepare_read_message(con);
1633 break;
1634 case CEPH_MSGR_TAG_ACK:
1635 prepare_read_ack(con);
1636 break;
1637 case CEPH_MSGR_TAG_CLOSE:
1638 set_bit(CLOSED, &con->state); /* fixme */
1639 goto done;
1640 default:
1641 goto bad_tag;
1642 }
1643 }
1644 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
1645 ret = read_partial_message(con);
1646 if (ret <= 0) {
1647 switch (ret) {
1648 case -EBADMSG:
1649 con->error_msg = "bad crc";
1650 ret = -EIO;
1651 goto out;
1652 case -EIO:
1653 con->error_msg = "io error";
1654 goto out;
1655 default:
1656 goto done;
1657 }
1658 }
1659 if (con->in_tag == CEPH_MSGR_TAG_READY)
1660 goto more;
1661 process_message(con);
1662 goto more;
1663 }
1664 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
1665 ret = read_partial_ack(con);
1666 if (ret <= 0)
1667 goto done;
1668 process_ack(con);
1669 goto more;
1670 }
1671
1672done:
1673 ret = 0;
1674out:
1675 dout("try_read done on %p\n", con);
1676 return ret;
1677
1678bad_tag:
1679 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
1680 con->error_msg = "protocol error, garbage tag";
1681 ret = -1;
1682 goto out;
1683}
1684
1685
1686/*
1687 * Atomically queue work on a connection. Bump @con reference to
1688 * avoid races with connection teardown.
1689 *
1690 * There is some trickery going on with QUEUED and BUSY because we
1691 * only want a _single_ thread operating on each connection at any
1692 * point in time, but we want to use all available CPUs.
1693 *
1694 * The worker thread only proceeds if it can atomically set BUSY. It
1695 * clears QUEUED and does it's thing. When it thinks it's done, it
1696 * clears BUSY, then rechecks QUEUED.. if it's set again, it loops
1697 * (tries again to set BUSY).
1698 *
1699 * To queue work, we first set QUEUED, _then_ if BUSY isn't set, we
1700 * try to queue work. If that fails (work is already queued, or BUSY)
1701 * we give up (work also already being done or is queued) but leave QUEUED
1702 * set so that the worker thread will loop if necessary.
1703 */
1704static void queue_con(struct ceph_connection *con)
1705{
1706 if (test_bit(DEAD, &con->state)) {
1707 dout("queue_con %p ignoring: DEAD\n",
1708 con);
1709 return;
1710 }
1711
1712 if (!con->ops->get(con)) {
1713 dout("queue_con %p ref count 0\n", con);
1714 return;
1715 }
1716
1717 set_bit(QUEUED, &con->state);
1718 if (test_bit(BUSY, &con->state)) {
1719 dout("queue_con %p - already BUSY\n", con);
1720 con->ops->put(con);
1721 } else if (!queue_work(ceph_msgr_wq, &con->work.work)) {
1722 dout("queue_con %p - already queued\n", con);
1723 con->ops->put(con);
1724 } else {
1725 dout("queue_con %p\n", con);
1726 }
1727}
1728
1729/*
1730 * Do some work on a connection. Drop a connection ref when we're done.
1731 */
1732static void con_work(struct work_struct *work)
1733{
1734 struct ceph_connection *con = container_of(work, struct ceph_connection,
1735 work.work);
1736 int backoff = 0;
1737
1738more:
1739 if (test_and_set_bit(BUSY, &con->state) != 0) {
1740 dout("con_work %p BUSY already set\n", con);
1741 goto out;
1742 }
1743 dout("con_work %p start, clearing QUEUED\n", con);
1744 clear_bit(QUEUED, &con->state);
1745
1746 if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
1747 dout("con_work CLOSED\n");
1748 con_close_socket(con);
1749 goto done;
1750 }
1751 if (test_and_clear_bit(OPENING, &con->state)) {
1752 /* reopen w/ new peer */
1753 dout("con_work OPENING\n");
1754 con_close_socket(con);
1755 }
1756
1757 if (test_and_clear_bit(SOCK_CLOSED, &con->state) ||
1758 try_read(con) < 0 ||
1759 try_write(con) < 0) {
1760 backoff = 1;
1761 ceph_fault(con); /* error/fault path */
1762 }
1763
1764done:
1765 clear_bit(BUSY, &con->state);
1766 dout("con->state=%lu\n", con->state);
1767 if (test_bit(QUEUED, &con->state)) {
1768 if (!backoff) {
1769 dout("con_work %p QUEUED reset, looping\n", con);
1770 goto more;
1771 }
1772 dout("con_work %p QUEUED reset, but just faulted\n", con);
1773 clear_bit(QUEUED, &con->state);
1774 }
1775 dout("con_work %p done\n", con);
1776
1777out:
1778 con->ops->put(con);
1779}
1780
1781
1782/*
1783 * Generic error/fault handler. A retry mechanism is used with
1784 * exponential backoff
1785 */
1786static void ceph_fault(struct ceph_connection *con)
1787{
1788 pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
1789 pr_addr(&con->peer_addr.in_addr), con->error_msg);
1790 dout("fault %p state %lu to peer %s\n",
1791 con, con->state, pr_addr(&con->peer_addr.in_addr));
1792
1793 if (test_bit(LOSSYTX, &con->state)) {
1794 dout("fault on LOSSYTX channel\n");
1795 goto out;
1796 }
1797
1798 clear_bit(BUSY, &con->state); /* to avoid an improbable race */
1799
1800 con_close_socket(con);
1801 con->in_msg = NULL;
1802
1803 /* If there are no messages in the queue, place the connection
1804 * in a STANDBY state (i.e., don't try to reconnect just yet). */
1805 mutex_lock(&con->out_mutex);
1806 if (list_empty(&con->out_queue) && !con->out_keepalive_pending) {
1807 dout("fault setting STANDBY\n");
1808 set_bit(STANDBY, &con->state);
1809 mutex_unlock(&con->out_mutex);
1810 goto out;
1811 }
1812
1813 /* Requeue anything that hasn't been acked, and retry after a
1814 * delay. */
1815 list_splice_init(&con->out_sent, &con->out_queue);
1816 mutex_unlock(&con->out_mutex);
1817
1818 if (con->delay == 0)
1819 con->delay = BASE_DELAY_INTERVAL;
1820 else if (con->delay < MAX_DELAY_INTERVAL)
1821 con->delay *= 2;
1822
1823 /* explicitly schedule work to try to reconnect again later. */
1824 dout("fault queueing %p delay %lu\n", con, con->delay);
1825 con->ops->get(con);
1826 if (queue_delayed_work(ceph_msgr_wq, &con->work,
1827 round_jiffies_relative(con->delay)) == 0)
1828 con->ops->put(con);
1829
1830out:
1831 if (con->ops->fault)
1832 con->ops->fault(con);
1833}
1834
1835
1836
1837/*
1838 * create a new messenger instance
1839 */
1840struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr)
1841{
1842 struct ceph_messenger *msgr;
1843
1844 msgr = kzalloc(sizeof(*msgr), GFP_KERNEL);
1845 if (msgr == NULL)
1846 return ERR_PTR(-ENOMEM);
1847
1848 spin_lock_init(&msgr->global_seq_lock);
1849
1850 /* the zero page is needed if a request is "canceled" while the message
1851 * is being written over the socket */
1852 msgr->zero_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1853 if (!msgr->zero_page) {
1854 kfree(msgr);
1855 return ERR_PTR(-ENOMEM);
1856 }
1857 kmap(msgr->zero_page);
1858
1859 if (myaddr)
1860 msgr->inst.addr = *myaddr;
1861
1862 /* select a random nonce */
1863 get_random_bytes(&msgr->inst.addr.nonce,
1864 sizeof(msgr->inst.addr.nonce));
63f2d211 1865 encode_my_addr(msgr);
31b8006e
SW		 1866
1867 dout("messenger_create %p\n", msgr);
1868 return msgr;
1869}
1870
1871void ceph_messenger_destroy(struct ceph_messenger *msgr)
1872{
1873 dout("destroy %p\n", msgr);
1874 kunmap(msgr->zero_page);
1875 __free_page(msgr->zero_page);
1876 kfree(msgr);
1877 dout("destroyed messenger %p\n", msgr);
1878}
1879
1880/*
1881 * Queue up an outgoing message on the given connection.
1882 */
1883void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
1884{
1885 if (test_bit(CLOSED, &con->state)) {
1886 dout("con_send %p closed, dropping %p\n", con, msg);
1887 ceph_msg_put(msg);
1888 return;
1889 }
1890
1891 /* set src+dst */
63f2d211
SW		 1892 msg->hdr.src.name = con->msgr->inst.name;
1893 msg->hdr.src.addr = con->msgr->my_enc_addr;
1894 msg->hdr.orig_src = msg->hdr.src;
31b8006e
SW		 1895 msg->hdr.dst_erank = con->peer_addr.erank;
1896
1897 /* queue */
1898 mutex_lock(&con->out_mutex);
1899 BUG_ON(!list_empty(&msg->list_head));
1900 list_add_tail(&msg->list_head, &con->out_queue);
1901 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
1902 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
1903 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1904 le32_to_cpu(msg->hdr.front_len),
1905 le32_to_cpu(msg->hdr.middle_len),
1906 le32_to_cpu(msg->hdr.data_len));
1907 mutex_unlock(&con->out_mutex);
1908
1909 /* if there wasn't anything waiting to send before, queue
1910 * new work */
1911 if (test_and_set_bit(WRITE_PENDING, &con->state) == 0)
1912 queue_con(con);
1913}
1914
1915/*
1916 * Revoke a message that was previously queued for send
1917 */
1918void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg)
1919{
1920 mutex_lock(&con->out_mutex);
1921 if (!list_empty(&msg->list_head)) {
1922 dout("con_revoke %p msg %p\n", con, msg);
1923 list_del_init(&msg->list_head);
1924 ceph_msg_put(msg);
1925 msg->hdr.seq = 0;
1926 if (con->out_msg == msg)
1927 con->out_msg = NULL;
1928 if (con->out_kvec_is_msg) {
1929 con->out_skip = con->out_kvec_bytes;
1930 con->out_kvec_is_msg = false;
1931 }
1932 } else {
1933 dout("con_revoke %p msg %p - not queued (sent?)\n", con, msg);
1934 }
1935 mutex_unlock(&con->out_mutex);
1936}
1937
1938/*
1939 * Queue a keepalive byte to ensure the tcp connection is alive.
1940 */
1941void ceph_con_keepalive(struct ceph_connection *con)
1942{
1943 if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 &&
1944 test_and_set_bit(WRITE_PENDING, &con->state) == 0)
1945 queue_con(con);
1946}
1947
1948
1949/*
1950 * construct a new message with given type, size
1951 * the new msg has a ref count of 1.
1952 */
1953struct ceph_msg *ceph_msg_new(int type, int front_len,
1954 int page_len, int page_off, struct page **pages)
1955{
1956 struct ceph_msg *m;
1957
1958 m = kmalloc(sizeof(*m), GFP_NOFS);
1959 if (m == NULL)
1960 goto out;
1961 atomic_set(&m->nref, 1);
1962 INIT_LIST_HEAD(&m->list_head);
1963
1964 m->hdr.type = cpu_to_le16(type);
1965 m->hdr.front_len = cpu_to_le32(front_len);
1966 m->hdr.middle_len = 0;
1967 m->hdr.data_len = cpu_to_le32(page_len);
1968 m->hdr.data_off = cpu_to_le16(page_off);
1969 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
1970 m->footer.front_crc = 0;
1971 m->footer.middle_crc = 0;
1972 m->footer.data_crc = 0;
1973 m->front_max = front_len;
1974 m->front_is_vmalloc = false;
1975 m->more_to_follow = false;
1976 m->pool = NULL;
1977
1978 /* front */
1979 if (front_len) {
1980 if (front_len > PAGE_CACHE_SIZE) {
1981 m->front.iov_base = __vmalloc(front_len, GFP_NOFS,
1982 PAGE_KERNEL);
1983 m->front_is_vmalloc = true;
1984 } else {
1985 m->front.iov_base = kmalloc(front_len, GFP_NOFS);
1986 }
1987 if (m->front.iov_base == NULL) {
1988 pr_err("msg_new can't allocate %d bytes\n",
1989 front_len);
1990 goto out2;
1991 }
1992 } else {
1993 m->front.iov_base = NULL;
1994 }
1995 m->front.iov_len = front_len;
1996
1997 /* middle */
1998 m->middle = NULL;
1999
2000 /* data */
2001 m->nr_pages = calc_pages_for(page_off, page_len);
2002 m->pages = pages;
2003
2004 dout("ceph_msg_new %p page %d~%d -> %d\n", m, page_off, page_len,
2005 m->nr_pages);
2006 return m;
2007
2008out2:
2009 ceph_msg_put(m);
2010out:
2011 pr_err("msg_new can't create type %d len %d\n", type, front_len);
2012 return ERR_PTR(-ENOMEM);
2013}
2014
2015/*
2016 * Generic message allocator, for incoming messages.
2017 */
2018struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
2019 struct ceph_msg_header *hdr)
2020{
2021 int type = le16_to_cpu(hdr->type);
2022 int front_len = le32_to_cpu(hdr->front_len);
2023 struct ceph_msg *msg = ceph_msg_new(type, front_len, 0, 0, NULL);
2024
2025 if (!msg) {
2026 pr_err("unable to allocate msg type %d len %d\n",
2027 type, front_len);
2028 return ERR_PTR(-ENOMEM);
2029 }
2030 return msg;
2031}
2032
2033/*
2034 * Allocate "middle" portion of a message, if it is needed and wasn't
2035 * allocated by alloc_msg. This allows us to read a small fixed-size
2036 * per-type header in the front and then gracefully fail (i.e.,
2037 * propagate the error to the caller based on info in the front) when
2038 * the middle is too large.
2039 */
2040int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2041{
2042 int type = le16_to_cpu(msg->hdr.type);
2043 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2044
2045 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2046 ceph_msg_type_name(type), middle_len);
2047 BUG_ON(!middle_len);
2048 BUG_ON(msg->middle);
2049
b6c1d5b8 2050 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
31b8006e
SW		 2051 if (!msg->middle)
2052 return -ENOMEM;
2053 return 0;
2054}
2055
2056
2057/*
2058 * Free a generically kmalloc'd message.
2059 */
2060void ceph_msg_kfree(struct ceph_msg *m)
2061{
2062 dout("msg_kfree %p\n", m);
2063 if (m->front_is_vmalloc)
2064 vfree(m->front.iov_base);
2065 else
2066 kfree(m->front.iov_base);
2067 kfree(m);
2068}
2069
2070/*
2071 * Drop a msg ref. Destroy as needed.
2072 */
2073void ceph_msg_put(struct ceph_msg *m)
2074{
2075 dout("ceph_msg_put %p %d -> %d\n", m, atomic_read(&m->nref),
2076 atomic_read(&m->nref)-1);
2077 if (atomic_read(&m->nref) <= 0) {
2078 pr_err("bad ceph_msg_put on %p %llu %d=%s %d+%d\n",
2079 m, le64_to_cpu(m->hdr.seq),
2080 le16_to_cpu(m->hdr.type),
2081 ceph_msg_type_name(le16_to_cpu(m->hdr.type)),
2082 le32_to_cpu(m->hdr.front_len),
2083 le32_to_cpu(m->hdr.data_len));
2084 WARN_ON(1);
2085 }
2086 if (atomic_dec_and_test(&m->nref)) {
2087 dout("ceph_msg_put last one on %p\n", m);
2088 WARN_ON(!list_empty(&m->list_head));
2089
2090 /* drop middle, data, if any */
2091 if (m->middle) {
2092 ceph_buffer_put(m->middle);
2093 m->middle = NULL;
2094 }
2095 m->nr_pages = 0;
2096 m->pages = NULL;
2097
2098 if (m->pool)
2099 ceph_msgpool_put(m->pool, m);
2100 else
2101 ceph_msg_kfree(m);
2102 }
2103}
Linux kernel - Deliverables
This page took 0.109554 seconds and 5 git commands to generate.