1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/bvec.h>
5 #include <linux/crc32c.h>
7 #include <linux/socket.h>
10 #include <linux/ceph/ceph_features.h>
11 #include <linux/ceph/decode.h>
12 #include <linux/ceph/libceph.h>
13 #include <linux/ceph/messenger.h>
15 /* static tag bytes (protocol control messages) */
16 static char tag_msg = CEPH_MSGR_TAG_MSG;
17 static char tag_ack = CEPH_MSGR_TAG_ACK;
18 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
19 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
22 * If @buf is NULL, discard up to @len bytes.
24 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
26 struct kvec iov = {buf, len};
27 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
31 msg.msg_flags |= MSG_TRUNC;
33 iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, len);
34 r = sock_recvmsg(sock, &msg, msg.msg_flags);
40 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
41 int page_offset, size_t length)
43 struct bio_vec bvec = {
45 .bv_offset = page_offset,
48 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
51 BUG_ON(page_offset + length > PAGE_SIZE);
52 iov_iter_bvec(&msg.msg_iter, READ, &bvec, 1, length);
53 r = sock_recvmsg(sock, &msg, msg.msg_flags);
60 * write something. @more is true if caller will be sending more data
63 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
64 size_t kvlen, size_t len, bool more)
66 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
70 msg.msg_flags |= MSG_MORE;
72 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
74 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
81 * @more: either or both of MSG_MORE and MSG_SENDPAGE_NOTLAST
83 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
84 int offset, size_t size, int more)
86 ssize_t (*sendpage)(struct socket *sock, struct page *page,
87 int offset, size_t size, int flags);
88 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more;
92 * sendpage cannot properly handle pages with page_count == 0,
93 * we need to fall back to sendmsg if that's the case.
95 * Same goes for slab pages: skb_can_coalesce() allows
96 * coalescing neighboring slab objects into a single frag which
97 * triggers one of hardened usercopy checks.
99 if (sendpage_ok(page))
100 sendpage = sock->ops->sendpage;
102 sendpage = sock_no_sendpage;
104 ret = sendpage(sock, page, offset, size, flags);
111 static void con_out_kvec_reset(struct ceph_connection *con)
113 BUG_ON(con->v1.out_skip);
115 con->v1.out_kvec_left = 0;
116 con->v1.out_kvec_bytes = 0;
117 con->v1.out_kvec_cur = &con->v1.out_kvec[0];
120 static void con_out_kvec_add(struct ceph_connection *con,
121 size_t size, void *data)
123 int index = con->v1.out_kvec_left;
125 BUG_ON(con->v1.out_skip);
126 BUG_ON(index >= ARRAY_SIZE(con->v1.out_kvec));
128 con->v1.out_kvec[index].iov_len = size;
129 con->v1.out_kvec[index].iov_base = data;
130 con->v1.out_kvec_left++;
131 con->v1.out_kvec_bytes += size;
135 * Chop off a kvec from the end. Return residual number of bytes for
136 * that kvec, i.e. how many bytes would have been written if the kvec
139 static int con_out_kvec_skip(struct ceph_connection *con)
143 if (con->v1.out_kvec_bytes > 0) {
144 skip = con->v1.out_kvec_cur[con->v1.out_kvec_left - 1].iov_len;
145 BUG_ON(con->v1.out_kvec_bytes < skip);
146 BUG_ON(!con->v1.out_kvec_left);
147 con->v1.out_kvec_bytes -= skip;
148 con->v1.out_kvec_left--;
154 static size_t sizeof_footer(struct ceph_connection *con)
156 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
157 sizeof(struct ceph_msg_footer) :
158 sizeof(struct ceph_msg_footer_old);
161 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
163 /* Initialize data cursor */
165 ceph_msg_data_cursor_init(&msg->cursor, msg, data_len);
169 * Prepare footer for currently outgoing message, and finish things
170 * off. Assumes out_kvec* are already valid.. we just add on to the end.
172 static void prepare_write_message_footer(struct ceph_connection *con)
174 struct ceph_msg *m = con->out_msg;
176 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
178 dout("prepare_write_message_footer %p\n", con);
179 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
180 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
181 if (con->ops->sign_message)
182 con->ops->sign_message(m);
186 m->old_footer.flags = m->footer.flags;
188 con->v1.out_more = m->more_to_follow;
189 con->v1.out_msg_done = true;
193 * Prepare headers for the next outgoing message.
195 static void prepare_write_message(struct ceph_connection *con)
200 con_out_kvec_reset(con);
201 con->v1.out_msg_done = false;
203 /* Sneak an ack in there first? If we can get it into the same
204 * TCP packet that's a good thing. */
205 if (con->in_seq > con->in_seq_acked) {
206 con->in_seq_acked = con->in_seq;
207 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
208 con->v1.out_temp_ack = cpu_to_le64(con->in_seq_acked);
209 con_out_kvec_add(con, sizeof(con->v1.out_temp_ack),
210 &con->v1.out_temp_ack);
213 ceph_con_get_out_msg(con);
216 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
217 m, con->out_seq, le16_to_cpu(m->hdr.type),
218 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
220 WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
221 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
223 /* tag + hdr + front + middle */
224 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
225 con_out_kvec_add(con, sizeof(con->v1.out_hdr), &con->v1.out_hdr);
226 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
229 con_out_kvec_add(con, m->middle->vec.iov_len,
230 m->middle->vec.iov_base);
232 /* fill in hdr crc and finalize hdr */
233 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
234 con->out_msg->hdr.crc = cpu_to_le32(crc);
235 memcpy(&con->v1.out_hdr, &con->out_msg->hdr, sizeof(con->v1.out_hdr));
237 /* fill in front and middle crc, footer */
238 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
239 con->out_msg->footer.front_crc = cpu_to_le32(crc);
241 crc = crc32c(0, m->middle->vec.iov_base,
242 m->middle->vec.iov_len);
243 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
245 con->out_msg->footer.middle_crc = 0;
246 dout("%s front_crc %u middle_crc %u\n", __func__,
247 le32_to_cpu(con->out_msg->footer.front_crc),
248 le32_to_cpu(con->out_msg->footer.middle_crc));
249 con->out_msg->footer.flags = 0;
251 /* is there a data payload? */
252 con->out_msg->footer.data_crc = 0;
253 if (m->data_length) {
254 prepare_message_data(con->out_msg, m->data_length);
255 con->v1.out_more = 1; /* data + footer will follow */
257 /* no, queue up footer too and be done */
258 prepare_write_message_footer(con);
261 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
267 static void prepare_write_ack(struct ceph_connection *con)
269 dout("prepare_write_ack %p %llu -> %llu\n", con,
270 con->in_seq_acked, con->in_seq);
271 con->in_seq_acked = con->in_seq;
273 con_out_kvec_reset(con);
275 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
277 con->v1.out_temp_ack = cpu_to_le64(con->in_seq_acked);
278 con_out_kvec_add(con, sizeof(con->v1.out_temp_ack),
279 &con->v1.out_temp_ack);
281 con->v1.out_more = 1; /* more will follow.. eventually.. */
282 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
286 * Prepare to share the seq during handshake
288 static void prepare_write_seq(struct ceph_connection *con)
290 dout("prepare_write_seq %p %llu -> %llu\n", con,
291 con->in_seq_acked, con->in_seq);
292 con->in_seq_acked = con->in_seq;
294 con_out_kvec_reset(con);
296 con->v1.out_temp_ack = cpu_to_le64(con->in_seq_acked);
297 con_out_kvec_add(con, sizeof(con->v1.out_temp_ack),
298 &con->v1.out_temp_ack);
300 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
304 * Prepare to write keepalive byte.
306 static void prepare_write_keepalive(struct ceph_connection *con)
308 dout("prepare_write_keepalive %p\n", con);
309 con_out_kvec_reset(con);
310 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
311 struct timespec64 now;
313 ktime_get_real_ts64(&now);
314 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
315 ceph_encode_timespec64(&con->v1.out_temp_keepalive2, &now);
316 con_out_kvec_add(con, sizeof(con->v1.out_temp_keepalive2),
317 &con->v1.out_temp_keepalive2);
319 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
321 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
325 * Connection negotiation.
328 static int get_connect_authorizer(struct ceph_connection *con)
330 struct ceph_auth_handshake *auth;
333 if (!con->ops->get_authorizer) {
335 con->v1.out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
336 con->v1.out_connect.authorizer_len = 0;
340 auth = con->ops->get_authorizer(con, &auth_proto, con->v1.auth_retry);
342 return PTR_ERR(auth);
345 con->v1.out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
346 con->v1.out_connect.authorizer_len =
347 cpu_to_le32(auth->authorizer_buf_len);
352 * We connected to a peer and are saying hello.
354 static void prepare_write_banner(struct ceph_connection *con)
356 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
357 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
358 &con->msgr->my_enc_addr);
360 con->v1.out_more = 0;
361 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
364 static void __prepare_write_connect(struct ceph_connection *con)
366 con_out_kvec_add(con, sizeof(con->v1.out_connect),
367 &con->v1.out_connect);
369 con_out_kvec_add(con, con->v1.auth->authorizer_buf_len,
370 con->v1.auth->authorizer_buf);
372 con->v1.out_more = 0;
373 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
376 static int prepare_write_connect(struct ceph_connection *con)
378 unsigned int global_seq = ceph_get_global_seq(con->msgr, 0);
382 switch (con->peer_name.type) {
383 case CEPH_ENTITY_TYPE_MON:
384 proto = CEPH_MONC_PROTOCOL;
386 case CEPH_ENTITY_TYPE_OSD:
387 proto = CEPH_OSDC_PROTOCOL;
389 case CEPH_ENTITY_TYPE_MDS:
390 proto = CEPH_MDSC_PROTOCOL;
396 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
397 con->v1.connect_seq, global_seq, proto);
399 con->v1.out_connect.features =
400 cpu_to_le64(from_msgr(con->msgr)->supported_features);
401 con->v1.out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
402 con->v1.out_connect.connect_seq = cpu_to_le32(con->v1.connect_seq);
403 con->v1.out_connect.global_seq = cpu_to_le32(global_seq);
404 con->v1.out_connect.protocol_version = cpu_to_le32(proto);
405 con->v1.out_connect.flags = 0;
407 ret = get_connect_authorizer(con);
411 __prepare_write_connect(con);
416 * write as much of pending kvecs to the socket as we can.
418 * 0 -> socket full, but more to do
421 static int write_partial_kvec(struct ceph_connection *con)
425 dout("write_partial_kvec %p %d left\n", con, con->v1.out_kvec_bytes);
426 while (con->v1.out_kvec_bytes > 0) {
427 ret = ceph_tcp_sendmsg(con->sock, con->v1.out_kvec_cur,
428 con->v1.out_kvec_left,
429 con->v1.out_kvec_bytes,
433 con->v1.out_kvec_bytes -= ret;
434 if (!con->v1.out_kvec_bytes)
437 /* account for full iov entries consumed */
438 while (ret >= con->v1.out_kvec_cur->iov_len) {
439 BUG_ON(!con->v1.out_kvec_left);
440 ret -= con->v1.out_kvec_cur->iov_len;
441 con->v1.out_kvec_cur++;
442 con->v1.out_kvec_left--;
444 /* and for a partially-consumed entry */
446 con->v1.out_kvec_cur->iov_len -= ret;
447 con->v1.out_kvec_cur->iov_base += ret;
450 con->v1.out_kvec_left = 0;
453 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
454 con->v1.out_kvec_bytes, con->v1.out_kvec_left, ret);
455 return ret; /* done! */
459 * Write as much message data payload as we can. If we finish, queue
461 * 1 -> done, footer is now queued in out_kvec[].
462 * 0 -> socket full, but more to do
465 static int write_partial_message_data(struct ceph_connection *con)
467 struct ceph_msg *msg = con->out_msg;
468 struct ceph_msg_data_cursor *cursor = &msg->cursor;
469 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
470 int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
473 dout("%s %p msg %p\n", __func__, con, msg);
475 if (!msg->num_data_items)
479 * Iterate through each page that contains data to be
480 * written, and send as much as possible for each.
482 * If we are calculating the data crc (the default), we will
483 * need to map the page. If we have no pages, they have
484 * been revoked, so use the zero page.
486 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
487 while (cursor->total_resid) {
493 if (!cursor->resid) {
494 ceph_msg_data_advance(cursor, 0);
498 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
499 if (length == cursor->total_resid)
501 ret = ceph_tcp_sendpage(con->sock, page, page_offset, length,
505 msg->footer.data_crc = cpu_to_le32(crc);
509 if (do_datacrc && cursor->need_crc)
510 crc = ceph_crc32c_page(crc, page, page_offset, length);
511 ceph_msg_data_advance(cursor, (size_t)ret);
514 dout("%s %p msg %p done\n", __func__, con, msg);
516 /* prepare and queue up footer, too */
518 msg->footer.data_crc = cpu_to_le32(crc);
520 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
521 con_out_kvec_reset(con);
522 prepare_write_message_footer(con);
524 return 1; /* must return > 0 to indicate success */
530 static int write_partial_skip(struct ceph_connection *con)
532 int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
535 dout("%s %p %d left\n", __func__, con, con->v1.out_skip);
536 while (con->v1.out_skip > 0) {
537 size_t size = min(con->v1.out_skip, (int)PAGE_SIZE);
539 if (size == con->v1.out_skip)
541 ret = ceph_tcp_sendpage(con->sock, ceph_zero_page, 0, size,
545 con->v1.out_skip -= ret;
553 * Prepare to read connection handshake, or an ack.
555 static void prepare_read_banner(struct ceph_connection *con)
557 dout("prepare_read_banner %p\n", con);
558 con->v1.in_base_pos = 0;
561 static void prepare_read_connect(struct ceph_connection *con)
563 dout("prepare_read_connect %p\n", con);
564 con->v1.in_base_pos = 0;
567 static void prepare_read_ack(struct ceph_connection *con)
569 dout("prepare_read_ack %p\n", con);
570 con->v1.in_base_pos = 0;
573 static void prepare_read_seq(struct ceph_connection *con)
575 dout("prepare_read_seq %p\n", con);
576 con->v1.in_base_pos = 0;
577 con->v1.in_tag = CEPH_MSGR_TAG_SEQ;
580 static void prepare_read_tag(struct ceph_connection *con)
582 dout("prepare_read_tag %p\n", con);
583 con->v1.in_base_pos = 0;
584 con->v1.in_tag = CEPH_MSGR_TAG_READY;
587 static void prepare_read_keepalive_ack(struct ceph_connection *con)
589 dout("prepare_read_keepalive_ack %p\n", con);
590 con->v1.in_base_pos = 0;
594 * Prepare to read a message.
596 static int prepare_read_message(struct ceph_connection *con)
598 dout("prepare_read_message %p\n", con);
599 BUG_ON(con->in_msg != NULL);
600 con->v1.in_base_pos = 0;
601 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
605 static int read_partial(struct ceph_connection *con,
606 int end, int size, void *object)
608 while (con->v1.in_base_pos < end) {
609 int left = end - con->v1.in_base_pos;
610 int have = size - left;
611 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
614 con->v1.in_base_pos += ret;
620 * Read all or part of the connect-side handshake on a new connection
622 static int read_partial_banner(struct ceph_connection *con)
628 dout("read_partial_banner %p at %d\n", con, con->v1.in_base_pos);
631 size = strlen(CEPH_BANNER);
633 ret = read_partial(con, end, size, con->v1.in_banner);
637 size = sizeof(con->v1.actual_peer_addr);
639 ret = read_partial(con, end, size, &con->v1.actual_peer_addr);
642 ceph_decode_banner_addr(&con->v1.actual_peer_addr);
644 size = sizeof(con->v1.peer_addr_for_me);
646 ret = read_partial(con, end, size, &con->v1.peer_addr_for_me);
649 ceph_decode_banner_addr(&con->v1.peer_addr_for_me);
655 static int read_partial_connect(struct ceph_connection *con)
661 dout("read_partial_connect %p at %d\n", con, con->v1.in_base_pos);
663 size = sizeof(con->v1.in_reply);
665 ret = read_partial(con, end, size, &con->v1.in_reply);
670 size = le32_to_cpu(con->v1.in_reply.authorizer_len);
671 if (size > con->v1.auth->authorizer_reply_buf_len) {
672 pr_err("authorizer reply too big: %d > %zu\n", size,
673 con->v1.auth->authorizer_reply_buf_len);
679 ret = read_partial(con, end, size,
680 con->v1.auth->authorizer_reply_buf);
685 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
686 con, con->v1.in_reply.tag,
687 le32_to_cpu(con->v1.in_reply.connect_seq),
688 le32_to_cpu(con->v1.in_reply.global_seq));
694 * Verify the hello banner looks okay.
696 static int verify_hello(struct ceph_connection *con)
698 if (memcmp(con->v1.in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
699 pr_err("connect to %s got bad banner\n",
700 ceph_pr_addr(&con->peer_addr));
701 con->error_msg = "protocol error, bad banner";
707 static int process_banner(struct ceph_connection *con)
709 struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
711 dout("process_banner on %p\n", con);
713 if (verify_hello(con) < 0)
717 * Make sure the other end is who we wanted. note that the other
718 * end may not yet know their ip address, so if it's 0.0.0.0, give
719 * them the benefit of the doubt.
721 if (memcmp(&con->peer_addr, &con->v1.actual_peer_addr,
722 sizeof(con->peer_addr)) != 0 &&
723 !(ceph_addr_is_blank(&con->v1.actual_peer_addr) &&
724 con->v1.actual_peer_addr.nonce == con->peer_addr.nonce)) {
725 pr_warn("wrong peer, want %s/%u, got %s/%u\n",
726 ceph_pr_addr(&con->peer_addr),
727 le32_to_cpu(con->peer_addr.nonce),
728 ceph_pr_addr(&con->v1.actual_peer_addr),
729 le32_to_cpu(con->v1.actual_peer_addr.nonce));
730 con->error_msg = "wrong peer at address";
735 * did we learn our address?
737 if (ceph_addr_is_blank(my_addr)) {
738 memcpy(&my_addr->in_addr,
739 &con->v1.peer_addr_for_me.in_addr,
740 sizeof(con->v1.peer_addr_for_me.in_addr));
741 ceph_addr_set_port(my_addr, 0);
742 ceph_encode_my_addr(con->msgr);
743 dout("process_banner learned my addr is %s\n",
744 ceph_pr_addr(my_addr));
750 static int process_connect(struct ceph_connection *con)
752 u64 sup_feat = from_msgr(con->msgr)->supported_features;
753 u64 req_feat = from_msgr(con->msgr)->required_features;
754 u64 server_feat = le64_to_cpu(con->v1.in_reply.features);
757 dout("process_connect on %p tag %d\n", con, con->v1.in_tag);
760 int len = le32_to_cpu(con->v1.in_reply.authorizer_len);
763 * Any connection that defines ->get_authorizer()
764 * should also define ->add_authorizer_challenge() and
765 * ->verify_authorizer_reply().
767 * See get_connect_authorizer().
769 if (con->v1.in_reply.tag ==
770 CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
771 ret = con->ops->add_authorizer_challenge(
772 con, con->v1.auth->authorizer_reply_buf, len);
776 con_out_kvec_reset(con);
777 __prepare_write_connect(con);
778 prepare_read_connect(con);
783 ret = con->ops->verify_authorizer_reply(con);
785 con->error_msg = "bad authorize reply";
791 switch (con->v1.in_reply.tag) {
792 case CEPH_MSGR_TAG_FEATURES:
793 pr_err("%s%lld %s feature set mismatch,"
794 " my %llx < server's %llx, missing %llx\n",
795 ENTITY_NAME(con->peer_name),
796 ceph_pr_addr(&con->peer_addr),
797 sup_feat, server_feat, server_feat & ~sup_feat);
798 con->error_msg = "missing required protocol features";
801 case CEPH_MSGR_TAG_BADPROTOVER:
802 pr_err("%s%lld %s protocol version mismatch,"
803 " my %d != server's %d\n",
804 ENTITY_NAME(con->peer_name),
805 ceph_pr_addr(&con->peer_addr),
806 le32_to_cpu(con->v1.out_connect.protocol_version),
807 le32_to_cpu(con->v1.in_reply.protocol_version));
808 con->error_msg = "protocol version mismatch";
811 case CEPH_MSGR_TAG_BADAUTHORIZER:
812 con->v1.auth_retry++;
813 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
815 if (con->v1.auth_retry == 2) {
816 con->error_msg = "connect authorization failure";
819 con_out_kvec_reset(con);
820 ret = prepare_write_connect(con);
823 prepare_read_connect(con);
826 case CEPH_MSGR_TAG_RESETSESSION:
828 * If we connected with a large connect_seq but the peer
829 * has no record of a session with us (no connection, or
830 * connect_seq == 0), they will send RESETSESION to indicate
831 * that they must have reset their session, and may have
834 dout("process_connect got RESET peer seq %u\n",
835 le32_to_cpu(con->v1.in_reply.connect_seq));
836 pr_info("%s%lld %s session reset\n",
837 ENTITY_NAME(con->peer_name),
838 ceph_pr_addr(&con->peer_addr));
839 ceph_con_reset_session(con);
840 con_out_kvec_reset(con);
841 ret = prepare_write_connect(con);
844 prepare_read_connect(con);
846 /* Tell ceph about it. */
847 mutex_unlock(&con->mutex);
848 if (con->ops->peer_reset)
849 con->ops->peer_reset(con);
850 mutex_lock(&con->mutex);
851 if (con->state != CEPH_CON_S_V1_CONNECT_MSG)
855 case CEPH_MSGR_TAG_RETRY_SESSION:
857 * If we sent a smaller connect_seq than the peer has, try
858 * again with a larger value.
860 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
861 le32_to_cpu(con->v1.out_connect.connect_seq),
862 le32_to_cpu(con->v1.in_reply.connect_seq));
863 con->v1.connect_seq = le32_to_cpu(con->v1.in_reply.connect_seq);
864 con_out_kvec_reset(con);
865 ret = prepare_write_connect(con);
868 prepare_read_connect(con);
871 case CEPH_MSGR_TAG_RETRY_GLOBAL:
873 * If we sent a smaller global_seq than the peer has, try
874 * again with a larger value.
876 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
877 con->v1.peer_global_seq,
878 le32_to_cpu(con->v1.in_reply.global_seq));
879 ceph_get_global_seq(con->msgr,
880 le32_to_cpu(con->v1.in_reply.global_seq));
881 con_out_kvec_reset(con);
882 ret = prepare_write_connect(con);
885 prepare_read_connect(con);
888 case CEPH_MSGR_TAG_SEQ:
889 case CEPH_MSGR_TAG_READY:
890 if (req_feat & ~server_feat) {
891 pr_err("%s%lld %s protocol feature mismatch,"
892 " my required %llx > server's %llx, need %llx\n",
893 ENTITY_NAME(con->peer_name),
894 ceph_pr_addr(&con->peer_addr),
895 req_feat, server_feat, req_feat & ~server_feat);
896 con->error_msg = "missing required protocol features";
900 WARN_ON(con->state != CEPH_CON_S_V1_CONNECT_MSG);
901 con->state = CEPH_CON_S_OPEN;
902 con->v1.auth_retry = 0; /* we authenticated; clear flag */
903 con->v1.peer_global_seq =
904 le32_to_cpu(con->v1.in_reply.global_seq);
905 con->v1.connect_seq++;
906 con->peer_features = server_feat;
907 dout("process_connect got READY gseq %d cseq %d (%d)\n",
908 con->v1.peer_global_seq,
909 le32_to_cpu(con->v1.in_reply.connect_seq),
910 con->v1.connect_seq);
911 WARN_ON(con->v1.connect_seq !=
912 le32_to_cpu(con->v1.in_reply.connect_seq));
914 if (con->v1.in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
915 ceph_con_flag_set(con, CEPH_CON_F_LOSSYTX);
917 con->delay = 0; /* reset backoff memory */
919 if (con->v1.in_reply.tag == CEPH_MSGR_TAG_SEQ) {
920 prepare_write_seq(con);
921 prepare_read_seq(con);
923 prepare_read_tag(con);
927 case CEPH_MSGR_TAG_WAIT:
929 * If there is a connection race (we are opening
930 * connections to each other), one of us may just have
931 * to WAIT. This shouldn't happen if we are the
934 con->error_msg = "protocol error, got WAIT as client";
938 con->error_msg = "protocol error, garbage tag during connect";
945 * read (part of) an ack
947 static int read_partial_ack(struct ceph_connection *con)
949 int size = sizeof(con->v1.in_temp_ack);
952 return read_partial(con, end, size, &con->v1.in_temp_ack);
956 * We can finally discard anything that's been acked.
958 static void process_ack(struct ceph_connection *con)
960 u64 ack = le64_to_cpu(con->v1.in_temp_ack);
962 if (con->v1.in_tag == CEPH_MSGR_TAG_ACK)
963 ceph_con_discard_sent(con, ack);
965 ceph_con_discard_requeued(con, ack);
967 prepare_read_tag(con);
970 static int read_partial_message_section(struct ceph_connection *con,
971 struct kvec *section,
972 unsigned int sec_len, u32 *crc)
978 while (section->iov_len < sec_len) {
979 BUG_ON(section->iov_base == NULL);
980 left = sec_len - section->iov_len;
981 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
982 section->iov_len, left);
985 section->iov_len += ret;
987 if (section->iov_len == sec_len)
988 *crc = crc32c(0, section->iov_base, section->iov_len);
993 static int read_partial_msg_data(struct ceph_connection *con)
995 struct ceph_msg *msg = con->in_msg;
996 struct ceph_msg_data_cursor *cursor = &msg->cursor;
997 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1004 if (!msg->num_data_items)
1008 crc = con->in_data_crc;
1009 while (cursor->total_resid) {
1010 if (!cursor->resid) {
1011 ceph_msg_data_advance(cursor, 0);
1015 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
1016 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
1019 con->in_data_crc = crc;
1025 crc = ceph_crc32c_page(crc, page, page_offset, ret);
1026 ceph_msg_data_advance(cursor, (size_t)ret);
1029 con->in_data_crc = crc;
1031 return 1; /* must return > 0 to indicate success */
1035 * read (part of) a message.
1037 static int read_partial_message(struct ceph_connection *con)
1039 struct ceph_msg *m = con->in_msg;
1043 unsigned int front_len, middle_len, data_len;
1044 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1045 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
1049 dout("read_partial_message con %p msg %p\n", con, m);
1052 size = sizeof(con->v1.in_hdr);
1054 ret = read_partial(con, end, size, &con->v1.in_hdr);
1058 crc = crc32c(0, &con->v1.in_hdr, offsetof(struct ceph_msg_header, crc));
1059 if (cpu_to_le32(crc) != con->v1.in_hdr.crc) {
1060 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
1061 crc, con->v1.in_hdr.crc);
1065 front_len = le32_to_cpu(con->v1.in_hdr.front_len);
1066 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1068 middle_len = le32_to_cpu(con->v1.in_hdr.middle_len);
1069 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
1071 data_len = le32_to_cpu(con->v1.in_hdr.data_len);
1072 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1076 seq = le64_to_cpu(con->v1.in_hdr.seq);
1077 if ((s64)seq - (s64)con->in_seq < 1) {
1078 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1079 ENTITY_NAME(con->peer_name),
1080 ceph_pr_addr(&con->peer_addr),
1081 seq, con->in_seq + 1);
1082 con->v1.in_base_pos = -front_len - middle_len - data_len -
1084 con->v1.in_tag = CEPH_MSGR_TAG_READY;
1086 } else if ((s64)seq - (s64)con->in_seq > 1) {
1087 pr_err("read_partial_message bad seq %lld expected %lld\n",
1088 seq, con->in_seq + 1);
1089 con->error_msg = "bad message sequence # for incoming message";
1093 /* allocate message? */
1097 dout("got hdr type %d front %d data %d\n", con->v1.in_hdr.type,
1098 front_len, data_len);
1099 ret = ceph_con_in_msg_alloc(con, &con->v1.in_hdr, &skip);
1103 BUG_ON((!con->in_msg) ^ skip);
1105 /* skip this message */
1106 dout("alloc_msg said skip message\n");
1107 con->v1.in_base_pos = -front_len - middle_len -
1108 data_len - sizeof_footer(con);
1109 con->v1.in_tag = CEPH_MSGR_TAG_READY;
1114 BUG_ON(!con->in_msg);
1115 BUG_ON(con->in_msg->con != con);
1117 m->front.iov_len = 0; /* haven't read it yet */
1119 m->middle->vec.iov_len = 0;
1121 /* prepare for data payload, if any */
1124 prepare_message_data(con->in_msg, data_len);
1128 ret = read_partial_message_section(con, &m->front, front_len,
1129 &con->in_front_crc);
1135 ret = read_partial_message_section(con, &m->middle->vec,
1137 &con->in_middle_crc);
1144 ret = read_partial_msg_data(con);
1150 size = sizeof_footer(con);
1152 ret = read_partial(con, end, size, &m->footer);
1157 m->footer.flags = m->old_footer.flags;
1161 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1162 m, front_len, m->footer.front_crc, middle_len,
1163 m->footer.middle_crc, data_len, m->footer.data_crc);
1166 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1167 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1168 m, con->in_front_crc, m->footer.front_crc);
1171 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1172 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1173 m, con->in_middle_crc, m->footer.middle_crc);
1177 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1178 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1179 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1180 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1184 if (need_sign && con->ops->check_message_signature &&
1185 con->ops->check_message_signature(m)) {
1186 pr_err("read_partial_message %p signature check failed\n", m);
1190 return 1; /* done! */
1193 static int read_keepalive_ack(struct ceph_connection *con)
1195 struct ceph_timespec ceph_ts;
1196 size_t size = sizeof(ceph_ts);
1197 int ret = read_partial(con, size, size, &ceph_ts);
1200 ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
1201 prepare_read_tag(con);
1206 * Read what we can from the socket.
1208 int ceph_con_v1_try_read(struct ceph_connection *con)
1213 dout("try_read start %p state %d\n", con, con->state);
1214 if (con->state != CEPH_CON_S_V1_BANNER &&
1215 con->state != CEPH_CON_S_V1_CONNECT_MSG &&
1216 con->state != CEPH_CON_S_OPEN)
1221 dout("try_read tag %d in_base_pos %d\n", con->v1.in_tag,
1222 con->v1.in_base_pos);
1224 if (con->state == CEPH_CON_S_V1_BANNER) {
1225 ret = read_partial_banner(con);
1228 ret = process_banner(con);
1232 con->state = CEPH_CON_S_V1_CONNECT_MSG;
1235 * Received banner is good, exchange connection info.
1236 * Do not reset out_kvec, as sending our banner raced
1237 * with receiving peer banner after connect completed.
1239 ret = prepare_write_connect(con);
1242 prepare_read_connect(con);
1244 /* Send connection info before awaiting response */
1248 if (con->state == CEPH_CON_S_V1_CONNECT_MSG) {
1249 ret = read_partial_connect(con);
1252 ret = process_connect(con);
1258 WARN_ON(con->state != CEPH_CON_S_OPEN);
1260 if (con->v1.in_base_pos < 0) {
1262 * skipping + discarding content.
1264 ret = ceph_tcp_recvmsg(con->sock, NULL, -con->v1.in_base_pos);
1267 dout("skipped %d / %d bytes\n", ret, -con->v1.in_base_pos);
1268 con->v1.in_base_pos += ret;
1269 if (con->v1.in_base_pos)
1272 if (con->v1.in_tag == CEPH_MSGR_TAG_READY) {
1276 ret = ceph_tcp_recvmsg(con->sock, &con->v1.in_tag, 1);
1279 dout("try_read got tag %d\n", con->v1.in_tag);
1280 switch (con->v1.in_tag) {
1281 case CEPH_MSGR_TAG_MSG:
1282 prepare_read_message(con);
1284 case CEPH_MSGR_TAG_ACK:
1285 prepare_read_ack(con);
1287 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
1288 prepare_read_keepalive_ack(con);
1290 case CEPH_MSGR_TAG_CLOSE:
1291 ceph_con_close_socket(con);
1292 con->state = CEPH_CON_S_CLOSED;
1298 if (con->v1.in_tag == CEPH_MSGR_TAG_MSG) {
1299 ret = read_partial_message(con);
1303 con->error_msg = "bad crc/signature";
1309 con->error_msg = "io error";
1314 if (con->v1.in_tag == CEPH_MSGR_TAG_READY)
1316 ceph_con_process_message(con);
1317 if (con->state == CEPH_CON_S_OPEN)
1318 prepare_read_tag(con);
1321 if (con->v1.in_tag == CEPH_MSGR_TAG_ACK ||
1322 con->v1.in_tag == CEPH_MSGR_TAG_SEQ) {
1324 * the final handshake seq exchange is semantically
1325 * equivalent to an ACK
1327 ret = read_partial_ack(con);
1333 if (con->v1.in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
1334 ret = read_keepalive_ack(con);
1341 dout("try_read done on %p ret %d\n", con, ret);
1345 pr_err("try_read bad tag %d\n", con->v1.in_tag);
1346 con->error_msg = "protocol error, garbage tag";
1352 * Write something to the socket. Called in a worker thread when the
1353 * socket appears to be writeable and we have something ready to send.
1355 int ceph_con_v1_try_write(struct ceph_connection *con)
1359 dout("try_write start %p state %d\n", con, con->state);
1360 if (con->state != CEPH_CON_S_PREOPEN &&
1361 con->state != CEPH_CON_S_V1_BANNER &&
1362 con->state != CEPH_CON_S_V1_CONNECT_MSG &&
1363 con->state != CEPH_CON_S_OPEN)
1366 /* open the socket first? */
1367 if (con->state == CEPH_CON_S_PREOPEN) {
1369 con->state = CEPH_CON_S_V1_BANNER;
1371 con_out_kvec_reset(con);
1372 prepare_write_banner(con);
1373 prepare_read_banner(con);
1375 BUG_ON(con->in_msg);
1376 con->v1.in_tag = CEPH_MSGR_TAG_READY;
1377 dout("try_write initiating connect on %p new state %d\n",
1379 ret = ceph_tcp_connect(con);
1381 con->error_msg = "connect error";
1387 dout("try_write out_kvec_bytes %d\n", con->v1.out_kvec_bytes);
1390 /* kvec data queued? */
1391 if (con->v1.out_kvec_left) {
1392 ret = write_partial_kvec(con);
1396 if (con->v1.out_skip) {
1397 ret = write_partial_skip(con);
1404 if (con->v1.out_msg_done) {
1405 ceph_msg_put(con->out_msg);
1406 con->out_msg = NULL; /* we're done with this one */
1410 ret = write_partial_message_data(con);
1412 goto more; /* we need to send the footer, too! */
1416 dout("try_write write_partial_message_data err %d\n",
1423 if (con->state == CEPH_CON_S_OPEN) {
1424 if (ceph_con_flag_test_and_clear(con,
1425 CEPH_CON_F_KEEPALIVE_PENDING)) {
1426 prepare_write_keepalive(con);
1429 /* is anything else pending? */
1430 if (!list_empty(&con->out_queue)) {
1431 prepare_write_message(con);
1434 if (con->in_seq > con->in_seq_acked) {
1435 prepare_write_ack(con);
1440 /* Nothing to do! */
1441 ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
1442 dout("try_write nothing else to write.\n");
1445 dout("try_write done on %p ret %d\n", con, ret);
1449 void ceph_con_v1_revoke(struct ceph_connection *con)
1451 struct ceph_msg *msg = con->out_msg;
1453 WARN_ON(con->v1.out_skip);
1455 if (con->v1.out_msg_done) {
1456 con->v1.out_skip += con_out_kvec_skip(con);
1458 WARN_ON(!msg->data_length);
1459 con->v1.out_skip += sizeof_footer(con);
1461 /* data, middle, front */
1462 if (msg->data_length)
1463 con->v1.out_skip += msg->cursor.total_resid;
1465 con->v1.out_skip += con_out_kvec_skip(con);
1466 con->v1.out_skip += con_out_kvec_skip(con);
1468 dout("%s con %p out_kvec_bytes %d out_skip %d\n", __func__, con,
1469 con->v1.out_kvec_bytes, con->v1.out_skip);
1472 void ceph_con_v1_revoke_incoming(struct ceph_connection *con)
1474 unsigned int front_len = le32_to_cpu(con->v1.in_hdr.front_len);
1475 unsigned int middle_len = le32_to_cpu(con->v1.in_hdr.middle_len);
1476 unsigned int data_len = le32_to_cpu(con->v1.in_hdr.data_len);
1478 /* skip rest of message */
1479 con->v1.in_base_pos = con->v1.in_base_pos -
1480 sizeof(struct ceph_msg_header) -
1484 sizeof(struct ceph_msg_footer);
1486 con->v1.in_tag = CEPH_MSGR_TAG_READY;
1489 dout("%s con %p in_base_pos %d\n", __func__, con, con->v1.in_base_pos);
1492 bool ceph_con_v1_opened(struct ceph_connection *con)
1494 return con->v1.connect_seq;
1497 void ceph_con_v1_reset_session(struct ceph_connection *con)
1499 con->v1.connect_seq = 0;
1500 con->v1.peer_global_seq = 0;
1503 void ceph_con_v1_reset_protocol(struct ceph_connection *con)
1505 con->v1.out_skip = 0;