1 // SPDX-License-Identifier: GPL-2.0
3 * Ceph msgr2 protocol implementation
5 * Copyright (C) 2020 Ilya Dryomov <idryomov@gmail.com>
8 #include <linux/ceph/ceph_debug.h>
10 #include <crypto/aead.h>
11 #include <crypto/algapi.h> /* for crypto_memneq() */
12 #include <crypto/hash.h>
13 #include <crypto/sha2.h>
14 #include <linux/bvec.h>
15 #include <linux/crc32c.h>
16 #include <linux/net.h>
17 #include <linux/scatterlist.h>
18 #include <linux/socket.h>
19 #include <linux/sched/mm.h>
23 #include <linux/ceph/ceph_features.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/libceph.h>
26 #include <linux/ceph/messenger.h>
28 #include "crypto.h" /* for CEPH_KEY_LEN and CEPH_MAX_CON_SECRET_LEN */
30 #define FRAME_TAG_HELLO 1
31 #define FRAME_TAG_AUTH_REQUEST 2
32 #define FRAME_TAG_AUTH_BAD_METHOD 3
33 #define FRAME_TAG_AUTH_REPLY_MORE 4
34 #define FRAME_TAG_AUTH_REQUEST_MORE 5
35 #define FRAME_TAG_AUTH_DONE 6
36 #define FRAME_TAG_AUTH_SIGNATURE 7
37 #define FRAME_TAG_CLIENT_IDENT 8
38 #define FRAME_TAG_SERVER_IDENT 9
39 #define FRAME_TAG_IDENT_MISSING_FEATURES 10
40 #define FRAME_TAG_SESSION_RECONNECT 11
41 #define FRAME_TAG_SESSION_RESET 12
42 #define FRAME_TAG_SESSION_RETRY 13
43 #define FRAME_TAG_SESSION_RETRY_GLOBAL 14
44 #define FRAME_TAG_SESSION_RECONNECT_OK 15
45 #define FRAME_TAG_WAIT 16
46 #define FRAME_TAG_MESSAGE 17
47 #define FRAME_TAG_KEEPALIVE2 18
48 #define FRAME_TAG_KEEPALIVE2_ACK 19
49 #define FRAME_TAG_ACK 20
51 #define FRAME_LATE_STATUS_ABORTED 0x1
52 #define FRAME_LATE_STATUS_COMPLETE 0xe
53 #define FRAME_LATE_STATUS_ABORTED_MASK 0xf
55 #define IN_S_HANDLE_PREAMBLE 1
56 #define IN_S_HANDLE_CONTROL 2
57 #define IN_S_HANDLE_CONTROL_REMAINDER 3
58 #define IN_S_PREPARE_READ_DATA 4
59 #define IN_S_PREPARE_READ_DATA_CONT 5
60 #define IN_S_PREPARE_READ_ENC_PAGE 6
61 #define IN_S_HANDLE_EPILOGUE 7
62 #define IN_S_FINISH_SKIP 8
64 #define OUT_S_QUEUE_DATA 1
65 #define OUT_S_QUEUE_DATA_CONT 2
66 #define OUT_S_QUEUE_ENC_PAGE 3
67 #define OUT_S_QUEUE_ZEROS 4
68 #define OUT_S_FINISH_MESSAGE 5
69 #define OUT_S_GET_NEXT 6
71 #define CTRL_BODY(p) ((void *)(p) + CEPH_PREAMBLE_LEN)
72 #define FRONT_PAD(p) ((void *)(p) + CEPH_EPILOGUE_SECURE_LEN)
73 #define MIDDLE_PAD(p) (FRONT_PAD(p) + CEPH_GCM_BLOCK_LEN)
74 #define DATA_PAD(p) (MIDDLE_PAD(p) + CEPH_GCM_BLOCK_LEN)
76 #define CEPH_MSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
78 static int do_recvmsg(struct socket *sock, struct iov_iter *it)
80 struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
84 while (iov_iter_count(it)) {
85 ret = sock_recvmsg(sock, &msg, msg.msg_flags);
92 iov_iter_advance(it, ret);
95 WARN_ON(msg_data_left(&msg));
100 * Read as much as possible.
103 * 1 - done, nothing (else) to read
104 * 0 - socket is empty, need to wait
107 static int ceph_tcp_recv(struct ceph_connection *con)
111 dout("%s con %p %s %zu\n", __func__, con,
112 iov_iter_is_discard(&con->v2.in_iter) ? "discard" : "need",
113 iov_iter_count(&con->v2.in_iter));
114 ret = do_recvmsg(con->sock, &con->v2.in_iter);
115 dout("%s con %p ret %d left %zu\n", __func__, con, ret,
116 iov_iter_count(&con->v2.in_iter));
120 static int do_sendmsg(struct socket *sock, struct iov_iter *it)
122 struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
126 while (iov_iter_count(it)) {
127 ret = sock_sendmsg(sock, &msg);
134 iov_iter_advance(it, ret);
137 WARN_ON(msg_data_left(&msg));
141 static int do_try_sendpage(struct socket *sock, struct iov_iter *it)
143 struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
147 if (WARN_ON(!iov_iter_is_bvec(it)))
150 while (iov_iter_count(it)) {
151 /* iov_iter_iovec() for ITER_BVEC */
152 bv.bv_page = it->bvec->bv_page;
153 bv.bv_offset = it->bvec->bv_offset + it->iov_offset;
154 bv.bv_len = min(iov_iter_count(it),
155 it->bvec->bv_len - it->iov_offset);
158 * sendpage cannot properly handle pages with
159 * page_count == 0, we need to fall back to sendmsg if
162 * Same goes for slab pages: skb_can_coalesce() allows
163 * coalescing neighboring slab objects into a single frag
164 * which triggers one of hardened usercopy checks.
166 if (sendpage_ok(bv.bv_page)) {
167 ret = sock->ops->sendpage(sock, bv.bv_page,
168 bv.bv_offset, bv.bv_len,
171 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bv, 1, bv.bv_len);
172 ret = sock_sendmsg(sock, &msg);
180 iov_iter_advance(it, ret);
187 * Write as much as possible. The socket is expected to be corked,
188 * so we don't bother with MSG_MORE/MSG_SENDPAGE_NOTLAST here.
191 * 1 - done, nothing (else) to write
192 * 0 - socket is full, need to wait
195 static int ceph_tcp_send(struct ceph_connection *con)
199 dout("%s con %p have %zu try_sendpage %d\n", __func__, con,
200 iov_iter_count(&con->v2.out_iter), con->v2.out_iter_sendpage);
201 if (con->v2.out_iter_sendpage)
202 ret = do_try_sendpage(con->sock, &con->v2.out_iter);
204 ret = do_sendmsg(con->sock, &con->v2.out_iter);
205 dout("%s con %p ret %d left %zu\n", __func__, con, ret,
206 iov_iter_count(&con->v2.out_iter));
210 static void add_in_kvec(struct ceph_connection *con, void *buf, int len)
212 BUG_ON(con->v2.in_kvec_cnt >= ARRAY_SIZE(con->v2.in_kvecs));
213 WARN_ON(!iov_iter_is_kvec(&con->v2.in_iter));
215 con->v2.in_kvecs[con->v2.in_kvec_cnt].iov_base = buf;
216 con->v2.in_kvecs[con->v2.in_kvec_cnt].iov_len = len;
217 con->v2.in_kvec_cnt++;
219 con->v2.in_iter.nr_segs++;
220 con->v2.in_iter.count += len;
223 static void reset_in_kvecs(struct ceph_connection *con)
225 WARN_ON(iov_iter_count(&con->v2.in_iter));
227 con->v2.in_kvec_cnt = 0;
228 iov_iter_kvec(&con->v2.in_iter, ITER_DEST, con->v2.in_kvecs, 0, 0);
231 static void set_in_bvec(struct ceph_connection *con, const struct bio_vec *bv)
233 WARN_ON(iov_iter_count(&con->v2.in_iter));
235 con->v2.in_bvec = *bv;
236 iov_iter_bvec(&con->v2.in_iter, ITER_DEST, &con->v2.in_bvec, 1, bv->bv_len);
239 static void set_in_skip(struct ceph_connection *con, int len)
241 WARN_ON(iov_iter_count(&con->v2.in_iter));
243 dout("%s con %p len %d\n", __func__, con, len);
244 iov_iter_discard(&con->v2.in_iter, ITER_DEST, len);
247 static void add_out_kvec(struct ceph_connection *con, void *buf, int len)
249 BUG_ON(con->v2.out_kvec_cnt >= ARRAY_SIZE(con->v2.out_kvecs));
250 WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
251 WARN_ON(con->v2.out_zero);
253 con->v2.out_kvecs[con->v2.out_kvec_cnt].iov_base = buf;
254 con->v2.out_kvecs[con->v2.out_kvec_cnt].iov_len = len;
255 con->v2.out_kvec_cnt++;
257 con->v2.out_iter.nr_segs++;
258 con->v2.out_iter.count += len;
261 static void reset_out_kvecs(struct ceph_connection *con)
263 WARN_ON(iov_iter_count(&con->v2.out_iter));
264 WARN_ON(con->v2.out_zero);
266 con->v2.out_kvec_cnt = 0;
268 iov_iter_kvec(&con->v2.out_iter, ITER_SOURCE, con->v2.out_kvecs, 0, 0);
269 con->v2.out_iter_sendpage = false;
272 static void set_out_bvec(struct ceph_connection *con, const struct bio_vec *bv,
275 WARN_ON(iov_iter_count(&con->v2.out_iter));
276 WARN_ON(con->v2.out_zero);
278 con->v2.out_bvec = *bv;
279 con->v2.out_iter_sendpage = zerocopy;
280 iov_iter_bvec(&con->v2.out_iter, ITER_SOURCE, &con->v2.out_bvec, 1,
281 con->v2.out_bvec.bv_len);
284 static void set_out_bvec_zero(struct ceph_connection *con)
286 WARN_ON(iov_iter_count(&con->v2.out_iter));
287 WARN_ON(!con->v2.out_zero);
289 con->v2.out_bvec.bv_page = ceph_zero_page;
290 con->v2.out_bvec.bv_offset = 0;
291 con->v2.out_bvec.bv_len = min(con->v2.out_zero, (int)PAGE_SIZE);
292 con->v2.out_iter_sendpage = true;
293 iov_iter_bvec(&con->v2.out_iter, ITER_SOURCE, &con->v2.out_bvec, 1,
294 con->v2.out_bvec.bv_len);
297 static void out_zero_add(struct ceph_connection *con, int len)
299 dout("%s con %p len %d\n", __func__, con, len);
300 con->v2.out_zero += len;
303 static void *alloc_conn_buf(struct ceph_connection *con, int len)
307 dout("%s con %p len %d\n", __func__, con, len);
309 if (WARN_ON(con->v2.conn_buf_cnt >= ARRAY_SIZE(con->v2.conn_bufs)))
312 buf = kvmalloc(len, GFP_NOIO);
316 con->v2.conn_bufs[con->v2.conn_buf_cnt++] = buf;
320 static void free_conn_bufs(struct ceph_connection *con)
322 while (con->v2.conn_buf_cnt)
323 kvfree(con->v2.conn_bufs[--con->v2.conn_buf_cnt]);
326 static void add_in_sign_kvec(struct ceph_connection *con, void *buf, int len)
328 BUG_ON(con->v2.in_sign_kvec_cnt >= ARRAY_SIZE(con->v2.in_sign_kvecs));
330 con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_base = buf;
331 con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_len = len;
332 con->v2.in_sign_kvec_cnt++;
335 static void clear_in_sign_kvecs(struct ceph_connection *con)
337 con->v2.in_sign_kvec_cnt = 0;
340 static void add_out_sign_kvec(struct ceph_connection *con, void *buf, int len)
342 BUG_ON(con->v2.out_sign_kvec_cnt >= ARRAY_SIZE(con->v2.out_sign_kvecs));
344 con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_base = buf;
345 con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_len = len;
346 con->v2.out_sign_kvec_cnt++;
349 static void clear_out_sign_kvecs(struct ceph_connection *con)
351 con->v2.out_sign_kvec_cnt = 0;
354 static bool con_secure(struct ceph_connection *con)
356 return con->v2.con_mode == CEPH_CON_MODE_SECURE;
359 static int front_len(const struct ceph_msg *msg)
361 return le32_to_cpu(msg->hdr.front_len);
364 static int middle_len(const struct ceph_msg *msg)
366 return le32_to_cpu(msg->hdr.middle_len);
369 static int data_len(const struct ceph_msg *msg)
371 return le32_to_cpu(msg->hdr.data_len);
374 static bool need_padding(int len)
376 return !IS_ALIGNED(len, CEPH_GCM_BLOCK_LEN);
379 static int padded_len(int len)
381 return ALIGN(len, CEPH_GCM_BLOCK_LEN);
384 static int padding_len(int len)
386 return padded_len(len) - len;
389 /* preamble + control segment */
390 static int head_onwire_len(int ctrl_len, bool secure)
395 BUG_ON(ctrl_len < 0 || ctrl_len > CEPH_MSG_MAX_CONTROL_LEN);
398 head_len = CEPH_PREAMBLE_SECURE_LEN;
399 if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) {
400 rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
401 head_len += padded_len(rem_len) + CEPH_GCM_TAG_LEN;
404 head_len = CEPH_PREAMBLE_PLAIN_LEN;
406 head_len += ctrl_len + CEPH_CRC_LEN;
411 /* front, middle and data segments + epilogue */
412 static int __tail_onwire_len(int front_len, int middle_len, int data_len,
415 BUG_ON(front_len < 0 || front_len > CEPH_MSG_MAX_FRONT_LEN ||
416 middle_len < 0 || middle_len > CEPH_MSG_MAX_MIDDLE_LEN ||
417 data_len < 0 || data_len > CEPH_MSG_MAX_DATA_LEN);
419 if (!front_len && !middle_len && !data_len)
423 return front_len + middle_len + data_len +
424 CEPH_EPILOGUE_PLAIN_LEN;
426 return padded_len(front_len) + padded_len(middle_len) +
427 padded_len(data_len) + CEPH_EPILOGUE_SECURE_LEN;
430 static int tail_onwire_len(const struct ceph_msg *msg, bool secure)
432 return __tail_onwire_len(front_len(msg), middle_len(msg),
433 data_len(msg), secure);
436 /* head_onwire_len(sizeof(struct ceph_msg_header2), false) */
437 #define MESSAGE_HEAD_PLAIN_LEN (CEPH_PREAMBLE_PLAIN_LEN + \
438 sizeof(struct ceph_msg_header2) + \
441 static const int frame_aligns[] = {
449 * Discards trailing empty segments, unless there is just one segment.
450 * A frame always has at least one (possibly empty) segment.
452 static int calc_segment_count(const int *lens, int len_cnt)
456 for (i = len_cnt - 1; i >= 0; i--) {
464 static void init_frame_desc(struct ceph_frame_desc *desc, int tag,
465 const int *lens, int len_cnt)
469 memset(desc, 0, sizeof(*desc));
472 desc->fd_seg_cnt = calc_segment_count(lens, len_cnt);
473 BUG_ON(desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT);
474 for (i = 0; i < desc->fd_seg_cnt; i++) {
475 desc->fd_lens[i] = lens[i];
476 desc->fd_aligns[i] = frame_aligns[i];
481 * Preamble crc covers everything up to itself (28 bytes) and
482 * is calculated and verified irrespective of the connection mode
483 * (i.e. even if the frame is encrypted).
485 static void encode_preamble(const struct ceph_frame_desc *desc, void *p)
487 void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN;
491 memset(p, 0, CEPH_PREAMBLE_LEN);
493 ceph_encode_8(&p, desc->fd_tag);
494 ceph_encode_8(&p, desc->fd_seg_cnt);
495 for (i = 0; i < desc->fd_seg_cnt; i++) {
496 ceph_encode_32(&p, desc->fd_lens[i]);
497 ceph_encode_16(&p, desc->fd_aligns[i]);
500 put_unaligned_le32(crc32c(0, start, crcp - start), crcp);
503 static int decode_preamble(void *p, struct ceph_frame_desc *desc)
505 void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN;
506 u32 crc, expected_crc;
509 crc = crc32c(0, p, crcp - p);
510 expected_crc = get_unaligned_le32(crcp);
511 if (crc != expected_crc) {
512 pr_err("bad preamble crc, calculated %u, expected %u\n",
517 memset(desc, 0, sizeof(*desc));
519 desc->fd_tag = ceph_decode_8(&p);
520 desc->fd_seg_cnt = ceph_decode_8(&p);
521 if (desc->fd_seg_cnt < 1 ||
522 desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT) {
523 pr_err("bad segment count %d\n", desc->fd_seg_cnt);
526 for (i = 0; i < desc->fd_seg_cnt; i++) {
527 desc->fd_lens[i] = ceph_decode_32(&p);
528 desc->fd_aligns[i] = ceph_decode_16(&p);
531 if (desc->fd_lens[0] < 0 ||
532 desc->fd_lens[0] > CEPH_MSG_MAX_CONTROL_LEN) {
533 pr_err("bad control segment length %d\n", desc->fd_lens[0]);
536 if (desc->fd_lens[1] < 0 ||
537 desc->fd_lens[1] > CEPH_MSG_MAX_FRONT_LEN) {
538 pr_err("bad front segment length %d\n", desc->fd_lens[1]);
541 if (desc->fd_lens[2] < 0 ||
542 desc->fd_lens[2] > CEPH_MSG_MAX_MIDDLE_LEN) {
543 pr_err("bad middle segment length %d\n", desc->fd_lens[2]);
546 if (desc->fd_lens[3] < 0 ||
547 desc->fd_lens[3] > CEPH_MSG_MAX_DATA_LEN) {
548 pr_err("bad data segment length %d\n", desc->fd_lens[3]);
553 * This would fire for FRAME_TAG_WAIT (it has one empty
554 * segment), but we should never get it as client.
556 if (!desc->fd_lens[desc->fd_seg_cnt - 1]) {
557 pr_err("last segment empty, segment count %d\n",
565 static void encode_epilogue_plain(struct ceph_connection *con, bool aborted)
567 con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED :
568 FRAME_LATE_STATUS_COMPLETE;
569 cpu_to_le32s(&con->v2.out_epil.front_crc);
570 cpu_to_le32s(&con->v2.out_epil.middle_crc);
571 cpu_to_le32s(&con->v2.out_epil.data_crc);
574 static void encode_epilogue_secure(struct ceph_connection *con, bool aborted)
576 memset(&con->v2.out_epil, 0, sizeof(con->v2.out_epil));
577 con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED :
578 FRAME_LATE_STATUS_COMPLETE;
581 static int decode_epilogue(void *p, u32 *front_crc, u32 *middle_crc,
586 late_status = ceph_decode_8(&p);
587 if ((late_status & FRAME_LATE_STATUS_ABORTED_MASK) !=
588 FRAME_LATE_STATUS_COMPLETE) {
589 /* we should never get an aborted message as client */
590 pr_err("bad late_status 0x%x\n", late_status);
594 if (front_crc && middle_crc && data_crc) {
595 *front_crc = ceph_decode_32(&p);
596 *middle_crc = ceph_decode_32(&p);
597 *data_crc = ceph_decode_32(&p);
603 static void fill_header(struct ceph_msg_header *hdr,
604 const struct ceph_msg_header2 *hdr2,
605 int front_len, int middle_len, int data_len,
606 const struct ceph_entity_name *peer_name)
608 hdr->seq = hdr2->seq;
609 hdr->tid = hdr2->tid;
610 hdr->type = hdr2->type;
611 hdr->priority = hdr2->priority;
612 hdr->version = hdr2->version;
613 hdr->front_len = cpu_to_le32(front_len);
614 hdr->middle_len = cpu_to_le32(middle_len);
615 hdr->data_len = cpu_to_le32(data_len);
616 hdr->data_off = hdr2->data_off;
617 hdr->src = *peer_name;
618 hdr->compat_version = hdr2->compat_version;
623 static void fill_header2(struct ceph_msg_header2 *hdr2,
624 const struct ceph_msg_header *hdr, u64 ack_seq)
626 hdr2->seq = hdr->seq;
627 hdr2->tid = hdr->tid;
628 hdr2->type = hdr->type;
629 hdr2->priority = hdr->priority;
630 hdr2->version = hdr->version;
631 hdr2->data_pre_padding_len = 0;
632 hdr2->data_off = hdr->data_off;
633 hdr2->ack_seq = cpu_to_le64(ack_seq);
635 hdr2->compat_version = hdr->compat_version;
639 static int verify_control_crc(struct ceph_connection *con)
641 int ctrl_len = con->v2.in_desc.fd_lens[0];
642 u32 crc, expected_crc;
644 WARN_ON(con->v2.in_kvecs[0].iov_len != ctrl_len);
645 WARN_ON(con->v2.in_kvecs[1].iov_len != CEPH_CRC_LEN);
647 crc = crc32c(-1, con->v2.in_kvecs[0].iov_base, ctrl_len);
648 expected_crc = get_unaligned_le32(con->v2.in_kvecs[1].iov_base);
649 if (crc != expected_crc) {
650 pr_err("bad control crc, calculated %u, expected %u\n",
658 static int verify_epilogue_crcs(struct ceph_connection *con, u32 front_crc,
659 u32 middle_crc, u32 data_crc)
661 if (front_len(con->in_msg)) {
662 con->in_front_crc = crc32c(-1, con->in_msg->front.iov_base,
663 front_len(con->in_msg));
665 WARN_ON(!middle_len(con->in_msg) && !data_len(con->in_msg));
666 con->in_front_crc = -1;
669 if (middle_len(con->in_msg))
670 con->in_middle_crc = crc32c(-1,
671 con->in_msg->middle->vec.iov_base,
672 middle_len(con->in_msg));
673 else if (data_len(con->in_msg))
674 con->in_middle_crc = -1;
676 con->in_middle_crc = 0;
678 if (!data_len(con->in_msg))
679 con->in_data_crc = 0;
681 dout("%s con %p msg %p crcs %u %u %u\n", __func__, con, con->in_msg,
682 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
684 if (con->in_front_crc != front_crc) {
685 pr_err("bad front crc, calculated %u, expected %u\n",
686 con->in_front_crc, front_crc);
689 if (con->in_middle_crc != middle_crc) {
690 pr_err("bad middle crc, calculated %u, expected %u\n",
691 con->in_middle_crc, middle_crc);
694 if (con->in_data_crc != data_crc) {
695 pr_err("bad data crc, calculated %u, expected %u\n",
696 con->in_data_crc, data_crc);
703 static int setup_crypto(struct ceph_connection *con,
704 const u8 *session_key, int session_key_len,
705 const u8 *con_secret, int con_secret_len)
707 unsigned int noio_flag;
710 dout("%s con %p con_mode %d session_key_len %d con_secret_len %d\n",
711 __func__, con, con->v2.con_mode, session_key_len, con_secret_len);
712 WARN_ON(con->v2.hmac_tfm || con->v2.gcm_tfm || con->v2.gcm_req);
714 if (con->v2.con_mode != CEPH_CON_MODE_CRC &&
715 con->v2.con_mode != CEPH_CON_MODE_SECURE) {
716 pr_err("bad con_mode %d\n", con->v2.con_mode);
720 if (!session_key_len) {
721 WARN_ON(con->v2.con_mode != CEPH_CON_MODE_CRC);
722 WARN_ON(con_secret_len);
723 return 0; /* auth_none */
726 noio_flag = memalloc_noio_save();
727 con->v2.hmac_tfm = crypto_alloc_shash("hmac(sha256)", 0, 0);
728 memalloc_noio_restore(noio_flag);
729 if (IS_ERR(con->v2.hmac_tfm)) {
730 ret = PTR_ERR(con->v2.hmac_tfm);
731 con->v2.hmac_tfm = NULL;
732 pr_err("failed to allocate hmac tfm context: %d\n", ret);
736 WARN_ON((unsigned long)session_key &
737 crypto_shash_alignmask(con->v2.hmac_tfm));
738 ret = crypto_shash_setkey(con->v2.hmac_tfm, session_key,
741 pr_err("failed to set hmac key: %d\n", ret);
745 if (con->v2.con_mode == CEPH_CON_MODE_CRC) {
746 WARN_ON(con_secret_len);
747 return 0; /* auth_x, plain mode */
750 if (con_secret_len < CEPH_GCM_KEY_LEN + 2 * CEPH_GCM_IV_LEN) {
751 pr_err("con_secret too small %d\n", con_secret_len);
755 noio_flag = memalloc_noio_save();
756 con->v2.gcm_tfm = crypto_alloc_aead("gcm(aes)", 0, 0);
757 memalloc_noio_restore(noio_flag);
758 if (IS_ERR(con->v2.gcm_tfm)) {
759 ret = PTR_ERR(con->v2.gcm_tfm);
760 con->v2.gcm_tfm = NULL;
761 pr_err("failed to allocate gcm tfm context: %d\n", ret);
765 WARN_ON((unsigned long)con_secret &
766 crypto_aead_alignmask(con->v2.gcm_tfm));
767 ret = crypto_aead_setkey(con->v2.gcm_tfm, con_secret, CEPH_GCM_KEY_LEN);
769 pr_err("failed to set gcm key: %d\n", ret);
773 WARN_ON(crypto_aead_ivsize(con->v2.gcm_tfm) != CEPH_GCM_IV_LEN);
774 ret = crypto_aead_setauthsize(con->v2.gcm_tfm, CEPH_GCM_TAG_LEN);
776 pr_err("failed to set gcm tag size: %d\n", ret);
780 con->v2.gcm_req = aead_request_alloc(con->v2.gcm_tfm, GFP_NOIO);
781 if (!con->v2.gcm_req) {
782 pr_err("failed to allocate gcm request\n");
786 crypto_init_wait(&con->v2.gcm_wait);
787 aead_request_set_callback(con->v2.gcm_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
788 crypto_req_done, &con->v2.gcm_wait);
790 memcpy(&con->v2.in_gcm_nonce, con_secret + CEPH_GCM_KEY_LEN,
792 memcpy(&con->v2.out_gcm_nonce,
793 con_secret + CEPH_GCM_KEY_LEN + CEPH_GCM_IV_LEN,
795 return 0; /* auth_x, secure mode */
798 static int hmac_sha256(struct ceph_connection *con, const struct kvec *kvecs,
799 int kvec_cnt, u8 *hmac)
801 SHASH_DESC_ON_STACK(desc, con->v2.hmac_tfm); /* tfm arg is ignored */
805 dout("%s con %p hmac_tfm %p kvec_cnt %d\n", __func__, con,
806 con->v2.hmac_tfm, kvec_cnt);
808 if (!con->v2.hmac_tfm) {
809 memset(hmac, 0, SHA256_DIGEST_SIZE);
810 return 0; /* auth_none */
813 desc->tfm = con->v2.hmac_tfm;
814 ret = crypto_shash_init(desc);
818 for (i = 0; i < kvec_cnt; i++) {
819 WARN_ON((unsigned long)kvecs[i].iov_base &
820 crypto_shash_alignmask(con->v2.hmac_tfm));
821 ret = crypto_shash_update(desc, kvecs[i].iov_base,
827 ret = crypto_shash_final(desc, hmac);
830 shash_desc_zero(desc);
831 return ret; /* auth_x, both plain and secure modes */
834 static void gcm_inc_nonce(struct ceph_gcm_nonce *nonce)
838 counter = le64_to_cpu(nonce->counter);
839 nonce->counter = cpu_to_le64(counter + 1);
842 static int gcm_crypt(struct ceph_connection *con, bool encrypt,
843 struct scatterlist *src, struct scatterlist *dst,
846 struct ceph_gcm_nonce *nonce;
849 nonce = encrypt ? &con->v2.out_gcm_nonce : &con->v2.in_gcm_nonce;
851 aead_request_set_ad(con->v2.gcm_req, 0); /* no AAD */
852 aead_request_set_crypt(con->v2.gcm_req, src, dst, src_len, (u8 *)nonce);
853 ret = crypto_wait_req(encrypt ? crypto_aead_encrypt(con->v2.gcm_req) :
854 crypto_aead_decrypt(con->v2.gcm_req),
859 gcm_inc_nonce(nonce);
863 static void get_bvec_at(struct ceph_msg_data_cursor *cursor,
869 WARN_ON(!cursor->total_resid);
871 /* skip zero-length data items */
872 while (!cursor->resid)
873 ceph_msg_data_advance(cursor, 0);
875 /* get a piece of data, cursor isn't advanced */
876 page = ceph_msg_data_next(cursor, &off, &len);
883 static int calc_sg_cnt(void *buf, int buf_len)
890 sg_cnt = need_padding(buf_len) ? 1 : 0;
891 if (is_vmalloc_addr(buf)) {
892 WARN_ON(offset_in_page(buf));
893 sg_cnt += PAGE_ALIGN(buf_len) >> PAGE_SHIFT;
901 static int calc_sg_cnt_cursor(struct ceph_msg_data_cursor *cursor)
903 int data_len = cursor->total_resid;
910 sg_cnt = need_padding(data_len) ? 1 : 0;
912 get_bvec_at(cursor, &bv);
915 ceph_msg_data_advance(cursor, bv.bv_len);
916 } while (cursor->total_resid);
921 static void init_sgs(struct scatterlist **sg, void *buf, int buf_len, u8 *pad)
923 void *end = buf + buf_len;
931 if (is_vmalloc_addr(buf)) {
934 page = vmalloc_to_page(p);
935 len = min_t(int, end - p, PAGE_SIZE);
936 WARN_ON(!page || !len || offset_in_page(p));
937 sg_set_page(*sg, page, len, 0);
942 sg_set_buf(*sg, buf, buf_len);
946 if (need_padding(buf_len)) {
947 sg_set_buf(*sg, pad, padding_len(buf_len));
952 static void init_sgs_cursor(struct scatterlist **sg,
953 struct ceph_msg_data_cursor *cursor, u8 *pad)
955 int data_len = cursor->total_resid;
962 get_bvec_at(cursor, &bv);
963 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
966 ceph_msg_data_advance(cursor, bv.bv_len);
967 } while (cursor->total_resid);
969 if (need_padding(data_len)) {
970 sg_set_buf(*sg, pad, padding_len(data_len));
975 static int setup_message_sgs(struct sg_table *sgt, struct ceph_msg *msg,
976 u8 *front_pad, u8 *middle_pad, u8 *data_pad,
977 void *epilogue, bool add_tag)
979 struct ceph_msg_data_cursor cursor;
980 struct scatterlist *cur_sg;
984 if (!front_len(msg) && !middle_len(msg) && !data_len(msg))
987 sg_cnt = 1; /* epilogue + [auth tag] */
989 sg_cnt += calc_sg_cnt(msg->front.iov_base,
992 sg_cnt += calc_sg_cnt(msg->middle->vec.iov_base,
995 ceph_msg_data_cursor_init(&cursor, msg, data_len(msg));
996 sg_cnt += calc_sg_cnt_cursor(&cursor);
999 ret = sg_alloc_table(sgt, sg_cnt, GFP_NOIO);
1005 init_sgs(&cur_sg, msg->front.iov_base, front_len(msg),
1007 if (middle_len(msg))
1008 init_sgs(&cur_sg, msg->middle->vec.iov_base, middle_len(msg),
1010 if (data_len(msg)) {
1011 ceph_msg_data_cursor_init(&cursor, msg, data_len(msg));
1012 init_sgs_cursor(&cur_sg, &cursor, data_pad);
1015 WARN_ON(!sg_is_last(cur_sg));
1016 sg_set_buf(cur_sg, epilogue,
1017 CEPH_GCM_BLOCK_LEN + (add_tag ? CEPH_GCM_TAG_LEN : 0));
1021 static int decrypt_preamble(struct ceph_connection *con)
1023 struct scatterlist sg;
1025 sg_init_one(&sg, con->v2.in_buf, CEPH_PREAMBLE_SECURE_LEN);
1026 return gcm_crypt(con, false, &sg, &sg, CEPH_PREAMBLE_SECURE_LEN);
1029 static int decrypt_control_remainder(struct ceph_connection *con)
1031 int ctrl_len = con->v2.in_desc.fd_lens[0];
1032 int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1033 int pt_len = padding_len(rem_len) + CEPH_GCM_TAG_LEN;
1034 struct scatterlist sgs[2];
1036 WARN_ON(con->v2.in_kvecs[0].iov_len != rem_len);
1037 WARN_ON(con->v2.in_kvecs[1].iov_len != pt_len);
1039 sg_init_table(sgs, 2);
1040 sg_set_buf(&sgs[0], con->v2.in_kvecs[0].iov_base, rem_len);
1041 sg_set_buf(&sgs[1], con->v2.in_buf, pt_len);
1043 return gcm_crypt(con, false, sgs, sgs,
1044 padded_len(rem_len) + CEPH_GCM_TAG_LEN);
1047 static int decrypt_tail(struct ceph_connection *con)
1049 struct sg_table enc_sgt = {};
1050 struct sg_table sgt = {};
1054 tail_len = tail_onwire_len(con->in_msg, true);
1055 ret = sg_alloc_table_from_pages(&enc_sgt, con->v2.in_enc_pages,
1056 con->v2.in_enc_page_cnt, 0, tail_len,
1061 ret = setup_message_sgs(&sgt, con->in_msg, FRONT_PAD(con->v2.in_buf),
1062 MIDDLE_PAD(con->v2.in_buf), DATA_PAD(con->v2.in_buf),
1063 con->v2.in_buf, true);
1067 dout("%s con %p msg %p enc_page_cnt %d sg_cnt %d\n", __func__, con,
1068 con->in_msg, con->v2.in_enc_page_cnt, sgt.orig_nents);
1069 ret = gcm_crypt(con, false, enc_sgt.sgl, sgt.sgl, tail_len);
1073 WARN_ON(!con->v2.in_enc_page_cnt);
1074 ceph_release_page_vector(con->v2.in_enc_pages,
1075 con->v2.in_enc_page_cnt);
1076 con->v2.in_enc_pages = NULL;
1077 con->v2.in_enc_page_cnt = 0;
1080 sg_free_table(&sgt);
1081 sg_free_table(&enc_sgt);
1085 static int prepare_banner(struct ceph_connection *con)
1087 int buf_len = CEPH_BANNER_V2_LEN + 2 + 8 + 8;
1090 buf = alloc_conn_buf(con, buf_len);
1095 ceph_encode_copy(&p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN);
1096 ceph_encode_16(&p, sizeof(u64) + sizeof(u64));
1097 ceph_encode_64(&p, CEPH_MSGR2_SUPPORTED_FEATURES);
1098 ceph_encode_64(&p, CEPH_MSGR2_REQUIRED_FEATURES);
1099 WARN_ON(p != buf + buf_len);
1101 add_out_kvec(con, buf, buf_len);
1102 add_out_sign_kvec(con, buf, buf_len);
1103 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1110 * control body (ctrl_len bytes)
1111 * space for control crc
1113 * extdata (optional):
1114 * control body (extdata_len bytes)
1116 * Compute control crc and gather base and extdata into:
1119 * control body (ctrl_len + extdata_len bytes)
1122 * Preamble should already be encoded at the start of base.
1124 static void prepare_head_plain(struct ceph_connection *con, void *base,
1125 int ctrl_len, void *extdata, int extdata_len,
1128 int base_len = CEPH_PREAMBLE_LEN + ctrl_len + CEPH_CRC_LEN;
1129 void *crcp = base + base_len - CEPH_CRC_LEN;
1132 crc = crc32c(-1, CTRL_BODY(base), ctrl_len);
1134 crc = crc32c(crc, extdata, extdata_len);
1135 put_unaligned_le32(crc, crcp);
1138 add_out_kvec(con, base, base_len);
1140 add_out_sign_kvec(con, base, base_len);
1144 add_out_kvec(con, base, crcp - base);
1145 add_out_kvec(con, extdata, extdata_len);
1146 add_out_kvec(con, crcp, CEPH_CRC_LEN);
1148 add_out_sign_kvec(con, base, crcp - base);
1149 add_out_sign_kvec(con, extdata, extdata_len);
1150 add_out_sign_kvec(con, crcp, CEPH_CRC_LEN);
1154 static int prepare_head_secure_small(struct ceph_connection *con,
1155 void *base, int ctrl_len)
1157 struct scatterlist sg;
1160 /* inline buffer padding? */
1161 if (ctrl_len < CEPH_PREAMBLE_INLINE_LEN)
1162 memset(CTRL_BODY(base) + ctrl_len, 0,
1163 CEPH_PREAMBLE_INLINE_LEN - ctrl_len);
1165 sg_init_one(&sg, base, CEPH_PREAMBLE_SECURE_LEN);
1166 ret = gcm_crypt(con, true, &sg, &sg,
1167 CEPH_PREAMBLE_SECURE_LEN - CEPH_GCM_TAG_LEN);
1171 add_out_kvec(con, base, CEPH_PREAMBLE_SECURE_LEN);
1178 * control body (ctrl_len bytes)
1179 * space for padding, if needed
1180 * space for control remainder auth tag
1181 * space for preamble auth tag
1183 * Encrypt preamble and the inline portion, then encrypt the remainder
1187 * control body (48 bytes)
1189 * control body (ctrl_len - 48 bytes)
1190 * zero padding, if needed
1191 * control remainder auth tag
1193 * Preamble should already be encoded at the start of base.
1195 static int prepare_head_secure_big(struct ceph_connection *con,
1196 void *base, int ctrl_len)
1198 int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1199 void *rem = CTRL_BODY(base) + CEPH_PREAMBLE_INLINE_LEN;
1200 void *rem_tag = rem + padded_len(rem_len);
1201 void *pmbl_tag = rem_tag + CEPH_GCM_TAG_LEN;
1202 struct scatterlist sgs[2];
1205 sg_init_table(sgs, 2);
1206 sg_set_buf(&sgs[0], base, rem - base);
1207 sg_set_buf(&sgs[1], pmbl_tag, CEPH_GCM_TAG_LEN);
1208 ret = gcm_crypt(con, true, sgs, sgs, rem - base);
1212 /* control remainder padding? */
1213 if (need_padding(rem_len))
1214 memset(rem + rem_len, 0, padding_len(rem_len));
1216 sg_init_one(&sgs[0], rem, pmbl_tag - rem);
1217 ret = gcm_crypt(con, true, sgs, sgs, rem_tag - rem);
1221 add_out_kvec(con, base, rem - base);
1222 add_out_kvec(con, pmbl_tag, CEPH_GCM_TAG_LEN);
1223 add_out_kvec(con, rem, pmbl_tag - rem);
1227 static int __prepare_control(struct ceph_connection *con, int tag,
1228 void *base, int ctrl_len, void *extdata,
1229 int extdata_len, bool to_be_signed)
1231 int total_len = ctrl_len + extdata_len;
1232 struct ceph_frame_desc desc;
1235 dout("%s con %p tag %d len %d (%d+%d)\n", __func__, con, tag,
1236 total_len, ctrl_len, extdata_len);
1238 /* extdata may be vmalloc'ed but not base */
1239 if (WARN_ON(is_vmalloc_addr(base) || !ctrl_len))
1242 init_frame_desc(&desc, tag, &total_len, 1);
1243 encode_preamble(&desc, base);
1245 if (con_secure(con)) {
1246 if (WARN_ON(extdata_len || to_be_signed))
1249 if (ctrl_len <= CEPH_PREAMBLE_INLINE_LEN)
1250 /* fully inlined, inline buffer may need padding */
1251 ret = prepare_head_secure_small(con, base, ctrl_len);
1253 /* partially inlined, inline buffer is full */
1254 ret = prepare_head_secure_big(con, base, ctrl_len);
1258 prepare_head_plain(con, base, ctrl_len, extdata, extdata_len,
1262 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1266 static int prepare_control(struct ceph_connection *con, int tag,
1267 void *base, int ctrl_len)
1269 return __prepare_control(con, tag, base, ctrl_len, NULL, 0, false);
1272 static int prepare_hello(struct ceph_connection *con)
1277 ctrl_len = 1 + ceph_entity_addr_encoding_len(&con->peer_addr);
1278 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1283 ceph_encode_8(&p, CEPH_ENTITY_TYPE_CLIENT);
1284 ceph_encode_entity_addr(&p, &con->peer_addr);
1285 WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1287 return __prepare_control(con, FRAME_TAG_HELLO, buf, ctrl_len,
1291 /* so that head_onwire_len(AUTH_BUF_LEN, false) is 512 */
1292 #define AUTH_BUF_LEN (512 - CEPH_CRC_LEN - CEPH_PREAMBLE_PLAIN_LEN)
1294 static int prepare_auth_request(struct ceph_connection *con)
1296 void *authorizer, *authorizer_copy;
1297 int ctrl_len, authorizer_len;
1301 ctrl_len = AUTH_BUF_LEN;
1302 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1306 mutex_unlock(&con->mutex);
1307 ret = con->ops->get_auth_request(con, CTRL_BODY(buf), &ctrl_len,
1308 &authorizer, &authorizer_len);
1309 mutex_lock(&con->mutex);
1310 if (con->state != CEPH_CON_S_V2_HELLO) {
1311 dout("%s con %p state changed to %d\n", __func__, con,
1316 dout("%s con %p get_auth_request ret %d\n", __func__, con, ret);
1320 authorizer_copy = alloc_conn_buf(con, authorizer_len);
1321 if (!authorizer_copy)
1324 memcpy(authorizer_copy, authorizer, authorizer_len);
1326 return __prepare_control(con, FRAME_TAG_AUTH_REQUEST, buf, ctrl_len,
1327 authorizer_copy, authorizer_len, true);
1330 static int prepare_auth_request_more(struct ceph_connection *con,
1331 void *reply, int reply_len)
1333 int ctrl_len, authorizer_len;
1338 ctrl_len = AUTH_BUF_LEN;
1339 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1343 mutex_unlock(&con->mutex);
1344 ret = con->ops->handle_auth_reply_more(con, reply, reply_len,
1345 CTRL_BODY(buf), &ctrl_len,
1346 &authorizer, &authorizer_len);
1347 mutex_lock(&con->mutex);
1348 if (con->state != CEPH_CON_S_V2_AUTH) {
1349 dout("%s con %p state changed to %d\n", __func__, con,
1354 dout("%s con %p handle_auth_reply_more ret %d\n", __func__, con, ret);
1358 return __prepare_control(con, FRAME_TAG_AUTH_REQUEST_MORE, buf,
1359 ctrl_len, authorizer, authorizer_len, true);
1362 static int prepare_auth_signature(struct ceph_connection *con)
1367 buf = alloc_conn_buf(con, head_onwire_len(SHA256_DIGEST_SIZE,
1372 ret = hmac_sha256(con, con->v2.in_sign_kvecs, con->v2.in_sign_kvec_cnt,
1377 return prepare_control(con, FRAME_TAG_AUTH_SIGNATURE, buf,
1378 SHA256_DIGEST_SIZE);
1381 static int prepare_client_ident(struct ceph_connection *con)
1383 struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
1384 struct ceph_client *client = from_msgr(con->msgr);
1385 u64 global_id = ceph_client_gid(client);
1389 WARN_ON(con->v2.server_cookie);
1390 WARN_ON(con->v2.connect_seq);
1391 WARN_ON(con->v2.peer_global_seq);
1393 if (!con->v2.client_cookie) {
1395 get_random_bytes(&con->v2.client_cookie,
1396 sizeof(con->v2.client_cookie));
1397 } while (!con->v2.client_cookie);
1398 dout("%s con %p generated cookie 0x%llx\n", __func__, con,
1399 con->v2.client_cookie);
1401 dout("%s con %p cookie already set 0x%llx\n", __func__, con,
1402 con->v2.client_cookie);
1405 dout("%s con %p my_addr %s/%u peer_addr %s/%u global_id %llu global_seq %llu features 0x%llx required_features 0x%llx cookie 0x%llx\n",
1406 __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce),
1407 ceph_pr_addr(&con->peer_addr), le32_to_cpu(con->peer_addr.nonce),
1408 global_id, con->v2.global_seq, client->supported_features,
1409 client->required_features, con->v2.client_cookie);
1411 ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) +
1412 ceph_entity_addr_encoding_len(&con->peer_addr) + 6 * 8;
1413 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con)));
1418 ceph_encode_8(&p, 2); /* addrvec marker */
1419 ceph_encode_32(&p, 1); /* addr_cnt */
1420 ceph_encode_entity_addr(&p, my_addr);
1421 ceph_encode_entity_addr(&p, &con->peer_addr);
1422 ceph_encode_64(&p, global_id);
1423 ceph_encode_64(&p, con->v2.global_seq);
1424 ceph_encode_64(&p, client->supported_features);
1425 ceph_encode_64(&p, client->required_features);
1426 ceph_encode_64(&p, 0); /* flags */
1427 ceph_encode_64(&p, con->v2.client_cookie);
1428 WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1430 return prepare_control(con, FRAME_TAG_CLIENT_IDENT, buf, ctrl_len);
1433 static int prepare_session_reconnect(struct ceph_connection *con)
1435 struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
1439 WARN_ON(!con->v2.client_cookie);
1440 WARN_ON(!con->v2.server_cookie);
1441 WARN_ON(!con->v2.connect_seq);
1442 WARN_ON(!con->v2.peer_global_seq);
1444 dout("%s con %p my_addr %s/%u client_cookie 0x%llx server_cookie 0x%llx global_seq %llu connect_seq %llu in_seq %llu\n",
1445 __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce),
1446 con->v2.client_cookie, con->v2.server_cookie, con->v2.global_seq,
1447 con->v2.connect_seq, con->in_seq);
1449 ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) + 5 * 8;
1450 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con)));
1455 ceph_encode_8(&p, 2); /* entity_addrvec_t marker */
1456 ceph_encode_32(&p, 1); /* my_addrs len */
1457 ceph_encode_entity_addr(&p, my_addr);
1458 ceph_encode_64(&p, con->v2.client_cookie);
1459 ceph_encode_64(&p, con->v2.server_cookie);
1460 ceph_encode_64(&p, con->v2.global_seq);
1461 ceph_encode_64(&p, con->v2.connect_seq);
1462 ceph_encode_64(&p, con->in_seq);
1463 WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1465 return prepare_control(con, FRAME_TAG_SESSION_RECONNECT, buf, ctrl_len);
1468 static int prepare_keepalive2(struct ceph_connection *con)
1470 struct ceph_timespec *ts = CTRL_BODY(con->v2.out_buf);
1471 struct timespec64 now;
1473 ktime_get_real_ts64(&now);
1474 dout("%s con %p timestamp %lld.%09ld\n", __func__, con, now.tv_sec,
1477 ceph_encode_timespec64(ts, &now);
1479 reset_out_kvecs(con);
1480 return prepare_control(con, FRAME_TAG_KEEPALIVE2, con->v2.out_buf,
1481 sizeof(struct ceph_timespec));
1484 static int prepare_ack(struct ceph_connection *con)
1488 dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con,
1489 con->in_seq_acked, con->in_seq);
1490 con->in_seq_acked = con->in_seq;
1492 p = CTRL_BODY(con->v2.out_buf);
1493 ceph_encode_64(&p, con->in_seq_acked);
1495 reset_out_kvecs(con);
1496 return prepare_control(con, FRAME_TAG_ACK, con->v2.out_buf, 8);
1499 static void prepare_epilogue_plain(struct ceph_connection *con, bool aborted)
1501 dout("%s con %p msg %p aborted %d crcs %u %u %u\n", __func__, con,
1502 con->out_msg, aborted, con->v2.out_epil.front_crc,
1503 con->v2.out_epil.middle_crc, con->v2.out_epil.data_crc);
1505 encode_epilogue_plain(con, aborted);
1506 add_out_kvec(con, &con->v2.out_epil, CEPH_EPILOGUE_PLAIN_LEN);
1510 * For "used" empty segments, crc is -1. For unused (trailing)
1511 * segments, crc is 0.
1513 static void prepare_message_plain(struct ceph_connection *con)
1515 struct ceph_msg *msg = con->out_msg;
1517 prepare_head_plain(con, con->v2.out_buf,
1518 sizeof(struct ceph_msg_header2), NULL, 0, false);
1520 if (!front_len(msg) && !middle_len(msg)) {
1521 if (!data_len(msg)) {
1523 * Empty message: once the head is written,
1524 * we are done -- there is no epilogue.
1526 con->v2.out_state = OUT_S_FINISH_MESSAGE;
1530 con->v2.out_epil.front_crc = -1;
1531 con->v2.out_epil.middle_crc = -1;
1532 con->v2.out_state = OUT_S_QUEUE_DATA;
1536 if (front_len(msg)) {
1537 con->v2.out_epil.front_crc = crc32c(-1, msg->front.iov_base,
1539 add_out_kvec(con, msg->front.iov_base, front_len(msg));
1541 /* middle (at least) is there, checked above */
1542 con->v2.out_epil.front_crc = -1;
1545 if (middle_len(msg)) {
1546 con->v2.out_epil.middle_crc =
1547 crc32c(-1, msg->middle->vec.iov_base, middle_len(msg));
1548 add_out_kvec(con, msg->middle->vec.iov_base, middle_len(msg));
1550 con->v2.out_epil.middle_crc = data_len(msg) ? -1 : 0;
1553 if (data_len(msg)) {
1554 con->v2.out_state = OUT_S_QUEUE_DATA;
1556 con->v2.out_epil.data_crc = 0;
1557 prepare_epilogue_plain(con, false);
1558 con->v2.out_state = OUT_S_FINISH_MESSAGE;
1563 * Unfortunately the kernel crypto API doesn't support streaming
1564 * (piecewise) operation for AEAD algorithms, so we can't get away
1565 * with a fixed size buffer and a couple sgs. Instead, we have to
1566 * allocate pages for the entire tail of the message (currently up
1567 * to ~32M) and two sgs arrays (up to ~256K each)...
1569 static int prepare_message_secure(struct ceph_connection *con)
1571 void *zerop = page_address(ceph_zero_page);
1572 struct sg_table enc_sgt = {};
1573 struct sg_table sgt = {};
1574 struct page **enc_pages;
1579 ret = prepare_head_secure_small(con, con->v2.out_buf,
1580 sizeof(struct ceph_msg_header2));
1584 tail_len = tail_onwire_len(con->out_msg, true);
1587 * Empty message: once the head is written,
1588 * we are done -- there is no epilogue.
1590 con->v2.out_state = OUT_S_FINISH_MESSAGE;
1594 encode_epilogue_secure(con, false);
1595 ret = setup_message_sgs(&sgt, con->out_msg, zerop, zerop, zerop,
1596 &con->v2.out_epil, false);
1600 enc_page_cnt = calc_pages_for(0, tail_len);
1601 enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO);
1602 if (IS_ERR(enc_pages)) {
1603 ret = PTR_ERR(enc_pages);
1607 WARN_ON(con->v2.out_enc_pages || con->v2.out_enc_page_cnt);
1608 con->v2.out_enc_pages = enc_pages;
1609 con->v2.out_enc_page_cnt = enc_page_cnt;
1610 con->v2.out_enc_resid = tail_len;
1611 con->v2.out_enc_i = 0;
1613 ret = sg_alloc_table_from_pages(&enc_sgt, enc_pages, enc_page_cnt,
1614 0, tail_len, GFP_NOIO);
1618 ret = gcm_crypt(con, true, sgt.sgl, enc_sgt.sgl,
1619 tail_len - CEPH_GCM_TAG_LEN);
1623 dout("%s con %p msg %p sg_cnt %d enc_page_cnt %d\n", __func__, con,
1624 con->out_msg, sgt.orig_nents, enc_page_cnt);
1625 con->v2.out_state = OUT_S_QUEUE_ENC_PAGE;
1628 sg_free_table(&sgt);
1629 sg_free_table(&enc_sgt);
1633 static int prepare_message(struct ceph_connection *con)
1636 sizeof(struct ceph_msg_header2),
1637 front_len(con->out_msg),
1638 middle_len(con->out_msg),
1639 data_len(con->out_msg)
1641 struct ceph_frame_desc desc;
1644 dout("%s con %p msg %p logical %d+%d+%d+%d\n", __func__, con,
1645 con->out_msg, lens[0], lens[1], lens[2], lens[3]);
1647 if (con->in_seq > con->in_seq_acked) {
1648 dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con,
1649 con->in_seq_acked, con->in_seq);
1650 con->in_seq_acked = con->in_seq;
1653 reset_out_kvecs(con);
1654 init_frame_desc(&desc, FRAME_TAG_MESSAGE, lens, 4);
1655 encode_preamble(&desc, con->v2.out_buf);
1656 fill_header2(CTRL_BODY(con->v2.out_buf), &con->out_msg->hdr,
1659 if (con_secure(con)) {
1660 ret = prepare_message_secure(con);
1664 prepare_message_plain(con);
1667 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1671 static int prepare_read_banner_prefix(struct ceph_connection *con)
1675 buf = alloc_conn_buf(con, CEPH_BANNER_V2_PREFIX_LEN);
1679 reset_in_kvecs(con);
1680 add_in_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN);
1681 add_in_sign_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN);
1682 con->state = CEPH_CON_S_V2_BANNER_PREFIX;
1686 static int prepare_read_banner_payload(struct ceph_connection *con,
1691 buf = alloc_conn_buf(con, payload_len);
1695 reset_in_kvecs(con);
1696 add_in_kvec(con, buf, payload_len);
1697 add_in_sign_kvec(con, buf, payload_len);
1698 con->state = CEPH_CON_S_V2_BANNER_PAYLOAD;
1702 static void prepare_read_preamble(struct ceph_connection *con)
1704 reset_in_kvecs(con);
1705 add_in_kvec(con, con->v2.in_buf,
1706 con_secure(con) ? CEPH_PREAMBLE_SECURE_LEN :
1707 CEPH_PREAMBLE_PLAIN_LEN);
1708 con->v2.in_state = IN_S_HANDLE_PREAMBLE;
1711 static int prepare_read_control(struct ceph_connection *con)
1713 int ctrl_len = con->v2.in_desc.fd_lens[0];
1717 reset_in_kvecs(con);
1718 if (con->state == CEPH_CON_S_V2_HELLO ||
1719 con->state == CEPH_CON_S_V2_AUTH) {
1720 head_len = head_onwire_len(ctrl_len, false);
1721 buf = alloc_conn_buf(con, head_len);
1725 /* preserve preamble */
1726 memcpy(buf, con->v2.in_buf, CEPH_PREAMBLE_LEN);
1728 add_in_kvec(con, CTRL_BODY(buf), ctrl_len);
1729 add_in_kvec(con, CTRL_BODY(buf) + ctrl_len, CEPH_CRC_LEN);
1730 add_in_sign_kvec(con, buf, head_len);
1732 if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) {
1733 buf = alloc_conn_buf(con, ctrl_len);
1737 add_in_kvec(con, buf, ctrl_len);
1739 add_in_kvec(con, CTRL_BODY(con->v2.in_buf), ctrl_len);
1741 add_in_kvec(con, con->v2.in_buf, CEPH_CRC_LEN);
1743 con->v2.in_state = IN_S_HANDLE_CONTROL;
1747 static int prepare_read_control_remainder(struct ceph_connection *con)
1749 int ctrl_len = con->v2.in_desc.fd_lens[0];
1750 int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1753 buf = alloc_conn_buf(con, ctrl_len);
1757 memcpy(buf, CTRL_BODY(con->v2.in_buf), CEPH_PREAMBLE_INLINE_LEN);
1759 reset_in_kvecs(con);
1760 add_in_kvec(con, buf + CEPH_PREAMBLE_INLINE_LEN, rem_len);
1761 add_in_kvec(con, con->v2.in_buf,
1762 padding_len(rem_len) + CEPH_GCM_TAG_LEN);
1763 con->v2.in_state = IN_S_HANDLE_CONTROL_REMAINDER;
1767 static int prepare_read_data(struct ceph_connection *con)
1771 con->in_data_crc = -1;
1772 ceph_msg_data_cursor_init(&con->v2.in_cursor, con->in_msg,
1773 data_len(con->in_msg));
1775 get_bvec_at(&con->v2.in_cursor, &bv);
1776 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1777 if (unlikely(!con->bounce_page)) {
1778 con->bounce_page = alloc_page(GFP_NOIO);
1779 if (!con->bounce_page) {
1780 pr_err("failed to allocate bounce page\n");
1785 bv.bv_page = con->bounce_page;
1788 set_in_bvec(con, &bv);
1789 con->v2.in_state = IN_S_PREPARE_READ_DATA_CONT;
1793 static void prepare_read_data_cont(struct ceph_connection *con)
1797 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1798 con->in_data_crc = crc32c(con->in_data_crc,
1799 page_address(con->bounce_page),
1800 con->v2.in_bvec.bv_len);
1802 get_bvec_at(&con->v2.in_cursor, &bv);
1803 memcpy_to_page(bv.bv_page, bv.bv_offset,
1804 page_address(con->bounce_page),
1805 con->v2.in_bvec.bv_len);
1807 con->in_data_crc = ceph_crc32c_page(con->in_data_crc,
1808 con->v2.in_bvec.bv_page,
1809 con->v2.in_bvec.bv_offset,
1810 con->v2.in_bvec.bv_len);
1813 ceph_msg_data_advance(&con->v2.in_cursor, con->v2.in_bvec.bv_len);
1814 if (con->v2.in_cursor.total_resid) {
1815 get_bvec_at(&con->v2.in_cursor, &bv);
1816 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1817 bv.bv_page = con->bounce_page;
1820 set_in_bvec(con, &bv);
1821 WARN_ON(con->v2.in_state != IN_S_PREPARE_READ_DATA_CONT);
1826 * We've read all data. Prepare to read epilogue.
1828 reset_in_kvecs(con);
1829 add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
1830 con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1833 static int prepare_read_tail_plain(struct ceph_connection *con)
1835 struct ceph_msg *msg = con->in_msg;
1837 if (!front_len(msg) && !middle_len(msg)) {
1838 WARN_ON(!data_len(msg));
1839 return prepare_read_data(con);
1842 reset_in_kvecs(con);
1843 if (front_len(msg)) {
1844 add_in_kvec(con, msg->front.iov_base, front_len(msg));
1845 WARN_ON(msg->front.iov_len != front_len(msg));
1847 if (middle_len(msg)) {
1848 add_in_kvec(con, msg->middle->vec.iov_base, middle_len(msg));
1849 WARN_ON(msg->middle->vec.iov_len != middle_len(msg));
1852 if (data_len(msg)) {
1853 con->v2.in_state = IN_S_PREPARE_READ_DATA;
1855 add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
1856 con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1861 static void prepare_read_enc_page(struct ceph_connection *con)
1865 dout("%s con %p i %d resid %d\n", __func__, con, con->v2.in_enc_i,
1866 con->v2.in_enc_resid);
1867 WARN_ON(!con->v2.in_enc_resid);
1869 bv.bv_page = con->v2.in_enc_pages[con->v2.in_enc_i];
1871 bv.bv_len = min(con->v2.in_enc_resid, (int)PAGE_SIZE);
1873 set_in_bvec(con, &bv);
1875 con->v2.in_enc_resid -= bv.bv_len;
1877 if (con->v2.in_enc_resid) {
1878 con->v2.in_state = IN_S_PREPARE_READ_ENC_PAGE;
1883 * We are set to read the last piece of ciphertext (ending
1884 * with epilogue) + auth tag.
1886 WARN_ON(con->v2.in_enc_i != con->v2.in_enc_page_cnt);
1887 con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1890 static int prepare_read_tail_secure(struct ceph_connection *con)
1892 struct page **enc_pages;
1896 tail_len = tail_onwire_len(con->in_msg, true);
1899 enc_page_cnt = calc_pages_for(0, tail_len);
1900 enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO);
1901 if (IS_ERR(enc_pages))
1902 return PTR_ERR(enc_pages);
1904 WARN_ON(con->v2.in_enc_pages || con->v2.in_enc_page_cnt);
1905 con->v2.in_enc_pages = enc_pages;
1906 con->v2.in_enc_page_cnt = enc_page_cnt;
1907 con->v2.in_enc_resid = tail_len;
1908 con->v2.in_enc_i = 0;
1910 prepare_read_enc_page(con);
1914 static void __finish_skip(struct ceph_connection *con)
1917 prepare_read_preamble(con);
1920 static void prepare_skip_message(struct ceph_connection *con)
1922 struct ceph_frame_desc *desc = &con->v2.in_desc;
1925 dout("%s con %p %d+%d+%d\n", __func__, con, desc->fd_lens[1],
1926 desc->fd_lens[2], desc->fd_lens[3]);
1928 tail_len = __tail_onwire_len(desc->fd_lens[1], desc->fd_lens[2],
1929 desc->fd_lens[3], con_secure(con));
1933 set_in_skip(con, tail_len);
1934 con->v2.in_state = IN_S_FINISH_SKIP;
1938 static int process_banner_prefix(struct ceph_connection *con)
1943 WARN_ON(con->v2.in_kvecs[0].iov_len != CEPH_BANNER_V2_PREFIX_LEN);
1945 p = con->v2.in_kvecs[0].iov_base;
1946 if (memcmp(p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN)) {
1947 if (!memcmp(p, CEPH_BANNER, CEPH_BANNER_LEN))
1948 con->error_msg = "server is speaking msgr1 protocol";
1950 con->error_msg = "protocol error, bad banner";
1954 p += CEPH_BANNER_V2_LEN;
1955 payload_len = ceph_decode_16(&p);
1956 dout("%s con %p payload_len %d\n", __func__, con, payload_len);
1958 return prepare_read_banner_payload(con, payload_len);
1961 static int process_banner_payload(struct ceph_connection *con)
1963 void *end = con->v2.in_kvecs[0].iov_base + con->v2.in_kvecs[0].iov_len;
1964 u64 feat = CEPH_MSGR2_SUPPORTED_FEATURES;
1965 u64 req_feat = CEPH_MSGR2_REQUIRED_FEATURES;
1966 u64 server_feat, server_req_feat;
1970 p = con->v2.in_kvecs[0].iov_base;
1971 ceph_decode_64_safe(&p, end, server_feat, bad);
1972 ceph_decode_64_safe(&p, end, server_req_feat, bad);
1974 dout("%s con %p server_feat 0x%llx server_req_feat 0x%llx\n",
1975 __func__, con, server_feat, server_req_feat);
1977 if (req_feat & ~server_feat) {
1978 pr_err("msgr2 feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n",
1979 server_feat, req_feat & ~server_feat);
1980 con->error_msg = "missing required protocol features";
1983 if (server_req_feat & ~feat) {
1984 pr_err("msgr2 feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n",
1985 feat, server_req_feat & ~feat);
1986 con->error_msg = "missing required protocol features";
1990 /* no reset_out_kvecs() as our banner may still be pending */
1991 ret = prepare_hello(con);
1993 pr_err("prepare_hello failed: %d\n", ret);
1997 con->state = CEPH_CON_S_V2_HELLO;
1998 prepare_read_preamble(con);
2002 pr_err("failed to decode banner payload\n");
2006 static int process_hello(struct ceph_connection *con, void *p, void *end)
2008 struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
2009 struct ceph_entity_addr addr_for_me;
2013 if (con->state != CEPH_CON_S_V2_HELLO) {
2014 con->error_msg = "protocol error, unexpected hello";
2018 ceph_decode_8_safe(&p, end, entity_type, bad);
2019 ret = ceph_decode_entity_addr(&p, end, &addr_for_me);
2021 pr_err("failed to decode addr_for_me: %d\n", ret);
2025 dout("%s con %p entity_type %d addr_for_me %s\n", __func__, con,
2026 entity_type, ceph_pr_addr(&addr_for_me));
2028 if (entity_type != con->peer_name.type) {
2029 pr_err("bad peer type, want %d, got %d\n",
2030 con->peer_name.type, entity_type);
2031 con->error_msg = "wrong peer at address";
2036 * Set our address to the address our first peer (i.e. monitor)
2037 * sees that we are connecting from. If we are behind some sort
2038 * of NAT and want to be identified by some private (not NATed)
2039 * address, ip option should be used.
2041 if (ceph_addr_is_blank(my_addr)) {
2042 memcpy(&my_addr->in_addr, &addr_for_me.in_addr,
2043 sizeof(my_addr->in_addr));
2044 ceph_addr_set_port(my_addr, 0);
2045 dout("%s con %p set my addr %s, as seen by peer %s\n",
2046 __func__, con, ceph_pr_addr(my_addr),
2047 ceph_pr_addr(&con->peer_addr));
2049 dout("%s con %p my addr already set %s\n",
2050 __func__, con, ceph_pr_addr(my_addr));
2053 WARN_ON(ceph_addr_is_blank(my_addr) || ceph_addr_port(my_addr));
2054 WARN_ON(my_addr->type != CEPH_ENTITY_ADDR_TYPE_ANY);
2055 WARN_ON(!my_addr->nonce);
2057 /* no reset_out_kvecs() as our hello may still be pending */
2058 ret = prepare_auth_request(con);
2061 pr_err("prepare_auth_request failed: %d\n", ret);
2065 con->state = CEPH_CON_S_V2_AUTH;
2069 pr_err("failed to decode hello\n");
2073 static int process_auth_bad_method(struct ceph_connection *con,
2076 int allowed_protos[8], allowed_modes[8];
2077 int allowed_proto_cnt, allowed_mode_cnt;
2078 int used_proto, result;
2082 if (con->state != CEPH_CON_S_V2_AUTH) {
2083 con->error_msg = "protocol error, unexpected auth_bad_method";
2087 ceph_decode_32_safe(&p, end, used_proto, bad);
2088 ceph_decode_32_safe(&p, end, result, bad);
2089 dout("%s con %p used_proto %d result %d\n", __func__, con, used_proto,
2092 ceph_decode_32_safe(&p, end, allowed_proto_cnt, bad);
2093 if (allowed_proto_cnt > ARRAY_SIZE(allowed_protos)) {
2094 pr_err("allowed_protos too big %d\n", allowed_proto_cnt);
2097 for (i = 0; i < allowed_proto_cnt; i++) {
2098 ceph_decode_32_safe(&p, end, allowed_protos[i], bad);
2099 dout("%s con %p allowed_protos[%d] %d\n", __func__, con,
2100 i, allowed_protos[i]);
2103 ceph_decode_32_safe(&p, end, allowed_mode_cnt, bad);
2104 if (allowed_mode_cnt > ARRAY_SIZE(allowed_modes)) {
2105 pr_err("allowed_modes too big %d\n", allowed_mode_cnt);
2108 for (i = 0; i < allowed_mode_cnt; i++) {
2109 ceph_decode_32_safe(&p, end, allowed_modes[i], bad);
2110 dout("%s con %p allowed_modes[%d] %d\n", __func__, con,
2111 i, allowed_modes[i]);
2114 mutex_unlock(&con->mutex);
2115 ret = con->ops->handle_auth_bad_method(con, used_proto, result,
2120 mutex_lock(&con->mutex);
2121 if (con->state != CEPH_CON_S_V2_AUTH) {
2122 dout("%s con %p state changed to %d\n", __func__, con,
2127 dout("%s con %p handle_auth_bad_method ret %d\n", __func__, con, ret);
2131 pr_err("failed to decode auth_bad_method\n");
2135 static int process_auth_reply_more(struct ceph_connection *con,
2141 if (con->state != CEPH_CON_S_V2_AUTH) {
2142 con->error_msg = "protocol error, unexpected auth_reply_more";
2146 ceph_decode_32_safe(&p, end, payload_len, bad);
2147 ceph_decode_need(&p, end, payload_len, bad);
2149 dout("%s con %p payload_len %d\n", __func__, con, payload_len);
2151 reset_out_kvecs(con);
2152 ret = prepare_auth_request_more(con, p, payload_len);
2155 pr_err("prepare_auth_request_more failed: %d\n", ret);
2162 pr_err("failed to decode auth_reply_more\n");
2167 * Align session_key and con_secret to avoid GFP_ATOMIC allocation
2168 * inside crypto_shash_setkey() and crypto_aead_setkey() called from
2169 * setup_crypto(). __aligned(16) isn't guaranteed to work for stack
2170 * objects, so do it by hand.
2172 static int process_auth_done(struct ceph_connection *con, void *p, void *end)
2174 u8 session_key_buf[CEPH_KEY_LEN + 16];
2175 u8 con_secret_buf[CEPH_MAX_CON_SECRET_LEN + 16];
2176 u8 *session_key = PTR_ALIGN(&session_key_buf[0], 16);
2177 u8 *con_secret = PTR_ALIGN(&con_secret_buf[0], 16);
2178 int session_key_len, con_secret_len;
2183 if (con->state != CEPH_CON_S_V2_AUTH) {
2184 con->error_msg = "protocol error, unexpected auth_done";
2188 ceph_decode_64_safe(&p, end, global_id, bad);
2189 ceph_decode_32_safe(&p, end, con->v2.con_mode, bad);
2190 ceph_decode_32_safe(&p, end, payload_len, bad);
2192 dout("%s con %p global_id %llu con_mode %d payload_len %d\n",
2193 __func__, con, global_id, con->v2.con_mode, payload_len);
2195 mutex_unlock(&con->mutex);
2196 session_key_len = 0;
2198 ret = con->ops->handle_auth_done(con, global_id, p, payload_len,
2199 session_key, &session_key_len,
2200 con_secret, &con_secret_len);
2201 mutex_lock(&con->mutex);
2202 if (con->state != CEPH_CON_S_V2_AUTH) {
2203 dout("%s con %p state changed to %d\n", __func__, con,
2209 dout("%s con %p handle_auth_done ret %d\n", __func__, con, ret);
2213 ret = setup_crypto(con, session_key, session_key_len, con_secret,
2218 reset_out_kvecs(con);
2219 ret = prepare_auth_signature(con);
2221 pr_err("prepare_auth_signature failed: %d\n", ret);
2225 con->state = CEPH_CON_S_V2_AUTH_SIGNATURE;
2228 memzero_explicit(session_key_buf, sizeof(session_key_buf));
2229 memzero_explicit(con_secret_buf, sizeof(con_secret_buf));
2233 pr_err("failed to decode auth_done\n");
2237 static int process_auth_signature(struct ceph_connection *con,
2240 u8 hmac[SHA256_DIGEST_SIZE];
2243 if (con->state != CEPH_CON_S_V2_AUTH_SIGNATURE) {
2244 con->error_msg = "protocol error, unexpected auth_signature";
2248 ret = hmac_sha256(con, con->v2.out_sign_kvecs,
2249 con->v2.out_sign_kvec_cnt, hmac);
2253 ceph_decode_need(&p, end, SHA256_DIGEST_SIZE, bad);
2254 if (crypto_memneq(p, hmac, SHA256_DIGEST_SIZE)) {
2255 con->error_msg = "integrity error, bad auth signature";
2259 dout("%s con %p auth signature ok\n", __func__, con);
2261 /* no reset_out_kvecs() as our auth_signature may still be pending */
2262 if (!con->v2.server_cookie) {
2263 ret = prepare_client_ident(con);
2265 pr_err("prepare_client_ident failed: %d\n", ret);
2269 con->state = CEPH_CON_S_V2_SESSION_CONNECT;
2271 ret = prepare_session_reconnect(con);
2273 pr_err("prepare_session_reconnect failed: %d\n", ret);
2277 con->state = CEPH_CON_S_V2_SESSION_RECONNECT;
2283 pr_err("failed to decode auth_signature\n");
2287 static int process_server_ident(struct ceph_connection *con,
2290 struct ceph_client *client = from_msgr(con->msgr);
2291 u64 features, required_features;
2292 struct ceph_entity_addr addr;
2299 if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) {
2300 con->error_msg = "protocol error, unexpected server_ident";
2304 ret = ceph_decode_entity_addrvec(&p, end, true, &addr);
2306 pr_err("failed to decode server addrs: %d\n", ret);
2310 ceph_decode_64_safe(&p, end, global_id, bad);
2311 ceph_decode_64_safe(&p, end, global_seq, bad);
2312 ceph_decode_64_safe(&p, end, features, bad);
2313 ceph_decode_64_safe(&p, end, required_features, bad);
2314 ceph_decode_64_safe(&p, end, flags, bad);
2315 ceph_decode_64_safe(&p, end, cookie, bad);
2317 dout("%s con %p addr %s/%u global_id %llu global_seq %llu features 0x%llx required_features 0x%llx flags 0x%llx cookie 0x%llx\n",
2318 __func__, con, ceph_pr_addr(&addr), le32_to_cpu(addr.nonce),
2319 global_id, global_seq, features, required_features, flags, cookie);
2321 /* is this who we intended to talk to? */
2322 if (memcmp(&addr, &con->peer_addr, sizeof(con->peer_addr))) {
2323 pr_err("bad peer addr/nonce, want %s/%u, got %s/%u\n",
2324 ceph_pr_addr(&con->peer_addr),
2325 le32_to_cpu(con->peer_addr.nonce),
2326 ceph_pr_addr(&addr), le32_to_cpu(addr.nonce));
2327 con->error_msg = "wrong peer at address";
2331 if (client->required_features & ~features) {
2332 pr_err("RADOS feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n",
2333 features, client->required_features & ~features);
2334 con->error_msg = "missing required protocol features";
2339 * Both name->type and name->num are set in ceph_con_open() but
2340 * name->num may be bogus in the initial monmap. name->type is
2341 * verified in handle_hello().
2343 WARN_ON(!con->peer_name.type);
2344 con->peer_name.num = cpu_to_le64(global_id);
2345 con->v2.peer_global_seq = global_seq;
2346 con->peer_features = features;
2347 WARN_ON(required_features & ~client->supported_features);
2348 con->v2.server_cookie = cookie;
2350 if (flags & CEPH_MSG_CONNECT_LOSSY) {
2351 ceph_con_flag_set(con, CEPH_CON_F_LOSSYTX);
2352 WARN_ON(con->v2.server_cookie);
2354 WARN_ON(!con->v2.server_cookie);
2357 clear_in_sign_kvecs(con);
2358 clear_out_sign_kvecs(con);
2359 free_conn_bufs(con);
2360 con->delay = 0; /* reset backoff memory */
2362 con->state = CEPH_CON_S_OPEN;
2363 con->v2.out_state = OUT_S_GET_NEXT;
2367 pr_err("failed to decode server_ident\n");
2371 static int process_ident_missing_features(struct ceph_connection *con,
2374 struct ceph_client *client = from_msgr(con->msgr);
2375 u64 missing_features;
2377 if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) {
2378 con->error_msg = "protocol error, unexpected ident_missing_features";
2382 ceph_decode_64_safe(&p, end, missing_features, bad);
2383 pr_err("RADOS feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n",
2384 client->supported_features, missing_features);
2385 con->error_msg = "missing required protocol features";
2389 pr_err("failed to decode ident_missing_features\n");
2393 static int process_session_reconnect_ok(struct ceph_connection *con,
2398 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2399 con->error_msg = "protocol error, unexpected session_reconnect_ok";
2403 ceph_decode_64_safe(&p, end, seq, bad);
2405 dout("%s con %p seq %llu\n", __func__, con, seq);
2406 ceph_con_discard_requeued(con, seq);
2408 clear_in_sign_kvecs(con);
2409 clear_out_sign_kvecs(con);
2410 free_conn_bufs(con);
2411 con->delay = 0; /* reset backoff memory */
2413 con->state = CEPH_CON_S_OPEN;
2414 con->v2.out_state = OUT_S_GET_NEXT;
2418 pr_err("failed to decode session_reconnect_ok\n");
2422 static int process_session_retry(struct ceph_connection *con,
2428 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2429 con->error_msg = "protocol error, unexpected session_retry";
2433 ceph_decode_64_safe(&p, end, connect_seq, bad);
2435 dout("%s con %p connect_seq %llu\n", __func__, con, connect_seq);
2436 WARN_ON(connect_seq <= con->v2.connect_seq);
2437 con->v2.connect_seq = connect_seq + 1;
2439 free_conn_bufs(con);
2441 reset_out_kvecs(con);
2442 ret = prepare_session_reconnect(con);
2444 pr_err("prepare_session_reconnect (cseq) failed: %d\n", ret);
2451 pr_err("failed to decode session_retry\n");
2455 static int process_session_retry_global(struct ceph_connection *con,
2461 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2462 con->error_msg = "protocol error, unexpected session_retry_global";
2466 ceph_decode_64_safe(&p, end, global_seq, bad);
2468 dout("%s con %p global_seq %llu\n", __func__, con, global_seq);
2469 WARN_ON(global_seq <= con->v2.global_seq);
2470 con->v2.global_seq = ceph_get_global_seq(con->msgr, global_seq);
2472 free_conn_bufs(con);
2474 reset_out_kvecs(con);
2475 ret = prepare_session_reconnect(con);
2477 pr_err("prepare_session_reconnect (gseq) failed: %d\n", ret);
2484 pr_err("failed to decode session_retry_global\n");
2488 static int process_session_reset(struct ceph_connection *con,
2494 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2495 con->error_msg = "protocol error, unexpected session_reset";
2499 ceph_decode_8_safe(&p, end, full, bad);
2501 con->error_msg = "protocol error, bad session_reset";
2505 pr_info("%s%lld %s session reset\n", ENTITY_NAME(con->peer_name),
2506 ceph_pr_addr(&con->peer_addr));
2507 ceph_con_reset_session(con);
2509 mutex_unlock(&con->mutex);
2510 if (con->ops->peer_reset)
2511 con->ops->peer_reset(con);
2512 mutex_lock(&con->mutex);
2513 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2514 dout("%s con %p state changed to %d\n", __func__, con,
2519 free_conn_bufs(con);
2521 reset_out_kvecs(con);
2522 ret = prepare_client_ident(con);
2524 pr_err("prepare_client_ident (rst) failed: %d\n", ret);
2528 con->state = CEPH_CON_S_V2_SESSION_CONNECT;
2532 pr_err("failed to decode session_reset\n");
2536 static int process_keepalive2_ack(struct ceph_connection *con,
2539 if (con->state != CEPH_CON_S_OPEN) {
2540 con->error_msg = "protocol error, unexpected keepalive2_ack";
2544 ceph_decode_need(&p, end, sizeof(struct ceph_timespec), bad);
2545 ceph_decode_timespec64(&con->last_keepalive_ack, p);
2547 dout("%s con %p timestamp %lld.%09ld\n", __func__, con,
2548 con->last_keepalive_ack.tv_sec, con->last_keepalive_ack.tv_nsec);
2553 pr_err("failed to decode keepalive2_ack\n");
2557 static int process_ack(struct ceph_connection *con, void *p, void *end)
2561 if (con->state != CEPH_CON_S_OPEN) {
2562 con->error_msg = "protocol error, unexpected ack";
2566 ceph_decode_64_safe(&p, end, seq, bad);
2568 dout("%s con %p seq %llu\n", __func__, con, seq);
2569 ceph_con_discard_sent(con, seq);
2573 pr_err("failed to decode ack\n");
2577 static int process_control(struct ceph_connection *con, void *p, void *end)
2579 int tag = con->v2.in_desc.fd_tag;
2582 dout("%s con %p tag %d len %d\n", __func__, con, tag, (int)(end - p));
2585 case FRAME_TAG_HELLO:
2586 ret = process_hello(con, p, end);
2588 case FRAME_TAG_AUTH_BAD_METHOD:
2589 ret = process_auth_bad_method(con, p, end);
2591 case FRAME_TAG_AUTH_REPLY_MORE:
2592 ret = process_auth_reply_more(con, p, end);
2594 case FRAME_TAG_AUTH_DONE:
2595 ret = process_auth_done(con, p, end);
2597 case FRAME_TAG_AUTH_SIGNATURE:
2598 ret = process_auth_signature(con, p, end);
2600 case FRAME_TAG_SERVER_IDENT:
2601 ret = process_server_ident(con, p, end);
2603 case FRAME_TAG_IDENT_MISSING_FEATURES:
2604 ret = process_ident_missing_features(con, p, end);
2606 case FRAME_TAG_SESSION_RECONNECT_OK:
2607 ret = process_session_reconnect_ok(con, p, end);
2609 case FRAME_TAG_SESSION_RETRY:
2610 ret = process_session_retry(con, p, end);
2612 case FRAME_TAG_SESSION_RETRY_GLOBAL:
2613 ret = process_session_retry_global(con, p, end);
2615 case FRAME_TAG_SESSION_RESET:
2616 ret = process_session_reset(con, p, end);
2618 case FRAME_TAG_KEEPALIVE2_ACK:
2619 ret = process_keepalive2_ack(con, p, end);
2622 ret = process_ack(con, p, end);
2625 pr_err("bad tag %d\n", tag);
2626 con->error_msg = "protocol error, bad tag";
2630 dout("%s con %p error %d\n", __func__, con, ret);
2634 prepare_read_preamble(con);
2640 * 1 - con->in_msg set, read message
2644 static int process_message_header(struct ceph_connection *con,
2647 struct ceph_frame_desc *desc = &con->v2.in_desc;
2648 struct ceph_msg_header2 *hdr2 = p;
2649 struct ceph_msg_header hdr;
2655 seq = le64_to_cpu(hdr2->seq);
2656 if ((s64)seq - (s64)con->in_seq < 1) {
2657 pr_info("%s%lld %s skipping old message: seq %llu, expected %llu\n",
2658 ENTITY_NAME(con->peer_name),
2659 ceph_pr_addr(&con->peer_addr),
2660 seq, con->in_seq + 1);
2663 if ((s64)seq - (s64)con->in_seq > 1) {
2664 pr_err("bad seq %llu, expected %llu\n", seq, con->in_seq + 1);
2665 con->error_msg = "bad message sequence # for incoming message";
2669 ceph_con_discard_sent(con, le64_to_cpu(hdr2->ack_seq));
2671 fill_header(&hdr, hdr2, desc->fd_lens[1], desc->fd_lens[2],
2672 desc->fd_lens[3], &con->peer_name);
2673 ret = ceph_con_in_msg_alloc(con, &hdr, &skip);
2677 WARN_ON(!con->in_msg ^ skip);
2681 WARN_ON(!con->in_msg);
2682 WARN_ON(con->in_msg->con != con);
2686 static int process_message(struct ceph_connection *con)
2688 ceph_con_process_message(con);
2691 * We could have been closed by ceph_con_close() because
2692 * ceph_con_process_message() temporarily drops con->mutex.
2694 if (con->state != CEPH_CON_S_OPEN) {
2695 dout("%s con %p state changed to %d\n", __func__, con,
2700 prepare_read_preamble(con);
2704 static int __handle_control(struct ceph_connection *con, void *p)
2706 void *end = p + con->v2.in_desc.fd_lens[0];
2707 struct ceph_msg *msg;
2710 if (con->v2.in_desc.fd_tag != FRAME_TAG_MESSAGE)
2711 return process_control(con, p, end);
2713 ret = process_message_header(con, p, end);
2717 prepare_skip_message(con);
2721 msg = con->in_msg; /* set in process_message_header() */
2722 if (front_len(msg)) {
2723 WARN_ON(front_len(msg) > msg->front_alloc_len);
2724 msg->front.iov_len = front_len(msg);
2726 msg->front.iov_len = 0;
2728 if (middle_len(msg)) {
2729 WARN_ON(middle_len(msg) > msg->middle->alloc_len);
2730 msg->middle->vec.iov_len = middle_len(msg);
2731 } else if (msg->middle) {
2732 msg->middle->vec.iov_len = 0;
2735 if (!front_len(msg) && !middle_len(msg) && !data_len(msg))
2736 return process_message(con);
2738 if (con_secure(con))
2739 return prepare_read_tail_secure(con);
2741 return prepare_read_tail_plain(con);
2744 static int handle_preamble(struct ceph_connection *con)
2746 struct ceph_frame_desc *desc = &con->v2.in_desc;
2749 if (con_secure(con)) {
2750 ret = decrypt_preamble(con);
2752 if (ret == -EBADMSG)
2753 con->error_msg = "integrity error, bad preamble auth tag";
2758 ret = decode_preamble(con->v2.in_buf, desc);
2760 if (ret == -EBADMSG)
2761 con->error_msg = "integrity error, bad crc";
2763 con->error_msg = "protocol error, bad preamble";
2767 dout("%s con %p tag %d seg_cnt %d %d+%d+%d+%d\n", __func__,
2768 con, desc->fd_tag, desc->fd_seg_cnt, desc->fd_lens[0],
2769 desc->fd_lens[1], desc->fd_lens[2], desc->fd_lens[3]);
2771 if (!con_secure(con))
2772 return prepare_read_control(con);
2774 if (desc->fd_lens[0] > CEPH_PREAMBLE_INLINE_LEN)
2775 return prepare_read_control_remainder(con);
2777 return __handle_control(con, CTRL_BODY(con->v2.in_buf));
2780 static int handle_control(struct ceph_connection *con)
2782 int ctrl_len = con->v2.in_desc.fd_lens[0];
2786 WARN_ON(con_secure(con));
2788 ret = verify_control_crc(con);
2790 con->error_msg = "integrity error, bad crc";
2794 if (con->state == CEPH_CON_S_V2_AUTH) {
2795 buf = alloc_conn_buf(con, ctrl_len);
2799 memcpy(buf, con->v2.in_kvecs[0].iov_base, ctrl_len);
2800 return __handle_control(con, buf);
2803 return __handle_control(con, con->v2.in_kvecs[0].iov_base);
2806 static int handle_control_remainder(struct ceph_connection *con)
2810 WARN_ON(!con_secure(con));
2812 ret = decrypt_control_remainder(con);
2814 if (ret == -EBADMSG)
2815 con->error_msg = "integrity error, bad control remainder auth tag";
2819 return __handle_control(con, con->v2.in_kvecs[0].iov_base -
2820 CEPH_PREAMBLE_INLINE_LEN);
2823 static int handle_epilogue(struct ceph_connection *con)
2825 u32 front_crc, middle_crc, data_crc;
2828 if (con_secure(con)) {
2829 ret = decrypt_tail(con);
2831 if (ret == -EBADMSG)
2832 con->error_msg = "integrity error, bad epilogue auth tag";
2836 /* just late_status */
2837 ret = decode_epilogue(con->v2.in_buf, NULL, NULL, NULL);
2839 con->error_msg = "protocol error, bad epilogue";
2843 ret = decode_epilogue(con->v2.in_buf, &front_crc,
2844 &middle_crc, &data_crc);
2846 con->error_msg = "protocol error, bad epilogue";
2850 ret = verify_epilogue_crcs(con, front_crc, middle_crc,
2853 con->error_msg = "integrity error, bad crc";
2858 return process_message(con);
2861 static void finish_skip(struct ceph_connection *con)
2863 dout("%s con %p\n", __func__, con);
2865 if (con_secure(con))
2866 gcm_inc_nonce(&con->v2.in_gcm_nonce);
2871 static int populate_in_iter(struct ceph_connection *con)
2875 dout("%s con %p state %d in_state %d\n", __func__, con, con->state,
2877 WARN_ON(iov_iter_count(&con->v2.in_iter));
2879 if (con->state == CEPH_CON_S_V2_BANNER_PREFIX) {
2880 ret = process_banner_prefix(con);
2881 } else if (con->state == CEPH_CON_S_V2_BANNER_PAYLOAD) {
2882 ret = process_banner_payload(con);
2883 } else if ((con->state >= CEPH_CON_S_V2_HELLO &&
2884 con->state <= CEPH_CON_S_V2_SESSION_RECONNECT) ||
2885 con->state == CEPH_CON_S_OPEN) {
2886 switch (con->v2.in_state) {
2887 case IN_S_HANDLE_PREAMBLE:
2888 ret = handle_preamble(con);
2890 case IN_S_HANDLE_CONTROL:
2891 ret = handle_control(con);
2893 case IN_S_HANDLE_CONTROL_REMAINDER:
2894 ret = handle_control_remainder(con);
2896 case IN_S_PREPARE_READ_DATA:
2897 ret = prepare_read_data(con);
2899 case IN_S_PREPARE_READ_DATA_CONT:
2900 prepare_read_data_cont(con);
2903 case IN_S_PREPARE_READ_ENC_PAGE:
2904 prepare_read_enc_page(con);
2907 case IN_S_HANDLE_EPILOGUE:
2908 ret = handle_epilogue(con);
2910 case IN_S_FINISH_SKIP:
2915 WARN(1, "bad in_state %d", con->v2.in_state);
2919 WARN(1, "bad state %d", con->state);
2923 dout("%s con %p error %d\n", __func__, con, ret);
2927 if (WARN_ON(!iov_iter_count(&con->v2.in_iter)))
2929 dout("%s con %p populated %zu\n", __func__, con,
2930 iov_iter_count(&con->v2.in_iter));
2934 int ceph_con_v2_try_read(struct ceph_connection *con)
2938 dout("%s con %p state %d need %zu\n", __func__, con, con->state,
2939 iov_iter_count(&con->v2.in_iter));
2941 if (con->state == CEPH_CON_S_PREOPEN)
2945 * We should always have something pending here. If not,
2946 * avoid calling populate_in_iter() as if we read something
2947 * (ceph_tcp_recv() would immediately return 1).
2949 if (WARN_ON(!iov_iter_count(&con->v2.in_iter)))
2953 ret = ceph_tcp_recv(con);
2957 ret = populate_in_iter(con);
2959 if (ret && ret != -EAGAIN && !con->error_msg)
2960 con->error_msg = "read processing error";
2966 static void queue_data(struct ceph_connection *con)
2970 con->v2.out_epil.data_crc = -1;
2971 ceph_msg_data_cursor_init(&con->v2.out_cursor, con->out_msg,
2972 data_len(con->out_msg));
2974 get_bvec_at(&con->v2.out_cursor, &bv);
2975 set_out_bvec(con, &bv, true);
2976 con->v2.out_state = OUT_S_QUEUE_DATA_CONT;
2979 static void queue_data_cont(struct ceph_connection *con)
2983 con->v2.out_epil.data_crc = ceph_crc32c_page(
2984 con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page,
2985 con->v2.out_bvec.bv_offset, con->v2.out_bvec.bv_len);
2987 ceph_msg_data_advance(&con->v2.out_cursor, con->v2.out_bvec.bv_len);
2988 if (con->v2.out_cursor.total_resid) {
2989 get_bvec_at(&con->v2.out_cursor, &bv);
2990 set_out_bvec(con, &bv, true);
2991 WARN_ON(con->v2.out_state != OUT_S_QUEUE_DATA_CONT);
2996 * We've written all data. Queue epilogue. Once it's written,
2999 reset_out_kvecs(con);
3000 prepare_epilogue_plain(con, false);
3001 con->v2.out_state = OUT_S_FINISH_MESSAGE;
3004 static void queue_enc_page(struct ceph_connection *con)
3008 dout("%s con %p i %d resid %d\n", __func__, con, con->v2.out_enc_i,
3009 con->v2.out_enc_resid);
3010 WARN_ON(!con->v2.out_enc_resid);
3012 bv.bv_page = con->v2.out_enc_pages[con->v2.out_enc_i];
3014 bv.bv_len = min(con->v2.out_enc_resid, (int)PAGE_SIZE);
3016 set_out_bvec(con, &bv, false);
3017 con->v2.out_enc_i++;
3018 con->v2.out_enc_resid -= bv.bv_len;
3020 if (con->v2.out_enc_resid) {
3021 WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE);
3026 * We've queued the last piece of ciphertext (ending with
3027 * epilogue) + auth tag. Once it's written, we are done.
3029 WARN_ON(con->v2.out_enc_i != con->v2.out_enc_page_cnt);
3030 con->v2.out_state = OUT_S_FINISH_MESSAGE;
3033 static void queue_zeros(struct ceph_connection *con)
3035 dout("%s con %p out_zero %d\n", __func__, con, con->v2.out_zero);
3037 if (con->v2.out_zero) {
3038 set_out_bvec_zero(con);
3039 con->v2.out_zero -= con->v2.out_bvec.bv_len;
3040 con->v2.out_state = OUT_S_QUEUE_ZEROS;
3045 * We've zero-filled everything up to epilogue. Queue epilogue
3046 * with late_status set to ABORTED and crcs adjusted for zeros.
3047 * Once it's written, we are done patching up for the revoke.
3049 reset_out_kvecs(con);
3050 prepare_epilogue_plain(con, true);
3051 con->v2.out_state = OUT_S_FINISH_MESSAGE;
3054 static void finish_message(struct ceph_connection *con)
3056 dout("%s con %p msg %p\n", __func__, con, con->out_msg);
3058 /* we end up here both plain and secure modes */
3059 if (con->v2.out_enc_pages) {
3060 WARN_ON(!con->v2.out_enc_page_cnt);
3061 ceph_release_page_vector(con->v2.out_enc_pages,
3062 con->v2.out_enc_page_cnt);
3063 con->v2.out_enc_pages = NULL;
3064 con->v2.out_enc_page_cnt = 0;
3066 /* message may have been revoked */
3068 ceph_msg_put(con->out_msg);
3069 con->out_msg = NULL;
3072 con->v2.out_state = OUT_S_GET_NEXT;
3075 static int populate_out_iter(struct ceph_connection *con)
3079 dout("%s con %p state %d out_state %d\n", __func__, con, con->state,
3081 WARN_ON(iov_iter_count(&con->v2.out_iter));
3083 if (con->state != CEPH_CON_S_OPEN) {
3084 WARN_ON(con->state < CEPH_CON_S_V2_BANNER_PREFIX ||
3085 con->state > CEPH_CON_S_V2_SESSION_RECONNECT);
3086 goto nothing_pending;
3089 switch (con->v2.out_state) {
3090 case OUT_S_QUEUE_DATA:
3091 WARN_ON(!con->out_msg);
3094 case OUT_S_QUEUE_DATA_CONT:
3095 WARN_ON(!con->out_msg);
3096 queue_data_cont(con);
3098 case OUT_S_QUEUE_ENC_PAGE:
3099 queue_enc_page(con);
3101 case OUT_S_QUEUE_ZEROS:
3102 WARN_ON(con->out_msg); /* revoked */
3105 case OUT_S_FINISH_MESSAGE:
3106 finish_message(con);
3108 case OUT_S_GET_NEXT:
3111 WARN(1, "bad out_state %d", con->v2.out_state);
3115 WARN_ON(con->v2.out_state != OUT_S_GET_NEXT);
3116 if (ceph_con_flag_test_and_clear(con, CEPH_CON_F_KEEPALIVE_PENDING)) {
3117 ret = prepare_keepalive2(con);
3119 pr_err("prepare_keepalive2 failed: %d\n", ret);
3122 } else if (!list_empty(&con->out_queue)) {
3123 ceph_con_get_out_msg(con);
3124 ret = prepare_message(con);
3126 pr_err("prepare_message failed: %d\n", ret);
3129 } else if (con->in_seq > con->in_seq_acked) {
3130 ret = prepare_ack(con);
3132 pr_err("prepare_ack failed: %d\n", ret);
3136 goto nothing_pending;
3140 if (WARN_ON(!iov_iter_count(&con->v2.out_iter)))
3142 dout("%s con %p populated %zu\n", __func__, con,
3143 iov_iter_count(&con->v2.out_iter));
3147 WARN_ON(iov_iter_count(&con->v2.out_iter));
3148 dout("%s con %p nothing pending\n", __func__, con);
3149 ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
3153 int ceph_con_v2_try_write(struct ceph_connection *con)
3157 dout("%s con %p state %d have %zu\n", __func__, con, con->state,
3158 iov_iter_count(&con->v2.out_iter));
3160 /* open the socket first? */
3161 if (con->state == CEPH_CON_S_PREOPEN) {
3162 WARN_ON(con->peer_addr.type != CEPH_ENTITY_ADDR_TYPE_MSGR2);
3165 * Always bump global_seq. Bump connect_seq only if
3166 * there is a session (i.e. we are reconnecting and will
3167 * send session_reconnect instead of client_ident).
3169 con->v2.global_seq = ceph_get_global_seq(con->msgr, 0);
3170 if (con->v2.server_cookie)
3171 con->v2.connect_seq++;
3173 ret = prepare_read_banner_prefix(con);
3175 pr_err("prepare_read_banner_prefix failed: %d\n", ret);
3176 con->error_msg = "connect error";
3180 reset_out_kvecs(con);
3181 ret = prepare_banner(con);
3183 pr_err("prepare_banner failed: %d\n", ret);
3184 con->error_msg = "connect error";
3188 ret = ceph_tcp_connect(con);
3190 pr_err("ceph_tcp_connect failed: %d\n", ret);
3191 con->error_msg = "connect error";
3196 if (!iov_iter_count(&con->v2.out_iter)) {
3197 ret = populate_out_iter(con);
3199 if (ret && ret != -EAGAIN && !con->error_msg)
3200 con->error_msg = "write processing error";
3205 tcp_sock_set_cork(con->sock->sk, true);
3207 ret = ceph_tcp_send(con);
3211 ret = populate_out_iter(con);
3213 if (ret && ret != -EAGAIN && !con->error_msg)
3214 con->error_msg = "write processing error";
3219 tcp_sock_set_cork(con->sock->sk, false);
3223 static u32 crc32c_zeros(u32 crc, int zero_len)
3228 len = min(zero_len, (int)PAGE_SIZE);
3229 crc = crc32c(crc, page_address(ceph_zero_page), len);
3236 static void prepare_zero_front(struct ceph_connection *con, int resid)
3240 WARN_ON(!resid || resid > front_len(con->out_msg));
3241 sent = front_len(con->out_msg) - resid;
3242 dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3245 con->v2.out_epil.front_crc =
3246 crc32c(-1, con->out_msg->front.iov_base, sent);
3247 con->v2.out_epil.front_crc =
3248 crc32c_zeros(con->v2.out_epil.front_crc, resid);
3250 con->v2.out_epil.front_crc = crc32c_zeros(-1, resid);
3253 con->v2.out_iter.count -= resid;
3254 out_zero_add(con, resid);
3257 static void prepare_zero_middle(struct ceph_connection *con, int resid)
3261 WARN_ON(!resid || resid > middle_len(con->out_msg));
3262 sent = middle_len(con->out_msg) - resid;
3263 dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3266 con->v2.out_epil.middle_crc =
3267 crc32c(-1, con->out_msg->middle->vec.iov_base, sent);
3268 con->v2.out_epil.middle_crc =
3269 crc32c_zeros(con->v2.out_epil.middle_crc, resid);
3271 con->v2.out_epil.middle_crc = crc32c_zeros(-1, resid);
3274 con->v2.out_iter.count -= resid;
3275 out_zero_add(con, resid);
3278 static void prepare_zero_data(struct ceph_connection *con)
3280 dout("%s con %p\n", __func__, con);
3281 con->v2.out_epil.data_crc = crc32c_zeros(-1, data_len(con->out_msg));
3282 out_zero_add(con, data_len(con->out_msg));
3285 static void revoke_at_queue_data(struct ceph_connection *con)
3290 WARN_ON(!data_len(con->out_msg));
3291 WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
3292 resid = iov_iter_count(&con->v2.out_iter);
3294 boundary = front_len(con->out_msg) + middle_len(con->out_msg);
3295 if (resid > boundary) {
3297 WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN);
3298 dout("%s con %p was sending head\n", __func__, con);
3299 if (front_len(con->out_msg))
3300 prepare_zero_front(con, front_len(con->out_msg));
3301 if (middle_len(con->out_msg))
3302 prepare_zero_middle(con, middle_len(con->out_msg));
3303 prepare_zero_data(con);
3304 WARN_ON(iov_iter_count(&con->v2.out_iter) != resid);
3305 con->v2.out_state = OUT_S_QUEUE_ZEROS;
3309 boundary = middle_len(con->out_msg);
3310 if (resid > boundary) {
3312 dout("%s con %p was sending front\n", __func__, con);
3313 prepare_zero_front(con, resid);
3314 if (middle_len(con->out_msg))
3315 prepare_zero_middle(con, middle_len(con->out_msg));
3316 prepare_zero_data(con);
3322 dout("%s con %p was sending middle\n", __func__, con);
3323 prepare_zero_middle(con, resid);
3324 prepare_zero_data(con);
3328 static void revoke_at_queue_data_cont(struct ceph_connection *con)
3330 int sent, resid; /* current piece of data */
3332 WARN_ON(!data_len(con->out_msg));
3333 WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter));
3334 resid = iov_iter_count(&con->v2.out_iter);
3335 WARN_ON(!resid || resid > con->v2.out_bvec.bv_len);
3336 sent = con->v2.out_bvec.bv_len - resid;
3337 dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3340 con->v2.out_epil.data_crc = ceph_crc32c_page(
3341 con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page,
3342 con->v2.out_bvec.bv_offset, sent);
3343 ceph_msg_data_advance(&con->v2.out_cursor, sent);
3345 WARN_ON(resid > con->v2.out_cursor.total_resid);
3346 con->v2.out_epil.data_crc = crc32c_zeros(con->v2.out_epil.data_crc,
3347 con->v2.out_cursor.total_resid);
3349 con->v2.out_iter.count -= resid;
3350 out_zero_add(con, con->v2.out_cursor.total_resid);
3354 static void revoke_at_finish_message(struct ceph_connection *con)
3359 WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
3360 resid = iov_iter_count(&con->v2.out_iter);
3362 if (!front_len(con->out_msg) && !middle_len(con->out_msg) &&
3363 !data_len(con->out_msg)) {
3364 WARN_ON(!resid || resid > MESSAGE_HEAD_PLAIN_LEN);
3365 dout("%s con %p was sending head (empty message) - noop\n",
3370 boundary = front_len(con->out_msg) + middle_len(con->out_msg) +
3371 CEPH_EPILOGUE_PLAIN_LEN;
3372 if (resid > boundary) {
3374 WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN);
3375 dout("%s con %p was sending head\n", __func__, con);
3376 if (front_len(con->out_msg))
3377 prepare_zero_front(con, front_len(con->out_msg));
3378 if (middle_len(con->out_msg))
3379 prepare_zero_middle(con, middle_len(con->out_msg));
3380 con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3381 WARN_ON(iov_iter_count(&con->v2.out_iter) != resid);
3382 con->v2.out_state = OUT_S_QUEUE_ZEROS;
3386 boundary = middle_len(con->out_msg) + CEPH_EPILOGUE_PLAIN_LEN;
3387 if (resid > boundary) {
3389 dout("%s con %p was sending front\n", __func__, con);
3390 prepare_zero_front(con, resid);
3391 if (middle_len(con->out_msg))
3392 prepare_zero_middle(con, middle_len(con->out_msg));
3393 con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3398 boundary = CEPH_EPILOGUE_PLAIN_LEN;
3399 if (resid > boundary) {
3401 dout("%s con %p was sending middle\n", __func__, con);
3402 prepare_zero_middle(con, resid);
3403 con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3409 dout("%s con %p was sending epilogue - noop\n", __func__, con);
3412 void ceph_con_v2_revoke(struct ceph_connection *con)
3414 WARN_ON(con->v2.out_zero);
3416 if (con_secure(con)) {
3417 WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE &&
3418 con->v2.out_state != OUT_S_FINISH_MESSAGE);
3419 dout("%s con %p secure - noop\n", __func__, con);
3423 switch (con->v2.out_state) {
3424 case OUT_S_QUEUE_DATA:
3425 revoke_at_queue_data(con);
3427 case OUT_S_QUEUE_DATA_CONT:
3428 revoke_at_queue_data_cont(con);
3430 case OUT_S_FINISH_MESSAGE:
3431 revoke_at_finish_message(con);
3434 WARN(1, "bad out_state %d", con->v2.out_state);
3439 static void revoke_at_prepare_read_data(struct ceph_connection *con)
3444 WARN_ON(con_secure(con));
3445 WARN_ON(!data_len(con->in_msg));
3446 WARN_ON(!iov_iter_is_kvec(&con->v2.in_iter));
3447 resid = iov_iter_count(&con->v2.in_iter);
3450 remaining = data_len(con->in_msg) + CEPH_EPILOGUE_PLAIN_LEN;
3451 dout("%s con %p resid %d remaining %d\n", __func__, con, resid,
3453 con->v2.in_iter.count -= resid;
3454 set_in_skip(con, resid + remaining);
3455 con->v2.in_state = IN_S_FINISH_SKIP;
3458 static void revoke_at_prepare_read_data_cont(struct ceph_connection *con)
3460 int recved, resid; /* current piece of data */
3463 WARN_ON(con_secure(con));
3464 WARN_ON(!data_len(con->in_msg));
3465 WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3466 resid = iov_iter_count(&con->v2.in_iter);
3467 WARN_ON(!resid || resid > con->v2.in_bvec.bv_len);
3468 recved = con->v2.in_bvec.bv_len - resid;
3469 dout("%s con %p recved %d resid %d\n", __func__, con, recved, resid);
3472 ceph_msg_data_advance(&con->v2.in_cursor, recved);
3473 WARN_ON(resid > con->v2.in_cursor.total_resid);
3475 remaining = CEPH_EPILOGUE_PLAIN_LEN;
3476 dout("%s con %p total_resid %zu remaining %d\n", __func__, con,
3477 con->v2.in_cursor.total_resid, remaining);
3478 con->v2.in_iter.count -= resid;
3479 set_in_skip(con, con->v2.in_cursor.total_resid + remaining);
3480 con->v2.in_state = IN_S_FINISH_SKIP;
3483 static void revoke_at_prepare_read_enc_page(struct ceph_connection *con)
3485 int resid; /* current enc page (not necessarily data) */
3487 WARN_ON(!con_secure(con));
3488 WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3489 resid = iov_iter_count(&con->v2.in_iter);
3490 WARN_ON(!resid || resid > con->v2.in_bvec.bv_len);
3492 dout("%s con %p resid %d enc_resid %d\n", __func__, con, resid,
3493 con->v2.in_enc_resid);
3494 con->v2.in_iter.count -= resid;
3495 set_in_skip(con, resid + con->v2.in_enc_resid);
3496 con->v2.in_state = IN_S_FINISH_SKIP;
3499 static void revoke_at_handle_epilogue(struct ceph_connection *con)
3503 resid = iov_iter_count(&con->v2.in_iter);
3506 dout("%s con %p resid %d\n", __func__, con, resid);
3507 con->v2.in_iter.count -= resid;
3508 set_in_skip(con, resid);
3509 con->v2.in_state = IN_S_FINISH_SKIP;
3512 void ceph_con_v2_revoke_incoming(struct ceph_connection *con)
3514 switch (con->v2.in_state) {
3515 case IN_S_PREPARE_READ_DATA:
3516 revoke_at_prepare_read_data(con);
3518 case IN_S_PREPARE_READ_DATA_CONT:
3519 revoke_at_prepare_read_data_cont(con);
3521 case IN_S_PREPARE_READ_ENC_PAGE:
3522 revoke_at_prepare_read_enc_page(con);
3524 case IN_S_HANDLE_EPILOGUE:
3525 revoke_at_handle_epilogue(con);
3528 WARN(1, "bad in_state %d", con->v2.in_state);
3533 bool ceph_con_v2_opened(struct ceph_connection *con)
3535 return con->v2.peer_global_seq;
3538 void ceph_con_v2_reset_session(struct ceph_connection *con)
3540 con->v2.client_cookie = 0;
3541 con->v2.server_cookie = 0;
3542 con->v2.global_seq = 0;
3543 con->v2.connect_seq = 0;
3544 con->v2.peer_global_seq = 0;
3547 void ceph_con_v2_reset_protocol(struct ceph_connection *con)
3549 iov_iter_truncate(&con->v2.in_iter, 0);
3550 iov_iter_truncate(&con->v2.out_iter, 0);
3551 con->v2.out_zero = 0;
3553 clear_in_sign_kvecs(con);
3554 clear_out_sign_kvecs(con);
3555 free_conn_bufs(con);
3557 if (con->v2.in_enc_pages) {
3558 WARN_ON(!con->v2.in_enc_page_cnt);
3559 ceph_release_page_vector(con->v2.in_enc_pages,
3560 con->v2.in_enc_page_cnt);
3561 con->v2.in_enc_pages = NULL;
3562 con->v2.in_enc_page_cnt = 0;
3564 if (con->v2.out_enc_pages) {
3565 WARN_ON(!con->v2.out_enc_page_cnt);
3566 ceph_release_page_vector(con->v2.out_enc_pages,
3567 con->v2.out_enc_page_cnt);
3568 con->v2.out_enc_pages = NULL;
3569 con->v2.out_enc_page_cnt = 0;
3572 con->v2.con_mode = CEPH_CON_MODE_UNKNOWN;
3573 memzero_explicit(&con->v2.in_gcm_nonce, CEPH_GCM_IV_LEN);
3574 memzero_explicit(&con->v2.out_gcm_nonce, CEPH_GCM_IV_LEN);
3576 if (con->v2.hmac_tfm) {
3577 crypto_free_shash(con->v2.hmac_tfm);
3578 con->v2.hmac_tfm = NULL;
3580 if (con->v2.gcm_req) {
3581 aead_request_free(con->v2.gcm_req);
3582 con->v2.gcm_req = NULL;
3584 if (con->v2.gcm_tfm) {
3585 crypto_free_aead(con->v2.gcm_tfm);
3586 con->v2.gcm_tfm = NULL;