1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Ceph msgr2 protocol implementation 4 * 5 * Copyright (C) 2020 Ilya Dryomov <idryomov@gmail.com> 6 */ 7 8 #include <linux/ceph/ceph_debug.h> 9 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> 20 #include <net/sock.h> 21 #include <net/tcp.h> 22 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> 27 28 #include "crypto.h" /* for CEPH_KEY_LEN and CEPH_MAX_CON_SECRET_LEN */ 29 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 50 51 #define FRAME_LATE_STATUS_ABORTED 0x1 52 #define FRAME_LATE_STATUS_COMPLETE 0xe 53 #define FRAME_LATE_STATUS_ABORTED_MASK 0xf 54 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 63 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 70 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) 75 76 #define CEPH_MSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 77 78 static int do_recvmsg(struct socket *sock, struct iov_iter *it) 79 { 80 struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS }; 81 int ret; 82 83 msg.msg_iter = *it; 84 while (iov_iter_count(it)) { 85 ret = sock_recvmsg(sock, &msg, msg.msg_flags); 86 if (ret <= 0) { 87 if (ret == -EAGAIN) 88 ret = 0; 89 return ret; 90 } 91 92 iov_iter_advance(it, ret); 93 } 94 95 WARN_ON(msg_data_left(&msg)); 96 return 1; 97 } 98 99 /* 100 * Read as much as possible. 101 * 102 * Return: 103 * 1 - done, nothing (else) to read 104 * 0 - socket is empty, need to wait 105 * <0 - error 106 */ 107 static int ceph_tcp_recv(struct ceph_connection *con) 108 { 109 int ret; 110 111 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)); 117 return ret; 118 } 119 120 static int do_sendmsg(struct socket *sock, struct iov_iter *it) 121 { 122 struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS }; 123 int ret; 124 125 msg.msg_iter = *it; 126 while (iov_iter_count(it)) { 127 ret = sock_sendmsg(sock, &msg); 128 if (ret <= 0) { 129 if (ret == -EAGAIN) 130 ret = 0; 131 return ret; 132 } 133 134 iov_iter_advance(it, ret); 135 } 136 137 WARN_ON(msg_data_left(&msg)); 138 return 1; 139 } 140 141 static int do_try_sendpage(struct socket *sock, struct iov_iter *it) 142 { 143 struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS }; 144 struct bio_vec bv; 145 int ret; 146 147 if (WARN_ON(!iov_iter_is_bvec(it))) 148 return -EINVAL; 149 150 while (iov_iter_count(it)) { 151 /* iov_iter_iovec() for ITER_BVEC */ 152 bvec_set_page(&bv, it->bvec->bv_page, 153 min(iov_iter_count(it), 154 it->bvec->bv_len - it->iov_offset), 155 it->bvec->bv_offset + it->iov_offset); 156 157 /* 158 * MSG_SPLICE_PAGES cannot properly handle pages with 159 * page_count == 0, we need to fall back to sendmsg if 160 * that's the case. 161 * 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. 165 */ 166 if (sendpage_ok(bv.bv_page)) 167 msg.msg_flags |= MSG_SPLICE_PAGES; 168 else 169 msg.msg_flags &= ~MSG_SPLICE_PAGES; 170 171 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bv, 1, bv.bv_len); 172 ret = sock_sendmsg(sock, &msg); 173 if (ret <= 0) { 174 if (ret == -EAGAIN) 175 ret = 0; 176 return ret; 177 } 178 179 iov_iter_advance(it, ret); 180 } 181 182 return 1; 183 } 184 185 /* 186 * Write as much as possible. The socket is expected to be corked, 187 * so we don't bother with MSG_MORE here. 188 * 189 * Return: 190 * 1 - done, nothing (else) to write 191 * 0 - socket is full, need to wait 192 * <0 - error 193 */ 194 static int ceph_tcp_send(struct ceph_connection *con) 195 { 196 int ret; 197 198 dout("%s con %p have %zu try_sendpage %d\n", __func__, con, 199 iov_iter_count(&con->v2.out_iter), con->v2.out_iter_sendpage); 200 if (con->v2.out_iter_sendpage) 201 ret = do_try_sendpage(con->sock, &con->v2.out_iter); 202 else 203 ret = do_sendmsg(con->sock, &con->v2.out_iter); 204 dout("%s con %p ret %d left %zu\n", __func__, con, ret, 205 iov_iter_count(&con->v2.out_iter)); 206 return ret; 207 } 208 209 static void add_in_kvec(struct ceph_connection *con, void *buf, int len) 210 { 211 BUG_ON(con->v2.in_kvec_cnt >= ARRAY_SIZE(con->v2.in_kvecs)); 212 WARN_ON(!iov_iter_is_kvec(&con->v2.in_iter)); 213 214 con->v2.in_kvecs[con->v2.in_kvec_cnt].iov_base = buf; 215 con->v2.in_kvecs[con->v2.in_kvec_cnt].iov_len = len; 216 con->v2.in_kvec_cnt++; 217 218 con->v2.in_iter.nr_segs++; 219 con->v2.in_iter.count += len; 220 } 221 222 static void reset_in_kvecs(struct ceph_connection *con) 223 { 224 WARN_ON(iov_iter_count(&con->v2.in_iter)); 225 226 con->v2.in_kvec_cnt = 0; 227 iov_iter_kvec(&con->v2.in_iter, ITER_DEST, con->v2.in_kvecs, 0, 0); 228 } 229 230 static void set_in_bvec(struct ceph_connection *con, const struct bio_vec *bv) 231 { 232 WARN_ON(iov_iter_count(&con->v2.in_iter)); 233 234 con->v2.in_bvec = *bv; 235 iov_iter_bvec(&con->v2.in_iter, ITER_DEST, &con->v2.in_bvec, 1, bv->bv_len); 236 } 237 238 static void set_in_skip(struct ceph_connection *con, int len) 239 { 240 WARN_ON(iov_iter_count(&con->v2.in_iter)); 241 242 dout("%s con %p len %d\n", __func__, con, len); 243 iov_iter_discard(&con->v2.in_iter, ITER_DEST, len); 244 } 245 246 static void add_out_kvec(struct ceph_connection *con, void *buf, int len) 247 { 248 BUG_ON(con->v2.out_kvec_cnt >= ARRAY_SIZE(con->v2.out_kvecs)); 249 WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter)); 250 WARN_ON(con->v2.out_zero); 251 252 con->v2.out_kvecs[con->v2.out_kvec_cnt].iov_base = buf; 253 con->v2.out_kvecs[con->v2.out_kvec_cnt].iov_len = len; 254 con->v2.out_kvec_cnt++; 255 256 con->v2.out_iter.nr_segs++; 257 con->v2.out_iter.count += len; 258 } 259 260 static void reset_out_kvecs(struct ceph_connection *con) 261 { 262 WARN_ON(iov_iter_count(&con->v2.out_iter)); 263 WARN_ON(con->v2.out_zero); 264 265 con->v2.out_kvec_cnt = 0; 266 267 iov_iter_kvec(&con->v2.out_iter, ITER_SOURCE, con->v2.out_kvecs, 0, 0); 268 con->v2.out_iter_sendpage = false; 269 } 270 271 static void set_out_bvec(struct ceph_connection *con, const struct bio_vec *bv, 272 bool zerocopy) 273 { 274 WARN_ON(iov_iter_count(&con->v2.out_iter)); 275 WARN_ON(con->v2.out_zero); 276 277 con->v2.out_bvec = *bv; 278 con->v2.out_iter_sendpage = zerocopy; 279 iov_iter_bvec(&con->v2.out_iter, ITER_SOURCE, &con->v2.out_bvec, 1, 280 con->v2.out_bvec.bv_len); 281 } 282 283 static void set_out_bvec_zero(struct ceph_connection *con) 284 { 285 WARN_ON(iov_iter_count(&con->v2.out_iter)); 286 WARN_ON(!con->v2.out_zero); 287 288 bvec_set_page(&con->v2.out_bvec, ceph_zero_page, 289 min(con->v2.out_zero, (int)PAGE_SIZE), 0); 290 con->v2.out_iter_sendpage = true; 291 iov_iter_bvec(&con->v2.out_iter, ITER_SOURCE, &con->v2.out_bvec, 1, 292 con->v2.out_bvec.bv_len); 293 } 294 295 static void out_zero_add(struct ceph_connection *con, int len) 296 { 297 dout("%s con %p len %d\n", __func__, con, len); 298 con->v2.out_zero += len; 299 } 300 301 static void *alloc_conn_buf(struct ceph_connection *con, int len) 302 { 303 void *buf; 304 305 dout("%s con %p len %d\n", __func__, con, len); 306 307 if (WARN_ON(con->v2.conn_buf_cnt >= ARRAY_SIZE(con->v2.conn_bufs))) 308 return NULL; 309 310 buf = kvmalloc(len, GFP_NOIO); 311 if (!buf) 312 return NULL; 313 314 con->v2.conn_bufs[con->v2.conn_buf_cnt++] = buf; 315 return buf; 316 } 317 318 static void free_conn_bufs(struct ceph_connection *con) 319 { 320 while (con->v2.conn_buf_cnt) 321 kvfree(con->v2.conn_bufs[--con->v2.conn_buf_cnt]); 322 } 323 324 static void add_in_sign_kvec(struct ceph_connection *con, void *buf, int len) 325 { 326 BUG_ON(con->v2.in_sign_kvec_cnt >= ARRAY_SIZE(con->v2.in_sign_kvecs)); 327 328 con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_base = buf; 329 con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_len = len; 330 con->v2.in_sign_kvec_cnt++; 331 } 332 333 static void clear_in_sign_kvecs(struct ceph_connection *con) 334 { 335 con->v2.in_sign_kvec_cnt = 0; 336 } 337 338 static void add_out_sign_kvec(struct ceph_connection *con, void *buf, int len) 339 { 340 BUG_ON(con->v2.out_sign_kvec_cnt >= ARRAY_SIZE(con->v2.out_sign_kvecs)); 341 342 con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_base = buf; 343 con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_len = len; 344 con->v2.out_sign_kvec_cnt++; 345 } 346 347 static void clear_out_sign_kvecs(struct ceph_connection *con) 348 { 349 con->v2.out_sign_kvec_cnt = 0; 350 } 351 352 static bool con_secure(struct ceph_connection *con) 353 { 354 return con->v2.con_mode == CEPH_CON_MODE_SECURE; 355 } 356 357 static int front_len(const struct ceph_msg *msg) 358 { 359 return le32_to_cpu(msg->hdr.front_len); 360 } 361 362 static int middle_len(const struct ceph_msg *msg) 363 { 364 return le32_to_cpu(msg->hdr.middle_len); 365 } 366 367 static int data_len(const struct ceph_msg *msg) 368 { 369 return le32_to_cpu(msg->hdr.data_len); 370 } 371 372 static bool need_padding(int len) 373 { 374 return !IS_ALIGNED(len, CEPH_GCM_BLOCK_LEN); 375 } 376 377 static int padded_len(int len) 378 { 379 return ALIGN(len, CEPH_GCM_BLOCK_LEN); 380 } 381 382 static int padding_len(int len) 383 { 384 return padded_len(len) - len; 385 } 386 387 /* preamble + control segment */ 388 static int head_onwire_len(int ctrl_len, bool secure) 389 { 390 int head_len; 391 int rem_len; 392 393 if (secure) { 394 head_len = CEPH_PREAMBLE_SECURE_LEN; 395 if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) { 396 rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN; 397 head_len += padded_len(rem_len) + CEPH_GCM_TAG_LEN; 398 } 399 } else { 400 head_len = CEPH_PREAMBLE_PLAIN_LEN; 401 if (ctrl_len) 402 head_len += ctrl_len + CEPH_CRC_LEN; 403 } 404 return head_len; 405 } 406 407 /* front, middle and data segments + epilogue */ 408 static int __tail_onwire_len(int front_len, int middle_len, int data_len, 409 bool secure) 410 { 411 if (!front_len && !middle_len && !data_len) 412 return 0; 413 414 if (!secure) 415 return front_len + middle_len + data_len + 416 CEPH_EPILOGUE_PLAIN_LEN; 417 418 return padded_len(front_len) + padded_len(middle_len) + 419 padded_len(data_len) + CEPH_EPILOGUE_SECURE_LEN; 420 } 421 422 static int tail_onwire_len(const struct ceph_msg *msg, bool secure) 423 { 424 return __tail_onwire_len(front_len(msg), middle_len(msg), 425 data_len(msg), secure); 426 } 427 428 /* head_onwire_len(sizeof(struct ceph_msg_header2), false) */ 429 #define MESSAGE_HEAD_PLAIN_LEN (CEPH_PREAMBLE_PLAIN_LEN + \ 430 sizeof(struct ceph_msg_header2) + \ 431 CEPH_CRC_LEN) 432 433 static const int frame_aligns[] = { 434 sizeof(void *), 435 sizeof(void *), 436 sizeof(void *), 437 PAGE_SIZE 438 }; 439 440 /* 441 * Discards trailing empty segments, unless there is just one segment. 442 * A frame always has at least one (possibly empty) segment. 443 */ 444 static int calc_segment_count(const int *lens, int len_cnt) 445 { 446 int i; 447 448 for (i = len_cnt - 1; i >= 0; i--) { 449 if (lens[i]) 450 return i + 1; 451 } 452 453 return 1; 454 } 455 456 static void init_frame_desc(struct ceph_frame_desc *desc, int tag, 457 const int *lens, int len_cnt) 458 { 459 int i; 460 461 memset(desc, 0, sizeof(*desc)); 462 463 desc->fd_tag = tag; 464 desc->fd_seg_cnt = calc_segment_count(lens, len_cnt); 465 BUG_ON(desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT); 466 for (i = 0; i < desc->fd_seg_cnt; i++) { 467 desc->fd_lens[i] = lens[i]; 468 desc->fd_aligns[i] = frame_aligns[i]; 469 } 470 } 471 472 /* 473 * Preamble crc covers everything up to itself (28 bytes) and 474 * is calculated and verified irrespective of the connection mode 475 * (i.e. even if the frame is encrypted). 476 */ 477 static void encode_preamble(const struct ceph_frame_desc *desc, void *p) 478 { 479 void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN; 480 void *start = p; 481 int i; 482 483 memset(p, 0, CEPH_PREAMBLE_LEN); 484 485 ceph_encode_8(&p, desc->fd_tag); 486 ceph_encode_8(&p, desc->fd_seg_cnt); 487 for (i = 0; i < desc->fd_seg_cnt; i++) { 488 ceph_encode_32(&p, desc->fd_lens[i]); 489 ceph_encode_16(&p, desc->fd_aligns[i]); 490 } 491 492 put_unaligned_le32(crc32c(0, start, crcp - start), crcp); 493 } 494 495 static int decode_preamble(void *p, struct ceph_frame_desc *desc) 496 { 497 void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN; 498 u32 crc, expected_crc; 499 int i; 500 501 crc = crc32c(0, p, crcp - p); 502 expected_crc = get_unaligned_le32(crcp); 503 if (crc != expected_crc) { 504 pr_err("bad preamble crc, calculated %u, expected %u\n", 505 crc, expected_crc); 506 return -EBADMSG; 507 } 508 509 memset(desc, 0, sizeof(*desc)); 510 511 desc->fd_tag = ceph_decode_8(&p); 512 desc->fd_seg_cnt = ceph_decode_8(&p); 513 if (desc->fd_seg_cnt < 1 || 514 desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT) { 515 pr_err("bad segment count %d\n", desc->fd_seg_cnt); 516 return -EINVAL; 517 } 518 for (i = 0; i < desc->fd_seg_cnt; i++) { 519 desc->fd_lens[i] = ceph_decode_32(&p); 520 desc->fd_aligns[i] = ceph_decode_16(&p); 521 } 522 523 /* 524 * This would fire for FRAME_TAG_WAIT (it has one empty 525 * segment), but we should never get it as client. 526 */ 527 if (!desc->fd_lens[desc->fd_seg_cnt - 1]) { 528 pr_err("last segment empty\n"); 529 return -EINVAL; 530 } 531 532 if (desc->fd_lens[0] > CEPH_MSG_MAX_CONTROL_LEN) { 533 pr_err("control segment too big %d\n", desc->fd_lens[0]); 534 return -EINVAL; 535 } 536 if (desc->fd_lens[1] > CEPH_MSG_MAX_FRONT_LEN) { 537 pr_err("front segment too big %d\n", desc->fd_lens[1]); 538 return -EINVAL; 539 } 540 if (desc->fd_lens[2] > CEPH_MSG_MAX_MIDDLE_LEN) { 541 pr_err("middle segment too big %d\n", desc->fd_lens[2]); 542 return -EINVAL; 543 } 544 if (desc->fd_lens[3] > CEPH_MSG_MAX_DATA_LEN) { 545 pr_err("data segment too big %d\n", desc->fd_lens[3]); 546 return -EINVAL; 547 } 548 549 return 0; 550 } 551 552 static void encode_epilogue_plain(struct ceph_connection *con, bool aborted) 553 { 554 con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED : 555 FRAME_LATE_STATUS_COMPLETE; 556 cpu_to_le32s(&con->v2.out_epil.front_crc); 557 cpu_to_le32s(&con->v2.out_epil.middle_crc); 558 cpu_to_le32s(&con->v2.out_epil.data_crc); 559 } 560 561 static void encode_epilogue_secure(struct ceph_connection *con, bool aborted) 562 { 563 memset(&con->v2.out_epil, 0, sizeof(con->v2.out_epil)); 564 con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED : 565 FRAME_LATE_STATUS_COMPLETE; 566 } 567 568 static int decode_epilogue(void *p, u32 *front_crc, u32 *middle_crc, 569 u32 *data_crc) 570 { 571 u8 late_status; 572 573 late_status = ceph_decode_8(&p); 574 if ((late_status & FRAME_LATE_STATUS_ABORTED_MASK) != 575 FRAME_LATE_STATUS_COMPLETE) { 576 /* we should never get an aborted message as client */ 577 pr_err("bad late_status 0x%x\n", late_status); 578 return -EINVAL; 579 } 580 581 if (front_crc && middle_crc && data_crc) { 582 *front_crc = ceph_decode_32(&p); 583 *middle_crc = ceph_decode_32(&p); 584 *data_crc = ceph_decode_32(&p); 585 } 586 587 return 0; 588 } 589 590 static void fill_header(struct ceph_msg_header *hdr, 591 const struct ceph_msg_header2 *hdr2, 592 int front_len, int middle_len, int data_len, 593 const struct ceph_entity_name *peer_name) 594 { 595 hdr->seq = hdr2->seq; 596 hdr->tid = hdr2->tid; 597 hdr->type = hdr2->type; 598 hdr->priority = hdr2->priority; 599 hdr->version = hdr2->version; 600 hdr->front_len = cpu_to_le32(front_len); 601 hdr->middle_len = cpu_to_le32(middle_len); 602 hdr->data_len = cpu_to_le32(data_len); 603 hdr->data_off = hdr2->data_off; 604 hdr->src = *peer_name; 605 hdr->compat_version = hdr2->compat_version; 606 hdr->reserved = 0; 607 hdr->crc = 0; 608 } 609 610 static void fill_header2(struct ceph_msg_header2 *hdr2, 611 const struct ceph_msg_header *hdr, u64 ack_seq) 612 { 613 hdr2->seq = hdr->seq; 614 hdr2->tid = hdr->tid; 615 hdr2->type = hdr->type; 616 hdr2->priority = hdr->priority; 617 hdr2->version = hdr->version; 618 hdr2->data_pre_padding_len = 0; 619 hdr2->data_off = hdr->data_off; 620 hdr2->ack_seq = cpu_to_le64(ack_seq); 621 hdr2->flags = 0; 622 hdr2->compat_version = hdr->compat_version; 623 hdr2->reserved = 0; 624 } 625 626 static int verify_control_crc(struct ceph_connection *con) 627 { 628 int ctrl_len = con->v2.in_desc.fd_lens[0]; 629 u32 crc, expected_crc; 630 631 WARN_ON(con->v2.in_kvecs[0].iov_len != ctrl_len); 632 WARN_ON(con->v2.in_kvecs[1].iov_len != CEPH_CRC_LEN); 633 634 crc = crc32c(-1, con->v2.in_kvecs[0].iov_base, ctrl_len); 635 expected_crc = get_unaligned_le32(con->v2.in_kvecs[1].iov_base); 636 if (crc != expected_crc) { 637 pr_err("bad control crc, calculated %u, expected %u\n", 638 crc, expected_crc); 639 return -EBADMSG; 640 } 641 642 return 0; 643 } 644 645 static int verify_epilogue_crcs(struct ceph_connection *con, u32 front_crc, 646 u32 middle_crc, u32 data_crc) 647 { 648 if (front_len(con->in_msg)) { 649 con->in_front_crc = crc32c(-1, con->in_msg->front.iov_base, 650 front_len(con->in_msg)); 651 } else { 652 WARN_ON(!middle_len(con->in_msg) && !data_len(con->in_msg)); 653 con->in_front_crc = -1; 654 } 655 656 if (middle_len(con->in_msg)) 657 con->in_middle_crc = crc32c(-1, 658 con->in_msg->middle->vec.iov_base, 659 middle_len(con->in_msg)); 660 else if (data_len(con->in_msg)) 661 con->in_middle_crc = -1; 662 else 663 con->in_middle_crc = 0; 664 665 if (!data_len(con->in_msg)) 666 con->in_data_crc = 0; 667 668 dout("%s con %p msg %p crcs %u %u %u\n", __func__, con, con->in_msg, 669 con->in_front_crc, con->in_middle_crc, con->in_data_crc); 670 671 if (con->in_front_crc != front_crc) { 672 pr_err("bad front crc, calculated %u, expected %u\n", 673 con->in_front_crc, front_crc); 674 return -EBADMSG; 675 } 676 if (con->in_middle_crc != middle_crc) { 677 pr_err("bad middle crc, calculated %u, expected %u\n", 678 con->in_middle_crc, middle_crc); 679 return -EBADMSG; 680 } 681 if (con->in_data_crc != data_crc) { 682 pr_err("bad data crc, calculated %u, expected %u\n", 683 con->in_data_crc, data_crc); 684 return -EBADMSG; 685 } 686 687 return 0; 688 } 689 690 static int setup_crypto(struct ceph_connection *con, 691 const u8 *session_key, int session_key_len, 692 const u8 *con_secret, int con_secret_len) 693 { 694 unsigned int noio_flag; 695 int ret; 696 697 dout("%s con %p con_mode %d session_key_len %d con_secret_len %d\n", 698 __func__, con, con->v2.con_mode, session_key_len, con_secret_len); 699 WARN_ON(con->v2.hmac_tfm || con->v2.gcm_tfm || con->v2.gcm_req); 700 701 if (con->v2.con_mode != CEPH_CON_MODE_CRC && 702 con->v2.con_mode != CEPH_CON_MODE_SECURE) { 703 pr_err("bad con_mode %d\n", con->v2.con_mode); 704 return -EINVAL; 705 } 706 707 if (!session_key_len) { 708 WARN_ON(con->v2.con_mode != CEPH_CON_MODE_CRC); 709 WARN_ON(con_secret_len); 710 return 0; /* auth_none */ 711 } 712 713 noio_flag = memalloc_noio_save(); 714 con->v2.hmac_tfm = crypto_alloc_shash("hmac(sha256)", 0, 0); 715 memalloc_noio_restore(noio_flag); 716 if (IS_ERR(con->v2.hmac_tfm)) { 717 ret = PTR_ERR(con->v2.hmac_tfm); 718 con->v2.hmac_tfm = NULL; 719 pr_err("failed to allocate hmac tfm context: %d\n", ret); 720 return ret; 721 } 722 723 WARN_ON((unsigned long)session_key & 724 crypto_shash_alignmask(con->v2.hmac_tfm)); 725 ret = crypto_shash_setkey(con->v2.hmac_tfm, session_key, 726 session_key_len); 727 if (ret) { 728 pr_err("failed to set hmac key: %d\n", ret); 729 return ret; 730 } 731 732 if (con->v2.con_mode == CEPH_CON_MODE_CRC) { 733 WARN_ON(con_secret_len); 734 return 0; /* auth_x, plain mode */ 735 } 736 737 if (con_secret_len < CEPH_GCM_KEY_LEN + 2 * CEPH_GCM_IV_LEN) { 738 pr_err("con_secret too small %d\n", con_secret_len); 739 return -EINVAL; 740 } 741 742 noio_flag = memalloc_noio_save(); 743 con->v2.gcm_tfm = crypto_alloc_aead("gcm(aes)", 0, 0); 744 memalloc_noio_restore(noio_flag); 745 if (IS_ERR(con->v2.gcm_tfm)) { 746 ret = PTR_ERR(con->v2.gcm_tfm); 747 con->v2.gcm_tfm = NULL; 748 pr_err("failed to allocate gcm tfm context: %d\n", ret); 749 return ret; 750 } 751 752 WARN_ON((unsigned long)con_secret & 753 crypto_aead_alignmask(con->v2.gcm_tfm)); 754 ret = crypto_aead_setkey(con->v2.gcm_tfm, con_secret, CEPH_GCM_KEY_LEN); 755 if (ret) { 756 pr_err("failed to set gcm key: %d\n", ret); 757 return ret; 758 } 759 760 WARN_ON(crypto_aead_ivsize(con->v2.gcm_tfm) != CEPH_GCM_IV_LEN); 761 ret = crypto_aead_setauthsize(con->v2.gcm_tfm, CEPH_GCM_TAG_LEN); 762 if (ret) { 763 pr_err("failed to set gcm tag size: %d\n", ret); 764 return ret; 765 } 766 767 con->v2.gcm_req = aead_request_alloc(con->v2.gcm_tfm, GFP_NOIO); 768 if (!con->v2.gcm_req) { 769 pr_err("failed to allocate gcm request\n"); 770 return -ENOMEM; 771 } 772 773 crypto_init_wait(&con->v2.gcm_wait); 774 aead_request_set_callback(con->v2.gcm_req, CRYPTO_TFM_REQ_MAY_BACKLOG, 775 crypto_req_done, &con->v2.gcm_wait); 776 777 memcpy(&con->v2.in_gcm_nonce, con_secret + CEPH_GCM_KEY_LEN, 778 CEPH_GCM_IV_LEN); 779 memcpy(&con->v2.out_gcm_nonce, 780 con_secret + CEPH_GCM_KEY_LEN + CEPH_GCM_IV_LEN, 781 CEPH_GCM_IV_LEN); 782 return 0; /* auth_x, secure mode */ 783 } 784 785 static int hmac_sha256(struct ceph_connection *con, const struct kvec *kvecs, 786 int kvec_cnt, u8 *hmac) 787 { 788 SHASH_DESC_ON_STACK(desc, con->v2.hmac_tfm); /* tfm arg is ignored */ 789 int ret; 790 int i; 791 792 dout("%s con %p hmac_tfm %p kvec_cnt %d\n", __func__, con, 793 con->v2.hmac_tfm, kvec_cnt); 794 795 if (!con->v2.hmac_tfm) { 796 memset(hmac, 0, SHA256_DIGEST_SIZE); 797 return 0; /* auth_none */ 798 } 799 800 desc->tfm = con->v2.hmac_tfm; 801 ret = crypto_shash_init(desc); 802 if (ret) 803 goto out; 804 805 for (i = 0; i < kvec_cnt; i++) { 806 WARN_ON((unsigned long)kvecs[i].iov_base & 807 crypto_shash_alignmask(con->v2.hmac_tfm)); 808 ret = crypto_shash_update(desc, kvecs[i].iov_base, 809 kvecs[i].iov_len); 810 if (ret) 811 goto out; 812 } 813 814 ret = crypto_shash_final(desc, hmac); 815 816 out: 817 shash_desc_zero(desc); 818 return ret; /* auth_x, both plain and secure modes */ 819 } 820 821 static void gcm_inc_nonce(struct ceph_gcm_nonce *nonce) 822 { 823 u64 counter; 824 825 counter = le64_to_cpu(nonce->counter); 826 nonce->counter = cpu_to_le64(counter + 1); 827 } 828 829 static int gcm_crypt(struct ceph_connection *con, bool encrypt, 830 struct scatterlist *src, struct scatterlist *dst, 831 int src_len) 832 { 833 struct ceph_gcm_nonce *nonce; 834 int ret; 835 836 nonce = encrypt ? &con->v2.out_gcm_nonce : &con->v2.in_gcm_nonce; 837 838 aead_request_set_ad(con->v2.gcm_req, 0); /* no AAD */ 839 aead_request_set_crypt(con->v2.gcm_req, src, dst, src_len, (u8 *)nonce); 840 ret = crypto_wait_req(encrypt ? crypto_aead_encrypt(con->v2.gcm_req) : 841 crypto_aead_decrypt(con->v2.gcm_req), 842 &con->v2.gcm_wait); 843 if (ret) 844 return ret; 845 846 gcm_inc_nonce(nonce); 847 return 0; 848 } 849 850 static void get_bvec_at(struct ceph_msg_data_cursor *cursor, 851 struct bio_vec *bv) 852 { 853 struct page *page; 854 size_t off, len; 855 856 WARN_ON(!cursor->total_resid); 857 858 /* skip zero-length data items */ 859 while (!cursor->resid) 860 ceph_msg_data_advance(cursor, 0); 861 862 /* get a piece of data, cursor isn't advanced */ 863 page = ceph_msg_data_next(cursor, &off, &len); 864 bvec_set_page(bv, page, len, off); 865 } 866 867 static int calc_sg_cnt(void *buf, int buf_len) 868 { 869 int sg_cnt; 870 871 if (!buf_len) 872 return 0; 873 874 sg_cnt = need_padding(buf_len) ? 1 : 0; 875 if (is_vmalloc_addr(buf)) { 876 WARN_ON(offset_in_page(buf)); 877 sg_cnt += PAGE_ALIGN(buf_len) >> PAGE_SHIFT; 878 } else { 879 sg_cnt++; 880 } 881 882 return sg_cnt; 883 } 884 885 static int calc_sg_cnt_cursor(struct ceph_msg_data_cursor *cursor) 886 { 887 int data_len = cursor->total_resid; 888 struct bio_vec bv; 889 int sg_cnt; 890 891 if (!data_len) 892 return 0; 893 894 sg_cnt = need_padding(data_len) ? 1 : 0; 895 do { 896 get_bvec_at(cursor, &bv); 897 sg_cnt++; 898 899 ceph_msg_data_advance(cursor, bv.bv_len); 900 } while (cursor->total_resid); 901 902 return sg_cnt; 903 } 904 905 static void init_sgs(struct scatterlist **sg, void *buf, int buf_len, u8 *pad) 906 { 907 void *end = buf + buf_len; 908 struct page *page; 909 int len; 910 void *p; 911 912 if (!buf_len) 913 return; 914 915 if (is_vmalloc_addr(buf)) { 916 p = buf; 917 do { 918 page = vmalloc_to_page(p); 919 len = min_t(int, end - p, PAGE_SIZE); 920 WARN_ON(!page || !len || offset_in_page(p)); 921 sg_set_page(*sg, page, len, 0); 922 *sg = sg_next(*sg); 923 p += len; 924 } while (p != end); 925 } else { 926 sg_set_buf(*sg, buf, buf_len); 927 *sg = sg_next(*sg); 928 } 929 930 if (need_padding(buf_len)) { 931 sg_set_buf(*sg, pad, padding_len(buf_len)); 932 *sg = sg_next(*sg); 933 } 934 } 935 936 static void init_sgs_cursor(struct scatterlist **sg, 937 struct ceph_msg_data_cursor *cursor, u8 *pad) 938 { 939 int data_len = cursor->total_resid; 940 struct bio_vec bv; 941 942 if (!data_len) 943 return; 944 945 do { 946 get_bvec_at(cursor, &bv); 947 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); 948 *sg = sg_next(*sg); 949 950 ceph_msg_data_advance(cursor, bv.bv_len); 951 } while (cursor->total_resid); 952 953 if (need_padding(data_len)) { 954 sg_set_buf(*sg, pad, padding_len(data_len)); 955 *sg = sg_next(*sg); 956 } 957 } 958 959 static int setup_message_sgs(struct sg_table *sgt, struct ceph_msg *msg, 960 u8 *front_pad, u8 *middle_pad, u8 *data_pad, 961 void *epilogue, bool add_tag) 962 { 963 struct ceph_msg_data_cursor cursor; 964 struct scatterlist *cur_sg; 965 int sg_cnt; 966 int ret; 967 968 if (!front_len(msg) && !middle_len(msg) && !data_len(msg)) 969 return 0; 970 971 sg_cnt = 1; /* epilogue + [auth tag] */ 972 if (front_len(msg)) 973 sg_cnt += calc_sg_cnt(msg->front.iov_base, 974 front_len(msg)); 975 if (middle_len(msg)) 976 sg_cnt += calc_sg_cnt(msg->middle->vec.iov_base, 977 middle_len(msg)); 978 if (data_len(msg)) { 979 ceph_msg_data_cursor_init(&cursor, msg, data_len(msg)); 980 sg_cnt += calc_sg_cnt_cursor(&cursor); 981 } 982 983 ret = sg_alloc_table(sgt, sg_cnt, GFP_NOIO); 984 if (ret) 985 return ret; 986 987 cur_sg = sgt->sgl; 988 if (front_len(msg)) 989 init_sgs(&cur_sg, msg->front.iov_base, front_len(msg), 990 front_pad); 991 if (middle_len(msg)) 992 init_sgs(&cur_sg, msg->middle->vec.iov_base, middle_len(msg), 993 middle_pad); 994 if (data_len(msg)) { 995 ceph_msg_data_cursor_init(&cursor, msg, data_len(msg)); 996 init_sgs_cursor(&cur_sg, &cursor, data_pad); 997 } 998 999 WARN_ON(!sg_is_last(cur_sg)); 1000 sg_set_buf(cur_sg, epilogue, 1001 CEPH_GCM_BLOCK_LEN + (add_tag ? CEPH_GCM_TAG_LEN : 0)); 1002 return 0; 1003 } 1004 1005 static int decrypt_preamble(struct ceph_connection *con) 1006 { 1007 struct scatterlist sg; 1008 1009 sg_init_one(&sg, con->v2.in_buf, CEPH_PREAMBLE_SECURE_LEN); 1010 return gcm_crypt(con, false, &sg, &sg, CEPH_PREAMBLE_SECURE_LEN); 1011 } 1012 1013 static int decrypt_control_remainder(struct ceph_connection *con) 1014 { 1015 int ctrl_len = con->v2.in_desc.fd_lens[0]; 1016 int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN; 1017 int pt_len = padding_len(rem_len) + CEPH_GCM_TAG_LEN; 1018 struct scatterlist sgs[2]; 1019 1020 WARN_ON(con->v2.in_kvecs[0].iov_len != rem_len); 1021 WARN_ON(con->v2.in_kvecs[1].iov_len != pt_len); 1022 1023 sg_init_table(sgs, 2); 1024 sg_set_buf(&sgs[0], con->v2.in_kvecs[0].iov_base, rem_len); 1025 sg_set_buf(&sgs[1], con->v2.in_buf, pt_len); 1026 1027 return gcm_crypt(con, false, sgs, sgs, 1028 padded_len(rem_len) + CEPH_GCM_TAG_LEN); 1029 } 1030 1031 static int decrypt_tail(struct ceph_connection *con) 1032 { 1033 struct sg_table enc_sgt = {}; 1034 struct sg_table sgt = {}; 1035 int tail_len; 1036 int ret; 1037 1038 tail_len = tail_onwire_len(con->in_msg, true); 1039 ret = sg_alloc_table_from_pages(&enc_sgt, con->v2.in_enc_pages, 1040 con->v2.in_enc_page_cnt, 0, tail_len, 1041 GFP_NOIO); 1042 if (ret) 1043 goto out; 1044 1045 ret = setup_message_sgs(&sgt, con->in_msg, FRONT_PAD(con->v2.in_buf), 1046 MIDDLE_PAD(con->v2.in_buf), DATA_PAD(con->v2.in_buf), 1047 con->v2.in_buf, true); 1048 if (ret) 1049 goto out; 1050 1051 dout("%s con %p msg %p enc_page_cnt %d sg_cnt %d\n", __func__, con, 1052 con->in_msg, con->v2.in_enc_page_cnt, sgt.orig_nents); 1053 ret = gcm_crypt(con, false, enc_sgt.sgl, sgt.sgl, tail_len); 1054 if (ret) 1055 goto out; 1056 1057 WARN_ON(!con->v2.in_enc_page_cnt); 1058 ceph_release_page_vector(con->v2.in_enc_pages, 1059 con->v2.in_enc_page_cnt); 1060 con->v2.in_enc_pages = NULL; 1061 con->v2.in_enc_page_cnt = 0; 1062 1063 out: 1064 sg_free_table(&sgt); 1065 sg_free_table(&enc_sgt); 1066 return ret; 1067 } 1068 1069 static int prepare_banner(struct ceph_connection *con) 1070 { 1071 int buf_len = CEPH_BANNER_V2_LEN + 2 + 8 + 8; 1072 void *buf, *p; 1073 1074 buf = alloc_conn_buf(con, buf_len); 1075 if (!buf) 1076 return -ENOMEM; 1077 1078 p = buf; 1079 ceph_encode_copy(&p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN); 1080 ceph_encode_16(&p, sizeof(u64) + sizeof(u64)); 1081 ceph_encode_64(&p, CEPH_MSGR2_SUPPORTED_FEATURES); 1082 ceph_encode_64(&p, CEPH_MSGR2_REQUIRED_FEATURES); 1083 WARN_ON(p != buf + buf_len); 1084 1085 add_out_kvec(con, buf, buf_len); 1086 add_out_sign_kvec(con, buf, buf_len); 1087 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING); 1088 return 0; 1089 } 1090 1091 /* 1092 * base: 1093 * preamble 1094 * control body (ctrl_len bytes) 1095 * space for control crc 1096 * 1097 * extdata (optional): 1098 * control body (extdata_len bytes) 1099 * 1100 * Compute control crc and gather base and extdata into: 1101 * 1102 * preamble 1103 * control body (ctrl_len + extdata_len bytes) 1104 * control crc 1105 * 1106 * Preamble should already be encoded at the start of base. 1107 */ 1108 static void prepare_head_plain(struct ceph_connection *con, void *base, 1109 int ctrl_len, void *extdata, int extdata_len, 1110 bool to_be_signed) 1111 { 1112 int base_len = CEPH_PREAMBLE_LEN + ctrl_len + CEPH_CRC_LEN; 1113 void *crcp = base + base_len - CEPH_CRC_LEN; 1114 u32 crc; 1115 1116 crc = crc32c(-1, CTRL_BODY(base), ctrl_len); 1117 if (extdata_len) 1118 crc = crc32c(crc, extdata, extdata_len); 1119 put_unaligned_le32(crc, crcp); 1120 1121 if (!extdata_len) { 1122 add_out_kvec(con, base, base_len); 1123 if (to_be_signed) 1124 add_out_sign_kvec(con, base, base_len); 1125 return; 1126 } 1127 1128 add_out_kvec(con, base, crcp - base); 1129 add_out_kvec(con, extdata, extdata_len); 1130 add_out_kvec(con, crcp, CEPH_CRC_LEN); 1131 if (to_be_signed) { 1132 add_out_sign_kvec(con, base, crcp - base); 1133 add_out_sign_kvec(con, extdata, extdata_len); 1134 add_out_sign_kvec(con, crcp, CEPH_CRC_LEN); 1135 } 1136 } 1137 1138 static int prepare_head_secure_small(struct ceph_connection *con, 1139 void *base, int ctrl_len) 1140 { 1141 struct scatterlist sg; 1142 int ret; 1143 1144 /* inline buffer padding? */ 1145 if (ctrl_len < CEPH_PREAMBLE_INLINE_LEN) 1146 memset(CTRL_BODY(base) + ctrl_len, 0, 1147 CEPH_PREAMBLE_INLINE_LEN - ctrl_len); 1148 1149 sg_init_one(&sg, base, CEPH_PREAMBLE_SECURE_LEN); 1150 ret = gcm_crypt(con, true, &sg, &sg, 1151 CEPH_PREAMBLE_SECURE_LEN - CEPH_GCM_TAG_LEN); 1152 if (ret) 1153 return ret; 1154 1155 add_out_kvec(con, base, CEPH_PREAMBLE_SECURE_LEN); 1156 return 0; 1157 } 1158 1159 /* 1160 * base: 1161 * preamble 1162 * control body (ctrl_len bytes) 1163 * space for padding, if needed 1164 * space for control remainder auth tag 1165 * space for preamble auth tag 1166 * 1167 * Encrypt preamble and the inline portion, then encrypt the remainder 1168 * and gather into: 1169 * 1170 * preamble 1171 * control body (48 bytes) 1172 * preamble auth tag 1173 * control body (ctrl_len - 48 bytes) 1174 * zero padding, if needed 1175 * control remainder auth tag 1176 * 1177 * Preamble should already be encoded at the start of base. 1178 */ 1179 static int prepare_head_secure_big(struct ceph_connection *con, 1180 void *base, int ctrl_len) 1181 { 1182 int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN; 1183 void *rem = CTRL_BODY(base) + CEPH_PREAMBLE_INLINE_LEN; 1184 void *rem_tag = rem + padded_len(rem_len); 1185 void *pmbl_tag = rem_tag + CEPH_GCM_TAG_LEN; 1186 struct scatterlist sgs[2]; 1187 int ret; 1188 1189 sg_init_table(sgs, 2); 1190 sg_set_buf(&sgs[0], base, rem - base); 1191 sg_set_buf(&sgs[1], pmbl_tag, CEPH_GCM_TAG_LEN); 1192 ret = gcm_crypt(con, true, sgs, sgs, rem - base); 1193 if (ret) 1194 return ret; 1195 1196 /* control remainder padding? */ 1197 if (need_padding(rem_len)) 1198 memset(rem + rem_len, 0, padding_len(rem_len)); 1199 1200 sg_init_one(&sgs[0], rem, pmbl_tag - rem); 1201 ret = gcm_crypt(con, true, sgs, sgs, rem_tag - rem); 1202 if (ret) 1203 return ret; 1204 1205 add_out_kvec(con, base, rem - base); 1206 add_out_kvec(con, pmbl_tag, CEPH_GCM_TAG_LEN); 1207 add_out_kvec(con, rem, pmbl_tag - rem); 1208 return 0; 1209 } 1210 1211 static int __prepare_control(struct ceph_connection *con, int tag, 1212 void *base, int ctrl_len, void *extdata, 1213 int extdata_len, bool to_be_signed) 1214 { 1215 int total_len = ctrl_len + extdata_len; 1216 struct ceph_frame_desc desc; 1217 int ret; 1218 1219 dout("%s con %p tag %d len %d (%d+%d)\n", __func__, con, tag, 1220 total_len, ctrl_len, extdata_len); 1221 1222 /* extdata may be vmalloc'ed but not base */ 1223 if (WARN_ON(is_vmalloc_addr(base) || !ctrl_len)) 1224 return -EINVAL; 1225 1226 init_frame_desc(&desc, tag, &total_len, 1); 1227 encode_preamble(&desc, base); 1228 1229 if (con_secure(con)) { 1230 if (WARN_ON(extdata_len || to_be_signed)) 1231 return -EINVAL; 1232 1233 if (ctrl_len <= CEPH_PREAMBLE_INLINE_LEN) 1234 /* fully inlined, inline buffer may need padding */ 1235 ret = prepare_head_secure_small(con, base, ctrl_len); 1236 else 1237 /* partially inlined, inline buffer is full */ 1238 ret = prepare_head_secure_big(con, base, ctrl_len); 1239 if (ret) 1240 return ret; 1241 } else { 1242 prepare_head_plain(con, base, ctrl_len, extdata, extdata_len, 1243 to_be_signed); 1244 } 1245 1246 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING); 1247 return 0; 1248 } 1249 1250 static int prepare_control(struct ceph_connection *con, int tag, 1251 void *base, int ctrl_len) 1252 { 1253 return __prepare_control(con, tag, base, ctrl_len, NULL, 0, false); 1254 } 1255 1256 static int prepare_hello(struct ceph_connection *con) 1257 { 1258 void *buf, *p; 1259 int ctrl_len; 1260 1261 ctrl_len = 1 + ceph_entity_addr_encoding_len(&con->peer_addr); 1262 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false)); 1263 if (!buf) 1264 return -ENOMEM; 1265 1266 p = CTRL_BODY(buf); 1267 ceph_encode_8(&p, CEPH_ENTITY_TYPE_CLIENT); 1268 ceph_encode_entity_addr(&p, &con->peer_addr); 1269 WARN_ON(p != CTRL_BODY(buf) + ctrl_len); 1270 1271 return __prepare_control(con, FRAME_TAG_HELLO, buf, ctrl_len, 1272 NULL, 0, true); 1273 } 1274 1275 /* so that head_onwire_len(AUTH_BUF_LEN, false) is 512 */ 1276 #define AUTH_BUF_LEN (512 - CEPH_CRC_LEN - CEPH_PREAMBLE_PLAIN_LEN) 1277 1278 static int prepare_auth_request(struct ceph_connection *con) 1279 { 1280 void *authorizer, *authorizer_copy; 1281 int ctrl_len, authorizer_len; 1282 void *buf; 1283 int ret; 1284 1285 ctrl_len = AUTH_BUF_LEN; 1286 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false)); 1287 if (!buf) 1288 return -ENOMEM; 1289 1290 mutex_unlock(&con->mutex); 1291 ret = con->ops->get_auth_request(con, CTRL_BODY(buf), &ctrl_len, 1292 &authorizer, &authorizer_len); 1293 mutex_lock(&con->mutex); 1294 if (con->state != CEPH_CON_S_V2_HELLO) { 1295 dout("%s con %p state changed to %d\n", __func__, con, 1296 con->state); 1297 return -EAGAIN; 1298 } 1299 1300 dout("%s con %p get_auth_request ret %d\n", __func__, con, ret); 1301 if (ret) 1302 return ret; 1303 1304 authorizer_copy = alloc_conn_buf(con, authorizer_len); 1305 if (!authorizer_copy) 1306 return -ENOMEM; 1307 1308 memcpy(authorizer_copy, authorizer, authorizer_len); 1309 1310 return __prepare_control(con, FRAME_TAG_AUTH_REQUEST, buf, ctrl_len, 1311 authorizer_copy, authorizer_len, true); 1312 } 1313 1314 static int prepare_auth_request_more(struct ceph_connection *con, 1315 void *reply, int reply_len) 1316 { 1317 int ctrl_len, authorizer_len; 1318 void *authorizer; 1319 void *buf; 1320 int ret; 1321 1322 ctrl_len = AUTH_BUF_LEN; 1323 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false)); 1324 if (!buf) 1325 return -ENOMEM; 1326 1327 mutex_unlock(&con->mutex); 1328 ret = con->ops->handle_auth_reply_more(con, reply, reply_len, 1329 CTRL_BODY(buf), &ctrl_len, 1330 &authorizer, &authorizer_len); 1331 mutex_lock(&con->mutex); 1332 if (con->state != CEPH_CON_S_V2_AUTH) { 1333 dout("%s con %p state changed to %d\n", __func__, con, 1334 con->state); 1335 return -EAGAIN; 1336 } 1337 1338 dout("%s con %p handle_auth_reply_more ret %d\n", __func__, con, ret); 1339 if (ret) 1340 return ret; 1341 1342 return __prepare_control(con, FRAME_TAG_AUTH_REQUEST_MORE, buf, 1343 ctrl_len, authorizer, authorizer_len, true); 1344 } 1345 1346 static int prepare_auth_signature(struct ceph_connection *con) 1347 { 1348 void *buf; 1349 int ret; 1350 1351 buf = alloc_conn_buf(con, head_onwire_len(SHA256_DIGEST_SIZE, 1352 con_secure(con))); 1353 if (!buf) 1354 return -ENOMEM; 1355 1356 ret = hmac_sha256(con, con->v2.in_sign_kvecs, con->v2.in_sign_kvec_cnt, 1357 CTRL_BODY(buf)); 1358 if (ret) 1359 return ret; 1360 1361 return prepare_control(con, FRAME_TAG_AUTH_SIGNATURE, buf, 1362 SHA256_DIGEST_SIZE); 1363 } 1364 1365 static int prepare_client_ident(struct ceph_connection *con) 1366 { 1367 struct ceph_entity_addr *my_addr = &con->msgr->inst.addr; 1368 struct ceph_client *client = from_msgr(con->msgr); 1369 u64 global_id = ceph_client_gid(client); 1370 void *buf, *p; 1371 int ctrl_len; 1372 1373 WARN_ON(con->v2.server_cookie); 1374 WARN_ON(con->v2.connect_seq); 1375 WARN_ON(con->v2.peer_global_seq); 1376 1377 if (!con->v2.client_cookie) { 1378 do { 1379 get_random_bytes(&con->v2.client_cookie, 1380 sizeof(con->v2.client_cookie)); 1381 } while (!con->v2.client_cookie); 1382 dout("%s con %p generated cookie 0x%llx\n", __func__, con, 1383 con->v2.client_cookie); 1384 } else { 1385 dout("%s con %p cookie already set 0x%llx\n", __func__, con, 1386 con->v2.client_cookie); 1387 } 1388 1389 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", 1390 __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce), 1391 ceph_pr_addr(&con->peer_addr), le32_to_cpu(con->peer_addr.nonce), 1392 global_id, con->v2.global_seq, client->supported_features, 1393 client->required_features, con->v2.client_cookie); 1394 1395 ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) + 1396 ceph_entity_addr_encoding_len(&con->peer_addr) + 6 * 8; 1397 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con))); 1398 if (!buf) 1399 return -ENOMEM; 1400 1401 p = CTRL_BODY(buf); 1402 ceph_encode_8(&p, 2); /* addrvec marker */ 1403 ceph_encode_32(&p, 1); /* addr_cnt */ 1404 ceph_encode_entity_addr(&p, my_addr); 1405 ceph_encode_entity_addr(&p, &con->peer_addr); 1406 ceph_encode_64(&p, global_id); 1407 ceph_encode_64(&p, con->v2.global_seq); 1408 ceph_encode_64(&p, client->supported_features); 1409 ceph_encode_64(&p, client->required_features); 1410 ceph_encode_64(&p, 0); /* flags */ 1411 ceph_encode_64(&p, con->v2.client_cookie); 1412 WARN_ON(p != CTRL_BODY(buf) + ctrl_len); 1413 1414 return prepare_control(con, FRAME_TAG_CLIENT_IDENT, buf, ctrl_len); 1415 } 1416 1417 static int prepare_session_reconnect(struct ceph_connection *con) 1418 { 1419 struct ceph_entity_addr *my_addr = &con->msgr->inst.addr; 1420 void *buf, *p; 1421 int ctrl_len; 1422 1423 WARN_ON(!con->v2.client_cookie); 1424 WARN_ON(!con->v2.server_cookie); 1425 WARN_ON(!con->v2.connect_seq); 1426 WARN_ON(!con->v2.peer_global_seq); 1427 1428 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", 1429 __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce), 1430 con->v2.client_cookie, con->v2.server_cookie, con->v2.global_seq, 1431 con->v2.connect_seq, con->in_seq); 1432 1433 ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) + 5 * 8; 1434 buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con))); 1435 if (!buf) 1436 return -ENOMEM; 1437 1438 p = CTRL_BODY(buf); 1439 ceph_encode_8(&p, 2); /* entity_addrvec_t marker */ 1440 ceph_encode_32(&p, 1); /* my_addrs len */ 1441 ceph_encode_entity_addr(&p, my_addr); 1442 ceph_encode_64(&p, con->v2.client_cookie); 1443 ceph_encode_64(&p, con->v2.server_cookie); 1444 ceph_encode_64(&p, con->v2.global_seq); 1445 ceph_encode_64(&p, con->v2.connect_seq); 1446 ceph_encode_64(&p, con->in_seq); 1447 WARN_ON(p != CTRL_BODY(buf) + ctrl_len); 1448 1449 return prepare_control(con, FRAME_TAG_SESSION_RECONNECT, buf, ctrl_len); 1450 } 1451 1452 static int prepare_keepalive2(struct ceph_connection *con) 1453 { 1454 struct ceph_timespec *ts = CTRL_BODY(con->v2.out_buf); 1455 struct timespec64 now; 1456 1457 ktime_get_real_ts64(&now); 1458 dout("%s con %p timestamp %lld.%09ld\n", __func__, con, now.tv_sec, 1459 now.tv_nsec); 1460 1461 ceph_encode_timespec64(ts, &now); 1462 1463 reset_out_kvecs(con); 1464 return prepare_control(con, FRAME_TAG_KEEPALIVE2, con->v2.out_buf, 1465 sizeof(struct ceph_timespec)); 1466 } 1467 1468 static int prepare_ack(struct ceph_connection *con) 1469 { 1470 void *p; 1471 1472 dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con, 1473 con->in_seq_acked, con->in_seq); 1474 con->in_seq_acked = con->in_seq; 1475 1476 p = CTRL_BODY(con->v2.out_buf); 1477 ceph_encode_64(&p, con->in_seq_acked); 1478 1479 reset_out_kvecs(con); 1480 return prepare_control(con, FRAME_TAG_ACK, con->v2.out_buf, 8); 1481 } 1482 1483 static void prepare_epilogue_plain(struct ceph_connection *con, bool aborted) 1484 { 1485 dout("%s con %p msg %p aborted %d crcs %u %u %u\n", __func__, con, 1486 con->out_msg, aborted, con->v2.out_epil.front_crc, 1487 con->v2.out_epil.middle_crc, con->v2.out_epil.data_crc); 1488 1489 encode_epilogue_plain(con, aborted); 1490 add_out_kvec(con, &con->v2.out_epil, CEPH_EPILOGUE_PLAIN_LEN); 1491 } 1492 1493 /* 1494 * For "used" empty segments, crc is -1. For unused (trailing) 1495 * segments, crc is 0. 1496 */ 1497 static void prepare_message_plain(struct ceph_connection *con) 1498 { 1499 struct ceph_msg *msg = con->out_msg; 1500 1501 prepare_head_plain(con, con->v2.out_buf, 1502 sizeof(struct ceph_msg_header2), NULL, 0, false); 1503 1504 if (!front_len(msg) && !middle_len(msg)) { 1505 if (!data_len(msg)) { 1506 /* 1507 * Empty message: once the head is written, 1508 * we are done -- there is no epilogue. 1509 */ 1510 con->v2.out_state = OUT_S_FINISH_MESSAGE; 1511 return; 1512 } 1513 1514 con->v2.out_epil.front_crc = -1; 1515 con->v2.out_epil.middle_crc = -1; 1516 con->v2.out_state = OUT_S_QUEUE_DATA; 1517 return; 1518 } 1519 1520 if (front_len(msg)) { 1521 con->v2.out_epil.front_crc = crc32c(-1, msg->front.iov_base, 1522 front_len(msg)); 1523 add_out_kvec(con, msg->front.iov_base, front_len(msg)); 1524 } else { 1525 /* middle (at least) is there, checked above */ 1526 con->v2.out_epil.front_crc = -1; 1527 } 1528 1529 if (middle_len(msg)) { 1530 con->v2.out_epil.middle_crc = 1531 crc32c(-1, msg->middle->vec.iov_base, middle_len(msg)); 1532 add_out_kvec(con, msg->middle->vec.iov_base, middle_len(msg)); 1533 } else { 1534 con->v2.out_epil.middle_crc = data_len(msg) ? -1 : 0; 1535 } 1536 1537 if (data_len(msg)) { 1538 con->v2.out_state = OUT_S_QUEUE_DATA; 1539 } else { 1540 con->v2.out_epil.data_crc = 0; 1541 prepare_epilogue_plain(con, false); 1542 con->v2.out_state = OUT_S_FINISH_MESSAGE; 1543 } 1544 } 1545 1546 /* 1547 * Unfortunately the kernel crypto API doesn't support streaming 1548 * (piecewise) operation for AEAD algorithms, so we can't get away 1549 * with a fixed size buffer and a couple sgs. Instead, we have to 1550 * allocate pages for the entire tail of the message (currently up 1551 * to ~32M) and two sgs arrays (up to ~256K each)... 1552 */ 1553 static int prepare_message_secure(struct ceph_connection *con) 1554 { 1555 void *zerop = page_address(ceph_zero_page); 1556 struct sg_table enc_sgt = {}; 1557 struct sg_table sgt = {}; 1558 struct page **enc_pages; 1559 int enc_page_cnt; 1560 int tail_len; 1561 int ret; 1562 1563 ret = prepare_head_secure_small(con, con->v2.out_buf, 1564 sizeof(struct ceph_msg_header2)); 1565 if (ret) 1566 return ret; 1567 1568 tail_len = tail_onwire_len(con->out_msg, true); 1569 if (!tail_len) { 1570 /* 1571 * Empty message: once the head is written, 1572 * we are done -- there is no epilogue. 1573 */ 1574 con->v2.out_state = OUT_S_FINISH_MESSAGE; 1575 return 0; 1576 } 1577 1578 encode_epilogue_secure(con, false); 1579 ret = setup_message_sgs(&sgt, con->out_msg, zerop, zerop, zerop, 1580 &con->v2.out_epil, false); 1581 if (ret) 1582 goto out; 1583 1584 enc_page_cnt = calc_pages_for(0, tail_len); 1585 enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO); 1586 if (IS_ERR(enc_pages)) { 1587 ret = PTR_ERR(enc_pages); 1588 goto out; 1589 } 1590 1591 WARN_ON(con->v2.out_enc_pages || con->v2.out_enc_page_cnt); 1592 con->v2.out_enc_pages = enc_pages; 1593 con->v2.out_enc_page_cnt = enc_page_cnt; 1594 con->v2.out_enc_resid = tail_len; 1595 con->v2.out_enc_i = 0; 1596 1597 ret = sg_alloc_table_from_pages(&enc_sgt, enc_pages, enc_page_cnt, 1598 0, tail_len, GFP_NOIO); 1599 if (ret) 1600 goto out; 1601 1602 ret = gcm_crypt(con, true, sgt.sgl, enc_sgt.sgl, 1603 tail_len - CEPH_GCM_TAG_LEN); 1604 if (ret) 1605 goto out; 1606 1607 dout("%s con %p msg %p sg_cnt %d enc_page_cnt %d\n", __func__, con, 1608 con->out_msg, sgt.orig_nents, enc_page_cnt); 1609 con->v2.out_state = OUT_S_QUEUE_ENC_PAGE; 1610 1611 out: 1612 sg_free_table(&sgt); 1613 sg_free_table(&enc_sgt); 1614 return ret; 1615 } 1616 1617 static int prepare_message(struct ceph_connection *con) 1618 { 1619 int lens[] = { 1620 sizeof(struct ceph_msg_header2), 1621 front_len(con->out_msg), 1622 middle_len(con->out_msg), 1623 data_len(con->out_msg) 1624 }; 1625 struct ceph_frame_desc desc; 1626 int ret; 1627 1628 dout("%s con %p msg %p logical %d+%d+%d+%d\n", __func__, con, 1629 con->out_msg, lens[0], lens[1], lens[2], lens[3]); 1630 1631 if (con->in_seq > con->in_seq_acked) { 1632 dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con, 1633 con->in_seq_acked, con->in_seq); 1634 con->in_seq_acked = con->in_seq; 1635 } 1636 1637 reset_out_kvecs(con); 1638 init_frame_desc(&desc, FRAME_TAG_MESSAGE, lens, 4); 1639 encode_preamble(&desc, con->v2.out_buf); 1640 fill_header2(CTRL_BODY(con->v2.out_buf), &con->out_msg->hdr, 1641 con->in_seq_acked); 1642 1643 if (con_secure(con)) { 1644 ret = prepare_message_secure(con); 1645 if (ret) 1646 return ret; 1647 } else { 1648 prepare_message_plain(con); 1649 } 1650 1651 ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING); 1652 return 0; 1653 } 1654 1655 static int prepare_read_banner_prefix(struct ceph_connection *con) 1656 { 1657 void *buf; 1658 1659 buf = alloc_conn_buf(con, CEPH_BANNER_V2_PREFIX_LEN); 1660 if (!buf) 1661 return -ENOMEM; 1662 1663 reset_in_kvecs(con); 1664 add_in_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN); 1665 add_in_sign_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN); 1666 con->state = CEPH_CON_S_V2_BANNER_PREFIX; 1667 return 0; 1668 } 1669 1670 static int prepare_read_banner_payload(struct ceph_connection *con, 1671 int payload_len) 1672 { 1673 void *buf; 1674 1675 buf = alloc_conn_buf(con, payload_len); 1676 if (!buf) 1677 return -ENOMEM; 1678 1679 reset_in_kvecs(con); 1680 add_in_kvec(con, buf, payload_len); 1681 add_in_sign_kvec(con, buf, payload_len); 1682 con->state = CEPH_CON_S_V2_BANNER_PAYLOAD; 1683 return 0; 1684 } 1685 1686 static void prepare_read_preamble(struct ceph_connection *con) 1687 { 1688 reset_in_kvecs(con); 1689 add_in_kvec(con, con->v2.in_buf, 1690 con_secure(con) ? CEPH_PREAMBLE_SECURE_LEN : 1691 CEPH_PREAMBLE_PLAIN_LEN); 1692 con->v2.in_state = IN_S_HANDLE_PREAMBLE; 1693 } 1694 1695 static int prepare_read_control(struct ceph_connection *con) 1696 { 1697 int ctrl_len = con->v2.in_desc.fd_lens[0]; 1698 int head_len; 1699 void *buf; 1700 1701 reset_in_kvecs(con); 1702 if (con->state == CEPH_CON_S_V2_HELLO || 1703 con->state == CEPH_CON_S_V2_AUTH) { 1704 head_len = head_onwire_len(ctrl_len, false); 1705 buf = alloc_conn_buf(con, head_len); 1706 if (!buf) 1707 return -ENOMEM; 1708 1709 /* preserve preamble */ 1710 memcpy(buf, con->v2.in_buf, CEPH_PREAMBLE_LEN); 1711 1712 add_in_kvec(con, CTRL_BODY(buf), ctrl_len); 1713 add_in_kvec(con, CTRL_BODY(buf) + ctrl_len, CEPH_CRC_LEN); 1714 add_in_sign_kvec(con, buf, head_len); 1715 } else { 1716 if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) { 1717 buf = alloc_conn_buf(con, ctrl_len); 1718 if (!buf) 1719 return -ENOMEM; 1720 1721 add_in_kvec(con, buf, ctrl_len); 1722 } else { 1723 add_in_kvec(con, CTRL_BODY(con->v2.in_buf), ctrl_len); 1724 } 1725 add_in_kvec(con, con->v2.in_buf, CEPH_CRC_LEN); 1726 } 1727 con->v2.in_state = IN_S_HANDLE_CONTROL; 1728 return 0; 1729 } 1730 1731 static int prepare_read_control_remainder(struct ceph_connection *con) 1732 { 1733 int ctrl_len = con->v2.in_desc.fd_lens[0]; 1734 int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN; 1735 void *buf; 1736 1737 buf = alloc_conn_buf(con, ctrl_len); 1738 if (!buf) 1739 return -ENOMEM; 1740 1741 memcpy(buf, CTRL_BODY(con->v2.in_buf), CEPH_PREAMBLE_INLINE_LEN); 1742 1743 reset_in_kvecs(con); 1744 add_in_kvec(con, buf + CEPH_PREAMBLE_INLINE_LEN, rem_len); 1745 add_in_kvec(con, con->v2.in_buf, 1746 padding_len(rem_len) + CEPH_GCM_TAG_LEN); 1747 con->v2.in_state = IN_S_HANDLE_CONTROL_REMAINDER; 1748 return 0; 1749 } 1750 1751 static int prepare_read_data(struct ceph_connection *con) 1752 { 1753 struct bio_vec bv; 1754 1755 con->in_data_crc = -1; 1756 ceph_msg_data_cursor_init(&con->v2.in_cursor, con->in_msg, 1757 data_len(con->in_msg)); 1758 1759 get_bvec_at(&con->v2.in_cursor, &bv); 1760 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) { 1761 if (unlikely(!con->bounce_page)) { 1762 con->bounce_page = alloc_page(GFP_NOIO); 1763 if (!con->bounce_page) { 1764 pr_err("failed to allocate bounce page\n"); 1765 return -ENOMEM; 1766 } 1767 } 1768 1769 bv.bv_page = con->bounce_page; 1770 bv.bv_offset = 0; 1771 } 1772 set_in_bvec(con, &bv); 1773 con->v2.in_state = IN_S_PREPARE_READ_DATA_CONT; 1774 return 0; 1775 } 1776 1777 static void prepare_read_data_cont(struct ceph_connection *con) 1778 { 1779 struct bio_vec bv; 1780 1781 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) { 1782 con->in_data_crc = crc32c(con->in_data_crc, 1783 page_address(con->bounce_page), 1784 con->v2.in_bvec.bv_len); 1785 1786 get_bvec_at(&con->v2.in_cursor, &bv); 1787 memcpy_to_page(bv.bv_page, bv.bv_offset, 1788 page_address(con->bounce_page), 1789 con->v2.in_bvec.bv_len); 1790 } else { 1791 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, 1792 con->v2.in_bvec.bv_page, 1793 con->v2.in_bvec.bv_offset, 1794 con->v2.in_bvec.bv_len); 1795 } 1796 1797 ceph_msg_data_advance(&con->v2.in_cursor, con->v2.in_bvec.bv_len); 1798 if (con->v2.in_cursor.total_resid) { 1799 get_bvec_at(&con->v2.in_cursor, &bv); 1800 if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) { 1801 bv.bv_page = con->bounce_page; 1802 bv.bv_offset = 0; 1803 } 1804 set_in_bvec(con, &bv); 1805 WARN_ON(con->v2.in_state != IN_S_PREPARE_READ_DATA_CONT); 1806 return; 1807 } 1808 1809 /* 1810 * We've read all data. Prepare to read epilogue. 1811 */ 1812 reset_in_kvecs(con); 1813 add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN); 1814 con->v2.in_state = IN_S_HANDLE_EPILOGUE; 1815 } 1816 1817 static int prepare_read_tail_plain(struct ceph_connection *con) 1818 { 1819 struct ceph_msg *msg = con->in_msg; 1820 1821 if (!front_len(msg) && !middle_len(msg)) { 1822 WARN_ON(!data_len(msg)); 1823 return prepare_read_data(con); 1824 } 1825 1826 reset_in_kvecs(con); 1827 if (front_len(msg)) { 1828 add_in_kvec(con, msg->front.iov_base, front_len(msg)); 1829 WARN_ON(msg->front.iov_len != front_len(msg)); 1830 } 1831 if (middle_len(msg)) { 1832 add_in_kvec(con, msg->middle->vec.iov_base, middle_len(msg)); 1833 WARN_ON(msg->middle->vec.iov_len != middle_len(msg)); 1834 } 1835 1836 if (data_len(msg)) { 1837 con->v2.in_state = IN_S_PREPARE_READ_DATA; 1838 } else { 1839 add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN); 1840 con->v2.in_state = IN_S_HANDLE_EPILOGUE; 1841 } 1842 return 0; 1843 } 1844 1845 static void prepare_read_enc_page(struct ceph_connection *con) 1846 { 1847 struct bio_vec bv; 1848 1849 dout("%s con %p i %d resid %d\n", __func__, con, con->v2.in_enc_i, 1850 con->v2.in_enc_resid); 1851 WARN_ON(!con->v2.in_enc_resid); 1852 1853 bvec_set_page(&bv, con->v2.in_enc_pages[con->v2.in_enc_i], 1854 min(con->v2.in_enc_resid, (int)PAGE_SIZE), 0); 1855 1856 set_in_bvec(con, &bv); 1857 con->v2.in_enc_i++; 1858 con->v2.in_enc_resid -= bv.bv_len; 1859 1860 if (con->v2.in_enc_resid) { 1861 con->v2.in_state = IN_S_PREPARE_READ_ENC_PAGE; 1862 return; 1863 } 1864 1865 /* 1866 * We are set to read the last piece of ciphertext (ending 1867 * with epilogue) + auth tag. 1868 */ 1869 WARN_ON(con->v2.in_enc_i != con->v2.in_enc_page_cnt); 1870 con->v2.in_state = IN_S_HANDLE_EPILOGUE; 1871 } 1872 1873 static int prepare_read_tail_secure(struct ceph_connection *con) 1874 { 1875 struct page **enc_pages; 1876 int enc_page_cnt; 1877 int tail_len; 1878 1879 tail_len = tail_onwire_len(con->in_msg, true); 1880 WARN_ON(!tail_len); 1881 1882 enc_page_cnt = calc_pages_for(0, tail_len); 1883 enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO); 1884 if (IS_ERR(enc_pages)) 1885 return PTR_ERR(enc_pages); 1886 1887 WARN_ON(con->v2.in_enc_pages || con->v2.in_enc_page_cnt); 1888 con->v2.in_enc_pages = enc_pages; 1889 con->v2.in_enc_page_cnt = enc_page_cnt; 1890 con->v2.in_enc_resid = tail_len; 1891 con->v2.in_enc_i = 0; 1892 1893 prepare_read_enc_page(con); 1894 return 0; 1895 } 1896 1897 static void __finish_skip(struct ceph_connection *con) 1898 { 1899 con->in_seq++; 1900 prepare_read_preamble(con); 1901 } 1902 1903 static void prepare_skip_message(struct ceph_connection *con) 1904 { 1905 struct ceph_frame_desc *desc = &con->v2.in_desc; 1906 int tail_len; 1907 1908 dout("%s con %p %d+%d+%d\n", __func__, con, desc->fd_lens[1], 1909 desc->fd_lens[2], desc->fd_lens[3]); 1910 1911 tail_len = __tail_onwire_len(desc->fd_lens[1], desc->fd_lens[2], 1912 desc->fd_lens[3], con_secure(con)); 1913 if (!tail_len) { 1914 __finish_skip(con); 1915 } else { 1916 set_in_skip(con, tail_len); 1917 con->v2.in_state = IN_S_FINISH_SKIP; 1918 } 1919 } 1920 1921 static int process_banner_prefix(struct ceph_connection *con) 1922 { 1923 int payload_len; 1924 void *p; 1925 1926 WARN_ON(con->v2.in_kvecs[0].iov_len != CEPH_BANNER_V2_PREFIX_LEN); 1927 1928 p = con->v2.in_kvecs[0].iov_base; 1929 if (memcmp(p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN)) { 1930 if (!memcmp(p, CEPH_BANNER, CEPH_BANNER_LEN)) 1931 con->error_msg = "server is speaking msgr1 protocol"; 1932 else 1933 con->error_msg = "protocol error, bad banner"; 1934 return -EINVAL; 1935 } 1936 1937 p += CEPH_BANNER_V2_LEN; 1938 payload_len = ceph_decode_16(&p); 1939 dout("%s con %p payload_len %d\n", __func__, con, payload_len); 1940 1941 return prepare_read_banner_payload(con, payload_len); 1942 } 1943 1944 static int process_banner_payload(struct ceph_connection *con) 1945 { 1946 void *end = con->v2.in_kvecs[0].iov_base + con->v2.in_kvecs[0].iov_len; 1947 u64 feat = CEPH_MSGR2_SUPPORTED_FEATURES; 1948 u64 req_feat = CEPH_MSGR2_REQUIRED_FEATURES; 1949 u64 server_feat, server_req_feat; 1950 void *p; 1951 int ret; 1952 1953 p = con->v2.in_kvecs[0].iov_base; 1954 ceph_decode_64_safe(&p, end, server_feat, bad); 1955 ceph_decode_64_safe(&p, end, server_req_feat, bad); 1956 1957 dout("%s con %p server_feat 0x%llx server_req_feat 0x%llx\n", 1958 __func__, con, server_feat, server_req_feat); 1959 1960 if (req_feat & ~server_feat) { 1961 pr_err("msgr2 feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n", 1962 server_feat, req_feat & ~server_feat); 1963 con->error_msg = "missing required protocol features"; 1964 return -EINVAL; 1965 } 1966 if (server_req_feat & ~feat) { 1967 pr_err("msgr2 feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n", 1968 feat, server_req_feat & ~feat); 1969 con->error_msg = "missing required protocol features"; 1970 return -EINVAL; 1971 } 1972 1973 /* no reset_out_kvecs() as our banner may still be pending */ 1974 ret = prepare_hello(con); 1975 if (ret) { 1976 pr_err("prepare_hello failed: %d\n", ret); 1977 return ret; 1978 } 1979 1980 con->state = CEPH_CON_S_V2_HELLO; 1981 prepare_read_preamble(con); 1982 return 0; 1983 1984 bad: 1985 pr_err("failed to decode banner payload\n"); 1986 return -EINVAL; 1987 } 1988 1989 static int process_hello(struct ceph_connection *con, void *p, void *end) 1990 { 1991 struct ceph_entity_addr *my_addr = &con->msgr->inst.addr; 1992 struct ceph_entity_addr addr_for_me; 1993 u8 entity_type; 1994 int ret; 1995 1996 if (con->state != CEPH_CON_S_V2_HELLO) { 1997 con->error_msg = "protocol error, unexpected hello"; 1998 return -EINVAL; 1999 } 2000 2001 ceph_decode_8_safe(&p, end, entity_type, bad); 2002 ret = ceph_decode_entity_addr(&p, end, &addr_for_me); 2003 if (ret) { 2004 pr_err("failed to decode addr_for_me: %d\n", ret); 2005 return ret; 2006 } 2007 2008 dout("%s con %p entity_type %d addr_for_me %s\n", __func__, con, 2009 entity_type, ceph_pr_addr(&addr_for_me)); 2010 2011 if (entity_type != con->peer_name.type) { 2012 pr_err("bad peer type, want %d, got %d\n", 2013 con->peer_name.type, entity_type); 2014 con->error_msg = "wrong peer at address"; 2015 return -EINVAL; 2016 } 2017 2018 /* 2019 * Set our address to the address our first peer (i.e. monitor) 2020 * sees that we are connecting from. If we are behind some sort 2021 * of NAT and want to be identified by some private (not NATed) 2022 * address, ip option should be used. 2023 */ 2024 if (ceph_addr_is_blank(my_addr)) { 2025 memcpy(&my_addr->in_addr, &addr_for_me.in_addr, 2026 sizeof(my_addr->in_addr)); 2027 ceph_addr_set_port(my_addr, 0); 2028 dout("%s con %p set my addr %s, as seen by peer %s\n", 2029 __func__, con, ceph_pr_addr(my_addr), 2030 ceph_pr_addr(&con->peer_addr)); 2031 } else { 2032 dout("%s con %p my addr already set %s\n", 2033 __func__, con, ceph_pr_addr(my_addr)); 2034 } 2035 2036 WARN_ON(ceph_addr_is_blank(my_addr) || ceph_addr_port(my_addr)); 2037 WARN_ON(my_addr->type != CEPH_ENTITY_ADDR_TYPE_ANY); 2038 WARN_ON(!my_addr->nonce); 2039 2040 /* no reset_out_kvecs() as our hello may still be pending */ 2041 ret = prepare_auth_request(con); 2042 if (ret) { 2043 if (ret != -EAGAIN) 2044 pr_err("prepare_auth_request failed: %d\n", ret); 2045 return ret; 2046 } 2047 2048 con->state = CEPH_CON_S_V2_AUTH; 2049 return 0; 2050 2051 bad: 2052 pr_err("failed to decode hello\n"); 2053 return -EINVAL; 2054 } 2055 2056 static int process_auth_bad_method(struct ceph_connection *con, 2057 void *p, void *end) 2058 { 2059 int allowed_protos[8], allowed_modes[8]; 2060 int allowed_proto_cnt, allowed_mode_cnt; 2061 int used_proto, result; 2062 int ret; 2063 int i; 2064 2065 if (con->state != CEPH_CON_S_V2_AUTH) { 2066 con->error_msg = "protocol error, unexpected auth_bad_method"; 2067 return -EINVAL; 2068 } 2069 2070 ceph_decode_32_safe(&p, end, used_proto, bad); 2071 ceph_decode_32_safe(&p, end, result, bad); 2072 dout("%s con %p used_proto %d result %d\n", __func__, con, used_proto, 2073 result); 2074 2075 ceph_decode_32_safe(&p, end, allowed_proto_cnt, bad); 2076 if (allowed_proto_cnt > ARRAY_SIZE(allowed_protos)) { 2077 pr_err("allowed_protos too big %d\n", allowed_proto_cnt); 2078 return -EINVAL; 2079 } 2080 for (i = 0; i < allowed_proto_cnt; i++) { 2081 ceph_decode_32_safe(&p, end, allowed_protos[i], bad); 2082 dout("%s con %p allowed_protos[%d] %d\n", __func__, con, 2083 i, allowed_protos[i]); 2084 } 2085 2086 ceph_decode_32_safe(&p, end, allowed_mode_cnt, bad); 2087 if (allowed_mode_cnt > ARRAY_SIZE(allowed_modes)) { 2088 pr_err("allowed_modes too big %d\n", allowed_mode_cnt); 2089 return -EINVAL; 2090 } 2091 for (i = 0; i < allowed_mode_cnt; i++) { 2092 ceph_decode_32_safe(&p, end, allowed_modes[i], bad); 2093 dout("%s con %p allowed_modes[%d] %d\n", __func__, con, 2094 i, allowed_modes[i]); 2095 } 2096 2097 mutex_unlock(&con->mutex); 2098 ret = con->ops->handle_auth_bad_method(con, used_proto, result, 2099 allowed_protos, 2100 allowed_proto_cnt, 2101 allowed_modes, 2102 allowed_mode_cnt); 2103 mutex_lock(&con->mutex); 2104 if (con->state != CEPH_CON_S_V2_AUTH) { 2105 dout("%s con %p state changed to %d\n", __func__, con, 2106 con->state); 2107 return -EAGAIN; 2108 } 2109 2110 dout("%s con %p handle_auth_bad_method ret %d\n", __func__, con, ret); 2111 return ret; 2112 2113 bad: 2114 pr_err("failed to decode auth_bad_method\n"); 2115 return -EINVAL; 2116 } 2117 2118 static int process_auth_reply_more(struct ceph_connection *con, 2119 void *p, void *end) 2120 { 2121 int payload_len; 2122 int ret; 2123 2124 if (con->state != CEPH_CON_S_V2_AUTH) { 2125 con->error_msg = "protocol error, unexpected auth_reply_more"; 2126 return -EINVAL; 2127 } 2128 2129 ceph_decode_32_safe(&p, end, payload_len, bad); 2130 ceph_decode_need(&p, end, payload_len, bad); 2131 2132 dout("%s con %p payload_len %d\n", __func__, con, payload_len); 2133 2134 reset_out_kvecs(con); 2135 ret = prepare_auth_request_more(con, p, payload_len); 2136 if (ret) { 2137 if (ret != -EAGAIN) 2138 pr_err("prepare_auth_request_more failed: %d\n", ret); 2139 return ret; 2140 } 2141 2142 return 0; 2143 2144 bad: 2145 pr_err("failed to decode auth_reply_more\n"); 2146 return -EINVAL; 2147 } 2148 2149 /* 2150 * Align session_key and con_secret to avoid GFP_ATOMIC allocation 2151 * inside crypto_shash_setkey() and crypto_aead_setkey() called from 2152 * setup_crypto(). __aligned(16) isn't guaranteed to work for stack 2153 * objects, so do it by hand. 2154 */ 2155 static int process_auth_done(struct ceph_connection *con, void *p, void *end) 2156 { 2157 u8 session_key_buf[CEPH_KEY_LEN + 16]; 2158 u8 con_secret_buf[CEPH_MAX_CON_SECRET_LEN + 16]; 2159 u8 *session_key = PTR_ALIGN(&session_key_buf[0], 16); 2160 u8 *con_secret = PTR_ALIGN(&con_secret_buf[0], 16); 2161 int session_key_len, con_secret_len; 2162 int payload_len; 2163 u64 global_id; 2164 int ret; 2165 2166 if (con->state != CEPH_CON_S_V2_AUTH) { 2167 con->error_msg = "protocol error, unexpected auth_done"; 2168 return -EINVAL; 2169 } 2170 2171 ceph_decode_64_safe(&p, end, global_id, bad); 2172 ceph_decode_32_safe(&p, end, con->v2.con_mode, bad); 2173 ceph_decode_32_safe(&p, end, payload_len, bad); 2174 2175 dout("%s con %p global_id %llu con_mode %d payload_len %d\n", 2176 __func__, con, global_id, con->v2.con_mode, payload_len); 2177 2178 mutex_unlock(&con->mutex); 2179 session_key_len = 0; 2180 con_secret_len = 0; 2181 ret = con->ops->handle_auth_done(con, global_id, p, payload_len, 2182 session_key, &session_key_len, 2183 con_secret, &con_secret_len); 2184 mutex_lock(&con->mutex); 2185 if (con->state != CEPH_CON_S_V2_AUTH) { 2186 dout("%s con %p state changed to %d\n", __func__, con, 2187 con->state); 2188 ret = -EAGAIN; 2189 goto out; 2190 } 2191 2192 dout("%s con %p handle_auth_done ret %d\n", __func__, con, ret); 2193 if (ret) 2194 goto out; 2195 2196 ret = setup_crypto(con, session_key, session_key_len, con_secret, 2197 con_secret_len); 2198 if (ret) 2199 goto out; 2200 2201 reset_out_kvecs(con); 2202 ret = prepare_auth_signature(con); 2203 if (ret) { 2204 pr_err("prepare_auth_signature failed: %d\n", ret); 2205 goto out; 2206 } 2207 2208 con->state = CEPH_CON_S_V2_AUTH_SIGNATURE; 2209 2210 out: 2211 memzero_explicit(session_key_buf, sizeof(session_key_buf)); 2212 memzero_explicit(con_secret_buf, sizeof(con_secret_buf)); 2213 return ret; 2214 2215 bad: 2216 pr_err("failed to decode auth_done\n"); 2217 return -EINVAL; 2218 } 2219 2220 static int process_auth_signature(struct ceph_connection *con, 2221 void *p, void *end) 2222 { 2223 u8 hmac[SHA256_DIGEST_SIZE]; 2224 int ret; 2225 2226 if (con->state != CEPH_CON_S_V2_AUTH_SIGNATURE) { 2227 con->error_msg = "protocol error, unexpected auth_signature"; 2228 return -EINVAL; 2229 } 2230 2231 ret = hmac_sha256(con, con->v2.out_sign_kvecs, 2232 con->v2.out_sign_kvec_cnt, hmac); 2233 if (ret) 2234 return ret; 2235 2236 ceph_decode_need(&p, end, SHA256_DIGEST_SIZE, bad); 2237 if (crypto_memneq(p, hmac, SHA256_DIGEST_SIZE)) { 2238 con->error_msg = "integrity error, bad auth signature"; 2239 return -EBADMSG; 2240 } 2241 2242 dout("%s con %p auth signature ok\n", __func__, con); 2243 2244 /* no reset_out_kvecs() as our auth_signature may still be pending */ 2245 if (!con->v2.server_cookie) { 2246 ret = prepare_client_ident(con); 2247 if (ret) { 2248 pr_err("prepare_client_ident failed: %d\n", ret); 2249 return ret; 2250 } 2251 2252 con->state = CEPH_CON_S_V2_SESSION_CONNECT; 2253 } else { 2254 ret = prepare_session_reconnect(con); 2255 if (ret) { 2256 pr_err("prepare_session_reconnect failed: %d\n", ret); 2257 return ret; 2258 } 2259 2260 con->state = CEPH_CON_S_V2_SESSION_RECONNECT; 2261 } 2262 2263 return 0; 2264 2265 bad: 2266 pr_err("failed to decode auth_signature\n"); 2267 return -EINVAL; 2268 } 2269 2270 static int process_server_ident(struct ceph_connection *con, 2271 void *p, void *end) 2272 { 2273 struct ceph_client *client = from_msgr(con->msgr); 2274 u64 features, required_features; 2275 struct ceph_entity_addr addr; 2276 u64 global_seq; 2277 u64 global_id; 2278 u64 cookie; 2279 u64 flags; 2280 int ret; 2281 2282 if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) { 2283 con->error_msg = "protocol error, unexpected server_ident"; 2284 return -EINVAL; 2285 } 2286 2287 ret = ceph_decode_entity_addrvec(&p, end, true, &addr); 2288 if (ret) { 2289 pr_err("failed to decode server addrs: %d\n", ret); 2290 return ret; 2291 } 2292 2293 ceph_decode_64_safe(&p, end, global_id, bad); 2294 ceph_decode_64_safe(&p, end, global_seq, bad); 2295 ceph_decode_64_safe(&p, end, features, bad); 2296 ceph_decode_64_safe(&p, end, required_features, bad); 2297 ceph_decode_64_safe(&p, end, flags, bad); 2298 ceph_decode_64_safe(&p, end, cookie, bad); 2299 2300 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", 2301 __func__, con, ceph_pr_addr(&addr), le32_to_cpu(addr.nonce), 2302 global_id, global_seq, features, required_features, flags, cookie); 2303 2304 /* is this who we intended to talk to? */ 2305 if (memcmp(&addr, &con->peer_addr, sizeof(con->peer_addr))) { 2306 pr_err("bad peer addr/nonce, want %s/%u, got %s/%u\n", 2307 ceph_pr_addr(&con->peer_addr), 2308 le32_to_cpu(con->peer_addr.nonce), 2309 ceph_pr_addr(&addr), le32_to_cpu(addr.nonce)); 2310 con->error_msg = "wrong peer at address"; 2311 return -EINVAL; 2312 } 2313 2314 if (client->required_features & ~features) { 2315 pr_err("RADOS feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n", 2316 features, client->required_features & ~features); 2317 con->error_msg = "missing required protocol features"; 2318 return -EINVAL; 2319 } 2320 2321 /* 2322 * Both name->type and name->num are set in ceph_con_open() but 2323 * name->num may be bogus in the initial monmap. name->type is 2324 * verified in handle_hello(). 2325 */ 2326 WARN_ON(!con->peer_name.type); 2327 con->peer_name.num = cpu_to_le64(global_id); 2328 con->v2.peer_global_seq = global_seq; 2329 con->peer_features = features; 2330 WARN_ON(required_features & ~client->supported_features); 2331 con->v2.server_cookie = cookie; 2332 2333 if (flags & CEPH_MSG_CONNECT_LOSSY) { 2334 ceph_con_flag_set(con, CEPH_CON_F_LOSSYTX); 2335 WARN_ON(con->v2.server_cookie); 2336 } else { 2337 WARN_ON(!con->v2.server_cookie); 2338 } 2339 2340 clear_in_sign_kvecs(con); 2341 clear_out_sign_kvecs(con); 2342 free_conn_bufs(con); 2343 con->delay = 0; /* reset backoff memory */ 2344 2345 con->state = CEPH_CON_S_OPEN; 2346 con->v2.out_state = OUT_S_GET_NEXT; 2347 return 0; 2348 2349 bad: 2350 pr_err("failed to decode server_ident\n"); 2351 return -EINVAL; 2352 } 2353 2354 static int process_ident_missing_features(struct ceph_connection *con, 2355 void *p, void *end) 2356 { 2357 struct ceph_client *client = from_msgr(con->msgr); 2358 u64 missing_features; 2359 2360 if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) { 2361 con->error_msg = "protocol error, unexpected ident_missing_features"; 2362 return -EINVAL; 2363 } 2364 2365 ceph_decode_64_safe(&p, end, missing_features, bad); 2366 pr_err("RADOS feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n", 2367 client->supported_features, missing_features); 2368 con->error_msg = "missing required protocol features"; 2369 return -EINVAL; 2370 2371 bad: 2372 pr_err("failed to decode ident_missing_features\n"); 2373 return -EINVAL; 2374 } 2375 2376 static int process_session_reconnect_ok(struct ceph_connection *con, 2377 void *p, void *end) 2378 { 2379 u64 seq; 2380 2381 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) { 2382 con->error_msg = "protocol error, unexpected session_reconnect_ok"; 2383 return -EINVAL; 2384 } 2385 2386 ceph_decode_64_safe(&p, end, seq, bad); 2387 2388 dout("%s con %p seq %llu\n", __func__, con, seq); 2389 ceph_con_discard_requeued(con, seq); 2390 2391 clear_in_sign_kvecs(con); 2392 clear_out_sign_kvecs(con); 2393 free_conn_bufs(con); 2394 con->delay = 0; /* reset backoff memory */ 2395 2396 con->state = CEPH_CON_S_OPEN; 2397 con->v2.out_state = OUT_S_GET_NEXT; 2398 return 0; 2399 2400 bad: 2401 pr_err("failed to decode session_reconnect_ok\n"); 2402 return -EINVAL; 2403 } 2404 2405 static int process_session_retry(struct ceph_connection *con, 2406 void *p, void *end) 2407 { 2408 u64 connect_seq; 2409 int ret; 2410 2411 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) { 2412 con->error_msg = "protocol error, unexpected session_retry"; 2413 return -EINVAL; 2414 } 2415 2416 ceph_decode_64_safe(&p, end, connect_seq, bad); 2417 2418 dout("%s con %p connect_seq %llu\n", __func__, con, connect_seq); 2419 WARN_ON(connect_seq <= con->v2.connect_seq); 2420 con->v2.connect_seq = connect_seq + 1; 2421 2422 free_conn_bufs(con); 2423 2424 reset_out_kvecs(con); 2425 ret = prepare_session_reconnect(con); 2426 if (ret) { 2427 pr_err("prepare_session_reconnect (cseq) failed: %d\n", ret); 2428 return ret; 2429 } 2430 2431 return 0; 2432 2433 bad: 2434 pr_err("failed to decode session_retry\n"); 2435 return -EINVAL; 2436 } 2437 2438 static int process_session_retry_global(struct ceph_connection *con, 2439 void *p, void *end) 2440 { 2441 u64 global_seq; 2442 int ret; 2443 2444 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) { 2445 con->error_msg = "protocol error, unexpected session_retry_global"; 2446 return -EINVAL; 2447 } 2448 2449 ceph_decode_64_safe(&p, end, global_seq, bad); 2450 2451 dout("%s con %p global_seq %llu\n", __func__, con, global_seq); 2452 WARN_ON(global_seq <= con->v2.global_seq); 2453 con->v2.global_seq = ceph_get_global_seq(con->msgr, global_seq); 2454 2455 free_conn_bufs(con); 2456 2457 reset_out_kvecs(con); 2458 ret = prepare_session_reconnect(con); 2459 if (ret) { 2460 pr_err("prepare_session_reconnect (gseq) failed: %d\n", ret); 2461 return ret; 2462 } 2463 2464 return 0; 2465 2466 bad: 2467 pr_err("failed to decode session_retry_global\n"); 2468 return -EINVAL; 2469 } 2470 2471 static int process_session_reset(struct ceph_connection *con, 2472 void *p, void *end) 2473 { 2474 bool full; 2475 int ret; 2476 2477 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) { 2478 con->error_msg = "protocol error, unexpected session_reset"; 2479 return -EINVAL; 2480 } 2481 2482 ceph_decode_8_safe(&p, end, full, bad); 2483 if (!full) { 2484 con->error_msg = "protocol error, bad session_reset"; 2485 return -EINVAL; 2486 } 2487 2488 pr_info("%s%lld %s session reset\n", ENTITY_NAME(con->peer_name), 2489 ceph_pr_addr(&con->peer_addr)); 2490 ceph_con_reset_session(con); 2491 2492 mutex_unlock(&con->mutex); 2493 if (con->ops->peer_reset) 2494 con->ops->peer_reset(con); 2495 mutex_lock(&con->mutex); 2496 if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) { 2497 dout("%s con %p state changed to %d\n", __func__, con, 2498 con->state); 2499 return -EAGAIN; 2500 } 2501 2502 free_conn_bufs(con); 2503 2504 reset_out_kvecs(con); 2505 ret = prepare_client_ident(con); 2506 if (ret) { 2507 pr_err("prepare_client_ident (rst) failed: %d\n", ret); 2508 return ret; 2509 } 2510 2511 con->state = CEPH_CON_S_V2_SESSION_CONNECT; 2512 return 0; 2513 2514 bad: 2515 pr_err("failed to decode session_reset\n"); 2516 return -EINVAL; 2517 } 2518 2519 static int process_keepalive2_ack(struct ceph_connection *con, 2520 void *p, void *end) 2521 { 2522 if (con->state != CEPH_CON_S_OPEN) { 2523 con->error_msg = "protocol error, unexpected keepalive2_ack"; 2524 return -EINVAL; 2525 } 2526 2527 ceph_decode_need(&p, end, sizeof(struct ceph_timespec), bad); 2528 ceph_decode_timespec64(&con->last_keepalive_ack, p); 2529 2530 dout("%s con %p timestamp %lld.%09ld\n", __func__, con, 2531 con->last_keepalive_ack.tv_sec, con->last_keepalive_ack.tv_nsec); 2532 2533 return 0; 2534 2535 bad: 2536 pr_err("failed to decode keepalive2_ack\n"); 2537 return -EINVAL; 2538 } 2539 2540 static int process_ack(struct ceph_connection *con, void *p, void *end) 2541 { 2542 u64 seq; 2543 2544 if (con->state != CEPH_CON_S_OPEN) { 2545 con->error_msg = "protocol error, unexpected ack"; 2546 return -EINVAL; 2547 } 2548 2549 ceph_decode_64_safe(&p, end, seq, bad); 2550 2551 dout("%s con %p seq %llu\n", __func__, con, seq); 2552 ceph_con_discard_sent(con, seq); 2553 return 0; 2554 2555 bad: 2556 pr_err("failed to decode ack\n"); 2557 return -EINVAL; 2558 } 2559 2560 static int process_control(struct ceph_connection *con, void *p, void *end) 2561 { 2562 int tag = con->v2.in_desc.fd_tag; 2563 int ret; 2564 2565 dout("%s con %p tag %d len %d\n", __func__, con, tag, (int)(end - p)); 2566 2567 switch (tag) { 2568 case FRAME_TAG_HELLO: 2569 ret = process_hello(con, p, end); 2570 break; 2571 case FRAME_TAG_AUTH_BAD_METHOD: 2572 ret = process_auth_bad_method(con, p, end); 2573 break; 2574 case FRAME_TAG_AUTH_REPLY_MORE: 2575 ret = process_auth_reply_more(con, p, end); 2576 break; 2577 case FRAME_TAG_AUTH_DONE: 2578 ret = process_auth_done(con, p, end); 2579 break; 2580 case FRAME_TAG_AUTH_SIGNATURE: 2581 ret = process_auth_signature(con, p, end); 2582 break; 2583 case FRAME_TAG_SERVER_IDENT: 2584 ret = process_server_ident(con, p, end); 2585 break; 2586 case FRAME_TAG_IDENT_MISSING_FEATURES: 2587 ret = process_ident_missing_features(con, p, end); 2588 break; 2589 case FRAME_TAG_SESSION_RECONNECT_OK: 2590 ret = process_session_reconnect_ok(con, p, end); 2591 break; 2592 case FRAME_TAG_SESSION_RETRY: 2593 ret = process_session_retry(con, p, end); 2594 break; 2595 case FRAME_TAG_SESSION_RETRY_GLOBAL: 2596 ret = process_session_retry_global(con, p, end); 2597 break; 2598 case FRAME_TAG_SESSION_RESET: 2599 ret = process_session_reset(con, p, end); 2600 break; 2601 case FRAME_TAG_KEEPALIVE2_ACK: 2602 ret = process_keepalive2_ack(con, p, end); 2603 break; 2604 case FRAME_TAG_ACK: 2605 ret = process_ack(con, p, end); 2606 break; 2607 default: 2608 pr_err("bad tag %d\n", tag); 2609 con->error_msg = "protocol error, bad tag"; 2610 return -EINVAL; 2611 } 2612 if (ret) { 2613 dout("%s con %p error %d\n", __func__, con, ret); 2614 return ret; 2615 } 2616 2617 prepare_read_preamble(con); 2618 return 0; 2619 } 2620 2621 /* 2622 * Return: 2623 * 1 - con->in_msg set, read message 2624 * 0 - skip message 2625 * <0 - error 2626 */ 2627 static int process_message_header(struct ceph_connection *con, 2628 void *p, void *end) 2629 { 2630 struct ceph_frame_desc *desc = &con->v2.in_desc; 2631 struct ceph_msg_header2 *hdr2 = p; 2632 struct ceph_msg_header hdr; 2633 int skip; 2634 int ret; 2635 u64 seq; 2636 2637 /* verify seq# */ 2638 seq = le64_to_cpu(hdr2->seq); 2639 if ((s64)seq - (s64)con->in_seq < 1) { 2640 pr_info("%s%lld %s skipping old message: seq %llu, expected %llu\n", 2641 ENTITY_NAME(con->peer_name), 2642 ceph_pr_addr(&con->peer_addr), 2643 seq, con->in_seq + 1); 2644 return 0; 2645 } 2646 if ((s64)seq - (s64)con->in_seq > 1) { 2647 pr_err("bad seq %llu, expected %llu\n", seq, con->in_seq + 1); 2648 con->error_msg = "bad message sequence # for incoming message"; 2649 return -EBADE; 2650 } 2651 2652 ceph_con_discard_sent(con, le64_to_cpu(hdr2->ack_seq)); 2653 2654 fill_header(&hdr, hdr2, desc->fd_lens[1], desc->fd_lens[2], 2655 desc->fd_lens[3], &con->peer_name); 2656 ret = ceph_con_in_msg_alloc(con, &hdr, &skip); 2657 if (ret) 2658 return ret; 2659 2660 WARN_ON(!con->in_msg ^ skip); 2661 if (skip) 2662 return 0; 2663 2664 WARN_ON(!con->in_msg); 2665 WARN_ON(con->in_msg->con != con); 2666 return 1; 2667 } 2668 2669 static int process_message(struct ceph_connection *con) 2670 { 2671 ceph_con_process_message(con); 2672 2673 /* 2674 * We could have been closed by ceph_con_close() because 2675 * ceph_con_process_message() temporarily drops con->mutex. 2676 */ 2677 if (con->state != CEPH_CON_S_OPEN) { 2678 dout("%s con %p state changed to %d\n", __func__, con, 2679 con->state); 2680 return -EAGAIN; 2681 } 2682 2683 prepare_read_preamble(con); 2684 return 0; 2685 } 2686 2687 static int __handle_control(struct ceph_connection *con, void *p) 2688 { 2689 void *end = p + con->v2.in_desc.fd_lens[0]; 2690 struct ceph_msg *msg; 2691 int ret; 2692 2693 if (con->v2.in_desc.fd_tag != FRAME_TAG_MESSAGE) 2694 return process_control(con, p, end); 2695 2696 ret = process_message_header(con, p, end); 2697 if (ret < 0) 2698 return ret; 2699 if (ret == 0) { 2700 prepare_skip_message(con); 2701 return 0; 2702 } 2703 2704 msg = con->in_msg; /* set in process_message_header() */ 2705 if (front_len(msg)) { 2706 WARN_ON(front_len(msg) > msg->front_alloc_len); 2707 msg->front.iov_len = front_len(msg); 2708 } else { 2709 msg->front.iov_len = 0; 2710 } 2711 if (middle_len(msg)) { 2712 WARN_ON(middle_len(msg) > msg->middle->alloc_len); 2713 msg->middle->vec.iov_len = middle_len(msg); 2714 } else if (msg->middle) { 2715 msg->middle->vec.iov_len = 0; 2716 } 2717 2718 if (!front_len(msg) && !middle_len(msg) && !data_len(msg)) 2719 return process_message(con); 2720 2721 if (con_secure(con)) 2722 return prepare_read_tail_secure(con); 2723 2724 return prepare_read_tail_plain(con); 2725 } 2726 2727 static int handle_preamble(struct ceph_connection *con) 2728 { 2729 struct ceph_frame_desc *desc = &con->v2.in_desc; 2730 int ret; 2731 2732 if (con_secure(con)) { 2733 ret = decrypt_preamble(con); 2734 if (ret) { 2735 if (ret == -EBADMSG) 2736 con->error_msg = "integrity error, bad preamble auth tag"; 2737 return ret; 2738 } 2739 } 2740 2741 ret = decode_preamble(con->v2.in_buf, desc); 2742 if (ret) { 2743 if (ret == -EBADMSG) 2744 con->error_msg = "integrity error, bad crc"; 2745 else 2746 con->error_msg = "protocol error, bad preamble"; 2747 return ret; 2748 } 2749 2750 dout("%s con %p tag %d seg_cnt %d %d+%d+%d+%d\n", __func__, 2751 con, desc->fd_tag, desc->fd_seg_cnt, desc->fd_lens[0], 2752 desc->fd_lens[1], desc->fd_lens[2], desc->fd_lens[3]); 2753 2754 if (!con_secure(con)) 2755 return prepare_read_control(con); 2756 2757 if (desc->fd_lens[0] > CEPH_PREAMBLE_INLINE_LEN) 2758 return prepare_read_control_remainder(con); 2759 2760 return __handle_control(con, CTRL_BODY(con->v2.in_buf)); 2761 } 2762 2763 static int handle_control(struct ceph_connection *con) 2764 { 2765 int ctrl_len = con->v2.in_desc.fd_lens[0]; 2766 void *buf; 2767 int ret; 2768 2769 WARN_ON(con_secure(con)); 2770 2771 ret = verify_control_crc(con); 2772 if (ret) { 2773 con->error_msg = "integrity error, bad crc"; 2774 return ret; 2775 } 2776 2777 if (con->state == CEPH_CON_S_V2_AUTH) { 2778 buf = alloc_conn_buf(con, ctrl_len); 2779 if (!buf) 2780 return -ENOMEM; 2781 2782 memcpy(buf, con->v2.in_kvecs[0].iov_base, ctrl_len); 2783 return __handle_control(con, buf); 2784 } 2785 2786 return __handle_control(con, con->v2.in_kvecs[0].iov_base); 2787 } 2788 2789 static int handle_control_remainder(struct ceph_connection *con) 2790 { 2791 int ret; 2792 2793 WARN_ON(!con_secure(con)); 2794 2795 ret = decrypt_control_remainder(con); 2796 if (ret) { 2797 if (ret == -EBADMSG) 2798 con->error_msg = "integrity error, bad control remainder auth tag"; 2799 return ret; 2800 } 2801 2802 return __handle_control(con, con->v2.in_kvecs[0].iov_base - 2803 CEPH_PREAMBLE_INLINE_LEN); 2804 } 2805 2806 static int handle_epilogue(struct ceph_connection *con) 2807 { 2808 u32 front_crc, middle_crc, data_crc; 2809 int ret; 2810 2811 if (con_secure(con)) { 2812 ret = decrypt_tail(con); 2813 if (ret) { 2814 if (ret == -EBADMSG) 2815 con->error_msg = "integrity error, bad epilogue auth tag"; 2816 return ret; 2817 } 2818 2819 /* just late_status */ 2820 ret = decode_epilogue(con->v2.in_buf, NULL, NULL, NULL); 2821 if (ret) { 2822 con->error_msg = "protocol error, bad epilogue"; 2823 return ret; 2824 } 2825 } else { 2826 ret = decode_epilogue(con->v2.in_buf, &front_crc, 2827 &middle_crc, &data_crc); 2828 if (ret) { 2829 con->error_msg = "protocol error, bad epilogue"; 2830 return ret; 2831 } 2832 2833 ret = verify_epilogue_crcs(con, front_crc, middle_crc, 2834 data_crc); 2835 if (ret) { 2836 con->error_msg = "integrity error, bad crc"; 2837 return ret; 2838 } 2839 } 2840 2841 return process_message(con); 2842 } 2843 2844 static void finish_skip(struct ceph_connection *con) 2845 { 2846 dout("%s con %p\n", __func__, con); 2847 2848 if (con_secure(con)) 2849 gcm_inc_nonce(&con->v2.in_gcm_nonce); 2850 2851 __finish_skip(con); 2852 } 2853 2854 static int populate_in_iter(struct ceph_connection *con) 2855 { 2856 int ret; 2857 2858 dout("%s con %p state %d in_state %d\n", __func__, con, con->state, 2859 con->v2.in_state); 2860 WARN_ON(iov_iter_count(&con->v2.in_iter)); 2861 2862 if (con->state == CEPH_CON_S_V2_BANNER_PREFIX) { 2863 ret = process_banner_prefix(con); 2864 } else if (con->state == CEPH_CON_S_V2_BANNER_PAYLOAD) { 2865 ret = process_banner_payload(con); 2866 } else if ((con->state >= CEPH_CON_S_V2_HELLO && 2867 con->state <= CEPH_CON_S_V2_SESSION_RECONNECT) || 2868 con->state == CEPH_CON_S_OPEN) { 2869 switch (con->v2.in_state) { 2870 case IN_S_HANDLE_PREAMBLE: 2871 ret = handle_preamble(con); 2872 break; 2873 case IN_S_HANDLE_CONTROL: 2874 ret = handle_control(con); 2875 break; 2876 case IN_S_HANDLE_CONTROL_REMAINDER: 2877 ret = handle_control_remainder(con); 2878 break; 2879 case IN_S_PREPARE_READ_DATA: 2880 ret = prepare_read_data(con); 2881 break; 2882 case IN_S_PREPARE_READ_DATA_CONT: 2883 prepare_read_data_cont(con); 2884 ret = 0; 2885 break; 2886 case IN_S_PREPARE_READ_ENC_PAGE: 2887 prepare_read_enc_page(con); 2888 ret = 0; 2889 break; 2890 case IN_S_HANDLE_EPILOGUE: 2891 ret = handle_epilogue(con); 2892 break; 2893 case IN_S_FINISH_SKIP: 2894 finish_skip(con); 2895 ret = 0; 2896 break; 2897 default: 2898 WARN(1, "bad in_state %d", con->v2.in_state); 2899 return -EINVAL; 2900 } 2901 } else { 2902 WARN(1, "bad state %d", con->state); 2903 return -EINVAL; 2904 } 2905 if (ret) { 2906 dout("%s con %p error %d\n", __func__, con, ret); 2907 return ret; 2908 } 2909 2910 if (WARN_ON(!iov_iter_count(&con->v2.in_iter))) 2911 return -ENODATA; 2912 dout("%s con %p populated %zu\n", __func__, con, 2913 iov_iter_count(&con->v2.in_iter)); 2914 return 1; 2915 } 2916 2917 int ceph_con_v2_try_read(struct ceph_connection *con) 2918 { 2919 int ret; 2920 2921 dout("%s con %p state %d need %zu\n", __func__, con, con->state, 2922 iov_iter_count(&con->v2.in_iter)); 2923 2924 if (con->state == CEPH_CON_S_PREOPEN) 2925 return 0; 2926 2927 /* 2928 * We should always have something pending here. If not, 2929 * avoid calling populate_in_iter() as if we read something 2930 * (ceph_tcp_recv() would immediately return 1). 2931 */ 2932 if (WARN_ON(!iov_iter_count(&con->v2.in_iter))) 2933 return -ENODATA; 2934 2935 for (;;) { 2936 ret = ceph_tcp_recv(con); 2937 if (ret <= 0) 2938 return ret; 2939 2940 ret = populate_in_iter(con); 2941 if (ret <= 0) { 2942 if (ret && ret != -EAGAIN && !con->error_msg) 2943 con->error_msg = "read processing error"; 2944 return ret; 2945 } 2946 } 2947 } 2948 2949 static void queue_data(struct ceph_connection *con) 2950 { 2951 struct bio_vec bv; 2952 2953 con->v2.out_epil.data_crc = -1; 2954 ceph_msg_data_cursor_init(&con->v2.out_cursor, con->out_msg, 2955 data_len(con->out_msg)); 2956 2957 get_bvec_at(&con->v2.out_cursor, &bv); 2958 set_out_bvec(con, &bv, true); 2959 con->v2.out_state = OUT_S_QUEUE_DATA_CONT; 2960 } 2961 2962 static void queue_data_cont(struct ceph_connection *con) 2963 { 2964 struct bio_vec bv; 2965 2966 con->v2.out_epil.data_crc = ceph_crc32c_page( 2967 con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page, 2968 con->v2.out_bvec.bv_offset, con->v2.out_bvec.bv_len); 2969 2970 ceph_msg_data_advance(&con->v2.out_cursor, con->v2.out_bvec.bv_len); 2971 if (con->v2.out_cursor.total_resid) { 2972 get_bvec_at(&con->v2.out_cursor, &bv); 2973 set_out_bvec(con, &bv, true); 2974 WARN_ON(con->v2.out_state != OUT_S_QUEUE_DATA_CONT); 2975 return; 2976 } 2977 2978 /* 2979 * We've written all data. Queue epilogue. Once it's written, 2980 * we are done. 2981 */ 2982 reset_out_kvecs(con); 2983 prepare_epilogue_plain(con, false); 2984 con->v2.out_state = OUT_S_FINISH_MESSAGE; 2985 } 2986 2987 static void queue_enc_page(struct ceph_connection *con) 2988 { 2989 struct bio_vec bv; 2990 2991 dout("%s con %p i %d resid %d\n", __func__, con, con->v2.out_enc_i, 2992 con->v2.out_enc_resid); 2993 WARN_ON(!con->v2.out_enc_resid); 2994 2995 bvec_set_page(&bv, con->v2.out_enc_pages[con->v2.out_enc_i], 2996 min(con->v2.out_enc_resid, (int)PAGE_SIZE), 0); 2997 2998 set_out_bvec(con, &bv, false); 2999 con->v2.out_enc_i++; 3000 con->v2.out_enc_resid -= bv.bv_len; 3001 3002 if (con->v2.out_enc_resid) { 3003 WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE); 3004 return; 3005 } 3006 3007 /* 3008 * We've queued the last piece of ciphertext (ending with 3009 * epilogue) + auth tag. Once it's written, we are done. 3010 */ 3011 WARN_ON(con->v2.out_enc_i != con->v2.out_enc_page_cnt); 3012 con->v2.out_state = OUT_S_FINISH_MESSAGE; 3013 } 3014 3015 static void queue_zeros(struct ceph_connection *con) 3016 { 3017 dout("%s con %p out_zero %d\n", __func__, con, con->v2.out_zero); 3018 3019 if (con->v2.out_zero) { 3020 set_out_bvec_zero(con); 3021 con->v2.out_zero -= con->v2.out_bvec.bv_len; 3022 con->v2.out_state = OUT_S_QUEUE_ZEROS; 3023 return; 3024 } 3025 3026 /* 3027 * We've zero-filled everything up to epilogue. Queue epilogue 3028 * with late_status set to ABORTED and crcs adjusted for zeros. 3029 * Once it's written, we are done patching up for the revoke. 3030 */ 3031 reset_out_kvecs(con); 3032 prepare_epilogue_plain(con, true); 3033 con->v2.out_state = OUT_S_FINISH_MESSAGE; 3034 } 3035 3036 static void finish_message(struct ceph_connection *con) 3037 { 3038 dout("%s con %p msg %p\n", __func__, con, con->out_msg); 3039 3040 /* we end up here both plain and secure modes */ 3041 if (con->v2.out_enc_pages) { 3042 WARN_ON(!con->v2.out_enc_page_cnt); 3043 ceph_release_page_vector(con->v2.out_enc_pages, 3044 con->v2.out_enc_page_cnt); 3045 con->v2.out_enc_pages = NULL; 3046 con->v2.out_enc_page_cnt = 0; 3047 } 3048 /* message may have been revoked */ 3049 if (con->out_msg) { 3050 ceph_msg_put(con->out_msg); 3051 con->out_msg = NULL; 3052 } 3053 3054 con->v2.out_state = OUT_S_GET_NEXT; 3055 } 3056 3057 static int populate_out_iter(struct ceph_connection *con) 3058 { 3059 int ret; 3060 3061 dout("%s con %p state %d out_state %d\n", __func__, con, con->state, 3062 con->v2.out_state); 3063 WARN_ON(iov_iter_count(&con->v2.out_iter)); 3064 3065 if (con->state != CEPH_CON_S_OPEN) { 3066 WARN_ON(con->state < CEPH_CON_S_V2_BANNER_PREFIX || 3067 con->state > CEPH_CON_S_V2_SESSION_RECONNECT); 3068 goto nothing_pending; 3069 } 3070 3071 switch (con->v2.out_state) { 3072 case OUT_S_QUEUE_DATA: 3073 WARN_ON(!con->out_msg); 3074 queue_data(con); 3075 goto populated; 3076 case OUT_S_QUEUE_DATA_CONT: 3077 WARN_ON(!con->out_msg); 3078 queue_data_cont(con); 3079 goto populated; 3080 case OUT_S_QUEUE_ENC_PAGE: 3081 queue_enc_page(con); 3082 goto populated; 3083 case OUT_S_QUEUE_ZEROS: 3084 WARN_ON(con->out_msg); /* revoked */ 3085 queue_zeros(con); 3086 goto populated; 3087 case OUT_S_FINISH_MESSAGE: 3088 finish_message(con); 3089 break; 3090 case OUT_S_GET_NEXT: 3091 break; 3092 default: 3093 WARN(1, "bad out_state %d", con->v2.out_state); 3094 return -EINVAL; 3095 } 3096 3097 WARN_ON(con->v2.out_state != OUT_S_GET_NEXT); 3098 if (ceph_con_flag_test_and_clear(con, CEPH_CON_F_KEEPALIVE_PENDING)) { 3099 ret = prepare_keepalive2(con); 3100 if (ret) { 3101 pr_err("prepare_keepalive2 failed: %d\n", ret); 3102 return ret; 3103 } 3104 } else if (!list_empty(&con->out_queue)) { 3105 ceph_con_get_out_msg(con); 3106 ret = prepare_message(con); 3107 if (ret) { 3108 pr_err("prepare_message failed: %d\n", ret); 3109 return ret; 3110 } 3111 } else if (con->in_seq > con->in_seq_acked) { 3112 ret = prepare_ack(con); 3113 if (ret) { 3114 pr_err("prepare_ack failed: %d\n", ret); 3115 return ret; 3116 } 3117 } else { 3118 goto nothing_pending; 3119 } 3120 3121 populated: 3122 if (WARN_ON(!iov_iter_count(&con->v2.out_iter))) 3123 return -ENODATA; 3124 dout("%s con %p populated %zu\n", __func__, con, 3125 iov_iter_count(&con->v2.out_iter)); 3126 return 1; 3127 3128 nothing_pending: 3129 WARN_ON(iov_iter_count(&con->v2.out_iter)); 3130 dout("%s con %p nothing pending\n", __func__, con); 3131 ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING); 3132 return 0; 3133 } 3134 3135 int ceph_con_v2_try_write(struct ceph_connection *con) 3136 { 3137 int ret; 3138 3139 dout("%s con %p state %d have %zu\n", __func__, con, con->state, 3140 iov_iter_count(&con->v2.out_iter)); 3141 3142 /* open the socket first? */ 3143 if (con->state == CEPH_CON_S_PREOPEN) { 3144 WARN_ON(con->peer_addr.type != CEPH_ENTITY_ADDR_TYPE_MSGR2); 3145 3146 /* 3147 * Always bump global_seq. Bump connect_seq only if 3148 * there is a session (i.e. we are reconnecting and will 3149 * send session_reconnect instead of client_ident). 3150 */ 3151 con->v2.global_seq = ceph_get_global_seq(con->msgr, 0); 3152 if (con->v2.server_cookie) 3153 con->v2.connect_seq++; 3154 3155 ret = prepare_read_banner_prefix(con); 3156 if (ret) { 3157 pr_err("prepare_read_banner_prefix failed: %d\n", ret); 3158 con->error_msg = "connect error"; 3159 return ret; 3160 } 3161 3162 reset_out_kvecs(con); 3163 ret = prepare_banner(con); 3164 if (ret) { 3165 pr_err("prepare_banner failed: %d\n", ret); 3166 con->error_msg = "connect error"; 3167 return ret; 3168 } 3169 3170 ret = ceph_tcp_connect(con); 3171 if (ret) { 3172 pr_err("ceph_tcp_connect failed: %d\n", ret); 3173 con->error_msg = "connect error"; 3174 return ret; 3175 } 3176 } 3177 3178 if (!iov_iter_count(&con->v2.out_iter)) { 3179 ret = populate_out_iter(con); 3180 if (ret <= 0) { 3181 if (ret && ret != -EAGAIN && !con->error_msg) 3182 con->error_msg = "write processing error"; 3183 return ret; 3184 } 3185 } 3186 3187 tcp_sock_set_cork(con->sock->sk, true); 3188 for (;;) { 3189 ret = ceph_tcp_send(con); 3190 if (ret <= 0) 3191 break; 3192 3193 ret = populate_out_iter(con); 3194 if (ret <= 0) { 3195 if (ret && ret != -EAGAIN && !con->error_msg) 3196 con->error_msg = "write processing error"; 3197 break; 3198 } 3199 } 3200 3201 tcp_sock_set_cork(con->sock->sk, false); 3202 return ret; 3203 } 3204 3205 static u32 crc32c_zeros(u32 crc, int zero_len) 3206 { 3207 int len; 3208 3209 while (zero_len) { 3210 len = min(zero_len, (int)PAGE_SIZE); 3211 crc = crc32c(crc, page_address(ceph_zero_page), len); 3212 zero_len -= len; 3213 } 3214 3215 return crc; 3216 } 3217 3218 static void prepare_zero_front(struct ceph_connection *con, int resid) 3219 { 3220 int sent; 3221 3222 WARN_ON(!resid || resid > front_len(con->out_msg)); 3223 sent = front_len(con->out_msg) - resid; 3224 dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid); 3225 3226 if (sent) { 3227 con->v2.out_epil.front_crc = 3228 crc32c(-1, con->out_msg->front.iov_base, sent); 3229 con->v2.out_epil.front_crc = 3230 crc32c_zeros(con->v2.out_epil.front_crc, resid); 3231 } else { 3232 con->v2.out_epil.front_crc = crc32c_zeros(-1, resid); 3233 } 3234 3235 con->v2.out_iter.count -= resid; 3236 out_zero_add(con, resid); 3237 } 3238 3239 static void prepare_zero_middle(struct ceph_connection *con, int resid) 3240 { 3241 int sent; 3242 3243 WARN_ON(!resid || resid > middle_len(con->out_msg)); 3244 sent = middle_len(con->out_msg) - resid; 3245 dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid); 3246 3247 if (sent) { 3248 con->v2.out_epil.middle_crc = 3249 crc32c(-1, con->out_msg->middle->vec.iov_base, sent); 3250 con->v2.out_epil.middle_crc = 3251 crc32c_zeros(con->v2.out_epil.middle_crc, resid); 3252 } else { 3253 con->v2.out_epil.middle_crc = crc32c_zeros(-1, resid); 3254 } 3255 3256 con->v2.out_iter.count -= resid; 3257 out_zero_add(con, resid); 3258 } 3259 3260 static void prepare_zero_data(struct ceph_connection *con) 3261 { 3262 dout("%s con %p\n", __func__, con); 3263 con->v2.out_epil.data_crc = crc32c_zeros(-1, data_len(con->out_msg)); 3264 out_zero_add(con, data_len(con->out_msg)); 3265 } 3266 3267 static void revoke_at_queue_data(struct ceph_connection *con) 3268 { 3269 int boundary; 3270 int resid; 3271 3272 WARN_ON(!data_len(con->out_msg)); 3273 WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter)); 3274 resid = iov_iter_count(&con->v2.out_iter); 3275 3276 boundary = front_len(con->out_msg) + middle_len(con->out_msg); 3277 if (resid > boundary) { 3278 resid -= boundary; 3279 WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN); 3280 dout("%s con %p was sending head\n", __func__, con); 3281 if (front_len(con->out_msg)) 3282 prepare_zero_front(con, front_len(con->out_msg)); 3283 if (middle_len(con->out_msg)) 3284 prepare_zero_middle(con, middle_len(con->out_msg)); 3285 prepare_zero_data(con); 3286 WARN_ON(iov_iter_count(&con->v2.out_iter) != resid); 3287 con->v2.out_state = OUT_S_QUEUE_ZEROS; 3288 return; 3289 } 3290 3291 boundary = middle_len(con->out_msg); 3292 if (resid > boundary) { 3293 resid -= boundary; 3294 dout("%s con %p was sending front\n", __func__, con); 3295 prepare_zero_front(con, resid); 3296 if (middle_len(con->out_msg)) 3297 prepare_zero_middle(con, middle_len(con->out_msg)); 3298 prepare_zero_data(con); 3299 queue_zeros(con); 3300 return; 3301 } 3302 3303 WARN_ON(!resid); 3304 dout("%s con %p was sending middle\n", __func__, con); 3305 prepare_zero_middle(con, resid); 3306 prepare_zero_data(con); 3307 queue_zeros(con); 3308 } 3309 3310 static void revoke_at_queue_data_cont(struct ceph_connection *con) 3311 { 3312 int sent, resid; /* current piece of data */ 3313 3314 WARN_ON(!data_len(con->out_msg)); 3315 WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter)); 3316 resid = iov_iter_count(&con->v2.out_iter); 3317 WARN_ON(!resid || resid > con->v2.out_bvec.bv_len); 3318 sent = con->v2.out_bvec.bv_len - resid; 3319 dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid); 3320 3321 if (sent) { 3322 con->v2.out_epil.data_crc = ceph_crc32c_page( 3323 con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page, 3324 con->v2.out_bvec.bv_offset, sent); 3325 ceph_msg_data_advance(&con->v2.out_cursor, sent); 3326 } 3327 WARN_ON(resid > con->v2.out_cursor.total_resid); 3328 con->v2.out_epil.data_crc = crc32c_zeros(con->v2.out_epil.data_crc, 3329 con->v2.out_cursor.total_resid); 3330 3331 con->v2.out_iter.count -= resid; 3332 out_zero_add(con, con->v2.out_cursor.total_resid); 3333 queue_zeros(con); 3334 } 3335 3336 static void revoke_at_finish_message(struct ceph_connection *con) 3337 { 3338 int boundary; 3339 int resid; 3340 3341 WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter)); 3342 resid = iov_iter_count(&con->v2.out_iter); 3343 3344 if (!front_len(con->out_msg) && !middle_len(con->out_msg) && 3345 !data_len(con->out_msg)) { 3346 WARN_ON(!resid || resid > MESSAGE_HEAD_PLAIN_LEN); 3347 dout("%s con %p was sending head (empty message) - noop\n", 3348 __func__, con); 3349 return; 3350 } 3351 3352 boundary = front_len(con->out_msg) + middle_len(con->out_msg) + 3353 CEPH_EPILOGUE_PLAIN_LEN; 3354 if (resid > boundary) { 3355 resid -= boundary; 3356 WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN); 3357 dout("%s con %p was sending head\n", __func__, con); 3358 if (front_len(con->out_msg)) 3359 prepare_zero_front(con, front_len(con->out_msg)); 3360 if (middle_len(con->out_msg)) 3361 prepare_zero_middle(con, middle_len(con->out_msg)); 3362 con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN; 3363 WARN_ON(iov_iter_count(&con->v2.out_iter) != resid); 3364 con->v2.out_state = OUT_S_QUEUE_ZEROS; 3365 return; 3366 } 3367 3368 boundary = middle_len(con->out_msg) + CEPH_EPILOGUE_PLAIN_LEN; 3369 if (resid > boundary) { 3370 resid -= boundary; 3371 dout("%s con %p was sending front\n", __func__, con); 3372 prepare_zero_front(con, resid); 3373 if (middle_len(con->out_msg)) 3374 prepare_zero_middle(con, middle_len(con->out_msg)); 3375 con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN; 3376 queue_zeros(con); 3377 return; 3378 } 3379 3380 boundary = CEPH_EPILOGUE_PLAIN_LEN; 3381 if (resid > boundary) { 3382 resid -= boundary; 3383 dout("%s con %p was sending middle\n", __func__, con); 3384 prepare_zero_middle(con, resid); 3385 con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN; 3386 queue_zeros(con); 3387 return; 3388 } 3389 3390 WARN_ON(!resid); 3391 dout("%s con %p was sending epilogue - noop\n", __func__, con); 3392 } 3393 3394 void ceph_con_v2_revoke(struct ceph_connection *con) 3395 { 3396 WARN_ON(con->v2.out_zero); 3397 3398 if (con_secure(con)) { 3399 WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE && 3400 con->v2.out_state != OUT_S_FINISH_MESSAGE); 3401 dout("%s con %p secure - noop\n", __func__, con); 3402 return; 3403 } 3404 3405 switch (con->v2.out_state) { 3406 case OUT_S_QUEUE_DATA: 3407 revoke_at_queue_data(con); 3408 break; 3409 case OUT_S_QUEUE_DATA_CONT: 3410 revoke_at_queue_data_cont(con); 3411 break; 3412 case OUT_S_FINISH_MESSAGE: 3413 revoke_at_finish_message(con); 3414 break; 3415 default: 3416 WARN(1, "bad out_state %d", con->v2.out_state); 3417 break; 3418 } 3419 } 3420 3421 static void revoke_at_prepare_read_data(struct ceph_connection *con) 3422 { 3423 int remaining; 3424 int resid; 3425 3426 WARN_ON(con_secure(con)); 3427 WARN_ON(!data_len(con->in_msg)); 3428 WARN_ON(!iov_iter_is_kvec(&con->v2.in_iter)); 3429 resid = iov_iter_count(&con->v2.in_iter); 3430 WARN_ON(!resid); 3431 3432 remaining = data_len(con->in_msg) + CEPH_EPILOGUE_PLAIN_LEN; 3433 dout("%s con %p resid %d remaining %d\n", __func__, con, resid, 3434 remaining); 3435 con->v2.in_iter.count -= resid; 3436 set_in_skip(con, resid + remaining); 3437 con->v2.in_state = IN_S_FINISH_SKIP; 3438 } 3439 3440 static void revoke_at_prepare_read_data_cont(struct ceph_connection *con) 3441 { 3442 int recved, resid; /* current piece of data */ 3443 int remaining; 3444 3445 WARN_ON(con_secure(con)); 3446 WARN_ON(!data_len(con->in_msg)); 3447 WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter)); 3448 resid = iov_iter_count(&con->v2.in_iter); 3449 WARN_ON(!resid || resid > con->v2.in_bvec.bv_len); 3450 recved = con->v2.in_bvec.bv_len - resid; 3451 dout("%s con %p recved %d resid %d\n", __func__, con, recved, resid); 3452 3453 if (recved) 3454 ceph_msg_data_advance(&con->v2.in_cursor, recved); 3455 WARN_ON(resid > con->v2.in_cursor.total_resid); 3456 3457 remaining = CEPH_EPILOGUE_PLAIN_LEN; 3458 dout("%s con %p total_resid %zu remaining %d\n", __func__, con, 3459 con->v2.in_cursor.total_resid, remaining); 3460 con->v2.in_iter.count -= resid; 3461 set_in_skip(con, con->v2.in_cursor.total_resid + remaining); 3462 con->v2.in_state = IN_S_FINISH_SKIP; 3463 } 3464 3465 static void revoke_at_prepare_read_enc_page(struct ceph_connection *con) 3466 { 3467 int resid; /* current enc page (not necessarily data) */ 3468 3469 WARN_ON(!con_secure(con)); 3470 WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter)); 3471 resid = iov_iter_count(&con->v2.in_iter); 3472 WARN_ON(!resid || resid > con->v2.in_bvec.bv_len); 3473 3474 dout("%s con %p resid %d enc_resid %d\n", __func__, con, resid, 3475 con->v2.in_enc_resid); 3476 con->v2.in_iter.count -= resid; 3477 set_in_skip(con, resid + con->v2.in_enc_resid); 3478 con->v2.in_state = IN_S_FINISH_SKIP; 3479 } 3480 3481 static void revoke_at_handle_epilogue(struct ceph_connection *con) 3482 { 3483 int resid; 3484 3485 resid = iov_iter_count(&con->v2.in_iter); 3486 WARN_ON(!resid); 3487 3488 dout("%s con %p resid %d\n", __func__, con, resid); 3489 con->v2.in_iter.count -= resid; 3490 set_in_skip(con, resid); 3491 con->v2.in_state = IN_S_FINISH_SKIP; 3492 } 3493 3494 void ceph_con_v2_revoke_incoming(struct ceph_connection *con) 3495 { 3496 switch (con->v2.in_state) { 3497 case IN_S_PREPARE_READ_DATA: 3498 revoke_at_prepare_read_data(con); 3499 break; 3500 case IN_S_PREPARE_READ_DATA_CONT: 3501 revoke_at_prepare_read_data_cont(con); 3502 break; 3503 case IN_S_PREPARE_READ_ENC_PAGE: 3504 revoke_at_prepare_read_enc_page(con); 3505 break; 3506 case IN_S_HANDLE_EPILOGUE: 3507 revoke_at_handle_epilogue(con); 3508 break; 3509 default: 3510 WARN(1, "bad in_state %d", con->v2.in_state); 3511 break; 3512 } 3513 } 3514 3515 bool ceph_con_v2_opened(struct ceph_connection *con) 3516 { 3517 return con->v2.peer_global_seq; 3518 } 3519 3520 void ceph_con_v2_reset_session(struct ceph_connection *con) 3521 { 3522 con->v2.client_cookie = 0; 3523 con->v2.server_cookie = 0; 3524 con->v2.global_seq = 0; 3525 con->v2.connect_seq = 0; 3526 con->v2.peer_global_seq = 0; 3527 } 3528 3529 void ceph_con_v2_reset_protocol(struct ceph_connection *con) 3530 { 3531 iov_iter_truncate(&con->v2.in_iter, 0); 3532 iov_iter_truncate(&con->v2.out_iter, 0); 3533 con->v2.out_zero = 0; 3534 3535 clear_in_sign_kvecs(con); 3536 clear_out_sign_kvecs(con); 3537 free_conn_bufs(con); 3538 3539 if (con->v2.in_enc_pages) { 3540 WARN_ON(!con->v2.in_enc_page_cnt); 3541 ceph_release_page_vector(con->v2.in_enc_pages, 3542 con->v2.in_enc_page_cnt); 3543 con->v2.in_enc_pages = NULL; 3544 con->v2.in_enc_page_cnt = 0; 3545 } 3546 if (con->v2.out_enc_pages) { 3547 WARN_ON(!con->v2.out_enc_page_cnt); 3548 ceph_release_page_vector(con->v2.out_enc_pages, 3549 con->v2.out_enc_page_cnt); 3550 con->v2.out_enc_pages = NULL; 3551 con->v2.out_enc_page_cnt = 0; 3552 } 3553 3554 con->v2.con_mode = CEPH_CON_MODE_UNKNOWN; 3555 memzero_explicit(&con->v2.in_gcm_nonce, CEPH_GCM_IV_LEN); 3556 memzero_explicit(&con->v2.out_gcm_nonce, CEPH_GCM_IV_LEN); 3557 3558 if (con->v2.hmac_tfm) { 3559 crypto_free_shash(con->v2.hmac_tfm); 3560 con->v2.hmac_tfm = NULL; 3561 } 3562 if (con->v2.gcm_req) { 3563 aead_request_free(con->v2.gcm_req); 3564 con->v2.gcm_req = NULL; 3565 } 3566 if (con->v2.gcm_tfm) { 3567 crypto_free_aead(con->v2.gcm_tfm); 3568 con->v2.gcm_tfm = NULL; 3569 } 3570 } 3571