1 /* 2 drbd_receiver.c 3 4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg. 5 6 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. 7 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. 8 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. 9 10 drbd is free software; you can redistribute it and/or modify 11 it under the terms of the GNU General Public License as published by 12 the Free Software Foundation; either version 2, or (at your option) 13 any later version. 14 15 drbd is distributed in the hope that it will be useful, 16 but WITHOUT ANY WARRANTY; without even the implied warranty of 17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 18 GNU General Public License for more details. 19 20 You should have received a copy of the GNU General Public License 21 along with drbd; see the file COPYING. If not, write to 22 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. 23 */ 24 25 26 #include <linux/module.h> 27 28 #include <asm/uaccess.h> 29 #include <net/sock.h> 30 31 #include <linux/drbd.h> 32 #include <linux/fs.h> 33 #include <linux/file.h> 34 #include <linux/in.h> 35 #include <linux/mm.h> 36 #include <linux/memcontrol.h> 37 #include <linux/mm_inline.h> 38 #include <linux/slab.h> 39 #include <linux/pkt_sched.h> 40 #define __KERNEL_SYSCALLS__ 41 #include <linux/unistd.h> 42 #include <linux/vmalloc.h> 43 #include <linux/random.h> 44 #include <linux/string.h> 45 #include <linux/scatterlist.h> 46 #include "drbd_int.h" 47 #include "drbd_req.h" 48 49 #include "drbd_vli.h" 50 51 enum finish_epoch { 52 FE_STILL_LIVE, 53 FE_DESTROYED, 54 FE_RECYCLED, 55 }; 56 57 static int drbd_do_handshake(struct drbd_conf *mdev); 58 static int drbd_do_auth(struct drbd_conf *mdev); 59 60 static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event); 61 static int e_end_block(struct drbd_conf *, struct drbd_work *, int); 62 63 64 #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) 65 66 /* 67 * some helper functions to deal with single linked page lists, 68 * page->private being our "next" pointer. 69 */ 70 71 /* If at least n pages are linked at head, get n pages off. 72 * Otherwise, don't modify head, and return NULL. 73 * Locking is the responsibility of the caller. 74 */ 75 static struct page *page_chain_del(struct page **head, int n) 76 { 77 struct page *page; 78 struct page *tmp; 79 80 BUG_ON(!n); 81 BUG_ON(!head); 82 83 page = *head; 84 85 if (!page) 86 return NULL; 87 88 while (page) { 89 tmp = page_chain_next(page); 90 if (--n == 0) 91 break; /* found sufficient pages */ 92 if (tmp == NULL) 93 /* insufficient pages, don't use any of them. */ 94 return NULL; 95 page = tmp; 96 } 97 98 /* add end of list marker for the returned list */ 99 set_page_private(page, 0); 100 /* actual return value, and adjustment of head */ 101 page = *head; 102 *head = tmp; 103 return page; 104 } 105 106 /* may be used outside of locks to find the tail of a (usually short) 107 * "private" page chain, before adding it back to a global chain head 108 * with page_chain_add() under a spinlock. */ 109 static struct page *page_chain_tail(struct page *page, int *len) 110 { 111 struct page *tmp; 112 int i = 1; 113 while ((tmp = page_chain_next(page))) 114 ++i, page = tmp; 115 if (len) 116 *len = i; 117 return page; 118 } 119 120 static int page_chain_free(struct page *page) 121 { 122 struct page *tmp; 123 int i = 0; 124 page_chain_for_each_safe(page, tmp) { 125 put_page(page); 126 ++i; 127 } 128 return i; 129 } 130 131 static void page_chain_add(struct page **head, 132 struct page *chain_first, struct page *chain_last) 133 { 134 #if 1 135 struct page *tmp; 136 tmp = page_chain_tail(chain_first, NULL); 137 BUG_ON(tmp != chain_last); 138 #endif 139 140 /* add chain to head */ 141 set_page_private(chain_last, (unsigned long)*head); 142 *head = chain_first; 143 } 144 145 static struct page *drbd_pp_first_pages_or_try_alloc(struct drbd_conf *mdev, int number) 146 { 147 struct page *page = NULL; 148 struct page *tmp = NULL; 149 int i = 0; 150 151 /* Yes, testing drbd_pp_vacant outside the lock is racy. 152 * So what. It saves a spin_lock. */ 153 if (drbd_pp_vacant >= number) { 154 spin_lock(&drbd_pp_lock); 155 page = page_chain_del(&drbd_pp_pool, number); 156 if (page) 157 drbd_pp_vacant -= number; 158 spin_unlock(&drbd_pp_lock); 159 if (page) 160 return page; 161 } 162 163 /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD 164 * "criss-cross" setup, that might cause write-out on some other DRBD, 165 * which in turn might block on the other node at this very place. */ 166 for (i = 0; i < number; i++) { 167 tmp = alloc_page(GFP_TRY); 168 if (!tmp) 169 break; 170 set_page_private(tmp, (unsigned long)page); 171 page = tmp; 172 } 173 174 if (i == number) 175 return page; 176 177 /* Not enough pages immediately available this time. 178 * No need to jump around here, drbd_pp_alloc will retry this 179 * function "soon". */ 180 if (page) { 181 tmp = page_chain_tail(page, NULL); 182 spin_lock(&drbd_pp_lock); 183 page_chain_add(&drbd_pp_pool, page, tmp); 184 drbd_pp_vacant += i; 185 spin_unlock(&drbd_pp_lock); 186 } 187 return NULL; 188 } 189 190 /* kick lower level device, if we have more than (arbitrary number) 191 * reference counts on it, which typically are locally submitted io 192 * requests. don't use unacked_cnt, so we speed up proto A and B, too. */ 193 static void maybe_kick_lo(struct drbd_conf *mdev) 194 { 195 if (atomic_read(&mdev->local_cnt) >= mdev->net_conf->unplug_watermark) 196 drbd_kick_lo(mdev); 197 } 198 199 static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed) 200 { 201 struct drbd_epoch_entry *e; 202 struct list_head *le, *tle; 203 204 /* The EEs are always appended to the end of the list. Since 205 they are sent in order over the wire, they have to finish 206 in order. As soon as we see the first not finished we can 207 stop to examine the list... */ 208 209 list_for_each_safe(le, tle, &mdev->net_ee) { 210 e = list_entry(le, struct drbd_epoch_entry, w.list); 211 if (drbd_ee_has_active_page(e)) 212 break; 213 list_move(le, to_be_freed); 214 } 215 } 216 217 static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev) 218 { 219 LIST_HEAD(reclaimed); 220 struct drbd_epoch_entry *e, *t; 221 222 maybe_kick_lo(mdev); 223 spin_lock_irq(&mdev->req_lock); 224 reclaim_net_ee(mdev, &reclaimed); 225 spin_unlock_irq(&mdev->req_lock); 226 227 list_for_each_entry_safe(e, t, &reclaimed, w.list) 228 drbd_free_net_ee(mdev, e); 229 } 230 231 /** 232 * drbd_pp_alloc() - Returns @number pages, retries forever (or until signalled) 233 * @mdev: DRBD device. 234 * @number: number of pages requested 235 * @retry: whether to retry, if not enough pages are available right now 236 * 237 * Tries to allocate number pages, first from our own page pool, then from 238 * the kernel, unless this allocation would exceed the max_buffers setting. 239 * Possibly retry until DRBD frees sufficient pages somewhere else. 240 * 241 * Returns a page chain linked via page->private. 242 */ 243 static struct page *drbd_pp_alloc(struct drbd_conf *mdev, unsigned number, bool retry) 244 { 245 struct page *page = NULL; 246 DEFINE_WAIT(wait); 247 248 /* Yes, we may run up to @number over max_buffers. If we 249 * follow it strictly, the admin will get it wrong anyways. */ 250 if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) 251 page = drbd_pp_first_pages_or_try_alloc(mdev, number); 252 253 while (page == NULL) { 254 prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); 255 256 drbd_kick_lo_and_reclaim_net(mdev); 257 258 if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) { 259 page = drbd_pp_first_pages_or_try_alloc(mdev, number); 260 if (page) 261 break; 262 } 263 264 if (!retry) 265 break; 266 267 if (signal_pending(current)) { 268 dev_warn(DEV, "drbd_pp_alloc interrupted!\n"); 269 break; 270 } 271 272 schedule(); 273 } 274 finish_wait(&drbd_pp_wait, &wait); 275 276 if (page) 277 atomic_add(number, &mdev->pp_in_use); 278 return page; 279 } 280 281 /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc. 282 * Is also used from inside an other spin_lock_irq(&mdev->req_lock); 283 * Either links the page chain back to the global pool, 284 * or returns all pages to the system. */ 285 static void drbd_pp_free(struct drbd_conf *mdev, struct page *page, int is_net) 286 { 287 atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use; 288 int i; 289 290 if (drbd_pp_vacant > (DRBD_MAX_SEGMENT_SIZE/PAGE_SIZE)*minor_count) 291 i = page_chain_free(page); 292 else { 293 struct page *tmp; 294 tmp = page_chain_tail(page, &i); 295 spin_lock(&drbd_pp_lock); 296 page_chain_add(&drbd_pp_pool, page, tmp); 297 drbd_pp_vacant += i; 298 spin_unlock(&drbd_pp_lock); 299 } 300 i = atomic_sub_return(i, a); 301 if (i < 0) 302 dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n", 303 is_net ? "pp_in_use_by_net" : "pp_in_use", i); 304 wake_up(&drbd_pp_wait); 305 } 306 307 /* 308 You need to hold the req_lock: 309 _drbd_wait_ee_list_empty() 310 311 You must not have the req_lock: 312 drbd_free_ee() 313 drbd_alloc_ee() 314 drbd_init_ee() 315 drbd_release_ee() 316 drbd_ee_fix_bhs() 317 drbd_process_done_ee() 318 drbd_clear_done_ee() 319 drbd_wait_ee_list_empty() 320 */ 321 322 struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev, 323 u64 id, 324 sector_t sector, 325 unsigned int data_size, 326 gfp_t gfp_mask) __must_hold(local) 327 { 328 struct drbd_epoch_entry *e; 329 struct page *page; 330 unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; 331 332 if (FAULT_ACTIVE(mdev, DRBD_FAULT_AL_EE)) 333 return NULL; 334 335 e = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); 336 if (!e) { 337 if (!(gfp_mask & __GFP_NOWARN)) 338 dev_err(DEV, "alloc_ee: Allocation of an EE failed\n"); 339 return NULL; 340 } 341 342 page = drbd_pp_alloc(mdev, nr_pages, (gfp_mask & __GFP_WAIT)); 343 if (!page) 344 goto fail; 345 346 INIT_HLIST_NODE(&e->colision); 347 e->epoch = NULL; 348 e->mdev = mdev; 349 e->pages = page; 350 atomic_set(&e->pending_bios, 0); 351 e->size = data_size; 352 e->flags = 0; 353 e->sector = sector; 354 e->block_id = id; 355 356 return e; 357 358 fail: 359 mempool_free(e, drbd_ee_mempool); 360 return NULL; 361 } 362 363 void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, int is_net) 364 { 365 if (e->flags & EE_HAS_DIGEST) 366 kfree(e->digest); 367 drbd_pp_free(mdev, e->pages, is_net); 368 D_ASSERT(atomic_read(&e->pending_bios) == 0); 369 D_ASSERT(hlist_unhashed(&e->colision)); 370 mempool_free(e, drbd_ee_mempool); 371 } 372 373 int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list) 374 { 375 LIST_HEAD(work_list); 376 struct drbd_epoch_entry *e, *t; 377 int count = 0; 378 int is_net = list == &mdev->net_ee; 379 380 spin_lock_irq(&mdev->req_lock); 381 list_splice_init(list, &work_list); 382 spin_unlock_irq(&mdev->req_lock); 383 384 list_for_each_entry_safe(e, t, &work_list, w.list) { 385 drbd_free_some_ee(mdev, e, is_net); 386 count++; 387 } 388 return count; 389 } 390 391 392 /* 393 * This function is called from _asender only_ 394 * but see also comments in _req_mod(,barrier_acked) 395 * and receive_Barrier. 396 * 397 * Move entries from net_ee to done_ee, if ready. 398 * Grab done_ee, call all callbacks, free the entries. 399 * The callbacks typically send out ACKs. 400 */ 401 static int drbd_process_done_ee(struct drbd_conf *mdev) 402 { 403 LIST_HEAD(work_list); 404 LIST_HEAD(reclaimed); 405 struct drbd_epoch_entry *e, *t; 406 int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS); 407 408 spin_lock_irq(&mdev->req_lock); 409 reclaim_net_ee(mdev, &reclaimed); 410 list_splice_init(&mdev->done_ee, &work_list); 411 spin_unlock_irq(&mdev->req_lock); 412 413 list_for_each_entry_safe(e, t, &reclaimed, w.list) 414 drbd_free_net_ee(mdev, e); 415 416 /* possible callbacks here: 417 * e_end_block, and e_end_resync_block, e_send_discard_ack. 418 * all ignore the last argument. 419 */ 420 list_for_each_entry_safe(e, t, &work_list, w.list) { 421 /* list_del not necessary, next/prev members not touched */ 422 ok = e->w.cb(mdev, &e->w, !ok) && ok; 423 drbd_free_ee(mdev, e); 424 } 425 wake_up(&mdev->ee_wait); 426 427 return ok; 428 } 429 430 void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) 431 { 432 DEFINE_WAIT(wait); 433 434 /* avoids spin_lock/unlock 435 * and calling prepare_to_wait in the fast path */ 436 while (!list_empty(head)) { 437 prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE); 438 spin_unlock_irq(&mdev->req_lock); 439 drbd_kick_lo(mdev); 440 schedule(); 441 finish_wait(&mdev->ee_wait, &wait); 442 spin_lock_irq(&mdev->req_lock); 443 } 444 } 445 446 void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) 447 { 448 spin_lock_irq(&mdev->req_lock); 449 _drbd_wait_ee_list_empty(mdev, head); 450 spin_unlock_irq(&mdev->req_lock); 451 } 452 453 /* see also kernel_accept; which is only present since 2.6.18. 454 * also we want to log which part of it failed, exactly */ 455 static int drbd_accept(struct drbd_conf *mdev, const char **what, 456 struct socket *sock, struct socket **newsock) 457 { 458 struct sock *sk = sock->sk; 459 int err = 0; 460 461 *what = "listen"; 462 err = sock->ops->listen(sock, 5); 463 if (err < 0) 464 goto out; 465 466 *what = "sock_create_lite"; 467 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol, 468 newsock); 469 if (err < 0) 470 goto out; 471 472 *what = "accept"; 473 err = sock->ops->accept(sock, *newsock, 0); 474 if (err < 0) { 475 sock_release(*newsock); 476 *newsock = NULL; 477 goto out; 478 } 479 (*newsock)->ops = sock->ops; 480 481 out: 482 return err; 483 } 484 485 static int drbd_recv_short(struct drbd_conf *mdev, struct socket *sock, 486 void *buf, size_t size, int flags) 487 { 488 mm_segment_t oldfs; 489 struct kvec iov = { 490 .iov_base = buf, 491 .iov_len = size, 492 }; 493 struct msghdr msg = { 494 .msg_iovlen = 1, 495 .msg_iov = (struct iovec *)&iov, 496 .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) 497 }; 498 int rv; 499 500 oldfs = get_fs(); 501 set_fs(KERNEL_DS); 502 rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); 503 set_fs(oldfs); 504 505 return rv; 506 } 507 508 static int drbd_recv(struct drbd_conf *mdev, void *buf, size_t size) 509 { 510 mm_segment_t oldfs; 511 struct kvec iov = { 512 .iov_base = buf, 513 .iov_len = size, 514 }; 515 struct msghdr msg = { 516 .msg_iovlen = 1, 517 .msg_iov = (struct iovec *)&iov, 518 .msg_flags = MSG_WAITALL | MSG_NOSIGNAL 519 }; 520 int rv; 521 522 oldfs = get_fs(); 523 set_fs(KERNEL_DS); 524 525 for (;;) { 526 rv = sock_recvmsg(mdev->data.socket, &msg, size, msg.msg_flags); 527 if (rv == size) 528 break; 529 530 /* Note: 531 * ECONNRESET other side closed the connection 532 * ERESTARTSYS (on sock) we got a signal 533 */ 534 535 if (rv < 0) { 536 if (rv == -ECONNRESET) 537 dev_info(DEV, "sock was reset by peer\n"); 538 else if (rv != -ERESTARTSYS) 539 dev_err(DEV, "sock_recvmsg returned %d\n", rv); 540 break; 541 } else if (rv == 0) { 542 dev_info(DEV, "sock was shut down by peer\n"); 543 break; 544 } else { 545 /* signal came in, or peer/link went down, 546 * after we read a partial message 547 */ 548 /* D_ASSERT(signal_pending(current)); */ 549 break; 550 } 551 }; 552 553 set_fs(oldfs); 554 555 if (rv != size) 556 drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE)); 557 558 return rv; 559 } 560 561 /* quoting tcp(7): 562 * On individual connections, the socket buffer size must be set prior to the 563 * listen(2) or connect(2) calls in order to have it take effect. 564 * This is our wrapper to do so. 565 */ 566 static void drbd_setbufsize(struct socket *sock, unsigned int snd, 567 unsigned int rcv) 568 { 569 /* open coded SO_SNDBUF, SO_RCVBUF */ 570 if (snd) { 571 sock->sk->sk_sndbuf = snd; 572 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 573 } 574 if (rcv) { 575 sock->sk->sk_rcvbuf = rcv; 576 sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 577 } 578 } 579 580 static struct socket *drbd_try_connect(struct drbd_conf *mdev) 581 { 582 const char *what; 583 struct socket *sock; 584 struct sockaddr_in6 src_in6; 585 int err; 586 int disconnect_on_error = 1; 587 588 if (!get_net_conf(mdev)) 589 return NULL; 590 591 what = "sock_create_kern"; 592 err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, 593 SOCK_STREAM, IPPROTO_TCP, &sock); 594 if (err < 0) { 595 sock = NULL; 596 goto out; 597 } 598 599 sock->sk->sk_rcvtimeo = 600 sock->sk->sk_sndtimeo = mdev->net_conf->try_connect_int*HZ; 601 drbd_setbufsize(sock, mdev->net_conf->sndbuf_size, 602 mdev->net_conf->rcvbuf_size); 603 604 /* explicitly bind to the configured IP as source IP 605 * for the outgoing connections. 606 * This is needed for multihomed hosts and to be 607 * able to use lo: interfaces for drbd. 608 * Make sure to use 0 as port number, so linux selects 609 * a free one dynamically. 610 */ 611 memcpy(&src_in6, mdev->net_conf->my_addr, 612 min_t(int, mdev->net_conf->my_addr_len, sizeof(src_in6))); 613 if (((struct sockaddr *)mdev->net_conf->my_addr)->sa_family == AF_INET6) 614 src_in6.sin6_port = 0; 615 else 616 ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ 617 618 what = "bind before connect"; 619 err = sock->ops->bind(sock, 620 (struct sockaddr *) &src_in6, 621 mdev->net_conf->my_addr_len); 622 if (err < 0) 623 goto out; 624 625 /* connect may fail, peer not yet available. 626 * stay C_WF_CONNECTION, don't go Disconnecting! */ 627 disconnect_on_error = 0; 628 what = "connect"; 629 err = sock->ops->connect(sock, 630 (struct sockaddr *)mdev->net_conf->peer_addr, 631 mdev->net_conf->peer_addr_len, 0); 632 633 out: 634 if (err < 0) { 635 if (sock) { 636 sock_release(sock); 637 sock = NULL; 638 } 639 switch (-err) { 640 /* timeout, busy, signal pending */ 641 case ETIMEDOUT: case EAGAIN: case EINPROGRESS: 642 case EINTR: case ERESTARTSYS: 643 /* peer not (yet) available, network problem */ 644 case ECONNREFUSED: case ENETUNREACH: 645 case EHOSTDOWN: case EHOSTUNREACH: 646 disconnect_on_error = 0; 647 break; 648 default: 649 dev_err(DEV, "%s failed, err = %d\n", what, err); 650 } 651 if (disconnect_on_error) 652 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 653 } 654 put_net_conf(mdev); 655 return sock; 656 } 657 658 static struct socket *drbd_wait_for_connect(struct drbd_conf *mdev) 659 { 660 int timeo, err; 661 struct socket *s_estab = NULL, *s_listen; 662 const char *what; 663 664 if (!get_net_conf(mdev)) 665 return NULL; 666 667 what = "sock_create_kern"; 668 err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, 669 SOCK_STREAM, IPPROTO_TCP, &s_listen); 670 if (err) { 671 s_listen = NULL; 672 goto out; 673 } 674 675 timeo = mdev->net_conf->try_connect_int * HZ; 676 timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */ 677 678 s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */ 679 s_listen->sk->sk_rcvtimeo = timeo; 680 s_listen->sk->sk_sndtimeo = timeo; 681 drbd_setbufsize(s_listen, mdev->net_conf->sndbuf_size, 682 mdev->net_conf->rcvbuf_size); 683 684 what = "bind before listen"; 685 err = s_listen->ops->bind(s_listen, 686 (struct sockaddr *) mdev->net_conf->my_addr, 687 mdev->net_conf->my_addr_len); 688 if (err < 0) 689 goto out; 690 691 err = drbd_accept(mdev, &what, s_listen, &s_estab); 692 693 out: 694 if (s_listen) 695 sock_release(s_listen); 696 if (err < 0) { 697 if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { 698 dev_err(DEV, "%s failed, err = %d\n", what, err); 699 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 700 } 701 } 702 put_net_conf(mdev); 703 704 return s_estab; 705 } 706 707 static int drbd_send_fp(struct drbd_conf *mdev, 708 struct socket *sock, enum drbd_packets cmd) 709 { 710 struct p_header80 *h = &mdev->data.sbuf.header.h80; 711 712 return _drbd_send_cmd(mdev, sock, cmd, h, sizeof(*h), 0); 713 } 714 715 static enum drbd_packets drbd_recv_fp(struct drbd_conf *mdev, struct socket *sock) 716 { 717 struct p_header80 *h = &mdev->data.rbuf.header.h80; 718 int rr; 719 720 rr = drbd_recv_short(mdev, sock, h, sizeof(*h), 0); 721 722 if (rr == sizeof(*h) && h->magic == BE_DRBD_MAGIC) 723 return be16_to_cpu(h->command); 724 725 return 0xffff; 726 } 727 728 /** 729 * drbd_socket_okay() - Free the socket if its connection is not okay 730 * @mdev: DRBD device. 731 * @sock: pointer to the pointer to the socket. 732 */ 733 static int drbd_socket_okay(struct drbd_conf *mdev, struct socket **sock) 734 { 735 int rr; 736 char tb[4]; 737 738 if (!*sock) 739 return FALSE; 740 741 rr = drbd_recv_short(mdev, *sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); 742 743 if (rr > 0 || rr == -EAGAIN) { 744 return TRUE; 745 } else { 746 sock_release(*sock); 747 *sock = NULL; 748 return FALSE; 749 } 750 } 751 752 /* 753 * return values: 754 * 1 yes, we have a valid connection 755 * 0 oops, did not work out, please try again 756 * -1 peer talks different language, 757 * no point in trying again, please go standalone. 758 * -2 We do not have a network config... 759 */ 760 static int drbd_connect(struct drbd_conf *mdev) 761 { 762 struct socket *s, *sock, *msock; 763 int try, h, ok; 764 765 D_ASSERT(!mdev->data.socket); 766 767 if (drbd_request_state(mdev, NS(conn, C_WF_CONNECTION)) < SS_SUCCESS) 768 return -2; 769 770 clear_bit(DISCARD_CONCURRENT, &mdev->flags); 771 772 sock = NULL; 773 msock = NULL; 774 775 do { 776 for (try = 0;;) { 777 /* 3 tries, this should take less than a second! */ 778 s = drbd_try_connect(mdev); 779 if (s || ++try >= 3) 780 break; 781 /* give the other side time to call bind() & listen() */ 782 __set_current_state(TASK_INTERRUPTIBLE); 783 schedule_timeout(HZ / 10); 784 } 785 786 if (s) { 787 if (!sock) { 788 drbd_send_fp(mdev, s, P_HAND_SHAKE_S); 789 sock = s; 790 s = NULL; 791 } else if (!msock) { 792 drbd_send_fp(mdev, s, P_HAND_SHAKE_M); 793 msock = s; 794 s = NULL; 795 } else { 796 dev_err(DEV, "Logic error in drbd_connect()\n"); 797 goto out_release_sockets; 798 } 799 } 800 801 if (sock && msock) { 802 __set_current_state(TASK_INTERRUPTIBLE); 803 schedule_timeout(HZ / 10); 804 ok = drbd_socket_okay(mdev, &sock); 805 ok = drbd_socket_okay(mdev, &msock) && ok; 806 if (ok) 807 break; 808 } 809 810 retry: 811 s = drbd_wait_for_connect(mdev); 812 if (s) { 813 try = drbd_recv_fp(mdev, s); 814 drbd_socket_okay(mdev, &sock); 815 drbd_socket_okay(mdev, &msock); 816 switch (try) { 817 case P_HAND_SHAKE_S: 818 if (sock) { 819 dev_warn(DEV, "initial packet S crossed\n"); 820 sock_release(sock); 821 } 822 sock = s; 823 break; 824 case P_HAND_SHAKE_M: 825 if (msock) { 826 dev_warn(DEV, "initial packet M crossed\n"); 827 sock_release(msock); 828 } 829 msock = s; 830 set_bit(DISCARD_CONCURRENT, &mdev->flags); 831 break; 832 default: 833 dev_warn(DEV, "Error receiving initial packet\n"); 834 sock_release(s); 835 if (random32() & 1) 836 goto retry; 837 } 838 } 839 840 if (mdev->state.conn <= C_DISCONNECTING) 841 goto out_release_sockets; 842 if (signal_pending(current)) { 843 flush_signals(current); 844 smp_rmb(); 845 if (get_t_state(&mdev->receiver) == Exiting) 846 goto out_release_sockets; 847 } 848 849 if (sock && msock) { 850 ok = drbd_socket_okay(mdev, &sock); 851 ok = drbd_socket_okay(mdev, &msock) && ok; 852 if (ok) 853 break; 854 } 855 } while (1); 856 857 msock->sk->sk_reuse = 1; /* SO_REUSEADDR */ 858 sock->sk->sk_reuse = 1; /* SO_REUSEADDR */ 859 860 sock->sk->sk_allocation = GFP_NOIO; 861 msock->sk->sk_allocation = GFP_NOIO; 862 863 sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; 864 msock->sk->sk_priority = TC_PRIO_INTERACTIVE; 865 866 /* NOT YET ... 867 * sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; 868 * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 869 * first set it to the P_HAND_SHAKE timeout, 870 * which we set to 4x the configured ping_timeout. */ 871 sock->sk->sk_sndtimeo = 872 sock->sk->sk_rcvtimeo = mdev->net_conf->ping_timeo*4*HZ/10; 873 874 msock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; 875 msock->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ; 876 877 /* we don't want delays. 878 * we use TCP_CORK where apropriate, though */ 879 drbd_tcp_nodelay(sock); 880 drbd_tcp_nodelay(msock); 881 882 mdev->data.socket = sock; 883 mdev->meta.socket = msock; 884 mdev->last_received = jiffies; 885 886 D_ASSERT(mdev->asender.task == NULL); 887 888 h = drbd_do_handshake(mdev); 889 if (h <= 0) 890 return h; 891 892 if (mdev->cram_hmac_tfm) { 893 /* drbd_request_state(mdev, NS(conn, WFAuth)); */ 894 switch (drbd_do_auth(mdev)) { 895 case -1: 896 dev_err(DEV, "Authentication of peer failed\n"); 897 return -1; 898 case 0: 899 dev_err(DEV, "Authentication of peer failed, trying again.\n"); 900 return 0; 901 } 902 } 903 904 if (drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS)) < SS_SUCCESS) 905 return 0; 906 907 sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; 908 sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 909 910 atomic_set(&mdev->packet_seq, 0); 911 mdev->peer_seq = 0; 912 913 drbd_thread_start(&mdev->asender); 914 915 if (mdev->agreed_pro_version < 95 && get_ldev(mdev)) { 916 drbd_setup_queue_param(mdev, DRBD_MAX_SIZE_H80_PACKET); 917 put_ldev(mdev); 918 } 919 920 if (!drbd_send_protocol(mdev)) 921 return -1; 922 drbd_send_sync_param(mdev, &mdev->sync_conf); 923 drbd_send_sizes(mdev, 0, 0); 924 drbd_send_uuids(mdev); 925 drbd_send_state(mdev); 926 clear_bit(USE_DEGR_WFC_T, &mdev->flags); 927 clear_bit(RESIZE_PENDING, &mdev->flags); 928 929 return 1; 930 931 out_release_sockets: 932 if (sock) 933 sock_release(sock); 934 if (msock) 935 sock_release(msock); 936 return -1; 937 } 938 939 static int drbd_recv_header(struct drbd_conf *mdev, enum drbd_packets *cmd, unsigned int *packet_size) 940 { 941 union p_header *h = &mdev->data.rbuf.header; 942 int r; 943 944 r = drbd_recv(mdev, h, sizeof(*h)); 945 if (unlikely(r != sizeof(*h))) { 946 dev_err(DEV, "short read expecting header on sock: r=%d\n", r); 947 return FALSE; 948 } 949 950 if (likely(h->h80.magic == BE_DRBD_MAGIC)) { 951 *cmd = be16_to_cpu(h->h80.command); 952 *packet_size = be16_to_cpu(h->h80.length); 953 } else if (h->h95.magic == BE_DRBD_MAGIC_BIG) { 954 *cmd = be16_to_cpu(h->h95.command); 955 *packet_size = be32_to_cpu(h->h95.length); 956 } else { 957 dev_err(DEV, "magic?? on data m: 0x%08x c: %d l: %d\n", 958 be32_to_cpu(h->h80.magic), 959 be16_to_cpu(h->h80.command), 960 be16_to_cpu(h->h80.length)); 961 return FALSE; 962 } 963 mdev->last_received = jiffies; 964 965 return TRUE; 966 } 967 968 static void drbd_flush(struct drbd_conf *mdev) 969 { 970 int rv; 971 972 if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) { 973 rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL, 974 NULL); 975 if (rv) { 976 dev_err(DEV, "local disk flush failed with status %d\n", rv); 977 /* would rather check on EOPNOTSUPP, but that is not reliable. 978 * don't try again for ANY return value != 0 979 * if (rv == -EOPNOTSUPP) */ 980 drbd_bump_write_ordering(mdev, WO_drain_io); 981 } 982 put_ldev(mdev); 983 } 984 } 985 986 /** 987 * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. 988 * @mdev: DRBD device. 989 * @epoch: Epoch object. 990 * @ev: Epoch event. 991 */ 992 static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev, 993 struct drbd_epoch *epoch, 994 enum epoch_event ev) 995 { 996 int epoch_size; 997 struct drbd_epoch *next_epoch; 998 enum finish_epoch rv = FE_STILL_LIVE; 999 1000 spin_lock(&mdev->epoch_lock); 1001 do { 1002 next_epoch = NULL; 1003 1004 epoch_size = atomic_read(&epoch->epoch_size); 1005 1006 switch (ev & ~EV_CLEANUP) { 1007 case EV_PUT: 1008 atomic_dec(&epoch->active); 1009 break; 1010 case EV_GOT_BARRIER_NR: 1011 set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); 1012 break; 1013 case EV_BECAME_LAST: 1014 /* nothing to do*/ 1015 break; 1016 } 1017 1018 if (epoch_size != 0 && 1019 atomic_read(&epoch->active) == 0 && 1020 test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) { 1021 if (!(ev & EV_CLEANUP)) { 1022 spin_unlock(&mdev->epoch_lock); 1023 drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size); 1024 spin_lock(&mdev->epoch_lock); 1025 } 1026 dec_unacked(mdev); 1027 1028 if (mdev->current_epoch != epoch) { 1029 next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); 1030 list_del(&epoch->list); 1031 ev = EV_BECAME_LAST | (ev & EV_CLEANUP); 1032 mdev->epochs--; 1033 kfree(epoch); 1034 1035 if (rv == FE_STILL_LIVE) 1036 rv = FE_DESTROYED; 1037 } else { 1038 epoch->flags = 0; 1039 atomic_set(&epoch->epoch_size, 0); 1040 /* atomic_set(&epoch->active, 0); is already zero */ 1041 if (rv == FE_STILL_LIVE) 1042 rv = FE_RECYCLED; 1043 wake_up(&mdev->ee_wait); 1044 } 1045 } 1046 1047 if (!next_epoch) 1048 break; 1049 1050 epoch = next_epoch; 1051 } while (1); 1052 1053 spin_unlock(&mdev->epoch_lock); 1054 1055 return rv; 1056 } 1057 1058 /** 1059 * drbd_bump_write_ordering() - Fall back to an other write ordering method 1060 * @mdev: DRBD device. 1061 * @wo: Write ordering method to try. 1062 */ 1063 void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local) 1064 { 1065 enum write_ordering_e pwo; 1066 static char *write_ordering_str[] = { 1067 [WO_none] = "none", 1068 [WO_drain_io] = "drain", 1069 [WO_bdev_flush] = "flush", 1070 }; 1071 1072 pwo = mdev->write_ordering; 1073 wo = min(pwo, wo); 1074 if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush) 1075 wo = WO_drain_io; 1076 if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain) 1077 wo = WO_none; 1078 mdev->write_ordering = wo; 1079 if (pwo != mdev->write_ordering || wo == WO_bdev_flush) 1080 dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]); 1081 } 1082 1083 /** 1084 * drbd_submit_ee() 1085 * @mdev: DRBD device. 1086 * @e: epoch entry 1087 * @rw: flag field, see bio->bi_rw 1088 */ 1089 /* TODO allocate from our own bio_set. */ 1090 int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, 1091 const unsigned rw, const int fault_type) 1092 { 1093 struct bio *bios = NULL; 1094 struct bio *bio; 1095 struct page *page = e->pages; 1096 sector_t sector = e->sector; 1097 unsigned ds = e->size; 1098 unsigned n_bios = 0; 1099 unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; 1100 1101 /* In most cases, we will only need one bio. But in case the lower 1102 * level restrictions happen to be different at this offset on this 1103 * side than those of the sending peer, we may need to submit the 1104 * request in more than one bio. */ 1105 next_bio: 1106 bio = bio_alloc(GFP_NOIO, nr_pages); 1107 if (!bio) { 1108 dev_err(DEV, "submit_ee: Allocation of a bio failed\n"); 1109 goto fail; 1110 } 1111 /* > e->sector, unless this is the first bio */ 1112 bio->bi_sector = sector; 1113 bio->bi_bdev = mdev->ldev->backing_bdev; 1114 /* we special case some flags in the multi-bio case, see below 1115 * (REQ_UNPLUG) */ 1116 bio->bi_rw = rw; 1117 bio->bi_private = e; 1118 bio->bi_end_io = drbd_endio_sec; 1119 1120 bio->bi_next = bios; 1121 bios = bio; 1122 ++n_bios; 1123 1124 page_chain_for_each(page) { 1125 unsigned len = min_t(unsigned, ds, PAGE_SIZE); 1126 if (!bio_add_page(bio, page, len, 0)) { 1127 /* a single page must always be possible! */ 1128 BUG_ON(bio->bi_vcnt == 0); 1129 goto next_bio; 1130 } 1131 ds -= len; 1132 sector += len >> 9; 1133 --nr_pages; 1134 } 1135 D_ASSERT(page == NULL); 1136 D_ASSERT(ds == 0); 1137 1138 atomic_set(&e->pending_bios, n_bios); 1139 do { 1140 bio = bios; 1141 bios = bios->bi_next; 1142 bio->bi_next = NULL; 1143 1144 /* strip off REQ_UNPLUG unless it is the last bio */ 1145 if (bios) 1146 bio->bi_rw &= ~REQ_UNPLUG; 1147 1148 drbd_generic_make_request(mdev, fault_type, bio); 1149 } while (bios); 1150 maybe_kick_lo(mdev); 1151 return 0; 1152 1153 fail: 1154 while (bios) { 1155 bio = bios; 1156 bios = bios->bi_next; 1157 bio_put(bio); 1158 } 1159 return -ENOMEM; 1160 } 1161 1162 static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 1163 { 1164 int rv; 1165 struct p_barrier *p = &mdev->data.rbuf.barrier; 1166 struct drbd_epoch *epoch; 1167 1168 inc_unacked(mdev); 1169 1170 if (mdev->net_conf->wire_protocol != DRBD_PROT_C) 1171 drbd_kick_lo(mdev); 1172 1173 mdev->current_epoch->barrier_nr = p->barrier; 1174 rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR); 1175 1176 /* P_BARRIER_ACK may imply that the corresponding extent is dropped from 1177 * the activity log, which means it would not be resynced in case the 1178 * R_PRIMARY crashes now. 1179 * Therefore we must send the barrier_ack after the barrier request was 1180 * completed. */ 1181 switch (mdev->write_ordering) { 1182 case WO_none: 1183 if (rv == FE_RECYCLED) 1184 return TRUE; 1185 1186 /* receiver context, in the writeout path of the other node. 1187 * avoid potential distributed deadlock */ 1188 epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); 1189 if (epoch) 1190 break; 1191 else 1192 dev_warn(DEV, "Allocation of an epoch failed, slowing down\n"); 1193 /* Fall through */ 1194 1195 case WO_bdev_flush: 1196 case WO_drain_io: 1197 drbd_wait_ee_list_empty(mdev, &mdev->active_ee); 1198 drbd_flush(mdev); 1199 1200 if (atomic_read(&mdev->current_epoch->epoch_size)) { 1201 epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); 1202 if (epoch) 1203 break; 1204 } 1205 1206 epoch = mdev->current_epoch; 1207 wait_event(mdev->ee_wait, atomic_read(&epoch->epoch_size) == 0); 1208 1209 D_ASSERT(atomic_read(&epoch->active) == 0); 1210 D_ASSERT(epoch->flags == 0); 1211 1212 return TRUE; 1213 default: 1214 dev_err(DEV, "Strangeness in mdev->write_ordering %d\n", mdev->write_ordering); 1215 return FALSE; 1216 } 1217 1218 epoch->flags = 0; 1219 atomic_set(&epoch->epoch_size, 0); 1220 atomic_set(&epoch->active, 0); 1221 1222 spin_lock(&mdev->epoch_lock); 1223 if (atomic_read(&mdev->current_epoch->epoch_size)) { 1224 list_add(&epoch->list, &mdev->current_epoch->list); 1225 mdev->current_epoch = epoch; 1226 mdev->epochs++; 1227 } else { 1228 /* The current_epoch got recycled while we allocated this one... */ 1229 kfree(epoch); 1230 } 1231 spin_unlock(&mdev->epoch_lock); 1232 1233 return TRUE; 1234 } 1235 1236 /* used from receive_RSDataReply (recv_resync_read) 1237 * and from receive_Data */ 1238 static struct drbd_epoch_entry * 1239 read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, int data_size) __must_hold(local) 1240 { 1241 const sector_t capacity = drbd_get_capacity(mdev->this_bdev); 1242 struct drbd_epoch_entry *e; 1243 struct page *page; 1244 int dgs, ds, rr; 1245 void *dig_in = mdev->int_dig_in; 1246 void *dig_vv = mdev->int_dig_vv; 1247 unsigned long *data; 1248 1249 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? 1250 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; 1251 1252 if (dgs) { 1253 rr = drbd_recv(mdev, dig_in, dgs); 1254 if (rr != dgs) { 1255 dev_warn(DEV, "short read receiving data digest: read %d expected %d\n", 1256 rr, dgs); 1257 return NULL; 1258 } 1259 } 1260 1261 data_size -= dgs; 1262 1263 ERR_IF(data_size & 0x1ff) return NULL; 1264 ERR_IF(data_size > DRBD_MAX_SEGMENT_SIZE) return NULL; 1265 1266 /* even though we trust out peer, 1267 * we sometimes have to double check. */ 1268 if (sector + (data_size>>9) > capacity) { 1269 dev_err(DEV, "capacity: %llus < sector: %llus + size: %u\n", 1270 (unsigned long long)capacity, 1271 (unsigned long long)sector, data_size); 1272 return NULL; 1273 } 1274 1275 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD 1276 * "criss-cross" setup, that might cause write-out on some other DRBD, 1277 * which in turn might block on the other node at this very place. */ 1278 e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO); 1279 if (!e) 1280 return NULL; 1281 1282 ds = data_size; 1283 page = e->pages; 1284 page_chain_for_each(page) { 1285 unsigned len = min_t(int, ds, PAGE_SIZE); 1286 data = kmap(page); 1287 rr = drbd_recv(mdev, data, len); 1288 if (FAULT_ACTIVE(mdev, DRBD_FAULT_RECEIVE)) { 1289 dev_err(DEV, "Fault injection: Corrupting data on receive\n"); 1290 data[0] = data[0] ^ (unsigned long)-1; 1291 } 1292 kunmap(page); 1293 if (rr != len) { 1294 drbd_free_ee(mdev, e); 1295 dev_warn(DEV, "short read receiving data: read %d expected %d\n", 1296 rr, len); 1297 return NULL; 1298 } 1299 ds -= rr; 1300 } 1301 1302 if (dgs) { 1303 drbd_csum_ee(mdev, mdev->integrity_r_tfm, e, dig_vv); 1304 if (memcmp(dig_in, dig_vv, dgs)) { 1305 dev_err(DEV, "Digest integrity check FAILED.\n"); 1306 drbd_bcast_ee(mdev, "digest failed", 1307 dgs, dig_in, dig_vv, e); 1308 drbd_free_ee(mdev, e); 1309 return NULL; 1310 } 1311 } 1312 mdev->recv_cnt += data_size>>9; 1313 return e; 1314 } 1315 1316 /* drbd_drain_block() just takes a data block 1317 * out of the socket input buffer, and discards it. 1318 */ 1319 static int drbd_drain_block(struct drbd_conf *mdev, int data_size) 1320 { 1321 struct page *page; 1322 int rr, rv = 1; 1323 void *data; 1324 1325 if (!data_size) 1326 return TRUE; 1327 1328 page = drbd_pp_alloc(mdev, 1, 1); 1329 1330 data = kmap(page); 1331 while (data_size) { 1332 rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE)); 1333 if (rr != min_t(int, data_size, PAGE_SIZE)) { 1334 rv = 0; 1335 dev_warn(DEV, "short read receiving data: read %d expected %d\n", 1336 rr, min_t(int, data_size, PAGE_SIZE)); 1337 break; 1338 } 1339 data_size -= rr; 1340 } 1341 kunmap(page); 1342 drbd_pp_free(mdev, page, 0); 1343 return rv; 1344 } 1345 1346 static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, 1347 sector_t sector, int data_size) 1348 { 1349 struct bio_vec *bvec; 1350 struct bio *bio; 1351 int dgs, rr, i, expect; 1352 void *dig_in = mdev->int_dig_in; 1353 void *dig_vv = mdev->int_dig_vv; 1354 1355 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? 1356 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; 1357 1358 if (dgs) { 1359 rr = drbd_recv(mdev, dig_in, dgs); 1360 if (rr != dgs) { 1361 dev_warn(DEV, "short read receiving data reply digest: read %d expected %d\n", 1362 rr, dgs); 1363 return 0; 1364 } 1365 } 1366 1367 data_size -= dgs; 1368 1369 /* optimistically update recv_cnt. if receiving fails below, 1370 * we disconnect anyways, and counters will be reset. */ 1371 mdev->recv_cnt += data_size>>9; 1372 1373 bio = req->master_bio; 1374 D_ASSERT(sector == bio->bi_sector); 1375 1376 bio_for_each_segment(bvec, bio, i) { 1377 expect = min_t(int, data_size, bvec->bv_len); 1378 rr = drbd_recv(mdev, 1379 kmap(bvec->bv_page)+bvec->bv_offset, 1380 expect); 1381 kunmap(bvec->bv_page); 1382 if (rr != expect) { 1383 dev_warn(DEV, "short read receiving data reply: " 1384 "read %d expected %d\n", 1385 rr, expect); 1386 return 0; 1387 } 1388 data_size -= rr; 1389 } 1390 1391 if (dgs) { 1392 drbd_csum_bio(mdev, mdev->integrity_r_tfm, bio, dig_vv); 1393 if (memcmp(dig_in, dig_vv, dgs)) { 1394 dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); 1395 return 0; 1396 } 1397 } 1398 1399 D_ASSERT(data_size == 0); 1400 return 1; 1401 } 1402 1403 /* e_end_resync_block() is called via 1404 * drbd_process_done_ee() by asender only */ 1405 static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused) 1406 { 1407 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; 1408 sector_t sector = e->sector; 1409 int ok; 1410 1411 D_ASSERT(hlist_unhashed(&e->colision)); 1412 1413 if (likely((e->flags & EE_WAS_ERROR) == 0)) { 1414 drbd_set_in_sync(mdev, sector, e->size); 1415 ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e); 1416 } else { 1417 /* Record failure to sync */ 1418 drbd_rs_failed_io(mdev, sector, e->size); 1419 1420 ok = drbd_send_ack(mdev, P_NEG_ACK, e); 1421 } 1422 dec_unacked(mdev); 1423 1424 return ok; 1425 } 1426 1427 static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) 1428 { 1429 struct drbd_epoch_entry *e; 1430 1431 e = read_in_block(mdev, ID_SYNCER, sector, data_size); 1432 if (!e) 1433 goto fail; 1434 1435 dec_rs_pending(mdev); 1436 1437 inc_unacked(mdev); 1438 /* corresponding dec_unacked() in e_end_resync_block() 1439 * respective _drbd_clear_done_ee */ 1440 1441 e->w.cb = e_end_resync_block; 1442 1443 spin_lock_irq(&mdev->req_lock); 1444 list_add(&e->w.list, &mdev->sync_ee); 1445 spin_unlock_irq(&mdev->req_lock); 1446 1447 atomic_add(data_size >> 9, &mdev->rs_sect_ev); 1448 if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_RS_WR) == 0) 1449 return TRUE; 1450 1451 /* drbd_submit_ee currently fails for one reason only: 1452 * not being able to allocate enough bios. 1453 * Is dropping the connection going to help? */ 1454 spin_lock_irq(&mdev->req_lock); 1455 list_del(&e->w.list); 1456 spin_unlock_irq(&mdev->req_lock); 1457 1458 drbd_free_ee(mdev, e); 1459 fail: 1460 put_ldev(mdev); 1461 return FALSE; 1462 } 1463 1464 static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 1465 { 1466 struct drbd_request *req; 1467 sector_t sector; 1468 int ok; 1469 struct p_data *p = &mdev->data.rbuf.data; 1470 1471 sector = be64_to_cpu(p->sector); 1472 1473 spin_lock_irq(&mdev->req_lock); 1474 req = _ar_id_to_req(mdev, p->block_id, sector); 1475 spin_unlock_irq(&mdev->req_lock); 1476 if (unlikely(!req)) { 1477 dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n"); 1478 return FALSE; 1479 } 1480 1481 /* hlist_del(&req->colision) is done in _req_may_be_done, to avoid 1482 * special casing it there for the various failure cases. 1483 * still no race with drbd_fail_pending_reads */ 1484 ok = recv_dless_read(mdev, req, sector, data_size); 1485 1486 if (ok) 1487 req_mod(req, data_received); 1488 /* else: nothing. handled from drbd_disconnect... 1489 * I don't think we may complete this just yet 1490 * in case we are "on-disconnect: freeze" */ 1491 1492 return ok; 1493 } 1494 1495 static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 1496 { 1497 sector_t sector; 1498 int ok; 1499 struct p_data *p = &mdev->data.rbuf.data; 1500 1501 sector = be64_to_cpu(p->sector); 1502 D_ASSERT(p->block_id == ID_SYNCER); 1503 1504 if (get_ldev(mdev)) { 1505 /* data is submitted to disk within recv_resync_read. 1506 * corresponding put_ldev done below on error, 1507 * or in drbd_endio_write_sec. */ 1508 ok = recv_resync_read(mdev, sector, data_size); 1509 } else { 1510 if (__ratelimit(&drbd_ratelimit_state)) 1511 dev_err(DEV, "Can not write resync data to local disk.\n"); 1512 1513 ok = drbd_drain_block(mdev, data_size); 1514 1515 drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); 1516 } 1517 1518 atomic_add(data_size >> 9, &mdev->rs_sect_in); 1519 1520 return ok; 1521 } 1522 1523 /* e_end_block() is called via drbd_process_done_ee(). 1524 * this means this function only runs in the asender thread 1525 */ 1526 static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1527 { 1528 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; 1529 sector_t sector = e->sector; 1530 int ok = 1, pcmd; 1531 1532 if (mdev->net_conf->wire_protocol == DRBD_PROT_C) { 1533 if (likely((e->flags & EE_WAS_ERROR) == 0)) { 1534 pcmd = (mdev->state.conn >= C_SYNC_SOURCE && 1535 mdev->state.conn <= C_PAUSED_SYNC_T && 1536 e->flags & EE_MAY_SET_IN_SYNC) ? 1537 P_RS_WRITE_ACK : P_WRITE_ACK; 1538 ok &= drbd_send_ack(mdev, pcmd, e); 1539 if (pcmd == P_RS_WRITE_ACK) 1540 drbd_set_in_sync(mdev, sector, e->size); 1541 } else { 1542 ok = drbd_send_ack(mdev, P_NEG_ACK, e); 1543 /* we expect it to be marked out of sync anyways... 1544 * maybe assert this? */ 1545 } 1546 dec_unacked(mdev); 1547 } 1548 /* we delete from the conflict detection hash _after_ we sent out the 1549 * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ 1550 if (mdev->net_conf->two_primaries) { 1551 spin_lock_irq(&mdev->req_lock); 1552 D_ASSERT(!hlist_unhashed(&e->colision)); 1553 hlist_del_init(&e->colision); 1554 spin_unlock_irq(&mdev->req_lock); 1555 } else { 1556 D_ASSERT(hlist_unhashed(&e->colision)); 1557 } 1558 1559 drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); 1560 1561 return ok; 1562 } 1563 1564 static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused) 1565 { 1566 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; 1567 int ok = 1; 1568 1569 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 1570 ok = drbd_send_ack(mdev, P_DISCARD_ACK, e); 1571 1572 spin_lock_irq(&mdev->req_lock); 1573 D_ASSERT(!hlist_unhashed(&e->colision)); 1574 hlist_del_init(&e->colision); 1575 spin_unlock_irq(&mdev->req_lock); 1576 1577 dec_unacked(mdev); 1578 1579 return ok; 1580 } 1581 1582 /* Called from receive_Data. 1583 * Synchronize packets on sock with packets on msock. 1584 * 1585 * This is here so even when a P_DATA packet traveling via sock overtook an Ack 1586 * packet traveling on msock, they are still processed in the order they have 1587 * been sent. 1588 * 1589 * Note: we don't care for Ack packets overtaking P_DATA packets. 1590 * 1591 * In case packet_seq is larger than mdev->peer_seq number, there are 1592 * outstanding packets on the msock. We wait for them to arrive. 1593 * In case we are the logically next packet, we update mdev->peer_seq 1594 * ourselves. Correctly handles 32bit wrap around. 1595 * 1596 * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, 1597 * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds 1598 * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have 1599 * 1<<9 == 512 seconds aka ages for the 32bit wrap around... 1600 * 1601 * returns 0 if we may process the packet, 1602 * -ERESTARTSYS if we were interrupted (by disconnect signal). */ 1603 static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq) 1604 { 1605 DEFINE_WAIT(wait); 1606 unsigned int p_seq; 1607 long timeout; 1608 int ret = 0; 1609 spin_lock(&mdev->peer_seq_lock); 1610 for (;;) { 1611 prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); 1612 if (seq_le(packet_seq, mdev->peer_seq+1)) 1613 break; 1614 if (signal_pending(current)) { 1615 ret = -ERESTARTSYS; 1616 break; 1617 } 1618 p_seq = mdev->peer_seq; 1619 spin_unlock(&mdev->peer_seq_lock); 1620 timeout = schedule_timeout(30*HZ); 1621 spin_lock(&mdev->peer_seq_lock); 1622 if (timeout == 0 && p_seq == mdev->peer_seq) { 1623 ret = -ETIMEDOUT; 1624 dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n"); 1625 break; 1626 } 1627 } 1628 finish_wait(&mdev->seq_wait, &wait); 1629 if (mdev->peer_seq+1 == packet_seq) 1630 mdev->peer_seq++; 1631 spin_unlock(&mdev->peer_seq_lock); 1632 return ret; 1633 } 1634 1635 static unsigned long write_flags_to_bio(struct drbd_conf *mdev, u32 dpf) 1636 { 1637 if (mdev->agreed_pro_version >= 95) 1638 return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | 1639 (dpf & DP_UNPLUG ? REQ_UNPLUG : 0) | 1640 (dpf & DP_FUA ? REQ_FUA : 0) | 1641 (dpf & DP_FLUSH ? REQ_FUA : 0) | 1642 (dpf & DP_DISCARD ? REQ_DISCARD : 0); 1643 else 1644 return dpf & DP_RW_SYNC ? (REQ_SYNC | REQ_UNPLUG) : 0; 1645 } 1646 1647 /* mirrored write */ 1648 static int receive_Data(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 1649 { 1650 sector_t sector; 1651 struct drbd_epoch_entry *e; 1652 struct p_data *p = &mdev->data.rbuf.data; 1653 int rw = WRITE; 1654 u32 dp_flags; 1655 1656 if (!get_ldev(mdev)) { 1657 if (__ratelimit(&drbd_ratelimit_state)) 1658 dev_err(DEV, "Can not write mirrored data block " 1659 "to local disk.\n"); 1660 spin_lock(&mdev->peer_seq_lock); 1661 if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num)) 1662 mdev->peer_seq++; 1663 spin_unlock(&mdev->peer_seq_lock); 1664 1665 drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); 1666 atomic_inc(&mdev->current_epoch->epoch_size); 1667 return drbd_drain_block(mdev, data_size); 1668 } 1669 1670 /* get_ldev(mdev) successful. 1671 * Corresponding put_ldev done either below (on various errors), 1672 * or in drbd_endio_write_sec, if we successfully submit the data at 1673 * the end of this function. */ 1674 1675 sector = be64_to_cpu(p->sector); 1676 e = read_in_block(mdev, p->block_id, sector, data_size); 1677 if (!e) { 1678 put_ldev(mdev); 1679 return FALSE; 1680 } 1681 1682 e->w.cb = e_end_block; 1683 1684 spin_lock(&mdev->epoch_lock); 1685 e->epoch = mdev->current_epoch; 1686 atomic_inc(&e->epoch->epoch_size); 1687 atomic_inc(&e->epoch->active); 1688 spin_unlock(&mdev->epoch_lock); 1689 1690 dp_flags = be32_to_cpu(p->dp_flags); 1691 rw |= write_flags_to_bio(mdev, dp_flags); 1692 1693 if (dp_flags & DP_MAY_SET_IN_SYNC) 1694 e->flags |= EE_MAY_SET_IN_SYNC; 1695 1696 /* I'm the receiver, I do hold a net_cnt reference. */ 1697 if (!mdev->net_conf->two_primaries) { 1698 spin_lock_irq(&mdev->req_lock); 1699 } else { 1700 /* don't get the req_lock yet, 1701 * we may sleep in drbd_wait_peer_seq */ 1702 const int size = e->size; 1703 const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags); 1704 DEFINE_WAIT(wait); 1705 struct drbd_request *i; 1706 struct hlist_node *n; 1707 struct hlist_head *slot; 1708 int first; 1709 1710 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 1711 BUG_ON(mdev->ee_hash == NULL); 1712 BUG_ON(mdev->tl_hash == NULL); 1713 1714 /* conflict detection and handling: 1715 * 1. wait on the sequence number, 1716 * in case this data packet overtook ACK packets. 1717 * 2. check our hash tables for conflicting requests. 1718 * we only need to walk the tl_hash, since an ee can not 1719 * have a conflict with an other ee: on the submitting 1720 * node, the corresponding req had already been conflicting, 1721 * and a conflicting req is never sent. 1722 * 1723 * Note: for two_primaries, we are protocol C, 1724 * so there cannot be any request that is DONE 1725 * but still on the transfer log. 1726 * 1727 * unconditionally add to the ee_hash. 1728 * 1729 * if no conflicting request is found: 1730 * submit. 1731 * 1732 * if any conflicting request is found 1733 * that has not yet been acked, 1734 * AND I have the "discard concurrent writes" flag: 1735 * queue (via done_ee) the P_DISCARD_ACK; OUT. 1736 * 1737 * if any conflicting request is found: 1738 * block the receiver, waiting on misc_wait 1739 * until no more conflicting requests are there, 1740 * or we get interrupted (disconnect). 1741 * 1742 * we do not just write after local io completion of those 1743 * requests, but only after req is done completely, i.e. 1744 * we wait for the P_DISCARD_ACK to arrive! 1745 * 1746 * then proceed normally, i.e. submit. 1747 */ 1748 if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num))) 1749 goto out_interrupted; 1750 1751 spin_lock_irq(&mdev->req_lock); 1752 1753 hlist_add_head(&e->colision, ee_hash_slot(mdev, sector)); 1754 1755 #define OVERLAPS overlaps(i->sector, i->size, sector, size) 1756 slot = tl_hash_slot(mdev, sector); 1757 first = 1; 1758 for (;;) { 1759 int have_unacked = 0; 1760 int have_conflict = 0; 1761 prepare_to_wait(&mdev->misc_wait, &wait, 1762 TASK_INTERRUPTIBLE); 1763 hlist_for_each_entry(i, n, slot, colision) { 1764 if (OVERLAPS) { 1765 /* only ALERT on first iteration, 1766 * we may be woken up early... */ 1767 if (first) 1768 dev_alert(DEV, "%s[%u] Concurrent local write detected!" 1769 " new: %llus +%u; pending: %llus +%u\n", 1770 current->comm, current->pid, 1771 (unsigned long long)sector, size, 1772 (unsigned long long)i->sector, i->size); 1773 if (i->rq_state & RQ_NET_PENDING) 1774 ++have_unacked; 1775 ++have_conflict; 1776 } 1777 } 1778 #undef OVERLAPS 1779 if (!have_conflict) 1780 break; 1781 1782 /* Discard Ack only for the _first_ iteration */ 1783 if (first && discard && have_unacked) { 1784 dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n", 1785 (unsigned long long)sector); 1786 inc_unacked(mdev); 1787 e->w.cb = e_send_discard_ack; 1788 list_add_tail(&e->w.list, &mdev->done_ee); 1789 1790 spin_unlock_irq(&mdev->req_lock); 1791 1792 /* we could probably send that P_DISCARD_ACK ourselves, 1793 * but I don't like the receiver using the msock */ 1794 1795 put_ldev(mdev); 1796 wake_asender(mdev); 1797 finish_wait(&mdev->misc_wait, &wait); 1798 return TRUE; 1799 } 1800 1801 if (signal_pending(current)) { 1802 hlist_del_init(&e->colision); 1803 1804 spin_unlock_irq(&mdev->req_lock); 1805 1806 finish_wait(&mdev->misc_wait, &wait); 1807 goto out_interrupted; 1808 } 1809 1810 spin_unlock_irq(&mdev->req_lock); 1811 if (first) { 1812 first = 0; 1813 dev_alert(DEV, "Concurrent write! [W AFTERWARDS] " 1814 "sec=%llus\n", (unsigned long long)sector); 1815 } else if (discard) { 1816 /* we had none on the first iteration. 1817 * there must be none now. */ 1818 D_ASSERT(have_unacked == 0); 1819 } 1820 schedule(); 1821 spin_lock_irq(&mdev->req_lock); 1822 } 1823 finish_wait(&mdev->misc_wait, &wait); 1824 } 1825 1826 list_add(&e->w.list, &mdev->active_ee); 1827 spin_unlock_irq(&mdev->req_lock); 1828 1829 switch (mdev->net_conf->wire_protocol) { 1830 case DRBD_PROT_C: 1831 inc_unacked(mdev); 1832 /* corresponding dec_unacked() in e_end_block() 1833 * respective _drbd_clear_done_ee */ 1834 break; 1835 case DRBD_PROT_B: 1836 /* I really don't like it that the receiver thread 1837 * sends on the msock, but anyways */ 1838 drbd_send_ack(mdev, P_RECV_ACK, e); 1839 break; 1840 case DRBD_PROT_A: 1841 /* nothing to do */ 1842 break; 1843 } 1844 1845 if (mdev->state.pdsk < D_INCONSISTENT) { 1846 /* In case we have the only disk of the cluster, */ 1847 drbd_set_out_of_sync(mdev, e->sector, e->size); 1848 e->flags |= EE_CALL_AL_COMPLETE_IO; 1849 e->flags &= ~EE_MAY_SET_IN_SYNC; 1850 drbd_al_begin_io(mdev, e->sector); 1851 } 1852 1853 if (drbd_submit_ee(mdev, e, rw, DRBD_FAULT_DT_WR) == 0) 1854 return TRUE; 1855 1856 /* drbd_submit_ee currently fails for one reason only: 1857 * not being able to allocate enough bios. 1858 * Is dropping the connection going to help? */ 1859 spin_lock_irq(&mdev->req_lock); 1860 list_del(&e->w.list); 1861 hlist_del_init(&e->colision); 1862 spin_unlock_irq(&mdev->req_lock); 1863 if (e->flags & EE_CALL_AL_COMPLETE_IO) 1864 drbd_al_complete_io(mdev, e->sector); 1865 1866 out_interrupted: 1867 /* yes, the epoch_size now is imbalanced. 1868 * but we drop the connection anyways, so we don't have a chance to 1869 * receive a barrier... atomic_inc(&mdev->epoch_size); */ 1870 put_ldev(mdev); 1871 drbd_free_ee(mdev, e); 1872 return FALSE; 1873 } 1874 1875 /* We may throttle resync, if the lower device seems to be busy, 1876 * and current sync rate is above c_min_rate. 1877 * 1878 * To decide whether or not the lower device is busy, we use a scheme similar 1879 * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" 1880 * (more than 64 sectors) of activity we cannot account for with our own resync 1881 * activity, it obviously is "busy". 1882 * 1883 * The current sync rate used here uses only the most recent two step marks, 1884 * to have a short time average so we can react faster. 1885 */ 1886 int drbd_rs_should_slow_down(struct drbd_conf *mdev) 1887 { 1888 struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk; 1889 unsigned long db, dt, dbdt; 1890 int curr_events; 1891 int throttle = 0; 1892 1893 /* feature disabled? */ 1894 if (mdev->sync_conf.c_min_rate == 0) 1895 return 0; 1896 1897 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + 1898 (int)part_stat_read(&disk->part0, sectors[1]) - 1899 atomic_read(&mdev->rs_sect_ev); 1900 if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) { 1901 unsigned long rs_left; 1902 int i; 1903 1904 mdev->rs_last_events = curr_events; 1905 1906 /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, 1907 * approx. */ 1908 i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-2) % DRBD_SYNC_MARKS; 1909 rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; 1910 1911 dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ; 1912 if (!dt) 1913 dt++; 1914 db = mdev->rs_mark_left[i] - rs_left; 1915 dbdt = Bit2KB(db/dt); 1916 1917 if (dbdt > mdev->sync_conf.c_min_rate) 1918 throttle = 1; 1919 } 1920 return throttle; 1921 } 1922 1923 1924 static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int digest_size) 1925 { 1926 sector_t sector; 1927 const sector_t capacity = drbd_get_capacity(mdev->this_bdev); 1928 struct drbd_epoch_entry *e; 1929 struct digest_info *di = NULL; 1930 int size, verb; 1931 unsigned int fault_type; 1932 struct p_block_req *p = &mdev->data.rbuf.block_req; 1933 1934 sector = be64_to_cpu(p->sector); 1935 size = be32_to_cpu(p->blksize); 1936 1937 if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_SEGMENT_SIZE) { 1938 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, 1939 (unsigned long long)sector, size); 1940 return FALSE; 1941 } 1942 if (sector + (size>>9) > capacity) { 1943 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, 1944 (unsigned long long)sector, size); 1945 return FALSE; 1946 } 1947 1948 if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { 1949 verb = 1; 1950 switch (cmd) { 1951 case P_DATA_REQUEST: 1952 drbd_send_ack_rp(mdev, P_NEG_DREPLY, p); 1953 break; 1954 case P_RS_DATA_REQUEST: 1955 case P_CSUM_RS_REQUEST: 1956 case P_OV_REQUEST: 1957 drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p); 1958 break; 1959 case P_OV_REPLY: 1960 verb = 0; 1961 dec_rs_pending(mdev); 1962 drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC); 1963 break; 1964 default: 1965 dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", 1966 cmdname(cmd)); 1967 } 1968 if (verb && __ratelimit(&drbd_ratelimit_state)) 1969 dev_err(DEV, "Can not satisfy peer's read request, " 1970 "no local data.\n"); 1971 1972 /* drain possibly payload */ 1973 return drbd_drain_block(mdev, digest_size); 1974 } 1975 1976 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD 1977 * "criss-cross" setup, that might cause write-out on some other DRBD, 1978 * which in turn might block on the other node at this very place. */ 1979 e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO); 1980 if (!e) { 1981 put_ldev(mdev); 1982 return FALSE; 1983 } 1984 1985 switch (cmd) { 1986 case P_DATA_REQUEST: 1987 e->w.cb = w_e_end_data_req; 1988 fault_type = DRBD_FAULT_DT_RD; 1989 /* application IO, don't drbd_rs_begin_io */ 1990 goto submit; 1991 1992 case P_RS_DATA_REQUEST: 1993 e->w.cb = w_e_end_rsdata_req; 1994 fault_type = DRBD_FAULT_RS_RD; 1995 break; 1996 1997 case P_OV_REPLY: 1998 case P_CSUM_RS_REQUEST: 1999 fault_type = DRBD_FAULT_RS_RD; 2000 di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO); 2001 if (!di) 2002 goto out_free_e; 2003 2004 di->digest_size = digest_size; 2005 di->digest = (((char *)di)+sizeof(struct digest_info)); 2006 2007 e->digest = di; 2008 e->flags |= EE_HAS_DIGEST; 2009 2010 if (drbd_recv(mdev, di->digest, digest_size) != digest_size) 2011 goto out_free_e; 2012 2013 if (cmd == P_CSUM_RS_REQUEST) { 2014 D_ASSERT(mdev->agreed_pro_version >= 89); 2015 e->w.cb = w_e_end_csum_rs_req; 2016 } else if (cmd == P_OV_REPLY) { 2017 e->w.cb = w_e_end_ov_reply; 2018 dec_rs_pending(mdev); 2019 /* drbd_rs_begin_io done when we sent this request, 2020 * but accounting still needs to be done. */ 2021 goto submit_for_resync; 2022 } 2023 break; 2024 2025 case P_OV_REQUEST: 2026 if (mdev->ov_start_sector == ~(sector_t)0 && 2027 mdev->agreed_pro_version >= 90) { 2028 mdev->ov_start_sector = sector; 2029 mdev->ov_position = sector; 2030 mdev->ov_left = mdev->rs_total - BM_SECT_TO_BIT(sector); 2031 dev_info(DEV, "Online Verify start sector: %llu\n", 2032 (unsigned long long)sector); 2033 } 2034 e->w.cb = w_e_end_ov_req; 2035 fault_type = DRBD_FAULT_RS_RD; 2036 break; 2037 2038 default: 2039 dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", 2040 cmdname(cmd)); 2041 fault_type = DRBD_FAULT_MAX; 2042 goto out_free_e; 2043 } 2044 2045 /* Throttle, drbd_rs_begin_io and submit should become asynchronous 2046 * wrt the receiver, but it is not as straightforward as it may seem. 2047 * Various places in the resync start and stop logic assume resync 2048 * requests are processed in order, requeuing this on the worker thread 2049 * introduces a bunch of new code for synchronization between threads. 2050 * 2051 * Unlimited throttling before drbd_rs_begin_io may stall the resync 2052 * "forever", throttling after drbd_rs_begin_io will lock that extent 2053 * for application writes for the same time. For now, just throttle 2054 * here, where the rest of the code expects the receiver to sleep for 2055 * a while, anyways. 2056 */ 2057 2058 /* Throttle before drbd_rs_begin_io, as that locks out application IO; 2059 * this defers syncer requests for some time, before letting at least 2060 * on request through. The resync controller on the receiving side 2061 * will adapt to the incoming rate accordingly. 2062 * 2063 * We cannot throttle here if remote is Primary/SyncTarget: 2064 * we would also throttle its application reads. 2065 * In that case, throttling is done on the SyncTarget only. 2066 */ 2067 if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev)) 2068 msleep(100); 2069 if (drbd_rs_begin_io(mdev, e->sector)) 2070 goto out_free_e; 2071 2072 submit_for_resync: 2073 atomic_add(size >> 9, &mdev->rs_sect_ev); 2074 2075 submit: 2076 inc_unacked(mdev); 2077 spin_lock_irq(&mdev->req_lock); 2078 list_add_tail(&e->w.list, &mdev->read_ee); 2079 spin_unlock_irq(&mdev->req_lock); 2080 2081 if (drbd_submit_ee(mdev, e, READ, fault_type) == 0) 2082 return TRUE; 2083 2084 /* drbd_submit_ee currently fails for one reason only: 2085 * not being able to allocate enough bios. 2086 * Is dropping the connection going to help? */ 2087 spin_lock_irq(&mdev->req_lock); 2088 list_del(&e->w.list); 2089 spin_unlock_irq(&mdev->req_lock); 2090 /* no drbd_rs_complete_io(), we are dropping the connection anyways */ 2091 2092 out_free_e: 2093 put_ldev(mdev); 2094 drbd_free_ee(mdev, e); 2095 return FALSE; 2096 } 2097 2098 static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) 2099 { 2100 int self, peer, rv = -100; 2101 unsigned long ch_self, ch_peer; 2102 2103 self = mdev->ldev->md.uuid[UI_BITMAP] & 1; 2104 peer = mdev->p_uuid[UI_BITMAP] & 1; 2105 2106 ch_peer = mdev->p_uuid[UI_SIZE]; 2107 ch_self = mdev->comm_bm_set; 2108 2109 switch (mdev->net_conf->after_sb_0p) { 2110 case ASB_CONSENSUS: 2111 case ASB_DISCARD_SECONDARY: 2112 case ASB_CALL_HELPER: 2113 dev_err(DEV, "Configuration error.\n"); 2114 break; 2115 case ASB_DISCONNECT: 2116 break; 2117 case ASB_DISCARD_YOUNGER_PRI: 2118 if (self == 0 && peer == 1) { 2119 rv = -1; 2120 break; 2121 } 2122 if (self == 1 && peer == 0) { 2123 rv = 1; 2124 break; 2125 } 2126 /* Else fall through to one of the other strategies... */ 2127 case ASB_DISCARD_OLDER_PRI: 2128 if (self == 0 && peer == 1) { 2129 rv = 1; 2130 break; 2131 } 2132 if (self == 1 && peer == 0) { 2133 rv = -1; 2134 break; 2135 } 2136 /* Else fall through to one of the other strategies... */ 2137 dev_warn(DEV, "Discard younger/older primary did not find a decision\n" 2138 "Using discard-least-changes instead\n"); 2139 case ASB_DISCARD_ZERO_CHG: 2140 if (ch_peer == 0 && ch_self == 0) { 2141 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) 2142 ? -1 : 1; 2143 break; 2144 } else { 2145 if (ch_peer == 0) { rv = 1; break; } 2146 if (ch_self == 0) { rv = -1; break; } 2147 } 2148 if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG) 2149 break; 2150 case ASB_DISCARD_LEAST_CHG: 2151 if (ch_self < ch_peer) 2152 rv = -1; 2153 else if (ch_self > ch_peer) 2154 rv = 1; 2155 else /* ( ch_self == ch_peer ) */ 2156 /* Well, then use something else. */ 2157 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) 2158 ? -1 : 1; 2159 break; 2160 case ASB_DISCARD_LOCAL: 2161 rv = -1; 2162 break; 2163 case ASB_DISCARD_REMOTE: 2164 rv = 1; 2165 } 2166 2167 return rv; 2168 } 2169 2170 static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) 2171 { 2172 int self, peer, hg, rv = -100; 2173 2174 self = mdev->ldev->md.uuid[UI_BITMAP] & 1; 2175 peer = mdev->p_uuid[UI_BITMAP] & 1; 2176 2177 switch (mdev->net_conf->after_sb_1p) { 2178 case ASB_DISCARD_YOUNGER_PRI: 2179 case ASB_DISCARD_OLDER_PRI: 2180 case ASB_DISCARD_LEAST_CHG: 2181 case ASB_DISCARD_LOCAL: 2182 case ASB_DISCARD_REMOTE: 2183 dev_err(DEV, "Configuration error.\n"); 2184 break; 2185 case ASB_DISCONNECT: 2186 break; 2187 case ASB_CONSENSUS: 2188 hg = drbd_asb_recover_0p(mdev); 2189 if (hg == -1 && mdev->state.role == R_SECONDARY) 2190 rv = hg; 2191 if (hg == 1 && mdev->state.role == R_PRIMARY) 2192 rv = hg; 2193 break; 2194 case ASB_VIOLENTLY: 2195 rv = drbd_asb_recover_0p(mdev); 2196 break; 2197 case ASB_DISCARD_SECONDARY: 2198 return mdev->state.role == R_PRIMARY ? 1 : -1; 2199 case ASB_CALL_HELPER: 2200 hg = drbd_asb_recover_0p(mdev); 2201 if (hg == -1 && mdev->state.role == R_PRIMARY) { 2202 self = drbd_set_role(mdev, R_SECONDARY, 0); 2203 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, 2204 * we might be here in C_WF_REPORT_PARAMS which is transient. 2205 * we do not need to wait for the after state change work either. */ 2206 self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); 2207 if (self != SS_SUCCESS) { 2208 drbd_khelper(mdev, "pri-lost-after-sb"); 2209 } else { 2210 dev_warn(DEV, "Successfully gave up primary role.\n"); 2211 rv = hg; 2212 } 2213 } else 2214 rv = hg; 2215 } 2216 2217 return rv; 2218 } 2219 2220 static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) 2221 { 2222 int self, peer, hg, rv = -100; 2223 2224 self = mdev->ldev->md.uuid[UI_BITMAP] & 1; 2225 peer = mdev->p_uuid[UI_BITMAP] & 1; 2226 2227 switch (mdev->net_conf->after_sb_2p) { 2228 case ASB_DISCARD_YOUNGER_PRI: 2229 case ASB_DISCARD_OLDER_PRI: 2230 case ASB_DISCARD_LEAST_CHG: 2231 case ASB_DISCARD_LOCAL: 2232 case ASB_DISCARD_REMOTE: 2233 case ASB_CONSENSUS: 2234 case ASB_DISCARD_SECONDARY: 2235 dev_err(DEV, "Configuration error.\n"); 2236 break; 2237 case ASB_VIOLENTLY: 2238 rv = drbd_asb_recover_0p(mdev); 2239 break; 2240 case ASB_DISCONNECT: 2241 break; 2242 case ASB_CALL_HELPER: 2243 hg = drbd_asb_recover_0p(mdev); 2244 if (hg == -1) { 2245 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, 2246 * we might be here in C_WF_REPORT_PARAMS which is transient. 2247 * we do not need to wait for the after state change work either. */ 2248 self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); 2249 if (self != SS_SUCCESS) { 2250 drbd_khelper(mdev, "pri-lost-after-sb"); 2251 } else { 2252 dev_warn(DEV, "Successfully gave up primary role.\n"); 2253 rv = hg; 2254 } 2255 } else 2256 rv = hg; 2257 } 2258 2259 return rv; 2260 } 2261 2262 static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, 2263 u64 bits, u64 flags) 2264 { 2265 if (!uuid) { 2266 dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); 2267 return; 2268 } 2269 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", 2270 text, 2271 (unsigned long long)uuid[UI_CURRENT], 2272 (unsigned long long)uuid[UI_BITMAP], 2273 (unsigned long long)uuid[UI_HISTORY_START], 2274 (unsigned long long)uuid[UI_HISTORY_END], 2275 (unsigned long long)bits, 2276 (unsigned long long)flags); 2277 } 2278 2279 /* 2280 100 after split brain try auto recover 2281 2 C_SYNC_SOURCE set BitMap 2282 1 C_SYNC_SOURCE use BitMap 2283 0 no Sync 2284 -1 C_SYNC_TARGET use BitMap 2285 -2 C_SYNC_TARGET set BitMap 2286 -100 after split brain, disconnect 2287 -1000 unrelated data 2288 */ 2289 static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) 2290 { 2291 u64 self, peer; 2292 int i, j; 2293 2294 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); 2295 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); 2296 2297 *rule_nr = 10; 2298 if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) 2299 return 0; 2300 2301 *rule_nr = 20; 2302 if ((self == UUID_JUST_CREATED || self == (u64)0) && 2303 peer != UUID_JUST_CREATED) 2304 return -2; 2305 2306 *rule_nr = 30; 2307 if (self != UUID_JUST_CREATED && 2308 (peer == UUID_JUST_CREATED || peer == (u64)0)) 2309 return 2; 2310 2311 if (self == peer) { 2312 int rct, dc; /* roles at crash time */ 2313 2314 if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { 2315 2316 if (mdev->agreed_pro_version < 91) 2317 return -1001; 2318 2319 if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && 2320 (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { 2321 dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); 2322 drbd_uuid_set_bm(mdev, 0UL); 2323 2324 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, 2325 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); 2326 *rule_nr = 34; 2327 } else { 2328 dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); 2329 *rule_nr = 36; 2330 } 2331 2332 return 1; 2333 } 2334 2335 if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { 2336 2337 if (mdev->agreed_pro_version < 91) 2338 return -1001; 2339 2340 if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && 2341 (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { 2342 dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); 2343 2344 mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; 2345 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; 2346 mdev->p_uuid[UI_BITMAP] = 0UL; 2347 2348 drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); 2349 *rule_nr = 35; 2350 } else { 2351 dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); 2352 *rule_nr = 37; 2353 } 2354 2355 return -1; 2356 } 2357 2358 /* Common power [off|failure] */ 2359 rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + 2360 (mdev->p_uuid[UI_FLAGS] & 2); 2361 /* lowest bit is set when we were primary, 2362 * next bit (weight 2) is set when peer was primary */ 2363 *rule_nr = 40; 2364 2365 switch (rct) { 2366 case 0: /* !self_pri && !peer_pri */ return 0; 2367 case 1: /* self_pri && !peer_pri */ return 1; 2368 case 2: /* !self_pri && peer_pri */ return -1; 2369 case 3: /* self_pri && peer_pri */ 2370 dc = test_bit(DISCARD_CONCURRENT, &mdev->flags); 2371 return dc ? -1 : 1; 2372 } 2373 } 2374 2375 *rule_nr = 50; 2376 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); 2377 if (self == peer) 2378 return -1; 2379 2380 *rule_nr = 51; 2381 peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); 2382 if (self == peer) { 2383 self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); 2384 peer = mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1); 2385 if (self == peer) { 2386 /* The last P_SYNC_UUID did not get though. Undo the last start of 2387 resync as sync source modifications of the peer's UUIDs. */ 2388 2389 if (mdev->agreed_pro_version < 91) 2390 return -1001; 2391 2392 mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; 2393 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; 2394 return -1; 2395 } 2396 } 2397 2398 *rule_nr = 60; 2399 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); 2400 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2401 peer = mdev->p_uuid[i] & ~((u64)1); 2402 if (self == peer) 2403 return -2; 2404 } 2405 2406 *rule_nr = 70; 2407 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); 2408 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); 2409 if (self == peer) 2410 return 1; 2411 2412 *rule_nr = 71; 2413 self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); 2414 if (self == peer) { 2415 self = mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1); 2416 peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); 2417 if (self == peer) { 2418 /* The last P_SYNC_UUID did not get though. Undo the last start of 2419 resync as sync source modifications of our UUIDs. */ 2420 2421 if (mdev->agreed_pro_version < 91) 2422 return -1001; 2423 2424 _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); 2425 _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); 2426 2427 dev_info(DEV, "Undid last start of resync:\n"); 2428 2429 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, 2430 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); 2431 2432 return 1; 2433 } 2434 } 2435 2436 2437 *rule_nr = 80; 2438 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); 2439 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2440 self = mdev->ldev->md.uuid[i] & ~((u64)1); 2441 if (self == peer) 2442 return 2; 2443 } 2444 2445 *rule_nr = 90; 2446 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); 2447 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); 2448 if (self == peer && self != ((u64)0)) 2449 return 100; 2450 2451 *rule_nr = 100; 2452 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2453 self = mdev->ldev->md.uuid[i] & ~((u64)1); 2454 for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { 2455 peer = mdev->p_uuid[j] & ~((u64)1); 2456 if (self == peer) 2457 return -100; 2458 } 2459 } 2460 2461 return -1000; 2462 } 2463 2464 /* drbd_sync_handshake() returns the new conn state on success, or 2465 CONN_MASK (-1) on failure. 2466 */ 2467 static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, 2468 enum drbd_disk_state peer_disk) __must_hold(local) 2469 { 2470 int hg, rule_nr; 2471 enum drbd_conns rv = C_MASK; 2472 enum drbd_disk_state mydisk; 2473 2474 mydisk = mdev->state.disk; 2475 if (mydisk == D_NEGOTIATING) 2476 mydisk = mdev->new_state_tmp.disk; 2477 2478 dev_info(DEV, "drbd_sync_handshake:\n"); 2479 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); 2480 drbd_uuid_dump(mdev, "peer", mdev->p_uuid, 2481 mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); 2482 2483 hg = drbd_uuid_compare(mdev, &rule_nr); 2484 2485 dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); 2486 2487 if (hg == -1000) { 2488 dev_alert(DEV, "Unrelated data, aborting!\n"); 2489 return C_MASK; 2490 } 2491 if (hg == -1001) { 2492 dev_alert(DEV, "To resolve this both sides have to support at least protocol\n"); 2493 return C_MASK; 2494 } 2495 2496 if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || 2497 (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { 2498 int f = (hg == -100) || abs(hg) == 2; 2499 hg = mydisk > D_INCONSISTENT ? 1 : -1; 2500 if (f) 2501 hg = hg*2; 2502 dev_info(DEV, "Becoming sync %s due to disk states.\n", 2503 hg > 0 ? "source" : "target"); 2504 } 2505 2506 if (abs(hg) == 100) 2507 drbd_khelper(mdev, "initial-split-brain"); 2508 2509 if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) { 2510 int pcount = (mdev->state.role == R_PRIMARY) 2511 + (peer_role == R_PRIMARY); 2512 int forced = (hg == -100); 2513 2514 switch (pcount) { 2515 case 0: 2516 hg = drbd_asb_recover_0p(mdev); 2517 break; 2518 case 1: 2519 hg = drbd_asb_recover_1p(mdev); 2520 break; 2521 case 2: 2522 hg = drbd_asb_recover_2p(mdev); 2523 break; 2524 } 2525 if (abs(hg) < 100) { 2526 dev_warn(DEV, "Split-Brain detected, %d primaries, " 2527 "automatically solved. Sync from %s node\n", 2528 pcount, (hg < 0) ? "peer" : "this"); 2529 if (forced) { 2530 dev_warn(DEV, "Doing a full sync, since" 2531 " UUIDs where ambiguous.\n"); 2532 hg = hg*2; 2533 } 2534 } 2535 } 2536 2537 if (hg == -100) { 2538 if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1)) 2539 hg = -1; 2540 if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1)) 2541 hg = 1; 2542 2543 if (abs(hg) < 100) 2544 dev_warn(DEV, "Split-Brain detected, manually solved. " 2545 "Sync from %s node\n", 2546 (hg < 0) ? "peer" : "this"); 2547 } 2548 2549 if (hg == -100) { 2550 /* FIXME this log message is not correct if we end up here 2551 * after an attempted attach on a diskless node. 2552 * We just refuse to attach -- well, we drop the "connection" 2553 * to that disk, in a way... */ 2554 dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n"); 2555 drbd_khelper(mdev, "split-brain"); 2556 return C_MASK; 2557 } 2558 2559 if (hg > 0 && mydisk <= D_INCONSISTENT) { 2560 dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); 2561 return C_MASK; 2562 } 2563 2564 if (hg < 0 && /* by intention we do not use mydisk here. */ 2565 mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { 2566 switch (mdev->net_conf->rr_conflict) { 2567 case ASB_CALL_HELPER: 2568 drbd_khelper(mdev, "pri-lost"); 2569 /* fall through */ 2570 case ASB_DISCONNECT: 2571 dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); 2572 return C_MASK; 2573 case ASB_VIOLENTLY: 2574 dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" 2575 "assumption\n"); 2576 } 2577 } 2578 2579 if (mdev->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) { 2580 if (hg == 0) 2581 dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n"); 2582 else 2583 dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.", 2584 drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), 2585 abs(hg) >= 2 ? "full" : "bit-map based"); 2586 return C_MASK; 2587 } 2588 2589 if (abs(hg) >= 2) { 2590 dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); 2591 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake")) 2592 return C_MASK; 2593 } 2594 2595 if (hg > 0) { /* become sync source. */ 2596 rv = C_WF_BITMAP_S; 2597 } else if (hg < 0) { /* become sync target */ 2598 rv = C_WF_BITMAP_T; 2599 } else { 2600 rv = C_CONNECTED; 2601 if (drbd_bm_total_weight(mdev)) { 2602 dev_info(DEV, "No resync, but %lu bits in bitmap!\n", 2603 drbd_bm_total_weight(mdev)); 2604 } 2605 } 2606 2607 return rv; 2608 } 2609 2610 /* returns 1 if invalid */ 2611 static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self) 2612 { 2613 /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ 2614 if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) || 2615 (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL)) 2616 return 0; 2617 2618 /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ 2619 if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL || 2620 self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL) 2621 return 1; 2622 2623 /* everything else is valid if they are equal on both sides. */ 2624 if (peer == self) 2625 return 0; 2626 2627 /* everything es is invalid. */ 2628 return 1; 2629 } 2630 2631 static int receive_protocol(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 2632 { 2633 struct p_protocol *p = &mdev->data.rbuf.protocol; 2634 int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; 2635 int p_want_lose, p_two_primaries, cf; 2636 char p_integrity_alg[SHARED_SECRET_MAX] = ""; 2637 2638 p_proto = be32_to_cpu(p->protocol); 2639 p_after_sb_0p = be32_to_cpu(p->after_sb_0p); 2640 p_after_sb_1p = be32_to_cpu(p->after_sb_1p); 2641 p_after_sb_2p = be32_to_cpu(p->after_sb_2p); 2642 p_two_primaries = be32_to_cpu(p->two_primaries); 2643 cf = be32_to_cpu(p->conn_flags); 2644 p_want_lose = cf & CF_WANT_LOSE; 2645 2646 clear_bit(CONN_DRY_RUN, &mdev->flags); 2647 2648 if (cf & CF_DRY_RUN) 2649 set_bit(CONN_DRY_RUN, &mdev->flags); 2650 2651 if (p_proto != mdev->net_conf->wire_protocol) { 2652 dev_err(DEV, "incompatible communication protocols\n"); 2653 goto disconnect; 2654 } 2655 2656 if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) { 2657 dev_err(DEV, "incompatible after-sb-0pri settings\n"); 2658 goto disconnect; 2659 } 2660 2661 if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) { 2662 dev_err(DEV, "incompatible after-sb-1pri settings\n"); 2663 goto disconnect; 2664 } 2665 2666 if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) { 2667 dev_err(DEV, "incompatible after-sb-2pri settings\n"); 2668 goto disconnect; 2669 } 2670 2671 if (p_want_lose && mdev->net_conf->want_lose) { 2672 dev_err(DEV, "both sides have the 'want_lose' flag set\n"); 2673 goto disconnect; 2674 } 2675 2676 if (p_two_primaries != mdev->net_conf->two_primaries) { 2677 dev_err(DEV, "incompatible setting of the two-primaries options\n"); 2678 goto disconnect; 2679 } 2680 2681 if (mdev->agreed_pro_version >= 87) { 2682 unsigned char *my_alg = mdev->net_conf->integrity_alg; 2683 2684 if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size) 2685 return FALSE; 2686 2687 p_integrity_alg[SHARED_SECRET_MAX-1] = 0; 2688 if (strcmp(p_integrity_alg, my_alg)) { 2689 dev_err(DEV, "incompatible setting of the data-integrity-alg\n"); 2690 goto disconnect; 2691 } 2692 dev_info(DEV, "data-integrity-alg: %s\n", 2693 my_alg[0] ? my_alg : (unsigned char *)"<not-used>"); 2694 } 2695 2696 return TRUE; 2697 2698 disconnect: 2699 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2700 return FALSE; 2701 } 2702 2703 /* helper function 2704 * input: alg name, feature name 2705 * return: NULL (alg name was "") 2706 * ERR_PTR(error) if something goes wrong 2707 * or the crypto hash ptr, if it worked out ok. */ 2708 struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, 2709 const char *alg, const char *name) 2710 { 2711 struct crypto_hash *tfm; 2712 2713 if (!alg[0]) 2714 return NULL; 2715 2716 tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); 2717 if (IS_ERR(tfm)) { 2718 dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", 2719 alg, name, PTR_ERR(tfm)); 2720 return tfm; 2721 } 2722 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { 2723 crypto_free_hash(tfm); 2724 dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name); 2725 return ERR_PTR(-EINVAL); 2726 } 2727 return tfm; 2728 } 2729 2730 static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int packet_size) 2731 { 2732 int ok = TRUE; 2733 struct p_rs_param_95 *p = &mdev->data.rbuf.rs_param_95; 2734 unsigned int header_size, data_size, exp_max_sz; 2735 struct crypto_hash *verify_tfm = NULL; 2736 struct crypto_hash *csums_tfm = NULL; 2737 const int apv = mdev->agreed_pro_version; 2738 int *rs_plan_s = NULL; 2739 int fifo_size = 0; 2740 2741 exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) 2742 : apv == 88 ? sizeof(struct p_rs_param) 2743 + SHARED_SECRET_MAX 2744 : apv <= 94 ? sizeof(struct p_rs_param_89) 2745 : /* apv >= 95 */ sizeof(struct p_rs_param_95); 2746 2747 if (packet_size > exp_max_sz) { 2748 dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", 2749 packet_size, exp_max_sz); 2750 return FALSE; 2751 } 2752 2753 if (apv <= 88) { 2754 header_size = sizeof(struct p_rs_param) - sizeof(struct p_header80); 2755 data_size = packet_size - header_size; 2756 } else if (apv <= 94) { 2757 header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header80); 2758 data_size = packet_size - header_size; 2759 D_ASSERT(data_size == 0); 2760 } else { 2761 header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header80); 2762 data_size = packet_size - header_size; 2763 D_ASSERT(data_size == 0); 2764 } 2765 2766 /* initialize verify_alg and csums_alg */ 2767 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); 2768 2769 if (drbd_recv(mdev, &p->head.payload, header_size) != header_size) 2770 return FALSE; 2771 2772 mdev->sync_conf.rate = be32_to_cpu(p->rate); 2773 2774 if (apv >= 88) { 2775 if (apv == 88) { 2776 if (data_size > SHARED_SECRET_MAX) { 2777 dev_err(DEV, "verify-alg too long, " 2778 "peer wants %u, accepting only %u byte\n", 2779 data_size, SHARED_SECRET_MAX); 2780 return FALSE; 2781 } 2782 2783 if (drbd_recv(mdev, p->verify_alg, data_size) != data_size) 2784 return FALSE; 2785 2786 /* we expect NUL terminated string */ 2787 /* but just in case someone tries to be evil */ 2788 D_ASSERT(p->verify_alg[data_size-1] == 0); 2789 p->verify_alg[data_size-1] = 0; 2790 2791 } else /* apv >= 89 */ { 2792 /* we still expect NUL terminated strings */ 2793 /* but just in case someone tries to be evil */ 2794 D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); 2795 D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); 2796 p->verify_alg[SHARED_SECRET_MAX-1] = 0; 2797 p->csums_alg[SHARED_SECRET_MAX-1] = 0; 2798 } 2799 2800 if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) { 2801 if (mdev->state.conn == C_WF_REPORT_PARAMS) { 2802 dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", 2803 mdev->sync_conf.verify_alg, p->verify_alg); 2804 goto disconnect; 2805 } 2806 verify_tfm = drbd_crypto_alloc_digest_safe(mdev, 2807 p->verify_alg, "verify-alg"); 2808 if (IS_ERR(verify_tfm)) { 2809 verify_tfm = NULL; 2810 goto disconnect; 2811 } 2812 } 2813 2814 if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) { 2815 if (mdev->state.conn == C_WF_REPORT_PARAMS) { 2816 dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", 2817 mdev->sync_conf.csums_alg, p->csums_alg); 2818 goto disconnect; 2819 } 2820 csums_tfm = drbd_crypto_alloc_digest_safe(mdev, 2821 p->csums_alg, "csums-alg"); 2822 if (IS_ERR(csums_tfm)) { 2823 csums_tfm = NULL; 2824 goto disconnect; 2825 } 2826 } 2827 2828 if (apv > 94) { 2829 mdev->sync_conf.rate = be32_to_cpu(p->rate); 2830 mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead); 2831 mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target); 2832 mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target); 2833 mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate); 2834 2835 fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ; 2836 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { 2837 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); 2838 if (!rs_plan_s) { 2839 dev_err(DEV, "kmalloc of fifo_buffer failed"); 2840 goto disconnect; 2841 } 2842 } 2843 } 2844 2845 spin_lock(&mdev->peer_seq_lock); 2846 /* lock against drbd_nl_syncer_conf() */ 2847 if (verify_tfm) { 2848 strcpy(mdev->sync_conf.verify_alg, p->verify_alg); 2849 mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1; 2850 crypto_free_hash(mdev->verify_tfm); 2851 mdev->verify_tfm = verify_tfm; 2852 dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); 2853 } 2854 if (csums_tfm) { 2855 strcpy(mdev->sync_conf.csums_alg, p->csums_alg); 2856 mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1; 2857 crypto_free_hash(mdev->csums_tfm); 2858 mdev->csums_tfm = csums_tfm; 2859 dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); 2860 } 2861 if (fifo_size != mdev->rs_plan_s.size) { 2862 kfree(mdev->rs_plan_s.values); 2863 mdev->rs_plan_s.values = rs_plan_s; 2864 mdev->rs_plan_s.size = fifo_size; 2865 mdev->rs_planed = 0; 2866 } 2867 spin_unlock(&mdev->peer_seq_lock); 2868 } 2869 2870 return ok; 2871 disconnect: 2872 /* just for completeness: actually not needed, 2873 * as this is not reached if csums_tfm was ok. */ 2874 crypto_free_hash(csums_tfm); 2875 /* but free the verify_tfm again, if csums_tfm did not work out */ 2876 crypto_free_hash(verify_tfm); 2877 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2878 return FALSE; 2879 } 2880 2881 static void drbd_setup_order_type(struct drbd_conf *mdev, int peer) 2882 { 2883 /* sorry, we currently have no working implementation 2884 * of distributed TCQ */ 2885 } 2886 2887 /* warn if the arguments differ by more than 12.5% */ 2888 static void warn_if_differ_considerably(struct drbd_conf *mdev, 2889 const char *s, sector_t a, sector_t b) 2890 { 2891 sector_t d; 2892 if (a == 0 || b == 0) 2893 return; 2894 d = (a > b) ? (a - b) : (b - a); 2895 if (d > (a>>3) || d > (b>>3)) 2896 dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, 2897 (unsigned long long)a, (unsigned long long)b); 2898 } 2899 2900 static int receive_sizes(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 2901 { 2902 struct p_sizes *p = &mdev->data.rbuf.sizes; 2903 enum determine_dev_size dd = unchanged; 2904 unsigned int max_seg_s; 2905 sector_t p_size, p_usize, my_usize; 2906 int ldsc = 0; /* local disk size changed */ 2907 enum dds_flags ddsf; 2908 2909 p_size = be64_to_cpu(p->d_size); 2910 p_usize = be64_to_cpu(p->u_size); 2911 2912 if (p_size == 0 && mdev->state.disk == D_DISKLESS) { 2913 dev_err(DEV, "some backing storage is needed\n"); 2914 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2915 return FALSE; 2916 } 2917 2918 /* just store the peer's disk size for now. 2919 * we still need to figure out whether we accept that. */ 2920 mdev->p_size = p_size; 2921 2922 if (get_ldev(mdev)) { 2923 warn_if_differ_considerably(mdev, "lower level device sizes", 2924 p_size, drbd_get_max_capacity(mdev->ldev)); 2925 warn_if_differ_considerably(mdev, "user requested size", 2926 p_usize, mdev->ldev->dc.disk_size); 2927 2928 /* if this is the first connect, or an otherwise expected 2929 * param exchange, choose the minimum */ 2930 if (mdev->state.conn == C_WF_REPORT_PARAMS) 2931 p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size, 2932 p_usize); 2933 2934 my_usize = mdev->ldev->dc.disk_size; 2935 2936 if (mdev->ldev->dc.disk_size != p_usize) { 2937 mdev->ldev->dc.disk_size = p_usize; 2938 dev_info(DEV, "Peer sets u_size to %lu sectors\n", 2939 (unsigned long)mdev->ldev->dc.disk_size); 2940 } 2941 2942 /* Never shrink a device with usable data during connect. 2943 But allow online shrinking if we are connected. */ 2944 if (drbd_new_dev_size(mdev, mdev->ldev, 0) < 2945 drbd_get_capacity(mdev->this_bdev) && 2946 mdev->state.disk >= D_OUTDATED && 2947 mdev->state.conn < C_CONNECTED) { 2948 dev_err(DEV, "The peer's disk size is too small!\n"); 2949 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2950 mdev->ldev->dc.disk_size = my_usize; 2951 put_ldev(mdev); 2952 return FALSE; 2953 } 2954 put_ldev(mdev); 2955 } 2956 #undef min_not_zero 2957 2958 ddsf = be16_to_cpu(p->dds_flags); 2959 if (get_ldev(mdev)) { 2960 dd = drbd_determin_dev_size(mdev, ddsf); 2961 put_ldev(mdev); 2962 if (dd == dev_size_error) 2963 return FALSE; 2964 drbd_md_sync(mdev); 2965 } else { 2966 /* I am diskless, need to accept the peer's size. */ 2967 drbd_set_my_capacity(mdev, p_size); 2968 } 2969 2970 if (get_ldev(mdev)) { 2971 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { 2972 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); 2973 ldsc = 1; 2974 } 2975 2976 if (mdev->agreed_pro_version < 94) 2977 max_seg_s = be32_to_cpu(p->max_segment_size); 2978 else if (mdev->agreed_pro_version == 94) 2979 max_seg_s = DRBD_MAX_SIZE_H80_PACKET; 2980 else /* drbd 8.3.8 onwards */ 2981 max_seg_s = DRBD_MAX_SEGMENT_SIZE; 2982 2983 if (max_seg_s != queue_max_segment_size(mdev->rq_queue)) 2984 drbd_setup_queue_param(mdev, max_seg_s); 2985 2986 drbd_setup_order_type(mdev, be16_to_cpu(p->queue_order_type)); 2987 put_ldev(mdev); 2988 } 2989 2990 if (mdev->state.conn > C_WF_REPORT_PARAMS) { 2991 if (be64_to_cpu(p->c_size) != 2992 drbd_get_capacity(mdev->this_bdev) || ldsc) { 2993 /* we have different sizes, probably peer 2994 * needs to know my new size... */ 2995 drbd_send_sizes(mdev, 0, ddsf); 2996 } 2997 if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || 2998 (dd == grew && mdev->state.conn == C_CONNECTED)) { 2999 if (mdev->state.pdsk >= D_INCONSISTENT && 3000 mdev->state.disk >= D_INCONSISTENT) { 3001 if (ddsf & DDSF_NO_RESYNC) 3002 dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n"); 3003 else 3004 resync_after_online_grow(mdev); 3005 } else 3006 set_bit(RESYNC_AFTER_NEG, &mdev->flags); 3007 } 3008 } 3009 3010 return TRUE; 3011 } 3012 3013 static int receive_uuids(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3014 { 3015 struct p_uuids *p = &mdev->data.rbuf.uuids; 3016 u64 *p_uuid; 3017 int i; 3018 3019 p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); 3020 3021 for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) 3022 p_uuid[i] = be64_to_cpu(p->uuid[i]); 3023 3024 kfree(mdev->p_uuid); 3025 mdev->p_uuid = p_uuid; 3026 3027 if (mdev->state.conn < C_CONNECTED && 3028 mdev->state.disk < D_INCONSISTENT && 3029 mdev->state.role == R_PRIMARY && 3030 (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { 3031 dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", 3032 (unsigned long long)mdev->ed_uuid); 3033 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 3034 return FALSE; 3035 } 3036 3037 if (get_ldev(mdev)) { 3038 int skip_initial_sync = 3039 mdev->state.conn == C_CONNECTED && 3040 mdev->agreed_pro_version >= 90 && 3041 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && 3042 (p_uuid[UI_FLAGS] & 8); 3043 if (skip_initial_sync) { 3044 dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); 3045 drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, 3046 "clear_n_write from receive_uuids"); 3047 _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); 3048 _drbd_uuid_set(mdev, UI_BITMAP, 0); 3049 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), 3050 CS_VERBOSE, NULL); 3051 drbd_md_sync(mdev); 3052 } 3053 put_ldev(mdev); 3054 } else if (mdev->state.disk < D_INCONSISTENT && 3055 mdev->state.role == R_PRIMARY) { 3056 /* I am a diskless primary, the peer just created a new current UUID 3057 for me. */ 3058 drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); 3059 } 3060 3061 /* Before we test for the disk state, we should wait until an eventually 3062 ongoing cluster wide state change is finished. That is important if 3063 we are primary and are detaching from our disk. We need to see the 3064 new disk state... */ 3065 wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags)); 3066 if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) 3067 drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); 3068 3069 return TRUE; 3070 } 3071 3072 /** 3073 * convert_state() - Converts the peer's view of the cluster state to our point of view 3074 * @ps: The state as seen by the peer. 3075 */ 3076 static union drbd_state convert_state(union drbd_state ps) 3077 { 3078 union drbd_state ms; 3079 3080 static enum drbd_conns c_tab[] = { 3081 [C_CONNECTED] = C_CONNECTED, 3082 3083 [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, 3084 [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, 3085 [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ 3086 [C_VERIFY_S] = C_VERIFY_T, 3087 [C_MASK] = C_MASK, 3088 }; 3089 3090 ms.i = ps.i; 3091 3092 ms.conn = c_tab[ps.conn]; 3093 ms.peer = ps.role; 3094 ms.role = ps.peer; 3095 ms.pdsk = ps.disk; 3096 ms.disk = ps.pdsk; 3097 ms.peer_isp = (ps.aftr_isp | ps.user_isp); 3098 3099 return ms; 3100 } 3101 3102 static int receive_req_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3103 { 3104 struct p_req_state *p = &mdev->data.rbuf.req_state; 3105 union drbd_state mask, val; 3106 int rv; 3107 3108 mask.i = be32_to_cpu(p->mask); 3109 val.i = be32_to_cpu(p->val); 3110 3111 if (test_bit(DISCARD_CONCURRENT, &mdev->flags) && 3112 test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) { 3113 drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); 3114 return TRUE; 3115 } 3116 3117 mask = convert_state(mask); 3118 val = convert_state(val); 3119 3120 rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); 3121 3122 drbd_send_sr_reply(mdev, rv); 3123 drbd_md_sync(mdev); 3124 3125 return TRUE; 3126 } 3127 3128 static int receive_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3129 { 3130 struct p_state *p = &mdev->data.rbuf.state; 3131 union drbd_state os, ns, peer_state; 3132 enum drbd_disk_state real_peer_disk; 3133 enum chg_state_flags cs_flags; 3134 int rv; 3135 3136 peer_state.i = be32_to_cpu(p->state); 3137 3138 real_peer_disk = peer_state.disk; 3139 if (peer_state.disk == D_NEGOTIATING) { 3140 real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; 3141 dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); 3142 } 3143 3144 spin_lock_irq(&mdev->req_lock); 3145 retry: 3146 os = ns = mdev->state; 3147 spin_unlock_irq(&mdev->req_lock); 3148 3149 /* peer says his disk is uptodate, while we think it is inconsistent, 3150 * and this happens while we think we have a sync going on. */ 3151 if (os.pdsk == D_INCONSISTENT && real_peer_disk == D_UP_TO_DATE && 3152 os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { 3153 /* If we are (becoming) SyncSource, but peer is still in sync 3154 * preparation, ignore its uptodate-ness to avoid flapping, it 3155 * will change to inconsistent once the peer reaches active 3156 * syncing states. 3157 * It may have changed syncer-paused flags, however, so we 3158 * cannot ignore this completely. */ 3159 if (peer_state.conn > C_CONNECTED && 3160 peer_state.conn < C_SYNC_SOURCE) 3161 real_peer_disk = D_INCONSISTENT; 3162 3163 /* if peer_state changes to connected at the same time, 3164 * it explicitly notifies us that it finished resync. 3165 * Maybe we should finish it up, too? */ 3166 else if (os.conn >= C_SYNC_SOURCE && 3167 peer_state.conn == C_CONNECTED) { 3168 if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) 3169 drbd_resync_finished(mdev); 3170 return TRUE; 3171 } 3172 } 3173 3174 /* peer says his disk is inconsistent, while we think it is uptodate, 3175 * and this happens while the peer still thinks we have a sync going on, 3176 * but we think we are already done with the sync. 3177 * We ignore this to avoid flapping pdsk. 3178 * This should not happen, if the peer is a recent version of drbd. */ 3179 if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && 3180 os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) 3181 real_peer_disk = D_UP_TO_DATE; 3182 3183 if (ns.conn == C_WF_REPORT_PARAMS) 3184 ns.conn = C_CONNECTED; 3185 3186 if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && 3187 get_ldev_if_state(mdev, D_NEGOTIATING)) { 3188 int cr; /* consider resync */ 3189 3190 /* if we established a new connection */ 3191 cr = (os.conn < C_CONNECTED); 3192 /* if we had an established connection 3193 * and one of the nodes newly attaches a disk */ 3194 cr |= (os.conn == C_CONNECTED && 3195 (peer_state.disk == D_NEGOTIATING || 3196 os.disk == D_NEGOTIATING)); 3197 /* if we have both been inconsistent, and the peer has been 3198 * forced to be UpToDate with --overwrite-data */ 3199 cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); 3200 /* if we had been plain connected, and the admin requested to 3201 * start a sync by "invalidate" or "invalidate-remote" */ 3202 cr |= (os.conn == C_CONNECTED && 3203 (peer_state.conn >= C_STARTING_SYNC_S && 3204 peer_state.conn <= C_WF_BITMAP_T)); 3205 3206 if (cr) 3207 ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); 3208 3209 put_ldev(mdev); 3210 if (ns.conn == C_MASK) { 3211 ns.conn = C_CONNECTED; 3212 if (mdev->state.disk == D_NEGOTIATING) { 3213 drbd_force_state(mdev, NS(disk, D_FAILED)); 3214 } else if (peer_state.disk == D_NEGOTIATING) { 3215 dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); 3216 peer_state.disk = D_DISKLESS; 3217 real_peer_disk = D_DISKLESS; 3218 } else { 3219 if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags)) 3220 return FALSE; 3221 D_ASSERT(os.conn == C_WF_REPORT_PARAMS); 3222 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 3223 return FALSE; 3224 } 3225 } 3226 } 3227 3228 spin_lock_irq(&mdev->req_lock); 3229 if (mdev->state.i != os.i) 3230 goto retry; 3231 clear_bit(CONSIDER_RESYNC, &mdev->flags); 3232 ns.peer = peer_state.role; 3233 ns.pdsk = real_peer_disk; 3234 ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); 3235 if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) 3236 ns.disk = mdev->new_state_tmp.disk; 3237 cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); 3238 if (ns.pdsk == D_CONSISTENT && is_susp(ns) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && 3239 test_bit(NEW_CUR_UUID, &mdev->flags)) { 3240 /* Do not allow tl_restart(resend) for a rebooted peer. We can only allow this 3241 for temporal network outages! */ 3242 spin_unlock_irq(&mdev->req_lock); 3243 dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); 3244 tl_clear(mdev); 3245 drbd_uuid_new_current(mdev); 3246 clear_bit(NEW_CUR_UUID, &mdev->flags); 3247 drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0)); 3248 return FALSE; 3249 } 3250 rv = _drbd_set_state(mdev, ns, cs_flags, NULL); 3251 ns = mdev->state; 3252 spin_unlock_irq(&mdev->req_lock); 3253 3254 if (rv < SS_SUCCESS) { 3255 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 3256 return FALSE; 3257 } 3258 3259 if (os.conn > C_WF_REPORT_PARAMS) { 3260 if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && 3261 peer_state.disk != D_NEGOTIATING ) { 3262 /* we want resync, peer has not yet decided to sync... */ 3263 /* Nowadays only used when forcing a node into primary role and 3264 setting its disk to UpToDate with that */ 3265 drbd_send_uuids(mdev); 3266 drbd_send_state(mdev); 3267 } 3268 } 3269 3270 mdev->net_conf->want_lose = 0; 3271 3272 drbd_md_sync(mdev); /* update connected indicator, la_size, ... */ 3273 3274 return TRUE; 3275 } 3276 3277 static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3278 { 3279 struct p_rs_uuid *p = &mdev->data.rbuf.rs_uuid; 3280 3281 wait_event(mdev->misc_wait, 3282 mdev->state.conn == C_WF_SYNC_UUID || 3283 mdev->state.conn < C_CONNECTED || 3284 mdev->state.disk < D_NEGOTIATING); 3285 3286 /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ 3287 3288 /* Here the _drbd_uuid_ functions are right, current should 3289 _not_ be rotated into the history */ 3290 if (get_ldev_if_state(mdev, D_NEGOTIATING)) { 3291 _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); 3292 _drbd_uuid_set(mdev, UI_BITMAP, 0UL); 3293 3294 drbd_start_resync(mdev, C_SYNC_TARGET); 3295 3296 put_ldev(mdev); 3297 } else 3298 dev_err(DEV, "Ignoring SyncUUID packet!\n"); 3299 3300 return TRUE; 3301 } 3302 3303 enum receive_bitmap_ret { OK, DONE, FAILED }; 3304 3305 static enum receive_bitmap_ret 3306 receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size, 3307 unsigned long *buffer, struct bm_xfer_ctx *c) 3308 { 3309 unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset); 3310 unsigned want = num_words * sizeof(long); 3311 3312 if (want != data_size) { 3313 dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size); 3314 return FAILED; 3315 } 3316 if (want == 0) 3317 return DONE; 3318 if (drbd_recv(mdev, buffer, want) != want) 3319 return FAILED; 3320 3321 drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer); 3322 3323 c->word_offset += num_words; 3324 c->bit_offset = c->word_offset * BITS_PER_LONG; 3325 if (c->bit_offset > c->bm_bits) 3326 c->bit_offset = c->bm_bits; 3327 3328 return OK; 3329 } 3330 3331 static enum receive_bitmap_ret 3332 recv_bm_rle_bits(struct drbd_conf *mdev, 3333 struct p_compressed_bm *p, 3334 struct bm_xfer_ctx *c) 3335 { 3336 struct bitstream bs; 3337 u64 look_ahead; 3338 u64 rl; 3339 u64 tmp; 3340 unsigned long s = c->bit_offset; 3341 unsigned long e; 3342 int len = be16_to_cpu(p->head.length) - (sizeof(*p) - sizeof(p->head)); 3343 int toggle = DCBP_get_start(p); 3344 int have; 3345 int bits; 3346 3347 bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p)); 3348 3349 bits = bitstream_get_bits(&bs, &look_ahead, 64); 3350 if (bits < 0) 3351 return FAILED; 3352 3353 for (have = bits; have > 0; s += rl, toggle = !toggle) { 3354 bits = vli_decode_bits(&rl, look_ahead); 3355 if (bits <= 0) 3356 return FAILED; 3357 3358 if (toggle) { 3359 e = s + rl -1; 3360 if (e >= c->bm_bits) { 3361 dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); 3362 return FAILED; 3363 } 3364 _drbd_bm_set_bits(mdev, s, e); 3365 } 3366 3367 if (have < bits) { 3368 dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", 3369 have, bits, look_ahead, 3370 (unsigned int)(bs.cur.b - p->code), 3371 (unsigned int)bs.buf_len); 3372 return FAILED; 3373 } 3374 look_ahead >>= bits; 3375 have -= bits; 3376 3377 bits = bitstream_get_bits(&bs, &tmp, 64 - have); 3378 if (bits < 0) 3379 return FAILED; 3380 look_ahead |= tmp << have; 3381 have += bits; 3382 } 3383 3384 c->bit_offset = s; 3385 bm_xfer_ctx_bit_to_word_offset(c); 3386 3387 return (s == c->bm_bits) ? DONE : OK; 3388 } 3389 3390 static enum receive_bitmap_ret 3391 decode_bitmap_c(struct drbd_conf *mdev, 3392 struct p_compressed_bm *p, 3393 struct bm_xfer_ctx *c) 3394 { 3395 if (DCBP_get_code(p) == RLE_VLI_Bits) 3396 return recv_bm_rle_bits(mdev, p, c); 3397 3398 /* other variants had been implemented for evaluation, 3399 * but have been dropped as this one turned out to be "best" 3400 * during all our tests. */ 3401 3402 dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); 3403 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); 3404 return FAILED; 3405 } 3406 3407 void INFO_bm_xfer_stats(struct drbd_conf *mdev, 3408 const char *direction, struct bm_xfer_ctx *c) 3409 { 3410 /* what would it take to transfer it "plaintext" */ 3411 unsigned plain = sizeof(struct p_header80) * 3412 ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1) 3413 + c->bm_words * sizeof(long); 3414 unsigned total = c->bytes[0] + c->bytes[1]; 3415 unsigned r; 3416 3417 /* total can not be zero. but just in case: */ 3418 if (total == 0) 3419 return; 3420 3421 /* don't report if not compressed */ 3422 if (total >= plain) 3423 return; 3424 3425 /* total < plain. check for overflow, still */ 3426 r = (total > UINT_MAX/1000) ? (total / (plain/1000)) 3427 : (1000 * total / plain); 3428 3429 if (r > 1000) 3430 r = 1000; 3431 3432 r = 1000 - r; 3433 dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " 3434 "total %u; compression: %u.%u%%\n", 3435 direction, 3436 c->bytes[1], c->packets[1], 3437 c->bytes[0], c->packets[0], 3438 total, r/10, r % 10); 3439 } 3440 3441 /* Since we are processing the bitfield from lower addresses to higher, 3442 it does not matter if the process it in 32 bit chunks or 64 bit 3443 chunks as long as it is little endian. (Understand it as byte stream, 3444 beginning with the lowest byte...) If we would use big endian 3445 we would need to process it from the highest address to the lowest, 3446 in order to be agnostic to the 32 vs 64 bits issue. 3447 3448 returns 0 on failure, 1 if we successfully received it. */ 3449 static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3450 { 3451 struct bm_xfer_ctx c; 3452 void *buffer; 3453 enum receive_bitmap_ret ret; 3454 int ok = FALSE; 3455 struct p_header80 *h = &mdev->data.rbuf.header.h80; 3456 3457 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt)); 3458 3459 drbd_bm_lock(mdev, "receive bitmap"); 3460 3461 /* maybe we should use some per thread scratch page, 3462 * and allocate that during initial device creation? */ 3463 buffer = (unsigned long *) __get_free_page(GFP_NOIO); 3464 if (!buffer) { 3465 dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__); 3466 goto out; 3467 } 3468 3469 c = (struct bm_xfer_ctx) { 3470 .bm_bits = drbd_bm_bits(mdev), 3471 .bm_words = drbd_bm_words(mdev), 3472 }; 3473 3474 do { 3475 if (cmd == P_BITMAP) { 3476 ret = receive_bitmap_plain(mdev, data_size, buffer, &c); 3477 } else if (cmd == P_COMPRESSED_BITMAP) { 3478 /* MAYBE: sanity check that we speak proto >= 90, 3479 * and the feature is enabled! */ 3480 struct p_compressed_bm *p; 3481 3482 if (data_size > BM_PACKET_PAYLOAD_BYTES) { 3483 dev_err(DEV, "ReportCBitmap packet too large\n"); 3484 goto out; 3485 } 3486 /* use the page buff */ 3487 p = buffer; 3488 memcpy(p, h, sizeof(*h)); 3489 if (drbd_recv(mdev, p->head.payload, data_size) != data_size) 3490 goto out; 3491 if (data_size <= (sizeof(*p) - sizeof(p->head))) { 3492 dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size); 3493 return FAILED; 3494 } 3495 ret = decode_bitmap_c(mdev, p, &c); 3496 } else { 3497 dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd); 3498 goto out; 3499 } 3500 3501 c.packets[cmd == P_BITMAP]++; 3502 c.bytes[cmd == P_BITMAP] += sizeof(struct p_header80) + data_size; 3503 3504 if (ret != OK) 3505 break; 3506 3507 if (!drbd_recv_header(mdev, &cmd, &data_size)) 3508 goto out; 3509 } while (ret == OK); 3510 if (ret == FAILED) 3511 goto out; 3512 3513 INFO_bm_xfer_stats(mdev, "receive", &c); 3514 3515 if (mdev->state.conn == C_WF_BITMAP_T) { 3516 ok = !drbd_send_bitmap(mdev); 3517 if (!ok) 3518 goto out; 3519 /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ 3520 ok = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); 3521 D_ASSERT(ok == SS_SUCCESS); 3522 } else if (mdev->state.conn != C_WF_BITMAP_S) { 3523 /* admin may have requested C_DISCONNECTING, 3524 * other threads may have noticed network errors */ 3525 dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", 3526 drbd_conn_str(mdev->state.conn)); 3527 } 3528 3529 ok = TRUE; 3530 out: 3531 drbd_bm_unlock(mdev); 3532 if (ok && mdev->state.conn == C_WF_BITMAP_S) 3533 drbd_start_resync(mdev, C_SYNC_SOURCE); 3534 free_page((unsigned long) buffer); 3535 return ok; 3536 } 3537 3538 static int receive_skip(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3539 { 3540 /* TODO zero copy sink :) */ 3541 static char sink[128]; 3542 int size, want, r; 3543 3544 dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n", 3545 cmd, data_size); 3546 3547 size = data_size; 3548 while (size > 0) { 3549 want = min_t(int, size, sizeof(sink)); 3550 r = drbd_recv(mdev, sink, want); 3551 ERR_IF(r <= 0) break; 3552 size -= r; 3553 } 3554 return size == 0; 3555 } 3556 3557 static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3558 { 3559 if (mdev->state.disk >= D_INCONSISTENT) 3560 drbd_kick_lo(mdev); 3561 3562 /* Make sure we've acked all the TCP data associated 3563 * with the data requests being unplugged */ 3564 drbd_tcp_quickack(mdev->data.socket); 3565 3566 return TRUE; 3567 } 3568 3569 typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packets cmd, unsigned int to_receive); 3570 3571 struct data_cmd { 3572 int expect_payload; 3573 size_t pkt_size; 3574 drbd_cmd_handler_f function; 3575 }; 3576 3577 static struct data_cmd drbd_cmd_handler[] = { 3578 [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, 3579 [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, 3580 [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , 3581 [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , 3582 [P_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } , 3583 [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } , 3584 [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header80), receive_UnplugRemote }, 3585 [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 3586 [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 3587 [P_SYNC_PARAM] = { 1, sizeof(struct p_header80), receive_SyncParam }, 3588 [P_SYNC_PARAM89] = { 1, sizeof(struct p_header80), receive_SyncParam }, 3589 [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, 3590 [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, 3591 [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, 3592 [P_STATE] = { 0, sizeof(struct p_state), receive_state }, 3593 [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, 3594 [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, 3595 [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 3596 [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, 3597 [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, 3598 [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, 3599 /* anything missing from this table is in 3600 * the asender_tbl, see get_asender_cmd */ 3601 [P_MAX_CMD] = { 0, 0, NULL }, 3602 }; 3603 3604 /* All handler functions that expect a sub-header get that sub-heder in 3605 mdev->data.rbuf.header.head.payload. 3606 3607 Usually in mdev->data.rbuf.header.head the callback can find the usual 3608 p_header, but they may not rely on that. Since there is also p_header95 ! 3609 */ 3610 3611 static void drbdd(struct drbd_conf *mdev) 3612 { 3613 union p_header *header = &mdev->data.rbuf.header; 3614 unsigned int packet_size; 3615 enum drbd_packets cmd; 3616 size_t shs; /* sub header size */ 3617 int rv; 3618 3619 while (get_t_state(&mdev->receiver) == Running) { 3620 drbd_thread_current_set_cpu(mdev); 3621 if (!drbd_recv_header(mdev, &cmd, &packet_size)) 3622 goto err_out; 3623 3624 if (unlikely(cmd >= P_MAX_CMD || !drbd_cmd_handler[cmd].function)) { 3625 dev_err(DEV, "unknown packet type %d, l: %d!\n", cmd, packet_size); 3626 goto err_out; 3627 } 3628 3629 shs = drbd_cmd_handler[cmd].pkt_size - sizeof(union p_header); 3630 if (packet_size - shs > 0 && !drbd_cmd_handler[cmd].expect_payload) { 3631 dev_err(DEV, "No payload expected %s l:%d\n", cmdname(cmd), packet_size); 3632 goto err_out; 3633 } 3634 3635 if (shs) { 3636 rv = drbd_recv(mdev, &header->h80.payload, shs); 3637 if (unlikely(rv != shs)) { 3638 dev_err(DEV, "short read while reading sub header: rv=%d\n", rv); 3639 goto err_out; 3640 } 3641 } 3642 3643 rv = drbd_cmd_handler[cmd].function(mdev, cmd, packet_size - shs); 3644 3645 if (unlikely(!rv)) { 3646 dev_err(DEV, "error receiving %s, l: %d!\n", 3647 cmdname(cmd), packet_size); 3648 goto err_out; 3649 } 3650 } 3651 3652 if (0) { 3653 err_out: 3654 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); 3655 } 3656 /* If we leave here, we probably want to update at least the 3657 * "Connected" indicator on stable storage. Do so explicitly here. */ 3658 drbd_md_sync(mdev); 3659 } 3660 3661 void drbd_flush_workqueue(struct drbd_conf *mdev) 3662 { 3663 struct drbd_wq_barrier barr; 3664 3665 barr.w.cb = w_prev_work_done; 3666 init_completion(&barr.done); 3667 drbd_queue_work(&mdev->data.work, &barr.w); 3668 wait_for_completion(&barr.done); 3669 } 3670 3671 void drbd_free_tl_hash(struct drbd_conf *mdev) 3672 { 3673 struct hlist_head *h; 3674 3675 spin_lock_irq(&mdev->req_lock); 3676 3677 if (!mdev->tl_hash || mdev->state.conn != C_STANDALONE) { 3678 spin_unlock_irq(&mdev->req_lock); 3679 return; 3680 } 3681 /* paranoia code */ 3682 for (h = mdev->ee_hash; h < mdev->ee_hash + mdev->ee_hash_s; h++) 3683 if (h->first) 3684 dev_err(DEV, "ASSERT FAILED ee_hash[%u].first == %p, expected NULL\n", 3685 (int)(h - mdev->ee_hash), h->first); 3686 kfree(mdev->ee_hash); 3687 mdev->ee_hash = NULL; 3688 mdev->ee_hash_s = 0; 3689 3690 /* paranoia code */ 3691 for (h = mdev->tl_hash; h < mdev->tl_hash + mdev->tl_hash_s; h++) 3692 if (h->first) 3693 dev_err(DEV, "ASSERT FAILED tl_hash[%u] == %p, expected NULL\n", 3694 (int)(h - mdev->tl_hash), h->first); 3695 kfree(mdev->tl_hash); 3696 mdev->tl_hash = NULL; 3697 mdev->tl_hash_s = 0; 3698 spin_unlock_irq(&mdev->req_lock); 3699 } 3700 3701 static void drbd_disconnect(struct drbd_conf *mdev) 3702 { 3703 enum drbd_fencing_p fp; 3704 union drbd_state os, ns; 3705 int rv = SS_UNKNOWN_ERROR; 3706 unsigned int i; 3707 3708 if (mdev->state.conn == C_STANDALONE) 3709 return; 3710 if (mdev->state.conn >= C_WF_CONNECTION) 3711 dev_err(DEV, "ASSERT FAILED cstate = %s, expected < WFConnection\n", 3712 drbd_conn_str(mdev->state.conn)); 3713 3714 /* asender does not clean up anything. it must not interfere, either */ 3715 drbd_thread_stop(&mdev->asender); 3716 drbd_free_sock(mdev); 3717 3718 /* wait for current activity to cease. */ 3719 spin_lock_irq(&mdev->req_lock); 3720 _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); 3721 _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); 3722 _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); 3723 spin_unlock_irq(&mdev->req_lock); 3724 3725 /* We do not have data structures that would allow us to 3726 * get the rs_pending_cnt down to 0 again. 3727 * * On C_SYNC_TARGET we do not have any data structures describing 3728 * the pending RSDataRequest's we have sent. 3729 * * On C_SYNC_SOURCE there is no data structure that tracks 3730 * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. 3731 * And no, it is not the sum of the reference counts in the 3732 * resync_LRU. The resync_LRU tracks the whole operation including 3733 * the disk-IO, while the rs_pending_cnt only tracks the blocks 3734 * on the fly. */ 3735 drbd_rs_cancel_all(mdev); 3736 mdev->rs_total = 0; 3737 mdev->rs_failed = 0; 3738 atomic_set(&mdev->rs_pending_cnt, 0); 3739 wake_up(&mdev->misc_wait); 3740 3741 /* make sure syncer is stopped and w_resume_next_sg queued */ 3742 del_timer_sync(&mdev->resync_timer); 3743 resync_timer_fn((unsigned long)mdev); 3744 3745 /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, 3746 * w_make_resync_request etc. which may still be on the worker queue 3747 * to be "canceled" */ 3748 drbd_flush_workqueue(mdev); 3749 3750 /* This also does reclaim_net_ee(). If we do this too early, we might 3751 * miss some resync ee and pages.*/ 3752 drbd_process_done_ee(mdev); 3753 3754 kfree(mdev->p_uuid); 3755 mdev->p_uuid = NULL; 3756 3757 if (!is_susp(mdev->state)) 3758 tl_clear(mdev); 3759 3760 dev_info(DEV, "Connection closed\n"); 3761 3762 drbd_md_sync(mdev); 3763 3764 fp = FP_DONT_CARE; 3765 if (get_ldev(mdev)) { 3766 fp = mdev->ldev->dc.fencing; 3767 put_ldev(mdev); 3768 } 3769 3770 if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN) 3771 drbd_try_outdate_peer_async(mdev); 3772 3773 spin_lock_irq(&mdev->req_lock); 3774 os = mdev->state; 3775 if (os.conn >= C_UNCONNECTED) { 3776 /* Do not restart in case we are C_DISCONNECTING */ 3777 ns = os; 3778 ns.conn = C_UNCONNECTED; 3779 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); 3780 } 3781 spin_unlock_irq(&mdev->req_lock); 3782 3783 if (os.conn == C_DISCONNECTING) { 3784 wait_event(mdev->net_cnt_wait, atomic_read(&mdev->net_cnt) == 0); 3785 3786 if (!is_susp(mdev->state)) { 3787 /* we must not free the tl_hash 3788 * while application io is still on the fly */ 3789 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt)); 3790 drbd_free_tl_hash(mdev); 3791 } 3792 3793 crypto_free_hash(mdev->cram_hmac_tfm); 3794 mdev->cram_hmac_tfm = NULL; 3795 3796 kfree(mdev->net_conf); 3797 mdev->net_conf = NULL; 3798 drbd_request_state(mdev, NS(conn, C_STANDALONE)); 3799 } 3800 3801 /* tcp_close and release of sendpage pages can be deferred. I don't 3802 * want to use SO_LINGER, because apparently it can be deferred for 3803 * more than 20 seconds (longest time I checked). 3804 * 3805 * Actually we don't care for exactly when the network stack does its 3806 * put_page(), but release our reference on these pages right here. 3807 */ 3808 i = drbd_release_ee(mdev, &mdev->net_ee); 3809 if (i) 3810 dev_info(DEV, "net_ee not empty, killed %u entries\n", i); 3811 i = atomic_read(&mdev->pp_in_use_by_net); 3812 if (i) 3813 dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i); 3814 i = atomic_read(&mdev->pp_in_use); 3815 if (i) 3816 dev_info(DEV, "pp_in_use = %d, expected 0\n", i); 3817 3818 D_ASSERT(list_empty(&mdev->read_ee)); 3819 D_ASSERT(list_empty(&mdev->active_ee)); 3820 D_ASSERT(list_empty(&mdev->sync_ee)); 3821 D_ASSERT(list_empty(&mdev->done_ee)); 3822 3823 /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ 3824 atomic_set(&mdev->current_epoch->epoch_size, 0); 3825 D_ASSERT(list_empty(&mdev->current_epoch->list)); 3826 } 3827 3828 /* 3829 * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version 3830 * we can agree on is stored in agreed_pro_version. 3831 * 3832 * feature flags and the reserved array should be enough room for future 3833 * enhancements of the handshake protocol, and possible plugins... 3834 * 3835 * for now, they are expected to be zero, but ignored. 3836 */ 3837 static int drbd_send_handshake(struct drbd_conf *mdev) 3838 { 3839 /* ASSERT current == mdev->receiver ... */ 3840 struct p_handshake *p = &mdev->data.sbuf.handshake; 3841 int ok; 3842 3843 if (mutex_lock_interruptible(&mdev->data.mutex)) { 3844 dev_err(DEV, "interrupted during initial handshake\n"); 3845 return 0; /* interrupted. not ok. */ 3846 } 3847 3848 if (mdev->data.socket == NULL) { 3849 mutex_unlock(&mdev->data.mutex); 3850 return 0; 3851 } 3852 3853 memset(p, 0, sizeof(*p)); 3854 p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); 3855 p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); 3856 ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE, 3857 (struct p_header80 *)p, sizeof(*p), 0 ); 3858 mutex_unlock(&mdev->data.mutex); 3859 return ok; 3860 } 3861 3862 /* 3863 * return values: 3864 * 1 yes, we have a valid connection 3865 * 0 oops, did not work out, please try again 3866 * -1 peer talks different language, 3867 * no point in trying again, please go standalone. 3868 */ 3869 static int drbd_do_handshake(struct drbd_conf *mdev) 3870 { 3871 /* ASSERT current == mdev->receiver ... */ 3872 struct p_handshake *p = &mdev->data.rbuf.handshake; 3873 const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80); 3874 unsigned int length; 3875 enum drbd_packets cmd; 3876 int rv; 3877 3878 rv = drbd_send_handshake(mdev); 3879 if (!rv) 3880 return 0; 3881 3882 rv = drbd_recv_header(mdev, &cmd, &length); 3883 if (!rv) 3884 return 0; 3885 3886 if (cmd != P_HAND_SHAKE) { 3887 dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n", 3888 cmdname(cmd), cmd); 3889 return -1; 3890 } 3891 3892 if (length != expect) { 3893 dev_err(DEV, "expected HandShake length: %u, received: %u\n", 3894 expect, length); 3895 return -1; 3896 } 3897 3898 rv = drbd_recv(mdev, &p->head.payload, expect); 3899 3900 if (rv != expect) { 3901 dev_err(DEV, "short read receiving handshake packet: l=%u\n", rv); 3902 return 0; 3903 } 3904 3905 p->protocol_min = be32_to_cpu(p->protocol_min); 3906 p->protocol_max = be32_to_cpu(p->protocol_max); 3907 if (p->protocol_max == 0) 3908 p->protocol_max = p->protocol_min; 3909 3910 if (PRO_VERSION_MAX < p->protocol_min || 3911 PRO_VERSION_MIN > p->protocol_max) 3912 goto incompat; 3913 3914 mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); 3915 3916 dev_info(DEV, "Handshake successful: " 3917 "Agreed network protocol version %d\n", mdev->agreed_pro_version); 3918 3919 return 1; 3920 3921 incompat: 3922 dev_err(DEV, "incompatible DRBD dialects: " 3923 "I support %d-%d, peer supports %d-%d\n", 3924 PRO_VERSION_MIN, PRO_VERSION_MAX, 3925 p->protocol_min, p->protocol_max); 3926 return -1; 3927 } 3928 3929 #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) 3930 static int drbd_do_auth(struct drbd_conf *mdev) 3931 { 3932 dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); 3933 dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); 3934 return -1; 3935 } 3936 #else 3937 #define CHALLENGE_LEN 64 3938 3939 /* Return value: 3940 1 - auth succeeded, 3941 0 - failed, try again (network error), 3942 -1 - auth failed, don't try again. 3943 */ 3944 3945 static int drbd_do_auth(struct drbd_conf *mdev) 3946 { 3947 char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ 3948 struct scatterlist sg; 3949 char *response = NULL; 3950 char *right_response = NULL; 3951 char *peers_ch = NULL; 3952 unsigned int key_len = strlen(mdev->net_conf->shared_secret); 3953 unsigned int resp_size; 3954 struct hash_desc desc; 3955 enum drbd_packets cmd; 3956 unsigned int length; 3957 int rv; 3958 3959 desc.tfm = mdev->cram_hmac_tfm; 3960 desc.flags = 0; 3961 3962 rv = crypto_hash_setkey(mdev->cram_hmac_tfm, 3963 (u8 *)mdev->net_conf->shared_secret, key_len); 3964 if (rv) { 3965 dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv); 3966 rv = -1; 3967 goto fail; 3968 } 3969 3970 get_random_bytes(my_challenge, CHALLENGE_LEN); 3971 3972 rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN); 3973 if (!rv) 3974 goto fail; 3975 3976 rv = drbd_recv_header(mdev, &cmd, &length); 3977 if (!rv) 3978 goto fail; 3979 3980 if (cmd != P_AUTH_CHALLENGE) { 3981 dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n", 3982 cmdname(cmd), cmd); 3983 rv = 0; 3984 goto fail; 3985 } 3986 3987 if (length > CHALLENGE_LEN * 2) { 3988 dev_err(DEV, "expected AuthChallenge payload too big.\n"); 3989 rv = -1; 3990 goto fail; 3991 } 3992 3993 peers_ch = kmalloc(length, GFP_NOIO); 3994 if (peers_ch == NULL) { 3995 dev_err(DEV, "kmalloc of peers_ch failed\n"); 3996 rv = -1; 3997 goto fail; 3998 } 3999 4000 rv = drbd_recv(mdev, peers_ch, length); 4001 4002 if (rv != length) { 4003 dev_err(DEV, "short read AuthChallenge: l=%u\n", rv); 4004 rv = 0; 4005 goto fail; 4006 } 4007 4008 resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm); 4009 response = kmalloc(resp_size, GFP_NOIO); 4010 if (response == NULL) { 4011 dev_err(DEV, "kmalloc of response failed\n"); 4012 rv = -1; 4013 goto fail; 4014 } 4015 4016 sg_init_table(&sg, 1); 4017 sg_set_buf(&sg, peers_ch, length); 4018 4019 rv = crypto_hash_digest(&desc, &sg, sg.length, response); 4020 if (rv) { 4021 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); 4022 rv = -1; 4023 goto fail; 4024 } 4025 4026 rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size); 4027 if (!rv) 4028 goto fail; 4029 4030 rv = drbd_recv_header(mdev, &cmd, &length); 4031 if (!rv) 4032 goto fail; 4033 4034 if (cmd != P_AUTH_RESPONSE) { 4035 dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n", 4036 cmdname(cmd), cmd); 4037 rv = 0; 4038 goto fail; 4039 } 4040 4041 if (length != resp_size) { 4042 dev_err(DEV, "expected AuthResponse payload of wrong size\n"); 4043 rv = 0; 4044 goto fail; 4045 } 4046 4047 rv = drbd_recv(mdev, response , resp_size); 4048 4049 if (rv != resp_size) { 4050 dev_err(DEV, "short read receiving AuthResponse: l=%u\n", rv); 4051 rv = 0; 4052 goto fail; 4053 } 4054 4055 right_response = kmalloc(resp_size, GFP_NOIO); 4056 if (right_response == NULL) { 4057 dev_err(DEV, "kmalloc of right_response failed\n"); 4058 rv = -1; 4059 goto fail; 4060 } 4061 4062 sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); 4063 4064 rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); 4065 if (rv) { 4066 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); 4067 rv = -1; 4068 goto fail; 4069 } 4070 4071 rv = !memcmp(response, right_response, resp_size); 4072 4073 if (rv) 4074 dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n", 4075 resp_size, mdev->net_conf->cram_hmac_alg); 4076 else 4077 rv = -1; 4078 4079 fail: 4080 kfree(peers_ch); 4081 kfree(response); 4082 kfree(right_response); 4083 4084 return rv; 4085 } 4086 #endif 4087 4088 int drbdd_init(struct drbd_thread *thi) 4089 { 4090 struct drbd_conf *mdev = thi->mdev; 4091 unsigned int minor = mdev_to_minor(mdev); 4092 int h; 4093 4094 sprintf(current->comm, "drbd%d_receiver", minor); 4095 4096 dev_info(DEV, "receiver (re)started\n"); 4097 4098 do { 4099 h = drbd_connect(mdev); 4100 if (h == 0) { 4101 drbd_disconnect(mdev); 4102 __set_current_state(TASK_INTERRUPTIBLE); 4103 schedule_timeout(HZ); 4104 } 4105 if (h == -1) { 4106 dev_warn(DEV, "Discarding network configuration.\n"); 4107 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 4108 } 4109 } while (h == 0); 4110 4111 if (h > 0) { 4112 if (get_net_conf(mdev)) { 4113 drbdd(mdev); 4114 put_net_conf(mdev); 4115 } 4116 } 4117 4118 drbd_disconnect(mdev); 4119 4120 dev_info(DEV, "receiver terminated\n"); 4121 return 0; 4122 } 4123 4124 /* ********* acknowledge sender ******** */ 4125 4126 static int got_RqSReply(struct drbd_conf *mdev, struct p_header80 *h) 4127 { 4128 struct p_req_state_reply *p = (struct p_req_state_reply *)h; 4129 4130 int retcode = be32_to_cpu(p->retcode); 4131 4132 if (retcode >= SS_SUCCESS) { 4133 set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); 4134 } else { 4135 set_bit(CL_ST_CHG_FAIL, &mdev->flags); 4136 dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", 4137 drbd_set_st_err_str(retcode), retcode); 4138 } 4139 wake_up(&mdev->state_wait); 4140 4141 return TRUE; 4142 } 4143 4144 static int got_Ping(struct drbd_conf *mdev, struct p_header80 *h) 4145 { 4146 return drbd_send_ping_ack(mdev); 4147 4148 } 4149 4150 static int got_PingAck(struct drbd_conf *mdev, struct p_header80 *h) 4151 { 4152 /* restore idle timeout */ 4153 mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ; 4154 if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags)) 4155 wake_up(&mdev->misc_wait); 4156 4157 return TRUE; 4158 } 4159 4160 static int got_IsInSync(struct drbd_conf *mdev, struct p_header80 *h) 4161 { 4162 struct p_block_ack *p = (struct p_block_ack *)h; 4163 sector_t sector = be64_to_cpu(p->sector); 4164 int blksize = be32_to_cpu(p->blksize); 4165 4166 D_ASSERT(mdev->agreed_pro_version >= 89); 4167 4168 update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4169 4170 if (get_ldev(mdev)) { 4171 drbd_rs_complete_io(mdev, sector); 4172 drbd_set_in_sync(mdev, sector, blksize); 4173 /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ 4174 mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); 4175 put_ldev(mdev); 4176 } 4177 dec_rs_pending(mdev); 4178 atomic_add(blksize >> 9, &mdev->rs_sect_in); 4179 4180 return TRUE; 4181 } 4182 4183 /* when we receive the ACK for a write request, 4184 * verify that we actually know about it */ 4185 static struct drbd_request *_ack_id_to_req(struct drbd_conf *mdev, 4186 u64 id, sector_t sector) 4187 { 4188 struct hlist_head *slot = tl_hash_slot(mdev, sector); 4189 struct hlist_node *n; 4190 struct drbd_request *req; 4191 4192 hlist_for_each_entry(req, n, slot, colision) { 4193 if ((unsigned long)req == (unsigned long)id) { 4194 if (req->sector != sector) { 4195 dev_err(DEV, "_ack_id_to_req: found req %p but it has " 4196 "wrong sector (%llus versus %llus)\n", req, 4197 (unsigned long long)req->sector, 4198 (unsigned long long)sector); 4199 break; 4200 } 4201 return req; 4202 } 4203 } 4204 dev_err(DEV, "_ack_id_to_req: failed to find req %p, sector %llus in list\n", 4205 (void *)(unsigned long)id, (unsigned long long)sector); 4206 return NULL; 4207 } 4208 4209 typedef struct drbd_request *(req_validator_fn) 4210 (struct drbd_conf *mdev, u64 id, sector_t sector); 4211 4212 static int validate_req_change_req_state(struct drbd_conf *mdev, 4213 u64 id, sector_t sector, req_validator_fn validator, 4214 const char *func, enum drbd_req_event what) 4215 { 4216 struct drbd_request *req; 4217 struct bio_and_error m; 4218 4219 spin_lock_irq(&mdev->req_lock); 4220 req = validator(mdev, id, sector); 4221 if (unlikely(!req)) { 4222 spin_unlock_irq(&mdev->req_lock); 4223 dev_err(DEV, "%s: got a corrupt block_id/sector pair\n", func); 4224 return FALSE; 4225 } 4226 __req_mod(req, what, &m); 4227 spin_unlock_irq(&mdev->req_lock); 4228 4229 if (m.bio) 4230 complete_master_bio(mdev, &m); 4231 return TRUE; 4232 } 4233 4234 static int got_BlockAck(struct drbd_conf *mdev, struct p_header80 *h) 4235 { 4236 struct p_block_ack *p = (struct p_block_ack *)h; 4237 sector_t sector = be64_to_cpu(p->sector); 4238 int blksize = be32_to_cpu(p->blksize); 4239 enum drbd_req_event what; 4240 4241 update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4242 4243 if (is_syncer_block_id(p->block_id)) { 4244 drbd_set_in_sync(mdev, sector, blksize); 4245 dec_rs_pending(mdev); 4246 return TRUE; 4247 } 4248 switch (be16_to_cpu(h->command)) { 4249 case P_RS_WRITE_ACK: 4250 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 4251 what = write_acked_by_peer_and_sis; 4252 break; 4253 case P_WRITE_ACK: 4254 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 4255 what = write_acked_by_peer; 4256 break; 4257 case P_RECV_ACK: 4258 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B); 4259 what = recv_acked_by_peer; 4260 break; 4261 case P_DISCARD_ACK: 4262 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 4263 what = conflict_discarded_by_peer; 4264 break; 4265 default: 4266 D_ASSERT(0); 4267 return FALSE; 4268 } 4269 4270 return validate_req_change_req_state(mdev, p->block_id, sector, 4271 _ack_id_to_req, __func__ , what); 4272 } 4273 4274 static int got_NegAck(struct drbd_conf *mdev, struct p_header80 *h) 4275 { 4276 struct p_block_ack *p = (struct p_block_ack *)h; 4277 sector_t sector = be64_to_cpu(p->sector); 4278 4279 if (__ratelimit(&drbd_ratelimit_state)) 4280 dev_warn(DEV, "Got NegAck packet. Peer is in troubles?\n"); 4281 4282 update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4283 4284 if (is_syncer_block_id(p->block_id)) { 4285 int size = be32_to_cpu(p->blksize); 4286 dec_rs_pending(mdev); 4287 drbd_rs_failed_io(mdev, sector, size); 4288 return TRUE; 4289 } 4290 return validate_req_change_req_state(mdev, p->block_id, sector, 4291 _ack_id_to_req, __func__ , neg_acked); 4292 } 4293 4294 static int got_NegDReply(struct drbd_conf *mdev, struct p_header80 *h) 4295 { 4296 struct p_block_ack *p = (struct p_block_ack *)h; 4297 sector_t sector = be64_to_cpu(p->sector); 4298 4299 update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4300 dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n", 4301 (unsigned long long)sector, be32_to_cpu(p->blksize)); 4302 4303 return validate_req_change_req_state(mdev, p->block_id, sector, 4304 _ar_id_to_req, __func__ , neg_acked); 4305 } 4306 4307 static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header80 *h) 4308 { 4309 sector_t sector; 4310 int size; 4311 struct p_block_ack *p = (struct p_block_ack *)h; 4312 4313 sector = be64_to_cpu(p->sector); 4314 size = be32_to_cpu(p->blksize); 4315 4316 update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4317 4318 dec_rs_pending(mdev); 4319 4320 if (get_ldev_if_state(mdev, D_FAILED)) { 4321 drbd_rs_complete_io(mdev, sector); 4322 drbd_rs_failed_io(mdev, sector, size); 4323 put_ldev(mdev); 4324 } 4325 4326 return TRUE; 4327 } 4328 4329 static int got_BarrierAck(struct drbd_conf *mdev, struct p_header80 *h) 4330 { 4331 struct p_barrier_ack *p = (struct p_barrier_ack *)h; 4332 4333 tl_release(mdev, p->barrier, be32_to_cpu(p->set_size)); 4334 4335 return TRUE; 4336 } 4337 4338 static int got_OVResult(struct drbd_conf *mdev, struct p_header80 *h) 4339 { 4340 struct p_block_ack *p = (struct p_block_ack *)h; 4341 struct drbd_work *w; 4342 sector_t sector; 4343 int size; 4344 4345 sector = be64_to_cpu(p->sector); 4346 size = be32_to_cpu(p->blksize); 4347 4348 update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4349 4350 if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) 4351 drbd_ov_oos_found(mdev, sector, size); 4352 else 4353 ov_oos_print(mdev); 4354 4355 if (!get_ldev(mdev)) 4356 return TRUE; 4357 4358 drbd_rs_complete_io(mdev, sector); 4359 dec_rs_pending(mdev); 4360 4361 if (--mdev->ov_left == 0) { 4362 w = kmalloc(sizeof(*w), GFP_NOIO); 4363 if (w) { 4364 w->cb = w_ov_finished; 4365 drbd_queue_work_front(&mdev->data.work, w); 4366 } else { 4367 dev_err(DEV, "kmalloc(w) failed."); 4368 ov_oos_print(mdev); 4369 drbd_resync_finished(mdev); 4370 } 4371 } 4372 put_ldev(mdev); 4373 return TRUE; 4374 } 4375 4376 static int got_skip(struct drbd_conf *mdev, struct p_header80 *h) 4377 { 4378 return TRUE; 4379 } 4380 4381 struct asender_cmd { 4382 size_t pkt_size; 4383 int (*process)(struct drbd_conf *mdev, struct p_header80 *h); 4384 }; 4385 4386 static struct asender_cmd *get_asender_cmd(int cmd) 4387 { 4388 static struct asender_cmd asender_tbl[] = { 4389 /* anything missing from this table is in 4390 * the drbd_cmd_handler (drbd_default_handler) table, 4391 * see the beginning of drbdd() */ 4392 [P_PING] = { sizeof(struct p_header80), got_Ping }, 4393 [P_PING_ACK] = { sizeof(struct p_header80), got_PingAck }, 4394 [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4395 [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4396 [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4397 [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4398 [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, 4399 [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, 4400 [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply}, 4401 [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, 4402 [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, 4403 [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, 4404 [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, 4405 [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, 4406 [P_MAX_CMD] = { 0, NULL }, 4407 }; 4408 if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL) 4409 return NULL; 4410 return &asender_tbl[cmd]; 4411 } 4412 4413 int drbd_asender(struct drbd_thread *thi) 4414 { 4415 struct drbd_conf *mdev = thi->mdev; 4416 struct p_header80 *h = &mdev->meta.rbuf.header.h80; 4417 struct asender_cmd *cmd = NULL; 4418 4419 int rv, len; 4420 void *buf = h; 4421 int received = 0; 4422 int expect = sizeof(struct p_header80); 4423 int empty; 4424 4425 sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev)); 4426 4427 current->policy = SCHED_RR; /* Make this a realtime task! */ 4428 current->rt_priority = 2; /* more important than all other tasks */ 4429 4430 while (get_t_state(thi) == Running) { 4431 drbd_thread_current_set_cpu(mdev); 4432 if (test_and_clear_bit(SEND_PING, &mdev->flags)) { 4433 ERR_IF(!drbd_send_ping(mdev)) goto reconnect; 4434 mdev->meta.socket->sk->sk_rcvtimeo = 4435 mdev->net_conf->ping_timeo*HZ/10; 4436 } 4437 4438 /* conditionally cork; 4439 * it may hurt latency if we cork without much to send */ 4440 if (!mdev->net_conf->no_cork && 4441 3 < atomic_read(&mdev->unacked_cnt)) 4442 drbd_tcp_cork(mdev->meta.socket); 4443 while (1) { 4444 clear_bit(SIGNAL_ASENDER, &mdev->flags); 4445 flush_signals(current); 4446 if (!drbd_process_done_ee(mdev)) 4447 goto reconnect; 4448 /* to avoid race with newly queued ACKs */ 4449 set_bit(SIGNAL_ASENDER, &mdev->flags); 4450 spin_lock_irq(&mdev->req_lock); 4451 empty = list_empty(&mdev->done_ee); 4452 spin_unlock_irq(&mdev->req_lock); 4453 /* new ack may have been queued right here, 4454 * but then there is also a signal pending, 4455 * and we start over... */ 4456 if (empty) 4457 break; 4458 } 4459 /* but unconditionally uncork unless disabled */ 4460 if (!mdev->net_conf->no_cork) 4461 drbd_tcp_uncork(mdev->meta.socket); 4462 4463 /* short circuit, recv_msg would return EINTR anyways. */ 4464 if (signal_pending(current)) 4465 continue; 4466 4467 rv = drbd_recv_short(mdev, mdev->meta.socket, 4468 buf, expect-received, 0); 4469 clear_bit(SIGNAL_ASENDER, &mdev->flags); 4470 4471 flush_signals(current); 4472 4473 /* Note: 4474 * -EINTR (on meta) we got a signal 4475 * -EAGAIN (on meta) rcvtimeo expired 4476 * -ECONNRESET other side closed the connection 4477 * -ERESTARTSYS (on data) we got a signal 4478 * rv < 0 other than above: unexpected error! 4479 * rv == expected: full header or command 4480 * rv < expected: "woken" by signal during receive 4481 * rv == 0 : "connection shut down by peer" 4482 */ 4483 if (likely(rv > 0)) { 4484 received += rv; 4485 buf += rv; 4486 } else if (rv == 0) { 4487 dev_err(DEV, "meta connection shut down by peer.\n"); 4488 goto reconnect; 4489 } else if (rv == -EAGAIN) { 4490 if (mdev->meta.socket->sk->sk_rcvtimeo == 4491 mdev->net_conf->ping_timeo*HZ/10) { 4492 dev_err(DEV, "PingAck did not arrive in time.\n"); 4493 goto reconnect; 4494 } 4495 set_bit(SEND_PING, &mdev->flags); 4496 continue; 4497 } else if (rv == -EINTR) { 4498 continue; 4499 } else { 4500 dev_err(DEV, "sock_recvmsg returned %d\n", rv); 4501 goto reconnect; 4502 } 4503 4504 if (received == expect && cmd == NULL) { 4505 if (unlikely(h->magic != BE_DRBD_MAGIC)) { 4506 dev_err(DEV, "magic?? on meta m: 0x%08x c: %d l: %d\n", 4507 be32_to_cpu(h->magic), 4508 be16_to_cpu(h->command), 4509 be16_to_cpu(h->length)); 4510 goto reconnect; 4511 } 4512 cmd = get_asender_cmd(be16_to_cpu(h->command)); 4513 len = be16_to_cpu(h->length); 4514 if (unlikely(cmd == NULL)) { 4515 dev_err(DEV, "unknown command?? on meta m: 0x%08x c: %d l: %d\n", 4516 be32_to_cpu(h->magic), 4517 be16_to_cpu(h->command), 4518 be16_to_cpu(h->length)); 4519 goto disconnect; 4520 } 4521 expect = cmd->pkt_size; 4522 ERR_IF(len != expect-sizeof(struct p_header80)) 4523 goto reconnect; 4524 } 4525 if (received == expect) { 4526 D_ASSERT(cmd != NULL); 4527 if (!cmd->process(mdev, h)) 4528 goto reconnect; 4529 4530 buf = h; 4531 received = 0; 4532 expect = sizeof(struct p_header80); 4533 cmd = NULL; 4534 } 4535 } 4536 4537 if (0) { 4538 reconnect: 4539 drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE)); 4540 drbd_md_sync(mdev); 4541 } 4542 if (0) { 4543 disconnect: 4544 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 4545 drbd_md_sync(mdev); 4546 } 4547 clear_bit(SIGNAL_ASENDER, &mdev->flags); 4548 4549 D_ASSERT(mdev->state.conn < C_CONNECTED); 4550 dev_info(DEV, "asender terminated\n"); 4551 4552 return 0; 4553 } 4554