drivers/block/drbd/drbd_receiver.c

   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/smp_lock.h>
  40 #include <linux/pkt_sched.h>
  41 #define __KERNEL_SYSCALLS__
  42 #include <linux/unistd.h>
  43 #include <linux/vmalloc.h>
  44 #include <linux/random.h>
  45 #include <linux/string.h>
  46 #include <linux/scatterlist.h>
  47 #include "drbd_int.h"
  48 #include "drbd_req.h"
  49
  50 #include "drbd_vli.h"
  51
  52 struct flush_work {
  53         struct drbd_work w;
  54         struct drbd_epoch *epoch;
  55 };
  56
  57 enum finish_epoch {
  58         FE_STILL_LIVE,
  59         FE_DESTROYED,
  60         FE_RECYCLED,
  61 };
  62
  63 static int drbd_do_handshake(struct drbd_conf *mdev);
  64 static int drbd_do_auth(struct drbd_conf *mdev);
  65
  66 static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event);
  67 static int e_end_block(struct drbd_conf *, struct drbd_work *, int);
  68
  69 static struct drbd_epoch *previous_epoch(struct drbd_conf *mdev, struct drbd_epoch *epoch)
  70 {
  71         struct drbd_epoch *prev;
  72         spin_lock(&mdev->epoch_lock);
  73         prev = list_entry(epoch->list.prev, struct drbd_epoch, list);
  74         if (prev == epoch || prev == mdev->current_epoch)
  75                 prev = NULL;
  76         spin_unlock(&mdev->epoch_lock);
  77         return prev;
  78 }
  79
  80 #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN)
  81
  82 /*
  83  * some helper functions to deal with single linked page lists,
  84  * page->private being our "next" pointer.
  85  */
  86
  87 /* If at least n pages are linked at head, get n pages off.
  88  * Otherwise, don't modify head, and return NULL.
  89  * Locking is the responsibility of the caller.
  90  */
  91 static struct page *page_chain_del(struct page **head, int n)
  92 {
  93         struct page *page;
  94         struct page *tmp;
  95
  96         BUG_ON(!n);
  97         BUG_ON(!head);
  98
  99         page = *head;
 100
 101         if (!page)
 102                 return NULL;
 103
 104         while (page) {
 105                 tmp = page_chain_next(page);
 106                 if (--n == 0)
 107                         break; /* found sufficient pages */
 108                 if (tmp == NULL)
 109                         /* insufficient pages, don't use any of them. */
 110                         return NULL;
 111                 page = tmp;
 112         }
 113
 114         /* add end of list marker for the returned list */
 115         set_page_private(page, 0);
 116         /* actual return value, and adjustment of head */
 117         page = *head;
 118         *head = tmp;
 119         return page;
 120 }
 121
 122 /* may be used outside of locks to find the tail of a (usually short)
 123  * "private" page chain, before adding it back to a global chain head
 124  * with page_chain_add() under a spinlock. */
 125 static struct page *page_chain_tail(struct page *page, int *len)
 126 {
 127         struct page *tmp;
 128         int i = 1;
 129         while ((tmp = page_chain_next(page)))
 130                 ++i, page = tmp;
 131         if (len)
 132                 *len = i;
 133         return page;
 134 }
 135
 136 static int page_chain_free(struct page *page)
 137 {
 138         struct page *tmp;
 139         int i = 0;
 140         page_chain_for_each_safe(page, tmp) {
 141                 put_page(page);
 142                 ++i;
 143         }
 144         return i;
 145 }
 146
 147 static void page_chain_add(struct page **head,
 148                 struct page *chain_first, struct page *chain_last)
 149 {
 150 #if 1
 151         struct page *tmp;
 152         tmp = page_chain_tail(chain_first, NULL);
 153         BUG_ON(tmp != chain_last);
 154 #endif
 155
 156         /* add chain to head */
 157         set_page_private(chain_last, (unsigned long)*head);
 158         *head = chain_first;
 159 }
 160
 161 static struct page *drbd_pp_first_pages_or_try_alloc(struct drbd_conf *mdev, int number)
 162 {
 163         struct page *page = NULL;
 164         struct page *tmp = NULL;
 165         int i = 0;
 166
 167         /* Yes, testing drbd_pp_vacant outside the lock is racy.
 168          * So what. It saves a spin_lock. */
 169         if (drbd_pp_vacant >= number) {
 170                 spin_lock(&drbd_pp_lock);
 171                 page = page_chain_del(&drbd_pp_pool, number);
 172                 if (page)
 173                         drbd_pp_vacant -= number;
 174                 spin_unlock(&drbd_pp_lock);
 175                 if (page)
 176                         return page;
 177         }
 178
 179         /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD
 180          * "criss-cross" setup, that might cause write-out on some other DRBD,
 181          * which in turn might block on the other node at this very place.  */
 182         for (i = 0; i < number; i++) {
 183                 tmp = alloc_page(GFP_TRY);
 184                 if (!tmp)
 185                         break;
 186                 set_page_private(tmp, (unsigned long)page);
 187                 page = tmp;
 188         }
 189
 190         if (i == number)
 191                 return page;
 192
 193         /* Not enough pages immediately available this time.
 194          * No need to jump around here, drbd_pp_alloc will retry this
 195          * function "soon". */
 196         if (page) {
 197                 tmp = page_chain_tail(page, NULL);
 198                 spin_lock(&drbd_pp_lock);
 199                 page_chain_add(&drbd_pp_pool, page, tmp);
 200                 drbd_pp_vacant += i;
 201                 spin_unlock(&drbd_pp_lock);
 202         }
 203         return NULL;
 204 }
 205
 206 /* kick lower level device, if we have more than (arbitrary number)
 207  * reference counts on it, which typically are locally submitted io
 208  * requests.  don't use unacked_cnt, so we speed up proto A and B, too. */
 209 static void maybe_kick_lo(struct drbd_conf *mdev)
 210 {
 211         if (atomic_read(&mdev->local_cnt) >= mdev->net_conf->unplug_watermark)
 212                 drbd_kick_lo(mdev);
 213 }
 214
 215 static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed)
 216 {
 217         struct drbd_epoch_entry *e;
 218         struct list_head *le, *tle;
 219
 220         /* The EEs are always appended to the end of the list. Since
 221            they are sent in order over the wire, they have to finish
 222            in order. As soon as we see the first not finished we can
 223            stop to examine the list... */
 224
 225         list_for_each_safe(le, tle, &mdev->net_ee) {
 226                 e = list_entry(le, struct drbd_epoch_entry, w.list);
 227                 if (drbd_ee_has_active_page(e))
 228                         break;
 229                 list_move(le, to_be_freed);
 230         }
 231 }
 232
 233 static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev)
 234 {
 235         LIST_HEAD(reclaimed);
 236         struct drbd_epoch_entry *e, *t;
 237
 238         maybe_kick_lo(mdev);
 239         spin_lock_irq(&mdev->req_lock);
 240         reclaim_net_ee(mdev, &reclaimed);
 241         spin_unlock_irq(&mdev->req_lock);
 242
 243         list_for_each_entry_safe(e, t, &reclaimed, w.list)
 244                 drbd_free_net_ee(mdev, e);
 245 }
 246
 247 /**
 248  * drbd_pp_alloc() - Returns @number pages, retries forever (or until signalled)
 249  * @mdev:       DRBD device.
 250  * @number:     number of pages requested
 251  * @retry:      whether to retry, if not enough pages are available right now
 252  *
 253  * Tries to allocate number pages, first from our own page pool, then from
 254  * the kernel, unless this allocation would exceed the max_buffers setting.
 255  * Possibly retry until DRBD frees sufficient pages somewhere else.
 256  *
 257  * Returns a page chain linked via page->private.
 258  */
 259 static struct page *drbd_pp_alloc(struct drbd_conf *mdev, unsigned number, bool retry)
 260 {
 261         struct page *page = NULL;
 262         DEFINE_WAIT(wait);
 263
 264         /* Yes, we may run up to @number over max_buffers. If we
 265          * follow it strictly, the admin will get it wrong anyways. */
 266         if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers)
 267                 page = drbd_pp_first_pages_or_try_alloc(mdev, number);
 268
 269         while (page == NULL) {
 270                 prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE);
 271
 272                 drbd_kick_lo_and_reclaim_net(mdev);
 273
 274                 if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) {
 275                         page = drbd_pp_first_pages_or_try_alloc(mdev, number);
 276                         if (page)
 277                                 break;
 278                 }
 279
 280                 if (!retry)
 281                         break;
 282
 283                 if (signal_pending(current)) {
 284                         dev_warn(DEV, "drbd_pp_alloc interrupted!\n");
 285                         break;
 286                 }
 287
 288                 schedule();
 289         }
 290         finish_wait(&drbd_pp_wait, &wait);
 291
 292         if (page)
 293                 atomic_add(number, &mdev->pp_in_use);
 294         return page;
 295 }
 296
 297 /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc.
 298  * Is also used from inside an other spin_lock_irq(&mdev->req_lock);
 299  * Either links the page chain back to the global pool,
 300  * or returns all pages to the system. */
 301 static void drbd_pp_free(struct drbd_conf *mdev, struct page *page, int is_net)
 302 {
 303         atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use;
 304         int i;
 305
 306         if (drbd_pp_vacant > (DRBD_MAX_SEGMENT_SIZE/PAGE_SIZE)*minor_count)
 307                 i = page_chain_free(page);
 308         else {
 309                 struct page *tmp;
 310                 tmp = page_chain_tail(page, &i);
 311                 spin_lock(&drbd_pp_lock);
 312                 page_chain_add(&drbd_pp_pool, page, tmp);
 313                 drbd_pp_vacant += i;
 314                 spin_unlock(&drbd_pp_lock);
 315         }
 316         i = atomic_sub_return(i, a);
 317         if (i < 0)
 318                 dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n",
 319                         is_net ? "pp_in_use_by_net" : "pp_in_use", i);
 320         wake_up(&drbd_pp_wait);
 321 }
 322
 323 /*
 324 You need to hold the req_lock:
 325  _drbd_wait_ee_list_empty()
 326
 327 You must not have the req_lock:
 328  drbd_free_ee()
 329  drbd_alloc_ee()
 330  drbd_init_ee()
 331  drbd_release_ee()
 332  drbd_ee_fix_bhs()
 333  drbd_process_done_ee()
 334  drbd_clear_done_ee()
 335  drbd_wait_ee_list_empty()
 336 */
 337
 338 struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev,
 339                                      u64 id,
 340                                      sector_t sector,
 341                                      unsigned int data_size,
 342                                      gfp_t gfp_mask) __must_hold(local)
 343 {
 344         struct drbd_epoch_entry *e;
 345         struct page *page;
 346         unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT;
 347
 348         if (FAULT_ACTIVE(mdev, DRBD_FAULT_AL_EE))
 349                 return NULL;
 350
 351         e = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM);
 352         if (!e) {
 353                 if (!(gfp_mask & __GFP_NOWARN))
 354                         dev_err(DEV, "alloc_ee: Allocation of an EE failed\n");
 355                 return NULL;
 356         }
 357
 358         page = drbd_pp_alloc(mdev, nr_pages, (gfp_mask & __GFP_WAIT));
 359         if (!page)
 360                 goto fail;
 361
 362         INIT_HLIST_NODE(&e->colision);
 363         e->epoch = NULL;
 364         e->mdev = mdev;
 365         e->pages = page;
 366         atomic_set(&e->pending_bios, 0);
 367         e->size = data_size;
 368         e->flags = 0;
 369         e->sector = sector;
 370         e->block_id = id;
 371
 372         return e;
 373
 374  fail:
 375         mempool_free(e, drbd_ee_mempool);
 376         return NULL;
 377 }
 378
 379 void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, int is_net)
 380 {
 381         if (e->flags & EE_HAS_DIGEST)
 382                 kfree(e->digest);
 383         drbd_pp_free(mdev, e->pages, is_net);
 384         D_ASSERT(atomic_read(&e->pending_bios) == 0);
 385         D_ASSERT(hlist_unhashed(&e->colision));
 386         mempool_free(e, drbd_ee_mempool);
 387 }
 388
 389 int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list)
 390 {
 391         LIST_HEAD(work_list);
 392         struct drbd_epoch_entry *e, *t;
 393         int count = 0;
 394         int is_net = list == &mdev->net_ee;
 395
 396         spin_lock_irq(&mdev->req_lock);
 397         list_splice_init(list, &work_list);
 398         spin_unlock_irq(&mdev->req_lock);
 399
 400         list_for_each_entry_safe(e, t, &work_list, w.list) {
 401                 drbd_free_some_ee(mdev, e, is_net);
 402                 count++;
 403         }
 404         return count;
 405 }
 406
 407
 408 /*
 409  * This function is called from _asender only_
 410  * but see also comments in _req_mod(,barrier_acked)
 411  * and receive_Barrier.
 412  *
 413  * Move entries from net_ee to done_ee, if ready.
 414  * Grab done_ee, call all callbacks, free the entries.
 415  * The callbacks typically send out ACKs.
 416  */
 417 static int drbd_process_done_ee(struct drbd_conf *mdev)
 418 {
 419         LIST_HEAD(work_list);
 420         LIST_HEAD(reclaimed);
 421         struct drbd_epoch_entry *e, *t;
 422         int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS);
 423
 424         spin_lock_irq(&mdev->req_lock);
 425         reclaim_net_ee(mdev, &reclaimed);
 426         list_splice_init(&mdev->done_ee, &work_list);
 427         spin_unlock_irq(&mdev->req_lock);
 428
 429         list_for_each_entry_safe(e, t, &reclaimed, w.list)
 430                 drbd_free_net_ee(mdev, e);
 431
 432         /* possible callbacks here:
 433          * e_end_block, and e_end_resync_block, e_send_discard_ack.
 434          * all ignore the last argument.
 435          */
 436         list_for_each_entry_safe(e, t, &work_list, w.list) {
 437                 /* list_del not necessary, next/prev members not touched */
 438                 ok = e->w.cb(mdev, &e->w, !ok) && ok;
 439                 drbd_free_ee(mdev, e);
 440         }
 441         wake_up(&mdev->ee_wait);
 442
 443         return ok;
 444 }
 445
 446 void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head)
 447 {
 448         DEFINE_WAIT(wait);
 449
 450         /* avoids spin_lock/unlock
 451          * and calling prepare_to_wait in the fast path */
 452         while (!list_empty(head)) {
 453                 prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE);
 454                 spin_unlock_irq(&mdev->req_lock);
 455                 drbd_kick_lo(mdev);
 456                 schedule();
 457                 finish_wait(&mdev->ee_wait, &wait);
 458                 spin_lock_irq(&mdev->req_lock);
 459         }
 460 }
 461
 462 void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head)
 463 {
 464         spin_lock_irq(&mdev->req_lock);
 465         _drbd_wait_ee_list_empty(mdev, head);
 466         spin_unlock_irq(&mdev->req_lock);
 467 }
 468
 469 /* see also kernel_accept; which is only present since 2.6.18.
 470  * also we want to log which part of it failed, exactly */
 471 static int drbd_accept(struct drbd_conf *mdev, const char **what,
 472                 struct socket *sock, struct socket **newsock)
 473 {
 474         struct sock *sk = sock->sk;
 475         int err = 0;
 476
 477         *what = "listen";
 478         err = sock->ops->listen(sock, 5);
 479         if (err < 0)
 480                 goto out;
 481
 482         *what = "sock_create_lite";
 483         err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
 484                                newsock);
 485         if (err < 0)
 486                 goto out;
 487
 488         *what = "accept";
 489         err = sock->ops->accept(sock, *newsock, 0);
 490         if (err < 0) {
 491                 sock_release(*newsock);
 492                 *newsock = NULL;
 493                 goto out;
 494         }
 495         (*newsock)->ops  = sock->ops;
 496
 497 out:
 498         return err;
 499 }
 500
 501 static int drbd_recv_short(struct drbd_conf *mdev, struct socket *sock,
 502                     void *buf, size_t size, int flags)
 503 {
 504         mm_segment_t oldfs;
 505         struct kvec iov = {
 506                 .iov_base = buf,
 507                 .iov_len = size,
 508         };
 509         struct msghdr msg = {
 510                 .msg_iovlen = 1,
 511                 .msg_iov = (struct iovec *)&iov,
 512                 .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL)
 513         };
 514         int rv;
 515
 516         oldfs = get_fs();
 517         set_fs(KERNEL_DS);
 518         rv = sock_recvmsg(sock, &msg, size, msg.msg_flags);
 519         set_fs(oldfs);
 520
 521         return rv;
 522 }
 523
 524 static int drbd_recv(struct drbd_conf *mdev, void *buf, size_t size)
 525 {
 526         mm_segment_t oldfs;
 527         struct kvec iov = {
 528                 .iov_base = buf,
 529                 .iov_len = size,
 530         };
 531         struct msghdr msg = {
 532                 .msg_iovlen = 1,
 533                 .msg_iov = (struct iovec *)&iov,
 534                 .msg_flags = MSG_WAITALL | MSG_NOSIGNAL
 535         };
 536         int rv;
 537
 538         oldfs = get_fs();
 539         set_fs(KERNEL_DS);
 540
 541         for (;;) {
 542                 rv = sock_recvmsg(mdev->data.socket, &msg, size, msg.msg_flags);
 543                 if (rv == size)
 544                         break;
 545
 546                 /* Note:
 547                  * ECONNRESET   other side closed the connection
 548                  * ERESTARTSYS  (on  sock) we got a signal
 549                  */
 550
 551                 if (rv < 0) {
 552                         if (rv == -ECONNRESET)
 553                                 dev_info(DEV, "sock was reset by peer\n");
 554                         else if (rv != -ERESTARTSYS)
 555                                 dev_err(DEV, "sock_recvmsg returned %d\n", rv);
 556                         break;
 557                 } else if (rv == 0) {
 558                         dev_info(DEV, "sock was shut down by peer\n");
 559                         break;
 560                 } else  {
 561                         /* signal came in, or peer/link went down,
 562                          * after we read a partial message
 563                          */
 564                         /* D_ASSERT(signal_pending(current)); */
 565                         break;
 566                 }
 567         };
 568
 569         set_fs(oldfs);
 570
 571         if (rv != size)
 572                 drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE));
 573
 574         return rv;
 575 }
 576
 577 /* quoting tcp(7):
 578  *   On individual connections, the socket buffer size must be set prior to the
 579  *   listen(2) or connect(2) calls in order to have it take effect.
 580  * This is our wrapper to do so.
 581  */
 582 static void drbd_setbufsize(struct socket *sock, unsigned int snd,
 583                 unsigned int rcv)
 584 {
 585         /* open coded SO_SNDBUF, SO_RCVBUF */
 586         if (snd) {
 587                 sock->sk->sk_sndbuf = snd;
 588                 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 589         }
 590         if (rcv) {
 591                 sock->sk->sk_rcvbuf = rcv;
 592                 sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 593         }
 594 }
 595
 596 static struct socket *drbd_try_connect(struct drbd_conf *mdev)
 597 {
 598         const char *what;
 599         struct socket *sock;
 600         struct sockaddr_in6 src_in6;
 601         int err;
 602         int disconnect_on_error = 1;
 603
 604         if (!get_net_conf(mdev))
 605                 return NULL;
 606
 607         what = "sock_create_kern";
 608         err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family,
 609                 SOCK_STREAM, IPPROTO_TCP, &sock);
 610         if (err < 0) {
 611                 sock = NULL;
 612                 goto out;
 613         }
 614
 615         sock->sk->sk_rcvtimeo =
 616         sock->sk->sk_sndtimeo =  mdev->net_conf->try_connect_int*HZ;
 617         drbd_setbufsize(sock, mdev->net_conf->sndbuf_size,
 618                         mdev->net_conf->rcvbuf_size);
 619
 620        /* explicitly bind to the configured IP as source IP
 621         *  for the outgoing connections.
 622         *  This is needed for multihomed hosts and to be
 623         *  able to use lo: interfaces for drbd.
 624         * Make sure to use 0 as port number, so linux selects
 625         *  a free one dynamically.
 626         */
 627         memcpy(&src_in6, mdev->net_conf->my_addr,
 628                min_t(int, mdev->net_conf->my_addr_len, sizeof(src_in6)));
 629         if (((struct sockaddr *)mdev->net_conf->my_addr)->sa_family == AF_INET6)
 630                 src_in6.sin6_port = 0;
 631         else
 632                 ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */
 633
 634         what = "bind before connect";
 635         err = sock->ops->bind(sock,
 636                               (struct sockaddr *) &src_in6,
 637                               mdev->net_conf->my_addr_len);
 638         if (err < 0)
 639                 goto out;
 640
 641         /* connect may fail, peer not yet available.
 642          * stay C_WF_CONNECTION, don't go Disconnecting! */
 643         disconnect_on_error = 0;
 644         what = "connect";
 645         err = sock->ops->connect(sock,
 646                                  (struct sockaddr *)mdev->net_conf->peer_addr,
 647                                  mdev->net_conf->peer_addr_len, 0);
 648
 649 out:
 650         if (err < 0) {
 651                 if (sock) {
 652                         sock_release(sock);
 653                         sock = NULL;
 654                 }
 655                 switch (-err) {
 656                         /* timeout, busy, signal pending */
 657                 case ETIMEDOUT: case EAGAIN: case EINPROGRESS:
 658                 case EINTR: case ERESTARTSYS:
 659                         /* peer not (yet) available, network problem */
 660                 case ECONNREFUSED: case ENETUNREACH:
 661                 case EHOSTDOWN:    case EHOSTUNREACH:
 662                         disconnect_on_error = 0;
 663                         break;
 664                 default:
 665                         dev_err(DEV, "%s failed, err = %d\n", what, err);
 666                 }
 667                 if (disconnect_on_error)
 668                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
 669         }
 670         put_net_conf(mdev);
 671         return sock;
 672 }
 673
 674 static struct socket *drbd_wait_for_connect(struct drbd_conf *mdev)
 675 {
 676         int timeo, err;
 677         struct socket *s_estab = NULL, *s_listen;
 678         const char *what;
 679
 680         if (!get_net_conf(mdev))
 681                 return NULL;
 682
 683         what = "sock_create_kern";
 684         err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family,
 685                 SOCK_STREAM, IPPROTO_TCP, &s_listen);
 686         if (err) {
 687                 s_listen = NULL;
 688                 goto out;
 689         }
 690
 691         timeo = mdev->net_conf->try_connect_int * HZ;
 692         timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */
 693
 694         s_listen->sk->sk_reuse    = 1; /* SO_REUSEADDR */
 695         s_listen->sk->sk_rcvtimeo = timeo;
 696         s_listen->sk->sk_sndtimeo = timeo;
 697         drbd_setbufsize(s_listen, mdev->net_conf->sndbuf_size,
 698                         mdev->net_conf->rcvbuf_size);
 699
 700         what = "bind before listen";
 701         err = s_listen->ops->bind(s_listen,
 702                               (struct sockaddr *) mdev->net_conf->my_addr,
 703                               mdev->net_conf->my_addr_len);
 704         if (err < 0)
 705                 goto out;
 706
 707         err = drbd_accept(mdev, &what, s_listen, &s_estab);
 708
 709 out:
 710         if (s_listen)
 711                 sock_release(s_listen);
 712         if (err < 0) {
 713                 if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) {
 714                         dev_err(DEV, "%s failed, err = %d\n", what, err);
 715                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
 716                 }
 717         }
 718         put_net_conf(mdev);
 719
 720         return s_estab;
 721 }
 722
 723 static int drbd_send_fp(struct drbd_conf *mdev,
 724         struct socket *sock, enum drbd_packets cmd)
 725 {
 726         struct p_header80 *h = &mdev->data.sbuf.header.h80;
 727
 728         return _drbd_send_cmd(mdev, sock, cmd, h, sizeof(*h), 0);
 729 }
 730
 731 static enum drbd_packets drbd_recv_fp(struct drbd_conf *mdev, struct socket *sock)
 732 {
 733         struct p_header80 *h = &mdev->data.rbuf.header.h80;
 734         int rr;
 735
 736         rr = drbd_recv_short(mdev, sock, h, sizeof(*h), 0);
 737
 738         if (rr == sizeof(*h) && h->magic == BE_DRBD_MAGIC)
 739                 return be16_to_cpu(h->command);
 740
 741         return 0xffff;
 742 }
 743
 744 /**
 745  * drbd_socket_okay() - Free the socket if its connection is not okay
 746  * @mdev:       DRBD device.
 747  * @sock:       pointer to the pointer to the socket.
 748  */
 749 static int drbd_socket_okay(struct drbd_conf *mdev, struct socket **sock)
 750 {
 751         int rr;
 752         char tb[4];
 753
 754         if (!*sock)
 755                 return FALSE;
 756
 757         rr = drbd_recv_short(mdev, *sock, tb, 4, MSG_DONTWAIT | MSG_PEEK);
 758
 759         if (rr > 0 || rr == -EAGAIN) {
 760                 return TRUE;
 761         } else {
 762                 sock_release(*sock);
 763                 *sock = NULL;
 764                 return FALSE;
 765         }
 766 }
 767
 768 /*
 769  * return values:
 770  *   1 yes, we have a valid connection
 771  *   0 oops, did not work out, please try again
 772  *  -1 peer talks different language,
 773  *     no point in trying again, please go standalone.
 774  *  -2 We do not have a network config...
 775  */
 776 static int drbd_connect(struct drbd_conf *mdev)
 777 {
 778         struct socket *s, *sock, *msock;
 779         int try, h, ok;
 780
 781         D_ASSERT(!mdev->data.socket);
 782
 783         if (drbd_request_state(mdev, NS(conn, C_WF_CONNECTION)) < SS_SUCCESS)
 784                 return -2;
 785
 786         clear_bit(DISCARD_CONCURRENT, &mdev->flags);
 787
 788         sock  = NULL;
 789         msock = NULL;
 790
 791         do {
 792                 for (try = 0;;) {
 793                         /* 3 tries, this should take less than a second! */
 794                         s = drbd_try_connect(mdev);
 795                         if (s || ++try >= 3)
 796                                 break;
 797                         /* give the other side time to call bind() & listen() */
 798                         __set_current_state(TASK_INTERRUPTIBLE);
 799                         schedule_timeout(HZ / 10);
 800                 }
 801
 802                 if (s) {
 803                         if (!sock) {
 804                                 drbd_send_fp(mdev, s, P_HAND_SHAKE_S);
 805                                 sock = s;
 806                                 s = NULL;
 807                         } else if (!msock) {
 808                                 drbd_send_fp(mdev, s, P_HAND_SHAKE_M);
 809                                 msock = s;
 810                                 s = NULL;
 811                         } else {
 812                                 dev_err(DEV, "Logic error in drbd_connect()\n");
 813                                 goto out_release_sockets;
 814                         }
 815                 }
 816
 817                 if (sock && msock) {
 818                         __set_current_state(TASK_INTERRUPTIBLE);
 819                         schedule_timeout(HZ / 10);
 820                         ok = drbd_socket_okay(mdev, &sock);
 821                         ok = drbd_socket_okay(mdev, &msock) && ok;
 822                         if (ok)
 823                                 break;
 824                 }
 825
 826 retry:
 827                 s = drbd_wait_for_connect(mdev);
 828                 if (s) {
 829                         try = drbd_recv_fp(mdev, s);
 830                         drbd_socket_okay(mdev, &sock);
 831                         drbd_socket_okay(mdev, &msock);
 832                         switch (try) {
 833                         case P_HAND_SHAKE_S:
 834                                 if (sock) {
 835                                         dev_warn(DEV, "initial packet S crossed\n");
 836                                         sock_release(sock);
 837                                 }
 838                                 sock = s;
 839                                 break;
 840                         case P_HAND_SHAKE_M:
 841                                 if (msock) {
 842                                         dev_warn(DEV, "initial packet M crossed\n");
 843                                         sock_release(msock);
 844                                 }
 845                                 msock = s;
 846                                 set_bit(DISCARD_CONCURRENT, &mdev->flags);
 847                                 break;
 848                         default:
 849                                 dev_warn(DEV, "Error receiving initial packet\n");
 850                                 sock_release(s);
 851                                 if (random32() & 1)
 852                                         goto retry;
 853                         }
 854                 }
 855
 856                 if (mdev->state.conn <= C_DISCONNECTING)
 857                         goto out_release_sockets;
 858                 if (signal_pending(current)) {
 859                         flush_signals(current);
 860                         smp_rmb();
 861                         if (get_t_state(&mdev->receiver) == Exiting)
 862                                 goto out_release_sockets;
 863                 }
 864
 865                 if (sock && msock) {
 866                         ok = drbd_socket_okay(mdev, &sock);
 867                         ok = drbd_socket_okay(mdev, &msock) && ok;
 868                         if (ok)
 869                                 break;
 870                 }
 871         } while (1);
 872
 873         msock->sk->sk_reuse = 1; /* SO_REUSEADDR */
 874         sock->sk->sk_reuse = 1; /* SO_REUSEADDR */
 875
 876         sock->sk->sk_allocation = GFP_NOIO;
 877         msock->sk->sk_allocation = GFP_NOIO;
 878
 879         sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK;
 880         msock->sk->sk_priority = TC_PRIO_INTERACTIVE;
 881
 882         /* NOT YET ...
 883          * sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
 884          * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
 885          * first set it to the P_HAND_SHAKE timeout,
 886          * which we set to 4x the configured ping_timeout. */
 887         sock->sk->sk_sndtimeo =
 888         sock->sk->sk_rcvtimeo = mdev->net_conf->ping_timeo*4*HZ/10;
 889
 890         msock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
 891         msock->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ;
 892
 893         /* we don't want delays.
 894          * we use TCP_CORK where apropriate, though */
 895         drbd_tcp_nodelay(sock);
 896         drbd_tcp_nodelay(msock);
 897
 898         mdev->data.socket = sock;
 899         mdev->meta.socket = msock;
 900         mdev->last_received = jiffies;
 901
 902         D_ASSERT(mdev->asender.task == NULL);
 903
 904         h = drbd_do_handshake(mdev);
 905         if (h <= 0)
 906                 return h;
 907
 908         if (mdev->cram_hmac_tfm) {
 909                 /* drbd_request_state(mdev, NS(conn, WFAuth)); */
 910                 switch (drbd_do_auth(mdev)) {
 911                 case -1:
 912                         dev_err(DEV, "Authentication of peer failed\n");
 913                         return -1;
 914                 case 0:
 915                         dev_err(DEV, "Authentication of peer failed, trying again.\n");
 916                         return 0;
 917                 }
 918         }
 919
 920         if (drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS)) < SS_SUCCESS)
 921                 return 0;
 922
 923         sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
 924         sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
 925
 926         atomic_set(&mdev->packet_seq, 0);
 927         mdev->peer_seq = 0;
 928
 929         drbd_thread_start(&mdev->asender);
 930
 931         if (mdev->agreed_pro_version < 95 && get_ldev(mdev)) {
 932                 drbd_setup_queue_param(mdev, DRBD_MAX_SIZE_H80_PACKET);
 933                 put_ldev(mdev);
 934         }
 935
 936         if (!drbd_send_protocol(mdev))
 937                 return -1;
 938         drbd_send_sync_param(mdev, &mdev->sync_conf);
 939         drbd_send_sizes(mdev, 0, 0);
 940         drbd_send_uuids(mdev);
 941         drbd_send_state(mdev);
 942         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
 943         clear_bit(RESIZE_PENDING, &mdev->flags);
 944
 945         return 1;
 946
 947 out_release_sockets:
 948         if (sock)
 949                 sock_release(sock);
 950         if (msock)
 951                 sock_release(msock);
 952         return -1;
 953 }
 954
 955 static int drbd_recv_header(struct drbd_conf *mdev, enum drbd_packets *cmd, unsigned int *packet_size)
 956 {
 957         union p_header *h = &mdev->data.rbuf.header;
 958         int r;
 959
 960         r = drbd_recv(mdev, h, sizeof(*h));
 961         if (unlikely(r != sizeof(*h))) {
 962                 dev_err(DEV, "short read expecting header on sock: r=%d\n", r);
 963                 return FALSE;
 964         }
 965
 966         if (likely(h->h80.magic == BE_DRBD_MAGIC)) {
 967                 *cmd = be16_to_cpu(h->h80.command);
 968                 *packet_size = be16_to_cpu(h->h80.length);
 969         } else if (h->h95.magic == BE_DRBD_MAGIC_BIG) {
 970                 *cmd = be16_to_cpu(h->h95.command);
 971                 *packet_size = be32_to_cpu(h->h95.length);
 972         } else {
 973                 dev_err(DEV, "magic?? on data m: 0x%08x c: %d l: %d\n",
 974                     be32_to_cpu(h->h80.magic),
 975                     be16_to_cpu(h->h80.command),
 976                     be16_to_cpu(h->h80.length));
 977                 return FALSE;
 978         }
 979         mdev->last_received = jiffies;
 980
 981         return TRUE;
 982 }
 983
 984 static enum finish_epoch drbd_flush_after_epoch(struct drbd_conf *mdev, struct drbd_epoch *epoch)
 985 {
 986         int rv;
 987
 988         if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) {
 989                 rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL,
 990                                         NULL, BLKDEV_IFL_WAIT);
 991                 if (rv) {
 992                         dev_err(DEV, "local disk flush failed with status %d\n", rv);
 993                         /* would rather check on EOPNOTSUPP, but that is not reliable.
 994                          * don't try again for ANY return value != 0
 995                          * if (rv == -EOPNOTSUPP) */
 996                         drbd_bump_write_ordering(mdev, WO_drain_io);
 997                 }
 998                 put_ldev(mdev);
 999         }
1000
1001         return drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE);
1002 }
1003
1004 static int w_flush(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1005 {
1006         struct flush_work *fw = (struct flush_work *)w;
1007         struct drbd_epoch *epoch = fw->epoch;
1008
1009         kfree(w);
1010
1011         if (!test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags))
1012                 drbd_flush_after_epoch(mdev, epoch);
1013
1014         drbd_may_finish_epoch(mdev, epoch, EV_PUT |
1015                               (mdev->state.conn < C_CONNECTED ? EV_CLEANUP : 0));
1016
1017         return 1;
1018 }
1019
1020 /**
1021  * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it.
1022  * @mdev:       DRBD device.
1023  * @epoch:      Epoch object.
1024  * @ev:         Epoch event.
1025  */
1026 static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev,
1027                                                struct drbd_epoch *epoch,
1028                                                enum epoch_event ev)
1029 {
1030         int finish, epoch_size;
1031         struct drbd_epoch *next_epoch;
1032         int schedule_flush = 0;
1033         enum finish_epoch rv = FE_STILL_LIVE;
1034
1035         spin_lock(&mdev->epoch_lock);
1036         do {
1037                 next_epoch = NULL;
1038                 finish = 0;
1039
1040                 epoch_size = atomic_read(&epoch->epoch_size);
1041
1042                 switch (ev & ~EV_CLEANUP) {
1043                 case EV_PUT:
1044                         atomic_dec(&epoch->active);
1045                         break;
1046                 case EV_GOT_BARRIER_NR:
1047                         set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags);
1048
1049                         /* Special case: If we just switched from WO_bio_barrier to
1050                            WO_bdev_flush we should not finish the current epoch */
1051                         if (test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags) && epoch_size == 1 &&
1052                             mdev->write_ordering != WO_bio_barrier &&
1053                             epoch == mdev->current_epoch)
1054                                 clear_bit(DE_CONTAINS_A_BARRIER, &epoch->flags);
1055                         break;
1056                 case EV_BARRIER_DONE:
1057                         set_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags);
1058                         break;
1059                 case EV_BECAME_LAST:
1060                         /* nothing to do*/
1061                         break;
1062                 }
1063
1064                 if (epoch_size != 0 &&
1065                     atomic_read(&epoch->active) == 0 &&
1066                     test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) &&
1067                     epoch->list.prev == &mdev->current_epoch->list &&
1068                     !test_bit(DE_IS_FINISHING, &epoch->flags)) {
1069                         /* Nearly all conditions are met to finish that epoch... */
1070                         if (test_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags) ||
1071                             mdev->write_ordering == WO_none ||
1072                             (epoch_size == 1 && test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) ||
1073                             ev & EV_CLEANUP) {
1074                                 finish = 1;
1075                                 set_bit(DE_IS_FINISHING, &epoch->flags);
1076                         } else if (!test_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags) &&
1077                                  mdev->write_ordering == WO_bio_barrier) {
1078                                 atomic_inc(&epoch->active);
1079                                 schedule_flush = 1;
1080                         }
1081                 }
1082                 if (finish) {
1083                         if (!(ev & EV_CLEANUP)) {
1084                                 spin_unlock(&mdev->epoch_lock);
1085                                 drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size);
1086                                 spin_lock(&mdev->epoch_lock);
1087                         }
1088                         dec_unacked(mdev);
1089
1090                         if (mdev->current_epoch != epoch) {
1091                                 next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list);
1092                                 list_del(&epoch->list);
1093                                 ev = EV_BECAME_LAST | (ev & EV_CLEANUP);
1094                                 mdev->epochs--;
1095                                 kfree(epoch);
1096
1097                                 if (rv == FE_STILL_LIVE)
1098                                         rv = FE_DESTROYED;
1099                         } else {
1100                                 epoch->flags = 0;
1101                                 atomic_set(&epoch->epoch_size, 0);
1102                                 /* atomic_set(&epoch->active, 0); is already zero */
1103                                 if (rv == FE_STILL_LIVE)
1104                                         rv = FE_RECYCLED;
1105                         }
1106                 }
1107
1108                 if (!next_epoch)
1109                         break;
1110
1111                 epoch = next_epoch;
1112         } while (1);
1113
1114         spin_unlock(&mdev->epoch_lock);
1115
1116         if (schedule_flush) {
1117                 struct flush_work *fw;
1118                 fw = kmalloc(sizeof(*fw), GFP_ATOMIC);
1119                 if (fw) {
1120                         fw->w.cb = w_flush;
1121                         fw->epoch = epoch;
1122                         drbd_queue_work(&mdev->data.work, &fw->w);
1123                 } else {
1124                         dev_warn(DEV, "Could not kmalloc a flush_work obj\n");
1125                         set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags);
1126                         /* That is not a recursion, only one level */
1127                         drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE);
1128                         drbd_may_finish_epoch(mdev, epoch, EV_PUT);
1129                 }
1130         }
1131
1132         return rv;
1133 }
1134
1135 /**
1136  * drbd_bump_write_ordering() - Fall back to an other write ordering method
1137  * @mdev:       DRBD device.
1138  * @wo:         Write ordering method to try.
1139  */
1140 void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local)
1141 {
1142         enum write_ordering_e pwo;
1143         static char *write_ordering_str[] = {
1144                 [WO_none] = "none",
1145                 [WO_drain_io] = "drain",
1146                 [WO_bdev_flush] = "flush",
1147                 [WO_bio_barrier] = "barrier",
1148         };
1149
1150         pwo = mdev->write_ordering;
1151         wo = min(pwo, wo);
1152         if (wo == WO_bio_barrier && mdev->ldev->dc.no_disk_barrier)
1153                 wo = WO_bdev_flush;
1154         if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush)
1155                 wo = WO_drain_io;
1156         if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain)
1157                 wo = WO_none;
1158         mdev->write_ordering = wo;
1159         if (pwo != mdev->write_ordering || wo == WO_bio_barrier)
1160                 dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]);
1161 }
1162
1163 /**
1164  * drbd_submit_ee()
1165  * @mdev:       DRBD device.
1166  * @e:          epoch entry
1167  * @rw:         flag field, see bio->bi_rw
1168  */
1169 /* TODO allocate from our own bio_set. */
1170 int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e,
1171                 const unsigned rw, const int fault_type)
1172 {
1173         struct bio *bios = NULL;
1174         struct bio *bio;
1175         struct page *page = e->pages;
1176         sector_t sector = e->sector;
1177         unsigned ds = e->size;
1178         unsigned n_bios = 0;
1179         unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT;
1180
1181         /* In most cases, we will only need one bio.  But in case the lower
1182          * level restrictions happen to be different at this offset on this
1183          * side than those of the sending peer, we may need to submit the
1184          * request in more than one bio. */
1185 next_bio:
1186         bio = bio_alloc(GFP_NOIO, nr_pages);
1187         if (!bio) {
1188                 dev_err(DEV, "submit_ee: Allocation of a bio failed\n");
1189                 goto fail;
1190         }
1191         /* > e->sector, unless this is the first bio */
1192         bio->bi_sector = sector;
1193         bio->bi_bdev = mdev->ldev->backing_bdev;
1194         /* we special case some flags in the multi-bio case, see below
1195          * (REQ_UNPLUG, REQ_HARDBARRIER) */
1196         bio->bi_rw = rw;
1197         bio->bi_private = e;
1198         bio->bi_end_io = drbd_endio_sec;
1199
1200         bio->bi_next = bios;
1201         bios = bio;
1202         ++n_bios;
1203
1204         page_chain_for_each(page) {
1205                 unsigned len = min_t(unsigned, ds, PAGE_SIZE);
1206                 if (!bio_add_page(bio, page, len, 0)) {
1207                         /* a single page must always be possible! */
1208                         BUG_ON(bio->bi_vcnt == 0);
1209                         goto next_bio;
1210                 }
1211                 ds -= len;
1212                 sector += len >> 9;
1213                 --nr_pages;
1214         }
1215         D_ASSERT(page == NULL);
1216         D_ASSERT(ds == 0);
1217
1218         atomic_set(&e->pending_bios, n_bios);
1219         do {
1220                 bio = bios;
1221                 bios = bios->bi_next;
1222                 bio->bi_next = NULL;
1223
1224                 /* strip off REQ_UNPLUG unless it is the last bio */
1225                 if (bios)
1226                         bio->bi_rw &= ~REQ_UNPLUG;
1227
1228                 drbd_generic_make_request(mdev, fault_type, bio);
1229
1230                 /* strip off REQ_HARDBARRIER,
1231                  * unless it is the first or last bio */
1232                 if (bios && bios->bi_next)
1233                         bios->bi_rw &= ~REQ_HARDBARRIER;
1234         } while (bios);
1235         maybe_kick_lo(mdev);
1236         return 0;
1237
1238 fail:
1239         while (bios) {
1240                 bio = bios;
1241                 bios = bios->bi_next;
1242                 bio_put(bio);
1243         }
1244         return -ENOMEM;
1245 }
1246
1247 /**
1248  * w_e_reissue() - Worker callback; Resubmit a bio, without REQ_HARDBARRIER set
1249  * @mdev:       DRBD device.
1250  * @w:          work object.
1251  * @cancel:     The connection will be closed anyways (unused in this callback)
1252  */
1253 int w_e_reissue(struct drbd_conf *mdev, struct drbd_work *w, int cancel) __releases(local)
1254 {
1255         struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1256         /* We leave DE_CONTAINS_A_BARRIER and EE_IS_BARRIER in place,
1257            (and DE_BARRIER_IN_NEXT_EPOCH_ISSUED in the previous Epoch)
1258            so that we can finish that epoch in drbd_may_finish_epoch().
1259            That is necessary if we already have a long chain of Epochs, before
1260            we realize that REQ_HARDBARRIER is actually not supported */
1261
1262         /* As long as the -ENOTSUPP on the barrier is reported immediately
1263            that will never trigger. If it is reported late, we will just
1264            print that warning and continue correctly for all future requests
1265            with WO_bdev_flush */
1266         if (previous_epoch(mdev, e->epoch))
1267                 dev_warn(DEV, "Write ordering was not enforced (one time event)\n");
1268
1269         /* we still have a local reference,
1270          * get_ldev was done in receive_Data. */
1271
1272         e->w.cb = e_end_block;
1273         if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_DT_WR) != 0) {
1274                 /* drbd_submit_ee fails for one reason only:
1275                  * if was not able to allocate sufficient bios.
1276                  * requeue, try again later. */
1277                 e->w.cb = w_e_reissue;
1278                 drbd_queue_work(&mdev->data.work, &e->w);
1279         }
1280         return 1;
1281 }
1282
1283 static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1284 {
1285         int rv, issue_flush;
1286         struct p_barrier *p = &mdev->data.rbuf.barrier;
1287         struct drbd_epoch *epoch;
1288
1289         inc_unacked(mdev);
1290
1291         if (mdev->net_conf->wire_protocol != DRBD_PROT_C)
1292                 drbd_kick_lo(mdev);
1293
1294         mdev->current_epoch->barrier_nr = p->barrier;
1295         rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR);
1296
1297         /* P_BARRIER_ACK may imply that the corresponding extent is dropped from
1298          * the activity log, which means it would not be resynced in case the
1299          * R_PRIMARY crashes now.
1300          * Therefore we must send the barrier_ack after the barrier request was
1301          * completed. */
1302         switch (mdev->write_ordering) {
1303         case WO_bio_barrier:
1304         case WO_none:
1305                 if (rv == FE_RECYCLED)
1306                         return TRUE;
1307                 break;
1308
1309         case WO_bdev_flush:
1310         case WO_drain_io:
1311                 if (rv == FE_STILL_LIVE) {
1312                         set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags);
1313                         drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
1314                         rv = drbd_flush_after_epoch(mdev, mdev->current_epoch);
1315                 }
1316                 if (rv == FE_RECYCLED)
1317                         return TRUE;
1318
1319                 /* The asender will send all the ACKs and barrier ACKs out, since
1320                    all EEs moved from the active_ee to the done_ee. We need to
1321                    provide a new epoch object for the EEs that come in soon */
1322                 break;
1323         }
1324
1325         /* receiver context, in the writeout path of the other node.
1326          * avoid potential distributed deadlock */
1327         epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO);
1328         if (!epoch) {
1329                 dev_warn(DEV, "Allocation of an epoch failed, slowing down\n");
1330                 issue_flush = !test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags);
1331                 drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
1332                 if (issue_flush) {
1333                         rv = drbd_flush_after_epoch(mdev, mdev->current_epoch);
1334                         if (rv == FE_RECYCLED)
1335                                 return TRUE;
1336                 }
1337
1338                 drbd_wait_ee_list_empty(mdev, &mdev->done_ee);
1339
1340                 return TRUE;
1341         }
1342
1343         epoch->flags = 0;
1344         atomic_set(&epoch->epoch_size, 0);
1345         atomic_set(&epoch->active, 0);
1346
1347         spin_lock(&mdev->epoch_lock);
1348         if (atomic_read(&mdev->current_epoch->epoch_size)) {
1349                 list_add(&epoch->list, &mdev->current_epoch->list);
1350                 mdev->current_epoch = epoch;
1351                 mdev->epochs++;
1352         } else {
1353                 /* The current_epoch got recycled while we allocated this one... */
1354                 kfree(epoch);
1355         }
1356         spin_unlock(&mdev->epoch_lock);
1357
1358         return TRUE;
1359 }
1360
1361 /* used from receive_RSDataReply (recv_resync_read)
1362  * and from receive_Data */
1363 static struct drbd_epoch_entry *
1364 read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, int data_size) __must_hold(local)
1365 {
1366         const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
1367         struct drbd_epoch_entry *e;
1368         struct page *page;
1369         int dgs, ds, rr;
1370         void *dig_in = mdev->int_dig_in;
1371         void *dig_vv = mdev->int_dig_vv;
1372         unsigned long *data;
1373
1374         dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ?
1375                 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0;
1376
1377         if (dgs) {
1378                 rr = drbd_recv(mdev, dig_in, dgs);
1379                 if (rr != dgs) {
1380                         dev_warn(DEV, "short read receiving data digest: read %d expected %d\n",
1381                              rr, dgs);
1382                         return NULL;
1383                 }
1384         }
1385
1386         data_size -= dgs;
1387
1388         ERR_IF(data_size &  0x1ff) return NULL;
1389         ERR_IF(data_size >  DRBD_MAX_SEGMENT_SIZE) return NULL;
1390
1391         /* even though we trust out peer,
1392          * we sometimes have to double check. */
1393         if (sector + (data_size>>9) > capacity) {
1394                 dev_err(DEV, "capacity: %llus < sector: %llus + size: %u\n",
1395                         (unsigned long long)capacity,
1396                         (unsigned long long)sector, data_size);
1397                 return NULL;
1398         }
1399
1400         /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
1401          * "criss-cross" setup, that might cause write-out on some other DRBD,
1402          * which in turn might block on the other node at this very place.  */
1403         e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO);
1404         if (!e)
1405                 return NULL;
1406
1407         ds = data_size;
1408         page = e->pages;
1409         page_chain_for_each(page) {
1410                 unsigned len = min_t(int, ds, PAGE_SIZE);
1411                 data = kmap(page);
1412                 rr = drbd_recv(mdev, data, len);
1413                 if (FAULT_ACTIVE(mdev, DRBD_FAULT_RECEIVE)) {
1414                         dev_err(DEV, "Fault injection: Corrupting data on receive\n");
1415                         data[0] = data[0] ^ (unsigned long)-1;
1416                 }
1417                 kunmap(page);
1418                 if (rr != len) {
1419                         drbd_free_ee(mdev, e);
1420                         dev_warn(DEV, "short read receiving data: read %d expected %d\n",
1421                              rr, len);
1422                         return NULL;
1423                 }
1424                 ds -= rr;
1425         }
1426
1427         if (dgs) {
1428                 drbd_csum_ee(mdev, mdev->integrity_r_tfm, e, dig_vv);
1429                 if (memcmp(dig_in, dig_vv, dgs)) {
1430                         dev_err(DEV, "Digest integrity check FAILED.\n");
1431                         drbd_bcast_ee(mdev, "digest failed",
1432                                         dgs, dig_in, dig_vv, e);
1433                         drbd_free_ee(mdev, e);
1434                         return NULL;
1435                 }
1436         }
1437         mdev->recv_cnt += data_size>>9;
1438         return e;
1439 }
1440
1441 /* drbd_drain_block() just takes a data block
1442  * out of the socket input buffer, and discards it.
1443  */
1444 static int drbd_drain_block(struct drbd_conf *mdev, int data_size)
1445 {
1446         struct page *page;
1447         int rr, rv = 1;
1448         void *data;
1449
1450         if (!data_size)
1451                 return TRUE;
1452
1453         page = drbd_pp_alloc(mdev, 1, 1);
1454
1455         data = kmap(page);
1456         while (data_size) {
1457                 rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE));
1458                 if (rr != min_t(int, data_size, PAGE_SIZE)) {
1459                         rv = 0;
1460                         dev_warn(DEV, "short read receiving data: read %d expected %d\n",
1461                              rr, min_t(int, data_size, PAGE_SIZE));
1462                         break;
1463                 }
1464                 data_size -= rr;
1465         }
1466         kunmap(page);
1467         drbd_pp_free(mdev, page, 0);
1468         return rv;
1469 }
1470
1471 static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req,
1472                            sector_t sector, int data_size)
1473 {
1474         struct bio_vec *bvec;
1475         struct bio *bio;
1476         int dgs, rr, i, expect;
1477         void *dig_in = mdev->int_dig_in;
1478         void *dig_vv = mdev->int_dig_vv;
1479
1480         dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ?
1481                 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0;
1482
1483         if (dgs) {
1484                 rr = drbd_recv(mdev, dig_in, dgs);
1485                 if (rr != dgs) {
1486                         dev_warn(DEV, "short read receiving data reply digest: read %d expected %d\n",
1487                              rr, dgs);
1488                         return 0;
1489                 }
1490         }
1491
1492         data_size -= dgs;
1493
1494         /* optimistically update recv_cnt.  if receiving fails below,
1495          * we disconnect anyways, and counters will be reset. */
1496         mdev->recv_cnt += data_size>>9;
1497
1498         bio = req->master_bio;
1499         D_ASSERT(sector == bio->bi_sector);
1500
1501         bio_for_each_segment(bvec, bio, i) {
1502                 expect = min_t(int, data_size, bvec->bv_len);
1503                 rr = drbd_recv(mdev,
1504                              kmap(bvec->bv_page)+bvec->bv_offset,
1505                              expect);
1506                 kunmap(bvec->bv_page);
1507                 if (rr != expect) {
1508                         dev_warn(DEV, "short read receiving data reply: "
1509                              "read %d expected %d\n",
1510                              rr, expect);
1511                         return 0;
1512                 }
1513                 data_size -= rr;
1514         }
1515
1516         if (dgs) {
1517                 drbd_csum_bio(mdev, mdev->integrity_r_tfm, bio, dig_vv);
1518                 if (memcmp(dig_in, dig_vv, dgs)) {
1519                         dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n");
1520                         return 0;
1521                 }
1522         }
1523
1524         D_ASSERT(data_size == 0);
1525         return 1;
1526 }
1527
1528 /* e_end_resync_block() is called via
1529  * drbd_process_done_ee() by asender only */
1530 static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1531 {
1532         struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1533         sector_t sector = e->sector;
1534         int ok;
1535
1536         D_ASSERT(hlist_unhashed(&e->colision));
1537
1538         if (likely((e->flags & EE_WAS_ERROR) == 0)) {
1539                 drbd_set_in_sync(mdev, sector, e->size);
1540                 ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e);
1541         } else {
1542                 /* Record failure to sync */
1543                 drbd_rs_failed_io(mdev, sector, e->size);
1544
1545                 ok  = drbd_send_ack(mdev, P_NEG_ACK, e);
1546         }
1547         dec_unacked(mdev);
1548
1549         return ok;
1550 }
1551
1552 static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local)
1553 {
1554         struct drbd_epoch_entry *e;
1555
1556         e = read_in_block(mdev, ID_SYNCER, sector, data_size);
1557         if (!e)
1558                 goto fail;
1559
1560         dec_rs_pending(mdev);
1561
1562         inc_unacked(mdev);
1563         /* corresponding dec_unacked() in e_end_resync_block()
1564          * respective _drbd_clear_done_ee */
1565
1566         e->w.cb = e_end_resync_block;
1567
1568         spin_lock_irq(&mdev->req_lock);
1569         list_add(&e->w.list, &mdev->sync_ee);
1570         spin_unlock_irq(&mdev->req_lock);
1571
1572         atomic_add(data_size >> 9, &mdev->rs_sect_ev);
1573         if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_RS_WR) == 0)
1574                 return TRUE;
1575
1576         /* drbd_submit_ee currently fails for one reason only:
1577          * not being able to allocate enough bios.
1578          * Is dropping the connection going to help? */
1579         spin_lock_irq(&mdev->req_lock);
1580         list_del(&e->w.list);
1581         spin_unlock_irq(&mdev->req_lock);
1582
1583         drbd_free_ee(mdev, e);
1584 fail:
1585         put_ldev(mdev);
1586         return FALSE;
1587 }
1588
1589 static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1590 {
1591         struct drbd_request *req;
1592         sector_t sector;
1593         int ok;
1594         struct p_data *p = &mdev->data.rbuf.data;
1595
1596         sector = be64_to_cpu(p->sector);
1597
1598         spin_lock_irq(&mdev->req_lock);
1599         req = _ar_id_to_req(mdev, p->block_id, sector);
1600         spin_unlock_irq(&mdev->req_lock);
1601         if (unlikely(!req)) {
1602                 dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n");
1603                 return FALSE;
1604         }
1605
1606         /* hlist_del(&req->colision) is done in _req_may_be_done, to avoid
1607          * special casing it there for the various failure cases.
1608          * still no race with drbd_fail_pending_reads */
1609         ok = recv_dless_read(mdev, req, sector, data_size);
1610
1611         if (ok)
1612                 req_mod(req, data_received);
1613         /* else: nothing. handled from drbd_disconnect...
1614          * I don't think we may complete this just yet
1615          * in case we are "on-disconnect: freeze" */
1616
1617         return ok;
1618 }
1619
1620 static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1621 {
1622         sector_t sector;
1623         int ok;
1624         struct p_data *p = &mdev->data.rbuf.data;
1625
1626         sector = be64_to_cpu(p->sector);
1627         D_ASSERT(p->block_id == ID_SYNCER);
1628
1629         if (get_ldev(mdev)) {
1630                 /* data is submitted to disk within recv_resync_read.
1631                  * corresponding put_ldev done below on error,
1632                  * or in drbd_endio_write_sec. */
1633                 ok = recv_resync_read(mdev, sector, data_size);
1634         } else {
1635                 if (__ratelimit(&drbd_ratelimit_state))
1636                         dev_err(DEV, "Can not write resync data to local disk.\n");
1637
1638                 ok = drbd_drain_block(mdev, data_size);
1639
1640                 drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size);
1641         }
1642
1643         atomic_add(data_size >> 9, &mdev->rs_sect_in);
1644
1645         return ok;
1646 }
1647
1648 /* e_end_block() is called via drbd_process_done_ee().
1649  * this means this function only runs in the asender thread
1650  */
1651 static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1652 {
1653         struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1654         sector_t sector = e->sector;
1655         struct drbd_epoch *epoch;
1656         int ok = 1, pcmd;
1657
1658         if (e->flags & EE_IS_BARRIER) {
1659                 epoch = previous_epoch(mdev, e->epoch);
1660                 if (epoch)
1661                         drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE + (cancel ? EV_CLEANUP : 0));
1662         }
1663
1664         if (mdev->net_conf->wire_protocol == DRBD_PROT_C) {
1665                 if (likely((e->flags & EE_WAS_ERROR) == 0)) {
1666                         pcmd = (mdev->state.conn >= C_SYNC_SOURCE &&
1667                                 mdev->state.conn <= C_PAUSED_SYNC_T &&
1668                                 e->flags & EE_MAY_SET_IN_SYNC) ?
1669                                 P_RS_WRITE_ACK : P_WRITE_ACK;
1670                         ok &= drbd_send_ack(mdev, pcmd, e);
1671                         if (pcmd == P_RS_WRITE_ACK)
1672                                 drbd_set_in_sync(mdev, sector, e->size);
1673                 } else {
1674                         ok  = drbd_send_ack(mdev, P_NEG_ACK, e);
1675                         /* we expect it to be marked out of sync anyways...
1676                          * maybe assert this?  */
1677                 }
1678                 dec_unacked(mdev);
1679         }
1680         /* we delete from the conflict detection hash _after_ we sent out the
1681          * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right.  */
1682         if (mdev->net_conf->two_primaries) {
1683                 spin_lock_irq(&mdev->req_lock);
1684                 D_ASSERT(!hlist_unhashed(&e->colision));
1685                 hlist_del_init(&e->colision);
1686                 spin_unlock_irq(&mdev->req_lock);
1687         } else {
1688                 D_ASSERT(hlist_unhashed(&e->colision));
1689         }
1690
1691         drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0));
1692
1693         return ok;
1694 }
1695
1696 static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1697 {
1698         struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1699         int ok = 1;
1700
1701         D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1702         ok = drbd_send_ack(mdev, P_DISCARD_ACK, e);
1703
1704         spin_lock_irq(&mdev->req_lock);
1705         D_ASSERT(!hlist_unhashed(&e->colision));
1706         hlist_del_init(&e->colision);
1707         spin_unlock_irq(&mdev->req_lock);
1708
1709         dec_unacked(mdev);
1710
1711         return ok;
1712 }
1713
1714 /* Called from receive_Data.
1715  * Synchronize packets on sock with packets on msock.
1716  *
1717  * This is here so even when a P_DATA packet traveling via sock overtook an Ack
1718  * packet traveling on msock, they are still processed in the order they have
1719  * been sent.
1720  *
1721  * Note: we don't care for Ack packets overtaking P_DATA packets.
1722  *
1723  * In case packet_seq is larger than mdev->peer_seq number, there are
1724  * outstanding packets on the msock. We wait for them to arrive.
1725  * In case we are the logically next packet, we update mdev->peer_seq
1726  * ourselves. Correctly handles 32bit wrap around.
1727  *
1728  * Assume we have a 10 GBit connection, that is about 1<<30 byte per second,
1729  * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds
1730  * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have
1731  * 1<<9 == 512 seconds aka ages for the 32bit wrap around...
1732  *
1733  * returns 0 if we may process the packet,
1734  * -ERESTARTSYS if we were interrupted (by disconnect signal). */
1735 static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq)
1736 {
1737         DEFINE_WAIT(wait);
1738         unsigned int p_seq;
1739         long timeout;
1740         int ret = 0;
1741         spin_lock(&mdev->peer_seq_lock);
1742         for (;;) {
1743                 prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE);
1744                 if (seq_le(packet_seq, mdev->peer_seq+1))
1745                         break;
1746                 if (signal_pending(current)) {
1747                         ret = -ERESTARTSYS;
1748                         break;
1749                 }
1750                 p_seq = mdev->peer_seq;
1751                 spin_unlock(&mdev->peer_seq_lock);
1752                 timeout = schedule_timeout(30*HZ);
1753                 spin_lock(&mdev->peer_seq_lock);
1754                 if (timeout == 0 && p_seq == mdev->peer_seq) {
1755                         ret = -ETIMEDOUT;
1756                         dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n");
1757                         break;
1758                 }
1759         }
1760         finish_wait(&mdev->seq_wait, &wait);
1761         if (mdev->peer_seq+1 == packet_seq)
1762                 mdev->peer_seq++;
1763         spin_unlock(&mdev->peer_seq_lock);
1764         return ret;
1765 }
1766
1767 static unsigned long write_flags_to_bio(struct drbd_conf *mdev, u32 dpf)
1768 {
1769         if (mdev->agreed_pro_version >= 95)
1770                 return  (dpf & DP_RW_SYNC ? REQ_SYNC : 0) |
1771                         (dpf & DP_UNPLUG ? REQ_UNPLUG : 0) |
1772                         (dpf & DP_FUA ? REQ_FUA : 0) |
1773                         (dpf & DP_FLUSH ? REQ_FUA : 0) |
1774                         (dpf & DP_DISCARD ? REQ_DISCARD : 0);
1775         else
1776                 return dpf & DP_RW_SYNC ? (REQ_SYNC | REQ_UNPLUG) : 0;
1777 }
1778
1779 /* mirrored write */
1780 static int receive_Data(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1781 {
1782         sector_t sector;
1783         struct drbd_epoch_entry *e;
1784         struct p_data *p = &mdev->data.rbuf.data;
1785         int rw = WRITE;
1786         u32 dp_flags;
1787
1788         if (!get_ldev(mdev)) {
1789                 if (__ratelimit(&drbd_ratelimit_state))
1790                         dev_err(DEV, "Can not write mirrored data block "
1791                             "to local disk.\n");
1792                 spin_lock(&mdev->peer_seq_lock);
1793                 if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num))
1794                         mdev->peer_seq++;
1795                 spin_unlock(&mdev->peer_seq_lock);
1796
1797                 drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size);
1798                 atomic_inc(&mdev->current_epoch->epoch_size);
1799                 return drbd_drain_block(mdev, data_size);
1800         }
1801
1802         /* get_ldev(mdev) successful.
1803          * Corresponding put_ldev done either below (on various errors),
1804          * or in drbd_endio_write_sec, if we successfully submit the data at
1805          * the end of this function. */
1806
1807         sector = be64_to_cpu(p->sector);
1808         e = read_in_block(mdev, p->block_id, sector, data_size);
1809         if (!e) {
1810                 put_ldev(mdev);
1811                 return FALSE;
1812         }
1813
1814         e->w.cb = e_end_block;
1815
1816         spin_lock(&mdev->epoch_lock);
1817         e->epoch = mdev->current_epoch;
1818         atomic_inc(&e->epoch->epoch_size);
1819         atomic_inc(&e->epoch->active);
1820
1821         if (mdev->write_ordering == WO_bio_barrier && atomic_read(&e->epoch->epoch_size) == 1) {
1822                 struct drbd_epoch *epoch;
1823                 /* Issue a barrier if we start a new epoch, and the previous epoch
1824                    was not a epoch containing a single request which already was
1825                    a Barrier. */
1826                 epoch = list_entry(e->epoch->list.prev, struct drbd_epoch, list);
1827                 if (epoch == e->epoch) {
1828                         set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags);
1829                         rw |= REQ_HARDBARRIER;
1830                         e->flags |= EE_IS_BARRIER;
1831                 } else {
1832                         if (atomic_read(&epoch->epoch_size) > 1 ||
1833                             !test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) {
1834                                 set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags);
1835                                 set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags);
1836                                 rw |= REQ_HARDBARRIER;
1837                                 e->flags |= EE_IS_BARRIER;
1838                         }
1839                 }
1840         }
1841         spin_unlock(&mdev->epoch_lock);
1842
1843         dp_flags = be32_to_cpu(p->dp_flags);
1844         rw |= write_flags_to_bio(mdev, dp_flags);
1845
1846         if (dp_flags & DP_MAY_SET_IN_SYNC)
1847                 e->flags |= EE_MAY_SET_IN_SYNC;
1848
1849         /* I'm the receiver, I do hold a net_cnt reference. */
1850         if (!mdev->net_conf->two_primaries) {
1851                 spin_lock_irq(&mdev->req_lock);
1852         } else {
1853                 /* don't get the req_lock yet,
1854                  * we may sleep in drbd_wait_peer_seq */
1855                 const int size = e->size;
1856                 const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1857                 DEFINE_WAIT(wait);
1858                 struct drbd_request *i;
1859                 struct hlist_node *n;
1860                 struct hlist_head *slot;
1861                 int first;
1862
1863                 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1864                 BUG_ON(mdev->ee_hash == NULL);
1865                 BUG_ON(mdev->tl_hash == NULL);
1866
1867                 /* conflict detection and handling:
1868                  * 1. wait on the sequence number,
1869                  *    in case this data packet overtook ACK packets.
1870                  * 2. check our hash tables for conflicting requests.
1871                  *    we only need to walk the tl_hash, since an ee can not
1872                  *    have a conflict with an other ee: on the submitting
1873                  *    node, the corresponding req had already been conflicting,
1874                  *    and a conflicting req is never sent.
1875                  *
1876                  * Note: for two_primaries, we are protocol C,
1877                  * so there cannot be any request that is DONE
1878                  * but still on the transfer log.
1879                  *
1880                  * unconditionally add to the ee_hash.
1881                  *
1882                  * if no conflicting request is found:
1883                  *    submit.
1884                  *
1885                  * if any conflicting request is found
1886                  * that has not yet been acked,
1887                  * AND I have the "discard concurrent writes" flag:
1888                  *       queue (via done_ee) the P_DISCARD_ACK; OUT.
1889                  *
1890                  * if any conflicting request is found:
1891                  *       block the receiver, waiting on misc_wait
1892                  *       until no more conflicting requests are there,
1893                  *       or we get interrupted (disconnect).
1894                  *
1895                  *       we do not just write after local io completion of those
1896                  *       requests, but only after req is done completely, i.e.
1897                  *       we wait for the P_DISCARD_ACK to arrive!
1898                  *
1899                  *       then proceed normally, i.e. submit.
1900                  */
1901                 if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num)))
1902                         goto out_interrupted;
1903
1904                 spin_lock_irq(&mdev->req_lock);
1905
1906                 hlist_add_head(&e->colision, ee_hash_slot(mdev, sector));
1907
1908 #define OVERLAPS overlaps(i->sector, i->size, sector, size)
1909                 slot = tl_hash_slot(mdev, sector);
1910                 first = 1;
1911                 for (;;) {
1912                         int have_unacked = 0;
1913                         int have_conflict = 0;
1914                         prepare_to_wait(&mdev->misc_wait, &wait,
1915                                 TASK_INTERRUPTIBLE);
1916                         hlist_for_each_entry(i, n, slot, colision) {
1917                                 if (OVERLAPS) {
1918                                         /* only ALERT on first iteration,
1919                                          * we may be woken up early... */
1920                                         if (first)
1921                                                 dev_alert(DEV, "%s[%u] Concurrent local write detected!"
1922                                                       " new: %llus +%u; pending: %llus +%u\n",
1923                                                       current->comm, current->pid,
1924                                                       (unsigned long long)sector, size,
1925                                                       (unsigned long long)i->sector, i->size);
1926                                         if (i->rq_state & RQ_NET_PENDING)
1927                                                 ++have_unacked;
1928                                         ++have_conflict;
1929                                 }
1930                         }
1931 #undef OVERLAPS
1932                         if (!have_conflict)
1933                                 break;
1934
1935                         /* Discard Ack only for the _first_ iteration */
1936                         if (first && discard && have_unacked) {
1937                                 dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n",
1938                                      (unsigned long long)sector);
1939                                 inc_unacked(mdev);
1940                                 e->w.cb = e_send_discard_ack;
1941                                 list_add_tail(&e->w.list, &mdev->done_ee);
1942
1943                                 spin_unlock_irq(&mdev->req_lock);
1944
1945                                 /* we could probably send that P_DISCARD_ACK ourselves,
1946                                  * but I don't like the receiver using the msock */
1947
1948                                 put_ldev(mdev);
1949                                 wake_asender(mdev);
1950                                 finish_wait(&mdev->misc_wait, &wait);
1951                                 return TRUE;
1952                         }
1953
1954                         if (signal_pending(current)) {
1955                                 hlist_del_init(&e->colision);
1956
1957                                 spin_unlock_irq(&mdev->req_lock);
1958
1959                                 finish_wait(&mdev->misc_wait, &wait);
1960                                 goto out_interrupted;
1961                         }
1962
1963                         spin_unlock_irq(&mdev->req_lock);
1964                         if (first) {
1965                                 first = 0;
1966                                 dev_alert(DEV, "Concurrent write! [W AFTERWARDS] "
1967                                      "sec=%llus\n", (unsigned long long)sector);
1968                         } else if (discard) {
1969                                 /* we had none on the first iteration.
1970                                  * there must be none now. */
1971                                 D_ASSERT(have_unacked == 0);
1972                         }
1973                         schedule();
1974                         spin_lock_irq(&mdev->req_lock);
1975                 }
1976                 finish_wait(&mdev->misc_wait, &wait);
1977         }
1978
1979         list_add(&e->w.list, &mdev->active_ee);
1980         spin_unlock_irq(&mdev->req_lock);
1981
1982         switch (mdev->net_conf->wire_protocol) {
1983         case DRBD_PROT_C:
1984                 inc_unacked(mdev);
1985                 /* corresponding dec_unacked() in e_end_block()
1986                  * respective _drbd_clear_done_ee */
1987                 break;
1988         case DRBD_PROT_B:
1989                 /* I really don't like it that the receiver thread
1990                  * sends on the msock, but anyways */
1991                 drbd_send_ack(mdev, P_RECV_ACK, e);
1992                 break;
1993         case DRBD_PROT_A:
1994                 /* nothing to do */
1995                 break;
1996         }
1997
1998         if (mdev->state.pdsk == D_DISKLESS) {
1999                 /* In case we have the only disk of the cluster, */
2000                 drbd_set_out_of_sync(mdev, e->sector, e->size);
2001                 e->flags |= EE_CALL_AL_COMPLETE_IO;
2002                 drbd_al_begin_io(mdev, e->sector);
2003         }
2004
2005         if (drbd_submit_ee(mdev, e, rw, DRBD_FAULT_DT_WR) == 0)
2006                 return TRUE;
2007
2008         /* drbd_submit_ee currently fails for one reason only:
2009          * not being able to allocate enough bios.
2010          * Is dropping the connection going to help? */
2011         spin_lock_irq(&mdev->req_lock);
2012         list_del(&e->w.list);
2013         hlist_del_init(&e->colision);
2014         spin_unlock_irq(&mdev->req_lock);
2015         if (e->flags & EE_CALL_AL_COMPLETE_IO)
2016                 drbd_al_complete_io(mdev, e->sector);
2017
2018 out_interrupted:
2019         /* yes, the epoch_size now is imbalanced.
2020          * but we drop the connection anyways, so we don't have a chance to
2021          * receive a barrier... atomic_inc(&mdev->epoch_size); */
2022         put_ldev(mdev);
2023         drbd_free_ee(mdev, e);
2024         return FALSE;
2025 }
2026
2027 /* We may throttle resync, if the lower device seems to be busy,
2028  * and current sync rate is above c_min_rate.
2029  *
2030  * To decide whether or not the lower device is busy, we use a scheme similar
2031  * to MD RAID is_mddev_idle(): if the partition stats reveal "significant"
2032  * (more than 64 sectors) of activity we cannot account for with our own resync
2033  * activity, it obviously is "busy".
2034  *
2035  * The current sync rate used here uses only the most recent two step marks,
2036  * to have a short time average so we can react faster.
2037  */
2038 int drbd_rs_should_slow_down(struct drbd_conf *mdev)
2039 {
2040         struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk;
2041         unsigned long db, dt, dbdt;
2042         int curr_events;
2043         int throttle = 0;
2044
2045         /* feature disabled? */
2046         if (mdev->sync_conf.c_min_rate == 0)
2047                 return 0;
2048
2049         curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
2050                       (int)part_stat_read(&disk->part0, sectors[1]) -
2051                         atomic_read(&mdev->rs_sect_ev);
2052         if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) {
2053                 unsigned long rs_left;
2054                 int i;
2055
2056                 mdev->rs_last_events = curr_events;
2057
2058                 /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP,
2059                  * approx. */
2060                 i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-2) % DRBD_SYNC_MARKS;
2061                 rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed;
2062
2063                 dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ;
2064                 if (!dt)
2065                         dt++;
2066                 db = mdev->rs_mark_left[i] - rs_left;
2067                 dbdt = Bit2KB(db/dt);
2068
2069                 if (dbdt > mdev->sync_conf.c_min_rate)
2070                         throttle = 1;
2071         }
2072         return throttle;
2073 }
2074
2075
2076 static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int digest_size)
2077 {
2078         sector_t sector;
2079         const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
2080         struct drbd_epoch_entry *e;
2081         struct digest_info *di = NULL;
2082         int size;
2083         unsigned int fault_type;
2084         struct p_block_req *p = &mdev->data.rbuf.block_req;
2085
2086         sector = be64_to_cpu(p->sector);
2087         size   = be32_to_cpu(p->blksize);
2088
2089         if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_SEGMENT_SIZE) {
2090                 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
2091                                 (unsigned long long)sector, size);
2092                 return FALSE;
2093         }
2094         if (sector + (size>>9) > capacity) {
2095                 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
2096                                 (unsigned long long)sector, size);
2097                 return FALSE;
2098         }
2099
2100         if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) {
2101                 if (__ratelimit(&drbd_ratelimit_state))
2102                         dev_err(DEV, "Can not satisfy peer's read request, "
2103                             "no local data.\n");
2104                 drbd_send_ack_rp(mdev, cmd == P_DATA_REQUEST ? P_NEG_DREPLY :
2105                                  P_NEG_RS_DREPLY , p);
2106                 /* drain possibly payload */
2107                 return drbd_drain_block(mdev, digest_size);
2108         }
2109
2110         /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
2111          * "criss-cross" setup, that might cause write-out on some other DRBD,
2112          * which in turn might block on the other node at this very place.  */
2113         e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO);
2114         if (!e) {
2115                 put_ldev(mdev);
2116                 return FALSE;
2117         }
2118
2119         switch (cmd) {
2120         case P_DATA_REQUEST:
2121                 e->w.cb = w_e_end_data_req;
2122                 fault_type = DRBD_FAULT_DT_RD;
2123                 /* application IO, don't drbd_rs_begin_io */
2124                 goto submit;
2125
2126         case P_RS_DATA_REQUEST:
2127                 e->w.cb = w_e_end_rsdata_req;
2128                 fault_type = DRBD_FAULT_RS_RD;
2129                 break;
2130
2131         case P_OV_REPLY:
2132         case P_CSUM_RS_REQUEST:
2133                 fault_type = DRBD_FAULT_RS_RD;
2134                 di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO);
2135                 if (!di)
2136                         goto out_free_e;
2137
2138                 di->digest_size = digest_size;
2139                 di->digest = (((char *)di)+sizeof(struct digest_info));
2140
2141                 e->digest = di;
2142                 e->flags |= EE_HAS_DIGEST;
2143
2144                 if (drbd_recv(mdev, di->digest, digest_size) != digest_size)
2145                         goto out_free_e;
2146
2147                 if (cmd == P_CSUM_RS_REQUEST) {
2148                         D_ASSERT(mdev->agreed_pro_version >= 89);
2149                         e->w.cb = w_e_end_csum_rs_req;
2150                 } else if (cmd == P_OV_REPLY) {
2151                         e->w.cb = w_e_end_ov_reply;
2152                         dec_rs_pending(mdev);
2153                         /* drbd_rs_begin_io done when we sent this request,
2154                          * but accounting still needs to be done. */
2155                         goto submit_for_resync;
2156                 }
2157                 break;
2158
2159         case P_OV_REQUEST:
2160                 if (mdev->state.conn >= C_CONNECTED &&
2161                     mdev->state.conn != C_VERIFY_T)
2162                         dev_warn(DEV, "ASSERT FAILED: got P_OV_REQUEST while being %s\n",
2163                                 drbd_conn_str(mdev->state.conn));
2164                 if (mdev->ov_start_sector == ~(sector_t)0 &&
2165                     mdev->agreed_pro_version >= 90) {
2166                         mdev->ov_start_sector = sector;
2167                         mdev->ov_position = sector;
2168                         mdev->ov_left = mdev->rs_total - BM_SECT_TO_BIT(sector);
2169                         dev_info(DEV, "Online Verify start sector: %llu\n",
2170                                         (unsigned long long)sector);
2171                 }
2172                 e->w.cb = w_e_end_ov_req;
2173                 fault_type = DRBD_FAULT_RS_RD;
2174                 break;
2175
2176         default:
2177                 dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n",
2178                     cmdname(cmd));
2179                 fault_type = DRBD_FAULT_MAX;
2180                 goto out_free_e;
2181         }
2182
2183         /* Throttle, drbd_rs_begin_io and submit should become asynchronous
2184          * wrt the receiver, but it is not as straightforward as it may seem.
2185          * Various places in the resync start and stop logic assume resync
2186          * requests are processed in order, requeuing this on the worker thread
2187          * introduces a bunch of new code for synchronization between threads.
2188          *
2189          * Unlimited throttling before drbd_rs_begin_io may stall the resync
2190          * "forever", throttling after drbd_rs_begin_io will lock that extent
2191          * for application writes for the same time.  For now, just throttle
2192          * here, where the rest of the code expects the receiver to sleep for
2193          * a while, anyways.
2194          */
2195
2196         /* Throttle before drbd_rs_begin_io, as that locks out application IO;
2197          * this defers syncer requests for some time, before letting at least
2198          * on request through.  The resync controller on the receiving side
2199          * will adapt to the incoming rate accordingly.
2200          *
2201          * We cannot throttle here if remote is Primary/SyncTarget:
2202          * we would also throttle its application reads.
2203          * In that case, throttling is done on the SyncTarget only.
2204          */
2205         if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev))
2206                 msleep(100);
2207         if (drbd_rs_begin_io(mdev, e->sector))
2208                 goto out_free_e;
2209
2210 submit_for_resync:
2211         atomic_add(size >> 9, &mdev->rs_sect_ev);
2212
2213 submit:
2214         inc_unacked(mdev);
2215         spin_lock_irq(&mdev->req_lock);
2216         list_add_tail(&e->w.list, &mdev->read_ee);
2217         spin_unlock_irq(&mdev->req_lock);
2218
2219         if (drbd_submit_ee(mdev, e, READ, fault_type) == 0)
2220                 return TRUE;
2221
2222         /* drbd_submit_ee currently fails for one reason only:
2223          * not being able to allocate enough bios.
2224          * Is dropping the connection going to help? */
2225         spin_lock_irq(&mdev->req_lock);
2226         list_del(&e->w.list);
2227         spin_unlock_irq(&mdev->req_lock);
2228         /* no drbd_rs_complete_io(), we are dropping the connection anyways */
2229
2230 out_free_e:
2231         put_ldev(mdev);
2232         drbd_free_ee(mdev, e);
2233         return FALSE;
2234 }
2235
2236 static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local)
2237 {
2238         int self, peer, rv = -100;
2239         unsigned long ch_self, ch_peer;
2240
2241         self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2242         peer = mdev->p_uuid[UI_BITMAP] & 1;
2243
2244         ch_peer = mdev->p_uuid[UI_SIZE];
2245         ch_self = mdev->comm_bm_set;
2246
2247         switch (mdev->net_conf->after_sb_0p) {
2248         case ASB_CONSENSUS:
2249         case ASB_DISCARD_SECONDARY:
2250         case ASB_CALL_HELPER:
2251                 dev_err(DEV, "Configuration error.\n");
2252                 break;
2253         case ASB_DISCONNECT:
2254                 break;
2255         case ASB_DISCARD_YOUNGER_PRI:
2256                 if (self == 0 && peer == 1) {
2257                         rv = -1;
2258                         break;
2259                 }
2260                 if (self == 1 && peer == 0) {
2261                         rv =  1;
2262                         break;
2263                 }
2264                 /* Else fall through to one of the other strategies... */
2265         case ASB_DISCARD_OLDER_PRI:
2266                 if (self == 0 && peer == 1) {
2267                         rv = 1;
2268                         break;
2269                 }
2270                 if (self == 1 && peer == 0) {
2271                         rv = -1;
2272                         break;
2273                 }
2274                 /* Else fall through to one of the other strategies... */
2275                 dev_warn(DEV, "Discard younger/older primary did not find a decision\n"
2276                      "Using discard-least-changes instead\n");
2277         case ASB_DISCARD_ZERO_CHG:
2278                 if (ch_peer == 0 && ch_self == 0) {
2279                         rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2280                                 ? -1 : 1;
2281                         break;
2282                 } else {
2283                         if (ch_peer == 0) { rv =  1; break; }
2284                         if (ch_self == 0) { rv = -1; break; }
2285                 }
2286                 if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG)
2287                         break;
2288         case ASB_DISCARD_LEAST_CHG:
2289                 if      (ch_self < ch_peer)
2290                         rv = -1;
2291                 else if (ch_self > ch_peer)
2292                         rv =  1;
2293                 else /* ( ch_self == ch_peer ) */
2294                      /* Well, then use something else. */
2295                         rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2296                                 ? -1 : 1;
2297                 break;
2298         case ASB_DISCARD_LOCAL:
2299                 rv = -1;
2300                 break;
2301         case ASB_DISCARD_REMOTE:
2302                 rv =  1;
2303         }
2304
2305         return rv;
2306 }
2307
2308 static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local)
2309 {
2310         int self, peer, hg, rv = -100;
2311
2312         self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2313         peer = mdev->p_uuid[UI_BITMAP] & 1;
2314
2315         switch (mdev->net_conf->after_sb_1p) {
2316         case ASB_DISCARD_YOUNGER_PRI:
2317         case ASB_DISCARD_OLDER_PRI:
2318         case ASB_DISCARD_LEAST_CHG:
2319         case ASB_DISCARD_LOCAL:
2320         case ASB_DISCARD_REMOTE:
2321                 dev_err(DEV, "Configuration error.\n");
2322                 break;
2323         case ASB_DISCONNECT:
2324                 break;
2325         case ASB_CONSENSUS:
2326                 hg = drbd_asb_recover_0p(mdev);
2327                 if (hg == -1 && mdev->state.role == R_SECONDARY)
2328                         rv = hg;
2329                 if (hg == 1  && mdev->state.role == R_PRIMARY)
2330                         rv = hg;
2331                 break;
2332         case ASB_VIOLENTLY:
2333                 rv = drbd_asb_recover_0p(mdev);
2334                 break;
2335         case ASB_DISCARD_SECONDARY:
2336                 return mdev->state.role == R_PRIMARY ? 1 : -1;
2337         case ASB_CALL_HELPER:
2338                 hg = drbd_asb_recover_0p(mdev);
2339                 if (hg == -1 && mdev->state.role == R_PRIMARY) {
2340                         self = drbd_set_role(mdev, R_SECONDARY, 0);
2341                          /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2342                           * we might be here in C_WF_REPORT_PARAMS which is transient.
2343                           * we do not need to wait for the after state change work either. */
2344                         self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2345                         if (self != SS_SUCCESS) {
2346                                 drbd_khelper(mdev, "pri-lost-after-sb");
2347                         } else {
2348                                 dev_warn(DEV, "Successfully gave up primary role.\n");
2349                                 rv = hg;
2350                         }
2351                 } else
2352                         rv = hg;
2353         }
2354
2355         return rv;
2356 }
2357
2358 static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local)
2359 {
2360         int self, peer, hg, rv = -100;
2361
2362         self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2363         peer = mdev->p_uuid[UI_BITMAP] & 1;
2364
2365         switch (mdev->net_conf->after_sb_2p) {
2366         case ASB_DISCARD_YOUNGER_PRI:
2367         case ASB_DISCARD_OLDER_PRI:
2368         case ASB_DISCARD_LEAST_CHG:
2369         case ASB_DISCARD_LOCAL:
2370         case ASB_DISCARD_REMOTE:
2371         case ASB_CONSENSUS:
2372         case ASB_DISCARD_SECONDARY:
2373                 dev_err(DEV, "Configuration error.\n");
2374                 break;
2375         case ASB_VIOLENTLY:
2376                 rv = drbd_asb_recover_0p(mdev);
2377                 break;
2378         case ASB_DISCONNECT:
2379                 break;
2380         case ASB_CALL_HELPER:
2381                 hg = drbd_asb_recover_0p(mdev);
2382                 if (hg == -1) {
2383                          /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2384                           * we might be here in C_WF_REPORT_PARAMS which is transient.
2385                           * we do not need to wait for the after state change work either. */
2386                         self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2387                         if (self != SS_SUCCESS) {
2388                                 drbd_khelper(mdev, "pri-lost-after-sb");
2389                         } else {
2390                                 dev_warn(DEV, "Successfully gave up primary role.\n");
2391                                 rv = hg;
2392                         }
2393                 } else
2394                         rv = hg;
2395         }
2396
2397         return rv;
2398 }
2399
2400 static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid,
2401                            u64 bits, u64 flags)
2402 {
2403         if (!uuid) {
2404                 dev_info(DEV, "%s uuid info vanished while I was looking!\n", text);
2405                 return;
2406         }
2407         dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n",
2408              text,
2409              (unsigned long long)uuid[UI_CURRENT],
2410              (unsigned long long)uuid[UI_BITMAP],
2411              (unsigned long long)uuid[UI_HISTORY_START],
2412              (unsigned long long)uuid[UI_HISTORY_END],
2413              (unsigned long long)bits,
2414              (unsigned long long)flags);
2415 }
2416
2417 /*
2418   100   after split brain try auto recover
2419     2   C_SYNC_SOURCE set BitMap
2420     1   C_SYNC_SOURCE use BitMap
2421     0   no Sync
2422    -1   C_SYNC_TARGET use BitMap
2423    -2   C_SYNC_TARGET set BitMap
2424  -100   after split brain, disconnect
2425 -1000   unrelated data
2426  */
2427 static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local)
2428 {
2429         u64 self, peer;
2430         int i, j;
2431
2432         self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2433         peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2434
2435         *rule_nr = 10;
2436         if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED)
2437                 return 0;
2438
2439         *rule_nr = 20;
2440         if ((self == UUID_JUST_CREATED || self == (u64)0) &&
2441              peer != UUID_JUST_CREATED)
2442                 return -2;
2443
2444         *rule_nr = 30;
2445         if (self != UUID_JUST_CREATED &&
2446             (peer == UUID_JUST_CREATED || peer == (u64)0))
2447                 return 2;
2448
2449         if (self == peer) {
2450                 int rct, dc; /* roles at crash time */
2451
2452                 if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) {
2453
2454                         if (mdev->agreed_pro_version < 91)
2455                                 return -1001;
2456
2457                         if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) &&
2458                             (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) {
2459                                 dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n");
2460                                 drbd_uuid_set_bm(mdev, 0UL);
2461
2462                                 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2463                                                mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2464                                 *rule_nr = 34;
2465                         } else {
2466                                 dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n");
2467                                 *rule_nr = 36;
2468                         }
2469
2470                         return 1;
2471                 }
2472
2473                 if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) {
2474
2475                         if (mdev->agreed_pro_version < 91)
2476                                 return -1001;
2477
2478                         if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) &&
2479                             (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) {
2480                                 dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n");
2481
2482                                 mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START];
2483                                 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP];
2484                                 mdev->p_uuid[UI_BITMAP] = 0UL;
2485
2486                                 drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2487                                 *rule_nr = 35;
2488                         } else {
2489                                 dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n");
2490                                 *rule_nr = 37;
2491                         }
2492
2493                         return -1;
2494                 }
2495
2496                 /* Common power [off|failure] */
2497                 rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) +
2498                         (mdev->p_uuid[UI_FLAGS] & 2);
2499                 /* lowest bit is set when we were primary,
2500                  * next bit (weight 2) is set when peer was primary */
2501                 *rule_nr = 40;
2502
2503                 switch (rct) {
2504                 case 0: /* !self_pri && !peer_pri */ return 0;
2505                 case 1: /*  self_pri && !peer_pri */ return 1;
2506                 case 2: /* !self_pri &&  peer_pri */ return -1;
2507                 case 3: /*  self_pri &&  peer_pri */
2508                         dc = test_bit(DISCARD_CONCURRENT, &mdev->flags);
2509                         return dc ? -1 : 1;
2510                 }
2511         }
2512
2513         *rule_nr = 50;
2514         peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2515         if (self == peer)
2516                 return -1;
2517
2518         *rule_nr = 51;
2519         peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1);
2520         if (self == peer) {
2521                 self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1);
2522                 peer = mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1);
2523                 if (self == peer) {
2524                         /* The last P_SYNC_UUID did not get though. Undo the last start of
2525                            resync as sync source modifications of the peer's UUIDs. */
2526
2527                         if (mdev->agreed_pro_version < 91)
2528                                 return -1001;
2529
2530                         mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START];
2531                         mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1];
2532                         return -1;
2533                 }
2534         }
2535
2536         *rule_nr = 60;
2537         self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2538         for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2539                 peer = mdev->p_uuid[i] & ~((u64)1);
2540                 if (self == peer)
2541                         return -2;
2542         }
2543
2544         *rule_nr = 70;
2545         self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2546         peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2547         if (self == peer)
2548                 return 1;
2549
2550         *rule_nr = 71;
2551         self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1);
2552         if (self == peer) {
2553                 self = mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1);
2554                 peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1);
2555                 if (self == peer) {
2556                         /* The last P_SYNC_UUID did not get though. Undo the last start of
2557                            resync as sync source modifications of our UUIDs. */
2558
2559                         if (mdev->agreed_pro_version < 91)
2560                                 return -1001;
2561
2562                         _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]);
2563                         _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]);
2564
2565                         dev_info(DEV, "Undid last start of resync:\n");
2566
2567                         drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2568                                        mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2569
2570                         return 1;
2571                 }
2572         }
2573
2574
2575         *rule_nr = 80;
2576         peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2577         for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2578                 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2579                 if (self == peer)
2580                         return 2;
2581         }
2582
2583         *rule_nr = 90;
2584         self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2585         peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2586         if (self == peer && self != ((u64)0))
2587                 return 100;
2588
2589         *rule_nr = 100;
2590         for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2591                 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2592                 for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) {
2593                         peer = mdev->p_uuid[j] & ~((u64)1);
2594                         if (self == peer)
2595                                 return -100;
2596                 }
2597         }
2598
2599         return -1000;
2600 }
2601
2602 /* drbd_sync_handshake() returns the new conn state on success, or
2603    CONN_MASK (-1) on failure.
2604  */
2605 static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role,
2606                                            enum drbd_disk_state peer_disk) __must_hold(local)
2607 {
2608         int hg, rule_nr;
2609         enum drbd_conns rv = C_MASK;
2610         enum drbd_disk_state mydisk;
2611
2612         mydisk = mdev->state.disk;
2613         if (mydisk == D_NEGOTIATING)
2614                 mydisk = mdev->new_state_tmp.disk;
2615
2616         dev_info(DEV, "drbd_sync_handshake:\n");
2617         drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0);
2618         drbd_uuid_dump(mdev, "peer", mdev->p_uuid,
2619                        mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2620
2621         hg = drbd_uuid_compare(mdev, &rule_nr);
2622
2623         dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr);
2624
2625         if (hg == -1000) {
2626                 dev_alert(DEV, "Unrelated data, aborting!\n");
2627                 return C_MASK;
2628         }
2629         if (hg == -1001) {
2630                 dev_alert(DEV, "To resolve this both sides have to support at least protocol\n");
2631                 return C_MASK;
2632         }
2633
2634         if    ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) ||
2635             (peer_disk == D_INCONSISTENT && mydisk    > D_INCONSISTENT)) {
2636                 int f = (hg == -100) || abs(hg) == 2;
2637                 hg = mydisk > D_INCONSISTENT ? 1 : -1;
2638                 if (f)
2639                         hg = hg*2;
2640                 dev_info(DEV, "Becoming sync %s due to disk states.\n",
2641                      hg > 0 ? "source" : "target");
2642         }
2643
2644         if (abs(hg) == 100)
2645                 drbd_khelper(mdev, "initial-split-brain");
2646
2647         if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) {
2648                 int pcount = (mdev->state.role == R_PRIMARY)
2649                            + (peer_role == R_PRIMARY);
2650                 int forced = (hg == -100);
2651
2652                 switch (pcount) {
2653                 case 0:
2654                         hg = drbd_asb_recover_0p(mdev);
2655                         break;
2656                 case 1:
2657                         hg = drbd_asb_recover_1p(mdev);
2658                         break;
2659                 case 2:
2660                         hg = drbd_asb_recover_2p(mdev);
2661                         break;
2662                 }
2663                 if (abs(hg) < 100) {
2664                         dev_warn(DEV, "Split-Brain detected, %d primaries, "
2665                              "automatically solved. Sync from %s node\n",
2666                              pcount, (hg < 0) ? "peer" : "this");
2667                         if (forced) {
2668                                 dev_warn(DEV, "Doing a full sync, since"
2669                                      " UUIDs where ambiguous.\n");
2670                                 hg = hg*2;
2671                         }
2672                 }
2673         }
2674
2675         if (hg == -100) {
2676                 if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1))
2677                         hg = -1;
2678                 if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1))
2679                         hg = 1;
2680
2681                 if (abs(hg) < 100)
2682                         dev_warn(DEV, "Split-Brain detected, manually solved. "
2683                              "Sync from %s node\n",
2684                              (hg < 0) ? "peer" : "this");
2685         }
2686
2687         if (hg == -100) {
2688                 /* FIXME this log message is not correct if we end up here
2689                  * after an attempted attach on a diskless node.
2690                  * We just refuse to attach -- well, we drop the "connection"
2691                  * to that disk, in a way... */
2692                 dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n");
2693                 drbd_khelper(mdev, "split-brain");
2694                 return C_MASK;
2695         }
2696
2697         if (hg > 0 && mydisk <= D_INCONSISTENT) {
2698                 dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n");
2699                 return C_MASK;
2700         }
2701
2702         if (hg < 0 && /* by intention we do not use mydisk here. */
2703             mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) {
2704                 switch (mdev->net_conf->rr_conflict) {
2705                 case ASB_CALL_HELPER:
2706                         drbd_khelper(mdev, "pri-lost");
2707                         /* fall through */
2708                 case ASB_DISCONNECT:
2709                         dev_err(DEV, "I shall become SyncTarget, but I am primary!\n");
2710                         return C_MASK;
2711                 case ASB_VIOLENTLY:
2712                         dev_warn(DEV, "Becoming SyncTarget, violating the stable-data"
2713                              "assumption\n");
2714                 }
2715         }
2716
2717         if (mdev->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) {
2718                 if (hg == 0)
2719                         dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n");
2720                 else
2721                         dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.",
2722                                  drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET),
2723                                  abs(hg) >= 2 ? "full" : "bit-map based");
2724                 return C_MASK;
2725         }
2726
2727         if (abs(hg) >= 2) {
2728                 dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n");
2729                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake"))
2730                         return C_MASK;
2731         }
2732
2733         if (hg > 0) { /* become sync source. */
2734                 rv = C_WF_BITMAP_S;
2735         } else if (hg < 0) { /* become sync target */
2736                 rv = C_WF_BITMAP_T;
2737         } else {
2738                 rv = C_CONNECTED;
2739                 if (drbd_bm_total_weight(mdev)) {
2740                         dev_info(DEV, "No resync, but %lu bits in bitmap!\n",
2741                              drbd_bm_total_weight(mdev));
2742                 }
2743         }
2744
2745         return rv;
2746 }
2747
2748 /* returns 1 if invalid */
2749 static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self)
2750 {
2751         /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */
2752         if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) ||
2753             (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL))
2754                 return 0;
2755
2756         /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */
2757         if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL ||
2758             self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL)
2759                 return 1;
2760
2761         /* everything else is valid if they are equal on both sides. */
2762         if (peer == self)
2763                 return 0;
2764
2765         /* everything es is invalid. */
2766         return 1;
2767 }
2768
2769 static int receive_protocol(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
2770 {
2771         struct p_protocol *p = &mdev->data.rbuf.protocol;
2772         int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p;
2773         int p_want_lose, p_two_primaries, cf;
2774         char p_integrity_alg[SHARED_SECRET_MAX] = "";
2775
2776         p_proto         = be32_to_cpu(p->protocol);
2777         p_after_sb_0p   = be32_to_cpu(p->after_sb_0p);
2778         p_after_sb_1p   = be32_to_cpu(p->after_sb_1p);
2779         p_after_sb_2p   = be32_to_cpu(p->after_sb_2p);
2780         p_two_primaries = be32_to_cpu(p->two_primaries);
2781         cf              = be32_to_cpu(p->conn_flags);
2782         p_want_lose = cf & CF_WANT_LOSE;
2783
2784         clear_bit(CONN_DRY_RUN, &mdev->flags);
2785
2786         if (cf & CF_DRY_RUN)
2787                 set_bit(CONN_DRY_RUN, &mdev->flags);
2788
2789         if (p_proto != mdev->net_conf->wire_protocol) {
2790                 dev_err(DEV, "incompatible communication protocols\n");
2791                 goto disconnect;
2792         }
2793
2794         if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) {
2795                 dev_err(DEV, "incompatible after-sb-0pri settings\n");
2796                 goto disconnect;
2797         }
2798
2799         if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) {
2800                 dev_err(DEV, "incompatible after-sb-1pri settings\n");
2801                 goto disconnect;
2802         }
2803
2804         if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) {
2805                 dev_err(DEV, "incompatible after-sb-2pri settings\n");
2806                 goto disconnect;
2807         }
2808
2809         if (p_want_lose && mdev->net_conf->want_lose) {
2810                 dev_err(DEV, "both sides have the 'want_lose' flag set\n");
2811                 goto disconnect;
2812         }
2813
2814         if (p_two_primaries != mdev->net_conf->two_primaries) {
2815                 dev_err(DEV, "incompatible setting of the two-primaries options\n");
2816                 goto disconnect;
2817         }
2818
2819         if (mdev->agreed_pro_version >= 87) {
2820                 unsigned char *my_alg = mdev->net_conf->integrity_alg;
2821
2822                 if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size)
2823                         return FALSE;
2824
2825                 p_integrity_alg[SHARED_SECRET_MAX-1] = 0;
2826                 if (strcmp(p_integrity_alg, my_alg)) {
2827                         dev_err(DEV, "incompatible setting of the data-integrity-alg\n");
2828                         goto disconnect;
2829                 }
2830                 dev_info(DEV, "data-integrity-alg: %s\n",
2831                      my_alg[0] ? my_alg : (unsigned char *)"<not-used>");
2832         }
2833
2834         return TRUE;
2835
2836 disconnect:
2837         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2838         return FALSE;
2839 }
2840
2841 /* helper function
2842  * input: alg name, feature name
2843  * return: NULL (alg name was "")
2844  *         ERR_PTR(error) if something goes wrong
2845  *         or the crypto hash ptr, if it worked out ok. */
2846 struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev,
2847                 const char *alg, const char *name)
2848 {
2849         struct crypto_hash *tfm;
2850
2851         if (!alg[0])
2852                 return NULL;
2853
2854         tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC);
2855         if (IS_ERR(tfm)) {
2856                 dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n",
2857                         alg, name, PTR_ERR(tfm));
2858                 return tfm;
2859         }
2860         if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
2861                 crypto_free_hash(tfm);
2862                 dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name);
2863                 return ERR_PTR(-EINVAL);
2864         }
2865         return tfm;
2866 }
2867
2868 static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int packet_size)
2869 {
2870         int ok = TRUE;
2871         struct p_rs_param_95 *p = &mdev->data.rbuf.rs_param_95;
2872         unsigned int header_size, data_size, exp_max_sz;
2873         struct crypto_hash *verify_tfm = NULL;
2874         struct crypto_hash *csums_tfm = NULL;
2875         const int apv = mdev->agreed_pro_version;
2876         int *rs_plan_s = NULL;
2877         int fifo_size = 0;
2878
2879         exp_max_sz  = apv <= 87 ? sizeof(struct p_rs_param)
2880                     : apv == 88 ? sizeof(struct p_rs_param)
2881                                         + SHARED_SECRET_MAX
2882                     : apv <= 94 ? sizeof(struct p_rs_param_89)
2883                     : /* apv >= 95 */ sizeof(struct p_rs_param_95);
2884
2885         if (packet_size > exp_max_sz) {
2886                 dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n",
2887                     packet_size, exp_max_sz);
2888                 return FALSE;
2889         }
2890
2891         if (apv <= 88) {
2892                 header_size = sizeof(struct p_rs_param) - sizeof(struct p_header80);
2893                 data_size   = packet_size  - header_size;
2894         } else if (apv <= 94) {
2895                 header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header80);
2896                 data_size   = packet_size  - header_size;
2897                 D_ASSERT(data_size == 0);
2898         } else {
2899                 header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header80);
2900                 data_size   = packet_size  - header_size;
2901                 D_ASSERT(data_size == 0);
2902         }
2903
2904         /* initialize verify_alg and csums_alg */
2905         memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
2906
2907         if (drbd_recv(mdev, &p->head.payload, header_size) != header_size)
2908                 return FALSE;
2909
2910         mdev->sync_conf.rate      = be32_to_cpu(p->rate);
2911
2912         if (apv >= 88) {
2913                 if (apv == 88) {
2914                         if (data_size > SHARED_SECRET_MAX) {
2915                                 dev_err(DEV, "verify-alg too long, "
2916                                     "peer wants %u, accepting only %u byte\n",
2917                                                 data_size, SHARED_SECRET_MAX);
2918                                 return FALSE;
2919                         }
2920
2921                         if (drbd_recv(mdev, p->verify_alg, data_size) != data_size)
2922                                 return FALSE;
2923
2924                         /* we expect NUL terminated string */
2925                         /* but just in case someone tries to be evil */
2926                         D_ASSERT(p->verify_alg[data_size-1] == 0);
2927                         p->verify_alg[data_size-1] = 0;
2928
2929                 } else /* apv >= 89 */ {
2930                         /* we still expect NUL terminated strings */
2931                         /* but just in case someone tries to be evil */
2932                         D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0);
2933                         D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0);
2934                         p->verify_alg[SHARED_SECRET_MAX-1] = 0;
2935                         p->csums_alg[SHARED_SECRET_MAX-1] = 0;
2936                 }
2937
2938                 if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) {
2939                         if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2940                                 dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n",
2941                                     mdev->sync_conf.verify_alg, p->verify_alg);
2942                                 goto disconnect;
2943                         }
2944                         verify_tfm = drbd_crypto_alloc_digest_safe(mdev,
2945                                         p->verify_alg, "verify-alg");
2946                         if (IS_ERR(verify_tfm)) {
2947                                 verify_tfm = NULL;
2948                                 goto disconnect;
2949                         }
2950                 }
2951
2952                 if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) {
2953                         if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2954                                 dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n",
2955                                     mdev->sync_conf.csums_alg, p->csums_alg);
2956                                 goto disconnect;
2957                         }
2958                         csums_tfm = drbd_crypto_alloc_digest_safe(mdev,
2959                                         p->csums_alg, "csums-alg");
2960                         if (IS_ERR(csums_tfm)) {
2961                                 csums_tfm = NULL;
2962                                 goto disconnect;
2963                         }
2964                 }
2965
2966                 if (apv > 94) {
2967                         mdev->sync_conf.rate      = be32_to_cpu(p->rate);
2968                         mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead);
2969                         mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target);
2970                         mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target);
2971                         mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate);
2972
2973                         fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
2974                         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
2975                                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
2976                                 if (!rs_plan_s) {
2977                                         dev_err(DEV, "kmalloc of fifo_buffer failed");
2978                                         goto disconnect;
2979                                 }
2980                         }
2981                 }
2982
2983                 spin_lock(&mdev->peer_seq_lock);
2984                 /* lock against drbd_nl_syncer_conf() */
2985                 if (verify_tfm) {
2986                         strcpy(mdev->sync_conf.verify_alg, p->verify_alg);
2987                         mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1;
2988                         crypto_free_hash(mdev->verify_tfm);
2989                         mdev->verify_tfm = verify_tfm;
2990                         dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg);
2991                 }
2992                 if (csums_tfm) {
2993                         strcpy(mdev->sync_conf.csums_alg, p->csums_alg);
2994                         mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1;
2995                         crypto_free_hash(mdev->csums_tfm);
2996                         mdev->csums_tfm = csums_tfm;
2997                         dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg);
2998                 }
2999                 if (fifo_size != mdev->rs_plan_s.size) {
3000                         kfree(mdev->rs_plan_s.values);
3001                         mdev->rs_plan_s.values = rs_plan_s;
3002                         mdev->rs_plan_s.size   = fifo_size;
3003                         mdev->rs_planed = 0;
3004                 }
3005                 spin_unlock(&mdev->peer_seq_lock);
3006         }
3007
3008         return ok;
3009 disconnect:
3010         /* just for completeness: actually not needed,
3011          * as this is not reached if csums_tfm was ok. */
3012         crypto_free_hash(csums_tfm);
3013         /* but free the verify_tfm again, if csums_tfm did not work out */
3014         crypto_free_hash(verify_tfm);
3015         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3016         return FALSE;
3017 }
3018
3019 static void drbd_setup_order_type(struct drbd_conf *mdev, int peer)
3020 {
3021         /* sorry, we currently have no working implementation
3022          * of distributed TCQ */
3023 }
3024
3025 /* warn if the arguments differ by more than 12.5% */
3026 static void warn_if_differ_considerably(struct drbd_conf *mdev,
3027         const char *s, sector_t a, sector_t b)
3028 {
3029         sector_t d;
3030         if (a == 0 || b == 0)
3031                 return;
3032         d = (a > b) ? (a - b) : (b - a);
3033         if (d > (a>>3) || d > (b>>3))
3034                 dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s,
3035                      (unsigned long long)a, (unsigned long long)b);
3036 }
3037
3038 static int receive_sizes(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3039 {
3040         struct p_sizes *p = &mdev->data.rbuf.sizes;
3041         enum determine_dev_size dd = unchanged;
3042         unsigned int max_seg_s;
3043         sector_t p_size, p_usize, my_usize;
3044         int ldsc = 0; /* local disk size changed */
3045         enum dds_flags ddsf;
3046
3047         p_size = be64_to_cpu(p->d_size);
3048         p_usize = be64_to_cpu(p->u_size);
3049
3050         if (p_size == 0 && mdev->state.disk == D_DISKLESS) {
3051                 dev_err(DEV, "some backing storage is needed\n");
3052                 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3053                 return FALSE;
3054         }
3055
3056         /* just store the peer's disk size for now.
3057          * we still need to figure out whether we accept that. */
3058         mdev->p_size = p_size;
3059
3060 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
3061         if (get_ldev(mdev)) {
3062                 warn_if_differ_considerably(mdev, "lower level device sizes",
3063                            p_size, drbd_get_max_capacity(mdev->ldev));
3064                 warn_if_differ_considerably(mdev, "user requested size",
3065                                             p_usize, mdev->ldev->dc.disk_size);
3066
3067                 /* if this is the first connect, or an otherwise expected
3068                  * param exchange, choose the minimum */
3069                 if (mdev->state.conn == C_WF_REPORT_PARAMS)
3070                         p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size,
3071                                              p_usize);
3072
3073                 my_usize = mdev->ldev->dc.disk_size;
3074
3075                 if (mdev->ldev->dc.disk_size != p_usize) {
3076                         mdev->ldev->dc.disk_size = p_usize;
3077                         dev_info(DEV, "Peer sets u_size to %lu sectors\n",
3078                              (unsigned long)mdev->ldev->dc.disk_size);
3079                 }
3080
3081                 /* Never shrink a device with usable data during connect.
3082                    But allow online shrinking if we are connected. */
3083                 if (drbd_new_dev_size(mdev, mdev->ldev, 0) <
3084                    drbd_get_capacity(mdev->this_bdev) &&
3085                    mdev->state.disk >= D_OUTDATED &&
3086                    mdev->state.conn < C_CONNECTED) {
3087                         dev_err(DEV, "The peer's disk size is too small!\n");
3088                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3089                         mdev->ldev->dc.disk_size = my_usize;
3090                         put_ldev(mdev);
3091                         return FALSE;
3092                 }
3093                 put_ldev(mdev);
3094         }
3095 #undef min_not_zero
3096
3097         ddsf = be16_to_cpu(p->dds_flags);
3098         if (get_ldev(mdev)) {
3099                 dd = drbd_determin_dev_size(mdev, ddsf);
3100                 put_ldev(mdev);
3101                 if (dd == dev_size_error)
3102                         return FALSE;
3103                 drbd_md_sync(mdev);
3104         } else {
3105                 /* I am diskless, need to accept the peer's size. */
3106                 drbd_set_my_capacity(mdev, p_size);
3107         }
3108
3109         if (get_ldev(mdev)) {
3110                 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) {
3111                         mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
3112                         ldsc = 1;
3113                 }
3114
3115                 if (mdev->agreed_pro_version < 94)
3116                         max_seg_s = be32_to_cpu(p->max_segment_size);
3117                 else if (mdev->agreed_pro_version == 94)
3118                         max_seg_s = DRBD_MAX_SIZE_H80_PACKET;
3119                 else /* drbd 8.3.8 onwards */
3120                         max_seg_s = DRBD_MAX_SEGMENT_SIZE;
3121
3122                 if (max_seg_s != queue_max_segment_size(mdev->rq_queue))
3123                         drbd_setup_queue_param(mdev, max_seg_s);
3124
3125                 drbd_setup_order_type(mdev, be16_to_cpu(p->queue_order_type));
3126                 put_ldev(mdev);
3127         }
3128
3129         if (mdev->state.conn > C_WF_REPORT_PARAMS) {
3130                 if (be64_to_cpu(p->c_size) !=
3131                     drbd_get_capacity(mdev->this_bdev) || ldsc) {
3132                         /* we have different sizes, probably peer
3133                          * needs to know my new size... */
3134                         drbd_send_sizes(mdev, 0, ddsf);
3135                 }
3136                 if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) ||
3137                     (dd == grew && mdev->state.conn == C_CONNECTED)) {
3138                         if (mdev->state.pdsk >= D_INCONSISTENT &&
3139                             mdev->state.disk >= D_INCONSISTENT) {
3140                                 if (ddsf & DDSF_NO_RESYNC)
3141                                         dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n");
3142                                 else
3143                                         resync_after_online_grow(mdev);
3144                         } else
3145                                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
3146                 }
3147         }
3148
3149         return TRUE;
3150 }
3151
3152 static int receive_uuids(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3153 {
3154         struct p_uuids *p = &mdev->data.rbuf.uuids;
3155         u64 *p_uuid;
3156         int i;
3157
3158         p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO);
3159
3160         for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++)
3161                 p_uuid[i] = be64_to_cpu(p->uuid[i]);
3162
3163         kfree(mdev->p_uuid);
3164         mdev->p_uuid = p_uuid;
3165
3166         if (mdev->state.conn < C_CONNECTED &&
3167             mdev->state.disk < D_INCONSISTENT &&
3168             mdev->state.role == R_PRIMARY &&
3169             (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) {
3170                 dev_err(DEV, "Can only connect to data with current UUID=%016llX\n",
3171                     (unsigned long long)mdev->ed_uuid);
3172                 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3173                 return FALSE;
3174         }
3175
3176         if (get_ldev(mdev)) {
3177                 int skip_initial_sync =
3178                         mdev->state.conn == C_CONNECTED &&
3179                         mdev->agreed_pro_version >= 90 &&
3180                         mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
3181                         (p_uuid[UI_FLAGS] & 8);
3182                 if (skip_initial_sync) {
3183                         dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n");
3184                         drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
3185                                         "clear_n_write from receive_uuids");
3186                         _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]);
3187                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
3188                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
3189                                         CS_VERBOSE, NULL);
3190                         drbd_md_sync(mdev);
3191                 }
3192                 put_ldev(mdev);
3193         } else if (mdev->state.disk < D_INCONSISTENT &&
3194                    mdev->state.role == R_PRIMARY) {
3195                 /* I am a diskless primary, the peer just created a new current UUID
3196                    for me. */
3197                 drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]);
3198         }
3199
3200         /* Before we test for the disk state, we should wait until an eventually
3201            ongoing cluster wide state change is finished. That is important if
3202            we are primary and are detaching from our disk. We need to see the
3203            new disk state... */
3204         wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags));
3205         if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT)
3206                 drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]);
3207
3208         return TRUE;
3209 }
3210
3211 /**
3212  * convert_state() - Converts the peer's view of the cluster state to our point of view
3213  * @ps:         The state as seen by the peer.
3214  */
3215 static union drbd_state convert_state(union drbd_state ps)
3216 {
3217         union drbd_state ms;
3218
3219         static enum drbd_conns c_tab[] = {
3220                 [C_CONNECTED] = C_CONNECTED,
3221
3222                 [C_STARTING_SYNC_S] = C_STARTING_SYNC_T,
3223                 [C_STARTING_SYNC_T] = C_STARTING_SYNC_S,
3224                 [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */
3225                 [C_VERIFY_S]       = C_VERIFY_T,
3226                 [C_MASK]   = C_MASK,
3227         };
3228
3229         ms.i = ps.i;
3230
3231         ms.conn = c_tab[ps.conn];
3232         ms.peer = ps.role;
3233         ms.role = ps.peer;
3234         ms.pdsk = ps.disk;
3235         ms.disk = ps.pdsk;
3236         ms.peer_isp = (ps.aftr_isp | ps.user_isp);
3237
3238         return ms;
3239 }
3240
3241 static int receive_req_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3242 {
3243         struct p_req_state *p = &mdev->data.rbuf.req_state;
3244         union drbd_state mask, val;
3245         int rv;
3246
3247         mask.i = be32_to_cpu(p->mask);
3248         val.i = be32_to_cpu(p->val);
3249
3250         if (test_bit(DISCARD_CONCURRENT, &mdev->flags) &&
3251             test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) {
3252                 drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG);
3253                 return TRUE;
3254         }
3255
3256         mask = convert_state(mask);
3257         val = convert_state(val);
3258
3259         rv = drbd_change_state(mdev, CS_VERBOSE, mask, val);
3260
3261         drbd_send_sr_reply(mdev, rv);
3262         drbd_md_sync(mdev);
3263
3264         return TRUE;
3265 }
3266
3267 static int receive_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3268 {
3269         struct p_state *p = &mdev->data.rbuf.state;
3270         union drbd_state os, ns, peer_state;
3271         enum drbd_disk_state real_peer_disk;
3272         enum chg_state_flags cs_flags;
3273         int rv;
3274
3275         peer_state.i = be32_to_cpu(p->state);
3276
3277         real_peer_disk = peer_state.disk;
3278         if (peer_state.disk == D_NEGOTIATING) {
3279                 real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT;
3280                 dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk));
3281         }
3282
3283         spin_lock_irq(&mdev->req_lock);
3284  retry:
3285         os = ns = mdev->state;
3286         spin_unlock_irq(&mdev->req_lock);
3287
3288         /* peer says his disk is uptodate, while we think it is inconsistent,
3289          * and this happens while we think we have a sync going on. */
3290         if (os.pdsk == D_INCONSISTENT && real_peer_disk == D_UP_TO_DATE &&
3291             os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) {
3292                 /* If we are (becoming) SyncSource, but peer is still in sync
3293                  * preparation, ignore its uptodate-ness to avoid flapping, it
3294                  * will change to inconsistent once the peer reaches active
3295                  * syncing states.
3296                  * It may have changed syncer-paused flags, however, so we
3297                  * cannot ignore this completely. */
3298                 if (peer_state.conn > C_CONNECTED &&
3299                     peer_state.conn < C_SYNC_SOURCE)
3300                         real_peer_disk = D_INCONSISTENT;
3301
3302                 /* if peer_state changes to connected at the same time,
3303                  * it explicitly notifies us that it finished resync.
3304                  * Maybe we should finish it up, too? */
3305                 else if (os.conn >= C_SYNC_SOURCE &&
3306                          peer_state.conn == C_CONNECTED) {
3307                         if (drbd_bm_total_weight(mdev) <= mdev->rs_failed)
3308                                 drbd_resync_finished(mdev);
3309                         return TRUE;
3310                 }
3311         }
3312
3313         /* peer says his disk is inconsistent, while we think it is uptodate,
3314          * and this happens while the peer still thinks we have a sync going on,
3315          * but we think we are already done with the sync.
3316          * We ignore this to avoid flapping pdsk.
3317          * This should not happen, if the peer is a recent version of drbd. */
3318         if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT &&
3319             os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE)
3320                 real_peer_disk = D_UP_TO_DATE;
3321
3322         if (ns.conn == C_WF_REPORT_PARAMS)
3323                 ns.conn = C_CONNECTED;
3324
3325         if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING &&
3326             get_ldev_if_state(mdev, D_NEGOTIATING)) {
3327                 int cr; /* consider resync */
3328
3329                 /* if we established a new connection */
3330                 cr  = (os.conn < C_CONNECTED);
3331                 /* if we had an established connection
3332                  * and one of the nodes newly attaches a disk */
3333                 cr |= (os.conn == C_CONNECTED &&
3334                        (peer_state.disk == D_NEGOTIATING ||
3335                         os.disk == D_NEGOTIATING));
3336                 /* if we have both been inconsistent, and the peer has been
3337                  * forced to be UpToDate with --overwrite-data */
3338                 cr |= test_bit(CONSIDER_RESYNC, &mdev->flags);
3339                 /* if we had been plain connected, and the admin requested to
3340                  * start a sync by "invalidate" or "invalidate-remote" */
3341                 cr |= (os.conn == C_CONNECTED &&
3342                                 (peer_state.conn >= C_STARTING_SYNC_S &&
3343                                  peer_state.conn <= C_WF_BITMAP_T));
3344
3345                 if (cr)
3346                         ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk);
3347
3348                 put_ldev(mdev);
3349                 if (ns.conn == C_MASK) {
3350                         ns.conn = C_CONNECTED;
3351                         if (mdev->state.disk == D_NEGOTIATING) {
3352                                 drbd_force_state(mdev, NS(disk, D_DISKLESS));
3353                         } else if (peer_state.disk == D_NEGOTIATING) {
3354                                 dev_err(DEV, "Disk attach process on the peer node was aborted.\n");
3355                                 peer_state.disk = D_DISKLESS;
3356                                 real_peer_disk = D_DISKLESS;
3357                         } else {
3358                                 if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags))
3359                                         return FALSE;
3360                                 D_ASSERT(os.conn == C_WF_REPORT_PARAMS);
3361                                 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3362                                 return FALSE;
3363                         }
3364                 }
3365         }
3366
3367         spin_lock_irq(&mdev->req_lock);
3368         if (mdev->state.i != os.i)
3369                 goto retry;
3370         clear_bit(CONSIDER_RESYNC, &mdev->flags);
3371         ns.peer = peer_state.role;
3372         ns.pdsk = real_peer_disk;
3373         ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp);
3374         if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING)
3375                 ns.disk = mdev->new_state_tmp.disk;
3376         cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD);
3377         if (ns.pdsk == D_CONSISTENT && is_susp(ns) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED &&
3378             test_bit(NEW_CUR_UUID, &mdev->flags)) {
3379                 /* Do not allow tl_restart(resend) for a rebooted peer. We can only allow this
3380                    for temporal network outages! */
3381                 spin_unlock_irq(&mdev->req_lock);
3382                 dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n");
3383                 tl_clear(mdev);
3384                 drbd_uuid_new_current(mdev);
3385                 clear_bit(NEW_CUR_UUID, &mdev->flags);
3386                 drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0));
3387                 return FALSE;
3388         }
3389         rv = _drbd_set_state(mdev, ns, cs_flags, NULL);
3390         ns = mdev->state;
3391         spin_unlock_irq(&mdev->req_lock);
3392
3393         if (rv < SS_SUCCESS) {
3394                 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3395                 return FALSE;
3396         }
3397
3398         if (os.conn > C_WF_REPORT_PARAMS) {
3399                 if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED &&
3400                     peer_state.disk != D_NEGOTIATING ) {
3401                         /* we want resync, peer has not yet decided to sync... */
3402                         /* Nowadays only used when forcing a node into primary role and
3403                            setting its disk to UpToDate with that */
3404                         drbd_send_uuids(mdev);
3405                         drbd_send_state(mdev);
3406                 }
3407         }
3408
3409         mdev->net_conf->want_lose = 0;
3410
3411         drbd_md_sync(mdev); /* update connected indicator, la_size, ... */
3412
3413         return TRUE;
3414 }
3415
3416 static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3417 {
3418         struct p_rs_uuid *p = &mdev->data.rbuf.rs_uuid;
3419
3420         wait_event(mdev->misc_wait,
3421                    mdev->state.conn == C_WF_SYNC_UUID ||
3422                    mdev->state.conn < C_CONNECTED ||
3423                    mdev->state.disk < D_NEGOTIATING);
3424
3425         /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */
3426
3427         /* Here the _drbd_uuid_ functions are right, current should
3428            _not_ be rotated into the history */
3429         if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
3430                 _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid));
3431                 _drbd_uuid_set(mdev, UI_BITMAP, 0UL);
3432
3433                 drbd_start_resync(mdev, C_SYNC_TARGET);
3434
3435                 put_ldev(mdev);
3436         } else
3437                 dev_err(DEV, "Ignoring SyncUUID packet!\n");
3438
3439         return TRUE;
3440 }
3441
3442 enum receive_bitmap_ret { OK, DONE, FAILED };
3443
3444 static enum receive_bitmap_ret
3445 receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size,
3446                      unsigned long *buffer, struct bm_xfer_ctx *c)
3447 {
3448         unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset);
3449         unsigned want = num_words * sizeof(long);
3450
3451         if (want != data_size) {
3452                 dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size);
3453                 return FAILED;
3454         }
3455         if (want == 0)
3456                 return DONE;
3457         if (drbd_recv(mdev, buffer, want) != want)
3458                 return FAILED;
3459
3460         drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer);
3461
3462         c->word_offset += num_words;
3463         c->bit_offset = c->word_offset * BITS_PER_LONG;
3464         if (c->bit_offset > c->bm_bits)
3465                 c->bit_offset = c->bm_bits;
3466
3467         return OK;
3468 }
3469
3470 static enum receive_bitmap_ret
3471 recv_bm_rle_bits(struct drbd_conf *mdev,
3472                 struct p_compressed_bm *p,
3473                 struct bm_xfer_ctx *c)
3474 {
3475         struct bitstream bs;
3476         u64 look_ahead;
3477         u64 rl;
3478         u64 tmp;
3479         unsigned long s = c->bit_offset;
3480         unsigned long e;
3481         int len = be16_to_cpu(p->head.length) - (sizeof(*p) - sizeof(p->head));
3482         int toggle = DCBP_get_start(p);
3483         int have;
3484         int bits;
3485
3486         bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p));
3487
3488         bits = bitstream_get_bits(&bs, &look_ahead, 64);
3489         if (bits < 0)
3490                 return FAILED;
3491
3492         for (have = bits; have > 0; s += rl, toggle = !toggle) {
3493                 bits = vli_decode_bits(&rl, look_ahead);
3494                 if (bits <= 0)
3495                         return FAILED;
3496
3497                 if (toggle) {
3498                         e = s + rl -1;
3499                         if (e >= c->bm_bits) {
3500                                 dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e);
3501                                 return FAILED;
3502                         }
3503                         _drbd_bm_set_bits(mdev, s, e);
3504                 }
3505
3506                 if (have < bits) {
3507                         dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n",
3508                                 have, bits, look_ahead,
3509                                 (unsigned int)(bs.cur.b - p->code),
3510                                 (unsigned int)bs.buf_len);
3511                         return FAILED;
3512                 }
3513                 look_ahead >>= bits;
3514                 have -= bits;
3515
3516                 bits = bitstream_get_bits(&bs, &tmp, 64 - have);
3517                 if (bits < 0)
3518                         return FAILED;
3519                 look_ahead |= tmp << have;
3520                 have += bits;
3521         }
3522
3523         c->bit_offset = s;
3524         bm_xfer_ctx_bit_to_word_offset(c);
3525
3526         return (s == c->bm_bits) ? DONE : OK;
3527 }
3528
3529 static enum receive_bitmap_ret
3530 decode_bitmap_c(struct drbd_conf *mdev,
3531                 struct p_compressed_bm *p,
3532                 struct bm_xfer_ctx *c)
3533 {
3534         if (DCBP_get_code(p) == RLE_VLI_Bits)
3535                 return recv_bm_rle_bits(mdev, p, c);
3536
3537         /* other variants had been implemented for evaluation,
3538          * but have been dropped as this one turned out to be "best"
3539          * during all our tests. */
3540
3541         dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding);
3542         drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3543         return FAILED;
3544 }
3545
3546 void INFO_bm_xfer_stats(struct drbd_conf *mdev,
3547                 const char *direction, struct bm_xfer_ctx *c)
3548 {
3549         /* what would it take to transfer it "plaintext" */
3550         unsigned plain = sizeof(struct p_header80) *
3551                 ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1)
3552                 + c->bm_words * sizeof(long);
3553         unsigned total = c->bytes[0] + c->bytes[1];
3554         unsigned r;
3555
3556         /* total can not be zero. but just in case: */
3557         if (total == 0)
3558                 return;
3559
3560         /* don't report if not compressed */
3561         if (total >= plain)
3562                 return;
3563
3564         /* total < plain. check for overflow, still */
3565         r = (total > UINT_MAX/1000) ? (total / (plain/1000))
3566                                     : (1000 * total / plain);
3567
3568         if (r > 1000)
3569                 r = 1000;
3570
3571         r = 1000 - r;
3572         dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), "
3573              "total %u; compression: %u.%u%%\n",
3574                         direction,
3575                         c->bytes[1], c->packets[1],
3576                         c->bytes[0], c->packets[0],
3577                         total, r/10, r % 10);
3578 }
3579
3580 /* Since we are processing the bitfield from lower addresses to higher,
3581    it does not matter if the process it in 32 bit chunks or 64 bit
3582    chunks as long as it is little endian. (Understand it as byte stream,
3583    beginning with the lowest byte...) If we would use big endian
3584    we would need to process it from the highest address to the lowest,
3585    in order to be agnostic to the 32 vs 64 bits issue.
3586
3587    returns 0 on failure, 1 if we successfully received it. */
3588 static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3589 {
3590         struct bm_xfer_ctx c;
3591         void *buffer;
3592         enum receive_bitmap_ret ret;
3593         int ok = FALSE;
3594         struct p_header80 *h = &mdev->data.rbuf.header.h80;
3595
3596         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
3597
3598         drbd_bm_lock(mdev, "receive bitmap");
3599
3600         /* maybe we should use some per thread scratch page,
3601          * and allocate that during initial device creation? */
3602         buffer   = (unsigned long *) __get_free_page(GFP_NOIO);
3603         if (!buffer) {
3604                 dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__);
3605                 goto out;
3606         }
3607
3608         c = (struct bm_xfer_ctx) {
3609                 .bm_bits = drbd_bm_bits(mdev),
3610                 .bm_words = drbd_bm_words(mdev),
3611         };
3612
3613         do {
3614                 if (cmd == P_BITMAP) {
3615                         ret = receive_bitmap_plain(mdev, data_size, buffer, &c);
3616                 } else if (cmd == P_COMPRESSED_BITMAP) {
3617                         /* MAYBE: sanity check that we speak proto >= 90,
3618                          * and the feature is enabled! */
3619                         struct p_compressed_bm *p;
3620
3621                         if (data_size > BM_PACKET_PAYLOAD_BYTES) {
3622                                 dev_err(DEV, "ReportCBitmap packet too large\n");
3623                                 goto out;
3624                         }
3625                         /* use the page buff */
3626                         p = buffer;
3627                         memcpy(p, h, sizeof(*h));
3628                         if (drbd_recv(mdev, p->head.payload, data_size) != data_size)
3629                                 goto out;
3630                         if (data_size <= (sizeof(*p) - sizeof(p->head))) {
3631                                 dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size);
3632                                 return FAILED;
3633                         }
3634                         ret = decode_bitmap_c(mdev, p, &c);
3635                 } else {
3636                         dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd);
3637                         goto out;
3638                 }
3639
3640                 c.packets[cmd == P_BITMAP]++;
3641                 c.bytes[cmd == P_BITMAP] += sizeof(struct p_header80) + data_size;
3642
3643                 if (ret != OK)
3644                         break;
3645
3646                 if (!drbd_recv_header(mdev, &cmd, &data_size))
3647                         goto out;
3648         } while (ret == OK);
3649         if (ret == FAILED)
3650                 goto out;
3651
3652         INFO_bm_xfer_stats(mdev, "receive", &c);
3653
3654         if (mdev->state.conn == C_WF_BITMAP_T) {
3655                 ok = !drbd_send_bitmap(mdev);
3656                 if (!ok)
3657                         goto out;
3658                 /* Omit CS_ORDERED with this state transition to avoid deadlocks. */
3659                 ok = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
3660                 D_ASSERT(ok == SS_SUCCESS);
3661         } else if (mdev->state.conn != C_WF_BITMAP_S) {
3662                 /* admin may have requested C_DISCONNECTING,
3663                  * other threads may have noticed network errors */
3664                 dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n",
3665                     drbd_conn_str(mdev->state.conn));
3666         }
3667
3668         ok = TRUE;
3669  out:
3670         drbd_bm_unlock(mdev);
3671         if (ok && mdev->state.conn == C_WF_BITMAP_S)
3672                 drbd_start_resync(mdev, C_SYNC_SOURCE);
3673         free_page((unsigned long) buffer);
3674         return ok;
3675 }
3676
3677 static int receive_skip(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3678 {
3679         /* TODO zero copy sink :) */
3680         static char sink[128];
3681         int size, want, r;
3682
3683         dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n",
3684                  cmd, data_size);
3685
3686         size = data_size;
3687         while (size > 0) {
3688                 want = min_t(int, size, sizeof(sink));
3689                 r = drbd_recv(mdev, sink, want);
3690                 ERR_IF(r <= 0) break;
3691                 size -= r;
3692         }
3693         return size == 0;
3694 }
3695
3696 static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3697 {
3698         if (mdev->state.disk >= D_INCONSISTENT)
3699                 drbd_kick_lo(mdev);
3700
3701         /* Make sure we've acked all the TCP data associated
3702          * with the data requests being unplugged */
3703         drbd_tcp_quickack(mdev->data.socket);
3704
3705         return TRUE;
3706 }
3707
3708 typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packets cmd, unsigned int to_receive);
3709
3710 struct data_cmd {
3711         int expect_payload;
3712         size_t pkt_size;
3713         drbd_cmd_handler_f function;
3714 };
3715
3716 static struct data_cmd drbd_cmd_handler[] = {
3717         [P_DATA]            = { 1, sizeof(struct p_data), receive_Data },
3718         [P_DATA_REPLY]      = { 1, sizeof(struct p_data), receive_DataReply },
3719         [P_RS_DATA_REPLY]   = { 1, sizeof(struct p_data), receive_RSDataReply } ,
3720         [P_BARRIER]         = { 0, sizeof(struct p_barrier), receive_Barrier } ,
3721         [P_BITMAP]          = { 1, sizeof(struct p_header80), receive_bitmap } ,
3722         [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } ,
3723         [P_UNPLUG_REMOTE]   = { 0, sizeof(struct p_header80), receive_UnplugRemote },
3724         [P_DATA_REQUEST]    = { 0, sizeof(struct p_block_req), receive_DataRequest },
3725         [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3726         [P_SYNC_PARAM]      = { 1, sizeof(struct p_header80), receive_SyncParam },
3727         [P_SYNC_PARAM89]    = { 1, sizeof(struct p_header80), receive_SyncParam },
3728         [P_PROTOCOL]        = { 1, sizeof(struct p_protocol), receive_protocol },
3729         [P_UUIDS]           = { 0, sizeof(struct p_uuids), receive_uuids },
3730         [P_SIZES]           = { 0, sizeof(struct p_sizes), receive_sizes },
3731         [P_STATE]           = { 0, sizeof(struct p_state), receive_state },
3732         [P_STATE_CHG_REQ]   = { 0, sizeof(struct p_req_state), receive_req_state },
3733         [P_SYNC_UUID]       = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid },
3734         [P_OV_REQUEST]      = { 0, sizeof(struct p_block_req), receive_DataRequest },
3735         [P_OV_REPLY]        = { 1, sizeof(struct p_block_req), receive_DataRequest },
3736         [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest },
3737         [P_DELAY_PROBE]     = { 0, sizeof(struct p_delay_probe93), receive_skip },
3738         /* anything missing from this table is in
3739          * the asender_tbl, see get_asender_cmd */
3740         [P_MAX_CMD]         = { 0, 0, NULL },
3741 };
3742
3743 /* All handler functions that expect a sub-header get that sub-heder in
3744    mdev->data.rbuf.header.head.payload.
3745
3746    Usually in mdev->data.rbuf.header.head the callback can find the usual
3747    p_header, but they may not rely on that. Since there is also p_header95 !
3748  */
3749
3750 static void drbdd(struct drbd_conf *mdev)
3751 {
3752         union p_header *header = &mdev->data.rbuf.header;
3753         unsigned int packet_size;
3754         enum drbd_packets cmd;
3755         size_t shs; /* sub header size */
3756         int rv;
3757
3758         while (get_t_state(&mdev->receiver) == Running) {
3759                 drbd_thread_current_set_cpu(mdev);
3760                 if (!drbd_recv_header(mdev, &cmd, &packet_size))
3761                         goto err_out;
3762
3763                 if (unlikely(cmd >= P_MAX_CMD || !drbd_cmd_handler[cmd].function)) {
3764                         dev_err(DEV, "unknown packet type %d, l: %d!\n", cmd, packet_size);
3765                         goto err_out;
3766                 }
3767
3768                 shs = drbd_cmd_handler[cmd].pkt_size - sizeof(union p_header);
3769                 rv = drbd_recv(mdev, &header->h80.payload, shs);
3770                 if (unlikely(rv != shs)) {
3771                         dev_err(DEV, "short read while reading sub header: rv=%d\n", rv);
3772                         goto err_out;
3773                 }
3774
3775                 if (packet_size - shs > 0 && !drbd_cmd_handler[cmd].expect_payload) {
3776                         dev_err(DEV, "No payload expected %s l:%d\n", cmdname(cmd), packet_size);
3777                         goto err_out;
3778                 }
3779
3780                 rv = drbd_cmd_handler[cmd].function(mdev, cmd, packet_size - shs);
3781
3782                 if (unlikely(!rv)) {
3783                         dev_err(DEV, "error receiving %s, l: %d!\n",
3784                             cmdname(cmd), packet_size);
3785                         goto err_out;
3786                 }
3787         }
3788
3789         if (0) {
3790         err_out:
3791                 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3792         }
3793 }
3794
3795 void drbd_flush_workqueue(struct drbd_conf *mdev)
3796 {
3797         struct drbd_wq_barrier barr;
3798
3799         barr.w.cb = w_prev_work_done;
3800         init_completion(&barr.done);
3801         drbd_queue_work(&mdev->data.work, &barr.w);
3802         wait_for_completion(&barr.done);
3803 }
3804
3805 void drbd_free_tl_hash(struct drbd_conf *mdev)
3806 {
3807         struct hlist_head *h;
3808
3809         spin_lock_irq(&mdev->req_lock);
3810
3811         if (!mdev->tl_hash || mdev->state.conn != C_STANDALONE) {
3812                 spin_unlock_irq(&mdev->req_lock);
3813                 return;
3814         }
3815         /* paranoia code */
3816         for (h = mdev->ee_hash; h < mdev->ee_hash + mdev->ee_hash_s; h++)
3817                 if (h->first)
3818                         dev_err(DEV, "ASSERT FAILED ee_hash[%u].first == %p, expected NULL\n",
3819                                 (int)(h - mdev->ee_hash), h->first);
3820         kfree(mdev->ee_hash);
3821         mdev->ee_hash = NULL;
3822         mdev->ee_hash_s = 0;
3823
3824         /* paranoia code */
3825         for (h = mdev->tl_hash; h < mdev->tl_hash + mdev->tl_hash_s; h++)
3826                 if (h->first)
3827                         dev_err(DEV, "ASSERT FAILED tl_hash[%u] == %p, expected NULL\n",
3828                                 (int)(h - mdev->tl_hash), h->first);
3829         kfree(mdev->tl_hash);
3830         mdev->tl_hash = NULL;
3831         mdev->tl_hash_s = 0;
3832         spin_unlock_irq(&mdev->req_lock);
3833 }
3834
3835 static void drbd_disconnect(struct drbd_conf *mdev)
3836 {
3837         enum drbd_fencing_p fp;
3838         union drbd_state os, ns;
3839         int rv = SS_UNKNOWN_ERROR;
3840         unsigned int i;
3841
3842         if (mdev->state.conn == C_STANDALONE)
3843                 return;
3844         if (mdev->state.conn >= C_WF_CONNECTION)
3845                 dev_err(DEV, "ASSERT FAILED cstate = %s, expected < WFConnection\n",
3846                                 drbd_conn_str(mdev->state.conn));
3847
3848         /* asender does not clean up anything. it must not interfere, either */
3849         drbd_thread_stop(&mdev->asender);
3850         drbd_free_sock(mdev);
3851
3852         /* wait for current activity to cease. */
3853         spin_lock_irq(&mdev->req_lock);
3854         _drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
3855         _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee);
3856         _drbd_wait_ee_list_empty(mdev, &mdev->read_ee);
3857         spin_unlock_irq(&mdev->req_lock);
3858
3859         /* We do not have data structures that would allow us to
3860          * get the rs_pending_cnt down to 0 again.
3861          *  * On C_SYNC_TARGET we do not have any data structures describing
3862          *    the pending RSDataRequest's we have sent.
3863          *  * On C_SYNC_SOURCE there is no data structure that tracks
3864          *    the P_RS_DATA_REPLY blocks that we sent to the SyncTarget.
3865          *  And no, it is not the sum of the reference counts in the
3866          *  resync_LRU. The resync_LRU tracks the whole operation including
3867          *  the disk-IO, while the rs_pending_cnt only tracks the blocks
3868          *  on the fly. */
3869         drbd_rs_cancel_all(mdev);
3870         mdev->rs_total = 0;
3871         mdev->rs_failed = 0;
3872         atomic_set(&mdev->rs_pending_cnt, 0);
3873         wake_up(&mdev->misc_wait);
3874
3875         /* make sure syncer is stopped and w_resume_next_sg queued */
3876         del_timer_sync(&mdev->resync_timer);
3877         resync_timer_fn((unsigned long)mdev);
3878
3879         /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier,
3880          * w_make_resync_request etc. which may still be on the worker queue
3881          * to be "canceled" */
3882         drbd_flush_workqueue(mdev);
3883
3884         /* This also does reclaim_net_ee().  If we do this too early, we might
3885          * miss some resync ee and pages.*/
3886         drbd_process_done_ee(mdev);
3887
3888         kfree(mdev->p_uuid);
3889         mdev->p_uuid = NULL;
3890
3891         if (!is_susp(mdev->state))
3892                 tl_clear(mdev);
3893
3894         dev_info(DEV, "Connection closed\n");
3895
3896         drbd_md_sync(mdev);
3897
3898         fp = FP_DONT_CARE;
3899         if (get_ldev(mdev)) {
3900                 fp = mdev->ldev->dc.fencing;
3901                 put_ldev(mdev);
3902         }
3903
3904         if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN)
3905                 drbd_try_outdate_peer_async(mdev);
3906
3907         spin_lock_irq(&mdev->req_lock);
3908         os = mdev->state;
3909         if (os.conn >= C_UNCONNECTED) {
3910                 /* Do not restart in case we are C_DISCONNECTING */
3911                 ns = os;
3912                 ns.conn = C_UNCONNECTED;
3913                 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
3914         }
3915         spin_unlock_irq(&mdev->req_lock);
3916
3917         if (os.conn == C_DISCONNECTING) {
3918                 wait_event(mdev->net_cnt_wait, atomic_read(&mdev->net_cnt) == 0);
3919
3920                 if (!is_susp(mdev->state)) {
3921                         /* we must not free the tl_hash
3922                          * while application io is still on the fly */
3923                         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
3924                         drbd_free_tl_hash(mdev);
3925                 }
3926
3927                 crypto_free_hash(mdev->cram_hmac_tfm);
3928                 mdev->cram_hmac_tfm = NULL;
3929
3930                 kfree(mdev->net_conf);
3931                 mdev->net_conf = NULL;
3932                 drbd_request_state(mdev, NS(conn, C_STANDALONE));
3933         }
3934
3935         /* tcp_close and release of sendpage pages can be deferred.  I don't
3936          * want to use SO_LINGER, because apparently it can be deferred for
3937          * more than 20 seconds (longest time I checked).
3938          *
3939          * Actually we don't care for exactly when the network stack does its
3940          * put_page(), but release our reference on these pages right here.
3941          */
3942         i = drbd_release_ee(mdev, &mdev->net_ee);
3943         if (i)
3944                 dev_info(DEV, "net_ee not empty, killed %u entries\n", i);
3945         i = atomic_read(&mdev->pp_in_use_by_net);
3946         if (i)
3947                 dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i);
3948         i = atomic_read(&mdev->pp_in_use);
3949         if (i)
3950                 dev_info(DEV, "pp_in_use = %d, expected 0\n", i);
3951
3952         D_ASSERT(list_empty(&mdev->read_ee));
3953         D_ASSERT(list_empty(&mdev->active_ee));
3954         D_ASSERT(list_empty(&mdev->sync_ee));
3955         D_ASSERT(list_empty(&mdev->done_ee));
3956
3957         /* ok, no more ee's on the fly, it is safe to reset the epoch_size */
3958         atomic_set(&mdev->current_epoch->epoch_size, 0);
3959         D_ASSERT(list_empty(&mdev->current_epoch->list));
3960 }
3961
3962 /*
3963  * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version
3964  * we can agree on is stored in agreed_pro_version.
3965  *
3966  * feature flags and the reserved array should be enough room for future
3967  * enhancements of the handshake protocol, and possible plugins...
3968  *
3969  * for now, they are expected to be zero, but ignored.
3970  */
3971 static int drbd_send_handshake(struct drbd_conf *mdev)
3972 {
3973         /* ASSERT current == mdev->receiver ... */
3974         struct p_handshake *p = &mdev->data.sbuf.handshake;
3975         int ok;
3976
3977         if (mutex_lock_interruptible(&mdev->data.mutex)) {
3978                 dev_err(DEV, "interrupted during initial handshake\n");
3979                 return 0; /* interrupted. not ok. */
3980         }
3981
3982         if (mdev->data.socket == NULL) {
3983                 mutex_unlock(&mdev->data.mutex);
3984                 return 0;
3985         }
3986
3987         memset(p, 0, sizeof(*p));
3988         p->protocol_min = cpu_to_be32(PRO_VERSION_MIN);
3989         p->protocol_max = cpu_to_be32(PRO_VERSION_MAX);
3990         ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE,
3991                              (struct p_header80 *)p, sizeof(*p), 0 );
3992         mutex_unlock(&mdev->data.mutex);
3993         return ok;
3994 }
3995
3996 /*
3997  * return values:
3998  *   1 yes, we have a valid connection
3999  *   0 oops, did not work out, please try again
4000  *  -1 peer talks different language,
4001  *     no point in trying again, please go standalone.
4002  */
4003 static int drbd_do_handshake(struct drbd_conf *mdev)
4004 {
4005         /* ASSERT current == mdev->receiver ... */
4006         struct p_handshake *p = &mdev->data.rbuf.handshake;
4007         const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80);
4008         unsigned int length;
4009         enum drbd_packets cmd;
4010         int rv;
4011
4012         rv = drbd_send_handshake(mdev);
4013         if (!rv)
4014                 return 0;
4015
4016         rv = drbd_recv_header(mdev, &cmd, &length);
4017         if (!rv)
4018                 return 0;
4019
4020         if (cmd != P_HAND_SHAKE) {
4021                 dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n",
4022                      cmdname(cmd), cmd);
4023                 return -1;
4024         }
4025
4026         if (length != expect) {
4027                 dev_err(DEV, "expected HandShake length: %u, received: %u\n",
4028                      expect, length);
4029                 return -1;
4030         }
4031
4032         rv = drbd_recv(mdev, &p->head.payload, expect);
4033
4034         if (rv != expect) {
4035                 dev_err(DEV, "short read receiving handshake packet: l=%u\n", rv);
4036                 return 0;
4037         }
4038
4039         p->protocol_min = be32_to_cpu(p->protocol_min);
4040         p->protocol_max = be32_to_cpu(p->protocol_max);
4041         if (p->protocol_max == 0)
4042                 p->protocol_max = p->protocol_min;
4043
4044         if (PRO_VERSION_MAX < p->protocol_min ||
4045             PRO_VERSION_MIN > p->protocol_max)
4046                 goto incompat;
4047
4048         mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max);
4049
4050         dev_info(DEV, "Handshake successful: "
4051              "Agreed network protocol version %d\n", mdev->agreed_pro_version);
4052
4053         return 1;
4054
4055  incompat:
4056         dev_err(DEV, "incompatible DRBD dialects: "
4057             "I support %d-%d, peer supports %d-%d\n",
4058             PRO_VERSION_MIN, PRO_VERSION_MAX,
4059             p->protocol_min, p->protocol_max);
4060         return -1;
4061 }
4062
4063 #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE)
4064 static int drbd_do_auth(struct drbd_conf *mdev)
4065 {
4066         dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n");
4067         dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n");
4068         return -1;
4069 }
4070 #else
4071 #define CHALLENGE_LEN 64
4072
4073 /* Return value:
4074         1 - auth succeeded,
4075         0 - failed, try again (network error),
4076         -1 - auth failed, don't try again.
4077 */
4078
4079 static int drbd_do_auth(struct drbd_conf *mdev)
4080 {
4081         char my_challenge[CHALLENGE_LEN];  /* 64 Bytes... */
4082         struct scatterlist sg;
4083         char *response = NULL;
4084         char *right_response = NULL;
4085         char *peers_ch = NULL;
4086         unsigned int key_len = strlen(mdev->net_conf->shared_secret);
4087         unsigned int resp_size;
4088         struct hash_desc desc;
4089         enum drbd_packets cmd;
4090         unsigned int length;
4091         int rv;
4092
4093         desc.tfm = mdev->cram_hmac_tfm;
4094         desc.flags = 0;
4095
4096         rv = crypto_hash_setkey(mdev->cram_hmac_tfm,
4097                                 (u8 *)mdev->net_conf->shared_secret, key_len);
4098         if (rv) {
4099                 dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv);
4100                 rv = -1;
4101                 goto fail;
4102         }
4103
4104         get_random_bytes(my_challenge, CHALLENGE_LEN);
4105
4106         rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN);
4107         if (!rv)
4108                 goto fail;
4109
4110         rv = drbd_recv_header(mdev, &cmd, &length);
4111         if (!rv)
4112                 goto fail;
4113
4114         if (cmd != P_AUTH_CHALLENGE) {
4115                 dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n",
4116                     cmdname(cmd), cmd);
4117                 rv = 0;
4118                 goto fail;
4119         }
4120
4121         if (length > CHALLENGE_LEN * 2) {
4122                 dev_err(DEV, "expected AuthChallenge payload too big.\n");
4123                 rv = -1;
4124                 goto fail;
4125         }
4126
4127         peers_ch = kmalloc(length, GFP_NOIO);
4128         if (peers_ch == NULL) {
4129                 dev_err(DEV, "kmalloc of peers_ch failed\n");
4130                 rv = -1;
4131                 goto fail;
4132         }
4133
4134         rv = drbd_recv(mdev, peers_ch, length);
4135
4136         if (rv != length) {
4137                 dev_err(DEV, "short read AuthChallenge: l=%u\n", rv);
4138                 rv = 0;
4139                 goto fail;
4140         }
4141
4142         resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm);
4143         response = kmalloc(resp_size, GFP_NOIO);
4144         if (response == NULL) {
4145                 dev_err(DEV, "kmalloc of response failed\n");
4146                 rv = -1;
4147                 goto fail;
4148         }
4149
4150         sg_init_table(&sg, 1);
4151         sg_set_buf(&sg, peers_ch, length);
4152
4153         rv = crypto_hash_digest(&desc, &sg, sg.length, response);
4154         if (rv) {
4155                 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
4156                 rv = -1;
4157                 goto fail;
4158         }
4159
4160         rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size);
4161         if (!rv)
4162                 goto fail;
4163
4164         rv = drbd_recv_header(mdev, &cmd, &length);
4165         if (!rv)
4166                 goto fail;
4167
4168         if (cmd != P_AUTH_RESPONSE) {
4169                 dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n",
4170                         cmdname(cmd), cmd);
4171                 rv = 0;
4172                 goto fail;
4173         }
4174
4175         if (length != resp_size) {
4176                 dev_err(DEV, "expected AuthResponse payload of wrong size\n");
4177                 rv = 0;
4178                 goto fail;
4179         }
4180
4181         rv = drbd_recv(mdev, response , resp_size);
4182
4183         if (rv != resp_size) {
4184                 dev_err(DEV, "short read receiving AuthResponse: l=%u\n", rv);
4185                 rv = 0;
4186                 goto fail;
4187         }
4188
4189         right_response = kmalloc(resp_size, GFP_NOIO);
4190         if (right_response == NULL) {
4191                 dev_err(DEV, "kmalloc of right_response failed\n");
4192                 rv = -1;
4193                 goto fail;
4194         }
4195
4196         sg_set_buf(&sg, my_challenge, CHALLENGE_LEN);
4197
4198         rv = crypto_hash_digest(&desc, &sg, sg.length, right_response);
4199         if (rv) {
4200                 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
4201                 rv = -1;
4202                 goto fail;
4203         }
4204
4205         rv = !memcmp(response, right_response, resp_size);
4206
4207         if (rv)
4208                 dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n",
4209                      resp_size, mdev->net_conf->cram_hmac_alg);
4210         else
4211                 rv = -1;
4212
4213  fail:
4214         kfree(peers_ch);
4215         kfree(response);
4216         kfree(right_response);
4217
4218         return rv;
4219 }
4220 #endif
4221
4222 int drbdd_init(struct drbd_thread *thi)
4223 {
4224         struct drbd_conf *mdev = thi->mdev;
4225         unsigned int minor = mdev_to_minor(mdev);
4226         int h;
4227
4228         sprintf(current->comm, "drbd%d_receiver", minor);
4229
4230         dev_info(DEV, "receiver (re)started\n");
4231
4232         do {
4233                 h = drbd_connect(mdev);
4234                 if (h == 0) {
4235                         drbd_disconnect(mdev);
4236                         __set_current_state(TASK_INTERRUPTIBLE);
4237                         schedule_timeout(HZ);
4238                 }
4239                 if (h == -1) {
4240                         dev_warn(DEV, "Discarding network configuration.\n");
4241                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4242                 }
4243         } while (h == 0);
4244
4245         if (h > 0) {
4246                 if (get_net_conf(mdev)) {
4247                         drbdd(mdev);
4248                         put_net_conf(mdev);
4249                 }
4250         }
4251
4252         drbd_disconnect(mdev);
4253
4254         dev_info(DEV, "receiver terminated\n");
4255         return 0;
4256 }
4257
4258 /* ********* acknowledge sender ******** */
4259
4260 static int got_RqSReply(struct drbd_conf *mdev, struct p_header80 *h)
4261 {
4262         struct p_req_state_reply *p = (struct p_req_state_reply *)h;
4263
4264         int retcode = be32_to_cpu(p->retcode);
4265
4266         if (retcode >= SS_SUCCESS) {
4267                 set_bit(CL_ST_CHG_SUCCESS, &mdev->flags);
4268         } else {
4269                 set_bit(CL_ST_CHG_FAIL, &mdev->flags);
4270                 dev_err(DEV, "Requested state change failed by peer: %s (%d)\n",
4271                     drbd_set_st_err_str(retcode), retcode);
4272         }
4273         wake_up(&mdev->state_wait);
4274
4275         return TRUE;
4276 }
4277
4278 static int got_Ping(struct drbd_conf *mdev, struct p_header80 *h)
4279 {
4280         return drbd_send_ping_ack(mdev);
4281
4282 }
4283
4284 static int got_PingAck(struct drbd_conf *mdev, struct p_header80 *h)
4285 {
4286         /* restore idle timeout */
4287         mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ;
4288         if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags))
4289                 wake_up(&mdev->misc_wait);
4290
4291         return TRUE;
4292 }
4293
4294 static int got_IsInSync(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         int blksize = be32_to_cpu(p->blksize);
4299
4300         D_ASSERT(mdev->agreed_pro_version >= 89);
4301
4302         update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4303
4304         if (get_ldev(mdev)) {
4305                 drbd_rs_complete_io(mdev, sector);
4306                 drbd_set_in_sync(mdev, sector, blksize);
4307                 /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */
4308                 mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT);
4309                 put_ldev(mdev);
4310         }
4311         dec_rs_pending(mdev);
4312         atomic_add(blksize >> 9, &mdev->rs_sect_in);
4313
4314         return TRUE;
4315 }
4316
4317 /* when we receive the ACK for a write request,
4318  * verify that we actually know about it */
4319 static struct drbd_request *_ack_id_to_req(struct drbd_conf *mdev,
4320         u64 id, sector_t sector)
4321 {
4322         struct hlist_head *slot = tl_hash_slot(mdev, sector);
4323         struct hlist_node *n;
4324         struct drbd_request *req;
4325
4326         hlist_for_each_entry(req, n, slot, colision) {
4327                 if ((unsigned long)req == (unsigned long)id) {
4328                         if (req->sector != sector) {
4329                                 dev_err(DEV, "_ack_id_to_req: found req %p but it has "
4330                                     "wrong sector (%llus versus %llus)\n", req,
4331                                     (unsigned long long)req->sector,
4332                                     (unsigned long long)sector);
4333                                 break;
4334                         }
4335                         return req;
4336                 }
4337         }
4338         dev_err(DEV, "_ack_id_to_req: failed to find req %p, sector %llus in list\n",
4339                 (void *)(unsigned long)id, (unsigned long long)sector);
4340         return NULL;
4341 }
4342
4343 typedef struct drbd_request *(req_validator_fn)
4344         (struct drbd_conf *mdev, u64 id, sector_t sector);
4345
4346 static int validate_req_change_req_state(struct drbd_conf *mdev,
4347         u64 id, sector_t sector, req_validator_fn validator,
4348         const char *func, enum drbd_req_event what)
4349 {
4350         struct drbd_request *req;
4351         struct bio_and_error m;
4352
4353         spin_lock_irq(&mdev->req_lock);
4354         req = validator(mdev, id, sector);
4355         if (unlikely(!req)) {
4356                 spin_unlock_irq(&mdev->req_lock);
4357                 dev_err(DEV, "%s: got a corrupt block_id/sector pair\n", func);
4358                 return FALSE;
4359         }
4360         __req_mod(req, what, &m);
4361         spin_unlock_irq(&mdev->req_lock);
4362
4363         if (m.bio)
4364                 complete_master_bio(mdev, &m);
4365         return TRUE;
4366 }
4367
4368 static int got_BlockAck(struct drbd_conf *mdev, struct p_header80 *h)
4369 {
4370         struct p_block_ack *p = (struct p_block_ack *)h;
4371         sector_t sector = be64_to_cpu(p->sector);
4372         int blksize = be32_to_cpu(p->blksize);
4373         enum drbd_req_event what;
4374
4375         update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4376
4377         if (is_syncer_block_id(p->block_id)) {
4378                 drbd_set_in_sync(mdev, sector, blksize);
4379                 dec_rs_pending(mdev);
4380                 return TRUE;
4381         }
4382         switch (be16_to_cpu(h->command)) {
4383         case P_RS_WRITE_ACK:
4384                 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4385                 what = write_acked_by_peer_and_sis;
4386                 break;
4387         case P_WRITE_ACK:
4388                 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4389                 what = write_acked_by_peer;
4390                 break;
4391         case P_RECV_ACK:
4392                 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B);
4393                 what = recv_acked_by_peer;
4394                 break;
4395         case P_DISCARD_ACK:
4396                 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4397                 what = conflict_discarded_by_peer;
4398                 break;
4399         default:
4400                 D_ASSERT(0);
4401                 return FALSE;
4402         }
4403
4404         return validate_req_change_req_state(mdev, p->block_id, sector,
4405                 _ack_id_to_req, __func__ , what);
4406 }
4407
4408 static int got_NegAck(struct drbd_conf *mdev, struct p_header80 *h)
4409 {
4410         struct p_block_ack *p = (struct p_block_ack *)h;
4411         sector_t sector = be64_to_cpu(p->sector);
4412
4413         if (__ratelimit(&drbd_ratelimit_state))
4414                 dev_warn(DEV, "Got NegAck packet. Peer is in troubles?\n");
4415
4416         update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4417
4418         if (is_syncer_block_id(p->block_id)) {
4419                 int size = be32_to_cpu(p->blksize);
4420                 dec_rs_pending(mdev);
4421                 drbd_rs_failed_io(mdev, sector, size);
4422                 return TRUE;
4423         }
4424         return validate_req_change_req_state(mdev, p->block_id, sector,
4425                 _ack_id_to_req, __func__ , neg_acked);
4426 }
4427
4428 static int got_NegDReply(struct drbd_conf *mdev, struct p_header80 *h)
4429 {
4430         struct p_block_ack *p = (struct p_block_ack *)h;
4431         sector_t sector = be64_to_cpu(p->sector);
4432
4433         update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4434         dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n",
4435             (unsigned long long)sector, be32_to_cpu(p->blksize));
4436
4437         return validate_req_change_req_state(mdev, p->block_id, sector,
4438                 _ar_id_to_req, __func__ , neg_acked);
4439 }
4440
4441 static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header80 *h)
4442 {
4443         sector_t sector;
4444         int size;
4445         struct p_block_ack *p = (struct p_block_ack *)h;
4446
4447         sector = be64_to_cpu(p->sector);
4448         size = be32_to_cpu(p->blksize);
4449
4450         update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4451
4452         dec_rs_pending(mdev);
4453
4454         if (get_ldev_if_state(mdev, D_FAILED)) {
4455                 drbd_rs_complete_io(mdev, sector);
4456                 drbd_rs_failed_io(mdev, sector, size);
4457                 put_ldev(mdev);
4458         }
4459
4460         return TRUE;
4461 }
4462
4463 static int got_BarrierAck(struct drbd_conf *mdev, struct p_header80 *h)
4464 {
4465         struct p_barrier_ack *p = (struct p_barrier_ack *)h;
4466
4467         tl_release(mdev, p->barrier, be32_to_cpu(p->set_size));
4468
4469         return TRUE;
4470 }
4471
4472 static int got_OVResult(struct drbd_conf *mdev, struct p_header80 *h)
4473 {
4474         struct p_block_ack *p = (struct p_block_ack *)h;
4475         struct drbd_work *w;
4476         sector_t sector;
4477         int size;
4478
4479         sector = be64_to_cpu(p->sector);
4480         size = be32_to_cpu(p->blksize);
4481
4482         update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4483
4484         if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC)
4485                 drbd_ov_oos_found(mdev, sector, size);
4486         else
4487                 ov_oos_print(mdev);
4488
4489         if (!get_ldev(mdev))
4490                 return TRUE;
4491
4492         drbd_rs_complete_io(mdev, sector);
4493         dec_rs_pending(mdev);
4494
4495         if (--mdev->ov_left == 0) {
4496                 w = kmalloc(sizeof(*w), GFP_NOIO);
4497                 if (w) {
4498                         w->cb = w_ov_finished;
4499                         drbd_queue_work_front(&mdev->data.work, w);
4500                 } else {
4501                         dev_err(DEV, "kmalloc(w) failed.");
4502                         ov_oos_print(mdev);
4503                         drbd_resync_finished(mdev);
4504                 }
4505         }
4506         put_ldev(mdev);
4507         return TRUE;
4508 }
4509
4510 static int got_skip(struct drbd_conf *mdev, struct p_header80 *h)
4511 {
4512         return TRUE;
4513 }
4514
4515 struct asender_cmd {
4516         size_t pkt_size;
4517         int (*process)(struct drbd_conf *mdev, struct p_header80 *h);
4518 };
4519
4520 static struct asender_cmd *get_asender_cmd(int cmd)
4521 {
4522         static struct asender_cmd asender_tbl[] = {
4523                 /* anything missing from this table is in
4524                  * the drbd_cmd_handler (drbd_default_handler) table,
4525                  * see the beginning of drbdd() */
4526         [P_PING]            = { sizeof(struct p_header80), got_Ping },
4527         [P_PING_ACK]        = { sizeof(struct p_header80), got_PingAck },
4528         [P_RECV_ACK]        = { sizeof(struct p_block_ack), got_BlockAck },
4529         [P_WRITE_ACK]       = { sizeof(struct p_block_ack), got_BlockAck },
4530         [P_RS_WRITE_ACK]    = { sizeof(struct p_block_ack), got_BlockAck },
4531         [P_DISCARD_ACK]     = { sizeof(struct p_block_ack), got_BlockAck },
4532         [P_NEG_ACK]         = { sizeof(struct p_block_ack), got_NegAck },
4533         [P_NEG_DREPLY]      = { sizeof(struct p_block_ack), got_NegDReply },
4534         [P_NEG_RS_DREPLY]   = { sizeof(struct p_block_ack), got_NegRSDReply},
4535         [P_OV_RESULT]       = { sizeof(struct p_block_ack), got_OVResult },
4536         [P_BARRIER_ACK]     = { sizeof(struct p_barrier_ack), got_BarrierAck },
4537         [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply },
4538         [P_RS_IS_IN_SYNC]   = { sizeof(struct p_block_ack), got_IsInSync },
4539         [P_DELAY_PROBE]     = { sizeof(struct p_delay_probe93), got_skip },
4540         [P_MAX_CMD]         = { 0, NULL },
4541         };
4542         if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL)
4543                 return NULL;
4544         return &asender_tbl[cmd];
4545 }
4546
4547 int drbd_asender(struct drbd_thread *thi)
4548 {
4549         struct drbd_conf *mdev = thi->mdev;
4550         struct p_header80 *h = &mdev->meta.rbuf.header.h80;
4551         struct asender_cmd *cmd = NULL;
4552
4553         int rv, len;
4554         void *buf    = h;
4555         int received = 0;
4556         int expect   = sizeof(struct p_header80);
4557         int empty;
4558
4559         sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev));
4560
4561         current->policy = SCHED_RR;  /* Make this a realtime task! */
4562         current->rt_priority = 2;    /* more important than all other tasks */
4563
4564         while (get_t_state(thi) == Running) {
4565                 drbd_thread_current_set_cpu(mdev);
4566                 if (test_and_clear_bit(SEND_PING, &mdev->flags)) {
4567                         ERR_IF(!drbd_send_ping(mdev)) goto reconnect;
4568                         mdev->meta.socket->sk->sk_rcvtimeo =
4569                                 mdev->net_conf->ping_timeo*HZ/10;
4570                 }
4571
4572                 /* conditionally cork;
4573                  * it may hurt latency if we cork without much to send */
4574                 if (!mdev->net_conf->no_cork &&
4575                         3 < atomic_read(&mdev->unacked_cnt))
4576                         drbd_tcp_cork(mdev->meta.socket);
4577                 while (1) {
4578                         clear_bit(SIGNAL_ASENDER, &mdev->flags);
4579                         flush_signals(current);
4580                         if (!drbd_process_done_ee(mdev)) {
4581                                 dev_err(DEV, "process_done_ee() = NOT_OK\n");
4582                                 goto reconnect;
4583                         }
4584                         /* to avoid race with newly queued ACKs */
4585                         set_bit(SIGNAL_ASENDER, &mdev->flags);
4586                         spin_lock_irq(&mdev->req_lock);
4587                         empty = list_empty(&mdev->done_ee);
4588                         spin_unlock_irq(&mdev->req_lock);
4589                         /* new ack may have been queued right here,
4590                          * but then there is also a signal pending,
4591                          * and we start over... */
4592                         if (empty)
4593                                 break;
4594                 }
4595                 /* but unconditionally uncork unless disabled */
4596                 if (!mdev->net_conf->no_cork)
4597                         drbd_tcp_uncork(mdev->meta.socket);
4598
4599                 /* short circuit, recv_msg would return EINTR anyways. */
4600                 if (signal_pending(current))
4601                         continue;
4602
4603                 rv = drbd_recv_short(mdev, mdev->meta.socket,
4604                                      buf, expect-received, 0);
4605                 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4606
4607                 flush_signals(current);
4608
4609                 /* Note:
4610                  * -EINTR        (on meta) we got a signal
4611                  * -EAGAIN       (on meta) rcvtimeo expired
4612                  * -ECONNRESET   other side closed the connection
4613                  * -ERESTARTSYS  (on data) we got a signal
4614                  * rv <  0       other than above: unexpected error!
4615                  * rv == expected: full header or command
4616                  * rv <  expected: "woken" by signal during receive
4617                  * rv == 0       : "connection shut down by peer"
4618                  */
4619                 if (likely(rv > 0)) {
4620                         received += rv;
4621                         buf      += rv;
4622                 } else if (rv == 0) {
4623                         dev_err(DEV, "meta connection shut down by peer.\n");
4624                         goto reconnect;
4625                 } else if (rv == -EAGAIN) {
4626                         if (mdev->meta.socket->sk->sk_rcvtimeo ==
4627                             mdev->net_conf->ping_timeo*HZ/10) {
4628                                 dev_err(DEV, "PingAck did not arrive in time.\n");
4629                                 goto reconnect;
4630                         }
4631                         set_bit(SEND_PING, &mdev->flags);
4632                         continue;
4633                 } else if (rv == -EINTR) {
4634                         continue;
4635                 } else {
4636                         dev_err(DEV, "sock_recvmsg returned %d\n", rv);
4637                         goto reconnect;
4638                 }
4639
4640                 if (received == expect && cmd == NULL) {
4641                         if (unlikely(h->magic != BE_DRBD_MAGIC)) {
4642                                 dev_err(DEV, "magic?? on meta m: 0x%08x c: %d l: %d\n",
4643                                     be32_to_cpu(h->magic),
4644                                     be16_to_cpu(h->command),
4645                                     be16_to_cpu(h->length));
4646                                 goto reconnect;
4647                         }
4648                         cmd = get_asender_cmd(be16_to_cpu(h->command));
4649                         len = be16_to_cpu(h->length);
4650                         if (unlikely(cmd == NULL)) {
4651                                 dev_err(DEV, "unknown command?? on meta m: 0x%08x c: %d l: %d\n",
4652                                     be32_to_cpu(h->magic),
4653                                     be16_to_cpu(h->command),
4654                                     be16_to_cpu(h->length));
4655                                 goto disconnect;
4656                         }
4657                         expect = cmd->pkt_size;
4658                         ERR_IF(len != expect-sizeof(struct p_header80))
4659                                 goto reconnect;
4660                 }
4661                 if (received == expect) {
4662                         D_ASSERT(cmd != NULL);
4663                         if (!cmd->process(mdev, h))
4664                                 goto reconnect;
4665
4666                         buf      = h;
4667                         received = 0;
4668                         expect   = sizeof(struct p_header80);
4669                         cmd      = NULL;
4670                 }
4671         }
4672
4673         if (0) {
4674 reconnect:
4675                 drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE));
4676         }
4677         if (0) {
4678 disconnect:
4679                 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4680         }
4681         clear_bit(SIGNAL_ASENDER, &mdev->flags);
4682
4683         D_ASSERT(mdev->state.conn < C_CONNECTED);
4684         dev_info(DEV, "asender terminated\n");
4685
4686         return 0;
4687 }