drivers/block/drbd/drbd_state.c

   1 /*
   2    drbd_state.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    Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
  11    from Logicworks, Inc. for making SDP replication support possible.
  12
  13    drbd is free software; you can redistribute it and/or modify
  14    it under the terms of the GNU General Public License as published by
  15    the Free Software Foundation; either version 2, or (at your option)
  16    any later version.
  17
  18    drbd is distributed in the hope that it will be useful,
  19    but WITHOUT ANY WARRANTY; without even the implied warranty of
  20    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  21    GNU General Public License for more details.
  22
  23    You should have received a copy of the GNU General Public License
  24    along with drbd; see the file COPYING.  If not, write to
  25    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  26  */
  27
  28 #include <linux/drbd_limits.h>
  29 #include "drbd_int.h"
  30 #include "drbd_req.h"
  31
  32 struct after_state_chg_work {
  33         struct drbd_work w;
  34         union drbd_state os;
  35         union drbd_state ns;
  36         enum chg_state_flags flags;
  37         struct completion *done;
  38 };
  39
  40
  41 extern void _tl_restart(struct drbd_conf *mdev, enum drbd_req_event what);
  42 int drbd_send_state_req(struct drbd_conf *, union drbd_state, union drbd_state);
  43 static int w_after_state_ch(struct drbd_work *w, int unused);
  44 static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
  45                            union drbd_state ns, enum chg_state_flags flags);
  46 static void after_conn_state_ch(struct drbd_tconn *tconn, union drbd_state os,
  47                                 union drbd_state ns, enum chg_state_flags flags);
  48 static enum drbd_state_rv is_valid_state(struct drbd_conf *, union drbd_state);
  49 static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state);
  50 static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns);
  51 static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state os,
  52                                        union drbd_state ns, const char **warn_sync_abort);
  53
  54 /**
  55  * cl_wide_st_chg() - true if the state change is a cluster wide one
  56  * @mdev:       DRBD device.
  57  * @os:         old (current) state.
  58  * @ns:         new (wanted) state.
  59  */
  60 static int cl_wide_st_chg(struct drbd_conf *mdev,
  61                           union drbd_state os, union drbd_state ns)
  62 {
  63         return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED &&
  64                  ((os.role != R_PRIMARY && ns.role == R_PRIMARY) ||
  65                   (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
  66                   (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) ||
  67                   (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))) ||
  68                 (os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) ||
  69                 (os.conn == C_CONNECTED && ns.conn == C_VERIFY_S);
  70 }
  71
  72 enum drbd_state_rv
  73 drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f,
  74                   union drbd_state mask, union drbd_state val)
  75 {
  76         unsigned long flags;
  77         union drbd_state os, ns;
  78         enum drbd_state_rv rv;
  79
  80         spin_lock_irqsave(&mdev->tconn->req_lock, flags);
  81         os = mdev->state;
  82         ns.i = (os.i & ~mask.i) | val.i;
  83         rv = _drbd_set_state(mdev, ns, f, NULL);
  84         ns = mdev->state;
  85         spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
  86
  87         return rv;
  88 }
  89
  90 /**
  91  * drbd_force_state() - Impose a change which happens outside our control on our state
  92  * @mdev:       DRBD device.
  93  * @mask:       mask of state bits to change.
  94  * @val:        value of new state bits.
  95  */
  96 void drbd_force_state(struct drbd_conf *mdev,
  97         union drbd_state mask, union drbd_state val)
  98 {
  99         drbd_change_state(mdev, CS_HARD, mask, val);
 100 }
 101
 102 static enum drbd_state_rv
 103 _req_st_cond(struct drbd_conf *mdev, union drbd_state mask,
 104              union drbd_state val)
 105 {
 106         union drbd_state os, ns;
 107         unsigned long flags;
 108         enum drbd_state_rv rv;
 109
 110         if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &mdev->flags))
 111                 return SS_CW_SUCCESS;
 112
 113         if (test_and_clear_bit(CL_ST_CHG_FAIL, &mdev->flags))
 114                 return SS_CW_FAILED_BY_PEER;
 115
 116         spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 117         os = mdev->state;
 118         ns.i = (os.i & ~mask.i) | val.i;
 119         ns = sanitize_state(mdev, os, ns, NULL);
 120         rv = is_valid_transition(os, ns);
 121         if (rv == SS_SUCCESS)
 122                 rv = SS_UNKNOWN_ERROR;  /* cont waiting, otherwise fail. */
 123
 124         if (!cl_wide_st_chg(mdev, os, ns))
 125                 rv = SS_CW_NO_NEED;
 126         if (rv == SS_UNKNOWN_ERROR) {
 127                 rv = is_valid_state(mdev, ns);
 128                 if (rv == SS_SUCCESS) {
 129                         rv = is_valid_soft_transition(os, ns);
 130                         if (rv == SS_SUCCESS)
 131                                 rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
 132                 }
 133         }
 134         spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 135
 136         return rv;
 137 }
 138
 139 /**
 140  * drbd_req_state() - Perform an eventually cluster wide state change
 141  * @mdev:       DRBD device.
 142  * @mask:       mask of state bits to change.
 143  * @val:        value of new state bits.
 144  * @f:          flags
 145  *
 146  * Should not be called directly, use drbd_request_state() or
 147  * _drbd_request_state().
 148  */
 149 static enum drbd_state_rv
 150 drbd_req_state(struct drbd_conf *mdev, union drbd_state mask,
 151                union drbd_state val, enum chg_state_flags f)
 152 {
 153         struct completion done;
 154         unsigned long flags;
 155         union drbd_state os, ns;
 156         enum drbd_state_rv rv;
 157
 158         init_completion(&done);
 159
 160         if (f & CS_SERIALIZE)
 161                 mutex_lock(&mdev->state_mutex);
 162
 163         spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 164         os = mdev->state;
 165         ns.i = (os.i & ~mask.i) | val.i;
 166         ns = sanitize_state(mdev, os, ns, NULL);
 167         rv = is_valid_transition(os, ns);
 168         if (rv < SS_SUCCESS)
 169                 goto abort;
 170
 171         if (cl_wide_st_chg(mdev, os, ns)) {
 172                 rv = is_valid_state(mdev, ns);
 173                 if (rv == SS_SUCCESS)
 174                         rv = is_valid_soft_transition(os, ns);
 175                 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 176
 177                 if (rv < SS_SUCCESS) {
 178                         if (f & CS_VERBOSE)
 179                                 print_st_err(mdev, os, ns, rv);
 180                         goto abort;
 181                 }
 182
 183                 drbd_state_lock(mdev);
 184                 if (!drbd_send_state_req(mdev, mask, val)) {
 185                         drbd_state_unlock(mdev);
 186                         rv = SS_CW_FAILED_BY_PEER;
 187                         if (f & CS_VERBOSE)
 188                                 print_st_err(mdev, os, ns, rv);
 189                         goto abort;
 190                 }
 191
 192                 wait_event(mdev->state_wait,
 193                         (rv = _req_st_cond(mdev, mask, val)));
 194
 195                 if (rv < SS_SUCCESS) {
 196                         drbd_state_unlock(mdev);
 197                         if (f & CS_VERBOSE)
 198                                 print_st_err(mdev, os, ns, rv);
 199                         goto abort;
 200                 }
 201                 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 202                 os = mdev->state;
 203                 ns.i = (os.i & ~mask.i) | val.i;
 204                 rv = _drbd_set_state(mdev, ns, f, &done);
 205                 drbd_state_unlock(mdev);
 206         } else {
 207                 rv = _drbd_set_state(mdev, ns, f, &done);
 208         }
 209
 210         spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 211
 212         if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) {
 213                 D_ASSERT(current != mdev->tconn->worker.task);
 214                 wait_for_completion(&done);
 215         }
 216
 217 abort:
 218         if (f & CS_SERIALIZE)
 219                 mutex_unlock(&mdev->state_mutex);
 220
 221         return rv;
 222 }
 223
 224 /**
 225  * _drbd_request_state() - Request a state change (with flags)
 226  * @mdev:       DRBD device.
 227  * @mask:       mask of state bits to change.
 228  * @val:        value of new state bits.
 229  * @f:          flags
 230  *
 231  * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE
 232  * flag, or when logging of failed state change requests is not desired.
 233  */
 234 enum drbd_state_rv
 235 _drbd_request_state(struct drbd_conf *mdev, union drbd_state mask,
 236                     union drbd_state val, enum chg_state_flags f)
 237 {
 238         enum drbd_state_rv rv;
 239
 240         wait_event(mdev->state_wait,
 241                    (rv = drbd_req_state(mdev, mask, val, f)) != SS_IN_TRANSIENT_STATE);
 242
 243         return rv;
 244 }
 245
 246 static void print_st(struct drbd_conf *mdev, char *name, union drbd_state ns)
 247 {
 248         dev_err(DEV, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n",
 249             name,
 250             drbd_conn_str(ns.conn),
 251             drbd_role_str(ns.role),
 252             drbd_role_str(ns.peer),
 253             drbd_disk_str(ns.disk),
 254             drbd_disk_str(ns.pdsk),
 255             is_susp(ns) ? 's' : 'r',
 256             ns.aftr_isp ? 'a' : '-',
 257             ns.peer_isp ? 'p' : '-',
 258             ns.user_isp ? 'u' : '-',
 259             ns.susp_fen ? 'F' : '-',
 260             ns.susp_nod ? 'N' : '-'
 261             );
 262 }
 263
 264 void print_st_err(struct drbd_conf *mdev, union drbd_state os,
 265                   union drbd_state ns, enum drbd_state_rv err)
 266 {
 267         if (err == SS_IN_TRANSIENT_STATE)
 268                 return;
 269         dev_err(DEV, "State change failed: %s\n", drbd_set_st_err_str(err));
 270         print_st(mdev, " state", os);
 271         print_st(mdev, "wanted", ns);
 272 }
 273
 274
 275 /**
 276  * is_valid_state() - Returns an SS_ error code if ns is not valid
 277  * @mdev:       DRBD device.
 278  * @ns:         State to consider.
 279  */
 280 static enum drbd_state_rv
 281 is_valid_state(struct drbd_conf *mdev, union drbd_state ns)
 282 {
 283         /* See drbd_state_sw_errors in drbd_strings.c */
 284
 285         enum drbd_fencing_p fp;
 286         enum drbd_state_rv rv = SS_SUCCESS;
 287
 288         fp = FP_DONT_CARE;
 289         if (get_ldev(mdev)) {
 290                 fp = mdev->ldev->dc.fencing;
 291                 put_ldev(mdev);
 292         }
 293
 294         if (get_net_conf(mdev->tconn)) {
 295                 if (!mdev->tconn->net_conf->two_primaries &&
 296                     ns.role == R_PRIMARY && ns.peer == R_PRIMARY)
 297                         rv = SS_TWO_PRIMARIES;
 298                 put_net_conf(mdev->tconn);
 299         }
 300
 301         if (rv <= 0)
 302                 /* already found a reason to abort */;
 303         else if (ns.role == R_SECONDARY && mdev->open_cnt)
 304                 rv = SS_DEVICE_IN_USE;
 305
 306         else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE)
 307                 rv = SS_NO_UP_TO_DATE_DISK;
 308
 309         else if (fp >= FP_RESOURCE &&
 310                  ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN)
 311                 rv = SS_PRIMARY_NOP;
 312
 313         else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT)
 314                 rv = SS_NO_UP_TO_DATE_DISK;
 315
 316         else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT)
 317                 rv = SS_NO_LOCAL_DISK;
 318
 319         else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT)
 320                 rv = SS_NO_REMOTE_DISK;
 321
 322         else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
 323                 rv = SS_NO_UP_TO_DATE_DISK;
 324
 325         else if ((ns.conn == C_CONNECTED ||
 326                   ns.conn == C_WF_BITMAP_S ||
 327                   ns.conn == C_SYNC_SOURCE ||
 328                   ns.conn == C_PAUSED_SYNC_S) &&
 329                   ns.disk == D_OUTDATED)
 330                 rv = SS_CONNECTED_OUTDATES;
 331
 332         else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 333                  (mdev->sync_conf.verify_alg[0] == 0))
 334                 rv = SS_NO_VERIFY_ALG;
 335
 336         else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 337                   mdev->tconn->agreed_pro_version < 88)
 338                 rv = SS_NOT_SUPPORTED;
 339
 340         else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN)
 341                 rv = SS_CONNECTED_OUTDATES;
 342
 343         return rv;
 344 }
 345
 346 /**
 347  * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible
 348  * This function limits state transitions that may be declined by DRBD. I.e.
 349  * user requests (aka soft transitions).
 350  * @mdev:       DRBD device.
 351  * @ns:         new state.
 352  * @os:         old state.
 353  */
 354 static enum drbd_state_rv
 355 is_valid_soft_transition(union drbd_state os, union drbd_state ns)
 356 {
 357         enum drbd_state_rv rv = SS_SUCCESS;
 358
 359         if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
 360             os.conn > C_CONNECTED)
 361                 rv = SS_RESYNC_RUNNING;
 362
 363         if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
 364                 rv = SS_ALREADY_STANDALONE;
 365
 366         if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
 367                 rv = SS_IS_DISKLESS;
 368
 369         if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
 370                 rv = SS_NO_NET_CONFIG;
 371
 372         if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
 373                 rv = SS_LOWER_THAN_OUTDATED;
 374
 375         if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
 376                 rv = SS_IN_TRANSIENT_STATE;
 377
 378         if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS)
 379                 rv = SS_IN_TRANSIENT_STATE;
 380
 381         if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
 382                 rv = SS_NEED_CONNECTION;
 383
 384         if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 385             ns.conn != os.conn && os.conn > C_CONNECTED)
 386                 rv = SS_RESYNC_RUNNING;
 387
 388         if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
 389             os.conn < C_CONNECTED)
 390                 rv = SS_NEED_CONNECTION;
 391
 392         if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)
 393             && os.conn < C_WF_REPORT_PARAMS)
 394                 rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */
 395
 396         return rv;
 397 }
 398
 399 /**
 400  * is_valid_transition() - Returns an SS_ error code if the state transition is not possible
 401  * This limits hard state transitions. Hard state transitions are facts there are
 402  * imposed on DRBD by the environment. E.g. disk broke or network broke down.
 403  * But those hard state transitions are still not allowed to do everything.
 404  * @ns:         new state.
 405  * @os:         old state.
 406  */
 407 static enum drbd_state_rv
 408 is_valid_transition(union drbd_state os, union drbd_state ns)
 409 {
 410         enum drbd_state_rv rv = SS_SUCCESS;
 411
 412         /* Disallow Network errors to configure a device's network part */
 413         if ((ns.conn >= C_TIMEOUT && ns.conn <= C_TEAR_DOWN) &&
 414             os.conn <= C_DISCONNECTING)
 415                 rv = SS_NEED_CONNECTION;
 416
 417         /* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */
 418         if (os.conn >= C_TIMEOUT && os.conn <= C_TEAR_DOWN &&
 419             ns.conn != C_UNCONNECTED && ns.conn != C_DISCONNECTING)
 420                 rv = SS_IN_TRANSIENT_STATE;
 421
 422         /* After C_DISCONNECTING only C_STANDALONE may follow */
 423         if (os.conn == C_DISCONNECTING && ns.conn != C_STANDALONE)
 424                 rv = SS_IN_TRANSIENT_STATE;
 425
 426         /* we cannot fail (again) if we already detached */
 427         if (ns.disk == D_FAILED && os.disk == D_DISKLESS)
 428                 rv = SS_IS_DISKLESS;
 429
 430         return rv;
 431 }
 432
 433 /**
 434  * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
 435  * @mdev:       DRBD device.
 436  * @os:         old state.
 437  * @ns:         new state.
 438  * @warn_sync_abort:
 439  *
 440  * When we loose connection, we have to set the state of the peers disk (pdsk)
 441  * to D_UNKNOWN. This rule and many more along those lines are in this function.
 442  */
 443 static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state os,
 444                                        union drbd_state ns, const char **warn_sync_abort)
 445 {
 446         enum drbd_fencing_p fp;
 447         enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max;
 448
 449         fp = FP_DONT_CARE;
 450         if (get_ldev(mdev)) {
 451                 fp = mdev->ldev->dc.fencing;
 452                 put_ldev(mdev);
 453         }
 454
 455         /* if we are only D_ATTACHING yet,
 456          * we can (and should) go directly to D_DISKLESS. */
 457         if (ns.disk == D_FAILED && os.disk == D_ATTACHING)
 458                 ns.disk = D_DISKLESS;
 459
 460         /* Implications from connection to peer and peer_isp */
 461         if (ns.conn < C_CONNECTED) {
 462                 ns.peer_isp = 0;
 463                 ns.peer = R_UNKNOWN;
 464                 if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
 465                         ns.pdsk = D_UNKNOWN;
 466         }
 467
 468         /* Clear the aftr_isp when becoming unconfigured */
 469         if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
 470                 ns.aftr_isp = 0;
 471
 472         /* Abort resync if a disk fails/detaches */
 473         if (os.conn > C_CONNECTED && ns.conn > C_CONNECTED &&
 474             (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
 475                 if (warn_sync_abort)
 476                         *warn_sync_abort =
 477                                 os.conn == C_VERIFY_S || os.conn == C_VERIFY_T ?
 478                                 "Online-verify" : "Resync";
 479                 ns.conn = C_CONNECTED;
 480         }
 481
 482         /* Connection breaks down before we finished "Negotiating" */
 483         if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
 484             get_ldev_if_state(mdev, D_NEGOTIATING)) {
 485                 if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) {
 486                         ns.disk = mdev->new_state_tmp.disk;
 487                         ns.pdsk = mdev->new_state_tmp.pdsk;
 488                 } else {
 489                         dev_alert(DEV, "Connection lost while negotiating, no data!\n");
 490                         ns.disk = D_DISKLESS;
 491                         ns.pdsk = D_UNKNOWN;
 492                 }
 493                 put_ldev(mdev);
 494         }
 495
 496         /* D_CONSISTENT and D_OUTDATED vanish when we get connected */
 497         if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) {
 498                 if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED)
 499                         ns.disk = D_UP_TO_DATE;
 500                 if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)
 501                         ns.pdsk = D_UP_TO_DATE;
 502         }
 503
 504         /* Implications of the connection stat on the disk states */
 505         disk_min = D_DISKLESS;
 506         disk_max = D_UP_TO_DATE;
 507         pdsk_min = D_INCONSISTENT;
 508         pdsk_max = D_UNKNOWN;
 509         switch ((enum drbd_conns)ns.conn) {
 510         case C_WF_BITMAP_T:
 511         case C_PAUSED_SYNC_T:
 512         case C_STARTING_SYNC_T:
 513         case C_WF_SYNC_UUID:
 514         case C_BEHIND:
 515                 disk_min = D_INCONSISTENT;
 516                 disk_max = D_OUTDATED;
 517                 pdsk_min = D_UP_TO_DATE;
 518                 pdsk_max = D_UP_TO_DATE;
 519                 break;
 520         case C_VERIFY_S:
 521         case C_VERIFY_T:
 522                 disk_min = D_UP_TO_DATE;
 523                 disk_max = D_UP_TO_DATE;
 524                 pdsk_min = D_UP_TO_DATE;
 525                 pdsk_max = D_UP_TO_DATE;
 526                 break;
 527         case C_CONNECTED:
 528                 disk_min = D_DISKLESS;
 529                 disk_max = D_UP_TO_DATE;
 530                 pdsk_min = D_DISKLESS;
 531                 pdsk_max = D_UP_TO_DATE;
 532                 break;
 533         case C_WF_BITMAP_S:
 534         case C_PAUSED_SYNC_S:
 535         case C_STARTING_SYNC_S:
 536         case C_AHEAD:
 537                 disk_min = D_UP_TO_DATE;
 538                 disk_max = D_UP_TO_DATE;
 539                 pdsk_min = D_INCONSISTENT;
 540                 pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/
 541                 break;
 542         case C_SYNC_TARGET:
 543                 disk_min = D_INCONSISTENT;
 544                 disk_max = D_INCONSISTENT;
 545                 pdsk_min = D_UP_TO_DATE;
 546                 pdsk_max = D_UP_TO_DATE;
 547                 break;
 548         case C_SYNC_SOURCE:
 549                 disk_min = D_UP_TO_DATE;
 550                 disk_max = D_UP_TO_DATE;
 551                 pdsk_min = D_INCONSISTENT;
 552                 pdsk_max = D_INCONSISTENT;
 553                 break;
 554         case C_STANDALONE:
 555         case C_DISCONNECTING:
 556         case C_UNCONNECTED:
 557         case C_TIMEOUT:
 558         case C_BROKEN_PIPE:
 559         case C_NETWORK_FAILURE:
 560         case C_PROTOCOL_ERROR:
 561         case C_TEAR_DOWN:
 562         case C_WF_CONNECTION:
 563         case C_WF_REPORT_PARAMS:
 564         case C_MASK:
 565                 break;
 566         }
 567         if (ns.disk > disk_max)
 568                 ns.disk = disk_max;
 569
 570         if (ns.disk < disk_min) {
 571                 dev_warn(DEV, "Implicitly set disk from %s to %s\n",
 572                          drbd_disk_str(ns.disk), drbd_disk_str(disk_min));
 573                 ns.disk = disk_min;
 574         }
 575         if (ns.pdsk > pdsk_max)
 576                 ns.pdsk = pdsk_max;
 577
 578         if (ns.pdsk < pdsk_min) {
 579                 dev_warn(DEV, "Implicitly set pdsk from %s to %s\n",
 580                          drbd_disk_str(ns.pdsk), drbd_disk_str(pdsk_min));
 581                 ns.pdsk = pdsk_min;
 582         }
 583
 584         if (fp == FP_STONITH &&
 585             (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED) &&
 586             !(os.role == R_PRIMARY && os.conn < C_CONNECTED && os.pdsk > D_OUTDATED))
 587                 ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */
 588
 589         if (mdev->sync_conf.on_no_data == OND_SUSPEND_IO &&
 590             (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE) &&
 591             !(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE))
 592                 ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */
 593
 594         if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
 595                 if (ns.conn == C_SYNC_SOURCE)
 596                         ns.conn = C_PAUSED_SYNC_S;
 597                 if (ns.conn == C_SYNC_TARGET)
 598                         ns.conn = C_PAUSED_SYNC_T;
 599         } else {
 600                 if (ns.conn == C_PAUSED_SYNC_S)
 601                         ns.conn = C_SYNC_SOURCE;
 602                 if (ns.conn == C_PAUSED_SYNC_T)
 603                         ns.conn = C_SYNC_TARGET;
 604         }
 605
 606         return ns;
 607 }
 608
 609 void drbd_resume_al(struct drbd_conf *mdev)
 610 {
 611         if (test_and_clear_bit(AL_SUSPENDED, &mdev->flags))
 612                 dev_info(DEV, "Resumed AL updates\n");
 613 }
 614
 615 /* helper for __drbd_set_state */
 616 static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs)
 617 {
 618         if (mdev->tconn->agreed_pro_version < 90)
 619                 mdev->ov_start_sector = 0;
 620         mdev->rs_total = drbd_bm_bits(mdev);
 621         mdev->ov_position = 0;
 622         if (cs == C_VERIFY_T) {
 623                 /* starting online verify from an arbitrary position
 624                  * does not fit well into the existing protocol.
 625                  * on C_VERIFY_T, we initialize ov_left and friends
 626                  * implicitly in receive_DataRequest once the
 627                  * first P_OV_REQUEST is received */
 628                 mdev->ov_start_sector = ~(sector_t)0;
 629         } else {
 630                 unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector);
 631                 if (bit >= mdev->rs_total) {
 632                         mdev->ov_start_sector =
 633                                 BM_BIT_TO_SECT(mdev->rs_total - 1);
 634                         mdev->rs_total = 1;
 635                 } else
 636                         mdev->rs_total -= bit;
 637                 mdev->ov_position = mdev->ov_start_sector;
 638         }
 639         mdev->ov_left = mdev->rs_total;
 640 }
 641
 642 /**
 643  * __drbd_set_state() - Set a new DRBD state
 644  * @mdev:       DRBD device.
 645  * @ns:         new state.
 646  * @flags:      Flags
 647  * @done:       Optional completion, that will get completed after the after_state_ch() finished
 648  *
 649  * Caller needs to hold req_lock, and global_state_lock. Do not call directly.
 650  */
 651 enum drbd_state_rv
 652 __drbd_set_state(struct drbd_conf *mdev, union drbd_state ns,
 653                  enum chg_state_flags flags, struct completion *done)
 654 {
 655         union drbd_state os;
 656         enum drbd_state_rv rv = SS_SUCCESS;
 657         const char *warn_sync_abort = NULL;
 658         struct after_state_chg_work *ascw;
 659
 660         os = mdev->state;
 661
 662         ns = sanitize_state(mdev, os, ns, &warn_sync_abort);
 663
 664         if (ns.i == os.i)
 665                 return SS_NOTHING_TO_DO;
 666
 667         rv = is_valid_transition(os, ns);
 668         if (rv < SS_SUCCESS)
 669                 return rv;
 670
 671         if (!(flags & CS_HARD)) {
 672                 /*  pre-state-change checks ; only look at ns  */
 673                 /* See drbd_state_sw_errors in drbd_strings.c */
 674
 675                 rv = is_valid_state(mdev, ns);
 676                 if (rv < SS_SUCCESS) {
 677                         /* If the old state was illegal as well, then let
 678                            this happen...*/
 679
 680                         if (is_valid_state(mdev, os) == rv)
 681                                 rv = is_valid_soft_transition(os, ns);
 682                 } else
 683                         rv = is_valid_soft_transition(os, ns);
 684         }
 685
 686         if (rv < SS_SUCCESS) {
 687                 if (flags & CS_VERBOSE)
 688                         print_st_err(mdev, os, ns, rv);
 689                 return rv;
 690         }
 691
 692         if (warn_sync_abort)
 693                 dev_warn(DEV, "%s aborted.\n", warn_sync_abort);
 694
 695         {
 696         char *pbp, pb[300];
 697         pbp = pb;
 698         *pbp = 0;
 699         if (ns.role != os.role)
 700                 pbp += sprintf(pbp, "role( %s -> %s ) ",
 701                                drbd_role_str(os.role),
 702                                drbd_role_str(ns.role));
 703         if (ns.peer != os.peer)
 704                 pbp += sprintf(pbp, "peer( %s -> %s ) ",
 705                                drbd_role_str(os.peer),
 706                                drbd_role_str(ns.peer));
 707         if (ns.conn != os.conn)
 708                 pbp += sprintf(pbp, "conn( %s -> %s ) ",
 709                                drbd_conn_str(os.conn),
 710                                drbd_conn_str(ns.conn));
 711         if (ns.disk != os.disk)
 712                 pbp += sprintf(pbp, "disk( %s -> %s ) ",
 713                                drbd_disk_str(os.disk),
 714                                drbd_disk_str(ns.disk));
 715         if (ns.pdsk != os.pdsk)
 716                 pbp += sprintf(pbp, "pdsk( %s -> %s ) ",
 717                                drbd_disk_str(os.pdsk),
 718                                drbd_disk_str(ns.pdsk));
 719         if (is_susp(ns) != is_susp(os))
 720                 pbp += sprintf(pbp, "susp( %d -> %d ) ",
 721                                is_susp(os),
 722                                is_susp(ns));
 723         if (ns.aftr_isp != os.aftr_isp)
 724                 pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ",
 725                                os.aftr_isp,
 726                                ns.aftr_isp);
 727         if (ns.peer_isp != os.peer_isp)
 728                 pbp += sprintf(pbp, "peer_isp( %d -> %d ) ",
 729                                os.peer_isp,
 730                                ns.peer_isp);
 731         if (ns.user_isp != os.user_isp)
 732                 pbp += sprintf(pbp, "user_isp( %d -> %d ) ",
 733                                os.user_isp,
 734                                ns.user_isp);
 735         dev_info(DEV, "%s\n", pb);
 736         }
 737
 738         /* solve the race between becoming unconfigured,
 739          * worker doing the cleanup, and
 740          * admin reconfiguring us:
 741          * on (re)configure, first set CONFIG_PENDING,
 742          * then wait for a potentially exiting worker,
 743          * start the worker, and schedule one no_op.
 744          * then proceed with configuration.
 745          */
 746         if (ns.disk == D_DISKLESS &&
 747             ns.conn == C_STANDALONE &&
 748             ns.role == R_SECONDARY &&
 749             !test_and_set_bit(CONFIG_PENDING, &mdev->flags))
 750                 set_bit(DEVICE_DYING, &mdev->flags);
 751
 752         /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
 753          * on the ldev here, to be sure the transition -> D_DISKLESS resp.
 754          * drbd_ldev_destroy() won't happen before our corresponding
 755          * after_state_ch works run, where we put_ldev again. */
 756         if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
 757             (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
 758                 atomic_inc(&mdev->local_cnt);
 759
 760         mdev->state = ns;
 761
 762         if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
 763                 drbd_print_uuids(mdev, "attached to UUIDs");
 764
 765         wake_up(&mdev->misc_wait);
 766         wake_up(&mdev->state_wait);
 767         wake_up(&mdev->tconn->ping_wait);
 768
 769         /* aborted verify run. log the last position */
 770         if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
 771             ns.conn < C_CONNECTED) {
 772                 mdev->ov_start_sector =
 773                         BM_BIT_TO_SECT(drbd_bm_bits(mdev) - mdev->ov_left);
 774                 dev_info(DEV, "Online Verify reached sector %llu\n",
 775                         (unsigned long long)mdev->ov_start_sector);
 776         }
 777
 778         if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
 779             (ns.conn == C_SYNC_TARGET  || ns.conn == C_SYNC_SOURCE)) {
 780                 dev_info(DEV, "Syncer continues.\n");
 781                 mdev->rs_paused += (long)jiffies
 782                                   -(long)mdev->rs_mark_time[mdev->rs_last_mark];
 783                 if (ns.conn == C_SYNC_TARGET)
 784                         mod_timer(&mdev->resync_timer, jiffies);
 785         }
 786
 787         if ((os.conn == C_SYNC_TARGET  || os.conn == C_SYNC_SOURCE) &&
 788             (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
 789                 dev_info(DEV, "Resync suspended\n");
 790                 mdev->rs_mark_time[mdev->rs_last_mark] = jiffies;
 791         }
 792
 793         if (os.conn == C_CONNECTED &&
 794             (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
 795                 unsigned long now = jiffies;
 796                 int i;
 797
 798                 set_ov_position(mdev, ns.conn);
 799                 mdev->rs_start = now;
 800                 mdev->rs_last_events = 0;
 801                 mdev->rs_last_sect_ev = 0;
 802                 mdev->ov_last_oos_size = 0;
 803                 mdev->ov_last_oos_start = 0;
 804
 805                 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
 806                         mdev->rs_mark_left[i] = mdev->ov_left;
 807                         mdev->rs_mark_time[i] = now;
 808                 }
 809
 810                 drbd_rs_controller_reset(mdev);
 811
 812                 if (ns.conn == C_VERIFY_S) {
 813                         dev_info(DEV, "Starting Online Verify from sector %llu\n",
 814                                         (unsigned long long)mdev->ov_position);
 815                         mod_timer(&mdev->resync_timer, jiffies);
 816                 }
 817         }
 818
 819         if (get_ldev(mdev)) {
 820                 u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
 821                                                  MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
 822                                                  MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
 823
 824                 if (test_bit(CRASHED_PRIMARY, &mdev->flags))
 825                         mdf |= MDF_CRASHED_PRIMARY;
 826                 if (mdev->state.role == R_PRIMARY ||
 827                     (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY))
 828                         mdf |= MDF_PRIMARY_IND;
 829                 if (mdev->state.conn > C_WF_REPORT_PARAMS)
 830                         mdf |= MDF_CONNECTED_IND;
 831                 if (mdev->state.disk > D_INCONSISTENT)
 832                         mdf |= MDF_CONSISTENT;
 833                 if (mdev->state.disk > D_OUTDATED)
 834                         mdf |= MDF_WAS_UP_TO_DATE;
 835                 if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT)
 836                         mdf |= MDF_PEER_OUT_DATED;
 837                 if (mdf != mdev->ldev->md.flags) {
 838                         mdev->ldev->md.flags = mdf;
 839                         drbd_md_mark_dirty(mdev);
 840                 }
 841                 if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
 842                         drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]);
 843                 put_ldev(mdev);
 844         }
 845
 846         /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
 847         if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
 848             os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
 849                 set_bit(CONSIDER_RESYNC, &mdev->flags);
 850
 851         /* Receiver should clean up itself */
 852         if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
 853                 drbd_thread_stop_nowait(&mdev->tconn->receiver);
 854
 855         /* Now the receiver finished cleaning up itself, it should die */
 856         if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
 857                 drbd_thread_stop_nowait(&mdev->tconn->receiver);
 858
 859         /* Upon network failure, we need to restart the receiver. */
 860         if (os.conn > C_TEAR_DOWN &&
 861             ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
 862                 drbd_thread_restart_nowait(&mdev->tconn->receiver);
 863
 864         /* Resume AL writing if we get a connection */
 865         if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
 866                 drbd_resume_al(mdev);
 867
 868         ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
 869         if (ascw) {
 870                 ascw->os = os;
 871                 ascw->ns = ns;
 872                 ascw->flags = flags;
 873                 ascw->w.cb = w_after_state_ch;
 874                 ascw->w.mdev = mdev;
 875                 ascw->done = done;
 876                 drbd_queue_work(&mdev->tconn->data.work, &ascw->w);
 877         } else {
 878                 dev_warn(DEV, "Could not kmalloc an ascw\n");
 879         }
 880
 881         return rv;
 882 }
 883
 884 static int w_after_state_ch(struct drbd_work *w, int unused)
 885 {
 886         struct after_state_chg_work *ascw =
 887                 container_of(w, struct after_state_chg_work, w);
 888         struct drbd_conf *mdev = w->mdev;
 889
 890         after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags);
 891         if (ascw->flags & CS_WAIT_COMPLETE) {
 892                 D_ASSERT(ascw->done != NULL);
 893                 complete(ascw->done);
 894         }
 895         kfree(ascw);
 896
 897         return 1;
 898 }
 899
 900 static void abw_start_sync(struct drbd_conf *mdev, int rv)
 901 {
 902         if (rv) {
 903                 dev_err(DEV, "Writing the bitmap failed not starting resync.\n");
 904                 _drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE);
 905                 return;
 906         }
 907
 908         switch (mdev->state.conn) {
 909         case C_STARTING_SYNC_T:
 910                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
 911                 break;
 912         case C_STARTING_SYNC_S:
 913                 drbd_start_resync(mdev, C_SYNC_SOURCE);
 914                 break;
 915         }
 916 }
 917
 918 int drbd_bitmap_io_from_worker(struct drbd_conf *mdev,
 919                 int (*io_fn)(struct drbd_conf *),
 920                 char *why, enum bm_flag flags)
 921 {
 922         int rv;
 923
 924         D_ASSERT(current == mdev->tconn->worker.task);
 925
 926         /* open coded non-blocking drbd_suspend_io(mdev); */
 927         set_bit(SUSPEND_IO, &mdev->flags);
 928
 929         drbd_bm_lock(mdev, why, flags);
 930         rv = io_fn(mdev);
 931         drbd_bm_unlock(mdev);
 932
 933         drbd_resume_io(mdev);
 934
 935         return rv;
 936 }
 937
 938 /**
 939  * after_state_ch() - Perform after state change actions that may sleep
 940  * @mdev:       DRBD device.
 941  * @os:         old state.
 942  * @ns:         new state.
 943  * @flags:      Flags
 944  */
 945 static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
 946                            union drbd_state ns, enum chg_state_flags flags)
 947 {
 948         enum drbd_fencing_p fp;
 949         enum drbd_req_event what = NOTHING;
 950         union drbd_state nsm = (union drbd_state){ .i = -1 };
 951
 952         if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) {
 953                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
 954                 if (mdev->p_uuid)
 955                         mdev->p_uuid[UI_FLAGS] &= ~((u64)2);
 956         }
 957
 958         fp = FP_DONT_CARE;
 959         if (get_ldev(mdev)) {
 960                 fp = mdev->ldev->dc.fencing;
 961                 put_ldev(mdev);
 962         }
 963
 964         /* Inform userspace about the change... */
 965         drbd_bcast_state(mdev, ns);
 966
 967         if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
 968             (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
 969                 drbd_khelper(mdev, "pri-on-incon-degr");
 970
 971         /* Here we have the actions that are performed after a
 972            state change. This function might sleep */
 973
 974         nsm.i = -1;
 975         if (ns.susp_nod) {
 976                 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
 977                         what = RESEND;
 978
 979                 if (os.disk == D_ATTACHING && ns.disk > D_ATTACHING)
 980                         what = RESTART_FROZEN_DISK_IO;
 981
 982                 if (what != NOTHING)
 983                         nsm.susp_nod = 0;
 984         }
 985
 986         if (ns.susp_fen) {
 987                 /* case1: The outdate peer handler is successful: */
 988                 if (os.pdsk > D_OUTDATED  && ns.pdsk <= D_OUTDATED) {
 989                         tl_clear(mdev);
 990                         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
 991                                 drbd_uuid_new_current(mdev);
 992                                 clear_bit(NEW_CUR_UUID, &mdev->flags);
 993                         }
 994                         spin_lock_irq(&mdev->tconn->req_lock);
 995                         _drbd_set_state(_NS(mdev, susp_fen, 0), CS_VERBOSE, NULL);
 996                         spin_unlock_irq(&mdev->tconn->req_lock);
 997                 }
 998                 /* case2: The connection was established again: */
 999                 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) {
1000                         clear_bit(NEW_CUR_UUID, &mdev->flags);
1001                         what = RESEND;
1002                         nsm.susp_fen = 0;
1003                 }
1004         }
1005
1006         if (what != NOTHING) {
1007                 spin_lock_irq(&mdev->tconn->req_lock);
1008                 _tl_restart(mdev, what);
1009                 nsm.i &= mdev->state.i;
1010                 _drbd_set_state(mdev, nsm, CS_VERBOSE, NULL);
1011                 spin_unlock_irq(&mdev->tconn->req_lock);
1012         }
1013
1014         /* Became sync source.  With protocol >= 96, we still need to send out
1015          * the sync uuid now. Need to do that before any drbd_send_state, or
1016          * the other side may go "paused sync" before receiving the sync uuids,
1017          * which is unexpected. */
1018         if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
1019             (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
1020             mdev->tconn->agreed_pro_version >= 96 && get_ldev(mdev)) {
1021                 drbd_gen_and_send_sync_uuid(mdev);
1022                 put_ldev(mdev);
1023         }
1024
1025         /* Do not change the order of the if above and the two below... */
1026         if (os.pdsk == D_DISKLESS && ns.pdsk > D_DISKLESS) {      /* attach on the peer */
1027                 drbd_send_uuids(mdev);
1028                 drbd_send_state(mdev);
1029         }
1030         /* No point in queuing send_bitmap if we don't have a connection
1031          * anymore, so check also the _current_ state, not only the new state
1032          * at the time this work was queued. */
1033         if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
1034             mdev->state.conn == C_WF_BITMAP_S)
1035                 drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL,
1036                                 "send_bitmap (WFBitMapS)",
1037                                 BM_LOCKED_TEST_ALLOWED);
1038
1039         /* Lost contact to peer's copy of the data */
1040         if ((os.pdsk >= D_INCONSISTENT &&
1041              os.pdsk != D_UNKNOWN &&
1042              os.pdsk != D_OUTDATED)
1043         &&  (ns.pdsk < D_INCONSISTENT ||
1044              ns.pdsk == D_UNKNOWN ||
1045              ns.pdsk == D_OUTDATED)) {
1046                 if (get_ldev(mdev)) {
1047                         if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1048                             mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1049                                 if (is_susp(mdev->state)) {
1050                                         set_bit(NEW_CUR_UUID, &mdev->flags);
1051                                 } else {
1052                                         drbd_uuid_new_current(mdev);
1053                                         drbd_send_uuids(mdev);
1054                                 }
1055                         }
1056                         put_ldev(mdev);
1057                 }
1058         }
1059
1060         if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) {
1061                 if (ns.peer == R_PRIMARY && mdev->ldev->md.uuid[UI_BITMAP] == 0) {
1062                         drbd_uuid_new_current(mdev);
1063                         drbd_send_uuids(mdev);
1064                 }
1065
1066                 /* D_DISKLESS Peer becomes secondary */
1067                 if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1068                         /* We may still be Primary ourselves.
1069                          * No harm done if the bitmap still changes,
1070                          * redirtied pages will follow later. */
1071                         drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1072                                 "demote diskless peer", BM_LOCKED_SET_ALLOWED);
1073                 put_ldev(mdev);
1074         }
1075
1076         /* Write out all changed bits on demote.
1077          * Though, no need to da that just yet
1078          * if there is a resync going on still */
1079         if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
1080                 mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) {
1081                 /* No changes to the bitmap expected this time, so assert that,
1082                  * even though no harm was done if it did change. */
1083                 drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1084                                 "demote", BM_LOCKED_TEST_ALLOWED);
1085                 put_ldev(mdev);
1086         }
1087
1088         /* Last part of the attaching process ... */
1089         if (ns.conn >= C_CONNECTED &&
1090             os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1091                 drbd_send_sizes(mdev, 0, 0);  /* to start sync... */
1092                 drbd_send_uuids(mdev);
1093                 drbd_send_state(mdev);
1094         }
1095
1096         /* We want to pause/continue resync, tell peer. */
1097         if (ns.conn >= C_CONNECTED &&
1098              ((os.aftr_isp != ns.aftr_isp) ||
1099               (os.user_isp != ns.user_isp)))
1100                 drbd_send_state(mdev);
1101
1102         /* In case one of the isp bits got set, suspend other devices. */
1103         if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1104             (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1105                 suspend_other_sg(mdev);
1106
1107         /* Make sure the peer gets informed about eventual state
1108            changes (ISP bits) while we were in WFReportParams. */
1109         if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1110                 drbd_send_state(mdev);
1111
1112         if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
1113                 drbd_send_state(mdev);
1114
1115         /* We are in the progress to start a full sync... */
1116         if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1117             (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1118                 /* no other bitmap changes expected during this phase */
1119                 drbd_queue_bitmap_io(mdev,
1120                         &drbd_bmio_set_n_write, &abw_start_sync,
1121                         "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED);
1122
1123         /* We are invalidating our self... */
1124         if (os.conn < C_CONNECTED && ns.conn < C_CONNECTED &&
1125             os.disk > D_INCONSISTENT && ns.disk == D_INCONSISTENT)
1126                 /* other bitmap operation expected during this phase */
1127                 drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, NULL,
1128                         "set_n_write from invalidate", BM_LOCKED_MASK);
1129
1130         /* first half of local IO error, failure to attach,
1131          * or administrative detach */
1132         if (os.disk != D_FAILED && ns.disk == D_FAILED) {
1133                 enum drbd_io_error_p eh;
1134                 int was_io_error;
1135                 /* corresponding get_ldev was in __drbd_set_state, to serialize
1136                  * our cleanup here with the transition to D_DISKLESS,
1137                  * so it is safe to dreference ldev here. */
1138                 eh = mdev->ldev->dc.on_io_error;
1139                 was_io_error = test_and_clear_bit(WAS_IO_ERROR, &mdev->flags);
1140
1141                 /* current state still has to be D_FAILED,
1142                  * there is only one way out: to D_DISKLESS,
1143                  * and that may only happen after our put_ldev below. */
1144                 if (mdev->state.disk != D_FAILED)
1145                         dev_err(DEV,
1146                                 "ASSERT FAILED: disk is %s during detach\n",
1147                                 drbd_disk_str(mdev->state.disk));
1148
1149                 if (drbd_send_state(mdev))
1150                         dev_warn(DEV, "Notified peer that I am detaching my disk\n");
1151                 else
1152                         dev_err(DEV, "Sending state for detaching disk failed\n");
1153
1154                 drbd_rs_cancel_all(mdev);
1155
1156                 /* In case we want to get something to stable storage still,
1157                  * this may be the last chance.
1158                  * Following put_ldev may transition to D_DISKLESS. */
1159                 drbd_md_sync(mdev);
1160                 put_ldev(mdev);
1161
1162                 if (was_io_error && eh == EP_CALL_HELPER)
1163                         drbd_khelper(mdev, "local-io-error");
1164         }
1165
1166         /* second half of local IO error, failure to attach,
1167          * or administrative detach,
1168          * after local_cnt references have reached zero again */
1169         if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
1170                 /* We must still be diskless,
1171                  * re-attach has to be serialized with this! */
1172                 if (mdev->state.disk != D_DISKLESS)
1173                         dev_err(DEV,
1174                                 "ASSERT FAILED: disk is %s while going diskless\n",
1175                                 drbd_disk_str(mdev->state.disk));
1176
1177                 mdev->rs_total = 0;
1178                 mdev->rs_failed = 0;
1179                 atomic_set(&mdev->rs_pending_cnt, 0);
1180
1181                 if (drbd_send_state(mdev))
1182                         dev_warn(DEV, "Notified peer that I'm now diskless.\n");
1183                 /* corresponding get_ldev in __drbd_set_state
1184                  * this may finally trigger drbd_ldev_destroy. */
1185                 put_ldev(mdev);
1186         }
1187
1188         /* Notify peer that I had a local IO error, and did not detached.. */
1189         if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT)
1190                 drbd_send_state(mdev);
1191
1192         /* Disks got bigger while they were detached */
1193         if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1194             test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) {
1195                 if (ns.conn == C_CONNECTED)
1196                         resync_after_online_grow(mdev);
1197         }
1198
1199         /* A resync finished or aborted, wake paused devices... */
1200         if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1201             (os.peer_isp && !ns.peer_isp) ||
1202             (os.user_isp && !ns.user_isp))
1203                 resume_next_sg(mdev);
1204
1205         /* sync target done with resync.  Explicitly notify peer, even though
1206          * it should (at least for non-empty resyncs) already know itself. */
1207         if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
1208                 drbd_send_state(mdev);
1209
1210         /* This triggers bitmap writeout of potentially still unwritten pages
1211          * if the resync finished cleanly, or aborted because of peer disk
1212          * failure, or because of connection loss.
1213          * For resync aborted because of local disk failure, we cannot do
1214          * any bitmap writeout anymore.
1215          * No harm done if some bits change during this phase.
1216          */
1217         if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) {
1218                 drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL,
1219                         "write from resync_finished", BM_LOCKED_SET_ALLOWED);
1220                 put_ldev(mdev);
1221         }
1222
1223         if (ns.disk == D_DISKLESS &&
1224             ns.conn == C_STANDALONE &&
1225             ns.role == R_SECONDARY) {
1226                 if (os.aftr_isp != ns.aftr_isp)
1227                         resume_next_sg(mdev);
1228         }
1229
1230         after_conn_state_ch(mdev->tconn, os, ns, flags);
1231         drbd_md_sync(mdev);
1232 }
1233
1234 static void after_conn_state_ch(struct drbd_tconn *tconn, union drbd_state os,
1235                                 union drbd_state ns, enum chg_state_flags flags)
1236 {
1237         /* Upon network configuration, we need to start the receiver */
1238         if (os.conn == C_STANDALONE && ns.conn == C_UNCONNECTED)
1239                 drbd_thread_start(&tconn->receiver);
1240
1241         if (ns.disk == D_DISKLESS &&
1242             ns.conn == C_STANDALONE &&
1243             ns.role == R_SECONDARY) {
1244                 /* if (test_bit(DEVICE_DYING, &mdev->flags)) TODO: DEVICE_DYING functionality */
1245                 drbd_thread_stop_nowait(&tconn->worker);
1246         }
1247 }