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 ns,
  52                                        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 static union drbd_state
  73 apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val)
  74 {
  75         union drbd_state ns;
  76         ns.i = (os.i & ~mask.i) | val.i;
  77         return ns;
  78 }
  79
  80 enum drbd_state_rv
  81 drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f,
  82                   union drbd_state mask, union drbd_state val)
  83 {
  84         unsigned long flags;
  85         union drbd_state ns;
  86         enum drbd_state_rv rv;
  87
  88         spin_lock_irqsave(&mdev->tconn->req_lock, flags);
  89         ns = apply_mask_val(mdev->state, mask, val);
  90         rv = _drbd_set_state(mdev, ns, f, NULL);
  91         ns = mdev->state;
  92         spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
  93
  94         return rv;
  95 }
  96
  97 /**
  98  * drbd_force_state() - Impose a change which happens outside our control on our state
  99  * @mdev:       DRBD device.
 100  * @mask:       mask of state bits to change.
 101  * @val:        value of new state bits.
 102  */
 103 void drbd_force_state(struct drbd_conf *mdev,
 104         union drbd_state mask, union drbd_state val)
 105 {
 106         drbd_change_state(mdev, CS_HARD, mask, val);
 107 }
 108
 109 static enum drbd_state_rv
 110 _req_st_cond(struct drbd_conf *mdev, union drbd_state mask,
 111              union drbd_state val)
 112 {
 113         union drbd_state os, ns;
 114         unsigned long flags;
 115         enum drbd_state_rv rv;
 116
 117         if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &mdev->flags))
 118                 return SS_CW_SUCCESS;
 119
 120         if (test_and_clear_bit(CL_ST_CHG_FAIL, &mdev->flags))
 121                 return SS_CW_FAILED_BY_PEER;
 122
 123         spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 124         os = mdev->state;
 125         ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
 126         rv = is_valid_transition(os, ns);
 127         if (rv == SS_SUCCESS)
 128                 rv = SS_UNKNOWN_ERROR;  /* cont waiting, otherwise fail. */
 129
 130         if (!cl_wide_st_chg(mdev, os, ns))
 131                 rv = SS_CW_NO_NEED;
 132         if (rv == SS_UNKNOWN_ERROR) {
 133                 rv = is_valid_state(mdev, ns);
 134                 if (rv == SS_SUCCESS) {
 135                         rv = is_valid_soft_transition(os, ns);
 136                         if (rv == SS_SUCCESS)
 137                                 rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
 138                 }
 139         }
 140         spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 141
 142         return rv;
 143 }
 144
 145 /**
 146  * drbd_req_state() - Perform an eventually cluster wide state change
 147  * @mdev:       DRBD device.
 148  * @mask:       mask of state bits to change.
 149  * @val:        value of new state bits.
 150  * @f:          flags
 151  *
 152  * Should not be called directly, use drbd_request_state() or
 153  * _drbd_request_state().
 154  */
 155 static enum drbd_state_rv
 156 drbd_req_state(struct drbd_conf *mdev, union drbd_state mask,
 157                union drbd_state val, enum chg_state_flags f)
 158 {
 159         struct completion done;
 160         unsigned long flags;
 161         union drbd_state os, ns;
 162         enum drbd_state_rv rv;
 163
 164         init_completion(&done);
 165
 166         if (f & CS_SERIALIZE)
 167                 mutex_lock(&mdev->state_mutex);
 168
 169         spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 170         os = mdev->state;
 171         ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
 172         rv = is_valid_transition(os, ns);
 173         if (rv < SS_SUCCESS)
 174                 goto abort;
 175
 176         if (cl_wide_st_chg(mdev, os, ns)) {
 177                 rv = is_valid_state(mdev, ns);
 178                 if (rv == SS_SUCCESS)
 179                         rv = is_valid_soft_transition(os, ns);
 180                 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 181
 182                 if (rv < SS_SUCCESS) {
 183                         if (f & CS_VERBOSE)
 184                                 print_st_err(mdev, os, ns, rv);
 185                         goto abort;
 186                 }
 187
 188                 drbd_state_lock(mdev);
 189                 if (!drbd_send_state_req(mdev, mask, val)) {
 190                         drbd_state_unlock(mdev);
 191                         rv = SS_CW_FAILED_BY_PEER;
 192                         if (f & CS_VERBOSE)
 193                                 print_st_err(mdev, os, ns, rv);
 194                         goto abort;
 195                 }
 196
 197                 wait_event(mdev->state_wait,
 198                         (rv = _req_st_cond(mdev, mask, val)));
 199
 200                 if (rv < SS_SUCCESS) {
 201                         drbd_state_unlock(mdev);
 202                         if (f & CS_VERBOSE)
 203                                 print_st_err(mdev, os, ns, rv);
 204                         goto abort;
 205                 }
 206                 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 207                 ns = apply_mask_val(mdev->state, mask, val);
 208                 rv = _drbd_set_state(mdev, ns, f, &done);
 209                 drbd_state_unlock(mdev);
 210         } else {
 211                 rv = _drbd_set_state(mdev, ns, f, &done);
 212         }
 213
 214         spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 215
 216         if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) {
 217                 D_ASSERT(current != mdev->tconn->worker.task);
 218                 wait_for_completion(&done);
 219         }
 220
 221 abort:
 222         if (f & CS_SERIALIZE)
 223                 mutex_unlock(&mdev->state_mutex);
 224
 225         return rv;
 226 }
 227
 228 /**
 229  * _drbd_request_state() - Request a state change (with flags)
 230  * @mdev:       DRBD device.
 231  * @mask:       mask of state bits to change.
 232  * @val:        value of new state bits.
 233  * @f:          flags
 234  *
 235  * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE
 236  * flag, or when logging of failed state change requests is not desired.
 237  */
 238 enum drbd_state_rv
 239 _drbd_request_state(struct drbd_conf *mdev, union drbd_state mask,
 240                     union drbd_state val, enum chg_state_flags f)
 241 {
 242         enum drbd_state_rv rv;
 243
 244         wait_event(mdev->state_wait,
 245                    (rv = drbd_req_state(mdev, mask, val, f)) != SS_IN_TRANSIENT_STATE);
 246
 247         return rv;
 248 }
 249
 250 static void print_st(struct drbd_conf *mdev, char *name, union drbd_state ns)
 251 {
 252         dev_err(DEV, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n",
 253             name,
 254             drbd_conn_str(ns.conn),
 255             drbd_role_str(ns.role),
 256             drbd_role_str(ns.peer),
 257             drbd_disk_str(ns.disk),
 258             drbd_disk_str(ns.pdsk),
 259             is_susp(ns) ? 's' : 'r',
 260             ns.aftr_isp ? 'a' : '-',
 261             ns.peer_isp ? 'p' : '-',
 262             ns.user_isp ? 'u' : '-',
 263             ns.susp_fen ? 'F' : '-',
 264             ns.susp_nod ? 'N' : '-'
 265             );
 266 }
 267
 268 void print_st_err(struct drbd_conf *mdev, union drbd_state os,
 269                   union drbd_state ns, enum drbd_state_rv err)
 270 {
 271         if (err == SS_IN_TRANSIENT_STATE)
 272                 return;
 273         dev_err(DEV, "State change failed: %s\n", drbd_set_st_err_str(err));
 274         print_st(mdev, " state", os);
 275         print_st(mdev, "wanted", ns);
 276 }
 277
 278
 279 /**
 280  * is_valid_state() - Returns an SS_ error code if ns is not valid
 281  * @mdev:       DRBD device.
 282  * @ns:         State to consider.
 283  */
 284 static enum drbd_state_rv
 285 is_valid_state(struct drbd_conf *mdev, union drbd_state ns)
 286 {
 287         /* See drbd_state_sw_errors in drbd_strings.c */
 288
 289         enum drbd_fencing_p fp;
 290         enum drbd_state_rv rv = SS_SUCCESS;
 291
 292         fp = FP_DONT_CARE;
 293         if (get_ldev(mdev)) {
 294                 fp = mdev->ldev->dc.fencing;
 295                 put_ldev(mdev);
 296         }
 297
 298         if (get_net_conf(mdev->tconn)) {
 299                 if (!mdev->tconn->net_conf->two_primaries &&
 300                     ns.role == R_PRIMARY && ns.peer == R_PRIMARY)
 301                         rv = SS_TWO_PRIMARIES;
 302                 put_net_conf(mdev->tconn);
 303         }
 304
 305         if (rv <= 0)
 306                 /* already found a reason to abort */;
 307         else if (ns.role == R_SECONDARY && mdev->open_cnt)
 308                 rv = SS_DEVICE_IN_USE;
 309
 310         else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE)
 311                 rv = SS_NO_UP_TO_DATE_DISK;
 312
 313         else if (fp >= FP_RESOURCE &&
 314                  ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN)
 315                 rv = SS_PRIMARY_NOP;
 316
 317         else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT)
 318                 rv = SS_NO_UP_TO_DATE_DISK;
 319
 320         else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT)
 321                 rv = SS_NO_LOCAL_DISK;
 322
 323         else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT)
 324                 rv = SS_NO_REMOTE_DISK;
 325
 326         else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
 327                 rv = SS_NO_UP_TO_DATE_DISK;
 328
 329         else if ((ns.conn == C_CONNECTED ||
 330                   ns.conn == C_WF_BITMAP_S ||
 331                   ns.conn == C_SYNC_SOURCE ||
 332                   ns.conn == C_PAUSED_SYNC_S) &&
 333                   ns.disk == D_OUTDATED)
 334                 rv = SS_CONNECTED_OUTDATES;
 335
 336         else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 337                  (mdev->sync_conf.verify_alg[0] == 0))
 338                 rv = SS_NO_VERIFY_ALG;
 339
 340         else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 341                   mdev->tconn->agreed_pro_version < 88)
 342                 rv = SS_NOT_SUPPORTED;
 343
 344         else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN)
 345                 rv = SS_CONNECTED_OUTDATES;
 346
 347         return rv;
 348 }
 349
 350 /**
 351  * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible
 352  * This function limits state transitions that may be declined by DRBD. I.e.
 353  * user requests (aka soft transitions).
 354  * @mdev:       DRBD device.
 355  * @ns:         new state.
 356  * @os:         old state.
 357  */
 358 static enum drbd_state_rv
 359 is_valid_soft_transition(union drbd_state os, union drbd_state ns)
 360 {
 361         enum drbd_state_rv rv = SS_SUCCESS;
 362
 363         if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
 364             os.conn > C_CONNECTED)
 365                 rv = SS_RESYNC_RUNNING;
 366
 367         if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
 368                 rv = SS_ALREADY_STANDALONE;
 369
 370         if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
 371                 rv = SS_IS_DISKLESS;
 372
 373         if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
 374                 rv = SS_NO_NET_CONFIG;
 375
 376         if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
 377                 rv = SS_LOWER_THAN_OUTDATED;
 378
 379         if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
 380                 rv = SS_IN_TRANSIENT_STATE;
 381
 382         if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS)
 383                 rv = SS_IN_TRANSIENT_STATE;
 384
 385         if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
 386                 rv = SS_NEED_CONNECTION;
 387
 388         if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 389             ns.conn != os.conn && os.conn > C_CONNECTED)
 390                 rv = SS_RESYNC_RUNNING;
 391
 392         if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
 393             os.conn < C_CONNECTED)
 394                 rv = SS_NEED_CONNECTION;
 395
 396         if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)
 397             && os.conn < C_WF_REPORT_PARAMS)
 398                 rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */
 399
 400         return rv;
 401 }
 402
 403 static enum drbd_state_rv
 404 is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc)
 405 {
 406         enum drbd_state_rv rv = SS_SUCCESS;
 407
 408         /* Disallow Network errors to configure a device's network part */
 409         if ((nc >= C_TIMEOUT && nc <= C_TEAR_DOWN) && oc <= C_DISCONNECTING)
 410                 rv = SS_NEED_CONNECTION;
 411
 412         /* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */
 413         if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING)
 414                 rv = SS_IN_TRANSIENT_STATE;
 415
 416         /* After C_DISCONNECTING only C_STANDALONE may follow */
 417         if (oc == C_DISCONNECTING && nc != C_STANDALONE)
 418                 rv = SS_IN_TRANSIENT_STATE;
 419
 420         return rv;
 421 }
 422
 423
 424 /**
 425  * is_valid_transition() - Returns an SS_ error code if the state transition is not possible
 426  * This limits hard state transitions. Hard state transitions are facts there are
 427  * imposed on DRBD by the environment. E.g. disk broke or network broke down.
 428  * But those hard state transitions are still not allowed to do everything.
 429  * @ns:         new state.
 430  * @os:         old state.
 431  */
 432 static enum drbd_state_rv
 433 is_valid_transition(union drbd_state os, union drbd_state ns)
 434 {
 435         enum drbd_state_rv rv;
 436
 437         rv = is_valid_conn_transition(os.conn, ns.conn);
 438
 439         /* we cannot fail (again) if we already detached */
 440         if (ns.disk == D_FAILED && os.disk == D_DISKLESS)
 441                 rv = SS_IS_DISKLESS;
 442
 443         /* if we are only D_ATTACHING yet,
 444          * we can (and should) go directly to D_DISKLESS. */
 445         if (ns.disk == D_FAILED && os.disk == D_ATTACHING) {
 446                 printk("TODO: FIX ME\n");
 447                 rv = SS_IS_DISKLESS;
 448         }
 449
 450         return rv;
 451 }
 452
 453 /**
 454  * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
 455  * @mdev:       DRBD device.
 456  * @os:         old state.
 457  * @ns:         new state.
 458  * @warn_sync_abort:
 459  *
 460  * When we loose connection, we have to set the state of the peers disk (pdsk)
 461  * to D_UNKNOWN. This rule and many more along those lines are in this function.
 462  */
 463 static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns,
 464                                        const char **warn_sync_abort)
 465 {
 466         enum drbd_fencing_p fp;
 467         enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max;
 468
 469         fp = FP_DONT_CARE;
 470         if (get_ldev(mdev)) {
 471                 fp = mdev->ldev->dc.fencing;
 472                 put_ldev(mdev);
 473         }
 474
 475         /* Implications from connection to peer and peer_isp */
 476         if (ns.conn < C_CONNECTED) {
 477                 ns.peer_isp = 0;
 478                 ns.peer = R_UNKNOWN;
 479                 if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
 480                         ns.pdsk = D_UNKNOWN;
 481         }
 482
 483         /* Clear the aftr_isp when becoming unconfigured */
 484         if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
 485                 ns.aftr_isp = 0;
 486
 487         /* An implication of the disk states onto the connection state */
 488         /* Abort resync if a disk fails/detaches */
 489         if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
 490                 if (warn_sync_abort)
 491                         *warn_sync_abort =
 492                                 ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ?
 493                                 "Online-verify" : "Resync";
 494                 ns.conn = C_CONNECTED;
 495         }
 496
 497         /* Connection breaks down before we finished "Negotiating" */
 498         if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
 499             get_ldev_if_state(mdev, D_NEGOTIATING)) {
 500                 if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) {
 501                         ns.disk = mdev->new_state_tmp.disk;
 502                         ns.pdsk = mdev->new_state_tmp.pdsk;
 503                 } else {
 504                         dev_alert(DEV, "Connection lost while negotiating, no data!\n");
 505                         ns.disk = D_DISKLESS;
 506                         ns.pdsk = D_UNKNOWN;
 507                 }
 508                 put_ldev(mdev);
 509         }
 510
 511         /* D_CONSISTENT and D_OUTDATED vanish when we get connected */
 512         if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) {
 513                 if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED)
 514                         ns.disk = D_UP_TO_DATE;
 515                 if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)
 516                         ns.pdsk = D_UP_TO_DATE;
 517         }
 518
 519         /* Implications of the connection stat on the disk states */
 520         disk_min = D_DISKLESS;
 521         disk_max = D_UP_TO_DATE;
 522         pdsk_min = D_INCONSISTENT;
 523         pdsk_max = D_UNKNOWN;
 524         switch ((enum drbd_conns)ns.conn) {
 525         case C_WF_BITMAP_T:
 526         case C_PAUSED_SYNC_T:
 527         case C_STARTING_SYNC_T:
 528         case C_WF_SYNC_UUID:
 529         case C_BEHIND:
 530                 disk_min = D_INCONSISTENT;
 531                 disk_max = D_OUTDATED;
 532                 pdsk_min = D_UP_TO_DATE;
 533                 pdsk_max = D_UP_TO_DATE;
 534                 break;
 535         case C_VERIFY_S:
 536         case C_VERIFY_T:
 537                 disk_min = D_UP_TO_DATE;
 538                 disk_max = D_UP_TO_DATE;
 539                 pdsk_min = D_UP_TO_DATE;
 540                 pdsk_max = D_UP_TO_DATE;
 541                 break;
 542         case C_CONNECTED:
 543                 disk_min = D_DISKLESS;
 544                 disk_max = D_UP_TO_DATE;
 545                 pdsk_min = D_DISKLESS;
 546                 pdsk_max = D_UP_TO_DATE;
 547                 break;
 548         case C_WF_BITMAP_S:
 549         case C_PAUSED_SYNC_S:
 550         case C_STARTING_SYNC_S:
 551         case C_AHEAD:
 552                 disk_min = D_UP_TO_DATE;
 553                 disk_max = D_UP_TO_DATE;
 554                 pdsk_min = D_INCONSISTENT;
 555                 pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/
 556                 break;
 557         case C_SYNC_TARGET:
 558                 disk_min = D_INCONSISTENT;
 559                 disk_max = D_INCONSISTENT;
 560                 pdsk_min = D_UP_TO_DATE;
 561                 pdsk_max = D_UP_TO_DATE;
 562                 break;
 563         case C_SYNC_SOURCE:
 564                 disk_min = D_UP_TO_DATE;
 565                 disk_max = D_UP_TO_DATE;
 566                 pdsk_min = D_INCONSISTENT;
 567                 pdsk_max = D_INCONSISTENT;
 568                 break;
 569         case C_STANDALONE:
 570         case C_DISCONNECTING:
 571         case C_UNCONNECTED:
 572         case C_TIMEOUT:
 573         case C_BROKEN_PIPE:
 574         case C_NETWORK_FAILURE:
 575         case C_PROTOCOL_ERROR:
 576         case C_TEAR_DOWN:
 577         case C_WF_CONNECTION:
 578         case C_WF_REPORT_PARAMS:
 579         case C_MASK:
 580                 break;
 581         }
 582         if (ns.disk > disk_max)
 583                 ns.disk = disk_max;
 584
 585         if (ns.disk < disk_min) {
 586                 dev_warn(DEV, "Implicitly set disk from %s to %s\n",
 587                          drbd_disk_str(ns.disk), drbd_disk_str(disk_min));
 588                 ns.disk = disk_min;
 589         }
 590         if (ns.pdsk > pdsk_max)
 591                 ns.pdsk = pdsk_max;
 592
 593         if (ns.pdsk < pdsk_min) {
 594                 dev_warn(DEV, "Implicitly set pdsk from %s to %s\n",
 595                          drbd_disk_str(ns.pdsk), drbd_disk_str(pdsk_min));
 596                 ns.pdsk = pdsk_min;
 597         }
 598
 599         if (fp == FP_STONITH &&
 600             (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED))
 601                 ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */
 602
 603         if (mdev->sync_conf.on_no_data == OND_SUSPEND_IO &&
 604             (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
 605                 ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */
 606
 607         if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
 608                 if (ns.conn == C_SYNC_SOURCE)
 609                         ns.conn = C_PAUSED_SYNC_S;
 610                 if (ns.conn == C_SYNC_TARGET)
 611                         ns.conn = C_PAUSED_SYNC_T;
 612         } else {
 613                 if (ns.conn == C_PAUSED_SYNC_S)
 614                         ns.conn = C_SYNC_SOURCE;
 615                 if (ns.conn == C_PAUSED_SYNC_T)
 616                         ns.conn = C_SYNC_TARGET;
 617         }
 618
 619         return ns;
 620 }
 621
 622 void drbd_resume_al(struct drbd_conf *mdev)
 623 {
 624         if (test_and_clear_bit(AL_SUSPENDED, &mdev->flags))
 625                 dev_info(DEV, "Resumed AL updates\n");
 626 }
 627
 628 /* helper for __drbd_set_state */
 629 static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs)
 630 {
 631         if (mdev->tconn->agreed_pro_version < 90)
 632                 mdev->ov_start_sector = 0;
 633         mdev->rs_total = drbd_bm_bits(mdev);
 634         mdev->ov_position = 0;
 635         if (cs == C_VERIFY_T) {
 636                 /* starting online verify from an arbitrary position
 637                  * does not fit well into the existing protocol.
 638                  * on C_VERIFY_T, we initialize ov_left and friends
 639                  * implicitly in receive_DataRequest once the
 640                  * first P_OV_REQUEST is received */
 641                 mdev->ov_start_sector = ~(sector_t)0;
 642         } else {
 643                 unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector);
 644                 if (bit >= mdev->rs_total) {
 645                         mdev->ov_start_sector =
 646                                 BM_BIT_TO_SECT(mdev->rs_total - 1);
 647                         mdev->rs_total = 1;
 648                 } else
 649                         mdev->rs_total -= bit;
 650                 mdev->ov_position = mdev->ov_start_sector;
 651         }
 652         mdev->ov_left = mdev->rs_total;
 653 }
 654
 655 /**
 656  * __drbd_set_state() - Set a new DRBD state
 657  * @mdev:       DRBD device.
 658  * @ns:         new state.
 659  * @flags:      Flags
 660  * @done:       Optional completion, that will get completed after the after_state_ch() finished
 661  *
 662  * Caller needs to hold req_lock, and global_state_lock. Do not call directly.
 663  */
 664 enum drbd_state_rv
 665 __drbd_set_state(struct drbd_conf *mdev, union drbd_state ns,
 666                  enum chg_state_flags flags, struct completion *done)
 667 {
 668         union drbd_state os;
 669         enum drbd_state_rv rv = SS_SUCCESS;
 670         const char *warn_sync_abort = NULL;
 671         struct after_state_chg_work *ascw;
 672
 673         os = mdev->state;
 674
 675         ns = sanitize_state(mdev, ns, &warn_sync_abort);
 676         if (ns.i == os.i)
 677                 return SS_NOTHING_TO_DO;
 678
 679         rv = is_valid_transition(os, ns);
 680         if (rv < SS_SUCCESS)
 681                 return rv;
 682
 683         if (!(flags & CS_HARD)) {
 684                 /*  pre-state-change checks ; only look at ns  */
 685                 /* See drbd_state_sw_errors in drbd_strings.c */
 686
 687                 rv = is_valid_state(mdev, ns);
 688                 if (rv < SS_SUCCESS) {
 689                         /* If the old state was illegal as well, then let
 690                            this happen...*/
 691
 692                         if (is_valid_state(mdev, os) == rv)
 693                                 rv = is_valid_soft_transition(os, ns);
 694                 } else
 695                         rv = is_valid_soft_transition(os, ns);
 696         }
 697
 698         if (rv < SS_SUCCESS) {
 699                 if (flags & CS_VERBOSE)
 700                         print_st_err(mdev, os, ns, rv);
 701                 return rv;
 702         }
 703
 704         if (warn_sync_abort)
 705                 dev_warn(DEV, "%s aborted.\n", warn_sync_abort);
 706
 707         {
 708         char *pbp, pb[300];
 709         pbp = pb;
 710         *pbp = 0;
 711         if (ns.role != os.role)
 712                 pbp += sprintf(pbp, "role( %s -> %s ) ",
 713                                drbd_role_str(os.role),
 714                                drbd_role_str(ns.role));
 715         if (ns.peer != os.peer)
 716                 pbp += sprintf(pbp, "peer( %s -> %s ) ",
 717                                drbd_role_str(os.peer),
 718                                drbd_role_str(ns.peer));
 719         if (ns.conn != os.conn)
 720                 pbp += sprintf(pbp, "conn( %s -> %s ) ",
 721                                drbd_conn_str(os.conn),
 722                                drbd_conn_str(ns.conn));
 723         if (ns.disk != os.disk)
 724                 pbp += sprintf(pbp, "disk( %s -> %s ) ",
 725                                drbd_disk_str(os.disk),
 726                                drbd_disk_str(ns.disk));
 727         if (ns.pdsk != os.pdsk)
 728                 pbp += sprintf(pbp, "pdsk( %s -> %s ) ",
 729                                drbd_disk_str(os.pdsk),
 730                                drbd_disk_str(ns.pdsk));
 731         if (is_susp(ns) != is_susp(os))
 732                 pbp += sprintf(pbp, "susp( %d -> %d ) ",
 733                                is_susp(os),
 734                                is_susp(ns));
 735         if (ns.aftr_isp != os.aftr_isp)
 736                 pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ",
 737                                os.aftr_isp,
 738                                ns.aftr_isp);
 739         if (ns.peer_isp != os.peer_isp)
 740                 pbp += sprintf(pbp, "peer_isp( %d -> %d ) ",
 741                                os.peer_isp,
 742                                ns.peer_isp);
 743         if (ns.user_isp != os.user_isp)
 744                 pbp += sprintf(pbp, "user_isp( %d -> %d ) ",
 745                                os.user_isp,
 746                                ns.user_isp);
 747         dev_info(DEV, "%s\n", pb);
 748         }
 749
 750         /* solve the race between becoming unconfigured,
 751          * worker doing the cleanup, and
 752          * admin reconfiguring us:
 753          * on (re)configure, first set CONFIG_PENDING,
 754          * then wait for a potentially exiting worker,
 755          * start the worker, and schedule one no_op.
 756          * then proceed with configuration.
 757          */
 758         if (ns.disk == D_DISKLESS &&
 759             ns.conn == C_STANDALONE &&
 760             ns.role == R_SECONDARY &&
 761             !test_and_set_bit(CONFIG_PENDING, &mdev->flags))
 762                 set_bit(DEVICE_DYING, &mdev->flags);
 763
 764         /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
 765          * on the ldev here, to be sure the transition -> D_DISKLESS resp.
 766          * drbd_ldev_destroy() won't happen before our corresponding
 767          * after_state_ch works run, where we put_ldev again. */
 768         if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
 769             (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
 770                 atomic_inc(&mdev->local_cnt);
 771
 772         mdev->state = ns;
 773
 774         if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
 775                 drbd_print_uuids(mdev, "attached to UUIDs");
 776
 777         wake_up(&mdev->misc_wait);
 778         wake_up(&mdev->state_wait);
 779         wake_up(&mdev->tconn->ping_wait);
 780
 781         /* aborted verify run. log the last position */
 782         if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
 783             ns.conn < C_CONNECTED) {
 784                 mdev->ov_start_sector =
 785                         BM_BIT_TO_SECT(drbd_bm_bits(mdev) - mdev->ov_left);
 786                 dev_info(DEV, "Online Verify reached sector %llu\n",
 787                         (unsigned long long)mdev->ov_start_sector);
 788         }
 789
 790         if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
 791             (ns.conn == C_SYNC_TARGET  || ns.conn == C_SYNC_SOURCE)) {
 792                 dev_info(DEV, "Syncer continues.\n");
 793                 mdev->rs_paused += (long)jiffies
 794                                   -(long)mdev->rs_mark_time[mdev->rs_last_mark];
 795                 if (ns.conn == C_SYNC_TARGET)
 796                         mod_timer(&mdev->resync_timer, jiffies);
 797         }
 798
 799         if ((os.conn == C_SYNC_TARGET  || os.conn == C_SYNC_SOURCE) &&
 800             (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
 801                 dev_info(DEV, "Resync suspended\n");
 802                 mdev->rs_mark_time[mdev->rs_last_mark] = jiffies;
 803         }
 804
 805         if (os.conn == C_CONNECTED &&
 806             (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
 807                 unsigned long now = jiffies;
 808                 int i;
 809
 810                 set_ov_position(mdev, ns.conn);
 811                 mdev->rs_start = now;
 812                 mdev->rs_last_events = 0;
 813                 mdev->rs_last_sect_ev = 0;
 814                 mdev->ov_last_oos_size = 0;
 815                 mdev->ov_last_oos_start = 0;
 816
 817                 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
 818                         mdev->rs_mark_left[i] = mdev->ov_left;
 819                         mdev->rs_mark_time[i] = now;
 820                 }
 821
 822                 drbd_rs_controller_reset(mdev);
 823
 824                 if (ns.conn == C_VERIFY_S) {
 825                         dev_info(DEV, "Starting Online Verify from sector %llu\n",
 826                                         (unsigned long long)mdev->ov_position);
 827                         mod_timer(&mdev->resync_timer, jiffies);
 828                 }
 829         }
 830
 831         if (get_ldev(mdev)) {
 832                 u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
 833                                                  MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
 834                                                  MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
 835
 836                 if (test_bit(CRASHED_PRIMARY, &mdev->flags))
 837                         mdf |= MDF_CRASHED_PRIMARY;
 838                 if (mdev->state.role == R_PRIMARY ||
 839                     (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY))
 840                         mdf |= MDF_PRIMARY_IND;
 841                 if (mdev->state.conn > C_WF_REPORT_PARAMS)
 842                         mdf |= MDF_CONNECTED_IND;
 843                 if (mdev->state.disk > D_INCONSISTENT)
 844                         mdf |= MDF_CONSISTENT;
 845                 if (mdev->state.disk > D_OUTDATED)
 846                         mdf |= MDF_WAS_UP_TO_DATE;
 847                 if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT)
 848                         mdf |= MDF_PEER_OUT_DATED;
 849                 if (mdf != mdev->ldev->md.flags) {
 850                         mdev->ldev->md.flags = mdf;
 851                         drbd_md_mark_dirty(mdev);
 852                 }
 853                 if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
 854                         drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]);
 855                 put_ldev(mdev);
 856         }
 857
 858         /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
 859         if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
 860             os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
 861                 set_bit(CONSIDER_RESYNC, &mdev->flags);
 862
 863         /* Receiver should clean up itself */
 864         if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
 865                 drbd_thread_stop_nowait(&mdev->tconn->receiver);
 866
 867         /* Now the receiver finished cleaning up itself, it should die */
 868         if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
 869                 drbd_thread_stop_nowait(&mdev->tconn->receiver);
 870
 871         /* Upon network failure, we need to restart the receiver. */
 872         if (os.conn > C_TEAR_DOWN &&
 873             ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
 874                 drbd_thread_restart_nowait(&mdev->tconn->receiver);
 875
 876         /* Resume AL writing if we get a connection */
 877         if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
 878                 drbd_resume_al(mdev);
 879
 880         ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
 881         if (ascw) {
 882                 ascw->os = os;
 883                 ascw->ns = ns;
 884                 ascw->flags = flags;
 885                 ascw->w.cb = w_after_state_ch;
 886                 ascw->w.mdev = mdev;
 887                 ascw->done = done;
 888                 drbd_queue_work(&mdev->tconn->data.work, &ascw->w);
 889         } else {
 890                 dev_warn(DEV, "Could not kmalloc an ascw\n");
 891         }
 892
 893         return rv;
 894 }
 895
 896 static int w_after_state_ch(struct drbd_work *w, int unused)
 897 {
 898         struct after_state_chg_work *ascw =
 899                 container_of(w, struct after_state_chg_work, w);
 900         struct drbd_conf *mdev = w->mdev;
 901
 902         after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags);
 903         if (ascw->flags & CS_WAIT_COMPLETE) {
 904                 D_ASSERT(ascw->done != NULL);
 905                 complete(ascw->done);
 906         }
 907         kfree(ascw);
 908
 909         return 1;
 910 }
 911
 912 static void abw_start_sync(struct drbd_conf *mdev, int rv)
 913 {
 914         if (rv) {
 915                 dev_err(DEV, "Writing the bitmap failed not starting resync.\n");
 916                 _drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE);
 917                 return;
 918         }
 919
 920         switch (mdev->state.conn) {
 921         case C_STARTING_SYNC_T:
 922                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
 923                 break;
 924         case C_STARTING_SYNC_S:
 925                 drbd_start_resync(mdev, C_SYNC_SOURCE);
 926                 break;
 927         }
 928 }
 929
 930 int drbd_bitmap_io_from_worker(struct drbd_conf *mdev,
 931                 int (*io_fn)(struct drbd_conf *),
 932                 char *why, enum bm_flag flags)
 933 {
 934         int rv;
 935
 936         D_ASSERT(current == mdev->tconn->worker.task);
 937
 938         /* open coded non-blocking drbd_suspend_io(mdev); */
 939         set_bit(SUSPEND_IO, &mdev->flags);
 940
 941         drbd_bm_lock(mdev, why, flags);
 942         rv = io_fn(mdev);
 943         drbd_bm_unlock(mdev);
 944
 945         drbd_resume_io(mdev);
 946
 947         return rv;
 948 }
 949
 950 /**
 951  * after_state_ch() - Perform after state change actions that may sleep
 952  * @mdev:       DRBD device.
 953  * @os:         old state.
 954  * @ns:         new state.
 955  * @flags:      Flags
 956  */
 957 static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
 958                            union drbd_state ns, enum chg_state_flags flags)
 959 {
 960         enum drbd_fencing_p fp;
 961         enum drbd_req_event what = NOTHING;
 962         union drbd_state nsm = (union drbd_state){ .i = -1 };
 963
 964         if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) {
 965                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
 966                 if (mdev->p_uuid)
 967                         mdev->p_uuid[UI_FLAGS] &= ~((u64)2);
 968         }
 969
 970         fp = FP_DONT_CARE;
 971         if (get_ldev(mdev)) {
 972                 fp = mdev->ldev->dc.fencing;
 973                 put_ldev(mdev);
 974         }
 975
 976         /* Inform userspace about the change... */
 977         drbd_bcast_state(mdev, ns);
 978
 979         if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
 980             (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
 981                 drbd_khelper(mdev, "pri-on-incon-degr");
 982
 983         /* Here we have the actions that are performed after a
 984            state change. This function might sleep */
 985
 986         nsm.i = -1;
 987         if (ns.susp_nod) {
 988                 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
 989                         what = RESEND;
 990
 991                 if (os.disk == D_ATTACHING && ns.disk > D_ATTACHING)
 992                         what = RESTART_FROZEN_DISK_IO;
 993
 994                 if (what != NOTHING)
 995                         nsm.susp_nod = 0;
 996         }
 997
 998         if (ns.susp_fen) {
 999                 /* case1: The outdate peer handler is successful: */
1000                 if (os.pdsk > D_OUTDATED  && ns.pdsk <= D_OUTDATED) {
1001                         tl_clear(mdev);
1002                         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1003                                 drbd_uuid_new_current(mdev);
1004                                 clear_bit(NEW_CUR_UUID, &mdev->flags);
1005                         }
1006                         spin_lock_irq(&mdev->tconn->req_lock);
1007                         _drbd_set_state(_NS(mdev, susp_fen, 0), CS_VERBOSE, NULL);
1008                         spin_unlock_irq(&mdev->tconn->req_lock);
1009                 }
1010                 /* case2: The connection was established again: */
1011                 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) {
1012                         clear_bit(NEW_CUR_UUID, &mdev->flags);
1013                         what = RESEND;
1014                         nsm.susp_fen = 0;
1015                 }
1016         }
1017
1018         if (what != NOTHING) {
1019                 spin_lock_irq(&mdev->tconn->req_lock);
1020                 _tl_restart(mdev, what);
1021                 nsm.i &= mdev->state.i;
1022                 _drbd_set_state(mdev, nsm, CS_VERBOSE, NULL);
1023                 spin_unlock_irq(&mdev->tconn->req_lock);
1024         }
1025
1026         /* Became sync source.  With protocol >= 96, we still need to send out
1027          * the sync uuid now. Need to do that before any drbd_send_state, or
1028          * the other side may go "paused sync" before receiving the sync uuids,
1029          * which is unexpected. */
1030         if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
1031             (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
1032             mdev->tconn->agreed_pro_version >= 96 && get_ldev(mdev)) {
1033                 drbd_gen_and_send_sync_uuid(mdev);
1034                 put_ldev(mdev);
1035         }
1036
1037         /* Do not change the order of the if above and the two below... */
1038         if (os.pdsk == D_DISKLESS && ns.pdsk > D_DISKLESS) {      /* attach on the peer */
1039                 drbd_send_uuids(mdev);
1040                 drbd_send_state(mdev);
1041         }
1042         /* No point in queuing send_bitmap if we don't have a connection
1043          * anymore, so check also the _current_ state, not only the new state
1044          * at the time this work was queued. */
1045         if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
1046             mdev->state.conn == C_WF_BITMAP_S)
1047                 drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL,
1048                                 "send_bitmap (WFBitMapS)",
1049                                 BM_LOCKED_TEST_ALLOWED);
1050
1051         /* Lost contact to peer's copy of the data */
1052         if ((os.pdsk >= D_INCONSISTENT &&
1053              os.pdsk != D_UNKNOWN &&
1054              os.pdsk != D_OUTDATED)
1055         &&  (ns.pdsk < D_INCONSISTENT ||
1056              ns.pdsk == D_UNKNOWN ||
1057              ns.pdsk == D_OUTDATED)) {
1058                 if (get_ldev(mdev)) {
1059                         if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1060                             mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1061                                 if (is_susp(mdev->state)) {
1062                                         set_bit(NEW_CUR_UUID, &mdev->flags);
1063                                 } else {
1064                                         drbd_uuid_new_current(mdev);
1065                                         drbd_send_uuids(mdev);
1066                                 }
1067                         }
1068                         put_ldev(mdev);
1069                 }
1070         }
1071
1072         if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) {
1073                 if (ns.peer == R_PRIMARY && mdev->ldev->md.uuid[UI_BITMAP] == 0) {
1074                         drbd_uuid_new_current(mdev);
1075                         drbd_send_uuids(mdev);
1076                 }
1077
1078                 /* D_DISKLESS Peer becomes secondary */
1079                 if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1080                         /* We may still be Primary ourselves.
1081                          * No harm done if the bitmap still changes,
1082                          * redirtied pages will follow later. */
1083                         drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1084                                 "demote diskless peer", BM_LOCKED_SET_ALLOWED);
1085                 put_ldev(mdev);
1086         }
1087
1088         /* Write out all changed bits on demote.
1089          * Though, no need to da that just yet
1090          * if there is a resync going on still */
1091         if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
1092                 mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) {
1093                 /* No changes to the bitmap expected this time, so assert that,
1094                  * even though no harm was done if it did change. */
1095                 drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1096                                 "demote", BM_LOCKED_TEST_ALLOWED);
1097                 put_ldev(mdev);
1098         }
1099
1100         /* Last part of the attaching process ... */
1101         if (ns.conn >= C_CONNECTED &&
1102             os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1103                 drbd_send_sizes(mdev, 0, 0);  /* to start sync... */
1104                 drbd_send_uuids(mdev);
1105                 drbd_send_state(mdev);
1106         }
1107
1108         /* We want to pause/continue resync, tell peer. */
1109         if (ns.conn >= C_CONNECTED &&
1110              ((os.aftr_isp != ns.aftr_isp) ||
1111               (os.user_isp != ns.user_isp)))
1112                 drbd_send_state(mdev);
1113
1114         /* In case one of the isp bits got set, suspend other devices. */
1115         if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1116             (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1117                 suspend_other_sg(mdev);
1118
1119         /* Make sure the peer gets informed about eventual state
1120            changes (ISP bits) while we were in WFReportParams. */
1121         if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1122                 drbd_send_state(mdev);
1123
1124         if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
1125                 drbd_send_state(mdev);
1126
1127         /* We are in the progress to start a full sync... */
1128         if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1129             (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1130                 /* no other bitmap changes expected during this phase */
1131                 drbd_queue_bitmap_io(mdev,
1132                         &drbd_bmio_set_n_write, &abw_start_sync,
1133                         "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED);
1134
1135         /* We are invalidating our self... */
1136         if (os.conn < C_CONNECTED && ns.conn < C_CONNECTED &&
1137             os.disk > D_INCONSISTENT && ns.disk == D_INCONSISTENT)
1138                 /* other bitmap operation expected during this phase */
1139                 drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, NULL,
1140                         "set_n_write from invalidate", BM_LOCKED_MASK);
1141
1142         /* first half of local IO error, failure to attach,
1143          * or administrative detach */
1144         if (os.disk != D_FAILED && ns.disk == D_FAILED) {
1145                 enum drbd_io_error_p eh;
1146                 int was_io_error;
1147                 /* corresponding get_ldev was in __drbd_set_state, to serialize
1148                  * our cleanup here with the transition to D_DISKLESS,
1149                  * so it is safe to dreference ldev here. */
1150                 eh = mdev->ldev->dc.on_io_error;
1151                 was_io_error = test_and_clear_bit(WAS_IO_ERROR, &mdev->flags);
1152
1153                 /* current state still has to be D_FAILED,
1154                  * there is only one way out: to D_DISKLESS,
1155                  * and that may only happen after our put_ldev below. */
1156                 if (mdev->state.disk != D_FAILED)
1157                         dev_err(DEV,
1158                                 "ASSERT FAILED: disk is %s during detach\n",
1159                                 drbd_disk_str(mdev->state.disk));
1160
1161                 if (drbd_send_state(mdev))
1162                         dev_warn(DEV, "Notified peer that I am detaching my disk\n");
1163                 else
1164                         dev_err(DEV, "Sending state for detaching disk failed\n");
1165
1166                 drbd_rs_cancel_all(mdev);
1167
1168                 /* In case we want to get something to stable storage still,
1169                  * this may be the last chance.
1170                  * Following put_ldev may transition to D_DISKLESS. */
1171                 drbd_md_sync(mdev);
1172                 put_ldev(mdev);
1173
1174                 if (was_io_error && eh == EP_CALL_HELPER)
1175                         drbd_khelper(mdev, "local-io-error");
1176         }
1177
1178         /* second half of local IO error, failure to attach,
1179          * or administrative detach,
1180          * after local_cnt references have reached zero again */
1181         if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
1182                 /* We must still be diskless,
1183                  * re-attach has to be serialized with this! */
1184                 if (mdev->state.disk != D_DISKLESS)
1185                         dev_err(DEV,
1186                                 "ASSERT FAILED: disk is %s while going diskless\n",
1187                                 drbd_disk_str(mdev->state.disk));
1188
1189                 mdev->rs_total = 0;
1190                 mdev->rs_failed = 0;
1191                 atomic_set(&mdev->rs_pending_cnt, 0);
1192
1193                 if (drbd_send_state(mdev))
1194                         dev_warn(DEV, "Notified peer that I'm now diskless.\n");
1195                 /* corresponding get_ldev in __drbd_set_state
1196                  * this may finally trigger drbd_ldev_destroy. */
1197                 put_ldev(mdev);
1198         }
1199
1200         /* Notify peer that I had a local IO error, and did not detached.. */
1201         if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT)
1202                 drbd_send_state(mdev);
1203
1204         /* Disks got bigger while they were detached */
1205         if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1206             test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) {
1207                 if (ns.conn == C_CONNECTED)
1208                         resync_after_online_grow(mdev);
1209         }
1210
1211         /* A resync finished or aborted, wake paused devices... */
1212         if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1213             (os.peer_isp && !ns.peer_isp) ||
1214             (os.user_isp && !ns.user_isp))
1215                 resume_next_sg(mdev);
1216
1217         /* sync target done with resync.  Explicitly notify peer, even though
1218          * it should (at least for non-empty resyncs) already know itself. */
1219         if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
1220                 drbd_send_state(mdev);
1221
1222         /* This triggers bitmap writeout of potentially still unwritten pages
1223          * if the resync finished cleanly, or aborted because of peer disk
1224          * failure, or because of connection loss.
1225          * For resync aborted because of local disk failure, we cannot do
1226          * any bitmap writeout anymore.
1227          * No harm done if some bits change during this phase.
1228          */
1229         if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) {
1230                 drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL,
1231                         "write from resync_finished", BM_LOCKED_SET_ALLOWED);
1232                 put_ldev(mdev);
1233         }
1234
1235         if (ns.disk == D_DISKLESS &&
1236             ns.conn == C_STANDALONE &&
1237             ns.role == R_SECONDARY) {
1238                 if (os.aftr_isp != ns.aftr_isp)
1239                         resume_next_sg(mdev);
1240         }
1241
1242         after_conn_state_ch(mdev->tconn, os, ns, flags);
1243         drbd_md_sync(mdev);
1244 }
1245
1246 static void after_conn_state_ch(struct drbd_tconn *tconn, union drbd_state os,
1247                                 union drbd_state ns, enum chg_state_flags flags)
1248 {
1249         /* Upon network configuration, we need to start the receiver */
1250         if (os.conn == C_STANDALONE && ns.conn == C_UNCONNECTED)
1251                 drbd_thread_start(&tconn->receiver);
1252
1253         if (ns.disk == D_DISKLESS &&
1254             ns.conn == C_STANDALONE &&
1255             ns.role == R_SECONDARY) {
1256                 /* if (test_bit(DEVICE_DYING, &mdev->flags)) TODO: DEVICE_DYING functionality */
1257                 drbd_thread_stop_nowait(&tconn->worker);
1258         }
1259 }