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 /* in drbd_main.c */
  33 extern void tl_abort_disk_io(struct drbd_conf *mdev);
  34
  35 struct after_state_chg_work {
  36         struct drbd_work w;
  37         union drbd_state os;
  38         union drbd_state ns;
  39         enum chg_state_flags flags;
  40         struct completion *done;
  41 };
  42
  43 enum sanitize_state_warnings {
  44         NO_WARNING,
  45         ABORTED_ONLINE_VERIFY,
  46         ABORTED_RESYNC,
  47         CONNECTION_LOST_NEGOTIATING,
  48         IMPLICITLY_UPGRADED_DISK,
  49         IMPLICITLY_UPGRADED_PDSK,
  50 };
  51
  52 static int w_after_state_ch(struct drbd_work *w, int unused);
  53 static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
  54                            union drbd_state ns, enum chg_state_flags flags);
  55 static enum drbd_state_rv is_valid_state(struct drbd_conf *, union drbd_state);
  56 static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state, struct drbd_tconn *);
  57 static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns);
  58 static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns,
  59                                        enum sanitize_state_warnings *warn);
  60
  61 static inline bool is_susp(union drbd_state s)
  62 {
  63         return s.susp || s.susp_nod || s.susp_fen;
  64 }
  65
  66 bool conn_all_vols_unconf(struct drbd_tconn *tconn)
  67 {
  68         struct drbd_conf *mdev;
  69         bool rv = true;
  70         int vnr;
  71
  72         rcu_read_lock();
  73         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
  74                 if (mdev->state.disk != D_DISKLESS ||
  75                     mdev->state.conn != C_STANDALONE ||
  76                     mdev->state.role != R_SECONDARY) {
  77                         rv = false;
  78                         break;
  79                 }
  80         }
  81         rcu_read_unlock();
  82
  83         return rv;
  84 }
  85
  86 /* Unfortunately the states where not correctly ordered, when
  87    they where defined. therefore can not use max_t() here. */
  88 static enum drbd_role max_role(enum drbd_role role1, enum drbd_role role2)
  89 {
  90         if (role1 == R_PRIMARY || role2 == R_PRIMARY)
  91                 return R_PRIMARY;
  92         if (role1 == R_SECONDARY || role2 == R_SECONDARY)
  93                 return R_SECONDARY;
  94         return R_UNKNOWN;
  95 }
  96 static enum drbd_role min_role(enum drbd_role role1, enum drbd_role role2)
  97 {
  98         if (role1 == R_UNKNOWN || role2 == R_UNKNOWN)
  99                 return R_UNKNOWN;
 100         if (role1 == R_SECONDARY || role2 == R_SECONDARY)
 101                 return R_SECONDARY;
 102         return R_PRIMARY;
 103 }
 104
 105 enum drbd_role conn_highest_role(struct drbd_tconn *tconn)
 106 {
 107         enum drbd_role role = R_UNKNOWN;
 108         struct drbd_conf *mdev;
 109         int vnr;
 110
 111         rcu_read_lock();
 112         idr_for_each_entry(&tconn->volumes, mdev, vnr)
 113                 role = max_role(role, mdev->state.role);
 114         rcu_read_unlock();
 115
 116         return role;
 117 }
 118
 119 enum drbd_role conn_highest_peer(struct drbd_tconn *tconn)
 120 {
 121         enum drbd_role peer = R_UNKNOWN;
 122         struct drbd_conf *mdev;
 123         int vnr;
 124
 125         rcu_read_lock();
 126         idr_for_each_entry(&tconn->volumes, mdev, vnr)
 127                 peer = max_role(peer, mdev->state.peer);
 128         rcu_read_unlock();
 129
 130         return peer;
 131 }
 132
 133 enum drbd_disk_state conn_highest_disk(struct drbd_tconn *tconn)
 134 {
 135         enum drbd_disk_state ds = D_DISKLESS;
 136         struct drbd_conf *mdev;
 137         int vnr;
 138
 139         rcu_read_lock();
 140         idr_for_each_entry(&tconn->volumes, mdev, vnr)
 141                 ds = max_t(enum drbd_disk_state, ds, mdev->state.disk);
 142         rcu_read_unlock();
 143
 144         return ds;
 145 }
 146
 147 enum drbd_disk_state conn_lowest_disk(struct drbd_tconn *tconn)
 148 {
 149         enum drbd_disk_state ds = D_MASK;
 150         struct drbd_conf *mdev;
 151         int vnr;
 152
 153         rcu_read_lock();
 154         idr_for_each_entry(&tconn->volumes, mdev, vnr)
 155                 ds = min_t(enum drbd_disk_state, ds, mdev->state.disk);
 156         rcu_read_unlock();
 157
 158         return ds;
 159 }
 160
 161 enum drbd_disk_state conn_highest_pdsk(struct drbd_tconn *tconn)
 162 {
 163         enum drbd_disk_state ds = D_DISKLESS;
 164         struct drbd_conf *mdev;
 165         int vnr;
 166
 167         rcu_read_lock();
 168         idr_for_each_entry(&tconn->volumes, mdev, vnr)
 169                 ds = max_t(enum drbd_disk_state, ds, mdev->state.pdsk);
 170         rcu_read_unlock();
 171
 172         return ds;
 173 }
 174
 175 enum drbd_conns conn_lowest_conn(struct drbd_tconn *tconn)
 176 {
 177         enum drbd_conns conn = C_MASK;
 178         struct drbd_conf *mdev;
 179         int vnr;
 180
 181         rcu_read_lock();
 182         idr_for_each_entry(&tconn->volumes, mdev, vnr)
 183                 conn = min_t(enum drbd_conns, conn, mdev->state.conn);
 184         rcu_read_unlock();
 185
 186         return conn;
 187 }
 188
 189 static bool no_peer_wf_report_params(struct drbd_tconn *tconn)
 190 {
 191         struct drbd_conf *mdev;
 192         int vnr;
 193         bool rv = true;
 194
 195         rcu_read_lock();
 196         idr_for_each_entry(&tconn->volumes, mdev, vnr)
 197                 if (mdev->state.conn == C_WF_REPORT_PARAMS) {
 198                         rv = false;
 199                         break;
 200                 }
 201         rcu_read_unlock();
 202
 203         return rv;
 204 }
 205
 206
 207 /**
 208  * cl_wide_st_chg() - true if the state change is a cluster wide one
 209  * @mdev:       DRBD device.
 210  * @os:         old (current) state.
 211  * @ns:         new (wanted) state.
 212  */
 213 static int cl_wide_st_chg(struct drbd_conf *mdev,
 214                           union drbd_state os, union drbd_state ns)
 215 {
 216         return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED &&
 217                  ((os.role != R_PRIMARY && ns.role == R_PRIMARY) ||
 218                   (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
 219                   (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) ||
 220                   (os.disk != D_FAILED && ns.disk == D_FAILED))) ||
 221                 (os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) ||
 222                 (os.conn == C_CONNECTED && ns.conn == C_VERIFY_S) ||
 223                 (os.conn == C_CONNECTED && ns.conn == C_WF_REPORT_PARAMS);
 224 }
 225
 226 static union drbd_state
 227 apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val)
 228 {
 229         union drbd_state ns;
 230         ns.i = (os.i & ~mask.i) | val.i;
 231         return ns;
 232 }
 233
 234 enum drbd_state_rv
 235 drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f,
 236                   union drbd_state mask, union drbd_state val)
 237 {
 238         unsigned long flags;
 239         union drbd_state ns;
 240         enum drbd_state_rv rv;
 241
 242         spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 243         ns = apply_mask_val(drbd_read_state(mdev), mask, val);
 244         rv = _drbd_set_state(mdev, ns, f, NULL);
 245         spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 246
 247         return rv;
 248 }
 249
 250 /**
 251  * drbd_force_state() - Impose a change which happens outside our control on our state
 252  * @mdev:       DRBD device.
 253  * @mask:       mask of state bits to change.
 254  * @val:        value of new state bits.
 255  */
 256 void drbd_force_state(struct drbd_conf *mdev,
 257         union drbd_state mask, union drbd_state val)
 258 {
 259         drbd_change_state(mdev, CS_HARD, mask, val);
 260 }
 261
 262 static enum drbd_state_rv
 263 _req_st_cond(struct drbd_conf *mdev, union drbd_state mask,
 264              union drbd_state val)
 265 {
 266         union drbd_state os, ns;
 267         unsigned long flags;
 268         enum drbd_state_rv rv;
 269
 270         if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &mdev->flags))
 271                 return SS_CW_SUCCESS;
 272
 273         if (test_and_clear_bit(CL_ST_CHG_FAIL, &mdev->flags))
 274                 return SS_CW_FAILED_BY_PEER;
 275
 276         spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 277         os = drbd_read_state(mdev);
 278         ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
 279         rv = is_valid_transition(os, ns);
 280         if (rv >= SS_SUCCESS)
 281                 rv = SS_UNKNOWN_ERROR;  /* cont waiting, otherwise fail. */
 282
 283         if (!cl_wide_st_chg(mdev, os, ns))
 284                 rv = SS_CW_NO_NEED;
 285         if (rv == SS_UNKNOWN_ERROR) {
 286                 rv = is_valid_state(mdev, ns);
 287                 if (rv >= SS_SUCCESS) {
 288                         rv = is_valid_soft_transition(os, ns, mdev->tconn);
 289                         if (rv >= SS_SUCCESS)
 290                                 rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
 291                 }
 292         }
 293         spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 294
 295         return rv;
 296 }
 297
 298 /**
 299  * drbd_req_state() - Perform an eventually cluster wide state change
 300  * @mdev:       DRBD device.
 301  * @mask:       mask of state bits to change.
 302  * @val:        value of new state bits.
 303  * @f:          flags
 304  *
 305  * Should not be called directly, use drbd_request_state() or
 306  * _drbd_request_state().
 307  */
 308 static enum drbd_state_rv
 309 drbd_req_state(struct drbd_conf *mdev, union drbd_state mask,
 310                union drbd_state val, enum chg_state_flags f)
 311 {
 312         struct completion done;
 313         unsigned long flags;
 314         union drbd_state os, ns;
 315         enum drbd_state_rv rv;
 316
 317         init_completion(&done);
 318
 319         if (f & CS_SERIALIZE)
 320                 mutex_lock(mdev->state_mutex);
 321
 322         spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 323         os = drbd_read_state(mdev);
 324         ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
 325         rv = is_valid_transition(os, ns);
 326         if (rv < SS_SUCCESS) {
 327                 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 328                 goto abort;
 329         }
 330
 331         if (cl_wide_st_chg(mdev, os, ns)) {
 332                 rv = is_valid_state(mdev, ns);
 333                 if (rv == SS_SUCCESS)
 334                         rv = is_valid_soft_transition(os, ns, mdev->tconn);
 335                 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 336
 337                 if (rv < SS_SUCCESS) {
 338                         if (f & CS_VERBOSE)
 339                                 print_st_err(mdev, os, ns, rv);
 340                         goto abort;
 341                 }
 342
 343                 if (drbd_send_state_req(mdev, mask, val)) {
 344                         rv = SS_CW_FAILED_BY_PEER;
 345                         if (f & CS_VERBOSE)
 346                                 print_st_err(mdev, os, ns, rv);
 347                         goto abort;
 348                 }
 349
 350                 wait_event(mdev->state_wait,
 351                         (rv = _req_st_cond(mdev, mask, val)));
 352
 353                 if (rv < SS_SUCCESS) {
 354                         if (f & CS_VERBOSE)
 355                                 print_st_err(mdev, os, ns, rv);
 356                         goto abort;
 357                 }
 358                 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
 359                 ns = apply_mask_val(drbd_read_state(mdev), mask, val);
 360                 rv = _drbd_set_state(mdev, ns, f, &done);
 361         } else {
 362                 rv = _drbd_set_state(mdev, ns, f, &done);
 363         }
 364
 365         spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
 366
 367         if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) {
 368                 D_ASSERT(current != mdev->tconn->worker.task);
 369                 wait_for_completion(&done);
 370         }
 371
 372 abort:
 373         if (f & CS_SERIALIZE)
 374                 mutex_unlock(mdev->state_mutex);
 375
 376         return rv;
 377 }
 378
 379 /**
 380  * _drbd_request_state() - Request a state change (with flags)
 381  * @mdev:       DRBD device.
 382  * @mask:       mask of state bits to change.
 383  * @val:        value of new state bits.
 384  * @f:          flags
 385  *
 386  * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE
 387  * flag, or when logging of failed state change requests is not desired.
 388  */
 389 enum drbd_state_rv
 390 _drbd_request_state(struct drbd_conf *mdev, union drbd_state mask,
 391                     union drbd_state val, enum chg_state_flags f)
 392 {
 393         enum drbd_state_rv rv;
 394
 395         wait_event(mdev->state_wait,
 396                    (rv = drbd_req_state(mdev, mask, val, f)) != SS_IN_TRANSIENT_STATE);
 397
 398         return rv;
 399 }
 400
 401 static void print_st(struct drbd_conf *mdev, char *name, union drbd_state ns)
 402 {
 403         dev_err(DEV, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n",
 404             name,
 405             drbd_conn_str(ns.conn),
 406             drbd_role_str(ns.role),
 407             drbd_role_str(ns.peer),
 408             drbd_disk_str(ns.disk),
 409             drbd_disk_str(ns.pdsk),
 410             is_susp(ns) ? 's' : 'r',
 411             ns.aftr_isp ? 'a' : '-',
 412             ns.peer_isp ? 'p' : '-',
 413             ns.user_isp ? 'u' : '-',
 414             ns.susp_fen ? 'F' : '-',
 415             ns.susp_nod ? 'N' : '-'
 416             );
 417 }
 418
 419 void print_st_err(struct drbd_conf *mdev, union drbd_state os,
 420                   union drbd_state ns, enum drbd_state_rv err)
 421 {
 422         if (err == SS_IN_TRANSIENT_STATE)
 423                 return;
 424         dev_err(DEV, "State change failed: %s\n", drbd_set_st_err_str(err));
 425         print_st(mdev, " state", os);
 426         print_st(mdev, "wanted", ns);
 427 }
 428
 429 static long print_state_change(char *pb, union drbd_state os, union drbd_state ns,
 430                                enum chg_state_flags flags)
 431 {
 432         char *pbp;
 433         pbp = pb;
 434         *pbp = 0;
 435
 436         if (ns.role != os.role && flags & CS_DC_ROLE)
 437                 pbp += sprintf(pbp, "role( %s -> %s ) ",
 438                                drbd_role_str(os.role),
 439                                drbd_role_str(ns.role));
 440         if (ns.peer != os.peer && flags & CS_DC_PEER)
 441                 pbp += sprintf(pbp, "peer( %s -> %s ) ",
 442                                drbd_role_str(os.peer),
 443                                drbd_role_str(ns.peer));
 444         if (ns.conn != os.conn && flags & CS_DC_CONN)
 445                 pbp += sprintf(pbp, "conn( %s -> %s ) ",
 446                                drbd_conn_str(os.conn),
 447                                drbd_conn_str(ns.conn));
 448         if (ns.disk != os.disk && flags & CS_DC_DISK)
 449                 pbp += sprintf(pbp, "disk( %s -> %s ) ",
 450                                drbd_disk_str(os.disk),
 451                                drbd_disk_str(ns.disk));
 452         if (ns.pdsk != os.pdsk && flags & CS_DC_PDSK)
 453                 pbp += sprintf(pbp, "pdsk( %s -> %s ) ",
 454                                drbd_disk_str(os.pdsk),
 455                                drbd_disk_str(ns.pdsk));
 456
 457         return pbp - pb;
 458 }
 459
 460 static void drbd_pr_state_change(struct drbd_conf *mdev, union drbd_state os, union drbd_state ns,
 461                                  enum chg_state_flags flags)
 462 {
 463         char pb[300];
 464         char *pbp = pb;
 465
 466         pbp += print_state_change(pbp, os, ns, flags ^ CS_DC_MASK);
 467
 468         if (ns.aftr_isp != os.aftr_isp)
 469                 pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ",
 470                                os.aftr_isp,
 471                                ns.aftr_isp);
 472         if (ns.peer_isp != os.peer_isp)
 473                 pbp += sprintf(pbp, "peer_isp( %d -> %d ) ",
 474                                os.peer_isp,
 475                                ns.peer_isp);
 476         if (ns.user_isp != os.user_isp)
 477                 pbp += sprintf(pbp, "user_isp( %d -> %d ) ",
 478                                os.user_isp,
 479                                ns.user_isp);
 480
 481         if (pbp != pb)
 482                 dev_info(DEV, "%s\n", pb);
 483 }
 484
 485 static void conn_pr_state_change(struct drbd_tconn *tconn, union drbd_state os, union drbd_state ns,
 486                                  enum chg_state_flags flags)
 487 {
 488         char pb[300];
 489         char *pbp = pb;
 490
 491         pbp += print_state_change(pbp, os, ns, flags);
 492
 493         if (is_susp(ns) != is_susp(os) && flags & CS_DC_SUSP)
 494                 pbp += sprintf(pbp, "susp( %d -> %d ) ",
 495                                is_susp(os),
 496                                is_susp(ns));
 497
 498         if (pbp != pb)
 499                 conn_info(tconn, "%s\n", pb);
 500 }
 501
 502
 503 /**
 504  * is_valid_state() - Returns an SS_ error code if ns is not valid
 505  * @mdev:       DRBD device.
 506  * @ns:         State to consider.
 507  */
 508 static enum drbd_state_rv
 509 is_valid_state(struct drbd_conf *mdev, union drbd_state ns)
 510 {
 511         /* See drbd_state_sw_errors in drbd_strings.c */
 512
 513         enum drbd_fencing_p fp;
 514         enum drbd_state_rv rv = SS_SUCCESS;
 515         struct net_conf *nc;
 516
 517         rcu_read_lock();
 518         fp = FP_DONT_CARE;
 519         if (get_ldev(mdev)) {
 520                 fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
 521                 put_ldev(mdev);
 522         }
 523
 524         nc = rcu_dereference(mdev->tconn->net_conf);
 525         if (nc) {
 526                 if (!nc->two_primaries && ns.role == R_PRIMARY) {
 527                         if (ns.peer == R_PRIMARY)
 528                                 rv = SS_TWO_PRIMARIES;
 529                         else if (conn_highest_peer(mdev->tconn) == R_PRIMARY)
 530                                 rv = SS_O_VOL_PEER_PRI;
 531                 }
 532         }
 533
 534         if (rv <= 0)
 535                 /* already found a reason to abort */;
 536         else if (ns.role == R_SECONDARY && mdev->open_cnt)
 537                 rv = SS_DEVICE_IN_USE;
 538
 539         else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE)
 540                 rv = SS_NO_UP_TO_DATE_DISK;
 541
 542         else if (fp >= FP_RESOURCE &&
 543                  ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN)
 544                 rv = SS_PRIMARY_NOP;
 545
 546         else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT)
 547                 rv = SS_NO_UP_TO_DATE_DISK;
 548
 549         else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT)
 550                 rv = SS_NO_LOCAL_DISK;
 551
 552         else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT)
 553                 rv = SS_NO_REMOTE_DISK;
 554
 555         else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
 556                 rv = SS_NO_UP_TO_DATE_DISK;
 557
 558         else if ((ns.conn == C_CONNECTED ||
 559                   ns.conn == C_WF_BITMAP_S ||
 560                   ns.conn == C_SYNC_SOURCE ||
 561                   ns.conn == C_PAUSED_SYNC_S) &&
 562                   ns.disk == D_OUTDATED)
 563                 rv = SS_CONNECTED_OUTDATES;
 564
 565         else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 566                  (nc->verify_alg[0] == 0))
 567                 rv = SS_NO_VERIFY_ALG;
 568
 569         else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 570                   mdev->tconn->agreed_pro_version < 88)
 571                 rv = SS_NOT_SUPPORTED;
 572
 573         else if (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
 574                 rv = SS_NO_UP_TO_DATE_DISK;
 575
 576         else if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
 577                  ns.pdsk == D_UNKNOWN)
 578                 rv = SS_NEED_CONNECTION;
 579
 580         else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN)
 581                 rv = SS_CONNECTED_OUTDATES;
 582
 583         rcu_read_unlock();
 584
 585         return rv;
 586 }
 587
 588 /**
 589  * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible
 590  * This function limits state transitions that may be declined by DRBD. I.e.
 591  * user requests (aka soft transitions).
 592  * @mdev:       DRBD device.
 593  * @ns:         new state.
 594  * @os:         old state.
 595  */
 596 static enum drbd_state_rv
 597 is_valid_soft_transition(union drbd_state os, union drbd_state ns, struct drbd_tconn *tconn)
 598 {
 599         enum drbd_state_rv rv = SS_SUCCESS;
 600
 601         if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
 602             os.conn > C_CONNECTED)
 603                 rv = SS_RESYNC_RUNNING;
 604
 605         if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
 606                 rv = SS_ALREADY_STANDALONE;
 607
 608         if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
 609                 rv = SS_IS_DISKLESS;
 610
 611         if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
 612                 rv = SS_NO_NET_CONFIG;
 613
 614         if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
 615                 rv = SS_LOWER_THAN_OUTDATED;
 616
 617         if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
 618                 rv = SS_IN_TRANSIENT_STATE;
 619
 620         /* if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS)
 621            rv = SS_IN_TRANSIENT_STATE; */
 622
 623         /* While establishing a connection only allow cstate to change.
 624            Delay/refuse role changes, detach attach etc... */
 625         if (test_bit(STATE_SENT, &tconn->flags) &&
 626             !(os.conn == C_WF_REPORT_PARAMS ||
 627               (ns.conn == C_WF_REPORT_PARAMS && os.conn == C_WF_CONNECTION)))
 628                 rv = SS_IN_TRANSIENT_STATE;
 629
 630         if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
 631                 rv = SS_NEED_CONNECTION;
 632
 633         if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 634             ns.conn != os.conn && os.conn > C_CONNECTED)
 635                 rv = SS_RESYNC_RUNNING;
 636
 637         if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
 638             os.conn < C_CONNECTED)
 639                 rv = SS_NEED_CONNECTION;
 640
 641         if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)
 642             && os.conn < C_WF_REPORT_PARAMS)
 643                 rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */
 644
 645         if (ns.conn == C_DISCONNECTING && ns.pdsk == D_OUTDATED &&
 646             os.conn < C_CONNECTED && os.pdsk > D_OUTDATED)
 647                 rv = SS_OUTDATE_WO_CONN;
 648
 649         return rv;
 650 }
 651
 652 static enum drbd_state_rv
 653 is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc)
 654 {
 655         /* no change -> nothing to do, at least for the connection part */
 656         if (oc == nc)
 657                 return SS_NOTHING_TO_DO;
 658
 659         /* disconnect of an unconfigured connection does not make sense */
 660         if (oc == C_STANDALONE && nc == C_DISCONNECTING)
 661                 return SS_ALREADY_STANDALONE;
 662
 663         /* from C_STANDALONE, we start with C_UNCONNECTED */
 664         if (oc == C_STANDALONE && nc != C_UNCONNECTED)
 665                 return SS_NEED_CONNECTION;
 666
 667         /* When establishing a connection we need to go through WF_REPORT_PARAMS!
 668            Necessary to do the right thing upon invalidate-remote on a disconnected resource */
 669         if (oc < C_WF_REPORT_PARAMS && nc >= C_CONNECTED)
 670                 return SS_NEED_CONNECTION;
 671
 672         /* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */
 673         if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING)
 674                 return SS_IN_TRANSIENT_STATE;
 675
 676         /* After C_DISCONNECTING only C_STANDALONE may follow */
 677         if (oc == C_DISCONNECTING && nc != C_STANDALONE)
 678                 return SS_IN_TRANSIENT_STATE;
 679
 680         return SS_SUCCESS;
 681 }
 682
 683
 684 /**
 685  * is_valid_transition() - Returns an SS_ error code if the state transition is not possible
 686  * This limits hard state transitions. Hard state transitions are facts there are
 687  * imposed on DRBD by the environment. E.g. disk broke or network broke down.
 688  * But those hard state transitions are still not allowed to do everything.
 689  * @ns:         new state.
 690  * @os:         old state.
 691  */
 692 static enum drbd_state_rv
 693 is_valid_transition(union drbd_state os, union drbd_state ns)
 694 {
 695         enum drbd_state_rv rv;
 696
 697         rv = is_valid_conn_transition(os.conn, ns.conn);
 698
 699         /* we cannot fail (again) if we already detached */
 700         if (ns.disk == D_FAILED && os.disk == D_DISKLESS)
 701                 rv = SS_IS_DISKLESS;
 702
 703         return rv;
 704 }
 705
 706 static void print_sanitize_warnings(struct drbd_conf *mdev, enum sanitize_state_warnings warn)
 707 {
 708         static const char *msg_table[] = {
 709                 [NO_WARNING] = "",
 710                 [ABORTED_ONLINE_VERIFY] = "Online-verify aborted.",
 711                 [ABORTED_RESYNC] = "Resync aborted.",
 712                 [CONNECTION_LOST_NEGOTIATING] = "Connection lost while negotiating, no data!",
 713                 [IMPLICITLY_UPGRADED_DISK] = "Implicitly upgraded disk",
 714                 [IMPLICITLY_UPGRADED_PDSK] = "Implicitly upgraded pdsk",
 715         };
 716
 717         if (warn != NO_WARNING)
 718                 dev_warn(DEV, "%s\n", msg_table[warn]);
 719 }
 720
 721 /**
 722  * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
 723  * @mdev:       DRBD device.
 724  * @os:         old state.
 725  * @ns:         new state.
 726  * @warn_sync_abort:
 727  *
 728  * When we loose connection, we have to set the state of the peers disk (pdsk)
 729  * to D_UNKNOWN. This rule and many more along those lines are in this function.
 730  */
 731 static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns,
 732                                        enum sanitize_state_warnings *warn)
 733 {
 734         enum drbd_fencing_p fp;
 735         enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max;
 736
 737         if (warn)
 738                 *warn = NO_WARNING;
 739
 740         fp = FP_DONT_CARE;
 741         if (get_ldev(mdev)) {
 742                 rcu_read_lock();
 743                 fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
 744                 rcu_read_unlock();
 745                 put_ldev(mdev);
 746         }
 747
 748         /* Implications from connection to peer and peer_isp */
 749         if (ns.conn < C_CONNECTED) {
 750                 ns.peer_isp = 0;
 751                 ns.peer = R_UNKNOWN;
 752                 if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
 753                         ns.pdsk = D_UNKNOWN;
 754         }
 755
 756         /* Clear the aftr_isp when becoming unconfigured */
 757         if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
 758                 ns.aftr_isp = 0;
 759
 760         /* An implication of the disk states onto the connection state */
 761         /* Abort resync if a disk fails/detaches */
 762         if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
 763                 if (warn)
 764                         *warn = ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ?
 765                                 ABORTED_ONLINE_VERIFY : ABORTED_RESYNC;
 766                 ns.conn = C_CONNECTED;
 767         }
 768
 769         /* Connection breaks down before we finished "Negotiating" */
 770         if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
 771             get_ldev_if_state(mdev, D_NEGOTIATING)) {
 772                 if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) {
 773                         ns.disk = mdev->new_state_tmp.disk;
 774                         ns.pdsk = mdev->new_state_tmp.pdsk;
 775                 } else {
 776                         if (warn)
 777                                 *warn = CONNECTION_LOST_NEGOTIATING;
 778                         ns.disk = D_DISKLESS;
 779                         ns.pdsk = D_UNKNOWN;
 780                 }
 781                 put_ldev(mdev);
 782         }
 783
 784         /* D_CONSISTENT and D_OUTDATED vanish when we get connected */
 785         if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) {
 786                 if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED)
 787                         ns.disk = D_UP_TO_DATE;
 788                 if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)
 789                         ns.pdsk = D_UP_TO_DATE;
 790         }
 791
 792         /* Implications of the connection stat on the disk states */
 793         disk_min = D_DISKLESS;
 794         disk_max = D_UP_TO_DATE;
 795         pdsk_min = D_INCONSISTENT;
 796         pdsk_max = D_UNKNOWN;
 797         switch ((enum drbd_conns)ns.conn) {
 798         case C_WF_BITMAP_T:
 799         case C_PAUSED_SYNC_T:
 800         case C_STARTING_SYNC_T:
 801         case C_WF_SYNC_UUID:
 802         case C_BEHIND:
 803                 disk_min = D_INCONSISTENT;
 804                 disk_max = D_OUTDATED;
 805                 pdsk_min = D_UP_TO_DATE;
 806                 pdsk_max = D_UP_TO_DATE;
 807                 break;
 808         case C_VERIFY_S:
 809         case C_VERIFY_T:
 810                 disk_min = D_UP_TO_DATE;
 811                 disk_max = D_UP_TO_DATE;
 812                 pdsk_min = D_UP_TO_DATE;
 813                 pdsk_max = D_UP_TO_DATE;
 814                 break;
 815         case C_CONNECTED:
 816                 disk_min = D_DISKLESS;
 817                 disk_max = D_UP_TO_DATE;
 818                 pdsk_min = D_DISKLESS;
 819                 pdsk_max = D_UP_TO_DATE;
 820                 break;
 821         case C_WF_BITMAP_S:
 822         case C_PAUSED_SYNC_S:
 823         case C_STARTING_SYNC_S:
 824         case C_AHEAD:
 825                 disk_min = D_UP_TO_DATE;
 826                 disk_max = D_UP_TO_DATE;
 827                 pdsk_min = D_INCONSISTENT;
 828                 pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/
 829                 break;
 830         case C_SYNC_TARGET:
 831                 disk_min = D_INCONSISTENT;
 832                 disk_max = D_INCONSISTENT;
 833                 pdsk_min = D_UP_TO_DATE;
 834                 pdsk_max = D_UP_TO_DATE;
 835                 break;
 836         case C_SYNC_SOURCE:
 837                 disk_min = D_UP_TO_DATE;
 838                 disk_max = D_UP_TO_DATE;
 839                 pdsk_min = D_INCONSISTENT;
 840                 pdsk_max = D_INCONSISTENT;
 841                 break;
 842         case C_STANDALONE:
 843         case C_DISCONNECTING:
 844         case C_UNCONNECTED:
 845         case C_TIMEOUT:
 846         case C_BROKEN_PIPE:
 847         case C_NETWORK_FAILURE:
 848         case C_PROTOCOL_ERROR:
 849         case C_TEAR_DOWN:
 850         case C_WF_CONNECTION:
 851         case C_WF_REPORT_PARAMS:
 852         case C_MASK:
 853                 break;
 854         }
 855         if (ns.disk > disk_max)
 856                 ns.disk = disk_max;
 857
 858         if (ns.disk < disk_min) {
 859                 if (warn)
 860                         *warn = IMPLICITLY_UPGRADED_DISK;
 861                 ns.disk = disk_min;
 862         }
 863         if (ns.pdsk > pdsk_max)
 864                 ns.pdsk = pdsk_max;
 865
 866         if (ns.pdsk < pdsk_min) {
 867                 if (warn)
 868                         *warn = IMPLICITLY_UPGRADED_PDSK;
 869                 ns.pdsk = pdsk_min;
 870         }
 871
 872         if (fp == FP_STONITH &&
 873             (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED))
 874                 ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */
 875
 876         if (mdev->tconn->res_opts.on_no_data == OND_SUSPEND_IO &&
 877             (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
 878                 ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */
 879
 880         if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
 881                 if (ns.conn == C_SYNC_SOURCE)
 882                         ns.conn = C_PAUSED_SYNC_S;
 883                 if (ns.conn == C_SYNC_TARGET)
 884                         ns.conn = C_PAUSED_SYNC_T;
 885         } else {
 886                 if (ns.conn == C_PAUSED_SYNC_S)
 887                         ns.conn = C_SYNC_SOURCE;
 888                 if (ns.conn == C_PAUSED_SYNC_T)
 889                         ns.conn = C_SYNC_TARGET;
 890         }
 891
 892         return ns;
 893 }
 894
 895 void drbd_resume_al(struct drbd_conf *mdev)
 896 {
 897         if (test_and_clear_bit(AL_SUSPENDED, &mdev->flags))
 898                 dev_info(DEV, "Resumed AL updates\n");
 899 }
 900
 901 /* helper for __drbd_set_state */
 902 static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs)
 903 {
 904         if (mdev->tconn->agreed_pro_version < 90)
 905                 mdev->ov_start_sector = 0;
 906         mdev->rs_total = drbd_bm_bits(mdev);
 907         mdev->ov_position = 0;
 908         if (cs == C_VERIFY_T) {
 909                 /* starting online verify from an arbitrary position
 910                  * does not fit well into the existing protocol.
 911                  * on C_VERIFY_T, we initialize ov_left and friends
 912                  * implicitly in receive_DataRequest once the
 913                  * first P_OV_REQUEST is received */
 914                 mdev->ov_start_sector = ~(sector_t)0;
 915         } else {
 916                 unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector);
 917                 if (bit >= mdev->rs_total) {
 918                         mdev->ov_start_sector =
 919                                 BM_BIT_TO_SECT(mdev->rs_total - 1);
 920                         mdev->rs_total = 1;
 921                 } else
 922                         mdev->rs_total -= bit;
 923                 mdev->ov_position = mdev->ov_start_sector;
 924         }
 925         mdev->ov_left = mdev->rs_total;
 926 }
 927
 928 /**
 929  * __drbd_set_state() - Set a new DRBD state
 930  * @mdev:       DRBD device.
 931  * @ns:         new state.
 932  * @flags:      Flags
 933  * @done:       Optional completion, that will get completed after the after_state_ch() finished
 934  *
 935  * Caller needs to hold req_lock, and global_state_lock. Do not call directly.
 936  */
 937 enum drbd_state_rv
 938 __drbd_set_state(struct drbd_conf *mdev, union drbd_state ns,
 939                  enum chg_state_flags flags, struct completion *done)
 940 {
 941         union drbd_state os;
 942         enum drbd_state_rv rv = SS_SUCCESS;
 943         enum sanitize_state_warnings ssw;
 944         struct after_state_chg_work *ascw;
 945         bool did_remote, should_do_remote;
 946
 947         os = drbd_read_state(mdev);
 948
 949         ns = sanitize_state(mdev, ns, &ssw);
 950         if (ns.i == os.i)
 951                 return SS_NOTHING_TO_DO;
 952
 953         rv = is_valid_transition(os, ns);
 954         if (rv < SS_SUCCESS)
 955                 return rv;
 956
 957         if (!(flags & CS_HARD)) {
 958                 /*  pre-state-change checks ; only look at ns  */
 959                 /* See drbd_state_sw_errors in drbd_strings.c */
 960
 961                 rv = is_valid_state(mdev, ns);
 962                 if (rv < SS_SUCCESS) {
 963                         /* If the old state was illegal as well, then let
 964                            this happen...*/
 965
 966                         if (is_valid_state(mdev, os) == rv)
 967                                 rv = is_valid_soft_transition(os, ns, mdev->tconn);
 968                 } else
 969                         rv = is_valid_soft_transition(os, ns, mdev->tconn);
 970         }
 971
 972         if (rv < SS_SUCCESS) {
 973                 if (flags & CS_VERBOSE)
 974                         print_st_err(mdev, os, ns, rv);
 975                 return rv;
 976         }
 977
 978         print_sanitize_warnings(mdev, ssw);
 979
 980         drbd_pr_state_change(mdev, os, ns, flags);
 981
 982         /* Display changes to the susp* flags that where caused by the call to
 983            sanitize_state(). Only display it here if we where not called from
 984            _conn_request_state() */
 985         if (!(flags & CS_DC_SUSP))
 986                 conn_pr_state_change(mdev->tconn, os, ns, (flags & ~CS_DC_MASK) | CS_DC_SUSP);
 987
 988         /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
 989          * on the ldev here, to be sure the transition -> D_DISKLESS resp.
 990          * drbd_ldev_destroy() won't happen before our corresponding
 991          * after_state_ch works run, where we put_ldev again. */
 992         if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
 993             (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
 994                 atomic_inc(&mdev->local_cnt);
 995
 996         did_remote = drbd_should_do_remote(mdev->state);
 997         mdev->state.i = ns.i;
 998         should_do_remote = drbd_should_do_remote(mdev->state);
 999         mdev->tconn->susp = ns.susp;
1000         mdev->tconn->susp_nod = ns.susp_nod;
1001         mdev->tconn->susp_fen = ns.susp_fen;
1002
1003         /* put replicated vs not-replicated requests in seperate epochs */
1004         if (did_remote != should_do_remote)
1005                 start_new_tl_epoch(mdev->tconn);
1006
1007         if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
1008                 drbd_print_uuids(mdev, "attached to UUIDs");
1009
1010         /* Wake up role changes, that were delayed because of connection establishing */
1011         if (os.conn == C_WF_REPORT_PARAMS && ns.conn != C_WF_REPORT_PARAMS &&
1012             no_peer_wf_report_params(mdev->tconn))
1013                 clear_bit(STATE_SENT, &mdev->tconn->flags);
1014
1015         wake_up(&mdev->misc_wait);
1016         wake_up(&mdev->state_wait);
1017         wake_up(&mdev->tconn->ping_wait);
1018
1019         /* Aborted verify run, or we reached the stop sector.
1020          * Log the last position, unless end-of-device. */
1021         if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
1022             ns.conn <= C_CONNECTED) {
1023                 mdev->ov_start_sector =
1024                         BM_BIT_TO_SECT(drbd_bm_bits(mdev) - mdev->ov_left);
1025                 if (mdev->ov_left)
1026                         dev_info(DEV, "Online Verify reached sector %llu\n",
1027                                 (unsigned long long)mdev->ov_start_sector);
1028         }
1029
1030         if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
1031             (ns.conn == C_SYNC_TARGET  || ns.conn == C_SYNC_SOURCE)) {
1032                 dev_info(DEV, "Syncer continues.\n");
1033                 mdev->rs_paused += (long)jiffies
1034                                   -(long)mdev->rs_mark_time[mdev->rs_last_mark];
1035                 if (ns.conn == C_SYNC_TARGET)
1036                         mod_timer(&mdev->resync_timer, jiffies);
1037         }
1038
1039         if ((os.conn == C_SYNC_TARGET  || os.conn == C_SYNC_SOURCE) &&
1040             (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
1041                 dev_info(DEV, "Resync suspended\n");
1042                 mdev->rs_mark_time[mdev->rs_last_mark] = jiffies;
1043         }
1044
1045         if (os.conn == C_CONNECTED &&
1046             (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
1047                 unsigned long now = jiffies;
1048                 int i;
1049
1050                 set_ov_position(mdev, ns.conn);
1051                 mdev->rs_start = now;
1052                 mdev->rs_last_events = 0;
1053                 mdev->rs_last_sect_ev = 0;
1054                 mdev->ov_last_oos_size = 0;
1055                 mdev->ov_last_oos_start = 0;
1056
1057                 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1058                         mdev->rs_mark_left[i] = mdev->ov_left;
1059                         mdev->rs_mark_time[i] = now;
1060                 }
1061
1062                 drbd_rs_controller_reset(mdev);
1063
1064                 if (ns.conn == C_VERIFY_S) {
1065                         dev_info(DEV, "Starting Online Verify from sector %llu\n",
1066                                         (unsigned long long)mdev->ov_position);
1067                         mod_timer(&mdev->resync_timer, jiffies);
1068                 }
1069         }
1070
1071         if (get_ldev(mdev)) {
1072                 u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
1073                                                  MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
1074                                                  MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
1075
1076                 mdf &= ~MDF_AL_CLEAN;
1077                 if (test_bit(CRASHED_PRIMARY, &mdev->flags))
1078                         mdf |= MDF_CRASHED_PRIMARY;
1079                 if (mdev->state.role == R_PRIMARY ||
1080                     (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY))
1081                         mdf |= MDF_PRIMARY_IND;
1082                 if (mdev->state.conn > C_WF_REPORT_PARAMS)
1083                         mdf |= MDF_CONNECTED_IND;
1084                 if (mdev->state.disk > D_INCONSISTENT)
1085                         mdf |= MDF_CONSISTENT;
1086                 if (mdev->state.disk > D_OUTDATED)
1087                         mdf |= MDF_WAS_UP_TO_DATE;
1088                 if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT)
1089                         mdf |= MDF_PEER_OUT_DATED;
1090                 if (mdf != mdev->ldev->md.flags) {
1091                         mdev->ldev->md.flags = mdf;
1092                         drbd_md_mark_dirty(mdev);
1093                 }
1094                 if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
1095                         drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]);
1096                 put_ldev(mdev);
1097         }
1098
1099         /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
1100         if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
1101             os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
1102                 set_bit(CONSIDER_RESYNC, &mdev->flags);
1103
1104         /* Receiver should clean up itself */
1105         if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
1106                 drbd_thread_stop_nowait(&mdev->tconn->receiver);
1107
1108         /* Now the receiver finished cleaning up itself, it should die */
1109         if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
1110                 drbd_thread_stop_nowait(&mdev->tconn->receiver);
1111
1112         /* Upon network failure, we need to restart the receiver. */
1113         if (os.conn > C_WF_CONNECTION &&
1114             ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
1115                 drbd_thread_restart_nowait(&mdev->tconn->receiver);
1116
1117         /* Resume AL writing if we get a connection */
1118         if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
1119                 drbd_resume_al(mdev);
1120
1121         /* remember last attach time so request_timer_fn() won't
1122          * kill newly established sessions while we are still trying to thaw
1123          * previously frozen IO */
1124         if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1125             ns.disk > D_NEGOTIATING)
1126                 mdev->last_reattach_jif = jiffies;
1127
1128         ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
1129         if (ascw) {
1130                 ascw->os = os;
1131                 ascw->ns = ns;
1132                 ascw->flags = flags;
1133                 ascw->w.cb = w_after_state_ch;
1134                 ascw->w.mdev = mdev;
1135                 ascw->done = done;
1136                 drbd_queue_work(&mdev->tconn->sender_work, &ascw->w);
1137         } else {
1138                 dev_err(DEV, "Could not kmalloc an ascw\n");
1139         }
1140
1141         return rv;
1142 }
1143
1144 static int w_after_state_ch(struct drbd_work *w, int unused)
1145 {
1146         struct after_state_chg_work *ascw =
1147                 container_of(w, struct after_state_chg_work, w);
1148         struct drbd_conf *mdev = w->mdev;
1149
1150         after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags);
1151         if (ascw->flags & CS_WAIT_COMPLETE) {
1152                 D_ASSERT(ascw->done != NULL);
1153                 complete(ascw->done);
1154         }
1155         kfree(ascw);
1156
1157         return 0;
1158 }
1159
1160 static void abw_start_sync(struct drbd_conf *mdev, int rv)
1161 {
1162         if (rv) {
1163                 dev_err(DEV, "Writing the bitmap failed not starting resync.\n");
1164                 _drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE);
1165                 return;
1166         }
1167
1168         switch (mdev->state.conn) {
1169         case C_STARTING_SYNC_T:
1170                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
1171                 break;
1172         case C_STARTING_SYNC_S:
1173                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1174                 break;
1175         }
1176 }
1177
1178 int drbd_bitmap_io_from_worker(struct drbd_conf *mdev,
1179                 int (*io_fn)(struct drbd_conf *),
1180                 char *why, enum bm_flag flags)
1181 {
1182         int rv;
1183
1184         D_ASSERT(current == mdev->tconn->worker.task);
1185
1186         /* open coded non-blocking drbd_suspend_io(mdev); */
1187         set_bit(SUSPEND_IO, &mdev->flags);
1188
1189         drbd_bm_lock(mdev, why, flags);
1190         rv = io_fn(mdev);
1191         drbd_bm_unlock(mdev);
1192
1193         drbd_resume_io(mdev);
1194
1195         return rv;
1196 }
1197
1198 /**
1199  * after_state_ch() - Perform after state change actions that may sleep
1200  * @mdev:       DRBD device.
1201  * @os:         old state.
1202  * @ns:         new state.
1203  * @flags:      Flags
1204  */
1205 static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
1206                            union drbd_state ns, enum chg_state_flags flags)
1207 {
1208         struct sib_info sib;
1209
1210         sib.sib_reason = SIB_STATE_CHANGE;
1211         sib.os = os;
1212         sib.ns = ns;
1213
1214         if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) {
1215                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1216                 if (mdev->p_uuid)
1217                         mdev->p_uuid[UI_FLAGS] &= ~((u64)2);
1218         }
1219
1220         /* Inform userspace about the change... */
1221         drbd_bcast_event(mdev, &sib);
1222
1223         if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
1224             (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
1225                 drbd_khelper(mdev, "pri-on-incon-degr");
1226
1227         /* Here we have the actions that are performed after a
1228            state change. This function might sleep */
1229
1230         if (ns.susp_nod) {
1231                 struct drbd_tconn *tconn = mdev->tconn;
1232                 enum drbd_req_event what = NOTHING;
1233
1234                 spin_lock_irq(&tconn->req_lock);
1235                 if (os.conn < C_CONNECTED && conn_lowest_conn(tconn) >= C_CONNECTED)
1236                         what = RESEND;
1237
1238                 if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1239                     conn_lowest_disk(tconn) > D_NEGOTIATING)
1240                         what = RESTART_FROZEN_DISK_IO;
1241
1242                 if (tconn->susp_nod && what != NOTHING) {
1243                         _tl_restart(tconn, what);
1244                         _conn_request_state(tconn,
1245                                             (union drbd_state) { { .susp_nod = 1 } },
1246                                             (union drbd_state) { { .susp_nod = 0 } },
1247                                             CS_VERBOSE);
1248                 }
1249                 spin_unlock_irq(&tconn->req_lock);
1250         }
1251
1252         if (ns.susp_fen) {
1253                 struct drbd_tconn *tconn = mdev->tconn;
1254
1255                 spin_lock_irq(&tconn->req_lock);
1256                 if (tconn->susp_fen && conn_lowest_conn(tconn) >= C_CONNECTED) {
1257                         /* case2: The connection was established again: */
1258                         struct drbd_conf *odev;
1259                         int vnr;
1260
1261                         rcu_read_lock();
1262                         idr_for_each_entry(&tconn->volumes, odev, vnr)
1263                                 clear_bit(NEW_CUR_UUID, &odev->flags);
1264                         rcu_read_unlock();
1265                         _tl_restart(tconn, RESEND);
1266                         _conn_request_state(tconn,
1267                                             (union drbd_state) { { .susp_fen = 1 } },
1268                                             (union drbd_state) { { .susp_fen = 0 } },
1269                                             CS_VERBOSE);
1270                 }
1271                 spin_unlock_irq(&tconn->req_lock);
1272         }
1273
1274         /* Became sync source.  With protocol >= 96, we still need to send out
1275          * the sync uuid now. Need to do that before any drbd_send_state, or
1276          * the other side may go "paused sync" before receiving the sync uuids,
1277          * which is unexpected. */
1278         if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
1279             (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
1280             mdev->tconn->agreed_pro_version >= 96 && get_ldev(mdev)) {
1281                 drbd_gen_and_send_sync_uuid(mdev);
1282                 put_ldev(mdev);
1283         }
1284
1285         /* Do not change the order of the if above and the two below... */
1286         if (os.pdsk == D_DISKLESS &&
1287             ns.pdsk > D_DISKLESS && ns.pdsk != D_UNKNOWN) {      /* attach on the peer */
1288                 /* we probably will start a resync soon.
1289                  * make sure those things are properly reset. */
1290                 mdev->rs_total = 0;
1291                 mdev->rs_failed = 0;
1292                 atomic_set(&mdev->rs_pending_cnt, 0);
1293                 drbd_rs_cancel_all(mdev);
1294
1295                 drbd_send_uuids(mdev);
1296                 drbd_send_state(mdev, ns);
1297         }
1298         /* No point in queuing send_bitmap if we don't have a connection
1299          * anymore, so check also the _current_ state, not only the new state
1300          * at the time this work was queued. */
1301         if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
1302             mdev->state.conn == C_WF_BITMAP_S)
1303                 drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL,
1304                                 "send_bitmap (WFBitMapS)",
1305                                 BM_LOCKED_TEST_ALLOWED);
1306
1307         /* Lost contact to peer's copy of the data */
1308         if ((os.pdsk >= D_INCONSISTENT &&
1309              os.pdsk != D_UNKNOWN &&
1310              os.pdsk != D_OUTDATED)
1311         &&  (ns.pdsk < D_INCONSISTENT ||
1312              ns.pdsk == D_UNKNOWN ||
1313              ns.pdsk == D_OUTDATED)) {
1314                 if (get_ldev(mdev)) {
1315                         if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1316                             mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1317                                 if (drbd_suspended(mdev)) {
1318                                         set_bit(NEW_CUR_UUID, &mdev->flags);
1319                                 } else {
1320                                         drbd_uuid_new_current(mdev);
1321                                         drbd_send_uuids(mdev);
1322                                 }
1323                         }
1324                         put_ldev(mdev);
1325                 }
1326         }
1327
1328         if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) {
1329                 if (os.peer == R_SECONDARY && ns.peer == R_PRIMARY &&
1330                     mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1331                         drbd_uuid_new_current(mdev);
1332                         drbd_send_uuids(mdev);
1333                 }
1334                 /* D_DISKLESS Peer becomes secondary */
1335                 if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1336                         /* We may still be Primary ourselves.
1337                          * No harm done if the bitmap still changes,
1338                          * redirtied pages will follow later. */
1339                         drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1340                                 "demote diskless peer", BM_LOCKED_SET_ALLOWED);
1341                 put_ldev(mdev);
1342         }
1343
1344         /* Write out all changed bits on demote.
1345          * Though, no need to da that just yet
1346          * if there is a resync going on still */
1347         if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
1348                 mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) {
1349                 /* No changes to the bitmap expected this time, so assert that,
1350                  * even though no harm was done if it did change. */
1351                 drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1352                                 "demote", BM_LOCKED_TEST_ALLOWED);
1353                 put_ldev(mdev);
1354         }
1355
1356         /* Last part of the attaching process ... */
1357         if (ns.conn >= C_CONNECTED &&
1358             os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1359                 drbd_send_sizes(mdev, 0, 0);  /* to start sync... */
1360                 drbd_send_uuids(mdev);
1361                 drbd_send_state(mdev, ns);
1362         }
1363
1364         /* We want to pause/continue resync, tell peer. */
1365         if (ns.conn >= C_CONNECTED &&
1366              ((os.aftr_isp != ns.aftr_isp) ||
1367               (os.user_isp != ns.user_isp)))
1368                 drbd_send_state(mdev, ns);
1369
1370         /* In case one of the isp bits got set, suspend other devices. */
1371         if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1372             (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1373                 suspend_other_sg(mdev);
1374
1375         /* Make sure the peer gets informed about eventual state
1376            changes (ISP bits) while we were in WFReportParams. */
1377         if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1378                 drbd_send_state(mdev, ns);
1379
1380         if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
1381                 drbd_send_state(mdev, ns);
1382
1383         /* We are in the progress to start a full sync... */
1384         if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1385             (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1386                 /* no other bitmap changes expected during this phase */
1387                 drbd_queue_bitmap_io(mdev,
1388                         &drbd_bmio_set_n_write, &abw_start_sync,
1389                         "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED);
1390
1391         /* first half of local IO error, failure to attach,
1392          * or administrative detach */
1393         if (os.disk != D_FAILED && ns.disk == D_FAILED) {
1394                 enum drbd_io_error_p eh = EP_PASS_ON;
1395                 int was_io_error = 0;
1396                 /* corresponding get_ldev was in __drbd_set_state, to serialize
1397                  * our cleanup here with the transition to D_DISKLESS.
1398                  * But is is still not save to dreference ldev here, since
1399                  * we might come from an failed Attach before ldev was set. */
1400                 if (mdev->ldev) {
1401                         rcu_read_lock();
1402                         eh = rcu_dereference(mdev->ldev->disk_conf)->on_io_error;
1403                         rcu_read_unlock();
1404
1405                         was_io_error = test_and_clear_bit(WAS_IO_ERROR, &mdev->flags);
1406
1407                         if (was_io_error && eh == EP_CALL_HELPER)
1408                                 drbd_khelper(mdev, "local-io-error");
1409
1410                         /* Immediately allow completion of all application IO,
1411                          * that waits for completion from the local disk,
1412                          * if this was a force-detach due to disk_timeout
1413                          * or administrator request (drbdsetup detach --force).
1414                          * Do NOT abort otherwise.
1415                          * Aborting local requests may cause serious problems,
1416                          * if requests are completed to upper layers already,
1417                          * and then later the already submitted local bio completes.
1418                          * This can cause DMA into former bio pages that meanwhile
1419                          * have been re-used for other things.
1420                          * So aborting local requests may cause crashes,
1421                          * or even worse, silent data corruption.
1422                          */
1423                         if (test_and_clear_bit(FORCE_DETACH, &mdev->flags))
1424                                 tl_abort_disk_io(mdev);
1425
1426                         /* current state still has to be D_FAILED,
1427                          * there is only one way out: to D_DISKLESS,
1428                          * and that may only happen after our put_ldev below. */
1429                         if (mdev->state.disk != D_FAILED)
1430                                 dev_err(DEV,
1431                                         "ASSERT FAILED: disk is %s during detach\n",
1432                                         drbd_disk_str(mdev->state.disk));
1433
1434                         if (ns.conn >= C_CONNECTED)
1435                                 drbd_send_state(mdev, ns);
1436
1437                         drbd_rs_cancel_all(mdev);
1438
1439                         /* In case we want to get something to stable storage still,
1440                          * this may be the last chance.
1441                          * Following put_ldev may transition to D_DISKLESS. */
1442                         drbd_md_sync(mdev);
1443                 }
1444                 put_ldev(mdev);
1445         }
1446
1447         /* second half of local IO error, failure to attach,
1448          * or administrative detach,
1449          * after local_cnt references have reached zero again */
1450         if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
1451                 /* We must still be diskless,
1452                  * re-attach has to be serialized with this! */
1453                 if (mdev->state.disk != D_DISKLESS)
1454                         dev_err(DEV,
1455                                 "ASSERT FAILED: disk is %s while going diskless\n",
1456                                 drbd_disk_str(mdev->state.disk));
1457
1458                 if (ns.conn >= C_CONNECTED)
1459                         drbd_send_state(mdev, ns);
1460                 /* corresponding get_ldev in __drbd_set_state
1461                  * this may finally trigger drbd_ldev_destroy. */
1462                 put_ldev(mdev);
1463         }
1464
1465         /* Notify peer that I had a local IO error, and did not detached.. */
1466         if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT && ns.conn >= C_CONNECTED)
1467                 drbd_send_state(mdev, ns);
1468
1469         /* Disks got bigger while they were detached */
1470         if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1471             test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) {
1472                 if (ns.conn == C_CONNECTED)
1473                         resync_after_online_grow(mdev);
1474         }
1475
1476         /* A resync finished or aborted, wake paused devices... */
1477         if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1478             (os.peer_isp && !ns.peer_isp) ||
1479             (os.user_isp && !ns.user_isp))
1480                 resume_next_sg(mdev);
1481
1482         /* sync target done with resync.  Explicitly notify peer, even though
1483          * it should (at least for non-empty resyncs) already know itself. */
1484         if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
1485                 drbd_send_state(mdev, ns);
1486
1487         /* Verify finished, or reached stop sector.  Peer did not know about
1488          * the stop sector, and we may even have changed the stop sector during
1489          * verify to interrupt/stop early.  Send the new state. */
1490         if (os.conn == C_VERIFY_S && ns.conn == C_CONNECTED
1491         && verify_can_do_stop_sector(mdev))
1492                 drbd_send_state(mdev, ns);
1493
1494         /* This triggers bitmap writeout of potentially still unwritten pages
1495          * if the resync finished cleanly, or aborted because of peer disk
1496          * failure, or because of connection loss.
1497          * For resync aborted because of local disk failure, we cannot do
1498          * any bitmap writeout anymore.
1499          * No harm done if some bits change during this phase.
1500          */
1501         if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) {
1502                 drbd_queue_bitmap_io(mdev, &drbd_bm_write_copy_pages, NULL,
1503                         "write from resync_finished", BM_LOCKED_CHANGE_ALLOWED);
1504                 put_ldev(mdev);
1505         }
1506
1507         if (ns.disk == D_DISKLESS &&
1508             ns.conn == C_STANDALONE &&
1509             ns.role == R_SECONDARY) {
1510                 if (os.aftr_isp != ns.aftr_isp)
1511                         resume_next_sg(mdev);
1512         }
1513
1514         drbd_md_sync(mdev);
1515 }
1516
1517 struct after_conn_state_chg_work {
1518         struct drbd_work w;
1519         enum drbd_conns oc;
1520         union drbd_state ns_min;
1521         union drbd_state ns_max; /* new, max state, over all mdevs */
1522         enum chg_state_flags flags;
1523 };
1524
1525 static int w_after_conn_state_ch(struct drbd_work *w, int unused)
1526 {
1527         struct after_conn_state_chg_work *acscw =
1528                 container_of(w, struct after_conn_state_chg_work, w);
1529         struct drbd_tconn *tconn = w->tconn;
1530         enum drbd_conns oc = acscw->oc;
1531         union drbd_state ns_max = acscw->ns_max;
1532         struct drbd_conf *mdev;
1533         int vnr;
1534
1535         kfree(acscw);
1536
1537         /* Upon network configuration, we need to start the receiver */
1538         if (oc == C_STANDALONE && ns_max.conn == C_UNCONNECTED)
1539                 drbd_thread_start(&tconn->receiver);
1540
1541         if (oc == C_DISCONNECTING && ns_max.conn == C_STANDALONE) {
1542                 struct net_conf *old_conf;
1543
1544                 mutex_lock(&tconn->conf_update);
1545                 old_conf = tconn->net_conf;
1546                 tconn->my_addr_len = 0;
1547                 tconn->peer_addr_len = 0;
1548                 rcu_assign_pointer(tconn->net_conf, NULL);
1549                 conn_free_crypto(tconn);
1550                 mutex_unlock(&tconn->conf_update);
1551
1552                 synchronize_rcu();
1553                 kfree(old_conf);
1554         }
1555
1556         if (ns_max.susp_fen) {
1557                 /* case1: The outdate peer handler is successful: */
1558                 if (ns_max.pdsk <= D_OUTDATED) {
1559                         rcu_read_lock();
1560                         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1561                                 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1562                                         drbd_uuid_new_current(mdev);
1563                                         clear_bit(NEW_CUR_UUID, &mdev->flags);
1564                                 }
1565                         }
1566                         rcu_read_unlock();
1567                         spin_lock_irq(&tconn->req_lock);
1568                         _tl_restart(tconn, CONNECTION_LOST_WHILE_PENDING);
1569                         _conn_request_state(tconn,
1570                                             (union drbd_state) { { .susp_fen = 1 } },
1571                                             (union drbd_state) { { .susp_fen = 0 } },
1572                                             CS_VERBOSE);
1573                         spin_unlock_irq(&tconn->req_lock);
1574                 }
1575         }
1576         kref_put(&tconn->kref, &conn_destroy);
1577
1578         conn_md_sync(tconn);
1579
1580         return 0;
1581 }
1582
1583 void conn_old_common_state(struct drbd_tconn *tconn, union drbd_state *pcs, enum chg_state_flags *pf)
1584 {
1585         enum chg_state_flags flags = ~0;
1586         struct drbd_conf *mdev;
1587         int vnr, first_vol = 1;
1588         union drbd_dev_state os, cs = {
1589                 { .role = R_SECONDARY,
1590                   .peer = R_UNKNOWN,
1591                   .conn = tconn->cstate,
1592                   .disk = D_DISKLESS,
1593                   .pdsk = D_UNKNOWN,
1594                 } };
1595
1596         rcu_read_lock();
1597         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1598                 os = mdev->state;
1599
1600                 if (first_vol) {
1601                         cs = os;
1602                         first_vol = 0;
1603                         continue;
1604                 }
1605
1606                 if (cs.role != os.role)
1607                         flags &= ~CS_DC_ROLE;
1608
1609                 if (cs.peer != os.peer)
1610                         flags &= ~CS_DC_PEER;
1611
1612                 if (cs.conn != os.conn)
1613                         flags &= ~CS_DC_CONN;
1614
1615                 if (cs.disk != os.disk)
1616                         flags &= ~CS_DC_DISK;
1617
1618                 if (cs.pdsk != os.pdsk)
1619                         flags &= ~CS_DC_PDSK;
1620         }
1621         rcu_read_unlock();
1622
1623         *pf |= CS_DC_MASK;
1624         *pf &= flags;
1625         (*pcs).i = cs.i;
1626 }
1627
1628 static enum drbd_state_rv
1629 conn_is_valid_transition(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1630                          enum chg_state_flags flags)
1631 {
1632         enum drbd_state_rv rv = SS_SUCCESS;
1633         union drbd_state ns, os;
1634         struct drbd_conf *mdev;
1635         int vnr;
1636
1637         rcu_read_lock();
1638         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1639                 os = drbd_read_state(mdev);
1640                 ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
1641
1642                 if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
1643                         ns.disk = os.disk;
1644
1645                 if (ns.i == os.i)
1646                         continue;
1647
1648                 rv = is_valid_transition(os, ns);
1649                 if (rv < SS_SUCCESS)
1650                         break;
1651
1652                 if (!(flags & CS_HARD)) {
1653                         rv = is_valid_state(mdev, ns);
1654                         if (rv < SS_SUCCESS) {
1655                                 if (is_valid_state(mdev, os) == rv)
1656                                         rv = is_valid_soft_transition(os, ns, tconn);
1657                         } else
1658                                 rv = is_valid_soft_transition(os, ns, tconn);
1659                 }
1660                 if (rv < SS_SUCCESS)
1661                         break;
1662         }
1663         rcu_read_unlock();
1664
1665         if (rv < SS_SUCCESS && flags & CS_VERBOSE)
1666                 print_st_err(mdev, os, ns, rv);
1667
1668         return rv;
1669 }
1670
1671 void
1672 conn_set_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1673                union drbd_state *pns_min, union drbd_state *pns_max, enum chg_state_flags flags)
1674 {
1675         union drbd_state ns, os, ns_max = { };
1676         union drbd_state ns_min = {
1677                 { .role = R_MASK,
1678                   .peer = R_MASK,
1679                   .conn = val.conn,
1680                   .disk = D_MASK,
1681                   .pdsk = D_MASK
1682                 } };
1683         struct drbd_conf *mdev;
1684         enum drbd_state_rv rv;
1685         int vnr, number_of_volumes = 0;
1686
1687         if (mask.conn == C_MASK) {
1688                 /* remember last connect time so request_timer_fn() won't
1689                  * kill newly established sessions while we are still trying to thaw
1690                  * previously frozen IO */
1691                 if (tconn->cstate != C_WF_REPORT_PARAMS && val.conn == C_WF_REPORT_PARAMS)
1692                         tconn->last_reconnect_jif = jiffies;
1693
1694                 tconn->cstate = val.conn;
1695         }
1696
1697         rcu_read_lock();
1698         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1699                 number_of_volumes++;
1700                 os = drbd_read_state(mdev);
1701                 ns = apply_mask_val(os, mask, val);
1702                 ns = sanitize_state(mdev, ns, NULL);
1703
1704                 if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
1705                         ns.disk = os.disk;
1706
1707                 rv = __drbd_set_state(mdev, ns, flags, NULL);
1708                 if (rv < SS_SUCCESS)
1709                         BUG();
1710
1711                 ns.i = mdev->state.i;
1712                 ns_max.role = max_role(ns.role, ns_max.role);
1713                 ns_max.peer = max_role(ns.peer, ns_max.peer);
1714                 ns_max.conn = max_t(enum drbd_conns, ns.conn, ns_max.conn);
1715                 ns_max.disk = max_t(enum drbd_disk_state, ns.disk, ns_max.disk);
1716                 ns_max.pdsk = max_t(enum drbd_disk_state, ns.pdsk, ns_max.pdsk);
1717
1718                 ns_min.role = min_role(ns.role, ns_min.role);
1719                 ns_min.peer = min_role(ns.peer, ns_min.peer);
1720                 ns_min.conn = min_t(enum drbd_conns, ns.conn, ns_min.conn);
1721                 ns_min.disk = min_t(enum drbd_disk_state, ns.disk, ns_min.disk);
1722                 ns_min.pdsk = min_t(enum drbd_disk_state, ns.pdsk, ns_min.pdsk);
1723         }
1724         rcu_read_unlock();
1725
1726         if (number_of_volumes == 0) {
1727                 ns_min = ns_max = (union drbd_state) { {
1728                                 .role = R_SECONDARY,
1729                                 .peer = R_UNKNOWN,
1730                                 .conn = val.conn,
1731                                 .disk = D_DISKLESS,
1732                                 .pdsk = D_UNKNOWN
1733                         } };
1734         }
1735
1736         ns_min.susp = ns_max.susp = tconn->susp;
1737         ns_min.susp_nod = ns_max.susp_nod = tconn->susp_nod;
1738         ns_min.susp_fen = ns_max.susp_fen = tconn->susp_fen;
1739
1740         *pns_min = ns_min;
1741         *pns_max = ns_max;
1742 }
1743
1744 static enum drbd_state_rv
1745 _conn_rq_cond(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val)
1746 {
1747         enum drbd_state_rv rv;
1748
1749         if (test_and_clear_bit(CONN_WD_ST_CHG_OKAY, &tconn->flags))
1750                 return SS_CW_SUCCESS;
1751
1752         if (test_and_clear_bit(CONN_WD_ST_CHG_FAIL, &tconn->flags))
1753                 return SS_CW_FAILED_BY_PEER;
1754
1755         rv = conn_is_valid_transition(tconn, mask, val, 0);
1756         if (rv == SS_SUCCESS && tconn->cstate == C_WF_REPORT_PARAMS)
1757                 rv = SS_UNKNOWN_ERROR; /* continue waiting */
1758
1759         return rv;
1760 }
1761
1762 enum drbd_state_rv
1763 _conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1764                     enum chg_state_flags flags)
1765 {
1766         enum drbd_state_rv rv = SS_SUCCESS;
1767         struct after_conn_state_chg_work *acscw;
1768         enum drbd_conns oc = tconn->cstate;
1769         union drbd_state ns_max, ns_min, os;
1770         bool have_mutex = false;
1771
1772         if (mask.conn) {
1773                 rv = is_valid_conn_transition(oc, val.conn);
1774                 if (rv < SS_SUCCESS)
1775                         goto abort;
1776         }
1777
1778         rv = conn_is_valid_transition(tconn, mask, val, flags);
1779         if (rv < SS_SUCCESS)
1780                 goto abort;
1781
1782         if (oc == C_WF_REPORT_PARAMS && val.conn == C_DISCONNECTING &&
1783             !(flags & (CS_LOCAL_ONLY | CS_HARD))) {
1784
1785                 /* This will be a cluster-wide state change.
1786                  * Need to give up the spinlock, grab the mutex,
1787                  * then send the state change request, ... */
1788                 spin_unlock_irq(&tconn->req_lock);
1789                 mutex_lock(&tconn->cstate_mutex);
1790                 have_mutex = true;
1791
1792                 set_bit(CONN_WD_ST_CHG_REQ, &tconn->flags);
1793                 if (conn_send_state_req(tconn, mask, val)) {
1794                         /* sending failed. */
1795                         clear_bit(CONN_WD_ST_CHG_REQ, &tconn->flags);
1796                         rv = SS_CW_FAILED_BY_PEER;
1797                         /* need to re-aquire the spin lock, though */
1798                         goto abort_unlocked;
1799                 }
1800
1801                 if (val.conn == C_DISCONNECTING)
1802                         set_bit(DISCONNECT_SENT, &tconn->flags);
1803
1804                 /* ... and re-aquire the spinlock.
1805                  * If _conn_rq_cond() returned >= SS_SUCCESS, we must call
1806                  * conn_set_state() within the same spinlock. */
1807                 spin_lock_irq(&tconn->req_lock);
1808                 wait_event_lock_irq(tconn->ping_wait,
1809                                 (rv = _conn_rq_cond(tconn, mask, val)),
1810                                 tconn->req_lock);
1811                 clear_bit(CONN_WD_ST_CHG_REQ, &tconn->flags);
1812                 if (rv < SS_SUCCESS)
1813                         goto abort;
1814         }
1815
1816         conn_old_common_state(tconn, &os, &flags);
1817         flags |= CS_DC_SUSP;
1818         conn_set_state(tconn, mask, val, &ns_min, &ns_max, flags);
1819         conn_pr_state_change(tconn, os, ns_max, flags);
1820
1821         acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC);
1822         if (acscw) {
1823                 acscw->oc = os.conn;
1824                 acscw->ns_min = ns_min;
1825                 acscw->ns_max = ns_max;
1826                 acscw->flags = flags;
1827                 acscw->w.cb = w_after_conn_state_ch;
1828                 kref_get(&tconn->kref);
1829                 acscw->w.tconn = tconn;
1830                 drbd_queue_work(&tconn->sender_work, &acscw->w);
1831         } else {
1832                 conn_err(tconn, "Could not kmalloc an acscw\n");
1833         }
1834
1835  abort:
1836         if (have_mutex) {
1837                 /* mutex_unlock() "... must not be used in interrupt context.",
1838                  * so give up the spinlock, then re-aquire it */
1839                 spin_unlock_irq(&tconn->req_lock);
1840  abort_unlocked:
1841                 mutex_unlock(&tconn->cstate_mutex);
1842                 spin_lock_irq(&tconn->req_lock);
1843         }
1844         if (rv < SS_SUCCESS && flags & CS_VERBOSE) {
1845                 conn_err(tconn, "State change failed: %s\n", drbd_set_st_err_str(rv));
1846                 conn_err(tconn, " mask = 0x%x val = 0x%x\n", mask.i, val.i);
1847                 conn_err(tconn, " old_conn:%s wanted_conn:%s\n", drbd_conn_str(oc), drbd_conn_str(val.conn));
1848         }
1849         return rv;
1850 }
1851
1852 enum drbd_state_rv
1853 conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1854                    enum chg_state_flags flags)
1855 {
1856         enum drbd_state_rv rv;
1857
1858         spin_lock_irq(&tconn->req_lock);
1859         rv = _conn_request_state(tconn, mask, val, flags);
1860         spin_unlock_irq(&tconn->req_lock);
1861
1862         return rv;
1863 }