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_protocol.h"
  31#include "drbd_req.h"
 
  32
  33struct after_state_chg_work {
  34	struct drbd_work w;
  35	struct drbd_device *device;
  36	union drbd_state os;
  37	union drbd_state ns;
  38	enum chg_state_flags flags;
  39	struct completion *done;
 
  40};
  41
  42enum sanitize_state_warnings {
  43	NO_WARNING,
  44	ABORTED_ONLINE_VERIFY,
  45	ABORTED_RESYNC,
  46	CONNECTION_LOST_NEGOTIATING,
  47	IMPLICITLY_UPGRADED_DISK,
  48	IMPLICITLY_UPGRADED_PDSK,
  49};
  50
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  51static int w_after_state_ch(struct drbd_work *w, int unused);
  52static void after_state_ch(struct drbd_device *device, union drbd_state os,
  53			   union drbd_state ns, enum chg_state_flags flags);
 
  54static enum drbd_state_rv is_valid_state(struct drbd_device *, union drbd_state);
  55static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state, struct drbd_connection *);
  56static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns);
  57static union drbd_state sanitize_state(struct drbd_device *device, union drbd_state ns,
  58				       enum sanitize_state_warnings *warn);
  59
  60static inline bool is_susp(union drbd_state s)
  61{
  62        return s.susp || s.susp_nod || s.susp_fen;
  63}
  64
  65bool conn_all_vols_unconf(struct drbd_connection *connection)
  66{
  67	struct drbd_peer_device *peer_device;
  68	bool rv = true;
  69	int vnr;
  70
  71	rcu_read_lock();
  72	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
  73		struct drbd_device *device = peer_device->device;
  74		if (device->state.disk != D_DISKLESS ||
  75		    device->state.conn != C_STANDALONE ||
  76		    device->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. */
  88static 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}
 
  96static 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
 105enum drbd_role conn_highest_role(struct drbd_connection *connection)
 106{
 107	enum drbd_role role = R_UNKNOWN;
 108	struct drbd_peer_device *peer_device;
 109	int vnr;
 110
 111	rcu_read_lock();
 112	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
 113		struct drbd_device *device = peer_device->device;
 114		role = max_role(role, device->state.role);
 115	}
 116	rcu_read_unlock();
 117
 118	return role;
 119}
 120
 121enum drbd_role conn_highest_peer(struct drbd_connection *connection)
 122{
 123	enum drbd_role peer = R_UNKNOWN;
 124	struct drbd_peer_device *peer_device;
 125	int vnr;
 126
 127	rcu_read_lock();
 128	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
 129		struct drbd_device *device = peer_device->device;
 130		peer = max_role(peer, device->state.peer);
 131	}
 132	rcu_read_unlock();
 133
 134	return peer;
 135}
 136
 137enum drbd_disk_state conn_highest_disk(struct drbd_connection *connection)
 138{
 139	enum drbd_disk_state ds = D_DISKLESS;
 140	struct drbd_peer_device *peer_device;
 141	int vnr;
 142
 143	rcu_read_lock();
 144	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
 145		struct drbd_device *device = peer_device->device;
 146		ds = max_t(enum drbd_disk_state, ds, device->state.disk);
 147	}
 148	rcu_read_unlock();
 149
 150	return ds;
 151}
 152
 153enum drbd_disk_state conn_lowest_disk(struct drbd_connection *connection)
 154{
 155	enum drbd_disk_state ds = D_MASK;
 156	struct drbd_peer_device *peer_device;
 157	int vnr;
 158
 159	rcu_read_lock();
 160	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
 161		struct drbd_device *device = peer_device->device;
 162		ds = min_t(enum drbd_disk_state, ds, device->state.disk);
 163	}
 164	rcu_read_unlock();
 165
 166	return ds;
 167}
 168
 169enum drbd_disk_state conn_highest_pdsk(struct drbd_connection *connection)
 170{
 171	enum drbd_disk_state ds = D_DISKLESS;
 172	struct drbd_peer_device *peer_device;
 173	int vnr;
 174
 175	rcu_read_lock();
 176	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
 177		struct drbd_device *device = peer_device->device;
 178		ds = max_t(enum drbd_disk_state, ds, device->state.pdsk);
 179	}
 180	rcu_read_unlock();
 181
 182	return ds;
 183}
 184
 185enum drbd_conns conn_lowest_conn(struct drbd_connection *connection)
 186{
 187	enum drbd_conns conn = C_MASK;
 188	struct drbd_peer_device *peer_device;
 189	int vnr;
 190
 191	rcu_read_lock();
 192	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
 193		struct drbd_device *device = peer_device->device;
 194		conn = min_t(enum drbd_conns, conn, device->state.conn);
 195	}
 196	rcu_read_unlock();
 197
 198	return conn;
 199}
 200
 201static bool no_peer_wf_report_params(struct drbd_connection *connection)
 202{
 203	struct drbd_peer_device *peer_device;
 204	int vnr;
 205	bool rv = true;
 206
 207	rcu_read_lock();
 208	idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
 209		if (peer_device->device->state.conn == C_WF_REPORT_PARAMS) {
 210			rv = false;
 211			break;
 212		}
 213	rcu_read_unlock();
 214
 215	return rv;
 216}
 217
 
 
 
 
 
 
 
 
 
 
 
 
 218
 219/**
 220 * cl_wide_st_chg() - true if the state change is a cluster wide one
 221 * @device:	DRBD device.
 222 * @os:		old (current) state.
 223 * @ns:		new (wanted) state.
 224 */
 225static int cl_wide_st_chg(struct drbd_device *device,
 226			  union drbd_state os, union drbd_state ns)
 227{
 228	return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED &&
 229		 ((os.role != R_PRIMARY && ns.role == R_PRIMARY) ||
 230		  (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
 231		  (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) ||
 232		  (os.disk != D_FAILED && ns.disk == D_FAILED))) ||
 233		(os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) ||
 234		(os.conn == C_CONNECTED && ns.conn == C_VERIFY_S) ||
 235		(os.conn == C_CONNECTED && ns.conn == C_WF_REPORT_PARAMS);
 236}
 237
 238static union drbd_state
 239apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val)
 240{
 241	union drbd_state ns;
 242	ns.i = (os.i & ~mask.i) | val.i;
 243	return ns;
 244}
 245
 246enum drbd_state_rv
 247drbd_change_state(struct drbd_device *device, enum chg_state_flags f,
 248		  union drbd_state mask, union drbd_state val)
 249{
 250	unsigned long flags;
 251	union drbd_state ns;
 252	enum drbd_state_rv rv;
 253
 254	spin_lock_irqsave(&device->resource->req_lock, flags);
 255	ns = apply_mask_val(drbd_read_state(device), mask, val);
 256	rv = _drbd_set_state(device, ns, f, NULL);
 257	spin_unlock_irqrestore(&device->resource->req_lock, flags);
 258
 259	return rv;
 260}
 261
 262/**
 263 * drbd_force_state() - Impose a change which happens outside our control on our state
 264 * @device:	DRBD device.
 265 * @mask:	mask of state bits to change.
 266 * @val:	value of new state bits.
 267 */
 268void drbd_force_state(struct drbd_device *device,
 269	union drbd_state mask, union drbd_state val)
 270{
 271	drbd_change_state(device, CS_HARD, mask, val);
 272}
 273
 274static enum drbd_state_rv
 275_req_st_cond(struct drbd_device *device, union drbd_state mask,
 276	     union drbd_state val)
 277{
 278	union drbd_state os, ns;
 279	unsigned long flags;
 280	enum drbd_state_rv rv;
 281
 282	if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &device->flags))
 283		return SS_CW_SUCCESS;
 284
 285	if (test_and_clear_bit(CL_ST_CHG_FAIL, &device->flags))
 286		return SS_CW_FAILED_BY_PEER;
 287
 288	spin_lock_irqsave(&device->resource->req_lock, flags);
 289	os = drbd_read_state(device);
 290	ns = sanitize_state(device, apply_mask_val(os, mask, val), NULL);
 291	rv = is_valid_transition(os, ns);
 292	if (rv >= SS_SUCCESS)
 293		rv = SS_UNKNOWN_ERROR;  /* cont waiting, otherwise fail. */
 294
 295	if (!cl_wide_st_chg(device, os, ns))
 296		rv = SS_CW_NO_NEED;
 297	if (rv == SS_UNKNOWN_ERROR) {
 298		rv = is_valid_state(device, ns);
 299		if (rv >= SS_SUCCESS) {
 300			rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection);
 301			if (rv >= SS_SUCCESS)
 302				rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
 303		}
 304	}
 305	spin_unlock_irqrestore(&device->resource->req_lock, flags);
 306
 307	return rv;
 308}
 309
 310/**
 311 * drbd_req_state() - Perform an eventually cluster wide state change
 312 * @device:	DRBD device.
 313 * @mask:	mask of state bits to change.
 314 * @val:	value of new state bits.
 315 * @f:		flags
 316 *
 317 * Should not be called directly, use drbd_request_state() or
 318 * _drbd_request_state().
 319 */
 320static enum drbd_state_rv
 321drbd_req_state(struct drbd_device *device, union drbd_state mask,
 322	       union drbd_state val, enum chg_state_flags f)
 323{
 324	struct completion done;
 325	unsigned long flags;
 326	union drbd_state os, ns;
 327	enum drbd_state_rv rv;
 328
 329	init_completion(&done);
 330
 331	if (f & CS_SERIALIZE)
 332		mutex_lock(device->state_mutex);
 333
 334	spin_lock_irqsave(&device->resource->req_lock, flags);
 335	os = drbd_read_state(device);
 336	ns = sanitize_state(device, apply_mask_val(os, mask, val), NULL);
 337	rv = is_valid_transition(os, ns);
 338	if (rv < SS_SUCCESS) {
 339		spin_unlock_irqrestore(&device->resource->req_lock, flags);
 340		goto abort;
 341	}
 342
 343	if (cl_wide_st_chg(device, os, ns)) {
 344		rv = is_valid_state(device, ns);
 345		if (rv == SS_SUCCESS)
 346			rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection);
 347		spin_unlock_irqrestore(&device->resource->req_lock, flags);
 348
 349		if (rv < SS_SUCCESS) {
 350			if (f & CS_VERBOSE)
 351				print_st_err(device, os, ns, rv);
 352			goto abort;
 353		}
 354
 355		if (drbd_send_state_req(first_peer_device(device), mask, val)) {
 356			rv = SS_CW_FAILED_BY_PEER;
 357			if (f & CS_VERBOSE)
 358				print_st_err(device, os, ns, rv);
 359			goto abort;
 360		}
 361
 362		wait_event(device->state_wait,
 363			(rv = _req_st_cond(device, mask, val)));
 364
 365		if (rv < SS_SUCCESS) {
 366			if (f & CS_VERBOSE)
 367				print_st_err(device, os, ns, rv);
 368			goto abort;
 369		}
 370		spin_lock_irqsave(&device->resource->req_lock, flags);
 371		ns = apply_mask_val(drbd_read_state(device), mask, val);
 372		rv = _drbd_set_state(device, ns, f, &done);
 373	} else {
 374		rv = _drbd_set_state(device, ns, f, &done);
 375	}
 376
 377	spin_unlock_irqrestore(&device->resource->req_lock, flags);
 378
 379	if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) {
 380		D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
 381		wait_for_completion(&done);
 382	}
 383
 384abort:
 385	if (f & CS_SERIALIZE)
 386		mutex_unlock(device->state_mutex);
 387
 388	return rv;
 389}
 390
 391/**
 392 * _drbd_request_state() - Request a state change (with flags)
 393 * @device:	DRBD device.
 394 * @mask:	mask of state bits to change.
 395 * @val:	value of new state bits.
 396 * @f:		flags
 397 *
 398 * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE
 399 * flag, or when logging of failed state change requests is not desired.
 400 */
 401enum drbd_state_rv
 402_drbd_request_state(struct drbd_device *device, union drbd_state mask,
 403		    union drbd_state val, enum chg_state_flags f)
 404{
 405	enum drbd_state_rv rv;
 406
 407	wait_event(device->state_wait,
 408		   (rv = drbd_req_state(device, mask, val, f)) != SS_IN_TRANSIENT_STATE);
 409
 410	return rv;
 411}
 412
 413static void print_st(struct drbd_device *device, char *name, union drbd_state ns)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 414{
 415	drbd_err(device, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n",
 416	    name,
 417	    drbd_conn_str(ns.conn),
 418	    drbd_role_str(ns.role),
 419	    drbd_role_str(ns.peer),
 420	    drbd_disk_str(ns.disk),
 421	    drbd_disk_str(ns.pdsk),
 422	    is_susp(ns) ? 's' : 'r',
 423	    ns.aftr_isp ? 'a' : '-',
 424	    ns.peer_isp ? 'p' : '-',
 425	    ns.user_isp ? 'u' : '-',
 426	    ns.susp_fen ? 'F' : '-',
 427	    ns.susp_nod ? 'N' : '-'
 428	    );
 429}
 430
 431void print_st_err(struct drbd_device *device, union drbd_state os,
 432	          union drbd_state ns, enum drbd_state_rv err)
 433{
 434	if (err == SS_IN_TRANSIENT_STATE)
 435		return;
 436	drbd_err(device, "State change failed: %s\n", drbd_set_st_err_str(err));
 437	print_st(device, " state", os);
 438	print_st(device, "wanted", ns);
 439}
 440
 441static long print_state_change(char *pb, union drbd_state os, union drbd_state ns,
 442			       enum chg_state_flags flags)
 443{
 444	char *pbp;
 445	pbp = pb;
 446	*pbp = 0;
 447
 448	if (ns.role != os.role && flags & CS_DC_ROLE)
 449		pbp += sprintf(pbp, "role( %s -> %s ) ",
 450			       drbd_role_str(os.role),
 451			       drbd_role_str(ns.role));
 452	if (ns.peer != os.peer && flags & CS_DC_PEER)
 453		pbp += sprintf(pbp, "peer( %s -> %s ) ",
 454			       drbd_role_str(os.peer),
 455			       drbd_role_str(ns.peer));
 456	if (ns.conn != os.conn && flags & CS_DC_CONN)
 457		pbp += sprintf(pbp, "conn( %s -> %s ) ",
 458			       drbd_conn_str(os.conn),
 459			       drbd_conn_str(ns.conn));
 460	if (ns.disk != os.disk && flags & CS_DC_DISK)
 461		pbp += sprintf(pbp, "disk( %s -> %s ) ",
 462			       drbd_disk_str(os.disk),
 463			       drbd_disk_str(ns.disk));
 464	if (ns.pdsk != os.pdsk && flags & CS_DC_PDSK)
 465		pbp += sprintf(pbp, "pdsk( %s -> %s ) ",
 466			       drbd_disk_str(os.pdsk),
 467			       drbd_disk_str(ns.pdsk));
 468
 469	return pbp - pb;
 470}
 471
 472static void drbd_pr_state_change(struct drbd_device *device, union drbd_state os, union drbd_state ns,
 473				 enum chg_state_flags flags)
 474{
 475	char pb[300];
 476	char *pbp = pb;
 477
 478	pbp += print_state_change(pbp, os, ns, flags ^ CS_DC_MASK);
 479
 480	if (ns.aftr_isp != os.aftr_isp)
 481		pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ",
 482			       os.aftr_isp,
 483			       ns.aftr_isp);
 484	if (ns.peer_isp != os.peer_isp)
 485		pbp += sprintf(pbp, "peer_isp( %d -> %d ) ",
 486			       os.peer_isp,
 487			       ns.peer_isp);
 488	if (ns.user_isp != os.user_isp)
 489		pbp += sprintf(pbp, "user_isp( %d -> %d ) ",
 490			       os.user_isp,
 491			       ns.user_isp);
 492
 493	if (pbp != pb)
 494		drbd_info(device, "%s\n", pb);
 495}
 496
 497static void conn_pr_state_change(struct drbd_connection *connection, union drbd_state os, union drbd_state ns,
 498				 enum chg_state_flags flags)
 499{
 500	char pb[300];
 501	char *pbp = pb;
 502
 503	pbp += print_state_change(pbp, os, ns, flags);
 504
 505	if (is_susp(ns) != is_susp(os) && flags & CS_DC_SUSP)
 506		pbp += sprintf(pbp, "susp( %d -> %d ) ",
 507			       is_susp(os),
 508			       is_susp(ns));
 509
 510	if (pbp != pb)
 511		drbd_info(connection, "%s\n", pb);
 512}
 513
 514
 515/**
 516 * is_valid_state() - Returns an SS_ error code if ns is not valid
 517 * @device:	DRBD device.
 518 * @ns:		State to consider.
 519 */
 520static enum drbd_state_rv
 521is_valid_state(struct drbd_device *device, union drbd_state ns)
 522{
 523	/* See drbd_state_sw_errors in drbd_strings.c */
 524
 525	enum drbd_fencing_p fp;
 526	enum drbd_state_rv rv = SS_SUCCESS;
 527	struct net_conf *nc;
 528
 529	rcu_read_lock();
 530	fp = FP_DONT_CARE;
 531	if (get_ldev(device)) {
 532		fp = rcu_dereference(device->ldev->disk_conf)->fencing;
 533		put_ldev(device);
 534	}
 535
 536	nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
 537	if (nc) {
 538		if (!nc->two_primaries && ns.role == R_PRIMARY) {
 539			if (ns.peer == R_PRIMARY)
 540				rv = SS_TWO_PRIMARIES;
 541			else if (conn_highest_peer(first_peer_device(device)->connection) == R_PRIMARY)
 542				rv = SS_O_VOL_PEER_PRI;
 543		}
 544	}
 545
 546	if (rv <= 0)
 547		/* already found a reason to abort */;
 548	else if (ns.role == R_SECONDARY && device->open_cnt)
 549		rv = SS_DEVICE_IN_USE;
 550
 551	else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE)
 552		rv = SS_NO_UP_TO_DATE_DISK;
 553
 554	else if (fp >= FP_RESOURCE &&
 555		 ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN)
 556		rv = SS_PRIMARY_NOP;
 557
 558	else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT)
 559		rv = SS_NO_UP_TO_DATE_DISK;
 560
 561	else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT)
 562		rv = SS_NO_LOCAL_DISK;
 563
 564	else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT)
 565		rv = SS_NO_REMOTE_DISK;
 566
 567	else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
 568		rv = SS_NO_UP_TO_DATE_DISK;
 569
 570	else if ((ns.conn == C_CONNECTED ||
 571		  ns.conn == C_WF_BITMAP_S ||
 572		  ns.conn == C_SYNC_SOURCE ||
 573		  ns.conn == C_PAUSED_SYNC_S) &&
 574		  ns.disk == D_OUTDATED)
 575		rv = SS_CONNECTED_OUTDATES;
 576
 577	else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 578		 (nc->verify_alg[0] == 0))
 579		rv = SS_NO_VERIFY_ALG;
 580
 581	else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 582		  first_peer_device(device)->connection->agreed_pro_version < 88)
 583		rv = SS_NOT_SUPPORTED;
 584
 585	else if (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
 586		rv = SS_NO_UP_TO_DATE_DISK;
 587
 588	else if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
 589                 ns.pdsk == D_UNKNOWN)
 590		rv = SS_NEED_CONNECTION;
 591
 592	else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN)
 593		rv = SS_CONNECTED_OUTDATES;
 594
 
 595	rcu_read_unlock();
 596
 597	return rv;
 598}
 599
 600/**
 601 * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible
 602 * This function limits state transitions that may be declined by DRBD. I.e.
 603 * user requests (aka soft transitions).
 604 * @device:	DRBD device.
 605 * @ns:		new state.
 606 * @os:		old state.
 607 */
 608static enum drbd_state_rv
 609is_valid_soft_transition(union drbd_state os, union drbd_state ns, struct drbd_connection *connection)
 610{
 611	enum drbd_state_rv rv = SS_SUCCESS;
 612
 613	if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
 614	    os.conn > C_CONNECTED)
 615		rv = SS_RESYNC_RUNNING;
 616
 617	if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
 618		rv = SS_ALREADY_STANDALONE;
 619
 620	if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
 621		rv = SS_IS_DISKLESS;
 622
 623	if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
 624		rv = SS_NO_NET_CONFIG;
 625
 626	if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
 627		rv = SS_LOWER_THAN_OUTDATED;
 628
 629	if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
 630		rv = SS_IN_TRANSIENT_STATE;
 631
 632	/* if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS)
 633	   rv = SS_IN_TRANSIENT_STATE; */
 634
 635	/* While establishing a connection only allow cstate to change.
 636	   Delay/refuse role changes, detach attach etc... */
 637	if (test_bit(STATE_SENT, &connection->flags) &&
 638	    !(os.conn == C_WF_REPORT_PARAMS ||
 639	      (ns.conn == C_WF_REPORT_PARAMS && os.conn == C_WF_CONNECTION)))
 640		rv = SS_IN_TRANSIENT_STATE;
 641
 
 
 
 
 
 
 
 
 
 642	if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
 643		rv = SS_NEED_CONNECTION;
 644
 645	if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
 646	    ns.conn != os.conn && os.conn > C_CONNECTED)
 647		rv = SS_RESYNC_RUNNING;
 648
 649	if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
 650	    os.conn < C_CONNECTED)
 651		rv = SS_NEED_CONNECTION;
 652
 653	if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)
 654	    && os.conn < C_WF_REPORT_PARAMS)
 655		rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */
 656
 657	if (ns.conn == C_DISCONNECTING && ns.pdsk == D_OUTDATED &&
 658	    os.conn < C_CONNECTED && os.pdsk > D_OUTDATED)
 659		rv = SS_OUTDATE_WO_CONN;
 660
 661	return rv;
 662}
 663
 664static enum drbd_state_rv
 665is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc)
 666{
 667	/* no change -> nothing to do, at least for the connection part */
 668	if (oc == nc)
 669		return SS_NOTHING_TO_DO;
 670
 671	/* disconnect of an unconfigured connection does not make sense */
 672	if (oc == C_STANDALONE && nc == C_DISCONNECTING)
 673		return SS_ALREADY_STANDALONE;
 674
 675	/* from C_STANDALONE, we start with C_UNCONNECTED */
 676	if (oc == C_STANDALONE && nc != C_UNCONNECTED)
 677		return SS_NEED_CONNECTION;
 678
 679	/* When establishing a connection we need to go through WF_REPORT_PARAMS!
 680	   Necessary to do the right thing upon invalidate-remote on a disconnected resource */
 681	if (oc < C_WF_REPORT_PARAMS && nc >= C_CONNECTED)
 682		return SS_NEED_CONNECTION;
 683
 684	/* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */
 685	if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING)
 686		return SS_IN_TRANSIENT_STATE;
 687
 688	/* After C_DISCONNECTING only C_STANDALONE may follow */
 689	if (oc == C_DISCONNECTING && nc != C_STANDALONE)
 690		return SS_IN_TRANSIENT_STATE;
 691
 692	return SS_SUCCESS;
 693}
 694
 695
 696/**
 697 * is_valid_transition() - Returns an SS_ error code if the state transition is not possible
 698 * This limits hard state transitions. Hard state transitions are facts there are
 699 * imposed on DRBD by the environment. E.g. disk broke or network broke down.
 700 * But those hard state transitions are still not allowed to do everything.
 701 * @ns:		new state.
 702 * @os:		old state.
 703 */
 704static enum drbd_state_rv
 705is_valid_transition(union drbd_state os, union drbd_state ns)
 706{
 707	enum drbd_state_rv rv;
 708
 709	rv = is_valid_conn_transition(os.conn, ns.conn);
 710
 711	/* we cannot fail (again) if we already detached */
 712	if (ns.disk == D_FAILED && os.disk == D_DISKLESS)
 713		rv = SS_IS_DISKLESS;
 714
 715	return rv;
 716}
 717
 718static void print_sanitize_warnings(struct drbd_device *device, enum sanitize_state_warnings warn)
 719{
 720	static const char *msg_table[] = {
 721		[NO_WARNING] = "",
 722		[ABORTED_ONLINE_VERIFY] = "Online-verify aborted.",
 723		[ABORTED_RESYNC] = "Resync aborted.",
 724		[CONNECTION_LOST_NEGOTIATING] = "Connection lost while negotiating, no data!",
 725		[IMPLICITLY_UPGRADED_DISK] = "Implicitly upgraded disk",
 726		[IMPLICITLY_UPGRADED_PDSK] = "Implicitly upgraded pdsk",
 727	};
 728
 729	if (warn != NO_WARNING)
 730		drbd_warn(device, "%s\n", msg_table[warn]);
 731}
 732
 733/**
 734 * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
 735 * @device:	DRBD device.
 736 * @os:		old state.
 737 * @ns:		new state.
 738 * @warn_sync_abort:
 739 *
 740 * When we loose connection, we have to set the state of the peers disk (pdsk)
 741 * to D_UNKNOWN. This rule and many more along those lines are in this function.
 742 */
 743static union drbd_state sanitize_state(struct drbd_device *device, union drbd_state ns,
 744				       enum sanitize_state_warnings *warn)
 745{
 746	enum drbd_fencing_p fp;
 747	enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max;
 748
 749	if (warn)
 750		*warn = NO_WARNING;
 751
 752	fp = FP_DONT_CARE;
 753	if (get_ldev(device)) {
 754		rcu_read_lock();
 755		fp = rcu_dereference(device->ldev->disk_conf)->fencing;
 756		rcu_read_unlock();
 757		put_ldev(device);
 758	}
 759
 760	/* Implications from connection to peer and peer_isp */
 761	if (ns.conn < C_CONNECTED) {
 762		ns.peer_isp = 0;
 763		ns.peer = R_UNKNOWN;
 764		if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
 765			ns.pdsk = D_UNKNOWN;
 766	}
 767
 768	/* Clear the aftr_isp when becoming unconfigured */
 769	if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
 770		ns.aftr_isp = 0;
 771
 772	/* An implication of the disk states onto the connection state */
 773	/* Abort resync if a disk fails/detaches */
 774	if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
 775		if (warn)
 776			*warn = ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ?
 777				ABORTED_ONLINE_VERIFY : ABORTED_RESYNC;
 778		ns.conn = C_CONNECTED;
 779	}
 780
 781	/* Connection breaks down before we finished "Negotiating" */
 782	if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
 783	    get_ldev_if_state(device, D_NEGOTIATING)) {
 784		if (device->ed_uuid == device->ldev->md.uuid[UI_CURRENT]) {
 785			ns.disk = device->new_state_tmp.disk;
 786			ns.pdsk = device->new_state_tmp.pdsk;
 787		} else {
 788			if (warn)
 789				*warn = CONNECTION_LOST_NEGOTIATING;
 790			ns.disk = D_DISKLESS;
 791			ns.pdsk = D_UNKNOWN;
 792		}
 793		put_ldev(device);
 794	}
 795
 796	/* D_CONSISTENT and D_OUTDATED vanish when we get connected */
 797	if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) {
 798		if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED)
 799			ns.disk = D_UP_TO_DATE;
 800		if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)
 801			ns.pdsk = D_UP_TO_DATE;
 802	}
 803
 804	/* Implications of the connection stat on the disk states */
 805	disk_min = D_DISKLESS;
 806	disk_max = D_UP_TO_DATE;
 807	pdsk_min = D_INCONSISTENT;
 808	pdsk_max = D_UNKNOWN;
 809	switch ((enum drbd_conns)ns.conn) {
 810	case C_WF_BITMAP_T:
 811	case C_PAUSED_SYNC_T:
 812	case C_STARTING_SYNC_T:
 813	case C_WF_SYNC_UUID:
 814	case C_BEHIND:
 815		disk_min = D_INCONSISTENT;
 816		disk_max = D_OUTDATED;
 817		pdsk_min = D_UP_TO_DATE;
 818		pdsk_max = D_UP_TO_DATE;
 819		break;
 820	case C_VERIFY_S:
 821	case C_VERIFY_T:
 822		disk_min = D_UP_TO_DATE;
 823		disk_max = D_UP_TO_DATE;
 824		pdsk_min = D_UP_TO_DATE;
 825		pdsk_max = D_UP_TO_DATE;
 826		break;
 827	case C_CONNECTED:
 828		disk_min = D_DISKLESS;
 829		disk_max = D_UP_TO_DATE;
 830		pdsk_min = D_DISKLESS;
 831		pdsk_max = D_UP_TO_DATE;
 832		break;
 833	case C_WF_BITMAP_S:
 834	case C_PAUSED_SYNC_S:
 835	case C_STARTING_SYNC_S:
 836	case C_AHEAD:
 837		disk_min = D_UP_TO_DATE;
 838		disk_max = D_UP_TO_DATE;
 839		pdsk_min = D_INCONSISTENT;
 840		pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/
 841		break;
 842	case C_SYNC_TARGET:
 843		disk_min = D_INCONSISTENT;
 844		disk_max = D_INCONSISTENT;
 845		pdsk_min = D_UP_TO_DATE;
 846		pdsk_max = D_UP_TO_DATE;
 847		break;
 848	case C_SYNC_SOURCE:
 849		disk_min = D_UP_TO_DATE;
 850		disk_max = D_UP_TO_DATE;
 851		pdsk_min = D_INCONSISTENT;
 852		pdsk_max = D_INCONSISTENT;
 853		break;
 854	case C_STANDALONE:
 855	case C_DISCONNECTING:
 856	case C_UNCONNECTED:
 857	case C_TIMEOUT:
 858	case C_BROKEN_PIPE:
 859	case C_NETWORK_FAILURE:
 860	case C_PROTOCOL_ERROR:
 861	case C_TEAR_DOWN:
 862	case C_WF_CONNECTION:
 863	case C_WF_REPORT_PARAMS:
 864	case C_MASK:
 865		break;
 866	}
 867	if (ns.disk > disk_max)
 868		ns.disk = disk_max;
 869
 870	if (ns.disk < disk_min) {
 871		if (warn)
 872			*warn = IMPLICITLY_UPGRADED_DISK;
 873		ns.disk = disk_min;
 874	}
 875	if (ns.pdsk > pdsk_max)
 876		ns.pdsk = pdsk_max;
 877
 878	if (ns.pdsk < pdsk_min) {
 879		if (warn)
 880			*warn = IMPLICITLY_UPGRADED_PDSK;
 881		ns.pdsk = pdsk_min;
 882	}
 883
 884	if (fp == FP_STONITH &&
 885	    (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED))
 
 886		ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */
 887
 888	if (device->resource->res_opts.on_no_data == OND_SUSPEND_IO &&
 889	    (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
 
 890		ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */
 891
 892	if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
 893		if (ns.conn == C_SYNC_SOURCE)
 894			ns.conn = C_PAUSED_SYNC_S;
 895		if (ns.conn == C_SYNC_TARGET)
 896			ns.conn = C_PAUSED_SYNC_T;
 897	} else {
 898		if (ns.conn == C_PAUSED_SYNC_S)
 899			ns.conn = C_SYNC_SOURCE;
 900		if (ns.conn == C_PAUSED_SYNC_T)
 901			ns.conn = C_SYNC_TARGET;
 902	}
 903
 904	return ns;
 905}
 906
 907void drbd_resume_al(struct drbd_device *device)
 908{
 909	if (test_and_clear_bit(AL_SUSPENDED, &device->flags))
 910		drbd_info(device, "Resumed AL updates\n");
 911}
 912
 913/* helper for __drbd_set_state */
 914static void set_ov_position(struct drbd_device *device, enum drbd_conns cs)
 915{
 916	if (first_peer_device(device)->connection->agreed_pro_version < 90)
 917		device->ov_start_sector = 0;
 918	device->rs_total = drbd_bm_bits(device);
 919	device->ov_position = 0;
 920	if (cs == C_VERIFY_T) {
 921		/* starting online verify from an arbitrary position
 922		 * does not fit well into the existing protocol.
 923		 * on C_VERIFY_T, we initialize ov_left and friends
 924		 * implicitly in receive_DataRequest once the
 925		 * first P_OV_REQUEST is received */
 926		device->ov_start_sector = ~(sector_t)0;
 927	} else {
 928		unsigned long bit = BM_SECT_TO_BIT(device->ov_start_sector);
 929		if (bit >= device->rs_total) {
 930			device->ov_start_sector =
 931				BM_BIT_TO_SECT(device->rs_total - 1);
 932			device->rs_total = 1;
 933		} else
 934			device->rs_total -= bit;
 935		device->ov_position = device->ov_start_sector;
 936	}
 937	device->ov_left = device->rs_total;
 938}
 939
 940/**
 941 * __drbd_set_state() - Set a new DRBD state
 942 * @device:	DRBD device.
 943 * @ns:		new state.
 944 * @flags:	Flags
 945 * @done:	Optional completion, that will get completed after the after_state_ch() finished
 946 *
 947 * Caller needs to hold req_lock, and global_state_lock. Do not call directly.
 948 */
 949enum drbd_state_rv
 950__drbd_set_state(struct drbd_device *device, union drbd_state ns,
 951	         enum chg_state_flags flags, struct completion *done)
 952{
 
 
 953	union drbd_state os;
 954	enum drbd_state_rv rv = SS_SUCCESS;
 955	enum sanitize_state_warnings ssw;
 956	struct after_state_chg_work *ascw;
 957	bool did_remote, should_do_remote;
 958
 959	os = drbd_read_state(device);
 960
 961	ns = sanitize_state(device, ns, &ssw);
 962	if (ns.i == os.i)
 963		return SS_NOTHING_TO_DO;
 964
 965	rv = is_valid_transition(os, ns);
 966	if (rv < SS_SUCCESS)
 967		return rv;
 968
 969	if (!(flags & CS_HARD)) {
 970		/*  pre-state-change checks ; only look at ns  */
 971		/* See drbd_state_sw_errors in drbd_strings.c */
 972
 973		rv = is_valid_state(device, ns);
 974		if (rv < SS_SUCCESS) {
 975			/* If the old state was illegal as well, then let
 976			   this happen...*/
 977
 978			if (is_valid_state(device, os) == rv)
 979				rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection);
 980		} else
 981			rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection);
 982	}
 983
 984	if (rv < SS_SUCCESS) {
 985		if (flags & CS_VERBOSE)
 986			print_st_err(device, os, ns, rv);
 987		return rv;
 988	}
 989
 990	print_sanitize_warnings(device, ssw);
 991
 992	drbd_pr_state_change(device, os, ns, flags);
 993
 994	/* Display changes to the susp* flags that where caused by the call to
 995	   sanitize_state(). Only display it here if we where not called from
 996	   _conn_request_state() */
 997	if (!(flags & CS_DC_SUSP))
 998		conn_pr_state_change(first_peer_device(device)->connection, os, ns,
 999				     (flags & ~CS_DC_MASK) | CS_DC_SUSP);
1000
1001	/* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
1002	 * on the ldev here, to be sure the transition -> D_DISKLESS resp.
1003	 * drbd_ldev_destroy() won't happen before our corresponding
1004	 * after_state_ch works run, where we put_ldev again. */
1005	if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
1006	    (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
1007		atomic_inc(&device->local_cnt);
1008
1009	did_remote = drbd_should_do_remote(device->state);
 
 
 
 
 
 
 
 
 
1010	device->state.i = ns.i;
1011	should_do_remote = drbd_should_do_remote(device->state);
1012	device->resource->susp = ns.susp;
1013	device->resource->susp_nod = ns.susp_nod;
1014	device->resource->susp_fen = ns.susp_fen;
 
 
 
1015
1016	/* put replicated vs not-replicated requests in seperate epochs */
1017	if (did_remote != should_do_remote)
1018		start_new_tl_epoch(first_peer_device(device)->connection);
 
1019
1020	if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
1021		drbd_print_uuids(device, "attached to UUIDs");
1022
1023	/* Wake up role changes, that were delayed because of connection establishing */
1024	if (os.conn == C_WF_REPORT_PARAMS && ns.conn != C_WF_REPORT_PARAMS &&
1025	    no_peer_wf_report_params(first_peer_device(device)->connection))
1026		clear_bit(STATE_SENT, &first_peer_device(device)->connection->flags);
 
 
1027
1028	wake_up(&device->misc_wait);
1029	wake_up(&device->state_wait);
1030	wake_up(&first_peer_device(device)->connection->ping_wait);
1031
1032	/* Aborted verify run, or we reached the stop sector.
1033	 * Log the last position, unless end-of-device. */
1034	if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
1035	    ns.conn <= C_CONNECTED) {
1036		device->ov_start_sector =
1037			BM_BIT_TO_SECT(drbd_bm_bits(device) - device->ov_left);
1038		if (device->ov_left)
1039			drbd_info(device, "Online Verify reached sector %llu\n",
1040				(unsigned long long)device->ov_start_sector);
1041	}
1042
1043	if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
1044	    (ns.conn == C_SYNC_TARGET  || ns.conn == C_SYNC_SOURCE)) {
1045		drbd_info(device, "Syncer continues.\n");
1046		device->rs_paused += (long)jiffies
1047				  -(long)device->rs_mark_time[device->rs_last_mark];
1048		if (ns.conn == C_SYNC_TARGET)
1049			mod_timer(&device->resync_timer, jiffies);
1050	}
1051
1052	if ((os.conn == C_SYNC_TARGET  || os.conn == C_SYNC_SOURCE) &&
1053	    (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
1054		drbd_info(device, "Resync suspended\n");
1055		device->rs_mark_time[device->rs_last_mark] = jiffies;
1056	}
1057
1058	if (os.conn == C_CONNECTED &&
1059	    (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
1060		unsigned long now = jiffies;
1061		int i;
1062
1063		set_ov_position(device, ns.conn);
1064		device->rs_start = now;
1065		device->rs_last_events = 0;
1066		device->rs_last_sect_ev = 0;
1067		device->ov_last_oos_size = 0;
1068		device->ov_last_oos_start = 0;
1069
1070		for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1071			device->rs_mark_left[i] = device->ov_left;
1072			device->rs_mark_time[i] = now;
1073		}
1074
1075		drbd_rs_controller_reset(device);
1076
1077		if (ns.conn == C_VERIFY_S) {
1078			drbd_info(device, "Starting Online Verify from sector %llu\n",
1079					(unsigned long long)device->ov_position);
1080			mod_timer(&device->resync_timer, jiffies);
1081		}
1082	}
1083
1084	if (get_ldev(device)) {
1085		u32 mdf = device->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
1086						 MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
1087						 MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
1088
1089		mdf &= ~MDF_AL_CLEAN;
1090		if (test_bit(CRASHED_PRIMARY, &device->flags))
1091			mdf |= MDF_CRASHED_PRIMARY;
1092		if (device->state.role == R_PRIMARY ||
1093		    (device->state.pdsk < D_INCONSISTENT && device->state.peer == R_PRIMARY))
1094			mdf |= MDF_PRIMARY_IND;
1095		if (device->state.conn > C_WF_REPORT_PARAMS)
1096			mdf |= MDF_CONNECTED_IND;
1097		if (device->state.disk > D_INCONSISTENT)
1098			mdf |= MDF_CONSISTENT;
1099		if (device->state.disk > D_OUTDATED)
1100			mdf |= MDF_WAS_UP_TO_DATE;
1101		if (device->state.pdsk <= D_OUTDATED && device->state.pdsk >= D_INCONSISTENT)
1102			mdf |= MDF_PEER_OUT_DATED;
1103		if (mdf != device->ldev->md.flags) {
1104			device->ldev->md.flags = mdf;
1105			drbd_md_mark_dirty(device);
1106		}
1107		if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
1108			drbd_set_ed_uuid(device, device->ldev->md.uuid[UI_CURRENT]);
1109		put_ldev(device);
1110	}
1111
1112	/* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
1113	if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
1114	    os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
1115		set_bit(CONSIDER_RESYNC, &device->flags);
1116
1117	/* Receiver should clean up itself */
1118	if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
1119		drbd_thread_stop_nowait(&first_peer_device(device)->connection->receiver);
1120
1121	/* Now the receiver finished cleaning up itself, it should die */
1122	if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
1123		drbd_thread_stop_nowait(&first_peer_device(device)->connection->receiver);
1124
1125	/* Upon network failure, we need to restart the receiver. */
1126	if (os.conn > C_WF_CONNECTION &&
1127	    ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
1128		drbd_thread_restart_nowait(&first_peer_device(device)->connection->receiver);
1129
1130	/* Resume AL writing if we get a connection */
1131	if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) {
1132		drbd_resume_al(device);
1133		first_peer_device(device)->connection->connect_cnt++;
1134	}
1135
1136	/* remember last attach time so request_timer_fn() won't
1137	 * kill newly established sessions while we are still trying to thaw
1138	 * previously frozen IO */
1139	if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1140	    ns.disk > D_NEGOTIATING)
1141		device->last_reattach_jif = jiffies;
1142
1143	ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
1144	if (ascw) {
1145		ascw->os = os;
1146		ascw->ns = ns;
1147		ascw->flags = flags;
1148		ascw->w.cb = w_after_state_ch;
1149		ascw->device = device;
1150		ascw->done = done;
1151		drbd_queue_work(&first_peer_device(device)->connection->sender_work,
 
1152				&ascw->w);
1153	} else {
1154		drbd_err(device, "Could not kmalloc an ascw\n");
1155	}
1156
1157	return rv;
1158}
1159
1160static int w_after_state_ch(struct drbd_work *w, int unused)
1161{
1162	struct after_state_chg_work *ascw =
1163		container_of(w, struct after_state_chg_work, w);
1164	struct drbd_device *device = ascw->device;
1165
1166	after_state_ch(device, ascw->os, ascw->ns, ascw->flags);
 
1167	if (ascw->flags & CS_WAIT_COMPLETE)
1168		complete(ascw->done);
1169	kfree(ascw);
1170
1171	return 0;
1172}
1173
1174static void abw_start_sync(struct drbd_device *device, int rv)
1175{
1176	if (rv) {
1177		drbd_err(device, "Writing the bitmap failed not starting resync.\n");
1178		_drbd_request_state(device, NS(conn, C_CONNECTED), CS_VERBOSE);
1179		return;
1180	}
1181
1182	switch (device->state.conn) {
1183	case C_STARTING_SYNC_T:
1184		_drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
1185		break;
1186	case C_STARTING_SYNC_S:
1187		drbd_start_resync(device, C_SYNC_SOURCE);
1188		break;
1189	}
1190}
1191
1192int drbd_bitmap_io_from_worker(struct drbd_device *device,
1193		int (*io_fn)(struct drbd_device *),
1194		char *why, enum bm_flag flags)
1195{
1196	int rv;
1197
1198	D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
1199
1200	/* open coded non-blocking drbd_suspend_io(device); */
1201	set_bit(SUSPEND_IO, &device->flags);
1202
1203	drbd_bm_lock(device, why, flags);
1204	rv = io_fn(device);
1205	drbd_bm_unlock(device);
1206
1207	drbd_resume_io(device);
1208
1209	return rv;
1210}
1211
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1212/**
1213 * after_state_ch() - Perform after state change actions that may sleep
1214 * @device:	DRBD device.
1215 * @os:		old state.
1216 * @ns:		new state.
1217 * @flags:	Flags
1218 */
1219static void after_state_ch(struct drbd_device *device, union drbd_state os,
1220			   union drbd_state ns, enum chg_state_flags flags)
 
1221{
1222	struct drbd_resource *resource = device->resource;
 
 
1223	struct sib_info sib;
1224
 
 
1225	sib.sib_reason = SIB_STATE_CHANGE;
1226	sib.os = os;
1227	sib.ns = ns;
1228
1229	if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) {
 
1230		clear_bit(CRASHED_PRIMARY, &device->flags);
1231		if (device->p_uuid)
1232			device->p_uuid[UI_FLAGS] &= ~((u64)2);
1233	}
1234
1235	/* Inform userspace about the change... */
1236	drbd_bcast_event(device, &sib);
1237
1238	if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
1239	    (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
1240		drbd_khelper(device, "pri-on-incon-degr");
1241
1242	/* Here we have the actions that are performed after a
1243	   state change. This function might sleep */
1244
1245	if (ns.susp_nod) {
1246		struct drbd_connection *connection = first_peer_device(device)->connection;
1247		enum drbd_req_event what = NOTHING;
1248
1249		spin_lock_irq(&device->resource->req_lock);
1250		if (os.conn < C_CONNECTED && conn_lowest_conn(connection) >= C_CONNECTED)
1251			what = RESEND;
1252
1253		if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1254		    conn_lowest_disk(connection) > D_NEGOTIATING)
1255			what = RESTART_FROZEN_DISK_IO;
1256
1257		if (resource->susp_nod && what != NOTHING) {
1258			_tl_restart(connection, what);
1259			_conn_request_state(connection,
1260					    (union drbd_state) { { .susp_nod = 1 } },
1261					    (union drbd_state) { { .susp_nod = 0 } },
1262					    CS_VERBOSE);
1263		}
1264		spin_unlock_irq(&device->resource->req_lock);
1265	}
1266
1267	if (ns.susp_fen) {
1268		struct drbd_connection *connection = first_peer_device(device)->connection;
1269
1270		spin_lock_irq(&device->resource->req_lock);
1271		if (resource->susp_fen && conn_lowest_conn(connection) >= C_CONNECTED) {
1272			/* case2: The connection was established again: */
1273			struct drbd_peer_device *peer_device;
1274			int vnr;
1275
1276			rcu_read_lock();
1277			idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
1278				clear_bit(NEW_CUR_UUID, &peer_device->device->flags);
1279			rcu_read_unlock();
 
 
 
 
 
 
 
1280			_tl_restart(connection, RESEND);
1281			_conn_request_state(connection,
1282					    (union drbd_state) { { .susp_fen = 1 } },
1283					    (union drbd_state) { { .susp_fen = 0 } },
1284					    CS_VERBOSE);
1285		}
1286		spin_unlock_irq(&device->resource->req_lock);
1287	}
1288
1289	/* Became sync source.  With protocol >= 96, we still need to send out
1290	 * the sync uuid now. Need to do that before any drbd_send_state, or
1291	 * the other side may go "paused sync" before receiving the sync uuids,
1292	 * which is unexpected. */
1293	if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
1294	    (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
1295	    first_peer_device(device)->connection->agreed_pro_version >= 96 && get_ldev(device)) {
1296		drbd_gen_and_send_sync_uuid(first_peer_device(device));
1297		put_ldev(device);
1298	}
1299
1300	/* Do not change the order of the if above and the two below... */
1301	if (os.pdsk == D_DISKLESS &&
1302	    ns.pdsk > D_DISKLESS && ns.pdsk != D_UNKNOWN) {      /* attach on the peer */
1303		/* we probably will start a resync soon.
1304		 * make sure those things are properly reset. */
1305		device->rs_total = 0;
1306		device->rs_failed = 0;
1307		atomic_set(&device->rs_pending_cnt, 0);
1308		drbd_rs_cancel_all(device);
1309
1310		drbd_send_uuids(first_peer_device(device));
1311		drbd_send_state(first_peer_device(device), ns);
1312	}
1313	/* No point in queuing send_bitmap if we don't have a connection
1314	 * anymore, so check also the _current_ state, not only the new state
1315	 * at the time this work was queued. */
1316	if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
1317	    device->state.conn == C_WF_BITMAP_S)
1318		drbd_queue_bitmap_io(device, &drbd_send_bitmap, NULL,
1319				"send_bitmap (WFBitMapS)",
1320				BM_LOCKED_TEST_ALLOWED);
1321
1322	/* Lost contact to peer's copy of the data */
1323	if ((os.pdsk >= D_INCONSISTENT &&
1324	     os.pdsk != D_UNKNOWN &&
1325	     os.pdsk != D_OUTDATED)
1326	&&  (ns.pdsk < D_INCONSISTENT ||
1327	     ns.pdsk == D_UNKNOWN ||
1328	     ns.pdsk == D_OUTDATED)) {
1329		if (get_ldev(device)) {
1330			if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1331			    device->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1332				if (drbd_suspended(device)) {
1333					set_bit(NEW_CUR_UUID, &device->flags);
1334				} else {
1335					drbd_uuid_new_current(device);
1336					drbd_send_uuids(first_peer_device(device));
1337				}
1338			}
1339			put_ldev(device);
1340		}
1341	}
1342
1343	if (ns.pdsk < D_INCONSISTENT && get_ldev(device)) {
1344		if (os.peer == R_SECONDARY && ns.peer == R_PRIMARY &&
1345		    device->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1346			drbd_uuid_new_current(device);
1347			drbd_send_uuids(first_peer_device(device));
1348		}
1349		/* D_DISKLESS Peer becomes secondary */
1350		if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1351			/* We may still be Primary ourselves.
1352			 * No harm done if the bitmap still changes,
1353			 * redirtied pages will follow later. */
1354			drbd_bitmap_io_from_worker(device, &drbd_bm_write,
1355				"demote diskless peer", BM_LOCKED_SET_ALLOWED);
1356		put_ldev(device);
1357	}
1358
1359	/* Write out all changed bits on demote.
1360	 * Though, no need to da that just yet
1361	 * if there is a resync going on still */
1362	if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
1363		device->state.conn <= C_CONNECTED && get_ldev(device)) {
1364		/* No changes to the bitmap expected this time, so assert that,
1365		 * even though no harm was done if it did change. */
1366		drbd_bitmap_io_from_worker(device, &drbd_bm_write,
1367				"demote", BM_LOCKED_TEST_ALLOWED);
1368		put_ldev(device);
1369	}
1370
1371	/* Last part of the attaching process ... */
1372	if (ns.conn >= C_CONNECTED &&
1373	    os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1374		drbd_send_sizes(first_peer_device(device), 0, 0);  /* to start sync... */
1375		drbd_send_uuids(first_peer_device(device));
1376		drbd_send_state(first_peer_device(device), ns);
1377	}
1378
1379	/* We want to pause/continue resync, tell peer. */
1380	if (ns.conn >= C_CONNECTED &&
1381	     ((os.aftr_isp != ns.aftr_isp) ||
1382	      (os.user_isp != ns.user_isp)))
1383		drbd_send_state(first_peer_device(device), ns);
1384
1385	/* In case one of the isp bits got set, suspend other devices. */
1386	if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1387	    (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1388		suspend_other_sg(device);
1389
1390	/* Make sure the peer gets informed about eventual state
1391	   changes (ISP bits) while we were in WFReportParams. */
1392	if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1393		drbd_send_state(first_peer_device(device), ns);
1394
1395	if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
1396		drbd_send_state(first_peer_device(device), ns);
1397
1398	/* We are in the progress to start a full sync... */
1399	if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1400	    (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1401		/* no other bitmap changes expected during this phase */
1402		drbd_queue_bitmap_io(device,
1403			&drbd_bmio_set_n_write, &abw_start_sync,
1404			"set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED);
1405
1406	/* first half of local IO error, failure to attach,
1407	 * or administrative detach */
1408	if (os.disk != D_FAILED && ns.disk == D_FAILED) {
1409		enum drbd_io_error_p eh = EP_PASS_ON;
1410		int was_io_error = 0;
1411		/* corresponding get_ldev was in __drbd_set_state, to serialize
1412		 * our cleanup here with the transition to D_DISKLESS.
1413		 * But is is still not save to dreference ldev here, since
1414		 * we might come from an failed Attach before ldev was set. */
1415		if (device->ldev) {
1416			rcu_read_lock();
1417			eh = rcu_dereference(device->ldev->disk_conf)->on_io_error;
1418			rcu_read_unlock();
1419
1420			was_io_error = test_and_clear_bit(WAS_IO_ERROR, &device->flags);
1421
 
 
 
 
1422			if (was_io_error && eh == EP_CALL_HELPER)
1423				drbd_khelper(device, "local-io-error");
1424
1425			/* Immediately allow completion of all application IO,
1426			 * that waits for completion from the local disk,
1427			 * if this was a force-detach due to disk_timeout
1428			 * or administrator request (drbdsetup detach --force).
1429			 * Do NOT abort otherwise.
1430			 * Aborting local requests may cause serious problems,
1431			 * if requests are completed to upper layers already,
1432			 * and then later the already submitted local bio completes.
1433			 * This can cause DMA into former bio pages that meanwhile
1434			 * have been re-used for other things.
1435			 * So aborting local requests may cause crashes,
1436			 * or even worse, silent data corruption.
1437			 */
1438			if (test_and_clear_bit(FORCE_DETACH, &device->flags))
1439				tl_abort_disk_io(device);
1440
1441			/* current state still has to be D_FAILED,
1442			 * there is only one way out: to D_DISKLESS,
1443			 * and that may only happen after our put_ldev below. */
1444			if (device->state.disk != D_FAILED)
1445				drbd_err(device,
1446					"ASSERT FAILED: disk is %s during detach\n",
1447					drbd_disk_str(device->state.disk));
1448
1449			if (ns.conn >= C_CONNECTED)
1450				drbd_send_state(first_peer_device(device), ns);
1451
1452			drbd_rs_cancel_all(device);
1453
1454			/* In case we want to get something to stable storage still,
1455			 * this may be the last chance.
1456			 * Following put_ldev may transition to D_DISKLESS. */
1457			drbd_md_sync(device);
1458		}
1459		put_ldev(device);
1460	}
1461
1462	/* second half of local IO error, failure to attach,
1463	 * or administrative detach,
1464	 * after local_cnt references have reached zero again */
1465	if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
1466		/* We must still be diskless,
1467		 * re-attach has to be serialized with this! */
1468		if (device->state.disk != D_DISKLESS)
1469			drbd_err(device,
1470				 "ASSERT FAILED: disk is %s while going diskless\n",
1471				 drbd_disk_str(device->state.disk));
1472
1473		if (ns.conn >= C_CONNECTED)
1474			drbd_send_state(first_peer_device(device), ns);
1475		/* corresponding get_ldev in __drbd_set_state
1476		 * this may finally trigger drbd_ldev_destroy. */
1477		put_ldev(device);
1478	}
1479
1480	/* Notify peer that I had a local IO error, and did not detached.. */
1481	if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT && ns.conn >= C_CONNECTED)
1482		drbd_send_state(first_peer_device(device), ns);
1483
1484	/* Disks got bigger while they were detached */
1485	if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1486	    test_and_clear_bit(RESYNC_AFTER_NEG, &device->flags)) {
1487		if (ns.conn == C_CONNECTED)
1488			resync_after_online_grow(device);
1489	}
1490
1491	/* A resync finished or aborted, wake paused devices... */
1492	if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1493	    (os.peer_isp && !ns.peer_isp) ||
1494	    (os.user_isp && !ns.user_isp))
1495		resume_next_sg(device);
1496
1497	/* sync target done with resync.  Explicitly notify peer, even though
1498	 * it should (at least for non-empty resyncs) already know itself. */
1499	if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
1500		drbd_send_state(first_peer_device(device), ns);
1501
1502	/* Verify finished, or reached stop sector.  Peer did not know about
1503	 * the stop sector, and we may even have changed the stop sector during
1504	 * verify to interrupt/stop early.  Send the new state. */
1505	if (os.conn == C_VERIFY_S && ns.conn == C_CONNECTED
1506	&& verify_can_do_stop_sector(device))
1507		drbd_send_state(first_peer_device(device), ns);
1508
1509	/* This triggers bitmap writeout of potentially still unwritten pages
1510	 * if the resync finished cleanly, or aborted because of peer disk
1511	 * failure, or because of connection loss.
 
 
 
1512	 * For resync aborted because of local disk failure, we cannot do
1513	 * any bitmap writeout anymore.
 
1514	 * No harm done if some bits change during this phase.
1515	 */
1516	if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(device)) {
 
1517		drbd_queue_bitmap_io(device, &drbd_bm_write_copy_pages, NULL,
1518			"write from resync_finished", BM_LOCKED_CHANGE_ALLOWED);
1519		put_ldev(device);
1520	}
1521
1522	if (ns.disk == D_DISKLESS &&
1523	    ns.conn == C_STANDALONE &&
1524	    ns.role == R_SECONDARY) {
1525		if (os.aftr_isp != ns.aftr_isp)
1526			resume_next_sg(device);
1527	}
1528
1529	drbd_md_sync(device);
1530}
1531
1532struct after_conn_state_chg_work {
1533	struct drbd_work w;
1534	enum drbd_conns oc;
1535	union drbd_state ns_min;
1536	union drbd_state ns_max; /* new, max state, over all devices */
1537	enum chg_state_flags flags;
1538	struct drbd_connection *connection;
 
1539};
1540
1541static int w_after_conn_state_ch(struct drbd_work *w, int unused)
1542{
1543	struct after_conn_state_chg_work *acscw =
1544		container_of(w, struct after_conn_state_chg_work, w);
1545	struct drbd_connection *connection = acscw->connection;
1546	enum drbd_conns oc = acscw->oc;
1547	union drbd_state ns_max = acscw->ns_max;
1548	struct drbd_peer_device *peer_device;
1549	int vnr;
1550
 
 
1551	kfree(acscw);
1552
1553	/* Upon network configuration, we need to start the receiver */
1554	if (oc == C_STANDALONE && ns_max.conn == C_UNCONNECTED)
1555		drbd_thread_start(&connection->receiver);
1556
1557	if (oc == C_DISCONNECTING && ns_max.conn == C_STANDALONE) {
1558		struct net_conf *old_conf;
1559
 
 
 
 
 
 
 
1560		mutex_lock(&connection->resource->conf_update);
1561		old_conf = connection->net_conf;
1562		connection->my_addr_len = 0;
1563		connection->peer_addr_len = 0;
1564		rcu_assign_pointer(connection->net_conf, NULL);
1565		conn_free_crypto(connection);
1566		mutex_unlock(&connection->resource->conf_update);
1567
1568		synchronize_rcu();
1569		kfree(old_conf);
1570	}
1571
1572	if (ns_max.susp_fen) {
1573		/* case1: The outdate peer handler is successful: */
1574		if (ns_max.pdsk <= D_OUTDATED) {
1575			rcu_read_lock();
1576			idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
1577				struct drbd_device *device = peer_device->device;
1578				if (test_bit(NEW_CUR_UUID, &device->flags)) {
1579					drbd_uuid_new_current(device);
1580					clear_bit(NEW_CUR_UUID, &device->flags);
1581				}
1582			}
1583			rcu_read_unlock();
1584			spin_lock_irq(&connection->resource->req_lock);
1585			_tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
1586			_conn_request_state(connection,
1587					    (union drbd_state) { { .susp_fen = 1 } },
1588					    (union drbd_state) { { .susp_fen = 0 } },
1589					    CS_VERBOSE);
1590			spin_unlock_irq(&connection->resource->req_lock);
1591		}
1592	}
1593	kref_put(&connection->kref, drbd_destroy_connection);
1594
1595	conn_md_sync(connection);
1596
1597	return 0;
1598}
1599
1600void conn_old_common_state(struct drbd_connection *connection, union drbd_state *pcs, enum chg_state_flags *pf)
1601{
1602	enum chg_state_flags flags = ~0;
1603	struct drbd_peer_device *peer_device;
1604	int vnr, first_vol = 1;
1605	union drbd_dev_state os, cs = {
1606		{ .role = R_SECONDARY,
1607		  .peer = R_UNKNOWN,
1608		  .conn = connection->cstate,
1609		  .disk = D_DISKLESS,
1610		  .pdsk = D_UNKNOWN,
1611		} };
1612
1613	rcu_read_lock();
1614	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
1615		struct drbd_device *device = peer_device->device;
1616		os = device->state;
1617
1618		if (first_vol) {
1619			cs = os;
1620			first_vol = 0;
1621			continue;
1622		}
1623
1624		if (cs.role != os.role)
1625			flags &= ~CS_DC_ROLE;
1626
1627		if (cs.peer != os.peer)
1628			flags &= ~CS_DC_PEER;
1629
1630		if (cs.conn != os.conn)
1631			flags &= ~CS_DC_CONN;
1632
1633		if (cs.disk != os.disk)
1634			flags &= ~CS_DC_DISK;
1635
1636		if (cs.pdsk != os.pdsk)
1637			flags &= ~CS_DC_PDSK;
1638	}
1639	rcu_read_unlock();
1640
1641	*pf |= CS_DC_MASK;
1642	*pf &= flags;
1643	(*pcs).i = cs.i;
1644}
1645
1646static enum drbd_state_rv
1647conn_is_valid_transition(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
1648			 enum chg_state_flags flags)
1649{
1650	enum drbd_state_rv rv = SS_SUCCESS;
1651	union drbd_state ns, os;
1652	struct drbd_peer_device *peer_device;
1653	int vnr;
1654
1655	rcu_read_lock();
1656	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
1657		struct drbd_device *device = peer_device->device;
1658		os = drbd_read_state(device);
1659		ns = sanitize_state(device, apply_mask_val(os, mask, val), NULL);
1660
1661		if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
1662			ns.disk = os.disk;
1663
1664		if (ns.i == os.i)
1665			continue;
1666
1667		rv = is_valid_transition(os, ns);
1668
1669		if (rv >= SS_SUCCESS && !(flags & CS_HARD)) {
1670			rv = is_valid_state(device, ns);
1671			if (rv < SS_SUCCESS) {
1672				if (is_valid_state(device, os) == rv)
1673					rv = is_valid_soft_transition(os, ns, connection);
1674			} else
1675				rv = is_valid_soft_transition(os, ns, connection);
1676		}
1677
1678		if (rv < SS_SUCCESS) {
1679			if (flags & CS_VERBOSE)
1680				print_st_err(device, os, ns, rv);
1681			break;
1682		}
1683	}
1684	rcu_read_unlock();
1685
1686	return rv;
1687}
1688
1689void
1690conn_set_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
1691	       union drbd_state *pns_min, union drbd_state *pns_max, enum chg_state_flags flags)
1692{
1693	union drbd_state ns, os, ns_max = { };
1694	union drbd_state ns_min = {
1695		{ .role = R_MASK,
1696		  .peer = R_MASK,
1697		  .conn = val.conn,
1698		  .disk = D_MASK,
1699		  .pdsk = D_MASK
1700		} };
1701	struct drbd_peer_device *peer_device;
1702	enum drbd_state_rv rv;
1703	int vnr, number_of_volumes = 0;
1704
1705	if (mask.conn == C_MASK) {
1706		/* remember last connect time so request_timer_fn() won't
1707		 * kill newly established sessions while we are still trying to thaw
1708		 * previously frozen IO */
1709		if (connection->cstate != C_WF_REPORT_PARAMS && val.conn == C_WF_REPORT_PARAMS)
1710			connection->last_reconnect_jif = jiffies;
1711
1712		connection->cstate = val.conn;
1713	}
1714
1715	rcu_read_lock();
1716	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
1717		struct drbd_device *device = peer_device->device;
1718		number_of_volumes++;
1719		os = drbd_read_state(device);
1720		ns = apply_mask_val(os, mask, val);
1721		ns = sanitize_state(device, ns, NULL);
1722
1723		if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
1724			ns.disk = os.disk;
1725
1726		rv = __drbd_set_state(device, ns, flags, NULL);
1727		if (rv < SS_SUCCESS)
1728			BUG();
1729
1730		ns.i = device->state.i;
1731		ns_max.role = max_role(ns.role, ns_max.role);
1732		ns_max.peer = max_role(ns.peer, ns_max.peer);
1733		ns_max.conn = max_t(enum drbd_conns, ns.conn, ns_max.conn);
1734		ns_max.disk = max_t(enum drbd_disk_state, ns.disk, ns_max.disk);
1735		ns_max.pdsk = max_t(enum drbd_disk_state, ns.pdsk, ns_max.pdsk);
1736
1737		ns_min.role = min_role(ns.role, ns_min.role);
1738		ns_min.peer = min_role(ns.peer, ns_min.peer);
1739		ns_min.conn = min_t(enum drbd_conns, ns.conn, ns_min.conn);
1740		ns_min.disk = min_t(enum drbd_disk_state, ns.disk, ns_min.disk);
1741		ns_min.pdsk = min_t(enum drbd_disk_state, ns.pdsk, ns_min.pdsk);
1742	}
1743	rcu_read_unlock();
1744
1745	if (number_of_volumes == 0) {
1746		ns_min = ns_max = (union drbd_state) { {
1747				.role = R_SECONDARY,
1748				.peer = R_UNKNOWN,
1749				.conn = val.conn,
1750				.disk = D_DISKLESS,
1751				.pdsk = D_UNKNOWN
1752			} };
1753	}
1754
1755	ns_min.susp = ns_max.susp = connection->resource->susp;
1756	ns_min.susp_nod = ns_max.susp_nod = connection->resource->susp_nod;
1757	ns_min.susp_fen = ns_max.susp_fen = connection->resource->susp_fen;
1758
1759	*pns_min = ns_min;
1760	*pns_max = ns_max;
1761}
1762
1763static enum drbd_state_rv
1764_conn_rq_cond(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1765{
1766	enum drbd_state_rv rv;
1767
1768	if (test_and_clear_bit(CONN_WD_ST_CHG_OKAY, &connection->flags))
1769		return SS_CW_SUCCESS;
1770
1771	if (test_and_clear_bit(CONN_WD_ST_CHG_FAIL, &connection->flags))
1772		return SS_CW_FAILED_BY_PEER;
1773
1774	rv = conn_is_valid_transition(connection, mask, val, 0);
1775	if (rv == SS_SUCCESS && connection->cstate == C_WF_REPORT_PARAMS)
1776		rv = SS_UNKNOWN_ERROR; /* continue waiting */
1777
1778	return rv;
1779}
1780
1781enum drbd_state_rv
1782_conn_request_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
1783		    enum chg_state_flags flags)
1784{
1785	enum drbd_state_rv rv = SS_SUCCESS;
1786	struct after_conn_state_chg_work *acscw;
1787	enum drbd_conns oc = connection->cstate;
1788	union drbd_state ns_max, ns_min, os;
1789	bool have_mutex = false;
 
1790
1791	if (mask.conn) {
1792		rv = is_valid_conn_transition(oc, val.conn);
1793		if (rv < SS_SUCCESS)
1794			goto abort;
1795	}
1796
1797	rv = conn_is_valid_transition(connection, mask, val, flags);
1798	if (rv < SS_SUCCESS)
1799		goto abort;
1800
1801	if (oc == C_WF_REPORT_PARAMS && val.conn == C_DISCONNECTING &&
1802	    !(flags & (CS_LOCAL_ONLY | CS_HARD))) {
1803
1804		/* This will be a cluster-wide state change.
1805		 * Need to give up the spinlock, grab the mutex,
1806		 * then send the state change request, ... */
1807		spin_unlock_irq(&connection->resource->req_lock);
1808		mutex_lock(&connection->cstate_mutex);
1809		have_mutex = true;
1810
1811		set_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
1812		if (conn_send_state_req(connection, mask, val)) {
1813			/* sending failed. */
1814			clear_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
1815			rv = SS_CW_FAILED_BY_PEER;
1816			/* need to re-aquire the spin lock, though */
1817			goto abort_unlocked;
1818		}
1819
1820		if (val.conn == C_DISCONNECTING)
1821			set_bit(DISCONNECT_SENT, &connection->flags);
1822
1823		/* ... and re-aquire the spinlock.
1824		 * If _conn_rq_cond() returned >= SS_SUCCESS, we must call
1825		 * conn_set_state() within the same spinlock. */
1826		spin_lock_irq(&connection->resource->req_lock);
1827		wait_event_lock_irq(connection->ping_wait,
1828				(rv = _conn_rq_cond(connection, mask, val)),
1829				connection->resource->req_lock);
1830		clear_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
1831		if (rv < SS_SUCCESS)
1832			goto abort;
1833	}
1834
 
1835	conn_old_common_state(connection, &os, &flags);
1836	flags |= CS_DC_SUSP;
1837	conn_set_state(connection, mask, val, &ns_min, &ns_max, flags);
1838	conn_pr_state_change(connection, os, ns_max, flags);
 
1839
1840	acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC);
1841	if (acscw) {
1842		acscw->oc = os.conn;
1843		acscw->ns_min = ns_min;
1844		acscw->ns_max = ns_max;
1845		acscw->flags = flags;
1846		acscw->w.cb = w_after_conn_state_ch;
1847		kref_get(&connection->kref);
1848		acscw->connection = connection;
 
1849		drbd_queue_work(&connection->sender_work, &acscw->w);
1850	} else {
1851		drbd_err(connection, "Could not kmalloc an acscw\n");
1852	}
1853
1854 abort:
1855	if (have_mutex) {
1856		/* mutex_unlock() "... must not be used in interrupt context.",
1857		 * so give up the spinlock, then re-aquire it */
1858		spin_unlock_irq(&connection->resource->req_lock);
1859 abort_unlocked:
1860		mutex_unlock(&connection->cstate_mutex);
1861		spin_lock_irq(&connection->resource->req_lock);
1862	}
1863	if (rv < SS_SUCCESS && flags & CS_VERBOSE) {
1864		drbd_err(connection, "State change failed: %s\n", drbd_set_st_err_str(rv));
1865		drbd_err(connection, " mask = 0x%x val = 0x%x\n", mask.i, val.i);
1866		drbd_err(connection, " old_conn:%s wanted_conn:%s\n", drbd_conn_str(oc), drbd_conn_str(val.conn));
1867	}
1868	return rv;
1869}
1870
1871enum drbd_state_rv
1872conn_request_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
1873		   enum chg_state_flags flags)
1874{
1875	enum drbd_state_rv rv;
1876
1877	spin_lock_irq(&connection->resource->req_lock);
1878	rv = _conn_request_state(connection, mask, val, flags);
1879	spin_unlock_irq(&connection->resource->req_lock);
1880
1881	return rv;
1882}