xref: /linux/drivers/block/drbd/drbd_state.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN)
574 		rv = SS_CONNECTED_OUTDATES;
575 
576 	rcu_read_unlock();
577 
578 	return rv;
579 }
580 
581 /**
582  * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible
583  * This function limits state transitions that may be declined by DRBD. I.e.
584  * user requests (aka soft transitions).
585  * @mdev:	DRBD device.
586  * @ns:		new state.
587  * @os:		old state.
588  */
589 static enum drbd_state_rv
590 is_valid_soft_transition(union drbd_state os, union drbd_state ns, struct drbd_tconn *tconn)
591 {
592 	enum drbd_state_rv rv = SS_SUCCESS;
593 
594 	if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
595 	    os.conn > C_CONNECTED)
596 		rv = SS_RESYNC_RUNNING;
597 
598 	if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
599 		rv = SS_ALREADY_STANDALONE;
600 
601 	if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
602 		rv = SS_IS_DISKLESS;
603 
604 	if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
605 		rv = SS_NO_NET_CONFIG;
606 
607 	if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
608 		rv = SS_LOWER_THAN_OUTDATED;
609 
610 	if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
611 		rv = SS_IN_TRANSIENT_STATE;
612 
613 	/* if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS)
614 	   rv = SS_IN_TRANSIENT_STATE; */
615 
616 	/* While establishing a connection only allow cstate to change.
617 	   Delay/refuse role changes, detach attach etc... */
618 	if (test_bit(STATE_SENT, &tconn->flags) &&
619 	    !(os.conn == C_WF_REPORT_PARAMS ||
620 	      (ns.conn == C_WF_REPORT_PARAMS && os.conn == C_WF_CONNECTION)))
621 		rv = SS_IN_TRANSIENT_STATE;
622 
623 	if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
624 		rv = SS_NEED_CONNECTION;
625 
626 	if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
627 	    ns.conn != os.conn && os.conn > C_CONNECTED)
628 		rv = SS_RESYNC_RUNNING;
629 
630 	if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
631 	    os.conn < C_CONNECTED)
632 		rv = SS_NEED_CONNECTION;
633 
634 	if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)
635 	    && os.conn < C_WF_REPORT_PARAMS)
636 		rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */
637 
638 	return rv;
639 }
640 
641 static enum drbd_state_rv
642 is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc)
643 {
644 	/* no change -> nothing to do, at least for the connection part */
645 	if (oc == nc)
646 		return SS_NOTHING_TO_DO;
647 
648 	/* disconnect of an unconfigured connection does not make sense */
649 	if (oc == C_STANDALONE && nc == C_DISCONNECTING)
650 		return SS_ALREADY_STANDALONE;
651 
652 	/* from C_STANDALONE, we start with C_UNCONNECTED */
653 	if (oc == C_STANDALONE && nc != C_UNCONNECTED)
654 		return SS_NEED_CONNECTION;
655 
656 	/* When establishing a connection we need to go through WF_REPORT_PARAMS!
657 	   Necessary to do the right thing upon invalidate-remote on a disconnected resource */
658 	if (oc < C_WF_REPORT_PARAMS && nc >= C_CONNECTED)
659 		return SS_NEED_CONNECTION;
660 
661 	/* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */
662 	if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING)
663 		return SS_IN_TRANSIENT_STATE;
664 
665 	/* After C_DISCONNECTING only C_STANDALONE may follow */
666 	if (oc == C_DISCONNECTING && nc != C_STANDALONE)
667 		return SS_IN_TRANSIENT_STATE;
668 
669 	return SS_SUCCESS;
670 }
671 
672 
673 /**
674  * is_valid_transition() - Returns an SS_ error code if the state transition is not possible
675  * This limits hard state transitions. Hard state transitions are facts there are
676  * imposed on DRBD by the environment. E.g. disk broke or network broke down.
677  * But those hard state transitions are still not allowed to do everything.
678  * @ns:		new state.
679  * @os:		old state.
680  */
681 static enum drbd_state_rv
682 is_valid_transition(union drbd_state os, union drbd_state ns)
683 {
684 	enum drbd_state_rv rv;
685 
686 	rv = is_valid_conn_transition(os.conn, ns.conn);
687 
688 	/* we cannot fail (again) if we already detached */
689 	if (ns.disk == D_FAILED && os.disk == D_DISKLESS)
690 		rv = SS_IS_DISKLESS;
691 
692 	return rv;
693 }
694 
695 static void print_sanitize_warnings(struct drbd_conf *mdev, enum sanitize_state_warnings warn)
696 {
697 	static const char *msg_table[] = {
698 		[NO_WARNING] = "",
699 		[ABORTED_ONLINE_VERIFY] = "Online-verify aborted.",
700 		[ABORTED_RESYNC] = "Resync aborted.",
701 		[CONNECTION_LOST_NEGOTIATING] = "Connection lost while negotiating, no data!",
702 		[IMPLICITLY_UPGRADED_DISK] = "Implicitly upgraded disk",
703 		[IMPLICITLY_UPGRADED_PDSK] = "Implicitly upgraded pdsk",
704 	};
705 
706 	if (warn != NO_WARNING)
707 		dev_warn(DEV, "%s\n", msg_table[warn]);
708 }
709 
710 /**
711  * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
712  * @mdev:	DRBD device.
713  * @os:		old state.
714  * @ns:		new state.
715  * @warn_sync_abort:
716  *
717  * When we loose connection, we have to set the state of the peers disk (pdsk)
718  * to D_UNKNOWN. This rule and many more along those lines are in this function.
719  */
720 static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns,
721 				       enum sanitize_state_warnings *warn)
722 {
723 	enum drbd_fencing_p fp;
724 	enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max;
725 
726 	if (warn)
727 		*warn = NO_WARNING;
728 
729 	fp = FP_DONT_CARE;
730 	if (get_ldev(mdev)) {
731 		rcu_read_lock();
732 		fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
733 		rcu_read_unlock();
734 		put_ldev(mdev);
735 	}
736 
737 	/* Implications from connection to peer and peer_isp */
738 	if (ns.conn < C_CONNECTED) {
739 		ns.peer_isp = 0;
740 		ns.peer = R_UNKNOWN;
741 		if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
742 			ns.pdsk = D_UNKNOWN;
743 	}
744 
745 	/* Clear the aftr_isp when becoming unconfigured */
746 	if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
747 		ns.aftr_isp = 0;
748 
749 	/* An implication of the disk states onto the connection state */
750 	/* Abort resync if a disk fails/detaches */
751 	if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
752 		if (warn)
753 			*warn = ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ?
754 				ABORTED_ONLINE_VERIFY : ABORTED_RESYNC;
755 		ns.conn = C_CONNECTED;
756 	}
757 
758 	/* Connection breaks down before we finished "Negotiating" */
759 	if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
760 	    get_ldev_if_state(mdev, D_NEGOTIATING)) {
761 		if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) {
762 			ns.disk = mdev->new_state_tmp.disk;
763 			ns.pdsk = mdev->new_state_tmp.pdsk;
764 		} else {
765 			if (warn)
766 				*warn = CONNECTION_LOST_NEGOTIATING;
767 			ns.disk = D_DISKLESS;
768 			ns.pdsk = D_UNKNOWN;
769 		}
770 		put_ldev(mdev);
771 	}
772 
773 	/* D_CONSISTENT and D_OUTDATED vanish when we get connected */
774 	if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) {
775 		if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED)
776 			ns.disk = D_UP_TO_DATE;
777 		if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)
778 			ns.pdsk = D_UP_TO_DATE;
779 	}
780 
781 	/* Implications of the connection stat on the disk states */
782 	disk_min = D_DISKLESS;
783 	disk_max = D_UP_TO_DATE;
784 	pdsk_min = D_INCONSISTENT;
785 	pdsk_max = D_UNKNOWN;
786 	switch ((enum drbd_conns)ns.conn) {
787 	case C_WF_BITMAP_T:
788 	case C_PAUSED_SYNC_T:
789 	case C_STARTING_SYNC_T:
790 	case C_WF_SYNC_UUID:
791 	case C_BEHIND:
792 		disk_min = D_INCONSISTENT;
793 		disk_max = D_OUTDATED;
794 		pdsk_min = D_UP_TO_DATE;
795 		pdsk_max = D_UP_TO_DATE;
796 		break;
797 	case C_VERIFY_S:
798 	case C_VERIFY_T:
799 		disk_min = D_UP_TO_DATE;
800 		disk_max = D_UP_TO_DATE;
801 		pdsk_min = D_UP_TO_DATE;
802 		pdsk_max = D_UP_TO_DATE;
803 		break;
804 	case C_CONNECTED:
805 		disk_min = D_DISKLESS;
806 		disk_max = D_UP_TO_DATE;
807 		pdsk_min = D_DISKLESS;
808 		pdsk_max = D_UP_TO_DATE;
809 		break;
810 	case C_WF_BITMAP_S:
811 	case C_PAUSED_SYNC_S:
812 	case C_STARTING_SYNC_S:
813 	case C_AHEAD:
814 		disk_min = D_UP_TO_DATE;
815 		disk_max = D_UP_TO_DATE;
816 		pdsk_min = D_INCONSISTENT;
817 		pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/
818 		break;
819 	case C_SYNC_TARGET:
820 		disk_min = D_INCONSISTENT;
821 		disk_max = D_INCONSISTENT;
822 		pdsk_min = D_UP_TO_DATE;
823 		pdsk_max = D_UP_TO_DATE;
824 		break;
825 	case C_SYNC_SOURCE:
826 		disk_min = D_UP_TO_DATE;
827 		disk_max = D_UP_TO_DATE;
828 		pdsk_min = D_INCONSISTENT;
829 		pdsk_max = D_INCONSISTENT;
830 		break;
831 	case C_STANDALONE:
832 	case C_DISCONNECTING:
833 	case C_UNCONNECTED:
834 	case C_TIMEOUT:
835 	case C_BROKEN_PIPE:
836 	case C_NETWORK_FAILURE:
837 	case C_PROTOCOL_ERROR:
838 	case C_TEAR_DOWN:
839 	case C_WF_CONNECTION:
840 	case C_WF_REPORT_PARAMS:
841 	case C_MASK:
842 		break;
843 	}
844 	if (ns.disk > disk_max)
845 		ns.disk = disk_max;
846 
847 	if (ns.disk < disk_min) {
848 		if (warn)
849 			*warn = IMPLICITLY_UPGRADED_DISK;
850 		ns.disk = disk_min;
851 	}
852 	if (ns.pdsk > pdsk_max)
853 		ns.pdsk = pdsk_max;
854 
855 	if (ns.pdsk < pdsk_min) {
856 		if (warn)
857 			*warn = IMPLICITLY_UPGRADED_PDSK;
858 		ns.pdsk = pdsk_min;
859 	}
860 
861 	if (fp == FP_STONITH &&
862 	    (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED))
863 		ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */
864 
865 	if (mdev->tconn->res_opts.on_no_data == OND_SUSPEND_IO &&
866 	    (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
867 		ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */
868 
869 	if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
870 		if (ns.conn == C_SYNC_SOURCE)
871 			ns.conn = C_PAUSED_SYNC_S;
872 		if (ns.conn == C_SYNC_TARGET)
873 			ns.conn = C_PAUSED_SYNC_T;
874 	} else {
875 		if (ns.conn == C_PAUSED_SYNC_S)
876 			ns.conn = C_SYNC_SOURCE;
877 		if (ns.conn == C_PAUSED_SYNC_T)
878 			ns.conn = C_SYNC_TARGET;
879 	}
880 
881 	return ns;
882 }
883 
884 void drbd_resume_al(struct drbd_conf *mdev)
885 {
886 	if (test_and_clear_bit(AL_SUSPENDED, &mdev->flags))
887 		dev_info(DEV, "Resumed AL updates\n");
888 }
889 
890 /* helper for __drbd_set_state */
891 static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs)
892 {
893 	if (mdev->tconn->agreed_pro_version < 90)
894 		mdev->ov_start_sector = 0;
895 	mdev->rs_total = drbd_bm_bits(mdev);
896 	mdev->ov_position = 0;
897 	if (cs == C_VERIFY_T) {
898 		/* starting online verify from an arbitrary position
899 		 * does not fit well into the existing protocol.
900 		 * on C_VERIFY_T, we initialize ov_left and friends
901 		 * implicitly in receive_DataRequest once the
902 		 * first P_OV_REQUEST is received */
903 		mdev->ov_start_sector = ~(sector_t)0;
904 	} else {
905 		unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector);
906 		if (bit >= mdev->rs_total) {
907 			mdev->ov_start_sector =
908 				BM_BIT_TO_SECT(mdev->rs_total - 1);
909 			mdev->rs_total = 1;
910 		} else
911 			mdev->rs_total -= bit;
912 		mdev->ov_position = mdev->ov_start_sector;
913 	}
914 	mdev->ov_left = mdev->rs_total;
915 }
916 
917 /**
918  * __drbd_set_state() - Set a new DRBD state
919  * @mdev:	DRBD device.
920  * @ns:		new state.
921  * @flags:	Flags
922  * @done:	Optional completion, that will get completed after the after_state_ch() finished
923  *
924  * Caller needs to hold req_lock, and global_state_lock. Do not call directly.
925  */
926 enum drbd_state_rv
927 __drbd_set_state(struct drbd_conf *mdev, union drbd_state ns,
928 	         enum chg_state_flags flags, struct completion *done)
929 {
930 	union drbd_state os;
931 	enum drbd_state_rv rv = SS_SUCCESS;
932 	enum sanitize_state_warnings ssw;
933 	struct after_state_chg_work *ascw;
934 
935 	os = drbd_read_state(mdev);
936 
937 	ns = sanitize_state(mdev, ns, &ssw);
938 	if (ns.i == os.i)
939 		return SS_NOTHING_TO_DO;
940 
941 	rv = is_valid_transition(os, ns);
942 	if (rv < SS_SUCCESS)
943 		return rv;
944 
945 	if (!(flags & CS_HARD)) {
946 		/*  pre-state-change checks ; only look at ns  */
947 		/* See drbd_state_sw_errors in drbd_strings.c */
948 
949 		rv = is_valid_state(mdev, ns);
950 		if (rv < SS_SUCCESS) {
951 			/* If the old state was illegal as well, then let
952 			   this happen...*/
953 
954 			if (is_valid_state(mdev, os) == rv)
955 				rv = is_valid_soft_transition(os, ns, mdev->tconn);
956 		} else
957 			rv = is_valid_soft_transition(os, ns, mdev->tconn);
958 	}
959 
960 	if (rv < SS_SUCCESS) {
961 		if (flags & CS_VERBOSE)
962 			print_st_err(mdev, os, ns, rv);
963 		return rv;
964 	}
965 
966 	print_sanitize_warnings(mdev, ssw);
967 
968 	drbd_pr_state_change(mdev, os, ns, flags);
969 
970 	/* Display changes to the susp* flags that where caused by the call to
971 	   sanitize_state(). Only display it here if we where not called from
972 	   _conn_request_state() */
973 	if (!(flags & CS_DC_SUSP))
974 		conn_pr_state_change(mdev->tconn, os, ns, (flags & ~CS_DC_MASK) | CS_DC_SUSP);
975 
976 	/* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
977 	 * on the ldev here, to be sure the transition -> D_DISKLESS resp.
978 	 * drbd_ldev_destroy() won't happen before our corresponding
979 	 * after_state_ch works run, where we put_ldev again. */
980 	if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
981 	    (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
982 		atomic_inc(&mdev->local_cnt);
983 
984 	mdev->state.i = ns.i;
985 	mdev->tconn->susp = ns.susp;
986 	mdev->tconn->susp_nod = ns.susp_nod;
987 	mdev->tconn->susp_fen = ns.susp_fen;
988 
989 	if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
990 		drbd_print_uuids(mdev, "attached to UUIDs");
991 
992 	/* Wake up role changes, that were delayed because of connection establishing */
993 	if (os.conn == C_WF_REPORT_PARAMS && ns.conn != C_WF_REPORT_PARAMS &&
994 	    no_peer_wf_report_params(mdev->tconn))
995 		clear_bit(STATE_SENT, &mdev->tconn->flags);
996 
997 	wake_up(&mdev->misc_wait);
998 	wake_up(&mdev->state_wait);
999 	wake_up(&mdev->tconn->ping_wait);
1000 
1001 	/* Aborted verify run, or we reached the stop sector.
1002 	 * Log the last position, unless end-of-device. */
1003 	if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
1004 	    ns.conn <= C_CONNECTED) {
1005 		mdev->ov_start_sector =
1006 			BM_BIT_TO_SECT(drbd_bm_bits(mdev) - mdev->ov_left);
1007 		if (mdev->ov_left)
1008 			dev_info(DEV, "Online Verify reached sector %llu\n",
1009 				(unsigned long long)mdev->ov_start_sector);
1010 	}
1011 
1012 	if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
1013 	    (ns.conn == C_SYNC_TARGET  || ns.conn == C_SYNC_SOURCE)) {
1014 		dev_info(DEV, "Syncer continues.\n");
1015 		mdev->rs_paused += (long)jiffies
1016 				  -(long)mdev->rs_mark_time[mdev->rs_last_mark];
1017 		if (ns.conn == C_SYNC_TARGET)
1018 			mod_timer(&mdev->resync_timer, jiffies);
1019 	}
1020 
1021 	if ((os.conn == C_SYNC_TARGET  || os.conn == C_SYNC_SOURCE) &&
1022 	    (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
1023 		dev_info(DEV, "Resync suspended\n");
1024 		mdev->rs_mark_time[mdev->rs_last_mark] = jiffies;
1025 	}
1026 
1027 	if (os.conn == C_CONNECTED &&
1028 	    (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
1029 		unsigned long now = jiffies;
1030 		int i;
1031 
1032 		set_ov_position(mdev, ns.conn);
1033 		mdev->rs_start = now;
1034 		mdev->rs_last_events = 0;
1035 		mdev->rs_last_sect_ev = 0;
1036 		mdev->ov_last_oos_size = 0;
1037 		mdev->ov_last_oos_start = 0;
1038 
1039 		for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1040 			mdev->rs_mark_left[i] = mdev->ov_left;
1041 			mdev->rs_mark_time[i] = now;
1042 		}
1043 
1044 		drbd_rs_controller_reset(mdev);
1045 
1046 		if (ns.conn == C_VERIFY_S) {
1047 			dev_info(DEV, "Starting Online Verify from sector %llu\n",
1048 					(unsigned long long)mdev->ov_position);
1049 			mod_timer(&mdev->resync_timer, jiffies);
1050 		}
1051 	}
1052 
1053 	if (get_ldev(mdev)) {
1054 		u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
1055 						 MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
1056 						 MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
1057 
1058 		mdf &= ~MDF_AL_CLEAN;
1059 		if (test_bit(CRASHED_PRIMARY, &mdev->flags))
1060 			mdf |= MDF_CRASHED_PRIMARY;
1061 		if (mdev->state.role == R_PRIMARY ||
1062 		    (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY))
1063 			mdf |= MDF_PRIMARY_IND;
1064 		if (mdev->state.conn > C_WF_REPORT_PARAMS)
1065 			mdf |= MDF_CONNECTED_IND;
1066 		if (mdev->state.disk > D_INCONSISTENT)
1067 			mdf |= MDF_CONSISTENT;
1068 		if (mdev->state.disk > D_OUTDATED)
1069 			mdf |= MDF_WAS_UP_TO_DATE;
1070 		if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT)
1071 			mdf |= MDF_PEER_OUT_DATED;
1072 		if (mdf != mdev->ldev->md.flags) {
1073 			mdev->ldev->md.flags = mdf;
1074 			drbd_md_mark_dirty(mdev);
1075 		}
1076 		if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
1077 			drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]);
1078 		put_ldev(mdev);
1079 	}
1080 
1081 	/* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
1082 	if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
1083 	    os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
1084 		set_bit(CONSIDER_RESYNC, &mdev->flags);
1085 
1086 	/* Receiver should clean up itself */
1087 	if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
1088 		drbd_thread_stop_nowait(&mdev->tconn->receiver);
1089 
1090 	/* Now the receiver finished cleaning up itself, it should die */
1091 	if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
1092 		drbd_thread_stop_nowait(&mdev->tconn->receiver);
1093 
1094 	/* Upon network failure, we need to restart the receiver. */
1095 	if (os.conn > C_WF_CONNECTION &&
1096 	    ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
1097 		drbd_thread_restart_nowait(&mdev->tconn->receiver);
1098 
1099 	/* Resume AL writing if we get a connection */
1100 	if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
1101 		drbd_resume_al(mdev);
1102 
1103 	/* remember last attach time so request_timer_fn() won't
1104 	 * kill newly established sessions while we are still trying to thaw
1105 	 * previously frozen IO */
1106 	if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1107 	    ns.disk > D_NEGOTIATING)
1108 		mdev->last_reattach_jif = jiffies;
1109 
1110 	ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
1111 	if (ascw) {
1112 		ascw->os = os;
1113 		ascw->ns = ns;
1114 		ascw->flags = flags;
1115 		ascw->w.cb = w_after_state_ch;
1116 		ascw->w.mdev = mdev;
1117 		ascw->done = done;
1118 		drbd_queue_work(&mdev->tconn->sender_work, &ascw->w);
1119 	} else {
1120 		dev_err(DEV, "Could not kmalloc an ascw\n");
1121 	}
1122 
1123 	return rv;
1124 }
1125 
1126 static int w_after_state_ch(struct drbd_work *w, int unused)
1127 {
1128 	struct after_state_chg_work *ascw =
1129 		container_of(w, struct after_state_chg_work, w);
1130 	struct drbd_conf *mdev = w->mdev;
1131 
1132 	after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags);
1133 	if (ascw->flags & CS_WAIT_COMPLETE) {
1134 		D_ASSERT(ascw->done != NULL);
1135 		complete(ascw->done);
1136 	}
1137 	kfree(ascw);
1138 
1139 	return 0;
1140 }
1141 
1142 static void abw_start_sync(struct drbd_conf *mdev, int rv)
1143 {
1144 	if (rv) {
1145 		dev_err(DEV, "Writing the bitmap failed not starting resync.\n");
1146 		_drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE);
1147 		return;
1148 	}
1149 
1150 	switch (mdev->state.conn) {
1151 	case C_STARTING_SYNC_T:
1152 		_drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
1153 		break;
1154 	case C_STARTING_SYNC_S:
1155 		drbd_start_resync(mdev, C_SYNC_SOURCE);
1156 		break;
1157 	}
1158 }
1159 
1160 int drbd_bitmap_io_from_worker(struct drbd_conf *mdev,
1161 		int (*io_fn)(struct drbd_conf *),
1162 		char *why, enum bm_flag flags)
1163 {
1164 	int rv;
1165 
1166 	D_ASSERT(current == mdev->tconn->worker.task);
1167 
1168 	/* open coded non-blocking drbd_suspend_io(mdev); */
1169 	set_bit(SUSPEND_IO, &mdev->flags);
1170 
1171 	drbd_bm_lock(mdev, why, flags);
1172 	rv = io_fn(mdev);
1173 	drbd_bm_unlock(mdev);
1174 
1175 	drbd_resume_io(mdev);
1176 
1177 	return rv;
1178 }
1179 
1180 /**
1181  * after_state_ch() - Perform after state change actions that may sleep
1182  * @mdev:	DRBD device.
1183  * @os:		old state.
1184  * @ns:		new state.
1185  * @flags:	Flags
1186  */
1187 static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
1188 			   union drbd_state ns, enum chg_state_flags flags)
1189 {
1190 	struct sib_info sib;
1191 
1192 	sib.sib_reason = SIB_STATE_CHANGE;
1193 	sib.os = os;
1194 	sib.ns = ns;
1195 
1196 	if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) {
1197 		clear_bit(CRASHED_PRIMARY, &mdev->flags);
1198 		if (mdev->p_uuid)
1199 			mdev->p_uuid[UI_FLAGS] &= ~((u64)2);
1200 	}
1201 
1202 	/* Inform userspace about the change... */
1203 	drbd_bcast_event(mdev, &sib);
1204 
1205 	if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
1206 	    (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
1207 		drbd_khelper(mdev, "pri-on-incon-degr");
1208 
1209 	/* Here we have the actions that are performed after a
1210 	   state change. This function might sleep */
1211 
1212 	if (ns.susp_nod) {
1213 		struct drbd_tconn *tconn = mdev->tconn;
1214 		enum drbd_req_event what = NOTHING;
1215 
1216 		spin_lock_irq(&tconn->req_lock);
1217 		if (os.conn < C_CONNECTED && conn_lowest_conn(tconn) >= C_CONNECTED)
1218 			what = RESEND;
1219 
1220 		if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1221 		    conn_lowest_disk(tconn) > D_NEGOTIATING)
1222 			what = RESTART_FROZEN_DISK_IO;
1223 
1224 		if (tconn->susp_nod && what != NOTHING) {
1225 			_tl_restart(tconn, what);
1226 			_conn_request_state(tconn,
1227 					    (union drbd_state) { { .susp_nod = 1 } },
1228 					    (union drbd_state) { { .susp_nod = 0 } },
1229 					    CS_VERBOSE);
1230 		}
1231 		spin_unlock_irq(&tconn->req_lock);
1232 	}
1233 
1234 	if (ns.susp_fen) {
1235 		struct drbd_tconn *tconn = mdev->tconn;
1236 
1237 		spin_lock_irq(&tconn->req_lock);
1238 		if (tconn->susp_fen && conn_lowest_conn(tconn) >= C_CONNECTED) {
1239 			/* case2: The connection was established again: */
1240 			struct drbd_conf *odev;
1241 			int vnr;
1242 
1243 			rcu_read_lock();
1244 			idr_for_each_entry(&tconn->volumes, odev, vnr)
1245 				clear_bit(NEW_CUR_UUID, &odev->flags);
1246 			rcu_read_unlock();
1247 			_tl_restart(tconn, RESEND);
1248 			_conn_request_state(tconn,
1249 					    (union drbd_state) { { .susp_fen = 1 } },
1250 					    (union drbd_state) { { .susp_fen = 0 } },
1251 					    CS_VERBOSE);
1252 		}
1253 		spin_unlock_irq(&tconn->req_lock);
1254 	}
1255 
1256 	/* Became sync source.  With protocol >= 96, we still need to send out
1257 	 * the sync uuid now. Need to do that before any drbd_send_state, or
1258 	 * the other side may go "paused sync" before receiving the sync uuids,
1259 	 * which is unexpected. */
1260 	if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
1261 	    (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
1262 	    mdev->tconn->agreed_pro_version >= 96 && get_ldev(mdev)) {
1263 		drbd_gen_and_send_sync_uuid(mdev);
1264 		put_ldev(mdev);
1265 	}
1266 
1267 	/* Do not change the order of the if above and the two below... */
1268 	if (os.pdsk == D_DISKLESS &&
1269 	    ns.pdsk > D_DISKLESS && ns.pdsk != D_UNKNOWN) {      /* attach on the peer */
1270 		/* we probably will start a resync soon.
1271 		 * make sure those things are properly reset. */
1272 		mdev->rs_total = 0;
1273 		mdev->rs_failed = 0;
1274 		atomic_set(&mdev->rs_pending_cnt, 0);
1275 		drbd_rs_cancel_all(mdev);
1276 
1277 		drbd_send_uuids(mdev);
1278 		drbd_send_state(mdev, ns);
1279 	}
1280 	/* No point in queuing send_bitmap if we don't have a connection
1281 	 * anymore, so check also the _current_ state, not only the new state
1282 	 * at the time this work was queued. */
1283 	if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
1284 	    mdev->state.conn == C_WF_BITMAP_S)
1285 		drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL,
1286 				"send_bitmap (WFBitMapS)",
1287 				BM_LOCKED_TEST_ALLOWED);
1288 
1289 	/* Lost contact to peer's copy of the data */
1290 	if ((os.pdsk >= D_INCONSISTENT &&
1291 	     os.pdsk != D_UNKNOWN &&
1292 	     os.pdsk != D_OUTDATED)
1293 	&&  (ns.pdsk < D_INCONSISTENT ||
1294 	     ns.pdsk == D_UNKNOWN ||
1295 	     ns.pdsk == D_OUTDATED)) {
1296 		if (get_ldev(mdev)) {
1297 			if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1298 			    mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1299 				if (drbd_suspended(mdev)) {
1300 					set_bit(NEW_CUR_UUID, &mdev->flags);
1301 				} else {
1302 					drbd_uuid_new_current(mdev);
1303 					drbd_send_uuids(mdev);
1304 				}
1305 			}
1306 			put_ldev(mdev);
1307 		}
1308 	}
1309 
1310 	if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) {
1311 		if (os.peer == R_SECONDARY && ns.peer == R_PRIMARY &&
1312 		    mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1313 			drbd_uuid_new_current(mdev);
1314 			drbd_send_uuids(mdev);
1315 		}
1316 		/* D_DISKLESS Peer becomes secondary */
1317 		if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1318 			/* We may still be Primary ourselves.
1319 			 * No harm done if the bitmap still changes,
1320 			 * redirtied pages will follow later. */
1321 			drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1322 				"demote diskless peer", BM_LOCKED_SET_ALLOWED);
1323 		put_ldev(mdev);
1324 	}
1325 
1326 	/* Write out all changed bits on demote.
1327 	 * Though, no need to da that just yet
1328 	 * if there is a resync going on still */
1329 	if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
1330 		mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) {
1331 		/* No changes to the bitmap expected this time, so assert that,
1332 		 * even though no harm was done if it did change. */
1333 		drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1334 				"demote", BM_LOCKED_TEST_ALLOWED);
1335 		put_ldev(mdev);
1336 	}
1337 
1338 	/* Last part of the attaching process ... */
1339 	if (ns.conn >= C_CONNECTED &&
1340 	    os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1341 		drbd_send_sizes(mdev, 0, 0);  /* to start sync... */
1342 		drbd_send_uuids(mdev);
1343 		drbd_send_state(mdev, ns);
1344 	}
1345 
1346 	/* We want to pause/continue resync, tell peer. */
1347 	if (ns.conn >= C_CONNECTED &&
1348 	     ((os.aftr_isp != ns.aftr_isp) ||
1349 	      (os.user_isp != ns.user_isp)))
1350 		drbd_send_state(mdev, ns);
1351 
1352 	/* In case one of the isp bits got set, suspend other devices. */
1353 	if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1354 	    (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1355 		suspend_other_sg(mdev);
1356 
1357 	/* Make sure the peer gets informed about eventual state
1358 	   changes (ISP bits) while we were in WFReportParams. */
1359 	if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1360 		drbd_send_state(mdev, ns);
1361 
1362 	if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
1363 		drbd_send_state(mdev, ns);
1364 
1365 	/* We are in the progress to start a full sync... */
1366 	if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1367 	    (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1368 		/* no other bitmap changes expected during this phase */
1369 		drbd_queue_bitmap_io(mdev,
1370 			&drbd_bmio_set_n_write, &abw_start_sync,
1371 			"set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED);
1372 
1373 	/* We are invalidating our self... */
1374 	if (os.conn < C_CONNECTED && ns.conn < C_CONNECTED &&
1375 	    os.disk > D_INCONSISTENT && ns.disk == D_INCONSISTENT)
1376 		/* other bitmap operation expected during this phase */
1377 		drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, NULL,
1378 			"set_n_write from invalidate", BM_LOCKED_MASK);
1379 
1380 	/* first half of local IO error, failure to attach,
1381 	 * or administrative detach */
1382 	if (os.disk != D_FAILED && ns.disk == D_FAILED) {
1383 		enum drbd_io_error_p eh = EP_PASS_ON;
1384 		int was_io_error = 0;
1385 		/* corresponding get_ldev was in __drbd_set_state, to serialize
1386 		 * our cleanup here with the transition to D_DISKLESS.
1387 		 * But is is still not save to dreference ldev here, since
1388 		 * we might come from an failed Attach before ldev was set. */
1389 		if (mdev->ldev) {
1390 			rcu_read_lock();
1391 			eh = rcu_dereference(mdev->ldev->disk_conf)->on_io_error;
1392 			rcu_read_unlock();
1393 
1394 			was_io_error = test_and_clear_bit(WAS_IO_ERROR, &mdev->flags);
1395 
1396 			if (was_io_error && eh == EP_CALL_HELPER)
1397 				drbd_khelper(mdev, "local-io-error");
1398 
1399 			/* Immediately allow completion of all application IO,
1400 			 * that waits for completion from the local disk,
1401 			 * if this was a force-detach due to disk_timeout
1402 			 * or administrator request (drbdsetup detach --force).
1403 			 * Do NOT abort otherwise.
1404 			 * Aborting local requests may cause serious problems,
1405 			 * if requests are completed to upper layers already,
1406 			 * and then later the already submitted local bio completes.
1407 			 * This can cause DMA into former bio pages that meanwhile
1408 			 * have been re-used for other things.
1409 			 * So aborting local requests may cause crashes,
1410 			 * or even worse, silent data corruption.
1411 			 */
1412 			if (test_and_clear_bit(FORCE_DETACH, &mdev->flags))
1413 				tl_abort_disk_io(mdev);
1414 
1415 			/* current state still has to be D_FAILED,
1416 			 * there is only one way out: to D_DISKLESS,
1417 			 * and that may only happen after our put_ldev below. */
1418 			if (mdev->state.disk != D_FAILED)
1419 				dev_err(DEV,
1420 					"ASSERT FAILED: disk is %s during detach\n",
1421 					drbd_disk_str(mdev->state.disk));
1422 
1423 			if (ns.conn >= C_CONNECTED)
1424 				drbd_send_state(mdev, ns);
1425 
1426 			drbd_rs_cancel_all(mdev);
1427 
1428 			/* In case we want to get something to stable storage still,
1429 			 * this may be the last chance.
1430 			 * Following put_ldev may transition to D_DISKLESS. */
1431 			drbd_md_sync(mdev);
1432 		}
1433 		put_ldev(mdev);
1434 	}
1435 
1436         /* second half of local IO error, failure to attach,
1437          * or administrative detach,
1438          * after local_cnt references have reached zero again */
1439         if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
1440                 /* We must still be diskless,
1441                  * re-attach has to be serialized with this! */
1442                 if (mdev->state.disk != D_DISKLESS)
1443                         dev_err(DEV,
1444                                 "ASSERT FAILED: disk is %s while going diskless\n",
1445                                 drbd_disk_str(mdev->state.disk));
1446 
1447 		if (ns.conn >= C_CONNECTED)
1448 			drbd_send_state(mdev, ns);
1449 		/* corresponding get_ldev in __drbd_set_state
1450 		 * this may finally trigger drbd_ldev_destroy. */
1451 		put_ldev(mdev);
1452 	}
1453 
1454 	/* Notify peer that I had a local IO error, and did not detached.. */
1455 	if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT && ns.conn >= C_CONNECTED)
1456 		drbd_send_state(mdev, ns);
1457 
1458 	/* Disks got bigger while they were detached */
1459 	if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1460 	    test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) {
1461 		if (ns.conn == C_CONNECTED)
1462 			resync_after_online_grow(mdev);
1463 	}
1464 
1465 	/* A resync finished or aborted, wake paused devices... */
1466 	if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1467 	    (os.peer_isp && !ns.peer_isp) ||
1468 	    (os.user_isp && !ns.user_isp))
1469 		resume_next_sg(mdev);
1470 
1471 	/* sync target done with resync.  Explicitly notify peer, even though
1472 	 * it should (at least for non-empty resyncs) already know itself. */
1473 	if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
1474 		drbd_send_state(mdev, ns);
1475 
1476 	/* Verify finished, or reached stop sector.  Peer did not know about
1477 	 * the stop sector, and we may even have changed the stop sector during
1478 	 * verify to interrupt/stop early.  Send the new state. */
1479 	if (os.conn == C_VERIFY_S && ns.conn == C_CONNECTED
1480 	&& verify_can_do_stop_sector(mdev))
1481 		drbd_send_state(mdev, ns);
1482 
1483 	/* This triggers bitmap writeout of potentially still unwritten pages
1484 	 * if the resync finished cleanly, or aborted because of peer disk
1485 	 * failure, or because of connection loss.
1486 	 * For resync aborted because of local disk failure, we cannot do
1487 	 * any bitmap writeout anymore.
1488 	 * No harm done if some bits change during this phase.
1489 	 */
1490 	if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) {
1491 		drbd_queue_bitmap_io(mdev, &drbd_bm_write_copy_pages, NULL,
1492 			"write from resync_finished", BM_LOCKED_CHANGE_ALLOWED);
1493 		put_ldev(mdev);
1494 	}
1495 
1496 	if (ns.disk == D_DISKLESS &&
1497 	    ns.conn == C_STANDALONE &&
1498 	    ns.role == R_SECONDARY) {
1499 		if (os.aftr_isp != ns.aftr_isp)
1500 			resume_next_sg(mdev);
1501 	}
1502 
1503 	drbd_md_sync(mdev);
1504 }
1505 
1506 struct after_conn_state_chg_work {
1507 	struct drbd_work w;
1508 	enum drbd_conns oc;
1509 	union drbd_state ns_min;
1510 	union drbd_state ns_max; /* new, max state, over all mdevs */
1511 	enum chg_state_flags flags;
1512 };
1513 
1514 static int w_after_conn_state_ch(struct drbd_work *w, int unused)
1515 {
1516 	struct after_conn_state_chg_work *acscw =
1517 		container_of(w, struct after_conn_state_chg_work, w);
1518 	struct drbd_tconn *tconn = w->tconn;
1519 	enum drbd_conns oc = acscw->oc;
1520 	union drbd_state ns_max = acscw->ns_max;
1521 	struct drbd_conf *mdev;
1522 	int vnr;
1523 
1524 	kfree(acscw);
1525 
1526 	/* Upon network configuration, we need to start the receiver */
1527 	if (oc == C_STANDALONE && ns_max.conn == C_UNCONNECTED)
1528 		drbd_thread_start(&tconn->receiver);
1529 
1530 	if (oc == C_DISCONNECTING && ns_max.conn == C_STANDALONE) {
1531 		struct net_conf *old_conf;
1532 
1533 		mutex_lock(&tconn->conf_update);
1534 		old_conf = tconn->net_conf;
1535 		tconn->my_addr_len = 0;
1536 		tconn->peer_addr_len = 0;
1537 		rcu_assign_pointer(tconn->net_conf, NULL);
1538 		conn_free_crypto(tconn);
1539 		mutex_unlock(&tconn->conf_update);
1540 
1541 		synchronize_rcu();
1542 		kfree(old_conf);
1543 	}
1544 
1545 	if (ns_max.susp_fen) {
1546 		/* case1: The outdate peer handler is successful: */
1547 		if (ns_max.pdsk <= D_OUTDATED) {
1548 			rcu_read_lock();
1549 			idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1550 				if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1551 					drbd_uuid_new_current(mdev);
1552 					clear_bit(NEW_CUR_UUID, &mdev->flags);
1553 				}
1554 			}
1555 			rcu_read_unlock();
1556 			spin_lock_irq(&tconn->req_lock);
1557 			_tl_restart(tconn, CONNECTION_LOST_WHILE_PENDING);
1558 			_conn_request_state(tconn,
1559 					    (union drbd_state) { { .susp_fen = 1 } },
1560 					    (union drbd_state) { { .susp_fen = 0 } },
1561 					    CS_VERBOSE);
1562 			spin_unlock_irq(&tconn->req_lock);
1563 		}
1564 	}
1565 	kref_put(&tconn->kref, &conn_destroy);
1566 
1567 	conn_md_sync(tconn);
1568 
1569 	return 0;
1570 }
1571 
1572 void conn_old_common_state(struct drbd_tconn *tconn, union drbd_state *pcs, enum chg_state_flags *pf)
1573 {
1574 	enum chg_state_flags flags = ~0;
1575 	struct drbd_conf *mdev;
1576 	int vnr, first_vol = 1;
1577 	union drbd_dev_state os, cs = {
1578 		{ .role = R_SECONDARY,
1579 		  .peer = R_UNKNOWN,
1580 		  .conn = tconn->cstate,
1581 		  .disk = D_DISKLESS,
1582 		  .pdsk = D_UNKNOWN,
1583 		} };
1584 
1585 	rcu_read_lock();
1586 	idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1587 		os = mdev->state;
1588 
1589 		if (first_vol) {
1590 			cs = os;
1591 			first_vol = 0;
1592 			continue;
1593 		}
1594 
1595 		if (cs.role != os.role)
1596 			flags &= ~CS_DC_ROLE;
1597 
1598 		if (cs.peer != os.peer)
1599 			flags &= ~CS_DC_PEER;
1600 
1601 		if (cs.conn != os.conn)
1602 			flags &= ~CS_DC_CONN;
1603 
1604 		if (cs.disk != os.disk)
1605 			flags &= ~CS_DC_DISK;
1606 
1607 		if (cs.pdsk != os.pdsk)
1608 			flags &= ~CS_DC_PDSK;
1609 	}
1610 	rcu_read_unlock();
1611 
1612 	*pf |= CS_DC_MASK;
1613 	*pf &= flags;
1614 	(*pcs).i = cs.i;
1615 }
1616 
1617 static enum drbd_state_rv
1618 conn_is_valid_transition(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1619 			 enum chg_state_flags flags)
1620 {
1621 	enum drbd_state_rv rv = SS_SUCCESS;
1622 	union drbd_state ns, os;
1623 	struct drbd_conf *mdev;
1624 	int vnr;
1625 
1626 	rcu_read_lock();
1627 	idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1628 		os = drbd_read_state(mdev);
1629 		ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
1630 
1631 		if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
1632 			ns.disk = os.disk;
1633 
1634 		if (ns.i == os.i)
1635 			continue;
1636 
1637 		rv = is_valid_transition(os, ns);
1638 		if (rv < SS_SUCCESS)
1639 			break;
1640 
1641 		if (!(flags & CS_HARD)) {
1642 			rv = is_valid_state(mdev, ns);
1643 			if (rv < SS_SUCCESS) {
1644 				if (is_valid_state(mdev, os) == rv)
1645 					rv = is_valid_soft_transition(os, ns, tconn);
1646 			} else
1647 				rv = is_valid_soft_transition(os, ns, tconn);
1648 		}
1649 		if (rv < SS_SUCCESS)
1650 			break;
1651 	}
1652 	rcu_read_unlock();
1653 
1654 	if (rv < SS_SUCCESS && flags & CS_VERBOSE)
1655 		print_st_err(mdev, os, ns, rv);
1656 
1657 	return rv;
1658 }
1659 
1660 void
1661 conn_set_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1662 	       union drbd_state *pns_min, union drbd_state *pns_max, enum chg_state_flags flags)
1663 {
1664 	union drbd_state ns, os, ns_max = { };
1665 	union drbd_state ns_min = {
1666 		{ .role = R_MASK,
1667 		  .peer = R_MASK,
1668 		  .conn = val.conn,
1669 		  .disk = D_MASK,
1670 		  .pdsk = D_MASK
1671 		} };
1672 	struct drbd_conf *mdev;
1673 	enum drbd_state_rv rv;
1674 	int vnr, number_of_volumes = 0;
1675 
1676 	if (mask.conn == C_MASK) {
1677 		/* remember last connect time so request_timer_fn() won't
1678 		 * kill newly established sessions while we are still trying to thaw
1679 		 * previously frozen IO */
1680 		if (tconn->cstate != C_WF_REPORT_PARAMS && val.conn == C_WF_REPORT_PARAMS)
1681 			tconn->last_reconnect_jif = jiffies;
1682 
1683 		tconn->cstate = val.conn;
1684 	}
1685 
1686 	rcu_read_lock();
1687 	idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1688 		number_of_volumes++;
1689 		os = drbd_read_state(mdev);
1690 		ns = apply_mask_val(os, mask, val);
1691 		ns = sanitize_state(mdev, ns, NULL);
1692 
1693 		if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
1694 			ns.disk = os.disk;
1695 
1696 		rv = __drbd_set_state(mdev, ns, flags, NULL);
1697 		if (rv < SS_SUCCESS)
1698 			BUG();
1699 
1700 		ns.i = mdev->state.i;
1701 		ns_max.role = max_role(ns.role, ns_max.role);
1702 		ns_max.peer = max_role(ns.peer, ns_max.peer);
1703 		ns_max.conn = max_t(enum drbd_conns, ns.conn, ns_max.conn);
1704 		ns_max.disk = max_t(enum drbd_disk_state, ns.disk, ns_max.disk);
1705 		ns_max.pdsk = max_t(enum drbd_disk_state, ns.pdsk, ns_max.pdsk);
1706 
1707 		ns_min.role = min_role(ns.role, ns_min.role);
1708 		ns_min.peer = min_role(ns.peer, ns_min.peer);
1709 		ns_min.conn = min_t(enum drbd_conns, ns.conn, ns_min.conn);
1710 		ns_min.disk = min_t(enum drbd_disk_state, ns.disk, ns_min.disk);
1711 		ns_min.pdsk = min_t(enum drbd_disk_state, ns.pdsk, ns_min.pdsk);
1712 	}
1713 	rcu_read_unlock();
1714 
1715 	if (number_of_volumes == 0) {
1716 		ns_min = ns_max = (union drbd_state) { {
1717 				.role = R_SECONDARY,
1718 				.peer = R_UNKNOWN,
1719 				.conn = val.conn,
1720 				.disk = D_DISKLESS,
1721 				.pdsk = D_UNKNOWN
1722 			} };
1723 	}
1724 
1725 	ns_min.susp = ns_max.susp = tconn->susp;
1726 	ns_min.susp_nod = ns_max.susp_nod = tconn->susp_nod;
1727 	ns_min.susp_fen = ns_max.susp_fen = tconn->susp_fen;
1728 
1729 	*pns_min = ns_min;
1730 	*pns_max = ns_max;
1731 }
1732 
1733 static enum drbd_state_rv
1734 _conn_rq_cond(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val)
1735 {
1736 	enum drbd_state_rv rv;
1737 
1738 	if (test_and_clear_bit(CONN_WD_ST_CHG_OKAY, &tconn->flags))
1739 		return SS_CW_SUCCESS;
1740 
1741 	if (test_and_clear_bit(CONN_WD_ST_CHG_FAIL, &tconn->flags))
1742 		return SS_CW_FAILED_BY_PEER;
1743 
1744 	rv = tconn->cstate != C_WF_REPORT_PARAMS ? SS_CW_NO_NEED : SS_UNKNOWN_ERROR;
1745 
1746 	if (rv == SS_UNKNOWN_ERROR)
1747 		rv = conn_is_valid_transition(tconn, mask, val, 0);
1748 
1749 	if (rv == SS_SUCCESS)
1750 		rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
1751 
1752 	return rv;
1753 }
1754 
1755 enum drbd_state_rv
1756 _conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1757 		    enum chg_state_flags flags)
1758 {
1759 	enum drbd_state_rv rv = SS_SUCCESS;
1760 	struct after_conn_state_chg_work *acscw;
1761 	enum drbd_conns oc = tconn->cstate;
1762 	union drbd_state ns_max, ns_min, os;
1763 	bool have_mutex = false;
1764 
1765 	if (mask.conn) {
1766 		rv = is_valid_conn_transition(oc, val.conn);
1767 		if (rv < SS_SUCCESS)
1768 			goto abort;
1769 	}
1770 
1771 	rv = conn_is_valid_transition(tconn, mask, val, flags);
1772 	if (rv < SS_SUCCESS)
1773 		goto abort;
1774 
1775 	if (oc == C_WF_REPORT_PARAMS && val.conn == C_DISCONNECTING &&
1776 	    !(flags & (CS_LOCAL_ONLY | CS_HARD))) {
1777 
1778 		/* This will be a cluster-wide state change.
1779 		 * Need to give up the spinlock, grab the mutex,
1780 		 * then send the state change request, ... */
1781 		spin_unlock_irq(&tconn->req_lock);
1782 		mutex_lock(&tconn->cstate_mutex);
1783 		have_mutex = true;
1784 
1785 		set_bit(CONN_WD_ST_CHG_REQ, &tconn->flags);
1786 		if (conn_send_state_req(tconn, mask, val)) {
1787 			/* sending failed. */
1788 			clear_bit(CONN_WD_ST_CHG_REQ, &tconn->flags);
1789 			rv = SS_CW_FAILED_BY_PEER;
1790 			/* need to re-aquire the spin lock, though */
1791 			goto abort_unlocked;
1792 		}
1793 
1794 		if (val.conn == C_DISCONNECTING)
1795 			set_bit(DISCONNECT_SENT, &tconn->flags);
1796 
1797 		/* ... and re-aquire the spinlock.
1798 		 * If _conn_rq_cond() returned >= SS_SUCCESS, we must call
1799 		 * conn_set_state() within the same spinlock. */
1800 		spin_lock_irq(&tconn->req_lock);
1801 		wait_event_lock_irq(tconn->ping_wait,
1802 				(rv = _conn_rq_cond(tconn, mask, val)),
1803 				tconn->req_lock);
1804 		clear_bit(CONN_WD_ST_CHG_REQ, &tconn->flags);
1805 		if (rv < SS_SUCCESS)
1806 			goto abort;
1807 	}
1808 
1809 	conn_old_common_state(tconn, &os, &flags);
1810 	flags |= CS_DC_SUSP;
1811 	conn_set_state(tconn, mask, val, &ns_min, &ns_max, flags);
1812 	conn_pr_state_change(tconn, os, ns_max, flags);
1813 
1814 	acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC);
1815 	if (acscw) {
1816 		acscw->oc = os.conn;
1817 		acscw->ns_min = ns_min;
1818 		acscw->ns_max = ns_max;
1819 		acscw->flags = flags;
1820 		acscw->w.cb = w_after_conn_state_ch;
1821 		kref_get(&tconn->kref);
1822 		acscw->w.tconn = tconn;
1823 		drbd_queue_work(&tconn->sender_work, &acscw->w);
1824 	} else {
1825 		conn_err(tconn, "Could not kmalloc an acscw\n");
1826 	}
1827 
1828  abort:
1829 	if (have_mutex) {
1830 		/* mutex_unlock() "... must not be used in interrupt context.",
1831 		 * so give up the spinlock, then re-aquire it */
1832 		spin_unlock_irq(&tconn->req_lock);
1833  abort_unlocked:
1834 		mutex_unlock(&tconn->cstate_mutex);
1835 		spin_lock_irq(&tconn->req_lock);
1836 	}
1837 	if (rv < SS_SUCCESS && flags & CS_VERBOSE) {
1838 		conn_err(tconn, "State change failed: %s\n", drbd_set_st_err_str(rv));
1839 		conn_err(tconn, " mask = 0x%x val = 0x%x\n", mask.i, val.i);
1840 		conn_err(tconn, " old_conn:%s wanted_conn:%s\n", drbd_conn_str(oc), drbd_conn_str(val.conn));
1841 	}
1842 	return rv;
1843 }
1844 
1845 enum drbd_state_rv
1846 conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1847 		   enum chg_state_flags flags)
1848 {
1849 	enum drbd_state_rv rv;
1850 
1851 	spin_lock_irq(&tconn->req_lock);
1852 	rv = _conn_request_state(tconn, mask, val, flags);
1853 	spin_unlock_irq(&tconn->req_lock);
1854 
1855 	return rv;
1856 }
1857