xref: /openbmc/linux/drivers/block/drbd/drbd_state.c (revision 82e6fdd6)
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 #include "drbd_state_change.h"
33 
34 struct after_state_chg_work {
35 	struct drbd_work w;
36 	struct drbd_device *device;
37 	union drbd_state os;
38 	union drbd_state ns;
39 	enum chg_state_flags flags;
40 	struct completion *done;
41 	struct drbd_state_change *state_change;
42 };
43 
44 enum sanitize_state_warnings {
45 	NO_WARNING,
46 	ABORTED_ONLINE_VERIFY,
47 	ABORTED_RESYNC,
48 	CONNECTION_LOST_NEGOTIATING,
49 	IMPLICITLY_UPGRADED_DISK,
50 	IMPLICITLY_UPGRADED_PDSK,
51 };
52 
53 static void count_objects(struct drbd_resource *resource,
54 			  unsigned int *n_devices,
55 			  unsigned int *n_connections)
56 {
57 	struct drbd_device *device;
58 	struct drbd_connection *connection;
59 	int vnr;
60 
61 	*n_devices = 0;
62 	*n_connections = 0;
63 
64 	idr_for_each_entry(&resource->devices, device, vnr)
65 		(*n_devices)++;
66 	for_each_connection(connection, resource)
67 		(*n_connections)++;
68 }
69 
70 static struct drbd_state_change *alloc_state_change(unsigned int n_devices, unsigned int n_connections, gfp_t gfp)
71 {
72 	struct drbd_state_change *state_change;
73 	unsigned int size, n;
74 
75 	size = sizeof(struct drbd_state_change) +
76 	       n_devices * sizeof(struct drbd_device_state_change) +
77 	       n_connections * sizeof(struct drbd_connection_state_change) +
78 	       n_devices * n_connections * sizeof(struct drbd_peer_device_state_change);
79 	state_change = kmalloc(size, gfp);
80 	if (!state_change)
81 		return NULL;
82 	state_change->n_devices = n_devices;
83 	state_change->n_connections = n_connections;
84 	state_change->devices = (void *)(state_change + 1);
85 	state_change->connections = (void *)&state_change->devices[n_devices];
86 	state_change->peer_devices = (void *)&state_change->connections[n_connections];
87 	state_change->resource->resource = NULL;
88 	for (n = 0; n < n_devices; n++)
89 		state_change->devices[n].device = NULL;
90 	for (n = 0; n < n_connections; n++)
91 		state_change->connections[n].connection = NULL;
92 	return state_change;
93 }
94 
95 struct drbd_state_change *remember_old_state(struct drbd_resource *resource, gfp_t gfp)
96 {
97 	struct drbd_state_change *state_change;
98 	struct drbd_device *device;
99 	unsigned int n_devices;
100 	struct drbd_connection *connection;
101 	unsigned int n_connections;
102 	int vnr;
103 
104 	struct drbd_device_state_change *device_state_change;
105 	struct drbd_peer_device_state_change *peer_device_state_change;
106 	struct drbd_connection_state_change *connection_state_change;
107 
108 	/* Caller holds req_lock spinlock.
109 	 * No state, no device IDR, no connections lists can change. */
110 	count_objects(resource, &n_devices, &n_connections);
111 	state_change = alloc_state_change(n_devices, n_connections, gfp);
112 	if (!state_change)
113 		return NULL;
114 
115 	kref_get(&resource->kref);
116 	state_change->resource->resource = resource;
117 	state_change->resource->role[OLD] =
118 		conn_highest_role(first_connection(resource));
119 	state_change->resource->susp[OLD] = resource->susp;
120 	state_change->resource->susp_nod[OLD] = resource->susp_nod;
121 	state_change->resource->susp_fen[OLD] = resource->susp_fen;
122 
123 	connection_state_change = state_change->connections;
124 	for_each_connection(connection, resource) {
125 		kref_get(&connection->kref);
126 		connection_state_change->connection = connection;
127 		connection_state_change->cstate[OLD] =
128 			connection->cstate;
129 		connection_state_change->peer_role[OLD] =
130 			conn_highest_peer(connection);
131 		connection_state_change++;
132 	}
133 
134 	device_state_change = state_change->devices;
135 	peer_device_state_change = state_change->peer_devices;
136 	idr_for_each_entry(&resource->devices, device, vnr) {
137 		kref_get(&device->kref);
138 		device_state_change->device = device;
139 		device_state_change->disk_state[OLD] = device->state.disk;
140 
141 		/* The peer_devices for each device have to be enumerated in
142 		   the order of the connections. We may not use for_each_peer_device() here. */
143 		for_each_connection(connection, resource) {
144 			struct drbd_peer_device *peer_device;
145 
146 			peer_device = conn_peer_device(connection, device->vnr);
147 			peer_device_state_change->peer_device = peer_device;
148 			peer_device_state_change->disk_state[OLD] =
149 				device->state.pdsk;
150 			peer_device_state_change->repl_state[OLD] =
151 				max_t(enum drbd_conns,
152 				      C_WF_REPORT_PARAMS, device->state.conn);
153 			peer_device_state_change->resync_susp_user[OLD] =
154 				device->state.user_isp;
155 			peer_device_state_change->resync_susp_peer[OLD] =
156 				device->state.peer_isp;
157 			peer_device_state_change->resync_susp_dependency[OLD] =
158 				device->state.aftr_isp;
159 			peer_device_state_change++;
160 		}
161 		device_state_change++;
162 	}
163 
164 	return state_change;
165 }
166 
167 static void remember_new_state(struct drbd_state_change *state_change)
168 {
169 	struct drbd_resource_state_change *resource_state_change;
170 	struct drbd_resource *resource;
171 	unsigned int n;
172 
173 	if (!state_change)
174 		return;
175 
176 	resource_state_change = &state_change->resource[0];
177 	resource = resource_state_change->resource;
178 
179 	resource_state_change->role[NEW] =
180 		conn_highest_role(first_connection(resource));
181 	resource_state_change->susp[NEW] = resource->susp;
182 	resource_state_change->susp_nod[NEW] = resource->susp_nod;
183 	resource_state_change->susp_fen[NEW] = resource->susp_fen;
184 
185 	for (n = 0; n < state_change->n_devices; n++) {
186 		struct drbd_device_state_change *device_state_change =
187 			&state_change->devices[n];
188 		struct drbd_device *device = device_state_change->device;
189 
190 		device_state_change->disk_state[NEW] = device->state.disk;
191 	}
192 
193 	for (n = 0; n < state_change->n_connections; n++) {
194 		struct drbd_connection_state_change *connection_state_change =
195 			&state_change->connections[n];
196 		struct drbd_connection *connection =
197 			connection_state_change->connection;
198 
199 		connection_state_change->cstate[NEW] = connection->cstate;
200 		connection_state_change->peer_role[NEW] =
201 			conn_highest_peer(connection);
202 	}
203 
204 	for (n = 0; n < state_change->n_devices * state_change->n_connections; n++) {
205 		struct drbd_peer_device_state_change *peer_device_state_change =
206 			&state_change->peer_devices[n];
207 		struct drbd_device *device =
208 			peer_device_state_change->peer_device->device;
209 		union drbd_dev_state state = device->state;
210 
211 		peer_device_state_change->disk_state[NEW] = state.pdsk;
212 		peer_device_state_change->repl_state[NEW] =
213 			max_t(enum drbd_conns, C_WF_REPORT_PARAMS, state.conn);
214 		peer_device_state_change->resync_susp_user[NEW] =
215 			state.user_isp;
216 		peer_device_state_change->resync_susp_peer[NEW] =
217 			state.peer_isp;
218 		peer_device_state_change->resync_susp_dependency[NEW] =
219 			state.aftr_isp;
220 	}
221 }
222 
223 void copy_old_to_new_state_change(struct drbd_state_change *state_change)
224 {
225 	struct drbd_resource_state_change *resource_state_change = &state_change->resource[0];
226 	unsigned int n_device, n_connection, n_peer_device, n_peer_devices;
227 
228 #define OLD_TO_NEW(x) \
229 	(x[NEW] = x[OLD])
230 
231 	OLD_TO_NEW(resource_state_change->role);
232 	OLD_TO_NEW(resource_state_change->susp);
233 	OLD_TO_NEW(resource_state_change->susp_nod);
234 	OLD_TO_NEW(resource_state_change->susp_fen);
235 
236 	for (n_connection = 0; n_connection < state_change->n_connections; n_connection++) {
237 		struct drbd_connection_state_change *connection_state_change =
238 				&state_change->connections[n_connection];
239 
240 		OLD_TO_NEW(connection_state_change->peer_role);
241 		OLD_TO_NEW(connection_state_change->cstate);
242 	}
243 
244 	for (n_device = 0; n_device < state_change->n_devices; n_device++) {
245 		struct drbd_device_state_change *device_state_change =
246 			&state_change->devices[n_device];
247 
248 		OLD_TO_NEW(device_state_change->disk_state);
249 	}
250 
251 	n_peer_devices = state_change->n_devices * state_change->n_connections;
252 	for (n_peer_device = 0; n_peer_device < n_peer_devices; n_peer_device++) {
253 		struct drbd_peer_device_state_change *p =
254 			&state_change->peer_devices[n_peer_device];
255 
256 		OLD_TO_NEW(p->disk_state);
257 		OLD_TO_NEW(p->repl_state);
258 		OLD_TO_NEW(p->resync_susp_user);
259 		OLD_TO_NEW(p->resync_susp_peer);
260 		OLD_TO_NEW(p->resync_susp_dependency);
261 	}
262 
263 #undef OLD_TO_NEW
264 }
265 
266 void forget_state_change(struct drbd_state_change *state_change)
267 {
268 	unsigned int n;
269 
270 	if (!state_change)
271 		return;
272 
273 	if (state_change->resource->resource)
274 		kref_put(&state_change->resource->resource->kref, drbd_destroy_resource);
275 	for (n = 0; n < state_change->n_devices; n++) {
276 		struct drbd_device *device = state_change->devices[n].device;
277 
278 		if (device)
279 			kref_put(&device->kref, drbd_destroy_device);
280 	}
281 	for (n = 0; n < state_change->n_connections; n++) {
282 		struct drbd_connection *connection =
283 			state_change->connections[n].connection;
284 
285 		if (connection)
286 			kref_put(&connection->kref, drbd_destroy_connection);
287 	}
288 	kfree(state_change);
289 }
290 
291 static int w_after_state_ch(struct drbd_work *w, int unused);
292 static void after_state_ch(struct drbd_device *device, union drbd_state os,
293 			   union drbd_state ns, enum chg_state_flags flags,
294 			   struct drbd_state_change *);
295 static enum drbd_state_rv is_valid_state(struct drbd_device *, union drbd_state);
296 static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state, struct drbd_connection *);
297 static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns);
298 static union drbd_state sanitize_state(struct drbd_device *device, union drbd_state os,
299 				       union drbd_state ns, enum sanitize_state_warnings *warn);
300 
301 static inline bool is_susp(union drbd_state s)
302 {
303         return s.susp || s.susp_nod || s.susp_fen;
304 }
305 
306 bool conn_all_vols_unconf(struct drbd_connection *connection)
307 {
308 	struct drbd_peer_device *peer_device;
309 	bool rv = true;
310 	int vnr;
311 
312 	rcu_read_lock();
313 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
314 		struct drbd_device *device = peer_device->device;
315 		if (device->state.disk != D_DISKLESS ||
316 		    device->state.conn != C_STANDALONE ||
317 		    device->state.role != R_SECONDARY) {
318 			rv = false;
319 			break;
320 		}
321 	}
322 	rcu_read_unlock();
323 
324 	return rv;
325 }
326 
327 /* Unfortunately the states where not correctly ordered, when
328    they where defined. therefore can not use max_t() here. */
329 static enum drbd_role max_role(enum drbd_role role1, enum drbd_role role2)
330 {
331 	if (role1 == R_PRIMARY || role2 == R_PRIMARY)
332 		return R_PRIMARY;
333 	if (role1 == R_SECONDARY || role2 == R_SECONDARY)
334 		return R_SECONDARY;
335 	return R_UNKNOWN;
336 }
337 
338 static enum drbd_role min_role(enum drbd_role role1, enum drbd_role role2)
339 {
340 	if (role1 == R_UNKNOWN || role2 == R_UNKNOWN)
341 		return R_UNKNOWN;
342 	if (role1 == R_SECONDARY || role2 == R_SECONDARY)
343 		return R_SECONDARY;
344 	return R_PRIMARY;
345 }
346 
347 enum drbd_role conn_highest_role(struct drbd_connection *connection)
348 {
349 	enum drbd_role role = R_SECONDARY;
350 	struct drbd_peer_device *peer_device;
351 	int vnr;
352 
353 	rcu_read_lock();
354 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
355 		struct drbd_device *device = peer_device->device;
356 		role = max_role(role, device->state.role);
357 	}
358 	rcu_read_unlock();
359 
360 	return role;
361 }
362 
363 enum drbd_role conn_highest_peer(struct drbd_connection *connection)
364 {
365 	enum drbd_role peer = R_UNKNOWN;
366 	struct drbd_peer_device *peer_device;
367 	int vnr;
368 
369 	rcu_read_lock();
370 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
371 		struct drbd_device *device = peer_device->device;
372 		peer = max_role(peer, device->state.peer);
373 	}
374 	rcu_read_unlock();
375 
376 	return peer;
377 }
378 
379 enum drbd_disk_state conn_highest_disk(struct drbd_connection *connection)
380 {
381 	enum drbd_disk_state disk_state = D_DISKLESS;
382 	struct drbd_peer_device *peer_device;
383 	int vnr;
384 
385 	rcu_read_lock();
386 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
387 		struct drbd_device *device = peer_device->device;
388 		disk_state = max_t(enum drbd_disk_state, disk_state, device->state.disk);
389 	}
390 	rcu_read_unlock();
391 
392 	return disk_state;
393 }
394 
395 enum drbd_disk_state conn_lowest_disk(struct drbd_connection *connection)
396 {
397 	enum drbd_disk_state disk_state = D_MASK;
398 	struct drbd_peer_device *peer_device;
399 	int vnr;
400 
401 	rcu_read_lock();
402 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
403 		struct drbd_device *device = peer_device->device;
404 		disk_state = min_t(enum drbd_disk_state, disk_state, device->state.disk);
405 	}
406 	rcu_read_unlock();
407 
408 	return disk_state;
409 }
410 
411 enum drbd_disk_state conn_highest_pdsk(struct drbd_connection *connection)
412 {
413 	enum drbd_disk_state disk_state = D_DISKLESS;
414 	struct drbd_peer_device *peer_device;
415 	int vnr;
416 
417 	rcu_read_lock();
418 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
419 		struct drbd_device *device = peer_device->device;
420 		disk_state = max_t(enum drbd_disk_state, disk_state, device->state.pdsk);
421 	}
422 	rcu_read_unlock();
423 
424 	return disk_state;
425 }
426 
427 enum drbd_conns conn_lowest_conn(struct drbd_connection *connection)
428 {
429 	enum drbd_conns conn = C_MASK;
430 	struct drbd_peer_device *peer_device;
431 	int vnr;
432 
433 	rcu_read_lock();
434 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
435 		struct drbd_device *device = peer_device->device;
436 		conn = min_t(enum drbd_conns, conn, device->state.conn);
437 	}
438 	rcu_read_unlock();
439 
440 	return conn;
441 }
442 
443 static bool no_peer_wf_report_params(struct drbd_connection *connection)
444 {
445 	struct drbd_peer_device *peer_device;
446 	int vnr;
447 	bool rv = true;
448 
449 	rcu_read_lock();
450 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
451 		if (peer_device->device->state.conn == C_WF_REPORT_PARAMS) {
452 			rv = false;
453 			break;
454 		}
455 	rcu_read_unlock();
456 
457 	return rv;
458 }
459 
460 static void wake_up_all_devices(struct drbd_connection *connection)
461 {
462 	struct drbd_peer_device *peer_device;
463 	int vnr;
464 
465 	rcu_read_lock();
466 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
467 		wake_up(&peer_device->device->state_wait);
468 	rcu_read_unlock();
469 
470 }
471 
472 
473 /**
474  * cl_wide_st_chg() - true if the state change is a cluster wide one
475  * @device:	DRBD device.
476  * @os:		old (current) state.
477  * @ns:		new (wanted) state.
478  */
479 static int cl_wide_st_chg(struct drbd_device *device,
480 			  union drbd_state os, union drbd_state ns)
481 {
482 	return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED &&
483 		 ((os.role != R_PRIMARY && ns.role == R_PRIMARY) ||
484 		  (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
485 		  (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) ||
486 		  (os.disk != D_FAILED && ns.disk == D_FAILED))) ||
487 		(os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) ||
488 		(os.conn == C_CONNECTED && ns.conn == C_VERIFY_S) ||
489 		(os.conn == C_CONNECTED && ns.conn == C_WF_REPORT_PARAMS);
490 }
491 
492 static union drbd_state
493 apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val)
494 {
495 	union drbd_state ns;
496 	ns.i = (os.i & ~mask.i) | val.i;
497 	return ns;
498 }
499 
500 enum drbd_state_rv
501 drbd_change_state(struct drbd_device *device, enum chg_state_flags f,
502 		  union drbd_state mask, union drbd_state val)
503 {
504 	unsigned long flags;
505 	union drbd_state ns;
506 	enum drbd_state_rv rv;
507 
508 	spin_lock_irqsave(&device->resource->req_lock, flags);
509 	ns = apply_mask_val(drbd_read_state(device), mask, val);
510 	rv = _drbd_set_state(device, ns, f, NULL);
511 	spin_unlock_irqrestore(&device->resource->req_lock, flags);
512 
513 	return rv;
514 }
515 
516 /**
517  * drbd_force_state() - Impose a change which happens outside our control on our state
518  * @device:	DRBD device.
519  * @mask:	mask of state bits to change.
520  * @val:	value of new state bits.
521  */
522 void drbd_force_state(struct drbd_device *device,
523 	union drbd_state mask, union drbd_state val)
524 {
525 	drbd_change_state(device, CS_HARD, mask, val);
526 }
527 
528 static enum drbd_state_rv
529 _req_st_cond(struct drbd_device *device, union drbd_state mask,
530 	     union drbd_state val)
531 {
532 	union drbd_state os, ns;
533 	unsigned long flags;
534 	enum drbd_state_rv rv;
535 
536 	if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &device->flags))
537 		return SS_CW_SUCCESS;
538 
539 	if (test_and_clear_bit(CL_ST_CHG_FAIL, &device->flags))
540 		return SS_CW_FAILED_BY_PEER;
541 
542 	spin_lock_irqsave(&device->resource->req_lock, flags);
543 	os = drbd_read_state(device);
544 	ns = sanitize_state(device, os, apply_mask_val(os, mask, val), NULL);
545 	rv = is_valid_transition(os, ns);
546 	if (rv >= SS_SUCCESS)
547 		rv = SS_UNKNOWN_ERROR;  /* cont waiting, otherwise fail. */
548 
549 	if (!cl_wide_st_chg(device, os, ns))
550 		rv = SS_CW_NO_NEED;
551 	if (rv == SS_UNKNOWN_ERROR) {
552 		rv = is_valid_state(device, ns);
553 		if (rv >= SS_SUCCESS) {
554 			rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection);
555 			if (rv >= SS_SUCCESS)
556 				rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
557 		}
558 	}
559 	spin_unlock_irqrestore(&device->resource->req_lock, flags);
560 
561 	return rv;
562 }
563 
564 /**
565  * drbd_req_state() - Perform an eventually cluster wide state change
566  * @device:	DRBD device.
567  * @mask:	mask of state bits to change.
568  * @val:	value of new state bits.
569  * @f:		flags
570  *
571  * Should not be called directly, use drbd_request_state() or
572  * _drbd_request_state().
573  */
574 static enum drbd_state_rv
575 drbd_req_state(struct drbd_device *device, union drbd_state mask,
576 	       union drbd_state val, enum chg_state_flags f)
577 {
578 	struct completion done;
579 	unsigned long flags;
580 	union drbd_state os, ns;
581 	enum drbd_state_rv rv;
582 	void *buffer = NULL;
583 
584 	init_completion(&done);
585 
586 	if (f & CS_SERIALIZE)
587 		mutex_lock(device->state_mutex);
588 	if (f & CS_INHIBIT_MD_IO)
589 		buffer = drbd_md_get_buffer(device, __func__);
590 
591 	spin_lock_irqsave(&device->resource->req_lock, flags);
592 	os = drbd_read_state(device);
593 	ns = sanitize_state(device, os, apply_mask_val(os, mask, val), NULL);
594 	rv = is_valid_transition(os, ns);
595 	if (rv < SS_SUCCESS) {
596 		spin_unlock_irqrestore(&device->resource->req_lock, flags);
597 		goto abort;
598 	}
599 
600 	if (cl_wide_st_chg(device, os, ns)) {
601 		rv = is_valid_state(device, ns);
602 		if (rv == SS_SUCCESS)
603 			rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection);
604 		spin_unlock_irqrestore(&device->resource->req_lock, flags);
605 
606 		if (rv < SS_SUCCESS) {
607 			if (f & CS_VERBOSE)
608 				print_st_err(device, os, ns, rv);
609 			goto abort;
610 		}
611 
612 		if (drbd_send_state_req(first_peer_device(device), mask, val)) {
613 			rv = SS_CW_FAILED_BY_PEER;
614 			if (f & CS_VERBOSE)
615 				print_st_err(device, os, ns, rv);
616 			goto abort;
617 		}
618 
619 		wait_event(device->state_wait,
620 			(rv = _req_st_cond(device, mask, val)));
621 
622 		if (rv < SS_SUCCESS) {
623 			if (f & CS_VERBOSE)
624 				print_st_err(device, os, ns, rv);
625 			goto abort;
626 		}
627 		spin_lock_irqsave(&device->resource->req_lock, flags);
628 		ns = apply_mask_val(drbd_read_state(device), mask, val);
629 		rv = _drbd_set_state(device, ns, f, &done);
630 	} else {
631 		rv = _drbd_set_state(device, ns, f, &done);
632 	}
633 
634 	spin_unlock_irqrestore(&device->resource->req_lock, flags);
635 
636 	if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) {
637 		D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
638 		wait_for_completion(&done);
639 	}
640 
641 abort:
642 	if (buffer)
643 		drbd_md_put_buffer(device);
644 	if (f & CS_SERIALIZE)
645 		mutex_unlock(device->state_mutex);
646 
647 	return rv;
648 }
649 
650 /**
651  * _drbd_request_state() - Request a state change (with flags)
652  * @device:	DRBD device.
653  * @mask:	mask of state bits to change.
654  * @val:	value of new state bits.
655  * @f:		flags
656  *
657  * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE
658  * flag, or when logging of failed state change requests is not desired.
659  */
660 enum drbd_state_rv
661 _drbd_request_state(struct drbd_device *device, union drbd_state mask,
662 		    union drbd_state val, enum chg_state_flags f)
663 {
664 	enum drbd_state_rv rv;
665 
666 	wait_event(device->state_wait,
667 		   (rv = drbd_req_state(device, mask, val, f)) != SS_IN_TRANSIENT_STATE);
668 
669 	return rv;
670 }
671 
672 /*
673  * We grab drbd_md_get_buffer(), because we don't want to "fail" the disk while
674  * there is IO in-flight: the transition into D_FAILED for detach purposes
675  * may get misinterpreted as actual IO error in a confused endio function.
676  *
677  * We wrap it all into wait_event(), to retry in case the drbd_req_state()
678  * returns SS_IN_TRANSIENT_STATE.
679  *
680  * To avoid potential deadlock with e.g. the receiver thread trying to grab
681  * drbd_md_get_buffer() while trying to get out of the "transient state", we
682  * need to grab and release the meta data buffer inside of that wait_event loop.
683  */
684 static enum drbd_state_rv
685 request_detach(struct drbd_device *device)
686 {
687 	return drbd_req_state(device, NS(disk, D_FAILED),
688 			CS_VERBOSE | CS_ORDERED | CS_INHIBIT_MD_IO);
689 }
690 
691 enum drbd_state_rv
692 drbd_request_detach_interruptible(struct drbd_device *device)
693 {
694 	enum drbd_state_rv rv;
695 	int ret;
696 
697 	drbd_suspend_io(device); /* so no-one is stuck in drbd_al_begin_io */
698 	wait_event_interruptible(device->state_wait,
699 		(rv = request_detach(device)) != SS_IN_TRANSIENT_STATE);
700 	drbd_resume_io(device);
701 
702 	ret = wait_event_interruptible(device->misc_wait,
703 			device->state.disk != D_FAILED);
704 
705 	if (rv == SS_IS_DISKLESS)
706 		rv = SS_NOTHING_TO_DO;
707 	if (ret)
708 		rv = ERR_INTR;
709 
710 	return rv;
711 }
712 
713 enum drbd_state_rv
714 _drbd_request_state_holding_state_mutex(struct drbd_device *device, union drbd_state mask,
715 		    union drbd_state val, enum chg_state_flags f)
716 {
717 	enum drbd_state_rv rv;
718 
719 	BUG_ON(f & CS_SERIALIZE);
720 
721 	wait_event_cmd(device->state_wait,
722 		       (rv = drbd_req_state(device, mask, val, f)) != SS_IN_TRANSIENT_STATE,
723 		       mutex_unlock(device->state_mutex),
724 		       mutex_lock(device->state_mutex));
725 
726 	return rv;
727 }
728 
729 static void print_st(struct drbd_device *device, const char *name, union drbd_state ns)
730 {
731 	drbd_err(device, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n",
732 	    name,
733 	    drbd_conn_str(ns.conn),
734 	    drbd_role_str(ns.role),
735 	    drbd_role_str(ns.peer),
736 	    drbd_disk_str(ns.disk),
737 	    drbd_disk_str(ns.pdsk),
738 	    is_susp(ns) ? 's' : 'r',
739 	    ns.aftr_isp ? 'a' : '-',
740 	    ns.peer_isp ? 'p' : '-',
741 	    ns.user_isp ? 'u' : '-',
742 	    ns.susp_fen ? 'F' : '-',
743 	    ns.susp_nod ? 'N' : '-'
744 	    );
745 }
746 
747 void print_st_err(struct drbd_device *device, union drbd_state os,
748 	          union drbd_state ns, enum drbd_state_rv err)
749 {
750 	if (err == SS_IN_TRANSIENT_STATE)
751 		return;
752 	drbd_err(device, "State change failed: %s\n", drbd_set_st_err_str(err));
753 	print_st(device, " state", os);
754 	print_st(device, "wanted", ns);
755 }
756 
757 static long print_state_change(char *pb, union drbd_state os, union drbd_state ns,
758 			       enum chg_state_flags flags)
759 {
760 	char *pbp;
761 	pbp = pb;
762 	*pbp = 0;
763 
764 	if (ns.role != os.role && flags & CS_DC_ROLE)
765 		pbp += sprintf(pbp, "role( %s -> %s ) ",
766 			       drbd_role_str(os.role),
767 			       drbd_role_str(ns.role));
768 	if (ns.peer != os.peer && flags & CS_DC_PEER)
769 		pbp += sprintf(pbp, "peer( %s -> %s ) ",
770 			       drbd_role_str(os.peer),
771 			       drbd_role_str(ns.peer));
772 	if (ns.conn != os.conn && flags & CS_DC_CONN)
773 		pbp += sprintf(pbp, "conn( %s -> %s ) ",
774 			       drbd_conn_str(os.conn),
775 			       drbd_conn_str(ns.conn));
776 	if (ns.disk != os.disk && flags & CS_DC_DISK)
777 		pbp += sprintf(pbp, "disk( %s -> %s ) ",
778 			       drbd_disk_str(os.disk),
779 			       drbd_disk_str(ns.disk));
780 	if (ns.pdsk != os.pdsk && flags & CS_DC_PDSK)
781 		pbp += sprintf(pbp, "pdsk( %s -> %s ) ",
782 			       drbd_disk_str(os.pdsk),
783 			       drbd_disk_str(ns.pdsk));
784 
785 	return pbp - pb;
786 }
787 
788 static void drbd_pr_state_change(struct drbd_device *device, union drbd_state os, union drbd_state ns,
789 				 enum chg_state_flags flags)
790 {
791 	char pb[300];
792 	char *pbp = pb;
793 
794 	pbp += print_state_change(pbp, os, ns, flags ^ CS_DC_MASK);
795 
796 	if (ns.aftr_isp != os.aftr_isp)
797 		pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ",
798 			       os.aftr_isp,
799 			       ns.aftr_isp);
800 	if (ns.peer_isp != os.peer_isp)
801 		pbp += sprintf(pbp, "peer_isp( %d -> %d ) ",
802 			       os.peer_isp,
803 			       ns.peer_isp);
804 	if (ns.user_isp != os.user_isp)
805 		pbp += sprintf(pbp, "user_isp( %d -> %d ) ",
806 			       os.user_isp,
807 			       ns.user_isp);
808 
809 	if (pbp != pb)
810 		drbd_info(device, "%s\n", pb);
811 }
812 
813 static void conn_pr_state_change(struct drbd_connection *connection, union drbd_state os, union drbd_state ns,
814 				 enum chg_state_flags flags)
815 {
816 	char pb[300];
817 	char *pbp = pb;
818 
819 	pbp += print_state_change(pbp, os, ns, flags);
820 
821 	if (is_susp(ns) != is_susp(os) && flags & CS_DC_SUSP)
822 		pbp += sprintf(pbp, "susp( %d -> %d ) ",
823 			       is_susp(os),
824 			       is_susp(ns));
825 
826 	if (pbp != pb)
827 		drbd_info(connection, "%s\n", pb);
828 }
829 
830 
831 /**
832  * is_valid_state() - Returns an SS_ error code if ns is not valid
833  * @device:	DRBD device.
834  * @ns:		State to consider.
835  */
836 static enum drbd_state_rv
837 is_valid_state(struct drbd_device *device, union drbd_state ns)
838 {
839 	/* See drbd_state_sw_errors in drbd_strings.c */
840 
841 	enum drbd_fencing_p fp;
842 	enum drbd_state_rv rv = SS_SUCCESS;
843 	struct net_conf *nc;
844 
845 	rcu_read_lock();
846 	fp = FP_DONT_CARE;
847 	if (get_ldev(device)) {
848 		fp = rcu_dereference(device->ldev->disk_conf)->fencing;
849 		put_ldev(device);
850 	}
851 
852 	nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
853 	if (nc) {
854 		if (!nc->two_primaries && ns.role == R_PRIMARY) {
855 			if (ns.peer == R_PRIMARY)
856 				rv = SS_TWO_PRIMARIES;
857 			else if (conn_highest_peer(first_peer_device(device)->connection) == R_PRIMARY)
858 				rv = SS_O_VOL_PEER_PRI;
859 		}
860 	}
861 
862 	if (rv <= 0)
863 		goto out; /* already found a reason to abort */
864 	else if (ns.role == R_SECONDARY && device->open_cnt)
865 		rv = SS_DEVICE_IN_USE;
866 
867 	else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE)
868 		rv = SS_NO_UP_TO_DATE_DISK;
869 
870 	else if (fp >= FP_RESOURCE &&
871 		 ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN)
872 		rv = SS_PRIMARY_NOP;
873 
874 	else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT)
875 		rv = SS_NO_UP_TO_DATE_DISK;
876 
877 	else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT)
878 		rv = SS_NO_LOCAL_DISK;
879 
880 	else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT)
881 		rv = SS_NO_REMOTE_DISK;
882 
883 	else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
884 		rv = SS_NO_UP_TO_DATE_DISK;
885 
886 	else if ((ns.conn == C_CONNECTED ||
887 		  ns.conn == C_WF_BITMAP_S ||
888 		  ns.conn == C_SYNC_SOURCE ||
889 		  ns.conn == C_PAUSED_SYNC_S) &&
890 		  ns.disk == D_OUTDATED)
891 		rv = SS_CONNECTED_OUTDATES;
892 
893 	else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
894 		 (nc->verify_alg[0] == 0))
895 		rv = SS_NO_VERIFY_ALG;
896 
897 	else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
898 		  first_peer_device(device)->connection->agreed_pro_version < 88)
899 		rv = SS_NOT_SUPPORTED;
900 
901 	else if (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
902 		rv = SS_NO_UP_TO_DATE_DISK;
903 
904 	else if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
905                  ns.pdsk == D_UNKNOWN)
906 		rv = SS_NEED_CONNECTION;
907 
908 	else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN)
909 		rv = SS_CONNECTED_OUTDATES;
910 
911 out:
912 	rcu_read_unlock();
913 
914 	return rv;
915 }
916 
917 /**
918  * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible
919  * This function limits state transitions that may be declined by DRBD. I.e.
920  * user requests (aka soft transitions).
921  * @device:	DRBD device.
922  * @ns:		new state.
923  * @os:		old state.
924  */
925 static enum drbd_state_rv
926 is_valid_soft_transition(union drbd_state os, union drbd_state ns, struct drbd_connection *connection)
927 {
928 	enum drbd_state_rv rv = SS_SUCCESS;
929 
930 	if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
931 	    os.conn > C_CONNECTED)
932 		rv = SS_RESYNC_RUNNING;
933 
934 	if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
935 		rv = SS_ALREADY_STANDALONE;
936 
937 	if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
938 		rv = SS_IS_DISKLESS;
939 
940 	if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
941 		rv = SS_NO_NET_CONFIG;
942 
943 	if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
944 		rv = SS_LOWER_THAN_OUTDATED;
945 
946 	if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
947 		rv = SS_IN_TRANSIENT_STATE;
948 
949 	/* While establishing a connection only allow cstate to change.
950 	   Delay/refuse role changes, detach attach etc... (they do not touch cstate) */
951 	if (test_bit(STATE_SENT, &connection->flags) &&
952 	    !((ns.conn == C_WF_REPORT_PARAMS && os.conn == C_WF_CONNECTION) ||
953 	      (ns.conn >= C_CONNECTED && os.conn == C_WF_REPORT_PARAMS)))
954 		rv = SS_IN_TRANSIENT_STATE;
955 
956 	/* Do not promote during resync handshake triggered by "force primary".
957 	 * This is a hack. It should really be rejected by the peer during the
958 	 * cluster wide state change request. */
959 	if (os.role != R_PRIMARY && ns.role == R_PRIMARY
960 		&& ns.pdsk == D_UP_TO_DATE
961 		&& ns.disk != D_UP_TO_DATE && ns.disk != D_DISKLESS
962 		&& (ns.conn <= C_WF_SYNC_UUID || ns.conn != os.conn))
963 			rv = SS_IN_TRANSIENT_STATE;
964 
965 	if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
966 		rv = SS_NEED_CONNECTION;
967 
968 	if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
969 	    ns.conn != os.conn && os.conn > C_CONNECTED)
970 		rv = SS_RESYNC_RUNNING;
971 
972 	if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
973 	    os.conn < C_CONNECTED)
974 		rv = SS_NEED_CONNECTION;
975 
976 	if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)
977 	    && os.conn < C_WF_REPORT_PARAMS)
978 		rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */
979 
980 	if (ns.conn == C_DISCONNECTING && ns.pdsk == D_OUTDATED &&
981 	    os.conn < C_CONNECTED && os.pdsk > D_OUTDATED)
982 		rv = SS_OUTDATE_WO_CONN;
983 
984 	return rv;
985 }
986 
987 static enum drbd_state_rv
988 is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc)
989 {
990 	/* no change -> nothing to do, at least for the connection part */
991 	if (oc == nc)
992 		return SS_NOTHING_TO_DO;
993 
994 	/* disconnect of an unconfigured connection does not make sense */
995 	if (oc == C_STANDALONE && nc == C_DISCONNECTING)
996 		return SS_ALREADY_STANDALONE;
997 
998 	/* from C_STANDALONE, we start with C_UNCONNECTED */
999 	if (oc == C_STANDALONE && nc != C_UNCONNECTED)
1000 		return SS_NEED_CONNECTION;
1001 
1002 	/* When establishing a connection we need to go through WF_REPORT_PARAMS!
1003 	   Necessary to do the right thing upon invalidate-remote on a disconnected resource */
1004 	if (oc < C_WF_REPORT_PARAMS && nc >= C_CONNECTED)
1005 		return SS_NEED_CONNECTION;
1006 
1007 	/* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */
1008 	if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING)
1009 		return SS_IN_TRANSIENT_STATE;
1010 
1011 	/* After C_DISCONNECTING only C_STANDALONE may follow */
1012 	if (oc == C_DISCONNECTING && nc != C_STANDALONE)
1013 		return SS_IN_TRANSIENT_STATE;
1014 
1015 	return SS_SUCCESS;
1016 }
1017 
1018 
1019 /**
1020  * is_valid_transition() - Returns an SS_ error code if the state transition is not possible
1021  * This limits hard state transitions. Hard state transitions are facts there are
1022  * imposed on DRBD by the environment. E.g. disk broke or network broke down.
1023  * But those hard state transitions are still not allowed to do everything.
1024  * @ns:		new state.
1025  * @os:		old state.
1026  */
1027 static enum drbd_state_rv
1028 is_valid_transition(union drbd_state os, union drbd_state ns)
1029 {
1030 	enum drbd_state_rv rv;
1031 
1032 	rv = is_valid_conn_transition(os.conn, ns.conn);
1033 
1034 	/* we cannot fail (again) if we already detached */
1035 	if (ns.disk == D_FAILED && os.disk == D_DISKLESS)
1036 		rv = SS_IS_DISKLESS;
1037 
1038 	return rv;
1039 }
1040 
1041 static void print_sanitize_warnings(struct drbd_device *device, enum sanitize_state_warnings warn)
1042 {
1043 	static const char *msg_table[] = {
1044 		[NO_WARNING] = "",
1045 		[ABORTED_ONLINE_VERIFY] = "Online-verify aborted.",
1046 		[ABORTED_RESYNC] = "Resync aborted.",
1047 		[CONNECTION_LOST_NEGOTIATING] = "Connection lost while negotiating, no data!",
1048 		[IMPLICITLY_UPGRADED_DISK] = "Implicitly upgraded disk",
1049 		[IMPLICITLY_UPGRADED_PDSK] = "Implicitly upgraded pdsk",
1050 	};
1051 
1052 	if (warn != NO_WARNING)
1053 		drbd_warn(device, "%s\n", msg_table[warn]);
1054 }
1055 
1056 /**
1057  * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
1058  * @device:	DRBD device.
1059  * @os:		old state.
1060  * @ns:		new state.
1061  * @warn_sync_abort:
1062  *
1063  * When we loose connection, we have to set the state of the peers disk (pdsk)
1064  * to D_UNKNOWN. This rule and many more along those lines are in this function.
1065  */
1066 static union drbd_state sanitize_state(struct drbd_device *device, union drbd_state os,
1067 				       union drbd_state ns, enum sanitize_state_warnings *warn)
1068 {
1069 	enum drbd_fencing_p fp;
1070 	enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max;
1071 
1072 	if (warn)
1073 		*warn = NO_WARNING;
1074 
1075 	fp = FP_DONT_CARE;
1076 	if (get_ldev(device)) {
1077 		rcu_read_lock();
1078 		fp = rcu_dereference(device->ldev->disk_conf)->fencing;
1079 		rcu_read_unlock();
1080 		put_ldev(device);
1081 	}
1082 
1083 	/* Implications from connection to peer and peer_isp */
1084 	if (ns.conn < C_CONNECTED) {
1085 		ns.peer_isp = 0;
1086 		ns.peer = R_UNKNOWN;
1087 		if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
1088 			ns.pdsk = D_UNKNOWN;
1089 	}
1090 
1091 	/* Clear the aftr_isp when becoming unconfigured */
1092 	if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
1093 		ns.aftr_isp = 0;
1094 
1095 	/* An implication of the disk states onto the connection state */
1096 	/* Abort resync if a disk fails/detaches */
1097 	if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
1098 		if (warn)
1099 			*warn = ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ?
1100 				ABORTED_ONLINE_VERIFY : ABORTED_RESYNC;
1101 		ns.conn = C_CONNECTED;
1102 	}
1103 
1104 	/* Connection breaks down before we finished "Negotiating" */
1105 	if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
1106 	    get_ldev_if_state(device, D_NEGOTIATING)) {
1107 		if (device->ed_uuid == device->ldev->md.uuid[UI_CURRENT]) {
1108 			ns.disk = device->new_state_tmp.disk;
1109 			ns.pdsk = device->new_state_tmp.pdsk;
1110 		} else {
1111 			if (warn)
1112 				*warn = CONNECTION_LOST_NEGOTIATING;
1113 			ns.disk = D_DISKLESS;
1114 			ns.pdsk = D_UNKNOWN;
1115 		}
1116 		put_ldev(device);
1117 	}
1118 
1119 	/* D_CONSISTENT and D_OUTDATED vanish when we get connected */
1120 	if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) {
1121 		if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED)
1122 			ns.disk = D_UP_TO_DATE;
1123 		if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)
1124 			ns.pdsk = D_UP_TO_DATE;
1125 	}
1126 
1127 	/* Implications of the connection stat on the disk states */
1128 	disk_min = D_DISKLESS;
1129 	disk_max = D_UP_TO_DATE;
1130 	pdsk_min = D_INCONSISTENT;
1131 	pdsk_max = D_UNKNOWN;
1132 	switch ((enum drbd_conns)ns.conn) {
1133 	case C_WF_BITMAP_T:
1134 	case C_PAUSED_SYNC_T:
1135 	case C_STARTING_SYNC_T:
1136 	case C_WF_SYNC_UUID:
1137 	case C_BEHIND:
1138 		disk_min = D_INCONSISTENT;
1139 		disk_max = D_OUTDATED;
1140 		pdsk_min = D_UP_TO_DATE;
1141 		pdsk_max = D_UP_TO_DATE;
1142 		break;
1143 	case C_VERIFY_S:
1144 	case C_VERIFY_T:
1145 		disk_min = D_UP_TO_DATE;
1146 		disk_max = D_UP_TO_DATE;
1147 		pdsk_min = D_UP_TO_DATE;
1148 		pdsk_max = D_UP_TO_DATE;
1149 		break;
1150 	case C_CONNECTED:
1151 		disk_min = D_DISKLESS;
1152 		disk_max = D_UP_TO_DATE;
1153 		pdsk_min = D_DISKLESS;
1154 		pdsk_max = D_UP_TO_DATE;
1155 		break;
1156 	case C_WF_BITMAP_S:
1157 	case C_PAUSED_SYNC_S:
1158 	case C_STARTING_SYNC_S:
1159 	case C_AHEAD:
1160 		disk_min = D_UP_TO_DATE;
1161 		disk_max = D_UP_TO_DATE;
1162 		pdsk_min = D_INCONSISTENT;
1163 		pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/
1164 		break;
1165 	case C_SYNC_TARGET:
1166 		disk_min = D_INCONSISTENT;
1167 		disk_max = D_INCONSISTENT;
1168 		pdsk_min = D_UP_TO_DATE;
1169 		pdsk_max = D_UP_TO_DATE;
1170 		break;
1171 	case C_SYNC_SOURCE:
1172 		disk_min = D_UP_TO_DATE;
1173 		disk_max = D_UP_TO_DATE;
1174 		pdsk_min = D_INCONSISTENT;
1175 		pdsk_max = D_INCONSISTENT;
1176 		break;
1177 	case C_STANDALONE:
1178 	case C_DISCONNECTING:
1179 	case C_UNCONNECTED:
1180 	case C_TIMEOUT:
1181 	case C_BROKEN_PIPE:
1182 	case C_NETWORK_FAILURE:
1183 	case C_PROTOCOL_ERROR:
1184 	case C_TEAR_DOWN:
1185 	case C_WF_CONNECTION:
1186 	case C_WF_REPORT_PARAMS:
1187 	case C_MASK:
1188 		break;
1189 	}
1190 	if (ns.disk > disk_max)
1191 		ns.disk = disk_max;
1192 
1193 	if (ns.disk < disk_min) {
1194 		if (warn)
1195 			*warn = IMPLICITLY_UPGRADED_DISK;
1196 		ns.disk = disk_min;
1197 	}
1198 	if (ns.pdsk > pdsk_max)
1199 		ns.pdsk = pdsk_max;
1200 
1201 	if (ns.pdsk < pdsk_min) {
1202 		if (warn)
1203 			*warn = IMPLICITLY_UPGRADED_PDSK;
1204 		ns.pdsk = pdsk_min;
1205 	}
1206 
1207 	if (fp == FP_STONITH &&
1208 	    (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED) &&
1209 	    !(os.role == R_PRIMARY && os.conn < C_CONNECTED && os.pdsk > D_OUTDATED))
1210 		ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */
1211 
1212 	if (device->resource->res_opts.on_no_data == OND_SUSPEND_IO &&
1213 	    (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE) &&
1214 	    !(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE))
1215 		ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */
1216 
1217 	if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
1218 		if (ns.conn == C_SYNC_SOURCE)
1219 			ns.conn = C_PAUSED_SYNC_S;
1220 		if (ns.conn == C_SYNC_TARGET)
1221 			ns.conn = C_PAUSED_SYNC_T;
1222 	} else {
1223 		if (ns.conn == C_PAUSED_SYNC_S)
1224 			ns.conn = C_SYNC_SOURCE;
1225 		if (ns.conn == C_PAUSED_SYNC_T)
1226 			ns.conn = C_SYNC_TARGET;
1227 	}
1228 
1229 	return ns;
1230 }
1231 
1232 void drbd_resume_al(struct drbd_device *device)
1233 {
1234 	if (test_and_clear_bit(AL_SUSPENDED, &device->flags))
1235 		drbd_info(device, "Resumed AL updates\n");
1236 }
1237 
1238 /* helper for _drbd_set_state */
1239 static void set_ov_position(struct drbd_device *device, enum drbd_conns cs)
1240 {
1241 	if (first_peer_device(device)->connection->agreed_pro_version < 90)
1242 		device->ov_start_sector = 0;
1243 	device->rs_total = drbd_bm_bits(device);
1244 	device->ov_position = 0;
1245 	if (cs == C_VERIFY_T) {
1246 		/* starting online verify from an arbitrary position
1247 		 * does not fit well into the existing protocol.
1248 		 * on C_VERIFY_T, we initialize ov_left and friends
1249 		 * implicitly in receive_DataRequest once the
1250 		 * first P_OV_REQUEST is received */
1251 		device->ov_start_sector = ~(sector_t)0;
1252 	} else {
1253 		unsigned long bit = BM_SECT_TO_BIT(device->ov_start_sector);
1254 		if (bit >= device->rs_total) {
1255 			device->ov_start_sector =
1256 				BM_BIT_TO_SECT(device->rs_total - 1);
1257 			device->rs_total = 1;
1258 		} else
1259 			device->rs_total -= bit;
1260 		device->ov_position = device->ov_start_sector;
1261 	}
1262 	device->ov_left = device->rs_total;
1263 }
1264 
1265 /**
1266  * _drbd_set_state() - Set a new DRBD state
1267  * @device:	DRBD device.
1268  * @ns:		new state.
1269  * @flags:	Flags
1270  * @done:	Optional completion, that will get completed after the after_state_ch() finished
1271  *
1272  * Caller needs to hold req_lock. Do not call directly.
1273  */
1274 enum drbd_state_rv
1275 _drbd_set_state(struct drbd_device *device, union drbd_state ns,
1276 	        enum chg_state_flags flags, struct completion *done)
1277 {
1278 	struct drbd_peer_device *peer_device = first_peer_device(device);
1279 	struct drbd_connection *connection = peer_device ? peer_device->connection : NULL;
1280 	union drbd_state os;
1281 	enum drbd_state_rv rv = SS_SUCCESS;
1282 	enum sanitize_state_warnings ssw;
1283 	struct after_state_chg_work *ascw;
1284 	struct drbd_state_change *state_change;
1285 
1286 	os = drbd_read_state(device);
1287 
1288 	ns = sanitize_state(device, os, ns, &ssw);
1289 	if (ns.i == os.i)
1290 		return SS_NOTHING_TO_DO;
1291 
1292 	rv = is_valid_transition(os, ns);
1293 	if (rv < SS_SUCCESS)
1294 		return rv;
1295 
1296 	if (!(flags & CS_HARD)) {
1297 		/*  pre-state-change checks ; only look at ns  */
1298 		/* See drbd_state_sw_errors in drbd_strings.c */
1299 
1300 		rv = is_valid_state(device, ns);
1301 		if (rv < SS_SUCCESS) {
1302 			/* If the old state was illegal as well, then let
1303 			   this happen...*/
1304 
1305 			if (is_valid_state(device, os) == rv)
1306 				rv = is_valid_soft_transition(os, ns, connection);
1307 		} else
1308 			rv = is_valid_soft_transition(os, ns, connection);
1309 	}
1310 
1311 	if (rv < SS_SUCCESS) {
1312 		if (flags & CS_VERBOSE)
1313 			print_st_err(device, os, ns, rv);
1314 		return rv;
1315 	}
1316 
1317 	print_sanitize_warnings(device, ssw);
1318 
1319 	drbd_pr_state_change(device, os, ns, flags);
1320 
1321 	/* Display changes to the susp* flags that where caused by the call to
1322 	   sanitize_state(). Only display it here if we where not called from
1323 	   _conn_request_state() */
1324 	if (!(flags & CS_DC_SUSP))
1325 		conn_pr_state_change(connection, os, ns,
1326 				     (flags & ~CS_DC_MASK) | CS_DC_SUSP);
1327 
1328 	/* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
1329 	 * on the ldev here, to be sure the transition -> D_DISKLESS resp.
1330 	 * drbd_ldev_destroy() won't happen before our corresponding
1331 	 * after_state_ch works run, where we put_ldev again. */
1332 	if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
1333 	    (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
1334 		atomic_inc(&device->local_cnt);
1335 
1336 	if (!is_sync_state(os.conn) && is_sync_state(ns.conn))
1337 		clear_bit(RS_DONE, &device->flags);
1338 
1339 	/* FIXME: Have any flags been set earlier in this function already? */
1340 	state_change = remember_old_state(device->resource, GFP_ATOMIC);
1341 
1342 	/* changes to local_cnt and device flags should be visible before
1343 	 * changes to state, which again should be visible before anything else
1344 	 * depending on that change happens. */
1345 	smp_wmb();
1346 	device->state.i = ns.i;
1347 	device->resource->susp = ns.susp;
1348 	device->resource->susp_nod = ns.susp_nod;
1349 	device->resource->susp_fen = ns.susp_fen;
1350 	smp_wmb();
1351 
1352 	remember_new_state(state_change);
1353 
1354 	/* put replicated vs not-replicated requests in seperate epochs */
1355 	if (drbd_should_do_remote((union drbd_dev_state)os.i) !=
1356 	    drbd_should_do_remote((union drbd_dev_state)ns.i))
1357 		start_new_tl_epoch(connection);
1358 
1359 	if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
1360 		drbd_print_uuids(device, "attached to UUIDs");
1361 
1362 	/* Wake up role changes, that were delayed because of connection establishing */
1363 	if (os.conn == C_WF_REPORT_PARAMS && ns.conn != C_WF_REPORT_PARAMS &&
1364 	    no_peer_wf_report_params(connection)) {
1365 		clear_bit(STATE_SENT, &connection->flags);
1366 		wake_up_all_devices(connection);
1367 	}
1368 
1369 	wake_up(&device->misc_wait);
1370 	wake_up(&device->state_wait);
1371 	wake_up(&connection->ping_wait);
1372 
1373 	/* Aborted verify run, or we reached the stop sector.
1374 	 * Log the last position, unless end-of-device. */
1375 	if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
1376 	    ns.conn <= C_CONNECTED) {
1377 		device->ov_start_sector =
1378 			BM_BIT_TO_SECT(drbd_bm_bits(device) - device->ov_left);
1379 		if (device->ov_left)
1380 			drbd_info(device, "Online Verify reached sector %llu\n",
1381 				(unsigned long long)device->ov_start_sector);
1382 	}
1383 
1384 	if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
1385 	    (ns.conn == C_SYNC_TARGET  || ns.conn == C_SYNC_SOURCE)) {
1386 		drbd_info(device, "Syncer continues.\n");
1387 		device->rs_paused += (long)jiffies
1388 				  -(long)device->rs_mark_time[device->rs_last_mark];
1389 		if (ns.conn == C_SYNC_TARGET)
1390 			mod_timer(&device->resync_timer, jiffies);
1391 	}
1392 
1393 	if ((os.conn == C_SYNC_TARGET  || os.conn == C_SYNC_SOURCE) &&
1394 	    (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
1395 		drbd_info(device, "Resync suspended\n");
1396 		device->rs_mark_time[device->rs_last_mark] = jiffies;
1397 	}
1398 
1399 	if (os.conn == C_CONNECTED &&
1400 	    (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
1401 		unsigned long now = jiffies;
1402 		int i;
1403 
1404 		set_ov_position(device, ns.conn);
1405 		device->rs_start = now;
1406 		device->rs_last_sect_ev = 0;
1407 		device->ov_last_oos_size = 0;
1408 		device->ov_last_oos_start = 0;
1409 
1410 		for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1411 			device->rs_mark_left[i] = device->ov_left;
1412 			device->rs_mark_time[i] = now;
1413 		}
1414 
1415 		drbd_rs_controller_reset(device);
1416 
1417 		if (ns.conn == C_VERIFY_S) {
1418 			drbd_info(device, "Starting Online Verify from sector %llu\n",
1419 					(unsigned long long)device->ov_position);
1420 			mod_timer(&device->resync_timer, jiffies);
1421 		}
1422 	}
1423 
1424 	if (get_ldev(device)) {
1425 		u32 mdf = device->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
1426 						 MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
1427 						 MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
1428 
1429 		mdf &= ~MDF_AL_CLEAN;
1430 		if (test_bit(CRASHED_PRIMARY, &device->flags))
1431 			mdf |= MDF_CRASHED_PRIMARY;
1432 		if (device->state.role == R_PRIMARY ||
1433 		    (device->state.pdsk < D_INCONSISTENT && device->state.peer == R_PRIMARY))
1434 			mdf |= MDF_PRIMARY_IND;
1435 		if (device->state.conn > C_WF_REPORT_PARAMS)
1436 			mdf |= MDF_CONNECTED_IND;
1437 		if (device->state.disk > D_INCONSISTENT)
1438 			mdf |= MDF_CONSISTENT;
1439 		if (device->state.disk > D_OUTDATED)
1440 			mdf |= MDF_WAS_UP_TO_DATE;
1441 		if (device->state.pdsk <= D_OUTDATED && device->state.pdsk >= D_INCONSISTENT)
1442 			mdf |= MDF_PEER_OUT_DATED;
1443 		if (mdf != device->ldev->md.flags) {
1444 			device->ldev->md.flags = mdf;
1445 			drbd_md_mark_dirty(device);
1446 		}
1447 		if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
1448 			drbd_set_ed_uuid(device, device->ldev->md.uuid[UI_CURRENT]);
1449 		put_ldev(device);
1450 	}
1451 
1452 	/* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
1453 	if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
1454 	    os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
1455 		set_bit(CONSIDER_RESYNC, &device->flags);
1456 
1457 	/* Receiver should clean up itself */
1458 	if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
1459 		drbd_thread_stop_nowait(&connection->receiver);
1460 
1461 	/* Now the receiver finished cleaning up itself, it should die */
1462 	if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
1463 		drbd_thread_stop_nowait(&connection->receiver);
1464 
1465 	/* Upon network failure, we need to restart the receiver. */
1466 	if (os.conn > C_WF_CONNECTION &&
1467 	    ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
1468 		drbd_thread_restart_nowait(&connection->receiver);
1469 
1470 	/* Resume AL writing if we get a connection */
1471 	if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) {
1472 		drbd_resume_al(device);
1473 		connection->connect_cnt++;
1474 	}
1475 
1476 	/* remember last attach time so request_timer_fn() won't
1477 	 * kill newly established sessions while we are still trying to thaw
1478 	 * previously frozen IO */
1479 	if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1480 	    ns.disk > D_NEGOTIATING)
1481 		device->last_reattach_jif = jiffies;
1482 
1483 	ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
1484 	if (ascw) {
1485 		ascw->os = os;
1486 		ascw->ns = ns;
1487 		ascw->flags = flags;
1488 		ascw->w.cb = w_after_state_ch;
1489 		ascw->device = device;
1490 		ascw->done = done;
1491 		ascw->state_change = state_change;
1492 		drbd_queue_work(&connection->sender_work,
1493 				&ascw->w);
1494 	} else {
1495 		drbd_err(device, "Could not kmalloc an ascw\n");
1496 	}
1497 
1498 	return rv;
1499 }
1500 
1501 static int w_after_state_ch(struct drbd_work *w, int unused)
1502 {
1503 	struct after_state_chg_work *ascw =
1504 		container_of(w, struct after_state_chg_work, w);
1505 	struct drbd_device *device = ascw->device;
1506 
1507 	after_state_ch(device, ascw->os, ascw->ns, ascw->flags, ascw->state_change);
1508 	forget_state_change(ascw->state_change);
1509 	if (ascw->flags & CS_WAIT_COMPLETE)
1510 		complete(ascw->done);
1511 	kfree(ascw);
1512 
1513 	return 0;
1514 }
1515 
1516 static void abw_start_sync(struct drbd_device *device, int rv)
1517 {
1518 	if (rv) {
1519 		drbd_err(device, "Writing the bitmap failed not starting resync.\n");
1520 		_drbd_request_state(device, NS(conn, C_CONNECTED), CS_VERBOSE);
1521 		return;
1522 	}
1523 
1524 	switch (device->state.conn) {
1525 	case C_STARTING_SYNC_T:
1526 		_drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
1527 		break;
1528 	case C_STARTING_SYNC_S:
1529 		drbd_start_resync(device, C_SYNC_SOURCE);
1530 		break;
1531 	}
1532 }
1533 
1534 int drbd_bitmap_io_from_worker(struct drbd_device *device,
1535 		int (*io_fn)(struct drbd_device *),
1536 		char *why, enum bm_flag flags)
1537 {
1538 	int rv;
1539 
1540 	D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
1541 
1542 	/* open coded non-blocking drbd_suspend_io(device); */
1543 	atomic_inc(&device->suspend_cnt);
1544 
1545 	drbd_bm_lock(device, why, flags);
1546 	rv = io_fn(device);
1547 	drbd_bm_unlock(device);
1548 
1549 	drbd_resume_io(device);
1550 
1551 	return rv;
1552 }
1553 
1554 void notify_resource_state_change(struct sk_buff *skb,
1555 				  unsigned int seq,
1556 				  struct drbd_resource_state_change *resource_state_change,
1557 				  enum drbd_notification_type type)
1558 {
1559 	struct drbd_resource *resource = resource_state_change->resource;
1560 	struct resource_info resource_info = {
1561 		.res_role = resource_state_change->role[NEW],
1562 		.res_susp = resource_state_change->susp[NEW],
1563 		.res_susp_nod = resource_state_change->susp_nod[NEW],
1564 		.res_susp_fen = resource_state_change->susp_fen[NEW],
1565 	};
1566 
1567 	notify_resource_state(skb, seq, resource, &resource_info, type);
1568 }
1569 
1570 void notify_connection_state_change(struct sk_buff *skb,
1571 				    unsigned int seq,
1572 				    struct drbd_connection_state_change *connection_state_change,
1573 				    enum drbd_notification_type type)
1574 {
1575 	struct drbd_connection *connection = connection_state_change->connection;
1576 	struct connection_info connection_info = {
1577 		.conn_connection_state = connection_state_change->cstate[NEW],
1578 		.conn_role = connection_state_change->peer_role[NEW],
1579 	};
1580 
1581 	notify_connection_state(skb, seq, connection, &connection_info, type);
1582 }
1583 
1584 void notify_device_state_change(struct sk_buff *skb,
1585 				unsigned int seq,
1586 				struct drbd_device_state_change *device_state_change,
1587 				enum drbd_notification_type type)
1588 {
1589 	struct drbd_device *device = device_state_change->device;
1590 	struct device_info device_info = {
1591 		.dev_disk_state = device_state_change->disk_state[NEW],
1592 	};
1593 
1594 	notify_device_state(skb, seq, device, &device_info, type);
1595 }
1596 
1597 void notify_peer_device_state_change(struct sk_buff *skb,
1598 				     unsigned int seq,
1599 				     struct drbd_peer_device_state_change *p,
1600 				     enum drbd_notification_type type)
1601 {
1602 	struct drbd_peer_device *peer_device = p->peer_device;
1603 	struct peer_device_info peer_device_info = {
1604 		.peer_repl_state = p->repl_state[NEW],
1605 		.peer_disk_state = p->disk_state[NEW],
1606 		.peer_resync_susp_user = p->resync_susp_user[NEW],
1607 		.peer_resync_susp_peer = p->resync_susp_peer[NEW],
1608 		.peer_resync_susp_dependency = p->resync_susp_dependency[NEW],
1609 	};
1610 
1611 	notify_peer_device_state(skb, seq, peer_device, &peer_device_info, type);
1612 }
1613 
1614 static void broadcast_state_change(struct drbd_state_change *state_change)
1615 {
1616 	struct drbd_resource_state_change *resource_state_change = &state_change->resource[0];
1617 	bool resource_state_has_changed;
1618 	unsigned int n_device, n_connection, n_peer_device, n_peer_devices;
1619 	void (*last_func)(struct sk_buff *, unsigned int, void *,
1620 			  enum drbd_notification_type) = NULL;
1621 	void *uninitialized_var(last_arg);
1622 
1623 #define HAS_CHANGED(state) ((state)[OLD] != (state)[NEW])
1624 #define FINAL_STATE_CHANGE(type) \
1625 	({ if (last_func) \
1626 		last_func(NULL, 0, last_arg, type); \
1627 	})
1628 #define REMEMBER_STATE_CHANGE(func, arg, type) \
1629 	({ FINAL_STATE_CHANGE(type | NOTIFY_CONTINUES); \
1630 	   last_func = (typeof(last_func))func; \
1631 	   last_arg = arg; \
1632 	 })
1633 
1634 	mutex_lock(&notification_mutex);
1635 
1636 	resource_state_has_changed =
1637 	    HAS_CHANGED(resource_state_change->role) ||
1638 	    HAS_CHANGED(resource_state_change->susp) ||
1639 	    HAS_CHANGED(resource_state_change->susp_nod) ||
1640 	    HAS_CHANGED(resource_state_change->susp_fen);
1641 
1642 	if (resource_state_has_changed)
1643 		REMEMBER_STATE_CHANGE(notify_resource_state_change,
1644 				      resource_state_change, NOTIFY_CHANGE);
1645 
1646 	for (n_connection = 0; n_connection < state_change->n_connections; n_connection++) {
1647 		struct drbd_connection_state_change *connection_state_change =
1648 				&state_change->connections[n_connection];
1649 
1650 		if (HAS_CHANGED(connection_state_change->peer_role) ||
1651 		    HAS_CHANGED(connection_state_change->cstate))
1652 			REMEMBER_STATE_CHANGE(notify_connection_state_change,
1653 					      connection_state_change, NOTIFY_CHANGE);
1654 	}
1655 
1656 	for (n_device = 0; n_device < state_change->n_devices; n_device++) {
1657 		struct drbd_device_state_change *device_state_change =
1658 			&state_change->devices[n_device];
1659 
1660 		if (HAS_CHANGED(device_state_change->disk_state))
1661 			REMEMBER_STATE_CHANGE(notify_device_state_change,
1662 					      device_state_change, NOTIFY_CHANGE);
1663 	}
1664 
1665 	n_peer_devices = state_change->n_devices * state_change->n_connections;
1666 	for (n_peer_device = 0; n_peer_device < n_peer_devices; n_peer_device++) {
1667 		struct drbd_peer_device_state_change *p =
1668 			&state_change->peer_devices[n_peer_device];
1669 
1670 		if (HAS_CHANGED(p->disk_state) ||
1671 		    HAS_CHANGED(p->repl_state) ||
1672 		    HAS_CHANGED(p->resync_susp_user) ||
1673 		    HAS_CHANGED(p->resync_susp_peer) ||
1674 		    HAS_CHANGED(p->resync_susp_dependency))
1675 			REMEMBER_STATE_CHANGE(notify_peer_device_state_change,
1676 					      p, NOTIFY_CHANGE);
1677 	}
1678 
1679 	FINAL_STATE_CHANGE(NOTIFY_CHANGE);
1680 	mutex_unlock(&notification_mutex);
1681 
1682 #undef HAS_CHANGED
1683 #undef FINAL_STATE_CHANGE
1684 #undef REMEMBER_STATE_CHANGE
1685 }
1686 
1687 /* takes old and new peer disk state */
1688 static bool lost_contact_to_peer_data(enum drbd_disk_state os, enum drbd_disk_state ns)
1689 {
1690 	if ((os >= D_INCONSISTENT && os != D_UNKNOWN && os != D_OUTDATED)
1691 	&&  (ns < D_INCONSISTENT || ns == D_UNKNOWN || ns == D_OUTDATED))
1692 		return true;
1693 
1694 	/* Scenario, starting with normal operation
1695 	 * Connected Primary/Secondary UpToDate/UpToDate
1696 	 * NetworkFailure Primary/Unknown UpToDate/DUnknown (frozen)
1697 	 * ...
1698 	 * Connected Primary/Secondary UpToDate/Diskless (resumed; needs to bump uuid!)
1699 	 */
1700 	if (os == D_UNKNOWN
1701 	&&  (ns == D_DISKLESS || ns == D_FAILED || ns == D_OUTDATED))
1702 		return true;
1703 
1704 	return false;
1705 }
1706 
1707 /**
1708  * after_state_ch() - Perform after state change actions that may sleep
1709  * @device:	DRBD device.
1710  * @os:		old state.
1711  * @ns:		new state.
1712  * @flags:	Flags
1713  */
1714 static void after_state_ch(struct drbd_device *device, union drbd_state os,
1715 			   union drbd_state ns, enum chg_state_flags flags,
1716 			   struct drbd_state_change *state_change)
1717 {
1718 	struct drbd_resource *resource = device->resource;
1719 	struct drbd_peer_device *peer_device = first_peer_device(device);
1720 	struct drbd_connection *connection = peer_device ? peer_device->connection : NULL;
1721 	struct sib_info sib;
1722 
1723 	broadcast_state_change(state_change);
1724 
1725 	sib.sib_reason = SIB_STATE_CHANGE;
1726 	sib.os = os;
1727 	sib.ns = ns;
1728 
1729 	if ((os.disk != D_UP_TO_DATE || os.pdsk != D_UP_TO_DATE)
1730 	&&  (ns.disk == D_UP_TO_DATE && ns.pdsk == D_UP_TO_DATE)) {
1731 		clear_bit(CRASHED_PRIMARY, &device->flags);
1732 		if (device->p_uuid)
1733 			device->p_uuid[UI_FLAGS] &= ~((u64)2);
1734 	}
1735 
1736 	/* Inform userspace about the change... */
1737 	drbd_bcast_event(device, &sib);
1738 
1739 	if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
1740 	    (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
1741 		drbd_khelper(device, "pri-on-incon-degr");
1742 
1743 	/* Here we have the actions that are performed after a
1744 	   state change. This function might sleep */
1745 
1746 	if (ns.susp_nod) {
1747 		enum drbd_req_event what = NOTHING;
1748 
1749 		spin_lock_irq(&device->resource->req_lock);
1750 		if (os.conn < C_CONNECTED && conn_lowest_conn(connection) >= C_CONNECTED)
1751 			what = RESEND;
1752 
1753 		if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1754 		    conn_lowest_disk(connection) == D_UP_TO_DATE)
1755 			what = RESTART_FROZEN_DISK_IO;
1756 
1757 		if (resource->susp_nod && what != NOTHING) {
1758 			_tl_restart(connection, what);
1759 			_conn_request_state(connection,
1760 					    (union drbd_state) { { .susp_nod = 1 } },
1761 					    (union drbd_state) { { .susp_nod = 0 } },
1762 					    CS_VERBOSE);
1763 		}
1764 		spin_unlock_irq(&device->resource->req_lock);
1765 	}
1766 
1767 	if (ns.susp_fen) {
1768 		spin_lock_irq(&device->resource->req_lock);
1769 		if (resource->susp_fen && conn_lowest_conn(connection) >= C_CONNECTED) {
1770 			/* case2: The connection was established again: */
1771 			struct drbd_peer_device *peer_device;
1772 			int vnr;
1773 
1774 			rcu_read_lock();
1775 			idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
1776 				clear_bit(NEW_CUR_UUID, &peer_device->device->flags);
1777 			rcu_read_unlock();
1778 
1779 			/* We should actively create a new uuid, _before_
1780 			 * we resume/resent, if the peer is diskless
1781 			 * (recovery from a multiple error scenario).
1782 			 * Currently, this happens with a slight delay
1783 			 * below when checking lost_contact_to_peer_data() ...
1784 			 */
1785 			_tl_restart(connection, RESEND);
1786 			_conn_request_state(connection,
1787 					    (union drbd_state) { { .susp_fen = 1 } },
1788 					    (union drbd_state) { { .susp_fen = 0 } },
1789 					    CS_VERBOSE);
1790 		}
1791 		spin_unlock_irq(&device->resource->req_lock);
1792 	}
1793 
1794 	/* Became sync source.  With protocol >= 96, we still need to send out
1795 	 * the sync uuid now. Need to do that before any drbd_send_state, or
1796 	 * the other side may go "paused sync" before receiving the sync uuids,
1797 	 * which is unexpected. */
1798 	if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
1799 	    (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
1800 	    connection->agreed_pro_version >= 96 && get_ldev(device)) {
1801 		drbd_gen_and_send_sync_uuid(peer_device);
1802 		put_ldev(device);
1803 	}
1804 
1805 	/* Do not change the order of the if above and the two below... */
1806 	if (os.pdsk == D_DISKLESS &&
1807 	    ns.pdsk > D_DISKLESS && ns.pdsk != D_UNKNOWN) {      /* attach on the peer */
1808 		/* we probably will start a resync soon.
1809 		 * make sure those things are properly reset. */
1810 		device->rs_total = 0;
1811 		device->rs_failed = 0;
1812 		atomic_set(&device->rs_pending_cnt, 0);
1813 		drbd_rs_cancel_all(device);
1814 
1815 		drbd_send_uuids(peer_device);
1816 		drbd_send_state(peer_device, ns);
1817 	}
1818 	/* No point in queuing send_bitmap if we don't have a connection
1819 	 * anymore, so check also the _current_ state, not only the new state
1820 	 * at the time this work was queued. */
1821 	if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
1822 	    device->state.conn == C_WF_BITMAP_S)
1823 		drbd_queue_bitmap_io(device, &drbd_send_bitmap, NULL,
1824 				"send_bitmap (WFBitMapS)",
1825 				BM_LOCKED_TEST_ALLOWED);
1826 
1827 	/* Lost contact to peer's copy of the data */
1828 	if (lost_contact_to_peer_data(os.pdsk, ns.pdsk)) {
1829 		if (get_ldev(device)) {
1830 			if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1831 			    device->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1832 				if (drbd_suspended(device)) {
1833 					set_bit(NEW_CUR_UUID, &device->flags);
1834 				} else {
1835 					drbd_uuid_new_current(device);
1836 					drbd_send_uuids(peer_device);
1837 				}
1838 			}
1839 			put_ldev(device);
1840 		}
1841 	}
1842 
1843 	if (ns.pdsk < D_INCONSISTENT && get_ldev(device)) {
1844 		if (os.peer != R_PRIMARY && ns.peer == R_PRIMARY &&
1845 		    device->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1846 			drbd_uuid_new_current(device);
1847 			drbd_send_uuids(peer_device);
1848 		}
1849 		/* D_DISKLESS Peer becomes secondary */
1850 		if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1851 			/* We may still be Primary ourselves.
1852 			 * No harm done if the bitmap still changes,
1853 			 * redirtied pages will follow later. */
1854 			drbd_bitmap_io_from_worker(device, &drbd_bm_write,
1855 				"demote diskless peer", BM_LOCKED_SET_ALLOWED);
1856 		put_ldev(device);
1857 	}
1858 
1859 	/* Write out all changed bits on demote.
1860 	 * Though, no need to da that just yet
1861 	 * if there is a resync going on still */
1862 	if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
1863 		device->state.conn <= C_CONNECTED && get_ldev(device)) {
1864 		/* No changes to the bitmap expected this time, so assert that,
1865 		 * even though no harm was done if it did change. */
1866 		drbd_bitmap_io_from_worker(device, &drbd_bm_write,
1867 				"demote", BM_LOCKED_TEST_ALLOWED);
1868 		put_ldev(device);
1869 	}
1870 
1871 	/* Last part of the attaching process ... */
1872 	if (ns.conn >= C_CONNECTED &&
1873 	    os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1874 		drbd_send_sizes(peer_device, 0, 0);  /* to start sync... */
1875 		drbd_send_uuids(peer_device);
1876 		drbd_send_state(peer_device, ns);
1877 	}
1878 
1879 	/* We want to pause/continue resync, tell peer. */
1880 	if (ns.conn >= C_CONNECTED &&
1881 	     ((os.aftr_isp != ns.aftr_isp) ||
1882 	      (os.user_isp != ns.user_isp)))
1883 		drbd_send_state(peer_device, ns);
1884 
1885 	/* In case one of the isp bits got set, suspend other devices. */
1886 	if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1887 	    (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1888 		suspend_other_sg(device);
1889 
1890 	/* Make sure the peer gets informed about eventual state
1891 	   changes (ISP bits) while we were in WFReportParams. */
1892 	if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1893 		drbd_send_state(peer_device, ns);
1894 
1895 	if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
1896 		drbd_send_state(peer_device, ns);
1897 
1898 	/* We are in the progress to start a full sync... */
1899 	if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1900 	    (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1901 		/* no other bitmap changes expected during this phase */
1902 		drbd_queue_bitmap_io(device,
1903 			&drbd_bmio_set_n_write, &abw_start_sync,
1904 			"set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED);
1905 
1906 	/* first half of local IO error, failure to attach,
1907 	 * or administrative detach */
1908 	if (os.disk != D_FAILED && ns.disk == D_FAILED) {
1909 		enum drbd_io_error_p eh = EP_PASS_ON;
1910 		int was_io_error = 0;
1911 		/* corresponding get_ldev was in _drbd_set_state, to serialize
1912 		 * our cleanup here with the transition to D_DISKLESS.
1913 		 * But is is still not save to dreference ldev here, since
1914 		 * we might come from an failed Attach before ldev was set. */
1915 		if (device->ldev) {
1916 			rcu_read_lock();
1917 			eh = rcu_dereference(device->ldev->disk_conf)->on_io_error;
1918 			rcu_read_unlock();
1919 
1920 			was_io_error = test_and_clear_bit(WAS_IO_ERROR, &device->flags);
1921 
1922 			/* Intentionally call this handler first, before drbd_send_state().
1923 			 * See: 2932204 drbd: call local-io-error handler early
1924 			 * People may chose to hard-reset the box from this handler.
1925 			 * It is useful if this looks like a "regular node crash". */
1926 			if (was_io_error && eh == EP_CALL_HELPER)
1927 				drbd_khelper(device, "local-io-error");
1928 
1929 			/* Immediately allow completion of all application IO,
1930 			 * that waits for completion from the local disk,
1931 			 * if this was a force-detach due to disk_timeout
1932 			 * or administrator request (drbdsetup detach --force).
1933 			 * Do NOT abort otherwise.
1934 			 * Aborting local requests may cause serious problems,
1935 			 * if requests are completed to upper layers already,
1936 			 * and then later the already submitted local bio completes.
1937 			 * This can cause DMA into former bio pages that meanwhile
1938 			 * have been re-used for other things.
1939 			 * So aborting local requests may cause crashes,
1940 			 * or even worse, silent data corruption.
1941 			 */
1942 			if (test_and_clear_bit(FORCE_DETACH, &device->flags))
1943 				tl_abort_disk_io(device);
1944 
1945 			/* current state still has to be D_FAILED,
1946 			 * there is only one way out: to D_DISKLESS,
1947 			 * and that may only happen after our put_ldev below. */
1948 			if (device->state.disk != D_FAILED)
1949 				drbd_err(device,
1950 					"ASSERT FAILED: disk is %s during detach\n",
1951 					drbd_disk_str(device->state.disk));
1952 
1953 			if (ns.conn >= C_CONNECTED)
1954 				drbd_send_state(peer_device, ns);
1955 
1956 			drbd_rs_cancel_all(device);
1957 
1958 			/* In case we want to get something to stable storage still,
1959 			 * this may be the last chance.
1960 			 * Following put_ldev may transition to D_DISKLESS. */
1961 			drbd_md_sync(device);
1962 		}
1963 		put_ldev(device);
1964 	}
1965 
1966 	/* second half of local IO error, failure to attach,
1967 	 * or administrative detach,
1968 	 * after local_cnt references have reached zero again */
1969 	if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
1970 		/* We must still be diskless,
1971 		 * re-attach has to be serialized with this! */
1972 		if (device->state.disk != D_DISKLESS)
1973 			drbd_err(device,
1974 				 "ASSERT FAILED: disk is %s while going diskless\n",
1975 				 drbd_disk_str(device->state.disk));
1976 
1977 		if (ns.conn >= C_CONNECTED)
1978 			drbd_send_state(peer_device, ns);
1979 		/* corresponding get_ldev in __drbd_set_state
1980 		 * this may finally trigger drbd_ldev_destroy. */
1981 		put_ldev(device);
1982 	}
1983 
1984 	/* Notify peer that I had a local IO error, and did not detached.. */
1985 	if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT && ns.conn >= C_CONNECTED)
1986 		drbd_send_state(peer_device, ns);
1987 
1988 	/* Disks got bigger while they were detached */
1989 	if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1990 	    test_and_clear_bit(RESYNC_AFTER_NEG, &device->flags)) {
1991 		if (ns.conn == C_CONNECTED)
1992 			resync_after_online_grow(device);
1993 	}
1994 
1995 	/* A resync finished or aborted, wake paused devices... */
1996 	if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1997 	    (os.peer_isp && !ns.peer_isp) ||
1998 	    (os.user_isp && !ns.user_isp))
1999 		resume_next_sg(device);
2000 
2001 	/* sync target done with resync.  Explicitly notify peer, even though
2002 	 * it should (at least for non-empty resyncs) already know itself. */
2003 	if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
2004 		drbd_send_state(peer_device, ns);
2005 
2006 	/* Verify finished, or reached stop sector.  Peer did not know about
2007 	 * the stop sector, and we may even have changed the stop sector during
2008 	 * verify to interrupt/stop early.  Send the new state. */
2009 	if (os.conn == C_VERIFY_S && ns.conn == C_CONNECTED
2010 	&& verify_can_do_stop_sector(device))
2011 		drbd_send_state(peer_device, ns);
2012 
2013 	/* This triggers bitmap writeout of potentially still unwritten pages
2014 	 * if the resync finished cleanly, or aborted because of peer disk
2015 	 * failure, or on transition from resync back to AHEAD/BEHIND.
2016 	 *
2017 	 * Connection loss is handled in drbd_disconnected() by the receiver.
2018 	 *
2019 	 * For resync aborted because of local disk failure, we cannot do
2020 	 * any bitmap writeout anymore.
2021 	 *
2022 	 * No harm done if some bits change during this phase.
2023 	 */
2024 	if ((os.conn > C_CONNECTED && os.conn < C_AHEAD) &&
2025 	    (ns.conn == C_CONNECTED || ns.conn >= C_AHEAD) && get_ldev(device)) {
2026 		drbd_queue_bitmap_io(device, &drbd_bm_write_copy_pages, NULL,
2027 			"write from resync_finished", BM_LOCKED_CHANGE_ALLOWED);
2028 		put_ldev(device);
2029 	}
2030 
2031 	if (ns.disk == D_DISKLESS &&
2032 	    ns.conn == C_STANDALONE &&
2033 	    ns.role == R_SECONDARY) {
2034 		if (os.aftr_isp != ns.aftr_isp)
2035 			resume_next_sg(device);
2036 	}
2037 
2038 	drbd_md_sync(device);
2039 }
2040 
2041 struct after_conn_state_chg_work {
2042 	struct drbd_work w;
2043 	enum drbd_conns oc;
2044 	union drbd_state ns_min;
2045 	union drbd_state ns_max; /* new, max state, over all devices */
2046 	enum chg_state_flags flags;
2047 	struct drbd_connection *connection;
2048 	struct drbd_state_change *state_change;
2049 };
2050 
2051 static int w_after_conn_state_ch(struct drbd_work *w, int unused)
2052 {
2053 	struct after_conn_state_chg_work *acscw =
2054 		container_of(w, struct after_conn_state_chg_work, w);
2055 	struct drbd_connection *connection = acscw->connection;
2056 	enum drbd_conns oc = acscw->oc;
2057 	union drbd_state ns_max = acscw->ns_max;
2058 	struct drbd_peer_device *peer_device;
2059 	int vnr;
2060 
2061 	broadcast_state_change(acscw->state_change);
2062 	forget_state_change(acscw->state_change);
2063 	kfree(acscw);
2064 
2065 	/* Upon network configuration, we need to start the receiver */
2066 	if (oc == C_STANDALONE && ns_max.conn == C_UNCONNECTED)
2067 		drbd_thread_start(&connection->receiver);
2068 
2069 	if (oc == C_DISCONNECTING && ns_max.conn == C_STANDALONE) {
2070 		struct net_conf *old_conf;
2071 
2072 		mutex_lock(&notification_mutex);
2073 		idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2074 			notify_peer_device_state(NULL, 0, peer_device, NULL,
2075 						 NOTIFY_DESTROY | NOTIFY_CONTINUES);
2076 		notify_connection_state(NULL, 0, connection, NULL, NOTIFY_DESTROY);
2077 		mutex_unlock(&notification_mutex);
2078 
2079 		mutex_lock(&connection->resource->conf_update);
2080 		old_conf = connection->net_conf;
2081 		connection->my_addr_len = 0;
2082 		connection->peer_addr_len = 0;
2083 		RCU_INIT_POINTER(connection->net_conf, NULL);
2084 		conn_free_crypto(connection);
2085 		mutex_unlock(&connection->resource->conf_update);
2086 
2087 		synchronize_rcu();
2088 		kfree(old_conf);
2089 	}
2090 
2091 	if (ns_max.susp_fen) {
2092 		/* case1: The outdate peer handler is successful: */
2093 		if (ns_max.pdsk <= D_OUTDATED) {
2094 			rcu_read_lock();
2095 			idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2096 				struct drbd_device *device = peer_device->device;
2097 				if (test_bit(NEW_CUR_UUID, &device->flags)) {
2098 					drbd_uuid_new_current(device);
2099 					clear_bit(NEW_CUR_UUID, &device->flags);
2100 				}
2101 			}
2102 			rcu_read_unlock();
2103 			spin_lock_irq(&connection->resource->req_lock);
2104 			_tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
2105 			_conn_request_state(connection,
2106 					    (union drbd_state) { { .susp_fen = 1 } },
2107 					    (union drbd_state) { { .susp_fen = 0 } },
2108 					    CS_VERBOSE);
2109 			spin_unlock_irq(&connection->resource->req_lock);
2110 		}
2111 	}
2112 	kref_put(&connection->kref, drbd_destroy_connection);
2113 
2114 	conn_md_sync(connection);
2115 
2116 	return 0;
2117 }
2118 
2119 static void conn_old_common_state(struct drbd_connection *connection, union drbd_state *pcs, enum chg_state_flags *pf)
2120 {
2121 	enum chg_state_flags flags = ~0;
2122 	struct drbd_peer_device *peer_device;
2123 	int vnr, first_vol = 1;
2124 	union drbd_dev_state os, cs = {
2125 		{ .role = R_SECONDARY,
2126 		  .peer = R_UNKNOWN,
2127 		  .conn = connection->cstate,
2128 		  .disk = D_DISKLESS,
2129 		  .pdsk = D_UNKNOWN,
2130 		} };
2131 
2132 	rcu_read_lock();
2133 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2134 		struct drbd_device *device = peer_device->device;
2135 		os = device->state;
2136 
2137 		if (first_vol) {
2138 			cs = os;
2139 			first_vol = 0;
2140 			continue;
2141 		}
2142 
2143 		if (cs.role != os.role)
2144 			flags &= ~CS_DC_ROLE;
2145 
2146 		if (cs.peer != os.peer)
2147 			flags &= ~CS_DC_PEER;
2148 
2149 		if (cs.conn != os.conn)
2150 			flags &= ~CS_DC_CONN;
2151 
2152 		if (cs.disk != os.disk)
2153 			flags &= ~CS_DC_DISK;
2154 
2155 		if (cs.pdsk != os.pdsk)
2156 			flags &= ~CS_DC_PDSK;
2157 	}
2158 	rcu_read_unlock();
2159 
2160 	*pf |= CS_DC_MASK;
2161 	*pf &= flags;
2162 	(*pcs).i = cs.i;
2163 }
2164 
2165 static enum drbd_state_rv
2166 conn_is_valid_transition(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2167 			 enum chg_state_flags flags)
2168 {
2169 	enum drbd_state_rv rv = SS_SUCCESS;
2170 	union drbd_state ns, os;
2171 	struct drbd_peer_device *peer_device;
2172 	int vnr;
2173 
2174 	rcu_read_lock();
2175 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2176 		struct drbd_device *device = peer_device->device;
2177 		os = drbd_read_state(device);
2178 		ns = sanitize_state(device, os, apply_mask_val(os, mask, val), NULL);
2179 
2180 		if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
2181 			ns.disk = os.disk;
2182 
2183 		if (ns.i == os.i)
2184 			continue;
2185 
2186 		rv = is_valid_transition(os, ns);
2187 
2188 		if (rv >= SS_SUCCESS && !(flags & CS_HARD)) {
2189 			rv = is_valid_state(device, ns);
2190 			if (rv < SS_SUCCESS) {
2191 				if (is_valid_state(device, os) == rv)
2192 					rv = is_valid_soft_transition(os, ns, connection);
2193 			} else
2194 				rv = is_valid_soft_transition(os, ns, connection);
2195 		}
2196 
2197 		if (rv < SS_SUCCESS) {
2198 			if (flags & CS_VERBOSE)
2199 				print_st_err(device, os, ns, rv);
2200 			break;
2201 		}
2202 	}
2203 	rcu_read_unlock();
2204 
2205 	return rv;
2206 }
2207 
2208 static void
2209 conn_set_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2210 	       union drbd_state *pns_min, union drbd_state *pns_max, enum chg_state_flags flags)
2211 {
2212 	union drbd_state ns, os, ns_max = { };
2213 	union drbd_state ns_min = {
2214 		{ .role = R_MASK,
2215 		  .peer = R_MASK,
2216 		  .conn = val.conn,
2217 		  .disk = D_MASK,
2218 		  .pdsk = D_MASK
2219 		} };
2220 	struct drbd_peer_device *peer_device;
2221 	enum drbd_state_rv rv;
2222 	int vnr, number_of_volumes = 0;
2223 
2224 	if (mask.conn == C_MASK) {
2225 		/* remember last connect time so request_timer_fn() won't
2226 		 * kill newly established sessions while we are still trying to thaw
2227 		 * previously frozen IO */
2228 		if (connection->cstate != C_WF_REPORT_PARAMS && val.conn == C_WF_REPORT_PARAMS)
2229 			connection->last_reconnect_jif = jiffies;
2230 
2231 		connection->cstate = val.conn;
2232 	}
2233 
2234 	rcu_read_lock();
2235 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2236 		struct drbd_device *device = peer_device->device;
2237 		number_of_volumes++;
2238 		os = drbd_read_state(device);
2239 		ns = apply_mask_val(os, mask, val);
2240 		ns = sanitize_state(device, os, ns, NULL);
2241 
2242 		if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
2243 			ns.disk = os.disk;
2244 
2245 		rv = _drbd_set_state(device, ns, flags, NULL);
2246 		BUG_ON(rv < SS_SUCCESS);
2247 		ns.i = device->state.i;
2248 		ns_max.role = max_role(ns.role, ns_max.role);
2249 		ns_max.peer = max_role(ns.peer, ns_max.peer);
2250 		ns_max.conn = max_t(enum drbd_conns, ns.conn, ns_max.conn);
2251 		ns_max.disk = max_t(enum drbd_disk_state, ns.disk, ns_max.disk);
2252 		ns_max.pdsk = max_t(enum drbd_disk_state, ns.pdsk, ns_max.pdsk);
2253 
2254 		ns_min.role = min_role(ns.role, ns_min.role);
2255 		ns_min.peer = min_role(ns.peer, ns_min.peer);
2256 		ns_min.conn = min_t(enum drbd_conns, ns.conn, ns_min.conn);
2257 		ns_min.disk = min_t(enum drbd_disk_state, ns.disk, ns_min.disk);
2258 		ns_min.pdsk = min_t(enum drbd_disk_state, ns.pdsk, ns_min.pdsk);
2259 	}
2260 	rcu_read_unlock();
2261 
2262 	if (number_of_volumes == 0) {
2263 		ns_min = ns_max = (union drbd_state) { {
2264 				.role = R_SECONDARY,
2265 				.peer = R_UNKNOWN,
2266 				.conn = val.conn,
2267 				.disk = D_DISKLESS,
2268 				.pdsk = D_UNKNOWN
2269 			} };
2270 	}
2271 
2272 	ns_min.susp = ns_max.susp = connection->resource->susp;
2273 	ns_min.susp_nod = ns_max.susp_nod = connection->resource->susp_nod;
2274 	ns_min.susp_fen = ns_max.susp_fen = connection->resource->susp_fen;
2275 
2276 	*pns_min = ns_min;
2277 	*pns_max = ns_max;
2278 }
2279 
2280 static enum drbd_state_rv
2281 _conn_rq_cond(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
2282 {
2283 	enum drbd_state_rv err, rv = SS_UNKNOWN_ERROR; /* continue waiting */;
2284 
2285 	if (test_and_clear_bit(CONN_WD_ST_CHG_OKAY, &connection->flags))
2286 		rv = SS_CW_SUCCESS;
2287 
2288 	if (test_and_clear_bit(CONN_WD_ST_CHG_FAIL, &connection->flags))
2289 		rv = SS_CW_FAILED_BY_PEER;
2290 
2291 	err = conn_is_valid_transition(connection, mask, val, 0);
2292 	if (err == SS_SUCCESS && connection->cstate == C_WF_REPORT_PARAMS)
2293 		return rv;
2294 
2295 	return err;
2296 }
2297 
2298 enum drbd_state_rv
2299 _conn_request_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2300 		    enum chg_state_flags flags)
2301 {
2302 	enum drbd_state_rv rv = SS_SUCCESS;
2303 	struct after_conn_state_chg_work *acscw;
2304 	enum drbd_conns oc = connection->cstate;
2305 	union drbd_state ns_max, ns_min, os;
2306 	bool have_mutex = false;
2307 	struct drbd_state_change *state_change;
2308 
2309 	if (mask.conn) {
2310 		rv = is_valid_conn_transition(oc, val.conn);
2311 		if (rv < SS_SUCCESS)
2312 			goto abort;
2313 	}
2314 
2315 	rv = conn_is_valid_transition(connection, mask, val, flags);
2316 	if (rv < SS_SUCCESS)
2317 		goto abort;
2318 
2319 	if (oc == C_WF_REPORT_PARAMS && val.conn == C_DISCONNECTING &&
2320 	    !(flags & (CS_LOCAL_ONLY | CS_HARD))) {
2321 
2322 		/* This will be a cluster-wide state change.
2323 		 * Need to give up the spinlock, grab the mutex,
2324 		 * then send the state change request, ... */
2325 		spin_unlock_irq(&connection->resource->req_lock);
2326 		mutex_lock(&connection->cstate_mutex);
2327 		have_mutex = true;
2328 
2329 		set_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
2330 		if (conn_send_state_req(connection, mask, val)) {
2331 			/* sending failed. */
2332 			clear_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
2333 			rv = SS_CW_FAILED_BY_PEER;
2334 			/* need to re-aquire the spin lock, though */
2335 			goto abort_unlocked;
2336 		}
2337 
2338 		if (val.conn == C_DISCONNECTING)
2339 			set_bit(DISCONNECT_SENT, &connection->flags);
2340 
2341 		/* ... and re-aquire the spinlock.
2342 		 * If _conn_rq_cond() returned >= SS_SUCCESS, we must call
2343 		 * conn_set_state() within the same spinlock. */
2344 		spin_lock_irq(&connection->resource->req_lock);
2345 		wait_event_lock_irq(connection->ping_wait,
2346 				(rv = _conn_rq_cond(connection, mask, val)),
2347 				connection->resource->req_lock);
2348 		clear_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
2349 		if (rv < SS_SUCCESS)
2350 			goto abort;
2351 	}
2352 
2353 	state_change = remember_old_state(connection->resource, GFP_ATOMIC);
2354 	conn_old_common_state(connection, &os, &flags);
2355 	flags |= CS_DC_SUSP;
2356 	conn_set_state(connection, mask, val, &ns_min, &ns_max, flags);
2357 	conn_pr_state_change(connection, os, ns_max, flags);
2358 	remember_new_state(state_change);
2359 
2360 	acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC);
2361 	if (acscw) {
2362 		acscw->oc = os.conn;
2363 		acscw->ns_min = ns_min;
2364 		acscw->ns_max = ns_max;
2365 		acscw->flags = flags;
2366 		acscw->w.cb = w_after_conn_state_ch;
2367 		kref_get(&connection->kref);
2368 		acscw->connection = connection;
2369 		acscw->state_change = state_change;
2370 		drbd_queue_work(&connection->sender_work, &acscw->w);
2371 	} else {
2372 		drbd_err(connection, "Could not kmalloc an acscw\n");
2373 	}
2374 
2375  abort:
2376 	if (have_mutex) {
2377 		/* mutex_unlock() "... must not be used in interrupt context.",
2378 		 * so give up the spinlock, then re-aquire it */
2379 		spin_unlock_irq(&connection->resource->req_lock);
2380  abort_unlocked:
2381 		mutex_unlock(&connection->cstate_mutex);
2382 		spin_lock_irq(&connection->resource->req_lock);
2383 	}
2384 	if (rv < SS_SUCCESS && flags & CS_VERBOSE) {
2385 		drbd_err(connection, "State change failed: %s\n", drbd_set_st_err_str(rv));
2386 		drbd_err(connection, " mask = 0x%x val = 0x%x\n", mask.i, val.i);
2387 		drbd_err(connection, " old_conn:%s wanted_conn:%s\n", drbd_conn_str(oc), drbd_conn_str(val.conn));
2388 	}
2389 	return rv;
2390 }
2391 
2392 enum drbd_state_rv
2393 conn_request_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2394 		   enum chg_state_flags flags)
2395 {
2396 	enum drbd_state_rv rv;
2397 
2398 	spin_lock_irq(&connection->resource->req_lock);
2399 	rv = _conn_request_state(connection, mask, val, flags);
2400 	spin_unlock_irq(&connection->resource->req_lock);
2401 
2402 	return rv;
2403 }
2404