1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3    drbd_state.c
4 
5    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6 
7    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
8    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
9    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
10 
11    Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
12    from Logicworks, Inc. for making SDP replication support possible.
13 
14  */
15 
16 #include <linux/drbd_limits.h>
17 #include "drbd_int.h"
18 #include "drbd_protocol.h"
19 #include "drbd_req.h"
20 #include "drbd_state_change.h"
21 
22 struct after_state_chg_work {
23 	struct drbd_work w;
24 	struct drbd_device *device;
25 	union drbd_state os;
26 	union drbd_state ns;
27 	enum chg_state_flags flags;
28 	struct completion *done;
29 	struct drbd_state_change *state_change;
30 };
31 
32 enum sanitize_state_warnings {
33 	NO_WARNING,
34 	ABORTED_ONLINE_VERIFY,
35 	ABORTED_RESYNC,
36 	CONNECTION_LOST_NEGOTIATING,
37 	IMPLICITLY_UPGRADED_DISK,
38 	IMPLICITLY_UPGRADED_PDSK,
39 };
40 
count_objects(struct drbd_resource * resource,unsigned int * n_devices,unsigned int * n_connections)41 static void count_objects(struct drbd_resource *resource,
42 			  unsigned int *n_devices,
43 			  unsigned int *n_connections)
44 {
45 	struct drbd_device *device;
46 	struct drbd_connection *connection;
47 	int vnr;
48 
49 	*n_devices = 0;
50 	*n_connections = 0;
51 
52 	idr_for_each_entry(&resource->devices, device, vnr)
53 		(*n_devices)++;
54 	for_each_connection(connection, resource)
55 		(*n_connections)++;
56 }
57 
alloc_state_change(unsigned int n_devices,unsigned int n_connections,gfp_t gfp)58 static struct drbd_state_change *alloc_state_change(unsigned int n_devices, unsigned int n_connections, gfp_t gfp)
59 {
60 	struct drbd_state_change *state_change;
61 	unsigned int size, n;
62 
63 	size = sizeof(struct drbd_state_change) +
64 	       n_devices * sizeof(struct drbd_device_state_change) +
65 	       n_connections * sizeof(struct drbd_connection_state_change) +
66 	       n_devices * n_connections * sizeof(struct drbd_peer_device_state_change);
67 	state_change = kmalloc(size, gfp);
68 	if (!state_change)
69 		return NULL;
70 	state_change->n_devices = n_devices;
71 	state_change->n_connections = n_connections;
72 	state_change->devices = (void *)(state_change + 1);
73 	state_change->connections = (void *)&state_change->devices[n_devices];
74 	state_change->peer_devices = (void *)&state_change->connections[n_connections];
75 	state_change->resource->resource = NULL;
76 	for (n = 0; n < n_devices; n++)
77 		state_change->devices[n].device = NULL;
78 	for (n = 0; n < n_connections; n++)
79 		state_change->connections[n].connection = NULL;
80 	return state_change;
81 }
82 
remember_old_state(struct drbd_resource * resource,gfp_t gfp)83 struct drbd_state_change *remember_old_state(struct drbd_resource *resource, gfp_t gfp)
84 {
85 	struct drbd_state_change *state_change;
86 	struct drbd_device *device;
87 	unsigned int n_devices;
88 	struct drbd_connection *connection;
89 	unsigned int n_connections;
90 	int vnr;
91 
92 	struct drbd_device_state_change *device_state_change;
93 	struct drbd_peer_device_state_change *peer_device_state_change;
94 	struct drbd_connection_state_change *connection_state_change;
95 
96 	/* Caller holds req_lock spinlock.
97 	 * No state, no device IDR, no connections lists can change. */
98 	count_objects(resource, &n_devices, &n_connections);
99 	state_change = alloc_state_change(n_devices, n_connections, gfp);
100 	if (!state_change)
101 		return NULL;
102 
103 	kref_get(&resource->kref);
104 	state_change->resource->resource = resource;
105 	state_change->resource->role[OLD] =
106 		conn_highest_role(first_connection(resource));
107 	state_change->resource->susp[OLD] = resource->susp;
108 	state_change->resource->susp_nod[OLD] = resource->susp_nod;
109 	state_change->resource->susp_fen[OLD] = resource->susp_fen;
110 
111 	connection_state_change = state_change->connections;
112 	for_each_connection(connection, resource) {
113 		kref_get(&connection->kref);
114 		connection_state_change->connection = connection;
115 		connection_state_change->cstate[OLD] =
116 			connection->cstate;
117 		connection_state_change->peer_role[OLD] =
118 			conn_highest_peer(connection);
119 		connection_state_change++;
120 	}
121 
122 	device_state_change = state_change->devices;
123 	peer_device_state_change = state_change->peer_devices;
124 	idr_for_each_entry(&resource->devices, device, vnr) {
125 		kref_get(&device->kref);
126 		device_state_change->device = device;
127 		device_state_change->disk_state[OLD] = device->state.disk;
128 
129 		/* The peer_devices for each device have to be enumerated in
130 		   the order of the connections. We may not use for_each_peer_device() here. */
131 		for_each_connection(connection, resource) {
132 			struct drbd_peer_device *peer_device;
133 
134 			peer_device = conn_peer_device(connection, device->vnr);
135 			peer_device_state_change->peer_device = peer_device;
136 			peer_device_state_change->disk_state[OLD] =
137 				device->state.pdsk;
138 			peer_device_state_change->repl_state[OLD] =
139 				max_t(enum drbd_conns,
140 				      C_WF_REPORT_PARAMS, device->state.conn);
141 			peer_device_state_change->resync_susp_user[OLD] =
142 				device->state.user_isp;
143 			peer_device_state_change->resync_susp_peer[OLD] =
144 				device->state.peer_isp;
145 			peer_device_state_change->resync_susp_dependency[OLD] =
146 				device->state.aftr_isp;
147 			peer_device_state_change++;
148 		}
149 		device_state_change++;
150 	}
151 
152 	return state_change;
153 }
154 
remember_new_state(struct drbd_state_change * state_change)155 static void remember_new_state(struct drbd_state_change *state_change)
156 {
157 	struct drbd_resource_state_change *resource_state_change;
158 	struct drbd_resource *resource;
159 	unsigned int n;
160 
161 	if (!state_change)
162 		return;
163 
164 	resource_state_change = &state_change->resource[0];
165 	resource = resource_state_change->resource;
166 
167 	resource_state_change->role[NEW] =
168 		conn_highest_role(first_connection(resource));
169 	resource_state_change->susp[NEW] = resource->susp;
170 	resource_state_change->susp_nod[NEW] = resource->susp_nod;
171 	resource_state_change->susp_fen[NEW] = resource->susp_fen;
172 
173 	for (n = 0; n < state_change->n_devices; n++) {
174 		struct drbd_device_state_change *device_state_change =
175 			&state_change->devices[n];
176 		struct drbd_device *device = device_state_change->device;
177 
178 		device_state_change->disk_state[NEW] = device->state.disk;
179 	}
180 
181 	for (n = 0; n < state_change->n_connections; n++) {
182 		struct drbd_connection_state_change *connection_state_change =
183 			&state_change->connections[n];
184 		struct drbd_connection *connection =
185 			connection_state_change->connection;
186 
187 		connection_state_change->cstate[NEW] = connection->cstate;
188 		connection_state_change->peer_role[NEW] =
189 			conn_highest_peer(connection);
190 	}
191 
192 	for (n = 0; n < state_change->n_devices * state_change->n_connections; n++) {
193 		struct drbd_peer_device_state_change *peer_device_state_change =
194 			&state_change->peer_devices[n];
195 		struct drbd_device *device =
196 			peer_device_state_change->peer_device->device;
197 		union drbd_dev_state state = device->state;
198 
199 		peer_device_state_change->disk_state[NEW] = state.pdsk;
200 		peer_device_state_change->repl_state[NEW] =
201 			max_t(enum drbd_conns, C_WF_REPORT_PARAMS, state.conn);
202 		peer_device_state_change->resync_susp_user[NEW] =
203 			state.user_isp;
204 		peer_device_state_change->resync_susp_peer[NEW] =
205 			state.peer_isp;
206 		peer_device_state_change->resync_susp_dependency[NEW] =
207 			state.aftr_isp;
208 	}
209 }
210 
copy_old_to_new_state_change(struct drbd_state_change * state_change)211 void copy_old_to_new_state_change(struct drbd_state_change *state_change)
212 {
213 	struct drbd_resource_state_change *resource_state_change = &state_change->resource[0];
214 	unsigned int n_device, n_connection, n_peer_device, n_peer_devices;
215 
216 #define OLD_TO_NEW(x) \
217 	(x[NEW] = x[OLD])
218 
219 	OLD_TO_NEW(resource_state_change->role);
220 	OLD_TO_NEW(resource_state_change->susp);
221 	OLD_TO_NEW(resource_state_change->susp_nod);
222 	OLD_TO_NEW(resource_state_change->susp_fen);
223 
224 	for (n_connection = 0; n_connection < state_change->n_connections; n_connection++) {
225 		struct drbd_connection_state_change *connection_state_change =
226 				&state_change->connections[n_connection];
227 
228 		OLD_TO_NEW(connection_state_change->peer_role);
229 		OLD_TO_NEW(connection_state_change->cstate);
230 	}
231 
232 	for (n_device = 0; n_device < state_change->n_devices; n_device++) {
233 		struct drbd_device_state_change *device_state_change =
234 			&state_change->devices[n_device];
235 
236 		OLD_TO_NEW(device_state_change->disk_state);
237 	}
238 
239 	n_peer_devices = state_change->n_devices * state_change->n_connections;
240 	for (n_peer_device = 0; n_peer_device < n_peer_devices; n_peer_device++) {
241 		struct drbd_peer_device_state_change *p =
242 			&state_change->peer_devices[n_peer_device];
243 
244 		OLD_TO_NEW(p->disk_state);
245 		OLD_TO_NEW(p->repl_state);
246 		OLD_TO_NEW(p->resync_susp_user);
247 		OLD_TO_NEW(p->resync_susp_peer);
248 		OLD_TO_NEW(p->resync_susp_dependency);
249 	}
250 
251 #undef OLD_TO_NEW
252 }
253 
forget_state_change(struct drbd_state_change * state_change)254 void forget_state_change(struct drbd_state_change *state_change)
255 {
256 	unsigned int n;
257 
258 	if (!state_change)
259 		return;
260 
261 	if (state_change->resource->resource)
262 		kref_put(&state_change->resource->resource->kref, drbd_destroy_resource);
263 	for (n = 0; n < state_change->n_devices; n++) {
264 		struct drbd_device *device = state_change->devices[n].device;
265 
266 		if (device)
267 			kref_put(&device->kref, drbd_destroy_device);
268 	}
269 	for (n = 0; n < state_change->n_connections; n++) {
270 		struct drbd_connection *connection =
271 			state_change->connections[n].connection;
272 
273 		if (connection)
274 			kref_put(&connection->kref, drbd_destroy_connection);
275 	}
276 	kfree(state_change);
277 }
278 
279 static int w_after_state_ch(struct drbd_work *w, int unused);
280 static void after_state_ch(struct drbd_device *device, union drbd_state os,
281 			   union drbd_state ns, enum chg_state_flags flags,
282 			   struct drbd_state_change *);
283 static enum drbd_state_rv is_valid_state(struct drbd_device *, union drbd_state);
284 static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state, struct drbd_connection *);
285 static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns);
286 static union drbd_state sanitize_state(struct drbd_device *device, union drbd_state os,
287 				       union drbd_state ns, enum sanitize_state_warnings *warn);
288 
is_susp(union drbd_state s)289 static inline bool is_susp(union drbd_state s)
290 {
291         return s.susp || s.susp_nod || s.susp_fen;
292 }
293 
conn_all_vols_unconf(struct drbd_connection * connection)294 bool conn_all_vols_unconf(struct drbd_connection *connection)
295 {
296 	struct drbd_peer_device *peer_device;
297 	bool rv = true;
298 	int vnr;
299 
300 	rcu_read_lock();
301 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
302 		struct drbd_device *device = peer_device->device;
303 		if (device->state.disk != D_DISKLESS ||
304 		    device->state.conn != C_STANDALONE ||
305 		    device->state.role != R_SECONDARY) {
306 			rv = false;
307 			break;
308 		}
309 	}
310 	rcu_read_unlock();
311 
312 	return rv;
313 }
314 
315 /* Unfortunately the states where not correctly ordered, when
316    they where defined. therefore can not use max_t() here. */
max_role(enum drbd_role role1,enum drbd_role role2)317 static enum drbd_role max_role(enum drbd_role role1, enum drbd_role role2)
318 {
319 	if (role1 == R_PRIMARY || role2 == R_PRIMARY)
320 		return R_PRIMARY;
321 	if (role1 == R_SECONDARY || role2 == R_SECONDARY)
322 		return R_SECONDARY;
323 	return R_UNKNOWN;
324 }
325 
min_role(enum drbd_role role1,enum drbd_role role2)326 static enum drbd_role min_role(enum drbd_role role1, enum drbd_role role2)
327 {
328 	if (role1 == R_UNKNOWN || role2 == R_UNKNOWN)
329 		return R_UNKNOWN;
330 	if (role1 == R_SECONDARY || role2 == R_SECONDARY)
331 		return R_SECONDARY;
332 	return R_PRIMARY;
333 }
334 
conn_highest_role(struct drbd_connection * connection)335 enum drbd_role conn_highest_role(struct drbd_connection *connection)
336 {
337 	enum drbd_role role = R_SECONDARY;
338 	struct drbd_peer_device *peer_device;
339 	int vnr;
340 
341 	rcu_read_lock();
342 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
343 		struct drbd_device *device = peer_device->device;
344 		role = max_role(role, device->state.role);
345 	}
346 	rcu_read_unlock();
347 
348 	return role;
349 }
350 
conn_highest_peer(struct drbd_connection * connection)351 enum drbd_role conn_highest_peer(struct drbd_connection *connection)
352 {
353 	enum drbd_role peer = R_UNKNOWN;
354 	struct drbd_peer_device *peer_device;
355 	int vnr;
356 
357 	rcu_read_lock();
358 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
359 		struct drbd_device *device = peer_device->device;
360 		peer = max_role(peer, device->state.peer);
361 	}
362 	rcu_read_unlock();
363 
364 	return peer;
365 }
366 
conn_highest_disk(struct drbd_connection * connection)367 enum drbd_disk_state conn_highest_disk(struct drbd_connection *connection)
368 {
369 	enum drbd_disk_state disk_state = D_DISKLESS;
370 	struct drbd_peer_device *peer_device;
371 	int vnr;
372 
373 	rcu_read_lock();
374 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
375 		struct drbd_device *device = peer_device->device;
376 		disk_state = max_t(enum drbd_disk_state, disk_state, device->state.disk);
377 	}
378 	rcu_read_unlock();
379 
380 	return disk_state;
381 }
382 
conn_lowest_disk(struct drbd_connection * connection)383 enum drbd_disk_state conn_lowest_disk(struct drbd_connection *connection)
384 {
385 	enum drbd_disk_state disk_state = D_MASK;
386 	struct drbd_peer_device *peer_device;
387 	int vnr;
388 
389 	rcu_read_lock();
390 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
391 		struct drbd_device *device = peer_device->device;
392 		disk_state = min_t(enum drbd_disk_state, disk_state, device->state.disk);
393 	}
394 	rcu_read_unlock();
395 
396 	return disk_state;
397 }
398 
conn_highest_pdsk(struct drbd_connection * connection)399 enum drbd_disk_state conn_highest_pdsk(struct drbd_connection *connection)
400 {
401 	enum drbd_disk_state disk_state = D_DISKLESS;
402 	struct drbd_peer_device *peer_device;
403 	int vnr;
404 
405 	rcu_read_lock();
406 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
407 		struct drbd_device *device = peer_device->device;
408 		disk_state = max_t(enum drbd_disk_state, disk_state, device->state.pdsk);
409 	}
410 	rcu_read_unlock();
411 
412 	return disk_state;
413 }
414 
conn_lowest_conn(struct drbd_connection * connection)415 enum drbd_conns conn_lowest_conn(struct drbd_connection *connection)
416 {
417 	enum drbd_conns conn = C_MASK;
418 	struct drbd_peer_device *peer_device;
419 	int vnr;
420 
421 	rcu_read_lock();
422 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
423 		struct drbd_device *device = peer_device->device;
424 		conn = min_t(enum drbd_conns, conn, device->state.conn);
425 	}
426 	rcu_read_unlock();
427 
428 	return conn;
429 }
430 
no_peer_wf_report_params(struct drbd_connection * connection)431 static bool no_peer_wf_report_params(struct drbd_connection *connection)
432 {
433 	struct drbd_peer_device *peer_device;
434 	int vnr;
435 	bool rv = true;
436 
437 	rcu_read_lock();
438 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
439 		if (peer_device->device->state.conn == C_WF_REPORT_PARAMS) {
440 			rv = false;
441 			break;
442 		}
443 	rcu_read_unlock();
444 
445 	return rv;
446 }
447 
wake_up_all_devices(struct drbd_connection * connection)448 static void wake_up_all_devices(struct drbd_connection *connection)
449 {
450 	struct drbd_peer_device *peer_device;
451 	int vnr;
452 
453 	rcu_read_lock();
454 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
455 		wake_up(&peer_device->device->state_wait);
456 	rcu_read_unlock();
457 
458 }
459 
460 
461 /**
462  * cl_wide_st_chg() - true if the state change is a cluster wide one
463  * @device:	DRBD device.
464  * @os:		old (current) state.
465  * @ns:		new (wanted) state.
466  */
cl_wide_st_chg(struct drbd_device * device,union drbd_state os,union drbd_state ns)467 static int cl_wide_st_chg(struct drbd_device *device,
468 			  union drbd_state os, union drbd_state ns)
469 {
470 	return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED &&
471 		 ((os.role != R_PRIMARY && ns.role == R_PRIMARY) ||
472 		  (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
473 		  (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) ||
474 		  (os.disk != D_FAILED && ns.disk == D_FAILED))) ||
475 		(os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) ||
476 		(os.conn == C_CONNECTED && ns.conn == C_VERIFY_S) ||
477 		(os.conn == C_CONNECTED && ns.conn == C_WF_REPORT_PARAMS);
478 }
479 
480 static union drbd_state
apply_mask_val(union drbd_state os,union drbd_state mask,union drbd_state val)481 apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val)
482 {
483 	union drbd_state ns;
484 	ns.i = (os.i & ~mask.i) | val.i;
485 	return ns;
486 }
487 
488 enum drbd_state_rv
drbd_change_state(struct drbd_device * device,enum chg_state_flags f,union drbd_state mask,union drbd_state val)489 drbd_change_state(struct drbd_device *device, enum chg_state_flags f,
490 		  union drbd_state mask, union drbd_state val)
491 {
492 	unsigned long flags;
493 	union drbd_state ns;
494 	enum drbd_state_rv rv;
495 
496 	spin_lock_irqsave(&device->resource->req_lock, flags);
497 	ns = apply_mask_val(drbd_read_state(device), mask, val);
498 	rv = _drbd_set_state(device, ns, f, NULL);
499 	spin_unlock_irqrestore(&device->resource->req_lock, flags);
500 
501 	return rv;
502 }
503 
504 /**
505  * drbd_force_state() - Impose a change which happens outside our control on our state
506  * @device:	DRBD device.
507  * @mask:	mask of state bits to change.
508  * @val:	value of new state bits.
509  */
drbd_force_state(struct drbd_device * device,union drbd_state mask,union drbd_state val)510 void drbd_force_state(struct drbd_device *device,
511 	union drbd_state mask, union drbd_state val)
512 {
513 	drbd_change_state(device, CS_HARD, mask, val);
514 }
515 
516 static enum drbd_state_rv
_req_st_cond(struct drbd_device * device,union drbd_state mask,union drbd_state val)517 _req_st_cond(struct drbd_device *device, union drbd_state mask,
518 	     union drbd_state val)
519 {
520 	union drbd_state os, ns;
521 	unsigned long flags;
522 	enum drbd_state_rv rv;
523 
524 	if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &device->flags))
525 		return SS_CW_SUCCESS;
526 
527 	if (test_and_clear_bit(CL_ST_CHG_FAIL, &device->flags))
528 		return SS_CW_FAILED_BY_PEER;
529 
530 	spin_lock_irqsave(&device->resource->req_lock, flags);
531 	os = drbd_read_state(device);
532 	ns = sanitize_state(device, os, apply_mask_val(os, mask, val), NULL);
533 	rv = is_valid_transition(os, ns);
534 	if (rv >= SS_SUCCESS)
535 		rv = SS_UNKNOWN_ERROR;  /* cont waiting, otherwise fail. */
536 
537 	if (!cl_wide_st_chg(device, os, ns))
538 		rv = SS_CW_NO_NEED;
539 	if (rv == SS_UNKNOWN_ERROR) {
540 		rv = is_valid_state(device, ns);
541 		if (rv >= SS_SUCCESS) {
542 			rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection);
543 			if (rv >= SS_SUCCESS)
544 				rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
545 		}
546 	}
547 	spin_unlock_irqrestore(&device->resource->req_lock, flags);
548 
549 	return rv;
550 }
551 
552 /**
553  * drbd_req_state() - Perform an eventually cluster wide state change
554  * @device:	DRBD device.
555  * @mask:	mask of state bits to change.
556  * @val:	value of new state bits.
557  * @f:		flags
558  *
559  * Should not be called directly, use drbd_request_state() or
560  * _drbd_request_state().
561  */
562 static enum drbd_state_rv
drbd_req_state(struct drbd_device * device,union drbd_state mask,union drbd_state val,enum chg_state_flags f)563 drbd_req_state(struct drbd_device *device, union drbd_state mask,
564 	       union drbd_state val, enum chg_state_flags f)
565 {
566 	struct completion done;
567 	unsigned long flags;
568 	union drbd_state os, ns;
569 	enum drbd_state_rv rv;
570 	void *buffer = NULL;
571 
572 	init_completion(&done);
573 
574 	if (f & CS_SERIALIZE)
575 		mutex_lock(device->state_mutex);
576 	if (f & CS_INHIBIT_MD_IO)
577 		buffer = drbd_md_get_buffer(device, __func__);
578 
579 	spin_lock_irqsave(&device->resource->req_lock, flags);
580 	os = drbd_read_state(device);
581 	ns = sanitize_state(device, os, apply_mask_val(os, mask, val), NULL);
582 	rv = is_valid_transition(os, ns);
583 	if (rv < SS_SUCCESS) {
584 		spin_unlock_irqrestore(&device->resource->req_lock, flags);
585 		goto abort;
586 	}
587 
588 	if (cl_wide_st_chg(device, os, ns)) {
589 		rv = is_valid_state(device, ns);
590 		if (rv == SS_SUCCESS)
591 			rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection);
592 		spin_unlock_irqrestore(&device->resource->req_lock, flags);
593 
594 		if (rv < SS_SUCCESS) {
595 			if (f & CS_VERBOSE)
596 				print_st_err(device, os, ns, rv);
597 			goto abort;
598 		}
599 
600 		if (drbd_send_state_req(first_peer_device(device), mask, val)) {
601 			rv = SS_CW_FAILED_BY_PEER;
602 			if (f & CS_VERBOSE)
603 				print_st_err(device, os, ns, rv);
604 			goto abort;
605 		}
606 
607 		wait_event(device->state_wait,
608 			(rv = _req_st_cond(device, mask, val)));
609 
610 		if (rv < SS_SUCCESS) {
611 			if (f & CS_VERBOSE)
612 				print_st_err(device, os, ns, rv);
613 			goto abort;
614 		}
615 		spin_lock_irqsave(&device->resource->req_lock, flags);
616 		ns = apply_mask_val(drbd_read_state(device), mask, val);
617 		rv = _drbd_set_state(device, ns, f, &done);
618 	} else {
619 		rv = _drbd_set_state(device, ns, f, &done);
620 	}
621 
622 	spin_unlock_irqrestore(&device->resource->req_lock, flags);
623 
624 	if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) {
625 		D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
626 		wait_for_completion(&done);
627 	}
628 
629 abort:
630 	if (buffer)
631 		drbd_md_put_buffer(device);
632 	if (f & CS_SERIALIZE)
633 		mutex_unlock(device->state_mutex);
634 
635 	return rv;
636 }
637 
638 /**
639  * _drbd_request_state() - Request a state change (with flags)
640  * @device:	DRBD device.
641  * @mask:	mask of state bits to change.
642  * @val:	value of new state bits.
643  * @f:		flags
644  *
645  * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE
646  * flag, or when logging of failed state change requests is not desired.
647  */
648 enum drbd_state_rv
_drbd_request_state(struct drbd_device * device,union drbd_state mask,union drbd_state val,enum chg_state_flags f)649 _drbd_request_state(struct drbd_device *device, union drbd_state mask,
650 		    union drbd_state val, enum chg_state_flags f)
651 {
652 	enum drbd_state_rv rv;
653 
654 	wait_event(device->state_wait,
655 		   (rv = drbd_req_state(device, mask, val, f)) != SS_IN_TRANSIENT_STATE);
656 
657 	return rv;
658 }
659 
660 /*
661  * We grab drbd_md_get_buffer(), because we don't want to "fail" the disk while
662  * there is IO in-flight: the transition into D_FAILED for detach purposes
663  * may get misinterpreted as actual IO error in a confused endio function.
664  *
665  * We wrap it all into wait_event(), to retry in case the drbd_req_state()
666  * returns SS_IN_TRANSIENT_STATE.
667  *
668  * To avoid potential deadlock with e.g. the receiver thread trying to grab
669  * drbd_md_get_buffer() while trying to get out of the "transient state", we
670  * need to grab and release the meta data buffer inside of that wait_event loop.
671  */
672 static enum drbd_state_rv
request_detach(struct drbd_device * device)673 request_detach(struct drbd_device *device)
674 {
675 	return drbd_req_state(device, NS(disk, D_FAILED),
676 			CS_VERBOSE | CS_ORDERED | CS_INHIBIT_MD_IO);
677 }
678 
drbd_request_detach_interruptible(struct drbd_device * device)679 int drbd_request_detach_interruptible(struct drbd_device *device)
680 {
681 	int ret, rv;
682 
683 	drbd_suspend_io(device); /* so no-one is stuck in drbd_al_begin_io */
684 	wait_event_interruptible(device->state_wait,
685 		(rv = request_detach(device)) != SS_IN_TRANSIENT_STATE);
686 	drbd_resume_io(device);
687 
688 	ret = wait_event_interruptible(device->misc_wait,
689 			device->state.disk != D_FAILED);
690 
691 	if (rv == SS_IS_DISKLESS)
692 		rv = SS_NOTHING_TO_DO;
693 	if (ret)
694 		rv = ERR_INTR;
695 
696 	return rv;
697 }
698 
699 enum drbd_state_rv
_drbd_request_state_holding_state_mutex(struct drbd_device * device,union drbd_state mask,union drbd_state val,enum chg_state_flags f)700 _drbd_request_state_holding_state_mutex(struct drbd_device *device, union drbd_state mask,
701 		    union drbd_state val, enum chg_state_flags f)
702 {
703 	enum drbd_state_rv rv;
704 
705 	BUG_ON(f & CS_SERIALIZE);
706 
707 	wait_event_cmd(device->state_wait,
708 		       (rv = drbd_req_state(device, mask, val, f)) != SS_IN_TRANSIENT_STATE,
709 		       mutex_unlock(device->state_mutex),
710 		       mutex_lock(device->state_mutex));
711 
712 	return rv;
713 }
714 
print_st(struct drbd_device * device,const char * name,union drbd_state ns)715 static void print_st(struct drbd_device *device, const char *name, union drbd_state ns)
716 {
717 	drbd_err(device, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n",
718 	    name,
719 	    drbd_conn_str(ns.conn),
720 	    drbd_role_str(ns.role),
721 	    drbd_role_str(ns.peer),
722 	    drbd_disk_str(ns.disk),
723 	    drbd_disk_str(ns.pdsk),
724 	    is_susp(ns) ? 's' : 'r',
725 	    ns.aftr_isp ? 'a' : '-',
726 	    ns.peer_isp ? 'p' : '-',
727 	    ns.user_isp ? 'u' : '-',
728 	    ns.susp_fen ? 'F' : '-',
729 	    ns.susp_nod ? 'N' : '-'
730 	    );
731 }
732 
print_st_err(struct drbd_device * device,union drbd_state os,union drbd_state ns,enum drbd_state_rv err)733 void print_st_err(struct drbd_device *device, union drbd_state os,
734 	          union drbd_state ns, enum drbd_state_rv err)
735 {
736 	if (err == SS_IN_TRANSIENT_STATE)
737 		return;
738 	drbd_err(device, "State change failed: %s\n", drbd_set_st_err_str(err));
739 	print_st(device, " state", os);
740 	print_st(device, "wanted", ns);
741 }
742 
print_state_change(char * pb,union drbd_state os,union drbd_state ns,enum chg_state_flags flags)743 static long print_state_change(char *pb, union drbd_state os, union drbd_state ns,
744 			       enum chg_state_flags flags)
745 {
746 	char *pbp;
747 	pbp = pb;
748 	*pbp = 0;
749 
750 	if (ns.role != os.role && flags & CS_DC_ROLE)
751 		pbp += sprintf(pbp, "role( %s -> %s ) ",
752 			       drbd_role_str(os.role),
753 			       drbd_role_str(ns.role));
754 	if (ns.peer != os.peer && flags & CS_DC_PEER)
755 		pbp += sprintf(pbp, "peer( %s -> %s ) ",
756 			       drbd_role_str(os.peer),
757 			       drbd_role_str(ns.peer));
758 	if (ns.conn != os.conn && flags & CS_DC_CONN)
759 		pbp += sprintf(pbp, "conn( %s -> %s ) ",
760 			       drbd_conn_str(os.conn),
761 			       drbd_conn_str(ns.conn));
762 	if (ns.disk != os.disk && flags & CS_DC_DISK)
763 		pbp += sprintf(pbp, "disk( %s -> %s ) ",
764 			       drbd_disk_str(os.disk),
765 			       drbd_disk_str(ns.disk));
766 	if (ns.pdsk != os.pdsk && flags & CS_DC_PDSK)
767 		pbp += sprintf(pbp, "pdsk( %s -> %s ) ",
768 			       drbd_disk_str(os.pdsk),
769 			       drbd_disk_str(ns.pdsk));
770 
771 	return pbp - pb;
772 }
773 
drbd_pr_state_change(struct drbd_device * device,union drbd_state os,union drbd_state ns,enum chg_state_flags flags)774 static void drbd_pr_state_change(struct drbd_device *device, union drbd_state os, union drbd_state ns,
775 				 enum chg_state_flags flags)
776 {
777 	char pb[300];
778 	char *pbp = pb;
779 
780 	pbp += print_state_change(pbp, os, ns, flags ^ CS_DC_MASK);
781 
782 	if (ns.aftr_isp != os.aftr_isp)
783 		pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ",
784 			       os.aftr_isp,
785 			       ns.aftr_isp);
786 	if (ns.peer_isp != os.peer_isp)
787 		pbp += sprintf(pbp, "peer_isp( %d -> %d ) ",
788 			       os.peer_isp,
789 			       ns.peer_isp);
790 	if (ns.user_isp != os.user_isp)
791 		pbp += sprintf(pbp, "user_isp( %d -> %d ) ",
792 			       os.user_isp,
793 			       ns.user_isp);
794 
795 	if (pbp != pb)
796 		drbd_info(device, "%s\n", pb);
797 }
798 
conn_pr_state_change(struct drbd_connection * connection,union drbd_state os,union drbd_state ns,enum chg_state_flags flags)799 static void conn_pr_state_change(struct drbd_connection *connection, union drbd_state os, union drbd_state ns,
800 				 enum chg_state_flags flags)
801 {
802 	char pb[300];
803 	char *pbp = pb;
804 
805 	pbp += print_state_change(pbp, os, ns, flags);
806 
807 	if (is_susp(ns) != is_susp(os) && flags & CS_DC_SUSP)
808 		pbp += sprintf(pbp, "susp( %d -> %d ) ",
809 			       is_susp(os),
810 			       is_susp(ns));
811 
812 	if (pbp != pb)
813 		drbd_info(connection, "%s\n", pb);
814 }
815 
816 
817 /**
818  * is_valid_state() - Returns an SS_ error code if ns is not valid
819  * @device:	DRBD device.
820  * @ns:		State to consider.
821  */
822 static enum drbd_state_rv
is_valid_state(struct drbd_device * device,union drbd_state ns)823 is_valid_state(struct drbd_device *device, union drbd_state ns)
824 {
825 	/* See drbd_state_sw_errors in drbd_strings.c */
826 
827 	enum drbd_fencing_p fp;
828 	enum drbd_state_rv rv = SS_SUCCESS;
829 	struct net_conf *nc;
830 
831 	rcu_read_lock();
832 	fp = FP_DONT_CARE;
833 	if (get_ldev(device)) {
834 		fp = rcu_dereference(device->ldev->disk_conf)->fencing;
835 		put_ldev(device);
836 	}
837 
838 	nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
839 	if (nc) {
840 		if (!nc->two_primaries && ns.role == R_PRIMARY) {
841 			if (ns.peer == R_PRIMARY)
842 				rv = SS_TWO_PRIMARIES;
843 			else if (conn_highest_peer(first_peer_device(device)->connection) == R_PRIMARY)
844 				rv = SS_O_VOL_PEER_PRI;
845 		}
846 	}
847 
848 	if (rv <= 0)
849 		goto out; /* already found a reason to abort */
850 	else if (ns.role == R_SECONDARY && device->open_cnt)
851 		rv = SS_DEVICE_IN_USE;
852 
853 	else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE)
854 		rv = SS_NO_UP_TO_DATE_DISK;
855 
856 	else if (fp >= FP_RESOURCE &&
857 		 ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN)
858 		rv = SS_PRIMARY_NOP;
859 
860 	else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT)
861 		rv = SS_NO_UP_TO_DATE_DISK;
862 
863 	else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT)
864 		rv = SS_NO_LOCAL_DISK;
865 
866 	else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT)
867 		rv = SS_NO_REMOTE_DISK;
868 
869 	else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
870 		rv = SS_NO_UP_TO_DATE_DISK;
871 
872 	else if ((ns.conn == C_CONNECTED ||
873 		  ns.conn == C_WF_BITMAP_S ||
874 		  ns.conn == C_SYNC_SOURCE ||
875 		  ns.conn == C_PAUSED_SYNC_S) &&
876 		  ns.disk == D_OUTDATED)
877 		rv = SS_CONNECTED_OUTDATES;
878 
879 	else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
880 		 (nc->verify_alg[0] == 0))
881 		rv = SS_NO_VERIFY_ALG;
882 
883 	else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
884 		  first_peer_device(device)->connection->agreed_pro_version < 88)
885 		rv = SS_NOT_SUPPORTED;
886 
887 	else if (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
888 		rv = SS_NO_UP_TO_DATE_DISK;
889 
890 	else if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
891                  ns.pdsk == D_UNKNOWN)
892 		rv = SS_NEED_CONNECTION;
893 
894 	else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN)
895 		rv = SS_CONNECTED_OUTDATES;
896 
897 out:
898 	rcu_read_unlock();
899 
900 	return rv;
901 }
902 
903 /**
904  * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible
905  * This function limits state transitions that may be declined by DRBD. I.e.
906  * user requests (aka soft transitions).
907  * @os:		old state.
908  * @ns:		new state.
909  * @connection:  DRBD connection.
910  */
911 static enum drbd_state_rv
is_valid_soft_transition(union drbd_state os,union drbd_state ns,struct drbd_connection * connection)912 is_valid_soft_transition(union drbd_state os, union drbd_state ns, struct drbd_connection *connection)
913 {
914 	enum drbd_state_rv rv = SS_SUCCESS;
915 
916 	if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
917 	    os.conn > C_CONNECTED)
918 		rv = SS_RESYNC_RUNNING;
919 
920 	if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
921 		rv = SS_ALREADY_STANDALONE;
922 
923 	if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
924 		rv = SS_IS_DISKLESS;
925 
926 	if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
927 		rv = SS_NO_NET_CONFIG;
928 
929 	if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
930 		rv = SS_LOWER_THAN_OUTDATED;
931 
932 	if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
933 		rv = SS_IN_TRANSIENT_STATE;
934 
935 	/* While establishing a connection only allow cstate to change.
936 	   Delay/refuse role changes, detach attach etc... (they do not touch cstate) */
937 	if (test_bit(STATE_SENT, &connection->flags) &&
938 	    !((ns.conn == C_WF_REPORT_PARAMS && os.conn == C_WF_CONNECTION) ||
939 	      (ns.conn >= C_CONNECTED && os.conn == C_WF_REPORT_PARAMS)))
940 		rv = SS_IN_TRANSIENT_STATE;
941 
942 	/* Do not promote during resync handshake triggered by "force primary".
943 	 * This is a hack. It should really be rejected by the peer during the
944 	 * cluster wide state change request. */
945 	if (os.role != R_PRIMARY && ns.role == R_PRIMARY
946 		&& ns.pdsk == D_UP_TO_DATE
947 		&& ns.disk != D_UP_TO_DATE && ns.disk != D_DISKLESS
948 		&& (ns.conn <= C_WF_SYNC_UUID || ns.conn != os.conn))
949 			rv = SS_IN_TRANSIENT_STATE;
950 
951 	if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
952 		rv = SS_NEED_CONNECTION;
953 
954 	if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
955 	    ns.conn != os.conn && os.conn > C_CONNECTED)
956 		rv = SS_RESYNC_RUNNING;
957 
958 	if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
959 	    os.conn < C_CONNECTED)
960 		rv = SS_NEED_CONNECTION;
961 
962 	if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)
963 	    && os.conn < C_WF_REPORT_PARAMS)
964 		rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */
965 
966 	if (ns.conn == C_DISCONNECTING && ns.pdsk == D_OUTDATED &&
967 	    os.conn < C_CONNECTED && os.pdsk > D_OUTDATED)
968 		rv = SS_OUTDATE_WO_CONN;
969 
970 	return rv;
971 }
972 
973 static enum drbd_state_rv
is_valid_conn_transition(enum drbd_conns oc,enum drbd_conns nc)974 is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc)
975 {
976 	/* no change -> nothing to do, at least for the connection part */
977 	if (oc == nc)
978 		return SS_NOTHING_TO_DO;
979 
980 	/* disconnect of an unconfigured connection does not make sense */
981 	if (oc == C_STANDALONE && nc == C_DISCONNECTING)
982 		return SS_ALREADY_STANDALONE;
983 
984 	/* from C_STANDALONE, we start with C_UNCONNECTED */
985 	if (oc == C_STANDALONE && nc != C_UNCONNECTED)
986 		return SS_NEED_CONNECTION;
987 
988 	/* When establishing a connection we need to go through WF_REPORT_PARAMS!
989 	   Necessary to do the right thing upon invalidate-remote on a disconnected resource */
990 	if (oc < C_WF_REPORT_PARAMS && nc >= C_CONNECTED)
991 		return SS_NEED_CONNECTION;
992 
993 	/* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */
994 	if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING)
995 		return SS_IN_TRANSIENT_STATE;
996 
997 	/* After C_DISCONNECTING only C_STANDALONE may follow */
998 	if (oc == C_DISCONNECTING && nc != C_STANDALONE)
999 		return SS_IN_TRANSIENT_STATE;
1000 
1001 	return SS_SUCCESS;
1002 }
1003 
1004 
1005 /**
1006  * is_valid_transition() - Returns an SS_ error code if the state transition is not possible
1007  * This limits hard state transitions. Hard state transitions are facts there are
1008  * imposed on DRBD by the environment. E.g. disk broke or network broke down.
1009  * But those hard state transitions are still not allowed to do everything.
1010  * @ns:		new state.
1011  * @os:		old state.
1012  */
1013 static enum drbd_state_rv
is_valid_transition(union drbd_state os,union drbd_state ns)1014 is_valid_transition(union drbd_state os, union drbd_state ns)
1015 {
1016 	enum drbd_state_rv rv;
1017 
1018 	rv = is_valid_conn_transition(os.conn, ns.conn);
1019 
1020 	/* we cannot fail (again) if we already detached */
1021 	if (ns.disk == D_FAILED && os.disk == D_DISKLESS)
1022 		rv = SS_IS_DISKLESS;
1023 
1024 	return rv;
1025 }
1026 
print_sanitize_warnings(struct drbd_device * device,enum sanitize_state_warnings warn)1027 static void print_sanitize_warnings(struct drbd_device *device, enum sanitize_state_warnings warn)
1028 {
1029 	static const char *msg_table[] = {
1030 		[NO_WARNING] = "",
1031 		[ABORTED_ONLINE_VERIFY] = "Online-verify aborted.",
1032 		[ABORTED_RESYNC] = "Resync aborted.",
1033 		[CONNECTION_LOST_NEGOTIATING] = "Connection lost while negotiating, no data!",
1034 		[IMPLICITLY_UPGRADED_DISK] = "Implicitly upgraded disk",
1035 		[IMPLICITLY_UPGRADED_PDSK] = "Implicitly upgraded pdsk",
1036 	};
1037 
1038 	if (warn != NO_WARNING)
1039 		drbd_warn(device, "%s\n", msg_table[warn]);
1040 }
1041 
1042 /**
1043  * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
1044  * @device:	DRBD device.
1045  * @os:		old state.
1046  * @ns:		new state.
1047  * @warn:	placeholder for returned state warning.
1048  *
1049  * When we loose connection, we have to set the state of the peers disk (pdsk)
1050  * to D_UNKNOWN. This rule and many more along those lines are in this function.
1051  */
sanitize_state(struct drbd_device * device,union drbd_state os,union drbd_state ns,enum sanitize_state_warnings * warn)1052 static union drbd_state sanitize_state(struct drbd_device *device, union drbd_state os,
1053 				       union drbd_state ns, enum sanitize_state_warnings *warn)
1054 {
1055 	enum drbd_fencing_p fp;
1056 	enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max;
1057 
1058 	if (warn)
1059 		*warn = NO_WARNING;
1060 
1061 	fp = FP_DONT_CARE;
1062 	if (get_ldev(device)) {
1063 		rcu_read_lock();
1064 		fp = rcu_dereference(device->ldev->disk_conf)->fencing;
1065 		rcu_read_unlock();
1066 		put_ldev(device);
1067 	}
1068 
1069 	/* Implications from connection to peer and peer_isp */
1070 	if (ns.conn < C_CONNECTED) {
1071 		ns.peer_isp = 0;
1072 		ns.peer = R_UNKNOWN;
1073 		if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
1074 			ns.pdsk = D_UNKNOWN;
1075 	}
1076 
1077 	/* Clear the aftr_isp when becoming unconfigured */
1078 	if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
1079 		ns.aftr_isp = 0;
1080 
1081 	/* An implication of the disk states onto the connection state */
1082 	/* Abort resync if a disk fails/detaches */
1083 	if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
1084 		if (warn)
1085 			*warn = ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ?
1086 				ABORTED_ONLINE_VERIFY : ABORTED_RESYNC;
1087 		ns.conn = C_CONNECTED;
1088 	}
1089 
1090 	/* Connection breaks down before we finished "Negotiating" */
1091 	if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
1092 	    get_ldev_if_state(device, D_NEGOTIATING)) {
1093 		if (device->ed_uuid == device->ldev->md.uuid[UI_CURRENT]) {
1094 			ns.disk = device->new_state_tmp.disk;
1095 			ns.pdsk = device->new_state_tmp.pdsk;
1096 		} else {
1097 			if (warn)
1098 				*warn = CONNECTION_LOST_NEGOTIATING;
1099 			ns.disk = D_DISKLESS;
1100 			ns.pdsk = D_UNKNOWN;
1101 		}
1102 		put_ldev(device);
1103 	}
1104 
1105 	/* D_CONSISTENT and D_OUTDATED vanish when we get connected */
1106 	if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) {
1107 		if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED)
1108 			ns.disk = D_UP_TO_DATE;
1109 		if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)
1110 			ns.pdsk = D_UP_TO_DATE;
1111 	}
1112 
1113 	/* Implications of the connection state on the disk states */
1114 	disk_min = D_DISKLESS;
1115 	disk_max = D_UP_TO_DATE;
1116 	pdsk_min = D_INCONSISTENT;
1117 	pdsk_max = D_UNKNOWN;
1118 	switch ((enum drbd_conns)ns.conn) {
1119 	case C_WF_BITMAP_T:
1120 	case C_PAUSED_SYNC_T:
1121 	case C_STARTING_SYNC_T:
1122 	case C_WF_SYNC_UUID:
1123 	case C_BEHIND:
1124 		disk_min = D_INCONSISTENT;
1125 		disk_max = D_OUTDATED;
1126 		pdsk_min = D_UP_TO_DATE;
1127 		pdsk_max = D_UP_TO_DATE;
1128 		break;
1129 	case C_VERIFY_S:
1130 	case C_VERIFY_T:
1131 		disk_min = D_UP_TO_DATE;
1132 		disk_max = D_UP_TO_DATE;
1133 		pdsk_min = D_UP_TO_DATE;
1134 		pdsk_max = D_UP_TO_DATE;
1135 		break;
1136 	case C_CONNECTED:
1137 		disk_min = D_DISKLESS;
1138 		disk_max = D_UP_TO_DATE;
1139 		pdsk_min = D_DISKLESS;
1140 		pdsk_max = D_UP_TO_DATE;
1141 		break;
1142 	case C_WF_BITMAP_S:
1143 	case C_PAUSED_SYNC_S:
1144 	case C_STARTING_SYNC_S:
1145 	case C_AHEAD:
1146 		disk_min = D_UP_TO_DATE;
1147 		disk_max = D_UP_TO_DATE;
1148 		pdsk_min = D_INCONSISTENT;
1149 		pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/
1150 		break;
1151 	case C_SYNC_TARGET:
1152 		disk_min = D_INCONSISTENT;
1153 		disk_max = D_INCONSISTENT;
1154 		pdsk_min = D_UP_TO_DATE;
1155 		pdsk_max = D_UP_TO_DATE;
1156 		break;
1157 	case C_SYNC_SOURCE:
1158 		disk_min = D_UP_TO_DATE;
1159 		disk_max = D_UP_TO_DATE;
1160 		pdsk_min = D_INCONSISTENT;
1161 		pdsk_max = D_INCONSISTENT;
1162 		break;
1163 	case C_STANDALONE:
1164 	case C_DISCONNECTING:
1165 	case C_UNCONNECTED:
1166 	case C_TIMEOUT:
1167 	case C_BROKEN_PIPE:
1168 	case C_NETWORK_FAILURE:
1169 	case C_PROTOCOL_ERROR:
1170 	case C_TEAR_DOWN:
1171 	case C_WF_CONNECTION:
1172 	case C_WF_REPORT_PARAMS:
1173 	case C_MASK:
1174 		break;
1175 	}
1176 	if (ns.disk > disk_max)
1177 		ns.disk = disk_max;
1178 
1179 	if (ns.disk < disk_min) {
1180 		if (warn)
1181 			*warn = IMPLICITLY_UPGRADED_DISK;
1182 		ns.disk = disk_min;
1183 	}
1184 	if (ns.pdsk > pdsk_max)
1185 		ns.pdsk = pdsk_max;
1186 
1187 	if (ns.pdsk < pdsk_min) {
1188 		if (warn)
1189 			*warn = IMPLICITLY_UPGRADED_PDSK;
1190 		ns.pdsk = pdsk_min;
1191 	}
1192 
1193 	if (fp == FP_STONITH &&
1194 	    (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED) &&
1195 	    !(os.role == R_PRIMARY && os.conn < C_CONNECTED && os.pdsk > D_OUTDATED))
1196 		ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */
1197 
1198 	if (device->resource->res_opts.on_no_data == OND_SUSPEND_IO &&
1199 	    (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE) &&
1200 	    !(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE))
1201 		ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */
1202 
1203 	if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
1204 		if (ns.conn == C_SYNC_SOURCE)
1205 			ns.conn = C_PAUSED_SYNC_S;
1206 		if (ns.conn == C_SYNC_TARGET)
1207 			ns.conn = C_PAUSED_SYNC_T;
1208 	} else {
1209 		if (ns.conn == C_PAUSED_SYNC_S)
1210 			ns.conn = C_SYNC_SOURCE;
1211 		if (ns.conn == C_PAUSED_SYNC_T)
1212 			ns.conn = C_SYNC_TARGET;
1213 	}
1214 
1215 	return ns;
1216 }
1217 
drbd_resume_al(struct drbd_device * device)1218 void drbd_resume_al(struct drbd_device *device)
1219 {
1220 	if (test_and_clear_bit(AL_SUSPENDED, &device->flags))
1221 		drbd_info(device, "Resumed AL updates\n");
1222 }
1223 
1224 /* helper for _drbd_set_state */
set_ov_position(struct drbd_peer_device * peer_device,enum drbd_conns cs)1225 static void set_ov_position(struct drbd_peer_device *peer_device, enum drbd_conns cs)
1226 {
1227 	struct drbd_device *device = peer_device->device;
1228 
1229 	if (peer_device->connection->agreed_pro_version < 90)
1230 		device->ov_start_sector = 0;
1231 	device->rs_total = drbd_bm_bits(device);
1232 	device->ov_position = 0;
1233 	if (cs == C_VERIFY_T) {
1234 		/* starting online verify from an arbitrary position
1235 		 * does not fit well into the existing protocol.
1236 		 * on C_VERIFY_T, we initialize ov_left and friends
1237 		 * implicitly in receive_DataRequest once the
1238 		 * first P_OV_REQUEST is received */
1239 		device->ov_start_sector = ~(sector_t)0;
1240 	} else {
1241 		unsigned long bit = BM_SECT_TO_BIT(device->ov_start_sector);
1242 		if (bit >= device->rs_total) {
1243 			device->ov_start_sector =
1244 				BM_BIT_TO_SECT(device->rs_total - 1);
1245 			device->rs_total = 1;
1246 		} else
1247 			device->rs_total -= bit;
1248 		device->ov_position = device->ov_start_sector;
1249 	}
1250 	device->ov_left = device->rs_total;
1251 }
1252 
1253 /**
1254  * _drbd_set_state() - Set a new DRBD state
1255  * @device:	DRBD device.
1256  * @ns:		new state.
1257  * @flags:	Flags
1258  * @done:	Optional completion, that will get completed after the after_state_ch() finished
1259  *
1260  * Caller needs to hold req_lock. Do not call directly.
1261  */
1262 enum drbd_state_rv
_drbd_set_state(struct drbd_device * device,union drbd_state ns,enum chg_state_flags flags,struct completion * done)1263 _drbd_set_state(struct drbd_device *device, union drbd_state ns,
1264 	        enum chg_state_flags flags, struct completion *done)
1265 {
1266 	struct drbd_peer_device *peer_device = first_peer_device(device);
1267 	struct drbd_connection *connection = peer_device ? peer_device->connection : NULL;
1268 	union drbd_state os;
1269 	enum drbd_state_rv rv = SS_SUCCESS;
1270 	enum sanitize_state_warnings ssw;
1271 	struct after_state_chg_work *ascw;
1272 	struct drbd_state_change *state_change;
1273 
1274 	os = drbd_read_state(device);
1275 
1276 	ns = sanitize_state(device, os, ns, &ssw);
1277 	if (ns.i == os.i)
1278 		return SS_NOTHING_TO_DO;
1279 
1280 	rv = is_valid_transition(os, ns);
1281 	if (rv < SS_SUCCESS)
1282 		return rv;
1283 
1284 	if (!(flags & CS_HARD)) {
1285 		/*  pre-state-change checks ; only look at ns  */
1286 		/* See drbd_state_sw_errors in drbd_strings.c */
1287 
1288 		rv = is_valid_state(device, ns);
1289 		if (rv < SS_SUCCESS) {
1290 			/* If the old state was illegal as well, then let
1291 			   this happen...*/
1292 
1293 			if (is_valid_state(device, os) == rv)
1294 				rv = is_valid_soft_transition(os, ns, connection);
1295 		} else
1296 			rv = is_valid_soft_transition(os, ns, connection);
1297 	}
1298 
1299 	if (rv < SS_SUCCESS) {
1300 		if (flags & CS_VERBOSE)
1301 			print_st_err(device, os, ns, rv);
1302 		return rv;
1303 	}
1304 
1305 	print_sanitize_warnings(device, ssw);
1306 
1307 	drbd_pr_state_change(device, os, ns, flags);
1308 
1309 	/* Display changes to the susp* flags that where caused by the call to
1310 	   sanitize_state(). Only display it here if we where not called from
1311 	   _conn_request_state() */
1312 	if (!(flags & CS_DC_SUSP))
1313 		conn_pr_state_change(connection, os, ns,
1314 				     (flags & ~CS_DC_MASK) | CS_DC_SUSP);
1315 
1316 	/* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
1317 	 * on the ldev here, to be sure the transition -> D_DISKLESS resp.
1318 	 * drbd_ldev_destroy() won't happen before our corresponding
1319 	 * after_state_ch works run, where we put_ldev again. */
1320 	if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
1321 	    (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
1322 		atomic_inc(&device->local_cnt);
1323 
1324 	if (!is_sync_state(os.conn) && is_sync_state(ns.conn))
1325 		clear_bit(RS_DONE, &device->flags);
1326 
1327 	/* FIXME: Have any flags been set earlier in this function already? */
1328 	state_change = remember_old_state(device->resource, GFP_ATOMIC);
1329 
1330 	/* changes to local_cnt and device flags should be visible before
1331 	 * changes to state, which again should be visible before anything else
1332 	 * depending on that change happens. */
1333 	smp_wmb();
1334 	device->state.i = ns.i;
1335 	device->resource->susp = ns.susp;
1336 	device->resource->susp_nod = ns.susp_nod;
1337 	device->resource->susp_fen = ns.susp_fen;
1338 	smp_wmb();
1339 
1340 	remember_new_state(state_change);
1341 
1342 	/* put replicated vs not-replicated requests in seperate epochs */
1343 	if (drbd_should_do_remote((union drbd_dev_state)os.i) !=
1344 	    drbd_should_do_remote((union drbd_dev_state)ns.i))
1345 		start_new_tl_epoch(connection);
1346 
1347 	if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
1348 		drbd_print_uuids(device, "attached to UUIDs");
1349 
1350 	/* Wake up role changes, that were delayed because of connection establishing */
1351 	if (os.conn == C_WF_REPORT_PARAMS && ns.conn != C_WF_REPORT_PARAMS &&
1352 	    no_peer_wf_report_params(connection)) {
1353 		clear_bit(STATE_SENT, &connection->flags);
1354 		wake_up_all_devices(connection);
1355 	}
1356 
1357 	wake_up(&device->misc_wait);
1358 	wake_up(&device->state_wait);
1359 	wake_up(&connection->ping_wait);
1360 
1361 	/* Aborted verify run, or we reached the stop sector.
1362 	 * Log the last position, unless end-of-device. */
1363 	if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
1364 	    ns.conn <= C_CONNECTED) {
1365 		device->ov_start_sector =
1366 			BM_BIT_TO_SECT(drbd_bm_bits(device) - device->ov_left);
1367 		if (device->ov_left)
1368 			drbd_info(device, "Online Verify reached sector %llu\n",
1369 				(unsigned long long)device->ov_start_sector);
1370 	}
1371 
1372 	if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
1373 	    (ns.conn == C_SYNC_TARGET  || ns.conn == C_SYNC_SOURCE)) {
1374 		drbd_info(device, "Syncer continues.\n");
1375 		device->rs_paused += (long)jiffies
1376 				  -(long)device->rs_mark_time[device->rs_last_mark];
1377 		if (ns.conn == C_SYNC_TARGET)
1378 			mod_timer(&device->resync_timer, jiffies);
1379 	}
1380 
1381 	if ((os.conn == C_SYNC_TARGET  || os.conn == C_SYNC_SOURCE) &&
1382 	    (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
1383 		drbd_info(device, "Resync suspended\n");
1384 		device->rs_mark_time[device->rs_last_mark] = jiffies;
1385 	}
1386 
1387 	if (os.conn == C_CONNECTED &&
1388 	    (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
1389 		unsigned long now = jiffies;
1390 		int i;
1391 
1392 		set_ov_position(peer_device, ns.conn);
1393 		device->rs_start = now;
1394 		device->rs_last_sect_ev = 0;
1395 		device->ov_last_oos_size = 0;
1396 		device->ov_last_oos_start = 0;
1397 
1398 		for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1399 			device->rs_mark_left[i] = device->ov_left;
1400 			device->rs_mark_time[i] = now;
1401 		}
1402 
1403 		drbd_rs_controller_reset(peer_device);
1404 
1405 		if (ns.conn == C_VERIFY_S) {
1406 			drbd_info(device, "Starting Online Verify from sector %llu\n",
1407 					(unsigned long long)device->ov_position);
1408 			mod_timer(&device->resync_timer, jiffies);
1409 		}
1410 	}
1411 
1412 	if (get_ldev(device)) {
1413 		u32 mdf = device->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
1414 						 MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
1415 						 MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
1416 
1417 		mdf &= ~MDF_AL_CLEAN;
1418 		if (test_bit(CRASHED_PRIMARY, &device->flags))
1419 			mdf |= MDF_CRASHED_PRIMARY;
1420 		if (device->state.role == R_PRIMARY ||
1421 		    (device->state.pdsk < D_INCONSISTENT && device->state.peer == R_PRIMARY))
1422 			mdf |= MDF_PRIMARY_IND;
1423 		if (device->state.conn > C_WF_REPORT_PARAMS)
1424 			mdf |= MDF_CONNECTED_IND;
1425 		if (device->state.disk > D_INCONSISTENT)
1426 			mdf |= MDF_CONSISTENT;
1427 		if (device->state.disk > D_OUTDATED)
1428 			mdf |= MDF_WAS_UP_TO_DATE;
1429 		if (device->state.pdsk <= D_OUTDATED && device->state.pdsk >= D_INCONSISTENT)
1430 			mdf |= MDF_PEER_OUT_DATED;
1431 		if (mdf != device->ldev->md.flags) {
1432 			device->ldev->md.flags = mdf;
1433 			drbd_md_mark_dirty(device);
1434 		}
1435 		if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
1436 			drbd_set_ed_uuid(device, device->ldev->md.uuid[UI_CURRENT]);
1437 		put_ldev(device);
1438 	}
1439 
1440 	/* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
1441 	if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
1442 	    os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
1443 		set_bit(CONSIDER_RESYNC, &device->flags);
1444 
1445 	/* Receiver should clean up itself */
1446 	if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
1447 		drbd_thread_stop_nowait(&connection->receiver);
1448 
1449 	/* Now the receiver finished cleaning up itself, it should die */
1450 	if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
1451 		drbd_thread_stop_nowait(&connection->receiver);
1452 
1453 	/* Upon network failure, we need to restart the receiver. */
1454 	if (os.conn > C_WF_CONNECTION &&
1455 	    ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
1456 		drbd_thread_restart_nowait(&connection->receiver);
1457 
1458 	/* Resume AL writing if we get a connection */
1459 	if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) {
1460 		drbd_resume_al(device);
1461 		connection->connect_cnt++;
1462 	}
1463 
1464 	/* remember last attach time so request_timer_fn() won't
1465 	 * kill newly established sessions while we are still trying to thaw
1466 	 * previously frozen IO */
1467 	if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1468 	    ns.disk > D_NEGOTIATING)
1469 		device->last_reattach_jif = jiffies;
1470 
1471 	ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
1472 	if (ascw) {
1473 		ascw->os = os;
1474 		ascw->ns = ns;
1475 		ascw->flags = flags;
1476 		ascw->w.cb = w_after_state_ch;
1477 		ascw->device = device;
1478 		ascw->done = done;
1479 		ascw->state_change = state_change;
1480 		drbd_queue_work(&connection->sender_work,
1481 				&ascw->w);
1482 	} else {
1483 		drbd_err(device, "Could not kmalloc an ascw\n");
1484 	}
1485 
1486 	return rv;
1487 }
1488 
w_after_state_ch(struct drbd_work * w,int unused)1489 static int w_after_state_ch(struct drbd_work *w, int unused)
1490 {
1491 	struct after_state_chg_work *ascw =
1492 		container_of(w, struct after_state_chg_work, w);
1493 	struct drbd_device *device = ascw->device;
1494 
1495 	after_state_ch(device, ascw->os, ascw->ns, ascw->flags, ascw->state_change);
1496 	forget_state_change(ascw->state_change);
1497 	if (ascw->flags & CS_WAIT_COMPLETE)
1498 		complete(ascw->done);
1499 	kfree(ascw);
1500 
1501 	return 0;
1502 }
1503 
abw_start_sync(struct drbd_device * device,int rv)1504 static void abw_start_sync(struct drbd_device *device, int rv)
1505 {
1506 	if (rv) {
1507 		drbd_err(device, "Writing the bitmap failed not starting resync.\n");
1508 		_drbd_request_state(device, NS(conn, C_CONNECTED), CS_VERBOSE);
1509 		return;
1510 	}
1511 
1512 	switch (device->state.conn) {
1513 	case C_STARTING_SYNC_T:
1514 		_drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
1515 		break;
1516 	case C_STARTING_SYNC_S:
1517 		drbd_start_resync(device, C_SYNC_SOURCE);
1518 		break;
1519 	}
1520 }
1521 
drbd_bitmap_io_from_worker(struct drbd_device * device,int (* io_fn)(struct drbd_device *,struct drbd_peer_device *),char * why,enum bm_flag flags,struct drbd_peer_device * peer_device)1522 int drbd_bitmap_io_from_worker(struct drbd_device *device,
1523 		int (*io_fn)(struct drbd_device *, struct drbd_peer_device *),
1524 		char *why, enum bm_flag flags,
1525 		struct drbd_peer_device *peer_device)
1526 {
1527 	int rv;
1528 
1529 	D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
1530 
1531 	/* open coded non-blocking drbd_suspend_io(device); */
1532 	atomic_inc(&device->suspend_cnt);
1533 
1534 	drbd_bm_lock(device, why, flags);
1535 	rv = io_fn(device, peer_device);
1536 	drbd_bm_unlock(device);
1537 
1538 	drbd_resume_io(device);
1539 
1540 	return rv;
1541 }
1542 
notify_resource_state_change(struct sk_buff * skb,unsigned int seq,struct drbd_resource_state_change * resource_state_change,enum drbd_notification_type type)1543 int notify_resource_state_change(struct sk_buff *skb,
1544 				  unsigned int seq,
1545 				  struct drbd_resource_state_change *resource_state_change,
1546 				  enum drbd_notification_type type)
1547 {
1548 	struct drbd_resource *resource = resource_state_change->resource;
1549 	struct resource_info resource_info = {
1550 		.res_role = resource_state_change->role[NEW],
1551 		.res_susp = resource_state_change->susp[NEW],
1552 		.res_susp_nod = resource_state_change->susp_nod[NEW],
1553 		.res_susp_fen = resource_state_change->susp_fen[NEW],
1554 	};
1555 
1556 	return notify_resource_state(skb, seq, resource, &resource_info, type);
1557 }
1558 
notify_connection_state_change(struct sk_buff * skb,unsigned int seq,struct drbd_connection_state_change * connection_state_change,enum drbd_notification_type type)1559 int notify_connection_state_change(struct sk_buff *skb,
1560 				    unsigned int seq,
1561 				    struct drbd_connection_state_change *connection_state_change,
1562 				    enum drbd_notification_type type)
1563 {
1564 	struct drbd_connection *connection = connection_state_change->connection;
1565 	struct connection_info connection_info = {
1566 		.conn_connection_state = connection_state_change->cstate[NEW],
1567 		.conn_role = connection_state_change->peer_role[NEW],
1568 	};
1569 
1570 	return notify_connection_state(skb, seq, connection, &connection_info, type);
1571 }
1572 
notify_device_state_change(struct sk_buff * skb,unsigned int seq,struct drbd_device_state_change * device_state_change,enum drbd_notification_type type)1573 int notify_device_state_change(struct sk_buff *skb,
1574 				unsigned int seq,
1575 				struct drbd_device_state_change *device_state_change,
1576 				enum drbd_notification_type type)
1577 {
1578 	struct drbd_device *device = device_state_change->device;
1579 	struct device_info device_info = {
1580 		.dev_disk_state = device_state_change->disk_state[NEW],
1581 	};
1582 
1583 	return notify_device_state(skb, seq, device, &device_info, type);
1584 }
1585 
notify_peer_device_state_change(struct sk_buff * skb,unsigned int seq,struct drbd_peer_device_state_change * p,enum drbd_notification_type type)1586 int notify_peer_device_state_change(struct sk_buff *skb,
1587 				     unsigned int seq,
1588 				     struct drbd_peer_device_state_change *p,
1589 				     enum drbd_notification_type type)
1590 {
1591 	struct drbd_peer_device *peer_device = p->peer_device;
1592 	struct peer_device_info peer_device_info = {
1593 		.peer_repl_state = p->repl_state[NEW],
1594 		.peer_disk_state = p->disk_state[NEW],
1595 		.peer_resync_susp_user = p->resync_susp_user[NEW],
1596 		.peer_resync_susp_peer = p->resync_susp_peer[NEW],
1597 		.peer_resync_susp_dependency = p->resync_susp_dependency[NEW],
1598 	};
1599 
1600 	return notify_peer_device_state(skb, seq, peer_device, &peer_device_info, type);
1601 }
1602 
broadcast_state_change(struct drbd_state_change * state_change)1603 static void broadcast_state_change(struct drbd_state_change *state_change)
1604 {
1605 	struct drbd_resource_state_change *resource_state_change = &state_change->resource[0];
1606 	bool resource_state_has_changed;
1607 	unsigned int n_device, n_connection, n_peer_device, n_peer_devices;
1608 	int (*last_func)(struct sk_buff *, unsigned int, void *,
1609 			  enum drbd_notification_type) = NULL;
1610 	void *last_arg = NULL;
1611 
1612 #define HAS_CHANGED(state) ((state)[OLD] != (state)[NEW])
1613 #define FINAL_STATE_CHANGE(type) \
1614 	({ if (last_func) \
1615 		last_func(NULL, 0, last_arg, type); \
1616 	})
1617 #define REMEMBER_STATE_CHANGE(func, arg, type) \
1618 	({ FINAL_STATE_CHANGE(type | NOTIFY_CONTINUES); \
1619 	   last_func = (typeof(last_func))func; \
1620 	   last_arg = arg; \
1621 	 })
1622 
1623 	mutex_lock(&notification_mutex);
1624 
1625 	resource_state_has_changed =
1626 	    HAS_CHANGED(resource_state_change->role) ||
1627 	    HAS_CHANGED(resource_state_change->susp) ||
1628 	    HAS_CHANGED(resource_state_change->susp_nod) ||
1629 	    HAS_CHANGED(resource_state_change->susp_fen);
1630 
1631 	if (resource_state_has_changed)
1632 		REMEMBER_STATE_CHANGE(notify_resource_state_change,
1633 				      resource_state_change, NOTIFY_CHANGE);
1634 
1635 	for (n_connection = 0; n_connection < state_change->n_connections; n_connection++) {
1636 		struct drbd_connection_state_change *connection_state_change =
1637 				&state_change->connections[n_connection];
1638 
1639 		if (HAS_CHANGED(connection_state_change->peer_role) ||
1640 		    HAS_CHANGED(connection_state_change->cstate))
1641 			REMEMBER_STATE_CHANGE(notify_connection_state_change,
1642 					      connection_state_change, NOTIFY_CHANGE);
1643 	}
1644 
1645 	for (n_device = 0; n_device < state_change->n_devices; n_device++) {
1646 		struct drbd_device_state_change *device_state_change =
1647 			&state_change->devices[n_device];
1648 
1649 		if (HAS_CHANGED(device_state_change->disk_state))
1650 			REMEMBER_STATE_CHANGE(notify_device_state_change,
1651 					      device_state_change, NOTIFY_CHANGE);
1652 	}
1653 
1654 	n_peer_devices = state_change->n_devices * state_change->n_connections;
1655 	for (n_peer_device = 0; n_peer_device < n_peer_devices; n_peer_device++) {
1656 		struct drbd_peer_device_state_change *p =
1657 			&state_change->peer_devices[n_peer_device];
1658 
1659 		if (HAS_CHANGED(p->disk_state) ||
1660 		    HAS_CHANGED(p->repl_state) ||
1661 		    HAS_CHANGED(p->resync_susp_user) ||
1662 		    HAS_CHANGED(p->resync_susp_peer) ||
1663 		    HAS_CHANGED(p->resync_susp_dependency))
1664 			REMEMBER_STATE_CHANGE(notify_peer_device_state_change,
1665 					      p, NOTIFY_CHANGE);
1666 	}
1667 
1668 	FINAL_STATE_CHANGE(NOTIFY_CHANGE);
1669 	mutex_unlock(&notification_mutex);
1670 
1671 #undef HAS_CHANGED
1672 #undef FINAL_STATE_CHANGE
1673 #undef REMEMBER_STATE_CHANGE
1674 }
1675 
1676 /* takes old and new peer disk state */
lost_contact_to_peer_data(enum drbd_disk_state os,enum drbd_disk_state ns)1677 static bool lost_contact_to_peer_data(enum drbd_disk_state os, enum drbd_disk_state ns)
1678 {
1679 	if ((os >= D_INCONSISTENT && os != D_UNKNOWN && os != D_OUTDATED)
1680 	&&  (ns < D_INCONSISTENT || ns == D_UNKNOWN || ns == D_OUTDATED))
1681 		return true;
1682 
1683 	/* Scenario, starting with normal operation
1684 	 * Connected Primary/Secondary UpToDate/UpToDate
1685 	 * NetworkFailure Primary/Unknown UpToDate/DUnknown (frozen)
1686 	 * ...
1687 	 * Connected Primary/Secondary UpToDate/Diskless (resumed; needs to bump uuid!)
1688 	 */
1689 	if (os == D_UNKNOWN
1690 	&&  (ns == D_DISKLESS || ns == D_FAILED || ns == D_OUTDATED))
1691 		return true;
1692 
1693 	return false;
1694 }
1695 
1696 /**
1697  * after_state_ch() - Perform after state change actions that may sleep
1698  * @device:	DRBD device.
1699  * @os:		old state.
1700  * @ns:		new state.
1701  * @flags:	Flags
1702  * @state_change: state change to broadcast
1703  */
after_state_ch(struct drbd_device * device,union drbd_state os,union drbd_state ns,enum chg_state_flags flags,struct drbd_state_change * state_change)1704 static void after_state_ch(struct drbd_device *device, union drbd_state os,
1705 			   union drbd_state ns, enum chg_state_flags flags,
1706 			   struct drbd_state_change *state_change)
1707 {
1708 	struct drbd_resource *resource = device->resource;
1709 	struct drbd_peer_device *peer_device = first_peer_device(device);
1710 	struct drbd_connection *connection = peer_device ? peer_device->connection : NULL;
1711 	struct sib_info sib;
1712 
1713 	broadcast_state_change(state_change);
1714 
1715 	sib.sib_reason = SIB_STATE_CHANGE;
1716 	sib.os = os;
1717 	sib.ns = ns;
1718 
1719 	if ((os.disk != D_UP_TO_DATE || os.pdsk != D_UP_TO_DATE)
1720 	&&  (ns.disk == D_UP_TO_DATE && ns.pdsk == D_UP_TO_DATE)) {
1721 		clear_bit(CRASHED_PRIMARY, &device->flags);
1722 		if (device->p_uuid)
1723 			device->p_uuid[UI_FLAGS] &= ~((u64)2);
1724 	}
1725 
1726 	/* Inform userspace about the change... */
1727 	drbd_bcast_event(device, &sib);
1728 
1729 	if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
1730 	    (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
1731 		drbd_khelper(device, "pri-on-incon-degr");
1732 
1733 	/* Here we have the actions that are performed after a
1734 	   state change. This function might sleep */
1735 
1736 	if (ns.susp_nod) {
1737 		enum drbd_req_event what = NOTHING;
1738 
1739 		spin_lock_irq(&device->resource->req_lock);
1740 		if (os.conn < C_CONNECTED && conn_lowest_conn(connection) >= C_CONNECTED)
1741 			what = RESEND;
1742 
1743 		if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1744 		    conn_lowest_disk(connection) == D_UP_TO_DATE)
1745 			what = RESTART_FROZEN_DISK_IO;
1746 
1747 		if (resource->susp_nod && what != NOTHING) {
1748 			_tl_restart(connection, what);
1749 			_conn_request_state(connection,
1750 					    (union drbd_state) { { .susp_nod = 1 } },
1751 					    (union drbd_state) { { .susp_nod = 0 } },
1752 					    CS_VERBOSE);
1753 		}
1754 		spin_unlock_irq(&device->resource->req_lock);
1755 	}
1756 
1757 	if (ns.susp_fen) {
1758 		spin_lock_irq(&device->resource->req_lock);
1759 		if (resource->susp_fen && conn_lowest_conn(connection) >= C_CONNECTED) {
1760 			/* case2: The connection was established again: */
1761 			struct drbd_peer_device *peer_device;
1762 			int vnr;
1763 
1764 			rcu_read_lock();
1765 			idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
1766 				clear_bit(NEW_CUR_UUID, &peer_device->device->flags);
1767 			rcu_read_unlock();
1768 
1769 			/* We should actively create a new uuid, _before_
1770 			 * we resume/resent, if the peer is diskless
1771 			 * (recovery from a multiple error scenario).
1772 			 * Currently, this happens with a slight delay
1773 			 * below when checking lost_contact_to_peer_data() ...
1774 			 */
1775 			_tl_restart(connection, RESEND);
1776 			_conn_request_state(connection,
1777 					    (union drbd_state) { { .susp_fen = 1 } },
1778 					    (union drbd_state) { { .susp_fen = 0 } },
1779 					    CS_VERBOSE);
1780 		}
1781 		spin_unlock_irq(&device->resource->req_lock);
1782 	}
1783 
1784 	/* Became sync source.  With protocol >= 96, we still need to send out
1785 	 * the sync uuid now. Need to do that before any drbd_send_state, or
1786 	 * the other side may go "paused sync" before receiving the sync uuids,
1787 	 * which is unexpected. */
1788 	if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
1789 	    (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
1790 	    connection->agreed_pro_version >= 96 && get_ldev(device)) {
1791 		drbd_gen_and_send_sync_uuid(peer_device);
1792 		put_ldev(device);
1793 	}
1794 
1795 	/* Do not change the order of the if above and the two below... */
1796 	if (os.pdsk == D_DISKLESS &&
1797 	    ns.pdsk > D_DISKLESS && ns.pdsk != D_UNKNOWN) {      /* attach on the peer */
1798 		/* we probably will start a resync soon.
1799 		 * make sure those things are properly reset. */
1800 		device->rs_total = 0;
1801 		device->rs_failed = 0;
1802 		atomic_set(&device->rs_pending_cnt, 0);
1803 		drbd_rs_cancel_all(device);
1804 
1805 		drbd_send_uuids(peer_device);
1806 		drbd_send_state(peer_device, ns);
1807 	}
1808 	/* No point in queuing send_bitmap if we don't have a connection
1809 	 * anymore, so check also the _current_ state, not only the new state
1810 	 * at the time this work was queued. */
1811 	if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
1812 	    device->state.conn == C_WF_BITMAP_S)
1813 		drbd_queue_bitmap_io(device, &drbd_send_bitmap, NULL,
1814 				"send_bitmap (WFBitMapS)",
1815 				BM_LOCKED_TEST_ALLOWED, peer_device);
1816 
1817 	/* Lost contact to peer's copy of the data */
1818 	if (lost_contact_to_peer_data(os.pdsk, ns.pdsk)) {
1819 		if (get_ldev(device)) {
1820 			if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1821 			    device->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1822 				if (drbd_suspended(device)) {
1823 					set_bit(NEW_CUR_UUID, &device->flags);
1824 				} else {
1825 					drbd_uuid_new_current(device);
1826 					drbd_send_uuids(peer_device);
1827 				}
1828 			}
1829 			put_ldev(device);
1830 		}
1831 	}
1832 
1833 	if (ns.pdsk < D_INCONSISTENT && get_ldev(device)) {
1834 		if (os.peer != R_PRIMARY && ns.peer == R_PRIMARY &&
1835 		    device->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1836 			drbd_uuid_new_current(device);
1837 			drbd_send_uuids(peer_device);
1838 		}
1839 		/* D_DISKLESS Peer becomes secondary */
1840 		if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1841 			/* We may still be Primary ourselves.
1842 			 * No harm done if the bitmap still changes,
1843 			 * redirtied pages will follow later. */
1844 			drbd_bitmap_io_from_worker(device, &drbd_bm_write,
1845 				"demote diskless peer", BM_LOCKED_SET_ALLOWED, peer_device);
1846 		put_ldev(device);
1847 	}
1848 
1849 	/* Write out all changed bits on demote.
1850 	 * Though, no need to da that just yet
1851 	 * if there is a resync going on still */
1852 	if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
1853 		device->state.conn <= C_CONNECTED && get_ldev(device)) {
1854 		/* No changes to the bitmap expected this time, so assert that,
1855 		 * even though no harm was done if it did change. */
1856 		drbd_bitmap_io_from_worker(device, &drbd_bm_write,
1857 				"demote", BM_LOCKED_TEST_ALLOWED, peer_device);
1858 		put_ldev(device);
1859 	}
1860 
1861 	/* Last part of the attaching process ... */
1862 	if (ns.conn >= C_CONNECTED &&
1863 	    os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1864 		drbd_send_sizes(peer_device, 0, 0);  /* to start sync... */
1865 		drbd_send_uuids(peer_device);
1866 		drbd_send_state(peer_device, ns);
1867 	}
1868 
1869 	/* We want to pause/continue resync, tell peer. */
1870 	if (ns.conn >= C_CONNECTED &&
1871 	     ((os.aftr_isp != ns.aftr_isp) ||
1872 	      (os.user_isp != ns.user_isp)))
1873 		drbd_send_state(peer_device, ns);
1874 
1875 	/* In case one of the isp bits got set, suspend other devices. */
1876 	if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1877 	    (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1878 		suspend_other_sg(device);
1879 
1880 	/* Make sure the peer gets informed about eventual state
1881 	   changes (ISP bits) while we were in WFReportParams. */
1882 	if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1883 		drbd_send_state(peer_device, ns);
1884 
1885 	if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
1886 		drbd_send_state(peer_device, ns);
1887 
1888 	/* We are in the progress to start a full sync... */
1889 	if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1890 	    (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1891 		/* no other bitmap changes expected during this phase */
1892 		drbd_queue_bitmap_io(device,
1893 			&drbd_bmio_set_n_write, &abw_start_sync,
1894 			"set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED,
1895 			peer_device);
1896 
1897 	/* first half of local IO error, failure to attach,
1898 	 * or administrative detach */
1899 	if (os.disk != D_FAILED && ns.disk == D_FAILED) {
1900 		enum drbd_io_error_p eh = EP_PASS_ON;
1901 		int was_io_error = 0;
1902 		/* corresponding get_ldev was in _drbd_set_state, to serialize
1903 		 * our cleanup here with the transition to D_DISKLESS.
1904 		 * But is is still not save to dreference ldev here, since
1905 		 * we might come from an failed Attach before ldev was set. */
1906 		if (device->ldev) {
1907 			rcu_read_lock();
1908 			eh = rcu_dereference(device->ldev->disk_conf)->on_io_error;
1909 			rcu_read_unlock();
1910 
1911 			was_io_error = test_and_clear_bit(WAS_IO_ERROR, &device->flags);
1912 
1913 			/* Intentionally call this handler first, before drbd_send_state().
1914 			 * See: 2932204 drbd: call local-io-error handler early
1915 			 * People may chose to hard-reset the box from this handler.
1916 			 * It is useful if this looks like a "regular node crash". */
1917 			if (was_io_error && eh == EP_CALL_HELPER)
1918 				drbd_khelper(device, "local-io-error");
1919 
1920 			/* Immediately allow completion of all application IO,
1921 			 * that waits for completion from the local disk,
1922 			 * if this was a force-detach due to disk_timeout
1923 			 * or administrator request (drbdsetup detach --force).
1924 			 * Do NOT abort otherwise.
1925 			 * Aborting local requests may cause serious problems,
1926 			 * if requests are completed to upper layers already,
1927 			 * and then later the already submitted local bio completes.
1928 			 * This can cause DMA into former bio pages that meanwhile
1929 			 * have been re-used for other things.
1930 			 * So aborting local requests may cause crashes,
1931 			 * or even worse, silent data corruption.
1932 			 */
1933 			if (test_and_clear_bit(FORCE_DETACH, &device->flags))
1934 				tl_abort_disk_io(device);
1935 
1936 			/* current state still has to be D_FAILED,
1937 			 * there is only one way out: to D_DISKLESS,
1938 			 * and that may only happen after our put_ldev below. */
1939 			if (device->state.disk != D_FAILED)
1940 				drbd_err(device,
1941 					"ASSERT FAILED: disk is %s during detach\n",
1942 					drbd_disk_str(device->state.disk));
1943 
1944 			if (ns.conn >= C_CONNECTED)
1945 				drbd_send_state(peer_device, ns);
1946 
1947 			drbd_rs_cancel_all(device);
1948 
1949 			/* In case we want to get something to stable storage still,
1950 			 * this may be the last chance.
1951 			 * Following put_ldev may transition to D_DISKLESS. */
1952 			drbd_md_sync(device);
1953 		}
1954 		put_ldev(device);
1955 	}
1956 
1957 	/* second half of local IO error, failure to attach,
1958 	 * or administrative detach,
1959 	 * after local_cnt references have reached zero again */
1960 	if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
1961 		/* We must still be diskless,
1962 		 * re-attach has to be serialized with this! */
1963 		if (device->state.disk != D_DISKLESS)
1964 			drbd_err(device,
1965 				 "ASSERT FAILED: disk is %s while going diskless\n",
1966 				 drbd_disk_str(device->state.disk));
1967 
1968 		if (ns.conn >= C_CONNECTED)
1969 			drbd_send_state(peer_device, ns);
1970 		/* corresponding get_ldev in __drbd_set_state
1971 		 * this may finally trigger drbd_ldev_destroy. */
1972 		put_ldev(device);
1973 	}
1974 
1975 	/* Notify peer that I had a local IO error, and did not detached.. */
1976 	if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT && ns.conn >= C_CONNECTED)
1977 		drbd_send_state(peer_device, ns);
1978 
1979 	/* Disks got bigger while they were detached */
1980 	if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1981 	    test_and_clear_bit(RESYNC_AFTER_NEG, &device->flags)) {
1982 		if (ns.conn == C_CONNECTED)
1983 			resync_after_online_grow(device);
1984 	}
1985 
1986 	/* A resync finished or aborted, wake paused devices... */
1987 	if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1988 	    (os.peer_isp && !ns.peer_isp) ||
1989 	    (os.user_isp && !ns.user_isp))
1990 		resume_next_sg(device);
1991 
1992 	/* sync target done with resync.  Explicitly notify peer, even though
1993 	 * it should (at least for non-empty resyncs) already know itself. */
1994 	if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
1995 		drbd_send_state(peer_device, ns);
1996 
1997 	/* Verify finished, or reached stop sector.  Peer did not know about
1998 	 * the stop sector, and we may even have changed the stop sector during
1999 	 * verify to interrupt/stop early.  Send the new state. */
2000 	if (os.conn == C_VERIFY_S && ns.conn == C_CONNECTED
2001 	&& verify_can_do_stop_sector(device))
2002 		drbd_send_state(peer_device, ns);
2003 
2004 	/* This triggers bitmap writeout of potentially still unwritten pages
2005 	 * if the resync finished cleanly, or aborted because of peer disk
2006 	 * failure, or on transition from resync back to AHEAD/BEHIND.
2007 	 *
2008 	 * Connection loss is handled in drbd_disconnected() by the receiver.
2009 	 *
2010 	 * For resync aborted because of local disk failure, we cannot do
2011 	 * any bitmap writeout anymore.
2012 	 *
2013 	 * No harm done if some bits change during this phase.
2014 	 */
2015 	if ((os.conn > C_CONNECTED && os.conn < C_AHEAD) &&
2016 	    (ns.conn == C_CONNECTED || ns.conn >= C_AHEAD) && get_ldev(device)) {
2017 		drbd_queue_bitmap_io(device, &drbd_bm_write_copy_pages, NULL,
2018 			"write from resync_finished", BM_LOCKED_CHANGE_ALLOWED,
2019 			peer_device);
2020 		put_ldev(device);
2021 	}
2022 
2023 	if (ns.disk == D_DISKLESS &&
2024 	    ns.conn == C_STANDALONE &&
2025 	    ns.role == R_SECONDARY) {
2026 		if (os.aftr_isp != ns.aftr_isp)
2027 			resume_next_sg(device);
2028 	}
2029 
2030 	drbd_md_sync(device);
2031 }
2032 
2033 struct after_conn_state_chg_work {
2034 	struct drbd_work w;
2035 	enum drbd_conns oc;
2036 	union drbd_state ns_min;
2037 	union drbd_state ns_max; /* new, max state, over all devices */
2038 	enum chg_state_flags flags;
2039 	struct drbd_connection *connection;
2040 	struct drbd_state_change *state_change;
2041 };
2042 
w_after_conn_state_ch(struct drbd_work * w,int unused)2043 static int w_after_conn_state_ch(struct drbd_work *w, int unused)
2044 {
2045 	struct after_conn_state_chg_work *acscw =
2046 		container_of(w, struct after_conn_state_chg_work, w);
2047 	struct drbd_connection *connection = acscw->connection;
2048 	enum drbd_conns oc = acscw->oc;
2049 	union drbd_state ns_max = acscw->ns_max;
2050 	struct drbd_peer_device *peer_device;
2051 	int vnr;
2052 
2053 	broadcast_state_change(acscw->state_change);
2054 	forget_state_change(acscw->state_change);
2055 	kfree(acscw);
2056 
2057 	/* Upon network configuration, we need to start the receiver */
2058 	if (oc == C_STANDALONE && ns_max.conn == C_UNCONNECTED)
2059 		drbd_thread_start(&connection->receiver);
2060 
2061 	if (oc == C_DISCONNECTING && ns_max.conn == C_STANDALONE) {
2062 		struct net_conf *old_conf;
2063 
2064 		mutex_lock(&notification_mutex);
2065 		idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2066 			notify_peer_device_state(NULL, 0, peer_device, NULL,
2067 						 NOTIFY_DESTROY | NOTIFY_CONTINUES);
2068 		notify_connection_state(NULL, 0, connection, NULL, NOTIFY_DESTROY);
2069 		mutex_unlock(&notification_mutex);
2070 
2071 		mutex_lock(&connection->resource->conf_update);
2072 		old_conf = connection->net_conf;
2073 		connection->my_addr_len = 0;
2074 		connection->peer_addr_len = 0;
2075 		RCU_INIT_POINTER(connection->net_conf, NULL);
2076 		conn_free_crypto(connection);
2077 		mutex_unlock(&connection->resource->conf_update);
2078 
2079 		kvfree_rcu_mightsleep(old_conf);
2080 	}
2081 
2082 	if (ns_max.susp_fen) {
2083 		/* case1: The outdate peer handler is successful: */
2084 		if (ns_max.pdsk <= D_OUTDATED) {
2085 			rcu_read_lock();
2086 			idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2087 				struct drbd_device *device = peer_device->device;
2088 				if (test_bit(NEW_CUR_UUID, &device->flags)) {
2089 					drbd_uuid_new_current(device);
2090 					clear_bit(NEW_CUR_UUID, &device->flags);
2091 				}
2092 			}
2093 			rcu_read_unlock();
2094 			spin_lock_irq(&connection->resource->req_lock);
2095 			_tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
2096 			_conn_request_state(connection,
2097 					    (union drbd_state) { { .susp_fen = 1 } },
2098 					    (union drbd_state) { { .susp_fen = 0 } },
2099 					    CS_VERBOSE);
2100 			spin_unlock_irq(&connection->resource->req_lock);
2101 		}
2102 	}
2103 	conn_md_sync(connection);
2104 	kref_put(&connection->kref, drbd_destroy_connection);
2105 
2106 	return 0;
2107 }
2108 
conn_old_common_state(struct drbd_connection * connection,union drbd_state * pcs,enum chg_state_flags * pf)2109 static void conn_old_common_state(struct drbd_connection *connection, union drbd_state *pcs, enum chg_state_flags *pf)
2110 {
2111 	enum chg_state_flags flags = ~0;
2112 	struct drbd_peer_device *peer_device;
2113 	int vnr, first_vol = 1;
2114 	union drbd_dev_state os, cs = {
2115 		{ .role = R_SECONDARY,
2116 		  .peer = R_UNKNOWN,
2117 		  .conn = connection->cstate,
2118 		  .disk = D_DISKLESS,
2119 		  .pdsk = D_UNKNOWN,
2120 		} };
2121 
2122 	rcu_read_lock();
2123 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2124 		struct drbd_device *device = peer_device->device;
2125 		os = device->state;
2126 
2127 		if (first_vol) {
2128 			cs = os;
2129 			first_vol = 0;
2130 			continue;
2131 		}
2132 
2133 		if (cs.role != os.role)
2134 			flags &= ~CS_DC_ROLE;
2135 
2136 		if (cs.peer != os.peer)
2137 			flags &= ~CS_DC_PEER;
2138 
2139 		if (cs.conn != os.conn)
2140 			flags &= ~CS_DC_CONN;
2141 
2142 		if (cs.disk != os.disk)
2143 			flags &= ~CS_DC_DISK;
2144 
2145 		if (cs.pdsk != os.pdsk)
2146 			flags &= ~CS_DC_PDSK;
2147 	}
2148 	rcu_read_unlock();
2149 
2150 	*pf |= CS_DC_MASK;
2151 	*pf &= flags;
2152 	(*pcs).i = cs.i;
2153 }
2154 
2155 static enum drbd_state_rv
conn_is_valid_transition(struct drbd_connection * connection,union drbd_state mask,union drbd_state val,enum chg_state_flags flags)2156 conn_is_valid_transition(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2157 			 enum chg_state_flags flags)
2158 {
2159 	enum drbd_state_rv rv = SS_SUCCESS;
2160 	union drbd_state ns, os;
2161 	struct drbd_peer_device *peer_device;
2162 	int vnr;
2163 
2164 	rcu_read_lock();
2165 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2166 		struct drbd_device *device = peer_device->device;
2167 		os = drbd_read_state(device);
2168 		ns = sanitize_state(device, os, apply_mask_val(os, mask, val), NULL);
2169 
2170 		if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
2171 			ns.disk = os.disk;
2172 
2173 		if (ns.i == os.i)
2174 			continue;
2175 
2176 		rv = is_valid_transition(os, ns);
2177 
2178 		if (rv >= SS_SUCCESS && !(flags & CS_HARD)) {
2179 			rv = is_valid_state(device, ns);
2180 			if (rv < SS_SUCCESS) {
2181 				if (is_valid_state(device, os) == rv)
2182 					rv = is_valid_soft_transition(os, ns, connection);
2183 			} else
2184 				rv = is_valid_soft_transition(os, ns, connection);
2185 		}
2186 
2187 		if (rv < SS_SUCCESS) {
2188 			if (flags & CS_VERBOSE)
2189 				print_st_err(device, os, ns, rv);
2190 			break;
2191 		}
2192 	}
2193 	rcu_read_unlock();
2194 
2195 	return rv;
2196 }
2197 
2198 static void
conn_set_state(struct drbd_connection * connection,union drbd_state mask,union drbd_state val,union drbd_state * pns_min,union drbd_state * pns_max,enum chg_state_flags flags)2199 conn_set_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2200 	       union drbd_state *pns_min, union drbd_state *pns_max, enum chg_state_flags flags)
2201 {
2202 	union drbd_state ns, os, ns_max = { };
2203 	union drbd_state ns_min = {
2204 		{ .role = R_MASK,
2205 		  .peer = R_MASK,
2206 		  .conn = val.conn,
2207 		  .disk = D_MASK,
2208 		  .pdsk = D_MASK
2209 		} };
2210 	struct drbd_peer_device *peer_device;
2211 	enum drbd_state_rv rv;
2212 	int vnr, number_of_volumes = 0;
2213 
2214 	if (mask.conn == C_MASK) {
2215 		/* remember last connect time so request_timer_fn() won't
2216 		 * kill newly established sessions while we are still trying to thaw
2217 		 * previously frozen IO */
2218 		if (connection->cstate != C_WF_REPORT_PARAMS && val.conn == C_WF_REPORT_PARAMS)
2219 			connection->last_reconnect_jif = jiffies;
2220 
2221 		connection->cstate = val.conn;
2222 	}
2223 
2224 	rcu_read_lock();
2225 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2226 		struct drbd_device *device = peer_device->device;
2227 		number_of_volumes++;
2228 		os = drbd_read_state(device);
2229 		ns = apply_mask_val(os, mask, val);
2230 		ns = sanitize_state(device, os, ns, NULL);
2231 
2232 		if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
2233 			ns.disk = os.disk;
2234 
2235 		rv = _drbd_set_state(device, ns, flags, NULL);
2236 		BUG_ON(rv < SS_SUCCESS);
2237 		ns.i = device->state.i;
2238 		ns_max.role = max_role(ns.role, ns_max.role);
2239 		ns_max.peer = max_role(ns.peer, ns_max.peer);
2240 		ns_max.conn = max_t(enum drbd_conns, ns.conn, ns_max.conn);
2241 		ns_max.disk = max_t(enum drbd_disk_state, ns.disk, ns_max.disk);
2242 		ns_max.pdsk = max_t(enum drbd_disk_state, ns.pdsk, ns_max.pdsk);
2243 
2244 		ns_min.role = min_role(ns.role, ns_min.role);
2245 		ns_min.peer = min_role(ns.peer, ns_min.peer);
2246 		ns_min.conn = min_t(enum drbd_conns, ns.conn, ns_min.conn);
2247 		ns_min.disk = min_t(enum drbd_disk_state, ns.disk, ns_min.disk);
2248 		ns_min.pdsk = min_t(enum drbd_disk_state, ns.pdsk, ns_min.pdsk);
2249 	}
2250 	rcu_read_unlock();
2251 
2252 	if (number_of_volumes == 0) {
2253 		ns_min = ns_max = (union drbd_state) { {
2254 				.role = R_SECONDARY,
2255 				.peer = R_UNKNOWN,
2256 				.conn = val.conn,
2257 				.disk = D_DISKLESS,
2258 				.pdsk = D_UNKNOWN
2259 			} };
2260 	}
2261 
2262 	ns_min.susp = ns_max.susp = connection->resource->susp;
2263 	ns_min.susp_nod = ns_max.susp_nod = connection->resource->susp_nod;
2264 	ns_min.susp_fen = ns_max.susp_fen = connection->resource->susp_fen;
2265 
2266 	*pns_min = ns_min;
2267 	*pns_max = ns_max;
2268 }
2269 
2270 static enum drbd_state_rv
_conn_rq_cond(struct drbd_connection * connection,union drbd_state mask,union drbd_state val)2271 _conn_rq_cond(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
2272 {
2273 	enum drbd_state_rv err, rv = SS_UNKNOWN_ERROR; /* continue waiting */;
2274 
2275 	if (test_and_clear_bit(CONN_WD_ST_CHG_OKAY, &connection->flags))
2276 		rv = SS_CW_SUCCESS;
2277 
2278 	if (test_and_clear_bit(CONN_WD_ST_CHG_FAIL, &connection->flags))
2279 		rv = SS_CW_FAILED_BY_PEER;
2280 
2281 	err = conn_is_valid_transition(connection, mask, val, 0);
2282 	if (err == SS_SUCCESS && connection->cstate == C_WF_REPORT_PARAMS)
2283 		return rv;
2284 
2285 	return err;
2286 }
2287 
2288 enum drbd_state_rv
_conn_request_state(struct drbd_connection * connection,union drbd_state mask,union drbd_state val,enum chg_state_flags flags)2289 _conn_request_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2290 		    enum chg_state_flags flags)
2291 {
2292 	enum drbd_state_rv rv = SS_SUCCESS;
2293 	struct after_conn_state_chg_work *acscw;
2294 	enum drbd_conns oc = connection->cstate;
2295 	union drbd_state ns_max, ns_min, os;
2296 	bool have_mutex = false;
2297 	struct drbd_state_change *state_change;
2298 
2299 	if (mask.conn) {
2300 		rv = is_valid_conn_transition(oc, val.conn);
2301 		if (rv < SS_SUCCESS)
2302 			goto abort;
2303 	}
2304 
2305 	rv = conn_is_valid_transition(connection, mask, val, flags);
2306 	if (rv < SS_SUCCESS)
2307 		goto abort;
2308 
2309 	if (oc == C_WF_REPORT_PARAMS && val.conn == C_DISCONNECTING &&
2310 	    !(flags & (CS_LOCAL_ONLY | CS_HARD))) {
2311 
2312 		/* This will be a cluster-wide state change.
2313 		 * Need to give up the spinlock, grab the mutex,
2314 		 * then send the state change request, ... */
2315 		spin_unlock_irq(&connection->resource->req_lock);
2316 		mutex_lock(&connection->cstate_mutex);
2317 		have_mutex = true;
2318 
2319 		set_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
2320 		if (conn_send_state_req(connection, mask, val)) {
2321 			/* sending failed. */
2322 			clear_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
2323 			rv = SS_CW_FAILED_BY_PEER;
2324 			/* need to re-aquire the spin lock, though */
2325 			goto abort_unlocked;
2326 		}
2327 
2328 		if (val.conn == C_DISCONNECTING)
2329 			set_bit(DISCONNECT_SENT, &connection->flags);
2330 
2331 		/* ... and re-aquire the spinlock.
2332 		 * If _conn_rq_cond() returned >= SS_SUCCESS, we must call
2333 		 * conn_set_state() within the same spinlock. */
2334 		spin_lock_irq(&connection->resource->req_lock);
2335 		wait_event_lock_irq(connection->ping_wait,
2336 				(rv = _conn_rq_cond(connection, mask, val)),
2337 				connection->resource->req_lock);
2338 		clear_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
2339 		if (rv < SS_SUCCESS)
2340 			goto abort;
2341 	}
2342 
2343 	state_change = remember_old_state(connection->resource, GFP_ATOMIC);
2344 	conn_old_common_state(connection, &os, &flags);
2345 	flags |= CS_DC_SUSP;
2346 	conn_set_state(connection, mask, val, &ns_min, &ns_max, flags);
2347 	conn_pr_state_change(connection, os, ns_max, flags);
2348 	remember_new_state(state_change);
2349 
2350 	acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC);
2351 	if (acscw) {
2352 		acscw->oc = os.conn;
2353 		acscw->ns_min = ns_min;
2354 		acscw->ns_max = ns_max;
2355 		acscw->flags = flags;
2356 		acscw->w.cb = w_after_conn_state_ch;
2357 		kref_get(&connection->kref);
2358 		acscw->connection = connection;
2359 		acscw->state_change = state_change;
2360 		drbd_queue_work(&connection->sender_work, &acscw->w);
2361 	} else {
2362 		drbd_err(connection, "Could not kmalloc an acscw\n");
2363 	}
2364 
2365  abort:
2366 	if (have_mutex) {
2367 		/* mutex_unlock() "... must not be used in interrupt context.",
2368 		 * so give up the spinlock, then re-aquire it */
2369 		spin_unlock_irq(&connection->resource->req_lock);
2370  abort_unlocked:
2371 		mutex_unlock(&connection->cstate_mutex);
2372 		spin_lock_irq(&connection->resource->req_lock);
2373 	}
2374 	if (rv < SS_SUCCESS && flags & CS_VERBOSE) {
2375 		drbd_err(connection, "State change failed: %s\n", drbd_set_st_err_str(rv));
2376 		drbd_err(connection, " mask = 0x%x val = 0x%x\n", mask.i, val.i);
2377 		drbd_err(connection, " old_conn:%s wanted_conn:%s\n", drbd_conn_str(oc), drbd_conn_str(val.conn));
2378 	}
2379 	return rv;
2380 }
2381 
2382 enum drbd_state_rv
conn_request_state(struct drbd_connection * connection,union drbd_state mask,union drbd_state val,enum chg_state_flags flags)2383 conn_request_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2384 		   enum chg_state_flags flags)
2385 {
2386 	enum drbd_state_rv rv;
2387 
2388 	spin_lock_irq(&connection->resource->req_lock);
2389 	rv = _conn_request_state(connection, mask, val, flags);
2390 	spin_unlock_irq(&connection->resource->req_lock);
2391 
2392 	return rv;
2393 }
2394