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