1 // SPDX-License-Identifier: GPL-2.0-or-later
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 * @device: DRBD device.
908 * @ns: new state.
909 * @os: old state.
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_sync_abort:
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_device * device,enum drbd_conns cs)1225 static void set_ov_position(struct drbd_device *device, enum drbd_conns cs)
1226 {
1227 if (first_peer_device(device)->connection->agreed_pro_version < 90)
1228 device->ov_start_sector = 0;
1229 device->rs_total = drbd_bm_bits(device);
1230 device->ov_position = 0;
1231 if (cs == C_VERIFY_T) {
1232 /* starting online verify from an arbitrary position
1233 * does not fit well into the existing protocol.
1234 * on C_VERIFY_T, we initialize ov_left and friends
1235 * implicitly in receive_DataRequest once the
1236 * first P_OV_REQUEST is received */
1237 device->ov_start_sector = ~(sector_t)0;
1238 } else {
1239 unsigned long bit = BM_SECT_TO_BIT(device->ov_start_sector);
1240 if (bit >= device->rs_total) {
1241 device->ov_start_sector =
1242 BM_BIT_TO_SECT(device->rs_total - 1);
1243 device->rs_total = 1;
1244 } else
1245 device->rs_total -= bit;
1246 device->ov_position = device->ov_start_sector;
1247 }
1248 device->ov_left = device->rs_total;
1249 }
1250
1251 /**
1252 * _drbd_set_state() - Set a new DRBD state
1253 * @device: DRBD device.
1254 * @ns: new state.
1255 * @flags: Flags
1256 * @done: Optional completion, that will get completed after the after_state_ch() finished
1257 *
1258 * Caller needs to hold req_lock. Do not call directly.
1259 */
1260 enum drbd_state_rv
_drbd_set_state(struct drbd_device * device,union drbd_state ns,enum chg_state_flags flags,struct completion * done)1261 _drbd_set_state(struct drbd_device *device, union drbd_state ns,
1262 enum chg_state_flags flags, struct completion *done)
1263 {
1264 struct drbd_peer_device *peer_device = first_peer_device(device);
1265 struct drbd_connection *connection = peer_device ? peer_device->connection : NULL;
1266 union drbd_state os;
1267 enum drbd_state_rv rv = SS_SUCCESS;
1268 enum sanitize_state_warnings ssw;
1269 struct after_state_chg_work *ascw;
1270 struct drbd_state_change *state_change;
1271
1272 os = drbd_read_state(device);
1273
1274 ns = sanitize_state(device, os, ns, &ssw);
1275 if (ns.i == os.i)
1276 return SS_NOTHING_TO_DO;
1277
1278 rv = is_valid_transition(os, ns);
1279 if (rv < SS_SUCCESS)
1280 return rv;
1281
1282 if (!(flags & CS_HARD)) {
1283 /* pre-state-change checks ; only look at ns */
1284 /* See drbd_state_sw_errors in drbd_strings.c */
1285
1286 rv = is_valid_state(device, ns);
1287 if (rv < SS_SUCCESS) {
1288 /* If the old state was illegal as well, then let
1289 this happen...*/
1290
1291 if (is_valid_state(device, os) == rv)
1292 rv = is_valid_soft_transition(os, ns, connection);
1293 } else
1294 rv = is_valid_soft_transition(os, ns, connection);
1295 }
1296
1297 if (rv < SS_SUCCESS) {
1298 if (flags & CS_VERBOSE)
1299 print_st_err(device, os, ns, rv);
1300 return rv;
1301 }
1302
1303 print_sanitize_warnings(device, ssw);
1304
1305 drbd_pr_state_change(device, os, ns, flags);
1306
1307 /* Display changes to the susp* flags that where caused by the call to
1308 sanitize_state(). Only display it here if we where not called from
1309 _conn_request_state() */
1310 if (!(flags & CS_DC_SUSP))
1311 conn_pr_state_change(connection, os, ns,
1312 (flags & ~CS_DC_MASK) | CS_DC_SUSP);
1313
1314 /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
1315 * on the ldev here, to be sure the transition -> D_DISKLESS resp.
1316 * drbd_ldev_destroy() won't happen before our corresponding
1317 * after_state_ch works run, where we put_ldev again. */
1318 if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
1319 (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
1320 atomic_inc(&device->local_cnt);
1321
1322 if (!is_sync_state(os.conn) && is_sync_state(ns.conn))
1323 clear_bit(RS_DONE, &device->flags);
1324
1325 /* FIXME: Have any flags been set earlier in this function already? */
1326 state_change = remember_old_state(device->resource, GFP_ATOMIC);
1327
1328 /* changes to local_cnt and device flags should be visible before
1329 * changes to state, which again should be visible before anything else
1330 * depending on that change happens. */
1331 smp_wmb();
1332 device->state.i = ns.i;
1333 device->resource->susp = ns.susp;
1334 device->resource->susp_nod = ns.susp_nod;
1335 device->resource->susp_fen = ns.susp_fen;
1336 smp_wmb();
1337
1338 remember_new_state(state_change);
1339
1340 /* put replicated vs not-replicated requests in seperate epochs */
1341 if (drbd_should_do_remote((union drbd_dev_state)os.i) !=
1342 drbd_should_do_remote((union drbd_dev_state)ns.i))
1343 start_new_tl_epoch(connection);
1344
1345 if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
1346 drbd_print_uuids(device, "attached to UUIDs");
1347
1348 /* Wake up role changes, that were delayed because of connection establishing */
1349 if (os.conn == C_WF_REPORT_PARAMS && ns.conn != C_WF_REPORT_PARAMS &&
1350 no_peer_wf_report_params(connection)) {
1351 clear_bit(STATE_SENT, &connection->flags);
1352 wake_up_all_devices(connection);
1353 }
1354
1355 wake_up(&device->misc_wait);
1356 wake_up(&device->state_wait);
1357 wake_up(&connection->ping_wait);
1358
1359 /* Aborted verify run, or we reached the stop sector.
1360 * Log the last position, unless end-of-device. */
1361 if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
1362 ns.conn <= C_CONNECTED) {
1363 device->ov_start_sector =
1364 BM_BIT_TO_SECT(drbd_bm_bits(device) - device->ov_left);
1365 if (device->ov_left)
1366 drbd_info(device, "Online Verify reached sector %llu\n",
1367 (unsigned long long)device->ov_start_sector);
1368 }
1369
1370 if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
1371 (ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)) {
1372 drbd_info(device, "Syncer continues.\n");
1373 device->rs_paused += (long)jiffies
1374 -(long)device->rs_mark_time[device->rs_last_mark];
1375 if (ns.conn == C_SYNC_TARGET)
1376 mod_timer(&device->resync_timer, jiffies);
1377 }
1378
1379 if ((os.conn == C_SYNC_TARGET || os.conn == C_SYNC_SOURCE) &&
1380 (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
1381 drbd_info(device, "Resync suspended\n");
1382 device->rs_mark_time[device->rs_last_mark] = jiffies;
1383 }
1384
1385 if (os.conn == C_CONNECTED &&
1386 (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
1387 unsigned long now = jiffies;
1388 int i;
1389
1390 set_ov_position(device, ns.conn);
1391 device->rs_start = now;
1392 device->rs_last_sect_ev = 0;
1393 device->ov_last_oos_size = 0;
1394 device->ov_last_oos_start = 0;
1395
1396 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1397 device->rs_mark_left[i] = device->ov_left;
1398 device->rs_mark_time[i] = now;
1399 }
1400
1401 drbd_rs_controller_reset(device);
1402
1403 if (ns.conn == C_VERIFY_S) {
1404 drbd_info(device, "Starting Online Verify from sector %llu\n",
1405 (unsigned long long)device->ov_position);
1406 mod_timer(&device->resync_timer, jiffies);
1407 }
1408 }
1409
1410 if (get_ldev(device)) {
1411 u32 mdf = device->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
1412 MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
1413 MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
1414
1415 mdf &= ~MDF_AL_CLEAN;
1416 if (test_bit(CRASHED_PRIMARY, &device->flags))
1417 mdf |= MDF_CRASHED_PRIMARY;
1418 if (device->state.role == R_PRIMARY ||
1419 (device->state.pdsk < D_INCONSISTENT && device->state.peer == R_PRIMARY))
1420 mdf |= MDF_PRIMARY_IND;
1421 if (device->state.conn > C_WF_REPORT_PARAMS)
1422 mdf |= MDF_CONNECTED_IND;
1423 if (device->state.disk > D_INCONSISTENT)
1424 mdf |= MDF_CONSISTENT;
1425 if (device->state.disk > D_OUTDATED)
1426 mdf |= MDF_WAS_UP_TO_DATE;
1427 if (device->state.pdsk <= D_OUTDATED && device->state.pdsk >= D_INCONSISTENT)
1428 mdf |= MDF_PEER_OUT_DATED;
1429 if (mdf != device->ldev->md.flags) {
1430 device->ldev->md.flags = mdf;
1431 drbd_md_mark_dirty(device);
1432 }
1433 if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
1434 drbd_set_ed_uuid(device, device->ldev->md.uuid[UI_CURRENT]);
1435 put_ldev(device);
1436 }
1437
1438 /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
1439 if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
1440 os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
1441 set_bit(CONSIDER_RESYNC, &device->flags);
1442
1443 /* Receiver should clean up itself */
1444 if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
1445 drbd_thread_stop_nowait(&connection->receiver);
1446
1447 /* Now the receiver finished cleaning up itself, it should die */
1448 if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
1449 drbd_thread_stop_nowait(&connection->receiver);
1450
1451 /* Upon network failure, we need to restart the receiver. */
1452 if (os.conn > C_WF_CONNECTION &&
1453 ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
1454 drbd_thread_restart_nowait(&connection->receiver);
1455
1456 /* Resume AL writing if we get a connection */
1457 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) {
1458 drbd_resume_al(device);
1459 connection->connect_cnt++;
1460 }
1461
1462 /* remember last attach time so request_timer_fn() won't
1463 * kill newly established sessions while we are still trying to thaw
1464 * previously frozen IO */
1465 if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1466 ns.disk > D_NEGOTIATING)
1467 device->last_reattach_jif = jiffies;
1468
1469 ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
1470 if (ascw) {
1471 ascw->os = os;
1472 ascw->ns = ns;
1473 ascw->flags = flags;
1474 ascw->w.cb = w_after_state_ch;
1475 ascw->device = device;
1476 ascw->done = done;
1477 ascw->state_change = state_change;
1478 drbd_queue_work(&connection->sender_work,
1479 &ascw->w);
1480 } else {
1481 drbd_err(device, "Could not kmalloc an ascw\n");
1482 }
1483
1484 return rv;
1485 }
1486
w_after_state_ch(struct drbd_work * w,int unused)1487 static int w_after_state_ch(struct drbd_work *w, int unused)
1488 {
1489 struct after_state_chg_work *ascw =
1490 container_of(w, struct after_state_chg_work, w);
1491 struct drbd_device *device = ascw->device;
1492
1493 after_state_ch(device, ascw->os, ascw->ns, ascw->flags, ascw->state_change);
1494 forget_state_change(ascw->state_change);
1495 if (ascw->flags & CS_WAIT_COMPLETE)
1496 complete(ascw->done);
1497 kfree(ascw);
1498
1499 return 0;
1500 }
1501
abw_start_sync(struct drbd_device * device,int rv)1502 static void abw_start_sync(struct drbd_device *device, int rv)
1503 {
1504 if (rv) {
1505 drbd_err(device, "Writing the bitmap failed not starting resync.\n");
1506 _drbd_request_state(device, NS(conn, C_CONNECTED), CS_VERBOSE);
1507 return;
1508 }
1509
1510 switch (device->state.conn) {
1511 case C_STARTING_SYNC_T:
1512 _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
1513 break;
1514 case C_STARTING_SYNC_S:
1515 drbd_start_resync(device, C_SYNC_SOURCE);
1516 break;
1517 }
1518 }
1519
drbd_bitmap_io_from_worker(struct drbd_device * device,int (* io_fn)(struct drbd_device *),char * why,enum bm_flag flags)1520 int drbd_bitmap_io_from_worker(struct drbd_device *device,
1521 int (*io_fn)(struct drbd_device *),
1522 char *why, enum bm_flag flags)
1523 {
1524 int rv;
1525
1526 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
1527
1528 /* open coded non-blocking drbd_suspend_io(device); */
1529 atomic_inc(&device->suspend_cnt);
1530
1531 drbd_bm_lock(device, why, flags);
1532 rv = io_fn(device);
1533 drbd_bm_unlock(device);
1534
1535 drbd_resume_io(device);
1536
1537 return rv;
1538 }
1539
notify_resource_state_change(struct sk_buff * skb,unsigned int seq,struct drbd_resource_state_change * resource_state_change,enum drbd_notification_type type)1540 int notify_resource_state_change(struct sk_buff *skb,
1541 unsigned int seq,
1542 struct drbd_resource_state_change *resource_state_change,
1543 enum drbd_notification_type type)
1544 {
1545 struct drbd_resource *resource = resource_state_change->resource;
1546 struct resource_info resource_info = {
1547 .res_role = resource_state_change->role[NEW],
1548 .res_susp = resource_state_change->susp[NEW],
1549 .res_susp_nod = resource_state_change->susp_nod[NEW],
1550 .res_susp_fen = resource_state_change->susp_fen[NEW],
1551 };
1552
1553 return notify_resource_state(skb, seq, resource, &resource_info, type);
1554 }
1555
notify_connection_state_change(struct sk_buff * skb,unsigned int seq,struct drbd_connection_state_change * connection_state_change,enum drbd_notification_type type)1556 int notify_connection_state_change(struct sk_buff *skb,
1557 unsigned int seq,
1558 struct drbd_connection_state_change *connection_state_change,
1559 enum drbd_notification_type type)
1560 {
1561 struct drbd_connection *connection = connection_state_change->connection;
1562 struct connection_info connection_info = {
1563 .conn_connection_state = connection_state_change->cstate[NEW],
1564 .conn_role = connection_state_change->peer_role[NEW],
1565 };
1566
1567 return notify_connection_state(skb, seq, connection, &connection_info, type);
1568 }
1569
notify_device_state_change(struct sk_buff * skb,unsigned int seq,struct drbd_device_state_change * device_state_change,enum drbd_notification_type type)1570 int notify_device_state_change(struct sk_buff *skb,
1571 unsigned int seq,
1572 struct drbd_device_state_change *device_state_change,
1573 enum drbd_notification_type type)
1574 {
1575 struct drbd_device *device = device_state_change->device;
1576 struct device_info device_info = {
1577 .dev_disk_state = device_state_change->disk_state[NEW],
1578 };
1579
1580 return notify_device_state(skb, seq, device, &device_info, type);
1581 }
1582
notify_peer_device_state_change(struct sk_buff * skb,unsigned int seq,struct drbd_peer_device_state_change * p,enum drbd_notification_type type)1583 int notify_peer_device_state_change(struct sk_buff *skb,
1584 unsigned int seq,
1585 struct drbd_peer_device_state_change *p,
1586 enum drbd_notification_type type)
1587 {
1588 struct drbd_peer_device *peer_device = p->peer_device;
1589 struct peer_device_info peer_device_info = {
1590 .peer_repl_state = p->repl_state[NEW],
1591 .peer_disk_state = p->disk_state[NEW],
1592 .peer_resync_susp_user = p->resync_susp_user[NEW],
1593 .peer_resync_susp_peer = p->resync_susp_peer[NEW],
1594 .peer_resync_susp_dependency = p->resync_susp_dependency[NEW],
1595 };
1596
1597 return notify_peer_device_state(skb, seq, peer_device, &peer_device_info, type);
1598 }
1599
broadcast_state_change(struct drbd_state_change * state_change)1600 static void broadcast_state_change(struct drbd_state_change *state_change)
1601 {
1602 struct drbd_resource_state_change *resource_state_change = &state_change->resource[0];
1603 bool resource_state_has_changed;
1604 unsigned int n_device, n_connection, n_peer_device, n_peer_devices;
1605 int (*last_func)(struct sk_buff *, unsigned int, void *,
1606 enum drbd_notification_type) = NULL;
1607 void *last_arg = NULL;
1608
1609 #define HAS_CHANGED(state) ((state)[OLD] != (state)[NEW])
1610 #define FINAL_STATE_CHANGE(type) \
1611 ({ if (last_func) \
1612 last_func(NULL, 0, last_arg, type); \
1613 })
1614 #define REMEMBER_STATE_CHANGE(func, arg, type) \
1615 ({ FINAL_STATE_CHANGE(type | NOTIFY_CONTINUES); \
1616 last_func = (typeof(last_func))func; \
1617 last_arg = arg; \
1618 })
1619
1620 mutex_lock(¬ification_mutex);
1621
1622 resource_state_has_changed =
1623 HAS_CHANGED(resource_state_change->role) ||
1624 HAS_CHANGED(resource_state_change->susp) ||
1625 HAS_CHANGED(resource_state_change->susp_nod) ||
1626 HAS_CHANGED(resource_state_change->susp_fen);
1627
1628 if (resource_state_has_changed)
1629 REMEMBER_STATE_CHANGE(notify_resource_state_change,
1630 resource_state_change, NOTIFY_CHANGE);
1631
1632 for (n_connection = 0; n_connection < state_change->n_connections; n_connection++) {
1633 struct drbd_connection_state_change *connection_state_change =
1634 &state_change->connections[n_connection];
1635
1636 if (HAS_CHANGED(connection_state_change->peer_role) ||
1637 HAS_CHANGED(connection_state_change->cstate))
1638 REMEMBER_STATE_CHANGE(notify_connection_state_change,
1639 connection_state_change, NOTIFY_CHANGE);
1640 }
1641
1642 for (n_device = 0; n_device < state_change->n_devices; n_device++) {
1643 struct drbd_device_state_change *device_state_change =
1644 &state_change->devices[n_device];
1645
1646 if (HAS_CHANGED(device_state_change->disk_state))
1647 REMEMBER_STATE_CHANGE(notify_device_state_change,
1648 device_state_change, NOTIFY_CHANGE);
1649 }
1650
1651 n_peer_devices = state_change->n_devices * state_change->n_connections;
1652 for (n_peer_device = 0; n_peer_device < n_peer_devices; n_peer_device++) {
1653 struct drbd_peer_device_state_change *p =
1654 &state_change->peer_devices[n_peer_device];
1655
1656 if (HAS_CHANGED(p->disk_state) ||
1657 HAS_CHANGED(p->repl_state) ||
1658 HAS_CHANGED(p->resync_susp_user) ||
1659 HAS_CHANGED(p->resync_susp_peer) ||
1660 HAS_CHANGED(p->resync_susp_dependency))
1661 REMEMBER_STATE_CHANGE(notify_peer_device_state_change,
1662 p, NOTIFY_CHANGE);
1663 }
1664
1665 FINAL_STATE_CHANGE(NOTIFY_CHANGE);
1666 mutex_unlock(¬ification_mutex);
1667
1668 #undef HAS_CHANGED
1669 #undef FINAL_STATE_CHANGE
1670 #undef REMEMBER_STATE_CHANGE
1671 }
1672
1673 /* takes old and new peer disk state */
lost_contact_to_peer_data(enum drbd_disk_state os,enum drbd_disk_state ns)1674 static bool lost_contact_to_peer_data(enum drbd_disk_state os, enum drbd_disk_state ns)
1675 {
1676 if ((os >= D_INCONSISTENT && os != D_UNKNOWN && os != D_OUTDATED)
1677 && (ns < D_INCONSISTENT || ns == D_UNKNOWN || ns == D_OUTDATED))
1678 return true;
1679
1680 /* Scenario, starting with normal operation
1681 * Connected Primary/Secondary UpToDate/UpToDate
1682 * NetworkFailure Primary/Unknown UpToDate/DUnknown (frozen)
1683 * ...
1684 * Connected Primary/Secondary UpToDate/Diskless (resumed; needs to bump uuid!)
1685 */
1686 if (os == D_UNKNOWN
1687 && (ns == D_DISKLESS || ns == D_FAILED || ns == D_OUTDATED))
1688 return true;
1689
1690 return false;
1691 }
1692
1693 /**
1694 * after_state_ch() - Perform after state change actions that may sleep
1695 * @device: DRBD device.
1696 * @os: old state.
1697 * @ns: new state.
1698 * @flags: Flags
1699 */
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)1700 static void after_state_ch(struct drbd_device *device, union drbd_state os,
1701 union drbd_state ns, enum chg_state_flags flags,
1702 struct drbd_state_change *state_change)
1703 {
1704 struct drbd_resource *resource = device->resource;
1705 struct drbd_peer_device *peer_device = first_peer_device(device);
1706 struct drbd_connection *connection = peer_device ? peer_device->connection : NULL;
1707 struct sib_info sib;
1708
1709 broadcast_state_change(state_change);
1710
1711 sib.sib_reason = SIB_STATE_CHANGE;
1712 sib.os = os;
1713 sib.ns = ns;
1714
1715 if ((os.disk != D_UP_TO_DATE || os.pdsk != D_UP_TO_DATE)
1716 && (ns.disk == D_UP_TO_DATE && ns.pdsk == D_UP_TO_DATE)) {
1717 clear_bit(CRASHED_PRIMARY, &device->flags);
1718 if (device->p_uuid)
1719 device->p_uuid[UI_FLAGS] &= ~((u64)2);
1720 }
1721
1722 /* Inform userspace about the change... */
1723 drbd_bcast_event(device, &sib);
1724
1725 if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
1726 (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
1727 drbd_khelper(device, "pri-on-incon-degr");
1728
1729 /* Here we have the actions that are performed after a
1730 state change. This function might sleep */
1731
1732 if (ns.susp_nod) {
1733 enum drbd_req_event what = NOTHING;
1734
1735 spin_lock_irq(&device->resource->req_lock);
1736 if (os.conn < C_CONNECTED && conn_lowest_conn(connection) >= C_CONNECTED)
1737 what = RESEND;
1738
1739 if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
1740 conn_lowest_disk(connection) == D_UP_TO_DATE)
1741 what = RESTART_FROZEN_DISK_IO;
1742
1743 if (resource->susp_nod && what != NOTHING) {
1744 _tl_restart(connection, what);
1745 _conn_request_state(connection,
1746 (union drbd_state) { { .susp_nod = 1 } },
1747 (union drbd_state) { { .susp_nod = 0 } },
1748 CS_VERBOSE);
1749 }
1750 spin_unlock_irq(&device->resource->req_lock);
1751 }
1752
1753 if (ns.susp_fen) {
1754 spin_lock_irq(&device->resource->req_lock);
1755 if (resource->susp_fen && conn_lowest_conn(connection) >= C_CONNECTED) {
1756 /* case2: The connection was established again: */
1757 struct drbd_peer_device *peer_device;
1758 int vnr;
1759
1760 rcu_read_lock();
1761 idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
1762 clear_bit(NEW_CUR_UUID, &peer_device->device->flags);
1763 rcu_read_unlock();
1764
1765 /* We should actively create a new uuid, _before_
1766 * we resume/resent, if the peer is diskless
1767 * (recovery from a multiple error scenario).
1768 * Currently, this happens with a slight delay
1769 * below when checking lost_contact_to_peer_data() ...
1770 */
1771 _tl_restart(connection, RESEND);
1772 _conn_request_state(connection,
1773 (union drbd_state) { { .susp_fen = 1 } },
1774 (union drbd_state) { { .susp_fen = 0 } },
1775 CS_VERBOSE);
1776 }
1777 spin_unlock_irq(&device->resource->req_lock);
1778 }
1779
1780 /* Became sync source. With protocol >= 96, we still need to send out
1781 * the sync uuid now. Need to do that before any drbd_send_state, or
1782 * the other side may go "paused sync" before receiving the sync uuids,
1783 * which is unexpected. */
1784 if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
1785 (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
1786 connection->agreed_pro_version >= 96 && get_ldev(device)) {
1787 drbd_gen_and_send_sync_uuid(peer_device);
1788 put_ldev(device);
1789 }
1790
1791 /* Do not change the order of the if above and the two below... */
1792 if (os.pdsk == D_DISKLESS &&
1793 ns.pdsk > D_DISKLESS && ns.pdsk != D_UNKNOWN) { /* attach on the peer */
1794 /* we probably will start a resync soon.
1795 * make sure those things are properly reset. */
1796 device->rs_total = 0;
1797 device->rs_failed = 0;
1798 atomic_set(&device->rs_pending_cnt, 0);
1799 drbd_rs_cancel_all(device);
1800
1801 drbd_send_uuids(peer_device);
1802 drbd_send_state(peer_device, ns);
1803 }
1804 /* No point in queuing send_bitmap if we don't have a connection
1805 * anymore, so check also the _current_ state, not only the new state
1806 * at the time this work was queued. */
1807 if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
1808 device->state.conn == C_WF_BITMAP_S)
1809 drbd_queue_bitmap_io(device, &drbd_send_bitmap, NULL,
1810 "send_bitmap (WFBitMapS)",
1811 BM_LOCKED_TEST_ALLOWED);
1812
1813 /* Lost contact to peer's copy of the data */
1814 if (lost_contact_to_peer_data(os.pdsk, ns.pdsk)) {
1815 if (get_ldev(device)) {
1816 if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1817 device->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1818 if (drbd_suspended(device)) {
1819 set_bit(NEW_CUR_UUID, &device->flags);
1820 } else {
1821 drbd_uuid_new_current(device);
1822 drbd_send_uuids(peer_device);
1823 }
1824 }
1825 put_ldev(device);
1826 }
1827 }
1828
1829 if (ns.pdsk < D_INCONSISTENT && get_ldev(device)) {
1830 if (os.peer != R_PRIMARY && ns.peer == R_PRIMARY &&
1831 device->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1832 drbd_uuid_new_current(device);
1833 drbd_send_uuids(peer_device);
1834 }
1835 /* D_DISKLESS Peer becomes secondary */
1836 if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1837 /* We may still be Primary ourselves.
1838 * No harm done if the bitmap still changes,
1839 * redirtied pages will follow later. */
1840 drbd_bitmap_io_from_worker(device, &drbd_bm_write,
1841 "demote diskless peer", BM_LOCKED_SET_ALLOWED);
1842 put_ldev(device);
1843 }
1844
1845 /* Write out all changed bits on demote.
1846 * Though, no need to da that just yet
1847 * if there is a resync going on still */
1848 if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
1849 device->state.conn <= C_CONNECTED && get_ldev(device)) {
1850 /* No changes to the bitmap expected this time, so assert that,
1851 * even though no harm was done if it did change. */
1852 drbd_bitmap_io_from_worker(device, &drbd_bm_write,
1853 "demote", BM_LOCKED_TEST_ALLOWED);
1854 put_ldev(device);
1855 }
1856
1857 /* Last part of the attaching process ... */
1858 if (ns.conn >= C_CONNECTED &&
1859 os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1860 drbd_send_sizes(peer_device, 0, 0); /* to start sync... */
1861 drbd_send_uuids(peer_device);
1862 drbd_send_state(peer_device, ns);
1863 }
1864
1865 /* We want to pause/continue resync, tell peer. */
1866 if (ns.conn >= C_CONNECTED &&
1867 ((os.aftr_isp != ns.aftr_isp) ||
1868 (os.user_isp != ns.user_isp)))
1869 drbd_send_state(peer_device, ns);
1870
1871 /* In case one of the isp bits got set, suspend other devices. */
1872 if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1873 (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1874 suspend_other_sg(device);
1875
1876 /* Make sure the peer gets informed about eventual state
1877 changes (ISP bits) while we were in WFReportParams. */
1878 if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1879 drbd_send_state(peer_device, ns);
1880
1881 if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
1882 drbd_send_state(peer_device, ns);
1883
1884 /* We are in the progress to start a full sync... */
1885 if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1886 (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1887 /* no other bitmap changes expected during this phase */
1888 drbd_queue_bitmap_io(device,
1889 &drbd_bmio_set_n_write, &abw_start_sync,
1890 "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED);
1891
1892 /* first half of local IO error, failure to attach,
1893 * or administrative detach */
1894 if (os.disk != D_FAILED && ns.disk == D_FAILED) {
1895 enum drbd_io_error_p eh = EP_PASS_ON;
1896 int was_io_error = 0;
1897 /* corresponding get_ldev was in _drbd_set_state, to serialize
1898 * our cleanup here with the transition to D_DISKLESS.
1899 * But is is still not save to dreference ldev here, since
1900 * we might come from an failed Attach before ldev was set. */
1901 if (device->ldev) {
1902 rcu_read_lock();
1903 eh = rcu_dereference(device->ldev->disk_conf)->on_io_error;
1904 rcu_read_unlock();
1905
1906 was_io_error = test_and_clear_bit(WAS_IO_ERROR, &device->flags);
1907
1908 /* Intentionally call this handler first, before drbd_send_state().
1909 * See: 2932204 drbd: call local-io-error handler early
1910 * People may chose to hard-reset the box from this handler.
1911 * It is useful if this looks like a "regular node crash". */
1912 if (was_io_error && eh == EP_CALL_HELPER)
1913 drbd_khelper(device, "local-io-error");
1914
1915 /* Immediately allow completion of all application IO,
1916 * that waits for completion from the local disk,
1917 * if this was a force-detach due to disk_timeout
1918 * or administrator request (drbdsetup detach --force).
1919 * Do NOT abort otherwise.
1920 * Aborting local requests may cause serious problems,
1921 * if requests are completed to upper layers already,
1922 * and then later the already submitted local bio completes.
1923 * This can cause DMA into former bio pages that meanwhile
1924 * have been re-used for other things.
1925 * So aborting local requests may cause crashes,
1926 * or even worse, silent data corruption.
1927 */
1928 if (test_and_clear_bit(FORCE_DETACH, &device->flags))
1929 tl_abort_disk_io(device);
1930
1931 /* current state still has to be D_FAILED,
1932 * there is only one way out: to D_DISKLESS,
1933 * and that may only happen after our put_ldev below. */
1934 if (device->state.disk != D_FAILED)
1935 drbd_err(device,
1936 "ASSERT FAILED: disk is %s during detach\n",
1937 drbd_disk_str(device->state.disk));
1938
1939 if (ns.conn >= C_CONNECTED)
1940 drbd_send_state(peer_device, ns);
1941
1942 drbd_rs_cancel_all(device);
1943
1944 /* In case we want to get something to stable storage still,
1945 * this may be the last chance.
1946 * Following put_ldev may transition to D_DISKLESS. */
1947 drbd_md_sync(device);
1948 }
1949 put_ldev(device);
1950 }
1951
1952 /* second half of local IO error, failure to attach,
1953 * or administrative detach,
1954 * after local_cnt references have reached zero again */
1955 if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
1956 /* We must still be diskless,
1957 * re-attach has to be serialized with this! */
1958 if (device->state.disk != D_DISKLESS)
1959 drbd_err(device,
1960 "ASSERT FAILED: disk is %s while going diskless\n",
1961 drbd_disk_str(device->state.disk));
1962
1963 if (ns.conn >= C_CONNECTED)
1964 drbd_send_state(peer_device, ns);
1965 /* corresponding get_ldev in __drbd_set_state
1966 * this may finally trigger drbd_ldev_destroy. */
1967 put_ldev(device);
1968 }
1969
1970 /* Notify peer that I had a local IO error, and did not detached.. */
1971 if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT && ns.conn >= C_CONNECTED)
1972 drbd_send_state(peer_device, ns);
1973
1974 /* Disks got bigger while they were detached */
1975 if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1976 test_and_clear_bit(RESYNC_AFTER_NEG, &device->flags)) {
1977 if (ns.conn == C_CONNECTED)
1978 resync_after_online_grow(device);
1979 }
1980
1981 /* A resync finished or aborted, wake paused devices... */
1982 if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1983 (os.peer_isp && !ns.peer_isp) ||
1984 (os.user_isp && !ns.user_isp))
1985 resume_next_sg(device);
1986
1987 /* sync target done with resync. Explicitly notify peer, even though
1988 * it should (at least for non-empty resyncs) already know itself. */
1989 if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
1990 drbd_send_state(peer_device, ns);
1991
1992 /* Verify finished, or reached stop sector. Peer did not know about
1993 * the stop sector, and we may even have changed the stop sector during
1994 * verify to interrupt/stop early. Send the new state. */
1995 if (os.conn == C_VERIFY_S && ns.conn == C_CONNECTED
1996 && verify_can_do_stop_sector(device))
1997 drbd_send_state(peer_device, ns);
1998
1999 /* This triggers bitmap writeout of potentially still unwritten pages
2000 * if the resync finished cleanly, or aborted because of peer disk
2001 * failure, or on transition from resync back to AHEAD/BEHIND.
2002 *
2003 * Connection loss is handled in drbd_disconnected() by the receiver.
2004 *
2005 * For resync aborted because of local disk failure, we cannot do
2006 * any bitmap writeout anymore.
2007 *
2008 * No harm done if some bits change during this phase.
2009 */
2010 if ((os.conn > C_CONNECTED && os.conn < C_AHEAD) &&
2011 (ns.conn == C_CONNECTED || ns.conn >= C_AHEAD) && get_ldev(device)) {
2012 drbd_queue_bitmap_io(device, &drbd_bm_write_copy_pages, NULL,
2013 "write from resync_finished", BM_LOCKED_CHANGE_ALLOWED);
2014 put_ldev(device);
2015 }
2016
2017 if (ns.disk == D_DISKLESS &&
2018 ns.conn == C_STANDALONE &&
2019 ns.role == R_SECONDARY) {
2020 if (os.aftr_isp != ns.aftr_isp)
2021 resume_next_sg(device);
2022 }
2023
2024 drbd_md_sync(device);
2025 }
2026
2027 struct after_conn_state_chg_work {
2028 struct drbd_work w;
2029 enum drbd_conns oc;
2030 union drbd_state ns_min;
2031 union drbd_state ns_max; /* new, max state, over all devices */
2032 enum chg_state_flags flags;
2033 struct drbd_connection *connection;
2034 struct drbd_state_change *state_change;
2035 };
2036
w_after_conn_state_ch(struct drbd_work * w,int unused)2037 static int w_after_conn_state_ch(struct drbd_work *w, int unused)
2038 {
2039 struct after_conn_state_chg_work *acscw =
2040 container_of(w, struct after_conn_state_chg_work, w);
2041 struct drbd_connection *connection = acscw->connection;
2042 enum drbd_conns oc = acscw->oc;
2043 union drbd_state ns_max = acscw->ns_max;
2044 struct drbd_peer_device *peer_device;
2045 int vnr;
2046
2047 broadcast_state_change(acscw->state_change);
2048 forget_state_change(acscw->state_change);
2049 kfree(acscw);
2050
2051 /* Upon network configuration, we need to start the receiver */
2052 if (oc == C_STANDALONE && ns_max.conn == C_UNCONNECTED)
2053 drbd_thread_start(&connection->receiver);
2054
2055 if (oc == C_DISCONNECTING && ns_max.conn == C_STANDALONE) {
2056 struct net_conf *old_conf;
2057
2058 mutex_lock(¬ification_mutex);
2059 idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2060 notify_peer_device_state(NULL, 0, peer_device, NULL,
2061 NOTIFY_DESTROY | NOTIFY_CONTINUES);
2062 notify_connection_state(NULL, 0, connection, NULL, NOTIFY_DESTROY);
2063 mutex_unlock(¬ification_mutex);
2064
2065 mutex_lock(&connection->resource->conf_update);
2066 old_conf = connection->net_conf;
2067 connection->my_addr_len = 0;
2068 connection->peer_addr_len = 0;
2069 RCU_INIT_POINTER(connection->net_conf, NULL);
2070 conn_free_crypto(connection);
2071 mutex_unlock(&connection->resource->conf_update);
2072
2073 synchronize_rcu();
2074 kfree(old_conf);
2075 }
2076
2077 if (ns_max.susp_fen) {
2078 /* case1: The outdate peer handler is successful: */
2079 if (ns_max.pdsk <= D_OUTDATED) {
2080 rcu_read_lock();
2081 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2082 struct drbd_device *device = peer_device->device;
2083 if (test_bit(NEW_CUR_UUID, &device->flags)) {
2084 drbd_uuid_new_current(device);
2085 clear_bit(NEW_CUR_UUID, &device->flags);
2086 }
2087 }
2088 rcu_read_unlock();
2089 spin_lock_irq(&connection->resource->req_lock);
2090 _tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
2091 _conn_request_state(connection,
2092 (union drbd_state) { { .susp_fen = 1 } },
2093 (union drbd_state) { { .susp_fen = 0 } },
2094 CS_VERBOSE);
2095 spin_unlock_irq(&connection->resource->req_lock);
2096 }
2097 }
2098 conn_md_sync(connection);
2099 kref_put(&connection->kref, drbd_destroy_connection);
2100
2101 return 0;
2102 }
2103
conn_old_common_state(struct drbd_connection * connection,union drbd_state * pcs,enum chg_state_flags * pf)2104 static void conn_old_common_state(struct drbd_connection *connection, union drbd_state *pcs, enum chg_state_flags *pf)
2105 {
2106 enum chg_state_flags flags = ~0;
2107 struct drbd_peer_device *peer_device;
2108 int vnr, first_vol = 1;
2109 union drbd_dev_state os, cs = {
2110 { .role = R_SECONDARY,
2111 .peer = R_UNKNOWN,
2112 .conn = connection->cstate,
2113 .disk = D_DISKLESS,
2114 .pdsk = D_UNKNOWN,
2115 } };
2116
2117 rcu_read_lock();
2118 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2119 struct drbd_device *device = peer_device->device;
2120 os = device->state;
2121
2122 if (first_vol) {
2123 cs = os;
2124 first_vol = 0;
2125 continue;
2126 }
2127
2128 if (cs.role != os.role)
2129 flags &= ~CS_DC_ROLE;
2130
2131 if (cs.peer != os.peer)
2132 flags &= ~CS_DC_PEER;
2133
2134 if (cs.conn != os.conn)
2135 flags &= ~CS_DC_CONN;
2136
2137 if (cs.disk != os.disk)
2138 flags &= ~CS_DC_DISK;
2139
2140 if (cs.pdsk != os.pdsk)
2141 flags &= ~CS_DC_PDSK;
2142 }
2143 rcu_read_unlock();
2144
2145 *pf |= CS_DC_MASK;
2146 *pf &= flags;
2147 (*pcs).i = cs.i;
2148 }
2149
2150 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)2151 conn_is_valid_transition(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2152 enum chg_state_flags flags)
2153 {
2154 enum drbd_state_rv rv = SS_SUCCESS;
2155 union drbd_state ns, os;
2156 struct drbd_peer_device *peer_device;
2157 int vnr;
2158
2159 rcu_read_lock();
2160 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2161 struct drbd_device *device = peer_device->device;
2162 os = drbd_read_state(device);
2163 ns = sanitize_state(device, os, apply_mask_val(os, mask, val), NULL);
2164
2165 if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
2166 ns.disk = os.disk;
2167
2168 if (ns.i == os.i)
2169 continue;
2170
2171 rv = is_valid_transition(os, ns);
2172
2173 if (rv >= SS_SUCCESS && !(flags & CS_HARD)) {
2174 rv = is_valid_state(device, ns);
2175 if (rv < SS_SUCCESS) {
2176 if (is_valid_state(device, os) == rv)
2177 rv = is_valid_soft_transition(os, ns, connection);
2178 } else
2179 rv = is_valid_soft_transition(os, ns, connection);
2180 }
2181
2182 if (rv < SS_SUCCESS) {
2183 if (flags & CS_VERBOSE)
2184 print_st_err(device, os, ns, rv);
2185 break;
2186 }
2187 }
2188 rcu_read_unlock();
2189
2190 return rv;
2191 }
2192
2193 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)2194 conn_set_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2195 union drbd_state *pns_min, union drbd_state *pns_max, enum chg_state_flags flags)
2196 {
2197 union drbd_state ns, os, ns_max = { };
2198 union drbd_state ns_min = {
2199 { .role = R_MASK,
2200 .peer = R_MASK,
2201 .conn = val.conn,
2202 .disk = D_MASK,
2203 .pdsk = D_MASK
2204 } };
2205 struct drbd_peer_device *peer_device;
2206 enum drbd_state_rv rv;
2207 int vnr, number_of_volumes = 0;
2208
2209 if (mask.conn == C_MASK) {
2210 /* remember last connect time so request_timer_fn() won't
2211 * kill newly established sessions while we are still trying to thaw
2212 * previously frozen IO */
2213 if (connection->cstate != C_WF_REPORT_PARAMS && val.conn == C_WF_REPORT_PARAMS)
2214 connection->last_reconnect_jif = jiffies;
2215
2216 connection->cstate = val.conn;
2217 }
2218
2219 rcu_read_lock();
2220 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2221 struct drbd_device *device = peer_device->device;
2222 number_of_volumes++;
2223 os = drbd_read_state(device);
2224 ns = apply_mask_val(os, mask, val);
2225 ns = sanitize_state(device, os, ns, NULL);
2226
2227 if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED)
2228 ns.disk = os.disk;
2229
2230 rv = _drbd_set_state(device, ns, flags, NULL);
2231 BUG_ON(rv < SS_SUCCESS);
2232 ns.i = device->state.i;
2233 ns_max.role = max_role(ns.role, ns_max.role);
2234 ns_max.peer = max_role(ns.peer, ns_max.peer);
2235 ns_max.conn = max_t(enum drbd_conns, ns.conn, ns_max.conn);
2236 ns_max.disk = max_t(enum drbd_disk_state, ns.disk, ns_max.disk);
2237 ns_max.pdsk = max_t(enum drbd_disk_state, ns.pdsk, ns_max.pdsk);
2238
2239 ns_min.role = min_role(ns.role, ns_min.role);
2240 ns_min.peer = min_role(ns.peer, ns_min.peer);
2241 ns_min.conn = min_t(enum drbd_conns, ns.conn, ns_min.conn);
2242 ns_min.disk = min_t(enum drbd_disk_state, ns.disk, ns_min.disk);
2243 ns_min.pdsk = min_t(enum drbd_disk_state, ns.pdsk, ns_min.pdsk);
2244 }
2245 rcu_read_unlock();
2246
2247 if (number_of_volumes == 0) {
2248 ns_min = ns_max = (union drbd_state) { {
2249 .role = R_SECONDARY,
2250 .peer = R_UNKNOWN,
2251 .conn = val.conn,
2252 .disk = D_DISKLESS,
2253 .pdsk = D_UNKNOWN
2254 } };
2255 }
2256
2257 ns_min.susp = ns_max.susp = connection->resource->susp;
2258 ns_min.susp_nod = ns_max.susp_nod = connection->resource->susp_nod;
2259 ns_min.susp_fen = ns_max.susp_fen = connection->resource->susp_fen;
2260
2261 *pns_min = ns_min;
2262 *pns_max = ns_max;
2263 }
2264
2265 static enum drbd_state_rv
_conn_rq_cond(struct drbd_connection * connection,union drbd_state mask,union drbd_state val)2266 _conn_rq_cond(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
2267 {
2268 enum drbd_state_rv err, rv = SS_UNKNOWN_ERROR; /* continue waiting */;
2269
2270 if (test_and_clear_bit(CONN_WD_ST_CHG_OKAY, &connection->flags))
2271 rv = SS_CW_SUCCESS;
2272
2273 if (test_and_clear_bit(CONN_WD_ST_CHG_FAIL, &connection->flags))
2274 rv = SS_CW_FAILED_BY_PEER;
2275
2276 err = conn_is_valid_transition(connection, mask, val, 0);
2277 if (err == SS_SUCCESS && connection->cstate == C_WF_REPORT_PARAMS)
2278 return rv;
2279
2280 return err;
2281 }
2282
2283 enum drbd_state_rv
_conn_request_state(struct drbd_connection * connection,union drbd_state mask,union drbd_state val,enum chg_state_flags flags)2284 _conn_request_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2285 enum chg_state_flags flags)
2286 {
2287 enum drbd_state_rv rv = SS_SUCCESS;
2288 struct after_conn_state_chg_work *acscw;
2289 enum drbd_conns oc = connection->cstate;
2290 union drbd_state ns_max, ns_min, os;
2291 bool have_mutex = false;
2292 struct drbd_state_change *state_change;
2293
2294 if (mask.conn) {
2295 rv = is_valid_conn_transition(oc, val.conn);
2296 if (rv < SS_SUCCESS)
2297 goto abort;
2298 }
2299
2300 rv = conn_is_valid_transition(connection, mask, val, flags);
2301 if (rv < SS_SUCCESS)
2302 goto abort;
2303
2304 if (oc == C_WF_REPORT_PARAMS && val.conn == C_DISCONNECTING &&
2305 !(flags & (CS_LOCAL_ONLY | CS_HARD))) {
2306
2307 /* This will be a cluster-wide state change.
2308 * Need to give up the spinlock, grab the mutex,
2309 * then send the state change request, ... */
2310 spin_unlock_irq(&connection->resource->req_lock);
2311 mutex_lock(&connection->cstate_mutex);
2312 have_mutex = true;
2313
2314 set_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
2315 if (conn_send_state_req(connection, mask, val)) {
2316 /* sending failed. */
2317 clear_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
2318 rv = SS_CW_FAILED_BY_PEER;
2319 /* need to re-aquire the spin lock, though */
2320 goto abort_unlocked;
2321 }
2322
2323 if (val.conn == C_DISCONNECTING)
2324 set_bit(DISCONNECT_SENT, &connection->flags);
2325
2326 /* ... and re-aquire the spinlock.
2327 * If _conn_rq_cond() returned >= SS_SUCCESS, we must call
2328 * conn_set_state() within the same spinlock. */
2329 spin_lock_irq(&connection->resource->req_lock);
2330 wait_event_lock_irq(connection->ping_wait,
2331 (rv = _conn_rq_cond(connection, mask, val)),
2332 connection->resource->req_lock);
2333 clear_bit(CONN_WD_ST_CHG_REQ, &connection->flags);
2334 if (rv < SS_SUCCESS)
2335 goto abort;
2336 }
2337
2338 state_change = remember_old_state(connection->resource, GFP_ATOMIC);
2339 conn_old_common_state(connection, &os, &flags);
2340 flags |= CS_DC_SUSP;
2341 conn_set_state(connection, mask, val, &ns_min, &ns_max, flags);
2342 conn_pr_state_change(connection, os, ns_max, flags);
2343 remember_new_state(state_change);
2344
2345 acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC);
2346 if (acscw) {
2347 acscw->oc = os.conn;
2348 acscw->ns_min = ns_min;
2349 acscw->ns_max = ns_max;
2350 acscw->flags = flags;
2351 acscw->w.cb = w_after_conn_state_ch;
2352 kref_get(&connection->kref);
2353 acscw->connection = connection;
2354 acscw->state_change = state_change;
2355 drbd_queue_work(&connection->sender_work, &acscw->w);
2356 } else {
2357 drbd_err(connection, "Could not kmalloc an acscw\n");
2358 }
2359
2360 abort:
2361 if (have_mutex) {
2362 /* mutex_unlock() "... must not be used in interrupt context.",
2363 * so give up the spinlock, then re-aquire it */
2364 spin_unlock_irq(&connection->resource->req_lock);
2365 abort_unlocked:
2366 mutex_unlock(&connection->cstate_mutex);
2367 spin_lock_irq(&connection->resource->req_lock);
2368 }
2369 if (rv < SS_SUCCESS && flags & CS_VERBOSE) {
2370 drbd_err(connection, "State change failed: %s\n", drbd_set_st_err_str(rv));
2371 drbd_err(connection, " mask = 0x%x val = 0x%x\n", mask.i, val.i);
2372 drbd_err(connection, " old_conn:%s wanted_conn:%s\n", drbd_conn_str(oc), drbd_conn_str(val.conn));
2373 }
2374 return rv;
2375 }
2376
2377 enum drbd_state_rv
conn_request_state(struct drbd_connection * connection,union drbd_state mask,union drbd_state val,enum chg_state_flags flags)2378 conn_request_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val,
2379 enum chg_state_flags flags)
2380 {
2381 enum drbd_state_rv rv;
2382
2383 spin_lock_irq(&connection->resource->req_lock);
2384 rv = _conn_request_state(connection, mask, val, flags);
2385 spin_unlock_irq(&connection->resource->req_lock);
2386
2387 return rv;
2388 }
2389