projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
drbd: Converted the transfer log from mdev to tconn
[linux-block.git] / drivers / block / drbd / drbd_state.c
CommitLineData
b8907339
PR		 1/*
2 drbd_state.c
3
4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
5
6 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
7 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
9
10 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
11 from Logicworks, Inc. for making SDP replication support possible.
12
13 drbd is free software; you can redistribute it and/or modify
14 it under the terms of the GNU General Public License as published by
15 the Free Software Foundation; either version 2, or (at your option)
16 any later version.
17
18 drbd is distributed in the hope that it will be useful,
19 but WITHOUT ANY WARRANTY; without even the implied warranty of
20 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 GNU General Public License for more details.
22
23 You should have received a copy of the GNU General Public License
24 along with drbd; see the file COPYING. If not, write to
25 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
26 */
27
28#include <linux/drbd_limits.h>
29#include "drbd_int.h"
30#include "drbd_req.h"
31
32struct after_state_chg_work {
33 struct drbd_work w;
34 union drbd_state os;
35 union drbd_state ns;
36 enum chg_state_flags flags;
37 struct completion *done;
38};
39
00d56944 40static int w_after_state_ch(struct drbd_work *w, int unused);
b8907339
PR		 41static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
42 union drbd_state ns, enum chg_state_flags flags);
0e29d163 43static void after_all_state_ch(struct drbd_tconn *tconn);
a75f34ad
PR		 44static enum drbd_state_rv is_valid_state(struct drbd_conf *, union drbd_state);
45static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state);
3509502d 46static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns);
4308a0a3
PR		 47static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns,
48 const char **warn_sync_abort);
b8907339 49
0e29d163
PR		 50int conn_all_vols_unconf(struct drbd_tconn *tconn)
51{
52 struct drbd_conf *mdev;
53 int minor, uncfg = 1;
54
55 idr_for_each_entry(&tconn->volumes, mdev, minor) {
56 uncfg &= (mdev->state.disk == D_DISKLESS &&
57 mdev->state.conn == C_STANDALONE &&
58 mdev->state.role == R_SECONDARY);
59 if (!uncfg)
60 break;
61 }
62
63 return uncfg;
64}
65
b8907339
PR		 66/**
67 * cl_wide_st_chg() - true if the state change is a cluster wide one
68 * @mdev: DRBD device.
69 * @os: old (current) state.
70 * @ns: new (wanted) state.
71 */
72static int cl_wide_st_chg(struct drbd_conf *mdev,
73 union drbd_state os, union drbd_state ns)
74{
75 return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED &&
76 ((os.role != R_PRIMARY && ns.role == R_PRIMARY) ||
77 (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
78 (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) ||
79 (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))) ||
80 (os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) ||
81 (os.conn == C_CONNECTED && ns.conn == C_VERIFY_S);
82}
83
56707f9e
PR		 84static union drbd_state
85apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val)
86{
87 union drbd_state ns;
88 ns.i = (os.i & ~mask.i) | val.i;
89 return ns;
90}
91
b8907339
PR		 92enum drbd_state_rv
93drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f,
94 union drbd_state mask, union drbd_state val)
95{
96 unsigned long flags;
56707f9e 97 union drbd_state ns;
b8907339
PR		 98 enum drbd_state_rv rv;
99
100 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
56707f9e 101 ns = apply_mask_val(mdev->state, mask, val);
b8907339
PR		 102 rv = _drbd_set_state(mdev, ns, f, NULL);
103 ns = mdev->state;
104 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
105
106 return rv;
107}
108
109/**
110 * drbd_force_state() - Impose a change which happens outside our control on our state
111 * @mdev: DRBD device.
112 * @mask: mask of state bits to change.
113 * @val: value of new state bits.
114 */
115void drbd_force_state(struct drbd_conf *mdev,
116 union drbd_state mask, union drbd_state val)
117{
118 drbd_change_state(mdev, CS_HARD, mask, val);
119}
120
121static enum drbd_state_rv
122_req_st_cond(struct drbd_conf *mdev, union drbd_state mask,
123 union drbd_state val)
124{
125 union drbd_state os, ns;
126 unsigned long flags;
127 enum drbd_state_rv rv;
128
129 if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &mdev->flags))
130 return SS_CW_SUCCESS;
131
132 if (test_and_clear_bit(CL_ST_CHG_FAIL, &mdev->flags))
133 return SS_CW_FAILED_BY_PEER;
134
b8907339
PR		 135 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
136 os = mdev->state;
56707f9e 137 ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
3509502d
PR		 138 rv = is_valid_transition(os, ns);
139 if (rv == SS_SUCCESS)
140 rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
b8907339
PR		 141
142 if (!cl_wide_st_chg(mdev, os, ns))
143 rv = SS_CW_NO_NEED;
3509502d 144 if (rv == SS_UNKNOWN_ERROR) {
b8907339
PR		 145 rv = is_valid_state(mdev, ns);
146 if (rv == SS_SUCCESS) {
a75f34ad 147 rv = is_valid_soft_transition(os, ns);
b8907339
PR		 148 if (rv == SS_SUCCESS)
149 rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
150 }
151 }
152 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
153
154 return rv;
155}
156
157/**
158 * drbd_req_state() - Perform an eventually cluster wide state change
159 * @mdev: DRBD device.
160 * @mask: mask of state bits to change.
161 * @val: value of new state bits.
162 * @f: flags
163 *
164 * Should not be called directly, use drbd_request_state() or
165 * _drbd_request_state().
166 */
167static enum drbd_state_rv
168drbd_req_state(struct drbd_conf *mdev, union drbd_state mask,
169 union drbd_state val, enum chg_state_flags f)
170{
171 struct completion done;
172 unsigned long flags;
173 union drbd_state os, ns;
174 enum drbd_state_rv rv;
175
176 init_completion(&done);
177
178 if (f & CS_SERIALIZE)
8410da8f 179 mutex_lock(mdev->state_mutex);
b8907339
PR		 180
181 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
182 os = mdev->state;
56707f9e 183 ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
3509502d
PR		 184 rv = is_valid_transition(os, ns);
185 if (rv < SS_SUCCESS)
186 goto abort;
b8907339
PR		 187
188 if (cl_wide_st_chg(mdev, os, ns)) {
189 rv = is_valid_state(mdev, ns);
190 if (rv == SS_SUCCESS)
a75f34ad 191 rv = is_valid_soft_transition(os, ns);
b8907339
PR		 192 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
193
194 if (rv < SS_SUCCESS) {
195 if (f & CS_VERBOSE)
196 print_st_err(mdev, os, ns, rv);
197 goto abort;
198 }
199
b8907339 200 if (!drbd_send_state_req(mdev, mask, val)) {
b8907339
PR		 201 rv = SS_CW_FAILED_BY_PEER;
202 if (f & CS_VERBOSE)
203 print_st_err(mdev, os, ns, rv);
204 goto abort;
205 }
206
207 wait_event(mdev->state_wait,
208 (rv = _req_st_cond(mdev, mask, val)));
209
210 if (rv < SS_SUCCESS) {
b8907339
PR		 211 if (f & CS_VERBOSE)
212 print_st_err(mdev, os, ns, rv);
213 goto abort;
214 }
215 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
56707f9e 216 ns = apply_mask_val(mdev->state, mask, val);
b8907339 217 rv = _drbd_set_state(mdev, ns, f, &done);
b8907339
PR		 218 } else {
219 rv = _drbd_set_state(mdev, ns, f, &done);
220 }
221
222 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
223
224 if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) {
225 D_ASSERT(current != mdev->tconn->worker.task);
226 wait_for_completion(&done);
227 }
228
229abort:
230 if (f & CS_SERIALIZE)
8410da8f 231 mutex_unlock(mdev->state_mutex);
b8907339
PR		 232
233 return rv;
234}
235
236/**
237 * _drbd_request_state() - Request a state change (with flags)
238 * @mdev: DRBD device.
239 * @mask: mask of state bits to change.
240 * @val: value of new state bits.
241 * @f: flags
242 *
243 * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE
244 * flag, or when logging of failed state change requests is not desired.
245 */
246enum drbd_state_rv
247_drbd_request_state(struct drbd_conf *mdev, union drbd_state mask,
248 union drbd_state val, enum chg_state_flags f)
249{
250 enum drbd_state_rv rv;
251
252 wait_event(mdev->state_wait,
253 (rv = drbd_req_state(mdev, mask, val, f)) != SS_IN_TRANSIENT_STATE);
254
255 return rv;
256}
257
258static void print_st(struct drbd_conf *mdev, char *name, union drbd_state ns)
259{
260 dev_err(DEV, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n",
261 name,
262 drbd_conn_str(ns.conn),
263 drbd_role_str(ns.role),
264 drbd_role_str(ns.peer),
265 drbd_disk_str(ns.disk),
266 drbd_disk_str(ns.pdsk),
267 is_susp(ns) ? 's' : 'r',
268 ns.aftr_isp ? 'a' : '-',
269 ns.peer_isp ? 'p' : '-',
270 ns.user_isp ? 'u' : '-',
271 ns.susp_fen ? 'F' : '-',
272 ns.susp_nod ? 'N' : '-'
273 );
274}
275
276void print_st_err(struct drbd_conf *mdev, union drbd_state os,
277 union drbd_state ns, enum drbd_state_rv err)
278{
279 if (err == SS_IN_TRANSIENT_STATE)
280 return;
281 dev_err(DEV, "State change failed: %s\n", drbd_set_st_err_str(err));
282 print_st(mdev, " state", os);
283 print_st(mdev, "wanted", ns);
284}
285
bbeb641c
PR		 286static void print_state_change(struct drbd_conf *mdev, union drbd_state os, union drbd_state ns,
287 enum chg_state_flags flags)
288{
289 char *pbp, pb[300];
290 pbp = pb;
291 *pbp = 0;
292 if (ns.role != os.role)
293 pbp += sprintf(pbp, "role( %s -> %s ) ",
294 drbd_role_str(os.role),
295 drbd_role_str(ns.role));
296 if (ns.peer != os.peer)
297 pbp += sprintf(pbp, "peer( %s -> %s ) ",
298 drbd_role_str(os.peer),
299 drbd_role_str(ns.peer));
300 if (ns.conn != os.conn && !(flags & CS_NO_CSTATE_CHG))
301 pbp += sprintf(pbp, "conn( %s -> %s ) ",
302 drbd_conn_str(os.conn),
303 drbd_conn_str(ns.conn));
304 if (ns.disk != os.disk)
305 pbp += sprintf(pbp, "disk( %s -> %s ) ",
306 drbd_disk_str(os.disk),
307 drbd_disk_str(ns.disk));
308 if (ns.pdsk != os.pdsk)
309 pbp += sprintf(pbp, "pdsk( %s -> %s ) ",
310 drbd_disk_str(os.pdsk),
311 drbd_disk_str(ns.pdsk));
312 if (is_susp(ns) != is_susp(os))
313 pbp += sprintf(pbp, "susp( %d -> %d ) ",
314 is_susp(os),
315 is_susp(ns));
316 if (ns.aftr_isp != os.aftr_isp)
317 pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ",
318 os.aftr_isp,
319 ns.aftr_isp);
320 if (ns.peer_isp != os.peer_isp)
321 pbp += sprintf(pbp, "peer_isp( %d -> %d ) ",
322 os.peer_isp,
323 ns.peer_isp);
324 if (ns.user_isp != os.user_isp)
325 pbp += sprintf(pbp, "user_isp( %d -> %d ) ",
326 os.user_isp,
327 ns.user_isp);
328 if (pbp != pb)
329 dev_info(DEV, "%s\n", pb);
330}
b8907339
PR		 331
332/**
333 * is_valid_state() - Returns an SS_ error code if ns is not valid
334 * @mdev: DRBD device.
335 * @ns: State to consider.
336 */
337static enum drbd_state_rv
338is_valid_state(struct drbd_conf *mdev, union drbd_state ns)
339{
340 /* See drbd_state_sw_errors in drbd_strings.c */
341
342 enum drbd_fencing_p fp;
343 enum drbd_state_rv rv = SS_SUCCESS;
344
345 fp = FP_DONT_CARE;
346 if (get_ldev(mdev)) {
347 fp = mdev->ldev->dc.fencing;
348 put_ldev(mdev);
349 }
350
351 if (get_net_conf(mdev->tconn)) {
352 if (!mdev->tconn->net_conf->two_primaries &&
353 ns.role == R_PRIMARY && ns.peer == R_PRIMARY)
354 rv = SS_TWO_PRIMARIES;
355 put_net_conf(mdev->tconn);
356 }
357
358 if (rv <= 0)
359 /* already found a reason to abort */;
360 else if (ns.role == R_SECONDARY && mdev->open_cnt)
361 rv = SS_DEVICE_IN_USE;
362
363 else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE)
364 rv = SS_NO_UP_TO_DATE_DISK;
365
366 else if (fp >= FP_RESOURCE &&
367 ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN)
368 rv = SS_PRIMARY_NOP;
369
370 else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT)
371 rv = SS_NO_UP_TO_DATE_DISK;
372
373 else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT)
374 rv = SS_NO_LOCAL_DISK;
375
376 else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT)
377 rv = SS_NO_REMOTE_DISK;
378
379 else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
380 rv = SS_NO_UP_TO_DATE_DISK;
381
382 else if ((ns.conn == C_CONNECTED ||
383 ns.conn == C_WF_BITMAP_S ||
384 ns.conn == C_SYNC_SOURCE ||
385 ns.conn == C_PAUSED_SYNC_S) &&
386 ns.disk == D_OUTDATED)
387 rv = SS_CONNECTED_OUTDATES;
388
389 else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
390 (mdev->sync_conf.verify_alg[0] == 0))
391 rv = SS_NO_VERIFY_ALG;
392
393 else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
394 mdev->tconn->agreed_pro_version < 88)
395 rv = SS_NOT_SUPPORTED;
396
397 else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN)
398 rv = SS_CONNECTED_OUTDATES;
399
400 return rv;
401}
402
403/**
a75f34ad 404 * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible
3509502d
PR		 405 * This function limits state transitions that may be declined by DRBD. I.e.
406 * user requests (aka soft transitions).
b8907339
PR		 407 * @mdev: DRBD device.
408 * @ns: new state.
409 * @os: old state.
410 */
411static enum drbd_state_rv
a75f34ad 412is_valid_soft_transition(union drbd_state os, union drbd_state ns)
b8907339
PR		 413{
414 enum drbd_state_rv rv = SS_SUCCESS;
415
416 if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
417 os.conn > C_CONNECTED)
418 rv = SS_RESYNC_RUNNING;
419
420 if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
421 rv = SS_ALREADY_STANDALONE;
422
423 if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
424 rv = SS_IS_DISKLESS;
425
426 if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
427 rv = SS_NO_NET_CONFIG;
428
429 if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
430 rv = SS_LOWER_THAN_OUTDATED;
431
432 if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
433 rv = SS_IN_TRANSIENT_STATE;
434
435 if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS)
436 rv = SS_IN_TRANSIENT_STATE;
437
438 if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
439 rv = SS_NEED_CONNECTION;
440
441 if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
442 ns.conn != os.conn && os.conn > C_CONNECTED)
443 rv = SS_RESYNC_RUNNING;
444
445 if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
446 os.conn < C_CONNECTED)
447 rv = SS_NEED_CONNECTION;
448
449 if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)
450 && os.conn < C_WF_REPORT_PARAMS)
451 rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */
452
453 return rv;
454}
455
3509502d 456static enum drbd_state_rv
fda74117 457is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc)
3509502d
PR		 458{
459 enum drbd_state_rv rv = SS_SUCCESS;
460
461 /* Disallow Network errors to configure a device's network part */
fda74117 462 if ((nc >= C_TIMEOUT && nc <= C_TEAR_DOWN) && oc <= C_DISCONNECTING)
3509502d
PR		 463 rv = SS_NEED_CONNECTION;
464
465 /* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */
fda74117 466 if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING)
3509502d
PR		 467 rv = SS_IN_TRANSIENT_STATE;
468
469 /* After C_DISCONNECTING only C_STANDALONE may follow */
fda74117 470 if (oc == C_DISCONNECTING && nc != C_STANDALONE)
3509502d
PR		 471 rv = SS_IN_TRANSIENT_STATE;
472
fda74117
PR		 473 return rv;
474}
475
476
477/**
478 * is_valid_transition() - Returns an SS_ error code if the state transition is not possible
479 * This limits hard state transitions. Hard state transitions are facts there are
480 * imposed on DRBD by the environment. E.g. disk broke or network broke down.
481 * But those hard state transitions are still not allowed to do everything.
482 * @ns: new state.
483 * @os: old state.
484 */
485static enum drbd_state_rv
486is_valid_transition(union drbd_state os, union drbd_state ns)
487{
488 enum drbd_state_rv rv;
489
490 rv = is_valid_conn_transition(os.conn, ns.conn);
491
3509502d
PR		 492 /* we cannot fail (again) if we already detached */
493 if (ns.disk == D_FAILED && os.disk == D_DISKLESS)
494 rv = SS_IS_DISKLESS;
495
4308a0a3
PR		 496 /* if we are only D_ATTACHING yet,
497 * we can (and should) go directly to D_DISKLESS. */
498 if (ns.disk == D_FAILED && os.disk == D_ATTACHING) {
499 printk("TODO: FIX ME\n");
500 rv = SS_IS_DISKLESS;
501 }
502
3509502d
PR		 503 return rv;
504}
505
b8907339
PR		 506/**
507 * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
508 * @mdev: DRBD device.
509 * @os: old state.
510 * @ns: new state.
511 * @warn_sync_abort:
512 *
513 * When we loose connection, we have to set the state of the peers disk (pdsk)
514 * to D_UNKNOWN. This rule and many more along those lines are in this function.
515 */
4308a0a3
PR		 516static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns,
517 const char **warn_sync_abort)
b8907339
PR		 518{
519 enum drbd_fencing_p fp;
520 enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max;
521
522 fp = FP_DONT_CARE;
523 if (get_ldev(mdev)) {
524 fp = mdev->ldev->dc.fencing;
525 put_ldev(mdev);
526 }
527
3509502d 528 /* Implications from connection to peer and peer_isp */
b8907339
PR		 529 if (ns.conn < C_CONNECTED) {
530 ns.peer_isp = 0;
531 ns.peer = R_UNKNOWN;
532 if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
533 ns.pdsk = D_UNKNOWN;
534 }
535
536 /* Clear the aftr_isp when becoming unconfigured */
537 if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
538 ns.aftr_isp = 0;
539
4308a0a3 540 /* An implication of the disk states onto the connection state */
b8907339 541 /* Abort resync if a disk fails/detaches */
4308a0a3 542 if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
b8907339
PR		 543 if (warn_sync_abort)
544 *warn_sync_abort =
4308a0a3 545 ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ?
b8907339
PR		 546 "Online-verify" : "Resync";
547 ns.conn = C_CONNECTED;
548 }
549
550 /* Connection breaks down before we finished "Negotiating" */
551 if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
552 get_ldev_if_state(mdev, D_NEGOTIATING)) {
553 if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) {
554 ns.disk = mdev->new_state_tmp.disk;
555 ns.pdsk = mdev->new_state_tmp.pdsk;
556 } else {
557 dev_alert(DEV, "Connection lost while negotiating, no data!\n");
558 ns.disk = D_DISKLESS;
559 ns.pdsk = D_UNKNOWN;
560 }
561 put_ldev(mdev);
562 }
563
564 /* D_CONSISTENT and D_OUTDATED vanish when we get connected */
565 if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) {
566 if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED)
567 ns.disk = D_UP_TO_DATE;
568 if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)
569 ns.pdsk = D_UP_TO_DATE;
570 }
571
572 /* Implications of the connection stat on the disk states */
573 disk_min = D_DISKLESS;
574 disk_max = D_UP_TO_DATE;
575 pdsk_min = D_INCONSISTENT;
576 pdsk_max = D_UNKNOWN;
577 switch ((enum drbd_conns)ns.conn) {
578 case C_WF_BITMAP_T:
579 case C_PAUSED_SYNC_T:
580 case C_STARTING_SYNC_T:
581 case C_WF_SYNC_UUID:
582 case C_BEHIND:
583 disk_min = D_INCONSISTENT;
584 disk_max = D_OUTDATED;
585 pdsk_min = D_UP_TO_DATE;
586 pdsk_max = D_UP_TO_DATE;
587 break;
588 case C_VERIFY_S:
589 case C_VERIFY_T:
590 disk_min = D_UP_TO_DATE;
591 disk_max = D_UP_TO_DATE;
592 pdsk_min = D_UP_TO_DATE;
593 pdsk_max = D_UP_TO_DATE;
594 break;
595 case C_CONNECTED:
596 disk_min = D_DISKLESS;
597 disk_max = D_UP_TO_DATE;
598 pdsk_min = D_DISKLESS;
599 pdsk_max = D_UP_TO_DATE;
600 break;
601 case C_WF_BITMAP_S:
602 case C_PAUSED_SYNC_S:
603 case C_STARTING_SYNC_S:
604 case C_AHEAD:
605 disk_min = D_UP_TO_DATE;
606 disk_max = D_UP_TO_DATE;
607 pdsk_min = D_INCONSISTENT;
608 pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/
609 break;
610 case C_SYNC_TARGET:
611 disk_min = D_INCONSISTENT;
612 disk_max = D_INCONSISTENT;
613 pdsk_min = D_UP_TO_DATE;
614 pdsk_max = D_UP_TO_DATE;
615 break;
616 case C_SYNC_SOURCE:
617 disk_min = D_UP_TO_DATE;
618 disk_max = D_UP_TO_DATE;
619 pdsk_min = D_INCONSISTENT;
620 pdsk_max = D_INCONSISTENT;
621 break;
622 case C_STANDALONE:
623 case C_DISCONNECTING:
624 case C_UNCONNECTED:
625 case C_TIMEOUT:
626 case C_BROKEN_PIPE:
627 case C_NETWORK_FAILURE:
628 case C_PROTOCOL_ERROR:
629 case C_TEAR_DOWN:
630 case C_WF_CONNECTION:
631 case C_WF_REPORT_PARAMS:
632 case C_MASK:
633 break;
634 }
635 if (ns.disk > disk_max)
636 ns.disk = disk_max;
637
638 if (ns.disk < disk_min) {
639 dev_warn(DEV, "Implicitly set disk from %s to %s\n",
640 drbd_disk_str(ns.disk), drbd_disk_str(disk_min));
641 ns.disk = disk_min;
642 }
643 if (ns.pdsk > pdsk_max)
644 ns.pdsk = pdsk_max;
645
646 if (ns.pdsk < pdsk_min) {
647 dev_warn(DEV, "Implicitly set pdsk from %s to %s\n",
648 drbd_disk_str(ns.pdsk), drbd_disk_str(pdsk_min));
649 ns.pdsk = pdsk_min;
650 }
651
652 if (fp == FP_STONITH &&
4308a0a3 653 (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED))
b8907339
PR		 654 ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */
655
656 if (mdev->sync_conf.on_no_data == OND_SUSPEND_IO &&
4308a0a3 657 (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
b8907339
PR		 658 ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */
659
660 if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
661 if (ns.conn == C_SYNC_SOURCE)
662 ns.conn = C_PAUSED_SYNC_S;
663 if (ns.conn == C_SYNC_TARGET)
664 ns.conn = C_PAUSED_SYNC_T;
665 } else {
666 if (ns.conn == C_PAUSED_SYNC_S)
667 ns.conn = C_SYNC_SOURCE;
668 if (ns.conn == C_PAUSED_SYNC_T)
669 ns.conn = C_SYNC_TARGET;
670 }
671
672 return ns;
673}
674
675void drbd_resume_al(struct drbd_conf *mdev)
676{
677 if (test_and_clear_bit(AL_SUSPENDED, &mdev->flags))
678 dev_info(DEV, "Resumed AL updates\n");
679}
680
681/* helper for __drbd_set_state */
682static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs)
683{
684 if (mdev->tconn->agreed_pro_version < 90)
685 mdev->ov_start_sector = 0;
686 mdev->rs_total = drbd_bm_bits(mdev);
687 mdev->ov_position = 0;
688 if (cs == C_VERIFY_T) {
689 /* starting online verify from an arbitrary position
690 * does not fit well into the existing protocol.
691 * on C_VERIFY_T, we initialize ov_left and friends
692 * implicitly in receive_DataRequest once the
693 * first P_OV_REQUEST is received */
694 mdev->ov_start_sector = ~(sector_t)0;
695 } else {
696 unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector);
697 if (bit >= mdev->rs_total) {
698 mdev->ov_start_sector =
699 BM_BIT_TO_SECT(mdev->rs_total - 1);
700 mdev->rs_total = 1;
701 } else
702 mdev->rs_total -= bit;
703 mdev->ov_position = mdev->ov_start_sector;
704 }
705 mdev->ov_left = mdev->rs_total;
706}
707
708/**
709 * __drbd_set_state() - Set a new DRBD state
710 * @mdev: DRBD device.
711 * @ns: new state.
712 * @flags: Flags
713 * @done: Optional completion, that will get completed after the after_state_ch() finished
714 *
715 * Caller needs to hold req_lock, and global_state_lock. Do not call directly.
716 */
717enum drbd_state_rv
718__drbd_set_state(struct drbd_conf *mdev, union drbd_state ns,
719 enum chg_state_flags flags, struct completion *done)
720{
721 union drbd_state os;
722 enum drbd_state_rv rv = SS_SUCCESS;
723 const char *warn_sync_abort = NULL;
724 struct after_state_chg_work *ascw;
725
726 os = mdev->state;
727
4308a0a3 728 ns = sanitize_state(mdev, ns, &warn_sync_abort);
b8907339
PR		 729 if (ns.i == os.i)
730 return SS_NOTHING_TO_DO;
731
3509502d
PR		 732 rv = is_valid_transition(os, ns);
733 if (rv < SS_SUCCESS)
734 return rv;
735
b8907339
PR		 736 if (!(flags & CS_HARD)) {
737 /* pre-state-change checks ; only look at ns */
738 /* See drbd_state_sw_errors in drbd_strings.c */
739
740 rv = is_valid_state(mdev, ns);
741 if (rv < SS_SUCCESS) {
742 /* If the old state was illegal as well, then let
743 this happen...*/
744
745 if (is_valid_state(mdev, os) == rv)
a75f34ad 746 rv = is_valid_soft_transition(os, ns);
b8907339 747 } else
a75f34ad 748 rv = is_valid_soft_transition(os, ns);
b8907339
PR		 749 }
750
751 if (rv < SS_SUCCESS) {
752 if (flags & CS_VERBOSE)
753 print_st_err(mdev, os, ns, rv);
754 return rv;
755 }
756
757 if (warn_sync_abort)
758 dev_warn(DEV, "%s aborted.\n", warn_sync_abort);
759
bbeb641c 760 print_state_change(mdev, os, ns, flags);
b8907339 761
b8907339
PR		 762 /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
763 * on the ldev here, to be sure the transition -> D_DISKLESS resp.
764 * drbd_ldev_destroy() won't happen before our corresponding
765 * after_state_ch works run, where we put_ldev again. */
766 if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
767 (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
768 atomic_inc(&mdev->local_cnt);
769
770 mdev->state = ns;
771
0e29d163
PR		 772 /* solve the race between becoming unconfigured,
773 * worker doing the cleanup, and
774 * admin reconfiguring us:
775 * on (re)configure, first set CONFIG_PENDING,
776 * then wait for a potentially exiting worker,
777 * start the worker, and schedule one no_op.
778 * then proceed with configuration.
779 */
780 if(conn_all_vols_unconf(mdev->tconn) &&
781 !test_and_set_bit(CONFIG_PENDING, &mdev->tconn->flags))
782 set_bit(OBJECT_DYING, &mdev->tconn->flags);
783
b8907339
PR		 784 if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
785 drbd_print_uuids(mdev, "attached to UUIDs");
786
787 wake_up(&mdev->misc_wait);
788 wake_up(&mdev->state_wait);
2a67d8b9 789 wake_up(&mdev->tconn->ping_wait);
b8907339
PR		 790
791 /* aborted verify run. log the last position */
792 if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
793 ns.conn < C_CONNECTED) {
794 mdev->ov_start_sector =
795 BM_BIT_TO_SECT(drbd_bm_bits(mdev) - mdev->ov_left);
796 dev_info(DEV, "Online Verify reached sector %llu\n",
797 (unsigned long long)mdev->ov_start_sector);
798 }
799
800 if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
801 (ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)) {
802 dev_info(DEV, "Syncer continues.\n");
803 mdev->rs_paused += (long)jiffies
804 -(long)mdev->rs_mark_time[mdev->rs_last_mark];
805 if (ns.conn == C_SYNC_TARGET)
806 mod_timer(&mdev->resync_timer, jiffies);
807 }
808
809 if ((os.conn == C_SYNC_TARGET || os.conn == C_SYNC_SOURCE) &&
810 (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
811 dev_info(DEV, "Resync suspended\n");
812 mdev->rs_mark_time[mdev->rs_last_mark] = jiffies;
813 }
814
815 if (os.conn == C_CONNECTED &&
816 (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
817 unsigned long now = jiffies;
818 int i;
819
820 set_ov_position(mdev, ns.conn);
821 mdev->rs_start = now;
822 mdev->rs_last_events = 0;
823 mdev->rs_last_sect_ev = 0;
824 mdev->ov_last_oos_size = 0;
825 mdev->ov_last_oos_start = 0;
826
827 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
828 mdev->rs_mark_left[i] = mdev->ov_left;
829 mdev->rs_mark_time[i] = now;
830 }
831
832 drbd_rs_controller_reset(mdev);
833
834 if (ns.conn == C_VERIFY_S) {
835 dev_info(DEV, "Starting Online Verify from sector %llu\n",
836 (unsigned long long)mdev->ov_position);
837 mod_timer(&mdev->resync_timer, jiffies);
838 }
839 }
840
841 if (get_ldev(mdev)) {
842 u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
843 MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
844 MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
845
846 if (test_bit(CRASHED_PRIMARY, &mdev->flags))
847 mdf |= MDF_CRASHED_PRIMARY;
848 if (mdev->state.role == R_PRIMARY ||
849 (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY))
850 mdf |= MDF_PRIMARY_IND;
851 if (mdev->state.conn > C_WF_REPORT_PARAMS)
852 mdf |= MDF_CONNECTED_IND;
853 if (mdev->state.disk > D_INCONSISTENT)
854 mdf |= MDF_CONSISTENT;
855 if (mdev->state.disk > D_OUTDATED)
856 mdf |= MDF_WAS_UP_TO_DATE;
857 if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT)
858 mdf |= MDF_PEER_OUT_DATED;
859 if (mdf != mdev->ldev->md.flags) {
860 mdev->ldev->md.flags = mdf;
861 drbd_md_mark_dirty(mdev);
862 }
863 if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
864 drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]);
865 put_ldev(mdev);
866 }
867
868 /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
869 if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
870 os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
871 set_bit(CONSIDER_RESYNC, &mdev->flags);
872
873 /* Receiver should clean up itself */
874 if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
875 drbd_thread_stop_nowait(&mdev->tconn->receiver);
876
877 /* Now the receiver finished cleaning up itself, it should die */
878 if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
879 drbd_thread_stop_nowait(&mdev->tconn->receiver);
880
881 /* Upon network failure, we need to restart the receiver. */
882 if (os.conn > C_TEAR_DOWN &&
883 ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
884 drbd_thread_restart_nowait(&mdev->tconn->receiver);
885
886 /* Resume AL writing if we get a connection */
887 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
888 drbd_resume_al(mdev);
889
890 ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
891 if (ascw) {
892 ascw->os = os;
893 ascw->ns = ns;
894 ascw->flags = flags;
895 ascw->w.cb = w_after_state_ch;
a21e9298 896 ascw->w.mdev = mdev;
b8907339
PR		 897 ascw->done = done;
898 drbd_queue_work(&mdev->tconn->data.work, &ascw->w);
899 } else {
bbeb641c 900 dev_err(DEV, "Could not kmalloc an ascw\n");
b8907339
PR		 901 }
902
903 return rv;
904}
905
00d56944 906static int w_after_state_ch(struct drbd_work *w, int unused)
b8907339
PR		 907{
908 struct after_state_chg_work *ascw =
909 container_of(w, struct after_state_chg_work, w);
00d56944 910 struct drbd_conf *mdev = w->mdev;
b8907339
PR		 911
912 after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags);
913 if (ascw->flags & CS_WAIT_COMPLETE) {
914 D_ASSERT(ascw->done != NULL);
915 complete(ascw->done);
916 }
917 kfree(ascw);
918
919 return 1;
920}
921
922static void abw_start_sync(struct drbd_conf *mdev, int rv)
923{
924 if (rv) {
925 dev_err(DEV, "Writing the bitmap failed not starting resync.\n");
926 _drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE);
927 return;
928 }
929
930 switch (mdev->state.conn) {
931 case C_STARTING_SYNC_T:
932 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
933 break;
934 case C_STARTING_SYNC_S:
935 drbd_start_resync(mdev, C_SYNC_SOURCE);
936 break;
937 }
938}
939
940int drbd_bitmap_io_from_worker(struct drbd_conf *mdev,
941 int (*io_fn)(struct drbd_conf *),
942 char *why, enum bm_flag flags)
943{
944 int rv;
945
946 D_ASSERT(current == mdev->tconn->worker.task);
947
948 /* open coded non-blocking drbd_suspend_io(mdev); */
949 set_bit(SUSPEND_IO, &mdev->flags);
950
951 drbd_bm_lock(mdev, why, flags);
952 rv = io_fn(mdev);
953 drbd_bm_unlock(mdev);
954
955 drbd_resume_io(mdev);
956
957 return rv;
958}
959
960/**
961 * after_state_ch() - Perform after state change actions that may sleep
962 * @mdev: DRBD device.
963 * @os: old state.
964 * @ns: new state.
965 * @flags: Flags
966 */
967static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
968 union drbd_state ns, enum chg_state_flags flags)
969{
970 enum drbd_fencing_p fp;
971 enum drbd_req_event what = NOTHING;
972 union drbd_state nsm = (union drbd_state){ .i = -1 };
973
974 if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) {
975 clear_bit(CRASHED_PRIMARY, &mdev->flags);
976 if (mdev->p_uuid)
977 mdev->p_uuid[UI_FLAGS] &= ~((u64)2);
978 }
979
980 fp = FP_DONT_CARE;
981 if (get_ldev(mdev)) {
982 fp = mdev->ldev->dc.fencing;
983 put_ldev(mdev);
984 }
985
986 /* Inform userspace about the change... */
987 drbd_bcast_state(mdev, ns);
988
989 if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
990 (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
991 drbd_khelper(mdev, "pri-on-incon-degr");
992
993 /* Here we have the actions that are performed after a
994 state change. This function might sleep */
995
996 nsm.i = -1;
997 if (ns.susp_nod) {
998 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
999 what = RESEND;
1000
1001 if (os.disk == D_ATTACHING && ns.disk > D_ATTACHING)
1002 what = RESTART_FROZEN_DISK_IO;
1003
1004 if (what != NOTHING)
1005 nsm.susp_nod = 0;
1006 }
1007
1008 if (ns.susp_fen) {
1009 /* case1: The outdate peer handler is successful: */
1010 if (os.pdsk > D_OUTDATED && ns.pdsk <= D_OUTDATED) {
2f5cdd0b 1011 tl_clear(mdev->tconn);
b8907339
PR		 1012 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1013 drbd_uuid_new_current(mdev);
1014 clear_bit(NEW_CUR_UUID, &mdev->flags);
1015 }
1016 spin_lock_irq(&mdev->tconn->req_lock);
1017 _drbd_set_state(_NS(mdev, susp_fen, 0), CS_VERBOSE, NULL);
1018 spin_unlock_irq(&mdev->tconn->req_lock);
1019 }
1020 /* case2: The connection was established again: */
1021 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) {
1022 clear_bit(NEW_CUR_UUID, &mdev->flags);
1023 what = RESEND;
1024 nsm.susp_fen = 0;
1025 }
1026 }
1027
1028 if (what != NOTHING) {
1029 spin_lock_irq(&mdev->tconn->req_lock);
2f5cdd0b 1030 _tl_restart(mdev->tconn, what);
b8907339
PR		 1031 nsm.i &= mdev->state.i;
1032 _drbd_set_state(mdev, nsm, CS_VERBOSE, NULL);
1033 spin_unlock_irq(&mdev->tconn->req_lock);
1034 }
1035
1036 /* Became sync source. With protocol >= 96, we still need to send out
1037 * the sync uuid now. Need to do that before any drbd_send_state, or
1038 * the other side may go "paused sync" before receiving the sync uuids,
1039 * which is unexpected. */
1040 if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
1041 (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
1042 mdev->tconn->agreed_pro_version >= 96 && get_ldev(mdev)) {
1043 drbd_gen_and_send_sync_uuid(mdev);
1044 put_ldev(mdev);
1045 }
1046
1047 /* Do not change the order of the if above and the two below... */
1048 if (os.pdsk == D_DISKLESS && ns.pdsk > D_DISKLESS) { /* attach on the peer */
1049 drbd_send_uuids(mdev);
1050 drbd_send_state(mdev);
1051 }
1052 /* No point in queuing send_bitmap if we don't have a connection
1053 * anymore, so check also the _current_ state, not only the new state
1054 * at the time this work was queued. */
1055 if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
1056 mdev->state.conn == C_WF_BITMAP_S)
1057 drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL,
1058 "send_bitmap (WFBitMapS)",
1059 BM_LOCKED_TEST_ALLOWED);
1060
1061 /* Lost contact to peer's copy of the data */
1062 if ((os.pdsk >= D_INCONSISTENT &&
1063 os.pdsk != D_UNKNOWN &&
1064 os.pdsk != D_OUTDATED)
1065 && (ns.pdsk < D_INCONSISTENT ||
1066 ns.pdsk == D_UNKNOWN ||
1067 ns.pdsk == D_OUTDATED)) {
1068 if (get_ldev(mdev)) {
1069 if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1070 mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1071 if (is_susp(mdev->state)) {
1072 set_bit(NEW_CUR_UUID, &mdev->flags);
1073 } else {
1074 drbd_uuid_new_current(mdev);
1075 drbd_send_uuids(mdev);
1076 }
1077 }
1078 put_ldev(mdev);
1079 }
1080 }
1081
1082 if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) {
1083 if (ns.peer == R_PRIMARY && mdev->ldev->md.uuid[UI_BITMAP] == 0) {
1084 drbd_uuid_new_current(mdev);
1085 drbd_send_uuids(mdev);
1086 }
1087
1088 /* D_DISKLESS Peer becomes secondary */
1089 if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1090 /* We may still be Primary ourselves.
1091 * No harm done if the bitmap still changes,
1092 * redirtied pages will follow later. */
1093 drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1094 "demote diskless peer", BM_LOCKED_SET_ALLOWED);
1095 put_ldev(mdev);
1096 }
1097
1098 /* Write out all changed bits on demote.
1099 * Though, no need to da that just yet
1100 * if there is a resync going on still */
1101 if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
1102 mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) {
1103 /* No changes to the bitmap expected this time, so assert that,
1104 * even though no harm was done if it did change. */
1105 drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1106 "demote", BM_LOCKED_TEST_ALLOWED);
1107 put_ldev(mdev);
1108 }
1109
1110 /* Last part of the attaching process ... */
1111 if (ns.conn >= C_CONNECTED &&
1112 os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1113 drbd_send_sizes(mdev, 0, 0); /* to start sync... */
1114 drbd_send_uuids(mdev);
1115 drbd_send_state(mdev);
1116 }
1117
1118 /* We want to pause/continue resync, tell peer. */
1119 if (ns.conn >= C_CONNECTED &&
1120 ((os.aftr_isp != ns.aftr_isp) ||
1121 (os.user_isp != ns.user_isp)))
1122 drbd_send_state(mdev);
1123
1124 /* In case one of the isp bits got set, suspend other devices. */
1125 if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1126 (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1127 suspend_other_sg(mdev);
1128
1129 /* Make sure the peer gets informed about eventual state
1130 changes (ISP bits) while we were in WFReportParams. */
1131 if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1132 drbd_send_state(mdev);
1133
1134 if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
1135 drbd_send_state(mdev);
1136
1137 /* We are in the progress to start a full sync... */
1138 if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1139 (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1140 /* no other bitmap changes expected during this phase */
1141 drbd_queue_bitmap_io(mdev,
1142 &drbd_bmio_set_n_write, &abw_start_sync,
1143 "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED);
1144
1145 /* We are invalidating our self... */
1146 if (os.conn < C_CONNECTED && ns.conn < C_CONNECTED &&
1147 os.disk > D_INCONSISTENT && ns.disk == D_INCONSISTENT)
1148 /* other bitmap operation expected during this phase */
1149 drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, NULL,
1150 "set_n_write from invalidate", BM_LOCKED_MASK);
1151
1152 /* first half of local IO error, failure to attach,
1153 * or administrative detach */
1154 if (os.disk != D_FAILED && ns.disk == D_FAILED) {
1155 enum drbd_io_error_p eh;
1156 int was_io_error;
1157 /* corresponding get_ldev was in __drbd_set_state, to serialize
1158 * our cleanup here with the transition to D_DISKLESS,
1159 * so it is safe to dreference ldev here. */
1160 eh = mdev->ldev->dc.on_io_error;
1161 was_io_error = test_and_clear_bit(WAS_IO_ERROR, &mdev->flags);
1162
1163 /* current state still has to be D_FAILED,
1164 * there is only one way out: to D_DISKLESS,
1165 * and that may only happen after our put_ldev below. */
1166 if (mdev->state.disk != D_FAILED)
1167 dev_err(DEV,
1168 "ASSERT FAILED: disk is %s during detach\n",
1169 drbd_disk_str(mdev->state.disk));
1170
1171 if (drbd_send_state(mdev))
1172 dev_warn(DEV, "Notified peer that I am detaching my disk\n");
1173 else
1174 dev_err(DEV, "Sending state for detaching disk failed\n");
1175
1176 drbd_rs_cancel_all(mdev);
1177
1178 /* In case we want to get something to stable storage still,
1179 * this may be the last chance.
1180 * Following put_ldev may transition to D_DISKLESS. */
1181 drbd_md_sync(mdev);
1182 put_ldev(mdev);
1183
1184 if (was_io_error && eh == EP_CALL_HELPER)
1185 drbd_khelper(mdev, "local-io-error");
1186 }
1187
1188 /* second half of local IO error, failure to attach,
1189 * or administrative detach,
1190 * after local_cnt references have reached zero again */
1191 if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
1192 /* We must still be diskless,
1193 * re-attach has to be serialized with this! */
1194 if (mdev->state.disk != D_DISKLESS)
1195 dev_err(DEV,
1196 "ASSERT FAILED: disk is %s while going diskless\n",
1197 drbd_disk_str(mdev->state.disk));
1198
1199 mdev->rs_total = 0;
1200 mdev->rs_failed = 0;
1201 atomic_set(&mdev->rs_pending_cnt, 0);
1202
1203 if (drbd_send_state(mdev))
1204 dev_warn(DEV, "Notified peer that I'm now diskless.\n");
1205 /* corresponding get_ldev in __drbd_set_state
1206 * this may finally trigger drbd_ldev_destroy. */
1207 put_ldev(mdev);
1208 }
1209
1210 /* Notify peer that I had a local IO error, and did not detached.. */
1211 if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT)
1212 drbd_send_state(mdev);
1213
1214 /* Disks got bigger while they were detached */
1215 if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1216 test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) {
1217 if (ns.conn == C_CONNECTED)
1218 resync_after_online_grow(mdev);
1219 }
1220
1221 /* A resync finished or aborted, wake paused devices... */
1222 if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1223 (os.peer_isp && !ns.peer_isp) ||
1224 (os.user_isp && !ns.user_isp))
1225 resume_next_sg(mdev);
1226
1227 /* sync target done with resync. Explicitly notify peer, even though
1228 * it should (at least for non-empty resyncs) already know itself. */
1229 if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
1230 drbd_send_state(mdev);
1231
1232 /* This triggers bitmap writeout of potentially still unwritten pages
1233 * if the resync finished cleanly, or aborted because of peer disk
1234 * failure, or because of connection loss.
1235 * For resync aborted because of local disk failure, we cannot do
1236 * any bitmap writeout anymore.
1237 * No harm done if some bits change during this phase.
1238 */
1239 if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) {
1240 drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL,
1241 "write from resync_finished", BM_LOCKED_SET_ALLOWED);
1242 put_ldev(mdev);
1243 }
1244
1245 if (ns.disk == D_DISKLESS &&
1246 ns.conn == C_STANDALONE &&
1247 ns.role == R_SECONDARY) {
1248 if (os.aftr_isp != ns.aftr_isp)
1249 resume_next_sg(mdev);
1250 }
1251
0e29d163 1252 after_all_state_ch(mdev->tconn);
bbeb641c 1253
b8907339
PR		 1254 drbd_md_sync(mdev);
1255}
1256
bbeb641c
PR		 1257struct after_conn_state_chg_work {
1258 struct drbd_work w;
1259 enum drbd_conns oc;
1260 union drbd_state nms; /* new, max state, over all mdevs */
1261 enum chg_state_flags flags;
1262};
1263
0e29d163 1264static void after_all_state_ch(struct drbd_tconn *tconn)
b8907339 1265{
0e29d163
PR		 1266 if (conn_all_vols_unconf(tconn) &&
1267 test_bit(OBJECT_DYING, &tconn->flags)) {
bbeb641c
PR		 1268 drbd_thread_stop_nowait(&tconn->worker);
1269 }
1270}
1271
1272static int w_after_conn_state_ch(struct drbd_work *w, int unused)
1273{
1274 struct after_conn_state_chg_work *acscw =
1275 container_of(w, struct after_conn_state_chg_work, w);
1276 struct drbd_tconn *tconn = w->tconn;
1277 enum drbd_conns oc = acscw->oc;
1278 union drbd_state nms = acscw->nms;
1279
1280 kfree(acscw);
1281
b8907339 1282 /* Upon network configuration, we need to start the receiver */
bbeb641c 1283 if (oc == C_STANDALONE && nms.conn == C_UNCONNECTED)
b8907339
PR		 1284 drbd_thread_start(&tconn->receiver);
1285
bbeb641c 1286 //conn_err(tconn, STATE_FMT, STATE_ARGS("nms", nms));
0e29d163 1287 after_all_state_ch(tconn);
bbeb641c
PR		 1288
1289 return 1;
1290}
1291
1292static void print_conn_state_change(struct drbd_tconn *tconn, enum drbd_conns oc, enum drbd_conns nc)
1293{
1294 char *pbp, pb[300];
1295 pbp = pb;
1296 *pbp = 0;
1297 if (nc != oc)
1298 pbp += sprintf(pbp, "conn( %s -> %s ) ",
1299 drbd_conn_str(oc),
1300 drbd_conn_str(nc));
1301
1302 conn_info(tconn, "%s\n", pb);
1303}
1304
1305struct _is_valid_itr_params {
1306 enum chg_state_flags flags;
1307 union drbd_state mask, val;
1308 union drbd_state ms; /* maximal state, over all mdevs */
1309 enum drbd_conns oc;
1310 enum {
1311 OC_UNINITIALIZED,
1312 OC_CONSISTENT,
1313 OC_INCONSISTENT,
1314 } oc_state;
1315};
1316
1317static int _is_valid_itr_fn(int vnr, void *p, void *data)
1318{
1319 struct drbd_conf *mdev = (struct drbd_conf *)p;
1320 struct _is_valid_itr_params *params = (struct _is_valid_itr_params *)data;
1321 enum chg_state_flags flags = params->flags;
1322 union drbd_state ns, os;
1323 enum drbd_state_rv rv;
1324
1325 os = mdev->state;
1326 ns = apply_mask_val(os, params->mask, params->val);
1327 ns = sanitize_state(mdev, ns, NULL);
1328 rv = is_valid_state(mdev, ns);
1329
1330 if (rv < SS_SUCCESS) {
1331 /* If the old state was illegal as well, then let this happen...*/
1332
1333 if (is_valid_state(mdev, os) == rv)
1334 rv = is_valid_soft_transition(os, ns);
1335 } else
1336 rv = is_valid_soft_transition(os, ns);
1337
1338 switch (params->oc_state) {
1339 case OC_UNINITIALIZED:
1340 params->oc = os.conn;
1341 params->oc_state = OC_CONSISTENT;
1342 break;
1343 case OC_CONSISTENT:
1344 if (params->oc != os.conn)
1345 params->oc_state = OC_INCONSISTENT;
1346 break;
1347 case OC_INCONSISTENT:
1348 break;
1349 }
1350
1351 if (rv < SS_SUCCESS) {
1352 if (flags & CS_VERBOSE)
1353 print_st_err(mdev, os, ns, rv);
1354 return rv;
1355 } else
1356 return 0;
1357}
1358
1359static int _set_state_itr_fn(int vnr, void *p, void *data)
1360{
1361 struct drbd_conf *mdev = (struct drbd_conf *)p;
1362 struct _is_valid_itr_params *params = (struct _is_valid_itr_params *)data;
1363 enum chg_state_flags flags = params->flags;
1364 union drbd_state os, ns, ms = params->ms;
1365 enum drbd_state_rv rv;
1366
1367 os = mdev->state;
1368 ns = apply_mask_val(os, params->mask, params->val);
1369 ns = sanitize_state(mdev, ns, NULL);
1370
1371 rv = __drbd_set_state(mdev, ns, flags, NULL);
1372
1373 ms.role = max_t(enum drbd_role, mdev->state.role, ms.role);
1374 ms.peer = max_t(enum drbd_role, mdev->state.peer, ms.peer);
1375 ms.disk = max_t(enum drbd_role, mdev->state.disk, ms.disk);
1376 ms.pdsk = max_t(enum drbd_role, mdev->state.pdsk, ms.pdsk);
1377 params->ms = ms;
1378
1379 return 0;
1380}
1381
df24aa45
PR		 1382static enum drbd_state_rv
1383_conn_rq_cond(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val)
1384{
1385 struct _is_valid_itr_params params;
1386 enum drbd_state_rv rv;
1387
1388 if (test_and_clear_bit(CONN_WD_ST_CHG_OKAY, &tconn->flags))
1389 return SS_CW_SUCCESS;
1390
1391 if (test_and_clear_bit(CONN_WD_ST_CHG_FAIL, &tconn->flags))
1392 return SS_CW_FAILED_BY_PEER;
1393
1394 params.flags = CS_NO_CSTATE_CHG; /* ΓΆΓΆ think */
1395 params.mask = mask;
1396 params.val = val;
1397
1398 spin_lock_irq(&tconn->req_lock);
1399 rv = tconn->cstate != C_WF_REPORT_PARAMS ? SS_CW_NO_NEED : SS_UNKNOWN_ERROR;
1400
1401 if (rv == SS_UNKNOWN_ERROR)
1402 rv = idr_for_each(&tconn->volumes, _is_valid_itr_fn, &params);
1403
1404 if (rv == 0) /* idr_for_each semantics */
1405 rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
1406
1407 spin_unlock_irq(&tconn->req_lock);
1408
1409 return rv;
1410}
1411
1412static enum drbd_state_rv
1413conn_cl_wide(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1414 enum chg_state_flags f)
1415{
1416 enum drbd_state_rv rv;
1417
1418 spin_unlock_irq(&tconn->req_lock);
1419 mutex_lock(&tconn->cstate_mutex);
1420
1421 if (!conn_send_state_req(tconn, mask, val)) {
1422 rv = SS_CW_FAILED_BY_PEER;
1423 /* if (f & CS_VERBOSE)
1424 print_st_err(mdev, os, ns, rv); */
1425 goto abort;
1426 }
1427
1428 wait_event(tconn->ping_wait, (rv = _conn_rq_cond(tconn, mask, val)));
1429
1430abort:
1431 mutex_unlock(&tconn->cstate_mutex);
1432 spin_lock_irq(&tconn->req_lock);
1433
1434 return rv;
1435}
1436
bbeb641c
PR		 1437enum drbd_state_rv
1438_conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1439 enum chg_state_flags flags)
1440{
1441 enum drbd_state_rv rv = SS_SUCCESS;
1442 struct _is_valid_itr_params params;
1443 struct after_conn_state_chg_work *acscw;
1444 enum drbd_conns oc = tconn->cstate;
1445
bbeb641c
PR		 1446 rv = is_valid_conn_transition(oc, val.conn);
1447 if (rv < SS_SUCCESS)
1448 goto abort;
1449
1450 params.flags = flags;
1451 params.mask = mask;
1452 params.val = val;
1453 params.oc_state = OC_UNINITIALIZED;
1454
1455 if (!(flags & CS_HARD))
1456 rv = idr_for_each(&tconn->volumes, _is_valid_itr_fn, &params);
1457
1458 if (rv == 0) /* idr_for_each semantics */
1459 rv = SS_SUCCESS;
1460
1461 if (rv < SS_SUCCESS)
1462 goto abort;
1463
df24aa45
PR		 1464 if (oc == C_WF_REPORT_PARAMS && val.conn == C_DISCONNECTING &&
1465 !(flags & (CS_LOCAL_ONLY | CS_HARD))) {
1466 rv = conn_cl_wide(tconn, mask, val, flags);
1467 if (rv < SS_SUCCESS)
1468 goto abort;
1469 }
1470
bbeb641c
PR		 1471 if (params.oc_state == OC_CONSISTENT) {
1472 oc = params.oc;
1473 print_conn_state_change(tconn, oc, val.conn);
1474 params.flags |= CS_NO_CSTATE_CHG;
1475 }
1476 tconn->cstate = val.conn;
1477 params.ms.i = 0;
1478 params.ms.conn = val.conn;
1479 idr_for_each(&tconn->volumes, _set_state_itr_fn, &params);
1480
1481 acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC);
1482 if (acscw) {
1483 acscw->oc = oc;
1484 acscw->nms = params.ms;
1485 acscw->flags = flags;
1486 acscw->w.cb = w_after_conn_state_ch;
1487 acscw->w.tconn = tconn;
1488 drbd_queue_work(&tconn->data.work, &acscw->w);
1489 } else {
1490 conn_err(tconn, "Could not kmalloc an acscw\n");
b8907339 1491 }
bbeb641c
PR		 1492
1493abort:
bbeb641c
PR		 1494 return rv;
1495}
1496
1497enum drbd_state_rv
1498conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1499 enum chg_state_flags flags)
1500{
1501 enum drbd_state_rv rv;
1502
1503 spin_lock_irq(&tconn->req_lock);
1504 rv = _conn_request_state(tconn, mask, val, flags);
1505 spin_unlock_irq(&tconn->req_lock);
1506
1507 return rv;
b8907339 1508}
Linux 6.x block layer and io_uring tree(s)