projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
drbd: Killed volume0; last step of multi-volume-enablement
[deliverable/linux.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
40
41extern void _tl_restart(struct drbd_conf *mdev, enum drbd_req_event what);
42int drbd_send_state_req(struct drbd_conf *, union drbd_state, union drbd_state);
00d56944 43static int w_after_state_ch(struct drbd_work *w, int unused);
b8907339
PR		 44static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
45 union drbd_state ns, enum chg_state_flags flags);
bbeb641c 46static void after_all_state_ch(struct drbd_tconn *tconn, union drbd_state ns);
a75f34ad
PR		 47static enum drbd_state_rv is_valid_state(struct drbd_conf *, union drbd_state);
48static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state);
3509502d 49static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns);
4308a0a3
PR		 50static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns,
51 const char **warn_sync_abort);
b8907339
PR		 52
53/**
54 * cl_wide_st_chg() - true if the state change is a cluster wide one
55 * @mdev: DRBD device.
56 * @os: old (current) state.
57 * @ns: new (wanted) state.
58 */
59static int cl_wide_st_chg(struct drbd_conf *mdev,
60 union drbd_state os, union drbd_state ns)
61{
62 return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED &&
63 ((os.role != R_PRIMARY && ns.role == R_PRIMARY) ||
64 (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
65 (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) ||
66 (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))) ||
67 (os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) ||
68 (os.conn == C_CONNECTED && ns.conn == C_VERIFY_S);
69}
70
56707f9e
PR		 71static union drbd_state
72apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val)
73{
74 union drbd_state ns;
75 ns.i = (os.i & ~mask.i) | val.i;
76 return ns;
77}
78
b8907339
PR		 79enum drbd_state_rv
80drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f,
81 union drbd_state mask, union drbd_state val)
82{
83 unsigned long flags;
56707f9e 84 union drbd_state ns;
b8907339
PR		 85 enum drbd_state_rv rv;
86
87 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
56707f9e 88 ns = apply_mask_val(mdev->state, mask, val);
b8907339
PR		 89 rv = _drbd_set_state(mdev, ns, f, NULL);
90 ns = mdev->state;
91 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
92
93 return rv;
94}
95
96/**
97 * drbd_force_state() - Impose a change which happens outside our control on our state
98 * @mdev: DRBD device.
99 * @mask: mask of state bits to change.
100 * @val: value of new state bits.
101 */
102void drbd_force_state(struct drbd_conf *mdev,
103 union drbd_state mask, union drbd_state val)
104{
105 drbd_change_state(mdev, CS_HARD, mask, val);
106}
107
108static enum drbd_state_rv
109_req_st_cond(struct drbd_conf *mdev, union drbd_state mask,
110 union drbd_state val)
111{
112 union drbd_state os, ns;
113 unsigned long flags;
114 enum drbd_state_rv rv;
115
116 if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &mdev->flags))
117 return SS_CW_SUCCESS;
118
119 if (test_and_clear_bit(CL_ST_CHG_FAIL, &mdev->flags))
120 return SS_CW_FAILED_BY_PEER;
121
b8907339
PR		 122 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
123 os = mdev->state;
56707f9e 124 ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
3509502d
PR		 125 rv = is_valid_transition(os, ns);
126 if (rv == SS_SUCCESS)
127 rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
b8907339
PR		 128
129 if (!cl_wide_st_chg(mdev, os, ns))
130 rv = SS_CW_NO_NEED;
3509502d 131 if (rv == SS_UNKNOWN_ERROR) {
b8907339
PR		 132 rv = is_valid_state(mdev, ns);
133 if (rv == SS_SUCCESS) {
a75f34ad 134 rv = is_valid_soft_transition(os, ns);
b8907339
PR		 135 if (rv == SS_SUCCESS)
136 rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
137 }
138 }
139 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
140
141 return rv;
142}
143
144/**
145 * drbd_req_state() - Perform an eventually cluster wide state change
146 * @mdev: DRBD device.
147 * @mask: mask of state bits to change.
148 * @val: value of new state bits.
149 * @f: flags
150 *
151 * Should not be called directly, use drbd_request_state() or
152 * _drbd_request_state().
153 */
154static enum drbd_state_rv
155drbd_req_state(struct drbd_conf *mdev, union drbd_state mask,
156 union drbd_state val, enum chg_state_flags f)
157{
158 struct completion done;
159 unsigned long flags;
160 union drbd_state os, ns;
161 enum drbd_state_rv rv;
162
163 init_completion(&done);
164
165 if (f & CS_SERIALIZE)
166 mutex_lock(&mdev->state_mutex);
167
168 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
169 os = mdev->state;
56707f9e 170 ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
3509502d
PR		 171 rv = is_valid_transition(os, ns);
172 if (rv < SS_SUCCESS)
173 goto abort;
b8907339
PR		 174
175 if (cl_wide_st_chg(mdev, os, ns)) {
176 rv = is_valid_state(mdev, ns);
177 if (rv == SS_SUCCESS)
a75f34ad 178 rv = is_valid_soft_transition(os, ns);
b8907339
PR		 179 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
180
181 if (rv < SS_SUCCESS) {
182 if (f & CS_VERBOSE)
183 print_st_err(mdev, os, ns, rv);
184 goto abort;
185 }
186
187 drbd_state_lock(mdev);
188 if (!drbd_send_state_req(mdev, mask, val)) {
189 drbd_state_unlock(mdev);
190 rv = SS_CW_FAILED_BY_PEER;
191 if (f & CS_VERBOSE)
192 print_st_err(mdev, os, ns, rv);
193 goto abort;
194 }
195
196 wait_event(mdev->state_wait,
197 (rv = _req_st_cond(mdev, mask, val)));
198
199 if (rv < SS_SUCCESS) {
200 drbd_state_unlock(mdev);
201 if (f & CS_VERBOSE)
202 print_st_err(mdev, os, ns, rv);
203 goto abort;
204 }
205 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
56707f9e 206 ns = apply_mask_val(mdev->state, mask, val);
b8907339
PR		 207 rv = _drbd_set_state(mdev, ns, f, &done);
208 drbd_state_unlock(mdev);
209 } else {
210 rv = _drbd_set_state(mdev, ns, f, &done);
211 }
212
213 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
214
215 if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) {
216 D_ASSERT(current != mdev->tconn->worker.task);
217 wait_for_completion(&done);
218 }
219
220abort:
221 if (f & CS_SERIALIZE)
222 mutex_unlock(&mdev->state_mutex);
223
224 return rv;
225}
226
227/**
228 * _drbd_request_state() - Request a state change (with flags)
229 * @mdev: DRBD device.
230 * @mask: mask of state bits to change.
231 * @val: value of new state bits.
232 * @f: flags
233 *
234 * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE
235 * flag, or when logging of failed state change requests is not desired.
236 */
237enum drbd_state_rv
238_drbd_request_state(struct drbd_conf *mdev, union drbd_state mask,
239 union drbd_state val, enum chg_state_flags f)
240{
241 enum drbd_state_rv rv;
242
243 wait_event(mdev->state_wait,
244 (rv = drbd_req_state(mdev, mask, val, f)) != SS_IN_TRANSIENT_STATE);
245
246 return rv;
247}
248
249static void print_st(struct drbd_conf *mdev, char *name, union drbd_state ns)
250{
251 dev_err(DEV, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n",
252 name,
253 drbd_conn_str(ns.conn),
254 drbd_role_str(ns.role),
255 drbd_role_str(ns.peer),
256 drbd_disk_str(ns.disk),
257 drbd_disk_str(ns.pdsk),
258 is_susp(ns) ? 's' : 'r',
259 ns.aftr_isp ? 'a' : '-',
260 ns.peer_isp ? 'p' : '-',
261 ns.user_isp ? 'u' : '-',
262 ns.susp_fen ? 'F' : '-',
263 ns.susp_nod ? 'N' : '-'
264 );
265}
266
267void print_st_err(struct drbd_conf *mdev, union drbd_state os,
268 union drbd_state ns, enum drbd_state_rv err)
269{
270 if (err == SS_IN_TRANSIENT_STATE)
271 return;
272 dev_err(DEV, "State change failed: %s\n", drbd_set_st_err_str(err));
273 print_st(mdev, " state", os);
274 print_st(mdev, "wanted", ns);
275}
276
bbeb641c
PR		 277static void print_state_change(struct drbd_conf *mdev, union drbd_state os, union drbd_state ns,
278 enum chg_state_flags flags)
279{
280 char *pbp, pb[300];
281 pbp = pb;
282 *pbp = 0;
283 if (ns.role != os.role)
284 pbp += sprintf(pbp, "role( %s -> %s ) ",
285 drbd_role_str(os.role),
286 drbd_role_str(ns.role));
287 if (ns.peer != os.peer)
288 pbp += sprintf(pbp, "peer( %s -> %s ) ",
289 drbd_role_str(os.peer),
290 drbd_role_str(ns.peer));
291 if (ns.conn != os.conn && !(flags & CS_NO_CSTATE_CHG))
292 pbp += sprintf(pbp, "conn( %s -> %s ) ",
293 drbd_conn_str(os.conn),
294 drbd_conn_str(ns.conn));
295 if (ns.disk != os.disk)
296 pbp += sprintf(pbp, "disk( %s -> %s ) ",
297 drbd_disk_str(os.disk),
298 drbd_disk_str(ns.disk));
299 if (ns.pdsk != os.pdsk)
300 pbp += sprintf(pbp, "pdsk( %s -> %s ) ",
301 drbd_disk_str(os.pdsk),
302 drbd_disk_str(ns.pdsk));
303 if (is_susp(ns) != is_susp(os))
304 pbp += sprintf(pbp, "susp( %d -> %d ) ",
305 is_susp(os),
306 is_susp(ns));
307 if (ns.aftr_isp != os.aftr_isp)
308 pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ",
309 os.aftr_isp,
310 ns.aftr_isp);
311 if (ns.peer_isp != os.peer_isp)
312 pbp += sprintf(pbp, "peer_isp( %d -> %d ) ",
313 os.peer_isp,
314 ns.peer_isp);
315 if (ns.user_isp != os.user_isp)
316 pbp += sprintf(pbp, "user_isp( %d -> %d ) ",
317 os.user_isp,
318 ns.user_isp);
319 if (pbp != pb)
320 dev_info(DEV, "%s\n", pb);
321}
b8907339
PR		 322
323/**
324 * is_valid_state() - Returns an SS_ error code if ns is not valid
325 * @mdev: DRBD device.
326 * @ns: State to consider.
327 */
328static enum drbd_state_rv
329is_valid_state(struct drbd_conf *mdev, union drbd_state ns)
330{
331 /* See drbd_state_sw_errors in drbd_strings.c */
332
333 enum drbd_fencing_p fp;
334 enum drbd_state_rv rv = SS_SUCCESS;
335
336 fp = FP_DONT_CARE;
337 if (get_ldev(mdev)) {
338 fp = mdev->ldev->dc.fencing;
339 put_ldev(mdev);
340 }
341
342 if (get_net_conf(mdev->tconn)) {
343 if (!mdev->tconn->net_conf->two_primaries &&
344 ns.role == R_PRIMARY && ns.peer == R_PRIMARY)
345 rv = SS_TWO_PRIMARIES;
346 put_net_conf(mdev->tconn);
347 }
348
349 if (rv <= 0)
350 /* already found a reason to abort */;
351 else if (ns.role == R_SECONDARY && mdev->open_cnt)
352 rv = SS_DEVICE_IN_USE;
353
354 else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE)
355 rv = SS_NO_UP_TO_DATE_DISK;
356
357 else if (fp >= FP_RESOURCE &&
358 ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN)
359 rv = SS_PRIMARY_NOP;
360
361 else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT)
362 rv = SS_NO_UP_TO_DATE_DISK;
363
364 else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT)
365 rv = SS_NO_LOCAL_DISK;
366
367 else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT)
368 rv = SS_NO_REMOTE_DISK;
369
370 else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
371 rv = SS_NO_UP_TO_DATE_DISK;
372
373 else if ((ns.conn == C_CONNECTED ||
374 ns.conn == C_WF_BITMAP_S ||
375 ns.conn == C_SYNC_SOURCE ||
376 ns.conn == C_PAUSED_SYNC_S) &&
377 ns.disk == D_OUTDATED)
378 rv = SS_CONNECTED_OUTDATES;
379
380 else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
381 (mdev->sync_conf.verify_alg[0] == 0))
382 rv = SS_NO_VERIFY_ALG;
383
384 else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
385 mdev->tconn->agreed_pro_version < 88)
386 rv = SS_NOT_SUPPORTED;
387
388 else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN)
389 rv = SS_CONNECTED_OUTDATES;
390
391 return rv;
392}
393
394/**
a75f34ad 395 * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible
3509502d
PR		 396 * This function limits state transitions that may be declined by DRBD. I.e.
397 * user requests (aka soft transitions).
b8907339
PR		 398 * @mdev: DRBD device.
399 * @ns: new state.
400 * @os: old state.
401 */
402static enum drbd_state_rv
a75f34ad 403is_valid_soft_transition(union drbd_state os, union drbd_state ns)
b8907339
PR		 404{
405 enum drbd_state_rv rv = SS_SUCCESS;
406
407 if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
408 os.conn > C_CONNECTED)
409 rv = SS_RESYNC_RUNNING;
410
411 if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
412 rv = SS_ALREADY_STANDALONE;
413
414 if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
415 rv = SS_IS_DISKLESS;
416
417 if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
418 rv = SS_NO_NET_CONFIG;
419
420 if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
421 rv = SS_LOWER_THAN_OUTDATED;
422
423 if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
424 rv = SS_IN_TRANSIENT_STATE;
425
426 if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS)
427 rv = SS_IN_TRANSIENT_STATE;
428
429 if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
430 rv = SS_NEED_CONNECTION;
431
432 if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
433 ns.conn != os.conn && os.conn > C_CONNECTED)
434 rv = SS_RESYNC_RUNNING;
435
436 if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
437 os.conn < C_CONNECTED)
438 rv = SS_NEED_CONNECTION;
439
440 if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)
441 && os.conn < C_WF_REPORT_PARAMS)
442 rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */
443
444 return rv;
445}
446
3509502d 447static enum drbd_state_rv
fda74117 448is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc)
3509502d
PR		 449{
450 enum drbd_state_rv rv = SS_SUCCESS;
451
452 /* Disallow Network errors to configure a device's network part */
fda74117 453 if ((nc >= C_TIMEOUT && nc <= C_TEAR_DOWN) && oc <= C_DISCONNECTING)
3509502d
PR		 454 rv = SS_NEED_CONNECTION;
455
456 /* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */
fda74117 457 if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING)
3509502d
PR		 458 rv = SS_IN_TRANSIENT_STATE;
459
460 /* After C_DISCONNECTING only C_STANDALONE may follow */
fda74117 461 if (oc == C_DISCONNECTING && nc != C_STANDALONE)
3509502d
PR		 462 rv = SS_IN_TRANSIENT_STATE;
463
fda74117
PR		 464 return rv;
465}
466
467
468/**
469 * is_valid_transition() - Returns an SS_ error code if the state transition is not possible
470 * This limits hard state transitions. Hard state transitions are facts there are
471 * imposed on DRBD by the environment. E.g. disk broke or network broke down.
472 * But those hard state transitions are still not allowed to do everything.
473 * @ns: new state.
474 * @os: old state.
475 */
476static enum drbd_state_rv
477is_valid_transition(union drbd_state os, union drbd_state ns)
478{
479 enum drbd_state_rv rv;
480
481 rv = is_valid_conn_transition(os.conn, ns.conn);
482
3509502d
PR		 483 /* we cannot fail (again) if we already detached */
484 if (ns.disk == D_FAILED && os.disk == D_DISKLESS)
485 rv = SS_IS_DISKLESS;
486
4308a0a3
PR		 487 /* if we are only D_ATTACHING yet,
488 * we can (and should) go directly to D_DISKLESS. */
489 if (ns.disk == D_FAILED && os.disk == D_ATTACHING) {
490 printk("TODO: FIX ME\n");
491 rv = SS_IS_DISKLESS;
492 }
493
3509502d
PR		 494 return rv;
495}
496
b8907339
PR		 497/**
498 * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
499 * @mdev: DRBD device.
500 * @os: old state.
501 * @ns: new state.
502 * @warn_sync_abort:
503 *
504 * When we loose connection, we have to set the state of the peers disk (pdsk)
505 * to D_UNKNOWN. This rule and many more along those lines are in this function.
506 */
4308a0a3
PR		 507static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns,
508 const char **warn_sync_abort)
b8907339
PR		 509{
510 enum drbd_fencing_p fp;
511 enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max;
512
513 fp = FP_DONT_CARE;
514 if (get_ldev(mdev)) {
515 fp = mdev->ldev->dc.fencing;
516 put_ldev(mdev);
517 }
518
3509502d 519 /* Implications from connection to peer and peer_isp */
b8907339
PR		 520 if (ns.conn < C_CONNECTED) {
521 ns.peer_isp = 0;
522 ns.peer = R_UNKNOWN;
523 if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
524 ns.pdsk = D_UNKNOWN;
525 }
526
527 /* Clear the aftr_isp when becoming unconfigured */
528 if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
529 ns.aftr_isp = 0;
530
4308a0a3 531 /* An implication of the disk states onto the connection state */
b8907339 532 /* Abort resync if a disk fails/detaches */
4308a0a3 533 if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
b8907339
PR		 534 if (warn_sync_abort)
535 *warn_sync_abort =
4308a0a3 536 ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ?
b8907339
PR		 537 "Online-verify" : "Resync";
538 ns.conn = C_CONNECTED;
539 }
540
541 /* Connection breaks down before we finished "Negotiating" */
542 if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
543 get_ldev_if_state(mdev, D_NEGOTIATING)) {
544 if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) {
545 ns.disk = mdev->new_state_tmp.disk;
546 ns.pdsk = mdev->new_state_tmp.pdsk;
547 } else {
548 dev_alert(DEV, "Connection lost while negotiating, no data!\n");
549 ns.disk = D_DISKLESS;
550 ns.pdsk = D_UNKNOWN;
551 }
552 put_ldev(mdev);
553 }
554
555 /* D_CONSISTENT and D_OUTDATED vanish when we get connected */
556 if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) {
557 if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED)
558 ns.disk = D_UP_TO_DATE;
559 if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)
560 ns.pdsk = D_UP_TO_DATE;
561 }
562
563 /* Implications of the connection stat on the disk states */
564 disk_min = D_DISKLESS;
565 disk_max = D_UP_TO_DATE;
566 pdsk_min = D_INCONSISTENT;
567 pdsk_max = D_UNKNOWN;
568 switch ((enum drbd_conns)ns.conn) {
569 case C_WF_BITMAP_T:
570 case C_PAUSED_SYNC_T:
571 case C_STARTING_SYNC_T:
572 case C_WF_SYNC_UUID:
573 case C_BEHIND:
574 disk_min = D_INCONSISTENT;
575 disk_max = D_OUTDATED;
576 pdsk_min = D_UP_TO_DATE;
577 pdsk_max = D_UP_TO_DATE;
578 break;
579 case C_VERIFY_S:
580 case C_VERIFY_T:
581 disk_min = D_UP_TO_DATE;
582 disk_max = D_UP_TO_DATE;
583 pdsk_min = D_UP_TO_DATE;
584 pdsk_max = D_UP_TO_DATE;
585 break;
586 case C_CONNECTED:
587 disk_min = D_DISKLESS;
588 disk_max = D_UP_TO_DATE;
589 pdsk_min = D_DISKLESS;
590 pdsk_max = D_UP_TO_DATE;
591 break;
592 case C_WF_BITMAP_S:
593 case C_PAUSED_SYNC_S:
594 case C_STARTING_SYNC_S:
595 case C_AHEAD:
596 disk_min = D_UP_TO_DATE;
597 disk_max = D_UP_TO_DATE;
598 pdsk_min = D_INCONSISTENT;
599 pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/
600 break;
601 case C_SYNC_TARGET:
602 disk_min = D_INCONSISTENT;
603 disk_max = D_INCONSISTENT;
604 pdsk_min = D_UP_TO_DATE;
605 pdsk_max = D_UP_TO_DATE;
606 break;
607 case C_SYNC_SOURCE:
608 disk_min = D_UP_TO_DATE;
609 disk_max = D_UP_TO_DATE;
610 pdsk_min = D_INCONSISTENT;
611 pdsk_max = D_INCONSISTENT;
612 break;
613 case C_STANDALONE:
614 case C_DISCONNECTING:
615 case C_UNCONNECTED:
616 case C_TIMEOUT:
617 case C_BROKEN_PIPE:
618 case C_NETWORK_FAILURE:
619 case C_PROTOCOL_ERROR:
620 case C_TEAR_DOWN:
621 case C_WF_CONNECTION:
622 case C_WF_REPORT_PARAMS:
623 case C_MASK:
624 break;
625 }
626 if (ns.disk > disk_max)
627 ns.disk = disk_max;
628
629 if (ns.disk < disk_min) {
630 dev_warn(DEV, "Implicitly set disk from %s to %s\n",
631 drbd_disk_str(ns.disk), drbd_disk_str(disk_min));
632 ns.disk = disk_min;
633 }
634 if (ns.pdsk > pdsk_max)
635 ns.pdsk = pdsk_max;
636
637 if (ns.pdsk < pdsk_min) {
638 dev_warn(DEV, "Implicitly set pdsk from %s to %s\n",
639 drbd_disk_str(ns.pdsk), drbd_disk_str(pdsk_min));
640 ns.pdsk = pdsk_min;
641 }
642
643 if (fp == FP_STONITH &&
4308a0a3 644 (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED))
b8907339
PR		 645 ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */
646
647 if (mdev->sync_conf.on_no_data == OND_SUSPEND_IO &&
4308a0a3 648 (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
b8907339
PR		 649 ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */
650
651 if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
652 if (ns.conn == C_SYNC_SOURCE)
653 ns.conn = C_PAUSED_SYNC_S;
654 if (ns.conn == C_SYNC_TARGET)
655 ns.conn = C_PAUSED_SYNC_T;
656 } else {
657 if (ns.conn == C_PAUSED_SYNC_S)
658 ns.conn = C_SYNC_SOURCE;
659 if (ns.conn == C_PAUSED_SYNC_T)
660 ns.conn = C_SYNC_TARGET;
661 }
662
663 return ns;
664}
665
666void drbd_resume_al(struct drbd_conf *mdev)
667{
668 if (test_and_clear_bit(AL_SUSPENDED, &mdev->flags))
669 dev_info(DEV, "Resumed AL updates\n");
670}
671
672/* helper for __drbd_set_state */
673static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs)
674{
675 if (mdev->tconn->agreed_pro_version < 90)
676 mdev->ov_start_sector = 0;
677 mdev->rs_total = drbd_bm_bits(mdev);
678 mdev->ov_position = 0;
679 if (cs == C_VERIFY_T) {
680 /* starting online verify from an arbitrary position
681 * does not fit well into the existing protocol.
682 * on C_VERIFY_T, we initialize ov_left and friends
683 * implicitly in receive_DataRequest once the
684 * first P_OV_REQUEST is received */
685 mdev->ov_start_sector = ~(sector_t)0;
686 } else {
687 unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector);
688 if (bit >= mdev->rs_total) {
689 mdev->ov_start_sector =
690 BM_BIT_TO_SECT(mdev->rs_total - 1);
691 mdev->rs_total = 1;
692 } else
693 mdev->rs_total -= bit;
694 mdev->ov_position = mdev->ov_start_sector;
695 }
696 mdev->ov_left = mdev->rs_total;
697}
698
699/**
700 * __drbd_set_state() - Set a new DRBD state
701 * @mdev: DRBD device.
702 * @ns: new state.
703 * @flags: Flags
704 * @done: Optional completion, that will get completed after the after_state_ch() finished
705 *
706 * Caller needs to hold req_lock, and global_state_lock. Do not call directly.
707 */
708enum drbd_state_rv
709__drbd_set_state(struct drbd_conf *mdev, union drbd_state ns,
710 enum chg_state_flags flags, struct completion *done)
711{
712 union drbd_state os;
713 enum drbd_state_rv rv = SS_SUCCESS;
714 const char *warn_sync_abort = NULL;
715 struct after_state_chg_work *ascw;
716
717 os = mdev->state;
718
4308a0a3 719 ns = sanitize_state(mdev, ns, &warn_sync_abort);
b8907339
PR		 720 if (ns.i == os.i)
721 return SS_NOTHING_TO_DO;
722
3509502d
PR		 723 rv = is_valid_transition(os, ns);
724 if (rv < SS_SUCCESS)
725 return rv;
726
b8907339
PR		 727 if (!(flags & CS_HARD)) {
728 /* pre-state-change checks ; only look at ns */
729 /* See drbd_state_sw_errors in drbd_strings.c */
730
731 rv = is_valid_state(mdev, ns);
732 if (rv < SS_SUCCESS) {
733 /* If the old state was illegal as well, then let
734 this happen...*/
735
736 if (is_valid_state(mdev, os) == rv)
a75f34ad 737 rv = is_valid_soft_transition(os, ns);
b8907339 738 } else
a75f34ad 739 rv = is_valid_soft_transition(os, ns);
b8907339
PR		 740 }
741
742 if (rv < SS_SUCCESS) {
743 if (flags & CS_VERBOSE)
744 print_st_err(mdev, os, ns, rv);
745 return rv;
746 }
747
748 if (warn_sync_abort)
749 dev_warn(DEV, "%s aborted.\n", warn_sync_abort);
750
bbeb641c 751 print_state_change(mdev, os, ns, flags);
b8907339
PR		 752
753 /* solve the race between becoming unconfigured,
754 * worker doing the cleanup, and
755 * admin reconfiguring us:
756 * on (re)configure, first set CONFIG_PENDING,
757 * then wait for a potentially exiting worker,
758 * start the worker, and schedule one no_op.
759 * then proceed with configuration.
760 */
761 if (ns.disk == D_DISKLESS &&
762 ns.conn == C_STANDALONE &&
763 ns.role == R_SECONDARY &&
764 !test_and_set_bit(CONFIG_PENDING, &mdev->flags))
765 set_bit(DEVICE_DYING, &mdev->flags);
766
767 /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
768 * on the ldev here, to be sure the transition -> D_DISKLESS resp.
769 * drbd_ldev_destroy() won't happen before our corresponding
770 * after_state_ch works run, where we put_ldev again. */
771 if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
772 (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
773 atomic_inc(&mdev->local_cnt);
774
775 mdev->state = ns;
776
777 if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
778 drbd_print_uuids(mdev, "attached to UUIDs");
779
780 wake_up(&mdev->misc_wait);
781 wake_up(&mdev->state_wait);
2a67d8b9 782 wake_up(&mdev->tconn->ping_wait);
b8907339
PR		 783
784 /* aborted verify run. log the last position */
785 if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
786 ns.conn < C_CONNECTED) {
787 mdev->ov_start_sector =
788 BM_BIT_TO_SECT(drbd_bm_bits(mdev) - mdev->ov_left);
789 dev_info(DEV, "Online Verify reached sector %llu\n",
790 (unsigned long long)mdev->ov_start_sector);
791 }
792
793 if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
794 (ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)) {
795 dev_info(DEV, "Syncer continues.\n");
796 mdev->rs_paused += (long)jiffies
797 -(long)mdev->rs_mark_time[mdev->rs_last_mark];
798 if (ns.conn == C_SYNC_TARGET)
799 mod_timer(&mdev->resync_timer, jiffies);
800 }
801
802 if ((os.conn == C_SYNC_TARGET || os.conn == C_SYNC_SOURCE) &&
803 (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
804 dev_info(DEV, "Resync suspended\n");
805 mdev->rs_mark_time[mdev->rs_last_mark] = jiffies;
806 }
807
808 if (os.conn == C_CONNECTED &&
809 (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
810 unsigned long now = jiffies;
811 int i;
812
813 set_ov_position(mdev, ns.conn);
814 mdev->rs_start = now;
815 mdev->rs_last_events = 0;
816 mdev->rs_last_sect_ev = 0;
817 mdev->ov_last_oos_size = 0;
818 mdev->ov_last_oos_start = 0;
819
820 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
821 mdev->rs_mark_left[i] = mdev->ov_left;
822 mdev->rs_mark_time[i] = now;
823 }
824
825 drbd_rs_controller_reset(mdev);
826
827 if (ns.conn == C_VERIFY_S) {
828 dev_info(DEV, "Starting Online Verify from sector %llu\n",
829 (unsigned long long)mdev->ov_position);
830 mod_timer(&mdev->resync_timer, jiffies);
831 }
832 }
833
834 if (get_ldev(mdev)) {
835 u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
836 MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
837 MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
838
839 if (test_bit(CRASHED_PRIMARY, &mdev->flags))
840 mdf |= MDF_CRASHED_PRIMARY;
841 if (mdev->state.role == R_PRIMARY ||
842 (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY))
843 mdf |= MDF_PRIMARY_IND;
844 if (mdev->state.conn > C_WF_REPORT_PARAMS)
845 mdf |= MDF_CONNECTED_IND;
846 if (mdev->state.disk > D_INCONSISTENT)
847 mdf |= MDF_CONSISTENT;
848 if (mdev->state.disk > D_OUTDATED)
849 mdf |= MDF_WAS_UP_TO_DATE;
850 if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT)
851 mdf |= MDF_PEER_OUT_DATED;
852 if (mdf != mdev->ldev->md.flags) {
853 mdev->ldev->md.flags = mdf;
854 drbd_md_mark_dirty(mdev);
855 }
856 if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
857 drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]);
858 put_ldev(mdev);
859 }
860
861 /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
862 if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
863 os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
864 set_bit(CONSIDER_RESYNC, &mdev->flags);
865
866 /* Receiver should clean up itself */
867 if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
868 drbd_thread_stop_nowait(&mdev->tconn->receiver);
869
870 /* Now the receiver finished cleaning up itself, it should die */
871 if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
872 drbd_thread_stop_nowait(&mdev->tconn->receiver);
873
874 /* Upon network failure, we need to restart the receiver. */
875 if (os.conn > C_TEAR_DOWN &&
876 ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
877 drbd_thread_restart_nowait(&mdev->tconn->receiver);
878
879 /* Resume AL writing if we get a connection */
880 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
881 drbd_resume_al(mdev);
882
883 ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
884 if (ascw) {
885 ascw->os = os;
886 ascw->ns = ns;
887 ascw->flags = flags;
888 ascw->w.cb = w_after_state_ch;
a21e9298 889 ascw->w.mdev = mdev;
b8907339
PR		 890 ascw->done = done;
891 drbd_queue_work(&mdev->tconn->data.work, &ascw->w);
892 } else {
bbeb641c 893 dev_err(DEV, "Could not kmalloc an ascw\n");
b8907339
PR		 894 }
895
896 return rv;
897}
898
00d56944 899static int w_after_state_ch(struct drbd_work *w, int unused)
b8907339
PR		 900{
901 struct after_state_chg_work *ascw =
902 container_of(w, struct after_state_chg_work, w);
00d56944 903 struct drbd_conf *mdev = w->mdev;
b8907339
PR		 904
905 after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags);
906 if (ascw->flags & CS_WAIT_COMPLETE) {
907 D_ASSERT(ascw->done != NULL);
908 complete(ascw->done);
909 }
910 kfree(ascw);
911
912 return 1;
913}
914
915static void abw_start_sync(struct drbd_conf *mdev, int rv)
916{
917 if (rv) {
918 dev_err(DEV, "Writing the bitmap failed not starting resync.\n");
919 _drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE);
920 return;
921 }
922
923 switch (mdev->state.conn) {
924 case C_STARTING_SYNC_T:
925 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
926 break;
927 case C_STARTING_SYNC_S:
928 drbd_start_resync(mdev, C_SYNC_SOURCE);
929 break;
930 }
931}
932
933int drbd_bitmap_io_from_worker(struct drbd_conf *mdev,
934 int (*io_fn)(struct drbd_conf *),
935 char *why, enum bm_flag flags)
936{
937 int rv;
938
939 D_ASSERT(current == mdev->tconn->worker.task);
940
941 /* open coded non-blocking drbd_suspend_io(mdev); */
942 set_bit(SUSPEND_IO, &mdev->flags);
943
944 drbd_bm_lock(mdev, why, flags);
945 rv = io_fn(mdev);
946 drbd_bm_unlock(mdev);
947
948 drbd_resume_io(mdev);
949
950 return rv;
951}
952
953/**
954 * after_state_ch() - Perform after state change actions that may sleep
955 * @mdev: DRBD device.
956 * @os: old state.
957 * @ns: new state.
958 * @flags: Flags
959 */
960static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
961 union drbd_state ns, enum chg_state_flags flags)
962{
963 enum drbd_fencing_p fp;
964 enum drbd_req_event what = NOTHING;
965 union drbd_state nsm = (union drbd_state){ .i = -1 };
966
967 if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) {
968 clear_bit(CRASHED_PRIMARY, &mdev->flags);
969 if (mdev->p_uuid)
970 mdev->p_uuid[UI_FLAGS] &= ~((u64)2);
971 }
972
973 fp = FP_DONT_CARE;
974 if (get_ldev(mdev)) {
975 fp = mdev->ldev->dc.fencing;
976 put_ldev(mdev);
977 }
978
979 /* Inform userspace about the change... */
980 drbd_bcast_state(mdev, ns);
981
982 if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
983 (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
984 drbd_khelper(mdev, "pri-on-incon-degr");
985
986 /* Here we have the actions that are performed after a
987 state change. This function might sleep */
988
989 nsm.i = -1;
990 if (ns.susp_nod) {
991 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
992 what = RESEND;
993
994 if (os.disk == D_ATTACHING && ns.disk > D_ATTACHING)
995 what = RESTART_FROZEN_DISK_IO;
996
997 if (what != NOTHING)
998 nsm.susp_nod = 0;
999 }
1000
1001 if (ns.susp_fen) {
1002 /* case1: The outdate peer handler is successful: */
1003 if (os.pdsk > D_OUTDATED && ns.pdsk <= D_OUTDATED) {
1004 tl_clear(mdev);
1005 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1006 drbd_uuid_new_current(mdev);
1007 clear_bit(NEW_CUR_UUID, &mdev->flags);
1008 }
1009 spin_lock_irq(&mdev->tconn->req_lock);
1010 _drbd_set_state(_NS(mdev, susp_fen, 0), CS_VERBOSE, NULL);
1011 spin_unlock_irq(&mdev->tconn->req_lock);
1012 }
1013 /* case2: The connection was established again: */
1014 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) {
1015 clear_bit(NEW_CUR_UUID, &mdev->flags);
1016 what = RESEND;
1017 nsm.susp_fen = 0;
1018 }
1019 }
1020
1021 if (what != NOTHING) {
1022 spin_lock_irq(&mdev->tconn->req_lock);
1023 _tl_restart(mdev, what);
1024 nsm.i &= mdev->state.i;
1025 _drbd_set_state(mdev, nsm, CS_VERBOSE, NULL);
1026 spin_unlock_irq(&mdev->tconn->req_lock);
1027 }
1028
1029 /* Became sync source. With protocol >= 96, we still need to send out
1030 * the sync uuid now. Need to do that before any drbd_send_state, or
1031 * the other side may go "paused sync" before receiving the sync uuids,
1032 * which is unexpected. */
1033 if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
1034 (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
1035 mdev->tconn->agreed_pro_version >= 96 && get_ldev(mdev)) {
1036 drbd_gen_and_send_sync_uuid(mdev);
1037 put_ldev(mdev);
1038 }
1039
1040 /* Do not change the order of the if above and the two below... */
1041 if (os.pdsk == D_DISKLESS && ns.pdsk > D_DISKLESS) { /* attach on the peer */
1042 drbd_send_uuids(mdev);
1043 drbd_send_state(mdev);
1044 }
1045 /* No point in queuing send_bitmap if we don't have a connection
1046 * anymore, so check also the _current_ state, not only the new state
1047 * at the time this work was queued. */
1048 if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
1049 mdev->state.conn == C_WF_BITMAP_S)
1050 drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL,
1051 "send_bitmap (WFBitMapS)",
1052 BM_LOCKED_TEST_ALLOWED);
1053
1054 /* Lost contact to peer's copy of the data */
1055 if ((os.pdsk >= D_INCONSISTENT &&
1056 os.pdsk != D_UNKNOWN &&
1057 os.pdsk != D_OUTDATED)
1058 && (ns.pdsk < D_INCONSISTENT ||
1059 ns.pdsk == D_UNKNOWN ||
1060 ns.pdsk == D_OUTDATED)) {
1061 if (get_ldev(mdev)) {
1062 if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1063 mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1064 if (is_susp(mdev->state)) {
1065 set_bit(NEW_CUR_UUID, &mdev->flags);
1066 } else {
1067 drbd_uuid_new_current(mdev);
1068 drbd_send_uuids(mdev);
1069 }
1070 }
1071 put_ldev(mdev);
1072 }
1073 }
1074
1075 if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) {
1076 if (ns.peer == R_PRIMARY && mdev->ldev->md.uuid[UI_BITMAP] == 0) {
1077 drbd_uuid_new_current(mdev);
1078 drbd_send_uuids(mdev);
1079 }
1080
1081 /* D_DISKLESS Peer becomes secondary */
1082 if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1083 /* We may still be Primary ourselves.
1084 * No harm done if the bitmap still changes,
1085 * redirtied pages will follow later. */
1086 drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1087 "demote diskless peer", BM_LOCKED_SET_ALLOWED);
1088 put_ldev(mdev);
1089 }
1090
1091 /* Write out all changed bits on demote.
1092 * Though, no need to da that just yet
1093 * if there is a resync going on still */
1094 if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
1095 mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) {
1096 /* No changes to the bitmap expected this time, so assert that,
1097 * even though no harm was done if it did change. */
1098 drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1099 "demote", BM_LOCKED_TEST_ALLOWED);
1100 put_ldev(mdev);
1101 }
1102
1103 /* Last part of the attaching process ... */
1104 if (ns.conn >= C_CONNECTED &&
1105 os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1106 drbd_send_sizes(mdev, 0, 0); /* to start sync... */
1107 drbd_send_uuids(mdev);
1108 drbd_send_state(mdev);
1109 }
1110
1111 /* We want to pause/continue resync, tell peer. */
1112 if (ns.conn >= C_CONNECTED &&
1113 ((os.aftr_isp != ns.aftr_isp) ||
1114 (os.user_isp != ns.user_isp)))
1115 drbd_send_state(mdev);
1116
1117 /* In case one of the isp bits got set, suspend other devices. */
1118 if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1119 (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1120 suspend_other_sg(mdev);
1121
1122 /* Make sure the peer gets informed about eventual state
1123 changes (ISP bits) while we were in WFReportParams. */
1124 if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1125 drbd_send_state(mdev);
1126
1127 if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
1128 drbd_send_state(mdev);
1129
1130 /* We are in the progress to start a full sync... */
1131 if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1132 (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1133 /* no other bitmap changes expected during this phase */
1134 drbd_queue_bitmap_io(mdev,
1135 &drbd_bmio_set_n_write, &abw_start_sync,
1136 "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED);
1137
1138 /* We are invalidating our self... */
1139 if (os.conn < C_CONNECTED && ns.conn < C_CONNECTED &&
1140 os.disk > D_INCONSISTENT && ns.disk == D_INCONSISTENT)
1141 /* other bitmap operation expected during this phase */
1142 drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, NULL,
1143 "set_n_write from invalidate", BM_LOCKED_MASK);
1144
1145 /* first half of local IO error, failure to attach,
1146 * or administrative detach */
1147 if (os.disk != D_FAILED && ns.disk == D_FAILED) {
1148 enum drbd_io_error_p eh;
1149 int was_io_error;
1150 /* corresponding get_ldev was in __drbd_set_state, to serialize
1151 * our cleanup here with the transition to D_DISKLESS,
1152 * so it is safe to dreference ldev here. */
1153 eh = mdev->ldev->dc.on_io_error;
1154 was_io_error = test_and_clear_bit(WAS_IO_ERROR, &mdev->flags);
1155
1156 /* current state still has to be D_FAILED,
1157 * there is only one way out: to D_DISKLESS,
1158 * and that may only happen after our put_ldev below. */
1159 if (mdev->state.disk != D_FAILED)
1160 dev_err(DEV,
1161 "ASSERT FAILED: disk is %s during detach\n",
1162 drbd_disk_str(mdev->state.disk));
1163
1164 if (drbd_send_state(mdev))
1165 dev_warn(DEV, "Notified peer that I am detaching my disk\n");
1166 else
1167 dev_err(DEV, "Sending state for detaching disk failed\n");
1168
1169 drbd_rs_cancel_all(mdev);
1170
1171 /* In case we want to get something to stable storage still,
1172 * this may be the last chance.
1173 * Following put_ldev may transition to D_DISKLESS. */
1174 drbd_md_sync(mdev);
1175 put_ldev(mdev);
1176
1177 if (was_io_error && eh == EP_CALL_HELPER)
1178 drbd_khelper(mdev, "local-io-error");
1179 }
1180
1181 /* second half of local IO error, failure to attach,
1182 * or administrative detach,
1183 * after local_cnt references have reached zero again */
1184 if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
1185 /* We must still be diskless,
1186 * re-attach has to be serialized with this! */
1187 if (mdev->state.disk != D_DISKLESS)
1188 dev_err(DEV,
1189 "ASSERT FAILED: disk is %s while going diskless\n",
1190 drbd_disk_str(mdev->state.disk));
1191
1192 mdev->rs_total = 0;
1193 mdev->rs_failed = 0;
1194 atomic_set(&mdev->rs_pending_cnt, 0);
1195
1196 if (drbd_send_state(mdev))
1197 dev_warn(DEV, "Notified peer that I'm now diskless.\n");
1198 /* corresponding get_ldev in __drbd_set_state
1199 * this may finally trigger drbd_ldev_destroy. */
1200 put_ldev(mdev);
1201 }
1202
1203 /* Notify peer that I had a local IO error, and did not detached.. */
1204 if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT)
1205 drbd_send_state(mdev);
1206
1207 /* Disks got bigger while they were detached */
1208 if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1209 test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) {
1210 if (ns.conn == C_CONNECTED)
1211 resync_after_online_grow(mdev);
1212 }
1213
1214 /* A resync finished or aborted, wake paused devices... */
1215 if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1216 (os.peer_isp && !ns.peer_isp) ||
1217 (os.user_isp && !ns.user_isp))
1218 resume_next_sg(mdev);
1219
1220 /* sync target done with resync. Explicitly notify peer, even though
1221 * it should (at least for non-empty resyncs) already know itself. */
1222 if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
1223 drbd_send_state(mdev);
1224
1225 /* This triggers bitmap writeout of potentially still unwritten pages
1226 * if the resync finished cleanly, or aborted because of peer disk
1227 * failure, or because of connection loss.
1228 * For resync aborted because of local disk failure, we cannot do
1229 * any bitmap writeout anymore.
1230 * No harm done if some bits change during this phase.
1231 */
1232 if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) {
1233 drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL,
1234 "write from resync_finished", BM_LOCKED_SET_ALLOWED);
1235 put_ldev(mdev);
1236 }
1237
1238 if (ns.disk == D_DISKLESS &&
1239 ns.conn == C_STANDALONE &&
1240 ns.role == R_SECONDARY) {
1241 if (os.aftr_isp != ns.aftr_isp)
1242 resume_next_sg(mdev);
1243 }
1244
bbeb641c
PR		 1245 after_all_state_ch(mdev->tconn, ns);
1246
b8907339
PR		 1247 drbd_md_sync(mdev);
1248}
1249
bbeb641c
PR		 1250struct after_conn_state_chg_work {
1251 struct drbd_work w;
1252 enum drbd_conns oc;
1253 union drbd_state nms; /* new, max state, over all mdevs */
1254 enum chg_state_flags flags;
1255};
1256
1257static void after_all_state_ch(struct drbd_tconn *tconn, union drbd_state ns)
b8907339 1258{
bbeb641c
PR		 1259 if (ns.disk == D_DISKLESS && ns.conn == C_STANDALONE && ns.role == R_SECONDARY) {
1260 /* if (test_bit(DEVICE_DYING, &mdev->flags)) TODO: DEVICE_DYING functionality */
1261 drbd_thread_stop_nowait(&tconn->worker);
1262 }
1263}
1264
1265static int w_after_conn_state_ch(struct drbd_work *w, int unused)
1266{
1267 struct after_conn_state_chg_work *acscw =
1268 container_of(w, struct after_conn_state_chg_work, w);
1269 struct drbd_tconn *tconn = w->tconn;
1270 enum drbd_conns oc = acscw->oc;
1271 union drbd_state nms = acscw->nms;
1272
1273 kfree(acscw);
1274
b8907339 1275 /* Upon network configuration, we need to start the receiver */
bbeb641c 1276 if (oc == C_STANDALONE && nms.conn == C_UNCONNECTED)
b8907339
PR		 1277 drbd_thread_start(&tconn->receiver);
1278
bbeb641c
PR		 1279 //conn_err(tconn, STATE_FMT, STATE_ARGS("nms", nms));
1280 after_all_state_ch(tconn, nms);
1281
1282 return 1;
1283}
1284
1285static void print_conn_state_change(struct drbd_tconn *tconn, enum drbd_conns oc, enum drbd_conns nc)
1286{
1287 char *pbp, pb[300];
1288 pbp = pb;
1289 *pbp = 0;
1290 if (nc != oc)
1291 pbp += sprintf(pbp, "conn( %s -> %s ) ",
1292 drbd_conn_str(oc),
1293 drbd_conn_str(nc));
1294
1295 conn_info(tconn, "%s\n", pb);
1296}
1297
1298struct _is_valid_itr_params {
1299 enum chg_state_flags flags;
1300 union drbd_state mask, val;
1301 union drbd_state ms; /* maximal state, over all mdevs */
1302 enum drbd_conns oc;
1303 enum {
1304 OC_UNINITIALIZED,
1305 OC_CONSISTENT,
1306 OC_INCONSISTENT,
1307 } oc_state;
1308};
1309
1310static int _is_valid_itr_fn(int vnr, void *p, void *data)
1311{
1312 struct drbd_conf *mdev = (struct drbd_conf *)p;
1313 struct _is_valid_itr_params *params = (struct _is_valid_itr_params *)data;
1314 enum chg_state_flags flags = params->flags;
1315 union drbd_state ns, os;
1316 enum drbd_state_rv rv;
1317
1318 os = mdev->state;
1319 ns = apply_mask_val(os, params->mask, params->val);
1320 ns = sanitize_state(mdev, ns, NULL);
1321 rv = is_valid_state(mdev, ns);
1322
1323 if (rv < SS_SUCCESS) {
1324 /* If the old state was illegal as well, then let this happen...*/
1325
1326 if (is_valid_state(mdev, os) == rv)
1327 rv = is_valid_soft_transition(os, ns);
1328 } else
1329 rv = is_valid_soft_transition(os, ns);
1330
1331 switch (params->oc_state) {
1332 case OC_UNINITIALIZED:
1333 params->oc = os.conn;
1334 params->oc_state = OC_CONSISTENT;
1335 break;
1336 case OC_CONSISTENT:
1337 if (params->oc != os.conn)
1338 params->oc_state = OC_INCONSISTENT;
1339 break;
1340 case OC_INCONSISTENT:
1341 break;
1342 }
1343
1344 if (rv < SS_SUCCESS) {
1345 if (flags & CS_VERBOSE)
1346 print_st_err(mdev, os, ns, rv);
1347 return rv;
1348 } else
1349 return 0;
1350}
1351
1352static int _set_state_itr_fn(int vnr, void *p, void *data)
1353{
1354 struct drbd_conf *mdev = (struct drbd_conf *)p;
1355 struct _is_valid_itr_params *params = (struct _is_valid_itr_params *)data;
1356 enum chg_state_flags flags = params->flags;
1357 union drbd_state os, ns, ms = params->ms;
1358 enum drbd_state_rv rv;
1359
1360 os = mdev->state;
1361 ns = apply_mask_val(os, params->mask, params->val);
1362 ns = sanitize_state(mdev, ns, NULL);
1363
1364 rv = __drbd_set_state(mdev, ns, flags, NULL);
1365
1366 ms.role = max_t(enum drbd_role, mdev->state.role, ms.role);
1367 ms.peer = max_t(enum drbd_role, mdev->state.peer, ms.peer);
1368 ms.disk = max_t(enum drbd_role, mdev->state.disk, ms.disk);
1369 ms.pdsk = max_t(enum drbd_role, mdev->state.pdsk, ms.pdsk);
1370 params->ms = ms;
1371
1372 return 0;
1373}
1374
1375enum drbd_state_rv
1376_conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1377 enum chg_state_flags flags)
1378{
1379 enum drbd_state_rv rv = SS_SUCCESS;
1380 struct _is_valid_itr_params params;
1381 struct after_conn_state_chg_work *acscw;
1382 enum drbd_conns oc = tconn->cstate;
1383
1384 read_lock(&global_state_lock);
1385
1386 rv = is_valid_conn_transition(oc, val.conn);
1387 if (rv < SS_SUCCESS)
1388 goto abort;
1389
1390 params.flags = flags;
1391 params.mask = mask;
1392 params.val = val;
1393 params.oc_state = OC_UNINITIALIZED;
1394
1395 if (!(flags & CS_HARD))
1396 rv = idr_for_each(&tconn->volumes, _is_valid_itr_fn, &params);
1397
1398 if (rv == 0) /* idr_for_each semantics */
1399 rv = SS_SUCCESS;
1400
1401 if (rv < SS_SUCCESS)
1402 goto abort;
1403
1404 if (params.oc_state == OC_CONSISTENT) {
1405 oc = params.oc;
1406 print_conn_state_change(tconn, oc, val.conn);
1407 params.flags |= CS_NO_CSTATE_CHG;
1408 }
1409 tconn->cstate = val.conn;
1410 params.ms.i = 0;
1411 params.ms.conn = val.conn;
1412 idr_for_each(&tconn->volumes, _set_state_itr_fn, &params);
1413
1414 acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC);
1415 if (acscw) {
1416 acscw->oc = oc;
1417 acscw->nms = params.ms;
1418 acscw->flags = flags;
1419 acscw->w.cb = w_after_conn_state_ch;
1420 acscw->w.tconn = tconn;
1421 drbd_queue_work(&tconn->data.work, &acscw->w);
1422 } else {
1423 conn_err(tconn, "Could not kmalloc an acscw\n");
b8907339 1424 }
bbeb641c
PR		 1425
1426abort:
1427 read_unlock(&global_state_lock);
1428
1429 return rv;
1430}
1431
1432enum drbd_state_rv
1433conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1434 enum chg_state_flags flags)
1435{
1436 enum drbd_state_rv rv;
1437
1438 spin_lock_irq(&tconn->req_lock);
1439 rv = _conn_request_state(tconn, mask, val, flags);
1440 spin_unlock_irq(&tconn->req_lock);
1441
1442 return rv;
b8907339 1443}
Linux kernel - Deliverables
This page took 0.164892 seconds and 5 git commands to generate.