4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
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>.
10 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
11 from Logicworks, Inc. for making SDP replication support possible.
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)
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.
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.
28 #include <linux/drbd_limits.h>
32 struct after_state_chg_work {
36 enum chg_state_flags flags;
37 struct completion *done;
41 extern void _tl_restart(struct drbd_conf *mdev, enum drbd_req_event what);
42 int drbd_send_state_req(struct drbd_conf *, union drbd_state, union drbd_state);
43 static int w_after_state_ch(struct drbd_work *w, int unused);
44 static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
45 union drbd_state ns, enum chg_state_flags flags);
46 static void after_all_state_ch(struct drbd_tconn *tconn, union drbd_state ns);
47 static enum drbd_state_rv is_valid_state(struct drbd_conf *, union drbd_state);
48 static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state);
49 static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns);
50 static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns,
51 const char **warn_sync_abort);
54 * cl_wide_st_chg() - true if the state change is a cluster wide one
56 * @os: old (current) state.
57 * @ns: new (wanted) state.
59 static int cl_wide_st_chg(struct drbd_conf *mdev,
60 union drbd_state os, union drbd_state ns)
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);
71 static union drbd_state
72 apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val)
75 ns.i = (os.i & ~mask.i) | val.i;
80 drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f,
81 union drbd_state mask, union drbd_state val)
85 enum drbd_state_rv rv;
87 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
88 ns = apply_mask_val(mdev->state, mask, val);
89 rv = _drbd_set_state(mdev, ns, f, NULL);
91 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
97 * drbd_force_state() - Impose a change which happens outside our control on our state
99 * @mask: mask of state bits to change.
100 * @val: value of new state bits.
102 void drbd_force_state(struct drbd_conf *mdev,
103 union drbd_state mask, union drbd_state val)
105 drbd_change_state(mdev, CS_HARD, mask, val);
108 static enum drbd_state_rv
109 _req_st_cond(struct drbd_conf *mdev, union drbd_state mask,
110 union drbd_state val)
112 union drbd_state os, ns;
114 enum drbd_state_rv rv;
116 if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &mdev->flags))
117 return SS_CW_SUCCESS;
119 if (test_and_clear_bit(CL_ST_CHG_FAIL, &mdev->flags))
120 return SS_CW_FAILED_BY_PEER;
122 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
124 ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
125 rv = is_valid_transition(os, ns);
126 if (rv == SS_SUCCESS)
127 rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
129 if (!cl_wide_st_chg(mdev, os, ns))
131 if (rv == SS_UNKNOWN_ERROR) {
132 rv = is_valid_state(mdev, ns);
133 if (rv == SS_SUCCESS) {
134 rv = is_valid_soft_transition(os, ns);
135 if (rv == SS_SUCCESS)
136 rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
139 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
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.
151 * Should not be called directly, use drbd_request_state() or
152 * _drbd_request_state().
154 static enum drbd_state_rv
155 drbd_req_state(struct drbd_conf *mdev, union drbd_state mask,
156 union drbd_state val, enum chg_state_flags f)
158 struct completion done;
160 union drbd_state os, ns;
161 enum drbd_state_rv rv;
163 init_completion(&done);
165 if (f & CS_SERIALIZE)
166 mutex_lock(mdev->state_mutex);
168 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
170 ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL);
171 rv = is_valid_transition(os, ns);
175 if (cl_wide_st_chg(mdev, os, ns)) {
176 rv = is_valid_state(mdev, ns);
177 if (rv == SS_SUCCESS)
178 rv = is_valid_soft_transition(os, ns);
179 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
181 if (rv < SS_SUCCESS) {
183 print_st_err(mdev, os, ns, rv);
187 if (!drbd_send_state_req(mdev, mask, val)) {
188 rv = SS_CW_FAILED_BY_PEER;
190 print_st_err(mdev, os, ns, rv);
194 wait_event(mdev->state_wait,
195 (rv = _req_st_cond(mdev, mask, val)));
197 if (rv < SS_SUCCESS) {
199 print_st_err(mdev, os, ns, rv);
202 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
203 ns = apply_mask_val(mdev->state, mask, val);
204 rv = _drbd_set_state(mdev, ns, f, &done);
206 rv = _drbd_set_state(mdev, ns, f, &done);
209 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
211 if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) {
212 D_ASSERT(current != mdev->tconn->worker.task);
213 wait_for_completion(&done);
217 if (f & CS_SERIALIZE)
218 mutex_unlock(mdev->state_mutex);
224 * _drbd_request_state() - Request a state change (with flags)
225 * @mdev: DRBD device.
226 * @mask: mask of state bits to change.
227 * @val: value of new state bits.
230 * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE
231 * flag, or when logging of failed state change requests is not desired.
234 _drbd_request_state(struct drbd_conf *mdev, union drbd_state mask,
235 union drbd_state val, enum chg_state_flags f)
237 enum drbd_state_rv rv;
239 wait_event(mdev->state_wait,
240 (rv = drbd_req_state(mdev, mask, val, f)) != SS_IN_TRANSIENT_STATE);
245 static void print_st(struct drbd_conf *mdev, char *name, union drbd_state ns)
247 dev_err(DEV, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n",
249 drbd_conn_str(ns.conn),
250 drbd_role_str(ns.role),
251 drbd_role_str(ns.peer),
252 drbd_disk_str(ns.disk),
253 drbd_disk_str(ns.pdsk),
254 is_susp(ns) ? 's' : 'r',
255 ns.aftr_isp ? 'a' : '-',
256 ns.peer_isp ? 'p' : '-',
257 ns.user_isp ? 'u' : '-',
258 ns.susp_fen ? 'F' : '-',
259 ns.susp_nod ? 'N' : '-'
263 void print_st_err(struct drbd_conf *mdev, union drbd_state os,
264 union drbd_state ns, enum drbd_state_rv err)
266 if (err == SS_IN_TRANSIENT_STATE)
268 dev_err(DEV, "State change failed: %s\n", drbd_set_st_err_str(err));
269 print_st(mdev, " state", os);
270 print_st(mdev, "wanted", ns);
273 static void print_state_change(struct drbd_conf *mdev, union drbd_state os, union drbd_state ns,
274 enum chg_state_flags flags)
279 if (ns.role != os.role)
280 pbp += sprintf(pbp, "role( %s -> %s ) ",
281 drbd_role_str(os.role),
282 drbd_role_str(ns.role));
283 if (ns.peer != os.peer)
284 pbp += sprintf(pbp, "peer( %s -> %s ) ",
285 drbd_role_str(os.peer),
286 drbd_role_str(ns.peer));
287 if (ns.conn != os.conn && !(flags & CS_NO_CSTATE_CHG))
288 pbp += sprintf(pbp, "conn( %s -> %s ) ",
289 drbd_conn_str(os.conn),
290 drbd_conn_str(ns.conn));
291 if (ns.disk != os.disk)
292 pbp += sprintf(pbp, "disk( %s -> %s ) ",
293 drbd_disk_str(os.disk),
294 drbd_disk_str(ns.disk));
295 if (ns.pdsk != os.pdsk)
296 pbp += sprintf(pbp, "pdsk( %s -> %s ) ",
297 drbd_disk_str(os.pdsk),
298 drbd_disk_str(ns.pdsk));
299 if (is_susp(ns) != is_susp(os))
300 pbp += sprintf(pbp, "susp( %d -> %d ) ",
303 if (ns.aftr_isp != os.aftr_isp)
304 pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ",
307 if (ns.peer_isp != os.peer_isp)
308 pbp += sprintf(pbp, "peer_isp( %d -> %d ) ",
311 if (ns.user_isp != os.user_isp)
312 pbp += sprintf(pbp, "user_isp( %d -> %d ) ",
316 dev_info(DEV, "%s\n", pb);
320 * is_valid_state() - Returns an SS_ error code if ns is not valid
321 * @mdev: DRBD device.
322 * @ns: State to consider.
324 static enum drbd_state_rv
325 is_valid_state(struct drbd_conf *mdev, union drbd_state ns)
327 /* See drbd_state_sw_errors in drbd_strings.c */
329 enum drbd_fencing_p fp;
330 enum drbd_state_rv rv = SS_SUCCESS;
333 if (get_ldev(mdev)) {
334 fp = mdev->ldev->dc.fencing;
338 if (get_net_conf(mdev->tconn)) {
339 if (!mdev->tconn->net_conf->two_primaries &&
340 ns.role == R_PRIMARY && ns.peer == R_PRIMARY)
341 rv = SS_TWO_PRIMARIES;
342 put_net_conf(mdev->tconn);
346 /* already found a reason to abort */;
347 else if (ns.role == R_SECONDARY && mdev->open_cnt)
348 rv = SS_DEVICE_IN_USE;
350 else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE)
351 rv = SS_NO_UP_TO_DATE_DISK;
353 else if (fp >= FP_RESOURCE &&
354 ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN)
357 else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT)
358 rv = SS_NO_UP_TO_DATE_DISK;
360 else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT)
361 rv = SS_NO_LOCAL_DISK;
363 else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT)
364 rv = SS_NO_REMOTE_DISK;
366 else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
367 rv = SS_NO_UP_TO_DATE_DISK;
369 else if ((ns.conn == C_CONNECTED ||
370 ns.conn == C_WF_BITMAP_S ||
371 ns.conn == C_SYNC_SOURCE ||
372 ns.conn == C_PAUSED_SYNC_S) &&
373 ns.disk == D_OUTDATED)
374 rv = SS_CONNECTED_OUTDATES;
376 else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
377 (mdev->sync_conf.verify_alg[0] == 0))
378 rv = SS_NO_VERIFY_ALG;
380 else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
381 mdev->tconn->agreed_pro_version < 88)
382 rv = SS_NOT_SUPPORTED;
384 else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN)
385 rv = SS_CONNECTED_OUTDATES;
391 * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible
392 * This function limits state transitions that may be declined by DRBD. I.e.
393 * user requests (aka soft transitions).
394 * @mdev: DRBD device.
398 static enum drbd_state_rv
399 is_valid_soft_transition(union drbd_state os, union drbd_state ns)
401 enum drbd_state_rv rv = SS_SUCCESS;
403 if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
404 os.conn > C_CONNECTED)
405 rv = SS_RESYNC_RUNNING;
407 if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
408 rv = SS_ALREADY_STANDALONE;
410 if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
413 if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
414 rv = SS_NO_NET_CONFIG;
416 if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
417 rv = SS_LOWER_THAN_OUTDATED;
419 if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
420 rv = SS_IN_TRANSIENT_STATE;
422 if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS)
423 rv = SS_IN_TRANSIENT_STATE;
425 if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
426 rv = SS_NEED_CONNECTION;
428 if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
429 ns.conn != os.conn && os.conn > C_CONNECTED)
430 rv = SS_RESYNC_RUNNING;
432 if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
433 os.conn < C_CONNECTED)
434 rv = SS_NEED_CONNECTION;
436 if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)
437 && os.conn < C_WF_REPORT_PARAMS)
438 rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */
443 static enum drbd_state_rv
444 is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc)
446 enum drbd_state_rv rv = SS_SUCCESS;
448 /* Disallow Network errors to configure a device's network part */
449 if ((nc >= C_TIMEOUT && nc <= C_TEAR_DOWN) && oc <= C_DISCONNECTING)
450 rv = SS_NEED_CONNECTION;
452 /* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */
453 if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING)
454 rv = SS_IN_TRANSIENT_STATE;
456 /* After C_DISCONNECTING only C_STANDALONE may follow */
457 if (oc == C_DISCONNECTING && nc != C_STANDALONE)
458 rv = SS_IN_TRANSIENT_STATE;
465 * is_valid_transition() - Returns an SS_ error code if the state transition is not possible
466 * This limits hard state transitions. Hard state transitions are facts there are
467 * imposed on DRBD by the environment. E.g. disk broke or network broke down.
468 * But those hard state transitions are still not allowed to do everything.
472 static enum drbd_state_rv
473 is_valid_transition(union drbd_state os, union drbd_state ns)
475 enum drbd_state_rv rv;
477 rv = is_valid_conn_transition(os.conn, ns.conn);
479 /* we cannot fail (again) if we already detached */
480 if (ns.disk == D_FAILED && os.disk == D_DISKLESS)
483 /* if we are only D_ATTACHING yet,
484 * we can (and should) go directly to D_DISKLESS. */
485 if (ns.disk == D_FAILED && os.disk == D_ATTACHING) {
486 printk("TODO: FIX ME\n");
494 * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
495 * @mdev: DRBD device.
500 * When we loose connection, we have to set the state of the peers disk (pdsk)
501 * to D_UNKNOWN. This rule and many more along those lines are in this function.
503 static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns,
504 const char **warn_sync_abort)
506 enum drbd_fencing_p fp;
507 enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max;
510 if (get_ldev(mdev)) {
511 fp = mdev->ldev->dc.fencing;
515 /* Implications from connection to peer and peer_isp */
516 if (ns.conn < C_CONNECTED) {
519 if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
523 /* Clear the aftr_isp when becoming unconfigured */
524 if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
527 /* An implication of the disk states onto the connection state */
528 /* Abort resync if a disk fails/detaches */
529 if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
532 ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ?
533 "Online-verify" : "Resync";
534 ns.conn = C_CONNECTED;
537 /* Connection breaks down before we finished "Negotiating" */
538 if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
539 get_ldev_if_state(mdev, D_NEGOTIATING)) {
540 if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) {
541 ns.disk = mdev->new_state_tmp.disk;
542 ns.pdsk = mdev->new_state_tmp.pdsk;
544 dev_alert(DEV, "Connection lost while negotiating, no data!\n");
545 ns.disk = D_DISKLESS;
551 /* D_CONSISTENT and D_OUTDATED vanish when we get connected */
552 if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) {
553 if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED)
554 ns.disk = D_UP_TO_DATE;
555 if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)
556 ns.pdsk = D_UP_TO_DATE;
559 /* Implications of the connection stat on the disk states */
560 disk_min = D_DISKLESS;
561 disk_max = D_UP_TO_DATE;
562 pdsk_min = D_INCONSISTENT;
563 pdsk_max = D_UNKNOWN;
564 switch ((enum drbd_conns)ns.conn) {
566 case C_PAUSED_SYNC_T:
567 case C_STARTING_SYNC_T:
570 disk_min = D_INCONSISTENT;
571 disk_max = D_OUTDATED;
572 pdsk_min = D_UP_TO_DATE;
573 pdsk_max = D_UP_TO_DATE;
577 disk_min = D_UP_TO_DATE;
578 disk_max = D_UP_TO_DATE;
579 pdsk_min = D_UP_TO_DATE;
580 pdsk_max = D_UP_TO_DATE;
583 disk_min = D_DISKLESS;
584 disk_max = D_UP_TO_DATE;
585 pdsk_min = D_DISKLESS;
586 pdsk_max = D_UP_TO_DATE;
589 case C_PAUSED_SYNC_S:
590 case C_STARTING_SYNC_S:
592 disk_min = D_UP_TO_DATE;
593 disk_max = D_UP_TO_DATE;
594 pdsk_min = D_INCONSISTENT;
595 pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/
598 disk_min = D_INCONSISTENT;
599 disk_max = D_INCONSISTENT;
600 pdsk_min = D_UP_TO_DATE;
601 pdsk_max = D_UP_TO_DATE;
604 disk_min = D_UP_TO_DATE;
605 disk_max = D_UP_TO_DATE;
606 pdsk_min = D_INCONSISTENT;
607 pdsk_max = D_INCONSISTENT;
610 case C_DISCONNECTING:
614 case C_NETWORK_FAILURE:
615 case C_PROTOCOL_ERROR:
617 case C_WF_CONNECTION:
618 case C_WF_REPORT_PARAMS:
622 if (ns.disk > disk_max)
625 if (ns.disk < disk_min) {
626 dev_warn(DEV, "Implicitly set disk from %s to %s\n",
627 drbd_disk_str(ns.disk), drbd_disk_str(disk_min));
630 if (ns.pdsk > pdsk_max)
633 if (ns.pdsk < pdsk_min) {
634 dev_warn(DEV, "Implicitly set pdsk from %s to %s\n",
635 drbd_disk_str(ns.pdsk), drbd_disk_str(pdsk_min));
639 if (fp == FP_STONITH &&
640 (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED))
641 ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */
643 if (mdev->sync_conf.on_no_data == OND_SUSPEND_IO &&
644 (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
645 ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */
647 if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
648 if (ns.conn == C_SYNC_SOURCE)
649 ns.conn = C_PAUSED_SYNC_S;
650 if (ns.conn == C_SYNC_TARGET)
651 ns.conn = C_PAUSED_SYNC_T;
653 if (ns.conn == C_PAUSED_SYNC_S)
654 ns.conn = C_SYNC_SOURCE;
655 if (ns.conn == C_PAUSED_SYNC_T)
656 ns.conn = C_SYNC_TARGET;
662 void drbd_resume_al(struct drbd_conf *mdev)
664 if (test_and_clear_bit(AL_SUSPENDED, &mdev->flags))
665 dev_info(DEV, "Resumed AL updates\n");
668 /* helper for __drbd_set_state */
669 static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs)
671 if (mdev->tconn->agreed_pro_version < 90)
672 mdev->ov_start_sector = 0;
673 mdev->rs_total = drbd_bm_bits(mdev);
674 mdev->ov_position = 0;
675 if (cs == C_VERIFY_T) {
676 /* starting online verify from an arbitrary position
677 * does not fit well into the existing protocol.
678 * on C_VERIFY_T, we initialize ov_left and friends
679 * implicitly in receive_DataRequest once the
680 * first P_OV_REQUEST is received */
681 mdev->ov_start_sector = ~(sector_t)0;
683 unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector);
684 if (bit >= mdev->rs_total) {
685 mdev->ov_start_sector =
686 BM_BIT_TO_SECT(mdev->rs_total - 1);
689 mdev->rs_total -= bit;
690 mdev->ov_position = mdev->ov_start_sector;
692 mdev->ov_left = mdev->rs_total;
696 * __drbd_set_state() - Set a new DRBD state
697 * @mdev: DRBD device.
700 * @done: Optional completion, that will get completed after the after_state_ch() finished
702 * Caller needs to hold req_lock, and global_state_lock. Do not call directly.
705 __drbd_set_state(struct drbd_conf *mdev, union drbd_state ns,
706 enum chg_state_flags flags, struct completion *done)
709 enum drbd_state_rv rv = SS_SUCCESS;
710 const char *warn_sync_abort = NULL;
711 struct after_state_chg_work *ascw;
715 ns = sanitize_state(mdev, ns, &warn_sync_abort);
717 return SS_NOTHING_TO_DO;
719 rv = is_valid_transition(os, ns);
723 if (!(flags & CS_HARD)) {
724 /* pre-state-change checks ; only look at ns */
725 /* See drbd_state_sw_errors in drbd_strings.c */
727 rv = is_valid_state(mdev, ns);
728 if (rv < SS_SUCCESS) {
729 /* If the old state was illegal as well, then let
732 if (is_valid_state(mdev, os) == rv)
733 rv = is_valid_soft_transition(os, ns);
735 rv = is_valid_soft_transition(os, ns);
738 if (rv < SS_SUCCESS) {
739 if (flags & CS_VERBOSE)
740 print_st_err(mdev, os, ns, rv);
745 dev_warn(DEV, "%s aborted.\n", warn_sync_abort);
747 print_state_change(mdev, os, ns, flags);
749 /* solve the race between becoming unconfigured,
750 * worker doing the cleanup, and
751 * admin reconfiguring us:
752 * on (re)configure, first set CONFIG_PENDING,
753 * then wait for a potentially exiting worker,
754 * start the worker, and schedule one no_op.
755 * then proceed with configuration.
757 if (ns.disk == D_DISKLESS &&
758 ns.conn == C_STANDALONE &&
759 ns.role == R_SECONDARY &&
760 !test_and_set_bit(CONFIG_PENDING, &mdev->flags))
761 set_bit(DEVICE_DYING, &mdev->flags);
763 /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
764 * on the ldev here, to be sure the transition -> D_DISKLESS resp.
765 * drbd_ldev_destroy() won't happen before our corresponding
766 * after_state_ch works run, where we put_ldev again. */
767 if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
768 (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
769 atomic_inc(&mdev->local_cnt);
773 if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
774 drbd_print_uuids(mdev, "attached to UUIDs");
776 wake_up(&mdev->misc_wait);
777 wake_up(&mdev->state_wait);
778 wake_up(&mdev->tconn->ping_wait);
780 /* aborted verify run. log the last position */
781 if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
782 ns.conn < C_CONNECTED) {
783 mdev->ov_start_sector =
784 BM_BIT_TO_SECT(drbd_bm_bits(mdev) - mdev->ov_left);
785 dev_info(DEV, "Online Verify reached sector %llu\n",
786 (unsigned long long)mdev->ov_start_sector);
789 if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
790 (ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)) {
791 dev_info(DEV, "Syncer continues.\n");
792 mdev->rs_paused += (long)jiffies
793 -(long)mdev->rs_mark_time[mdev->rs_last_mark];
794 if (ns.conn == C_SYNC_TARGET)
795 mod_timer(&mdev->resync_timer, jiffies);
798 if ((os.conn == C_SYNC_TARGET || os.conn == C_SYNC_SOURCE) &&
799 (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
800 dev_info(DEV, "Resync suspended\n");
801 mdev->rs_mark_time[mdev->rs_last_mark] = jiffies;
804 if (os.conn == C_CONNECTED &&
805 (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
806 unsigned long now = jiffies;
809 set_ov_position(mdev, ns.conn);
810 mdev->rs_start = now;
811 mdev->rs_last_events = 0;
812 mdev->rs_last_sect_ev = 0;
813 mdev->ov_last_oos_size = 0;
814 mdev->ov_last_oos_start = 0;
816 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
817 mdev->rs_mark_left[i] = mdev->ov_left;
818 mdev->rs_mark_time[i] = now;
821 drbd_rs_controller_reset(mdev);
823 if (ns.conn == C_VERIFY_S) {
824 dev_info(DEV, "Starting Online Verify from sector %llu\n",
825 (unsigned long long)mdev->ov_position);
826 mod_timer(&mdev->resync_timer, jiffies);
830 if (get_ldev(mdev)) {
831 u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
832 MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
833 MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
835 if (test_bit(CRASHED_PRIMARY, &mdev->flags))
836 mdf |= MDF_CRASHED_PRIMARY;
837 if (mdev->state.role == R_PRIMARY ||
838 (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY))
839 mdf |= MDF_PRIMARY_IND;
840 if (mdev->state.conn > C_WF_REPORT_PARAMS)
841 mdf |= MDF_CONNECTED_IND;
842 if (mdev->state.disk > D_INCONSISTENT)
843 mdf |= MDF_CONSISTENT;
844 if (mdev->state.disk > D_OUTDATED)
845 mdf |= MDF_WAS_UP_TO_DATE;
846 if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT)
847 mdf |= MDF_PEER_OUT_DATED;
848 if (mdf != mdev->ldev->md.flags) {
849 mdev->ldev->md.flags = mdf;
850 drbd_md_mark_dirty(mdev);
852 if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
853 drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]);
857 /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
858 if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
859 os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
860 set_bit(CONSIDER_RESYNC, &mdev->flags);
862 /* Receiver should clean up itself */
863 if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
864 drbd_thread_stop_nowait(&mdev->tconn->receiver);
866 /* Now the receiver finished cleaning up itself, it should die */
867 if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
868 drbd_thread_stop_nowait(&mdev->tconn->receiver);
870 /* Upon network failure, we need to restart the receiver. */
871 if (os.conn > C_TEAR_DOWN &&
872 ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
873 drbd_thread_restart_nowait(&mdev->tconn->receiver);
875 /* Resume AL writing if we get a connection */
876 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
877 drbd_resume_al(mdev);
879 ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
884 ascw->w.cb = w_after_state_ch;
887 drbd_queue_work(&mdev->tconn->data.work, &ascw->w);
889 dev_err(DEV, "Could not kmalloc an ascw\n");
895 static int w_after_state_ch(struct drbd_work *w, int unused)
897 struct after_state_chg_work *ascw =
898 container_of(w, struct after_state_chg_work, w);
899 struct drbd_conf *mdev = w->mdev;
901 after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags);
902 if (ascw->flags & CS_WAIT_COMPLETE) {
903 D_ASSERT(ascw->done != NULL);
904 complete(ascw->done);
911 static void abw_start_sync(struct drbd_conf *mdev, int rv)
914 dev_err(DEV, "Writing the bitmap failed not starting resync.\n");
915 _drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE);
919 switch (mdev->state.conn) {
920 case C_STARTING_SYNC_T:
921 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
923 case C_STARTING_SYNC_S:
924 drbd_start_resync(mdev, C_SYNC_SOURCE);
929 int drbd_bitmap_io_from_worker(struct drbd_conf *mdev,
930 int (*io_fn)(struct drbd_conf *),
931 char *why, enum bm_flag flags)
935 D_ASSERT(current == mdev->tconn->worker.task);
937 /* open coded non-blocking drbd_suspend_io(mdev); */
938 set_bit(SUSPEND_IO, &mdev->flags);
940 drbd_bm_lock(mdev, why, flags);
942 drbd_bm_unlock(mdev);
944 drbd_resume_io(mdev);
950 * after_state_ch() - Perform after state change actions that may sleep
951 * @mdev: DRBD device.
956 static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
957 union drbd_state ns, enum chg_state_flags flags)
959 enum drbd_fencing_p fp;
960 enum drbd_req_event what = NOTHING;
961 union drbd_state nsm = (union drbd_state){ .i = -1 };
963 if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) {
964 clear_bit(CRASHED_PRIMARY, &mdev->flags);
966 mdev->p_uuid[UI_FLAGS] &= ~((u64)2);
970 if (get_ldev(mdev)) {
971 fp = mdev->ldev->dc.fencing;
975 /* Inform userspace about the change... */
976 drbd_bcast_state(mdev, ns);
978 if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
979 (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
980 drbd_khelper(mdev, "pri-on-incon-degr");
982 /* Here we have the actions that are performed after a
983 state change. This function might sleep */
987 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
990 if (os.disk == D_ATTACHING && ns.disk > D_ATTACHING)
991 what = RESTART_FROZEN_DISK_IO;
998 /* case1: The outdate peer handler is successful: */
999 if (os.pdsk > D_OUTDATED && ns.pdsk <= D_OUTDATED) {
1001 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1002 drbd_uuid_new_current(mdev);
1003 clear_bit(NEW_CUR_UUID, &mdev->flags);
1005 spin_lock_irq(&mdev->tconn->req_lock);
1006 _drbd_set_state(_NS(mdev, susp_fen, 0), CS_VERBOSE, NULL);
1007 spin_unlock_irq(&mdev->tconn->req_lock);
1009 /* case2: The connection was established again: */
1010 if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) {
1011 clear_bit(NEW_CUR_UUID, &mdev->flags);
1017 if (what != NOTHING) {
1018 spin_lock_irq(&mdev->tconn->req_lock);
1019 _tl_restart(mdev, what);
1020 nsm.i &= mdev->state.i;
1021 _drbd_set_state(mdev, nsm, CS_VERBOSE, NULL);
1022 spin_unlock_irq(&mdev->tconn->req_lock);
1025 /* Became sync source. With protocol >= 96, we still need to send out
1026 * the sync uuid now. Need to do that before any drbd_send_state, or
1027 * the other side may go "paused sync" before receiving the sync uuids,
1028 * which is unexpected. */
1029 if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
1030 (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
1031 mdev->tconn->agreed_pro_version >= 96 && get_ldev(mdev)) {
1032 drbd_gen_and_send_sync_uuid(mdev);
1036 /* Do not change the order of the if above and the two below... */
1037 if (os.pdsk == D_DISKLESS && ns.pdsk > D_DISKLESS) { /* attach on the peer */
1038 drbd_send_uuids(mdev);
1039 drbd_send_state(mdev);
1041 /* No point in queuing send_bitmap if we don't have a connection
1042 * anymore, so check also the _current_ state, not only the new state
1043 * at the time this work was queued. */
1044 if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
1045 mdev->state.conn == C_WF_BITMAP_S)
1046 drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL,
1047 "send_bitmap (WFBitMapS)",
1048 BM_LOCKED_TEST_ALLOWED);
1050 /* Lost contact to peer's copy of the data */
1051 if ((os.pdsk >= D_INCONSISTENT &&
1052 os.pdsk != D_UNKNOWN &&
1053 os.pdsk != D_OUTDATED)
1054 && (ns.pdsk < D_INCONSISTENT ||
1055 ns.pdsk == D_UNKNOWN ||
1056 ns.pdsk == D_OUTDATED)) {
1057 if (get_ldev(mdev)) {
1058 if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
1059 mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
1060 if (is_susp(mdev->state)) {
1061 set_bit(NEW_CUR_UUID, &mdev->flags);
1063 drbd_uuid_new_current(mdev);
1064 drbd_send_uuids(mdev);
1071 if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) {
1072 if (ns.peer == R_PRIMARY && mdev->ldev->md.uuid[UI_BITMAP] == 0) {
1073 drbd_uuid_new_current(mdev);
1074 drbd_send_uuids(mdev);
1077 /* D_DISKLESS Peer becomes secondary */
1078 if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
1079 /* We may still be Primary ourselves.
1080 * No harm done if the bitmap still changes,
1081 * redirtied pages will follow later. */
1082 drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1083 "demote diskless peer", BM_LOCKED_SET_ALLOWED);
1087 /* Write out all changed bits on demote.
1088 * Though, no need to da that just yet
1089 * if there is a resync going on still */
1090 if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
1091 mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) {
1092 /* No changes to the bitmap expected this time, so assert that,
1093 * even though no harm was done if it did change. */
1094 drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
1095 "demote", BM_LOCKED_TEST_ALLOWED);
1099 /* Last part of the attaching process ... */
1100 if (ns.conn >= C_CONNECTED &&
1101 os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
1102 drbd_send_sizes(mdev, 0, 0); /* to start sync... */
1103 drbd_send_uuids(mdev);
1104 drbd_send_state(mdev);
1107 /* We want to pause/continue resync, tell peer. */
1108 if (ns.conn >= C_CONNECTED &&
1109 ((os.aftr_isp != ns.aftr_isp) ||
1110 (os.user_isp != ns.user_isp)))
1111 drbd_send_state(mdev);
1113 /* In case one of the isp bits got set, suspend other devices. */
1114 if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
1115 (ns.aftr_isp || ns.peer_isp || ns.user_isp))
1116 suspend_other_sg(mdev);
1118 /* Make sure the peer gets informed about eventual state
1119 changes (ISP bits) while we were in WFReportParams. */
1120 if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
1121 drbd_send_state(mdev);
1123 if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
1124 drbd_send_state(mdev);
1126 /* We are in the progress to start a full sync... */
1127 if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
1128 (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
1129 /* no other bitmap changes expected during this phase */
1130 drbd_queue_bitmap_io(mdev,
1131 &drbd_bmio_set_n_write, &abw_start_sync,
1132 "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED);
1134 /* We are invalidating our self... */
1135 if (os.conn < C_CONNECTED && ns.conn < C_CONNECTED &&
1136 os.disk > D_INCONSISTENT && ns.disk == D_INCONSISTENT)
1137 /* other bitmap operation expected during this phase */
1138 drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, NULL,
1139 "set_n_write from invalidate", BM_LOCKED_MASK);
1141 /* first half of local IO error, failure to attach,
1142 * or administrative detach */
1143 if (os.disk != D_FAILED && ns.disk == D_FAILED) {
1144 enum drbd_io_error_p eh;
1146 /* corresponding get_ldev was in __drbd_set_state, to serialize
1147 * our cleanup here with the transition to D_DISKLESS,
1148 * so it is safe to dreference ldev here. */
1149 eh = mdev->ldev->dc.on_io_error;
1150 was_io_error = test_and_clear_bit(WAS_IO_ERROR, &mdev->flags);
1152 /* current state still has to be D_FAILED,
1153 * there is only one way out: to D_DISKLESS,
1154 * and that may only happen after our put_ldev below. */
1155 if (mdev->state.disk != D_FAILED)
1157 "ASSERT FAILED: disk is %s during detach\n",
1158 drbd_disk_str(mdev->state.disk));
1160 if (drbd_send_state(mdev))
1161 dev_warn(DEV, "Notified peer that I am detaching my disk\n");
1163 dev_err(DEV, "Sending state for detaching disk failed\n");
1165 drbd_rs_cancel_all(mdev);
1167 /* In case we want to get something to stable storage still,
1168 * this may be the last chance.
1169 * Following put_ldev may transition to D_DISKLESS. */
1173 if (was_io_error && eh == EP_CALL_HELPER)
1174 drbd_khelper(mdev, "local-io-error");
1177 /* second half of local IO error, failure to attach,
1178 * or administrative detach,
1179 * after local_cnt references have reached zero again */
1180 if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
1181 /* We must still be diskless,
1182 * re-attach has to be serialized with this! */
1183 if (mdev->state.disk != D_DISKLESS)
1185 "ASSERT FAILED: disk is %s while going diskless\n",
1186 drbd_disk_str(mdev->state.disk));
1189 mdev->rs_failed = 0;
1190 atomic_set(&mdev->rs_pending_cnt, 0);
1192 if (drbd_send_state(mdev))
1193 dev_warn(DEV, "Notified peer that I'm now diskless.\n");
1194 /* corresponding get_ldev in __drbd_set_state
1195 * this may finally trigger drbd_ldev_destroy. */
1199 /* Notify peer that I had a local IO error, and did not detached.. */
1200 if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT)
1201 drbd_send_state(mdev);
1203 /* Disks got bigger while they were detached */
1204 if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
1205 test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) {
1206 if (ns.conn == C_CONNECTED)
1207 resync_after_online_grow(mdev);
1210 /* A resync finished or aborted, wake paused devices... */
1211 if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
1212 (os.peer_isp && !ns.peer_isp) ||
1213 (os.user_isp && !ns.user_isp))
1214 resume_next_sg(mdev);
1216 /* sync target done with resync. Explicitly notify peer, even though
1217 * it should (at least for non-empty resyncs) already know itself. */
1218 if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
1219 drbd_send_state(mdev);
1221 /* This triggers bitmap writeout of potentially still unwritten pages
1222 * if the resync finished cleanly, or aborted because of peer disk
1223 * failure, or because of connection loss.
1224 * For resync aborted because of local disk failure, we cannot do
1225 * any bitmap writeout anymore.
1226 * No harm done if some bits change during this phase.
1228 if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) {
1229 drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL,
1230 "write from resync_finished", BM_LOCKED_SET_ALLOWED);
1234 if (ns.disk == D_DISKLESS &&
1235 ns.conn == C_STANDALONE &&
1236 ns.role == R_SECONDARY) {
1237 if (os.aftr_isp != ns.aftr_isp)
1238 resume_next_sg(mdev);
1241 after_all_state_ch(mdev->tconn, ns);
1246 struct after_conn_state_chg_work {
1249 union drbd_state nms; /* new, max state, over all mdevs */
1250 enum chg_state_flags flags;
1253 static void after_all_state_ch(struct drbd_tconn *tconn, union drbd_state ns)
1255 if (ns.disk == D_DISKLESS && ns.conn == C_STANDALONE && ns.role == R_SECONDARY) {
1256 /* if (test_bit(DEVICE_DYING, &mdev->flags)) TODO: DEVICE_DYING functionality */
1257 drbd_thread_stop_nowait(&tconn->worker);
1261 static int w_after_conn_state_ch(struct drbd_work *w, int unused)
1263 struct after_conn_state_chg_work *acscw =
1264 container_of(w, struct after_conn_state_chg_work, w);
1265 struct drbd_tconn *tconn = w->tconn;
1266 enum drbd_conns oc = acscw->oc;
1267 union drbd_state nms = acscw->nms;
1271 /* Upon network configuration, we need to start the receiver */
1272 if (oc == C_STANDALONE && nms.conn == C_UNCONNECTED)
1273 drbd_thread_start(&tconn->receiver);
1275 //conn_err(tconn, STATE_FMT, STATE_ARGS("nms", nms));
1276 after_all_state_ch(tconn, nms);
1281 static void print_conn_state_change(struct drbd_tconn *tconn, enum drbd_conns oc, enum drbd_conns nc)
1287 pbp += sprintf(pbp, "conn( %s -> %s ) ",
1291 conn_info(tconn, "%s\n", pb);
1294 struct _is_valid_itr_params {
1295 enum chg_state_flags flags;
1296 union drbd_state mask, val;
1297 union drbd_state ms; /* maximal state, over all mdevs */
1306 static int _is_valid_itr_fn(int vnr, void *p, void *data)
1308 struct drbd_conf *mdev = (struct drbd_conf *)p;
1309 struct _is_valid_itr_params *params = (struct _is_valid_itr_params *)data;
1310 enum chg_state_flags flags = params->flags;
1311 union drbd_state ns, os;
1312 enum drbd_state_rv rv;
1315 ns = apply_mask_val(os, params->mask, params->val);
1316 ns = sanitize_state(mdev, ns, NULL);
1317 rv = is_valid_state(mdev, ns);
1319 if (rv < SS_SUCCESS) {
1320 /* If the old state was illegal as well, then let this happen...*/
1322 if (is_valid_state(mdev, os) == rv)
1323 rv = is_valid_soft_transition(os, ns);
1325 rv = is_valid_soft_transition(os, ns);
1327 switch (params->oc_state) {
1328 case OC_UNINITIALIZED:
1329 params->oc = os.conn;
1330 params->oc_state = OC_CONSISTENT;
1333 if (params->oc != os.conn)
1334 params->oc_state = OC_INCONSISTENT;
1336 case OC_INCONSISTENT:
1340 if (rv < SS_SUCCESS) {
1341 if (flags & CS_VERBOSE)
1342 print_st_err(mdev, os, ns, rv);
1348 static int _set_state_itr_fn(int vnr, void *p, void *data)
1350 struct drbd_conf *mdev = (struct drbd_conf *)p;
1351 struct _is_valid_itr_params *params = (struct _is_valid_itr_params *)data;
1352 enum chg_state_flags flags = params->flags;
1353 union drbd_state os, ns, ms = params->ms;
1354 enum drbd_state_rv rv;
1357 ns = apply_mask_val(os, params->mask, params->val);
1358 ns = sanitize_state(mdev, ns, NULL);
1360 rv = __drbd_set_state(mdev, ns, flags, NULL);
1362 ms.role = max_t(enum drbd_role, mdev->state.role, ms.role);
1363 ms.peer = max_t(enum drbd_role, mdev->state.peer, ms.peer);
1364 ms.disk = max_t(enum drbd_role, mdev->state.disk, ms.disk);
1365 ms.pdsk = max_t(enum drbd_role, mdev->state.pdsk, ms.pdsk);
1372 _conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1373 enum chg_state_flags flags)
1375 enum drbd_state_rv rv = SS_SUCCESS;
1376 struct _is_valid_itr_params params;
1377 struct after_conn_state_chg_work *acscw;
1378 enum drbd_conns oc = tconn->cstate;
1380 read_lock(&global_state_lock);
1382 rv = is_valid_conn_transition(oc, val.conn);
1383 if (rv < SS_SUCCESS)
1386 params.flags = flags;
1389 params.oc_state = OC_UNINITIALIZED;
1391 if (!(flags & CS_HARD))
1392 rv = idr_for_each(&tconn->volumes, _is_valid_itr_fn, ¶ms);
1394 if (rv == 0) /* idr_for_each semantics */
1397 if (rv < SS_SUCCESS)
1400 if (params.oc_state == OC_CONSISTENT) {
1402 print_conn_state_change(tconn, oc, val.conn);
1403 params.flags |= CS_NO_CSTATE_CHG;
1405 tconn->cstate = val.conn;
1407 params.ms.conn = val.conn;
1408 idr_for_each(&tconn->volumes, _set_state_itr_fn, ¶ms);
1410 acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC);
1413 acscw->nms = params.ms;
1414 acscw->flags = flags;
1415 acscw->w.cb = w_after_conn_state_ch;
1416 acscw->w.tconn = tconn;
1417 drbd_queue_work(&tconn->data.work, &acscw->w);
1419 conn_err(tconn, "Could not kmalloc an acscw\n");
1423 read_unlock(&global_state_lock);
1429 conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val,
1430 enum chg_state_flags flags)
1432 enum drbd_state_rv rv;
1434 spin_lock_irq(&tconn->req_lock);
1435 rv = _conn_request_state(tconn, mask, val, flags);
1436 spin_unlock_irq(&tconn->req_lock);