Commit | Line | Data |
---|---|---|
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 | ||
32 | struct 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 | 40 | static int w_after_state_ch(struct drbd_work *w, int unused); |
b8907339 PR |
41 | static void after_state_ch(struct drbd_conf *mdev, union drbd_state os, |
42 | union drbd_state ns, enum chg_state_flags flags); | |
0e29d163 | 43 | static void after_all_state_ch(struct drbd_tconn *tconn); |
a75f34ad PR |
44 | static enum drbd_state_rv is_valid_state(struct drbd_conf *, union drbd_state); |
45 | static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state); | |
3509502d | 46 | static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns); |
4308a0a3 PR |
47 | static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns, |
48 | const char **warn_sync_abort); | |
b8907339 | 49 | |
d0456c72 | 50 | bool conn_all_vols_unconf(struct drbd_tconn *tconn) |
0e29d163 PR |
51 | { |
52 | struct drbd_conf *mdev; | |
e90285e0 | 53 | int vnr; |
0e29d163 | 54 | |
e90285e0 | 55 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
d0456c72 LE |
56 | if (mdev->state.disk != D_DISKLESS || |
57 | mdev->state.conn != C_STANDALONE || | |
58 | mdev->state.role != R_SECONDARY) | |
59 | return false; | |
0e29d163 | 60 | } |
d0456c72 | 61 | return true; |
0e29d163 PR |
62 | } |
63 | ||
b8907339 PR |
64 | /** |
65 | * cl_wide_st_chg() - true if the state change is a cluster wide one | |
66 | * @mdev: DRBD device. | |
67 | * @os: old (current) state. | |
68 | * @ns: new (wanted) state. | |
69 | */ | |
70 | static int cl_wide_st_chg(struct drbd_conf *mdev, | |
71 | union drbd_state os, union drbd_state ns) | |
72 | { | |
73 | return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED && | |
74 | ((os.role != R_PRIMARY && ns.role == R_PRIMARY) || | |
75 | (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) || | |
76 | (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) || | |
77 | (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))) || | |
78 | (os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) || | |
79 | (os.conn == C_CONNECTED && ns.conn == C_VERIFY_S); | |
80 | } | |
81 | ||
56707f9e PR |
82 | static union drbd_state |
83 | apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val) | |
84 | { | |
85 | union drbd_state ns; | |
86 | ns.i = (os.i & ~mask.i) | val.i; | |
87 | return ns; | |
88 | } | |
89 | ||
b8907339 PR |
90 | enum drbd_state_rv |
91 | drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f, | |
92 | union drbd_state mask, union drbd_state val) | |
93 | { | |
94 | unsigned long flags; | |
56707f9e | 95 | union drbd_state ns; |
b8907339 PR |
96 | enum drbd_state_rv rv; |
97 | ||
98 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); | |
56707f9e | 99 | ns = apply_mask_val(mdev->state, mask, val); |
b8907339 PR |
100 | rv = _drbd_set_state(mdev, ns, f, NULL); |
101 | ns = mdev->state; | |
102 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
103 | ||
104 | return rv; | |
105 | } | |
106 | ||
107 | /** | |
108 | * drbd_force_state() - Impose a change which happens outside our control on our state | |
109 | * @mdev: DRBD device. | |
110 | * @mask: mask of state bits to change. | |
111 | * @val: value of new state bits. | |
112 | */ | |
113 | void drbd_force_state(struct drbd_conf *mdev, | |
114 | union drbd_state mask, union drbd_state val) | |
115 | { | |
116 | drbd_change_state(mdev, CS_HARD, mask, val); | |
117 | } | |
118 | ||
119 | static enum drbd_state_rv | |
120 | _req_st_cond(struct drbd_conf *mdev, union drbd_state mask, | |
121 | union drbd_state val) | |
122 | { | |
123 | union drbd_state os, ns; | |
124 | unsigned long flags; | |
125 | enum drbd_state_rv rv; | |
126 | ||
127 | if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &mdev->flags)) | |
128 | return SS_CW_SUCCESS; | |
129 | ||
130 | if (test_and_clear_bit(CL_ST_CHG_FAIL, &mdev->flags)) | |
131 | return SS_CW_FAILED_BY_PEER; | |
132 | ||
b8907339 PR |
133 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
134 | os = mdev->state; | |
56707f9e | 135 | ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL); |
3509502d PR |
136 | rv = is_valid_transition(os, ns); |
137 | if (rv == SS_SUCCESS) | |
138 | rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */ | |
b8907339 PR |
139 | |
140 | if (!cl_wide_st_chg(mdev, os, ns)) | |
141 | rv = SS_CW_NO_NEED; | |
3509502d | 142 | if (rv == SS_UNKNOWN_ERROR) { |
b8907339 PR |
143 | rv = is_valid_state(mdev, ns); |
144 | if (rv == SS_SUCCESS) { | |
a75f34ad | 145 | rv = is_valid_soft_transition(os, ns); |
b8907339 PR |
146 | if (rv == SS_SUCCESS) |
147 | rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */ | |
148 | } | |
149 | } | |
150 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
151 | ||
152 | return rv; | |
153 | } | |
154 | ||
155 | /** | |
156 | * drbd_req_state() - Perform an eventually cluster wide state change | |
157 | * @mdev: DRBD device. | |
158 | * @mask: mask of state bits to change. | |
159 | * @val: value of new state bits. | |
160 | * @f: flags | |
161 | * | |
162 | * Should not be called directly, use drbd_request_state() or | |
163 | * _drbd_request_state(). | |
164 | */ | |
165 | static enum drbd_state_rv | |
166 | drbd_req_state(struct drbd_conf *mdev, union drbd_state mask, | |
167 | union drbd_state val, enum chg_state_flags f) | |
168 | { | |
169 | struct completion done; | |
170 | unsigned long flags; | |
171 | union drbd_state os, ns; | |
172 | enum drbd_state_rv rv; | |
173 | ||
174 | init_completion(&done); | |
175 | ||
176 | if (f & CS_SERIALIZE) | |
8410da8f | 177 | mutex_lock(mdev->state_mutex); |
b8907339 PR |
178 | |
179 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); | |
180 | os = mdev->state; | |
56707f9e | 181 | ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL); |
3509502d | 182 | rv = is_valid_transition(os, ns); |
3c5e5f6a LE |
183 | if (rv < SS_SUCCESS) { |
184 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
3509502d | 185 | goto abort; |
3c5e5f6a | 186 | } |
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 | ||
229 | abort: | |
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 | */ | |
246 | enum 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 | ||
258 | static 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 | ||
276 | void 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 |
286 | static 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 | 331 | |
047e95e2 PR |
332 | static bool vol_has_primary_peer(struct drbd_tconn *tconn) |
333 | { | |
334 | struct drbd_conf *mdev; | |
e90285e0 | 335 | int vnr; |
047e95e2 | 336 | |
e90285e0 | 337 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
047e95e2 PR |
338 | if (mdev->state.peer == R_PRIMARY) |
339 | return true; | |
340 | } | |
341 | return false; | |
342 | } | |
343 | ||
b8907339 PR |
344 | /** |
345 | * is_valid_state() - Returns an SS_ error code if ns is not valid | |
346 | * @mdev: DRBD device. | |
347 | * @ns: State to consider. | |
348 | */ | |
349 | static enum drbd_state_rv | |
350 | is_valid_state(struct drbd_conf *mdev, union drbd_state ns) | |
351 | { | |
352 | /* See drbd_state_sw_errors in drbd_strings.c */ | |
353 | ||
354 | enum drbd_fencing_p fp; | |
355 | enum drbd_state_rv rv = SS_SUCCESS; | |
356 | ||
357 | fp = FP_DONT_CARE; | |
358 | if (get_ldev(mdev)) { | |
359 | fp = mdev->ldev->dc.fencing; | |
360 | put_ldev(mdev); | |
361 | } | |
362 | ||
363 | if (get_net_conf(mdev->tconn)) { | |
047e95e2 PR |
364 | if (!mdev->tconn->net_conf->two_primaries && ns.role == R_PRIMARY) { |
365 | if (ns.peer == R_PRIMARY) | |
366 | rv = SS_TWO_PRIMARIES; | |
367 | else if (vol_has_primary_peer(mdev->tconn)) | |
368 | rv = SS_O_VOL_PEER_PRI; | |
369 | } | |
b8907339 PR |
370 | put_net_conf(mdev->tconn); |
371 | } | |
372 | ||
373 | if (rv <= 0) | |
374 | /* already found a reason to abort */; | |
375 | else if (ns.role == R_SECONDARY && mdev->open_cnt) | |
376 | rv = SS_DEVICE_IN_USE; | |
377 | ||
378 | else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE) | |
379 | rv = SS_NO_UP_TO_DATE_DISK; | |
380 | ||
381 | else if (fp >= FP_RESOURCE && | |
382 | ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN) | |
383 | rv = SS_PRIMARY_NOP; | |
384 | ||
385 | else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT) | |
386 | rv = SS_NO_UP_TO_DATE_DISK; | |
387 | ||
388 | else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT) | |
389 | rv = SS_NO_LOCAL_DISK; | |
390 | ||
391 | else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT) | |
392 | rv = SS_NO_REMOTE_DISK; | |
393 | ||
394 | else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE) | |
395 | rv = SS_NO_UP_TO_DATE_DISK; | |
396 | ||
397 | else if ((ns.conn == C_CONNECTED || | |
398 | ns.conn == C_WF_BITMAP_S || | |
399 | ns.conn == C_SYNC_SOURCE || | |
400 | ns.conn == C_PAUSED_SYNC_S) && | |
401 | ns.disk == D_OUTDATED) | |
402 | rv = SS_CONNECTED_OUTDATES; | |
403 | ||
404 | else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
f399002e | 405 | (mdev->tconn->net_conf->verify_alg[0] == 0)) |
b8907339 PR |
406 | rv = SS_NO_VERIFY_ALG; |
407 | ||
408 | else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
409 | mdev->tconn->agreed_pro_version < 88) | |
410 | rv = SS_NOT_SUPPORTED; | |
411 | ||
412 | else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN) | |
413 | rv = SS_CONNECTED_OUTDATES; | |
414 | ||
415 | return rv; | |
416 | } | |
417 | ||
418 | /** | |
a75f34ad | 419 | * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible |
3509502d PR |
420 | * This function limits state transitions that may be declined by DRBD. I.e. |
421 | * user requests (aka soft transitions). | |
b8907339 PR |
422 | * @mdev: DRBD device. |
423 | * @ns: new state. | |
424 | * @os: old state. | |
425 | */ | |
426 | static enum drbd_state_rv | |
a75f34ad | 427 | is_valid_soft_transition(union drbd_state os, union drbd_state ns) |
b8907339 PR |
428 | { |
429 | enum drbd_state_rv rv = SS_SUCCESS; | |
430 | ||
431 | if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) && | |
432 | os.conn > C_CONNECTED) | |
433 | rv = SS_RESYNC_RUNNING; | |
434 | ||
435 | if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE) | |
436 | rv = SS_ALREADY_STANDALONE; | |
437 | ||
438 | if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS) | |
439 | rv = SS_IS_DISKLESS; | |
440 | ||
441 | if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED) | |
442 | rv = SS_NO_NET_CONFIG; | |
443 | ||
444 | if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING) | |
445 | rv = SS_LOWER_THAN_OUTDATED; | |
446 | ||
447 | if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED) | |
448 | rv = SS_IN_TRANSIENT_STATE; | |
449 | ||
2325eb66 PR |
450 | /* if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS) |
451 | rv = SS_IN_TRANSIENT_STATE; */ | |
b8907339 PR |
452 | |
453 | if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED) | |
454 | rv = SS_NEED_CONNECTION; | |
455 | ||
456 | if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
457 | ns.conn != os.conn && os.conn > C_CONNECTED) | |
458 | rv = SS_RESYNC_RUNNING; | |
459 | ||
460 | if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) && | |
461 | os.conn < C_CONNECTED) | |
462 | rv = SS_NEED_CONNECTION; | |
463 | ||
464 | if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE) | |
465 | && os.conn < C_WF_REPORT_PARAMS) | |
466 | rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */ | |
467 | ||
468 | return rv; | |
469 | } | |
470 | ||
3509502d | 471 | static enum drbd_state_rv |
fda74117 | 472 | is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc) |
3509502d PR |
473 | { |
474 | enum drbd_state_rv rv = SS_SUCCESS; | |
475 | ||
476 | /* Disallow Network errors to configure a device's network part */ | |
fda74117 | 477 | if ((nc >= C_TIMEOUT && nc <= C_TEAR_DOWN) && oc <= C_DISCONNECTING) |
3509502d PR |
478 | rv = SS_NEED_CONNECTION; |
479 | ||
480 | /* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */ | |
fda74117 | 481 | if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING) |
3509502d PR |
482 | rv = SS_IN_TRANSIENT_STATE; |
483 | ||
484 | /* After C_DISCONNECTING only C_STANDALONE may follow */ | |
fda74117 | 485 | if (oc == C_DISCONNECTING && nc != C_STANDALONE) |
3509502d PR |
486 | rv = SS_IN_TRANSIENT_STATE; |
487 | ||
fda74117 PR |
488 | return rv; |
489 | } | |
490 | ||
491 | ||
492 | /** | |
493 | * is_valid_transition() - Returns an SS_ error code if the state transition is not possible | |
494 | * This limits hard state transitions. Hard state transitions are facts there are | |
495 | * imposed on DRBD by the environment. E.g. disk broke or network broke down. | |
496 | * But those hard state transitions are still not allowed to do everything. | |
497 | * @ns: new state. | |
498 | * @os: old state. | |
499 | */ | |
500 | static enum drbd_state_rv | |
501 | is_valid_transition(union drbd_state os, union drbd_state ns) | |
502 | { | |
503 | enum drbd_state_rv rv; | |
504 | ||
505 | rv = is_valid_conn_transition(os.conn, ns.conn); | |
506 | ||
3509502d PR |
507 | /* we cannot fail (again) if we already detached */ |
508 | if (ns.disk == D_FAILED && os.disk == D_DISKLESS) | |
509 | rv = SS_IS_DISKLESS; | |
510 | ||
4308a0a3 PR |
511 | /* if we are only D_ATTACHING yet, |
512 | * we can (and should) go directly to D_DISKLESS. */ | |
513 | if (ns.disk == D_FAILED && os.disk == D_ATTACHING) { | |
514 | printk("TODO: FIX ME\n"); | |
515 | rv = SS_IS_DISKLESS; | |
516 | } | |
517 | ||
3509502d PR |
518 | return rv; |
519 | } | |
520 | ||
b8907339 PR |
521 | /** |
522 | * sanitize_state() - Resolves implicitly necessary additional changes to a state transition | |
523 | * @mdev: DRBD device. | |
524 | * @os: old state. | |
525 | * @ns: new state. | |
526 | * @warn_sync_abort: | |
527 | * | |
528 | * When we loose connection, we have to set the state of the peers disk (pdsk) | |
529 | * to D_UNKNOWN. This rule and many more along those lines are in this function. | |
530 | */ | |
4308a0a3 PR |
531 | static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns, |
532 | const char **warn_sync_abort) | |
b8907339 PR |
533 | { |
534 | enum drbd_fencing_p fp; | |
535 | enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max; | |
536 | ||
537 | fp = FP_DONT_CARE; | |
538 | if (get_ldev(mdev)) { | |
539 | fp = mdev->ldev->dc.fencing; | |
540 | put_ldev(mdev); | |
541 | } | |
542 | ||
3509502d | 543 | /* Implications from connection to peer and peer_isp */ |
b8907339 PR |
544 | if (ns.conn < C_CONNECTED) { |
545 | ns.peer_isp = 0; | |
546 | ns.peer = R_UNKNOWN; | |
547 | if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT) | |
548 | ns.pdsk = D_UNKNOWN; | |
549 | } | |
550 | ||
551 | /* Clear the aftr_isp when becoming unconfigured */ | |
552 | if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY) | |
553 | ns.aftr_isp = 0; | |
554 | ||
4308a0a3 | 555 | /* An implication of the disk states onto the connection state */ |
b8907339 | 556 | /* Abort resync if a disk fails/detaches */ |
4308a0a3 | 557 | if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) { |
b8907339 PR |
558 | if (warn_sync_abort) |
559 | *warn_sync_abort = | |
4308a0a3 | 560 | ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ? |
b8907339 PR |
561 | "Online-verify" : "Resync"; |
562 | ns.conn = C_CONNECTED; | |
563 | } | |
564 | ||
565 | /* Connection breaks down before we finished "Negotiating" */ | |
566 | if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING && | |
567 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
568 | if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) { | |
569 | ns.disk = mdev->new_state_tmp.disk; | |
570 | ns.pdsk = mdev->new_state_tmp.pdsk; | |
571 | } else { | |
572 | dev_alert(DEV, "Connection lost while negotiating, no data!\n"); | |
573 | ns.disk = D_DISKLESS; | |
574 | ns.pdsk = D_UNKNOWN; | |
575 | } | |
576 | put_ldev(mdev); | |
577 | } | |
578 | ||
579 | /* D_CONSISTENT and D_OUTDATED vanish when we get connected */ | |
580 | if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) { | |
581 | if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED) | |
582 | ns.disk = D_UP_TO_DATE; | |
583 | if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED) | |
584 | ns.pdsk = D_UP_TO_DATE; | |
585 | } | |
586 | ||
587 | /* Implications of the connection stat on the disk states */ | |
588 | disk_min = D_DISKLESS; | |
589 | disk_max = D_UP_TO_DATE; | |
590 | pdsk_min = D_INCONSISTENT; | |
591 | pdsk_max = D_UNKNOWN; | |
592 | switch ((enum drbd_conns)ns.conn) { | |
593 | case C_WF_BITMAP_T: | |
594 | case C_PAUSED_SYNC_T: | |
595 | case C_STARTING_SYNC_T: | |
596 | case C_WF_SYNC_UUID: | |
597 | case C_BEHIND: | |
598 | disk_min = D_INCONSISTENT; | |
599 | disk_max = D_OUTDATED; | |
600 | pdsk_min = D_UP_TO_DATE; | |
601 | pdsk_max = D_UP_TO_DATE; | |
602 | break; | |
603 | case C_VERIFY_S: | |
604 | case C_VERIFY_T: | |
605 | disk_min = D_UP_TO_DATE; | |
606 | disk_max = D_UP_TO_DATE; | |
607 | pdsk_min = D_UP_TO_DATE; | |
608 | pdsk_max = D_UP_TO_DATE; | |
609 | break; | |
610 | case C_CONNECTED: | |
611 | disk_min = D_DISKLESS; | |
612 | disk_max = D_UP_TO_DATE; | |
613 | pdsk_min = D_DISKLESS; | |
614 | pdsk_max = D_UP_TO_DATE; | |
615 | break; | |
616 | case C_WF_BITMAP_S: | |
617 | case C_PAUSED_SYNC_S: | |
618 | case C_STARTING_SYNC_S: | |
619 | case C_AHEAD: | |
620 | disk_min = D_UP_TO_DATE; | |
621 | disk_max = D_UP_TO_DATE; | |
622 | pdsk_min = D_INCONSISTENT; | |
623 | pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/ | |
624 | break; | |
625 | case C_SYNC_TARGET: | |
626 | disk_min = D_INCONSISTENT; | |
627 | disk_max = D_INCONSISTENT; | |
628 | pdsk_min = D_UP_TO_DATE; | |
629 | pdsk_max = D_UP_TO_DATE; | |
630 | break; | |
631 | case C_SYNC_SOURCE: | |
632 | disk_min = D_UP_TO_DATE; | |
633 | disk_max = D_UP_TO_DATE; | |
634 | pdsk_min = D_INCONSISTENT; | |
635 | pdsk_max = D_INCONSISTENT; | |
636 | break; | |
637 | case C_STANDALONE: | |
638 | case C_DISCONNECTING: | |
639 | case C_UNCONNECTED: | |
640 | case C_TIMEOUT: | |
641 | case C_BROKEN_PIPE: | |
642 | case C_NETWORK_FAILURE: | |
643 | case C_PROTOCOL_ERROR: | |
644 | case C_TEAR_DOWN: | |
645 | case C_WF_CONNECTION: | |
646 | case C_WF_REPORT_PARAMS: | |
647 | case C_MASK: | |
648 | break; | |
649 | } | |
650 | if (ns.disk > disk_max) | |
651 | ns.disk = disk_max; | |
652 | ||
653 | if (ns.disk < disk_min) { | |
654 | dev_warn(DEV, "Implicitly set disk from %s to %s\n", | |
655 | drbd_disk_str(ns.disk), drbd_disk_str(disk_min)); | |
656 | ns.disk = disk_min; | |
657 | } | |
658 | if (ns.pdsk > pdsk_max) | |
659 | ns.pdsk = pdsk_max; | |
660 | ||
661 | if (ns.pdsk < pdsk_min) { | |
662 | dev_warn(DEV, "Implicitly set pdsk from %s to %s\n", | |
663 | drbd_disk_str(ns.pdsk), drbd_disk_str(pdsk_min)); | |
664 | ns.pdsk = pdsk_min; | |
665 | } | |
666 | ||
667 | if (fp == FP_STONITH && | |
4308a0a3 | 668 | (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED)) |
b8907339 PR |
669 | ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */ |
670 | ||
f399002e | 671 | if (mdev->tconn->res_opts.on_no_data == OND_SUSPEND_IO && |
4308a0a3 | 672 | (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)) |
b8907339 PR |
673 | ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */ |
674 | ||
675 | if (ns.aftr_isp || ns.peer_isp || ns.user_isp) { | |
676 | if (ns.conn == C_SYNC_SOURCE) | |
677 | ns.conn = C_PAUSED_SYNC_S; | |
678 | if (ns.conn == C_SYNC_TARGET) | |
679 | ns.conn = C_PAUSED_SYNC_T; | |
680 | } else { | |
681 | if (ns.conn == C_PAUSED_SYNC_S) | |
682 | ns.conn = C_SYNC_SOURCE; | |
683 | if (ns.conn == C_PAUSED_SYNC_T) | |
684 | ns.conn = C_SYNC_TARGET; | |
685 | } | |
686 | ||
687 | return ns; | |
688 | } | |
689 | ||
690 | void drbd_resume_al(struct drbd_conf *mdev) | |
691 | { | |
692 | if (test_and_clear_bit(AL_SUSPENDED, &mdev->flags)) | |
693 | dev_info(DEV, "Resumed AL updates\n"); | |
694 | } | |
695 | ||
696 | /* helper for __drbd_set_state */ | |
697 | static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs) | |
698 | { | |
699 | if (mdev->tconn->agreed_pro_version < 90) | |
700 | mdev->ov_start_sector = 0; | |
701 | mdev->rs_total = drbd_bm_bits(mdev); | |
702 | mdev->ov_position = 0; | |
703 | if (cs == C_VERIFY_T) { | |
704 | /* starting online verify from an arbitrary position | |
705 | * does not fit well into the existing protocol. | |
706 | * on C_VERIFY_T, we initialize ov_left and friends | |
707 | * implicitly in receive_DataRequest once the | |
708 | * first P_OV_REQUEST is received */ | |
709 | mdev->ov_start_sector = ~(sector_t)0; | |
710 | } else { | |
711 | unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector); | |
712 | if (bit >= mdev->rs_total) { | |
713 | mdev->ov_start_sector = | |
714 | BM_BIT_TO_SECT(mdev->rs_total - 1); | |
715 | mdev->rs_total = 1; | |
716 | } else | |
717 | mdev->rs_total -= bit; | |
718 | mdev->ov_position = mdev->ov_start_sector; | |
719 | } | |
720 | mdev->ov_left = mdev->rs_total; | |
721 | } | |
722 | ||
723 | /** | |
724 | * __drbd_set_state() - Set a new DRBD state | |
725 | * @mdev: DRBD device. | |
726 | * @ns: new state. | |
727 | * @flags: Flags | |
728 | * @done: Optional completion, that will get completed after the after_state_ch() finished | |
729 | * | |
730 | * Caller needs to hold req_lock, and global_state_lock. Do not call directly. | |
731 | */ | |
732 | enum drbd_state_rv | |
733 | __drbd_set_state(struct drbd_conf *mdev, union drbd_state ns, | |
734 | enum chg_state_flags flags, struct completion *done) | |
735 | { | |
736 | union drbd_state os; | |
737 | enum drbd_state_rv rv = SS_SUCCESS; | |
738 | const char *warn_sync_abort = NULL; | |
739 | struct after_state_chg_work *ascw; | |
740 | ||
741 | os = mdev->state; | |
742 | ||
4308a0a3 | 743 | ns = sanitize_state(mdev, ns, &warn_sync_abort); |
b8907339 PR |
744 | if (ns.i == os.i) |
745 | return SS_NOTHING_TO_DO; | |
746 | ||
3509502d PR |
747 | rv = is_valid_transition(os, ns); |
748 | if (rv < SS_SUCCESS) | |
749 | return rv; | |
750 | ||
b8907339 PR |
751 | if (!(flags & CS_HARD)) { |
752 | /* pre-state-change checks ; only look at ns */ | |
753 | /* See drbd_state_sw_errors in drbd_strings.c */ | |
754 | ||
755 | rv = is_valid_state(mdev, ns); | |
756 | if (rv < SS_SUCCESS) { | |
757 | /* If the old state was illegal as well, then let | |
758 | this happen...*/ | |
759 | ||
760 | if (is_valid_state(mdev, os) == rv) | |
a75f34ad | 761 | rv = is_valid_soft_transition(os, ns); |
b8907339 | 762 | } else |
a75f34ad | 763 | rv = is_valid_soft_transition(os, ns); |
b8907339 PR |
764 | } |
765 | ||
766 | if (rv < SS_SUCCESS) { | |
767 | if (flags & CS_VERBOSE) | |
768 | print_st_err(mdev, os, ns, rv); | |
769 | return rv; | |
770 | } | |
771 | ||
772 | if (warn_sync_abort) | |
773 | dev_warn(DEV, "%s aborted.\n", warn_sync_abort); | |
774 | ||
bbeb641c | 775 | print_state_change(mdev, os, ns, flags); |
b8907339 | 776 | |
b8907339 PR |
777 | /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference |
778 | * on the ldev here, to be sure the transition -> D_DISKLESS resp. | |
779 | * drbd_ldev_destroy() won't happen before our corresponding | |
780 | * after_state_ch works run, where we put_ldev again. */ | |
781 | if ((os.disk != D_FAILED && ns.disk == D_FAILED) || | |
782 | (os.disk != D_DISKLESS && ns.disk == D_DISKLESS)) | |
783 | atomic_inc(&mdev->local_cnt); | |
784 | ||
785 | mdev->state = ns; | |
786 | ||
0e29d163 PR |
787 | /* solve the race between becoming unconfigured, |
788 | * worker doing the cleanup, and | |
789 | * admin reconfiguring us: | |
790 | * on (re)configure, first set CONFIG_PENDING, | |
791 | * then wait for a potentially exiting worker, | |
792 | * start the worker, and schedule one no_op. | |
793 | * then proceed with configuration. | |
794 | */ | |
795 | if(conn_all_vols_unconf(mdev->tconn) && | |
796 | !test_and_set_bit(CONFIG_PENDING, &mdev->tconn->flags)) | |
797 | set_bit(OBJECT_DYING, &mdev->tconn->flags); | |
798 | ||
b8907339 PR |
799 | if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING) |
800 | drbd_print_uuids(mdev, "attached to UUIDs"); | |
801 | ||
802 | wake_up(&mdev->misc_wait); | |
803 | wake_up(&mdev->state_wait); | |
2a67d8b9 | 804 | wake_up(&mdev->tconn->ping_wait); |
b8907339 PR |
805 | |
806 | /* aborted verify run. log the last position */ | |
807 | if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) && | |
808 | ns.conn < C_CONNECTED) { | |
809 | mdev->ov_start_sector = | |
810 | BM_BIT_TO_SECT(drbd_bm_bits(mdev) - mdev->ov_left); | |
811 | dev_info(DEV, "Online Verify reached sector %llu\n", | |
812 | (unsigned long long)mdev->ov_start_sector); | |
813 | } | |
814 | ||
815 | if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) && | |
816 | (ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)) { | |
817 | dev_info(DEV, "Syncer continues.\n"); | |
818 | mdev->rs_paused += (long)jiffies | |
819 | -(long)mdev->rs_mark_time[mdev->rs_last_mark]; | |
820 | if (ns.conn == C_SYNC_TARGET) | |
821 | mod_timer(&mdev->resync_timer, jiffies); | |
822 | } | |
823 | ||
824 | if ((os.conn == C_SYNC_TARGET || os.conn == C_SYNC_SOURCE) && | |
825 | (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) { | |
826 | dev_info(DEV, "Resync suspended\n"); | |
827 | mdev->rs_mark_time[mdev->rs_last_mark] = jiffies; | |
828 | } | |
829 | ||
830 | if (os.conn == C_CONNECTED && | |
831 | (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) { | |
832 | unsigned long now = jiffies; | |
833 | int i; | |
834 | ||
835 | set_ov_position(mdev, ns.conn); | |
836 | mdev->rs_start = now; | |
837 | mdev->rs_last_events = 0; | |
838 | mdev->rs_last_sect_ev = 0; | |
839 | mdev->ov_last_oos_size = 0; | |
840 | mdev->ov_last_oos_start = 0; | |
841 | ||
842 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { | |
843 | mdev->rs_mark_left[i] = mdev->ov_left; | |
844 | mdev->rs_mark_time[i] = now; | |
845 | } | |
846 | ||
847 | drbd_rs_controller_reset(mdev); | |
848 | ||
849 | if (ns.conn == C_VERIFY_S) { | |
850 | dev_info(DEV, "Starting Online Verify from sector %llu\n", | |
851 | (unsigned long long)mdev->ov_position); | |
852 | mod_timer(&mdev->resync_timer, jiffies); | |
853 | } | |
854 | } | |
855 | ||
856 | if (get_ldev(mdev)) { | |
857 | u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND| | |
858 | MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE| | |
859 | MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY); | |
860 | ||
861 | if (test_bit(CRASHED_PRIMARY, &mdev->flags)) | |
862 | mdf |= MDF_CRASHED_PRIMARY; | |
863 | if (mdev->state.role == R_PRIMARY || | |
864 | (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY)) | |
865 | mdf |= MDF_PRIMARY_IND; | |
866 | if (mdev->state.conn > C_WF_REPORT_PARAMS) | |
867 | mdf |= MDF_CONNECTED_IND; | |
868 | if (mdev->state.disk > D_INCONSISTENT) | |
869 | mdf |= MDF_CONSISTENT; | |
870 | if (mdev->state.disk > D_OUTDATED) | |
871 | mdf |= MDF_WAS_UP_TO_DATE; | |
872 | if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT) | |
873 | mdf |= MDF_PEER_OUT_DATED; | |
874 | if (mdf != mdev->ldev->md.flags) { | |
875 | mdev->ldev->md.flags = mdf; | |
876 | drbd_md_mark_dirty(mdev); | |
877 | } | |
878 | if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT) | |
879 | drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]); | |
880 | put_ldev(mdev); | |
881 | } | |
882 | ||
883 | /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */ | |
884 | if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT && | |
885 | os.peer == R_SECONDARY && ns.peer == R_PRIMARY) | |
886 | set_bit(CONSIDER_RESYNC, &mdev->flags); | |
887 | ||
888 | /* Receiver should clean up itself */ | |
889 | if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING) | |
890 | drbd_thread_stop_nowait(&mdev->tconn->receiver); | |
891 | ||
892 | /* Now the receiver finished cleaning up itself, it should die */ | |
893 | if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE) | |
894 | drbd_thread_stop_nowait(&mdev->tconn->receiver); | |
895 | ||
896 | /* Upon network failure, we need to restart the receiver. */ | |
897 | if (os.conn > C_TEAR_DOWN && | |
898 | ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT) | |
899 | drbd_thread_restart_nowait(&mdev->tconn->receiver); | |
900 | ||
901 | /* Resume AL writing if we get a connection */ | |
902 | if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) | |
903 | drbd_resume_al(mdev); | |
904 | ||
905 | ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC); | |
906 | if (ascw) { | |
907 | ascw->os = os; | |
908 | ascw->ns = ns; | |
909 | ascw->flags = flags; | |
910 | ascw->w.cb = w_after_state_ch; | |
a21e9298 | 911 | ascw->w.mdev = mdev; |
b8907339 PR |
912 | ascw->done = done; |
913 | drbd_queue_work(&mdev->tconn->data.work, &ascw->w); | |
914 | } else { | |
bbeb641c | 915 | dev_err(DEV, "Could not kmalloc an ascw\n"); |
b8907339 PR |
916 | } |
917 | ||
918 | return rv; | |
919 | } | |
920 | ||
00d56944 | 921 | static int w_after_state_ch(struct drbd_work *w, int unused) |
b8907339 PR |
922 | { |
923 | struct after_state_chg_work *ascw = | |
924 | container_of(w, struct after_state_chg_work, w); | |
00d56944 | 925 | struct drbd_conf *mdev = w->mdev; |
b8907339 PR |
926 | |
927 | after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags); | |
928 | if (ascw->flags & CS_WAIT_COMPLETE) { | |
929 | D_ASSERT(ascw->done != NULL); | |
930 | complete(ascw->done); | |
931 | } | |
932 | kfree(ascw); | |
933 | ||
934 | return 1; | |
935 | } | |
936 | ||
937 | static void abw_start_sync(struct drbd_conf *mdev, int rv) | |
938 | { | |
939 | if (rv) { | |
940 | dev_err(DEV, "Writing the bitmap failed not starting resync.\n"); | |
941 | _drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE); | |
942 | return; | |
943 | } | |
944 | ||
945 | switch (mdev->state.conn) { | |
946 | case C_STARTING_SYNC_T: | |
947 | _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); | |
948 | break; | |
949 | case C_STARTING_SYNC_S: | |
950 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
951 | break; | |
952 | } | |
953 | } | |
954 | ||
955 | int drbd_bitmap_io_from_worker(struct drbd_conf *mdev, | |
956 | int (*io_fn)(struct drbd_conf *), | |
957 | char *why, enum bm_flag flags) | |
958 | { | |
959 | int rv; | |
960 | ||
961 | D_ASSERT(current == mdev->tconn->worker.task); | |
962 | ||
963 | /* open coded non-blocking drbd_suspend_io(mdev); */ | |
964 | set_bit(SUSPEND_IO, &mdev->flags); | |
965 | ||
966 | drbd_bm_lock(mdev, why, flags); | |
967 | rv = io_fn(mdev); | |
968 | drbd_bm_unlock(mdev); | |
969 | ||
970 | drbd_resume_io(mdev); | |
971 | ||
972 | return rv; | |
973 | } | |
974 | ||
975 | /** | |
976 | * after_state_ch() - Perform after state change actions that may sleep | |
977 | * @mdev: DRBD device. | |
978 | * @os: old state. | |
979 | * @ns: new state. | |
980 | * @flags: Flags | |
981 | */ | |
982 | static void after_state_ch(struct drbd_conf *mdev, union drbd_state os, | |
983 | union drbd_state ns, enum chg_state_flags flags) | |
984 | { | |
985 | enum drbd_fencing_p fp; | |
986 | enum drbd_req_event what = NOTHING; | |
987 | union drbd_state nsm = (union drbd_state){ .i = -1 }; | |
3b98c0c2 LE |
988 | struct sib_info sib; |
989 | ||
990 | sib.sib_reason = SIB_STATE_CHANGE; | |
991 | sib.os = os; | |
992 | sib.ns = ns; | |
b8907339 PR |
993 | |
994 | if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) { | |
995 | clear_bit(CRASHED_PRIMARY, &mdev->flags); | |
996 | if (mdev->p_uuid) | |
997 | mdev->p_uuid[UI_FLAGS] &= ~((u64)2); | |
998 | } | |
999 | ||
1000 | fp = FP_DONT_CARE; | |
1001 | if (get_ldev(mdev)) { | |
1002 | fp = mdev->ldev->dc.fencing; | |
1003 | put_ldev(mdev); | |
1004 | } | |
1005 | ||
1006 | /* Inform userspace about the change... */ | |
3b98c0c2 | 1007 | drbd_bcast_event(mdev, &sib); |
b8907339 PR |
1008 | |
1009 | if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) && | |
1010 | (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)) | |
1011 | drbd_khelper(mdev, "pri-on-incon-degr"); | |
1012 | ||
1013 | /* Here we have the actions that are performed after a | |
1014 | state change. This function might sleep */ | |
1015 | ||
1016 | nsm.i = -1; | |
1017 | if (ns.susp_nod) { | |
1018 | if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) | |
1019 | what = RESEND; | |
1020 | ||
1021 | if (os.disk == D_ATTACHING && ns.disk > D_ATTACHING) | |
1022 | what = RESTART_FROZEN_DISK_IO; | |
1023 | ||
1024 | if (what != NOTHING) | |
1025 | nsm.susp_nod = 0; | |
1026 | } | |
1027 | ||
1028 | if (ns.susp_fen) { | |
1029 | /* case1: The outdate peer handler is successful: */ | |
1030 | if (os.pdsk > D_OUTDATED && ns.pdsk <= D_OUTDATED) { | |
2f5cdd0b | 1031 | tl_clear(mdev->tconn); |
b8907339 PR |
1032 | if (test_bit(NEW_CUR_UUID, &mdev->flags)) { |
1033 | drbd_uuid_new_current(mdev); | |
1034 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
1035 | } | |
1036 | spin_lock_irq(&mdev->tconn->req_lock); | |
1037 | _drbd_set_state(_NS(mdev, susp_fen, 0), CS_VERBOSE, NULL); | |
1038 | spin_unlock_irq(&mdev->tconn->req_lock); | |
1039 | } | |
1040 | /* case2: The connection was established again: */ | |
1041 | if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) { | |
1042 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
1043 | what = RESEND; | |
1044 | nsm.susp_fen = 0; | |
1045 | } | |
1046 | } | |
1047 | ||
1048 | if (what != NOTHING) { | |
1049 | spin_lock_irq(&mdev->tconn->req_lock); | |
2f5cdd0b | 1050 | _tl_restart(mdev->tconn, what); |
b8907339 PR |
1051 | nsm.i &= mdev->state.i; |
1052 | _drbd_set_state(mdev, nsm, CS_VERBOSE, NULL); | |
1053 | spin_unlock_irq(&mdev->tconn->req_lock); | |
1054 | } | |
1055 | ||
1056 | /* Became sync source. With protocol >= 96, we still need to send out | |
1057 | * the sync uuid now. Need to do that before any drbd_send_state, or | |
1058 | * the other side may go "paused sync" before receiving the sync uuids, | |
1059 | * which is unexpected. */ | |
1060 | if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) && | |
1061 | (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) && | |
1062 | mdev->tconn->agreed_pro_version >= 96 && get_ldev(mdev)) { | |
1063 | drbd_gen_and_send_sync_uuid(mdev); | |
1064 | put_ldev(mdev); | |
1065 | } | |
1066 | ||
1067 | /* Do not change the order of the if above and the two below... */ | |
1068 | if (os.pdsk == D_DISKLESS && ns.pdsk > D_DISKLESS) { /* attach on the peer */ | |
1069 | drbd_send_uuids(mdev); | |
1070 | drbd_send_state(mdev); | |
1071 | } | |
1072 | /* No point in queuing send_bitmap if we don't have a connection | |
1073 | * anymore, so check also the _current_ state, not only the new state | |
1074 | * at the time this work was queued. */ | |
1075 | if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S && | |
1076 | mdev->state.conn == C_WF_BITMAP_S) | |
1077 | drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL, | |
1078 | "send_bitmap (WFBitMapS)", | |
1079 | BM_LOCKED_TEST_ALLOWED); | |
1080 | ||
1081 | /* Lost contact to peer's copy of the data */ | |
1082 | if ((os.pdsk >= D_INCONSISTENT && | |
1083 | os.pdsk != D_UNKNOWN && | |
1084 | os.pdsk != D_OUTDATED) | |
1085 | && (ns.pdsk < D_INCONSISTENT || | |
1086 | ns.pdsk == D_UNKNOWN || | |
1087 | ns.pdsk == D_OUTDATED)) { | |
1088 | if (get_ldev(mdev)) { | |
1089 | if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) && | |
1090 | mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) { | |
1091 | if (is_susp(mdev->state)) { | |
1092 | set_bit(NEW_CUR_UUID, &mdev->flags); | |
1093 | } else { | |
1094 | drbd_uuid_new_current(mdev); | |
1095 | drbd_send_uuids(mdev); | |
1096 | } | |
1097 | } | |
1098 | put_ldev(mdev); | |
1099 | } | |
1100 | } | |
1101 | ||
1102 | if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) { | |
1103 | if (ns.peer == R_PRIMARY && mdev->ldev->md.uuid[UI_BITMAP] == 0) { | |
1104 | drbd_uuid_new_current(mdev); | |
1105 | drbd_send_uuids(mdev); | |
1106 | } | |
1107 | ||
1108 | /* D_DISKLESS Peer becomes secondary */ | |
1109 | if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY) | |
1110 | /* We may still be Primary ourselves. | |
1111 | * No harm done if the bitmap still changes, | |
1112 | * redirtied pages will follow later. */ | |
1113 | drbd_bitmap_io_from_worker(mdev, &drbd_bm_write, | |
1114 | "demote diskless peer", BM_LOCKED_SET_ALLOWED); | |
1115 | put_ldev(mdev); | |
1116 | } | |
1117 | ||
1118 | /* Write out all changed bits on demote. | |
1119 | * Though, no need to da that just yet | |
1120 | * if there is a resync going on still */ | |
1121 | if (os.role == R_PRIMARY && ns.role == R_SECONDARY && | |
1122 | mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) { | |
1123 | /* No changes to the bitmap expected this time, so assert that, | |
1124 | * even though no harm was done if it did change. */ | |
1125 | drbd_bitmap_io_from_worker(mdev, &drbd_bm_write, | |
1126 | "demote", BM_LOCKED_TEST_ALLOWED); | |
1127 | put_ldev(mdev); | |
1128 | } | |
1129 | ||
1130 | /* Last part of the attaching process ... */ | |
1131 | if (ns.conn >= C_CONNECTED && | |
1132 | os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) { | |
1133 | drbd_send_sizes(mdev, 0, 0); /* to start sync... */ | |
1134 | drbd_send_uuids(mdev); | |
1135 | drbd_send_state(mdev); | |
1136 | } | |
1137 | ||
1138 | /* We want to pause/continue resync, tell peer. */ | |
1139 | if (ns.conn >= C_CONNECTED && | |
1140 | ((os.aftr_isp != ns.aftr_isp) || | |
1141 | (os.user_isp != ns.user_isp))) | |
1142 | drbd_send_state(mdev); | |
1143 | ||
1144 | /* In case one of the isp bits got set, suspend other devices. */ | |
1145 | if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) && | |
1146 | (ns.aftr_isp || ns.peer_isp || ns.user_isp)) | |
1147 | suspend_other_sg(mdev); | |
1148 | ||
1149 | /* Make sure the peer gets informed about eventual state | |
1150 | changes (ISP bits) while we were in WFReportParams. */ | |
1151 | if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED) | |
1152 | drbd_send_state(mdev); | |
1153 | ||
1154 | if (os.conn != C_AHEAD && ns.conn == C_AHEAD) | |
1155 | drbd_send_state(mdev); | |
1156 | ||
1157 | /* We are in the progress to start a full sync... */ | |
1158 | if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) || | |
1159 | (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S)) | |
1160 | /* no other bitmap changes expected during this phase */ | |
1161 | drbd_queue_bitmap_io(mdev, | |
1162 | &drbd_bmio_set_n_write, &abw_start_sync, | |
1163 | "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED); | |
1164 | ||
1165 | /* We are invalidating our self... */ | |
1166 | if (os.conn < C_CONNECTED && ns.conn < C_CONNECTED && | |
1167 | os.disk > D_INCONSISTENT && ns.disk == D_INCONSISTENT) | |
1168 | /* other bitmap operation expected during this phase */ | |
1169 | drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, NULL, | |
1170 | "set_n_write from invalidate", BM_LOCKED_MASK); | |
1171 | ||
1172 | /* first half of local IO error, failure to attach, | |
1173 | * or administrative detach */ | |
1174 | if (os.disk != D_FAILED && ns.disk == D_FAILED) { | |
1175 | enum drbd_io_error_p eh; | |
1176 | int was_io_error; | |
1177 | /* corresponding get_ldev was in __drbd_set_state, to serialize | |
1178 | * our cleanup here with the transition to D_DISKLESS, | |
1179 | * so it is safe to dreference ldev here. */ | |
1180 | eh = mdev->ldev->dc.on_io_error; | |
1181 | was_io_error = test_and_clear_bit(WAS_IO_ERROR, &mdev->flags); | |
1182 | ||
1183 | /* current state still has to be D_FAILED, | |
1184 | * there is only one way out: to D_DISKLESS, | |
1185 | * and that may only happen after our put_ldev below. */ | |
1186 | if (mdev->state.disk != D_FAILED) | |
1187 | dev_err(DEV, | |
1188 | "ASSERT FAILED: disk is %s during detach\n", | |
1189 | drbd_disk_str(mdev->state.disk)); | |
1190 | ||
927036f9 | 1191 | if (!drbd_send_state(mdev)) |
b8907339 PR |
1192 | dev_warn(DEV, "Notified peer that I am detaching my disk\n"); |
1193 | else | |
1194 | dev_err(DEV, "Sending state for detaching disk failed\n"); | |
1195 | ||
1196 | drbd_rs_cancel_all(mdev); | |
1197 | ||
1198 | /* In case we want to get something to stable storage still, | |
1199 | * this may be the last chance. | |
1200 | * Following put_ldev may transition to D_DISKLESS. */ | |
1201 | drbd_md_sync(mdev); | |
1202 | put_ldev(mdev); | |
1203 | ||
1204 | if (was_io_error && eh == EP_CALL_HELPER) | |
1205 | drbd_khelper(mdev, "local-io-error"); | |
1206 | } | |
1207 | ||
1208 | /* second half of local IO error, failure to attach, | |
1209 | * or administrative detach, | |
1210 | * after local_cnt references have reached zero again */ | |
1211 | if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) { | |
1212 | /* We must still be diskless, | |
1213 | * re-attach has to be serialized with this! */ | |
1214 | if (mdev->state.disk != D_DISKLESS) | |
1215 | dev_err(DEV, | |
1216 | "ASSERT FAILED: disk is %s while going diskless\n", | |
1217 | drbd_disk_str(mdev->state.disk)); | |
1218 | ||
1219 | mdev->rs_total = 0; | |
1220 | mdev->rs_failed = 0; | |
1221 | atomic_set(&mdev->rs_pending_cnt, 0); | |
1222 | ||
927036f9 | 1223 | if (!drbd_send_state(mdev)) |
b8907339 PR |
1224 | dev_warn(DEV, "Notified peer that I'm now diskless.\n"); |
1225 | /* corresponding get_ldev in __drbd_set_state | |
1226 | * this may finally trigger drbd_ldev_destroy. */ | |
1227 | put_ldev(mdev); | |
1228 | } | |
1229 | ||
1230 | /* Notify peer that I had a local IO error, and did not detached.. */ | |
1231 | if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT) | |
1232 | drbd_send_state(mdev); | |
1233 | ||
1234 | /* Disks got bigger while they were detached */ | |
1235 | if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING && | |
1236 | test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) { | |
1237 | if (ns.conn == C_CONNECTED) | |
1238 | resync_after_online_grow(mdev); | |
1239 | } | |
1240 | ||
1241 | /* A resync finished or aborted, wake paused devices... */ | |
1242 | if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) || | |
1243 | (os.peer_isp && !ns.peer_isp) || | |
1244 | (os.user_isp && !ns.user_isp)) | |
1245 | resume_next_sg(mdev); | |
1246 | ||
1247 | /* sync target done with resync. Explicitly notify peer, even though | |
1248 | * it should (at least for non-empty resyncs) already know itself. */ | |
1249 | if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED) | |
1250 | drbd_send_state(mdev); | |
1251 | ||
1252 | /* This triggers bitmap writeout of potentially still unwritten pages | |
1253 | * if the resync finished cleanly, or aborted because of peer disk | |
1254 | * failure, or because of connection loss. | |
1255 | * For resync aborted because of local disk failure, we cannot do | |
1256 | * any bitmap writeout anymore. | |
1257 | * No harm done if some bits change during this phase. | |
1258 | */ | |
1259 | if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) { | |
1260 | drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL, | |
1261 | "write from resync_finished", BM_LOCKED_SET_ALLOWED); | |
1262 | put_ldev(mdev); | |
1263 | } | |
1264 | ||
1265 | if (ns.disk == D_DISKLESS && | |
1266 | ns.conn == C_STANDALONE && | |
1267 | ns.role == R_SECONDARY) { | |
1268 | if (os.aftr_isp != ns.aftr_isp) | |
1269 | resume_next_sg(mdev); | |
1270 | } | |
1271 | ||
0e29d163 | 1272 | after_all_state_ch(mdev->tconn); |
bbeb641c | 1273 | |
b8907339 PR |
1274 | drbd_md_sync(mdev); |
1275 | } | |
1276 | ||
bbeb641c PR |
1277 | struct after_conn_state_chg_work { |
1278 | struct drbd_work w; | |
1279 | enum drbd_conns oc; | |
1280 | union drbd_state nms; /* new, max state, over all mdevs */ | |
1281 | enum chg_state_flags flags; | |
1282 | }; | |
1283 | ||
0e29d163 | 1284 | static void after_all_state_ch(struct drbd_tconn *tconn) |
b8907339 | 1285 | { |
0e29d163 PR |
1286 | if (conn_all_vols_unconf(tconn) && |
1287 | test_bit(OBJECT_DYING, &tconn->flags)) { | |
bbeb641c PR |
1288 | drbd_thread_stop_nowait(&tconn->worker); |
1289 | } | |
1290 | } | |
1291 | ||
1292 | static int w_after_conn_state_ch(struct drbd_work *w, int unused) | |
1293 | { | |
1294 | struct after_conn_state_chg_work *acscw = | |
1295 | container_of(w, struct after_conn_state_chg_work, w); | |
1296 | struct drbd_tconn *tconn = w->tconn; | |
1297 | enum drbd_conns oc = acscw->oc; | |
1298 | union drbd_state nms = acscw->nms; | |
1299 | ||
1300 | kfree(acscw); | |
1301 | ||
b8907339 | 1302 | /* Upon network configuration, we need to start the receiver */ |
bbeb641c | 1303 | if (oc == C_STANDALONE && nms.conn == C_UNCONNECTED) |
b8907339 PR |
1304 | drbd_thread_start(&tconn->receiver); |
1305 | ||
bbeb641c | 1306 | //conn_err(tconn, STATE_FMT, STATE_ARGS("nms", nms)); |
0e29d163 | 1307 | after_all_state_ch(tconn); |
bbeb641c PR |
1308 | |
1309 | return 1; | |
1310 | } | |
1311 | ||
1312 | static void print_conn_state_change(struct drbd_tconn *tconn, enum drbd_conns oc, enum drbd_conns nc) | |
1313 | { | |
1314 | char *pbp, pb[300]; | |
1315 | pbp = pb; | |
1316 | *pbp = 0; | |
1317 | if (nc != oc) | |
1318 | pbp += sprintf(pbp, "conn( %s -> %s ) ", | |
1319 | drbd_conn_str(oc), | |
1320 | drbd_conn_str(nc)); | |
1321 | ||
1322 | conn_info(tconn, "%s\n", pb); | |
1323 | } | |
1324 | ||
1325 | struct _is_valid_itr_params { | |
1326 | enum chg_state_flags flags; | |
1327 | union drbd_state mask, val; | |
1328 | union drbd_state ms; /* maximal state, over all mdevs */ | |
1329 | enum drbd_conns oc; | |
1330 | enum { | |
1331 | OC_UNINITIALIZED, | |
1332 | OC_CONSISTENT, | |
1333 | OC_INCONSISTENT, | |
1334 | } oc_state; | |
1335 | }; | |
1336 | ||
1337 | static int _is_valid_itr_fn(int vnr, void *p, void *data) | |
1338 | { | |
1339 | struct drbd_conf *mdev = (struct drbd_conf *)p; | |
1340 | struct _is_valid_itr_params *params = (struct _is_valid_itr_params *)data; | |
1341 | enum chg_state_flags flags = params->flags; | |
1342 | union drbd_state ns, os; | |
1343 | enum drbd_state_rv rv; | |
1344 | ||
1345 | os = mdev->state; | |
1346 | ns = apply_mask_val(os, params->mask, params->val); | |
1347 | ns = sanitize_state(mdev, ns, NULL); | |
1348 | rv = is_valid_state(mdev, ns); | |
1349 | ||
1350 | if (rv < SS_SUCCESS) { | |
1351 | /* If the old state was illegal as well, then let this happen...*/ | |
1352 | ||
1353 | if (is_valid_state(mdev, os) == rv) | |
1354 | rv = is_valid_soft_transition(os, ns); | |
1355 | } else | |
1356 | rv = is_valid_soft_transition(os, ns); | |
1357 | ||
1358 | switch (params->oc_state) { | |
1359 | case OC_UNINITIALIZED: | |
1360 | params->oc = os.conn; | |
1361 | params->oc_state = OC_CONSISTENT; | |
1362 | break; | |
1363 | case OC_CONSISTENT: | |
1364 | if (params->oc != os.conn) | |
1365 | params->oc_state = OC_INCONSISTENT; | |
1366 | break; | |
1367 | case OC_INCONSISTENT: | |
1368 | break; | |
1369 | } | |
1370 | ||
1371 | if (rv < SS_SUCCESS) { | |
1372 | if (flags & CS_VERBOSE) | |
1373 | print_st_err(mdev, os, ns, rv); | |
1374 | return rv; | |
1375 | } else | |
1376 | return 0; | |
1377 | } | |
1378 | ||
1379 | static int _set_state_itr_fn(int vnr, void *p, void *data) | |
1380 | { | |
1381 | struct drbd_conf *mdev = (struct drbd_conf *)p; | |
1382 | struct _is_valid_itr_params *params = (struct _is_valid_itr_params *)data; | |
1383 | enum chg_state_flags flags = params->flags; | |
1384 | union drbd_state os, ns, ms = params->ms; | |
1385 | enum drbd_state_rv rv; | |
1386 | ||
1387 | os = mdev->state; | |
1388 | ns = apply_mask_val(os, params->mask, params->val); | |
1389 | ns = sanitize_state(mdev, ns, NULL); | |
1390 | ||
1391 | rv = __drbd_set_state(mdev, ns, flags, NULL); | |
1392 | ||
1393 | ms.role = max_t(enum drbd_role, mdev->state.role, ms.role); | |
1394 | ms.peer = max_t(enum drbd_role, mdev->state.peer, ms.peer); | |
1395 | ms.disk = max_t(enum drbd_role, mdev->state.disk, ms.disk); | |
1396 | ms.pdsk = max_t(enum drbd_role, mdev->state.pdsk, ms.pdsk); | |
1397 | params->ms = ms; | |
1398 | ||
1399 | return 0; | |
1400 | } | |
1401 | ||
df24aa45 PR |
1402 | static enum drbd_state_rv |
1403 | _conn_rq_cond(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val) | |
1404 | { | |
1405 | struct _is_valid_itr_params params; | |
1406 | enum drbd_state_rv rv; | |
1407 | ||
1408 | if (test_and_clear_bit(CONN_WD_ST_CHG_OKAY, &tconn->flags)) | |
1409 | return SS_CW_SUCCESS; | |
1410 | ||
1411 | if (test_and_clear_bit(CONN_WD_ST_CHG_FAIL, &tconn->flags)) | |
1412 | return SS_CW_FAILED_BY_PEER; | |
1413 | ||
1414 | params.flags = CS_NO_CSTATE_CHG; /* ΓΆΓΆ think */ | |
1415 | params.mask = mask; | |
1416 | params.val = val; | |
1417 | ||
1418 | spin_lock_irq(&tconn->req_lock); | |
1419 | rv = tconn->cstate != C_WF_REPORT_PARAMS ? SS_CW_NO_NEED : SS_UNKNOWN_ERROR; | |
1420 | ||
1421 | if (rv == SS_UNKNOWN_ERROR) | |
1422 | rv = idr_for_each(&tconn->volumes, _is_valid_itr_fn, ¶ms); | |
1423 | ||
1424 | if (rv == 0) /* idr_for_each semantics */ | |
1425 | rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */ | |
1426 | ||
1427 | spin_unlock_irq(&tconn->req_lock); | |
1428 | ||
1429 | return rv; | |
1430 | } | |
1431 | ||
1432 | static enum drbd_state_rv | |
1433 | conn_cl_wide(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, | |
1434 | enum chg_state_flags f) | |
1435 | { | |
1436 | enum drbd_state_rv rv; | |
1437 | ||
1438 | spin_unlock_irq(&tconn->req_lock); | |
1439 | mutex_lock(&tconn->cstate_mutex); | |
1440 | ||
caee1c3a | 1441 | if (conn_send_state_req(tconn, mask, val)) { |
df24aa45 PR |
1442 | rv = SS_CW_FAILED_BY_PEER; |
1443 | /* if (f & CS_VERBOSE) | |
1444 | print_st_err(mdev, os, ns, rv); */ | |
1445 | goto abort; | |
1446 | } | |
1447 | ||
1448 | wait_event(tconn->ping_wait, (rv = _conn_rq_cond(tconn, mask, val))); | |
1449 | ||
1450 | abort: | |
1451 | mutex_unlock(&tconn->cstate_mutex); | |
1452 | spin_lock_irq(&tconn->req_lock); | |
1453 | ||
1454 | return rv; | |
1455 | } | |
1456 | ||
bbeb641c PR |
1457 | enum drbd_state_rv |
1458 | _conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, | |
1459 | enum chg_state_flags flags) | |
1460 | { | |
1461 | enum drbd_state_rv rv = SS_SUCCESS; | |
1462 | struct _is_valid_itr_params params; | |
1463 | struct after_conn_state_chg_work *acscw; | |
1464 | enum drbd_conns oc = tconn->cstate; | |
1465 | ||
bbeb641c PR |
1466 | rv = is_valid_conn_transition(oc, val.conn); |
1467 | if (rv < SS_SUCCESS) | |
1468 | goto abort; | |
1469 | ||
1470 | params.flags = flags; | |
1471 | params.mask = mask; | |
1472 | params.val = val; | |
1473 | params.oc_state = OC_UNINITIALIZED; | |
1474 | ||
1475 | if (!(flags & CS_HARD)) | |
1476 | rv = idr_for_each(&tconn->volumes, _is_valid_itr_fn, ¶ms); | |
1477 | ||
1478 | if (rv == 0) /* idr_for_each semantics */ | |
1479 | rv = SS_SUCCESS; | |
1480 | ||
1481 | if (rv < SS_SUCCESS) | |
1482 | goto abort; | |
1483 | ||
df24aa45 PR |
1484 | if (oc == C_WF_REPORT_PARAMS && val.conn == C_DISCONNECTING && |
1485 | !(flags & (CS_LOCAL_ONLY | CS_HARD))) { | |
1486 | rv = conn_cl_wide(tconn, mask, val, flags); | |
1487 | if (rv < SS_SUCCESS) | |
1488 | goto abort; | |
1489 | } | |
1490 | ||
bbeb641c PR |
1491 | if (params.oc_state == OC_CONSISTENT) { |
1492 | oc = params.oc; | |
1493 | print_conn_state_change(tconn, oc, val.conn); | |
1494 | params.flags |= CS_NO_CSTATE_CHG; | |
1495 | } | |
1496 | tconn->cstate = val.conn; | |
1497 | params.ms.i = 0; | |
1498 | params.ms.conn = val.conn; | |
1499 | idr_for_each(&tconn->volumes, _set_state_itr_fn, ¶ms); | |
1500 | ||
1501 | acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC); | |
1502 | if (acscw) { | |
1503 | acscw->oc = oc; | |
1504 | acscw->nms = params.ms; | |
1505 | acscw->flags = flags; | |
1506 | acscw->w.cb = w_after_conn_state_ch; | |
1507 | acscw->w.tconn = tconn; | |
1508 | drbd_queue_work(&tconn->data.work, &acscw->w); | |
1509 | } else { | |
1510 | conn_err(tconn, "Could not kmalloc an acscw\n"); | |
b8907339 | 1511 | } |
bbeb641c PR |
1512 | |
1513 | abort: | |
bbeb641c PR |
1514 | return rv; |
1515 | } | |
1516 | ||
1517 | enum drbd_state_rv | |
1518 | conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, | |
1519 | enum chg_state_flags flags) | |
1520 | { | |
1521 | enum drbd_state_rv rv; | |
1522 | ||
1523 | spin_lock_irq(&tconn->req_lock); | |
1524 | rv = _conn_request_state(tconn, mask, val, flags); | |
1525 | spin_unlock_irq(&tconn->req_lock); | |
1526 | ||
1527 | return rv; | |
b8907339 | 1528 | } |