]>
Commit | Line | Data |
---|---|---|
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_protocol.h" | |
31 | #include "drbd_req.h" | |
32 | ||
33 | struct after_state_chg_work { | |
34 | struct drbd_work w; | |
35 | union drbd_state os; | |
36 | union drbd_state ns; | |
37 | enum chg_state_flags flags; | |
38 | struct completion *done; | |
39 | }; | |
40 | ||
41 | enum sanitize_state_warnings { | |
42 | NO_WARNING, | |
43 | ABORTED_ONLINE_VERIFY, | |
44 | ABORTED_RESYNC, | |
45 | CONNECTION_LOST_NEGOTIATING, | |
46 | IMPLICITLY_UPGRADED_DISK, | |
47 | IMPLICITLY_UPGRADED_PDSK, | |
48 | }; | |
49 | ||
50 | static int w_after_state_ch(struct drbd_work *w, int unused); | |
51 | static void after_state_ch(struct drbd_device *device, union drbd_state os, | |
52 | union drbd_state ns, enum chg_state_flags flags); | |
53 | static enum drbd_state_rv is_valid_state(struct drbd_device *, union drbd_state); | |
54 | static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state, struct drbd_connection *); | |
55 | static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns); | |
56 | static union drbd_state sanitize_state(struct drbd_device *device, union drbd_state ns, | |
57 | enum sanitize_state_warnings *warn); | |
58 | ||
59 | static inline bool is_susp(union drbd_state s) | |
60 | { | |
61 | return s.susp || s.susp_nod || s.susp_fen; | |
62 | } | |
63 | ||
64 | bool conn_all_vols_unconf(struct drbd_connection *connection) | |
65 | { | |
66 | struct drbd_peer_device *peer_device; | |
67 | bool rv = true; | |
68 | int vnr; | |
69 | ||
70 | rcu_read_lock(); | |
71 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
72 | struct drbd_device *device = peer_device->device; | |
73 | if (device->state.disk != D_DISKLESS || | |
74 | device->state.conn != C_STANDALONE || | |
75 | device->state.role != R_SECONDARY) { | |
76 | rv = false; | |
77 | break; | |
78 | } | |
79 | } | |
80 | rcu_read_unlock(); | |
81 | ||
82 | return rv; | |
83 | } | |
84 | ||
85 | /* Unfortunately the states where not correctly ordered, when | |
86 | they where defined. therefore can not use max_t() here. */ | |
87 | static enum drbd_role max_role(enum drbd_role role1, enum drbd_role role2) | |
88 | { | |
89 | if (role1 == R_PRIMARY || role2 == R_PRIMARY) | |
90 | return R_PRIMARY; | |
91 | if (role1 == R_SECONDARY || role2 == R_SECONDARY) | |
92 | return R_SECONDARY; | |
93 | return R_UNKNOWN; | |
94 | } | |
95 | static enum drbd_role min_role(enum drbd_role role1, enum drbd_role role2) | |
96 | { | |
97 | if (role1 == R_UNKNOWN || role2 == R_UNKNOWN) | |
98 | return R_UNKNOWN; | |
99 | if (role1 == R_SECONDARY || role2 == R_SECONDARY) | |
100 | return R_SECONDARY; | |
101 | return R_PRIMARY; | |
102 | } | |
103 | ||
104 | enum drbd_role conn_highest_role(struct drbd_connection *connection) | |
105 | { | |
106 | enum drbd_role role = R_UNKNOWN; | |
107 | struct drbd_peer_device *peer_device; | |
108 | int vnr; | |
109 | ||
110 | rcu_read_lock(); | |
111 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
112 | struct drbd_device *device = peer_device->device; | |
113 | role = max_role(role, device->state.role); | |
114 | } | |
115 | rcu_read_unlock(); | |
116 | ||
117 | return role; | |
118 | } | |
119 | ||
120 | enum drbd_role conn_highest_peer(struct drbd_connection *connection) | |
121 | { | |
122 | enum drbd_role peer = R_UNKNOWN; | |
123 | struct drbd_peer_device *peer_device; | |
124 | int vnr; | |
125 | ||
126 | rcu_read_lock(); | |
127 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
128 | struct drbd_device *device = peer_device->device; | |
129 | peer = max_role(peer, device->state.peer); | |
130 | } | |
131 | rcu_read_unlock(); | |
132 | ||
133 | return peer; | |
134 | } | |
135 | ||
136 | enum drbd_disk_state conn_highest_disk(struct drbd_connection *connection) | |
137 | { | |
138 | enum drbd_disk_state ds = D_DISKLESS; | |
139 | struct drbd_peer_device *peer_device; | |
140 | int vnr; | |
141 | ||
142 | rcu_read_lock(); | |
143 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
144 | struct drbd_device *device = peer_device->device; | |
145 | ds = max_t(enum drbd_disk_state, ds, device->state.disk); | |
146 | } | |
147 | rcu_read_unlock(); | |
148 | ||
149 | return ds; | |
150 | } | |
151 | ||
152 | enum drbd_disk_state conn_lowest_disk(struct drbd_connection *connection) | |
153 | { | |
154 | enum drbd_disk_state ds = D_MASK; | |
155 | struct drbd_peer_device *peer_device; | |
156 | int vnr; | |
157 | ||
158 | rcu_read_lock(); | |
159 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
160 | struct drbd_device *device = peer_device->device; | |
161 | ds = min_t(enum drbd_disk_state, ds, device->state.disk); | |
162 | } | |
163 | rcu_read_unlock(); | |
164 | ||
165 | return ds; | |
166 | } | |
167 | ||
168 | enum drbd_disk_state conn_highest_pdsk(struct drbd_connection *connection) | |
169 | { | |
170 | enum drbd_disk_state ds = D_DISKLESS; | |
171 | struct drbd_peer_device *peer_device; | |
172 | int vnr; | |
173 | ||
174 | rcu_read_lock(); | |
175 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
176 | struct drbd_device *device = peer_device->device; | |
177 | ds = max_t(enum drbd_disk_state, ds, device->state.pdsk); | |
178 | } | |
179 | rcu_read_unlock(); | |
180 | ||
181 | return ds; | |
182 | } | |
183 | ||
184 | enum drbd_conns conn_lowest_conn(struct drbd_connection *connection) | |
185 | { | |
186 | enum drbd_conns conn = C_MASK; | |
187 | struct drbd_peer_device *peer_device; | |
188 | int vnr; | |
189 | ||
190 | rcu_read_lock(); | |
191 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
192 | struct drbd_device *device = peer_device->device; | |
193 | conn = min_t(enum drbd_conns, conn, device->state.conn); | |
194 | } | |
195 | rcu_read_unlock(); | |
196 | ||
197 | return conn; | |
198 | } | |
199 | ||
200 | static bool no_peer_wf_report_params(struct drbd_connection *connection) | |
201 | { | |
202 | struct drbd_peer_device *peer_device; | |
203 | int vnr; | |
204 | bool rv = true; | |
205 | ||
206 | rcu_read_lock(); | |
207 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) | |
208 | if (peer_device->device->state.conn == C_WF_REPORT_PARAMS) { | |
209 | rv = false; | |
210 | break; | |
211 | } | |
212 | rcu_read_unlock(); | |
213 | ||
214 | return rv; | |
215 | } | |
216 | ||
217 | ||
218 | /** | |
219 | * cl_wide_st_chg() - true if the state change is a cluster wide one | |
220 | * @device: DRBD device. | |
221 | * @os: old (current) state. | |
222 | * @ns: new (wanted) state. | |
223 | */ | |
224 | static int cl_wide_st_chg(struct drbd_device *device, | |
225 | union drbd_state os, union drbd_state ns) | |
226 | { | |
227 | return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED && | |
228 | ((os.role != R_PRIMARY && ns.role == R_PRIMARY) || | |
229 | (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) || | |
230 | (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) || | |
231 | (os.disk != D_FAILED && ns.disk == D_FAILED))) || | |
232 | (os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) || | |
233 | (os.conn == C_CONNECTED && ns.conn == C_VERIFY_S) || | |
234 | (os.conn == C_CONNECTED && ns.conn == C_WF_REPORT_PARAMS); | |
235 | } | |
236 | ||
237 | static union drbd_state | |
238 | apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val) | |
239 | { | |
240 | union drbd_state ns; | |
241 | ns.i = (os.i & ~mask.i) | val.i; | |
242 | return ns; | |
243 | } | |
244 | ||
245 | enum drbd_state_rv | |
246 | drbd_change_state(struct drbd_device *device, enum chg_state_flags f, | |
247 | union drbd_state mask, union drbd_state val) | |
248 | { | |
249 | unsigned long flags; | |
250 | union drbd_state ns; | |
251 | enum drbd_state_rv rv; | |
252 | ||
253 | spin_lock_irqsave(&device->resource->req_lock, flags); | |
254 | ns = apply_mask_val(drbd_read_state(device), mask, val); | |
255 | rv = _drbd_set_state(device, ns, f, NULL); | |
256 | spin_unlock_irqrestore(&device->resource->req_lock, flags); | |
257 | ||
258 | return rv; | |
259 | } | |
260 | ||
261 | /** | |
262 | * drbd_force_state() - Impose a change which happens outside our control on our state | |
263 | * @device: DRBD device. | |
264 | * @mask: mask of state bits to change. | |
265 | * @val: value of new state bits. | |
266 | */ | |
267 | void drbd_force_state(struct drbd_device *device, | |
268 | union drbd_state mask, union drbd_state val) | |
269 | { | |
270 | drbd_change_state(device, CS_HARD, mask, val); | |
271 | } | |
272 | ||
273 | static enum drbd_state_rv | |
274 | _req_st_cond(struct drbd_device *device, union drbd_state mask, | |
275 | union drbd_state val) | |
276 | { | |
277 | union drbd_state os, ns; | |
278 | unsigned long flags; | |
279 | enum drbd_state_rv rv; | |
280 | ||
281 | if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &device->flags)) | |
282 | return SS_CW_SUCCESS; | |
283 | ||
284 | if (test_and_clear_bit(CL_ST_CHG_FAIL, &device->flags)) | |
285 | return SS_CW_FAILED_BY_PEER; | |
286 | ||
287 | spin_lock_irqsave(&device->resource->req_lock, flags); | |
288 | os = drbd_read_state(device); | |
289 | ns = sanitize_state(device, apply_mask_val(os, mask, val), NULL); | |
290 | rv = is_valid_transition(os, ns); | |
291 | if (rv >= SS_SUCCESS) | |
292 | rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */ | |
293 | ||
294 | if (!cl_wide_st_chg(device, os, ns)) | |
295 | rv = SS_CW_NO_NEED; | |
296 | if (rv == SS_UNKNOWN_ERROR) { | |
297 | rv = is_valid_state(device, ns); | |
298 | if (rv >= SS_SUCCESS) { | |
299 | rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection); | |
300 | if (rv >= SS_SUCCESS) | |
301 | rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */ | |
302 | } | |
303 | } | |
304 | spin_unlock_irqrestore(&device->resource->req_lock, flags); | |
305 | ||
306 | return rv; | |
307 | } | |
308 | ||
309 | /** | |
310 | * drbd_req_state() - Perform an eventually cluster wide state change | |
311 | * @device: DRBD device. | |
312 | * @mask: mask of state bits to change. | |
313 | * @val: value of new state bits. | |
314 | * @f: flags | |
315 | * | |
316 | * Should not be called directly, use drbd_request_state() or | |
317 | * _drbd_request_state(). | |
318 | */ | |
319 | static enum drbd_state_rv | |
320 | drbd_req_state(struct drbd_device *device, union drbd_state mask, | |
321 | union drbd_state val, enum chg_state_flags f) | |
322 | { | |
323 | struct completion done; | |
324 | unsigned long flags; | |
325 | union drbd_state os, ns; | |
326 | enum drbd_state_rv rv; | |
327 | ||
328 | init_completion(&done); | |
329 | ||
330 | if (f & CS_SERIALIZE) | |
331 | mutex_lock(device->state_mutex); | |
332 | ||
333 | spin_lock_irqsave(&device->resource->req_lock, flags); | |
334 | os = drbd_read_state(device); | |
335 | ns = sanitize_state(device, apply_mask_val(os, mask, val), NULL); | |
336 | rv = is_valid_transition(os, ns); | |
337 | if (rv < SS_SUCCESS) { | |
338 | spin_unlock_irqrestore(&device->resource->req_lock, flags); | |
339 | goto abort; | |
340 | } | |
341 | ||
342 | if (cl_wide_st_chg(device, os, ns)) { | |
343 | rv = is_valid_state(device, ns); | |
344 | if (rv == SS_SUCCESS) | |
345 | rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection); | |
346 | spin_unlock_irqrestore(&device->resource->req_lock, flags); | |
347 | ||
348 | if (rv < SS_SUCCESS) { | |
349 | if (f & CS_VERBOSE) | |
350 | print_st_err(device, os, ns, rv); | |
351 | goto abort; | |
352 | } | |
353 | ||
354 | if (drbd_send_state_req(device, mask, val)) { | |
355 | rv = SS_CW_FAILED_BY_PEER; | |
356 | if (f & CS_VERBOSE) | |
357 | print_st_err(device, os, ns, rv); | |
358 | goto abort; | |
359 | } | |
360 | ||
361 | wait_event(device->state_wait, | |
362 | (rv = _req_st_cond(device, mask, val))); | |
363 | ||
364 | if (rv < SS_SUCCESS) { | |
365 | if (f & CS_VERBOSE) | |
366 | print_st_err(device, os, ns, rv); | |
367 | goto abort; | |
368 | } | |
369 | spin_lock_irqsave(&device->resource->req_lock, flags); | |
370 | ns = apply_mask_val(drbd_read_state(device), mask, val); | |
371 | rv = _drbd_set_state(device, ns, f, &done); | |
372 | } else { | |
373 | rv = _drbd_set_state(device, ns, f, &done); | |
374 | } | |
375 | ||
376 | spin_unlock_irqrestore(&device->resource->req_lock, flags); | |
377 | ||
378 | if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) { | |
379 | D_ASSERT(device, current != first_peer_device(device)->connection->worker.task); | |
380 | wait_for_completion(&done); | |
381 | } | |
382 | ||
383 | abort: | |
384 | if (f & CS_SERIALIZE) | |
385 | mutex_unlock(device->state_mutex); | |
386 | ||
387 | return rv; | |
388 | } | |
389 | ||
390 | /** | |
391 | * _drbd_request_state() - Request a state change (with flags) | |
392 | * @device: DRBD device. | |
393 | * @mask: mask of state bits to change. | |
394 | * @val: value of new state bits. | |
395 | * @f: flags | |
396 | * | |
397 | * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE | |
398 | * flag, or when logging of failed state change requests is not desired. | |
399 | */ | |
400 | enum drbd_state_rv | |
401 | _drbd_request_state(struct drbd_device *device, union drbd_state mask, | |
402 | union drbd_state val, enum chg_state_flags f) | |
403 | { | |
404 | enum drbd_state_rv rv; | |
405 | ||
406 | wait_event(device->state_wait, | |
407 | (rv = drbd_req_state(device, mask, val, f)) != SS_IN_TRANSIENT_STATE); | |
408 | ||
409 | return rv; | |
410 | } | |
411 | ||
412 | static void print_st(struct drbd_device *device, char *name, union drbd_state ns) | |
413 | { | |
414 | drbd_err(device, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n", | |
415 | name, | |
416 | drbd_conn_str(ns.conn), | |
417 | drbd_role_str(ns.role), | |
418 | drbd_role_str(ns.peer), | |
419 | drbd_disk_str(ns.disk), | |
420 | drbd_disk_str(ns.pdsk), | |
421 | is_susp(ns) ? 's' : 'r', | |
422 | ns.aftr_isp ? 'a' : '-', | |
423 | ns.peer_isp ? 'p' : '-', | |
424 | ns.user_isp ? 'u' : '-', | |
425 | ns.susp_fen ? 'F' : '-', | |
426 | ns.susp_nod ? 'N' : '-' | |
427 | ); | |
428 | } | |
429 | ||
430 | void print_st_err(struct drbd_device *device, union drbd_state os, | |
431 | union drbd_state ns, enum drbd_state_rv err) | |
432 | { | |
433 | if (err == SS_IN_TRANSIENT_STATE) | |
434 | return; | |
435 | drbd_err(device, "State change failed: %s\n", drbd_set_st_err_str(err)); | |
436 | print_st(device, " state", os); | |
437 | print_st(device, "wanted", ns); | |
438 | } | |
439 | ||
440 | static long print_state_change(char *pb, union drbd_state os, union drbd_state ns, | |
441 | enum chg_state_flags flags) | |
442 | { | |
443 | char *pbp; | |
444 | pbp = pb; | |
445 | *pbp = 0; | |
446 | ||
447 | if (ns.role != os.role && flags & CS_DC_ROLE) | |
448 | pbp += sprintf(pbp, "role( %s -> %s ) ", | |
449 | drbd_role_str(os.role), | |
450 | drbd_role_str(ns.role)); | |
451 | if (ns.peer != os.peer && flags & CS_DC_PEER) | |
452 | pbp += sprintf(pbp, "peer( %s -> %s ) ", | |
453 | drbd_role_str(os.peer), | |
454 | drbd_role_str(ns.peer)); | |
455 | if (ns.conn != os.conn && flags & CS_DC_CONN) | |
456 | pbp += sprintf(pbp, "conn( %s -> %s ) ", | |
457 | drbd_conn_str(os.conn), | |
458 | drbd_conn_str(ns.conn)); | |
459 | if (ns.disk != os.disk && flags & CS_DC_DISK) | |
460 | pbp += sprintf(pbp, "disk( %s -> %s ) ", | |
461 | drbd_disk_str(os.disk), | |
462 | drbd_disk_str(ns.disk)); | |
463 | if (ns.pdsk != os.pdsk && flags & CS_DC_PDSK) | |
464 | pbp += sprintf(pbp, "pdsk( %s -> %s ) ", | |
465 | drbd_disk_str(os.pdsk), | |
466 | drbd_disk_str(ns.pdsk)); | |
467 | ||
468 | return pbp - pb; | |
469 | } | |
470 | ||
471 | static void drbd_pr_state_change(struct drbd_device *device, union drbd_state os, union drbd_state ns, | |
472 | enum chg_state_flags flags) | |
473 | { | |
474 | char pb[300]; | |
475 | char *pbp = pb; | |
476 | ||
477 | pbp += print_state_change(pbp, os, ns, flags ^ CS_DC_MASK); | |
478 | ||
479 | if (ns.aftr_isp != os.aftr_isp) | |
480 | pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ", | |
481 | os.aftr_isp, | |
482 | ns.aftr_isp); | |
483 | if (ns.peer_isp != os.peer_isp) | |
484 | pbp += sprintf(pbp, "peer_isp( %d -> %d ) ", | |
485 | os.peer_isp, | |
486 | ns.peer_isp); | |
487 | if (ns.user_isp != os.user_isp) | |
488 | pbp += sprintf(pbp, "user_isp( %d -> %d ) ", | |
489 | os.user_isp, | |
490 | ns.user_isp); | |
491 | ||
492 | if (pbp != pb) | |
493 | drbd_info(device, "%s\n", pb); | |
494 | } | |
495 | ||
496 | static void conn_pr_state_change(struct drbd_connection *connection, union drbd_state os, union drbd_state ns, | |
497 | enum chg_state_flags flags) | |
498 | { | |
499 | char pb[300]; | |
500 | char *pbp = pb; | |
501 | ||
502 | pbp += print_state_change(pbp, os, ns, flags); | |
503 | ||
504 | if (is_susp(ns) != is_susp(os) && flags & CS_DC_SUSP) | |
505 | pbp += sprintf(pbp, "susp( %d -> %d ) ", | |
506 | is_susp(os), | |
507 | is_susp(ns)); | |
508 | ||
509 | if (pbp != pb) | |
510 | drbd_info(connection, "%s\n", pb); | |
511 | } | |
512 | ||
513 | ||
514 | /** | |
515 | * is_valid_state() - Returns an SS_ error code if ns is not valid | |
516 | * @device: DRBD device. | |
517 | * @ns: State to consider. | |
518 | */ | |
519 | static enum drbd_state_rv | |
520 | is_valid_state(struct drbd_device *device, union drbd_state ns) | |
521 | { | |
522 | /* See drbd_state_sw_errors in drbd_strings.c */ | |
523 | ||
524 | enum drbd_fencing_p fp; | |
525 | enum drbd_state_rv rv = SS_SUCCESS; | |
526 | struct net_conf *nc; | |
527 | ||
528 | rcu_read_lock(); | |
529 | fp = FP_DONT_CARE; | |
530 | if (get_ldev(device)) { | |
531 | fp = rcu_dereference(device->ldev->disk_conf)->fencing; | |
532 | put_ldev(device); | |
533 | } | |
534 | ||
535 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); | |
536 | if (nc) { | |
537 | if (!nc->two_primaries && ns.role == R_PRIMARY) { | |
538 | if (ns.peer == R_PRIMARY) | |
539 | rv = SS_TWO_PRIMARIES; | |
540 | else if (conn_highest_peer(first_peer_device(device)->connection) == R_PRIMARY) | |
541 | rv = SS_O_VOL_PEER_PRI; | |
542 | } | |
543 | } | |
544 | ||
545 | if (rv <= 0) | |
546 | /* already found a reason to abort */; | |
547 | else if (ns.role == R_SECONDARY && device->open_cnt) | |
548 | rv = SS_DEVICE_IN_USE; | |
549 | ||
550 | else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE) | |
551 | rv = SS_NO_UP_TO_DATE_DISK; | |
552 | ||
553 | else if (fp >= FP_RESOURCE && | |
554 | ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN) | |
555 | rv = SS_PRIMARY_NOP; | |
556 | ||
557 | else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT) | |
558 | rv = SS_NO_UP_TO_DATE_DISK; | |
559 | ||
560 | else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT) | |
561 | rv = SS_NO_LOCAL_DISK; | |
562 | ||
563 | else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT) | |
564 | rv = SS_NO_REMOTE_DISK; | |
565 | ||
566 | else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE) | |
567 | rv = SS_NO_UP_TO_DATE_DISK; | |
568 | ||
569 | else if ((ns.conn == C_CONNECTED || | |
570 | ns.conn == C_WF_BITMAP_S || | |
571 | ns.conn == C_SYNC_SOURCE || | |
572 | ns.conn == C_PAUSED_SYNC_S) && | |
573 | ns.disk == D_OUTDATED) | |
574 | rv = SS_CONNECTED_OUTDATES; | |
575 | ||
576 | else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
577 | (nc->verify_alg[0] == 0)) | |
578 | rv = SS_NO_VERIFY_ALG; | |
579 | ||
580 | else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
581 | first_peer_device(device)->connection->agreed_pro_version < 88) | |
582 | rv = SS_NOT_SUPPORTED; | |
583 | ||
584 | else if (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE) | |
585 | rv = SS_NO_UP_TO_DATE_DISK; | |
586 | ||
587 | else if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) && | |
588 | ns.pdsk == D_UNKNOWN) | |
589 | rv = SS_NEED_CONNECTION; | |
590 | ||
591 | else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN) | |
592 | rv = SS_CONNECTED_OUTDATES; | |
593 | ||
594 | rcu_read_unlock(); | |
595 | ||
596 | return rv; | |
597 | } | |
598 | ||
599 | /** | |
600 | * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible | |
601 | * This function limits state transitions that may be declined by DRBD. I.e. | |
602 | * user requests (aka soft transitions). | |
603 | * @device: DRBD device. | |
604 | * @ns: new state. | |
605 | * @os: old state. | |
606 | */ | |
607 | static enum drbd_state_rv | |
608 | is_valid_soft_transition(union drbd_state os, union drbd_state ns, struct drbd_connection *connection) | |
609 | { | |
610 | enum drbd_state_rv rv = SS_SUCCESS; | |
611 | ||
612 | if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) && | |
613 | os.conn > C_CONNECTED) | |
614 | rv = SS_RESYNC_RUNNING; | |
615 | ||
616 | if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE) | |
617 | rv = SS_ALREADY_STANDALONE; | |
618 | ||
619 | if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS) | |
620 | rv = SS_IS_DISKLESS; | |
621 | ||
622 | if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED) | |
623 | rv = SS_NO_NET_CONFIG; | |
624 | ||
625 | if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING) | |
626 | rv = SS_LOWER_THAN_OUTDATED; | |
627 | ||
628 | if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED) | |
629 | rv = SS_IN_TRANSIENT_STATE; | |
630 | ||
631 | /* if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS) | |
632 | rv = SS_IN_TRANSIENT_STATE; */ | |
633 | ||
634 | /* While establishing a connection only allow cstate to change. | |
635 | Delay/refuse role changes, detach attach etc... */ | |
636 | if (test_bit(STATE_SENT, &connection->flags) && | |
637 | !(os.conn == C_WF_REPORT_PARAMS || | |
638 | (ns.conn == C_WF_REPORT_PARAMS && os.conn == C_WF_CONNECTION))) | |
639 | rv = SS_IN_TRANSIENT_STATE; | |
640 | ||
641 | if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED) | |
642 | rv = SS_NEED_CONNECTION; | |
643 | ||
644 | if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
645 | ns.conn != os.conn && os.conn > C_CONNECTED) | |
646 | rv = SS_RESYNC_RUNNING; | |
647 | ||
648 | if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) && | |
649 | os.conn < C_CONNECTED) | |
650 | rv = SS_NEED_CONNECTION; | |
651 | ||
652 | if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE) | |
653 | && os.conn < C_WF_REPORT_PARAMS) | |
654 | rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */ | |
655 | ||
656 | if (ns.conn == C_DISCONNECTING && ns.pdsk == D_OUTDATED && | |
657 | os.conn < C_CONNECTED && os.pdsk > D_OUTDATED) | |
658 | rv = SS_OUTDATE_WO_CONN; | |
659 | ||
660 | return rv; | |
661 | } | |
662 | ||
663 | static enum drbd_state_rv | |
664 | is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc) | |
665 | { | |
666 | /* no change -> nothing to do, at least for the connection part */ | |
667 | if (oc == nc) | |
668 | return SS_NOTHING_TO_DO; | |
669 | ||
670 | /* disconnect of an unconfigured connection does not make sense */ | |
671 | if (oc == C_STANDALONE && nc == C_DISCONNECTING) | |
672 | return SS_ALREADY_STANDALONE; | |
673 | ||
674 | /* from C_STANDALONE, we start with C_UNCONNECTED */ | |
675 | if (oc == C_STANDALONE && nc != C_UNCONNECTED) | |
676 | return SS_NEED_CONNECTION; | |
677 | ||
678 | /* When establishing a connection we need to go through WF_REPORT_PARAMS! | |
679 | Necessary to do the right thing upon invalidate-remote on a disconnected resource */ | |
680 | if (oc < C_WF_REPORT_PARAMS && nc >= C_CONNECTED) | |
681 | return SS_NEED_CONNECTION; | |
682 | ||
683 | /* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */ | |
684 | if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING) | |
685 | return SS_IN_TRANSIENT_STATE; | |
686 | ||
687 | /* After C_DISCONNECTING only C_STANDALONE may follow */ | |
688 | if (oc == C_DISCONNECTING && nc != C_STANDALONE) | |
689 | return SS_IN_TRANSIENT_STATE; | |
690 | ||
691 | return SS_SUCCESS; | |
692 | } | |
693 | ||
694 | ||
695 | /** | |
696 | * is_valid_transition() - Returns an SS_ error code if the state transition is not possible | |
697 | * This limits hard state transitions. Hard state transitions are facts there are | |
698 | * imposed on DRBD by the environment. E.g. disk broke or network broke down. | |
699 | * But those hard state transitions are still not allowed to do everything. | |
700 | * @ns: new state. | |
701 | * @os: old state. | |
702 | */ | |
703 | static enum drbd_state_rv | |
704 | is_valid_transition(union drbd_state os, union drbd_state ns) | |
705 | { | |
706 | enum drbd_state_rv rv; | |
707 | ||
708 | rv = is_valid_conn_transition(os.conn, ns.conn); | |
709 | ||
710 | /* we cannot fail (again) if we already detached */ | |
711 | if (ns.disk == D_FAILED && os.disk == D_DISKLESS) | |
712 | rv = SS_IS_DISKLESS; | |
713 | ||
714 | return rv; | |
715 | } | |
716 | ||
717 | static void print_sanitize_warnings(struct drbd_device *device, enum sanitize_state_warnings warn) | |
718 | { | |
719 | static const char *msg_table[] = { | |
720 | [NO_WARNING] = "", | |
721 | [ABORTED_ONLINE_VERIFY] = "Online-verify aborted.", | |
722 | [ABORTED_RESYNC] = "Resync aborted.", | |
723 | [CONNECTION_LOST_NEGOTIATING] = "Connection lost while negotiating, no data!", | |
724 | [IMPLICITLY_UPGRADED_DISK] = "Implicitly upgraded disk", | |
725 | [IMPLICITLY_UPGRADED_PDSK] = "Implicitly upgraded pdsk", | |
726 | }; | |
727 | ||
728 | if (warn != NO_WARNING) | |
729 | drbd_warn(device, "%s\n", msg_table[warn]); | |
730 | } | |
731 | ||
732 | /** | |
733 | * sanitize_state() - Resolves implicitly necessary additional changes to a state transition | |
734 | * @device: DRBD device. | |
735 | * @os: old state. | |
736 | * @ns: new state. | |
737 | * @warn_sync_abort: | |
738 | * | |
739 | * When we loose connection, we have to set the state of the peers disk (pdsk) | |
740 | * to D_UNKNOWN. This rule and many more along those lines are in this function. | |
741 | */ | |
742 | static union drbd_state sanitize_state(struct drbd_device *device, union drbd_state ns, | |
743 | enum sanitize_state_warnings *warn) | |
744 | { | |
745 | enum drbd_fencing_p fp; | |
746 | enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max; | |
747 | ||
748 | if (warn) | |
749 | *warn = NO_WARNING; | |
750 | ||
751 | fp = FP_DONT_CARE; | |
752 | if (get_ldev(device)) { | |
753 | rcu_read_lock(); | |
754 | fp = rcu_dereference(device->ldev->disk_conf)->fencing; | |
755 | rcu_read_unlock(); | |
756 | put_ldev(device); | |
757 | } | |
758 | ||
759 | /* Implications from connection to peer and peer_isp */ | |
760 | if (ns.conn < C_CONNECTED) { | |
761 | ns.peer_isp = 0; | |
762 | ns.peer = R_UNKNOWN; | |
763 | if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT) | |
764 | ns.pdsk = D_UNKNOWN; | |
765 | } | |
766 | ||
767 | /* Clear the aftr_isp when becoming unconfigured */ | |
768 | if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY) | |
769 | ns.aftr_isp = 0; | |
770 | ||
771 | /* An implication of the disk states onto the connection state */ | |
772 | /* Abort resync if a disk fails/detaches */ | |
773 | if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) { | |
774 | if (warn) | |
775 | *warn = ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ? | |
776 | ABORTED_ONLINE_VERIFY : ABORTED_RESYNC; | |
777 | ns.conn = C_CONNECTED; | |
778 | } | |
779 | ||
780 | /* Connection breaks down before we finished "Negotiating" */ | |
781 | if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING && | |
782 | get_ldev_if_state(device, D_NEGOTIATING)) { | |
783 | if (device->ed_uuid == device->ldev->md.uuid[UI_CURRENT]) { | |
784 | ns.disk = device->new_state_tmp.disk; | |
785 | ns.pdsk = device->new_state_tmp.pdsk; | |
786 | } else { | |
787 | if (warn) | |
788 | *warn = CONNECTION_LOST_NEGOTIATING; | |
789 | ns.disk = D_DISKLESS; | |
790 | ns.pdsk = D_UNKNOWN; | |
791 | } | |
792 | put_ldev(device); | |
793 | } | |
794 | ||
795 | /* D_CONSISTENT and D_OUTDATED vanish when we get connected */ | |
796 | if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) { | |
797 | if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED) | |
798 | ns.disk = D_UP_TO_DATE; | |
799 | if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED) | |
800 | ns.pdsk = D_UP_TO_DATE; | |
801 | } | |
802 | ||
803 | /* Implications of the connection stat on the disk states */ | |
804 | disk_min = D_DISKLESS; | |
805 | disk_max = D_UP_TO_DATE; | |
806 | pdsk_min = D_INCONSISTENT; | |
807 | pdsk_max = D_UNKNOWN; | |
808 | switch ((enum drbd_conns)ns.conn) { | |
809 | case C_WF_BITMAP_T: | |
810 | case C_PAUSED_SYNC_T: | |
811 | case C_STARTING_SYNC_T: | |
812 | case C_WF_SYNC_UUID: | |
813 | case C_BEHIND: | |
814 | disk_min = D_INCONSISTENT; | |
815 | disk_max = D_OUTDATED; | |
816 | pdsk_min = D_UP_TO_DATE; | |
817 | pdsk_max = D_UP_TO_DATE; | |
818 | break; | |
819 | case C_VERIFY_S: | |
820 | case C_VERIFY_T: | |
821 | disk_min = D_UP_TO_DATE; | |
822 | disk_max = D_UP_TO_DATE; | |
823 | pdsk_min = D_UP_TO_DATE; | |
824 | pdsk_max = D_UP_TO_DATE; | |
825 | break; | |
826 | case C_CONNECTED: | |
827 | disk_min = D_DISKLESS; | |
828 | disk_max = D_UP_TO_DATE; | |
829 | pdsk_min = D_DISKLESS; | |
830 | pdsk_max = D_UP_TO_DATE; | |
831 | break; | |
832 | case C_WF_BITMAP_S: | |
833 | case C_PAUSED_SYNC_S: | |
834 | case C_STARTING_SYNC_S: | |
835 | case C_AHEAD: | |
836 | disk_min = D_UP_TO_DATE; | |
837 | disk_max = D_UP_TO_DATE; | |
838 | pdsk_min = D_INCONSISTENT; | |
839 | pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/ | |
840 | break; | |
841 | case C_SYNC_TARGET: | |
842 | disk_min = D_INCONSISTENT; | |
843 | disk_max = D_INCONSISTENT; | |
844 | pdsk_min = D_UP_TO_DATE; | |
845 | pdsk_max = D_UP_TO_DATE; | |
846 | break; | |
847 | case C_SYNC_SOURCE: | |
848 | disk_min = D_UP_TO_DATE; | |
849 | disk_max = D_UP_TO_DATE; | |
850 | pdsk_min = D_INCONSISTENT; | |
851 | pdsk_max = D_INCONSISTENT; | |
852 | break; | |
853 | case C_STANDALONE: | |
854 | case C_DISCONNECTING: | |
855 | case C_UNCONNECTED: | |
856 | case C_TIMEOUT: | |
857 | case C_BROKEN_PIPE: | |
858 | case C_NETWORK_FAILURE: | |
859 | case C_PROTOCOL_ERROR: | |
860 | case C_TEAR_DOWN: | |
861 | case C_WF_CONNECTION: | |
862 | case C_WF_REPORT_PARAMS: | |
863 | case C_MASK: | |
864 | break; | |
865 | } | |
866 | if (ns.disk > disk_max) | |
867 | ns.disk = disk_max; | |
868 | ||
869 | if (ns.disk < disk_min) { | |
870 | if (warn) | |
871 | *warn = IMPLICITLY_UPGRADED_DISK; | |
872 | ns.disk = disk_min; | |
873 | } | |
874 | if (ns.pdsk > pdsk_max) | |
875 | ns.pdsk = pdsk_max; | |
876 | ||
877 | if (ns.pdsk < pdsk_min) { | |
878 | if (warn) | |
879 | *warn = IMPLICITLY_UPGRADED_PDSK; | |
880 | ns.pdsk = pdsk_min; | |
881 | } | |
882 | ||
883 | if (fp == FP_STONITH && | |
884 | (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED)) | |
885 | ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */ | |
886 | ||
887 | if (device->resource->res_opts.on_no_data == OND_SUSPEND_IO && | |
888 | (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)) | |
889 | ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */ | |
890 | ||
891 | if (ns.aftr_isp || ns.peer_isp || ns.user_isp) { | |
892 | if (ns.conn == C_SYNC_SOURCE) | |
893 | ns.conn = C_PAUSED_SYNC_S; | |
894 | if (ns.conn == C_SYNC_TARGET) | |
895 | ns.conn = C_PAUSED_SYNC_T; | |
896 | } else { | |
897 | if (ns.conn == C_PAUSED_SYNC_S) | |
898 | ns.conn = C_SYNC_SOURCE; | |
899 | if (ns.conn == C_PAUSED_SYNC_T) | |
900 | ns.conn = C_SYNC_TARGET; | |
901 | } | |
902 | ||
903 | return ns; | |
904 | } | |
905 | ||
906 | void drbd_resume_al(struct drbd_device *device) | |
907 | { | |
908 | if (test_and_clear_bit(AL_SUSPENDED, &device->flags)) | |
909 | drbd_info(device, "Resumed AL updates\n"); | |
910 | } | |
911 | ||
912 | /* helper for __drbd_set_state */ | |
913 | static void set_ov_position(struct drbd_device *device, enum drbd_conns cs) | |
914 | { | |
915 | if (first_peer_device(device)->connection->agreed_pro_version < 90) | |
916 | device->ov_start_sector = 0; | |
917 | device->rs_total = drbd_bm_bits(device); | |
918 | device->ov_position = 0; | |
919 | if (cs == C_VERIFY_T) { | |
920 | /* starting online verify from an arbitrary position | |
921 | * does not fit well into the existing protocol. | |
922 | * on C_VERIFY_T, we initialize ov_left and friends | |
923 | * implicitly in receive_DataRequest once the | |
924 | * first P_OV_REQUEST is received */ | |
925 | device->ov_start_sector = ~(sector_t)0; | |
926 | } else { | |
927 | unsigned long bit = BM_SECT_TO_BIT(device->ov_start_sector); | |
928 | if (bit >= device->rs_total) { | |
929 | device->ov_start_sector = | |
930 | BM_BIT_TO_SECT(device->rs_total - 1); | |
931 | device->rs_total = 1; | |
932 | } else | |
933 | device->rs_total -= bit; | |
934 | device->ov_position = device->ov_start_sector; | |
935 | } | |
936 | device->ov_left = device->rs_total; | |
937 | } | |
938 | ||
939 | /** | |
940 | * __drbd_set_state() - Set a new DRBD state | |
941 | * @device: DRBD device. | |
942 | * @ns: new state. | |
943 | * @flags: Flags | |
944 | * @done: Optional completion, that will get completed after the after_state_ch() finished | |
945 | * | |
946 | * Caller needs to hold req_lock, and global_state_lock. Do not call directly. | |
947 | */ | |
948 | enum drbd_state_rv | |
949 | __drbd_set_state(struct drbd_device *device, union drbd_state ns, | |
950 | enum chg_state_flags flags, struct completion *done) | |
951 | { | |
952 | union drbd_state os; | |
953 | enum drbd_state_rv rv = SS_SUCCESS; | |
954 | enum sanitize_state_warnings ssw; | |
955 | struct after_state_chg_work *ascw; | |
956 | bool did_remote, should_do_remote; | |
957 | ||
958 | os = drbd_read_state(device); | |
959 | ||
960 | ns = sanitize_state(device, ns, &ssw); | |
961 | if (ns.i == os.i) | |
962 | return SS_NOTHING_TO_DO; | |
963 | ||
964 | rv = is_valid_transition(os, ns); | |
965 | if (rv < SS_SUCCESS) | |
966 | return rv; | |
967 | ||
968 | if (!(flags & CS_HARD)) { | |
969 | /* pre-state-change checks ; only look at ns */ | |
970 | /* See drbd_state_sw_errors in drbd_strings.c */ | |
971 | ||
972 | rv = is_valid_state(device, ns); | |
973 | if (rv < SS_SUCCESS) { | |
974 | /* If the old state was illegal as well, then let | |
975 | this happen...*/ | |
976 | ||
977 | if (is_valid_state(device, os) == rv) | |
978 | rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection); | |
979 | } else | |
980 | rv = is_valid_soft_transition(os, ns, first_peer_device(device)->connection); | |
981 | } | |
982 | ||
983 | if (rv < SS_SUCCESS) { | |
984 | if (flags & CS_VERBOSE) | |
985 | print_st_err(device, os, ns, rv); | |
986 | return rv; | |
987 | } | |
988 | ||
989 | print_sanitize_warnings(device, ssw); | |
990 | ||
991 | drbd_pr_state_change(device, os, ns, flags); | |
992 | ||
993 | /* Display changes to the susp* flags that where caused by the call to | |
994 | sanitize_state(). Only display it here if we where not called from | |
995 | _conn_request_state() */ | |
996 | if (!(flags & CS_DC_SUSP)) | |
997 | conn_pr_state_change(first_peer_device(device)->connection, os, ns, | |
998 | (flags & ~CS_DC_MASK) | CS_DC_SUSP); | |
999 | ||
1000 | /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference | |
1001 | * on the ldev here, to be sure the transition -> D_DISKLESS resp. | |
1002 | * drbd_ldev_destroy() won't happen before our corresponding | |
1003 | * after_state_ch works run, where we put_ldev again. */ | |
1004 | if ((os.disk != D_FAILED && ns.disk == D_FAILED) || | |
1005 | (os.disk != D_DISKLESS && ns.disk == D_DISKLESS)) | |
1006 | atomic_inc(&device->local_cnt); | |
1007 | ||
1008 | did_remote = drbd_should_do_remote(device->state); | |
1009 | device->state.i = ns.i; | |
1010 | should_do_remote = drbd_should_do_remote(device->state); | |
1011 | first_peer_device(device)->connection->susp = ns.susp; | |
1012 | first_peer_device(device)->connection->susp_nod = ns.susp_nod; | |
1013 | first_peer_device(device)->connection->susp_fen = ns.susp_fen; | |
1014 | ||
1015 | /* put replicated vs not-replicated requests in seperate epochs */ | |
1016 | if (did_remote != should_do_remote) | |
1017 | start_new_tl_epoch(first_peer_device(device)->connection); | |
1018 | ||
1019 | if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING) | |
1020 | drbd_print_uuids(device, "attached to UUIDs"); | |
1021 | ||
1022 | /* Wake up role changes, that were delayed because of connection establishing */ | |
1023 | if (os.conn == C_WF_REPORT_PARAMS && ns.conn != C_WF_REPORT_PARAMS && | |
1024 | no_peer_wf_report_params(first_peer_device(device)->connection)) | |
1025 | clear_bit(STATE_SENT, &first_peer_device(device)->connection->flags); | |
1026 | ||
1027 | wake_up(&device->misc_wait); | |
1028 | wake_up(&device->state_wait); | |
1029 | wake_up(&first_peer_device(device)->connection->ping_wait); | |
1030 | ||
1031 | /* Aborted verify run, or we reached the stop sector. | |
1032 | * Log the last position, unless end-of-device. */ | |
1033 | if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) && | |
1034 | ns.conn <= C_CONNECTED) { | |
1035 | device->ov_start_sector = | |
1036 | BM_BIT_TO_SECT(drbd_bm_bits(device) - device->ov_left); | |
1037 | if (device->ov_left) | |
1038 | drbd_info(device, "Online Verify reached sector %llu\n", | |
1039 | (unsigned long long)device->ov_start_sector); | |
1040 | } | |
1041 | ||
1042 | if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) && | |
1043 | (ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)) { | |
1044 | drbd_info(device, "Syncer continues.\n"); | |
1045 | device->rs_paused += (long)jiffies | |
1046 | -(long)device->rs_mark_time[device->rs_last_mark]; | |
1047 | if (ns.conn == C_SYNC_TARGET) | |
1048 | mod_timer(&device->resync_timer, jiffies); | |
1049 | } | |
1050 | ||
1051 | if ((os.conn == C_SYNC_TARGET || os.conn == C_SYNC_SOURCE) && | |
1052 | (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) { | |
1053 | drbd_info(device, "Resync suspended\n"); | |
1054 | device->rs_mark_time[device->rs_last_mark] = jiffies; | |
1055 | } | |
1056 | ||
1057 | if (os.conn == C_CONNECTED && | |
1058 | (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) { | |
1059 | unsigned long now = jiffies; | |
1060 | int i; | |
1061 | ||
1062 | set_ov_position(device, ns.conn); | |
1063 | device->rs_start = now; | |
1064 | device->rs_last_events = 0; | |
1065 | device->rs_last_sect_ev = 0; | |
1066 | device->ov_last_oos_size = 0; | |
1067 | device->ov_last_oos_start = 0; | |
1068 | ||
1069 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { | |
1070 | device->rs_mark_left[i] = device->ov_left; | |
1071 | device->rs_mark_time[i] = now; | |
1072 | } | |
1073 | ||
1074 | drbd_rs_controller_reset(device); | |
1075 | ||
1076 | if (ns.conn == C_VERIFY_S) { | |
1077 | drbd_info(device, "Starting Online Verify from sector %llu\n", | |
1078 | (unsigned long long)device->ov_position); | |
1079 | mod_timer(&device->resync_timer, jiffies); | |
1080 | } | |
1081 | } | |
1082 | ||
1083 | if (get_ldev(device)) { | |
1084 | u32 mdf = device->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND| | |
1085 | MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE| | |
1086 | MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY); | |
1087 | ||
1088 | mdf &= ~MDF_AL_CLEAN; | |
1089 | if (test_bit(CRASHED_PRIMARY, &device->flags)) | |
1090 | mdf |= MDF_CRASHED_PRIMARY; | |
1091 | if (device->state.role == R_PRIMARY || | |
1092 | (device->state.pdsk < D_INCONSISTENT && device->state.peer == R_PRIMARY)) | |
1093 | mdf |= MDF_PRIMARY_IND; | |
1094 | if (device->state.conn > C_WF_REPORT_PARAMS) | |
1095 | mdf |= MDF_CONNECTED_IND; | |
1096 | if (device->state.disk > D_INCONSISTENT) | |
1097 | mdf |= MDF_CONSISTENT; | |
1098 | if (device->state.disk > D_OUTDATED) | |
1099 | mdf |= MDF_WAS_UP_TO_DATE; | |
1100 | if (device->state.pdsk <= D_OUTDATED && device->state.pdsk >= D_INCONSISTENT) | |
1101 | mdf |= MDF_PEER_OUT_DATED; | |
1102 | if (mdf != device->ldev->md.flags) { | |
1103 | device->ldev->md.flags = mdf; | |
1104 | drbd_md_mark_dirty(device); | |
1105 | } | |
1106 | if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT) | |
1107 | drbd_set_ed_uuid(device, device->ldev->md.uuid[UI_CURRENT]); | |
1108 | put_ldev(device); | |
1109 | } | |
1110 | ||
1111 | /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */ | |
1112 | if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT && | |
1113 | os.peer == R_SECONDARY && ns.peer == R_PRIMARY) | |
1114 | set_bit(CONSIDER_RESYNC, &device->flags); | |
1115 | ||
1116 | /* Receiver should clean up itself */ | |
1117 | if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING) | |
1118 | drbd_thread_stop_nowait(&first_peer_device(device)->connection->receiver); | |
1119 | ||
1120 | /* Now the receiver finished cleaning up itself, it should die */ | |
1121 | if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE) | |
1122 | drbd_thread_stop_nowait(&first_peer_device(device)->connection->receiver); | |
1123 | ||
1124 | /* Upon network failure, we need to restart the receiver. */ | |
1125 | if (os.conn > C_WF_CONNECTION && | |
1126 | ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT) | |
1127 | drbd_thread_restart_nowait(&first_peer_device(device)->connection->receiver); | |
1128 | ||
1129 | /* Resume AL writing if we get a connection */ | |
1130 | if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) { | |
1131 | drbd_resume_al(device); | |
1132 | first_peer_device(device)->connection->connect_cnt++; | |
1133 | } | |
1134 | ||
1135 | /* remember last attach time so request_timer_fn() won't | |
1136 | * kill newly established sessions while we are still trying to thaw | |
1137 | * previously frozen IO */ | |
1138 | if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) && | |
1139 | ns.disk > D_NEGOTIATING) | |
1140 | device->last_reattach_jif = jiffies; | |
1141 | ||
1142 | ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC); | |
1143 | if (ascw) { | |
1144 | ascw->os = os; | |
1145 | ascw->ns = ns; | |
1146 | ascw->flags = flags; | |
1147 | ascw->w.cb = w_after_state_ch; | |
1148 | ascw->w.device = device; | |
1149 | ascw->done = done; | |
1150 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, &ascw->w); | |
1151 | } else { | |
1152 | drbd_err(device, "Could not kmalloc an ascw\n"); | |
1153 | } | |
1154 | ||
1155 | return rv; | |
1156 | } | |
1157 | ||
1158 | static int w_after_state_ch(struct drbd_work *w, int unused) | |
1159 | { | |
1160 | struct after_state_chg_work *ascw = | |
1161 | container_of(w, struct after_state_chg_work, w); | |
1162 | struct drbd_device *device = w->device; | |
1163 | ||
1164 | after_state_ch(device, ascw->os, ascw->ns, ascw->flags); | |
1165 | if (ascw->flags & CS_WAIT_COMPLETE) { | |
1166 | D_ASSERT(device, ascw->done != NULL); | |
1167 | complete(ascw->done); | |
1168 | } | |
1169 | kfree(ascw); | |
1170 | ||
1171 | return 0; | |
1172 | } | |
1173 | ||
1174 | static void abw_start_sync(struct drbd_device *device, int rv) | |
1175 | { | |
1176 | if (rv) { | |
1177 | drbd_err(device, "Writing the bitmap failed not starting resync.\n"); | |
1178 | _drbd_request_state(device, NS(conn, C_CONNECTED), CS_VERBOSE); | |
1179 | return; | |
1180 | } | |
1181 | ||
1182 | switch (device->state.conn) { | |
1183 | case C_STARTING_SYNC_T: | |
1184 | _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); | |
1185 | break; | |
1186 | case C_STARTING_SYNC_S: | |
1187 | drbd_start_resync(device, C_SYNC_SOURCE); | |
1188 | break; | |
1189 | } | |
1190 | } | |
1191 | ||
1192 | int drbd_bitmap_io_from_worker(struct drbd_device *device, | |
1193 | int (*io_fn)(struct drbd_device *), | |
1194 | char *why, enum bm_flag flags) | |
1195 | { | |
1196 | int rv; | |
1197 | ||
1198 | D_ASSERT(device, current == first_peer_device(device)->connection->worker.task); | |
1199 | ||
1200 | /* open coded non-blocking drbd_suspend_io(device); */ | |
1201 | set_bit(SUSPEND_IO, &device->flags); | |
1202 | ||
1203 | drbd_bm_lock(device, why, flags); | |
1204 | rv = io_fn(device); | |
1205 | drbd_bm_unlock(device); | |
1206 | ||
1207 | drbd_resume_io(device); | |
1208 | ||
1209 | return rv; | |
1210 | } | |
1211 | ||
1212 | /** | |
1213 | * after_state_ch() - Perform after state change actions that may sleep | |
1214 | * @device: DRBD device. | |
1215 | * @os: old state. | |
1216 | * @ns: new state. | |
1217 | * @flags: Flags | |
1218 | */ | |
1219 | static void after_state_ch(struct drbd_device *device, union drbd_state os, | |
1220 | union drbd_state ns, enum chg_state_flags flags) | |
1221 | { | |
1222 | struct sib_info sib; | |
1223 | ||
1224 | sib.sib_reason = SIB_STATE_CHANGE; | |
1225 | sib.os = os; | |
1226 | sib.ns = ns; | |
1227 | ||
1228 | if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) { | |
1229 | clear_bit(CRASHED_PRIMARY, &device->flags); | |
1230 | if (device->p_uuid) | |
1231 | device->p_uuid[UI_FLAGS] &= ~((u64)2); | |
1232 | } | |
1233 | ||
1234 | /* Inform userspace about the change... */ | |
1235 | drbd_bcast_event(device, &sib); | |
1236 | ||
1237 | if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) && | |
1238 | (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)) | |
1239 | drbd_khelper(device, "pri-on-incon-degr"); | |
1240 | ||
1241 | /* Here we have the actions that are performed after a | |
1242 | state change. This function might sleep */ | |
1243 | ||
1244 | if (ns.susp_nod) { | |
1245 | struct drbd_connection *connection = first_peer_device(device)->connection; | |
1246 | enum drbd_req_event what = NOTHING; | |
1247 | ||
1248 | spin_lock_irq(&device->resource->req_lock); | |
1249 | if (os.conn < C_CONNECTED && conn_lowest_conn(connection) >= C_CONNECTED) | |
1250 | what = RESEND; | |
1251 | ||
1252 | if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) && | |
1253 | conn_lowest_disk(connection) > D_NEGOTIATING) | |
1254 | what = RESTART_FROZEN_DISK_IO; | |
1255 | ||
1256 | if (connection->susp_nod && what != NOTHING) { | |
1257 | _tl_restart(connection, what); | |
1258 | _conn_request_state(connection, | |
1259 | (union drbd_state) { { .susp_nod = 1 } }, | |
1260 | (union drbd_state) { { .susp_nod = 0 } }, | |
1261 | CS_VERBOSE); | |
1262 | } | |
1263 | spin_unlock_irq(&device->resource->req_lock); | |
1264 | } | |
1265 | ||
1266 | if (ns.susp_fen) { | |
1267 | struct drbd_connection *connection = first_peer_device(device)->connection; | |
1268 | ||
1269 | spin_lock_irq(&device->resource->req_lock); | |
1270 | if (connection->susp_fen && conn_lowest_conn(connection) >= C_CONNECTED) { | |
1271 | /* case2: The connection was established again: */ | |
1272 | struct drbd_peer_device *peer_device; | |
1273 | int vnr; | |
1274 | ||
1275 | rcu_read_lock(); | |
1276 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) | |
1277 | clear_bit(NEW_CUR_UUID, &peer_device->device->flags); | |
1278 | rcu_read_unlock(); | |
1279 | _tl_restart(connection, RESEND); | |
1280 | _conn_request_state(connection, | |
1281 | (union drbd_state) { { .susp_fen = 1 } }, | |
1282 | (union drbd_state) { { .susp_fen = 0 } }, | |
1283 | CS_VERBOSE); | |
1284 | } | |
1285 | spin_unlock_irq(&device->resource->req_lock); | |
1286 | } | |
1287 | ||
1288 | /* Became sync source. With protocol >= 96, we still need to send out | |
1289 | * the sync uuid now. Need to do that before any drbd_send_state, or | |
1290 | * the other side may go "paused sync" before receiving the sync uuids, | |
1291 | * which is unexpected. */ | |
1292 | if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) && | |
1293 | (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) && | |
1294 | first_peer_device(device)->connection->agreed_pro_version >= 96 && get_ldev(device)) { | |
1295 | drbd_gen_and_send_sync_uuid(device); | |
1296 | put_ldev(device); | |
1297 | } | |
1298 | ||
1299 | /* Do not change the order of the if above and the two below... */ | |
1300 | if (os.pdsk == D_DISKLESS && | |
1301 | ns.pdsk > D_DISKLESS && ns.pdsk != D_UNKNOWN) { /* attach on the peer */ | |
1302 | /* we probably will start a resync soon. | |
1303 | * make sure those things are properly reset. */ | |
1304 | device->rs_total = 0; | |
1305 | device->rs_failed = 0; | |
1306 | atomic_set(&device->rs_pending_cnt, 0); | |
1307 | drbd_rs_cancel_all(device); | |
1308 | ||
1309 | drbd_send_uuids(device); | |
1310 | drbd_send_state(device, ns); | |
1311 | } | |
1312 | /* No point in queuing send_bitmap if we don't have a connection | |
1313 | * anymore, so check also the _current_ state, not only the new state | |
1314 | * at the time this work was queued. */ | |
1315 | if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S && | |
1316 | device->state.conn == C_WF_BITMAP_S) | |
1317 | drbd_queue_bitmap_io(device, &drbd_send_bitmap, NULL, | |
1318 | "send_bitmap (WFBitMapS)", | |
1319 | BM_LOCKED_TEST_ALLOWED); | |
1320 | ||
1321 | /* Lost contact to peer's copy of the data */ | |
1322 | if ((os.pdsk >= D_INCONSISTENT && | |
1323 | os.pdsk != D_UNKNOWN && | |
1324 | os.pdsk != D_OUTDATED) | |
1325 | && (ns.pdsk < D_INCONSISTENT || | |
1326 | ns.pdsk == D_UNKNOWN || | |
1327 | ns.pdsk == D_OUTDATED)) { | |
1328 | if (get_ldev(device)) { | |
1329 | if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) && | |
1330 | device->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) { | |
1331 | if (drbd_suspended(device)) { | |
1332 | set_bit(NEW_CUR_UUID, &device->flags); | |
1333 | } else { | |
1334 | drbd_uuid_new_current(device); | |
1335 | drbd_send_uuids(device); | |
1336 | } | |
1337 | } | |
1338 | put_ldev(device); | |
1339 | } | |
1340 | } | |
1341 | ||
1342 | if (ns.pdsk < D_INCONSISTENT && get_ldev(device)) { | |
1343 | if (os.peer == R_SECONDARY && ns.peer == R_PRIMARY && | |
1344 | device->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) { | |
1345 | drbd_uuid_new_current(device); | |
1346 | drbd_send_uuids(device); | |
1347 | } | |
1348 | /* D_DISKLESS Peer becomes secondary */ | |
1349 | if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY) | |
1350 | /* We may still be Primary ourselves. | |
1351 | * No harm done if the bitmap still changes, | |
1352 | * redirtied pages will follow later. */ | |
1353 | drbd_bitmap_io_from_worker(device, &drbd_bm_write, | |
1354 | "demote diskless peer", BM_LOCKED_SET_ALLOWED); | |
1355 | put_ldev(device); | |
1356 | } | |
1357 | ||
1358 | /* Write out all changed bits on demote. | |
1359 | * Though, no need to da that just yet | |
1360 | * if there is a resync going on still */ | |
1361 | if (os.role == R_PRIMARY && ns.role == R_SECONDARY && | |
1362 | device->state.conn <= C_CONNECTED && get_ldev(device)) { | |
1363 | /* No changes to the bitmap expected this time, so assert that, | |
1364 | * even though no harm was done if it did change. */ | |
1365 | drbd_bitmap_io_from_worker(device, &drbd_bm_write, | |
1366 | "demote", BM_LOCKED_TEST_ALLOWED); | |
1367 | put_ldev(device); | |
1368 | } | |
1369 | ||
1370 | /* Last part of the attaching process ... */ | |
1371 | if (ns.conn >= C_CONNECTED && | |
1372 | os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) { | |
1373 | drbd_send_sizes(device, 0, 0); /* to start sync... */ | |
1374 | drbd_send_uuids(device); | |
1375 | drbd_send_state(device, ns); | |
1376 | } | |
1377 | ||
1378 | /* We want to pause/continue resync, tell peer. */ | |
1379 | if (ns.conn >= C_CONNECTED && | |
1380 | ((os.aftr_isp != ns.aftr_isp) || | |
1381 | (os.user_isp != ns.user_isp))) | |
1382 | drbd_send_state(device, ns); | |
1383 | ||
1384 | /* In case one of the isp bits got set, suspend other devices. */ | |
1385 | if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) && | |
1386 | (ns.aftr_isp || ns.peer_isp || ns.user_isp)) | |
1387 | suspend_other_sg(device); | |
1388 | ||
1389 | /* Make sure the peer gets informed about eventual state | |
1390 | changes (ISP bits) while we were in WFReportParams. */ | |
1391 | if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED) | |
1392 | drbd_send_state(device, ns); | |
1393 | ||
1394 | if (os.conn != C_AHEAD && ns.conn == C_AHEAD) | |
1395 | drbd_send_state(device, ns); | |
1396 | ||
1397 | /* We are in the progress to start a full sync... */ | |
1398 | if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) || | |
1399 | (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S)) | |
1400 | /* no other bitmap changes expected during this phase */ | |
1401 | drbd_queue_bitmap_io(device, | |
1402 | &drbd_bmio_set_n_write, &abw_start_sync, | |
1403 | "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED); | |
1404 | ||
1405 | /* first half of local IO error, failure to attach, | |
1406 | * or administrative detach */ | |
1407 | if (os.disk != D_FAILED && ns.disk == D_FAILED) { | |
1408 | enum drbd_io_error_p eh = EP_PASS_ON; | |
1409 | int was_io_error = 0; | |
1410 | /* corresponding get_ldev was in __drbd_set_state, to serialize | |
1411 | * our cleanup here with the transition to D_DISKLESS. | |
1412 | * But is is still not save to dreference ldev here, since | |
1413 | * we might come from an failed Attach before ldev was set. */ | |
1414 | if (device->ldev) { | |
1415 | rcu_read_lock(); | |
1416 | eh = rcu_dereference(device->ldev->disk_conf)->on_io_error; | |
1417 | rcu_read_unlock(); | |
1418 | ||
1419 | was_io_error = test_and_clear_bit(WAS_IO_ERROR, &device->flags); | |
1420 | ||
1421 | if (was_io_error && eh == EP_CALL_HELPER) | |
1422 | drbd_khelper(device, "local-io-error"); | |
1423 | ||
1424 | /* Immediately allow completion of all application IO, | |
1425 | * that waits for completion from the local disk, | |
1426 | * if this was a force-detach due to disk_timeout | |
1427 | * or administrator request (drbdsetup detach --force). | |
1428 | * Do NOT abort otherwise. | |
1429 | * Aborting local requests may cause serious problems, | |
1430 | * if requests are completed to upper layers already, | |
1431 | * and then later the already submitted local bio completes. | |
1432 | * This can cause DMA into former bio pages that meanwhile | |
1433 | * have been re-used for other things. | |
1434 | * So aborting local requests may cause crashes, | |
1435 | * or even worse, silent data corruption. | |
1436 | */ | |
1437 | if (test_and_clear_bit(FORCE_DETACH, &device->flags)) | |
1438 | tl_abort_disk_io(device); | |
1439 | ||
1440 | /* current state still has to be D_FAILED, | |
1441 | * there is only one way out: to D_DISKLESS, | |
1442 | * and that may only happen after our put_ldev below. */ | |
1443 | if (device->state.disk != D_FAILED) | |
1444 | drbd_err(device, | |
1445 | "ASSERT FAILED: disk is %s during detach\n", | |
1446 | drbd_disk_str(device->state.disk)); | |
1447 | ||
1448 | if (ns.conn >= C_CONNECTED) | |
1449 | drbd_send_state(device, ns); | |
1450 | ||
1451 | drbd_rs_cancel_all(device); | |
1452 | ||
1453 | /* In case we want to get something to stable storage still, | |
1454 | * this may be the last chance. | |
1455 | * Following put_ldev may transition to D_DISKLESS. */ | |
1456 | drbd_md_sync(device); | |
1457 | } | |
1458 | put_ldev(device); | |
1459 | } | |
1460 | ||
1461 | /* second half of local IO error, failure to attach, | |
1462 | * or administrative detach, | |
1463 | * after local_cnt references have reached zero again */ | |
1464 | if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) { | |
1465 | /* We must still be diskless, | |
1466 | * re-attach has to be serialized with this! */ | |
1467 | if (device->state.disk != D_DISKLESS) | |
1468 | drbd_err(device, | |
1469 | "ASSERT FAILED: disk is %s while going diskless\n", | |
1470 | drbd_disk_str(device->state.disk)); | |
1471 | ||
1472 | if (ns.conn >= C_CONNECTED) | |
1473 | drbd_send_state(device, ns); | |
1474 | /* corresponding get_ldev in __drbd_set_state | |
1475 | * this may finally trigger drbd_ldev_destroy. */ | |
1476 | put_ldev(device); | |
1477 | } | |
1478 | ||
1479 | /* Notify peer that I had a local IO error, and did not detached.. */ | |
1480 | if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT && ns.conn >= C_CONNECTED) | |
1481 | drbd_send_state(device, ns); | |
1482 | ||
1483 | /* Disks got bigger while they were detached */ | |
1484 | if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING && | |
1485 | test_and_clear_bit(RESYNC_AFTER_NEG, &device->flags)) { | |
1486 | if (ns.conn == C_CONNECTED) | |
1487 | resync_after_online_grow(device); | |
1488 | } | |
1489 | ||
1490 | /* A resync finished or aborted, wake paused devices... */ | |
1491 | if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) || | |
1492 | (os.peer_isp && !ns.peer_isp) || | |
1493 | (os.user_isp && !ns.user_isp)) | |
1494 | resume_next_sg(device); | |
1495 | ||
1496 | /* sync target done with resync. Explicitly notify peer, even though | |
1497 | * it should (at least for non-empty resyncs) already know itself. */ | |
1498 | if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED) | |
1499 | drbd_send_state(device, ns); | |
1500 | ||
1501 | /* Verify finished, or reached stop sector. Peer did not know about | |
1502 | * the stop sector, and we may even have changed the stop sector during | |
1503 | * verify to interrupt/stop early. Send the new state. */ | |
1504 | if (os.conn == C_VERIFY_S && ns.conn == C_CONNECTED | |
1505 | && verify_can_do_stop_sector(device)) | |
1506 | drbd_send_state(device, ns); | |
1507 | ||
1508 | /* This triggers bitmap writeout of potentially still unwritten pages | |
1509 | * if the resync finished cleanly, or aborted because of peer disk | |
1510 | * failure, or because of connection loss. | |
1511 | * For resync aborted because of local disk failure, we cannot do | |
1512 | * any bitmap writeout anymore. | |
1513 | * No harm done if some bits change during this phase. | |
1514 | */ | |
1515 | if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(device)) { | |
1516 | drbd_queue_bitmap_io(device, &drbd_bm_write_copy_pages, NULL, | |
1517 | "write from resync_finished", BM_LOCKED_CHANGE_ALLOWED); | |
1518 | put_ldev(device); | |
1519 | } | |
1520 | ||
1521 | if (ns.disk == D_DISKLESS && | |
1522 | ns.conn == C_STANDALONE && | |
1523 | ns.role == R_SECONDARY) { | |
1524 | if (os.aftr_isp != ns.aftr_isp) | |
1525 | resume_next_sg(device); | |
1526 | } | |
1527 | ||
1528 | drbd_md_sync(device); | |
1529 | } | |
1530 | ||
1531 | struct after_conn_state_chg_work { | |
1532 | struct drbd_work w; | |
1533 | enum drbd_conns oc; | |
1534 | union drbd_state ns_min; | |
1535 | union drbd_state ns_max; /* new, max state, over all devices */ | |
1536 | enum chg_state_flags flags; | |
1537 | }; | |
1538 | ||
1539 | static int w_after_conn_state_ch(struct drbd_work *w, int unused) | |
1540 | { | |
1541 | struct after_conn_state_chg_work *acscw = | |
1542 | container_of(w, struct after_conn_state_chg_work, w); | |
1543 | struct drbd_connection *connection = w->connection; | |
1544 | enum drbd_conns oc = acscw->oc; | |
1545 | union drbd_state ns_max = acscw->ns_max; | |
1546 | struct drbd_peer_device *peer_device; | |
1547 | int vnr; | |
1548 | ||
1549 | kfree(acscw); | |
1550 | ||
1551 | /* Upon network configuration, we need to start the receiver */ | |
1552 | if (oc == C_STANDALONE && ns_max.conn == C_UNCONNECTED) | |
1553 | drbd_thread_start(&connection->receiver); | |
1554 | ||
1555 | if (oc == C_DISCONNECTING && ns_max.conn == C_STANDALONE) { | |
1556 | struct net_conf *old_conf; | |
1557 | ||
1558 | mutex_lock(&connection->resource->conf_update); | |
1559 | old_conf = connection->net_conf; | |
1560 | connection->my_addr_len = 0; | |
1561 | connection->peer_addr_len = 0; | |
1562 | rcu_assign_pointer(connection->net_conf, NULL); | |
1563 | conn_free_crypto(connection); | |
1564 | mutex_unlock(&connection->resource->conf_update); | |
1565 | ||
1566 | synchronize_rcu(); | |
1567 | kfree(old_conf); | |
1568 | } | |
1569 | ||
1570 | if (ns_max.susp_fen) { | |
1571 | /* case1: The outdate peer handler is successful: */ | |
1572 | if (ns_max.pdsk <= D_OUTDATED) { | |
1573 | rcu_read_lock(); | |
1574 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
1575 | struct drbd_device *device = peer_device->device; | |
1576 | if (test_bit(NEW_CUR_UUID, &device->flags)) { | |
1577 | drbd_uuid_new_current(device); | |
1578 | clear_bit(NEW_CUR_UUID, &device->flags); | |
1579 | } | |
1580 | } | |
1581 | rcu_read_unlock(); | |
1582 | spin_lock_irq(&connection->resource->req_lock); | |
1583 | _tl_restart(connection, CONNECTION_LOST_WHILE_PENDING); | |
1584 | _conn_request_state(connection, | |
1585 | (union drbd_state) { { .susp_fen = 1 } }, | |
1586 | (union drbd_state) { { .susp_fen = 0 } }, | |
1587 | CS_VERBOSE); | |
1588 | spin_unlock_irq(&connection->resource->req_lock); | |
1589 | } | |
1590 | } | |
1591 | kref_put(&connection->kref, drbd_destroy_connection); | |
1592 | ||
1593 | conn_md_sync(connection); | |
1594 | ||
1595 | return 0; | |
1596 | } | |
1597 | ||
1598 | void conn_old_common_state(struct drbd_connection *connection, union drbd_state *pcs, enum chg_state_flags *pf) | |
1599 | { | |
1600 | enum chg_state_flags flags = ~0; | |
1601 | struct drbd_peer_device *peer_device; | |
1602 | int vnr, first_vol = 1; | |
1603 | union drbd_dev_state os, cs = { | |
1604 | { .role = R_SECONDARY, | |
1605 | .peer = R_UNKNOWN, | |
1606 | .conn = connection->cstate, | |
1607 | .disk = D_DISKLESS, | |
1608 | .pdsk = D_UNKNOWN, | |
1609 | } }; | |
1610 | ||
1611 | rcu_read_lock(); | |
1612 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
1613 | struct drbd_device *device = peer_device->device; | |
1614 | os = device->state; | |
1615 | ||
1616 | if (first_vol) { | |
1617 | cs = os; | |
1618 | first_vol = 0; | |
1619 | continue; | |
1620 | } | |
1621 | ||
1622 | if (cs.role != os.role) | |
1623 | flags &= ~CS_DC_ROLE; | |
1624 | ||
1625 | if (cs.peer != os.peer) | |
1626 | flags &= ~CS_DC_PEER; | |
1627 | ||
1628 | if (cs.conn != os.conn) | |
1629 | flags &= ~CS_DC_CONN; | |
1630 | ||
1631 | if (cs.disk != os.disk) | |
1632 | flags &= ~CS_DC_DISK; | |
1633 | ||
1634 | if (cs.pdsk != os.pdsk) | |
1635 | flags &= ~CS_DC_PDSK; | |
1636 | } | |
1637 | rcu_read_unlock(); | |
1638 | ||
1639 | *pf |= CS_DC_MASK; | |
1640 | *pf &= flags; | |
1641 | (*pcs).i = cs.i; | |
1642 | } | |
1643 | ||
1644 | static enum drbd_state_rv | |
1645 | conn_is_valid_transition(struct drbd_connection *connection, union drbd_state mask, union drbd_state val, | |
1646 | enum chg_state_flags flags) | |
1647 | { | |
1648 | enum drbd_state_rv rv = SS_SUCCESS; | |
1649 | union drbd_state ns, os; | |
1650 | struct drbd_peer_device *peer_device; | |
1651 | int vnr; | |
1652 | ||
1653 | rcu_read_lock(); | |
1654 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
1655 | struct drbd_device *device = peer_device->device; | |
1656 | os = drbd_read_state(device); | |
1657 | ns = sanitize_state(device, apply_mask_val(os, mask, val), NULL); | |
1658 | ||
1659 | if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED) | |
1660 | ns.disk = os.disk; | |
1661 | ||
1662 | if (ns.i == os.i) | |
1663 | continue; | |
1664 | ||
1665 | rv = is_valid_transition(os, ns); | |
1666 | ||
1667 | if (rv >= SS_SUCCESS && !(flags & CS_HARD)) { | |
1668 | rv = is_valid_state(device, ns); | |
1669 | if (rv < SS_SUCCESS) { | |
1670 | if (is_valid_state(device, os) == rv) | |
1671 | rv = is_valid_soft_transition(os, ns, connection); | |
1672 | } else | |
1673 | rv = is_valid_soft_transition(os, ns, connection); | |
1674 | } | |
1675 | ||
1676 | if (rv < SS_SUCCESS) { | |
1677 | if (flags & CS_VERBOSE) | |
1678 | print_st_err(device, os, ns, rv); | |
1679 | break; | |
1680 | } | |
1681 | } | |
1682 | rcu_read_unlock(); | |
1683 | ||
1684 | return rv; | |
1685 | } | |
1686 | ||
1687 | void | |
1688 | conn_set_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val, | |
1689 | union drbd_state *pns_min, union drbd_state *pns_max, enum chg_state_flags flags) | |
1690 | { | |
1691 | union drbd_state ns, os, ns_max = { }; | |
1692 | union drbd_state ns_min = { | |
1693 | { .role = R_MASK, | |
1694 | .peer = R_MASK, | |
1695 | .conn = val.conn, | |
1696 | .disk = D_MASK, | |
1697 | .pdsk = D_MASK | |
1698 | } }; | |
1699 | struct drbd_peer_device *peer_device; | |
1700 | enum drbd_state_rv rv; | |
1701 | int vnr, number_of_volumes = 0; | |
1702 | ||
1703 | if (mask.conn == C_MASK) { | |
1704 | /* remember last connect time so request_timer_fn() won't | |
1705 | * kill newly established sessions while we are still trying to thaw | |
1706 | * previously frozen IO */ | |
1707 | if (connection->cstate != C_WF_REPORT_PARAMS && val.conn == C_WF_REPORT_PARAMS) | |
1708 | connection->last_reconnect_jif = jiffies; | |
1709 | ||
1710 | connection->cstate = val.conn; | |
1711 | } | |
1712 | ||
1713 | rcu_read_lock(); | |
1714 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { | |
1715 | struct drbd_device *device = peer_device->device; | |
1716 | number_of_volumes++; | |
1717 | os = drbd_read_state(device); | |
1718 | ns = apply_mask_val(os, mask, val); | |
1719 | ns = sanitize_state(device, ns, NULL); | |
1720 | ||
1721 | if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED) | |
1722 | ns.disk = os.disk; | |
1723 | ||
1724 | rv = __drbd_set_state(device, ns, flags, NULL); | |
1725 | if (rv < SS_SUCCESS) | |
1726 | BUG(); | |
1727 | ||
1728 | ns.i = device->state.i; | |
1729 | ns_max.role = max_role(ns.role, ns_max.role); | |
1730 | ns_max.peer = max_role(ns.peer, ns_max.peer); | |
1731 | ns_max.conn = max_t(enum drbd_conns, ns.conn, ns_max.conn); | |
1732 | ns_max.disk = max_t(enum drbd_disk_state, ns.disk, ns_max.disk); | |
1733 | ns_max.pdsk = max_t(enum drbd_disk_state, ns.pdsk, ns_max.pdsk); | |
1734 | ||
1735 | ns_min.role = min_role(ns.role, ns_min.role); | |
1736 | ns_min.peer = min_role(ns.peer, ns_min.peer); | |
1737 | ns_min.conn = min_t(enum drbd_conns, ns.conn, ns_min.conn); | |
1738 | ns_min.disk = min_t(enum drbd_disk_state, ns.disk, ns_min.disk); | |
1739 | ns_min.pdsk = min_t(enum drbd_disk_state, ns.pdsk, ns_min.pdsk); | |
1740 | } | |
1741 | rcu_read_unlock(); | |
1742 | ||
1743 | if (number_of_volumes == 0) { | |
1744 | ns_min = ns_max = (union drbd_state) { { | |
1745 | .role = R_SECONDARY, | |
1746 | .peer = R_UNKNOWN, | |
1747 | .conn = val.conn, | |
1748 | .disk = D_DISKLESS, | |
1749 | .pdsk = D_UNKNOWN | |
1750 | } }; | |
1751 | } | |
1752 | ||
1753 | ns_min.susp = ns_max.susp = connection->susp; | |
1754 | ns_min.susp_nod = ns_max.susp_nod = connection->susp_nod; | |
1755 | ns_min.susp_fen = ns_max.susp_fen = connection->susp_fen; | |
1756 | ||
1757 | *pns_min = ns_min; | |
1758 | *pns_max = ns_max; | |
1759 | } | |
1760 | ||
1761 | static enum drbd_state_rv | |
1762 | _conn_rq_cond(struct drbd_connection *connection, union drbd_state mask, union drbd_state val) | |
1763 | { | |
1764 | enum drbd_state_rv rv; | |
1765 | ||
1766 | if (test_and_clear_bit(CONN_WD_ST_CHG_OKAY, &connection->flags)) | |
1767 | return SS_CW_SUCCESS; | |
1768 | ||
1769 | if (test_and_clear_bit(CONN_WD_ST_CHG_FAIL, &connection->flags)) | |
1770 | return SS_CW_FAILED_BY_PEER; | |
1771 | ||
1772 | rv = conn_is_valid_transition(connection, mask, val, 0); | |
1773 | if (rv == SS_SUCCESS && connection->cstate == C_WF_REPORT_PARAMS) | |
1774 | rv = SS_UNKNOWN_ERROR; /* continue waiting */ | |
1775 | ||
1776 | return rv; | |
1777 | } | |
1778 | ||
1779 | enum drbd_state_rv | |
1780 | _conn_request_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val, | |
1781 | enum chg_state_flags flags) | |
1782 | { | |
1783 | enum drbd_state_rv rv = SS_SUCCESS; | |
1784 | struct after_conn_state_chg_work *acscw; | |
1785 | enum drbd_conns oc = connection->cstate; | |
1786 | union drbd_state ns_max, ns_min, os; | |
1787 | bool have_mutex = false; | |
1788 | ||
1789 | if (mask.conn) { | |
1790 | rv = is_valid_conn_transition(oc, val.conn); | |
1791 | if (rv < SS_SUCCESS) | |
1792 | goto abort; | |
1793 | } | |
1794 | ||
1795 | rv = conn_is_valid_transition(connection, mask, val, flags); | |
1796 | if (rv < SS_SUCCESS) | |
1797 | goto abort; | |
1798 | ||
1799 | if (oc == C_WF_REPORT_PARAMS && val.conn == C_DISCONNECTING && | |
1800 | !(flags & (CS_LOCAL_ONLY | CS_HARD))) { | |
1801 | ||
1802 | /* This will be a cluster-wide state change. | |
1803 | * Need to give up the spinlock, grab the mutex, | |
1804 | * then send the state change request, ... */ | |
1805 | spin_unlock_irq(&connection->resource->req_lock); | |
1806 | mutex_lock(&connection->cstate_mutex); | |
1807 | have_mutex = true; | |
1808 | ||
1809 | set_bit(CONN_WD_ST_CHG_REQ, &connection->flags); | |
1810 | if (conn_send_state_req(connection, mask, val)) { | |
1811 | /* sending failed. */ | |
1812 | clear_bit(CONN_WD_ST_CHG_REQ, &connection->flags); | |
1813 | rv = SS_CW_FAILED_BY_PEER; | |
1814 | /* need to re-aquire the spin lock, though */ | |
1815 | goto abort_unlocked; | |
1816 | } | |
1817 | ||
1818 | if (val.conn == C_DISCONNECTING) | |
1819 | set_bit(DISCONNECT_SENT, &connection->flags); | |
1820 | ||
1821 | /* ... and re-aquire the spinlock. | |
1822 | * If _conn_rq_cond() returned >= SS_SUCCESS, we must call | |
1823 | * conn_set_state() within the same spinlock. */ | |
1824 | spin_lock_irq(&connection->resource->req_lock); | |
1825 | wait_event_lock_irq(connection->ping_wait, | |
1826 | (rv = _conn_rq_cond(connection, mask, val)), | |
1827 | connection->resource->req_lock); | |
1828 | clear_bit(CONN_WD_ST_CHG_REQ, &connection->flags); | |
1829 | if (rv < SS_SUCCESS) | |
1830 | goto abort; | |
1831 | } | |
1832 | ||
1833 | conn_old_common_state(connection, &os, &flags); | |
1834 | flags |= CS_DC_SUSP; | |
1835 | conn_set_state(connection, mask, val, &ns_min, &ns_max, flags); | |
1836 | conn_pr_state_change(connection, os, ns_max, flags); | |
1837 | ||
1838 | acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC); | |
1839 | if (acscw) { | |
1840 | acscw->oc = os.conn; | |
1841 | acscw->ns_min = ns_min; | |
1842 | acscw->ns_max = ns_max; | |
1843 | acscw->flags = flags; | |
1844 | acscw->w.cb = w_after_conn_state_ch; | |
1845 | kref_get(&connection->kref); | |
1846 | acscw->w.connection = connection; | |
1847 | drbd_queue_work(&connection->sender_work, &acscw->w); | |
1848 | } else { | |
1849 | drbd_err(connection, "Could not kmalloc an acscw\n"); | |
1850 | } | |
1851 | ||
1852 | abort: | |
1853 | if (have_mutex) { | |
1854 | /* mutex_unlock() "... must not be used in interrupt context.", | |
1855 | * so give up the spinlock, then re-aquire it */ | |
1856 | spin_unlock_irq(&connection->resource->req_lock); | |
1857 | abort_unlocked: | |
1858 | mutex_unlock(&connection->cstate_mutex); | |
1859 | spin_lock_irq(&connection->resource->req_lock); | |
1860 | } | |
1861 | if (rv < SS_SUCCESS && flags & CS_VERBOSE) { | |
1862 | drbd_err(connection, "State change failed: %s\n", drbd_set_st_err_str(rv)); | |
1863 | drbd_err(connection, " mask = 0x%x val = 0x%x\n", mask.i, val.i); | |
1864 | drbd_err(connection, " old_conn:%s wanted_conn:%s\n", drbd_conn_str(oc), drbd_conn_str(val.conn)); | |
1865 | } | |
1866 | return rv; | |
1867 | } | |
1868 | ||
1869 | enum drbd_state_rv | |
1870 | conn_request_state(struct drbd_connection *connection, union drbd_state mask, union drbd_state val, | |
1871 | enum chg_state_flags flags) | |
1872 | { | |
1873 | enum drbd_state_rv rv; | |
1874 | ||
1875 | spin_lock_irq(&connection->resource->req_lock); | |
1876 | rv = _conn_request_state(connection, mask, val, flags); | |
1877 | spin_unlock_irq(&connection->resource->req_lock); | |
1878 | ||
1879 | return rv; | |
1880 | } |