]>
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 | ||
99920dc5 | 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 | |
2aebfabb PR |
50 | static inline bool is_susp(union drbd_state s) |
51 | { | |
52 | return s.susp || s.susp_nod || s.susp_fen; | |
53 | } | |
54 | ||
d0456c72 | 55 | bool conn_all_vols_unconf(struct drbd_tconn *tconn) |
0e29d163 PR |
56 | { |
57 | struct drbd_conf *mdev; | |
695d08fa | 58 | bool rv = true; |
e90285e0 | 59 | int vnr; |
0e29d163 | 60 | |
695d08fa | 61 | rcu_read_lock(); |
e90285e0 | 62 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
d0456c72 LE |
63 | if (mdev->state.disk != D_DISKLESS || |
64 | mdev->state.conn != C_STANDALONE || | |
695d08fa PR |
65 | mdev->state.role != R_SECONDARY) { |
66 | rv = false; | |
67 | break; | |
68 | } | |
0e29d163 | 69 | } |
695d08fa PR |
70 | rcu_read_unlock(); |
71 | ||
72 | return rv; | |
0e29d163 PR |
73 | } |
74 | ||
cb703454 PR |
75 | /* Unfortunately the states where not correctly ordered, when |
76 | they where defined. therefore can not use max_t() here. */ | |
77 | static enum drbd_role max_role(enum drbd_role role1, enum drbd_role role2) | |
78 | { | |
79 | if (role1 == R_PRIMARY || role2 == R_PRIMARY) | |
80 | return R_PRIMARY; | |
81 | if (role1 == R_SECONDARY || role2 == R_SECONDARY) | |
82 | return R_SECONDARY; | |
83 | return R_UNKNOWN; | |
84 | } | |
85 | static enum drbd_role min_role(enum drbd_role role1, enum drbd_role role2) | |
86 | { | |
87 | if (role1 == R_UNKNOWN || role2 == R_UNKNOWN) | |
88 | return R_UNKNOWN; | |
89 | if (role1 == R_SECONDARY || role2 == R_SECONDARY) | |
90 | return R_SECONDARY; | |
91 | return R_PRIMARY; | |
92 | } | |
93 | ||
94 | enum drbd_role conn_highest_role(struct drbd_tconn *tconn) | |
95 | { | |
96 | enum drbd_role role = R_UNKNOWN; | |
97 | struct drbd_conf *mdev; | |
98 | int vnr; | |
99 | ||
695d08fa | 100 | rcu_read_lock(); |
cb703454 PR |
101 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
102 | role = max_role(role, mdev->state.role); | |
695d08fa | 103 | rcu_read_unlock(); |
cb703454 PR |
104 | |
105 | return role; | |
106 | } | |
107 | ||
108 | enum drbd_role conn_highest_peer(struct drbd_tconn *tconn) | |
109 | { | |
110 | enum drbd_role peer = R_UNKNOWN; | |
111 | struct drbd_conf *mdev; | |
112 | int vnr; | |
113 | ||
695d08fa | 114 | rcu_read_lock(); |
cb703454 PR |
115 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
116 | peer = max_role(peer, mdev->state.peer); | |
695d08fa | 117 | rcu_read_unlock(); |
cb703454 PR |
118 | |
119 | return peer; | |
120 | } | |
121 | ||
122 | enum drbd_disk_state conn_highest_disk(struct drbd_tconn *tconn) | |
123 | { | |
124 | enum drbd_disk_state ds = D_DISKLESS; | |
125 | struct drbd_conf *mdev; | |
126 | int vnr; | |
127 | ||
695d08fa | 128 | rcu_read_lock(); |
cb703454 PR |
129 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
130 | ds = max_t(enum drbd_disk_state, ds, mdev->state.disk); | |
695d08fa | 131 | rcu_read_unlock(); |
cb703454 PR |
132 | |
133 | return ds; | |
134 | } | |
135 | ||
4669265a PR |
136 | enum drbd_disk_state conn_lowest_disk(struct drbd_tconn *tconn) |
137 | { | |
138 | enum drbd_disk_state ds = D_MASK; | |
139 | struct drbd_conf *mdev; | |
140 | int vnr; | |
141 | ||
695d08fa | 142 | rcu_read_lock(); |
4669265a PR |
143 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
144 | ds = min_t(enum drbd_disk_state, ds, mdev->state.disk); | |
695d08fa | 145 | rcu_read_unlock(); |
4669265a PR |
146 | |
147 | return ds; | |
148 | } | |
149 | ||
cb703454 PR |
150 | enum drbd_disk_state conn_highest_pdsk(struct drbd_tconn *tconn) |
151 | { | |
152 | enum drbd_disk_state ds = D_DISKLESS; | |
153 | struct drbd_conf *mdev; | |
154 | int vnr; | |
155 | ||
695d08fa | 156 | rcu_read_lock(); |
cb703454 PR |
157 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
158 | ds = max_t(enum drbd_disk_state, ds, mdev->state.pdsk); | |
695d08fa | 159 | rcu_read_unlock(); |
cb703454 PR |
160 | |
161 | return ds; | |
162 | } | |
163 | ||
19f83c76 PR |
164 | enum drbd_conns conn_lowest_conn(struct drbd_tconn *tconn) |
165 | { | |
166 | enum drbd_conns conn = C_MASK; | |
167 | struct drbd_conf *mdev; | |
168 | int vnr; | |
169 | ||
695d08fa | 170 | rcu_read_lock(); |
19f83c76 PR |
171 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
172 | conn = min_t(enum drbd_conns, conn, mdev->state.conn); | |
695d08fa | 173 | rcu_read_unlock(); |
19f83c76 PR |
174 | |
175 | return conn; | |
176 | } | |
177 | ||
b8907339 PR |
178 | /** |
179 | * cl_wide_st_chg() - true if the state change is a cluster wide one | |
180 | * @mdev: DRBD device. | |
181 | * @os: old (current) state. | |
182 | * @ns: new (wanted) state. | |
183 | */ | |
184 | static int cl_wide_st_chg(struct drbd_conf *mdev, | |
185 | union drbd_state os, union drbd_state ns) | |
186 | { | |
187 | return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED && | |
188 | ((os.role != R_PRIMARY && ns.role == R_PRIMARY) || | |
189 | (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) || | |
190 | (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) || | |
191 | (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))) || | |
192 | (os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) || | |
193 | (os.conn == C_CONNECTED && ns.conn == C_VERIFY_S); | |
194 | } | |
195 | ||
56707f9e PR |
196 | static union drbd_state |
197 | apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val) | |
198 | { | |
199 | union drbd_state ns; | |
200 | ns.i = (os.i & ~mask.i) | val.i; | |
201 | return ns; | |
202 | } | |
203 | ||
b8907339 PR |
204 | enum drbd_state_rv |
205 | drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f, | |
206 | union drbd_state mask, union drbd_state val) | |
207 | { | |
208 | unsigned long flags; | |
56707f9e | 209 | union drbd_state ns; |
b8907339 PR |
210 | enum drbd_state_rv rv; |
211 | ||
212 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); | |
78bae59b | 213 | ns = apply_mask_val(drbd_read_state(mdev), mask, val); |
b8907339 | 214 | rv = _drbd_set_state(mdev, ns, f, NULL); |
b8907339 PR |
215 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
216 | ||
217 | return rv; | |
218 | } | |
219 | ||
220 | /** | |
221 | * drbd_force_state() - Impose a change which happens outside our control on our state | |
222 | * @mdev: DRBD device. | |
223 | * @mask: mask of state bits to change. | |
224 | * @val: value of new state bits. | |
225 | */ | |
226 | void drbd_force_state(struct drbd_conf *mdev, | |
227 | union drbd_state mask, union drbd_state val) | |
228 | { | |
229 | drbd_change_state(mdev, CS_HARD, mask, val); | |
230 | } | |
231 | ||
232 | static enum drbd_state_rv | |
233 | _req_st_cond(struct drbd_conf *mdev, union drbd_state mask, | |
234 | union drbd_state val) | |
235 | { | |
236 | union drbd_state os, ns; | |
237 | unsigned long flags; | |
238 | enum drbd_state_rv rv; | |
239 | ||
240 | if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &mdev->flags)) | |
241 | return SS_CW_SUCCESS; | |
242 | ||
243 | if (test_and_clear_bit(CL_ST_CHG_FAIL, &mdev->flags)) | |
244 | return SS_CW_FAILED_BY_PEER; | |
245 | ||
b8907339 | 246 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
78bae59b | 247 | os = drbd_read_state(mdev); |
56707f9e | 248 | ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL); |
3509502d PR |
249 | rv = is_valid_transition(os, ns); |
250 | if (rv == SS_SUCCESS) | |
251 | rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */ | |
b8907339 PR |
252 | |
253 | if (!cl_wide_st_chg(mdev, os, ns)) | |
254 | rv = SS_CW_NO_NEED; | |
3509502d | 255 | if (rv == SS_UNKNOWN_ERROR) { |
b8907339 PR |
256 | rv = is_valid_state(mdev, ns); |
257 | if (rv == SS_SUCCESS) { | |
a75f34ad | 258 | rv = is_valid_soft_transition(os, ns); |
b8907339 PR |
259 | if (rv == SS_SUCCESS) |
260 | rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */ | |
261 | } | |
262 | } | |
263 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
264 | ||
265 | return rv; | |
266 | } | |
267 | ||
268 | /** | |
269 | * drbd_req_state() - Perform an eventually cluster wide state change | |
270 | * @mdev: DRBD device. | |
271 | * @mask: mask of state bits to change. | |
272 | * @val: value of new state bits. | |
273 | * @f: flags | |
274 | * | |
275 | * Should not be called directly, use drbd_request_state() or | |
276 | * _drbd_request_state(). | |
277 | */ | |
278 | static enum drbd_state_rv | |
279 | drbd_req_state(struct drbd_conf *mdev, union drbd_state mask, | |
280 | union drbd_state val, enum chg_state_flags f) | |
281 | { | |
282 | struct completion done; | |
283 | unsigned long flags; | |
284 | union drbd_state os, ns; | |
285 | enum drbd_state_rv rv; | |
286 | ||
287 | init_completion(&done); | |
288 | ||
289 | if (f & CS_SERIALIZE) | |
8410da8f | 290 | mutex_lock(mdev->state_mutex); |
b8907339 PR |
291 | |
292 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); | |
78bae59b | 293 | os = drbd_read_state(mdev); |
56707f9e | 294 | ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL); |
3509502d | 295 | rv = is_valid_transition(os, ns); |
3c5e5f6a LE |
296 | if (rv < SS_SUCCESS) { |
297 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
3509502d | 298 | goto abort; |
3c5e5f6a | 299 | } |
b8907339 PR |
300 | |
301 | if (cl_wide_st_chg(mdev, os, ns)) { | |
302 | rv = is_valid_state(mdev, ns); | |
303 | if (rv == SS_SUCCESS) | |
a75f34ad | 304 | rv = is_valid_soft_transition(os, ns); |
b8907339 PR |
305 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
306 | ||
307 | if (rv < SS_SUCCESS) { | |
308 | if (f & CS_VERBOSE) | |
309 | print_st_err(mdev, os, ns, rv); | |
310 | goto abort; | |
311 | } | |
312 | ||
d24ae219 | 313 | if (drbd_send_state_req(mdev, mask, val)) { |
b8907339 PR |
314 | rv = SS_CW_FAILED_BY_PEER; |
315 | if (f & CS_VERBOSE) | |
316 | print_st_err(mdev, os, ns, rv); | |
317 | goto abort; | |
318 | } | |
319 | ||
320 | wait_event(mdev->state_wait, | |
321 | (rv = _req_st_cond(mdev, mask, val))); | |
322 | ||
323 | if (rv < SS_SUCCESS) { | |
b8907339 PR |
324 | if (f & CS_VERBOSE) |
325 | print_st_err(mdev, os, ns, rv); | |
326 | goto abort; | |
327 | } | |
328 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); | |
78bae59b | 329 | ns = apply_mask_val(drbd_read_state(mdev), mask, val); |
b8907339 | 330 | rv = _drbd_set_state(mdev, ns, f, &done); |
b8907339 PR |
331 | } else { |
332 | rv = _drbd_set_state(mdev, ns, f, &done); | |
333 | } | |
334 | ||
335 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
336 | ||
337 | if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) { | |
338 | D_ASSERT(current != mdev->tconn->worker.task); | |
339 | wait_for_completion(&done); | |
340 | } | |
341 | ||
342 | abort: | |
343 | if (f & CS_SERIALIZE) | |
8410da8f | 344 | mutex_unlock(mdev->state_mutex); |
b8907339 PR |
345 | |
346 | return rv; | |
347 | } | |
348 | ||
349 | /** | |
350 | * _drbd_request_state() - Request a state change (with flags) | |
351 | * @mdev: DRBD device. | |
352 | * @mask: mask of state bits to change. | |
353 | * @val: value of new state bits. | |
354 | * @f: flags | |
355 | * | |
356 | * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE | |
357 | * flag, or when logging of failed state change requests is not desired. | |
358 | */ | |
359 | enum drbd_state_rv | |
360 | _drbd_request_state(struct drbd_conf *mdev, union drbd_state mask, | |
361 | union drbd_state val, enum chg_state_flags f) | |
362 | { | |
363 | enum drbd_state_rv rv; | |
364 | ||
365 | wait_event(mdev->state_wait, | |
366 | (rv = drbd_req_state(mdev, mask, val, f)) != SS_IN_TRANSIENT_STATE); | |
367 | ||
368 | return rv; | |
369 | } | |
370 | ||
371 | static void print_st(struct drbd_conf *mdev, char *name, union drbd_state ns) | |
372 | { | |
373 | dev_err(DEV, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n", | |
374 | name, | |
375 | drbd_conn_str(ns.conn), | |
376 | drbd_role_str(ns.role), | |
377 | drbd_role_str(ns.peer), | |
378 | drbd_disk_str(ns.disk), | |
379 | drbd_disk_str(ns.pdsk), | |
380 | is_susp(ns) ? 's' : 'r', | |
381 | ns.aftr_isp ? 'a' : '-', | |
382 | ns.peer_isp ? 'p' : '-', | |
383 | ns.user_isp ? 'u' : '-', | |
384 | ns.susp_fen ? 'F' : '-', | |
385 | ns.susp_nod ? 'N' : '-' | |
386 | ); | |
387 | } | |
388 | ||
389 | void print_st_err(struct drbd_conf *mdev, union drbd_state os, | |
390 | union drbd_state ns, enum drbd_state_rv err) | |
391 | { | |
392 | if (err == SS_IN_TRANSIENT_STATE) | |
393 | return; | |
394 | dev_err(DEV, "State change failed: %s\n", drbd_set_st_err_str(err)); | |
395 | print_st(mdev, " state", os); | |
396 | print_st(mdev, "wanted", ns); | |
397 | } | |
398 | ||
435693e8 | 399 | static long print_state_change(char *pb, union drbd_state os, union drbd_state ns, |
bbeb641c PR |
400 | enum chg_state_flags flags) |
401 | { | |
435693e8 | 402 | char *pbp; |
bbeb641c PR |
403 | pbp = pb; |
404 | *pbp = 0; | |
706cb24c | 405 | |
435693e8 | 406 | if (ns.role != os.role && flags & CS_DC_ROLE) |
bbeb641c PR |
407 | pbp += sprintf(pbp, "role( %s -> %s ) ", |
408 | drbd_role_str(os.role), | |
409 | drbd_role_str(ns.role)); | |
435693e8 | 410 | if (ns.peer != os.peer && flags & CS_DC_PEER) |
bbeb641c PR |
411 | pbp += sprintf(pbp, "peer( %s -> %s ) ", |
412 | drbd_role_str(os.peer), | |
413 | drbd_role_str(ns.peer)); | |
435693e8 | 414 | if (ns.conn != os.conn && flags & CS_DC_CONN) |
bbeb641c PR |
415 | pbp += sprintf(pbp, "conn( %s -> %s ) ", |
416 | drbd_conn_str(os.conn), | |
417 | drbd_conn_str(ns.conn)); | |
435693e8 | 418 | if (ns.disk != os.disk && flags & CS_DC_DISK) |
bbeb641c PR |
419 | pbp += sprintf(pbp, "disk( %s -> %s ) ", |
420 | drbd_disk_str(os.disk), | |
421 | drbd_disk_str(ns.disk)); | |
435693e8 | 422 | if (ns.pdsk != os.pdsk && flags & CS_DC_PDSK) |
bbeb641c PR |
423 | pbp += sprintf(pbp, "pdsk( %s -> %s ) ", |
424 | drbd_disk_str(os.pdsk), | |
425 | drbd_disk_str(ns.pdsk)); | |
706cb24c PR |
426 | |
427 | return pbp - pb; | |
428 | } | |
429 | ||
430 | static void drbd_pr_state_change(struct drbd_conf *mdev, union drbd_state os, union drbd_state ns, | |
431 | enum chg_state_flags flags) | |
432 | { | |
433 | char pb[300]; | |
434 | char *pbp = pb; | |
435 | ||
436 | pbp += print_state_change(pbp, os, ns, flags ^ CS_DC_MASK); | |
437 | ||
bbeb641c PR |
438 | if (ns.aftr_isp != os.aftr_isp) |
439 | pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ", | |
440 | os.aftr_isp, | |
441 | ns.aftr_isp); | |
442 | if (ns.peer_isp != os.peer_isp) | |
443 | pbp += sprintf(pbp, "peer_isp( %d -> %d ) ", | |
444 | os.peer_isp, | |
445 | ns.peer_isp); | |
446 | if (ns.user_isp != os.user_isp) | |
447 | pbp += sprintf(pbp, "user_isp( %d -> %d ) ", | |
448 | os.user_isp, | |
449 | ns.user_isp); | |
435693e8 | 450 | |
706cb24c | 451 | if (pbp != pb) |
bbeb641c PR |
452 | dev_info(DEV, "%s\n", pb); |
453 | } | |
b8907339 | 454 | |
435693e8 PR |
455 | static void conn_pr_state_change(struct drbd_tconn *tconn, union drbd_state os, union drbd_state ns, |
456 | enum chg_state_flags flags) | |
457 | { | |
458 | char pb[300]; | |
706cb24c PR |
459 | char *pbp = pb; |
460 | ||
461 | pbp += print_state_change(pbp, os, ns, flags); | |
435693e8 | 462 | |
706cb24c PR |
463 | if (is_susp(ns) != is_susp(os) && flags & CS_DC_SUSP) |
464 | pbp += sprintf(pbp, "susp( %d -> %d ) ", | |
465 | is_susp(os), | |
466 | is_susp(ns)); | |
467 | ||
468 | if (pbp != pb) | |
435693e8 PR |
469 | conn_info(tconn, "%s\n", pb); |
470 | } | |
471 | ||
472 | ||
b8907339 PR |
473 | /** |
474 | * is_valid_state() - Returns an SS_ error code if ns is not valid | |
475 | * @mdev: DRBD device. | |
476 | * @ns: State to consider. | |
477 | */ | |
478 | static enum drbd_state_rv | |
479 | is_valid_state(struct drbd_conf *mdev, union drbd_state ns) | |
480 | { | |
481 | /* See drbd_state_sw_errors in drbd_strings.c */ | |
482 | ||
483 | enum drbd_fencing_p fp; | |
484 | enum drbd_state_rv rv = SS_SUCCESS; | |
44ed167d | 485 | struct net_conf *nc; |
b8907339 PR |
486 | |
487 | fp = FP_DONT_CARE; | |
488 | if (get_ldev(mdev)) { | |
489 | fp = mdev->ldev->dc.fencing; | |
490 | put_ldev(mdev); | |
491 | } | |
492 | ||
44ed167d PR |
493 | rcu_read_lock(); |
494 | nc = rcu_dereference(mdev->tconn->net_conf); | |
495 | if (nc) { | |
496 | if (!nc->two_primaries && ns.role == R_PRIMARY) { | |
047e95e2 PR |
497 | if (ns.peer == R_PRIMARY) |
498 | rv = SS_TWO_PRIMARIES; | |
cb703454 | 499 | else if (conn_highest_peer(mdev->tconn) == R_PRIMARY) |
047e95e2 | 500 | rv = SS_O_VOL_PEER_PRI; |
44ed167d | 501 | } |
b8907339 PR |
502 | } |
503 | ||
504 | if (rv <= 0) | |
505 | /* already found a reason to abort */; | |
506 | else if (ns.role == R_SECONDARY && mdev->open_cnt) | |
507 | rv = SS_DEVICE_IN_USE; | |
508 | ||
509 | else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE) | |
510 | rv = SS_NO_UP_TO_DATE_DISK; | |
511 | ||
512 | else if (fp >= FP_RESOURCE && | |
513 | ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN) | |
514 | rv = SS_PRIMARY_NOP; | |
515 | ||
516 | else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT) | |
517 | rv = SS_NO_UP_TO_DATE_DISK; | |
518 | ||
519 | else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT) | |
520 | rv = SS_NO_LOCAL_DISK; | |
521 | ||
522 | else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT) | |
523 | rv = SS_NO_REMOTE_DISK; | |
524 | ||
525 | else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE) | |
526 | rv = SS_NO_UP_TO_DATE_DISK; | |
527 | ||
528 | else if ((ns.conn == C_CONNECTED || | |
529 | ns.conn == C_WF_BITMAP_S || | |
530 | ns.conn == C_SYNC_SOURCE || | |
531 | ns.conn == C_PAUSED_SYNC_S) && | |
532 | ns.disk == D_OUTDATED) | |
533 | rv = SS_CONNECTED_OUTDATES; | |
534 | ||
535 | else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
44ed167d | 536 | (nc->verify_alg[0] == 0)) |
b8907339 PR |
537 | rv = SS_NO_VERIFY_ALG; |
538 | ||
539 | else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
540 | mdev->tconn->agreed_pro_version < 88) | |
541 | rv = SS_NOT_SUPPORTED; | |
542 | ||
543 | else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN) | |
544 | rv = SS_CONNECTED_OUTDATES; | |
545 | ||
44ed167d PR |
546 | rcu_read_unlock(); |
547 | ||
b8907339 PR |
548 | return rv; |
549 | } | |
550 | ||
551 | /** | |
a75f34ad | 552 | * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible |
3509502d PR |
553 | * This function limits state transitions that may be declined by DRBD. I.e. |
554 | * user requests (aka soft transitions). | |
b8907339 PR |
555 | * @mdev: DRBD device. |
556 | * @ns: new state. | |
557 | * @os: old state. | |
558 | */ | |
559 | static enum drbd_state_rv | |
a75f34ad | 560 | is_valid_soft_transition(union drbd_state os, union drbd_state ns) |
b8907339 PR |
561 | { |
562 | enum drbd_state_rv rv = SS_SUCCESS; | |
563 | ||
564 | if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) && | |
565 | os.conn > C_CONNECTED) | |
566 | rv = SS_RESYNC_RUNNING; | |
567 | ||
568 | if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE) | |
569 | rv = SS_ALREADY_STANDALONE; | |
570 | ||
571 | if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS) | |
572 | rv = SS_IS_DISKLESS; | |
573 | ||
574 | if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED) | |
575 | rv = SS_NO_NET_CONFIG; | |
576 | ||
577 | if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING) | |
578 | rv = SS_LOWER_THAN_OUTDATED; | |
579 | ||
580 | if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED) | |
581 | rv = SS_IN_TRANSIENT_STATE; | |
582 | ||
2325eb66 PR |
583 | /* if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS) |
584 | rv = SS_IN_TRANSIENT_STATE; */ | |
b8907339 PR |
585 | |
586 | if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED) | |
587 | rv = SS_NEED_CONNECTION; | |
588 | ||
589 | if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
590 | ns.conn != os.conn && os.conn > C_CONNECTED) | |
591 | rv = SS_RESYNC_RUNNING; | |
592 | ||
593 | if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) && | |
594 | os.conn < C_CONNECTED) | |
595 | rv = SS_NEED_CONNECTION; | |
596 | ||
597 | if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE) | |
598 | && os.conn < C_WF_REPORT_PARAMS) | |
599 | rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */ | |
600 | ||
601 | return rv; | |
602 | } | |
603 | ||
3509502d | 604 | static enum drbd_state_rv |
fda74117 | 605 | is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc) |
3509502d PR |
606 | { |
607 | enum drbd_state_rv rv = SS_SUCCESS; | |
608 | ||
609 | /* Disallow Network errors to configure a device's network part */ | |
fda74117 | 610 | if ((nc >= C_TIMEOUT && nc <= C_TEAR_DOWN) && oc <= C_DISCONNECTING) |
3509502d PR |
611 | rv = SS_NEED_CONNECTION; |
612 | ||
613 | /* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */ | |
fda74117 | 614 | if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING) |
3509502d PR |
615 | rv = SS_IN_TRANSIENT_STATE; |
616 | ||
617 | /* After C_DISCONNECTING only C_STANDALONE may follow */ | |
fda74117 | 618 | if (oc == C_DISCONNECTING && nc != C_STANDALONE) |
3509502d PR |
619 | rv = SS_IN_TRANSIENT_STATE; |
620 | ||
fda74117 PR |
621 | return rv; |
622 | } | |
623 | ||
624 | ||
625 | /** | |
626 | * is_valid_transition() - Returns an SS_ error code if the state transition is not possible | |
627 | * This limits hard state transitions. Hard state transitions are facts there are | |
628 | * imposed on DRBD by the environment. E.g. disk broke or network broke down. | |
629 | * But those hard state transitions are still not allowed to do everything. | |
630 | * @ns: new state. | |
631 | * @os: old state. | |
632 | */ | |
633 | static enum drbd_state_rv | |
634 | is_valid_transition(union drbd_state os, union drbd_state ns) | |
635 | { | |
636 | enum drbd_state_rv rv; | |
637 | ||
638 | rv = is_valid_conn_transition(os.conn, ns.conn); | |
639 | ||
3509502d PR |
640 | /* we cannot fail (again) if we already detached */ |
641 | if (ns.disk == D_FAILED && os.disk == D_DISKLESS) | |
642 | rv = SS_IS_DISKLESS; | |
643 | ||
4308a0a3 PR |
644 | /* if we are only D_ATTACHING yet, |
645 | * we can (and should) go directly to D_DISKLESS. */ | |
646 | if (ns.disk == D_FAILED && os.disk == D_ATTACHING) { | |
647 | printk("TODO: FIX ME\n"); | |
648 | rv = SS_IS_DISKLESS; | |
649 | } | |
650 | ||
3509502d PR |
651 | return rv; |
652 | } | |
653 | ||
b8907339 PR |
654 | /** |
655 | * sanitize_state() - Resolves implicitly necessary additional changes to a state transition | |
656 | * @mdev: DRBD device. | |
657 | * @os: old state. | |
658 | * @ns: new state. | |
659 | * @warn_sync_abort: | |
660 | * | |
661 | * When we loose connection, we have to set the state of the peers disk (pdsk) | |
662 | * to D_UNKNOWN. This rule and many more along those lines are in this function. | |
663 | */ | |
4308a0a3 PR |
664 | static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns, |
665 | const char **warn_sync_abort) | |
b8907339 PR |
666 | { |
667 | enum drbd_fencing_p fp; | |
668 | enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max; | |
669 | ||
670 | fp = FP_DONT_CARE; | |
671 | if (get_ldev(mdev)) { | |
672 | fp = mdev->ldev->dc.fencing; | |
673 | put_ldev(mdev); | |
674 | } | |
675 | ||
3509502d | 676 | /* Implications from connection to peer and peer_isp */ |
b8907339 PR |
677 | if (ns.conn < C_CONNECTED) { |
678 | ns.peer_isp = 0; | |
679 | ns.peer = R_UNKNOWN; | |
680 | if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT) | |
681 | ns.pdsk = D_UNKNOWN; | |
682 | } | |
683 | ||
684 | /* Clear the aftr_isp when becoming unconfigured */ | |
685 | if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY) | |
686 | ns.aftr_isp = 0; | |
687 | ||
4308a0a3 | 688 | /* An implication of the disk states onto the connection state */ |
b8907339 | 689 | /* Abort resync if a disk fails/detaches */ |
4308a0a3 | 690 | if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) { |
b8907339 PR |
691 | if (warn_sync_abort) |
692 | *warn_sync_abort = | |
4308a0a3 | 693 | ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ? |
b8907339 PR |
694 | "Online-verify" : "Resync"; |
695 | ns.conn = C_CONNECTED; | |
696 | } | |
697 | ||
698 | /* Connection breaks down before we finished "Negotiating" */ | |
699 | if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING && | |
700 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
701 | if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) { | |
702 | ns.disk = mdev->new_state_tmp.disk; | |
703 | ns.pdsk = mdev->new_state_tmp.pdsk; | |
704 | } else { | |
705 | dev_alert(DEV, "Connection lost while negotiating, no data!\n"); | |
706 | ns.disk = D_DISKLESS; | |
707 | ns.pdsk = D_UNKNOWN; | |
708 | } | |
709 | put_ldev(mdev); | |
710 | } | |
711 | ||
712 | /* D_CONSISTENT and D_OUTDATED vanish when we get connected */ | |
713 | if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) { | |
714 | if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED) | |
715 | ns.disk = D_UP_TO_DATE; | |
716 | if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED) | |
717 | ns.pdsk = D_UP_TO_DATE; | |
718 | } | |
719 | ||
720 | /* Implications of the connection stat on the disk states */ | |
721 | disk_min = D_DISKLESS; | |
722 | disk_max = D_UP_TO_DATE; | |
723 | pdsk_min = D_INCONSISTENT; | |
724 | pdsk_max = D_UNKNOWN; | |
725 | switch ((enum drbd_conns)ns.conn) { | |
726 | case C_WF_BITMAP_T: | |
727 | case C_PAUSED_SYNC_T: | |
728 | case C_STARTING_SYNC_T: | |
729 | case C_WF_SYNC_UUID: | |
730 | case C_BEHIND: | |
731 | disk_min = D_INCONSISTENT; | |
732 | disk_max = D_OUTDATED; | |
733 | pdsk_min = D_UP_TO_DATE; | |
734 | pdsk_max = D_UP_TO_DATE; | |
735 | break; | |
736 | case C_VERIFY_S: | |
737 | case C_VERIFY_T: | |
738 | disk_min = D_UP_TO_DATE; | |
739 | disk_max = D_UP_TO_DATE; | |
740 | pdsk_min = D_UP_TO_DATE; | |
741 | pdsk_max = D_UP_TO_DATE; | |
742 | break; | |
743 | case C_CONNECTED: | |
744 | disk_min = D_DISKLESS; | |
745 | disk_max = D_UP_TO_DATE; | |
746 | pdsk_min = D_DISKLESS; | |
747 | pdsk_max = D_UP_TO_DATE; | |
748 | break; | |
749 | case C_WF_BITMAP_S: | |
750 | case C_PAUSED_SYNC_S: | |
751 | case C_STARTING_SYNC_S: | |
752 | case C_AHEAD: | |
753 | disk_min = D_UP_TO_DATE; | |
754 | disk_max = D_UP_TO_DATE; | |
755 | pdsk_min = D_INCONSISTENT; | |
756 | pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/ | |
757 | break; | |
758 | case C_SYNC_TARGET: | |
759 | disk_min = D_INCONSISTENT; | |
760 | disk_max = D_INCONSISTENT; | |
761 | pdsk_min = D_UP_TO_DATE; | |
762 | pdsk_max = D_UP_TO_DATE; | |
763 | break; | |
764 | case C_SYNC_SOURCE: | |
765 | disk_min = D_UP_TO_DATE; | |
766 | disk_max = D_UP_TO_DATE; | |
767 | pdsk_min = D_INCONSISTENT; | |
768 | pdsk_max = D_INCONSISTENT; | |
769 | break; | |
770 | case C_STANDALONE: | |
771 | case C_DISCONNECTING: | |
772 | case C_UNCONNECTED: | |
773 | case C_TIMEOUT: | |
774 | case C_BROKEN_PIPE: | |
775 | case C_NETWORK_FAILURE: | |
776 | case C_PROTOCOL_ERROR: | |
777 | case C_TEAR_DOWN: | |
778 | case C_WF_CONNECTION: | |
779 | case C_WF_REPORT_PARAMS: | |
780 | case C_MASK: | |
781 | break; | |
782 | } | |
783 | if (ns.disk > disk_max) | |
784 | ns.disk = disk_max; | |
785 | ||
786 | if (ns.disk < disk_min) { | |
787 | dev_warn(DEV, "Implicitly set disk from %s to %s\n", | |
788 | drbd_disk_str(ns.disk), drbd_disk_str(disk_min)); | |
789 | ns.disk = disk_min; | |
790 | } | |
791 | if (ns.pdsk > pdsk_max) | |
792 | ns.pdsk = pdsk_max; | |
793 | ||
794 | if (ns.pdsk < pdsk_min) { | |
795 | dev_warn(DEV, "Implicitly set pdsk from %s to %s\n", | |
796 | drbd_disk_str(ns.pdsk), drbd_disk_str(pdsk_min)); | |
797 | ns.pdsk = pdsk_min; | |
798 | } | |
799 | ||
800 | if (fp == FP_STONITH && | |
4308a0a3 | 801 | (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED)) |
b8907339 PR |
802 | ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */ |
803 | ||
f399002e | 804 | if (mdev->tconn->res_opts.on_no_data == OND_SUSPEND_IO && |
4308a0a3 | 805 | (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)) |
b8907339 PR |
806 | ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */ |
807 | ||
808 | if (ns.aftr_isp || ns.peer_isp || ns.user_isp) { | |
809 | if (ns.conn == C_SYNC_SOURCE) | |
810 | ns.conn = C_PAUSED_SYNC_S; | |
811 | if (ns.conn == C_SYNC_TARGET) | |
812 | ns.conn = C_PAUSED_SYNC_T; | |
813 | } else { | |
814 | if (ns.conn == C_PAUSED_SYNC_S) | |
815 | ns.conn = C_SYNC_SOURCE; | |
816 | if (ns.conn == C_PAUSED_SYNC_T) | |
817 | ns.conn = C_SYNC_TARGET; | |
818 | } | |
819 | ||
820 | return ns; | |
821 | } | |
822 | ||
823 | void drbd_resume_al(struct drbd_conf *mdev) | |
824 | { | |
825 | if (test_and_clear_bit(AL_SUSPENDED, &mdev->flags)) | |
826 | dev_info(DEV, "Resumed AL updates\n"); | |
827 | } | |
828 | ||
829 | /* helper for __drbd_set_state */ | |
830 | static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs) | |
831 | { | |
832 | if (mdev->tconn->agreed_pro_version < 90) | |
833 | mdev->ov_start_sector = 0; | |
834 | mdev->rs_total = drbd_bm_bits(mdev); | |
835 | mdev->ov_position = 0; | |
836 | if (cs == C_VERIFY_T) { | |
837 | /* starting online verify from an arbitrary position | |
838 | * does not fit well into the existing protocol. | |
839 | * on C_VERIFY_T, we initialize ov_left and friends | |
840 | * implicitly in receive_DataRequest once the | |
841 | * first P_OV_REQUEST is received */ | |
842 | mdev->ov_start_sector = ~(sector_t)0; | |
843 | } else { | |
844 | unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector); | |
845 | if (bit >= mdev->rs_total) { | |
846 | mdev->ov_start_sector = | |
847 | BM_BIT_TO_SECT(mdev->rs_total - 1); | |
848 | mdev->rs_total = 1; | |
849 | } else | |
850 | mdev->rs_total -= bit; | |
851 | mdev->ov_position = mdev->ov_start_sector; | |
852 | } | |
853 | mdev->ov_left = mdev->rs_total; | |
854 | } | |
855 | ||
856 | /** | |
857 | * __drbd_set_state() - Set a new DRBD state | |
858 | * @mdev: DRBD device. | |
859 | * @ns: new state. | |
860 | * @flags: Flags | |
861 | * @done: Optional completion, that will get completed after the after_state_ch() finished | |
862 | * | |
863 | * Caller needs to hold req_lock, and global_state_lock. Do not call directly. | |
864 | */ | |
865 | enum drbd_state_rv | |
866 | __drbd_set_state(struct drbd_conf *mdev, union drbd_state ns, | |
867 | enum chg_state_flags flags, struct completion *done) | |
868 | { | |
869 | union drbd_state os; | |
870 | enum drbd_state_rv rv = SS_SUCCESS; | |
871 | const char *warn_sync_abort = NULL; | |
872 | struct after_state_chg_work *ascw; | |
873 | ||
78bae59b | 874 | os = drbd_read_state(mdev); |
b8907339 | 875 | |
4308a0a3 | 876 | ns = sanitize_state(mdev, ns, &warn_sync_abort); |
b8907339 PR |
877 | if (ns.i == os.i) |
878 | return SS_NOTHING_TO_DO; | |
879 | ||
3509502d PR |
880 | rv = is_valid_transition(os, ns); |
881 | if (rv < SS_SUCCESS) | |
882 | return rv; | |
883 | ||
b8907339 PR |
884 | if (!(flags & CS_HARD)) { |
885 | /* pre-state-change checks ; only look at ns */ | |
886 | /* See drbd_state_sw_errors in drbd_strings.c */ | |
887 | ||
888 | rv = is_valid_state(mdev, ns); | |
889 | if (rv < SS_SUCCESS) { | |
890 | /* If the old state was illegal as well, then let | |
891 | this happen...*/ | |
892 | ||
893 | if (is_valid_state(mdev, os) == rv) | |
a75f34ad | 894 | rv = is_valid_soft_transition(os, ns); |
b8907339 | 895 | } else |
a75f34ad | 896 | rv = is_valid_soft_transition(os, ns); |
b8907339 PR |
897 | } |
898 | ||
899 | if (rv < SS_SUCCESS) { | |
900 | if (flags & CS_VERBOSE) | |
901 | print_st_err(mdev, os, ns, rv); | |
902 | return rv; | |
903 | } | |
904 | ||
905 | if (warn_sync_abort) | |
906 | dev_warn(DEV, "%s aborted.\n", warn_sync_abort); | |
907 | ||
435693e8 | 908 | drbd_pr_state_change(mdev, os, ns, flags); |
b8907339 | 909 | |
706cb24c PR |
910 | /* Display changes to the susp* flags that where caused by the call to |
911 | sanitize_state(). Only display it here if we where not called from | |
912 | _conn_request_state() */ | |
913 | if (!(flags & CS_DC_SUSP)) | |
914 | conn_pr_state_change(mdev->tconn, os, ns, (flags & ~CS_DC_MASK) | CS_DC_SUSP); | |
915 | ||
b8907339 PR |
916 | /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference |
917 | * on the ldev here, to be sure the transition -> D_DISKLESS resp. | |
918 | * drbd_ldev_destroy() won't happen before our corresponding | |
919 | * after_state_ch works run, where we put_ldev again. */ | |
920 | if ((os.disk != D_FAILED && ns.disk == D_FAILED) || | |
921 | (os.disk != D_DISKLESS && ns.disk == D_DISKLESS)) | |
922 | atomic_inc(&mdev->local_cnt); | |
923 | ||
da9fbc27 | 924 | mdev->state.i = ns.i; |
8e0af25f PR |
925 | mdev->tconn->susp = ns.susp; |
926 | mdev->tconn->susp_nod = ns.susp_nod; | |
927 | mdev->tconn->susp_fen = ns.susp_fen; | |
b8907339 PR |
928 | |
929 | if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING) | |
930 | drbd_print_uuids(mdev, "attached to UUIDs"); | |
931 | ||
932 | wake_up(&mdev->misc_wait); | |
933 | wake_up(&mdev->state_wait); | |
2a67d8b9 | 934 | wake_up(&mdev->tconn->ping_wait); |
b8907339 PR |
935 | |
936 | /* aborted verify run. log the last position */ | |
937 | if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) && | |
938 | ns.conn < C_CONNECTED) { | |
939 | mdev->ov_start_sector = | |
940 | BM_BIT_TO_SECT(drbd_bm_bits(mdev) - mdev->ov_left); | |
941 | dev_info(DEV, "Online Verify reached sector %llu\n", | |
942 | (unsigned long long)mdev->ov_start_sector); | |
943 | } | |
944 | ||
945 | if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) && | |
946 | (ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)) { | |
947 | dev_info(DEV, "Syncer continues.\n"); | |
948 | mdev->rs_paused += (long)jiffies | |
949 | -(long)mdev->rs_mark_time[mdev->rs_last_mark]; | |
950 | if (ns.conn == C_SYNC_TARGET) | |
951 | mod_timer(&mdev->resync_timer, jiffies); | |
952 | } | |
953 | ||
954 | if ((os.conn == C_SYNC_TARGET || os.conn == C_SYNC_SOURCE) && | |
955 | (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) { | |
956 | dev_info(DEV, "Resync suspended\n"); | |
957 | mdev->rs_mark_time[mdev->rs_last_mark] = jiffies; | |
958 | } | |
959 | ||
960 | if (os.conn == C_CONNECTED && | |
961 | (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) { | |
962 | unsigned long now = jiffies; | |
963 | int i; | |
964 | ||
965 | set_ov_position(mdev, ns.conn); | |
966 | mdev->rs_start = now; | |
967 | mdev->rs_last_events = 0; | |
968 | mdev->rs_last_sect_ev = 0; | |
969 | mdev->ov_last_oos_size = 0; | |
970 | mdev->ov_last_oos_start = 0; | |
971 | ||
972 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { | |
973 | mdev->rs_mark_left[i] = mdev->ov_left; | |
974 | mdev->rs_mark_time[i] = now; | |
975 | } | |
976 | ||
977 | drbd_rs_controller_reset(mdev); | |
978 | ||
979 | if (ns.conn == C_VERIFY_S) { | |
980 | dev_info(DEV, "Starting Online Verify from sector %llu\n", | |
981 | (unsigned long long)mdev->ov_position); | |
982 | mod_timer(&mdev->resync_timer, jiffies); | |
983 | } | |
984 | } | |
985 | ||
986 | if (get_ldev(mdev)) { | |
987 | u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND| | |
988 | MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE| | |
989 | MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY); | |
990 | ||
991 | if (test_bit(CRASHED_PRIMARY, &mdev->flags)) | |
992 | mdf |= MDF_CRASHED_PRIMARY; | |
993 | if (mdev->state.role == R_PRIMARY || | |
994 | (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY)) | |
995 | mdf |= MDF_PRIMARY_IND; | |
996 | if (mdev->state.conn > C_WF_REPORT_PARAMS) | |
997 | mdf |= MDF_CONNECTED_IND; | |
998 | if (mdev->state.disk > D_INCONSISTENT) | |
999 | mdf |= MDF_CONSISTENT; | |
1000 | if (mdev->state.disk > D_OUTDATED) | |
1001 | mdf |= MDF_WAS_UP_TO_DATE; | |
1002 | if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT) | |
1003 | mdf |= MDF_PEER_OUT_DATED; | |
1004 | if (mdf != mdev->ldev->md.flags) { | |
1005 | mdev->ldev->md.flags = mdf; | |
1006 | drbd_md_mark_dirty(mdev); | |
1007 | } | |
1008 | if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT) | |
1009 | drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]); | |
1010 | put_ldev(mdev); | |
1011 | } | |
1012 | ||
1013 | /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */ | |
1014 | if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT && | |
1015 | os.peer == R_SECONDARY && ns.peer == R_PRIMARY) | |
1016 | set_bit(CONSIDER_RESYNC, &mdev->flags); | |
1017 | ||
1018 | /* Receiver should clean up itself */ | |
1019 | if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING) | |
1020 | drbd_thread_stop_nowait(&mdev->tconn->receiver); | |
1021 | ||
1022 | /* Now the receiver finished cleaning up itself, it should die */ | |
1023 | if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE) | |
1024 | drbd_thread_stop_nowait(&mdev->tconn->receiver); | |
1025 | ||
1026 | /* Upon network failure, we need to restart the receiver. */ | |
1027 | if (os.conn > C_TEAR_DOWN && | |
1028 | ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT) | |
1029 | drbd_thread_restart_nowait(&mdev->tconn->receiver); | |
1030 | ||
1031 | /* Resume AL writing if we get a connection */ | |
1032 | if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) | |
1033 | drbd_resume_al(mdev); | |
1034 | ||
1035 | ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC); | |
1036 | if (ascw) { | |
1037 | ascw->os = os; | |
1038 | ascw->ns = ns; | |
1039 | ascw->flags = flags; | |
1040 | ascw->w.cb = w_after_state_ch; | |
a21e9298 | 1041 | ascw->w.mdev = mdev; |
b8907339 PR |
1042 | ascw->done = done; |
1043 | drbd_queue_work(&mdev->tconn->data.work, &ascw->w); | |
1044 | } else { | |
bbeb641c | 1045 | dev_err(DEV, "Could not kmalloc an ascw\n"); |
b8907339 PR |
1046 | } |
1047 | ||
1048 | return rv; | |
1049 | } | |
1050 | ||
99920dc5 | 1051 | static int w_after_state_ch(struct drbd_work *w, int unused) |
b8907339 PR |
1052 | { |
1053 | struct after_state_chg_work *ascw = | |
1054 | container_of(w, struct after_state_chg_work, w); | |
00d56944 | 1055 | struct drbd_conf *mdev = w->mdev; |
b8907339 PR |
1056 | |
1057 | after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags); | |
1058 | if (ascw->flags & CS_WAIT_COMPLETE) { | |
1059 | D_ASSERT(ascw->done != NULL); | |
1060 | complete(ascw->done); | |
1061 | } | |
1062 | kfree(ascw); | |
1063 | ||
99920dc5 | 1064 | return 0; |
b8907339 PR |
1065 | } |
1066 | ||
1067 | static void abw_start_sync(struct drbd_conf *mdev, int rv) | |
1068 | { | |
1069 | if (rv) { | |
1070 | dev_err(DEV, "Writing the bitmap failed not starting resync.\n"); | |
1071 | _drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE); | |
1072 | return; | |
1073 | } | |
1074 | ||
1075 | switch (mdev->state.conn) { | |
1076 | case C_STARTING_SYNC_T: | |
1077 | _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); | |
1078 | break; | |
1079 | case C_STARTING_SYNC_S: | |
1080 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
1081 | break; | |
1082 | } | |
1083 | } | |
1084 | ||
1085 | int drbd_bitmap_io_from_worker(struct drbd_conf *mdev, | |
1086 | int (*io_fn)(struct drbd_conf *), | |
1087 | char *why, enum bm_flag flags) | |
1088 | { | |
1089 | int rv; | |
1090 | ||
1091 | D_ASSERT(current == mdev->tconn->worker.task); | |
1092 | ||
1093 | /* open coded non-blocking drbd_suspend_io(mdev); */ | |
1094 | set_bit(SUSPEND_IO, &mdev->flags); | |
1095 | ||
1096 | drbd_bm_lock(mdev, why, flags); | |
1097 | rv = io_fn(mdev); | |
1098 | drbd_bm_unlock(mdev); | |
1099 | ||
1100 | drbd_resume_io(mdev); | |
1101 | ||
1102 | return rv; | |
1103 | } | |
1104 | ||
1105 | /** | |
1106 | * after_state_ch() - Perform after state change actions that may sleep | |
1107 | * @mdev: DRBD device. | |
1108 | * @os: old state. | |
1109 | * @ns: new state. | |
1110 | * @flags: Flags | |
1111 | */ | |
1112 | static void after_state_ch(struct drbd_conf *mdev, union drbd_state os, | |
1113 | union drbd_state ns, enum chg_state_flags flags) | |
1114 | { | |
1115 | enum drbd_fencing_p fp; | |
3b98c0c2 LE |
1116 | struct sib_info sib; |
1117 | ||
1118 | sib.sib_reason = SIB_STATE_CHANGE; | |
1119 | sib.os = os; | |
1120 | sib.ns = ns; | |
b8907339 PR |
1121 | |
1122 | if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) { | |
1123 | clear_bit(CRASHED_PRIMARY, &mdev->flags); | |
1124 | if (mdev->p_uuid) | |
1125 | mdev->p_uuid[UI_FLAGS] &= ~((u64)2); | |
1126 | } | |
1127 | ||
1128 | fp = FP_DONT_CARE; | |
1129 | if (get_ldev(mdev)) { | |
1130 | fp = mdev->ldev->dc.fencing; | |
1131 | put_ldev(mdev); | |
1132 | } | |
1133 | ||
1134 | /* Inform userspace about the change... */ | |
3b98c0c2 | 1135 | drbd_bcast_event(mdev, &sib); |
b8907339 PR |
1136 | |
1137 | if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) && | |
1138 | (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)) | |
1139 | drbd_khelper(mdev, "pri-on-incon-degr"); | |
1140 | ||
1141 | /* Here we have the actions that are performed after a | |
1142 | state change. This function might sleep */ | |
1143 | ||
b8907339 | 1144 | if (ns.susp_nod) { |
a6d00c8e PR |
1145 | enum drbd_req_event what = NOTHING; |
1146 | ||
1147 | if (os.conn < C_CONNECTED && conn_lowest_conn(mdev->tconn) >= C_CONNECTED) | |
b8907339 PR |
1148 | what = RESEND; |
1149 | ||
a6d00c8e | 1150 | if (os.disk == D_ATTACHING && conn_lowest_disk(mdev->tconn) > D_ATTACHING) |
b8907339 PR |
1151 | what = RESTART_FROZEN_DISK_IO; |
1152 | ||
a6d00c8e | 1153 | if (what != NOTHING) { |
b8907339 | 1154 | spin_lock_irq(&mdev->tconn->req_lock); |
a6d00c8e PR |
1155 | _tl_restart(mdev->tconn, what); |
1156 | _drbd_set_state(_NS(mdev, susp_nod, 0), CS_VERBOSE, NULL); | |
b8907339 PR |
1157 | spin_unlock_irq(&mdev->tconn->req_lock); |
1158 | } | |
b8907339 PR |
1159 | } |
1160 | ||
1161 | /* Became sync source. With protocol >= 96, we still need to send out | |
1162 | * the sync uuid now. Need to do that before any drbd_send_state, or | |
1163 | * the other side may go "paused sync" before receiving the sync uuids, | |
1164 | * which is unexpected. */ | |
1165 | if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) && | |
1166 | (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) && | |
1167 | mdev->tconn->agreed_pro_version >= 96 && get_ldev(mdev)) { | |
1168 | drbd_gen_and_send_sync_uuid(mdev); | |
1169 | put_ldev(mdev); | |
1170 | } | |
1171 | ||
1172 | /* Do not change the order of the if above and the two below... */ | |
1173 | if (os.pdsk == D_DISKLESS && ns.pdsk > D_DISKLESS) { /* attach on the peer */ | |
1174 | drbd_send_uuids(mdev); | |
1175 | drbd_send_state(mdev); | |
1176 | } | |
1177 | /* No point in queuing send_bitmap if we don't have a connection | |
1178 | * anymore, so check also the _current_ state, not only the new state | |
1179 | * at the time this work was queued. */ | |
1180 | if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S && | |
1181 | mdev->state.conn == C_WF_BITMAP_S) | |
1182 | drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL, | |
1183 | "send_bitmap (WFBitMapS)", | |
1184 | BM_LOCKED_TEST_ALLOWED); | |
1185 | ||
1186 | /* Lost contact to peer's copy of the data */ | |
1187 | if ((os.pdsk >= D_INCONSISTENT && | |
1188 | os.pdsk != D_UNKNOWN && | |
1189 | os.pdsk != D_OUTDATED) | |
1190 | && (ns.pdsk < D_INCONSISTENT || | |
1191 | ns.pdsk == D_UNKNOWN || | |
1192 | ns.pdsk == D_OUTDATED)) { | |
1193 | if (get_ldev(mdev)) { | |
1194 | if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) && | |
1195 | mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) { | |
2aebfabb | 1196 | if (drbd_suspended(mdev)) { |
b8907339 PR |
1197 | set_bit(NEW_CUR_UUID, &mdev->flags); |
1198 | } else { | |
1199 | drbd_uuid_new_current(mdev); | |
1200 | drbd_send_uuids(mdev); | |
1201 | } | |
1202 | } | |
1203 | put_ldev(mdev); | |
1204 | } | |
1205 | } | |
1206 | ||
1207 | if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) { | |
1208 | if (ns.peer == R_PRIMARY && mdev->ldev->md.uuid[UI_BITMAP] == 0) { | |
1209 | drbd_uuid_new_current(mdev); | |
1210 | drbd_send_uuids(mdev); | |
1211 | } | |
1212 | ||
1213 | /* D_DISKLESS Peer becomes secondary */ | |
1214 | if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY) | |
1215 | /* We may still be Primary ourselves. | |
1216 | * No harm done if the bitmap still changes, | |
1217 | * redirtied pages will follow later. */ | |
1218 | drbd_bitmap_io_from_worker(mdev, &drbd_bm_write, | |
1219 | "demote diskless peer", BM_LOCKED_SET_ALLOWED); | |
1220 | put_ldev(mdev); | |
1221 | } | |
1222 | ||
1223 | /* Write out all changed bits on demote. | |
1224 | * Though, no need to da that just yet | |
1225 | * if there is a resync going on still */ | |
1226 | if (os.role == R_PRIMARY && ns.role == R_SECONDARY && | |
1227 | mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) { | |
1228 | /* No changes to the bitmap expected this time, so assert that, | |
1229 | * even though no harm was done if it did change. */ | |
1230 | drbd_bitmap_io_from_worker(mdev, &drbd_bm_write, | |
1231 | "demote", BM_LOCKED_TEST_ALLOWED); | |
1232 | put_ldev(mdev); | |
1233 | } | |
1234 | ||
1235 | /* Last part of the attaching process ... */ | |
1236 | if (ns.conn >= C_CONNECTED && | |
1237 | os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) { | |
1238 | drbd_send_sizes(mdev, 0, 0); /* to start sync... */ | |
1239 | drbd_send_uuids(mdev); | |
1240 | drbd_send_state(mdev); | |
1241 | } | |
1242 | ||
1243 | /* We want to pause/continue resync, tell peer. */ | |
1244 | if (ns.conn >= C_CONNECTED && | |
1245 | ((os.aftr_isp != ns.aftr_isp) || | |
1246 | (os.user_isp != ns.user_isp))) | |
1247 | drbd_send_state(mdev); | |
1248 | ||
1249 | /* In case one of the isp bits got set, suspend other devices. */ | |
1250 | if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) && | |
1251 | (ns.aftr_isp || ns.peer_isp || ns.user_isp)) | |
1252 | suspend_other_sg(mdev); | |
1253 | ||
1254 | /* Make sure the peer gets informed about eventual state | |
1255 | changes (ISP bits) while we were in WFReportParams. */ | |
1256 | if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED) | |
1257 | drbd_send_state(mdev); | |
1258 | ||
1259 | if (os.conn != C_AHEAD && ns.conn == C_AHEAD) | |
1260 | drbd_send_state(mdev); | |
1261 | ||
1262 | /* We are in the progress to start a full sync... */ | |
1263 | if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) || | |
1264 | (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S)) | |
1265 | /* no other bitmap changes expected during this phase */ | |
1266 | drbd_queue_bitmap_io(mdev, | |
1267 | &drbd_bmio_set_n_write, &abw_start_sync, | |
1268 | "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED); | |
1269 | ||
1270 | /* We are invalidating our self... */ | |
1271 | if (os.conn < C_CONNECTED && ns.conn < C_CONNECTED && | |
1272 | os.disk > D_INCONSISTENT && ns.disk == D_INCONSISTENT) | |
1273 | /* other bitmap operation expected during this phase */ | |
1274 | drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, NULL, | |
1275 | "set_n_write from invalidate", BM_LOCKED_MASK); | |
1276 | ||
1277 | /* first half of local IO error, failure to attach, | |
1278 | * or administrative detach */ | |
1279 | if (os.disk != D_FAILED && ns.disk == D_FAILED) { | |
1280 | enum drbd_io_error_p eh; | |
1281 | int was_io_error; | |
1282 | /* corresponding get_ldev was in __drbd_set_state, to serialize | |
1283 | * our cleanup here with the transition to D_DISKLESS, | |
1284 | * so it is safe to dreference ldev here. */ | |
1285 | eh = mdev->ldev->dc.on_io_error; | |
1286 | was_io_error = test_and_clear_bit(WAS_IO_ERROR, &mdev->flags); | |
1287 | ||
1288 | /* current state still has to be D_FAILED, | |
1289 | * there is only one way out: to D_DISKLESS, | |
1290 | * and that may only happen after our put_ldev below. */ | |
1291 | if (mdev->state.disk != D_FAILED) | |
1292 | dev_err(DEV, | |
1293 | "ASSERT FAILED: disk is %s during detach\n", | |
1294 | drbd_disk_str(mdev->state.disk)); | |
1295 | ||
927036f9 | 1296 | if (!drbd_send_state(mdev)) |
b8907339 PR |
1297 | dev_warn(DEV, "Notified peer that I am detaching my disk\n"); |
1298 | else | |
1299 | dev_err(DEV, "Sending state for detaching disk failed\n"); | |
1300 | ||
1301 | drbd_rs_cancel_all(mdev); | |
1302 | ||
1303 | /* In case we want to get something to stable storage still, | |
1304 | * this may be the last chance. | |
1305 | * Following put_ldev may transition to D_DISKLESS. */ | |
1306 | drbd_md_sync(mdev); | |
1307 | put_ldev(mdev); | |
1308 | ||
1309 | if (was_io_error && eh == EP_CALL_HELPER) | |
1310 | drbd_khelper(mdev, "local-io-error"); | |
1311 | } | |
1312 | ||
1313 | /* second half of local IO error, failure to attach, | |
1314 | * or administrative detach, | |
1315 | * after local_cnt references have reached zero again */ | |
1316 | if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) { | |
1317 | /* We must still be diskless, | |
1318 | * re-attach has to be serialized with this! */ | |
1319 | if (mdev->state.disk != D_DISKLESS) | |
1320 | dev_err(DEV, | |
1321 | "ASSERT FAILED: disk is %s while going diskless\n", | |
1322 | drbd_disk_str(mdev->state.disk)); | |
1323 | ||
1324 | mdev->rs_total = 0; | |
1325 | mdev->rs_failed = 0; | |
1326 | atomic_set(&mdev->rs_pending_cnt, 0); | |
1327 | ||
927036f9 | 1328 | if (!drbd_send_state(mdev)) |
b8907339 PR |
1329 | dev_warn(DEV, "Notified peer that I'm now diskless.\n"); |
1330 | /* corresponding get_ldev in __drbd_set_state | |
1331 | * this may finally trigger drbd_ldev_destroy. */ | |
1332 | put_ldev(mdev); | |
1333 | } | |
1334 | ||
1335 | /* Notify peer that I had a local IO error, and did not detached.. */ | |
1336 | if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT) | |
1337 | drbd_send_state(mdev); | |
1338 | ||
1339 | /* Disks got bigger while they were detached */ | |
1340 | if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING && | |
1341 | test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) { | |
1342 | if (ns.conn == C_CONNECTED) | |
1343 | resync_after_online_grow(mdev); | |
1344 | } | |
1345 | ||
1346 | /* A resync finished or aborted, wake paused devices... */ | |
1347 | if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) || | |
1348 | (os.peer_isp && !ns.peer_isp) || | |
1349 | (os.user_isp && !ns.user_isp)) | |
1350 | resume_next_sg(mdev); | |
1351 | ||
1352 | /* sync target done with resync. Explicitly notify peer, even though | |
1353 | * it should (at least for non-empty resyncs) already know itself. */ | |
1354 | if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED) | |
1355 | drbd_send_state(mdev); | |
1356 | ||
1357 | /* This triggers bitmap writeout of potentially still unwritten pages | |
1358 | * if the resync finished cleanly, or aborted because of peer disk | |
1359 | * failure, or because of connection loss. | |
1360 | * For resync aborted because of local disk failure, we cannot do | |
1361 | * any bitmap writeout anymore. | |
1362 | * No harm done if some bits change during this phase. | |
1363 | */ | |
1364 | if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) { | |
1365 | drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL, | |
1366 | "write from resync_finished", BM_LOCKED_SET_ALLOWED); | |
1367 | put_ldev(mdev); | |
1368 | } | |
1369 | ||
1370 | if (ns.disk == D_DISKLESS && | |
1371 | ns.conn == C_STANDALONE && | |
1372 | ns.role == R_SECONDARY) { | |
1373 | if (os.aftr_isp != ns.aftr_isp) | |
1374 | resume_next_sg(mdev); | |
1375 | } | |
1376 | ||
0e29d163 | 1377 | after_all_state_ch(mdev->tconn); |
bbeb641c | 1378 | |
b8907339 PR |
1379 | drbd_md_sync(mdev); |
1380 | } | |
1381 | ||
bbeb641c PR |
1382 | struct after_conn_state_chg_work { |
1383 | struct drbd_work w; | |
1384 | enum drbd_conns oc; | |
8c7e16c3 | 1385 | union drbd_state ns_min; |
5f082f98 | 1386 | union drbd_state ns_max; /* new, max state, over all mdevs */ |
bbeb641c PR |
1387 | enum chg_state_flags flags; |
1388 | }; | |
1389 | ||
0e29d163 | 1390 | static void after_all_state_ch(struct drbd_tconn *tconn) |
b8907339 | 1391 | { |
a18e9d1e | 1392 | if (conn_all_vols_unconf(tconn)) |
bbeb641c | 1393 | drbd_thread_stop_nowait(&tconn->worker); |
bbeb641c PR |
1394 | } |
1395 | ||
99920dc5 | 1396 | static int w_after_conn_state_ch(struct drbd_work *w, int unused) |
bbeb641c PR |
1397 | { |
1398 | struct after_conn_state_chg_work *acscw = | |
1399 | container_of(w, struct after_conn_state_chg_work, w); | |
1400 | struct drbd_tconn *tconn = w->tconn; | |
1401 | enum drbd_conns oc = acscw->oc; | |
5f082f98 | 1402 | union drbd_state ns_max = acscw->ns_max; |
a6d00c8e PR |
1403 | union drbd_state ns_min = acscw->ns_min; |
1404 | struct drbd_conf *mdev; | |
1405 | int vnr; | |
bbeb641c PR |
1406 | |
1407 | kfree(acscw); | |
1408 | ||
b8907339 | 1409 | /* Upon network configuration, we need to start the receiver */ |
5f082f98 | 1410 | if (oc == C_STANDALONE && ns_max.conn == C_UNCONNECTED) |
b8907339 PR |
1411 | drbd_thread_start(&tconn->receiver); |
1412 | ||
a6d00c8e PR |
1413 | if (ns_max.susp_fen) { |
1414 | /* case1: The outdate peer handler is successful: */ | |
1415 | if (ns_max.pdsk <= D_OUTDATED) { | |
1416 | tl_clear(tconn); | |
695d08fa | 1417 | rcu_read_lock(); |
a6d00c8e PR |
1418 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
1419 | if (test_bit(NEW_CUR_UUID, &mdev->flags)) { | |
1420 | drbd_uuid_new_current(mdev); | |
1421 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
1422 | } | |
1423 | } | |
695d08fa | 1424 | rcu_read_unlock(); |
a6d00c8e PR |
1425 | conn_request_state(tconn, |
1426 | (union drbd_state) { { .susp_fen = 1 } }, | |
1427 | (union drbd_state) { { .susp_fen = 0 } }, | |
1428 | CS_VERBOSE); | |
1429 | } | |
1430 | /* case2: The connection was established again: */ | |
1431 | if (ns_min.conn >= C_CONNECTED) { | |
695d08fa | 1432 | rcu_read_lock(); |
a6d00c8e PR |
1433 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
1434 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
695d08fa | 1435 | rcu_read_unlock(); |
a6d00c8e PR |
1436 | spin_lock_irq(&tconn->req_lock); |
1437 | _tl_restart(tconn, RESEND); | |
1438 | _conn_request_state(tconn, | |
1439 | (union drbd_state) { { .susp_fen = 1 } }, | |
1440 | (union drbd_state) { { .susp_fen = 0 } }, | |
1441 | CS_VERBOSE); | |
1442 | spin_unlock_irq(&tconn->req_lock); | |
1443 | } | |
1444 | } | |
1445 | ||
1446 | ||
bbeb641c | 1447 | //conn_err(tconn, STATE_FMT, STATE_ARGS("nms", nms)); |
0e29d163 | 1448 | after_all_state_ch(tconn); |
9dc9fbb3 | 1449 | kref_put(&tconn->kref, &conn_destroy); |
bbeb641c | 1450 | |
99920dc5 | 1451 | return 0; |
bbeb641c PR |
1452 | } |
1453 | ||
435693e8 | 1454 | void conn_old_common_state(struct drbd_tconn *tconn, union drbd_state *pcs, enum chg_state_flags *pf) |
88ef594e | 1455 | { |
435693e8 | 1456 | enum chg_state_flags flags = ~0; |
da9fbc27 | 1457 | union drbd_dev_state os, cs = {}; /* old_state, common_state */ |
88ef594e | 1458 | struct drbd_conf *mdev; |
435693e8 | 1459 | int vnr, first_vol = 1; |
88ef594e | 1460 | |
695d08fa | 1461 | rcu_read_lock(); |
88ef594e PR |
1462 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
1463 | os = mdev->state; | |
1464 | ||
435693e8 PR |
1465 | if (first_vol) { |
1466 | cs = os; | |
1467 | first_vol = 0; | |
1468 | continue; | |
1469 | } | |
1470 | ||
1471 | if (cs.role != os.role) | |
1472 | flags &= ~CS_DC_ROLE; | |
1473 | ||
1474 | if (cs.peer != os.peer) | |
1475 | flags &= ~CS_DC_PEER; | |
1476 | ||
1477 | if (cs.conn != os.conn) | |
1478 | flags &= ~CS_DC_CONN; | |
88ef594e | 1479 | |
435693e8 PR |
1480 | if (cs.disk != os.disk) |
1481 | flags &= ~CS_DC_DISK; | |
88ef594e | 1482 | |
435693e8 PR |
1483 | if (cs.pdsk != os.pdsk) |
1484 | flags &= ~CS_DC_PDSK; | |
1485 | } | |
695d08fa | 1486 | rcu_read_unlock(); |
435693e8 PR |
1487 | |
1488 | *pf |= CS_DC_MASK; | |
1489 | *pf &= flags; | |
da9fbc27 | 1490 | (*pcs).i = cs.i; |
88ef594e | 1491 | } |
bbeb641c | 1492 | |
bd0c824a PR |
1493 | static enum drbd_state_rv |
1494 | conn_is_valid_transition(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, | |
88ef594e | 1495 | enum chg_state_flags flags) |
bbeb641c | 1496 | { |
bd0c824a | 1497 | enum drbd_state_rv rv = SS_SUCCESS; |
bbeb641c | 1498 | union drbd_state ns, os; |
bd0c824a PR |
1499 | struct drbd_conf *mdev; |
1500 | int vnr; | |
bbeb641c | 1501 | |
695d08fa | 1502 | rcu_read_lock(); |
bd0c824a | 1503 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
78bae59b | 1504 | os = drbd_read_state(mdev); |
bd0c824a PR |
1505 | ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL); |
1506 | ||
778bcf2e PR |
1507 | if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED) |
1508 | ns.disk = os.disk; | |
1509 | ||
bd0c824a PR |
1510 | if (ns.i == os.i) |
1511 | continue; | |
bbeb641c | 1512 | |
bd0c824a PR |
1513 | rv = is_valid_transition(os, ns); |
1514 | if (rv < SS_SUCCESS) | |
1515 | break; | |
1516 | ||
1517 | if (!(flags & CS_HARD)) { | |
1518 | rv = is_valid_state(mdev, ns); | |
1519 | if (rv < SS_SUCCESS) { | |
1520 | if (is_valid_state(mdev, os) == rv) | |
1521 | rv = is_valid_soft_transition(os, ns); | |
1522 | } else | |
1523 | rv = is_valid_soft_transition(os, ns); | |
1524 | } | |
1525 | if (rv < SS_SUCCESS) | |
1526 | break; | |
bbeb641c | 1527 | } |
695d08fa | 1528 | rcu_read_unlock(); |
bbeb641c | 1529 | |
bd0c824a PR |
1530 | if (rv < SS_SUCCESS && flags & CS_VERBOSE) |
1531 | print_st_err(mdev, os, ns, rv); | |
1532 | ||
1533 | return rv; | |
bbeb641c PR |
1534 | } |
1535 | ||
8c7e16c3 | 1536 | void |
bd0c824a | 1537 | conn_set_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, |
8c7e16c3 | 1538 | union drbd_state *pns_min, union drbd_state *pns_max, enum chg_state_flags flags) |
bbeb641c | 1539 | { |
8c7e16c3 PR |
1540 | union drbd_state ns, os, ns_max = { }; |
1541 | union drbd_state ns_min = { | |
1542 | { .role = R_MASK, | |
1543 | .peer = R_MASK, | |
1544 | .disk = D_MASK, | |
1545 | .pdsk = D_MASK | |
1546 | } }; | |
bd0c824a | 1547 | struct drbd_conf *mdev; |
bbeb641c | 1548 | enum drbd_state_rv rv; |
bd0c824a | 1549 | int vnr; |
bbeb641c | 1550 | |
bd0c824a PR |
1551 | if (mask.conn == C_MASK) |
1552 | tconn->cstate = val.conn; | |
1553 | ||
695d08fa | 1554 | rcu_read_lock(); |
bd0c824a | 1555 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
78bae59b | 1556 | os = drbd_read_state(mdev); |
bd0c824a PR |
1557 | ns = apply_mask_val(os, mask, val); |
1558 | ns = sanitize_state(mdev, ns, NULL); | |
bbeb641c | 1559 | |
778bcf2e PR |
1560 | if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED) |
1561 | ns.disk = os.disk; | |
1562 | ||
bd0c824a PR |
1563 | rv = __drbd_set_state(mdev, ns, flags, NULL); |
1564 | if (rv < SS_SUCCESS) | |
1565 | BUG(); | |
bbeb641c | 1566 | |
8c7e16c3 PR |
1567 | ns.i = mdev->state.i; |
1568 | ns_max.role = max_role(ns.role, ns_max.role); | |
1569 | ns_max.peer = max_role(ns.peer, ns_max.peer); | |
1570 | ns_max.conn = max_t(enum drbd_conns, ns.conn, ns_max.conn); | |
1571 | ns_max.disk = max_t(enum drbd_disk_state, ns.disk, ns_max.disk); | |
1572 | ns_max.pdsk = max_t(enum drbd_disk_state, ns.pdsk, ns_max.pdsk); | |
1573 | ||
1574 | ns_min.role = min_role(ns.role, ns_min.role); | |
1575 | ns_min.peer = min_role(ns.peer, ns_min.peer); | |
1576 | ns_min.conn = min_t(enum drbd_conns, ns.conn, ns_min.conn); | |
1577 | ns_min.disk = min_t(enum drbd_disk_state, ns.disk, ns_min.disk); | |
1578 | ns_min.pdsk = min_t(enum drbd_disk_state, ns.pdsk, ns_min.pdsk); | |
bd0c824a | 1579 | } |
695d08fa | 1580 | rcu_read_unlock(); |
bbeb641c | 1581 | |
8c7e16c3 PR |
1582 | ns_min.susp = ns_max.susp = tconn->susp; |
1583 | ns_min.susp_nod = ns_max.susp_nod = tconn->susp_nod; | |
1584 | ns_min.susp_fen = ns_max.susp_fen = tconn->susp_fen; | |
1585 | ||
1586 | *pns_min = ns_min; | |
1587 | *pns_max = ns_max; | |
bbeb641c PR |
1588 | } |
1589 | ||
df24aa45 PR |
1590 | static enum drbd_state_rv |
1591 | _conn_rq_cond(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val) | |
1592 | { | |
df24aa45 PR |
1593 | enum drbd_state_rv rv; |
1594 | ||
1595 | if (test_and_clear_bit(CONN_WD_ST_CHG_OKAY, &tconn->flags)) | |
1596 | return SS_CW_SUCCESS; | |
1597 | ||
1598 | if (test_and_clear_bit(CONN_WD_ST_CHG_FAIL, &tconn->flags)) | |
1599 | return SS_CW_FAILED_BY_PEER; | |
1600 | ||
df24aa45 PR |
1601 | spin_lock_irq(&tconn->req_lock); |
1602 | rv = tconn->cstate != C_WF_REPORT_PARAMS ? SS_CW_NO_NEED : SS_UNKNOWN_ERROR; | |
1603 | ||
1604 | if (rv == SS_UNKNOWN_ERROR) | |
435693e8 | 1605 | rv = conn_is_valid_transition(tconn, mask, val, 0); |
df24aa45 | 1606 | |
bd0c824a PR |
1607 | if (rv == SS_SUCCESS) |
1608 | rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */ | |
df24aa45 PR |
1609 | |
1610 | spin_unlock_irq(&tconn->req_lock); | |
1611 | ||
1612 | return rv; | |
1613 | } | |
1614 | ||
1615 | static enum drbd_state_rv | |
1616 | conn_cl_wide(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, | |
1617 | enum chg_state_flags f) | |
1618 | { | |
1619 | enum drbd_state_rv rv; | |
1620 | ||
1621 | spin_unlock_irq(&tconn->req_lock); | |
1622 | mutex_lock(&tconn->cstate_mutex); | |
1623 | ||
caee1c3a | 1624 | if (conn_send_state_req(tconn, mask, val)) { |
df24aa45 PR |
1625 | rv = SS_CW_FAILED_BY_PEER; |
1626 | /* if (f & CS_VERBOSE) | |
1627 | print_st_err(mdev, os, ns, rv); */ | |
1628 | goto abort; | |
1629 | } | |
1630 | ||
1631 | wait_event(tconn->ping_wait, (rv = _conn_rq_cond(tconn, mask, val))); | |
1632 | ||
1633 | abort: | |
1634 | mutex_unlock(&tconn->cstate_mutex); | |
1635 | spin_lock_irq(&tconn->req_lock); | |
1636 | ||
1637 | return rv; | |
1638 | } | |
1639 | ||
bbeb641c PR |
1640 | enum drbd_state_rv |
1641 | _conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, | |
1642 | enum chg_state_flags flags) | |
1643 | { | |
1644 | enum drbd_state_rv rv = SS_SUCCESS; | |
bbeb641c PR |
1645 | struct after_conn_state_chg_work *acscw; |
1646 | enum drbd_conns oc = tconn->cstate; | |
8c7e16c3 | 1647 | union drbd_state ns_max, ns_min, os; |
bbeb641c | 1648 | |
bbeb641c PR |
1649 | rv = is_valid_conn_transition(oc, val.conn); |
1650 | if (rv < SS_SUCCESS) | |
1651 | goto abort; | |
1652 | ||
88ef594e | 1653 | rv = conn_is_valid_transition(tconn, mask, val, flags); |
bbeb641c PR |
1654 | if (rv < SS_SUCCESS) |
1655 | goto abort; | |
1656 | ||
df24aa45 PR |
1657 | if (oc == C_WF_REPORT_PARAMS && val.conn == C_DISCONNECTING && |
1658 | !(flags & (CS_LOCAL_ONLY | CS_HARD))) { | |
1659 | rv = conn_cl_wide(tconn, mask, val, flags); | |
1660 | if (rv < SS_SUCCESS) | |
1661 | goto abort; | |
1662 | } | |
1663 | ||
435693e8 | 1664 | conn_old_common_state(tconn, &os, &flags); |
706cb24c | 1665 | flags |= CS_DC_SUSP; |
8c7e16c3 | 1666 | conn_set_state(tconn, mask, val, &ns_min, &ns_max, flags); |
5f082f98 | 1667 | conn_pr_state_change(tconn, os, ns_max, flags); |
bbeb641c PR |
1668 | |
1669 | acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC); | |
1670 | if (acscw) { | |
435693e8 | 1671 | acscw->oc = os.conn; |
8c7e16c3 | 1672 | acscw->ns_min = ns_min; |
5f082f98 | 1673 | acscw->ns_max = ns_max; |
bbeb641c PR |
1674 | acscw->flags = flags; |
1675 | acscw->w.cb = w_after_conn_state_ch; | |
9dc9fbb3 | 1676 | kref_get(&tconn->kref); |
bbeb641c PR |
1677 | acscw->w.tconn = tconn; |
1678 | drbd_queue_work(&tconn->data.work, &acscw->w); | |
1679 | } else { | |
1680 | conn_err(tconn, "Could not kmalloc an acscw\n"); | |
b8907339 | 1681 | } |
bbeb641c PR |
1682 | |
1683 | abort: | |
bbeb641c PR |
1684 | return rv; |
1685 | } | |
1686 | ||
1687 | enum drbd_state_rv | |
1688 | conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, | |
1689 | enum chg_state_flags flags) | |
1690 | { | |
1691 | enum drbd_state_rv rv; | |
1692 | ||
1693 | spin_lock_irq(&tconn->req_lock); | |
1694 | rv = _conn_request_state(tconn, mask, val, flags); | |
1695 | spin_unlock_irq(&tconn->req_lock); | |
1696 | ||
1697 | return rv; | |
b8907339 | 1698 | } |