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); | |
a75f34ad PR |
43 | static enum drbd_state_rv is_valid_state(struct drbd_conf *, union drbd_state); |
44 | static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state); | |
3509502d | 45 | static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns); |
4308a0a3 PR |
46 | static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns, |
47 | const char **warn_sync_abort); | |
b8907339 | 48 | |
2aebfabb PR |
49 | static inline bool is_susp(union drbd_state s) |
50 | { | |
51 | return s.susp || s.susp_nod || s.susp_fen; | |
52 | } | |
53 | ||
d0456c72 | 54 | bool conn_all_vols_unconf(struct drbd_tconn *tconn) |
0e29d163 PR |
55 | { |
56 | struct drbd_conf *mdev; | |
695d08fa | 57 | bool rv = true; |
e90285e0 | 58 | int vnr; |
0e29d163 | 59 | |
695d08fa | 60 | rcu_read_lock(); |
e90285e0 | 61 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
d0456c72 LE |
62 | if (mdev->state.disk != D_DISKLESS || |
63 | mdev->state.conn != C_STANDALONE || | |
695d08fa PR |
64 | mdev->state.role != R_SECONDARY) { |
65 | rv = false; | |
66 | break; | |
67 | } | |
0e29d163 | 68 | } |
695d08fa PR |
69 | rcu_read_unlock(); |
70 | ||
71 | return rv; | |
0e29d163 PR |
72 | } |
73 | ||
cb703454 PR |
74 | /* Unfortunately the states where not correctly ordered, when |
75 | they where defined. therefore can not use max_t() here. */ | |
76 | static enum drbd_role max_role(enum drbd_role role1, enum drbd_role role2) | |
77 | { | |
78 | if (role1 == R_PRIMARY || role2 == R_PRIMARY) | |
79 | return R_PRIMARY; | |
80 | if (role1 == R_SECONDARY || role2 == R_SECONDARY) | |
81 | return R_SECONDARY; | |
82 | return R_UNKNOWN; | |
83 | } | |
84 | static enum drbd_role min_role(enum drbd_role role1, enum drbd_role role2) | |
85 | { | |
86 | if (role1 == R_UNKNOWN || role2 == R_UNKNOWN) | |
87 | return R_UNKNOWN; | |
88 | if (role1 == R_SECONDARY || role2 == R_SECONDARY) | |
89 | return R_SECONDARY; | |
90 | return R_PRIMARY; | |
91 | } | |
92 | ||
93 | enum drbd_role conn_highest_role(struct drbd_tconn *tconn) | |
94 | { | |
95 | enum drbd_role role = R_UNKNOWN; | |
96 | struct drbd_conf *mdev; | |
97 | int vnr; | |
98 | ||
695d08fa | 99 | rcu_read_lock(); |
cb703454 PR |
100 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
101 | role = max_role(role, mdev->state.role); | |
695d08fa | 102 | rcu_read_unlock(); |
cb703454 PR |
103 | |
104 | return role; | |
105 | } | |
106 | ||
107 | enum drbd_role conn_highest_peer(struct drbd_tconn *tconn) | |
108 | { | |
109 | enum drbd_role peer = R_UNKNOWN; | |
110 | struct drbd_conf *mdev; | |
111 | int vnr; | |
112 | ||
695d08fa | 113 | rcu_read_lock(); |
cb703454 PR |
114 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
115 | peer = max_role(peer, mdev->state.peer); | |
695d08fa | 116 | rcu_read_unlock(); |
cb703454 PR |
117 | |
118 | return peer; | |
119 | } | |
120 | ||
121 | enum drbd_disk_state conn_highest_disk(struct drbd_tconn *tconn) | |
122 | { | |
123 | enum drbd_disk_state ds = D_DISKLESS; | |
124 | struct drbd_conf *mdev; | |
125 | int vnr; | |
126 | ||
695d08fa | 127 | rcu_read_lock(); |
cb703454 PR |
128 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
129 | ds = max_t(enum drbd_disk_state, ds, mdev->state.disk); | |
695d08fa | 130 | rcu_read_unlock(); |
cb703454 PR |
131 | |
132 | return ds; | |
133 | } | |
134 | ||
4669265a PR |
135 | enum drbd_disk_state conn_lowest_disk(struct drbd_tconn *tconn) |
136 | { | |
137 | enum drbd_disk_state ds = D_MASK; | |
138 | struct drbd_conf *mdev; | |
139 | int vnr; | |
140 | ||
695d08fa | 141 | rcu_read_lock(); |
4669265a PR |
142 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
143 | ds = min_t(enum drbd_disk_state, ds, mdev->state.disk); | |
695d08fa | 144 | rcu_read_unlock(); |
4669265a PR |
145 | |
146 | return ds; | |
147 | } | |
148 | ||
cb703454 PR |
149 | enum drbd_disk_state conn_highest_pdsk(struct drbd_tconn *tconn) |
150 | { | |
151 | enum drbd_disk_state ds = D_DISKLESS; | |
152 | struct drbd_conf *mdev; | |
153 | int vnr; | |
154 | ||
695d08fa | 155 | rcu_read_lock(); |
cb703454 PR |
156 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
157 | ds = max_t(enum drbd_disk_state, ds, mdev->state.pdsk); | |
695d08fa | 158 | rcu_read_unlock(); |
cb703454 PR |
159 | |
160 | return ds; | |
161 | } | |
162 | ||
19f83c76 PR |
163 | enum drbd_conns conn_lowest_conn(struct drbd_tconn *tconn) |
164 | { | |
165 | enum drbd_conns conn = C_MASK; | |
166 | struct drbd_conf *mdev; | |
167 | int vnr; | |
168 | ||
695d08fa | 169 | rcu_read_lock(); |
19f83c76 PR |
170 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
171 | conn = min_t(enum drbd_conns, conn, mdev->state.conn); | |
695d08fa | 172 | rcu_read_unlock(); |
19f83c76 PR |
173 | |
174 | return conn; | |
175 | } | |
176 | ||
b8907339 PR |
177 | /** |
178 | * cl_wide_st_chg() - true if the state change is a cluster wide one | |
179 | * @mdev: DRBD device. | |
180 | * @os: old (current) state. | |
181 | * @ns: new (wanted) state. | |
182 | */ | |
183 | static int cl_wide_st_chg(struct drbd_conf *mdev, | |
184 | union drbd_state os, union drbd_state ns) | |
185 | { | |
186 | return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED && | |
187 | ((os.role != R_PRIMARY && ns.role == R_PRIMARY) || | |
188 | (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) || | |
189 | (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) || | |
190 | (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))) || | |
191 | (os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) || | |
192 | (os.conn == C_CONNECTED && ns.conn == C_VERIFY_S); | |
193 | } | |
194 | ||
56707f9e PR |
195 | static union drbd_state |
196 | apply_mask_val(union drbd_state os, union drbd_state mask, union drbd_state val) | |
197 | { | |
198 | union drbd_state ns; | |
199 | ns.i = (os.i & ~mask.i) | val.i; | |
200 | return ns; | |
201 | } | |
202 | ||
b8907339 PR |
203 | enum drbd_state_rv |
204 | drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f, | |
205 | union drbd_state mask, union drbd_state val) | |
206 | { | |
207 | unsigned long flags; | |
56707f9e | 208 | union drbd_state ns; |
b8907339 PR |
209 | enum drbd_state_rv rv; |
210 | ||
211 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); | |
78bae59b | 212 | ns = apply_mask_val(drbd_read_state(mdev), mask, val); |
b8907339 | 213 | rv = _drbd_set_state(mdev, ns, f, NULL); |
b8907339 PR |
214 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
215 | ||
216 | return rv; | |
217 | } | |
218 | ||
219 | /** | |
220 | * drbd_force_state() - Impose a change which happens outside our control on our state | |
221 | * @mdev: DRBD device. | |
222 | * @mask: mask of state bits to change. | |
223 | * @val: value of new state bits. | |
224 | */ | |
225 | void drbd_force_state(struct drbd_conf *mdev, | |
226 | union drbd_state mask, union drbd_state val) | |
227 | { | |
228 | drbd_change_state(mdev, CS_HARD, mask, val); | |
229 | } | |
230 | ||
231 | static enum drbd_state_rv | |
232 | _req_st_cond(struct drbd_conf *mdev, union drbd_state mask, | |
233 | union drbd_state val) | |
234 | { | |
235 | union drbd_state os, ns; | |
236 | unsigned long flags; | |
237 | enum drbd_state_rv rv; | |
238 | ||
239 | if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &mdev->flags)) | |
240 | return SS_CW_SUCCESS; | |
241 | ||
242 | if (test_and_clear_bit(CL_ST_CHG_FAIL, &mdev->flags)) | |
243 | return SS_CW_FAILED_BY_PEER; | |
244 | ||
b8907339 | 245 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
78bae59b | 246 | os = drbd_read_state(mdev); |
56707f9e | 247 | ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL); |
3509502d PR |
248 | rv = is_valid_transition(os, ns); |
249 | if (rv == SS_SUCCESS) | |
250 | rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */ | |
b8907339 PR |
251 | |
252 | if (!cl_wide_st_chg(mdev, os, ns)) | |
253 | rv = SS_CW_NO_NEED; | |
3509502d | 254 | if (rv == SS_UNKNOWN_ERROR) { |
b8907339 PR |
255 | rv = is_valid_state(mdev, ns); |
256 | if (rv == SS_SUCCESS) { | |
a75f34ad | 257 | rv = is_valid_soft_transition(os, ns); |
b8907339 PR |
258 | if (rv == SS_SUCCESS) |
259 | rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */ | |
260 | } | |
261 | } | |
262 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
263 | ||
264 | return rv; | |
265 | } | |
266 | ||
267 | /** | |
268 | * drbd_req_state() - Perform an eventually cluster wide state change | |
269 | * @mdev: DRBD device. | |
270 | * @mask: mask of state bits to change. | |
271 | * @val: value of new state bits. | |
272 | * @f: flags | |
273 | * | |
274 | * Should not be called directly, use drbd_request_state() or | |
275 | * _drbd_request_state(). | |
276 | */ | |
277 | static enum drbd_state_rv | |
278 | drbd_req_state(struct drbd_conf *mdev, union drbd_state mask, | |
279 | union drbd_state val, enum chg_state_flags f) | |
280 | { | |
281 | struct completion done; | |
282 | unsigned long flags; | |
283 | union drbd_state os, ns; | |
284 | enum drbd_state_rv rv; | |
285 | ||
286 | init_completion(&done); | |
287 | ||
288 | if (f & CS_SERIALIZE) | |
8410da8f | 289 | mutex_lock(mdev->state_mutex); |
b8907339 PR |
290 | |
291 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); | |
78bae59b | 292 | os = drbd_read_state(mdev); |
56707f9e | 293 | ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL); |
3509502d | 294 | rv = is_valid_transition(os, ns); |
3c5e5f6a LE |
295 | if (rv < SS_SUCCESS) { |
296 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
3509502d | 297 | goto abort; |
3c5e5f6a | 298 | } |
b8907339 PR |
299 | |
300 | if (cl_wide_st_chg(mdev, os, ns)) { | |
301 | rv = is_valid_state(mdev, ns); | |
302 | if (rv == SS_SUCCESS) | |
a75f34ad | 303 | rv = is_valid_soft_transition(os, ns); |
b8907339 PR |
304 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
305 | ||
306 | if (rv < SS_SUCCESS) { | |
307 | if (f & CS_VERBOSE) | |
308 | print_st_err(mdev, os, ns, rv); | |
309 | goto abort; | |
310 | } | |
311 | ||
d24ae219 | 312 | if (drbd_send_state_req(mdev, mask, val)) { |
b8907339 PR |
313 | rv = SS_CW_FAILED_BY_PEER; |
314 | if (f & CS_VERBOSE) | |
315 | print_st_err(mdev, os, ns, rv); | |
316 | goto abort; | |
317 | } | |
318 | ||
319 | wait_event(mdev->state_wait, | |
320 | (rv = _req_st_cond(mdev, mask, val))); | |
321 | ||
322 | if (rv < SS_SUCCESS) { | |
b8907339 PR |
323 | if (f & CS_VERBOSE) |
324 | print_st_err(mdev, os, ns, rv); | |
325 | goto abort; | |
326 | } | |
327 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); | |
78bae59b | 328 | ns = apply_mask_val(drbd_read_state(mdev), mask, val); |
b8907339 | 329 | rv = _drbd_set_state(mdev, ns, f, &done); |
b8907339 PR |
330 | } else { |
331 | rv = _drbd_set_state(mdev, ns, f, &done); | |
332 | } | |
333 | ||
334 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
335 | ||
336 | if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) { | |
337 | D_ASSERT(current != mdev->tconn->worker.task); | |
338 | wait_for_completion(&done); | |
339 | } | |
340 | ||
341 | abort: | |
342 | if (f & CS_SERIALIZE) | |
8410da8f | 343 | mutex_unlock(mdev->state_mutex); |
b8907339 PR |
344 | |
345 | return rv; | |
346 | } | |
347 | ||
348 | /** | |
349 | * _drbd_request_state() - Request a state change (with flags) | |
350 | * @mdev: DRBD device. | |
351 | * @mask: mask of state bits to change. | |
352 | * @val: value of new state bits. | |
353 | * @f: flags | |
354 | * | |
355 | * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE | |
356 | * flag, or when logging of failed state change requests is not desired. | |
357 | */ | |
358 | enum drbd_state_rv | |
359 | _drbd_request_state(struct drbd_conf *mdev, union drbd_state mask, | |
360 | union drbd_state val, enum chg_state_flags f) | |
361 | { | |
362 | enum drbd_state_rv rv; | |
363 | ||
364 | wait_event(mdev->state_wait, | |
365 | (rv = drbd_req_state(mdev, mask, val, f)) != SS_IN_TRANSIENT_STATE); | |
366 | ||
367 | return rv; | |
368 | } | |
369 | ||
370 | static void print_st(struct drbd_conf *mdev, char *name, union drbd_state ns) | |
371 | { | |
372 | dev_err(DEV, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n", | |
373 | name, | |
374 | drbd_conn_str(ns.conn), | |
375 | drbd_role_str(ns.role), | |
376 | drbd_role_str(ns.peer), | |
377 | drbd_disk_str(ns.disk), | |
378 | drbd_disk_str(ns.pdsk), | |
379 | is_susp(ns) ? 's' : 'r', | |
380 | ns.aftr_isp ? 'a' : '-', | |
381 | ns.peer_isp ? 'p' : '-', | |
382 | ns.user_isp ? 'u' : '-', | |
383 | ns.susp_fen ? 'F' : '-', | |
384 | ns.susp_nod ? 'N' : '-' | |
385 | ); | |
386 | } | |
387 | ||
388 | void print_st_err(struct drbd_conf *mdev, union drbd_state os, | |
389 | union drbd_state ns, enum drbd_state_rv err) | |
390 | { | |
391 | if (err == SS_IN_TRANSIENT_STATE) | |
392 | return; | |
393 | dev_err(DEV, "State change failed: %s\n", drbd_set_st_err_str(err)); | |
394 | print_st(mdev, " state", os); | |
395 | print_st(mdev, "wanted", ns); | |
396 | } | |
397 | ||
435693e8 | 398 | static long print_state_change(char *pb, union drbd_state os, union drbd_state ns, |
bbeb641c PR |
399 | enum chg_state_flags flags) |
400 | { | |
435693e8 | 401 | char *pbp; |
bbeb641c PR |
402 | pbp = pb; |
403 | *pbp = 0; | |
706cb24c | 404 | |
435693e8 | 405 | if (ns.role != os.role && flags & CS_DC_ROLE) |
bbeb641c PR |
406 | pbp += sprintf(pbp, "role( %s -> %s ) ", |
407 | drbd_role_str(os.role), | |
408 | drbd_role_str(ns.role)); | |
435693e8 | 409 | if (ns.peer != os.peer && flags & CS_DC_PEER) |
bbeb641c PR |
410 | pbp += sprintf(pbp, "peer( %s -> %s ) ", |
411 | drbd_role_str(os.peer), | |
412 | drbd_role_str(ns.peer)); | |
435693e8 | 413 | if (ns.conn != os.conn && flags & CS_DC_CONN) |
bbeb641c PR |
414 | pbp += sprintf(pbp, "conn( %s -> %s ) ", |
415 | drbd_conn_str(os.conn), | |
416 | drbd_conn_str(ns.conn)); | |
435693e8 | 417 | if (ns.disk != os.disk && flags & CS_DC_DISK) |
bbeb641c PR |
418 | pbp += sprintf(pbp, "disk( %s -> %s ) ", |
419 | drbd_disk_str(os.disk), | |
420 | drbd_disk_str(ns.disk)); | |
435693e8 | 421 | if (ns.pdsk != os.pdsk && flags & CS_DC_PDSK) |
bbeb641c PR |
422 | pbp += sprintf(pbp, "pdsk( %s -> %s ) ", |
423 | drbd_disk_str(os.pdsk), | |
424 | drbd_disk_str(ns.pdsk)); | |
706cb24c PR |
425 | |
426 | return pbp - pb; | |
427 | } | |
428 | ||
429 | static void drbd_pr_state_change(struct drbd_conf *mdev, union drbd_state os, union drbd_state ns, | |
430 | enum chg_state_flags flags) | |
431 | { | |
432 | char pb[300]; | |
433 | char *pbp = pb; | |
434 | ||
435 | pbp += print_state_change(pbp, os, ns, flags ^ CS_DC_MASK); | |
436 | ||
bbeb641c PR |
437 | if (ns.aftr_isp != os.aftr_isp) |
438 | pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ", | |
439 | os.aftr_isp, | |
440 | ns.aftr_isp); | |
441 | if (ns.peer_isp != os.peer_isp) | |
442 | pbp += sprintf(pbp, "peer_isp( %d -> %d ) ", | |
443 | os.peer_isp, | |
444 | ns.peer_isp); | |
445 | if (ns.user_isp != os.user_isp) | |
446 | pbp += sprintf(pbp, "user_isp( %d -> %d ) ", | |
447 | os.user_isp, | |
448 | ns.user_isp); | |
435693e8 | 449 | |
706cb24c | 450 | if (pbp != pb) |
bbeb641c PR |
451 | dev_info(DEV, "%s\n", pb); |
452 | } | |
b8907339 | 453 | |
435693e8 PR |
454 | static void conn_pr_state_change(struct drbd_tconn *tconn, union drbd_state os, union drbd_state ns, |
455 | enum chg_state_flags flags) | |
456 | { | |
457 | char pb[300]; | |
706cb24c PR |
458 | char *pbp = pb; |
459 | ||
460 | pbp += print_state_change(pbp, os, ns, flags); | |
435693e8 | 461 | |
706cb24c PR |
462 | if (is_susp(ns) != is_susp(os) && flags & CS_DC_SUSP) |
463 | pbp += sprintf(pbp, "susp( %d -> %d ) ", | |
464 | is_susp(os), | |
465 | is_susp(ns)); | |
466 | ||
467 | if (pbp != pb) | |
435693e8 PR |
468 | conn_info(tconn, "%s\n", pb); |
469 | } | |
470 | ||
471 | ||
b8907339 PR |
472 | /** |
473 | * is_valid_state() - Returns an SS_ error code if ns is not valid | |
474 | * @mdev: DRBD device. | |
475 | * @ns: State to consider. | |
476 | */ | |
477 | static enum drbd_state_rv | |
478 | is_valid_state(struct drbd_conf *mdev, union drbd_state ns) | |
479 | { | |
480 | /* See drbd_state_sw_errors in drbd_strings.c */ | |
481 | ||
482 | enum drbd_fencing_p fp; | |
483 | enum drbd_state_rv rv = SS_SUCCESS; | |
44ed167d | 484 | struct net_conf *nc; |
b8907339 PR |
485 | |
486 | fp = FP_DONT_CARE; | |
487 | if (get_ldev(mdev)) { | |
488 | fp = mdev->ldev->dc.fencing; | |
489 | put_ldev(mdev); | |
490 | } | |
491 | ||
44ed167d PR |
492 | rcu_read_lock(); |
493 | nc = rcu_dereference(mdev->tconn->net_conf); | |
494 | if (nc) { | |
495 | if (!nc->two_primaries && ns.role == R_PRIMARY) { | |
047e95e2 PR |
496 | if (ns.peer == R_PRIMARY) |
497 | rv = SS_TWO_PRIMARIES; | |
cb703454 | 498 | else if (conn_highest_peer(mdev->tconn) == R_PRIMARY) |
047e95e2 | 499 | rv = SS_O_VOL_PEER_PRI; |
44ed167d | 500 | } |
b8907339 PR |
501 | } |
502 | ||
503 | if (rv <= 0) | |
504 | /* already found a reason to abort */; | |
505 | else if (ns.role == R_SECONDARY && mdev->open_cnt) | |
506 | rv = SS_DEVICE_IN_USE; | |
507 | ||
508 | else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE) | |
509 | rv = SS_NO_UP_TO_DATE_DISK; | |
510 | ||
511 | else if (fp >= FP_RESOURCE && | |
512 | ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN) | |
513 | rv = SS_PRIMARY_NOP; | |
514 | ||
515 | else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT) | |
516 | rv = SS_NO_UP_TO_DATE_DISK; | |
517 | ||
518 | else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT) | |
519 | rv = SS_NO_LOCAL_DISK; | |
520 | ||
521 | else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT) | |
522 | rv = SS_NO_REMOTE_DISK; | |
523 | ||
524 | else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE) | |
525 | rv = SS_NO_UP_TO_DATE_DISK; | |
526 | ||
527 | else if ((ns.conn == C_CONNECTED || | |
528 | ns.conn == C_WF_BITMAP_S || | |
529 | ns.conn == C_SYNC_SOURCE || | |
530 | ns.conn == C_PAUSED_SYNC_S) && | |
531 | ns.disk == D_OUTDATED) | |
532 | rv = SS_CONNECTED_OUTDATES; | |
533 | ||
534 | else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
44ed167d | 535 | (nc->verify_alg[0] == 0)) |
b8907339 PR |
536 | rv = SS_NO_VERIFY_ALG; |
537 | ||
538 | else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
539 | mdev->tconn->agreed_pro_version < 88) | |
540 | rv = SS_NOT_SUPPORTED; | |
541 | ||
542 | else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN) | |
543 | rv = SS_CONNECTED_OUTDATES; | |
544 | ||
44ed167d PR |
545 | rcu_read_unlock(); |
546 | ||
b8907339 PR |
547 | return rv; |
548 | } | |
549 | ||
550 | /** | |
a75f34ad | 551 | * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible |
3509502d PR |
552 | * This function limits state transitions that may be declined by DRBD. I.e. |
553 | * user requests (aka soft transitions). | |
b8907339 PR |
554 | * @mdev: DRBD device. |
555 | * @ns: new state. | |
556 | * @os: old state. | |
557 | */ | |
558 | static enum drbd_state_rv | |
a75f34ad | 559 | is_valid_soft_transition(union drbd_state os, union drbd_state ns) |
b8907339 PR |
560 | { |
561 | enum drbd_state_rv rv = SS_SUCCESS; | |
562 | ||
563 | if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) && | |
564 | os.conn > C_CONNECTED) | |
565 | rv = SS_RESYNC_RUNNING; | |
566 | ||
567 | if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE) | |
568 | rv = SS_ALREADY_STANDALONE; | |
569 | ||
570 | if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS) | |
571 | rv = SS_IS_DISKLESS; | |
572 | ||
573 | if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED) | |
574 | rv = SS_NO_NET_CONFIG; | |
575 | ||
576 | if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING) | |
577 | rv = SS_LOWER_THAN_OUTDATED; | |
578 | ||
579 | if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED) | |
580 | rv = SS_IN_TRANSIENT_STATE; | |
581 | ||
2325eb66 PR |
582 | /* if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS) |
583 | rv = SS_IN_TRANSIENT_STATE; */ | |
b8907339 PR |
584 | |
585 | if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED) | |
586 | rv = SS_NEED_CONNECTION; | |
587 | ||
588 | if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && | |
589 | ns.conn != os.conn && os.conn > C_CONNECTED) | |
590 | rv = SS_RESYNC_RUNNING; | |
591 | ||
592 | if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) && | |
593 | os.conn < C_CONNECTED) | |
594 | rv = SS_NEED_CONNECTION; | |
595 | ||
596 | if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE) | |
597 | && os.conn < C_WF_REPORT_PARAMS) | |
598 | rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */ | |
599 | ||
600 | return rv; | |
601 | } | |
602 | ||
3509502d | 603 | static enum drbd_state_rv |
fda74117 | 604 | is_valid_conn_transition(enum drbd_conns oc, enum drbd_conns nc) |
3509502d | 605 | { |
d9cc6e23 LE |
606 | /* no change -> nothing to do, at least for the connection part */ |
607 | if (oc == nc) | |
608 | return SS_NOTHING_TO_DO; | |
3509502d | 609 | |
d9cc6e23 LE |
610 | /* disconnect of an unconfigured connection does not make sense */ |
611 | if (oc == C_STANDALONE && nc == C_DISCONNECTING) | |
612 | return SS_ALREADY_STANDALONE; | |
613 | ||
614 | /* from C_STANDALONE, we start with C_UNCONNECTED */ | |
615 | if (oc == C_STANDALONE && nc != C_UNCONNECTED) | |
616 | return SS_NEED_CONNECTION; | |
3509502d PR |
617 | |
618 | /* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */ | |
fda74117 | 619 | if (oc >= C_TIMEOUT && oc <= C_TEAR_DOWN && nc != C_UNCONNECTED && nc != C_DISCONNECTING) |
d9cc6e23 | 620 | return SS_IN_TRANSIENT_STATE; |
3509502d PR |
621 | |
622 | /* After C_DISCONNECTING only C_STANDALONE may follow */ | |
fda74117 | 623 | if (oc == C_DISCONNECTING && nc != C_STANDALONE) |
d9cc6e23 | 624 | return SS_IN_TRANSIENT_STATE; |
3509502d | 625 | |
d9cc6e23 | 626 | return SS_SUCCESS; |
fda74117 PR |
627 | } |
628 | ||
629 | ||
630 | /** | |
631 | * is_valid_transition() - Returns an SS_ error code if the state transition is not possible | |
632 | * This limits hard state transitions. Hard state transitions are facts there are | |
633 | * imposed on DRBD by the environment. E.g. disk broke or network broke down. | |
634 | * But those hard state transitions are still not allowed to do everything. | |
635 | * @ns: new state. | |
636 | * @os: old state. | |
637 | */ | |
638 | static enum drbd_state_rv | |
639 | is_valid_transition(union drbd_state os, union drbd_state ns) | |
640 | { | |
641 | enum drbd_state_rv rv; | |
642 | ||
643 | rv = is_valid_conn_transition(os.conn, ns.conn); | |
644 | ||
3509502d PR |
645 | /* we cannot fail (again) if we already detached */ |
646 | if (ns.disk == D_FAILED && os.disk == D_DISKLESS) | |
647 | rv = SS_IS_DISKLESS; | |
648 | ||
4308a0a3 PR |
649 | /* if we are only D_ATTACHING yet, |
650 | * we can (and should) go directly to D_DISKLESS. */ | |
651 | if (ns.disk == D_FAILED && os.disk == D_ATTACHING) { | |
652 | printk("TODO: FIX ME\n"); | |
653 | rv = SS_IS_DISKLESS; | |
654 | } | |
655 | ||
3509502d PR |
656 | return rv; |
657 | } | |
658 | ||
b8907339 PR |
659 | /** |
660 | * sanitize_state() - Resolves implicitly necessary additional changes to a state transition | |
661 | * @mdev: DRBD device. | |
662 | * @os: old state. | |
663 | * @ns: new state. | |
664 | * @warn_sync_abort: | |
665 | * | |
666 | * When we loose connection, we have to set the state of the peers disk (pdsk) | |
667 | * to D_UNKNOWN. This rule and many more along those lines are in this function. | |
668 | */ | |
4308a0a3 PR |
669 | static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state ns, |
670 | const char **warn_sync_abort) | |
b8907339 PR |
671 | { |
672 | enum drbd_fencing_p fp; | |
673 | enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max; | |
674 | ||
675 | fp = FP_DONT_CARE; | |
676 | if (get_ldev(mdev)) { | |
677 | fp = mdev->ldev->dc.fencing; | |
678 | put_ldev(mdev); | |
679 | } | |
680 | ||
3509502d | 681 | /* Implications from connection to peer and peer_isp */ |
b8907339 PR |
682 | if (ns.conn < C_CONNECTED) { |
683 | ns.peer_isp = 0; | |
684 | ns.peer = R_UNKNOWN; | |
685 | if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT) | |
686 | ns.pdsk = D_UNKNOWN; | |
687 | } | |
688 | ||
689 | /* Clear the aftr_isp when becoming unconfigured */ | |
690 | if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY) | |
691 | ns.aftr_isp = 0; | |
692 | ||
4308a0a3 | 693 | /* An implication of the disk states onto the connection state */ |
b8907339 | 694 | /* Abort resync if a disk fails/detaches */ |
4308a0a3 | 695 | if (ns.conn > C_CONNECTED && (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) { |
b8907339 PR |
696 | if (warn_sync_abort) |
697 | *warn_sync_abort = | |
4308a0a3 | 698 | ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T ? |
b8907339 PR |
699 | "Online-verify" : "Resync"; |
700 | ns.conn = C_CONNECTED; | |
701 | } | |
702 | ||
703 | /* Connection breaks down before we finished "Negotiating" */ | |
704 | if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING && | |
705 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
706 | if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) { | |
707 | ns.disk = mdev->new_state_tmp.disk; | |
708 | ns.pdsk = mdev->new_state_tmp.pdsk; | |
709 | } else { | |
710 | dev_alert(DEV, "Connection lost while negotiating, no data!\n"); | |
711 | ns.disk = D_DISKLESS; | |
712 | ns.pdsk = D_UNKNOWN; | |
713 | } | |
714 | put_ldev(mdev); | |
715 | } | |
716 | ||
717 | /* D_CONSISTENT and D_OUTDATED vanish when we get connected */ | |
718 | if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) { | |
719 | if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED) | |
720 | ns.disk = D_UP_TO_DATE; | |
721 | if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED) | |
722 | ns.pdsk = D_UP_TO_DATE; | |
723 | } | |
724 | ||
725 | /* Implications of the connection stat on the disk states */ | |
726 | disk_min = D_DISKLESS; | |
727 | disk_max = D_UP_TO_DATE; | |
728 | pdsk_min = D_INCONSISTENT; | |
729 | pdsk_max = D_UNKNOWN; | |
730 | switch ((enum drbd_conns)ns.conn) { | |
731 | case C_WF_BITMAP_T: | |
732 | case C_PAUSED_SYNC_T: | |
733 | case C_STARTING_SYNC_T: | |
734 | case C_WF_SYNC_UUID: | |
735 | case C_BEHIND: | |
736 | disk_min = D_INCONSISTENT; | |
737 | disk_max = D_OUTDATED; | |
738 | pdsk_min = D_UP_TO_DATE; | |
739 | pdsk_max = D_UP_TO_DATE; | |
740 | break; | |
741 | case C_VERIFY_S: | |
742 | case C_VERIFY_T: | |
743 | disk_min = D_UP_TO_DATE; | |
744 | disk_max = D_UP_TO_DATE; | |
745 | pdsk_min = D_UP_TO_DATE; | |
746 | pdsk_max = D_UP_TO_DATE; | |
747 | break; | |
748 | case C_CONNECTED: | |
749 | disk_min = D_DISKLESS; | |
750 | disk_max = D_UP_TO_DATE; | |
751 | pdsk_min = D_DISKLESS; | |
752 | pdsk_max = D_UP_TO_DATE; | |
753 | break; | |
754 | case C_WF_BITMAP_S: | |
755 | case C_PAUSED_SYNC_S: | |
756 | case C_STARTING_SYNC_S: | |
757 | case C_AHEAD: | |
758 | disk_min = D_UP_TO_DATE; | |
759 | disk_max = D_UP_TO_DATE; | |
760 | pdsk_min = D_INCONSISTENT; | |
761 | pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/ | |
762 | break; | |
763 | case C_SYNC_TARGET: | |
764 | disk_min = D_INCONSISTENT; | |
765 | disk_max = D_INCONSISTENT; | |
766 | pdsk_min = D_UP_TO_DATE; | |
767 | pdsk_max = D_UP_TO_DATE; | |
768 | break; | |
769 | case C_SYNC_SOURCE: | |
770 | disk_min = D_UP_TO_DATE; | |
771 | disk_max = D_UP_TO_DATE; | |
772 | pdsk_min = D_INCONSISTENT; | |
773 | pdsk_max = D_INCONSISTENT; | |
774 | break; | |
775 | case C_STANDALONE: | |
776 | case C_DISCONNECTING: | |
777 | case C_UNCONNECTED: | |
778 | case C_TIMEOUT: | |
779 | case C_BROKEN_PIPE: | |
780 | case C_NETWORK_FAILURE: | |
781 | case C_PROTOCOL_ERROR: | |
782 | case C_TEAR_DOWN: | |
783 | case C_WF_CONNECTION: | |
784 | case C_WF_REPORT_PARAMS: | |
785 | case C_MASK: | |
786 | break; | |
787 | } | |
788 | if (ns.disk > disk_max) | |
789 | ns.disk = disk_max; | |
790 | ||
791 | if (ns.disk < disk_min) { | |
792 | dev_warn(DEV, "Implicitly set disk from %s to %s\n", | |
793 | drbd_disk_str(ns.disk), drbd_disk_str(disk_min)); | |
794 | ns.disk = disk_min; | |
795 | } | |
796 | if (ns.pdsk > pdsk_max) | |
797 | ns.pdsk = pdsk_max; | |
798 | ||
799 | if (ns.pdsk < pdsk_min) { | |
800 | dev_warn(DEV, "Implicitly set pdsk from %s to %s\n", | |
801 | drbd_disk_str(ns.pdsk), drbd_disk_str(pdsk_min)); | |
802 | ns.pdsk = pdsk_min; | |
803 | } | |
804 | ||
805 | if (fp == FP_STONITH && | |
4308a0a3 | 806 | (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED)) |
b8907339 PR |
807 | ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */ |
808 | ||
f399002e | 809 | if (mdev->tconn->res_opts.on_no_data == OND_SUSPEND_IO && |
4308a0a3 | 810 | (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)) |
b8907339 PR |
811 | ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */ |
812 | ||
813 | if (ns.aftr_isp || ns.peer_isp || ns.user_isp) { | |
814 | if (ns.conn == C_SYNC_SOURCE) | |
815 | ns.conn = C_PAUSED_SYNC_S; | |
816 | if (ns.conn == C_SYNC_TARGET) | |
817 | ns.conn = C_PAUSED_SYNC_T; | |
818 | } else { | |
819 | if (ns.conn == C_PAUSED_SYNC_S) | |
820 | ns.conn = C_SYNC_SOURCE; | |
821 | if (ns.conn == C_PAUSED_SYNC_T) | |
822 | ns.conn = C_SYNC_TARGET; | |
823 | } | |
824 | ||
825 | return ns; | |
826 | } | |
827 | ||
828 | void drbd_resume_al(struct drbd_conf *mdev) | |
829 | { | |
830 | if (test_and_clear_bit(AL_SUSPENDED, &mdev->flags)) | |
831 | dev_info(DEV, "Resumed AL updates\n"); | |
832 | } | |
833 | ||
834 | /* helper for __drbd_set_state */ | |
835 | static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs) | |
836 | { | |
837 | if (mdev->tconn->agreed_pro_version < 90) | |
838 | mdev->ov_start_sector = 0; | |
839 | mdev->rs_total = drbd_bm_bits(mdev); | |
840 | mdev->ov_position = 0; | |
841 | if (cs == C_VERIFY_T) { | |
842 | /* starting online verify from an arbitrary position | |
843 | * does not fit well into the existing protocol. | |
844 | * on C_VERIFY_T, we initialize ov_left and friends | |
845 | * implicitly in receive_DataRequest once the | |
846 | * first P_OV_REQUEST is received */ | |
847 | mdev->ov_start_sector = ~(sector_t)0; | |
848 | } else { | |
849 | unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector); | |
850 | if (bit >= mdev->rs_total) { | |
851 | mdev->ov_start_sector = | |
852 | BM_BIT_TO_SECT(mdev->rs_total - 1); | |
853 | mdev->rs_total = 1; | |
854 | } else | |
855 | mdev->rs_total -= bit; | |
856 | mdev->ov_position = mdev->ov_start_sector; | |
857 | } | |
858 | mdev->ov_left = mdev->rs_total; | |
859 | } | |
860 | ||
861 | /** | |
862 | * __drbd_set_state() - Set a new DRBD state | |
863 | * @mdev: DRBD device. | |
864 | * @ns: new state. | |
865 | * @flags: Flags | |
866 | * @done: Optional completion, that will get completed after the after_state_ch() finished | |
867 | * | |
868 | * Caller needs to hold req_lock, and global_state_lock. Do not call directly. | |
869 | */ | |
870 | enum drbd_state_rv | |
871 | __drbd_set_state(struct drbd_conf *mdev, union drbd_state ns, | |
872 | enum chg_state_flags flags, struct completion *done) | |
873 | { | |
874 | union drbd_state os; | |
875 | enum drbd_state_rv rv = SS_SUCCESS; | |
876 | const char *warn_sync_abort = NULL; | |
877 | struct after_state_chg_work *ascw; | |
878 | ||
78bae59b | 879 | os = drbd_read_state(mdev); |
b8907339 | 880 | |
4308a0a3 | 881 | ns = sanitize_state(mdev, ns, &warn_sync_abort); |
b8907339 PR |
882 | if (ns.i == os.i) |
883 | return SS_NOTHING_TO_DO; | |
884 | ||
3509502d PR |
885 | rv = is_valid_transition(os, ns); |
886 | if (rv < SS_SUCCESS) | |
887 | return rv; | |
888 | ||
b8907339 PR |
889 | if (!(flags & CS_HARD)) { |
890 | /* pre-state-change checks ; only look at ns */ | |
891 | /* See drbd_state_sw_errors in drbd_strings.c */ | |
892 | ||
893 | rv = is_valid_state(mdev, ns); | |
894 | if (rv < SS_SUCCESS) { | |
895 | /* If the old state was illegal as well, then let | |
896 | this happen...*/ | |
897 | ||
898 | if (is_valid_state(mdev, os) == rv) | |
a75f34ad | 899 | rv = is_valid_soft_transition(os, ns); |
b8907339 | 900 | } else |
a75f34ad | 901 | rv = is_valid_soft_transition(os, ns); |
b8907339 PR |
902 | } |
903 | ||
904 | if (rv < SS_SUCCESS) { | |
905 | if (flags & CS_VERBOSE) | |
906 | print_st_err(mdev, os, ns, rv); | |
907 | return rv; | |
908 | } | |
909 | ||
910 | if (warn_sync_abort) | |
911 | dev_warn(DEV, "%s aborted.\n", warn_sync_abort); | |
912 | ||
435693e8 | 913 | drbd_pr_state_change(mdev, os, ns, flags); |
b8907339 | 914 | |
706cb24c PR |
915 | /* Display changes to the susp* flags that where caused by the call to |
916 | sanitize_state(). Only display it here if we where not called from | |
917 | _conn_request_state() */ | |
918 | if (!(flags & CS_DC_SUSP)) | |
919 | conn_pr_state_change(mdev->tconn, os, ns, (flags & ~CS_DC_MASK) | CS_DC_SUSP); | |
920 | ||
b8907339 PR |
921 | /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference |
922 | * on the ldev here, to be sure the transition -> D_DISKLESS resp. | |
923 | * drbd_ldev_destroy() won't happen before our corresponding | |
924 | * after_state_ch works run, where we put_ldev again. */ | |
925 | if ((os.disk != D_FAILED && ns.disk == D_FAILED) || | |
926 | (os.disk != D_DISKLESS && ns.disk == D_DISKLESS)) | |
927 | atomic_inc(&mdev->local_cnt); | |
928 | ||
da9fbc27 | 929 | mdev->state.i = ns.i; |
8e0af25f PR |
930 | mdev->tconn->susp = ns.susp; |
931 | mdev->tconn->susp_nod = ns.susp_nod; | |
932 | mdev->tconn->susp_fen = ns.susp_fen; | |
b8907339 PR |
933 | |
934 | if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING) | |
935 | drbd_print_uuids(mdev, "attached to UUIDs"); | |
936 | ||
937 | wake_up(&mdev->misc_wait); | |
938 | wake_up(&mdev->state_wait); | |
2a67d8b9 | 939 | wake_up(&mdev->tconn->ping_wait); |
b8907339 PR |
940 | |
941 | /* aborted verify run. log the last position */ | |
942 | if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) && | |
943 | ns.conn < C_CONNECTED) { | |
944 | mdev->ov_start_sector = | |
945 | BM_BIT_TO_SECT(drbd_bm_bits(mdev) - mdev->ov_left); | |
946 | dev_info(DEV, "Online Verify reached sector %llu\n", | |
947 | (unsigned long long)mdev->ov_start_sector); | |
948 | } | |
949 | ||
950 | if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) && | |
951 | (ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)) { | |
952 | dev_info(DEV, "Syncer continues.\n"); | |
953 | mdev->rs_paused += (long)jiffies | |
954 | -(long)mdev->rs_mark_time[mdev->rs_last_mark]; | |
955 | if (ns.conn == C_SYNC_TARGET) | |
956 | mod_timer(&mdev->resync_timer, jiffies); | |
957 | } | |
958 | ||
959 | if ((os.conn == C_SYNC_TARGET || os.conn == C_SYNC_SOURCE) && | |
960 | (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) { | |
961 | dev_info(DEV, "Resync suspended\n"); | |
962 | mdev->rs_mark_time[mdev->rs_last_mark] = jiffies; | |
963 | } | |
964 | ||
965 | if (os.conn == C_CONNECTED && | |
966 | (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) { | |
967 | unsigned long now = jiffies; | |
968 | int i; | |
969 | ||
970 | set_ov_position(mdev, ns.conn); | |
971 | mdev->rs_start = now; | |
972 | mdev->rs_last_events = 0; | |
973 | mdev->rs_last_sect_ev = 0; | |
974 | mdev->ov_last_oos_size = 0; | |
975 | mdev->ov_last_oos_start = 0; | |
976 | ||
977 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { | |
978 | mdev->rs_mark_left[i] = mdev->ov_left; | |
979 | mdev->rs_mark_time[i] = now; | |
980 | } | |
981 | ||
982 | drbd_rs_controller_reset(mdev); | |
983 | ||
984 | if (ns.conn == C_VERIFY_S) { | |
985 | dev_info(DEV, "Starting Online Verify from sector %llu\n", | |
986 | (unsigned long long)mdev->ov_position); | |
987 | mod_timer(&mdev->resync_timer, jiffies); | |
988 | } | |
989 | } | |
990 | ||
991 | if (get_ldev(mdev)) { | |
992 | u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND| | |
993 | MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE| | |
994 | MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY); | |
995 | ||
996 | if (test_bit(CRASHED_PRIMARY, &mdev->flags)) | |
997 | mdf |= MDF_CRASHED_PRIMARY; | |
998 | if (mdev->state.role == R_PRIMARY || | |
999 | (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY)) | |
1000 | mdf |= MDF_PRIMARY_IND; | |
1001 | if (mdev->state.conn > C_WF_REPORT_PARAMS) | |
1002 | mdf |= MDF_CONNECTED_IND; | |
1003 | if (mdev->state.disk > D_INCONSISTENT) | |
1004 | mdf |= MDF_CONSISTENT; | |
1005 | if (mdev->state.disk > D_OUTDATED) | |
1006 | mdf |= MDF_WAS_UP_TO_DATE; | |
1007 | if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT) | |
1008 | mdf |= MDF_PEER_OUT_DATED; | |
1009 | if (mdf != mdev->ldev->md.flags) { | |
1010 | mdev->ldev->md.flags = mdf; | |
1011 | drbd_md_mark_dirty(mdev); | |
1012 | } | |
1013 | if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT) | |
1014 | drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]); | |
1015 | put_ldev(mdev); | |
1016 | } | |
1017 | ||
1018 | /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */ | |
1019 | if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT && | |
1020 | os.peer == R_SECONDARY && ns.peer == R_PRIMARY) | |
1021 | set_bit(CONSIDER_RESYNC, &mdev->flags); | |
1022 | ||
1023 | /* Receiver should clean up itself */ | |
1024 | if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING) | |
1025 | drbd_thread_stop_nowait(&mdev->tconn->receiver); | |
1026 | ||
1027 | /* Now the receiver finished cleaning up itself, it should die */ | |
1028 | if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE) | |
1029 | drbd_thread_stop_nowait(&mdev->tconn->receiver); | |
1030 | ||
1031 | /* Upon network failure, we need to restart the receiver. */ | |
1032 | if (os.conn > C_TEAR_DOWN && | |
1033 | ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT) | |
1034 | drbd_thread_restart_nowait(&mdev->tconn->receiver); | |
1035 | ||
1036 | /* Resume AL writing if we get a connection */ | |
1037 | if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) | |
1038 | drbd_resume_al(mdev); | |
1039 | ||
1040 | ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC); | |
1041 | if (ascw) { | |
1042 | ascw->os = os; | |
1043 | ascw->ns = ns; | |
1044 | ascw->flags = flags; | |
1045 | ascw->w.cb = w_after_state_ch; | |
a21e9298 | 1046 | ascw->w.mdev = mdev; |
b8907339 PR |
1047 | ascw->done = done; |
1048 | drbd_queue_work(&mdev->tconn->data.work, &ascw->w); | |
1049 | } else { | |
bbeb641c | 1050 | dev_err(DEV, "Could not kmalloc an ascw\n"); |
b8907339 PR |
1051 | } |
1052 | ||
1053 | return rv; | |
1054 | } | |
1055 | ||
99920dc5 | 1056 | static int w_after_state_ch(struct drbd_work *w, int unused) |
b8907339 PR |
1057 | { |
1058 | struct after_state_chg_work *ascw = | |
1059 | container_of(w, struct after_state_chg_work, w); | |
00d56944 | 1060 | struct drbd_conf *mdev = w->mdev; |
b8907339 PR |
1061 | |
1062 | after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags); | |
1063 | if (ascw->flags & CS_WAIT_COMPLETE) { | |
1064 | D_ASSERT(ascw->done != NULL); | |
1065 | complete(ascw->done); | |
1066 | } | |
1067 | kfree(ascw); | |
1068 | ||
99920dc5 | 1069 | return 0; |
b8907339 PR |
1070 | } |
1071 | ||
1072 | static void abw_start_sync(struct drbd_conf *mdev, int rv) | |
1073 | { | |
1074 | if (rv) { | |
1075 | dev_err(DEV, "Writing the bitmap failed not starting resync.\n"); | |
1076 | _drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE); | |
1077 | return; | |
1078 | } | |
1079 | ||
1080 | switch (mdev->state.conn) { | |
1081 | case C_STARTING_SYNC_T: | |
1082 | _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); | |
1083 | break; | |
1084 | case C_STARTING_SYNC_S: | |
1085 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
1086 | break; | |
1087 | } | |
1088 | } | |
1089 | ||
1090 | int drbd_bitmap_io_from_worker(struct drbd_conf *mdev, | |
1091 | int (*io_fn)(struct drbd_conf *), | |
1092 | char *why, enum bm_flag flags) | |
1093 | { | |
1094 | int rv; | |
1095 | ||
1096 | D_ASSERT(current == mdev->tconn->worker.task); | |
1097 | ||
1098 | /* open coded non-blocking drbd_suspend_io(mdev); */ | |
1099 | set_bit(SUSPEND_IO, &mdev->flags); | |
1100 | ||
1101 | drbd_bm_lock(mdev, why, flags); | |
1102 | rv = io_fn(mdev); | |
1103 | drbd_bm_unlock(mdev); | |
1104 | ||
1105 | drbd_resume_io(mdev); | |
1106 | ||
1107 | return rv; | |
1108 | } | |
1109 | ||
1110 | /** | |
1111 | * after_state_ch() - Perform after state change actions that may sleep | |
1112 | * @mdev: DRBD device. | |
1113 | * @os: old state. | |
1114 | * @ns: new state. | |
1115 | * @flags: Flags | |
1116 | */ | |
1117 | static void after_state_ch(struct drbd_conf *mdev, union drbd_state os, | |
1118 | union drbd_state ns, enum chg_state_flags flags) | |
1119 | { | |
1120 | enum drbd_fencing_p fp; | |
3b98c0c2 LE |
1121 | struct sib_info sib; |
1122 | ||
1123 | sib.sib_reason = SIB_STATE_CHANGE; | |
1124 | sib.os = os; | |
1125 | sib.ns = ns; | |
b8907339 PR |
1126 | |
1127 | if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) { | |
1128 | clear_bit(CRASHED_PRIMARY, &mdev->flags); | |
1129 | if (mdev->p_uuid) | |
1130 | mdev->p_uuid[UI_FLAGS] &= ~((u64)2); | |
1131 | } | |
1132 | ||
1133 | fp = FP_DONT_CARE; | |
1134 | if (get_ldev(mdev)) { | |
1135 | fp = mdev->ldev->dc.fencing; | |
1136 | put_ldev(mdev); | |
1137 | } | |
1138 | ||
1139 | /* Inform userspace about the change... */ | |
3b98c0c2 | 1140 | drbd_bcast_event(mdev, &sib); |
b8907339 PR |
1141 | |
1142 | if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) && | |
1143 | (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)) | |
1144 | drbd_khelper(mdev, "pri-on-incon-degr"); | |
1145 | ||
1146 | /* Here we have the actions that are performed after a | |
1147 | state change. This function might sleep */ | |
1148 | ||
b8907339 | 1149 | if (ns.susp_nod) { |
a6d00c8e PR |
1150 | enum drbd_req_event what = NOTHING; |
1151 | ||
1152 | if (os.conn < C_CONNECTED && conn_lowest_conn(mdev->tconn) >= C_CONNECTED) | |
b8907339 PR |
1153 | what = RESEND; |
1154 | ||
a6d00c8e | 1155 | if (os.disk == D_ATTACHING && conn_lowest_disk(mdev->tconn) > D_ATTACHING) |
b8907339 PR |
1156 | what = RESTART_FROZEN_DISK_IO; |
1157 | ||
a6d00c8e | 1158 | if (what != NOTHING) { |
b8907339 | 1159 | spin_lock_irq(&mdev->tconn->req_lock); |
a6d00c8e PR |
1160 | _tl_restart(mdev->tconn, what); |
1161 | _drbd_set_state(_NS(mdev, susp_nod, 0), CS_VERBOSE, NULL); | |
b8907339 PR |
1162 | spin_unlock_irq(&mdev->tconn->req_lock); |
1163 | } | |
b8907339 PR |
1164 | } |
1165 | ||
1166 | /* Became sync source. With protocol >= 96, we still need to send out | |
1167 | * the sync uuid now. Need to do that before any drbd_send_state, or | |
1168 | * the other side may go "paused sync" before receiving the sync uuids, | |
1169 | * which is unexpected. */ | |
1170 | if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) && | |
1171 | (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) && | |
1172 | mdev->tconn->agreed_pro_version >= 96 && get_ldev(mdev)) { | |
1173 | drbd_gen_and_send_sync_uuid(mdev); | |
1174 | put_ldev(mdev); | |
1175 | } | |
1176 | ||
1177 | /* Do not change the order of the if above and the two below... */ | |
1178 | if (os.pdsk == D_DISKLESS && ns.pdsk > D_DISKLESS) { /* attach on the peer */ | |
1179 | drbd_send_uuids(mdev); | |
1180 | drbd_send_state(mdev); | |
1181 | } | |
1182 | /* No point in queuing send_bitmap if we don't have a connection | |
1183 | * anymore, so check also the _current_ state, not only the new state | |
1184 | * at the time this work was queued. */ | |
1185 | if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S && | |
1186 | mdev->state.conn == C_WF_BITMAP_S) | |
1187 | drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL, | |
1188 | "send_bitmap (WFBitMapS)", | |
1189 | BM_LOCKED_TEST_ALLOWED); | |
1190 | ||
1191 | /* Lost contact to peer's copy of the data */ | |
1192 | if ((os.pdsk >= D_INCONSISTENT && | |
1193 | os.pdsk != D_UNKNOWN && | |
1194 | os.pdsk != D_OUTDATED) | |
1195 | && (ns.pdsk < D_INCONSISTENT || | |
1196 | ns.pdsk == D_UNKNOWN || | |
1197 | ns.pdsk == D_OUTDATED)) { | |
1198 | if (get_ldev(mdev)) { | |
1199 | if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) && | |
1200 | mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) { | |
2aebfabb | 1201 | if (drbd_suspended(mdev)) { |
b8907339 PR |
1202 | set_bit(NEW_CUR_UUID, &mdev->flags); |
1203 | } else { | |
1204 | drbd_uuid_new_current(mdev); | |
1205 | drbd_send_uuids(mdev); | |
1206 | } | |
1207 | } | |
1208 | put_ldev(mdev); | |
1209 | } | |
1210 | } | |
1211 | ||
1212 | if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) { | |
1213 | if (ns.peer == R_PRIMARY && mdev->ldev->md.uuid[UI_BITMAP] == 0) { | |
1214 | drbd_uuid_new_current(mdev); | |
1215 | drbd_send_uuids(mdev); | |
1216 | } | |
1217 | ||
1218 | /* D_DISKLESS Peer becomes secondary */ | |
1219 | if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY) | |
1220 | /* We may still be Primary ourselves. | |
1221 | * No harm done if the bitmap still changes, | |
1222 | * redirtied pages will follow later. */ | |
1223 | drbd_bitmap_io_from_worker(mdev, &drbd_bm_write, | |
1224 | "demote diskless peer", BM_LOCKED_SET_ALLOWED); | |
1225 | put_ldev(mdev); | |
1226 | } | |
1227 | ||
1228 | /* Write out all changed bits on demote. | |
1229 | * Though, no need to da that just yet | |
1230 | * if there is a resync going on still */ | |
1231 | if (os.role == R_PRIMARY && ns.role == R_SECONDARY && | |
1232 | mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) { | |
1233 | /* No changes to the bitmap expected this time, so assert that, | |
1234 | * even though no harm was done if it did change. */ | |
1235 | drbd_bitmap_io_from_worker(mdev, &drbd_bm_write, | |
1236 | "demote", BM_LOCKED_TEST_ALLOWED); | |
1237 | put_ldev(mdev); | |
1238 | } | |
1239 | ||
1240 | /* Last part of the attaching process ... */ | |
1241 | if (ns.conn >= C_CONNECTED && | |
1242 | os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) { | |
1243 | drbd_send_sizes(mdev, 0, 0); /* to start sync... */ | |
1244 | drbd_send_uuids(mdev); | |
1245 | drbd_send_state(mdev); | |
1246 | } | |
1247 | ||
1248 | /* We want to pause/continue resync, tell peer. */ | |
1249 | if (ns.conn >= C_CONNECTED && | |
1250 | ((os.aftr_isp != ns.aftr_isp) || | |
1251 | (os.user_isp != ns.user_isp))) | |
1252 | drbd_send_state(mdev); | |
1253 | ||
1254 | /* In case one of the isp bits got set, suspend other devices. */ | |
1255 | if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) && | |
1256 | (ns.aftr_isp || ns.peer_isp || ns.user_isp)) | |
1257 | suspend_other_sg(mdev); | |
1258 | ||
1259 | /* Make sure the peer gets informed about eventual state | |
1260 | changes (ISP bits) while we were in WFReportParams. */ | |
1261 | if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED) | |
1262 | drbd_send_state(mdev); | |
1263 | ||
1264 | if (os.conn != C_AHEAD && ns.conn == C_AHEAD) | |
1265 | drbd_send_state(mdev); | |
1266 | ||
1267 | /* We are in the progress to start a full sync... */ | |
1268 | if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) || | |
1269 | (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S)) | |
1270 | /* no other bitmap changes expected during this phase */ | |
1271 | drbd_queue_bitmap_io(mdev, | |
1272 | &drbd_bmio_set_n_write, &abw_start_sync, | |
1273 | "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED); | |
1274 | ||
1275 | /* We are invalidating our self... */ | |
1276 | if (os.conn < C_CONNECTED && ns.conn < C_CONNECTED && | |
1277 | os.disk > D_INCONSISTENT && ns.disk == D_INCONSISTENT) | |
1278 | /* other bitmap operation expected during this phase */ | |
1279 | drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, NULL, | |
1280 | "set_n_write from invalidate", BM_LOCKED_MASK); | |
1281 | ||
1282 | /* first half of local IO error, failure to attach, | |
1283 | * or administrative detach */ | |
1284 | if (os.disk != D_FAILED && ns.disk == D_FAILED) { | |
1285 | enum drbd_io_error_p eh; | |
1286 | int was_io_error; | |
1287 | /* corresponding get_ldev was in __drbd_set_state, to serialize | |
1288 | * our cleanup here with the transition to D_DISKLESS, | |
1289 | * so it is safe to dreference ldev here. */ | |
1290 | eh = mdev->ldev->dc.on_io_error; | |
1291 | was_io_error = test_and_clear_bit(WAS_IO_ERROR, &mdev->flags); | |
1292 | ||
1293 | /* current state still has to be D_FAILED, | |
1294 | * there is only one way out: to D_DISKLESS, | |
1295 | * and that may only happen after our put_ldev below. */ | |
1296 | if (mdev->state.disk != D_FAILED) | |
1297 | dev_err(DEV, | |
1298 | "ASSERT FAILED: disk is %s during detach\n", | |
1299 | drbd_disk_str(mdev->state.disk)); | |
1300 | ||
927036f9 | 1301 | if (!drbd_send_state(mdev)) |
b8907339 PR |
1302 | dev_warn(DEV, "Notified peer that I am detaching my disk\n"); |
1303 | else | |
1304 | dev_err(DEV, "Sending state for detaching disk failed\n"); | |
1305 | ||
1306 | drbd_rs_cancel_all(mdev); | |
1307 | ||
1308 | /* In case we want to get something to stable storage still, | |
1309 | * this may be the last chance. | |
1310 | * Following put_ldev may transition to D_DISKLESS. */ | |
1311 | drbd_md_sync(mdev); | |
1312 | put_ldev(mdev); | |
1313 | ||
1314 | if (was_io_error && eh == EP_CALL_HELPER) | |
1315 | drbd_khelper(mdev, "local-io-error"); | |
1316 | } | |
1317 | ||
1318 | /* second half of local IO error, failure to attach, | |
1319 | * or administrative detach, | |
1320 | * after local_cnt references have reached zero again */ | |
1321 | if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) { | |
1322 | /* We must still be diskless, | |
1323 | * re-attach has to be serialized with this! */ | |
1324 | if (mdev->state.disk != D_DISKLESS) | |
1325 | dev_err(DEV, | |
1326 | "ASSERT FAILED: disk is %s while going diskless\n", | |
1327 | drbd_disk_str(mdev->state.disk)); | |
1328 | ||
1329 | mdev->rs_total = 0; | |
1330 | mdev->rs_failed = 0; | |
1331 | atomic_set(&mdev->rs_pending_cnt, 0); | |
1332 | ||
927036f9 | 1333 | if (!drbd_send_state(mdev)) |
b8907339 PR |
1334 | dev_warn(DEV, "Notified peer that I'm now diskless.\n"); |
1335 | /* corresponding get_ldev in __drbd_set_state | |
1336 | * this may finally trigger drbd_ldev_destroy. */ | |
1337 | put_ldev(mdev); | |
1338 | } | |
1339 | ||
1340 | /* Notify peer that I had a local IO error, and did not detached.. */ | |
1341 | if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT) | |
1342 | drbd_send_state(mdev); | |
1343 | ||
1344 | /* Disks got bigger while they were detached */ | |
1345 | if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING && | |
1346 | test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) { | |
1347 | if (ns.conn == C_CONNECTED) | |
1348 | resync_after_online_grow(mdev); | |
1349 | } | |
1350 | ||
1351 | /* A resync finished or aborted, wake paused devices... */ | |
1352 | if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) || | |
1353 | (os.peer_isp && !ns.peer_isp) || | |
1354 | (os.user_isp && !ns.user_isp)) | |
1355 | resume_next_sg(mdev); | |
1356 | ||
1357 | /* sync target done with resync. Explicitly notify peer, even though | |
1358 | * it should (at least for non-empty resyncs) already know itself. */ | |
1359 | if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED) | |
1360 | drbd_send_state(mdev); | |
1361 | ||
1362 | /* This triggers bitmap writeout of potentially still unwritten pages | |
1363 | * if the resync finished cleanly, or aborted because of peer disk | |
1364 | * failure, or because of connection loss. | |
1365 | * For resync aborted because of local disk failure, we cannot do | |
1366 | * any bitmap writeout anymore. | |
1367 | * No harm done if some bits change during this phase. | |
1368 | */ | |
1369 | if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) { | |
1370 | drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL, | |
1371 | "write from resync_finished", BM_LOCKED_SET_ALLOWED); | |
1372 | put_ldev(mdev); | |
1373 | } | |
1374 | ||
1375 | if (ns.disk == D_DISKLESS && | |
1376 | ns.conn == C_STANDALONE && | |
1377 | ns.role == R_SECONDARY) { | |
1378 | if (os.aftr_isp != ns.aftr_isp) | |
1379 | resume_next_sg(mdev); | |
1380 | } | |
1381 | ||
b8907339 PR |
1382 | drbd_md_sync(mdev); |
1383 | } | |
1384 | ||
bbeb641c PR |
1385 | struct after_conn_state_chg_work { |
1386 | struct drbd_work w; | |
1387 | enum drbd_conns oc; | |
8c7e16c3 | 1388 | union drbd_state ns_min; |
5f082f98 | 1389 | union drbd_state ns_max; /* new, max state, over all mdevs */ |
bbeb641c PR |
1390 | enum chg_state_flags flags; |
1391 | }; | |
1392 | ||
99920dc5 | 1393 | static int w_after_conn_state_ch(struct drbd_work *w, int unused) |
bbeb641c PR |
1394 | { |
1395 | struct after_conn_state_chg_work *acscw = | |
1396 | container_of(w, struct after_conn_state_chg_work, w); | |
1397 | struct drbd_tconn *tconn = w->tconn; | |
1398 | enum drbd_conns oc = acscw->oc; | |
5f082f98 | 1399 | union drbd_state ns_max = acscw->ns_max; |
a6d00c8e PR |
1400 | union drbd_state ns_min = acscw->ns_min; |
1401 | struct drbd_conf *mdev; | |
1402 | int vnr; | |
bbeb641c PR |
1403 | |
1404 | kfree(acscw); | |
1405 | ||
b8907339 | 1406 | /* Upon network configuration, we need to start the receiver */ |
5f082f98 | 1407 | if (oc == C_STANDALONE && ns_max.conn == C_UNCONNECTED) |
b8907339 PR |
1408 | drbd_thread_start(&tconn->receiver); |
1409 | ||
f3dfa40a LE |
1410 | if (oc == C_DISCONNECTING && ns_max.conn == C_STANDALONE) { |
1411 | struct net_conf *old_conf; | |
1412 | ||
a0095508 | 1413 | mutex_lock(&tconn->conf_update); |
f3dfa40a LE |
1414 | old_conf = tconn->net_conf; |
1415 | rcu_assign_pointer(tconn->net_conf, NULL); | |
1416 | conn_free_crypto(tconn); | |
a0095508 | 1417 | mutex_unlock(&tconn->conf_update); |
f3dfa40a LE |
1418 | |
1419 | synchronize_rcu(); | |
1420 | kfree(old_conf); | |
1421 | } | |
1422 | ||
a6d00c8e PR |
1423 | if (ns_max.susp_fen) { |
1424 | /* case1: The outdate peer handler is successful: */ | |
1425 | if (ns_max.pdsk <= D_OUTDATED) { | |
1426 | tl_clear(tconn); | |
695d08fa | 1427 | rcu_read_lock(); |
a6d00c8e PR |
1428 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
1429 | if (test_bit(NEW_CUR_UUID, &mdev->flags)) { | |
1430 | drbd_uuid_new_current(mdev); | |
1431 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
1432 | } | |
1433 | } | |
695d08fa | 1434 | rcu_read_unlock(); |
a6d00c8e PR |
1435 | conn_request_state(tconn, |
1436 | (union drbd_state) { { .susp_fen = 1 } }, | |
1437 | (union drbd_state) { { .susp_fen = 0 } }, | |
1438 | CS_VERBOSE); | |
1439 | } | |
1440 | /* case2: The connection was established again: */ | |
1441 | if (ns_min.conn >= C_CONNECTED) { | |
695d08fa | 1442 | rcu_read_lock(); |
a6d00c8e PR |
1443 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
1444 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
695d08fa | 1445 | rcu_read_unlock(); |
a6d00c8e PR |
1446 | spin_lock_irq(&tconn->req_lock); |
1447 | _tl_restart(tconn, RESEND); | |
1448 | _conn_request_state(tconn, | |
1449 | (union drbd_state) { { .susp_fen = 1 } }, | |
1450 | (union drbd_state) { { .susp_fen = 0 } }, | |
1451 | CS_VERBOSE); | |
1452 | spin_unlock_irq(&tconn->req_lock); | |
1453 | } | |
1454 | } | |
9dc9fbb3 | 1455 | kref_put(&tconn->kref, &conn_destroy); |
99920dc5 | 1456 | return 0; |
bbeb641c PR |
1457 | } |
1458 | ||
435693e8 | 1459 | void conn_old_common_state(struct drbd_tconn *tconn, union drbd_state *pcs, enum chg_state_flags *pf) |
88ef594e | 1460 | { |
435693e8 | 1461 | enum chg_state_flags flags = ~0; |
da9fbc27 | 1462 | union drbd_dev_state os, cs = {}; /* old_state, common_state */ |
88ef594e | 1463 | struct drbd_conf *mdev; |
435693e8 | 1464 | int vnr, first_vol = 1; |
88ef594e | 1465 | |
695d08fa | 1466 | rcu_read_lock(); |
88ef594e PR |
1467 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
1468 | os = mdev->state; | |
1469 | ||
435693e8 PR |
1470 | if (first_vol) { |
1471 | cs = os; | |
1472 | first_vol = 0; | |
1473 | continue; | |
1474 | } | |
1475 | ||
1476 | if (cs.role != os.role) | |
1477 | flags &= ~CS_DC_ROLE; | |
1478 | ||
1479 | if (cs.peer != os.peer) | |
1480 | flags &= ~CS_DC_PEER; | |
1481 | ||
1482 | if (cs.conn != os.conn) | |
1483 | flags &= ~CS_DC_CONN; | |
88ef594e | 1484 | |
435693e8 PR |
1485 | if (cs.disk != os.disk) |
1486 | flags &= ~CS_DC_DISK; | |
88ef594e | 1487 | |
435693e8 PR |
1488 | if (cs.pdsk != os.pdsk) |
1489 | flags &= ~CS_DC_PDSK; | |
1490 | } | |
695d08fa | 1491 | rcu_read_unlock(); |
435693e8 PR |
1492 | |
1493 | *pf |= CS_DC_MASK; | |
1494 | *pf &= flags; | |
da9fbc27 | 1495 | (*pcs).i = cs.i; |
88ef594e | 1496 | } |
bbeb641c | 1497 | |
bd0c824a PR |
1498 | static enum drbd_state_rv |
1499 | conn_is_valid_transition(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, | |
88ef594e | 1500 | enum chg_state_flags flags) |
bbeb641c | 1501 | { |
bd0c824a | 1502 | enum drbd_state_rv rv = SS_SUCCESS; |
bbeb641c | 1503 | union drbd_state ns, os; |
bd0c824a PR |
1504 | struct drbd_conf *mdev; |
1505 | int vnr; | |
bbeb641c | 1506 | |
695d08fa | 1507 | rcu_read_lock(); |
bd0c824a | 1508 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
78bae59b | 1509 | os = drbd_read_state(mdev); |
bd0c824a PR |
1510 | ns = sanitize_state(mdev, apply_mask_val(os, mask, val), NULL); |
1511 | ||
778bcf2e PR |
1512 | if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED) |
1513 | ns.disk = os.disk; | |
1514 | ||
bd0c824a PR |
1515 | if (ns.i == os.i) |
1516 | continue; | |
bbeb641c | 1517 | |
bd0c824a PR |
1518 | rv = is_valid_transition(os, ns); |
1519 | if (rv < SS_SUCCESS) | |
1520 | break; | |
1521 | ||
1522 | if (!(flags & CS_HARD)) { | |
1523 | rv = is_valid_state(mdev, ns); | |
1524 | if (rv < SS_SUCCESS) { | |
1525 | if (is_valid_state(mdev, os) == rv) | |
1526 | rv = is_valid_soft_transition(os, ns); | |
1527 | } else | |
1528 | rv = is_valid_soft_transition(os, ns); | |
1529 | } | |
1530 | if (rv < SS_SUCCESS) | |
1531 | break; | |
bbeb641c | 1532 | } |
695d08fa | 1533 | rcu_read_unlock(); |
bbeb641c | 1534 | |
bd0c824a PR |
1535 | if (rv < SS_SUCCESS && flags & CS_VERBOSE) |
1536 | print_st_err(mdev, os, ns, rv); | |
1537 | ||
1538 | return rv; | |
bbeb641c PR |
1539 | } |
1540 | ||
8c7e16c3 | 1541 | void |
bd0c824a | 1542 | conn_set_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, |
8c7e16c3 | 1543 | union drbd_state *pns_min, union drbd_state *pns_max, enum chg_state_flags flags) |
bbeb641c | 1544 | { |
8c7e16c3 PR |
1545 | union drbd_state ns, os, ns_max = { }; |
1546 | union drbd_state ns_min = { | |
1547 | { .role = R_MASK, | |
1548 | .peer = R_MASK, | |
1549 | .disk = D_MASK, | |
1550 | .pdsk = D_MASK | |
1551 | } }; | |
bd0c824a | 1552 | struct drbd_conf *mdev; |
bbeb641c | 1553 | enum drbd_state_rv rv; |
bd0c824a | 1554 | int vnr; |
bbeb641c | 1555 | |
bd0c824a PR |
1556 | if (mask.conn == C_MASK) |
1557 | tconn->cstate = val.conn; | |
1558 | ||
695d08fa | 1559 | rcu_read_lock(); |
bd0c824a | 1560 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
78bae59b | 1561 | os = drbd_read_state(mdev); |
bd0c824a PR |
1562 | ns = apply_mask_val(os, mask, val); |
1563 | ns = sanitize_state(mdev, ns, NULL); | |
bbeb641c | 1564 | |
778bcf2e PR |
1565 | if (flags & CS_IGN_OUTD_FAIL && ns.disk == D_OUTDATED && os.disk < D_OUTDATED) |
1566 | ns.disk = os.disk; | |
1567 | ||
bd0c824a PR |
1568 | rv = __drbd_set_state(mdev, ns, flags, NULL); |
1569 | if (rv < SS_SUCCESS) | |
1570 | BUG(); | |
bbeb641c | 1571 | |
8c7e16c3 PR |
1572 | ns.i = mdev->state.i; |
1573 | ns_max.role = max_role(ns.role, ns_max.role); | |
1574 | ns_max.peer = max_role(ns.peer, ns_max.peer); | |
1575 | ns_max.conn = max_t(enum drbd_conns, ns.conn, ns_max.conn); | |
1576 | ns_max.disk = max_t(enum drbd_disk_state, ns.disk, ns_max.disk); | |
1577 | ns_max.pdsk = max_t(enum drbd_disk_state, ns.pdsk, ns_max.pdsk); | |
1578 | ||
1579 | ns_min.role = min_role(ns.role, ns_min.role); | |
1580 | ns_min.peer = min_role(ns.peer, ns_min.peer); | |
1581 | ns_min.conn = min_t(enum drbd_conns, ns.conn, ns_min.conn); | |
1582 | ns_min.disk = min_t(enum drbd_disk_state, ns.disk, ns_min.disk); | |
1583 | ns_min.pdsk = min_t(enum drbd_disk_state, ns.pdsk, ns_min.pdsk); | |
bd0c824a | 1584 | } |
695d08fa | 1585 | rcu_read_unlock(); |
bbeb641c | 1586 | |
8c7e16c3 PR |
1587 | ns_min.susp = ns_max.susp = tconn->susp; |
1588 | ns_min.susp_nod = ns_max.susp_nod = tconn->susp_nod; | |
1589 | ns_min.susp_fen = ns_max.susp_fen = tconn->susp_fen; | |
1590 | ||
1591 | *pns_min = ns_min; | |
1592 | *pns_max = ns_max; | |
bbeb641c PR |
1593 | } |
1594 | ||
df24aa45 PR |
1595 | static enum drbd_state_rv |
1596 | _conn_rq_cond(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val) | |
1597 | { | |
df24aa45 PR |
1598 | enum drbd_state_rv rv; |
1599 | ||
1600 | if (test_and_clear_bit(CONN_WD_ST_CHG_OKAY, &tconn->flags)) | |
1601 | return SS_CW_SUCCESS; | |
1602 | ||
1603 | if (test_and_clear_bit(CONN_WD_ST_CHG_FAIL, &tconn->flags)) | |
1604 | return SS_CW_FAILED_BY_PEER; | |
1605 | ||
df24aa45 PR |
1606 | spin_lock_irq(&tconn->req_lock); |
1607 | rv = tconn->cstate != C_WF_REPORT_PARAMS ? SS_CW_NO_NEED : SS_UNKNOWN_ERROR; | |
1608 | ||
1609 | if (rv == SS_UNKNOWN_ERROR) | |
435693e8 | 1610 | rv = conn_is_valid_transition(tconn, mask, val, 0); |
df24aa45 | 1611 | |
bd0c824a PR |
1612 | if (rv == SS_SUCCESS) |
1613 | rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */ | |
df24aa45 PR |
1614 | |
1615 | spin_unlock_irq(&tconn->req_lock); | |
1616 | ||
1617 | return rv; | |
1618 | } | |
1619 | ||
1620 | static enum drbd_state_rv | |
1621 | conn_cl_wide(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, | |
1622 | enum chg_state_flags f) | |
1623 | { | |
1624 | enum drbd_state_rv rv; | |
1625 | ||
1626 | spin_unlock_irq(&tconn->req_lock); | |
1627 | mutex_lock(&tconn->cstate_mutex); | |
1628 | ||
caee1c3a | 1629 | if (conn_send_state_req(tconn, mask, val)) { |
df24aa45 PR |
1630 | rv = SS_CW_FAILED_BY_PEER; |
1631 | /* if (f & CS_VERBOSE) | |
1632 | print_st_err(mdev, os, ns, rv); */ | |
1633 | goto abort; | |
1634 | } | |
1635 | ||
1636 | wait_event(tconn->ping_wait, (rv = _conn_rq_cond(tconn, mask, val))); | |
1637 | ||
1638 | abort: | |
1639 | mutex_unlock(&tconn->cstate_mutex); | |
1640 | spin_lock_irq(&tconn->req_lock); | |
1641 | ||
1642 | return rv; | |
1643 | } | |
1644 | ||
bbeb641c PR |
1645 | enum drbd_state_rv |
1646 | _conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, | |
1647 | enum chg_state_flags flags) | |
1648 | { | |
1649 | enum drbd_state_rv rv = SS_SUCCESS; | |
bbeb641c PR |
1650 | struct after_conn_state_chg_work *acscw; |
1651 | enum drbd_conns oc = tconn->cstate; | |
8c7e16c3 | 1652 | union drbd_state ns_max, ns_min, os; |
bbeb641c | 1653 | |
bbeb641c PR |
1654 | rv = is_valid_conn_transition(oc, val.conn); |
1655 | if (rv < SS_SUCCESS) | |
1656 | goto abort; | |
1657 | ||
88ef594e | 1658 | rv = conn_is_valid_transition(tconn, mask, val, flags); |
bbeb641c PR |
1659 | if (rv < SS_SUCCESS) |
1660 | goto abort; | |
1661 | ||
df24aa45 PR |
1662 | if (oc == C_WF_REPORT_PARAMS && val.conn == C_DISCONNECTING && |
1663 | !(flags & (CS_LOCAL_ONLY | CS_HARD))) { | |
1664 | rv = conn_cl_wide(tconn, mask, val, flags); | |
1665 | if (rv < SS_SUCCESS) | |
1666 | goto abort; | |
1667 | } | |
1668 | ||
435693e8 | 1669 | conn_old_common_state(tconn, &os, &flags); |
706cb24c | 1670 | flags |= CS_DC_SUSP; |
8c7e16c3 | 1671 | conn_set_state(tconn, mask, val, &ns_min, &ns_max, flags); |
5f082f98 | 1672 | conn_pr_state_change(tconn, os, ns_max, flags); |
bbeb641c PR |
1673 | |
1674 | acscw = kmalloc(sizeof(*acscw), GFP_ATOMIC); | |
1675 | if (acscw) { | |
435693e8 | 1676 | acscw->oc = os.conn; |
8c7e16c3 | 1677 | acscw->ns_min = ns_min; |
5f082f98 | 1678 | acscw->ns_max = ns_max; |
bbeb641c PR |
1679 | acscw->flags = flags; |
1680 | acscw->w.cb = w_after_conn_state_ch; | |
9dc9fbb3 | 1681 | kref_get(&tconn->kref); |
bbeb641c PR |
1682 | acscw->w.tconn = tconn; |
1683 | drbd_queue_work(&tconn->data.work, &acscw->w); | |
1684 | } else { | |
1685 | conn_err(tconn, "Could not kmalloc an acscw\n"); | |
b8907339 | 1686 | } |
bbeb641c PR |
1687 | |
1688 | abort: | |
bbeb641c PR |
1689 | return rv; |
1690 | } | |
1691 | ||
1692 | enum drbd_state_rv | |
1693 | conn_request_state(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val, | |
1694 | enum chg_state_flags flags) | |
1695 | { | |
1696 | enum drbd_state_rv rv; | |
1697 | ||
1698 | spin_lock_irq(&tconn->req_lock); | |
1699 | rv = _conn_request_state(tconn, mask, val, flags); | |
1700 | spin_unlock_irq(&tconn->req_lock); | |
1701 | ||
1702 | return rv; | |
b8907339 | 1703 | } |