drbd_state.c 37 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247
  1. /*
  2. drbd_state.c
  3. This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
  4. Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
  5. Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
  6. Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
  7. Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
  8. from Logicworks, Inc. for making SDP replication support possible.
  9. drbd is free software; you can redistribute it and/or modify
  10. it under the terms of the GNU General Public License as published by
  11. the Free Software Foundation; either version 2, or (at your option)
  12. any later version.
  13. drbd is distributed in the hope that it will be useful,
  14. but WITHOUT ANY WARRANTY; without even the implied warranty of
  15. MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  16. GNU General Public License for more details.
  17. You should have received a copy of the GNU General Public License
  18. along with drbd; see the file COPYING. If not, write to
  19. the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  20. */
  21. #include <linux/drbd_limits.h>
  22. #include "drbd_int.h"
  23. #include "drbd_req.h"
  24. struct after_state_chg_work {
  25. struct drbd_work w;
  26. union drbd_state os;
  27. union drbd_state ns;
  28. enum chg_state_flags flags;
  29. struct completion *done;
  30. };
  31. extern void _tl_restart(struct drbd_conf *mdev, enum drbd_req_event what);
  32. int drbd_send_state_req(struct drbd_conf *, union drbd_state, union drbd_state);
  33. static int w_after_state_ch(struct drbd_work *w, int unused);
  34. static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
  35. union drbd_state ns, enum chg_state_flags flags);
  36. static void after_conn_state_ch(struct drbd_tconn *tconn, union drbd_state os,
  37. union drbd_state ns, enum chg_state_flags flags);
  38. static enum drbd_state_rv is_valid_state(struct drbd_conf *, union drbd_state);
  39. static enum drbd_state_rv is_valid_soft_transition(union drbd_state, union drbd_state);
  40. static enum drbd_state_rv is_valid_transition(union drbd_state os, union drbd_state ns);
  41. static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state os,
  42. union drbd_state ns, const char **warn_sync_abort);
  43. /**
  44. * cl_wide_st_chg() - true if the state change is a cluster wide one
  45. * @mdev: DRBD device.
  46. * @os: old (current) state.
  47. * @ns: new (wanted) state.
  48. */
  49. static int cl_wide_st_chg(struct drbd_conf *mdev,
  50. union drbd_state os, union drbd_state ns)
  51. {
  52. return (os.conn >= C_CONNECTED && ns.conn >= C_CONNECTED &&
  53. ((os.role != R_PRIMARY && ns.role == R_PRIMARY) ||
  54. (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
  55. (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) ||
  56. (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))) ||
  57. (os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) ||
  58. (os.conn == C_CONNECTED && ns.conn == C_VERIFY_S);
  59. }
  60. enum drbd_state_rv
  61. drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f,
  62. union drbd_state mask, union drbd_state val)
  63. {
  64. unsigned long flags;
  65. union drbd_state os, ns;
  66. enum drbd_state_rv rv;
  67. spin_lock_irqsave(&mdev->tconn->req_lock, flags);
  68. os = mdev->state;
  69. ns.i = (os.i & ~mask.i) | val.i;
  70. rv = _drbd_set_state(mdev, ns, f, NULL);
  71. ns = mdev->state;
  72. spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
  73. return rv;
  74. }
  75. /**
  76. * drbd_force_state() - Impose a change which happens outside our control on our state
  77. * @mdev: DRBD device.
  78. * @mask: mask of state bits to change.
  79. * @val: value of new state bits.
  80. */
  81. void drbd_force_state(struct drbd_conf *mdev,
  82. union drbd_state mask, union drbd_state val)
  83. {
  84. drbd_change_state(mdev, CS_HARD, mask, val);
  85. }
  86. static enum drbd_state_rv
  87. _req_st_cond(struct drbd_conf *mdev, union drbd_state mask,
  88. union drbd_state val)
  89. {
  90. union drbd_state os, ns;
  91. unsigned long flags;
  92. enum drbd_state_rv rv;
  93. if (test_and_clear_bit(CL_ST_CHG_SUCCESS, &mdev->flags))
  94. return SS_CW_SUCCESS;
  95. if (test_and_clear_bit(CL_ST_CHG_FAIL, &mdev->flags))
  96. return SS_CW_FAILED_BY_PEER;
  97. spin_lock_irqsave(&mdev->tconn->req_lock, flags);
  98. os = mdev->state;
  99. ns.i = (os.i & ~mask.i) | val.i;
  100. ns = sanitize_state(mdev, os, ns, NULL);
  101. rv = is_valid_transition(os, ns);
  102. if (rv == SS_SUCCESS)
  103. rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
  104. if (!cl_wide_st_chg(mdev, os, ns))
  105. rv = SS_CW_NO_NEED;
  106. if (rv == SS_UNKNOWN_ERROR) {
  107. rv = is_valid_state(mdev, ns);
  108. if (rv == SS_SUCCESS) {
  109. rv = is_valid_soft_transition(os, ns);
  110. if (rv == SS_SUCCESS)
  111. rv = SS_UNKNOWN_ERROR; /* cont waiting, otherwise fail. */
  112. }
  113. }
  114. spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
  115. return rv;
  116. }
  117. /**
  118. * drbd_req_state() - Perform an eventually cluster wide state change
  119. * @mdev: DRBD device.
  120. * @mask: mask of state bits to change.
  121. * @val: value of new state bits.
  122. * @f: flags
  123. *
  124. * Should not be called directly, use drbd_request_state() or
  125. * _drbd_request_state().
  126. */
  127. static enum drbd_state_rv
  128. drbd_req_state(struct drbd_conf *mdev, union drbd_state mask,
  129. union drbd_state val, enum chg_state_flags f)
  130. {
  131. struct completion done;
  132. unsigned long flags;
  133. union drbd_state os, ns;
  134. enum drbd_state_rv rv;
  135. init_completion(&done);
  136. if (f & CS_SERIALIZE)
  137. mutex_lock(&mdev->state_mutex);
  138. spin_lock_irqsave(&mdev->tconn->req_lock, flags);
  139. os = mdev->state;
  140. ns.i = (os.i & ~mask.i) | val.i;
  141. ns = sanitize_state(mdev, os, ns, NULL);
  142. rv = is_valid_transition(os, ns);
  143. if (rv < SS_SUCCESS)
  144. goto abort;
  145. if (cl_wide_st_chg(mdev, os, ns)) {
  146. rv = is_valid_state(mdev, ns);
  147. if (rv == SS_SUCCESS)
  148. rv = is_valid_soft_transition(os, ns);
  149. spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
  150. if (rv < SS_SUCCESS) {
  151. if (f & CS_VERBOSE)
  152. print_st_err(mdev, os, ns, rv);
  153. goto abort;
  154. }
  155. drbd_state_lock(mdev);
  156. if (!drbd_send_state_req(mdev, mask, val)) {
  157. drbd_state_unlock(mdev);
  158. rv = SS_CW_FAILED_BY_PEER;
  159. if (f & CS_VERBOSE)
  160. print_st_err(mdev, os, ns, rv);
  161. goto abort;
  162. }
  163. wait_event(mdev->state_wait,
  164. (rv = _req_st_cond(mdev, mask, val)));
  165. if (rv < SS_SUCCESS) {
  166. drbd_state_unlock(mdev);
  167. if (f & CS_VERBOSE)
  168. print_st_err(mdev, os, ns, rv);
  169. goto abort;
  170. }
  171. spin_lock_irqsave(&mdev->tconn->req_lock, flags);
  172. os = mdev->state;
  173. ns.i = (os.i & ~mask.i) | val.i;
  174. rv = _drbd_set_state(mdev, ns, f, &done);
  175. drbd_state_unlock(mdev);
  176. } else {
  177. rv = _drbd_set_state(mdev, ns, f, &done);
  178. }
  179. spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
  180. if (f & CS_WAIT_COMPLETE && rv == SS_SUCCESS) {
  181. D_ASSERT(current != mdev->tconn->worker.task);
  182. wait_for_completion(&done);
  183. }
  184. abort:
  185. if (f & CS_SERIALIZE)
  186. mutex_unlock(&mdev->state_mutex);
  187. return rv;
  188. }
  189. /**
  190. * _drbd_request_state() - Request a state change (with flags)
  191. * @mdev: DRBD device.
  192. * @mask: mask of state bits to change.
  193. * @val: value of new state bits.
  194. * @f: flags
  195. *
  196. * Cousin of drbd_request_state(), useful with the CS_WAIT_COMPLETE
  197. * flag, or when logging of failed state change requests is not desired.
  198. */
  199. enum drbd_state_rv
  200. _drbd_request_state(struct drbd_conf *mdev, union drbd_state mask,
  201. union drbd_state val, enum chg_state_flags f)
  202. {
  203. enum drbd_state_rv rv;
  204. wait_event(mdev->state_wait,
  205. (rv = drbd_req_state(mdev, mask, val, f)) != SS_IN_TRANSIENT_STATE);
  206. return rv;
  207. }
  208. static void print_st(struct drbd_conf *mdev, char *name, union drbd_state ns)
  209. {
  210. dev_err(DEV, " %s = { cs:%s ro:%s/%s ds:%s/%s %c%c%c%c%c%c }\n",
  211. name,
  212. drbd_conn_str(ns.conn),
  213. drbd_role_str(ns.role),
  214. drbd_role_str(ns.peer),
  215. drbd_disk_str(ns.disk),
  216. drbd_disk_str(ns.pdsk),
  217. is_susp(ns) ? 's' : 'r',
  218. ns.aftr_isp ? 'a' : '-',
  219. ns.peer_isp ? 'p' : '-',
  220. ns.user_isp ? 'u' : '-',
  221. ns.susp_fen ? 'F' : '-',
  222. ns.susp_nod ? 'N' : '-'
  223. );
  224. }
  225. void print_st_err(struct drbd_conf *mdev, union drbd_state os,
  226. union drbd_state ns, enum drbd_state_rv err)
  227. {
  228. if (err == SS_IN_TRANSIENT_STATE)
  229. return;
  230. dev_err(DEV, "State change failed: %s\n", drbd_set_st_err_str(err));
  231. print_st(mdev, " state", os);
  232. print_st(mdev, "wanted", ns);
  233. }
  234. /**
  235. * is_valid_state() - Returns an SS_ error code if ns is not valid
  236. * @mdev: DRBD device.
  237. * @ns: State to consider.
  238. */
  239. static enum drbd_state_rv
  240. is_valid_state(struct drbd_conf *mdev, union drbd_state ns)
  241. {
  242. /* See drbd_state_sw_errors in drbd_strings.c */
  243. enum drbd_fencing_p fp;
  244. enum drbd_state_rv rv = SS_SUCCESS;
  245. fp = FP_DONT_CARE;
  246. if (get_ldev(mdev)) {
  247. fp = mdev->ldev->dc.fencing;
  248. put_ldev(mdev);
  249. }
  250. if (get_net_conf(mdev->tconn)) {
  251. if (!mdev->tconn->net_conf->two_primaries &&
  252. ns.role == R_PRIMARY && ns.peer == R_PRIMARY)
  253. rv = SS_TWO_PRIMARIES;
  254. put_net_conf(mdev->tconn);
  255. }
  256. if (rv <= 0)
  257. /* already found a reason to abort */;
  258. else if (ns.role == R_SECONDARY && mdev->open_cnt)
  259. rv = SS_DEVICE_IN_USE;
  260. else if (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.disk < D_UP_TO_DATE)
  261. rv = SS_NO_UP_TO_DATE_DISK;
  262. else if (fp >= FP_RESOURCE &&
  263. ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk >= D_UNKNOWN)
  264. rv = SS_PRIMARY_NOP;
  265. else if (ns.role == R_PRIMARY && ns.disk <= D_INCONSISTENT && ns.pdsk <= D_INCONSISTENT)
  266. rv = SS_NO_UP_TO_DATE_DISK;
  267. else if (ns.conn > C_CONNECTED && ns.disk < D_INCONSISTENT)
  268. rv = SS_NO_LOCAL_DISK;
  269. else if (ns.conn > C_CONNECTED && ns.pdsk < D_INCONSISTENT)
  270. rv = SS_NO_REMOTE_DISK;
  271. else if (ns.conn > C_CONNECTED && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE)
  272. rv = SS_NO_UP_TO_DATE_DISK;
  273. else if ((ns.conn == C_CONNECTED ||
  274. ns.conn == C_WF_BITMAP_S ||
  275. ns.conn == C_SYNC_SOURCE ||
  276. ns.conn == C_PAUSED_SYNC_S) &&
  277. ns.disk == D_OUTDATED)
  278. rv = SS_CONNECTED_OUTDATES;
  279. else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
  280. (mdev->sync_conf.verify_alg[0] == 0))
  281. rv = SS_NO_VERIFY_ALG;
  282. else if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
  283. mdev->tconn->agreed_pro_version < 88)
  284. rv = SS_NOT_SUPPORTED;
  285. else if (ns.conn >= C_CONNECTED && ns.pdsk == D_UNKNOWN)
  286. rv = SS_CONNECTED_OUTDATES;
  287. return rv;
  288. }
  289. /**
  290. * is_valid_soft_transition() - Returns an SS_ error code if the state transition is not possible
  291. * This function limits state transitions that may be declined by DRBD. I.e.
  292. * user requests (aka soft transitions).
  293. * @mdev: DRBD device.
  294. * @ns: new state.
  295. * @os: old state.
  296. */
  297. static enum drbd_state_rv
  298. is_valid_soft_transition(union drbd_state os, union drbd_state ns)
  299. {
  300. enum drbd_state_rv rv = SS_SUCCESS;
  301. if ((ns.conn == C_STARTING_SYNC_T || ns.conn == C_STARTING_SYNC_S) &&
  302. os.conn > C_CONNECTED)
  303. rv = SS_RESYNC_RUNNING;
  304. if (ns.conn == C_DISCONNECTING && os.conn == C_STANDALONE)
  305. rv = SS_ALREADY_STANDALONE;
  306. if (ns.disk > D_ATTACHING && os.disk == D_DISKLESS)
  307. rv = SS_IS_DISKLESS;
  308. if (ns.conn == C_WF_CONNECTION && os.conn < C_UNCONNECTED)
  309. rv = SS_NO_NET_CONFIG;
  310. if (ns.disk == D_OUTDATED && os.disk < D_OUTDATED && os.disk != D_ATTACHING)
  311. rv = SS_LOWER_THAN_OUTDATED;
  312. if (ns.conn == C_DISCONNECTING && os.conn == C_UNCONNECTED)
  313. rv = SS_IN_TRANSIENT_STATE;
  314. if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS)
  315. rv = SS_IN_TRANSIENT_STATE;
  316. if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
  317. rv = SS_NEED_CONNECTION;
  318. if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) &&
  319. ns.conn != os.conn && os.conn > C_CONNECTED)
  320. rv = SS_RESYNC_RUNNING;
  321. if ((ns.conn == C_STARTING_SYNC_S || ns.conn == C_STARTING_SYNC_T) &&
  322. os.conn < C_CONNECTED)
  323. rv = SS_NEED_CONNECTION;
  324. if ((ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)
  325. && os.conn < C_WF_REPORT_PARAMS)
  326. rv = SS_NEED_CONNECTION; /* No NetworkFailure -> SyncTarget etc... */
  327. return rv;
  328. }
  329. /**
  330. * is_valid_transition() - Returns an SS_ error code if the state transition is not possible
  331. * This limits hard state transitions. Hard state transitions are facts there are
  332. * imposed on DRBD by the environment. E.g. disk broke or network broke down.
  333. * But those hard state transitions are still not allowed to do everything.
  334. * @ns: new state.
  335. * @os: old state.
  336. */
  337. static enum drbd_state_rv
  338. is_valid_transition(union drbd_state os, union drbd_state ns)
  339. {
  340. enum drbd_state_rv rv = SS_SUCCESS;
  341. /* Disallow Network errors to configure a device's network part */
  342. if ((ns.conn >= C_TIMEOUT && ns.conn <= C_TEAR_DOWN) &&
  343. os.conn <= C_DISCONNECTING)
  344. rv = SS_NEED_CONNECTION;
  345. /* After a network error only C_UNCONNECTED or C_DISCONNECTING may follow. */
  346. if (os.conn >= C_TIMEOUT && os.conn <= C_TEAR_DOWN &&
  347. ns.conn != C_UNCONNECTED && ns.conn != C_DISCONNECTING)
  348. rv = SS_IN_TRANSIENT_STATE;
  349. /* After C_DISCONNECTING only C_STANDALONE may follow */
  350. if (os.conn == C_DISCONNECTING && ns.conn != C_STANDALONE)
  351. rv = SS_IN_TRANSIENT_STATE;
  352. /* we cannot fail (again) if we already detached */
  353. if (ns.disk == D_FAILED && os.disk == D_DISKLESS)
  354. rv = SS_IS_DISKLESS;
  355. return rv;
  356. }
  357. /**
  358. * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
  359. * @mdev: DRBD device.
  360. * @os: old state.
  361. * @ns: new state.
  362. * @warn_sync_abort:
  363. *
  364. * When we loose connection, we have to set the state of the peers disk (pdsk)
  365. * to D_UNKNOWN. This rule and many more along those lines are in this function.
  366. */
  367. static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state os,
  368. union drbd_state ns, const char **warn_sync_abort)
  369. {
  370. enum drbd_fencing_p fp;
  371. enum drbd_disk_state disk_min, disk_max, pdsk_min, pdsk_max;
  372. fp = FP_DONT_CARE;
  373. if (get_ldev(mdev)) {
  374. fp = mdev->ldev->dc.fencing;
  375. put_ldev(mdev);
  376. }
  377. /* if we are only D_ATTACHING yet,
  378. * we can (and should) go directly to D_DISKLESS. */
  379. if (ns.disk == D_FAILED && os.disk == D_ATTACHING)
  380. ns.disk = D_DISKLESS;
  381. /* Implications from connection to peer and peer_isp */
  382. if (ns.conn < C_CONNECTED) {
  383. ns.peer_isp = 0;
  384. ns.peer = R_UNKNOWN;
  385. if (ns.pdsk > D_UNKNOWN || ns.pdsk < D_INCONSISTENT)
  386. ns.pdsk = D_UNKNOWN;
  387. }
  388. /* Clear the aftr_isp when becoming unconfigured */
  389. if (ns.conn == C_STANDALONE && ns.disk == D_DISKLESS && ns.role == R_SECONDARY)
  390. ns.aftr_isp = 0;
  391. /* Abort resync if a disk fails/detaches */
  392. if (os.conn > C_CONNECTED && ns.conn > C_CONNECTED &&
  393. (ns.disk <= D_FAILED || ns.pdsk <= D_FAILED)) {
  394. if (warn_sync_abort)
  395. *warn_sync_abort =
  396. os.conn == C_VERIFY_S || os.conn == C_VERIFY_T ?
  397. "Online-verify" : "Resync";
  398. ns.conn = C_CONNECTED;
  399. }
  400. /* Connection breaks down before we finished "Negotiating" */
  401. if (ns.conn < C_CONNECTED && ns.disk == D_NEGOTIATING &&
  402. get_ldev_if_state(mdev, D_NEGOTIATING)) {
  403. if (mdev->ed_uuid == mdev->ldev->md.uuid[UI_CURRENT]) {
  404. ns.disk = mdev->new_state_tmp.disk;
  405. ns.pdsk = mdev->new_state_tmp.pdsk;
  406. } else {
  407. dev_alert(DEV, "Connection lost while negotiating, no data!\n");
  408. ns.disk = D_DISKLESS;
  409. ns.pdsk = D_UNKNOWN;
  410. }
  411. put_ldev(mdev);
  412. }
  413. /* D_CONSISTENT and D_OUTDATED vanish when we get connected */
  414. if (ns.conn >= C_CONNECTED && ns.conn < C_AHEAD) {
  415. if (ns.disk == D_CONSISTENT || ns.disk == D_OUTDATED)
  416. ns.disk = D_UP_TO_DATE;
  417. if (ns.pdsk == D_CONSISTENT || ns.pdsk == D_OUTDATED)
  418. ns.pdsk = D_UP_TO_DATE;
  419. }
  420. /* Implications of the connection stat on the disk states */
  421. disk_min = D_DISKLESS;
  422. disk_max = D_UP_TO_DATE;
  423. pdsk_min = D_INCONSISTENT;
  424. pdsk_max = D_UNKNOWN;
  425. switch ((enum drbd_conns)ns.conn) {
  426. case C_WF_BITMAP_T:
  427. case C_PAUSED_SYNC_T:
  428. case C_STARTING_SYNC_T:
  429. case C_WF_SYNC_UUID:
  430. case C_BEHIND:
  431. disk_min = D_INCONSISTENT;
  432. disk_max = D_OUTDATED;
  433. pdsk_min = D_UP_TO_DATE;
  434. pdsk_max = D_UP_TO_DATE;
  435. break;
  436. case C_VERIFY_S:
  437. case C_VERIFY_T:
  438. disk_min = D_UP_TO_DATE;
  439. disk_max = D_UP_TO_DATE;
  440. pdsk_min = D_UP_TO_DATE;
  441. pdsk_max = D_UP_TO_DATE;
  442. break;
  443. case C_CONNECTED:
  444. disk_min = D_DISKLESS;
  445. disk_max = D_UP_TO_DATE;
  446. pdsk_min = D_DISKLESS;
  447. pdsk_max = D_UP_TO_DATE;
  448. break;
  449. case C_WF_BITMAP_S:
  450. case C_PAUSED_SYNC_S:
  451. case C_STARTING_SYNC_S:
  452. case C_AHEAD:
  453. disk_min = D_UP_TO_DATE;
  454. disk_max = D_UP_TO_DATE;
  455. pdsk_min = D_INCONSISTENT;
  456. pdsk_max = D_CONSISTENT; /* D_OUTDATED would be nice. But explicit outdate necessary*/
  457. break;
  458. case C_SYNC_TARGET:
  459. disk_min = D_INCONSISTENT;
  460. disk_max = D_INCONSISTENT;
  461. pdsk_min = D_UP_TO_DATE;
  462. pdsk_max = D_UP_TO_DATE;
  463. break;
  464. case C_SYNC_SOURCE:
  465. disk_min = D_UP_TO_DATE;
  466. disk_max = D_UP_TO_DATE;
  467. pdsk_min = D_INCONSISTENT;
  468. pdsk_max = D_INCONSISTENT;
  469. break;
  470. case C_STANDALONE:
  471. case C_DISCONNECTING:
  472. case C_UNCONNECTED:
  473. case C_TIMEOUT:
  474. case C_BROKEN_PIPE:
  475. case C_NETWORK_FAILURE:
  476. case C_PROTOCOL_ERROR:
  477. case C_TEAR_DOWN:
  478. case C_WF_CONNECTION:
  479. case C_WF_REPORT_PARAMS:
  480. case C_MASK:
  481. break;
  482. }
  483. if (ns.disk > disk_max)
  484. ns.disk = disk_max;
  485. if (ns.disk < disk_min) {
  486. dev_warn(DEV, "Implicitly set disk from %s to %s\n",
  487. drbd_disk_str(ns.disk), drbd_disk_str(disk_min));
  488. ns.disk = disk_min;
  489. }
  490. if (ns.pdsk > pdsk_max)
  491. ns.pdsk = pdsk_max;
  492. if (ns.pdsk < pdsk_min) {
  493. dev_warn(DEV, "Implicitly set pdsk from %s to %s\n",
  494. drbd_disk_str(ns.pdsk), drbd_disk_str(pdsk_min));
  495. ns.pdsk = pdsk_min;
  496. }
  497. if (fp == FP_STONITH &&
  498. (ns.role == R_PRIMARY && ns.conn < C_CONNECTED && ns.pdsk > D_OUTDATED) &&
  499. !(os.role == R_PRIMARY && os.conn < C_CONNECTED && os.pdsk > D_OUTDATED))
  500. ns.susp_fen = 1; /* Suspend IO while fence-peer handler runs (peer lost) */
  501. if (mdev->sync_conf.on_no_data == OND_SUSPEND_IO &&
  502. (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE) &&
  503. !(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE))
  504. ns.susp_nod = 1; /* Suspend IO while no data available (no accessible data available) */
  505. if (ns.aftr_isp || ns.peer_isp || ns.user_isp) {
  506. if (ns.conn == C_SYNC_SOURCE)
  507. ns.conn = C_PAUSED_SYNC_S;
  508. if (ns.conn == C_SYNC_TARGET)
  509. ns.conn = C_PAUSED_SYNC_T;
  510. } else {
  511. if (ns.conn == C_PAUSED_SYNC_S)
  512. ns.conn = C_SYNC_SOURCE;
  513. if (ns.conn == C_PAUSED_SYNC_T)
  514. ns.conn = C_SYNC_TARGET;
  515. }
  516. return ns;
  517. }
  518. void drbd_resume_al(struct drbd_conf *mdev)
  519. {
  520. if (test_and_clear_bit(AL_SUSPENDED, &mdev->flags))
  521. dev_info(DEV, "Resumed AL updates\n");
  522. }
  523. /* helper for __drbd_set_state */
  524. static void set_ov_position(struct drbd_conf *mdev, enum drbd_conns cs)
  525. {
  526. if (mdev->tconn->agreed_pro_version < 90)
  527. mdev->ov_start_sector = 0;
  528. mdev->rs_total = drbd_bm_bits(mdev);
  529. mdev->ov_position = 0;
  530. if (cs == C_VERIFY_T) {
  531. /* starting online verify from an arbitrary position
  532. * does not fit well into the existing protocol.
  533. * on C_VERIFY_T, we initialize ov_left and friends
  534. * implicitly in receive_DataRequest once the
  535. * first P_OV_REQUEST is received */
  536. mdev->ov_start_sector = ~(sector_t)0;
  537. } else {
  538. unsigned long bit = BM_SECT_TO_BIT(mdev->ov_start_sector);
  539. if (bit >= mdev->rs_total) {
  540. mdev->ov_start_sector =
  541. BM_BIT_TO_SECT(mdev->rs_total - 1);
  542. mdev->rs_total = 1;
  543. } else
  544. mdev->rs_total -= bit;
  545. mdev->ov_position = mdev->ov_start_sector;
  546. }
  547. mdev->ov_left = mdev->rs_total;
  548. }
  549. /**
  550. * __drbd_set_state() - Set a new DRBD state
  551. * @mdev: DRBD device.
  552. * @ns: new state.
  553. * @flags: Flags
  554. * @done: Optional completion, that will get completed after the after_state_ch() finished
  555. *
  556. * Caller needs to hold req_lock, and global_state_lock. Do not call directly.
  557. */
  558. enum drbd_state_rv
  559. __drbd_set_state(struct drbd_conf *mdev, union drbd_state ns,
  560. enum chg_state_flags flags, struct completion *done)
  561. {
  562. union drbd_state os;
  563. enum drbd_state_rv rv = SS_SUCCESS;
  564. const char *warn_sync_abort = NULL;
  565. struct after_state_chg_work *ascw;
  566. os = mdev->state;
  567. ns = sanitize_state(mdev, os, ns, &warn_sync_abort);
  568. if (ns.i == os.i)
  569. return SS_NOTHING_TO_DO;
  570. rv = is_valid_transition(os, ns);
  571. if (rv < SS_SUCCESS)
  572. return rv;
  573. if (!(flags & CS_HARD)) {
  574. /* pre-state-change checks ; only look at ns */
  575. /* See drbd_state_sw_errors in drbd_strings.c */
  576. rv = is_valid_state(mdev, ns);
  577. if (rv < SS_SUCCESS) {
  578. /* If the old state was illegal as well, then let
  579. this happen...*/
  580. if (is_valid_state(mdev, os) == rv)
  581. rv = is_valid_soft_transition(os, ns);
  582. } else
  583. rv = is_valid_soft_transition(os, ns);
  584. }
  585. if (rv < SS_SUCCESS) {
  586. if (flags & CS_VERBOSE)
  587. print_st_err(mdev, os, ns, rv);
  588. return rv;
  589. }
  590. if (warn_sync_abort)
  591. dev_warn(DEV, "%s aborted.\n", warn_sync_abort);
  592. {
  593. char *pbp, pb[300];
  594. pbp = pb;
  595. *pbp = 0;
  596. if (ns.role != os.role)
  597. pbp += sprintf(pbp, "role( %s -> %s ) ",
  598. drbd_role_str(os.role),
  599. drbd_role_str(ns.role));
  600. if (ns.peer != os.peer)
  601. pbp += sprintf(pbp, "peer( %s -> %s ) ",
  602. drbd_role_str(os.peer),
  603. drbd_role_str(ns.peer));
  604. if (ns.conn != os.conn)
  605. pbp += sprintf(pbp, "conn( %s -> %s ) ",
  606. drbd_conn_str(os.conn),
  607. drbd_conn_str(ns.conn));
  608. if (ns.disk != os.disk)
  609. pbp += sprintf(pbp, "disk( %s -> %s ) ",
  610. drbd_disk_str(os.disk),
  611. drbd_disk_str(ns.disk));
  612. if (ns.pdsk != os.pdsk)
  613. pbp += sprintf(pbp, "pdsk( %s -> %s ) ",
  614. drbd_disk_str(os.pdsk),
  615. drbd_disk_str(ns.pdsk));
  616. if (is_susp(ns) != is_susp(os))
  617. pbp += sprintf(pbp, "susp( %d -> %d ) ",
  618. is_susp(os),
  619. is_susp(ns));
  620. if (ns.aftr_isp != os.aftr_isp)
  621. pbp += sprintf(pbp, "aftr_isp( %d -> %d ) ",
  622. os.aftr_isp,
  623. ns.aftr_isp);
  624. if (ns.peer_isp != os.peer_isp)
  625. pbp += sprintf(pbp, "peer_isp( %d -> %d ) ",
  626. os.peer_isp,
  627. ns.peer_isp);
  628. if (ns.user_isp != os.user_isp)
  629. pbp += sprintf(pbp, "user_isp( %d -> %d ) ",
  630. os.user_isp,
  631. ns.user_isp);
  632. dev_info(DEV, "%s\n", pb);
  633. }
  634. /* solve the race between becoming unconfigured,
  635. * worker doing the cleanup, and
  636. * admin reconfiguring us:
  637. * on (re)configure, first set CONFIG_PENDING,
  638. * then wait for a potentially exiting worker,
  639. * start the worker, and schedule one no_op.
  640. * then proceed with configuration.
  641. */
  642. if (ns.disk == D_DISKLESS &&
  643. ns.conn == C_STANDALONE &&
  644. ns.role == R_SECONDARY &&
  645. !test_and_set_bit(CONFIG_PENDING, &mdev->flags))
  646. set_bit(DEVICE_DYING, &mdev->flags);
  647. /* if we are going -> D_FAILED or D_DISKLESS, grab one extra reference
  648. * on the ldev here, to be sure the transition -> D_DISKLESS resp.
  649. * drbd_ldev_destroy() won't happen before our corresponding
  650. * after_state_ch works run, where we put_ldev again. */
  651. if ((os.disk != D_FAILED && ns.disk == D_FAILED) ||
  652. (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))
  653. atomic_inc(&mdev->local_cnt);
  654. mdev->state = ns;
  655. if (os.disk == D_ATTACHING && ns.disk >= D_NEGOTIATING)
  656. drbd_print_uuids(mdev, "attached to UUIDs");
  657. wake_up(&mdev->misc_wait);
  658. wake_up(&mdev->state_wait);
  659. wake_up(&mdev->tconn->ping_wait);
  660. /* aborted verify run. log the last position */
  661. if ((os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) &&
  662. ns.conn < C_CONNECTED) {
  663. mdev->ov_start_sector =
  664. BM_BIT_TO_SECT(drbd_bm_bits(mdev) - mdev->ov_left);
  665. dev_info(DEV, "Online Verify reached sector %llu\n",
  666. (unsigned long long)mdev->ov_start_sector);
  667. }
  668. if ((os.conn == C_PAUSED_SYNC_T || os.conn == C_PAUSED_SYNC_S) &&
  669. (ns.conn == C_SYNC_TARGET || ns.conn == C_SYNC_SOURCE)) {
  670. dev_info(DEV, "Syncer continues.\n");
  671. mdev->rs_paused += (long)jiffies
  672. -(long)mdev->rs_mark_time[mdev->rs_last_mark];
  673. if (ns.conn == C_SYNC_TARGET)
  674. mod_timer(&mdev->resync_timer, jiffies);
  675. }
  676. if ((os.conn == C_SYNC_TARGET || os.conn == C_SYNC_SOURCE) &&
  677. (ns.conn == C_PAUSED_SYNC_T || ns.conn == C_PAUSED_SYNC_S)) {
  678. dev_info(DEV, "Resync suspended\n");
  679. mdev->rs_mark_time[mdev->rs_last_mark] = jiffies;
  680. }
  681. if (os.conn == C_CONNECTED &&
  682. (ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T)) {
  683. unsigned long now = jiffies;
  684. int i;
  685. set_ov_position(mdev, ns.conn);
  686. mdev->rs_start = now;
  687. mdev->rs_last_events = 0;
  688. mdev->rs_last_sect_ev = 0;
  689. mdev->ov_last_oos_size = 0;
  690. mdev->ov_last_oos_start = 0;
  691. for (i = 0; i < DRBD_SYNC_MARKS; i++) {
  692. mdev->rs_mark_left[i] = mdev->ov_left;
  693. mdev->rs_mark_time[i] = now;
  694. }
  695. drbd_rs_controller_reset(mdev);
  696. if (ns.conn == C_VERIFY_S) {
  697. dev_info(DEV, "Starting Online Verify from sector %llu\n",
  698. (unsigned long long)mdev->ov_position);
  699. mod_timer(&mdev->resync_timer, jiffies);
  700. }
  701. }
  702. if (get_ldev(mdev)) {
  703. u32 mdf = mdev->ldev->md.flags & ~(MDF_CONSISTENT|MDF_PRIMARY_IND|
  704. MDF_CONNECTED_IND|MDF_WAS_UP_TO_DATE|
  705. MDF_PEER_OUT_DATED|MDF_CRASHED_PRIMARY);
  706. if (test_bit(CRASHED_PRIMARY, &mdev->flags))
  707. mdf |= MDF_CRASHED_PRIMARY;
  708. if (mdev->state.role == R_PRIMARY ||
  709. (mdev->state.pdsk < D_INCONSISTENT && mdev->state.peer == R_PRIMARY))
  710. mdf |= MDF_PRIMARY_IND;
  711. if (mdev->state.conn > C_WF_REPORT_PARAMS)
  712. mdf |= MDF_CONNECTED_IND;
  713. if (mdev->state.disk > D_INCONSISTENT)
  714. mdf |= MDF_CONSISTENT;
  715. if (mdev->state.disk > D_OUTDATED)
  716. mdf |= MDF_WAS_UP_TO_DATE;
  717. if (mdev->state.pdsk <= D_OUTDATED && mdev->state.pdsk >= D_INCONSISTENT)
  718. mdf |= MDF_PEER_OUT_DATED;
  719. if (mdf != mdev->ldev->md.flags) {
  720. mdev->ldev->md.flags = mdf;
  721. drbd_md_mark_dirty(mdev);
  722. }
  723. if (os.disk < D_CONSISTENT && ns.disk >= D_CONSISTENT)
  724. drbd_set_ed_uuid(mdev, mdev->ldev->md.uuid[UI_CURRENT]);
  725. put_ldev(mdev);
  726. }
  727. /* Peer was forced D_UP_TO_DATE & R_PRIMARY, consider to resync */
  728. if (os.disk == D_INCONSISTENT && os.pdsk == D_INCONSISTENT &&
  729. os.peer == R_SECONDARY && ns.peer == R_PRIMARY)
  730. set_bit(CONSIDER_RESYNC, &mdev->flags);
  731. /* Receiver should clean up itself */
  732. if (os.conn != C_DISCONNECTING && ns.conn == C_DISCONNECTING)
  733. drbd_thread_stop_nowait(&mdev->tconn->receiver);
  734. /* Now the receiver finished cleaning up itself, it should die */
  735. if (os.conn != C_STANDALONE && ns.conn == C_STANDALONE)
  736. drbd_thread_stop_nowait(&mdev->tconn->receiver);
  737. /* Upon network failure, we need to restart the receiver. */
  738. if (os.conn > C_TEAR_DOWN &&
  739. ns.conn <= C_TEAR_DOWN && ns.conn >= C_TIMEOUT)
  740. drbd_thread_restart_nowait(&mdev->tconn->receiver);
  741. /* Resume AL writing if we get a connection */
  742. if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
  743. drbd_resume_al(mdev);
  744. ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
  745. if (ascw) {
  746. ascw->os = os;
  747. ascw->ns = ns;
  748. ascw->flags = flags;
  749. ascw->w.cb = w_after_state_ch;
  750. ascw->w.mdev = mdev;
  751. ascw->done = done;
  752. drbd_queue_work(&mdev->tconn->data.work, &ascw->w);
  753. } else {
  754. dev_warn(DEV, "Could not kmalloc an ascw\n");
  755. }
  756. return rv;
  757. }
  758. static int w_after_state_ch(struct drbd_work *w, int unused)
  759. {
  760. struct after_state_chg_work *ascw =
  761. container_of(w, struct after_state_chg_work, w);
  762. struct drbd_conf *mdev = w->mdev;
  763. after_state_ch(mdev, ascw->os, ascw->ns, ascw->flags);
  764. if (ascw->flags & CS_WAIT_COMPLETE) {
  765. D_ASSERT(ascw->done != NULL);
  766. complete(ascw->done);
  767. }
  768. kfree(ascw);
  769. return 1;
  770. }
  771. static void abw_start_sync(struct drbd_conf *mdev, int rv)
  772. {
  773. if (rv) {
  774. dev_err(DEV, "Writing the bitmap failed not starting resync.\n");
  775. _drbd_request_state(mdev, NS(conn, C_CONNECTED), CS_VERBOSE);
  776. return;
  777. }
  778. switch (mdev->state.conn) {
  779. case C_STARTING_SYNC_T:
  780. _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
  781. break;
  782. case C_STARTING_SYNC_S:
  783. drbd_start_resync(mdev, C_SYNC_SOURCE);
  784. break;
  785. }
  786. }
  787. int drbd_bitmap_io_from_worker(struct drbd_conf *mdev,
  788. int (*io_fn)(struct drbd_conf *),
  789. char *why, enum bm_flag flags)
  790. {
  791. int rv;
  792. D_ASSERT(current == mdev->tconn->worker.task);
  793. /* open coded non-blocking drbd_suspend_io(mdev); */
  794. set_bit(SUSPEND_IO, &mdev->flags);
  795. drbd_bm_lock(mdev, why, flags);
  796. rv = io_fn(mdev);
  797. drbd_bm_unlock(mdev);
  798. drbd_resume_io(mdev);
  799. return rv;
  800. }
  801. /**
  802. * after_state_ch() - Perform after state change actions that may sleep
  803. * @mdev: DRBD device.
  804. * @os: old state.
  805. * @ns: new state.
  806. * @flags: Flags
  807. */
  808. static void after_state_ch(struct drbd_conf *mdev, union drbd_state os,
  809. union drbd_state ns, enum chg_state_flags flags)
  810. {
  811. enum drbd_fencing_p fp;
  812. enum drbd_req_event what = NOTHING;
  813. union drbd_state nsm = (union drbd_state){ .i = -1 };
  814. if (os.conn != C_CONNECTED && ns.conn == C_CONNECTED) {
  815. clear_bit(CRASHED_PRIMARY, &mdev->flags);
  816. if (mdev->p_uuid)
  817. mdev->p_uuid[UI_FLAGS] &= ~((u64)2);
  818. }
  819. fp = FP_DONT_CARE;
  820. if (get_ldev(mdev)) {
  821. fp = mdev->ldev->dc.fencing;
  822. put_ldev(mdev);
  823. }
  824. /* Inform userspace about the change... */
  825. drbd_bcast_state(mdev, ns);
  826. if (!(os.role == R_PRIMARY && os.disk < D_UP_TO_DATE && os.pdsk < D_UP_TO_DATE) &&
  827. (ns.role == R_PRIMARY && ns.disk < D_UP_TO_DATE && ns.pdsk < D_UP_TO_DATE))
  828. drbd_khelper(mdev, "pri-on-incon-degr");
  829. /* Here we have the actions that are performed after a
  830. state change. This function might sleep */
  831. nsm.i = -1;
  832. if (ns.susp_nod) {
  833. if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
  834. what = RESEND;
  835. if (os.disk == D_ATTACHING && ns.disk > D_ATTACHING)
  836. what = RESTART_FROZEN_DISK_IO;
  837. if (what != NOTHING)
  838. nsm.susp_nod = 0;
  839. }
  840. if (ns.susp_fen) {
  841. /* case1: The outdate peer handler is successful: */
  842. if (os.pdsk > D_OUTDATED && ns.pdsk <= D_OUTDATED) {
  843. tl_clear(mdev);
  844. if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
  845. drbd_uuid_new_current(mdev);
  846. clear_bit(NEW_CUR_UUID, &mdev->flags);
  847. }
  848. spin_lock_irq(&mdev->tconn->req_lock);
  849. _drbd_set_state(_NS(mdev, susp_fen, 0), CS_VERBOSE, NULL);
  850. spin_unlock_irq(&mdev->tconn->req_lock);
  851. }
  852. /* case2: The connection was established again: */
  853. if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED) {
  854. clear_bit(NEW_CUR_UUID, &mdev->flags);
  855. what = RESEND;
  856. nsm.susp_fen = 0;
  857. }
  858. }
  859. if (what != NOTHING) {
  860. spin_lock_irq(&mdev->tconn->req_lock);
  861. _tl_restart(mdev, what);
  862. nsm.i &= mdev->state.i;
  863. _drbd_set_state(mdev, nsm, CS_VERBOSE, NULL);
  864. spin_unlock_irq(&mdev->tconn->req_lock);
  865. }
  866. /* Became sync source. With protocol >= 96, we still need to send out
  867. * the sync uuid now. Need to do that before any drbd_send_state, or
  868. * the other side may go "paused sync" before receiving the sync uuids,
  869. * which is unexpected. */
  870. if ((os.conn != C_SYNC_SOURCE && os.conn != C_PAUSED_SYNC_S) &&
  871. (ns.conn == C_SYNC_SOURCE || ns.conn == C_PAUSED_SYNC_S) &&
  872. mdev->tconn->agreed_pro_version >= 96 && get_ldev(mdev)) {
  873. drbd_gen_and_send_sync_uuid(mdev);
  874. put_ldev(mdev);
  875. }
  876. /* Do not change the order of the if above and the two below... */
  877. if (os.pdsk == D_DISKLESS && ns.pdsk > D_DISKLESS) { /* attach on the peer */
  878. drbd_send_uuids(mdev);
  879. drbd_send_state(mdev);
  880. }
  881. /* No point in queuing send_bitmap if we don't have a connection
  882. * anymore, so check also the _current_ state, not only the new state
  883. * at the time this work was queued. */
  884. if (os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S &&
  885. mdev->state.conn == C_WF_BITMAP_S)
  886. drbd_queue_bitmap_io(mdev, &drbd_send_bitmap, NULL,
  887. "send_bitmap (WFBitMapS)",
  888. BM_LOCKED_TEST_ALLOWED);
  889. /* Lost contact to peer's copy of the data */
  890. if ((os.pdsk >= D_INCONSISTENT &&
  891. os.pdsk != D_UNKNOWN &&
  892. os.pdsk != D_OUTDATED)
  893. && (ns.pdsk < D_INCONSISTENT ||
  894. ns.pdsk == D_UNKNOWN ||
  895. ns.pdsk == D_OUTDATED)) {
  896. if (get_ldev(mdev)) {
  897. if ((ns.role == R_PRIMARY || ns.peer == R_PRIMARY) &&
  898. mdev->ldev->md.uuid[UI_BITMAP] == 0 && ns.disk >= D_UP_TO_DATE) {
  899. if (is_susp(mdev->state)) {
  900. set_bit(NEW_CUR_UUID, &mdev->flags);
  901. } else {
  902. drbd_uuid_new_current(mdev);
  903. drbd_send_uuids(mdev);
  904. }
  905. }
  906. put_ldev(mdev);
  907. }
  908. }
  909. if (ns.pdsk < D_INCONSISTENT && get_ldev(mdev)) {
  910. if (ns.peer == R_PRIMARY && mdev->ldev->md.uuid[UI_BITMAP] == 0) {
  911. drbd_uuid_new_current(mdev);
  912. drbd_send_uuids(mdev);
  913. }
  914. /* D_DISKLESS Peer becomes secondary */
  915. if (os.peer == R_PRIMARY && ns.peer == R_SECONDARY)
  916. /* We may still be Primary ourselves.
  917. * No harm done if the bitmap still changes,
  918. * redirtied pages will follow later. */
  919. drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
  920. "demote diskless peer", BM_LOCKED_SET_ALLOWED);
  921. put_ldev(mdev);
  922. }
  923. /* Write out all changed bits on demote.
  924. * Though, no need to da that just yet
  925. * if there is a resync going on still */
  926. if (os.role == R_PRIMARY && ns.role == R_SECONDARY &&
  927. mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) {
  928. /* No changes to the bitmap expected this time, so assert that,
  929. * even though no harm was done if it did change. */
  930. drbd_bitmap_io_from_worker(mdev, &drbd_bm_write,
  931. "demote", BM_LOCKED_TEST_ALLOWED);
  932. put_ldev(mdev);
  933. }
  934. /* Last part of the attaching process ... */
  935. if (ns.conn >= C_CONNECTED &&
  936. os.disk == D_ATTACHING && ns.disk == D_NEGOTIATING) {
  937. drbd_send_sizes(mdev, 0, 0); /* to start sync... */
  938. drbd_send_uuids(mdev);
  939. drbd_send_state(mdev);
  940. }
  941. /* We want to pause/continue resync, tell peer. */
  942. if (ns.conn >= C_CONNECTED &&
  943. ((os.aftr_isp != ns.aftr_isp) ||
  944. (os.user_isp != ns.user_isp)))
  945. drbd_send_state(mdev);
  946. /* In case one of the isp bits got set, suspend other devices. */
  947. if ((!os.aftr_isp && !os.peer_isp && !os.user_isp) &&
  948. (ns.aftr_isp || ns.peer_isp || ns.user_isp))
  949. suspend_other_sg(mdev);
  950. /* Make sure the peer gets informed about eventual state
  951. changes (ISP bits) while we were in WFReportParams. */
  952. if (os.conn == C_WF_REPORT_PARAMS && ns.conn >= C_CONNECTED)
  953. drbd_send_state(mdev);
  954. if (os.conn != C_AHEAD && ns.conn == C_AHEAD)
  955. drbd_send_state(mdev);
  956. /* We are in the progress to start a full sync... */
  957. if ((os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
  958. (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S))
  959. /* no other bitmap changes expected during this phase */
  960. drbd_queue_bitmap_io(mdev,
  961. &drbd_bmio_set_n_write, &abw_start_sync,
  962. "set_n_write from StartingSync", BM_LOCKED_TEST_ALLOWED);
  963. /* We are invalidating our self... */
  964. if (os.conn < C_CONNECTED && ns.conn < C_CONNECTED &&
  965. os.disk > D_INCONSISTENT && ns.disk == D_INCONSISTENT)
  966. /* other bitmap operation expected during this phase */
  967. drbd_queue_bitmap_io(mdev, &drbd_bmio_set_n_write, NULL,
  968. "set_n_write from invalidate", BM_LOCKED_MASK);
  969. /* first half of local IO error, failure to attach,
  970. * or administrative detach */
  971. if (os.disk != D_FAILED && ns.disk == D_FAILED) {
  972. enum drbd_io_error_p eh;
  973. int was_io_error;
  974. /* corresponding get_ldev was in __drbd_set_state, to serialize
  975. * our cleanup here with the transition to D_DISKLESS,
  976. * so it is safe to dreference ldev here. */
  977. eh = mdev->ldev->dc.on_io_error;
  978. was_io_error = test_and_clear_bit(WAS_IO_ERROR, &mdev->flags);
  979. /* current state still has to be D_FAILED,
  980. * there is only one way out: to D_DISKLESS,
  981. * and that may only happen after our put_ldev below. */
  982. if (mdev->state.disk != D_FAILED)
  983. dev_err(DEV,
  984. "ASSERT FAILED: disk is %s during detach\n",
  985. drbd_disk_str(mdev->state.disk));
  986. if (drbd_send_state(mdev))
  987. dev_warn(DEV, "Notified peer that I am detaching my disk\n");
  988. else
  989. dev_err(DEV, "Sending state for detaching disk failed\n");
  990. drbd_rs_cancel_all(mdev);
  991. /* In case we want to get something to stable storage still,
  992. * this may be the last chance.
  993. * Following put_ldev may transition to D_DISKLESS. */
  994. drbd_md_sync(mdev);
  995. put_ldev(mdev);
  996. if (was_io_error && eh == EP_CALL_HELPER)
  997. drbd_khelper(mdev, "local-io-error");
  998. }
  999. /* second half of local IO error, failure to attach,
  1000. * or administrative detach,
  1001. * after local_cnt references have reached zero again */
  1002. if (os.disk != D_DISKLESS && ns.disk == D_DISKLESS) {
  1003. /* We must still be diskless,
  1004. * re-attach has to be serialized with this! */
  1005. if (mdev->state.disk != D_DISKLESS)
  1006. dev_err(DEV,
  1007. "ASSERT FAILED: disk is %s while going diskless\n",
  1008. drbd_disk_str(mdev->state.disk));
  1009. mdev->rs_total = 0;
  1010. mdev->rs_failed = 0;
  1011. atomic_set(&mdev->rs_pending_cnt, 0);
  1012. if (drbd_send_state(mdev))
  1013. dev_warn(DEV, "Notified peer that I'm now diskless.\n");
  1014. /* corresponding get_ldev in __drbd_set_state
  1015. * this may finally trigger drbd_ldev_destroy. */
  1016. put_ldev(mdev);
  1017. }
  1018. /* Notify peer that I had a local IO error, and did not detached.. */
  1019. if (os.disk == D_UP_TO_DATE && ns.disk == D_INCONSISTENT)
  1020. drbd_send_state(mdev);
  1021. /* Disks got bigger while they were detached */
  1022. if (ns.disk > D_NEGOTIATING && ns.pdsk > D_NEGOTIATING &&
  1023. test_and_clear_bit(RESYNC_AFTER_NEG, &mdev->flags)) {
  1024. if (ns.conn == C_CONNECTED)
  1025. resync_after_online_grow(mdev);
  1026. }
  1027. /* A resync finished or aborted, wake paused devices... */
  1028. if ((os.conn > C_CONNECTED && ns.conn <= C_CONNECTED) ||
  1029. (os.peer_isp && !ns.peer_isp) ||
  1030. (os.user_isp && !ns.user_isp))
  1031. resume_next_sg(mdev);
  1032. /* sync target done with resync. Explicitly notify peer, even though
  1033. * it should (at least for non-empty resyncs) already know itself. */
  1034. if (os.disk < D_UP_TO_DATE && os.conn >= C_SYNC_SOURCE && ns.conn == C_CONNECTED)
  1035. drbd_send_state(mdev);
  1036. /* This triggers bitmap writeout of potentially still unwritten pages
  1037. * if the resync finished cleanly, or aborted because of peer disk
  1038. * failure, or because of connection loss.
  1039. * For resync aborted because of local disk failure, we cannot do
  1040. * any bitmap writeout anymore.
  1041. * No harm done if some bits change during this phase.
  1042. */
  1043. if (os.conn > C_CONNECTED && ns.conn <= C_CONNECTED && get_ldev(mdev)) {
  1044. drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL,
  1045. "write from resync_finished", BM_LOCKED_SET_ALLOWED);
  1046. put_ldev(mdev);
  1047. }
  1048. if (ns.disk == D_DISKLESS &&
  1049. ns.conn == C_STANDALONE &&
  1050. ns.role == R_SECONDARY) {
  1051. if (os.aftr_isp != ns.aftr_isp)
  1052. resume_next_sg(mdev);
  1053. }
  1054. after_conn_state_ch(mdev->tconn, os, ns, flags);
  1055. drbd_md_sync(mdev);
  1056. }
  1057. static void after_conn_state_ch(struct drbd_tconn *tconn, union drbd_state os,
  1058. union drbd_state ns, enum chg_state_flags flags)
  1059. {
  1060. /* Upon network configuration, we need to start the receiver */
  1061. if (os.conn == C_STANDALONE && ns.conn == C_UNCONNECTED)
  1062. drbd_thread_start(&tconn->receiver);
  1063. if (ns.disk == D_DISKLESS &&
  1064. ns.conn == C_STANDALONE &&
  1065. ns.role == R_SECONDARY) {
  1066. /* if (test_bit(DEVICE_DYING, &mdev->flags)) TODO: DEVICE_DYING functionality */
  1067. drbd_thread_stop_nowait(&tconn->worker);
  1068. }
  1069. }