xfs_log_cil.c 35 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130
  1. /*
  2. * Copyright (c) 2010 Red Hat, Inc. All Rights Reserved.
  3. *
  4. * This program is free software; you can redistribute it and/or
  5. * modify it under the terms of the GNU General Public License as
  6. * published by the Free Software Foundation.
  7. *
  8. * This program is distributed in the hope that it would be useful,
  9. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  11. * GNU General Public License for more details.
  12. *
  13. * You should have received a copy of the GNU General Public License
  14. * along with this program; if not, write the Free Software Foundation,
  15. * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  16. */
  17. #include "xfs.h"
  18. #include "xfs_fs.h"
  19. #include "xfs_format.h"
  20. #include "xfs_log_format.h"
  21. #include "xfs_shared.h"
  22. #include "xfs_trans_resv.h"
  23. #include "xfs_mount.h"
  24. #include "xfs_error.h"
  25. #include "xfs_alloc.h"
  26. #include "xfs_extent_busy.h"
  27. #include "xfs_discard.h"
  28. #include "xfs_trans.h"
  29. #include "xfs_trans_priv.h"
  30. #include "xfs_log.h"
  31. #include "xfs_log_priv.h"
  32. /*
  33. * Allocate a new ticket. Failing to get a new ticket makes it really hard to
  34. * recover, so we don't allow failure here. Also, we allocate in a context that
  35. * we don't want to be issuing transactions from, so we need to tell the
  36. * allocation code this as well.
  37. *
  38. * We don't reserve any space for the ticket - we are going to steal whatever
  39. * space we require from transactions as they commit. To ensure we reserve all
  40. * the space required, we need to set the current reservation of the ticket to
  41. * zero so that we know to steal the initial transaction overhead from the
  42. * first transaction commit.
  43. */
  44. static struct xlog_ticket *
  45. xlog_cil_ticket_alloc(
  46. struct xlog *log)
  47. {
  48. struct xlog_ticket *tic;
  49. tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0,
  50. KM_SLEEP|KM_NOFS);
  51. /*
  52. * set the current reservation to zero so we know to steal the basic
  53. * transaction overhead reservation from the first transaction commit.
  54. */
  55. tic->t_curr_res = 0;
  56. return tic;
  57. }
  58. /*
  59. * After the first stage of log recovery is done, we know where the head and
  60. * tail of the log are. We need this log initialisation done before we can
  61. * initialise the first CIL checkpoint context.
  62. *
  63. * Here we allocate a log ticket to track space usage during a CIL push. This
  64. * ticket is passed to xlog_write() directly so that we don't slowly leak log
  65. * space by failing to account for space used by log headers and additional
  66. * region headers for split regions.
  67. */
  68. void
  69. xlog_cil_init_post_recovery(
  70. struct xlog *log)
  71. {
  72. log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log);
  73. log->l_cilp->xc_ctx->sequence = 1;
  74. }
  75. static inline int
  76. xlog_cil_iovec_space(
  77. uint niovecs)
  78. {
  79. return round_up((sizeof(struct xfs_log_vec) +
  80. niovecs * sizeof(struct xfs_log_iovec)),
  81. sizeof(uint64_t));
  82. }
  83. /*
  84. * Allocate or pin log vector buffers for CIL insertion.
  85. *
  86. * The CIL currently uses disposable buffers for copying a snapshot of the
  87. * modified items into the log during a push. The biggest problem with this is
  88. * the requirement to allocate the disposable buffer during the commit if:
  89. * a) does not exist; or
  90. * b) it is too small
  91. *
  92. * If we do this allocation within xlog_cil_insert_format_items(), it is done
  93. * under the xc_ctx_lock, which means that a CIL push cannot occur during
  94. * the memory allocation. This means that we have a potential deadlock situation
  95. * under low memory conditions when we have lots of dirty metadata pinned in
  96. * the CIL and we need a CIL commit to occur to free memory.
  97. *
  98. * To avoid this, we need to move the memory allocation outside the
  99. * xc_ctx_lock, but because the log vector buffers are disposable, that opens
  100. * up a TOCTOU race condition w.r.t. the CIL committing and removing the log
  101. * vector buffers between the check and the formatting of the item into the
  102. * log vector buffer within the xc_ctx_lock.
  103. *
  104. * Because the log vector buffer needs to be unchanged during the CIL push
  105. * process, we cannot share the buffer between the transaction commit (which
  106. * modifies the buffer) and the CIL push context that is writing the changes
  107. * into the log. This means skipping preallocation of buffer space is
  108. * unreliable, but we most definitely do not want to be allocating and freeing
  109. * buffers unnecessarily during commits when overwrites can be done safely.
  110. *
  111. * The simplest solution to this problem is to allocate a shadow buffer when a
  112. * log item is committed for the second time, and then to only use this buffer
  113. * if necessary. The buffer can remain attached to the log item until such time
  114. * it is needed, and this is the buffer that is reallocated to match the size of
  115. * the incoming modification. Then during the formatting of the item we can swap
  116. * the active buffer with the new one if we can't reuse the existing buffer. We
  117. * don't free the old buffer as it may be reused on the next modification if
  118. * it's size is right, otherwise we'll free and reallocate it at that point.
  119. *
  120. * This function builds a vector for the changes in each log item in the
  121. * transaction. It then works out the length of the buffer needed for each log
  122. * item, allocates them and attaches the vector to the log item in preparation
  123. * for the formatting step which occurs under the xc_ctx_lock.
  124. *
  125. * While this means the memory footprint goes up, it avoids the repeated
  126. * alloc/free pattern that repeated modifications of an item would otherwise
  127. * cause, and hence minimises the CPU overhead of such behaviour.
  128. */
  129. static void
  130. xlog_cil_alloc_shadow_bufs(
  131. struct xlog *log,
  132. struct xfs_trans *tp)
  133. {
  134. struct xfs_log_item_desc *lidp;
  135. list_for_each_entry(lidp, &tp->t_items, lid_trans) {
  136. struct xfs_log_item *lip = lidp->lid_item;
  137. struct xfs_log_vec *lv;
  138. int niovecs = 0;
  139. int nbytes = 0;
  140. int buf_size;
  141. bool ordered = false;
  142. /* Skip items which aren't dirty in this transaction. */
  143. if (!(lidp->lid_flags & XFS_LID_DIRTY))
  144. continue;
  145. /* get number of vecs and size of data to be stored */
  146. lip->li_ops->iop_size(lip, &niovecs, &nbytes);
  147. /*
  148. * Ordered items need to be tracked but we do not wish to write
  149. * them. We need a logvec to track the object, but we do not
  150. * need an iovec or buffer to be allocated for copying data.
  151. */
  152. if (niovecs == XFS_LOG_VEC_ORDERED) {
  153. ordered = true;
  154. niovecs = 0;
  155. nbytes = 0;
  156. }
  157. /*
  158. * We 64-bit align the length of each iovec so that the start
  159. * of the next one is naturally aligned. We'll need to
  160. * account for that slack space here. Then round nbytes up
  161. * to 64-bit alignment so that the initial buffer alignment is
  162. * easy to calculate and verify.
  163. */
  164. nbytes += niovecs * sizeof(uint64_t);
  165. nbytes = round_up(nbytes, sizeof(uint64_t));
  166. /*
  167. * The data buffer needs to start 64-bit aligned, so round up
  168. * that space to ensure we can align it appropriately and not
  169. * overrun the buffer.
  170. */
  171. buf_size = nbytes + xlog_cil_iovec_space(niovecs);
  172. /*
  173. * if we have no shadow buffer, or it is too small, we need to
  174. * reallocate it.
  175. */
  176. if (!lip->li_lv_shadow ||
  177. buf_size > lip->li_lv_shadow->lv_size) {
  178. /*
  179. * We free and allocate here as a realloc would copy
  180. * unecessary data. We don't use kmem_zalloc() for the
  181. * same reason - we don't need to zero the data area in
  182. * the buffer, only the log vector header and the iovec
  183. * storage.
  184. */
  185. kmem_free(lip->li_lv_shadow);
  186. lv = kmem_alloc(buf_size, KM_SLEEP|KM_NOFS);
  187. memset(lv, 0, xlog_cil_iovec_space(niovecs));
  188. lv->lv_item = lip;
  189. lv->lv_size = buf_size;
  190. if (ordered)
  191. lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
  192. else
  193. lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1];
  194. lip->li_lv_shadow = lv;
  195. } else {
  196. /* same or smaller, optimise common overwrite case */
  197. lv = lip->li_lv_shadow;
  198. if (ordered)
  199. lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
  200. else
  201. lv->lv_buf_len = 0;
  202. lv->lv_bytes = 0;
  203. lv->lv_next = NULL;
  204. }
  205. /* Ensure the lv is set up according to ->iop_size */
  206. lv->lv_niovecs = niovecs;
  207. /* The allocated data region lies beyond the iovec region */
  208. lv->lv_buf = (char *)lv + xlog_cil_iovec_space(niovecs);
  209. }
  210. }
  211. /*
  212. * Prepare the log item for insertion into the CIL. Calculate the difference in
  213. * log space and vectors it will consume, and if it is a new item pin it as
  214. * well.
  215. */
  216. STATIC void
  217. xfs_cil_prepare_item(
  218. struct xlog *log,
  219. struct xfs_log_vec *lv,
  220. struct xfs_log_vec *old_lv,
  221. int *diff_len,
  222. int *diff_iovecs)
  223. {
  224. /* Account for the new LV being passed in */
  225. if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) {
  226. *diff_len += lv->lv_bytes;
  227. *diff_iovecs += lv->lv_niovecs;
  228. }
  229. /*
  230. * If there is no old LV, this is the first time we've seen the item in
  231. * this CIL context and so we need to pin it. If we are replacing the
  232. * old_lv, then remove the space it accounts for and make it the shadow
  233. * buffer for later freeing. In both cases we are now switching to the
  234. * shadow buffer, so update the the pointer to it appropriately.
  235. */
  236. if (!old_lv) {
  237. lv->lv_item->li_ops->iop_pin(lv->lv_item);
  238. lv->lv_item->li_lv_shadow = NULL;
  239. } else if (old_lv != lv) {
  240. ASSERT(lv->lv_buf_len != XFS_LOG_VEC_ORDERED);
  241. *diff_len -= old_lv->lv_bytes;
  242. *diff_iovecs -= old_lv->lv_niovecs;
  243. lv->lv_item->li_lv_shadow = old_lv;
  244. }
  245. /* attach new log vector to log item */
  246. lv->lv_item->li_lv = lv;
  247. /*
  248. * If this is the first time the item is being committed to the
  249. * CIL, store the sequence number on the log item so we can
  250. * tell in future commits whether this is the first checkpoint
  251. * the item is being committed into.
  252. */
  253. if (!lv->lv_item->li_seq)
  254. lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence;
  255. }
  256. /*
  257. * Format log item into a flat buffers
  258. *
  259. * For delayed logging, we need to hold a formatted buffer containing all the
  260. * changes on the log item. This enables us to relog the item in memory and
  261. * write it out asynchronously without needing to relock the object that was
  262. * modified at the time it gets written into the iclog.
  263. *
  264. * This function takes the prepared log vectors attached to each log item, and
  265. * formats the changes into the log vector buffer. The buffer it uses is
  266. * dependent on the current state of the vector in the CIL - the shadow lv is
  267. * guaranteed to be large enough for the current modification, but we will only
  268. * use that if we can't reuse the existing lv. If we can't reuse the existing
  269. * lv, then simple swap it out for the shadow lv. We don't free it - that is
  270. * done lazily either by th enext modification or the freeing of the log item.
  271. *
  272. * We don't set up region headers during this process; we simply copy the
  273. * regions into the flat buffer. We can do this because we still have to do a
  274. * formatting step to write the regions into the iclog buffer. Writing the
  275. * ophdrs during the iclog write means that we can support splitting large
  276. * regions across iclog boundares without needing a change in the format of the
  277. * item/region encapsulation.
  278. *
  279. * Hence what we need to do now is change the rewrite the vector array to point
  280. * to the copied region inside the buffer we just allocated. This allows us to
  281. * format the regions into the iclog as though they are being formatted
  282. * directly out of the objects themselves.
  283. */
  284. static void
  285. xlog_cil_insert_format_items(
  286. struct xlog *log,
  287. struct xfs_trans *tp,
  288. int *diff_len,
  289. int *diff_iovecs)
  290. {
  291. struct xfs_log_item_desc *lidp;
  292. /* Bail out if we didn't find a log item. */
  293. if (list_empty(&tp->t_items)) {
  294. ASSERT(0);
  295. return;
  296. }
  297. list_for_each_entry(lidp, &tp->t_items, lid_trans) {
  298. struct xfs_log_item *lip = lidp->lid_item;
  299. struct xfs_log_vec *lv;
  300. struct xfs_log_vec *old_lv = NULL;
  301. struct xfs_log_vec *shadow;
  302. bool ordered = false;
  303. /* Skip items which aren't dirty in this transaction. */
  304. if (!(lidp->lid_flags & XFS_LID_DIRTY))
  305. continue;
  306. /*
  307. * The formatting size information is already attached to
  308. * the shadow lv on the log item.
  309. */
  310. shadow = lip->li_lv_shadow;
  311. if (shadow->lv_buf_len == XFS_LOG_VEC_ORDERED)
  312. ordered = true;
  313. /* Skip items that do not have any vectors for writing */
  314. if (!shadow->lv_niovecs && !ordered)
  315. continue;
  316. /* compare to existing item size */
  317. old_lv = lip->li_lv;
  318. if (lip->li_lv && shadow->lv_size <= lip->li_lv->lv_size) {
  319. /* same or smaller, optimise common overwrite case */
  320. lv = lip->li_lv;
  321. lv->lv_next = NULL;
  322. if (ordered)
  323. goto insert;
  324. /*
  325. * set the item up as though it is a new insertion so
  326. * that the space reservation accounting is correct.
  327. */
  328. *diff_iovecs -= lv->lv_niovecs;
  329. *diff_len -= lv->lv_bytes;
  330. /* Ensure the lv is set up according to ->iop_size */
  331. lv->lv_niovecs = shadow->lv_niovecs;
  332. /* reset the lv buffer information for new formatting */
  333. lv->lv_buf_len = 0;
  334. lv->lv_bytes = 0;
  335. lv->lv_buf = (char *)lv +
  336. xlog_cil_iovec_space(lv->lv_niovecs);
  337. } else {
  338. /* switch to shadow buffer! */
  339. lv = shadow;
  340. lv->lv_item = lip;
  341. if (ordered) {
  342. /* track as an ordered logvec */
  343. ASSERT(lip->li_lv == NULL);
  344. goto insert;
  345. }
  346. }
  347. ASSERT(IS_ALIGNED((unsigned long)lv->lv_buf, sizeof(uint64_t)));
  348. lip->li_ops->iop_format(lip, lv);
  349. insert:
  350. xfs_cil_prepare_item(log, lv, old_lv, diff_len, diff_iovecs);
  351. }
  352. }
  353. /*
  354. * Insert the log items into the CIL and calculate the difference in space
  355. * consumed by the item. Add the space to the checkpoint ticket and calculate
  356. * if the change requires additional log metadata. If it does, take that space
  357. * as well. Remove the amount of space we added to the checkpoint ticket from
  358. * the current transaction ticket so that the accounting works out correctly.
  359. */
  360. static void
  361. xlog_cil_insert_items(
  362. struct xlog *log,
  363. struct xfs_trans *tp)
  364. {
  365. struct xfs_cil *cil = log->l_cilp;
  366. struct xfs_cil_ctx *ctx = cil->xc_ctx;
  367. struct xfs_log_item_desc *lidp;
  368. int len = 0;
  369. int diff_iovecs = 0;
  370. int iclog_space;
  371. ASSERT(tp);
  372. /*
  373. * We can do this safely because the context can't checkpoint until we
  374. * are done so it doesn't matter exactly how we update the CIL.
  375. */
  376. xlog_cil_insert_format_items(log, tp, &len, &diff_iovecs);
  377. /*
  378. * Now (re-)position everything modified at the tail of the CIL.
  379. * We do this here so we only need to take the CIL lock once during
  380. * the transaction commit.
  381. */
  382. spin_lock(&cil->xc_cil_lock);
  383. list_for_each_entry(lidp, &tp->t_items, lid_trans) {
  384. struct xfs_log_item *lip = lidp->lid_item;
  385. /* Skip items which aren't dirty in this transaction. */
  386. if (!(lidp->lid_flags & XFS_LID_DIRTY))
  387. continue;
  388. /*
  389. * Only move the item if it isn't already at the tail. This is
  390. * to prevent a transient list_empty() state when reinserting
  391. * an item that is already the only item in the CIL.
  392. */
  393. if (!list_is_last(&lip->li_cil, &cil->xc_cil))
  394. list_move_tail(&lip->li_cil, &cil->xc_cil);
  395. }
  396. /* account for space used by new iovec headers */
  397. len += diff_iovecs * sizeof(xlog_op_header_t);
  398. ctx->nvecs += diff_iovecs;
  399. /* attach the transaction to the CIL if it has any busy extents */
  400. if (!list_empty(&tp->t_busy))
  401. list_splice_init(&tp->t_busy, &ctx->busy_extents);
  402. /*
  403. * Now transfer enough transaction reservation to the context ticket
  404. * for the checkpoint. The context ticket is special - the unit
  405. * reservation has to grow as well as the current reservation as we
  406. * steal from tickets so we can correctly determine the space used
  407. * during the transaction commit.
  408. */
  409. if (ctx->ticket->t_curr_res == 0) {
  410. ctx->ticket->t_curr_res = ctx->ticket->t_unit_res;
  411. tp->t_ticket->t_curr_res -= ctx->ticket->t_unit_res;
  412. }
  413. /* do we need space for more log record headers? */
  414. iclog_space = log->l_iclog_size - log->l_iclog_hsize;
  415. if (len > 0 && (ctx->space_used / iclog_space !=
  416. (ctx->space_used + len) / iclog_space)) {
  417. int hdrs;
  418. hdrs = (len + iclog_space - 1) / iclog_space;
  419. /* need to take into account split region headers, too */
  420. hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
  421. ctx->ticket->t_unit_res += hdrs;
  422. ctx->ticket->t_curr_res += hdrs;
  423. tp->t_ticket->t_curr_res -= hdrs;
  424. ASSERT(tp->t_ticket->t_curr_res >= len);
  425. }
  426. tp->t_ticket->t_curr_res -= len;
  427. ctx->space_used += len;
  428. spin_unlock(&cil->xc_cil_lock);
  429. }
  430. static void
  431. xlog_cil_free_logvec(
  432. struct xfs_log_vec *log_vector)
  433. {
  434. struct xfs_log_vec *lv;
  435. for (lv = log_vector; lv; ) {
  436. struct xfs_log_vec *next = lv->lv_next;
  437. kmem_free(lv);
  438. lv = next;
  439. }
  440. }
  441. /*
  442. * Mark all items committed and clear busy extents. We free the log vector
  443. * chains in a separate pass so that we unpin the log items as quickly as
  444. * possible.
  445. */
  446. static void
  447. xlog_cil_committed(
  448. void *args,
  449. int abort)
  450. {
  451. struct xfs_cil_ctx *ctx = args;
  452. struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
  453. xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain,
  454. ctx->start_lsn, abort);
  455. xfs_extent_busy_sort(&ctx->busy_extents);
  456. xfs_extent_busy_clear(mp, &ctx->busy_extents,
  457. (mp->m_flags & XFS_MOUNT_DISCARD) && !abort);
  458. /*
  459. * If we are aborting the commit, wake up anyone waiting on the
  460. * committing list. If we don't, then a shutdown we can leave processes
  461. * waiting in xlog_cil_force_lsn() waiting on a sequence commit that
  462. * will never happen because we aborted it.
  463. */
  464. spin_lock(&ctx->cil->xc_push_lock);
  465. if (abort)
  466. wake_up_all(&ctx->cil->xc_commit_wait);
  467. list_del(&ctx->committing);
  468. spin_unlock(&ctx->cil->xc_push_lock);
  469. xlog_cil_free_logvec(ctx->lv_chain);
  470. if (!list_empty(&ctx->busy_extents)) {
  471. ASSERT(mp->m_flags & XFS_MOUNT_DISCARD);
  472. xfs_discard_extents(mp, &ctx->busy_extents);
  473. xfs_extent_busy_clear(mp, &ctx->busy_extents, false);
  474. }
  475. kmem_free(ctx);
  476. }
  477. /*
  478. * Push the Committed Item List to the log. If @push_seq flag is zero, then it
  479. * is a background flush and so we can chose to ignore it. Otherwise, if the
  480. * current sequence is the same as @push_seq we need to do a flush. If
  481. * @push_seq is less than the current sequence, then it has already been
  482. * flushed and we don't need to do anything - the caller will wait for it to
  483. * complete if necessary.
  484. *
  485. * @push_seq is a value rather than a flag because that allows us to do an
  486. * unlocked check of the sequence number for a match. Hence we can allows log
  487. * forces to run racily and not issue pushes for the same sequence twice. If we
  488. * get a race between multiple pushes for the same sequence they will block on
  489. * the first one and then abort, hence avoiding needless pushes.
  490. */
  491. STATIC int
  492. xlog_cil_push(
  493. struct xlog *log)
  494. {
  495. struct xfs_cil *cil = log->l_cilp;
  496. struct xfs_log_vec *lv;
  497. struct xfs_cil_ctx *ctx;
  498. struct xfs_cil_ctx *new_ctx;
  499. struct xlog_in_core *commit_iclog;
  500. struct xlog_ticket *tic;
  501. int num_iovecs;
  502. int error = 0;
  503. struct xfs_trans_header thdr;
  504. struct xfs_log_iovec lhdr;
  505. struct xfs_log_vec lvhdr = { NULL };
  506. xfs_lsn_t commit_lsn;
  507. xfs_lsn_t push_seq;
  508. if (!cil)
  509. return 0;
  510. new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS);
  511. new_ctx->ticket = xlog_cil_ticket_alloc(log);
  512. down_write(&cil->xc_ctx_lock);
  513. ctx = cil->xc_ctx;
  514. spin_lock(&cil->xc_push_lock);
  515. push_seq = cil->xc_push_seq;
  516. ASSERT(push_seq <= ctx->sequence);
  517. /*
  518. * Check if we've anything to push. If there is nothing, then we don't
  519. * move on to a new sequence number and so we have to be able to push
  520. * this sequence again later.
  521. */
  522. if (list_empty(&cil->xc_cil)) {
  523. cil->xc_push_seq = 0;
  524. spin_unlock(&cil->xc_push_lock);
  525. goto out_skip;
  526. }
  527. /* check for a previously pushed seqeunce */
  528. if (push_seq < cil->xc_ctx->sequence) {
  529. spin_unlock(&cil->xc_push_lock);
  530. goto out_skip;
  531. }
  532. /*
  533. * We are now going to push this context, so add it to the committing
  534. * list before we do anything else. This ensures that anyone waiting on
  535. * this push can easily detect the difference between a "push in
  536. * progress" and "CIL is empty, nothing to do".
  537. *
  538. * IOWs, a wait loop can now check for:
  539. * the current sequence not being found on the committing list;
  540. * an empty CIL; and
  541. * an unchanged sequence number
  542. * to detect a push that had nothing to do and therefore does not need
  543. * waiting on. If the CIL is not empty, we get put on the committing
  544. * list before emptying the CIL and bumping the sequence number. Hence
  545. * an empty CIL and an unchanged sequence number means we jumped out
  546. * above after doing nothing.
  547. *
  548. * Hence the waiter will either find the commit sequence on the
  549. * committing list or the sequence number will be unchanged and the CIL
  550. * still dirty. In that latter case, the push has not yet started, and
  551. * so the waiter will have to continue trying to check the CIL
  552. * committing list until it is found. In extreme cases of delay, the
  553. * sequence may fully commit between the attempts the wait makes to wait
  554. * on the commit sequence.
  555. */
  556. list_add(&ctx->committing, &cil->xc_committing);
  557. spin_unlock(&cil->xc_push_lock);
  558. /*
  559. * pull all the log vectors off the items in the CIL, and
  560. * remove the items from the CIL. We don't need the CIL lock
  561. * here because it's only needed on the transaction commit
  562. * side which is currently locked out by the flush lock.
  563. */
  564. lv = NULL;
  565. num_iovecs = 0;
  566. while (!list_empty(&cil->xc_cil)) {
  567. struct xfs_log_item *item;
  568. item = list_first_entry(&cil->xc_cil,
  569. struct xfs_log_item, li_cil);
  570. list_del_init(&item->li_cil);
  571. if (!ctx->lv_chain)
  572. ctx->lv_chain = item->li_lv;
  573. else
  574. lv->lv_next = item->li_lv;
  575. lv = item->li_lv;
  576. item->li_lv = NULL;
  577. num_iovecs += lv->lv_niovecs;
  578. }
  579. /*
  580. * initialise the new context and attach it to the CIL. Then attach
  581. * the current context to the CIL committing lsit so it can be found
  582. * during log forces to extract the commit lsn of the sequence that
  583. * needs to be forced.
  584. */
  585. INIT_LIST_HEAD(&new_ctx->committing);
  586. INIT_LIST_HEAD(&new_ctx->busy_extents);
  587. new_ctx->sequence = ctx->sequence + 1;
  588. new_ctx->cil = cil;
  589. cil->xc_ctx = new_ctx;
  590. /*
  591. * The switch is now done, so we can drop the context lock and move out
  592. * of a shared context. We can't just go straight to the commit record,
  593. * though - we need to synchronise with previous and future commits so
  594. * that the commit records are correctly ordered in the log to ensure
  595. * that we process items during log IO completion in the correct order.
  596. *
  597. * For example, if we get an EFI in one checkpoint and the EFD in the
  598. * next (e.g. due to log forces), we do not want the checkpoint with
  599. * the EFD to be committed before the checkpoint with the EFI. Hence
  600. * we must strictly order the commit records of the checkpoints so
  601. * that: a) the checkpoint callbacks are attached to the iclogs in the
  602. * correct order; and b) the checkpoints are replayed in correct order
  603. * in log recovery.
  604. *
  605. * Hence we need to add this context to the committing context list so
  606. * that higher sequences will wait for us to write out a commit record
  607. * before they do.
  608. *
  609. * xfs_log_force_lsn requires us to mirror the new sequence into the cil
  610. * structure atomically with the addition of this sequence to the
  611. * committing list. This also ensures that we can do unlocked checks
  612. * against the current sequence in log forces without risking
  613. * deferencing a freed context pointer.
  614. */
  615. spin_lock(&cil->xc_push_lock);
  616. cil->xc_current_sequence = new_ctx->sequence;
  617. spin_unlock(&cil->xc_push_lock);
  618. up_write(&cil->xc_ctx_lock);
  619. /*
  620. * Build a checkpoint transaction header and write it to the log to
  621. * begin the transaction. We need to account for the space used by the
  622. * transaction header here as it is not accounted for in xlog_write().
  623. *
  624. * The LSN we need to pass to the log items on transaction commit is
  625. * the LSN reported by the first log vector write. If we use the commit
  626. * record lsn then we can move the tail beyond the grant write head.
  627. */
  628. tic = ctx->ticket;
  629. thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
  630. thdr.th_type = XFS_TRANS_CHECKPOINT;
  631. thdr.th_tid = tic->t_tid;
  632. thdr.th_num_items = num_iovecs;
  633. lhdr.i_addr = &thdr;
  634. lhdr.i_len = sizeof(xfs_trans_header_t);
  635. lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
  636. tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
  637. lvhdr.lv_niovecs = 1;
  638. lvhdr.lv_iovecp = &lhdr;
  639. lvhdr.lv_next = ctx->lv_chain;
  640. error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0);
  641. if (error)
  642. goto out_abort_free_ticket;
  643. /*
  644. * now that we've written the checkpoint into the log, strictly
  645. * order the commit records so replay will get them in the right order.
  646. */
  647. restart:
  648. spin_lock(&cil->xc_push_lock);
  649. list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
  650. /*
  651. * Avoid getting stuck in this loop because we were woken by the
  652. * shutdown, but then went back to sleep once already in the
  653. * shutdown state.
  654. */
  655. if (XLOG_FORCED_SHUTDOWN(log)) {
  656. spin_unlock(&cil->xc_push_lock);
  657. goto out_abort_free_ticket;
  658. }
  659. /*
  660. * Higher sequences will wait for this one so skip them.
  661. * Don't wait for our own sequence, either.
  662. */
  663. if (new_ctx->sequence >= ctx->sequence)
  664. continue;
  665. if (!new_ctx->commit_lsn) {
  666. /*
  667. * It is still being pushed! Wait for the push to
  668. * complete, then start again from the beginning.
  669. */
  670. xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
  671. goto restart;
  672. }
  673. }
  674. spin_unlock(&cil->xc_push_lock);
  675. /* xfs_log_done always frees the ticket on error. */
  676. commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, false);
  677. if (commit_lsn == -1)
  678. goto out_abort;
  679. /* attach all the transactions w/ busy extents to iclog */
  680. ctx->log_cb.cb_func = xlog_cil_committed;
  681. ctx->log_cb.cb_arg = ctx;
  682. error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb);
  683. if (error)
  684. goto out_abort;
  685. /*
  686. * now the checkpoint commit is complete and we've attached the
  687. * callbacks to the iclog we can assign the commit LSN to the context
  688. * and wake up anyone who is waiting for the commit to complete.
  689. */
  690. spin_lock(&cil->xc_push_lock);
  691. ctx->commit_lsn = commit_lsn;
  692. wake_up_all(&cil->xc_commit_wait);
  693. spin_unlock(&cil->xc_push_lock);
  694. /* release the hounds! */
  695. return xfs_log_release_iclog(log->l_mp, commit_iclog);
  696. out_skip:
  697. up_write(&cil->xc_ctx_lock);
  698. xfs_log_ticket_put(new_ctx->ticket);
  699. kmem_free(new_ctx);
  700. return 0;
  701. out_abort_free_ticket:
  702. xfs_log_ticket_put(tic);
  703. out_abort:
  704. xlog_cil_committed(ctx, XFS_LI_ABORTED);
  705. return -EIO;
  706. }
  707. static void
  708. xlog_cil_push_work(
  709. struct work_struct *work)
  710. {
  711. struct xfs_cil *cil = container_of(work, struct xfs_cil,
  712. xc_push_work);
  713. xlog_cil_push(cil->xc_log);
  714. }
  715. /*
  716. * We need to push CIL every so often so we don't cache more than we can fit in
  717. * the log. The limit really is that a checkpoint can't be more than half the
  718. * log (the current checkpoint is not allowed to overwrite the previous
  719. * checkpoint), but commit latency and memory usage limit this to a smaller
  720. * size.
  721. */
  722. static void
  723. xlog_cil_push_background(
  724. struct xlog *log)
  725. {
  726. struct xfs_cil *cil = log->l_cilp;
  727. /*
  728. * The cil won't be empty because we are called while holding the
  729. * context lock so whatever we added to the CIL will still be there
  730. */
  731. ASSERT(!list_empty(&cil->xc_cil));
  732. /*
  733. * don't do a background push if we haven't used up all the
  734. * space available yet.
  735. */
  736. if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log))
  737. return;
  738. spin_lock(&cil->xc_push_lock);
  739. if (cil->xc_push_seq < cil->xc_current_sequence) {
  740. cil->xc_push_seq = cil->xc_current_sequence;
  741. queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
  742. }
  743. spin_unlock(&cil->xc_push_lock);
  744. }
  745. /*
  746. * xlog_cil_push_now() is used to trigger an immediate CIL push to the sequence
  747. * number that is passed. When it returns, the work will be queued for
  748. * @push_seq, but it won't be completed. The caller is expected to do any
  749. * waiting for push_seq to complete if it is required.
  750. */
  751. static void
  752. xlog_cil_push_now(
  753. struct xlog *log,
  754. xfs_lsn_t push_seq)
  755. {
  756. struct xfs_cil *cil = log->l_cilp;
  757. if (!cil)
  758. return;
  759. ASSERT(push_seq && push_seq <= cil->xc_current_sequence);
  760. /* start on any pending background push to minimise wait time on it */
  761. flush_work(&cil->xc_push_work);
  762. /*
  763. * If the CIL is empty or we've already pushed the sequence then
  764. * there's no work we need to do.
  765. */
  766. spin_lock(&cil->xc_push_lock);
  767. if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) {
  768. spin_unlock(&cil->xc_push_lock);
  769. return;
  770. }
  771. cil->xc_push_seq = push_seq;
  772. queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
  773. spin_unlock(&cil->xc_push_lock);
  774. }
  775. bool
  776. xlog_cil_empty(
  777. struct xlog *log)
  778. {
  779. struct xfs_cil *cil = log->l_cilp;
  780. bool empty = false;
  781. spin_lock(&cil->xc_push_lock);
  782. if (list_empty(&cil->xc_cil))
  783. empty = true;
  784. spin_unlock(&cil->xc_push_lock);
  785. return empty;
  786. }
  787. /*
  788. * Commit a transaction with the given vector to the Committed Item List.
  789. *
  790. * To do this, we need to format the item, pin it in memory if required and
  791. * account for the space used by the transaction. Once we have done that we
  792. * need to release the unused reservation for the transaction, attach the
  793. * transaction to the checkpoint context so we carry the busy extents through
  794. * to checkpoint completion, and then unlock all the items in the transaction.
  795. *
  796. * Called with the context lock already held in read mode to lock out
  797. * background commit, returns without it held once background commits are
  798. * allowed again.
  799. */
  800. void
  801. xfs_log_commit_cil(
  802. struct xfs_mount *mp,
  803. struct xfs_trans *tp,
  804. xfs_lsn_t *commit_lsn,
  805. bool regrant)
  806. {
  807. struct xlog *log = mp->m_log;
  808. struct xfs_cil *cil = log->l_cilp;
  809. /*
  810. * Do all necessary memory allocation before we lock the CIL.
  811. * This ensures the allocation does not deadlock with a CIL
  812. * push in memory reclaim (e.g. from kswapd).
  813. */
  814. xlog_cil_alloc_shadow_bufs(log, tp);
  815. /* lock out background commit */
  816. down_read(&cil->xc_ctx_lock);
  817. xlog_cil_insert_items(log, tp);
  818. /* check we didn't blow the reservation */
  819. if (tp->t_ticket->t_curr_res < 0)
  820. xlog_print_tic_res(mp, tp->t_ticket);
  821. tp->t_commit_lsn = cil->xc_ctx->sequence;
  822. if (commit_lsn)
  823. *commit_lsn = tp->t_commit_lsn;
  824. xfs_log_done(mp, tp->t_ticket, NULL, regrant);
  825. xfs_trans_unreserve_and_mod_sb(tp);
  826. /*
  827. * Once all the items of the transaction have been copied to the CIL,
  828. * the items can be unlocked and freed.
  829. *
  830. * This needs to be done before we drop the CIL context lock because we
  831. * have to update state in the log items and unlock them before they go
  832. * to disk. If we don't, then the CIL checkpoint can race with us and
  833. * we can run checkpoint completion before we've updated and unlocked
  834. * the log items. This affects (at least) processing of stale buffers,
  835. * inodes and EFIs.
  836. */
  837. xfs_trans_free_items(tp, tp->t_commit_lsn, false);
  838. xlog_cil_push_background(log);
  839. up_read(&cil->xc_ctx_lock);
  840. }
  841. /*
  842. * Conditionally push the CIL based on the sequence passed in.
  843. *
  844. * We only need to push if we haven't already pushed the sequence
  845. * number given. Hence the only time we will trigger a push here is
  846. * if the push sequence is the same as the current context.
  847. *
  848. * We return the current commit lsn to allow the callers to determine if a
  849. * iclog flush is necessary following this call.
  850. */
  851. xfs_lsn_t
  852. xlog_cil_force_lsn(
  853. struct xlog *log,
  854. xfs_lsn_t sequence)
  855. {
  856. struct xfs_cil *cil = log->l_cilp;
  857. struct xfs_cil_ctx *ctx;
  858. xfs_lsn_t commit_lsn = NULLCOMMITLSN;
  859. ASSERT(sequence <= cil->xc_current_sequence);
  860. /*
  861. * check to see if we need to force out the current context.
  862. * xlog_cil_push() handles racing pushes for the same sequence,
  863. * so no need to deal with it here.
  864. */
  865. restart:
  866. xlog_cil_push_now(log, sequence);
  867. /*
  868. * See if we can find a previous sequence still committing.
  869. * We need to wait for all previous sequence commits to complete
  870. * before allowing the force of push_seq to go ahead. Hence block
  871. * on commits for those as well.
  872. */
  873. spin_lock(&cil->xc_push_lock);
  874. list_for_each_entry(ctx, &cil->xc_committing, committing) {
  875. /*
  876. * Avoid getting stuck in this loop because we were woken by the
  877. * shutdown, but then went back to sleep once already in the
  878. * shutdown state.
  879. */
  880. if (XLOG_FORCED_SHUTDOWN(log))
  881. goto out_shutdown;
  882. if (ctx->sequence > sequence)
  883. continue;
  884. if (!ctx->commit_lsn) {
  885. /*
  886. * It is still being pushed! Wait for the push to
  887. * complete, then start again from the beginning.
  888. */
  889. xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
  890. goto restart;
  891. }
  892. if (ctx->sequence != sequence)
  893. continue;
  894. /* found it! */
  895. commit_lsn = ctx->commit_lsn;
  896. }
  897. /*
  898. * The call to xlog_cil_push_now() executes the push in the background.
  899. * Hence by the time we have got here it our sequence may not have been
  900. * pushed yet. This is true if the current sequence still matches the
  901. * push sequence after the above wait loop and the CIL still contains
  902. * dirty objects. This is guaranteed by the push code first adding the
  903. * context to the committing list before emptying the CIL.
  904. *
  905. * Hence if we don't find the context in the committing list and the
  906. * current sequence number is unchanged then the CIL contents are
  907. * significant. If the CIL is empty, if means there was nothing to push
  908. * and that means there is nothing to wait for. If the CIL is not empty,
  909. * it means we haven't yet started the push, because if it had started
  910. * we would have found the context on the committing list.
  911. */
  912. if (sequence == cil->xc_current_sequence &&
  913. !list_empty(&cil->xc_cil)) {
  914. spin_unlock(&cil->xc_push_lock);
  915. goto restart;
  916. }
  917. spin_unlock(&cil->xc_push_lock);
  918. return commit_lsn;
  919. /*
  920. * We detected a shutdown in progress. We need to trigger the log force
  921. * to pass through it's iclog state machine error handling, even though
  922. * we are already in a shutdown state. Hence we can't return
  923. * NULLCOMMITLSN here as that has special meaning to log forces (i.e.
  924. * LSN is already stable), so we return a zero LSN instead.
  925. */
  926. out_shutdown:
  927. spin_unlock(&cil->xc_push_lock);
  928. return 0;
  929. }
  930. /*
  931. * Check if the current log item was first committed in this sequence.
  932. * We can't rely on just the log item being in the CIL, we have to check
  933. * the recorded commit sequence number.
  934. *
  935. * Note: for this to be used in a non-racy manner, it has to be called with
  936. * CIL flushing locked out. As a result, it should only be used during the
  937. * transaction commit process when deciding what to format into the item.
  938. */
  939. bool
  940. xfs_log_item_in_current_chkpt(
  941. struct xfs_log_item *lip)
  942. {
  943. struct xfs_cil_ctx *ctx;
  944. if (list_empty(&lip->li_cil))
  945. return false;
  946. ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
  947. /*
  948. * li_seq is written on the first commit of a log item to record the
  949. * first checkpoint it is written to. Hence if it is different to the
  950. * current sequence, we're in a new checkpoint.
  951. */
  952. if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0)
  953. return false;
  954. return true;
  955. }
  956. /*
  957. * Perform initial CIL structure initialisation.
  958. */
  959. int
  960. xlog_cil_init(
  961. struct xlog *log)
  962. {
  963. struct xfs_cil *cil;
  964. struct xfs_cil_ctx *ctx;
  965. cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL);
  966. if (!cil)
  967. return -ENOMEM;
  968. ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL);
  969. if (!ctx) {
  970. kmem_free(cil);
  971. return -ENOMEM;
  972. }
  973. INIT_WORK(&cil->xc_push_work, xlog_cil_push_work);
  974. INIT_LIST_HEAD(&cil->xc_cil);
  975. INIT_LIST_HEAD(&cil->xc_committing);
  976. spin_lock_init(&cil->xc_cil_lock);
  977. spin_lock_init(&cil->xc_push_lock);
  978. init_rwsem(&cil->xc_ctx_lock);
  979. init_waitqueue_head(&cil->xc_commit_wait);
  980. INIT_LIST_HEAD(&ctx->committing);
  981. INIT_LIST_HEAD(&ctx->busy_extents);
  982. ctx->sequence = 1;
  983. ctx->cil = cil;
  984. cil->xc_ctx = ctx;
  985. cil->xc_current_sequence = ctx->sequence;
  986. cil->xc_log = log;
  987. log->l_cilp = cil;
  988. return 0;
  989. }
  990. void
  991. xlog_cil_destroy(
  992. struct xlog *log)
  993. {
  994. if (log->l_cilp->xc_ctx) {
  995. if (log->l_cilp->xc_ctx->ticket)
  996. xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
  997. kmem_free(log->l_cilp->xc_ctx);
  998. }
  999. ASSERT(list_empty(&log->l_cilp->xc_cil));
  1000. kmem_free(log->l_cilp);
  1001. }