xfs_log_cil.c 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724
  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_types.h"
  20. #include "xfs_bit.h"
  21. #include "xfs_log.h"
  22. #include "xfs_inum.h"
  23. #include "xfs_trans.h"
  24. #include "xfs_trans_priv.h"
  25. #include "xfs_log_priv.h"
  26. #include "xfs_sb.h"
  27. #include "xfs_ag.h"
  28. #include "xfs_dir2.h"
  29. #include "xfs_mount.h"
  30. #include "xfs_error.h"
  31. #include "xfs_alloc.h"
  32. /*
  33. * Perform initial CIL structure initialisation. If the CIL is not
  34. * enabled in this filesystem, ensure the log->l_cilp is null so
  35. * we can check this conditional to determine if we are doing delayed
  36. * logging or not.
  37. */
  38. int
  39. xlog_cil_init(
  40. struct log *log)
  41. {
  42. struct xfs_cil *cil;
  43. struct xfs_cil_ctx *ctx;
  44. log->l_cilp = NULL;
  45. if (!(log->l_mp->m_flags & XFS_MOUNT_DELAYLOG))
  46. return 0;
  47. cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL);
  48. if (!cil)
  49. return ENOMEM;
  50. ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL);
  51. if (!ctx) {
  52. kmem_free(cil);
  53. return ENOMEM;
  54. }
  55. INIT_LIST_HEAD(&cil->xc_cil);
  56. INIT_LIST_HEAD(&cil->xc_committing);
  57. spin_lock_init(&cil->xc_cil_lock);
  58. init_rwsem(&cil->xc_ctx_lock);
  59. sv_init(&cil->xc_commit_wait, SV_DEFAULT, "cilwait");
  60. INIT_LIST_HEAD(&ctx->committing);
  61. INIT_LIST_HEAD(&ctx->busy_extents);
  62. ctx->sequence = 1;
  63. ctx->cil = cil;
  64. cil->xc_ctx = ctx;
  65. cil->xc_log = log;
  66. log->l_cilp = cil;
  67. return 0;
  68. }
  69. void
  70. xlog_cil_destroy(
  71. struct log *log)
  72. {
  73. if (!log->l_cilp)
  74. return;
  75. if (log->l_cilp->xc_ctx) {
  76. if (log->l_cilp->xc_ctx->ticket)
  77. xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
  78. kmem_free(log->l_cilp->xc_ctx);
  79. }
  80. ASSERT(list_empty(&log->l_cilp->xc_cil));
  81. kmem_free(log->l_cilp);
  82. }
  83. /*
  84. * Allocate a new ticket. Failing to get a new ticket makes it really hard to
  85. * recover, so we don't allow failure here. Also, we allocate in a context that
  86. * we don't want to be issuing transactions from, so we need to tell the
  87. * allocation code this as well.
  88. *
  89. * We don't reserve any space for the ticket - we are going to steal whatever
  90. * space we require from transactions as they commit. To ensure we reserve all
  91. * the space required, we need to set the current reservation of the ticket to
  92. * zero so that we know to steal the initial transaction overhead from the
  93. * first transaction commit.
  94. */
  95. static struct xlog_ticket *
  96. xlog_cil_ticket_alloc(
  97. struct log *log)
  98. {
  99. struct xlog_ticket *tic;
  100. tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0,
  101. KM_SLEEP|KM_NOFS);
  102. tic->t_trans_type = XFS_TRANS_CHECKPOINT;
  103. /*
  104. * set the current reservation to zero so we know to steal the basic
  105. * transaction overhead reservation from the first transaction commit.
  106. */
  107. tic->t_curr_res = 0;
  108. return tic;
  109. }
  110. /*
  111. * After the first stage of log recovery is done, we know where the head and
  112. * tail of the log are. We need this log initialisation done before we can
  113. * initialise the first CIL checkpoint context.
  114. *
  115. * Here we allocate a log ticket to track space usage during a CIL push. This
  116. * ticket is passed to xlog_write() directly so that we don't slowly leak log
  117. * space by failing to account for space used by log headers and additional
  118. * region headers for split regions.
  119. */
  120. void
  121. xlog_cil_init_post_recovery(
  122. struct log *log)
  123. {
  124. if (!log->l_cilp)
  125. return;
  126. log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log);
  127. log->l_cilp->xc_ctx->sequence = 1;
  128. log->l_cilp->xc_ctx->commit_lsn = xlog_assign_lsn(log->l_curr_cycle,
  129. log->l_curr_block);
  130. }
  131. /*
  132. * Insert the log item into the CIL and calculate the difference in space
  133. * consumed by the item. Add the space to the checkpoint ticket and calculate
  134. * if the change requires additional log metadata. If it does, take that space
  135. * as well. Remove the amount of space we addded to the checkpoint ticket from
  136. * the current transaction ticket so that the accounting works out correctly.
  137. *
  138. * If this is the first time the item is being placed into the CIL in this
  139. * context, pin it so it can't be written to disk until the CIL is flushed to
  140. * the iclog and the iclog written to disk.
  141. */
  142. static void
  143. xlog_cil_insert(
  144. struct log *log,
  145. struct xlog_ticket *ticket,
  146. struct xfs_log_item *item,
  147. struct xfs_log_vec *lv)
  148. {
  149. struct xfs_cil *cil = log->l_cilp;
  150. struct xfs_log_vec *old = lv->lv_item->li_lv;
  151. struct xfs_cil_ctx *ctx = cil->xc_ctx;
  152. int len;
  153. int diff_iovecs;
  154. int iclog_space;
  155. if (old) {
  156. /* existing lv on log item, space used is a delta */
  157. ASSERT(!list_empty(&item->li_cil));
  158. ASSERT(old->lv_buf && old->lv_buf_len && old->lv_niovecs);
  159. len = lv->lv_buf_len - old->lv_buf_len;
  160. diff_iovecs = lv->lv_niovecs - old->lv_niovecs;
  161. kmem_free(old->lv_buf);
  162. kmem_free(old);
  163. } else {
  164. /* new lv, must pin the log item */
  165. ASSERT(!lv->lv_item->li_lv);
  166. ASSERT(list_empty(&item->li_cil));
  167. len = lv->lv_buf_len;
  168. diff_iovecs = lv->lv_niovecs;
  169. IOP_PIN(lv->lv_item);
  170. }
  171. len += diff_iovecs * sizeof(xlog_op_header_t);
  172. /* attach new log vector to log item */
  173. lv->lv_item->li_lv = lv;
  174. spin_lock(&cil->xc_cil_lock);
  175. list_move_tail(&item->li_cil, &cil->xc_cil);
  176. ctx->nvecs += diff_iovecs;
  177. /*
  178. * If this is the first time the item is being committed to the CIL,
  179. * store the sequence number on the log item so we can tell
  180. * in future commits whether this is the first checkpoint the item is
  181. * being committed into.
  182. */
  183. if (!item->li_seq)
  184. item->li_seq = ctx->sequence;
  185. /*
  186. * Now transfer enough transaction reservation to the context ticket
  187. * for the checkpoint. The context ticket is special - the unit
  188. * reservation has to grow as well as the current reservation as we
  189. * steal from tickets so we can correctly determine the space used
  190. * during the transaction commit.
  191. */
  192. if (ctx->ticket->t_curr_res == 0) {
  193. /* first commit in checkpoint, steal the header reservation */
  194. ASSERT(ticket->t_curr_res >= ctx->ticket->t_unit_res + len);
  195. ctx->ticket->t_curr_res = ctx->ticket->t_unit_res;
  196. ticket->t_curr_res -= ctx->ticket->t_unit_res;
  197. }
  198. /* do we need space for more log record headers? */
  199. iclog_space = log->l_iclog_size - log->l_iclog_hsize;
  200. if (len > 0 && (ctx->space_used / iclog_space !=
  201. (ctx->space_used + len) / iclog_space)) {
  202. int hdrs;
  203. hdrs = (len + iclog_space - 1) / iclog_space;
  204. /* need to take into account split region headers, too */
  205. hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
  206. ctx->ticket->t_unit_res += hdrs;
  207. ctx->ticket->t_curr_res += hdrs;
  208. ticket->t_curr_res -= hdrs;
  209. ASSERT(ticket->t_curr_res >= len);
  210. }
  211. ticket->t_curr_res -= len;
  212. ctx->space_used += len;
  213. spin_unlock(&cil->xc_cil_lock);
  214. }
  215. /*
  216. * Format log item into a flat buffers
  217. *
  218. * For delayed logging, we need to hold a formatted buffer containing all the
  219. * changes on the log item. This enables us to relog the item in memory and
  220. * write it out asynchronously without needing to relock the object that was
  221. * modified at the time it gets written into the iclog.
  222. *
  223. * This function builds a vector for the changes in each log item in the
  224. * transaction. It then works out the length of the buffer needed for each log
  225. * item, allocates them and formats the vector for the item into the buffer.
  226. * The buffer is then attached to the log item are then inserted into the
  227. * Committed Item List for tracking until the next checkpoint is written out.
  228. *
  229. * We don't set up region headers during this process; we simply copy the
  230. * regions into the flat buffer. We can do this because we still have to do a
  231. * formatting step to write the regions into the iclog buffer. Writing the
  232. * ophdrs during the iclog write means that we can support splitting large
  233. * regions across iclog boundares without needing a change in the format of the
  234. * item/region encapsulation.
  235. *
  236. * Hence what we need to do now is change the rewrite the vector array to point
  237. * to the copied region inside the buffer we just allocated. This allows us to
  238. * format the regions into the iclog as though they are being formatted
  239. * directly out of the objects themselves.
  240. */
  241. static void
  242. xlog_cil_format_items(
  243. struct log *log,
  244. struct xfs_log_vec *log_vector,
  245. struct xlog_ticket *ticket,
  246. xfs_lsn_t *start_lsn)
  247. {
  248. struct xfs_log_vec *lv;
  249. if (start_lsn)
  250. *start_lsn = log->l_cilp->xc_ctx->sequence;
  251. ASSERT(log_vector);
  252. for (lv = log_vector; lv; lv = lv->lv_next) {
  253. void *ptr;
  254. int index;
  255. int len = 0;
  256. /* build the vector array and calculate it's length */
  257. IOP_FORMAT(lv->lv_item, lv->lv_iovecp);
  258. for (index = 0; index < lv->lv_niovecs; index++)
  259. len += lv->lv_iovecp[index].i_len;
  260. lv->lv_buf_len = len;
  261. lv->lv_buf = kmem_zalloc(lv->lv_buf_len, KM_SLEEP|KM_NOFS);
  262. ptr = lv->lv_buf;
  263. for (index = 0; index < lv->lv_niovecs; index++) {
  264. struct xfs_log_iovec *vec = &lv->lv_iovecp[index];
  265. memcpy(ptr, vec->i_addr, vec->i_len);
  266. vec->i_addr = ptr;
  267. ptr += vec->i_len;
  268. }
  269. ASSERT(ptr == lv->lv_buf + lv->lv_buf_len);
  270. xlog_cil_insert(log, ticket, lv->lv_item, lv);
  271. }
  272. }
  273. static void
  274. xlog_cil_free_logvec(
  275. struct xfs_log_vec *log_vector)
  276. {
  277. struct xfs_log_vec *lv;
  278. for (lv = log_vector; lv; ) {
  279. struct xfs_log_vec *next = lv->lv_next;
  280. kmem_free(lv->lv_buf);
  281. kmem_free(lv);
  282. lv = next;
  283. }
  284. }
  285. /*
  286. * Commit a transaction with the given vector to the Committed Item List.
  287. *
  288. * To do this, we need to format the item, pin it in memory if required and
  289. * account for the space used by the transaction. Once we have done that we
  290. * need to release the unused reservation for the transaction, attach the
  291. * transaction to the checkpoint context so we carry the busy extents through
  292. * to checkpoint completion, and then unlock all the items in the transaction.
  293. *
  294. * For more specific information about the order of operations in
  295. * xfs_log_commit_cil() please refer to the comments in
  296. * xfs_trans_commit_iclog().
  297. *
  298. * Called with the context lock already held in read mode to lock out
  299. * background commit, returns without it held once background commits are
  300. * allowed again.
  301. */
  302. int
  303. xfs_log_commit_cil(
  304. struct xfs_mount *mp,
  305. struct xfs_trans *tp,
  306. struct xfs_log_vec *log_vector,
  307. xfs_lsn_t *commit_lsn,
  308. int flags)
  309. {
  310. struct log *log = mp->m_log;
  311. int log_flags = 0;
  312. int push = 0;
  313. if (flags & XFS_TRANS_RELEASE_LOG_RES)
  314. log_flags = XFS_LOG_REL_PERM_RESERV;
  315. if (XLOG_FORCED_SHUTDOWN(log)) {
  316. xlog_cil_free_logvec(log_vector);
  317. return XFS_ERROR(EIO);
  318. }
  319. /* lock out background commit */
  320. down_read(&log->l_cilp->xc_ctx_lock);
  321. xlog_cil_format_items(log, log_vector, tp->t_ticket, commit_lsn);
  322. /* check we didn't blow the reservation */
  323. if (tp->t_ticket->t_curr_res < 0)
  324. xlog_print_tic_res(log->l_mp, tp->t_ticket);
  325. /* attach the transaction to the CIL if it has any busy extents */
  326. if (!list_empty(&tp->t_busy)) {
  327. spin_lock(&log->l_cilp->xc_cil_lock);
  328. list_splice_init(&tp->t_busy,
  329. &log->l_cilp->xc_ctx->busy_extents);
  330. spin_unlock(&log->l_cilp->xc_cil_lock);
  331. }
  332. tp->t_commit_lsn = *commit_lsn;
  333. xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
  334. xfs_trans_unreserve_and_mod_sb(tp);
  335. /* check for background commit before unlock */
  336. if (log->l_cilp->xc_ctx->space_used > XLOG_CIL_SPACE_LIMIT(log))
  337. push = 1;
  338. up_read(&log->l_cilp->xc_ctx_lock);
  339. /*
  340. * We need to push CIL every so often so we don't cache more than we
  341. * can fit in the log. The limit really is that a checkpoint can't be
  342. * more than half the log (the current checkpoint is not allowed to
  343. * overwrite the previous checkpoint), but commit latency and memory
  344. * usage limit this to a smaller size in most cases.
  345. */
  346. if (push)
  347. xlog_cil_push(log, 0);
  348. return 0;
  349. }
  350. /*
  351. * Mark all items committed and clear busy extents. We free the log vector
  352. * chains in a separate pass so that we unpin the log items as quickly as
  353. * possible.
  354. */
  355. static void
  356. xlog_cil_committed(
  357. void *args,
  358. int abort)
  359. {
  360. struct xfs_cil_ctx *ctx = args;
  361. struct xfs_log_vec *lv;
  362. int abortflag = abort ? XFS_LI_ABORTED : 0;
  363. struct xfs_busy_extent *busyp, *n;
  364. /* unpin all the log items */
  365. for (lv = ctx->lv_chain; lv; lv = lv->lv_next ) {
  366. xfs_trans_item_committed(lv->lv_item, ctx->start_lsn,
  367. abortflag);
  368. }
  369. list_for_each_entry_safe(busyp, n, &ctx->busy_extents, list)
  370. xfs_alloc_busy_clear(ctx->cil->xc_log->l_mp, busyp);
  371. spin_lock(&ctx->cil->xc_cil_lock);
  372. list_del(&ctx->committing);
  373. spin_unlock(&ctx->cil->xc_cil_lock);
  374. xlog_cil_free_logvec(ctx->lv_chain);
  375. kmem_free(ctx);
  376. }
  377. /*
  378. * Push the Committed Item List to the log. If the push_now flag is not set,
  379. * then it is a background flush and so we can chose to ignore it.
  380. */
  381. int
  382. xlog_cil_push(
  383. struct log *log,
  384. int push_now)
  385. {
  386. struct xfs_cil *cil = log->l_cilp;
  387. struct xfs_log_vec *lv;
  388. struct xfs_cil_ctx *ctx;
  389. struct xfs_cil_ctx *new_ctx;
  390. struct xlog_in_core *commit_iclog;
  391. struct xlog_ticket *tic;
  392. int num_lv;
  393. int num_iovecs;
  394. int len;
  395. int error = 0;
  396. struct xfs_trans_header thdr;
  397. struct xfs_log_iovec lhdr;
  398. struct xfs_log_vec lvhdr = { NULL };
  399. xfs_lsn_t commit_lsn;
  400. if (!cil)
  401. return 0;
  402. new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS);
  403. new_ctx->ticket = xlog_cil_ticket_alloc(log);
  404. /* lock out transaction commit, but don't block on background push */
  405. if (!down_write_trylock(&cil->xc_ctx_lock)) {
  406. if (!push_now)
  407. goto out_free_ticket;
  408. down_write(&cil->xc_ctx_lock);
  409. }
  410. ctx = cil->xc_ctx;
  411. /* check if we've anything to push */
  412. if (list_empty(&cil->xc_cil))
  413. goto out_skip;
  414. /* check for spurious background flush */
  415. if (!push_now && cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log))
  416. goto out_skip;
  417. /*
  418. * pull all the log vectors off the items in the CIL, and
  419. * remove the items from the CIL. We don't need the CIL lock
  420. * here because it's only needed on the transaction commit
  421. * side which is currently locked out by the flush lock.
  422. */
  423. lv = NULL;
  424. num_lv = 0;
  425. num_iovecs = 0;
  426. len = 0;
  427. while (!list_empty(&cil->xc_cil)) {
  428. struct xfs_log_item *item;
  429. int i;
  430. item = list_first_entry(&cil->xc_cil,
  431. struct xfs_log_item, li_cil);
  432. list_del_init(&item->li_cil);
  433. if (!ctx->lv_chain)
  434. ctx->lv_chain = item->li_lv;
  435. else
  436. lv->lv_next = item->li_lv;
  437. lv = item->li_lv;
  438. item->li_lv = NULL;
  439. num_lv++;
  440. num_iovecs += lv->lv_niovecs;
  441. for (i = 0; i < lv->lv_niovecs; i++)
  442. len += lv->lv_iovecp[i].i_len;
  443. }
  444. /*
  445. * initialise the new context and attach it to the CIL. Then attach
  446. * the current context to the CIL committing lsit so it can be found
  447. * during log forces to extract the commit lsn of the sequence that
  448. * needs to be forced.
  449. */
  450. INIT_LIST_HEAD(&new_ctx->committing);
  451. INIT_LIST_HEAD(&new_ctx->busy_extents);
  452. new_ctx->sequence = ctx->sequence + 1;
  453. new_ctx->cil = cil;
  454. cil->xc_ctx = new_ctx;
  455. /*
  456. * The switch is now done, so we can drop the context lock and move out
  457. * of a shared context. We can't just go straight to the commit record,
  458. * though - we need to synchronise with previous and future commits so
  459. * that the commit records are correctly ordered in the log to ensure
  460. * that we process items during log IO completion in the correct order.
  461. *
  462. * For example, if we get an EFI in one checkpoint and the EFD in the
  463. * next (e.g. due to log forces), we do not want the checkpoint with
  464. * the EFD to be committed before the checkpoint with the EFI. Hence
  465. * we must strictly order the commit records of the checkpoints so
  466. * that: a) the checkpoint callbacks are attached to the iclogs in the
  467. * correct order; and b) the checkpoints are replayed in correct order
  468. * in log recovery.
  469. *
  470. * Hence we need to add this context to the committing context list so
  471. * that higher sequences will wait for us to write out a commit record
  472. * before they do.
  473. */
  474. spin_lock(&cil->xc_cil_lock);
  475. list_add(&ctx->committing, &cil->xc_committing);
  476. spin_unlock(&cil->xc_cil_lock);
  477. up_write(&cil->xc_ctx_lock);
  478. /*
  479. * Build a checkpoint transaction header and write it to the log to
  480. * begin the transaction. We need to account for the space used by the
  481. * transaction header here as it is not accounted for in xlog_write().
  482. *
  483. * The LSN we need to pass to the log items on transaction commit is
  484. * the LSN reported by the first log vector write. If we use the commit
  485. * record lsn then we can move the tail beyond the grant write head.
  486. */
  487. tic = ctx->ticket;
  488. thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
  489. thdr.th_type = XFS_TRANS_CHECKPOINT;
  490. thdr.th_tid = tic->t_tid;
  491. thdr.th_num_items = num_iovecs;
  492. lhdr.i_addr = (xfs_caddr_t)&thdr;
  493. lhdr.i_len = sizeof(xfs_trans_header_t);
  494. lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
  495. tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
  496. lvhdr.lv_niovecs = 1;
  497. lvhdr.lv_iovecp = &lhdr;
  498. lvhdr.lv_next = ctx->lv_chain;
  499. error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0);
  500. if (error)
  501. goto out_abort;
  502. /*
  503. * now that we've written the checkpoint into the log, strictly
  504. * order the commit records so replay will get them in the right order.
  505. */
  506. restart:
  507. spin_lock(&cil->xc_cil_lock);
  508. list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
  509. /*
  510. * Higher sequences will wait for this one so skip them.
  511. * Don't wait for own own sequence, either.
  512. */
  513. if (new_ctx->sequence >= ctx->sequence)
  514. continue;
  515. if (!new_ctx->commit_lsn) {
  516. /*
  517. * It is still being pushed! Wait for the push to
  518. * complete, then start again from the beginning.
  519. */
  520. sv_wait(&cil->xc_commit_wait, 0, &cil->xc_cil_lock, 0);
  521. goto restart;
  522. }
  523. }
  524. spin_unlock(&cil->xc_cil_lock);
  525. commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0);
  526. if (error || commit_lsn == -1)
  527. goto out_abort;
  528. /* attach all the transactions w/ busy extents to iclog */
  529. ctx->log_cb.cb_func = xlog_cil_committed;
  530. ctx->log_cb.cb_arg = ctx;
  531. error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb);
  532. if (error)
  533. goto out_abort;
  534. /*
  535. * now the checkpoint commit is complete and we've attached the
  536. * callbacks to the iclog we can assign the commit LSN to the context
  537. * and wake up anyone who is waiting for the commit to complete.
  538. */
  539. spin_lock(&cil->xc_cil_lock);
  540. ctx->commit_lsn = commit_lsn;
  541. sv_broadcast(&cil->xc_commit_wait);
  542. spin_unlock(&cil->xc_cil_lock);
  543. /* release the hounds! */
  544. return xfs_log_release_iclog(log->l_mp, commit_iclog);
  545. out_skip:
  546. up_write(&cil->xc_ctx_lock);
  547. out_free_ticket:
  548. xfs_log_ticket_put(new_ctx->ticket);
  549. kmem_free(new_ctx);
  550. return 0;
  551. out_abort:
  552. xlog_cil_committed(ctx, XFS_LI_ABORTED);
  553. return XFS_ERROR(EIO);
  554. }
  555. /*
  556. * Conditionally push the CIL based on the sequence passed in.
  557. *
  558. * We only need to push if we haven't already pushed the sequence
  559. * number given. Hence the only time we will trigger a push here is
  560. * if the push sequence is the same as the current context.
  561. *
  562. * We return the current commit lsn to allow the callers to determine if a
  563. * iclog flush is necessary following this call.
  564. *
  565. * XXX: Initially, just push the CIL unconditionally and return whatever
  566. * commit lsn is there. It'll be empty, so this is broken for now.
  567. */
  568. xfs_lsn_t
  569. xlog_cil_push_lsn(
  570. struct log *log,
  571. xfs_lsn_t push_seq)
  572. {
  573. struct xfs_cil *cil = log->l_cilp;
  574. struct xfs_cil_ctx *ctx;
  575. xfs_lsn_t commit_lsn = NULLCOMMITLSN;
  576. restart:
  577. down_write(&cil->xc_ctx_lock);
  578. ASSERT(push_seq <= cil->xc_ctx->sequence);
  579. /* check to see if we need to force out the current context */
  580. if (push_seq == cil->xc_ctx->sequence) {
  581. up_write(&cil->xc_ctx_lock);
  582. xlog_cil_push(log, 1);
  583. goto restart;
  584. }
  585. /*
  586. * See if we can find a previous sequence still committing.
  587. * We can drop the flush lock as soon as we have the cil lock
  588. * because we are now only comparing contexts protected by
  589. * the cil lock.
  590. *
  591. * We need to wait for all previous sequence commits to complete
  592. * before allowing the force of push_seq to go ahead. Hence block
  593. * on commits for those as well.
  594. */
  595. spin_lock(&cil->xc_cil_lock);
  596. up_write(&cil->xc_ctx_lock);
  597. list_for_each_entry(ctx, &cil->xc_committing, committing) {
  598. if (ctx->sequence > push_seq)
  599. continue;
  600. if (!ctx->commit_lsn) {
  601. /*
  602. * It is still being pushed! Wait for the push to
  603. * complete, then start again from the beginning.
  604. */
  605. sv_wait(&cil->xc_commit_wait, 0, &cil->xc_cil_lock, 0);
  606. goto restart;
  607. }
  608. if (ctx->sequence != push_seq)
  609. continue;
  610. /* found it! */
  611. commit_lsn = ctx->commit_lsn;
  612. }
  613. spin_unlock(&cil->xc_cil_lock);
  614. return commit_lsn;
  615. }
  616. /*
  617. * Check if the current log item was first committed in this sequence.
  618. * We can't rely on just the log item being in the CIL, we have to check
  619. * the recorded commit sequence number.
  620. *
  621. * Note: for this to be used in a non-racy manner, it has to be called with
  622. * CIL flushing locked out. As a result, it should only be used during the
  623. * transaction commit process when deciding what to format into the item.
  624. */
  625. bool
  626. xfs_log_item_in_current_chkpt(
  627. struct xfs_log_item *lip)
  628. {
  629. struct xfs_cil_ctx *ctx;
  630. if (!(lip->li_mountp->m_flags & XFS_MOUNT_DELAYLOG))
  631. return false;
  632. if (list_empty(&lip->li_cil))
  633. return false;
  634. ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
  635. /*
  636. * li_seq is written on the first commit of a log item to record the
  637. * first checkpoint it is written to. Hence if it is different to the
  638. * current sequence, we're in a new checkpoint.
  639. */
  640. if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0)
  641. return false;
  642. return true;
  643. }