X-Git-Url: https://git.camperquake.de/gitweb.cgi?a=blobdiff_plain;f=module%2Fzfs%2Fzil.c;h=64de856f03577e0f2439179a6d7ab7fa37cb49eb;hb=c99c90015ece64746e20b74245caca41d1dbefe1;hp=4aa4d10b07e80a8ecd61024bb244a3a78b025369;hpb=428870ff734fdaccc342b33fc53cf94724409a46;p=zfs.git diff --git a/module/zfs/zil.c b/module/zfs/zil.c index 4aa4d10..64de856 100644 --- a/module/zfs/zil.c +++ b/module/zfs/zil.c @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ /* Portions Copyright 2010 Robert Milkowski */ @@ -34,9 +35,10 @@ #include #include #include -#include +#include #include #include +#include /* * The zfs intent log (ZIL) saves transaction records of system calls @@ -65,6 +67,27 @@ */ /* + * See zil.h for more information about these fields. + */ +zil_stats_t zil_stats = { + { "zil_commit_count", KSTAT_DATA_UINT64 }, + { "zil_commit_writer_count", KSTAT_DATA_UINT64 }, + { "zil_itx_count", KSTAT_DATA_UINT64 }, + { "zil_itx_indirect_count", KSTAT_DATA_UINT64 }, + { "zil_itx_indirect_bytes", KSTAT_DATA_UINT64 }, + { "zil_itx_copied_count", KSTAT_DATA_UINT64 }, + { "zil_itx_copied_bytes", KSTAT_DATA_UINT64 }, + { "zil_itx_needcopy_count", KSTAT_DATA_UINT64 }, + { "zil_itx_needcopy_bytes", KSTAT_DATA_UINT64 }, + { "zil_itx_metaslab_normal_count", KSTAT_DATA_UINT64 }, + { "zil_itx_metaslab_normal_bytes", KSTAT_DATA_UINT64 }, + { "zil_itx_metaslab_slog_count", KSTAT_DATA_UINT64 }, + { "zil_itx_metaslab_slog_bytes", KSTAT_DATA_UINT64 }, +}; + +static kstat_t *zil_ksp; + +/* * This global ZIL switch affects all pools */ int zil_replay_disable = 0; /* disable intent logging replay */ @@ -74,16 +97,25 @@ int zil_replay_disable = 0; /* disable intent logging replay */ * zfs_nocacheflush will cause corruption on power loss if a volatile * out-of-order write cache is enabled. */ -boolean_t zfs_nocacheflush = B_FALSE; +int zfs_nocacheflush = 0; static kmem_cache_t *zil_lwb_cache; -static boolean_t zil_empty(zilog_t *zilog); +static void zil_async_to_sync(zilog_t *zilog, uint64_t foid); #define LWB_EMPTY(lwb) ((BP_GET_LSIZE(&lwb->lwb_blk) - \ sizeof (zil_chain_t)) == (lwb->lwb_sz - lwb->lwb_nused)) +/* + * ziltest is by and large an ugly hack, but very useful in + * checking replay without tedious work. + * When running ziltest we want to keep all itx's and so maintain + * a single list in the zl_itxg[] that uses a high txg: ZILTEST_TXG + * We subtract TXG_CONCURRENT_STATES to allow for common code. + */ +#define ZILTEST_TXG (UINT64_MAX - TXG_CONCURRENT_STATES) + static int zil_bp_compare(const void *x1, const void *x2) { @@ -134,7 +166,7 @@ zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp) if (avl_find(t, dva, &where) != NULL) return (EEXIST); - zn = kmem_alloc(sizeof (zil_bp_node_t), KM_SLEEP); + zn = kmem_alloc(sizeof (zil_bp_node_t), KM_PUSHPAGE); zn->zn_dva = *dva; avl_insert(t, zn, where); @@ -287,6 +319,8 @@ zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, char *lrbuf, *lrp; int error = 0; + bzero(&next_blk, sizeof(blkptr_t)); + /* * Old logs didn't record the maximum zh_claim_lr_seq. */ @@ -308,7 +342,7 @@ zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, for (blk = zh->zh_log; !BP_IS_HOLE(&blk); blk = next_blk) { uint64_t blk_seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ]; int reclen; - char *end; + char *end = NULL; if (blk_seq > claim_blk_seq) break; @@ -418,13 +452,14 @@ zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg) } static lwb_t * -zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, uint64_t txg) +zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, uint64_t txg, boolean_t fastwrite) { lwb_t *lwb; - lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP); + lwb = kmem_cache_alloc(zil_lwb_cache, KM_PUSHPAGE); lwb->lwb_zilog = zilog; lwb->lwb_blk = *bp; + lwb->lwb_fastwrite = fastwrite; lwb->lwb_buf = zio_buf_alloc(BP_GET_LSIZE(bp)); lwb->lwb_max_txg = txg; lwb->lwb_zio = NULL; @@ -445,6 +480,38 @@ zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, uint64_t txg) } /* + * Called when we create in-memory log transactions so that we know + * to cleanup the itxs at the end of spa_sync(). + */ +void +zilog_dirty(zilog_t *zilog, uint64_t txg) +{ + dsl_pool_t *dp = zilog->zl_dmu_pool; + dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); + + if (dsl_dataset_is_snapshot(ds)) + panic("dirtying snapshot!"); + + if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg) == 0) { + /* up the hold count until we can be written out */ + dmu_buf_add_ref(ds->ds_dbuf, zilog); + } +} + +boolean_t +zilog_is_dirty(zilog_t *zilog) +{ + dsl_pool_t *dp = zilog->zl_dmu_pool; + int t; + + for (t = 0; t < TXG_SIZE; t++) { + if (txg_list_member(&dp->dp_dirty_zilogs, zilog, t)) + return (B_TRUE); + } + return (B_FALSE); +} + +/* * Create an on-disk intent log. */ static lwb_t * @@ -456,6 +523,7 @@ zil_create(zilog_t *zilog) dmu_tx_t *tx = NULL; blkptr_t blk; int error = 0; + boolean_t fastwrite = FALSE; /* * Wait for any previous destroy to complete. @@ -483,8 +551,9 @@ zil_create(zilog_t *zilog) BP_ZERO(&blk); } - error = zio_alloc_zil(zilog->zl_spa, txg, &blk, NULL, - ZIL_MIN_BLKSZ, zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); + error = zio_alloc_zil(zilog->zl_spa, txg, &blk, + ZIL_MIN_BLKSZ, B_TRUE); + fastwrite = TRUE; if (error == 0) zil_init_log_chain(zilog, &blk); @@ -494,7 +563,7 @@ zil_create(zilog_t *zilog) * Allocate a log write buffer (lwb) for the first log block. */ if (error == 0) - lwb = zil_alloc_lwb(zilog, &blk, txg); + lwb = zil_alloc_lwb(zilog, &blk, txg, fastwrite); /* * If we just allocated the first log block, commit our transaction @@ -551,8 +620,12 @@ zil_destroy(zilog_t *zilog, boolean_t keep_first) if (!list_is_empty(&zilog->zl_lwb_list)) { ASSERT(zh->zh_claim_txg == 0); - ASSERT(!keep_first); + VERIFY(!keep_first); while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { + ASSERT(lwb->lwb_zio == NULL); + if (lwb->lwb_fastwrite) + metaslab_fastwrite_unmark(zilog->zl_spa, + &lwb->lwb_blk); list_remove(&zilog->zl_lwb_list, lwb); if (lwb->lwb_buf != NULL) zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); @@ -560,14 +633,21 @@ zil_destroy(zilog_t *zilog, boolean_t keep_first) kmem_cache_free(zil_lwb_cache, lwb); } } else if (!keep_first) { - (void) zil_parse(zilog, zil_free_log_block, - zil_free_log_record, tx, zh->zh_claim_txg); + zil_destroy_sync(zilog, tx); } mutex_exit(&zilog->zl_lock); dmu_tx_commit(tx); } +void +zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx) +{ + ASSERT(list_is_empty(&zilog->zl_lwb_list)); + (void) zil_parse(zilog, zil_free_log_block, + zil_free_log_record, tx, zilog->zl_header->zh_claim_txg); +} + int zil_claim(const char *osname, void *txarg) { @@ -631,6 +711,7 @@ zil_check_log_chain(const char *osname, void *tx) { zilog_t *zilog; objset_t *os; + blkptr_t *bp; int error; ASSERT(tx == NULL); @@ -642,6 +723,29 @@ zil_check_log_chain(const char *osname, void *tx) } zilog = dmu_objset_zil(os); + bp = (blkptr_t *)&zilog->zl_header->zh_log; + + /* + * Check the first block and determine if it's on a log device + * which may have been removed or faulted prior to loading this + * pool. If so, there's no point in checking the rest of the log + * as its content should have already been synced to the pool. + */ + if (!BP_IS_HOLE(bp)) { + vdev_t *vd; + boolean_t valid = B_TRUE; + + spa_config_enter(os->os_spa, SCL_STATE, FTAG, RW_READER); + vd = vdev_lookup_top(os->os_spa, DVA_GET_VDEV(&bp->blk_dva[0])); + if (vd->vdev_islog && vdev_is_dead(vd)) + valid = vdev_log_state_valid(vd); + spa_config_exit(os->os_spa, SCL_STATE, FTAG); + + if (!valid) { + dmu_objset_rele(os, FTAG); + return (0); + } + } /* * Because tx == NULL, zil_claim_log_block() will not actually claim @@ -661,8 +765,8 @@ zil_check_log_chain(const char *osname, void *tx) static int zil_vdev_compare(const void *x1, const void *x2) { - uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev; - uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev; + const uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev; + const uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev; if (v1 < v2) return (-1); @@ -695,7 +799,7 @@ zil_add_block(zilog_t *zilog, const blkptr_t *bp) for (i = 0; i < ndvas; i++) { zvsearch.zv_vdev = DVA_GET_VDEV(&bp->blk_dva[i]); if (avl_find(t, &zvsearch, &where) == NULL) { - zv = kmem_alloc(sizeof (*zv), KM_SLEEP); + zv = kmem_alloc(sizeof (*zv), KM_PUSHPAGE); zv->zv_vdev = zvsearch.zv_vdev; avl_insert(t, zv, where); } @@ -703,7 +807,7 @@ zil_add_block(zilog_t *zilog, const blkptr_t *bp) mutex_exit(&zilog->zl_vdev_lock); } -void +static void zil_flush_vdevs(zilog_t *zilog) { spa_t *spa = zilog->zl_spa; @@ -769,6 +873,8 @@ zil_lwb_write_done(zio_t *zio) */ zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); mutex_enter(&zilog->zl_lock); + lwb->lwb_zio = NULL; + lwb->lwb_fastwrite = FALSE; lwb->lwb_buf = NULL; lwb->lwb_tx = NULL; mutex_exit(&zilog->zl_lock); @@ -797,12 +903,21 @@ zil_lwb_write_init(zilog_t *zilog, lwb_t *lwb) zilog->zl_root_zio = zio_root(zilog->zl_spa, NULL, NULL, ZIO_FLAG_CANFAIL); } + + /* Lock so zil_sync() doesn't fastwrite_unmark after zio is created */ + mutex_enter(&zilog->zl_lock); if (lwb->lwb_zio == NULL) { + if (!lwb->lwb_fastwrite) { + metaslab_fastwrite_mark(zilog->zl_spa, &lwb->lwb_blk); + lwb->lwb_fastwrite = 1; + } lwb->lwb_zio = zio_rewrite(zilog->zl_root_zio, zilog->zl_spa, 0, &lwb->lwb_blk, lwb->lwb_buf, BP_GET_LSIZE(&lwb->lwb_blk), zil_lwb_write_done, lwb, ZIO_PRIORITY_LOG_WRITE, - ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE, &zb); + ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | + ZIO_FLAG_FASTWRITE, &zb); } + mutex_exit(&zilog->zl_lock); } /* @@ -819,14 +934,13 @@ uint64_t zil_block_buckets[] = { }; /* - * Use the slog as long as the logbias is 'latency' and the current commit size - * is less than the limit or the total list size is less than 2X the limit. - * Limit checking is disabled by setting zil_slog_limit to UINT64_MAX. + * Use the slog as long as the current commit size is less than the + * limit or the total list size is less than 2X the limit. Limit + * checking is disabled by setting zil_slog_limit to UINT64_MAX. */ -uint64_t zil_slog_limit = 1024 * 1024; -#define USE_SLOG(zilog) (((zilog)->zl_logbias == ZFS_LOGBIAS_LATENCY) && \ - (((zilog)->zl_cur_used < zil_slog_limit) || \ - ((zilog)->zl_itx_list_sz < (zil_slog_limit << 1)))) +unsigned long zil_slog_limit = 1024 * 1024; +#define USE_SLOG(zilog) (((zilog)->zl_cur_used < zil_slog_limit) || \ + ((zilog)->zl_itx_list_sz < (zil_slog_limit << 1))) /* * Start a log block write and advance to the next log block. @@ -843,6 +957,7 @@ zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb) uint64_t txg; uint64_t zil_blksz, wsz; int i, error; + boolean_t use_slog; if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) { zilc = (zil_chain_t *)lwb->lwb_buf; @@ -898,9 +1013,18 @@ zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb) zilog->zl_prev_rotor = (zilog->zl_prev_rotor + 1) & (ZIL_PREV_BLKS - 1); BP_ZERO(bp); - /* pass the old blkptr in order to spread log blocks across devs */ - error = zio_alloc_zil(spa, txg, bp, &lwb->lwb_blk, zil_blksz, - USE_SLOG(zilog)); + use_slog = USE_SLOG(zilog); + error = zio_alloc_zil(spa, txg, bp, zil_blksz, USE_SLOG(zilog)); + if (use_slog) + { + ZIL_STAT_BUMP(zil_itx_metaslab_slog_count); + ZIL_STAT_INCR(zil_itx_metaslab_slog_bytes, lwb->lwb_nused); + } + else + { + ZIL_STAT_BUMP(zil_itx_metaslab_normal_count); + ZIL_STAT_INCR(zil_itx_metaslab_normal_bytes, lwb->lwb_nused); + } if (!error) { ASSERT3U(bp->blk_birth, ==, txg); bp->blk_cksum = lwb->lwb_blk.blk_cksum; @@ -909,7 +1033,7 @@ zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb) /* * Allocate a new log write buffer (lwb). */ - nlwb = zil_alloc_lwb(zilog, bp, txg); + nlwb = zil_alloc_lwb(zilog, bp, txg, TRUE); /* Record the block for later vdev flushing */ zil_add_block(zilog, &lwb->lwb_blk); @@ -957,6 +1081,8 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) return (NULL); ASSERT(lwb->lwb_buf != NULL); + ASSERT(zilog_is_dirty(zilog) || + spa_freeze_txg(zilog->zl_spa) != UINT64_MAX); if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY) dlen = P2ROUNDUP_TYPED( @@ -986,13 +1112,18 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) lrc = (lr_t *)lr_buf; lrw = (lr_write_t *)lrc; + ZIL_STAT_BUMP(zil_itx_count); + /* * If it's a write, fetch the data or get its blkptr as appropriate. */ if (lrc->lrc_txtype == TX_WRITE) { if (txg > spa_freeze_txg(zilog->zl_spa)) txg_wait_synced(zilog->zl_dmu_pool, txg); - if (itx->itx_wr_state != WR_COPIED) { + if (itx->itx_wr_state == WR_COPIED) { + ZIL_STAT_BUMP(zil_itx_copied_count); + ZIL_STAT_INCR(zil_itx_copied_bytes, lrw->lr_length); + } else { char *dbuf; int error; @@ -1000,9 +1131,13 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) ASSERT(itx->itx_wr_state == WR_NEED_COPY); dbuf = lr_buf + reclen; lrw->lr_common.lrc_reclen += dlen; + ZIL_STAT_BUMP(zil_itx_needcopy_count); + ZIL_STAT_INCR(zil_itx_needcopy_bytes, lrw->lr_length); } else { ASSERT(itx->itx_wr_state == WR_INDIRECT); dbuf = NULL; + ZIL_STAT_BUMP(zil_itx_indirect_count); + ZIL_STAT_INCR(zil_itx_indirect_bytes, lrw->lr_length); } error = zilog->zl_get_data( itx->itx_private, lrw, dbuf, lwb->lwb_zio); @@ -1028,7 +1163,7 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) lwb->lwb_nused += reclen + dlen; lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg); ASSERT3U(lwb->lwb_nused, <=, lwb->lwb_sz); - ASSERT3U(P2PHASE(lwb->lwb_nused, sizeof (uint64_t)), ==, 0); + ASSERT0(P2PHASE(lwb->lwb_nused, sizeof (uint64_t))); return (lwb); } @@ -1040,11 +1175,13 @@ zil_itx_create(uint64_t txtype, size_t lrsize) lrsize = P2ROUNDUP_TYPED(lrsize, sizeof (uint64_t), size_t); - itx = kmem_alloc(offsetof(itx_t, itx_lr) + lrsize, KM_SLEEP); + itx = kmem_alloc(offsetof(itx_t, itx_lr) + lrsize, + KM_PUSHPAGE | KM_NODEBUG); itx->itx_lr.lrc_txtype = txtype; itx->itx_lr.lrc_reclen = lrsize; itx->itx_sod = lrsize; /* if write & WR_NEED_COPY will be increased */ itx->itx_lr.lrc_seq = 0; /* defensive */ + itx->itx_sync = B_TRUE; /* default is synchronous */ return (itx); } @@ -1055,190 +1192,366 @@ zil_itx_destroy(itx_t *itx) kmem_free(itx, offsetof(itx_t, itx_lr) + itx->itx_lr.lrc_reclen); } -uint64_t -zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx) +/* + * Free up the sync and async itxs. The itxs_t has already been detached + * so no locks are needed. + */ +static void +zil_itxg_clean(itxs_t *itxs) { - uint64_t seq; + itx_t *itx; + list_t *list; + avl_tree_t *t; + void *cookie; + itx_async_node_t *ian; + + list = &itxs->i_sync_list; + while ((itx = list_head(list)) != NULL) { + list_remove(list, itx); + kmem_free(itx, offsetof(itx_t, itx_lr) + + itx->itx_lr.lrc_reclen); + } - ASSERT(itx->itx_lr.lrc_seq == 0); - ASSERT(!zilog->zl_replay); + cookie = NULL; + t = &itxs->i_async_tree; + while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) { + list = &ian->ia_list; + while ((itx = list_head(list)) != NULL) { + list_remove(list, itx); + kmem_free(itx, offsetof(itx_t, itx_lr) + + itx->itx_lr.lrc_reclen); + } + list_destroy(list); + kmem_free(ian, sizeof (itx_async_node_t)); + } + avl_destroy(t); - mutex_enter(&zilog->zl_lock); - list_insert_tail(&zilog->zl_itx_list, itx); - zilog->zl_itx_list_sz += itx->itx_sod; - itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx); - itx->itx_lr.lrc_seq = seq = ++zilog->zl_itx_seq; - mutex_exit(&zilog->zl_lock); + kmem_free(itxs, sizeof (itxs_t)); +} + +static int +zil_aitx_compare(const void *x1, const void *x2) +{ + const uint64_t o1 = ((itx_async_node_t *)x1)->ia_foid; + const uint64_t o2 = ((itx_async_node_t *)x2)->ia_foid; + + if (o1 < o2) + return (-1); + if (o1 > o2) + return (1); - return (seq); + return (0); } /* - * Free up all in-memory intent log transactions that have now been synced. + * Remove all async itx with the given oid. */ static void -zil_itx_clean(zilog_t *zilog) +zil_remove_async(zilog_t *zilog, uint64_t oid) { - uint64_t synced_txg = spa_last_synced_txg(zilog->zl_spa); - uint64_t freeze_txg = spa_freeze_txg(zilog->zl_spa); + uint64_t otxg, txg; + itx_async_node_t *ian; + avl_tree_t *t; + avl_index_t where; list_t clean_list; itx_t *itx; + ASSERT(oid != 0); list_create(&clean_list, sizeof (itx_t), offsetof(itx_t, itx_node)); - mutex_enter(&zilog->zl_lock); - /* wait for a log writer to finish walking list */ - while (zilog->zl_writer) { - cv_wait(&zilog->zl_cv_writer, &zilog->zl_lock); - } + if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ + otxg = ZILTEST_TXG; + else + otxg = spa_last_synced_txg(zilog->zl_spa) + 1; - /* - * Move the sync'd log transactions to a separate list so we can call - * kmem_free without holding the zl_lock. - * - * There is no need to set zl_writer as we don't drop zl_lock here - */ - while ((itx = list_head(&zilog->zl_itx_list)) != NULL && - itx->itx_lr.lrc_txg <= MIN(synced_txg, freeze_txg)) { - list_remove(&zilog->zl_itx_list, itx); - zilog->zl_itx_list_sz -= itx->itx_sod; - list_insert_tail(&clean_list, itx); - } - cv_broadcast(&zilog->zl_cv_writer); - mutex_exit(&zilog->zl_lock); + for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { + itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; + + mutex_enter(&itxg->itxg_lock); + if (itxg->itxg_txg != txg) { + mutex_exit(&itxg->itxg_lock); + continue; + } - /* destroy sync'd log transactions */ + /* + * Locate the object node and append its list. + */ + t = &itxg->itxg_itxs->i_async_tree; + ian = avl_find(t, &oid, &where); + if (ian != NULL) + list_move_tail(&clean_list, &ian->ia_list); + mutex_exit(&itxg->itxg_lock); + } while ((itx = list_head(&clean_list)) != NULL) { list_remove(&clean_list, itx); - zil_itx_destroy(itx); + kmem_free(itx, offsetof(itx_t, itx_lr) + + itx->itx_lr.lrc_reclen); } list_destroy(&clean_list); } +void +zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx) +{ + uint64_t txg; + itxg_t *itxg; + itxs_t *itxs, *clean = NULL; + + /* + * Object ids can be re-instantiated in the next txg so + * remove any async transactions to avoid future leaks. + * This can happen if a fsync occurs on the re-instantiated + * object for a WR_INDIRECT or WR_NEED_COPY write, which gets + * the new file data and flushes a write record for the old object. + */ + if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_REMOVE) + zil_remove_async(zilog, itx->itx_oid); + + /* + * Ensure the data of a renamed file is committed before the rename. + */ + if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_RENAME) + zil_async_to_sync(zilog, itx->itx_oid); + + if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) + txg = ZILTEST_TXG; + else + txg = dmu_tx_get_txg(tx); + + itxg = &zilog->zl_itxg[txg & TXG_MASK]; + mutex_enter(&itxg->itxg_lock); + itxs = itxg->itxg_itxs; + if (itxg->itxg_txg != txg) { + if (itxs != NULL) { + /* + * The zil_clean callback hasn't got around to cleaning + * this itxg. Save the itxs for release below. + * This should be rare. + */ + atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod); + itxg->itxg_sod = 0; + clean = itxg->itxg_itxs; + } + ASSERT(itxg->itxg_sod == 0); + itxg->itxg_txg = txg; + itxs = itxg->itxg_itxs = kmem_zalloc(sizeof (itxs_t), KM_PUSHPAGE); + + list_create(&itxs->i_sync_list, sizeof (itx_t), + offsetof(itx_t, itx_node)); + avl_create(&itxs->i_async_tree, zil_aitx_compare, + sizeof (itx_async_node_t), + offsetof(itx_async_node_t, ia_node)); + } + if (itx->itx_sync) { + list_insert_tail(&itxs->i_sync_list, itx); + atomic_add_64(&zilog->zl_itx_list_sz, itx->itx_sod); + itxg->itxg_sod += itx->itx_sod; + } else { + avl_tree_t *t = &itxs->i_async_tree; + uint64_t foid = ((lr_ooo_t *)&itx->itx_lr)->lr_foid; + itx_async_node_t *ian; + avl_index_t where; + + ian = avl_find(t, &foid, &where); + if (ian == NULL) { + ian = kmem_alloc(sizeof (itx_async_node_t), KM_PUSHPAGE); + list_create(&ian->ia_list, sizeof (itx_t), + offsetof(itx_t, itx_node)); + ian->ia_foid = foid; + avl_insert(t, ian, where); + } + list_insert_tail(&ian->ia_list, itx); + } + + itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx); + zilog_dirty(zilog, txg); + mutex_exit(&itxg->itxg_lock); + + /* Release the old itxs now we've dropped the lock */ + if (clean != NULL) + zil_itxg_clean(clean); +} + /* * If there are any in-memory intent log transactions which have now been - * synced then start up a taskq to free them. + * synced then start up a taskq to free them. We should only do this after we + * have written out the uberblocks (i.e. txg has been comitted) so that + * don't inadvertently clean out in-memory log records that would be required + * by zil_commit(). */ void -zil_clean(zilog_t *zilog) +zil_clean(zilog_t *zilog, uint64_t synced_txg) { - itx_t *itx; + itxg_t *itxg = &zilog->zl_itxg[synced_txg & TXG_MASK]; + itxs_t *clean_me; - mutex_enter(&zilog->zl_lock); - itx = list_head(&zilog->zl_itx_list); - if ((itx != NULL) && - (itx->itx_lr.lrc_txg <= spa_last_synced_txg(zilog->zl_spa))) { - (void) taskq_dispatch(zilog->zl_clean_taskq, - (task_func_t *)zil_itx_clean, zilog, TQ_NOSLEEP); + mutex_enter(&itxg->itxg_lock); + if (itxg->itxg_itxs == NULL || itxg->itxg_txg == ZILTEST_TXG) { + mutex_exit(&itxg->itxg_lock); + return; } - mutex_exit(&zilog->zl_lock); + ASSERT3U(itxg->itxg_txg, <=, synced_txg); + ASSERT(itxg->itxg_txg != 0); + ASSERT(zilog->zl_clean_taskq != NULL); + atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod); + itxg->itxg_sod = 0; + clean_me = itxg->itxg_itxs; + itxg->itxg_itxs = NULL; + itxg->itxg_txg = 0; + mutex_exit(&itxg->itxg_lock); + /* + * Preferably start a task queue to free up the old itxs but + * if taskq_dispatch can't allocate resources to do that then + * free it in-line. This should be rare. Note, using TQ_SLEEP + * created a bad performance problem. + */ + if (taskq_dispatch(zilog->zl_clean_taskq, + (void (*)(void *))zil_itxg_clean, clean_me, TQ_NOSLEEP) == 0) + zil_itxg_clean(clean_me); } +/* + * Get the list of itxs to commit into zl_itx_commit_list. + */ +static void +zil_get_commit_list(zilog_t *zilog) +{ + uint64_t otxg, txg; + list_t *commit_list = &zilog->zl_itx_commit_list; + uint64_t push_sod = 0; + + if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ + otxg = ZILTEST_TXG; + else + otxg = spa_last_synced_txg(zilog->zl_spa) + 1; + + for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { + itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; + + mutex_enter(&itxg->itxg_lock); + if (itxg->itxg_txg != txg) { + mutex_exit(&itxg->itxg_lock); + continue; + } + + list_move_tail(commit_list, &itxg->itxg_itxs->i_sync_list); + push_sod += itxg->itxg_sod; + itxg->itxg_sod = 0; + + mutex_exit(&itxg->itxg_lock); + } + atomic_add_64(&zilog->zl_itx_list_sz, -push_sod); +} + +/* + * Move the async itxs for a specified object to commit into sync lists. + */ static void -zil_commit_writer(zilog_t *zilog, uint64_t seq, uint64_t foid) +zil_async_to_sync(zilog_t *zilog, uint64_t foid) +{ + uint64_t otxg, txg; + itx_async_node_t *ian; + avl_tree_t *t; + avl_index_t where; + + if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ + otxg = ZILTEST_TXG; + else + otxg = spa_last_synced_txg(zilog->zl_spa) + 1; + + for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { + itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; + + mutex_enter(&itxg->itxg_lock); + if (itxg->itxg_txg != txg) { + mutex_exit(&itxg->itxg_lock); + continue; + } + + /* + * If a foid is specified then find that node and append its + * list. Otherwise walk the tree appending all the lists + * to the sync list. We add to the end rather than the + * beginning to ensure the create has happened. + */ + t = &itxg->itxg_itxs->i_async_tree; + if (foid != 0) { + ian = avl_find(t, &foid, &where); + if (ian != NULL) { + list_move_tail(&itxg->itxg_itxs->i_sync_list, + &ian->ia_list); + } + } else { + void *cookie = NULL; + + while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) { + list_move_tail(&itxg->itxg_itxs->i_sync_list, + &ian->ia_list); + list_destroy(&ian->ia_list); + kmem_free(ian, sizeof (itx_async_node_t)); + } + } + mutex_exit(&itxg->itxg_lock); + } +} + +static void +zil_commit_writer(zilog_t *zilog) { uint64_t txg; - uint64_t commit_seq = 0; - itx_t *itx, *itx_next; + itx_t *itx; lwb_t *lwb; - spa_t *spa; + spa_t *spa = zilog->zl_spa; int error = 0; - zilog->zl_writer = B_TRUE; ASSERT(zilog->zl_root_zio == NULL); - spa = zilog->zl_spa; + + mutex_exit(&zilog->zl_lock); + + zil_get_commit_list(zilog); + + /* + * Return if there's nothing to commit before we dirty the fs by + * calling zil_create(). + */ + if (list_head(&zilog->zl_itx_commit_list) == NULL) { + mutex_enter(&zilog->zl_lock); + return; + } if (zilog->zl_suspend) { lwb = NULL; } else { lwb = list_tail(&zilog->zl_lwb_list); - if (lwb == NULL) { - /* - * Return if there's nothing to flush before we - * dirty the fs by calling zil_create() - */ - if (list_is_empty(&zilog->zl_itx_list)) { - zilog->zl_writer = B_FALSE; - return; - } - mutex_exit(&zilog->zl_lock); + if (lwb == NULL) lwb = zil_create(zilog); - mutex_enter(&zilog->zl_lock); - } } - ASSERT(lwb == NULL || lwb->lwb_zio == NULL); - /* Loop through in-memory log transactions filling log blocks. */ DTRACE_PROBE1(zil__cw1, zilog_t *, zilog); - - for (itx = list_head(&zilog->zl_itx_list); itx; itx = itx_next) { - /* - * Save the next pointer. Even though we drop zl_lock below, - * all threads that can remove itx list entries (other writers - * and zil_itx_clean()) can't do so until they have zl_writer. - */ - itx_next = list_next(&zilog->zl_itx_list, itx); - - /* - * Determine whether to push this itx. - * Push all transactions related to specified foid and - * all other transactions except those that can be logged - * out of order (TX_WRITE, TX_TRUNCATE, TX_SETATTR, TX_ACL) - * for all other files. - * - * If foid == 0 (meaning "push all foids") or - * itx->itx_sync is set (meaning O_[D]SYNC), push regardless. - */ - if (foid != 0 && !itx->itx_sync && - TX_OOO(itx->itx_lr.lrc_txtype) && - ((lr_ooo_t *)&itx->itx_lr)->lr_foid != foid) - continue; /* skip this record */ - - if ((itx->itx_lr.lrc_seq > seq) && - ((lwb == NULL) || (LWB_EMPTY(lwb)) || - (lwb->lwb_nused + itx->itx_sod > lwb->lwb_sz))) - break; - - list_remove(&zilog->zl_itx_list, itx); - zilog->zl_itx_list_sz -= itx->itx_sod; - - mutex_exit(&zilog->zl_lock); - + while ((itx = list_head(&zilog->zl_itx_commit_list))) { txg = itx->itx_lr.lrc_txg; ASSERT(txg); - if (txg > spa_last_synced_txg(spa) || - txg > spa_freeze_txg(spa)) + if (txg > spa_last_synced_txg(spa) || txg > spa_freeze_txg(spa)) lwb = zil_lwb_commit(zilog, itx, lwb); - - zil_itx_destroy(itx); - - mutex_enter(&zilog->zl_lock); + list_remove(&zilog->zl_itx_commit_list, itx); + kmem_free(itx, offsetof(itx_t, itx_lr) + + itx->itx_lr.lrc_reclen); } DTRACE_PROBE1(zil__cw2, zilog_t *, zilog); - /* determine commit sequence number */ - itx = list_head(&zilog->zl_itx_list); - if (itx) - commit_seq = itx->itx_lr.lrc_seq - 1; - else - commit_seq = zilog->zl_itx_seq; - mutex_exit(&zilog->zl_lock); /* write the last block out */ if (lwb != NULL && lwb->lwb_zio != NULL) lwb = zil_lwb_write_start(zilog, lwb); - zilog->zl_prev_used = zilog->zl_cur_used; zilog->zl_cur_used = 0; /* * Wait if necessary for the log blocks to be on stable storage. */ if (zilog->zl_root_zio) { - DTRACE_PROBE1(zil__cw3, zilog_t *, zilog); error = zio_wait(zilog->zl_root_zio); zilog->zl_root_zio = NULL; - DTRACE_PROBE1(zil__cw4, zilog_t *, zilog); zil_flush_vdevs(zilog); } @@ -1246,10 +1559,6 @@ zil_commit_writer(zilog_t *zilog, uint64_t seq, uint64_t foid) txg_wait_synced(zilog->zl_dmu_pool, 0); mutex_enter(&zilog->zl_lock); - zilog->zl_writer = B_FALSE; - - ASSERT3U(commit_seq, >=, zilog->zl_commit_seq); - zilog->zl_commit_seq = commit_seq; /* * Remember the highest committed log sequence number for ztest. @@ -1261,58 +1570,65 @@ zil_commit_writer(zilog_t *zilog, uint64_t seq, uint64_t foid) } /* - * Push zfs transactions to stable storage up to the supplied sequence number. + * Commit zfs transactions to stable storage. * If foid is 0 push out all transactions, otherwise push only those - * for that file or might have been used to create that file. + * for that object or might reference that object. + * + * itxs are committed in batches. In a heavily stressed zil there will be + * a commit writer thread who is writing out a bunch of itxs to the log + * for a set of committing threads (cthreads) in the same batch as the writer. + * Those cthreads are all waiting on the same cv for that batch. + * + * There will also be a different and growing batch of threads that are + * waiting to commit (qthreads). When the committing batch completes + * a transition occurs such that the cthreads exit and the qthreads become + * cthreads. One of the new cthreads becomes the writer thread for the + * batch. Any new threads arriving become new qthreads. + * + * Only 2 condition variables are needed and there's no transition + * between the two cvs needed. They just flip-flop between qthreads + * and cthreads. + * + * Using this scheme we can efficiently wakeup up only those threads + * that have been committed. */ void -zil_commit(zilog_t *zilog, uint64_t seq, uint64_t foid) +zil_commit(zilog_t *zilog, uint64_t foid) { - if (zilog->zl_sync == ZFS_SYNC_DISABLED || seq == 0) + uint64_t mybatch; + + if (zilog->zl_sync == ZFS_SYNC_DISABLED) return; - mutex_enter(&zilog->zl_lock); + ZIL_STAT_BUMP(zil_commit_count); - seq = MIN(seq, zilog->zl_itx_seq); /* cap seq at largest itx seq */ + /* move the async itxs for the foid to the sync queues */ + zil_async_to_sync(zilog, foid); + mutex_enter(&zilog->zl_lock); + mybatch = zilog->zl_next_batch; while (zilog->zl_writer) { - cv_wait(&zilog->zl_cv_writer, &zilog->zl_lock); - if (seq <= zilog->zl_commit_seq) { + cv_wait(&zilog->zl_cv_batch[mybatch & 1], &zilog->zl_lock); + if (mybatch <= zilog->zl_com_batch) { mutex_exit(&zilog->zl_lock); return; } } - zil_commit_writer(zilog, seq, foid); /* drops zl_lock */ - /* wake up others waiting on the commit */ - cv_broadcast(&zilog->zl_cv_writer); - mutex_exit(&zilog->zl_lock); -} -/* - * Report whether all transactions are committed. - */ -static boolean_t -zil_is_committed(zilog_t *zilog) -{ - lwb_t *lwb; - boolean_t committed; - - mutex_enter(&zilog->zl_lock); + zilog->zl_next_batch++; + zilog->zl_writer = B_TRUE; + ZIL_STAT_BUMP(zil_commit_writer_count); + zil_commit_writer(zilog); + zilog->zl_com_batch = mybatch; + zilog->zl_writer = B_FALSE; - while (zilog->zl_writer) - cv_wait(&zilog->zl_cv_writer, &zilog->zl_lock); + /* wake up one thread to become the next writer */ + cv_signal(&zilog->zl_cv_batch[(mybatch+1) & 1]); - if (!list_is_empty(&zilog->zl_itx_list)) - committed = B_FALSE; /* unpushed transactions */ - else if ((lwb = list_head(&zilog->zl_lwb_list)) == NULL) - committed = B_TRUE; /* intent log never used */ - else if (list_next(&zilog->zl_lwb_list, lwb) != NULL) - committed = B_FALSE; /* zil_sync() not done yet */ - else - committed = B_TRUE; /* everything synced */ + /* wake up all threads waiting for this batch to be committed */ + cv_broadcast(&zilog->zl_cv_batch[mybatch & 1]); mutex_exit(&zilog->zl_lock); - return (committed); } /* @@ -1370,6 +1686,9 @@ zil_sync(zilog_t *zilog, dmu_tx_t *tx) zh->zh_log = lwb->lwb_blk; if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg) break; + + ASSERT(lwb->lwb_zio == NULL); + list_remove(&zilog->zl_lwb_list, lwb); zio_free_zil(spa, txg, &lwb->lwb_blk); kmem_cache_free(zil_lwb_cache, lwb); @@ -1383,6 +1702,19 @@ zil_sync(zilog_t *zilog, dmu_tx_t *tx) if (list_head(&zilog->zl_lwb_list) == NULL) BP_ZERO(&zh->zh_log); } + + /* + * Remove fastwrite on any blocks that have been pre-allocated for + * the next commit. This prevents fastwrite counter pollution by + * unused, long-lived LWBs. + */ + for (; lwb != NULL; lwb = list_next(&zilog->zl_lwb_list, lwb)) { + if (lwb->lwb_fastwrite && !lwb->lwb_zio) { + metaslab_fastwrite_unmark(zilog->zl_spa, &lwb->lwb_blk); + lwb->lwb_fastwrite = 0; + } + } + mutex_exit(&zilog->zl_lock); } @@ -1391,12 +1723,26 @@ zil_init(void) { zil_lwb_cache = kmem_cache_create("zil_lwb_cache", sizeof (struct lwb), 0, NULL, NULL, NULL, NULL, NULL, 0); + + zil_ksp = kstat_create("zfs", 0, "zil", "misc", + KSTAT_TYPE_NAMED, sizeof(zil_stats) / sizeof(kstat_named_t), + KSTAT_FLAG_VIRTUAL); + + if (zil_ksp != NULL) { + zil_ksp->ks_data = &zil_stats; + kstat_install(zil_ksp); + } } void zil_fini(void) { kmem_cache_destroy(zil_lwb_cache); + + if (zil_ksp != NULL) { + kstat_delete(zil_ksp); + zil_ksp = NULL; + } } void @@ -1415,8 +1761,9 @@ zilog_t * zil_alloc(objset_t *os, zil_header_t *zh_phys) { zilog_t *zilog; + int i; - zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP); + zilog = kmem_zalloc(sizeof (zilog_t), KM_PUSHPAGE); zilog->zl_header = zh_phys; zilog->zl_os = os; @@ -1425,15 +1772,21 @@ zil_alloc(objset_t *os, zil_header_t *zh_phys) zilog->zl_destroy_txg = TXG_INITIAL - 1; zilog->zl_logbias = dmu_objset_logbias(os); zilog->zl_sync = dmu_objset_syncprop(os); + zilog->zl_next_batch = 1; mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL); - list_create(&zilog->zl_itx_list, sizeof (itx_t), - offsetof(itx_t, itx_node)); + for (i = 0; i < TXG_SIZE; i++) { + mutex_init(&zilog->zl_itxg[i].itxg_lock, NULL, + MUTEX_DEFAULT, NULL); + } list_create(&zilog->zl_lwb_list, sizeof (lwb_t), offsetof(lwb_t, lwb_node)); + list_create(&zilog->zl_itx_commit_list, sizeof (itx_t), + offsetof(itx_t, itx_node)); + mutex_init(&zilog->zl_vdev_lock, NULL, MUTEX_DEFAULT, NULL); avl_create(&zilog->zl_vdev_tree, zil_vdev_compare, @@ -1441,6 +1794,8 @@ zil_alloc(objset_t *os, zil_header_t *zh_phys) cv_init(&zilog->zl_cv_writer, NULL, CV_DEFAULT, NULL); cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL); + cv_init(&zilog->zl_cv_batch[0], NULL, CV_DEFAULT, NULL); + cv_init(&zilog->zl_cv_batch[1], NULL, CV_DEFAULT, NULL); return (zilog); } @@ -1448,27 +1803,38 @@ zil_alloc(objset_t *os, zil_header_t *zh_phys) void zil_free(zilog_t *zilog) { - lwb_t *lwb; + int i; zilog->zl_stop_sync = 1; - while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { - list_remove(&zilog->zl_lwb_list, lwb); - if (lwb->lwb_buf != NULL) - zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); - kmem_cache_free(zil_lwb_cache, lwb); - } + ASSERT(list_is_empty(&zilog->zl_lwb_list)); list_destroy(&zilog->zl_lwb_list); avl_destroy(&zilog->zl_vdev_tree); mutex_destroy(&zilog->zl_vdev_lock); - ASSERT(list_head(&zilog->zl_itx_list) == NULL); - list_destroy(&zilog->zl_itx_list); + ASSERT(list_is_empty(&zilog->zl_itx_commit_list)); + list_destroy(&zilog->zl_itx_commit_list); + + for (i = 0; i < TXG_SIZE; i++) { + /* + * It's possible for an itx to be generated that doesn't dirty + * a txg (e.g. ztest TX_TRUNCATE). So there's no zil_clean() + * callback to remove the entry. We remove those here. + * + * Also free up the ziltest itxs. + */ + if (zilog->zl_itxg[i].itxg_itxs) + zil_itxg_clean(zilog->zl_itxg[i].itxg_itxs); + mutex_destroy(&zilog->zl_itxg[i].itxg_lock); + } + mutex_destroy(&zilog->zl_lock); cv_destroy(&zilog->zl_cv_writer); cv_destroy(&zilog->zl_cv_suspend); + cv_destroy(&zilog->zl_cv_batch[0]); + cv_destroy(&zilog->zl_cv_batch[1]); kmem_free(zilog, sizeof (zilog_t)); } @@ -1481,6 +1847,10 @@ zil_open(objset_t *os, zil_get_data_t *get_data) { zilog_t *zilog = dmu_objset_zil(os); + ASSERT(zilog->zl_clean_taskq == NULL); + ASSERT(zilog->zl_get_data == NULL); + ASSERT(list_is_empty(&zilog->zl_lwb_list)); + zilog->zl_get_data = get_data; zilog->zl_clean_taskq = taskq_create("zil_clean", 1, minclsyspri, 2, 2, TASKQ_PREPOPULATE); @@ -1494,26 +1864,45 @@ zil_open(objset_t *os, zil_get_data_t *get_data) void zil_close(zilog_t *zilog) { + lwb_t *lwb; + uint64_t txg = 0; + + zil_commit(zilog, 0); /* commit all itx */ + /* - * If the log isn't already committed, mark the objset dirty - * (so zil_sync() will be called) and wait for that txg to sync. + * The lwb_max_txg for the stubby lwb will reflect the last activity + * for the zil. After a txg_wait_synced() on the txg we know all the + * callbacks have occurred that may clean the zil. Only then can we + * destroy the zl_clean_taskq. */ - if (!zil_is_committed(zilog)) { - uint64_t txg; - dmu_tx_t *tx = dmu_tx_create(zilog->zl_os); - VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0); - dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); - txg = dmu_tx_get_txg(tx); - dmu_tx_commit(tx); + mutex_enter(&zilog->zl_lock); + lwb = list_tail(&zilog->zl_lwb_list); + if (lwb != NULL) + txg = lwb->lwb_max_txg; + mutex_exit(&zilog->zl_lock); + if (txg) txg_wait_synced(zilog->zl_dmu_pool, txg); - } + ASSERT(!zilog_is_dirty(zilog)); taskq_destroy(zilog->zl_clean_taskq); zilog->zl_clean_taskq = NULL; zilog->zl_get_data = NULL; - zil_itx_clean(zilog); - ASSERT(list_head(&zilog->zl_itx_list) == NULL); + /* + * We should have only one LWB left on the list; remove it now. + */ + mutex_enter(&zilog->zl_lock); + lwb = list_head(&zilog->zl_lwb_list); + if (lwb != NULL) { + ASSERT(lwb == list_tail(&zilog->zl_lwb_list)); + ASSERT(lwb->lwb_zio == NULL); + if (lwb->lwb_fastwrite) + metaslab_fastwrite_unmark(zilog->zl_spa, &lwb->lwb_blk); + list_remove(&zilog->zl_lwb_list, lwb); + zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); + kmem_cache_free(zil_lwb_cache, lwb); + } + mutex_exit(&zilog->zl_lock); } /* @@ -1545,15 +1934,7 @@ zil_suspend(zilog_t *zilog) zilog->zl_suspending = B_TRUE; mutex_exit(&zilog->zl_lock); - zil_commit(zilog, UINT64_MAX, 0); - - /* - * Wait for any in-flight log writes to complete. - */ - mutex_enter(&zilog->zl_lock); - while (zilog->zl_writer) - cv_wait(&zilog->zl_cv_writer, &zilog->zl_lock); - mutex_exit(&zilog->zl_lock); + zil_commit(zilog, 0); zil_destroy(zilog, B_FALSE); @@ -1575,7 +1956,7 @@ zil_resume(zilog_t *zilog) } typedef struct zil_replay_arg { - zil_replay_func_t **zr_replay; + zil_replay_func_t *zr_replay; void *zr_arg; boolean_t zr_byteswap; char *zr_lr; @@ -1694,7 +2075,7 @@ zil_incr_blks(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) * If this dataset has a non-empty intent log, replay it and destroy it. */ void -zil_replay(objset_t *os, void *arg, zil_replay_func_t *replay_func[TX_MAX_TYPE]) +zil_replay(objset_t *os, void *arg, zil_replay_func_t replay_func[TX_MAX_TYPE]) { zilog_t *zilog = dmu_objset_zil(os); const zil_header_t *zh = zilog->zl_header; @@ -1708,7 +2089,7 @@ zil_replay(objset_t *os, void *arg, zil_replay_func_t *replay_func[TX_MAX_TYPE]) zr.zr_replay = replay_func; zr.zr_arg = arg; zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log); - zr.zr_lr = kmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP); + zr.zr_lr = vmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_PUSHPAGE); /* * Wait for in-progress removes to sync before starting replay. @@ -1720,7 +2101,7 @@ zil_replay(objset_t *os, void *arg, zil_replay_func_t *replay_func[TX_MAX_TYPE]) ASSERT(zilog->zl_replay_blks == 0); (void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr, zh->zh_claim_txg); - kmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE); + vmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE); zil_destroy(zilog, B_FALSE); txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); @@ -1763,3 +2144,14 @@ zil_vdev_offline(const char *osname, void *arg) dmu_objset_rele(os, FTAG); return (error); } + +#if defined(_KERNEL) && defined(HAVE_SPL) +module_param(zil_replay_disable, int, 0644); +MODULE_PARM_DESC(zil_replay_disable, "Disable intent logging replay"); + +module_param(zfs_nocacheflush, int, 0644); +MODULE_PARM_DESC(zfs_nocacheflush, "Disable cache flushes"); + +module_param(zil_slog_limit, ulong, 0644); +MODULE_PARM_DESC(zil_slog_limit, "Max commit bytes to separate log device"); +#endif