Date: Thu, 17 Nov 2016 14:27:16 -0800 From: Alexander Motin <mav@FreeBSD.org> To: Steven Hartland <steven.hartland@multiplay.co.uk>, src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r308782 - in head: cddl/contrib/opensolaris/cmd/ztest sys/cddl/contrib/opensolaris/uts/common/fs/zfs sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys Message-ID: <71f021c6-5847-5808-20d2-4cc84925af46@FreeBSD.org> In-Reply-To: <7675c768-c1b1-108b-b651-28b6184f8176@multiplay.co.uk> References: <201611172101.uAHL1R2O009871@repo.freebsd.org> <79c521bc-3186-10a5-d553-12072523f439@multiplay.co.uk> <725f293e-3d3b-1248-4900-acee5659d382@FreeBSD.org> <7675c768-c1b1-108b-b651-28b6184f8176@multiplay.co.uk>
next in thread | previous in thread | raw e-mail | index | archive | help
Rebased, but that is not the first time. What it really needs is reviews. On 17.11.2016 14:22, Steven Hartland wrote: > Thanks, looks like the PR needs a rebase before it can be merged. > > On 17/11/2016 22:11, Alexander Motin wrote: >> It is in OpenZFS review queue now: >> https://github.com/openzfs/openzfs/pull/219 Welcome to comment there to >> speed up the process. >> >> On 17.11.2016 13:43, Steven Hartland wrote: >>> Is this something that should be upstreamed? >>> >>> On 17/11/2016 21:01, Alexander Motin wrote: >>>> Author: mav >>>> Date: Thu Nov 17 21:01:27 2016 >>>> New Revision: 308782 >>>> URL: https://svnweb.freebsd.org/changeset/base/308782 >>>> >>>> Log: >>>> After some ZIL changes 6 years ago zil_slog_limit got partially broken >>>> due to zl_itx_list_sz not updated when async itx'es upgraded to sync. >>>> Actually because of other changes about that time zl_itx_list_sz is not >>>> really required to implement the functionality, so this patch removes >>>> some unneeded broken code and variables. >>>> >>>> Original idea of zil_slog_limit was to reduce chance of SLOG abuse by >>>> single heavy logger, that increased latency for other (more latency critical) >>>> loggers, by pushing heavy log out into the main pool instead of SLOG. Beside >>>> huge latency increase for heavy writers, this implementation caused double >>>> write of all data, since the log records were explicitly prepared for SLOG. >>>> Since we now have I/O scheduler, I've found it can be much more efficient >>>> to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE >>>> to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. >>>> >>>> Existing ZIL implementation had problem with space efficiency when it >>>> has to write large chunks of data into log blocks of limited size. In some >>>> cases efficiency stopped to almost as low as 50%. In case of ZIL stored on >>>> spinning rust, that also reduced log write speed in half, since head had to >>>> uselessly fly over allocated but not written areas. This change improves >>>> the situation by offloading problematic operations from z*_log_write() to >>>> zil_lwb_commit(), which knows real situation of log blocks allocation and >>>> can split large requests into pieces much more efficiently. Also as side >>>> effect it removes one of two data copy operations done by ZIL code WR_COPIED >>>> case. >>>> >>>> While there, untangle and unify code of z*_log_write() functions. >>>> Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing >>>> block boundary, that may also improve efficiency if ZPL is made to do that. >>>> >>>> Sponsored by: iXsystems, Inc. >>>> >>>> Modified: >>>> head/cddl/contrib/opensolaris/cmd/ztest/ztest.c >>>> head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zil.h >>>> head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zil_impl.h >>>> head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h >>>> head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_log.c >>>> head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c >>>> head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c >>>> head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c >>>> >>>> Modified: head/cddl/contrib/opensolaris/cmd/ztest/ztest.c >>>> ============================================================================== >>>> --- head/cddl/contrib/opensolaris/cmd/ztest/ztest.c Thu Nov 17 20:44:51 2016 (r308781) >>>> +++ head/cddl/contrib/opensolaris/cmd/ztest/ztest.c Thu Nov 17 21:01:27 2016 (r308782) >>>> @@ -1371,7 +1371,6 @@ ztest_log_write(ztest_ds_t *zd, dmu_tx_t >>>> itx->itx_private = zd; >>>> itx->itx_wr_state = write_state; >>>> itx->itx_sync = (ztest_random(8) == 0); >>>> - itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0); >>>> >>>> bcopy(&lr->lr_common + 1, &itx->itx_lr + 1, >>>> sizeof (*lr) - sizeof (lr_t)); >>>> >>>> Modified: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zil.h >>>> ============================================================================== >>>> --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zil.h Thu Nov 17 20:44:51 2016 (r308781) >>>> +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zil.h Thu Nov 17 21:01:27 2016 (r308782) >>>> @@ -369,7 +369,6 @@ typedef struct itx { >>>> void *itx_private; /* type-specific opaque data */ >>>> itx_wr_state_t itx_wr_state; /* write state */ >>>> uint8_t itx_sync; /* synchronous transaction */ >>>> - uint64_t itx_sod; /* record size on disk */ >>>> uint64_t itx_oid; /* object id */ >>>> lr_t itx_lr; /* common part of log record */ >>>> /* followed by type-specific part of lr_xx_t and its immediate data */ >>>> >>>> Modified: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zil_impl.h >>>> ============================================================================== >>>> --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zil_impl.h Thu Nov 17 20:44:51 2016 (r308781) >>>> +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zil_impl.h Thu Nov 17 21:01:27 2016 (r308782) >>>> @@ -42,6 +42,7 @@ extern "C" { >>>> typedef struct lwb { >>>> zilog_t *lwb_zilog; /* back pointer to log struct */ >>>> blkptr_t lwb_blk; /* on disk address of this log blk */ >>>> + boolean_t lwb_slog; /* lwb_blk is on SLOG device */ >>>> int lwb_nused; /* # used bytes in buffer */ >>>> int lwb_sz; /* size of block and buffer */ >>>> char *lwb_buf; /* log write buffer */ >>>> @@ -62,7 +63,6 @@ typedef struct itxs { >>>> typedef struct itxg { >>>> kmutex_t itxg_lock; /* lock for this structure */ >>>> uint64_t itxg_txg; /* txg for this chain */ >>>> - uint64_t itxg_sod; /* total size on disk for this txg */ >>>> itxs_t *itxg_itxs; /* sync and async itxs */ >>>> } itxg_t; >>>> >>>> @@ -120,7 +120,6 @@ struct zilog { >>>> kcondvar_t zl_cv_batch[2]; /* batch condition variables */ >>>> itxg_t zl_itxg[TXG_SIZE]; /* intent log txg chains */ >>>> list_t zl_itx_commit_list; /* itx list to be committed */ >>>> - uint64_t zl_itx_list_sz; /* total size of records on list */ >>>> uint64_t zl_cur_used; /* current commit log size used */ >>>> list_t zl_lwb_list; /* in-flight log write list */ >>>> kmutex_t zl_vdev_lock; /* protects zl_vdev_tree */ >>>> @@ -142,6 +141,8 @@ typedef struct zil_bp_node { >>>> >>>> #define ZIL_MAX_LOG_DATA (SPA_OLD_MAXBLOCKSIZE - sizeof (zil_chain_t) - \ >>>> sizeof (lr_write_t)) >>>> +#define ZIL_MAX_COPIED_DATA \ >>>> + ((SPA_OLD_MAXBLOCKSIZE - sizeof (zil_chain_t)) / 2 - sizeof (lr_write_t)) >>>> >>>> #ifdef __cplusplus >>>> } >>>> >>>> Modified: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h >>>> ============================================================================== >>>> --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h Thu Nov 17 20:44:51 2016 (r308781) >>>> +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h Thu Nov 17 21:01:27 2016 (r308782) >>>> @@ -547,7 +547,7 @@ extern zio_t *zio_free_sync(zio_t *pio, >>>> const blkptr_t *bp, uint64_t size, enum zio_flag flags); >>>> >>>> extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, >>>> - blkptr_t *old_bp, uint64_t size, boolean_t use_slog); >>>> + blkptr_t *old_bp, uint64_t size, boolean_t *slog); >>>> extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp); >>>> extern void zio_flush(zio_t *zio, vdev_t *vd); >>>> extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset, >>>> >>>> Modified: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_log.c >>>> ============================================================================== >>>> --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_log.c Thu Nov 17 20:44:51 2016 (r308781) >>>> +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_log.c Thu Nov 17 21:01:27 2016 (r308782) >>>> @@ -464,20 +464,17 @@ void >>>> zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype, >>>> znode_t *zp, offset_t off, ssize_t resid, int ioflag) >>>> { >>>> + uint32_t blocksize = zp->z_blksz; >>>> itx_wr_state_t write_state; >>>> - boolean_t slogging; >>>> uintptr_t fsync_cnt; >>>> - ssize_t immediate_write_sz; >>>> >>>> if (zil_replaying(zilog, tx) || zp->z_unlinked) >>>> return; >>>> >>>> - immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) >>>> - ? 0 : zfs_immediate_write_sz; >>>> - >>>> - slogging = spa_has_slogs(zilog->zl_spa) && >>>> - (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); >>>> - if (resid > immediate_write_sz && !slogging && resid <= zp->z_blksz) >>>> + if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) >>>> + write_state = WR_INDIRECT; >>>> + else if (!spa_has_slogs(zilog->zl_spa) && >>>> + resid >= zfs_immediate_write_sz) >>>> write_state = WR_INDIRECT; >>>> else if (ioflag & (FSYNC | FDSYNC)) >>>> write_state = WR_COPIED; >>>> @@ -491,30 +488,26 @@ zfs_log_write(zilog_t *zilog, dmu_tx_t * >>>> while (resid) { >>>> itx_t *itx; >>>> lr_write_t *lr; >>>> - ssize_t len; >>>> + itx_wr_state_t wr_state = write_state; >>>> + ssize_t len = resid; >>>> >>>> - /* >>>> - * If the write would overflow the largest block then split it. >>>> - */ >>>> - if (write_state != WR_INDIRECT && resid > ZIL_MAX_LOG_DATA) >>>> - len = SPA_OLD_MAXBLOCKSIZE >> 1; >>>> - else >>>> - len = resid; >>>> + if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA) >>>> + wr_state = WR_NEED_COPY; >>>> + else if (wr_state == WR_INDIRECT) >>>> + len = MIN(blocksize - P2PHASE(off, blocksize), resid); >>>> >>>> itx = zil_itx_create(txtype, sizeof (*lr) + >>>> - (write_state == WR_COPIED ? len : 0)); >>>> + (wr_state == WR_COPIED ? len : 0)); >>>> lr = (lr_write_t *)&itx->itx_lr; >>>> - if (write_state == WR_COPIED && dmu_read(zp->z_zfsvfs->z_os, >>>> + if (wr_state == WR_COPIED && dmu_read(zp->z_zfsvfs->z_os, >>>> zp->z_id, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { >>>> zil_itx_destroy(itx); >>>> itx = zil_itx_create(txtype, sizeof (*lr)); >>>> lr = (lr_write_t *)&itx->itx_lr; >>>> - write_state = WR_NEED_COPY; >>>> + wr_state = WR_NEED_COPY; >>>> } >>>> >>>> - itx->itx_wr_state = write_state; >>>> - if (write_state == WR_NEED_COPY) >>>> - itx->itx_sod += len; >>>> + itx->itx_wr_state = wr_state; >>>> lr->lr_foid = zp->z_id; >>>> lr->lr_offset = off; >>>> lr->lr_length = len; >>>> >>>> Modified: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c >>>> ============================================================================== >>>> --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c Thu Nov 17 20:44:51 2016 (r308781) >>>> +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c Thu Nov 17 21:01:27 2016 (r308782) >>>> @@ -88,6 +88,15 @@ SYSCTL_DECL(_vfs_zfs_trim); >>>> SYSCTL_INT(_vfs_zfs_trim, OID_AUTO, enabled, CTLFLAG_RDTUN, &zfs_trim_enabled, 0, >>>> "Enable ZFS TRIM"); >>>> >>>> +/* >>>> + * Limit SLOG write size per commit executed with synchronous priority. >>>> + * Any writes above that executed with lower (asynchronous) priority to >>>> + * limit potential SLOG device abuse by single active ZIL writer. >>>> + */ >>>> +uint64_t zil_slog_limit = 768 * 1024; >>>> +SYSCTL_QUAD(_vfs_zfs, OID_AUTO, zil_slog_limit, CTLFLAG_RWTUN, >>>> + &zil_slog_limit, 0, "Maximal SLOG commit size with sync priority"); >>>> + >>>> static kmem_cache_t *zil_lwb_cache; >>>> >>>> #define LWB_EMPTY(lwb) ((BP_GET_LSIZE(&lwb->lwb_blk) - \ >>>> @@ -447,13 +456,14 @@ zil_free_log_record(zilog_t *zilog, lr_t >>>> } >>>> >>>> static lwb_t * >>>> -zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, uint64_t txg) >>>> +zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, boolean_t slog, uint64_t txg) >>>> { >>>> lwb_t *lwb; >>>> >>>> lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP); >>>> lwb->lwb_zilog = zilog; >>>> lwb->lwb_blk = *bp; >>>> + lwb->lwb_slog = slog; >>>> lwb->lwb_buf = zio_buf_alloc(BP_GET_LSIZE(bp)); >>>> lwb->lwb_max_txg = txg; >>>> lwb->lwb_zio = NULL; >>>> @@ -516,6 +526,7 @@ zil_create(zilog_t *zilog) >>>> dmu_tx_t *tx = NULL; >>>> blkptr_t blk; >>>> int error = 0; >>>> + boolean_t slog = FALSE; >>>> >>>> /* >>>> * Wait for any previous destroy to complete. >>>> @@ -544,7 +555,7 @@ zil_create(zilog_t *zilog) >>>> } >>>> >>>> error = zio_alloc_zil(zilog->zl_spa, txg, &blk, NULL, >>>> - ZIL_MIN_BLKSZ, zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); >>>> + ZIL_MIN_BLKSZ, &slog); >>>> >>>> if (error == 0) >>>> zil_init_log_chain(zilog, &blk); >>>> @@ -554,7 +565,7 @@ 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, slog, txg); >>>> >>>> /* >>>> * If we just allocated the first log block, commit our transaction >>>> @@ -885,6 +896,7 @@ static void >>>> zil_lwb_write_init(zilog_t *zilog, lwb_t *lwb) >>>> { >>>> zbookmark_phys_t zb; >>>> + zio_priority_t prio; >>>> >>>> SET_BOOKMARK(&zb, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET], >>>> ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, >>>> @@ -895,9 +907,13 @@ zil_lwb_write_init(zilog_t *zilog, lwb_t >>>> ZIO_FLAG_CANFAIL); >>>> } >>>> if (lwb->lwb_zio == NULL) { >>>> + if (zilog->zl_cur_used <= zil_slog_limit || !lwb->lwb_slog) >>>> + prio = ZIO_PRIORITY_SYNC_WRITE; >>>> + else >>>> + prio = ZIO_PRIORITY_ASYNC_WRITE; >>>> 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_SYNC_WRITE, >>>> + zil_lwb_write_done, lwb, prio, >>>> ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE, &zb); >>>> } >>>> } >>>> @@ -917,18 +933,6 @@ 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. >>>> - */ >>>> -uint64_t zil_slog_limit = 1024 * 1024; >>>> -SYSCTL_QUAD(_vfs_zfs, OID_AUTO, zil_slog_limit, CTLFLAG_RWTUN, >>>> - &zil_slog_limit, 0, "Maximal commit size to use SLOG"); >>>> -#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)))) >>>> - >>>> -/* >>>> * Start a log block write and advance to the next log block. >>>> * Calls are serialized. >>>> */ >>>> @@ -943,6 +947,7 @@ zil_lwb_write_start(zilog_t *zilog, lwb_ >>>> uint64_t txg; >>>> uint64_t zil_blksz, wsz; >>>> int i, error; >>>> + boolean_t slog; >>>> >>>> if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) { >>>> zilc = (zil_chain_t *)lwb->lwb_buf; >>>> @@ -999,8 +1004,7 @@ zil_lwb_write_start(zilog_t *zilog, lwb_ >>>> >>>> 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)); >>>> + error = zio_alloc_zil(spa, txg, bp, &lwb->lwb_blk, zil_blksz, &slog); >>>> if (error == 0) { >>>> ASSERT3U(bp->blk_birth, ==, txg); >>>> bp->blk_cksum = lwb->lwb_blk.blk_cksum; >>>> @@ -1009,7 +1013,7 @@ zil_lwb_write_start(zilog_t *zilog, lwb_ >>>> /* >>>> * Allocate a new log write buffer (lwb). >>>> */ >>>> - nlwb = zil_alloc_lwb(zilog, bp, txg); >>>> + nlwb = zil_alloc_lwb(zilog, bp, slog, txg); >>>> >>>> /* Record the block for later vdev flushing */ >>>> zil_add_block(zilog, &lwb->lwb_blk); >>>> @@ -1046,12 +1050,13 @@ zil_lwb_write_start(zilog_t *zilog, lwb_ >>>> static lwb_t * >>>> zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) >>>> { >>>> - lr_t *lrc = &itx->itx_lr; /* common log record */ >>>> - lr_write_t *lrw = (lr_write_t *)lrc; >>>> + lr_t *lrcb, *lrc = &itx->itx_lr; /* common log record */ >>>> + lr_write_t *lrwb, *lrw = (lr_write_t *)lrc; >>>> char *lr_buf; >>>> uint64_t txg = lrc->lrc_txg; >>>> uint64_t reclen = lrc->lrc_reclen; >>>> uint64_t dlen = 0; >>>> + uint64_t dnow, lwb_sp; >>>> >>>> if (lwb == NULL) >>>> return (NULL); >>>> @@ -1068,25 +1073,30 @@ zil_lwb_commit(zilog_t *zilog, itx_t *it >>>> >>>> zil_lwb_write_init(zilog, lwb); >>>> >>>> +cont: >>>> /* >>>> * If this record won't fit in the current log block, start a new one. >>>> + * For WR_NEED_COPY optimize layout for minimal number of chunks, but >>>> + * try to keep wasted space withing reasonable range (12%). >>>> */ >>>> - if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) { >>>> + lwb_sp = lwb->lwb_sz - lwb->lwb_nused; >>>> + if (reclen > lwb_sp || (reclen + dlen > lwb_sp && >>>> + lwb_sp < ZIL_MAX_LOG_DATA / 8 && (dlen % ZIL_MAX_LOG_DATA == 0 || >>>> + lwb_sp < reclen + dlen % ZIL_MAX_LOG_DATA))) { >>>> lwb = zil_lwb_write_start(zilog, lwb); >>>> if (lwb == NULL) >>>> return (NULL); >>>> zil_lwb_write_init(zilog, lwb); >>>> ASSERT(LWB_EMPTY(lwb)); >>>> - if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) { >>>> - txg_wait_synced(zilog->zl_dmu_pool, txg); >>>> - return (lwb); >>>> - } >>>> + lwb_sp = lwb->lwb_sz - lwb->lwb_nused; >>>> + ASSERT3U(reclen + MIN(dlen, sizeof(uint64_t)), <=, lwb_sp); >>>> } >>>> >>>> + dnow = MIN(dlen, lwb_sp - reclen); >>>> lr_buf = lwb->lwb_buf + lwb->lwb_nused; >>>> bcopy(lrc, lr_buf, reclen); >>>> - lrc = (lr_t *)lr_buf; >>>> - lrw = (lr_write_t *)lrc; >>>> + lrcb = (lr_t *)lr_buf; >>>> + lrwb = (lr_write_t *)lrcb; >>>> >>>> /* >>>> * If it's a write, fetch the data or get its blkptr as appropriate. >>>> @@ -1098,16 +1108,19 @@ zil_lwb_commit(zilog_t *zilog, itx_t *it >>>> char *dbuf; >>>> int error; >>>> >>>> - if (dlen) { >>>> - ASSERT(itx->itx_wr_state == WR_NEED_COPY); >>>> + if (itx->itx_wr_state == WR_NEED_COPY) { >>>> dbuf = lr_buf + reclen; >>>> - lrw->lr_common.lrc_reclen += dlen; >>>> + lrcb->lrc_reclen += dnow; >>>> + if (lrwb->lr_length > dnow) >>>> + lrwb->lr_length = dnow; >>>> + lrw->lr_offset += dnow; >>>> + lrw->lr_length -= dnow; >>>> } else { >>>> ASSERT(itx->itx_wr_state == WR_INDIRECT); >>>> dbuf = NULL; >>>> } >>>> error = zilog->zl_get_data( >>>> - itx->itx_private, lrw, dbuf, lwb->lwb_zio); >>>> + itx->itx_private, lrwb, dbuf, lwb->lwb_zio); >>>> if (error == EIO) { >>>> txg_wait_synced(zilog->zl_dmu_pool, txg); >>>> return (lwb); >>>> @@ -1126,12 +1139,18 @@ zil_lwb_commit(zilog_t *zilog, itx_t *it >>>> * equal to the itx sequence number because not all transactions >>>> * are synchronous, and sometimes spa_sync() gets there first. >>>> */ >>>> - lrc->lrc_seq = ++zilog->zl_lr_seq; /* we are single threaded */ >>>> - lwb->lwb_nused += reclen + dlen; >>>> + lrcb->lrc_seq = ++zilog->zl_lr_seq; /* we are single threaded */ >>>> + lwb->lwb_nused += reclen + dnow; >>>> lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg); >>>> ASSERT3U(lwb->lwb_nused, <=, lwb->lwb_sz); >>>> ASSERT0(P2PHASE(lwb->lwb_nused, sizeof (uint64_t))); >>>> >>>> + dlen -= dnow; >>>> + if (dlen > 0) { >>>> + zilog->zl_cur_used += reclen; >>>> + goto cont; >>>> + } >>>> + >>>> return (lwb); >>>> } >>>> >>>> @@ -1145,7 +1164,6 @@ zil_itx_create(uint64_t txtype, size_t l >>>> itx = kmem_alloc(offsetof(itx_t, itx_lr) + lrsize, KM_SLEEP); >>>> 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 */ >>>> >>>> @@ -1294,11 +1312,8 @@ zil_itx_assign(zilog_t *zilog, itx_t *it >>>> * 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_SLEEP); >>>> >>>> @@ -1310,8 +1325,6 @@ zil_itx_assign(zilog_t *zilog, itx_t *it >>>> } >>>> 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; >>>> @@ -1359,8 +1372,6 @@ zil_clean(zilog_t *zilog, uint64_t synce >>>> 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; >>>> @@ -1384,7 +1395,6 @@ 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; >>>> @@ -1401,12 +1411,9 @@ zil_get_commit_list(zilog_t *zilog) >>>> } >>>> >>>> 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); >>>> } >>>> >>>> /* >>>> >>>> Modified: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c >>>> ============================================================================== >>>> --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c Thu Nov 17 20:44:51 2016 (r308781) >>>> +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c Thu Nov 17 21:01:27 2016 (r308782) >>>> @@ -2908,20 +2908,21 @@ zio_dva_unallocate(zio_t *zio, zio_gang_ >>>> */ >>>> int >>>> zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, blkptr_t *old_bp, >>>> - uint64_t size, boolean_t use_slog) >>>> + uint64_t size, boolean_t *slog) >>>> { >>>> int error = 1; >>>> >>>> ASSERT(txg > spa_syncing_txg(spa)); >>>> >>>> - if (use_slog) { >>>> - error = metaslab_alloc(spa, spa_log_class(spa), size, >>>> - new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID, NULL); >>>> - } >>>> - >>>> - if (error) { >>>> + error = metaslab_alloc(spa, spa_log_class(spa), size, >>>> + new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID, NULL); >>>> + if (error == 0) { >>>> + *slog = TRUE; >>>> + } else { >>>> error = metaslab_alloc(spa, spa_normal_class(spa), size, >>>> new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID, NULL); >>>> + if (error == 0) >>>> + *slog = FALSE; >>>> } >>>> >>>> if (error == 0) { >>>> >>>> Modified: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c >>>> ============================================================================== >>>> --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c Thu Nov 17 20:44:51 2016 (r308781) >>>> +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c Thu Nov 17 21:01:27 2016 (r308782) >>>> @@ -1387,54 +1387,44 @@ zvol_log_write(zvol_state_t *zv, dmu_tx_ >>>> { >>>> uint32_t blocksize = zv->zv_volblocksize; >>>> zilog_t *zilog = zv->zv_zilog; >>>> - boolean_t slogging; >>>> - ssize_t immediate_write_sz; >>>> + itx_wr_state_t write_state; >>>> >>>> if (zil_replaying(zilog, tx)) >>>> return; >>>> >>>> - immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) >>>> - ? 0 : zvol_immediate_write_sz; >>>> - >>>> - slogging = spa_has_slogs(zilog->zl_spa) && >>>> - (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); >>>> + if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) >>>> + write_state = WR_INDIRECT; >>>> + else if (!spa_has_slogs(zilog->zl_spa) && >>>> + resid >= blocksize && blocksize > zvol_immediate_write_sz) >>>> + write_state = WR_INDIRECT; >>>> + else if (sync) >>>> + write_state = WR_COPIED; >>>> + else >>>> + write_state = WR_NEED_COPY; >>>> >>>> while (resid) { >>>> itx_t *itx; >>>> lr_write_t *lr; >>>> - ssize_t len; >>>> - itx_wr_state_t write_state; >>>> + itx_wr_state_t wr_state = write_state; >>>> + ssize_t len = resid; >>>> >>>> - /* >>>> - * Unlike zfs_log_write() we can be called with >>>> - * upto DMU_MAX_ACCESS/2 (5MB) writes. >>>> - */ >>>> - if (blocksize > immediate_write_sz && !slogging && >>>> - resid >= blocksize && off % blocksize == 0) { >>>> - write_state = WR_INDIRECT; /* uses dmu_sync */ >>>> - len = blocksize; >>>> - } else if (sync) { >>>> - write_state = WR_COPIED; >>>> - len = MIN(ZIL_MAX_LOG_DATA, resid); >>>> - } else { >>>> - write_state = WR_NEED_COPY; >>>> - len = MIN(ZIL_MAX_LOG_DATA, resid); >>>> - } >>>> + if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA) >>>> + wr_state = WR_NEED_COPY; >>>> + else if (wr_state == WR_INDIRECT) >>>> + len = MIN(blocksize - P2PHASE(off, blocksize), resid); >>>> >>>> itx = zil_itx_create(TX_WRITE, sizeof (*lr) + >>>> - (write_state == WR_COPIED ? len : 0)); >>>> + (wr_state == WR_COPIED ? len : 0)); >>>> lr = (lr_write_t *)&itx->itx_lr; >>>> - if (write_state == WR_COPIED && dmu_read(zv->zv_objset, >>>> + if (wr_state == WR_COPIED && dmu_read(zv->zv_objset, >>>> ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { >>>> zil_itx_destroy(itx); >>>> itx = zil_itx_create(TX_WRITE, sizeof (*lr)); >>>> lr = (lr_write_t *)&itx->itx_lr; >>>> - write_state = WR_NEED_COPY; >>>> + wr_state = WR_NEED_COPY; >>>> } >>>> >>>> - itx->itx_wr_state = write_state; >>>> - if (write_state == WR_NEED_COPY) >>>> - itx->itx_sod += len; >>>> + itx->itx_wr_state = wr_state; >>>> lr->lr_foid = ZVOL_OBJ; >>>> lr->lr_offset = off; >>>> lr->lr_length = len; >>>> > -- Alexander Motin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?71f021c6-5847-5808-20d2-4cc84925af46>