From 47ae44b999cb927ab3964587ca4d506d9ce2b146 Mon Sep 17 00:00:00 2001 From: mav Date: Mon, 22 Jan 2018 05:57:14 +0000 Subject: [PATCH] MFV r328253: 8835 Speculative prefetch in ZFS not working for misaligned reads illumos/illumos-gate@5cb8d943bc8513c6230589aad5a409d58b0297cb https://www.illumos.org/issues/8835: Sequential reads not aligned to block size are not detected by ZFS prefetcher as sequential, killing prefetch and severely hurting performance. It is caused by dmu_zfetch() in case of misaligned sequential accesses being called with overlap of one block. Reviewed by: Matthew Ahrens Reviewed by: Allan Jude Approved by: Gordon Ross Author: Alexander Motin --- .../uts/common/fs/zfs/dmu_zfetch.c | 24 +++++++++++++++---- 1 file changed, 19 insertions(+), 5 deletions(-) diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_zfetch.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_zfetch.c index 4e3bee9b3fc2..093740cd591a 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_zfetch.c +++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_zfetch.c @@ -240,19 +240,33 @@ dmu_zfetch(zfetch_t *zf, uint64_t blkid, uint64_t nblks, boolean_t fetch_data) rw_enter(&zf->zf_rwlock, RW_READER); + /* + * Find matching prefetch stream. Depending on whether the accesses + * are block-aligned, first block of the new access may either follow + * the last block of the previous access, or be equal to it. + */ for (zs = list_head(&zf->zf_stream); zs != NULL; zs = list_next(&zf->zf_stream, zs)) { - if (blkid == zs->zs_blkid) { + if (blkid == zs->zs_blkid || blkid + 1 == zs->zs_blkid) { mutex_enter(&zs->zs_lock); /* * zs_blkid could have changed before we * acquired zs_lock; re-check them here. */ - if (blkid != zs->zs_blkid) { - mutex_exit(&zs->zs_lock); - continue; + if (blkid == zs->zs_blkid) { + break; + } else if (blkid + 1 == zs->zs_blkid) { + blkid++; + nblks--; + if (nblks == 0) { + /* Already prefetched this before. */ + mutex_exit(&zs->zs_lock); + rw_exit(&zf->zf_rwlock); + return; + } + break; } - break; + mutex_exit(&zs->zs_lock); } }