From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:54372) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XgvO6-0006pe-RF for qemu-devel@nongnu.org; Wed, 22 Oct 2014 08:51:45 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1XgvO0-0004O9-DR for qemu-devel@nongnu.org; Wed, 22 Oct 2014 08:51:38 -0400 Received: from mx1.redhat.com ([209.132.183.28]:35748) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XgvO0-0004O4-64 for qemu-devel@nongnu.org; Wed, 22 Oct 2014 08:51:32 -0400 Received: from int-mx09.intmail.prod.int.phx2.redhat.com (int-mx09.intmail.prod.int.phx2.redhat.com [10.5.11.22]) by mx1.redhat.com (8.14.4/8.14.4) with ESMTP id s9MCpVjw015998 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-GCM-SHA384 bits=256 verify=FAIL) for ; Wed, 22 Oct 2014 08:51:31 -0400 From: Max Reitz Date: Wed, 22 Oct 2014 14:51:12 +0200 Message-Id: <1413982283-10186-4-git-send-email-mreitz@redhat.com> In-Reply-To: <1413982283-10186-1-git-send-email-mreitz@redhat.com> References: <1413982283-10186-1-git-send-email-mreitz@redhat.com> Subject: [Qemu-devel] [PATCH v13 03/14] qcow2: Optimize bdrv_make_empty() List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: qemu-devel@nongnu.org Cc: Kevin Wolf , Stefan Hajnoczi , Max Reitz bdrv_make_empty() is currently only called if the current image represents an external snapshot that has been committed to its base image; it is therefore unlikely to have internal snapshots. In this case, bdrv_make_empty() can be greatly sped up by emptying the L1 and refcount table (while having the dirty flag set, which only works for compat=1.1) and creating a trivial refcount structure. If there are snapshots or for compat=0.10, fall back to the simple implementation (discard all clusters). Signed-off-by: Max Reitz --- block/blkdebug.c | 2 + block/qcow2.c | 143 ++++++++++++++++++++++++++++++++++++++++++++------ include/block/block.h | 2 + 3 files changed, 132 insertions(+), 15 deletions(-) diff --git a/block/blkdebug.c b/block/blkdebug.c index e046b92..862d93b 100644 --- a/block/blkdebug.c +++ b/block/blkdebug.c @@ -195,6 +195,8 @@ static const char *event_names[BLKDBG_EVENT_MAX] = { [BLKDBG_PWRITEV] = "pwritev", [BLKDBG_PWRITEV_ZERO] = "pwritev_zero", [BLKDBG_PWRITEV_DONE] = "pwritev_done", + + [BLKDBG_EMPTY_IMAGE_PREPARE] = "empty_image_prepare", }; static int get_event_by_name(const char *name, BlkDebugEvent *event) diff --git a/block/qcow2.c b/block/qcow2.c index 1ef3a5f..16dece2 100644 --- a/block/qcow2.c +++ b/block/qcow2.c @@ -2232,24 +2232,137 @@ fail: static int qcow2_make_empty(BlockDriverState *bs) { + BDRVQcowState *s = bs->opaque; int ret = 0; - uint64_t start_sector; - int sector_step = INT_MAX / BDRV_SECTOR_SIZE; - for (start_sector = 0; start_sector < bs->total_sectors; - start_sector += sector_step) - { - /* As this function is generally used after committing an external - * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the - * default action for this kind of discard is to pass the discard, - * which will ideally result in an actually smaller image file, as - * is probably desired. */ - ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE, - MIN(sector_step, - bs->total_sectors - start_sector), - QCOW2_DISCARD_SNAPSHOT, true); + if (s->snapshots || s->qcow_version < 3) { + uint64_t start_sector; + int sector_step = INT_MAX / BDRV_SECTOR_SIZE; + + /* If there are snapshots, every active cluster has to be discarded; and + * because compat=0.10 does not support setting the dirty flag, we have + * to use this fallback there as well */ + + for (start_sector = 0; start_sector < bs->total_sectors; + start_sector += sector_step) + { + /* As this function is generally used after committing an external + * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the + * default action for this kind of discard is to pass the discard, + * which will ideally result in an actually smaller image file, as + * is probably desired. */ + ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE, + MIN(sector_step, + bs->total_sectors - start_sector), + QCOW2_DISCARD_SNAPSHOT, true); + if (ret < 0) { + break; + } + } + } else { + int l1_clusters; + int64_t offset; + uint64_t *new_reftable; + uint64_t rt_entry; + struct { + uint64_t l1_offset; + uint64_t reftable_offset; + uint32_t reftable_clusters; + } QEMU_PACKED l1_ofs_rt_ofs_cls; + + ret = qcow2_cache_empty(bs, s->l2_table_cache); if (ret < 0) { - break; + return ret; + } + + ret = qcow2_cache_empty(bs, s->refcount_block_cache); + if (ret < 0) { + return ret; + } + + /* Refcounts will be broken utterly */ + ret = qcow2_mark_dirty(bs); + if (ret < 0) { + return ret; + } + + l1_clusters = DIV_ROUND_UP(s->l1_size, + s->cluster_size / sizeof(uint64_t)); + new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t)); + if (!new_reftable) { + return -ENOMEM; + } + + BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); + + /* Overwrite enough clusters at the beginning of the sectors to place + * the refcount table, a refcount block and the L1 table in; this may + * overwrite parts of the existing refcount and L1 table, which is not + * an issue because the dirty flag is set, complete data loss is in fact + * desired and partial data loss is consequently fine as well */ + ret = bdrv_write_zeroes(bs->file, s->cluster_size / BDRV_SECTOR_SIZE, + (2 + l1_clusters) * s->cluster_size / + BDRV_SECTOR_SIZE, 0); + if (ret < 0) { + g_free(new_reftable); + return ret; + } + + BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); + BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); + + /* "Create" an empty reftable (one cluster) directly after the image + * header and an empty L1 table three clusters after the image header; + * the cluster between those two will be used as the first refblock */ + cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size); + cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size); + cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1); + ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), + &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); + if (ret < 0) { + g_free(new_reftable); + return ret; + } + + s->l1_table_offset = 3 * s->cluster_size; + memset(s->l1_table, 0, s->l1_size * sizeof(uint64_t)); + + s->refcount_table_offset = s->cluster_size; + s->refcount_table_size = s->cluster_size / sizeof(uint64_t); + + g_free(s->refcount_table); + s->refcount_table = new_reftable; + + BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); + + /* Enter the first refblock into the reftable */ + rt_entry = cpu_to_be64(2 * s->cluster_size); + ret = bdrv_pwrite_sync(bs->file, s->cluster_size, + &rt_entry, sizeof(rt_entry)); + if (ret < 0) { + return ret; + } + + s->refcount_table[0] = 2 * s->cluster_size; + + ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size); + if (ret < 0) { + return ret; + } + + s->free_cluster_index = 0; + offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + + s->l1_size * sizeof(uint64_t)); + if (offset < 0) { + return offset; + } else if (offset > 0) { + error_report("First cluster in emptied image is in use"); + abort(); + } + + ret = qcow2_mark_clean(bs); + if (ret < 0) { + return ret; } } diff --git a/include/block/block.h b/include/block/block.h index 341054d..b1f4385 100644 --- a/include/block/block.h +++ b/include/block/block.h @@ -498,6 +498,8 @@ typedef enum { BLKDBG_PWRITEV_ZERO, BLKDBG_PWRITEV_DONE, + BLKDBG_EMPTY_IMAGE_PREPARE, + BLKDBG_EVENT_MAX, } BlkDebugEvent; -- 1.9.3