Patchwork [3/6] add-cow file format

login
register
mail settings
Submitter Robert Wang
Date July 31, 2012, 4:51 p.m.
Message ID <1343753510-24661-3-git-send-email-wdongxu@linux.vnet.ibm.com>
Download mbox | patch
Permalink /patch/174291/
State New
Headers show

Comments

Robert Wang - July 31, 2012, 4:51 p.m.
This is the implementation code for add-cow file format. Because image_file
might be very huge, then we can't read entire bitmap into memory, we must use
a cache. Since qcow-cache.c has implemted cache code, we can create our cache
code based on it.

Signed-off-by: Dong Xu Wang <wdongxu@linux.vnet.ibm.com>
---
 block/Makefile.objs   |    1 +
 block/add-cow-cache.c |  206 +++++++++++++++++
 block/add-cow.c       |  599 +++++++++++++++++++++++++++++++++++++++++++++++++
 block/add-cow.h       |  101 +++++++++
 block_int.h           |    2 +
 5 files changed, 909 insertions(+), 0 deletions(-)
 create mode 100644 block/add-cow-cache.c
 create mode 100644 block/add-cow.c
 create mode 100644 block/add-cow.h
Eric Blake - Aug. 1, 2012, 1:57 p.m.
On 07/31/2012 10:51 AM, Dong Xu Wang wrote:
> This is the implementation code for add-cow file format. Because image_file
> might be very huge, then we can't read entire bitmap into memory, we must use
> a cache. Since qcow-cache.c has implemted cache code, we can create our cache

s/implemted/implemented/

> code based on it.

Just wondering if Paolo's HBitmap code for drive-mirror might be a more
efficient way to implement your caching.

> 
> Signed-off-by: Dong Xu Wang <wdongxu@linux.vnet.ibm.com>
> ---
>  block/Makefile.objs   |    1 +
>  block/add-cow-cache.c |  206 +++++++++++++++++
>  block/add-cow.c       |  599 +++++++++++++++++++++++++++++++++++++++++++++++++
>  block/add-cow.h       |  101 +++++++++
>  block_int.h           |    2 +

Rather than adding a new implementation for code duplication, can you
refactor the existing implementation to be reusable, and update
qcow-cache.c to call into the common refactored code?
Stefan Hajnoczi - Aug. 1, 2012, 2:14 p.m.
On Wed, Aug 1, 2012 at 2:57 PM, Eric Blake <eblake@redhat.com> wrote:
> On 07/31/2012 10:51 AM, Dong Xu Wang wrote:
>> This is the implementation code for add-cow file format. Because image_file
>> might be very huge, then we can't read entire bitmap into memory, we must use
>> a cache. Since qcow-cache.c has implemted cache code, we can create our cache
>
> s/implemted/implemented/
>
>> code based on it.
>
> Just wondering if Paolo's HBitmap code for drive-mirror might be a more
> efficient way to implement your caching.
>
>>
>> Signed-off-by: Dong Xu Wang <wdongxu@linux.vnet.ibm.com>
>> ---
>>  block/Makefile.objs   |    1 +
>>  block/add-cow-cache.c |  206 +++++++++++++++++
>>  block/add-cow.c       |  599 +++++++++++++++++++++++++++++++++++++++++++++++++
>>  block/add-cow.h       |  101 +++++++++
>>  block_int.h           |    2 +
>
> Rather than adding a new implementation for code duplication, can you
> refactor the existing implementation to be reusable, and update
> qcow-cache.c to call into the common refactored code?

I looked at this and block/qcow2-cache.c seems fairly clean and
generic.  The only things that need to be parameterized are the
s->l2_table_cache/s->refcount_block_cache and s->cluster_size.

These can be fixed by adding fields to Qcow2Cache:
size_t table_size; /* cached table size in bytes */
BlkDebugEvent flush_event;
BlkDebugEvent load_event;

Or maybe a table type enum (L2/refcount blocks/bitmap blocks).

If Kevin agrees with extracting this from qcow2 (i.e. we can't add
more qcow2-specific behavior easily in the future), then this looks
doable.

Stefan
Kevin Wolf - Aug. 1, 2012, 2:21 p.m.
Am 01.08.2012 16:14, schrieb Stefan Hajnoczi:
> On Wed, Aug 1, 2012 at 2:57 PM, Eric Blake <eblake@redhat.com> wrote:
>> On 07/31/2012 10:51 AM, Dong Xu Wang wrote:
>>> This is the implementation code for add-cow file format. Because image_file
>>> might be very huge, then we can't read entire bitmap into memory, we must use
>>> a cache. Since qcow-cache.c has implemted cache code, we can create our cache
>>
>> s/implemted/implemented/
>>
>>> code based on it.
>>
>> Just wondering if Paolo's HBitmap code for drive-mirror might be a more
>> efficient way to implement your caching.
>>
>>>
>>> Signed-off-by: Dong Xu Wang <wdongxu@linux.vnet.ibm.com>
>>> ---
>>>  block/Makefile.objs   |    1 +
>>>  block/add-cow-cache.c |  206 +++++++++++++++++
>>>  block/add-cow.c       |  599 +++++++++++++++++++++++++++++++++++++++++++++++++
>>>  block/add-cow.h       |  101 +++++++++
>>>  block_int.h           |    2 +
>>
>> Rather than adding a new implementation for code duplication, can you
>> refactor the existing implementation to be reusable, and update
>> qcow-cache.c to call into the common refactored code?
> 
> I looked at this and block/qcow2-cache.c seems fairly clean and
> generic.  The only things that need to be parameterized are the
> s->l2_table_cache/s->refcount_block_cache and s->cluster_size.
> 
> These can be fixed by adding fields to Qcow2Cache:
> size_t table_size; /* cached table size in bytes */
> BlkDebugEvent flush_event;
> BlkDebugEvent load_event;
> 
> Or maybe a table type enum (L2/refcount blocks/bitmap blocks).
> 
> If Kevin agrees with extracting this from qcow2 (i.e. we can't add
> more qcow2-specific behavior easily in the future), then this looks
> doable.

I don't think we'd want to add qcow2 specific stuff there, and I've been
meaning to generalise it for quite a while, so feel free to do it.

One thing to note is that it currently depends on qcow2's s->lock being
taken while it's called. Eventually we'll want finer grained locking in
the cache itself and I believe this will also require some changes in
the interface, but for now add-cow can just do the same qcow2 and take
its global CoMutex.

Kevin
Stefan Hajnoczi - Aug. 1, 2012, 3:31 p.m.
On Tue, Jul 31, 2012 at 5:51 PM, Dong Xu Wang
<wdongxu@linux.vnet.ibm.com> wrote:
> +    if (backing_filename) {
> +        header.features |= ADD_COW_F_BACKING_FILE;
> +        header.backing_filename_offset = sizeof(header) + 16 * 2;

It's not obvious what 16 * 2 is.

> +    ret = bdrv_pwrite(bs, sizeof(le_header) + 16,
> +        image_format ? image_format : "raw",
> +        image_format ? sizeof(image_format) : sizeof("raw"));

sizeof(image_format)?  I think this should be strlen(image_format).

Stefan
Robert Wang - Aug. 2, 2012, 7:20 a.m.
On Wed, Aug 1, 2012 at 9:57 PM, Eric Blake <eblake@redhat.com> wrote:
> On 07/31/2012 10:51 AM, Dong Xu Wang wrote:
>> This is the implementation code for add-cow file format. Because image_file
>> might be very huge, then we can't read entire bitmap into memory, we must use
>> a cache. Since qcow-cache.c has implemted cache code, we can create our cache
>
> s/implemted/implemented/
>

Okay.

>> code based on it.
>
> Just wondering if Paolo's HBitmap code for drive-mirror might be a more
> efficient way to implement your caching.
>
>>
>> Signed-off-by: Dong Xu Wang <wdongxu@linux.vnet.ibm.com>
>> ---
>>  block/Makefile.objs   |    1 +
>>  block/add-cow-cache.c |  206 +++++++++++++++++
>>  block/add-cow.c       |  599 +++++++++++++++++++++++++++++++++++++++++++++++++
>>  block/add-cow.h       |  101 +++++++++
>>  block_int.h           |    2 +
>
> Rather than adding a new implementation for code duplication, can you
> refactor the existing implementation to be reusable, and update
> qcow-cache.c to call into the common refactored code?
>

Okay. thanks.
> --
> Eric Blake   eblake@redhat.com    +1-919-301-3266
> Libvirt virtualization library http://libvirt.org
>
Robert Wang - Aug. 2, 2012, 7:20 a.m.
On Wed, Aug 1, 2012 at 11:31 PM, Stefan Hajnoczi <stefanha@gmail.com> wrote:
> On Tue, Jul 31, 2012 at 5:51 PM, Dong Xu Wang
> <wdongxu@linux.vnet.ibm.com> wrote:
>> +    if (backing_filename) {
>> +        header.features |= ADD_COW_F_BACKING_FILE;
>> +        header.backing_filename_offset = sizeof(header) + 16 * 2;
>
> It's not obvious what 16 * 2 is.

Okay we give meaningful constant later.

>
>> +    ret = bdrv_pwrite(bs, sizeof(le_header) + 16,
>> +        image_format ? image_format : "raw",
>> +        image_format ? sizeof(image_format) : sizeof("raw"));
>
> sizeof(image_format)?  I think this should be strlen(image_format).

Okay. thanks.

>
> Stefan
>

Patch

diff --git a/block/Makefile.objs b/block/Makefile.objs
index b5754d3..357a3b1 100644
--- a/block/Makefile.objs
+++ b/block/Makefile.objs
@@ -2,6 +2,7 @@  block-obj-y += raw.o cow.o qcow.o vdi.o vmdk.o cloop.o dmg.o bochs.o vpc.o vvfat
 block-obj-y += qcow2.o qcow2-refcount.o qcow2-cluster.o qcow2-snapshot.o qcow2-cache.o
 block-obj-y += qed.o qed-gencb.o qed-l2-cache.o qed-table.o qed-cluster.o
 block-obj-y += qed-check.o
+block-obj-y += add-cow.o add-cow-cache.o
 block-obj-y += parallels.o nbd.o blkdebug.o sheepdog.o blkverify.o
 block-obj-y += stream.o
 block-obj-$(CONFIG_WIN32) += raw-win32.o
diff --git a/block/add-cow-cache.c b/block/add-cow-cache.c
new file mode 100644
index 0000000..5d5e8a4
--- /dev/null
+++ b/block/add-cow-cache.c
@@ -0,0 +1,206 @@ 
+/*
+ * Cache For QEMU ADD-COW Disk Format
+ *
+ * Copyright IBM, Corp. 2012
+ *
+ * Authors:
+ *  Dong Xu Wang <wdongxu@linux.vnet.ibm.com>
+ *
+ * This file is based on qcow2-cache.c, see its copyrights below:
+ *
+ * L2/refcount table cache for the QCOW2 format
+ *
+ * Copyright (c) 2010 Kevin Wolf <kwolf@redhat.com>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#include "block_int.h"
+#include "qemu-common.h"
+#include "add-cow.h"
+
+AddCowCache *add_cow_cache_create(BlockDriverState *bs, int num_tables)
+{
+    AddCowCache *c;
+    int i;
+
+    c = g_malloc0(sizeof(*c));
+    c->size = num_tables;
+    c->entries = g_malloc0(sizeof(*c->entries) * num_tables);
+    c->entry_size = ADD_COW_CACHE_ENTRY_SIZE;
+
+    for (i = 0; i < c->size; i++) {
+        c->entries[i].table = qemu_blockalign(bs, c->entry_size);
+        c->entries[i].offset = -1;
+    }
+
+    return c;
+}
+
+int add_cow_cache_destroy(BlockDriverState *bs, AddCowCache *c)
+{
+    int i;
+
+    for (i = 0; i < c->size; i++) {
+        qemu_vfree(c->entries[i].table);
+    }
+
+    g_free(c->entries);
+    g_free(c);
+
+    return 0;
+}
+
+static int add_cow_cache_entry_flush(BlockDriverState *bs,
+    AddCowCache *c,
+    int i)
+{
+    BDRVAddCowState *s = bs->opaque;
+    int ret = 0;
+
+    if (!c->entries[i].dirty || -1 == c->entries[i].offset) {
+        return ret;
+    }
+
+    ret = bdrv_pwrite(bs->file,
+        ADD_COW_PAGE_SIZE * s->header.header_pages_size + c->entries[i].offset,
+        c->entries[i].table,
+        MIN(c->entry_size, s->bitmap_size - c->entries[i].offset));
+    if (ret < 0) {
+        return ret;
+    }
+
+    c->entries[i].dirty = false;
+
+    return 0;
+}
+
+int add_cow_cache_flush(BlockDriverState *bs, AddCowCache *c)
+{
+    BDRVAddCowState *s = bs->opaque;
+    int result = 0;
+    int ret;
+    int i;
+
+    ret = bdrv_flush(s->image_hd);
+    if (ret < 0) {
+        return result;
+    }
+
+    for (i = 0; i < c->size; i++) {
+        ret = add_cow_cache_entry_flush(bs, c, i);
+        if (ret < 0 && result != -ENOSPC) {
+            result = ret;
+        }
+    }
+
+    if (result == 0) {
+        ret = bdrv_flush(bs->file);
+        if (ret < 0) {
+            result = ret;
+        }
+    }
+
+    return result;
+}
+
+static int add_cow_cache_find_entry_to_replace(AddCowCache *c)
+{
+    int i;
+    int min_count = INT_MAX;
+    int min_index = -1;
+
+
+    for (i = 0; i < c->size; i++) {
+        if (c->entries[i].cache_hits < min_count) {
+            min_index = i;
+            min_count = c->entries[i].cache_hits;
+        }
+
+        c->entries[i].cache_hits /= 2;
+    }
+
+    if (min_index == -1) {
+        abort();
+    }
+    return min_index;
+}
+
+static int add_cow_cache_do_get(BlockDriverState *bs, AddCowCache *c,
+    uint64_t offset, void **table)
+{
+    int i, ret;
+    BDRVAddCowState *s = bs->opaque;
+
+    for (i = 0; i < c->size; i++) {
+        if (c->entries[i].offset == offset) {
+            goto found;
+        }
+    }
+
+    i = add_cow_cache_find_entry_to_replace(c);
+    if (i < 0) {
+        return i;
+    }
+
+    ret = add_cow_cache_entry_flush(bs, c, i);
+    if (ret < 0) {
+        return ret;
+    }
+    c->entries[i].offset = -1;
+    ret = bdrv_pread(bs->file,
+        s->header.header_pages_size * ADD_COW_PAGE_SIZE + offset,
+        c->entries[i].table,
+        c->entry_size);
+    if (ret < 0) {
+        return ret;
+    }
+
+    c->entries[i].cache_hits = 32;
+    c->entries[i].offset = offset;
+
+found:
+    c->entries[i].cache_hits++;
+    *table = c->entries[i].table;
+
+    return 0;
+}
+
+int add_cow_cache_get(BlockDriverState *bs, AddCowCache *c, uint64_t sector_num,
+    void **table)
+{
+    /* each byte in bitmap indicates 8 * SECTORS_PER_CLUSTER clusters */
+    uint64_t offset = offset_in_bitmap(sector_num) & (~(c->entry_size - 1));
+    return add_cow_cache_do_get(bs, c, offset, table);
+}
+
+void add_cow_cache_entry_mark_dirty(AddCowCache *c, void *table)
+{
+    int i;
+
+    for (i = 0; i < c->size; i++) {
+        if (c->entries[i].table == table) {
+            goto found;
+        }
+    }
+    abort();
+
+found:
+    c->entries[i].dirty = true;
+}
diff --git a/block/add-cow.c b/block/add-cow.c
new file mode 100644
index 0000000..5f6b708
--- /dev/null
+++ b/block/add-cow.c
@@ -0,0 +1,599 @@ 
+/*
+ * QEMU ADD-COW Disk Format
+ *
+ * Copyright IBM, Corp. 2012
+ *
+ * Authors:
+ *  Dong Xu Wang <wdongxu@linux.vnet.ibm.com>
+ *
+ * This work is licensed under the terms of the GNU LGPL, version 2 or later.
+ * See the COPYING.LIB file in the top-level directory.
+ *
+ */
+
+#include "qemu-common.h"
+#include "block_int.h"
+#include "module.h"
+#include "add-cow.h"
+
+static void add_cow_header_le_to_cpu(const AddCowHeader *le, AddCowHeader *cpu)
+{
+    cpu->magic                      = le64_to_cpu(le->magic);
+    cpu->version                    = le32_to_cpu(le->version);
+
+    cpu->backing_filename_offset    = le32_to_cpu(le->backing_filename_offset);
+    cpu->backing_filename_size      = le32_to_cpu(le->backing_filename_size);
+
+    cpu->image_filename_offset      = le32_to_cpu(le->image_filename_offset);
+    cpu->image_filename_size        = le32_to_cpu(le->image_filename_size);
+
+    cpu->features                   = le64_to_cpu(le->features);
+    cpu->optional_features          = le64_to_cpu(le->optional_features);
+    cpu->header_pages_size          = le32_to_cpu(le->header_pages_size);
+}
+
+static void add_cow_header_cpu_to_le(const AddCowHeader *cpu, AddCowHeader *le)
+{
+    le->magic                       = cpu_to_le64(cpu->magic);
+    le->version                     = cpu_to_le32(cpu->version);
+
+    le->backing_filename_offset     = cpu_to_le32(cpu->backing_filename_offset);
+    le->backing_filename_size       = cpu_to_le32(cpu->backing_filename_size);
+
+    le->image_filename_offset       = cpu_to_le32(cpu->image_filename_offset);
+    le->image_filename_size         = cpu_to_le32(cpu->image_filename_size);
+
+    le->features                    = cpu_to_le64(cpu->features);
+    le->optional_features           = cpu_to_le64(cpu->optional_features);
+    le->header_pages_size           = cpu_to_le32(cpu->header_pages_size);
+}
+
+static int add_cow_probe(const uint8_t *buf, int buf_size, const char *filename)
+{
+    const AddCowHeader *header = (const AddCowHeader *)buf;
+
+    if (le64_to_cpu(header->magic) == ADD_COW_MAGIC &&
+        le32_to_cpu(header->version) == ADD_COW_VERSION) {
+        return 100;
+    } else {
+        return 0;
+    }
+}
+
+static int add_cow_create(const char *filename, QEMUOptionParameter *options)
+{
+    AddCowHeader header = {
+        .magic = ADD_COW_MAGIC,
+        .version = ADD_COW_VERSION,
+        .features = 0,
+        .optional_features = 0,
+        .header_pages_size = ADD_COW_DEFAULT_PAGE_SIZE,
+    };
+    AddCowHeader le_header;
+    int64_t backing_len = 0;
+    const char *backing_filename = NULL;
+    const char *backing_fmt = NULL;
+    const char *image_filename = NULL;
+    const char *image_format = NULL;
+    BlockDriverState *bs, *image_bs = NULL, *backing_bs =  NULL;
+    BlockDriver *drv = bdrv_find_format("add-cow");
+    int ret;
+
+    while (options && options->name) {
+        if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
+            backing_len = options->value.n;
+        } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
+            backing_filename = options->value.s;
+        } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
+            backing_fmt = options->value.s;
+        } else if (!strcmp(options->name, BLOCK_OPT_IMAGE_FILE)) {
+            image_filename = options->value.s;
+        } else if (!strcmp(options->name, BLOCK_OPT_IMAGE_FORMAT)) {
+            image_format = options->value.s;
+        }
+        options++;
+    }
+
+    if (backing_filename) {
+        header.features |= ADD_COW_F_BACKING_FILE;
+        header.backing_filename_offset = sizeof(header) + 16 * 2;
+        header.backing_filename_size = strlen(backing_filename);
+
+        if (!backing_fmt) {
+            backing_bs = bdrv_new("image");
+            ret = bdrv_open(backing_bs, backing_filename, BDRV_O_RDWR
+                    | BDRV_O_CACHE_WB, NULL);
+            if (ret < 0) {
+                return ret;
+            }
+            backing_fmt = bdrv_get_format_name(backing_bs);
+            bdrv_delete(backing_bs);
+        }
+    } else {
+        header.features |= ADD_COW_F_All_ALLOCATED;
+    }
+
+    if (image_filename) {
+        header.features |= ADD_COW_F_IMAGE_FILE;
+        header.image_filename_offset =
+            sizeof(header) + 16 * 2 + header.backing_filename_size;
+        header.image_filename_size = strlen(image_filename);
+    } else {
+        error_report("image_file should be given.");
+        return -EINVAL;
+    }
+
+    if (backing_filename && !strcmp(backing_filename, image_filename)) {
+        error_report("Error: Trying to create an image with the "
+                     "same backing file name as the image file name");
+        return -EINVAL;
+    }
+
+    if (!strcmp(filename, image_filename)) {
+        error_report("Error: Trying to create an image with the "
+                     "same filename as the image file name");
+        return -EINVAL;
+    }
+
+    if (header.image_filename_offset + header.image_filename_size
+            > ADD_COW_PAGE_SIZE * ADD_COW_DEFAULT_PAGE_SIZE) {
+        error_report("image_file name or backing_file name too long.");
+        return -ENOSPC;
+    }
+
+    ret = bdrv_file_open(&image_bs, image_filename, BDRV_O_RDWR);
+    if (ret < 0) {
+        return ret;
+    }
+    bdrv_delete(image_bs);
+
+    ret = bdrv_create_file(filename, NULL);
+    if (ret < 0) {
+        return ret;
+    }
+
+    ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
+    if (ret < 0) {
+        return ret;
+    }
+    add_cow_header_cpu_to_le(&header, &le_header);
+    ret = bdrv_pwrite(bs, 0, &le_header, sizeof(le_header));
+    if (ret < 0) {
+        bdrv_delete(bs);
+        return ret;
+    }
+
+    ret = bdrv_pwrite(bs, sizeof(le_header), backing_fmt ? backing_fmt : "",
+        backing_fmt ? sizeof(backing_fmt) : 0);
+    if (ret < 0) {
+        bdrv_delete(bs);
+        return ret;
+    }
+
+    ret = bdrv_pwrite(bs, sizeof(le_header) + 16,
+        image_format ? image_format : "raw",
+        image_format ? sizeof(image_format) : sizeof("raw"));
+    if (ret < 0) {
+        bdrv_delete(bs);
+        return ret;
+    }
+
+    if (backing_filename) {
+        ret = bdrv_pwrite(bs,
+            header.backing_filename_offset,
+            backing_filename,
+            header.backing_filename_size);
+        if (ret < 0) {
+            bdrv_delete(bs);
+            return ret;
+        }
+    }
+
+    ret = bdrv_pwrite(bs,
+        header.image_filename_offset,
+        image_filename,
+        header.image_filename_size);
+    if (ret < 0) {
+        bdrv_delete(bs);
+        return ret;
+    }
+
+    ret = bdrv_open(bs, filename, BDRV_O_RDWR | BDRV_O_NO_FLUSH, drv);
+    if (ret < 0) {
+        bdrv_delete(bs);
+        return ret;
+    }
+
+    ret = bdrv_truncate(bs, backing_len);
+    bdrv_delete(bs);
+    return ret;
+}
+
+static int add_cow_open(BlockDriverState *bs, int flags)
+{
+    char                image_filename[ADD_COW_FILE_LEN];
+    char                tmp_name[ADD_COW_FILE_LEN];
+    BlockDriver         *image_drv = NULL;
+    BlockDriverState    *backing_bs = NULL;
+    int                 ret;
+    int                 sector_per_byte;
+    BDRVAddCowState     *s = bs->opaque;
+    AddCowHeader        le_header;
+
+    ret = bdrv_pread(bs->file, 0, &le_header, sizeof(le_header));
+    if (ret != sizeof(s->header)) {
+        goto fail;
+    }
+
+    add_cow_header_le_to_cpu(&le_header, &s->header);
+
+    if (le64_to_cpu(s->header.magic) != ADD_COW_MAGIC) {
+        ret = -EINVAL;
+        goto fail;
+    }
+
+    if (s->header.version != ADD_COW_VERSION) {
+        char version[64];
+        snprintf(version, sizeof(version), "ADD-COW version %d",
+            s->header.version);
+        qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
+            bs->device_name, "add-cow", version);
+        ret = -ENOTSUP;
+        goto fail;
+    }
+
+    if (s->header.features & ~ADD_COW_FEATURE_MASK) {
+        char buf[64];
+        snprintf(buf, sizeof(buf), "%" PRIx64,
+            s->header.features & ~ADD_COW_FEATURE_MASK);
+        qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
+            bs->device_name, "add-cow", buf);
+        return -ENOTSUP;
+    }
+    if ((s->header.features & ADD_COW_F_IMAGE_FILE) == 0) {
+        ret = -ENOTSUP;
+        goto fail;
+    }
+
+    if ((s->header.features & ADD_COW_F_All_ALLOCATED) == 0) {
+        ret = bdrv_read_string(bs->file, sizeof(s->header),
+                      15, s->backing_file_format,
+                      sizeof(s->backing_file_format));
+        if (ret < 0) {
+            goto fail;
+        }
+    }
+
+    ret = bdrv_read_string(bs->file, sizeof(s->header) + 16,
+                  15, s->image_file_format,
+                  sizeof(s->image_file_format));
+    if (ret < 0) {
+        goto fail;
+    }
+
+    if ((s->header.features & ADD_COW_F_All_ALLOCATED) == 0) {
+        ret = bdrv_read_string(bs->file, s->header.backing_filename_offset,
+                          s->header.backing_filename_size, bs->backing_file,
+                          sizeof(bs->backing_file));
+        if (ret < 0) {
+            goto fail;
+        }
+    }
+
+    ret = bdrv_read_string(bs->file, s->header.image_filename_offset,
+                      s->header.image_filename_size, tmp_name,
+                      sizeof(tmp_name));
+    if (ret < 0) {
+        goto fail;
+    }
+
+    s->image_hd = bdrv_new("");
+    if (path_has_protocol(image_filename)) {
+        pstrcpy(image_filename, sizeof(image_filename), tmp_name);
+    } else {
+        path_combine(image_filename, sizeof(image_filename),
+                     bs->filename, tmp_name);
+    }
+
+    ret = bdrv_open(s->image_hd, image_filename, flags, image_drv);
+    if (ret < 0) {
+        bdrv_delete(s->image_hd);
+        goto fail;
+    }
+
+    if (s->header.features & ADD_COW_F_All_ALLOCATED) {
+        bs->total_sectors = bdrv_getlength(s->image_hd) >> 9;
+    } else {
+        backing_bs = bdrv_new("image");
+        ret = bdrv_open(backing_bs, bs->backing_file, BDRV_O_RDWR
+                | BDRV_O_CACHE_WB, NULL);
+        if (ret < 0) {
+            return ret;
+        }
+        bs->total_sectors = bdrv_getlength(backing_bs) >> 9;
+        bdrv_delete(backing_bs);
+    }
+
+    s->cluster_size = ADD_COW_CLUSTER_SIZE;
+    sector_per_byte = SECTORS_PER_CLUSTER * 8;
+    s->bitmap_size =
+        (bs->total_sectors + sector_per_byte - 1) / sector_per_byte;
+    s->bitmap_cache =
+        add_cow_cache_create(bs, ADD_COW_CACHE_SIZE);
+
+    qemu_co_mutex_init(&s->lock);
+    return 0;
+fail:
+    if (s->bitmap_cache) {
+        add_cow_cache_destroy(bs, s->bitmap_cache);
+    }
+    return ret;
+}
+
+static void add_cow_close(BlockDriverState *bs)
+{
+    BDRVAddCowState *s = bs->opaque;
+    add_cow_cache_destroy(bs, s->bitmap_cache);
+    bdrv_delete(s->image_hd);
+}
+
+static bool is_allocated(BlockDriverState *bs, int64_t sector_num)
+{
+    BDRVAddCowState *s  = bs->opaque;
+    int64_t cluster_num = sector_num / SECTORS_PER_CLUSTER;
+    uint8_t *table      = NULL;
+    int ret = add_cow_cache_get(bs, s->bitmap_cache,
+        sector_num, (void **)&table);
+
+    if (ret < 0) {
+        return ret;
+    }
+    return table[cluster_num / 8 % ADD_COW_CACHE_ENTRY_SIZE]
+        & (1 << (cluster_num % 8));
+}
+
+static coroutine_fn int add_cow_is_allocated(BlockDriverState *bs,
+        int64_t sector_num, int nb_sectors, int *num_same)
+{
+    BDRVAddCowState *s = bs->opaque;
+    int changed;
+
+    if (nb_sectors == 0) {
+        *num_same = 0;
+        return 0;
+    }
+
+    if (s->header.features & ADD_COW_F_All_ALLOCATED) {
+        *num_same = nb_sectors - 1;
+        return 1;
+    }
+    changed = is_allocated(bs, sector_num);
+
+    for (*num_same = 1; *num_same < nb_sectors; (*num_same)++) {
+        if (is_allocated(bs, sector_num + *num_same) != changed) {
+            break;
+        }
+    }
+    return changed;
+}
+
+static coroutine_fn int add_cow_co_readv(BlockDriverState *bs,
+    int64_t sector_num, int remaining_sectors, QEMUIOVector *qiov)
+{
+    BDRVAddCowState *s  = bs->opaque;
+    int cur_nr_sectors;
+    uint64_t bytes_done = 0;
+    QEMUIOVector hd_qiov;
+    int n, ret = 0;
+
+    qemu_iovec_init(&hd_qiov, qiov->niov);
+    qemu_co_mutex_lock(&s->lock);
+    while (remaining_sectors != 0) {
+        cur_nr_sectors = remaining_sectors;
+        if (add_cow_is_allocated(bs, sector_num, cur_nr_sectors, &n)) {
+            cur_nr_sectors = n;
+            qemu_iovec_reset(&hd_qiov);
+            qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
+                            cur_nr_sectors * BDRV_SECTOR_SIZE);
+            qemu_co_mutex_unlock(&s->lock);
+            ret = bdrv_co_readv(s->image_hd, sector_num, n, &hd_qiov);
+            qemu_co_mutex_lock(&s->lock);
+            if (ret < 0) {
+                goto fail;
+            }
+        } else {
+            cur_nr_sectors = n;
+            if (bs->backing_hd) {
+                qemu_iovec_reset(&hd_qiov);
+                qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
+                            cur_nr_sectors * BDRV_SECTOR_SIZE);
+                qemu_co_mutex_unlock(&s->lock);
+                ret = bdrv_co_readv(bs->backing_hd, sector_num,
+                                    n, &hd_qiov);
+                qemu_co_mutex_lock(&s->lock);
+                if (ret < 0) {
+                    goto fail;
+                }
+            } else {
+                qemu_iovec_memset(&hd_qiov, 0, 0,
+                    BDRV_SECTOR_SIZE * cur_nr_sectors);
+            }
+        }
+        remaining_sectors -= cur_nr_sectors;
+        sector_num += cur_nr_sectors;
+        bytes_done += cur_nr_sectors * BDRV_SECTOR_SIZE;
+    }
+fail:
+    qemu_co_mutex_unlock(&s->lock);
+    qemu_iovec_destroy(&hd_qiov);
+    return ret;
+}
+
+static int coroutine_fn copy_sectors(BlockDriverState *bs,
+                                     int n_start, int n_end)
+{
+    BDRVAddCowState *s = bs->opaque;
+    QEMUIOVector qiov;
+    struct iovec iov;
+    int n, ret;
+
+    n = n_end - n_start;
+    if (n <= 0) {
+        return 0;
+    }
+
+    iov.iov_len = n * BDRV_SECTOR_SIZE;
+    iov.iov_base = qemu_blockalign(bs, iov.iov_len);
+
+    qemu_iovec_init_external(&qiov, &iov, 1);
+
+    ret = bdrv_co_readv(bs->backing_hd, n_start, n, &qiov);
+    if (ret < 0) {
+        goto out;
+    }
+    ret = bdrv_co_writev(s->image_hd, n_start, n, &qiov);
+    if (ret < 0) {
+        goto out;
+    }
+
+    ret = 0;
+out:
+    qemu_vfree(iov.iov_base);
+    return ret;
+}
+
+static coroutine_fn int add_cow_co_writev(BlockDriverState *bs,
+        int64_t sector_num, int remaining_sectors, QEMUIOVector *qiov)
+{
+    BDRVAddCowState *s = bs->opaque;
+    int ret = 0, i;
+    QEMUIOVector hd_qiov;
+    uint8_t *table;
+
+    qemu_co_mutex_lock(&s->lock);
+    qemu_iovec_init(&hd_qiov, qiov->niov);
+    ret = bdrv_co_writev(s->image_hd,
+                     sector_num,
+                     remaining_sectors, qiov);
+
+    if (ret < 0) {
+        goto fail;
+    }
+    if ((s->header.features & ADD_COW_F_All_ALLOCATED) == 0) {
+        /* Copy content of unmodified sectors */
+        if (!is_cluster_head(sector_num) && !is_allocated(bs, sector_num)) {
+            ret = copy_sectors(bs, sector_num & ~(SECTORS_PER_CLUSTER - 1),
+                sector_num);
+            if (ret < 0) {
+                goto fail;
+            }
+        }
+
+        if (!is_cluster_tail(sector_num + remaining_sectors - 1)
+            && !is_allocated(bs, sector_num + remaining_sectors - 1)) {
+            ret = copy_sectors(bs, sector_num + remaining_sectors,
+                ((sector_num + remaining_sectors) | (SECTORS_PER_CLUSTER - 1)) + 1);
+            if (ret < 0) {
+                goto fail;
+            }
+        }
+
+        for (i = sector_num / SECTORS_PER_CLUSTER;
+            i <= (sector_num + remaining_sectors - 1) / SECTORS_PER_CLUSTER;
+            i++) {
+            ret = add_cow_cache_get(bs, s->bitmap_cache,
+                i * SECTORS_PER_CLUSTER, (void **)&table);
+            if (ret < 0) {
+                goto fail;
+            }
+            if ((table[i / 8] & (1 << (i % 8))) == 0) {
+                table[i / 8] |= (1 << (i % 8));
+                add_cow_cache_entry_mark_dirty(s->bitmap_cache, table);
+            }
+        }
+    }
+    ret = 0;
+fail:
+    qemu_co_mutex_unlock(&s->lock);
+    qemu_iovec_destroy(&hd_qiov);
+    return ret;
+}
+
+static int bdrv_add_cow_truncate(BlockDriverState *bs, int64_t size)
+{
+    BDRVAddCowState *s = bs->opaque;
+    int sector_per_byte = SECTORS_PER_CLUSTER * 8;
+    int ret;
+    uint32_t bitmap_pos = s->header.header_pages_size * ADD_COW_PAGE_SIZE;
+    int64_t bitmap_size =
+        (size / BDRV_SECTOR_SIZE + sector_per_byte - 1) / sector_per_byte;
+
+    ret = bdrv_truncate(bs->file, bitmap_pos + bitmap_size);
+    if (ret < 0) {
+        return ret;
+    }
+    return 0;
+}
+
+static coroutine_fn int add_cow_co_flush(BlockDriverState *bs)
+{
+    BDRVAddCowState *s = bs->opaque;
+    int ret;
+
+    qemu_co_mutex_lock(&s->lock);
+    ret = add_cow_cache_flush(bs, s->bitmap_cache);
+    qemu_co_mutex_unlock(&s->lock);
+    return ret;
+}
+
+static QEMUOptionParameter add_cow_create_options[] = {
+    {
+        .name = BLOCK_OPT_SIZE,
+        .type = OPT_SIZE,
+        .help = "Virtual disk size"
+    },
+    {
+        .name = BLOCK_OPT_BACKING_FILE,
+        .type = OPT_STRING,
+        .help = "File name of a base image"
+    },
+    {
+        .name = BLOCK_OPT_BACKING_FMT,
+        .type = OPT_STRING,
+        .help = "Image format of the base image"
+    },
+    {
+        .name = BLOCK_OPT_IMAGE_FILE,
+        .type = OPT_STRING,
+        .help = "File name of a image file"
+    },
+    {
+        .name = BLOCK_OPT_IMAGE_FORMAT,
+        .type = OPT_STRING,
+        .help = "Image format of the image file"
+    },
+    { NULL }
+};
+
+static BlockDriver bdrv_add_cow = {
+    .format_name                = "add-cow",
+    .instance_size              = sizeof(BDRVAddCowState),
+    .bdrv_probe                 = add_cow_probe,
+    .bdrv_open                  = add_cow_open,
+    .bdrv_close                 = add_cow_close,
+    .bdrv_create                = add_cow_create,
+    .bdrv_co_readv              = add_cow_co_readv,
+    .bdrv_co_writev             = add_cow_co_writev,
+    .bdrv_truncate              = bdrv_add_cow_truncate,
+    .bdrv_co_is_allocated       = add_cow_is_allocated,
+
+    .create_options             = add_cow_create_options,
+    .bdrv_co_flush_to_os        = add_cow_co_flush,
+};
+
+static void bdrv_add_cow_init(void)
+{
+    bdrv_register(&bdrv_add_cow);
+}
+
+block_init(bdrv_add_cow_init);
diff --git a/block/add-cow.h b/block/add-cow.h
new file mode 100644
index 0000000..738249c
--- /dev/null
+++ b/block/add-cow.h
@@ -0,0 +1,101 @@ 
+/*
+ * QEMU ADD-COW Disk Format
+ *
+ * Copyright IBM, Corp. 2012
+ *
+ * Authors:
+ *  Dong Xu Wang <wdongxu@linux.vnet.ibm.com>
+ *
+ * This work is licensed under the terms of the GNU LGPL, version 2 or later.
+ * See the COPYING.LIB file in the top-level directory.
+ *
+ */
+
+#ifndef BLOCK_ADD_COW_H
+#define BLOCK_ADD_COW_H
+
+enum {
+    ADD_COW_F_BACKING_FILE      = 0x01,
+    ADD_COW_F_IMAGE_FILE        = 0x02,
+    ADD_COW_F_All_ALLOCATED     = 0X04,
+    ADD_COW_FEATURE_MASK        = ADD_COW_F_BACKING_FILE | \
+                                    ADD_COW_F_IMAGE_FILE | \
+                                    ADD_COW_F_All_ALLOCATED,
+
+    ADD_COW_MAGIC = (((uint64_t)'A' << 56) | ((uint64_t)'D' << 48) | \
+                    ((uint64_t)'D' << 40) | ((uint64_t)'_' << 32) | \
+                    ((uint64_t)'C' << 24) | ((uint64_t)'O' << 16) | \
+                    ((uint64_t)'W' << 8) | 0xFF),
+    ADD_COW_VERSION             = 1,
+    ADD_COW_FILE_LEN            = 1024,
+    ADD_COW_CACHE_SIZE          = 16,
+    ADD_COW_CACHE_ENTRY_SIZE    = 65536,
+    ADD_COW_CLUSTER_SIZE        = 65536,
+    SECTORS_PER_CLUSTER         = (ADD_COW_CLUSTER_SIZE / BDRV_SECTOR_SIZE),
+    ADD_COW_PAGE_SIZE           = 4096,
+    ADD_COW_DEFAULT_PAGE_SIZE   = 1
+};
+
+typedef struct AddCowHeader {
+    uint64_t        magic;
+    uint32_t        version;
+
+    uint32_t        backing_filename_offset;
+    uint32_t        backing_filename_size;
+
+    uint32_t        image_filename_offset;
+    uint32_t        image_filename_size;
+
+    uint64_t        features;
+    uint64_t        optional_features;
+    uint32_t        header_pages_size;
+} QEMU_PACKED AddCowHeader;
+
+typedef struct AddCowCachedTable {
+    void    *table;
+    int64_t offset;
+    bool    dirty;
+    int     cache_hits;
+} AddCowCachedTable;
+
+typedef struct AddCowCache {
+    AddCowCachedTable       *entries;
+    int                     entry_size;
+    int                     size;
+} AddCowCache;
+
+typedef struct BDRVAddCowState {
+    BlockDriverState    *image_hd;
+    CoMutex             lock;
+    int                 cluster_size;
+    AddCowCache         *bitmap_cache;
+    uint64_t            bitmap_size;
+    AddCowHeader        header;
+    char                backing_file_format[16];
+    char                image_file_format[16];
+} BDRVAddCowState;
+
+/* Convert sector_num to offset in bitmap */
+static inline int64_t offset_in_bitmap(int64_t sector_num)
+{
+    int64_t cluster_num = sector_num / SECTORS_PER_CLUSTER;
+    return cluster_num / 8;
+}
+
+static inline bool is_cluster_head(int64_t sector_num)
+{
+    return sector_num % SECTORS_PER_CLUSTER == 0;
+}
+
+static inline bool is_cluster_tail(int64_t sector_num)
+{
+    return (sector_num + 1) % SECTORS_PER_CLUSTER == 0;
+}
+
+AddCowCache *add_cow_cache_create(BlockDriverState *bs, int num_tables);
+int add_cow_cache_destroy(BlockDriverState *bs, AddCowCache *c);
+void add_cow_cache_entry_mark_dirty(AddCowCache *c, void *table);
+int add_cow_cache_get(BlockDriverState *bs, AddCowCache *c, uint64_t offset,
+    void **table);
+int add_cow_cache_flush(BlockDriverState *bs, AddCowCache *c);
+#endif
diff --git a/block_int.h b/block_int.h
index d72317f..8f8bb2b 100644
--- a/block_int.h
+++ b/block_int.h
@@ -51,6 +51,8 @@ 
 #define BLOCK_OPT_PREALLOC      "preallocation"
 #define BLOCK_OPT_SUBFMT        "subformat"
 #define BLOCK_OPT_COMPAT_LEVEL  "compat"
+#define BLOCK_OPT_IMAGE_FILE    "image_file"
+#define BLOCK_OPT_IMAGE_FORMAT  "image_format"
 
 typedef struct BdrvTrackedRequest BdrvTrackedRequest;