Re: [Qemu-devel] [PATCH 11/22] qcow2-bitmap: add qcow2_store_persistent_bitmaps()

[Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>]

[-- Attachment #1: Type: text/plain, Size: 10445 bytes --]

On 30.09.2016 12:53, Vladimir Sementsov-Ogievskiy wrote:
> Realize block bitmap stroing interface, to allow qcow2 images store
> persistent bitmaps.
>
> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> ---
>  block/qcow2-bitmap.c | 241 +++++++++++++++++++++++++++++++++++++++++++++++++++
>  block/qcow2.c        |   2 +
>  block/qcow2.h        |   2 +
>  3 files changed, 245 insertions(+)
>
> diff --git a/block/qcow2-bitmap.c b/block/qcow2-bitmap.c
> index 81520cd..a5be25a 100644
> --- a/block/qcow2-bitmap.c
> +++ b/block/qcow2-bitmap.c
> @@ -27,6 +27,7 @@
>
>  #include "qemu/osdep.h"
>  #include "qapi/error.h"
> +#include "qemu/cutils.h"
>
>  #include "block/block_int.h"
>  #include "block/qcow2.h"
> @@ -96,6 +97,15 @@ static inline void bitmap_table_to_cpu(uint64_t *bitmap_table, size_t size)
>      }
>  }
>
> +static inline void bitmap_table_to_be(uint64_t *bitmap_table, size_t size)
> +{
> +    size_t i;
> +
> +    for (i = 0; i < size; ++i) {
> +        cpu_to_be64s(&bitmap_table[i]);
> +    }
> +}
> +
>  static inline int calc_dir_entry_size(size_t name_size, size_t extra_data_size)
>  {
>      return align_offset(sizeof(Qcow2BitmapDirEntry) +
> @@ -564,3 +574,234 @@ out:
>
>      return ret;
>  }
> +
> +/* store_bitmap_data()
> + * Store bitmap to image, filling bitamp table accordingly.


> + */
> +static int store_bitmap_data(BlockDriverState *bs, BdrvDirtyBitmap *bitmap,
> +                             uint64_t *bitmap_table, uint32_t bitmap_table_size)
> +{
> +    int ret;
> +    BDRVQcow2State *s = bs->opaque;
> +    uint64_t sector, dsc;
> +    uint64_t bm_size = bdrv_dirty_bitmap_size(bitmap);
> +    int cl_size = s->cluster_size;



> +    uint8_t *buf = NULL;
> +    uint32_t tb_size =
> +            size_to_clusters(s,
> +                bdrv_dirty_bitmap_serialization_size(bitmap, 0, bm_size));


> +
> +    BdrvDirtyBitmapIter *dbi;
> +
> +    if (tb_size != bitmap_table_size) {
> +        return -EINVAL;
> +    }
> +
> +    memset(bitmap_table, 0, bitmap_table_size * sizeof(bitmap_table[0]));


> +
> +    dbi = bdrv_dirty_iter_new(bitmap, 0);
> +    buf = g_malloc(cl_size);
> +    dsc = dirty_sectors_in_cluster(s, bitmap);
> +
> +    while ((sector = bdrv_dirty_iter_next(dbi)) != -1) {


> +        uint64_t cluster = sector / dsc;
> +        sector = cluster * dsc;



> +        uint64_t end = MIN(bm_size, sector + dsc);
> +        uint64_t write_size =
> +            bdrv_dirty_bitmap_serialization_size(bitmap, sector, end - sector);
> +
> +        int64_t off = qcow2_alloc_clusters(bs, cl_size);
> +        if (off < 0) {
> +            ret = off;
> +            goto finish;
> +        }
> +        bitmap_table[cluster] = off;
> +
> +        bdrv_dirty_bitmap_serialize_part(bitmap, buf, sector, end);



> +        if (write_size < cl_size) {
> +            memset(buf + write_size, 0, cl_size - write_size);
> +        }
> +


> +        ret = bdrv_pwrite(bs->file, off, buf, cl_size);
> +        if (ret < 0) {
> +            goto finish;
> +        }
> +
> +        if (end >= bm_size) {
> +            break;
> +        }
> +
> +        bdrv_set_dirty_iter(dbi, end);
> +    }
> +    ret = 0; /* writes */

What is that comment supposed to mean?

> +
> +finish:
> +    if (ret < 0) {
> +        clear_bitmap_table(bs, bitmap_table, bitmap_table_size);
> +    }
> +    g_free(buf);
> +    bdrv_dirty_iter_free(dbi);
> +
> +    return ret;

In case you decide to keep BME_MAX_PHYS_SIZE, this function should check
somewhere that the physical size of the bitmap does not exceed that value.

> +}
> +
> +/* store_bitmap()
> + * Store bitmap to qcow2 and set bitmap_table. bitmap_table itself is not
> + * stored to qcow2.

First of all, there is no parameter called "bitmap_table", and second,
yes, the bitmap table is written to the qcow2 file.

> + */
> +static int store_bitmap(BlockDriverState *bs,
> +                        BdrvDirtyBitmap *bitmap,
> +                        Qcow2BitmapDirEntry *entry)
> +{
> +    int ret;
> +    BDRVQcow2State *s = bs->opaque;
> +    uint64_t bm_size = bdrv_dirty_bitmap_size(bitmap);
> +    const char *bm_name = bdrv_dirty_bitmap_name(bitmap);
> +
> +    uint64_t *tb;
> +    int64_t tb_offset;
> +    uint32_t tb_size =
> +            size_to_clusters(s,
> +                bdrv_dirty_bitmap_serialization_size(bitmap, 0, bm_size));

As above, this variable should be of type uint64_t.

Also, you have to check that it does not exceed BME_MAX_TABLE_SIZE.

> +
> +    tb = g_try_new(uint64_t, tb_size);
> +    if (tb == NULL) {
> +        return -ENOMEM;
> +    }
> +
> +    ret = store_bitmap_data(bs, bitmap, tb, tb_size);
> +    if (ret < 0) {
> +        g_free(tb);
> +        return ret;
> +    }
> +
> +    tb_offset = qcow2_alloc_clusters(bs, tb_size * sizeof(tb[0]));

If you don't limit tb_size, then this multiplication can overflow on 32
bit machines.

> +    if (tb_offset < 0) {
> +        ret = tb_offset;
> +        goto fail;
> +    }
> +

There should be a metadata overlap check here.

> +    bitmap_table_to_be(tb, tb_size);
> +    ret = bdrv_pwrite(bs->file, tb_offset, tb, tb_size * sizeof(tb[0]));
> +    if (ret < 0) {
> +        goto fail;
> +    }
> +
> +    g_free(tb);
> +
> +    entry->bitmap_table_offset = tb_offset;
> +    entry->bitmap_table_size = tb_size;
> +    entry->flags = bdrv_dirty_bitmap_granularity(bitmap) ? BME_FLAG_AUTO : 0;

s/granularity/get_autoload/

> +    entry->type = BT_DIRTY_TRACKING_BITMAP;
> +    entry->granularity_bits = ctz32(bdrv_dirty_bitmap_granularity(bitmap));

You should probably check somewhere that the resulting value for
entry->granularity_bits is in the BME_{MIN,MAX}_GRANULARITY_BITS range.

> +    entry->name_size = strlen(bm_name);

And that this length does not exceed BME_MAX_NAME_SIZE.

> +    entry->extra_data_size = 0;
> +    memcpy(entry + 1, bm_name, entry->name_size);
> +
> +    return 0;
> +
> +fail:
> +    clear_bitmap_table(bs, tb, tb_size);
> +
> +    if (tb_offset > 0) {
> +        qcow2_free_clusters(bs, tb_offset, tb_size, QCOW2_DISCARD_ALWAYS);

As before, I'd vote for QCOW2_DISCARD_OTHER.

> +    }
> +
> +    g_free(tb);
> +
> +    return ret;
> +}
> +
> +static Qcow2BitmapDirEntry *find_bitmap_by_name(uint8_t *bitmap_directory,
> +                                                size_t size, const char *name)
> +{
> +    Qcow2BitmapDirEntry *e;
> +
> +    for_each_bitmap_dir_entry(e, bitmap_directory, size) {
> +        if (strncmp((char *)(e + 1), name, e->name_size) == 0) {
> +            return e;
> +        }
> +    }
> +
> +    return NULL;
> +}
> +
> +void qcow2_store_persistent_bitmaps(BlockDriverState *bs, Error **errp)
> +{
> +    BdrvDirtyBitmap *bm;
> +    BDRVQcow2State *s = bs->opaque;
> +    uint32_t new_nb_bitmaps = s->nb_bitmaps;
> +    uint64_t new_dir_size = s->bitmap_directory_size;
> +    uint8_t *dir = NULL, *new_dir = NULL;
> +    int ret;
> +    Qcow2BitmapDirEntry *new_pos;
> +
> +    if (s->nb_bitmaps > 0) {
> +        dir = directory_read(bs, s->bitmap_directory_offset,
> +                             s->bitmap_directory_size, errp);
> +        if (dir == NULL) {
> +            goto out;
> +        }
> +    }
> +
> +    for (bm = bdrv_dirty_bitmap_next(bs, NULL); bm != NULL;
> +            bm = bdrv_dirty_bitmap_next(bs, bm)) {
> +        const char *name = bdrv_dirty_bitmap_name(bm);
> +
> +        if (!bdrv_dirty_bitmap_get_persistance(bm)) {
> +            continue;
> +        }
> +
> +        if (s->nb_bitmaps > 0 &&
> +                find_bitmap_by_name(dir, s->bitmap_directory_size, name)) {
> +            error_setg(errp,
> +                       "Can't store bitmap '%s' to '%s', as it already exists",
> +                       name, bdrv_get_device_or_node_name(bs));
> +            goto out;
> +        }
> +
> +        new_nb_bitmaps++;
> +        new_dir_size += calc_dir_entry_size(strlen(name), 0);
> +    }
> +
> +    if (s->nb_bitmaps == new_nb_bitmaps) {
> +        /* No new bitmaps - nothing to do */
> +        goto out;
> +    }
> +
> +    new_dir = g_try_malloc0(new_dir_size);
> +    if (new_dir == NULL) {
> +        error_setg(errp, "Can't allocate space for bitmap directory.");
> +        goto out;
> +    }
> +
> +    memcpy(new_dir, dir, s->bitmap_directory_size);
> +    new_pos = (Qcow2BitmapDirEntry *)(new_dir + s->bitmap_directory_size);
> +
> +    for (bm = bdrv_dirty_bitmap_next(bs, NULL); bm != NULL;
> +            bm = bdrv_dirty_bitmap_next(bs, bm)) {
> +        if (!bdrv_dirty_bitmap_get_persistance(bm)) {
> +            continue;
> +        }
> +
> +        ret = store_bitmap(bs, bm, new_pos);
> +        if (ret < 0) {
> +            error_setg_errno(errp, -ret, "Can't store bitmap '%s' to '%s'",
> +                             bdrv_dirty_bitmap_name(bm),
> +                             bdrv_get_device_or_node_name(bs));
> +            goto out;
> +        }
> +        new_pos = next_dir_entry(new_pos);
> +    }
> +
> +    ret = directory_update(bs, new_dir, new_dir_size, new_nb_bitmaps);
> +    if (ret < 0) {
> +        error_setg_errno(errp, -ret, "Can't update bitmap directory in '%s'",
> +                         bdrv_get_device_or_node_name(bs));
> +        goto out;
> +    }
> +
> +out:
> +    g_free(new_dir);
> +    g_free(dir);

This error path leaks all the bitmaps that have been written
successfully (if any). I guess this is more or less fine if
directory_update() failed (because you can't really tell the state of
the image header after directory_update(), so better be safe) but it's
not so fine if just some store_bitmap() failed.

Max

> +}
> diff --git a/block/qcow2.c b/block/qcow2.c
> index 02ec224..8238205 100644
> --- a/block/qcow2.c
> +++ b/block/qcow2.c
> @@ -3493,6 +3493,8 @@ BlockDriver bdrv_qcow2 = {
>
>      .bdrv_detach_aio_context  = qcow2_detach_aio_context,
>      .bdrv_attach_aio_context  = qcow2_attach_aio_context,
> +
> +    .bdrv_store_persistent_bitmaps = qcow2_store_persistent_bitmaps,
>  };
>
>  static void bdrv_qcow2_init(void)
> diff --git a/block/qcow2.h b/block/qcow2.h
> index 482a29f..dfcf4c6 100644
> --- a/block/qcow2.h
> +++ b/block/qcow2.h
> @@ -627,4 +627,6 @@ int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
>      void **table);
>  void qcow2_cache_put(BlockDriverState *bs, Qcow2Cache *c, void **table);
>
> +void qcow2_store_persistent_bitmaps(BlockDriverState *bs, Error **errp);
> +
>  #endif
>




[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 481 bytes --]