diff mbox series

[PULL,v2,6/8] vmdk: Add read-only support for seSparse snapshots

Message ID 20190624144740.5338-7-mreitz@redhat.com (mailing list archive)
State New, archived
Headers show
Series [PULL,v2,1/8] nvme: do not advertise support for unsupported arbitration mechanism | expand

Commit Message

Max Reitz June 24, 2019, 2:47 p.m. UTC 
From: Sam Eiderman <shmuel.eiderman@oracle.com>

Until ESXi 6.5 VMware used the vmfsSparse format for snapshots (VMDK3 in
QEMU).

This format was lacking in the following:

    * Grain directory (L1) and grain table (L2) entries were 32-bit,
      allowing access to only 2TB (slightly less) of data.
    * The grain size (default) was 512 bytes - leading to data
      fragmentation and many grain tables.
    * For space reclamation purposes, it was necessary to find all the
      grains which are not pointed to by any grain table - so a reverse
      mapping of "offset of grain in vmdk" to "grain table" must be
      constructed - which takes large amounts of CPU/RAM.

The format specification can be found in VMware's documentation:
https://www.vmware.com/support/developer/vddk/vmdk_50_technote.pdf

In ESXi 6.5, to support snapshot files larger than 2TB, a new format was
introduced: SESparse (Space Efficient).

This format fixes the above issues:

    * All entries are now 64-bit.
    * The grain size (default) is 4KB.
    * Grain directory and grain tables are now located at the beginning
      of the file.
      + seSparse format reserves space for all grain tables.
      + Grain tables can be addressed using an index.
      + Grains are located in the end of the file and can also be
        addressed with an index.
      - seSparse vmdks of large disks (64TB) have huge preallocated
        headers - mainly due to L2 tables, even for empty snapshots.
    * The header contains a reverse mapping ("backmap") of "offset of
      grain in vmdk" to "grain table" and a bitmap ("free bitmap") which
      specifies for each grain - whether it is allocated or not.
      Using these data structures we can implement space reclamation
      efficiently.
    * Due to the fact that the header now maintains two mappings:
        * The regular one (grain directory & grain tables)
        * A reverse one (backmap and free bitmap)
      These data structures can lose consistency upon crash and result
      in a corrupted VMDK.
      Therefore, a journal is also added to the VMDK and is replayed
      when the VMware reopens the file after a crash.

Since ESXi 6.7 - SESparse is the only snapshot format available.

Unfortunately, VMware does not provide documentation regarding the new
seSparse format.

This commit is based on black-box research of the seSparse format.
Various in-guest block operations and their effect on the snapshot file
were tested.

The only VMware provided source of information (regarding the underlying
implementation) was a log file on the ESXi:

    /var/log/hostd.log

Whenever an seSparse snapshot is created - the log is being populated
with seSparse records.

Relevant log records are of the form:

[...] Const Header:
[...]  constMagic     = 0xcafebabe
[...]  version        = 2.1
[...]  capacity       = 204800
[...]  grainSize      = 8
[...]  grainTableSize = 64
[...]  flags          = 0
[...] Extents:
[...]  Header         : <1 : 1>
[...]  JournalHdr     : <2 : 2>
[...]  Journal        : <2048 : 2048>
[...]  GrainDirectory : <4096 : 2048>
[...]  GrainTables    : <6144 : 2048>
[...]  FreeBitmap     : <8192 : 2048>
[...]  BackMap        : <10240 : 2048>
[...]  Grain          : <12288 : 204800>
[...] Volatile Header:
[...] volatileMagic     = 0xcafecafe
[...] FreeGTNumber      = 0
[...] nextTxnSeqNumber  = 0
[...] replayJournal     = 0

The sizes that are seen in the log file are in sectors.
Extents are of the following format: <offset : size>

This commit is a strict implementation which enforces:
    * magics
    * version number 2.1
    * grain size of 8 sectors  (4KB)
    * grain table size of 64 sectors
    * zero flags
    * extent locations

Additionally, this commit proivdes only a subset of the functionality
offered by seSparse's format:
    * Read-only
    * No journal replay
    * No space reclamation
    * No unmap support

Hence, journal header, journal, free bitmap and backmap extents are
unused, only the "classic" (L1 -> L2 -> data) grain access is
implemented.

However there are several differences in the grain access itself.
Grain directory (L1):
    * Grain directory entries are indexes (not offsets) to grain
      tables.
    * Valid grain directory entries have their highest nibble set to
      0x1.
    * Since grain tables are always located in the beginning of the
      file - the index can fit into 32 bits - so we can use its low
      part if it's valid.
Grain table (L2):
    * Grain table entries are indexes (not offsets) to grains.
    * If the highest nibble of the entry is:
        0x0:
            The grain in not allocated.
            The rest of the bytes are 0.
        0x1:
            The grain is unmapped - guest sees a zero grain.
            The rest of the bits point to the previously mapped grain,
            see 0x3 case.
        0x2:
            The grain is zero.
        0x3:
            The grain is allocated - to get the index calculate:
            ((entry & 0x0fff000000000000) >> 48) |
            ((entry & 0x0000ffffffffffff) << 12)
    * The difference between 0x1 and 0x2 is that 0x1 is an unallocated
      grain which results from the guest using sg_unmap to unmap the
      grain - but the grain itself still exists in the grain extent - a
      space reclamation procedure should delete it.
      Unmapping a zero grain has no effect (0x2 will not change to 0x1)
      but unmapping an unallocated grain will (0x0 to 0x1) - naturally.

In order to implement seSparse some fields had to be changed to support
both 32-bit and 64-bit entry sizes.

Reviewed-by: Karl Heubaum <karl.heubaum@oracle.com>
Reviewed-by: Eyal Moscovici <eyal.moscovici@oracle.com>
Reviewed-by: Arbel Moshe <arbel.moshe@oracle.com>
Signed-off-by: Sam Eiderman <shmuel.eiderman@oracle.com>
Message-id: 20190620091057.47441-4-shmuel.eiderman@oracle.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
---
 block/vmdk.c | 358 ++++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 342 insertions(+), 16 deletions(-)

Comments

Peter Maydell July 2, 2019, 7:13 p.m. UTC | #1 
On Mon, 24 Jun 2019 at 15:48, Max Reitz <mreitz@redhat.com> wrote:
>
> From: Sam Eiderman <shmuel.eiderman@oracle.com>
>
> Until ESXi 6.5 VMware used the vmfsSparse format for snapshots (VMDK3 in
> QEMU).
>
> This format was lacking in the following:
>
>     * Grain directory (L1) and grain table (L2) entries were 32-bit,
>       allowing access to only 2TB (slightly less) of data.
>     * The grain size (default) was 512 bytes - leading to data
>       fragmentation and many grain tables.
>     * For space reclamation purposes, it was necessary to find all the
>       grains which are not pointed to by any grain table - so a reverse
>       mapping of "offset of grain in vmdk" to "grain table" must be
>       constructed - which takes large amounts of CPU/RAM.
>

Hi; this change has confused Coverity a bit (CID 1402786):

> @@ -481,7 +529,7 @@ static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
>      int i;
>
>      /* read the L1 table */
> -    l1_size = extent->l1_size * sizeof(uint32_t);
> +    l1_size = extent->l1_size * extent->entry_size;
>      extent->l1_table = g_try_malloc(l1_size);
>      if (l1_size && extent->l1_table == NULL) {
>          return -ENOMEM;

>      }
>
>      if (extent->l1_backup_table_offset) {
> +        assert(!extent->sesparse);
>          extent->l1_backup_table = g_try_malloc(l1_size);
>          if (l1_size && extent->l1_backup_table == NULL) {
>              ret = -ENOMEM;

Here we allocate extent->l1_backup_table via g_try_malloc(),
and our "ENOMEM" guard checks l1_size. However later on we do:

        for (i = 0; i < extent->l1_size; i++) {
            le32_to_cpus(&extent->l1_backup_table[i]);
        }

which is a dereference of l1_backup_table whose guard
is looking at extent->l1_size. Previously Coverity could
(probably) see this was OK because we were setting
   l1_size = extent->l1_size * sizeof(uint32_t)
so l1_size is 0 if and only if extent->l1_size is zero.
But now there's another possibility because we have changed
the initialization to
   l1_size = extent->l1_size * extent->entry_size;
so if extent->entry_size is zero then l1_size could be 0
(getting us past the ENOMEM check) but extent->l1_size
non-zero (causing us to try to dereference a NULL l1_backup_table
pointer).

This can't in fact happen because if extent->l1_size is
non-zero we'll assert that extent->entry_size is
either 8 or 4, but the logic has become sufficiently
convoluted that it's no longer clear to Coverity that this
is an impossible case. I could mark this as a false positive,
but maybe there's a way of rephrasing this code that makes
the logic a little clearer to both humans and robots?

Supplementary:

(1) this code:
    for (i = 0; i < extent->l1_size; i++) {
        if (extent->entry_size == sizeof(uint64_t)) {
            le64_to_cpus((uint64_t *)extent->l1_table + i);
        } else {
            assert(extent->entry_size == sizeof(uint32_t));
            le32_to_cpus((uint32_t *)extent->l1_table + i);
        }
    }
only asserts that extent->entry_size is sane if the l1_size
is non-zero, and it does it inside the loop so we assert
once for each entry. Hiding the assert like this might be
part of what's confusing Coverity.

(2) we allocate the l1_backup_table() with a size of l1_size,
which will differ depending on extent->entry_size; but then
we do the endianness conversion on it using
        for (i = 0; i < extent->l1_size; i++) {
            le32_to_cpus(&extent->l1_backup_table[i]);
        }
which assumes always 32-bit entries. Is that correct?

thanks
-- PMM
Max Reitz July 2, 2019, 7:46 p.m. UTC | #2 
On 02.07.19 21:13, Peter Maydell wrote:
> On Mon, 24 Jun 2019 at 15:48, Max Reitz <mreitz@redhat.com> wrote:
>>
>> From: Sam Eiderman <shmuel.eiderman@oracle.com>
>>
>> Until ESXi 6.5 VMware used the vmfsSparse format for snapshots (VMDK3 in
>> QEMU).
>>
>> This format was lacking in the following:
>>
>>     * Grain directory (L1) and grain table (L2) entries were 32-bit,
>>       allowing access to only 2TB (slightly less) of data.
>>     * The grain size (default) was 512 bytes - leading to data
>>       fragmentation and many grain tables.
>>     * For space reclamation purposes, it was necessary to find all the
>>       grains which are not pointed to by any grain table - so a reverse
>>       mapping of "offset of grain in vmdk" to "grain table" must be
>>       constructed - which takes large amounts of CPU/RAM.
>>
> 
> Hi; this change has confused Coverity a bit (CID 1402786):
> 
>> @@ -481,7 +529,7 @@ static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
>>      int i;
>>
>>      /* read the L1 table */
>> -    l1_size = extent->l1_size * sizeof(uint32_t);
>> +    l1_size = extent->l1_size * extent->entry_size;
>>      extent->l1_table = g_try_malloc(l1_size);
>>      if (l1_size && extent->l1_table == NULL) {
>>          return -ENOMEM;
> 
>>      }
>>
>>      if (extent->l1_backup_table_offset) {
>> +        assert(!extent->sesparse);
>>          extent->l1_backup_table = g_try_malloc(l1_size);
>>          if (l1_size && extent->l1_backup_table == NULL) {
>>              ret = -ENOMEM;
> 
> Here we allocate extent->l1_backup_table via g_try_malloc(),
> and our "ENOMEM" guard checks l1_size. However later on we do:
> 
>         for (i = 0; i < extent->l1_size; i++) {
>             le32_to_cpus(&extent->l1_backup_table[i]);
>         }
> 
> which is a dereference of l1_backup_table whose guard
> is looking at extent->l1_size. Previously Coverity could
> (probably) see this was OK because we were setting
>    l1_size = extent->l1_size * sizeof(uint32_t)
> so l1_size is 0 if and only if extent->l1_size is zero.
> But now there's another possibility because we have changed
> the initialization to
>    l1_size = extent->l1_size * extent->entry_size;
> so if extent->entry_size is zero then l1_size could be 0
> (getting us past the ENOMEM check) but extent->l1_size
> non-zero (causing us to try to dereference a NULL l1_backup_table
> pointer).
> 
> This can't in fact happen because if extent->l1_size is
> non-zero we'll assert that extent->entry_size is
> either 8 or 4, but the logic has become sufficiently
> convoluted that it's no longer clear to Coverity that this
> is an impossible case. I could mark this as a false positive,
> but maybe there's a way of rephrasing this code that makes
> the logic a little clearer to both humans and robots?

There is also the fact that entry_size is initialized to 4
(sizeof(uint32_t)) and updated to 8 (sizeof(uint64_t)) when the image is
in the seSparse subformat.  So entry_size can only ever be 4 or 8.

The only substantial thing I could imagine that might in some way make
it clearer would be to drop entry_size and always decide based on
extent->sesparse (because extent->entry_size == (extent->sesparse ? 8 : 4)).

But to me personally, that would not really be clearer.  I think it’s
reasonable to store the size of L1 entries in an extent attribute, and
remember that it’s always 4 or 8.

Of course, there is also the less substantial thing, which is adding a
comment “Size of L1 entries, can only be sizeof(uint32_t) or
sizeof(uint64_t)” above the entry_size definition.


I think Coverity is actually confused by aliasing.  The report says this:

> 1. Condition l1_size, taking false branch.

So l1_size must be 0, which means either extent->l1_size or
extent->entry_size must be 0.

It then says:

> 3. Condition i < extent->l1_size, taking true branch.

So extent->l1_size must be greater than 0.  (To be exact, it prints this
two times, so extent->l1_size must be 2.)

This is followed by:

> 4. Condition extent->entry_size == 8UL /* sizeof (uint64_t) */, taking true branch.

Er, so, extent->entry_size is 8?

That actually makes l1_size 16.  So somewhere between conditions 1 and
3/4, *extent must have changed its contents.

So it looks to me like Coverity just thinks that *extent may be used
concurrently.  Short of adding a “restrict”, I don’t know what to do
this but to close the report as a false positive.

> Supplementary:
> 
> (1) this code:
>     for (i = 0; i < extent->l1_size; i++) {
>         if (extent->entry_size == sizeof(uint64_t)) {
>             le64_to_cpus((uint64_t *)extent->l1_table + i);
>         } else {
>             assert(extent->entry_size == sizeof(uint32_t));
>             le32_to_cpus((uint32_t *)extent->l1_table + i);
>         }
>     }
> only asserts that extent->entry_size is sane if the l1_size
> is non-zero, and it does it inside the loop so we assert
> once for each entry. Hiding the assert like this might be
> part of what's confusing Coverity.
> 
> (2) we allocate the l1_backup_table() with a size of l1_size,
> which will differ depending on extent->entry_size; but then
> we do the endianness conversion on it using
>         for (i = 0; i < extent->l1_size; i++) {
>             le32_to_cpus(&extent->l1_backup_table[i]);
>         }
> which assumes always 32-bit entries. Is that correct?

Yes, because 64 bit entries only occur for the seSparse format.  At
least our current (RO) implementation does not use a backup table when
accessing seSparse images.  (Hence the assertion against that at the
beginning of the enclosing if block.)

Max
Peter Maydell July 4, 2019, 4:59 p.m. UTC | #3 
On Tue, 2 Jul 2019 at 20:46, Max Reitz <mreitz@redhat.com> wrote:
> So it looks to me like Coverity just thinks that *extent may be used
> concurrently.

Or it's just not smart enough to notice that it's assumed
several mutually contradictory things...

>  Short of adding a “restrict”, I don’t know what to do
> this but to close the report as a false positive.

OK; I'll mark it a false positive.

thanks
-- PMM
diff mbox series

Patch

diff --git a/block/vmdk.c b/block/vmdk.c
index 931eb2759c..bd36ece125 100644
--- a/block/vmdk.c
+++ b/block/vmdk.c
@@ -91,6 +91,44 @@  typedef struct {
     uint16_t compressAlgorithm;
 } QEMU_PACKED VMDK4Header;
 
+typedef struct VMDKSESparseConstHeader {
+    uint64_t magic;
+    uint64_t version;
+    uint64_t capacity;
+    uint64_t grain_size;
+    uint64_t grain_table_size;
+    uint64_t flags;
+    uint64_t reserved1;
+    uint64_t reserved2;
+    uint64_t reserved3;
+    uint64_t reserved4;
+    uint64_t volatile_header_offset;
+    uint64_t volatile_header_size;
+    uint64_t journal_header_offset;
+    uint64_t journal_header_size;
+    uint64_t journal_offset;
+    uint64_t journal_size;
+    uint64_t grain_dir_offset;
+    uint64_t grain_dir_size;
+    uint64_t grain_tables_offset;
+    uint64_t grain_tables_size;
+    uint64_t free_bitmap_offset;
+    uint64_t free_bitmap_size;
+    uint64_t backmap_offset;
+    uint64_t backmap_size;
+    uint64_t grains_offset;
+    uint64_t grains_size;
+    uint8_t pad[304];
+} QEMU_PACKED VMDKSESparseConstHeader;
+
+typedef struct VMDKSESparseVolatileHeader {
+    uint64_t magic;
+    uint64_t free_gt_number;
+    uint64_t next_txn_seq_number;
+    uint64_t replay_journal;
+    uint8_t pad[480];
+} QEMU_PACKED VMDKSESparseVolatileHeader;
+
 #define L2_CACHE_SIZE 16
 
 typedef struct VmdkExtent {
@@ -99,19 +137,23 @@  typedef struct VmdkExtent {
     bool compressed;
     bool has_marker;
     bool has_zero_grain;
+    bool sesparse;
+    uint64_t sesparse_l2_tables_offset;
+    uint64_t sesparse_clusters_offset;
+    int32_t entry_size;
     int version;
     int64_t sectors;
     int64_t end_sector;
     int64_t flat_start_offset;
     int64_t l1_table_offset;
     int64_t l1_backup_table_offset;
-    uint32_t *l1_table;
+    void *l1_table;
     uint32_t *l1_backup_table;
     unsigned int l1_size;
     uint32_t l1_entry_sectors;
 
     unsigned int l2_size;
-    uint32_t *l2_cache;
+    void *l2_cache;
     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
     uint32_t l2_cache_counts[L2_CACHE_SIZE];
 
@@ -435,6 +477,11 @@  static int vmdk_add_extent(BlockDriverState *bs,
          *            minimal L2 table size: 512 entries
          *            8 TB is still more than the maximal value supported for
          *            VMDK3 & VMDK4 which is 2TB.
+         *     64TB - for "ESXi seSparse Extent"
+         *            minimal cluster size: 512B (default is 4KB)
+         *            L2 table size: 4096 entries (const).
+         *            64TB is more than the maximal value supported for
+         *            seSparse VMDKs (which is slightly less than 64TB)
          */
         error_setg(errp, "L1 size too big");
         return -EFBIG;
@@ -460,6 +507,7 @@  static int vmdk_add_extent(BlockDriverState *bs,
     extent->l2_size = l2_size;
     extent->cluster_sectors = flat ? sectors : cluster_sectors;
     extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
+    extent->entry_size = sizeof(uint32_t);
 
     if (s->num_extents > 1) {
         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
@@ -481,7 +529,7 @@  static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
     int i;
 
     /* read the L1 table */
-    l1_size = extent->l1_size * sizeof(uint32_t);
+    l1_size = extent->l1_size * extent->entry_size;
     extent->l1_table = g_try_malloc(l1_size);
     if (l1_size && extent->l1_table == NULL) {
         return -ENOMEM;
@@ -499,10 +547,16 @@  static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
         goto fail_l1;
     }
     for (i = 0; i < extent->l1_size; i++) {
-        le32_to_cpus(&extent->l1_table[i]);
+        if (extent->entry_size == sizeof(uint64_t)) {
+            le64_to_cpus((uint64_t *)extent->l1_table + i);
+        } else {
+            assert(extent->entry_size == sizeof(uint32_t));
+            le32_to_cpus((uint32_t *)extent->l1_table + i);
+        }
     }
 
     if (extent->l1_backup_table_offset) {
+        assert(!extent->sesparse);
         extent->l1_backup_table = g_try_malloc(l1_size);
         if (l1_size && extent->l1_backup_table == NULL) {
             ret = -ENOMEM;
@@ -525,7 +579,7 @@  static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
     }
 
     extent->l2_cache =
-        g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
+        g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE);
     return 0;
  fail_l1b:
     g_free(extent->l1_backup_table);
@@ -571,6 +625,205 @@  static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
     return ret;
 }
 
+#define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe)
+#define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe)
+
+/* Strict checks - format not officially documented */
+static int check_se_sparse_const_header(VMDKSESparseConstHeader *header,
+                                        Error **errp)
+{
+    header->magic = le64_to_cpu(header->magic);
+    header->version = le64_to_cpu(header->version);
+    header->grain_size = le64_to_cpu(header->grain_size);
+    header->grain_table_size = le64_to_cpu(header->grain_table_size);
+    header->flags = le64_to_cpu(header->flags);
+    header->reserved1 = le64_to_cpu(header->reserved1);
+    header->reserved2 = le64_to_cpu(header->reserved2);
+    header->reserved3 = le64_to_cpu(header->reserved3);
+    header->reserved4 = le64_to_cpu(header->reserved4);
+
+    header->volatile_header_offset =
+        le64_to_cpu(header->volatile_header_offset);
+    header->volatile_header_size = le64_to_cpu(header->volatile_header_size);
+
+    header->journal_header_offset = le64_to_cpu(header->journal_header_offset);
+    header->journal_header_size = le64_to_cpu(header->journal_header_size);
+
+    header->journal_offset = le64_to_cpu(header->journal_offset);
+    header->journal_size = le64_to_cpu(header->journal_size);
+
+    header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset);
+    header->grain_dir_size = le64_to_cpu(header->grain_dir_size);
+
+    header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset);
+    header->grain_tables_size = le64_to_cpu(header->grain_tables_size);
+
+    header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset);
+    header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size);
+
+    header->backmap_offset = le64_to_cpu(header->backmap_offset);
+    header->backmap_size = le64_to_cpu(header->backmap_size);
+
+    header->grains_offset = le64_to_cpu(header->grains_offset);
+    header->grains_size = le64_to_cpu(header->grains_size);
+
+    if (header->magic != SESPARSE_CONST_HEADER_MAGIC) {
+        error_setg(errp, "Bad const header magic: 0x%016" PRIx64,
+                   header->magic);
+        return -EINVAL;
+    }
+
+    if (header->version != 0x0000000200000001) {
+        error_setg(errp, "Unsupported version: 0x%016" PRIx64,
+                   header->version);
+        return -ENOTSUP;
+    }
+
+    if (header->grain_size != 8) {
+        error_setg(errp, "Unsupported grain size: %" PRIu64,
+                   header->grain_size);
+        return -ENOTSUP;
+    }
+
+    if (header->grain_table_size != 64) {
+        error_setg(errp, "Unsupported grain table size: %" PRIu64,
+                   header->grain_table_size);
+        return -ENOTSUP;
+    }
+
+    if (header->flags != 0) {
+        error_setg(errp, "Unsupported flags: 0x%016" PRIx64,
+                   header->flags);
+        return -ENOTSUP;
+    }
+
+    if (header->reserved1 != 0 || header->reserved2 != 0 ||
+        header->reserved3 != 0 || header->reserved4 != 0) {
+        error_setg(errp, "Unsupported reserved bits:"
+                   " 0x%016" PRIx64 " 0x%016" PRIx64
+                   " 0x%016" PRIx64 " 0x%016" PRIx64,
+                   header->reserved1, header->reserved2,
+                   header->reserved3, header->reserved4);
+        return -ENOTSUP;
+    }
+
+    /* check that padding is 0 */
+    if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
+        error_setg(errp, "Unsupported non-zero const header padding");
+        return -ENOTSUP;
+    }
+
+    return 0;
+}
+
+static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header,
+                                           Error **errp)
+{
+    header->magic = le64_to_cpu(header->magic);
+    header->free_gt_number = le64_to_cpu(header->free_gt_number);
+    header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number);
+    header->replay_journal = le64_to_cpu(header->replay_journal);
+
+    if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) {
+        error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64,
+                   header->magic);
+        return -EINVAL;
+    }
+
+    if (header->replay_journal) {
+        error_setg(errp, "Image is dirty, Replaying journal not supported");
+        return -ENOTSUP;
+    }
+
+    /* check that padding is 0 */
+    if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
+        error_setg(errp, "Unsupported non-zero volatile header padding");
+        return -ENOTSUP;
+    }
+
+    return 0;
+}
+
+static int vmdk_open_se_sparse(BlockDriverState *bs,
+                               BdrvChild *file,
+                               int flags, Error **errp)
+{
+    int ret;
+    VMDKSESparseConstHeader const_header;
+    VMDKSESparseVolatileHeader volatile_header;
+    VmdkExtent *extent;
+
+    ret = bdrv_apply_auto_read_only(bs,
+            "No write support for seSparse images available", errp);
+    if (ret < 0) {
+        return ret;
+    }
+
+    assert(sizeof(const_header) == SECTOR_SIZE);
+
+    ret = bdrv_pread(file, 0, &const_header, sizeof(const_header));
+    if (ret < 0) {
+        bdrv_refresh_filename(file->bs);
+        error_setg_errno(errp, -ret,
+                         "Could not read const header from file '%s'",
+                         file->bs->filename);
+        return ret;
+    }
+
+    /* check const header */
+    ret = check_se_sparse_const_header(&const_header, errp);
+    if (ret < 0) {
+        return ret;
+    }
+
+    assert(sizeof(volatile_header) == SECTOR_SIZE);
+
+    ret = bdrv_pread(file,
+                     const_header.volatile_header_offset * SECTOR_SIZE,
+                     &volatile_header, sizeof(volatile_header));
+    if (ret < 0) {
+        bdrv_refresh_filename(file->bs);
+        error_setg_errno(errp, -ret,
+                         "Could not read volatile header from file '%s'",
+                         file->bs->filename);
+        return ret;
+    }
+
+    /* check volatile header */
+    ret = check_se_sparse_volatile_header(&volatile_header, errp);
+    if (ret < 0) {
+        return ret;
+    }
+
+    ret = vmdk_add_extent(bs, file, false,
+                          const_header.capacity,
+                          const_header.grain_dir_offset * SECTOR_SIZE,
+                          0,
+                          const_header.grain_dir_size *
+                          SECTOR_SIZE / sizeof(uint64_t),
+                          const_header.grain_table_size *
+                          SECTOR_SIZE / sizeof(uint64_t),
+                          const_header.grain_size,
+                          &extent,
+                          errp);
+    if (ret < 0) {
+        return ret;
+    }
+
+    extent->sesparse = true;
+    extent->sesparse_l2_tables_offset = const_header.grain_tables_offset;
+    extent->sesparse_clusters_offset = const_header.grains_offset;
+    extent->entry_size = sizeof(uint64_t);
+
+    ret = vmdk_init_tables(bs, extent, errp);
+    if (ret) {
+        /* free extent allocated by vmdk_add_extent */
+        vmdk_free_last_extent(bs);
+    }
+
+    return ret;
+}
+
 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
                                QDict *options, Error **errp);
 
@@ -848,6 +1101,7 @@  static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
          * RW [size in sectors] SPARSE "file-name.vmdk"
          * RW [size in sectors] VMFS "file-name.vmdk"
          * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
+         * RW [size in sectors] SESPARSE "file-name.vmdk"
          */
         flat_offset = -1;
         matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
@@ -870,7 +1124,8 @@  static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
 
         if (sectors <= 0 ||
             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
-             strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
+             strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
+             strcmp(type, "SESPARSE")) ||
             (strcmp(access, "RW"))) {
             continue;
         }
@@ -923,6 +1178,13 @@  static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
                 return ret;
             }
             extent = &s->extents[s->num_extents - 1];
+        } else if (!strcmp(type, "SESPARSE")) {
+            ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
+            if (ret) {
+                bdrv_unref_child(bs, extent_file);
+                return ret;
+            }
+            extent = &s->extents[s->num_extents - 1];
         } else {
             error_setg(errp, "Unsupported extent type '%s'", type);
             bdrv_unref_child(bs, extent_file);
@@ -957,6 +1219,7 @@  static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
     if (strcmp(ct, "monolithicFlat") &&
         strcmp(ct, "vmfs") &&
         strcmp(ct, "vmfsSparse") &&
+        strcmp(ct, "seSparse") &&
         strcmp(ct, "twoGbMaxExtentSparse") &&
         strcmp(ct, "twoGbMaxExtentFlat")) {
         error_setg(errp, "Unsupported image type '%s'", ct);
@@ -1207,10 +1470,12 @@  static int get_cluster_offset(BlockDriverState *bs,
 {
     unsigned int l1_index, l2_offset, l2_index;
     int min_index, i, j;
-    uint32_t min_count, *l2_table;
+    uint32_t min_count;
+    void *l2_table;
     bool zeroed = false;
     int64_t ret;
     int64_t cluster_sector;
+    unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
 
     if (m_data) {
         m_data->valid = 0;
@@ -1225,7 +1490,36 @@  static int get_cluster_offset(BlockDriverState *bs,
     if (l1_index >= extent->l1_size) {
         return VMDK_ERROR;
     }
-    l2_offset = extent->l1_table[l1_index];
+    if (extent->sesparse) {
+        uint64_t l2_offset_u64;
+
+        assert(extent->entry_size == sizeof(uint64_t));
+
+        l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
+        if (l2_offset_u64 == 0) {
+            l2_offset = 0;
+        } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
+            /*
+             * Top most nibble is 0x1 if grain table is allocated.
+             * strict check - top most 4 bytes must be 0x10000000 since max
+             * supported size is 64TB for disk - so no more than 64TB / 16MB
+             * grain directories which is smaller than uint32,
+             * where 16MB is the only supported default grain table coverage.
+             */
+            return VMDK_ERROR;
+        } else {
+            l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
+            l2_offset_u64 = extent->sesparse_l2_tables_offset +
+                l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
+            if (l2_offset_u64 > 0x00000000ffffffff) {
+                return VMDK_ERROR;
+            }
+            l2_offset = (unsigned int)(l2_offset_u64);
+        }
+    } else {
+        assert(extent->entry_size == sizeof(uint32_t));
+        l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
+    }
     if (!l2_offset) {
         return VMDK_UNALLOC;
     }
@@ -1237,7 +1531,7 @@  static int get_cluster_offset(BlockDriverState *bs,
                     extent->l2_cache_counts[j] >>= 1;
                 }
             }
-            l2_table = extent->l2_cache + (i * extent->l2_size);
+            l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
             goto found;
         }
     }
@@ -1250,13 +1544,13 @@  static int get_cluster_offset(BlockDriverState *bs,
             min_index = i;
         }
     }
-    l2_table = extent->l2_cache + (min_index * extent->l2_size);
+    l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
     BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
     if (bdrv_pread(extent->file,
                 (int64_t)l2_offset * 512,
                 l2_table,
-                extent->l2_size * sizeof(uint32_t)
-            ) != extent->l2_size * sizeof(uint32_t)) {
+                l2_size_bytes
+            ) != l2_size_bytes) {
         return VMDK_ERROR;
     }
 
@@ -1264,16 +1558,45 @@  static int get_cluster_offset(BlockDriverState *bs,
     extent->l2_cache_counts[min_index] = 1;
  found:
     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
-    cluster_sector = le32_to_cpu(l2_table[l2_index]);
 
-    if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
-        zeroed = true;
+    if (extent->sesparse) {
+        cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
+        switch (cluster_sector & 0xf000000000000000) {
+        case 0x0000000000000000:
+            /* unallocated grain */
+            if (cluster_sector != 0) {
+                return VMDK_ERROR;
+            }
+            break;
+        case 0x1000000000000000:
+            /* scsi-unmapped grain - fallthrough */
+        case 0x2000000000000000:
+            /* zero grain */
+            zeroed = true;
+            break;
+        case 0x3000000000000000:
+            /* allocated grain */
+            cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
+                              ((cluster_sector & 0x0000ffffffffffff) << 12));
+            cluster_sector = extent->sesparse_clusters_offset +
+                cluster_sector * extent->cluster_sectors;
+            break;
+        default:
+            return VMDK_ERROR;
+        }
+    } else {
+        cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
+
+        if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
+            zeroed = true;
+        }
     }
 
     if (!cluster_sector || zeroed) {
         if (!allocate) {
             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
         }
+        assert(!extent->sesparse);
 
         if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
             return VMDK_ERROR;
@@ -1297,7 +1620,7 @@  static int get_cluster_offset(BlockDriverState *bs,
             m_data->l1_index = l1_index;
             m_data->l2_index = l2_index;
             m_data->l2_offset = l2_offset;
-            m_data->l2_cache_entry = &l2_table[l2_index];
+            m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
         }
     }
     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
@@ -1623,6 +1946,9 @@  static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
         if (!extent) {
             return -EIO;
         }
+        if (extent->sesparse) {
+            return -ENOTSUP;
+        }
         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
                              - offset_in_cluster);