A qcow2 image file is organized in units of constant size, which are called (host) clusters. A cluster is the unit in which all allocations are done, both for actual guest data and for image metadata.
Likewise, the virtual disk as seen by the guest is divided into (guest) clusters of the same size.
All numbers in qcow2 are stored in Big Endian byte order.
The first cluster of a qcow2 image contains the file header:
Byte 0 - 3: magic QCOW magic string ("QFI\xfb") 4 - 7: version Version number (valid values are 2 and 3) 8 - 15: backing_file_offset Offset into the image file at which the backing file name is stored (NB: The string is not null terminated). 0 if the image doesn't have a backing file. 16 - 19: backing_file_size Length of the backing file name in bytes. Must not be longer than 1023 bytes. Undefined if the image doesn't have a backing file. 20 - 23: cluster_bits Number of bits that are used for addressing an offset within a cluster (1 << cluster_bits is the cluster size). Must not be less than 9 (i.e. 512 byte clusters). Note: qemu as of today has an implementation limit of 2 MB as the maximum cluster size and won't be able to open images with larger cluster sizes. 24 - 31: size Virtual disk size in bytes 32 - 35: crypt_method 0 for no encryption 1 for AES encryption 2 for LUKS encryption 36 - 39: l1_size Number of entries in the active L1 table 40 - 47: l1_table_offset Offset into the image file at which the active L1 table starts. Must be aligned to a cluster boundary. 48 - 55: refcount_table_offset Offset into the image file at which the refcount table starts. Must be aligned to a cluster boundary. 56 - 59: refcount_table_clusters Number of clusters that the refcount table occupies 60 - 63: nb_snapshots Number of snapshots contained in the image 64 - 71: snapshots_offset Offset into the image file at which the snapshot table starts. Must be aligned to a cluster boundary.
If the version is 3 or higher, the header has the following additional fields. For version 2, the values are assumed to be zero, unless specified otherwise in the description of a field.
72 - 79: incompatible_features Bitmask of incompatible features. An implementation must fail to open an image if an unknown bit is set. Bit 0: Dirty bit. If this bit is set then refcounts may be inconsistent, make sure to scan L1/L2 tables to repair refcounts before accessing the image. Bit 1: Corrupt bit. If this bit is set then any data structure may be corrupt and the image must not be written to (unless for regaining consistency). Bits 2-63: Reserved (set to 0) 80 - 87: compatible_features Bitmask of compatible features. An implementation can safely ignore any unknown bits that are set. Bit 0: Lazy refcounts bit. If this bit is set then lazy refcount updates can be used. This means marking the image file dirty and postponing refcount metadata updates. Bits 1-63: Reserved (set to 0) 88 - 95: autoclear_features Bitmask of auto-clear features. An implementation may only write to an image with unknown auto-clear features if it clears the respective bits from this field first. Bit 0: Bitmaps extension bit This bit indicates consistency for the bitmaps extension data. It is an error if this bit is set without the bitmaps extension present. If the bitmaps extension is present but this bit is unset, the bitmaps extension data must be considered inconsistent. Bits 1-63: Reserved (set to 0) 96 - 99: refcount_order Describes the width of a reference count block entry (width in bits: refcount_bits = 1 << refcount_order). For version 2 images, the order is always assumed to be 4 (i.e. refcount_bits = 16). This value may not exceed 6 (i.e. refcount_bits = 64). 100 - 103: header_length Length of the header structure in bytes. For version 2 images, the length is always assumed to be 72 bytes.
Directly after the image header, optional sections called header extensions can be stored. Each extension has a structure like the following:
Byte 0 - 3: Header extension type: 0x00000000 - End of the header extension area 0xE2792ACA - Backing file format name 0x6803f857 - Feature name table 0x23852875 - Bitmaps extension 0x0537be77 - Full disk encryption header pointer other - Unknown header extension, can be safely ignored 4 - 7: Length of the header extension data 8 - n: Header extension data n - m: Padding to round up the header extension size to the next multiple of 8.
Unless stated otherwise, each header extension type shall appear at most once in the same image.
If the image has a backing file then the backing file name should be stored in the remaining space between the end of the header extension area and the end of the first cluster. It is not allowed to store other data here, so that an implementation can safely modify the header and add extensions without harming data of compatible features that it doesn’t support. Compatible features that need space for additional data can use a header extension.
The feature name table is an optional header extension that contains the name for features used by the image. It can be used by applications that don’t know the respective feature (e.g. because the feature was introduced only later) to display a useful error message.
The number of entries in the feature name table is determined by the length of the header extension data. Each entry look like this:
Byte 0: Type of feature (select feature bitmap) 0: Incompatible feature 1: Compatible feature 2: Autoclear feature 1: Bit number within the selected feature bitmap (valid values: 0-63) 2 - 47: Feature name (padded with zeros, but not necessarily null terminated if it has full length)
The bitmaps extension is an optional header extension. It provides the ability to store bitmaps related to a virtual disk. For now, there is only one bitmap type: the dirty tracking bitmap, which tracks virtual disk changes from some point in time.
The data of the extension should be considered consistent only if the corresponding auto-clear feature bit is set, see autoclear_features above.
The fields of the bitmaps extension are:
Byte 0 - 3: nb_bitmaps The number of bitmaps contained in the image. Must be greater than or equal to 1. Note: Qemu currently only supports up to 65535 bitmaps per image. 4 - 7: Reserved, must be zero. 8 - 15: bitmap_directory_size Size of the bitmap directory in bytes. It is the cumulative size of all (nb_bitmaps) bitmap directory entries. 16 - 23: bitmap_directory_offset Offset into the image file at which the bitmap directory starts. Must be aligned to a cluster boundary.
The full disk encryption header must be present if, and only if, the ‘crypt_method’ header requires metadata. Currently this is only true of the ‘LUKS’ crypt method. The header extension must be absent for other methods.
This header provides the offset at which the crypt method can store its additional data, as well as the length of such data.
Byte 0 - 7: Offset into the image file at which the encryption header starts in bytes. Must be aligned to a cluster boundary. Byte 8 - 15: Length of the written encryption header in bytes. Note actual space allocated in the qcow2 file may be larger than this value, since it will be rounded to the nearest multiple of the cluster size. Any unused bytes in the allocated space will be initialized to 0.
For the LUKS crypt method, the encryption header works as follows.
The first 592 bytes of the header clusters will contain the LUKS partition header. This is then followed by the key material data areas. The size of the key material data areas is determined by the number of stripes in the key slot and key size. Refer to the LUKS format specification (‘docs/on-disk-format.pdf’ in the cryptsetup source package) for details of the LUKS partition header format.
In the LUKS partition header, the “payload-offset” field will be calculated as normal for the LUKS spec. ie the size of the LUKS header, plus key material regions, plus padding, relative to the start of the LUKS header. This offset value is not required to be qcow2 cluster aligned. Its value is currently never used in the context of qcow2, since the qcow2 file format itself defines where the real payload offset is, but none the less a valid payload offset should always be present.
In the LUKS key slots header, the “key-material-offset” is relative to the start of the LUKS header clusters in the qcow2 container, not the start of the qcow2 file.
Logically the layout looks like:
+-----------------------------+ | QCow2 header | | QCow2 header extension X | | QCow2 header extension FDE | | QCow2 header extension ... | | QCow2 header extension Z | +-----------------------------+ | ....other QCow2 tables.... | . . . . +-----------------------------+ | +-------------------------+ | | | LUKS partition header | | | +-------------------------+ | | | LUKS key material 1 | | | +-------------------------+ | | | LUKS key material 2 | | | +-------------------------+ | | | LUKS key material ... | | | +-------------------------+ | | | LUKS key material 8 | | | +-------------------------+ | +-----------------------------+ | QCow2 cluster payload | . . . . . . | | +-----------------------------+
When an encryption method is requested in the header, the image payload data must be encrypted/decrypted on every write/read. The image headers and metadata are never encrypted.
The algorithms used for encryption vary depending on the method
The AES cipher, in CBC mode, with 256 bit keys.
Initialization vectors generated using plain64 method, with the virtual disk sector as the input tweak.
This format is no longer supported in QEMU system emulators, due to a number of design flaws affecting its security. It is only supported in the command line tools for the sake of back compatibility and data liberation.
The algorithms are specified in the LUKS header.
Initialization vectors generated using the method specified in the LUKS header, with the physical disk sector as the input tweak.
qcow2 manages the allocation of host clusters by maintaining a reference count for each host cluster. A refcount of 0 means that the cluster is free, 1 means that it is used, and >= 2 means that it is used and any write access must perform a COW (copy on write) operation.
The refcounts are managed in a two-level table. The first level is called refcount table and has a variable size (which is stored in the header). The refcount table can cover multiple clusters, however it needs to be contiguous in the image file.
It contains pointers to the second level structures which are called refcount blocks and are exactly one cluster in size.
Given a offset into the image file, the refcount of its cluster can be obtained as follows:
refcount_block_entries = (cluster_size * 8 / refcount_bits) refcount_block_index = (offset / cluster_size) % refcount_block_entries refcount_table_index = (offset / cluster_size) / refcount_block_entries refcount_block = load_cluster(refcount_table[refcount_table_index]); return refcount_block[refcount_block_index];
Refcount table entry:
Bit 0 - 8: Reserved (set to 0) 9 - 63: Bits 9-63 of the offset into the image file at which the refcount block starts. Must be aligned to a cluster boundary. If this is 0, the corresponding refcount block has not yet been allocated. All refcounts managed by this refcount block are 0.
Refcount block entry (x = refcount_bits - 1):
Bit 0 - x: Reference count of the cluster. If refcount_bits implies a sub-byte width, note that bit 0 means the least significant bit in this context.
Just as for refcounts, qcow2 uses a two-level structure for the mapping of guest clusters to host clusters. They are called L1 and L2 table.
The L1 table has a variable size (stored in the header) and may use multiple clusters, however it must be contiguous in the image file. L2 tables are exactly one cluster in size.
Given a offset into the virtual disk, the offset into the image file can be obtained as follows:
l2_entries = (cluster_size / sizeof(uint64_t)) l2_index = (offset / cluster_size) % l2_entries l1_index = (offset / cluster_size) / l2_entries l2_table = load_cluster(l1_table[l1_index]); cluster_offset = l2_table[l2_index]; return cluster_offset + (offset % cluster_size)
L1 table entry:
Bit 0 - 8: Reserved (set to 0) 9 - 55: Bits 9-55 of the offset into the image file at which the L2 table starts. Must be aligned to a cluster boundary. If the offset is 0, the L2 table and all clusters described by this L2 table are unallocated. 56 - 62: Reserved (set to 0) 63: 0 for an L2 table that is unused or requires COW, 1 if its refcount is exactly one. This information is only accurate in the active L1 table.
L2 table entry:
Bit 0 - 61: Cluster descriptor 62: 0 for standard clusters 1 for compressed clusters 63: 0 for a cluster that is unused or requires COW, 1 if its refcount is exactly one. This information is only accurate in L2 tables that are reachable from the active L1 table.
Standard Cluster Descriptor:
Bit 0: If set to 1, the cluster reads as all zeros. The host cluster offset can be used to describe a preallocation, but it won't be used for reading data from this cluster, nor is data read from the backing file if the cluster is unallocated. With version 2, this is always 0. 1 - 8: Reserved (set to 0) 9 - 55: Bits 9-55 of host cluster offset. Must be aligned to a cluster boundary. If the offset is 0, the cluster is unallocated. 56 - 61: Reserved (set to 0)
Compressed Clusters Descriptor (x = 62 - (cluster_bits - 8)):
Bit 0 - x: Host cluster offset. This is usually _not_ aligned to a cluster boundary! x+1 - 61: Compressed size of the images in sectors of 512 bytes
If a cluster is unallocated, read requests shall read the data from the backing file (except if bit 0 in the Standard Cluster Descriptor is set). If there is no backing file or the backing file is smaller than the image, they shall read zeros for all parts that are not covered by the backing file.
qcow2 supports internal snapshots. Their basic principle of operation is to switch the active L1 table, so that a different set of host clusters are exposed to the guest.
When creating a snapshot, the L1 table should be copied and the refcount of all L2 tables and clusters reachable from this L1 table must be increased, so that a write causes a COW and isn’t visible in other snapshots.
When loading a snapshot, bit 63 of all entries in the new active L1 table and all L2 tables referenced by it must be reconstructed from the refcount table as it doesn’t need to be accurate in inactive L1 tables.
A directory of all snapshots is stored in the snapshot table, a contiguous area in the image file, whose starting offset and length are given by the header fields snapshots_offset and nb_snapshots. The entries of the snapshot table have variable length, depending on the length of ID, name and extra data.
Snapshot table entry:
Byte 0 - 7: Offset into the image file at which the L1 table for the snapshot starts. Must be aligned to a cluster boundary. 8 - 11: Number of entries in the L1 table of the snapshots 12 - 13: Length of the unique ID string describing the snapshot 14 - 15: Length of the name of the snapshot 16 - 19: Time at which the snapshot was taken in seconds since the Epoch 20 - 23: Subsecond part of the time at which the snapshot was taken in nanoseconds 24 - 31: Time that the guest was running until the snapshot was taken in nanoseconds 32 - 35: Size of the VM state in bytes. 0 if no VM state is saved. If there is VM state, it starts at the first cluster described by first L1 table entry that doesn't describe a regular guest cluster (i.e. VM state is stored like guest disk content, except that it is stored at offsets that are larger than the virtual disk presented to the guest) 36 - 39: Size of extra data in the table entry (used for future extensions of the format) variable: Extra data for future extensions. Unknown fields must be ignored. Currently defined are (offset relative to snapshot table entry): Byte 40 - 47: Size of the VM state in bytes. 0 if no VM state is saved. If this field is present, the 32-bit value in bytes 32-35 is ignored. Byte 48 - 55: Virtual disk size of the snapshot in bytes Version 3 images must include extra data at least up to byte 55. variable: Unique ID string for the snapshot (not null terminated) variable: Name of the snapshot (not null terminated) variable: Padding to round up the snapshot table entry size to the next multiple of 8.
As mentioned above, the bitmaps extension provides the ability to store bitmaps related to a virtual disk. This section describes how these bitmaps are stored.
All stored bitmaps are related to the virtual disk stored in the same image, so each bitmap size is equal to the virtual disk size.
Each bit of the bitmap is responsible for strictly defined range of the virtual disk. For bit number bit_nr the corresponding range (in bytes) will be:
[bit_nr * bitmap_granularity .. (bit_nr + 1) * bitmap_granularity - 1]
Granularity is a property of the concrete bitmap, see below.
Each bitmap saved in the image is described in a bitmap directory entry. The bitmap directory is a contiguous area in the image file, whose starting offset and length are given by the header extension fields bitmap_directory_offset and bitmap_directory_size. The entries of the bitmap directory have variable length, depending on the lengths of the bitmap name and extra data.
Structure of a bitmap directory entry:
Byte 0 - 7: bitmap_table_offset Offset into the image file at which the bitmap table (described below) for the bitmap starts. Must be aligned to a cluster boundary. 8 - 11: bitmap_table_size Number of entries in the bitmap table of the bitmap. 12 - 15: flags Bit 0: in_use The bitmap was not saved correctly and may be inconsistent. 1: auto The bitmap must reflect all changes of the virtual disk by any application that would write to this qcow2 file (including writes, snapshot switching, etc.). The type of this bitmap must be 'dirty tracking bitmap'. 2: extra_data_compatible This flags is meaningful when the extra data is unknown to the software (currently any extra data is unknown to Qemu). If it is set, the bitmap may be used as expected, extra data must be left as is. If it is not set, the bitmap must not be used, but both it and its extra data be left as is. Bits 3 - 31 are reserved and must be 0. 16: type This field describes the sort of the bitmap. Values: 1: Dirty tracking bitmap Values 0, 2 - 255 are reserved. 17: granularity_bits Granularity bits. Valid values: 0 - 63. Note: Qemu currently supports only values 9 - 31. Granularity is calculated as granularity = 1 << granularity_bits A bitmap's granularity is how many bytes of the image accounts for one bit of the bitmap. 18 - 19: name_size Size of the bitmap name. Must be non-zero. Note: Qemu currently doesn't support values greater than 1023. 20 - 23: extra_data_size Size of type-specific extra data. For now, as no extra data is defined, extra_data_size is reserved and should be zero. If it is non-zero the behavior is defined by extra_data_compatible flag. variable: extra_data Extra data for the bitmap, occupying extra_data_size bytes. Extra data must never contain references to clusters or in some other way allocate additional clusters. variable: name The name of the bitmap (not null terminated), occupying name_size bytes. Must be unique among all bitmap names within the bitmaps extension. variable: Padding to round up the bitmap directory entry size to the next multiple of 8. All bytes of the padding must be zero.
Each bitmap is stored using a one-level structure (as opposed to two-level structures like for refcounts and guest clusters mapping) for the mapping of bitmap data to host clusters. This structure is called the bitmap table.
Each bitmap table has a variable size (stored in the bitmap directory entry) and may use multiple clusters, however, it must be contiguous in the image file.
Structure of a bitmap table entry:
Bit 0: Reserved and must be zero if bits 9 - 55 are non-zero. If bits 9 - 55 are zero: 0: Cluster should be read as all zeros. 1: Cluster should be read as all ones. 1 - 8: Reserved and must be zero. 9 - 55: Bits 9 - 55 of the host cluster offset. Must be aligned to a cluster boundary. If the offset is 0, the cluster is unallocated; in that case, bit 0 determines how this cluster should be treated during reads. 56 - 63: Reserved and must be zero.
As noted above, bitmap data is stored in separate clusters, described by the bitmap table. Given an offset (in bytes) into the bitmap data, the offset into the image file can be obtained as follows:
image_offset(bitmap_data_offset) = bitmap_table[bitmap_data_offset / cluster_size] + (bitmap_data_offset % cluster_size)
This offset is not defined if bits 9 - 55 of bitmap table entry are zero (see above).
Given an offset byte_nr into the virtual disk and the bitmap’s granularity, the bit offset into the image file to the corresponding bit of the bitmap can be calculated like this:
bit_offset(byte_nr) = image_offset(byte_nr / granularity / 8) * 8 + (byte_nr / granularity) % 8
If the size of the bitmap data is not a multiple of the cluster size then the last cluster of the bitmap data contains some unused tail bits. These bits must be zero.
Bitmaps with ‘type’ field equal to one are dirty tracking bitmaps.
When the virtual disk is in use dirty tracking bitmap may be ‘enabled’ or ‘disabled’. While the bitmap is ‘enabled’, all writes to the virtual disk should be reflected in the bitmap. A set bit in the bitmap means that the corresponding range of the virtual disk (see above) was written to while the bitmap was ‘enabled’. An unset bit means that this range was not written to.
The software doesn’t have to sync the bitmap in the image file with its representation in RAM after each write. Flag ‘in_use’ should be set while the bitmap is not synced.
In the image file the ‘enabled’ state is reflected by the ‘auto’ flag. If this flag is set, the software must consider the bitmap as ‘enabled’ and start tracking virtual disk changes to this bitmap from the first write to the virtual disk. If this flag is not set then the bitmap is disabled.