nbdkit-protocol - which parts of the NBD protocol nbdkit supports
nbdkit [-n|--newstyle] [--mask-handshake MASK] [--no-mc|--no-meta-contexts]
[--no-sr|--no-structured-replies] [-o|--oldstyle] [...]
This page documents the level of support in nbdkit for various parts of the NBD protocol.
The NBD protocol comes in two incompatible forms that we call "oldstyle" and "newstyle". Unfortunately which protocol you should use depends on the client and cannot be known in advance, nor can it be negotiated from the server side.
nbdkit defaults to the newstyle protocol since nbdkit ≥ 1.3 (the command line flag -n or --newstyle is ignored for backwards compatibility with older versions). The newstyle protocol is better in every respect than the oldstyle protocol and you should prefer it if possible. The newstyle protocol also includes extensions where a client may request structured replies for even more capabilities, such as sparse reads or meta contexts for obtaining block status. By default, nbdkit advertises as many features as it can support (in some cases, this can be limited by what callbacks the plugin handles), even if the client does not negotiate to use all advertised features.
Nbdkit also includes some options that are useful mainly when performing integration tests, for proving whether clients have sane fallback behavior when dealing various older servers permitted by the NBD protocol. The following options intentionally disable optional parts of the NBD protocol, with successively larger impacts:
Force the newstyle protocol to treat all requests for meta context negotiation from the client as unsupported; the client will be unable to query block status. By default, nbdkit gracefully handles all meta context requests, even though it currently supports only the base:allocation
context (possibly by synthesizing a context that represents an all-data disk when the plugin lacks support for extents); but treating meta contexts as unsupported is useful for emulating nbd-server 3.25.
Force the newstyle protocol to decline any client request for structured replies; this is stronger than --no-meta-contexts in that it also disables the opportunity for sparse reads. This is useful for emulating nbd-server 3.24.
Alter the mask of which particular global features are advertised during new-style handshake (defaulting to all supported bits set). Clearing bit 0 (the low order bit) limits a client to using just NBD_OPT_EXPORT_NAME
(incompatible with TLS, and includes all effects of --no-sr); clearing bit 1 causes the handshake to send more padding bytes in response to NBD_OPT_EXPORT_NAME
. Other bits in the mask will only have an effect if the NBD protocol is extended in the future to define other global bits.
Force the oldstyle protocol. In this mode, --no-meta-contexts, --no-sr and --mask-handshake have no effect.
Client Protocol
------------------------------------------------------------
qemu <= 2.5 without exportname oldstyle
qemu <= 2.5 with exportname newstyle
qemu >= 2.6 client can talk either protocol
qemu >= 2.11 client tries structured replies
nbd-client < 3.10 client can talk either protocol
nbd-client >= 3.10 newstyle, no structured replies
any TLS (encrypted) client newstyle
nbdkit nbd plugin client can talk either protocol
nbdkit >= 1.13.3 nbd plugin tries structured replies
libnbd either protocol, tries structured replies
If you use qemu ≤ 2.5 without the exportname field against a newstyle server, it will give the error:
Server requires an export name
If you use qemu ≤ 2.5 with the exportname field against an oldstyle server, it will give the error:
Server does not support export names
If you use the oldstyle protocol with nbd-client ≥ 3.10, it will give the error:
Error: It looks like you're trying to connect to an oldstyle server.
Port 10809/tcp is reserved by IANA for the NBD protocol, but you can use nbdkit on any port or on Unix domain sockets.
The NBD protocol specification claims that you should always use newstyle when using port 10809, and use oldstyle on all other ports, but this claim is not based on the reality of what NBD servers do, and nbdkit does not require or encourage this.
Supported in nbdkit ≥ 1.1.12, and the default in nbdkit ≥ 1.3.
Partially supported in nbdkit ≥ 1.1.12. Support for plugins to read the client export name added in nbdkit ≥ 1.15.2.
Versions of nbdkit before 1.16 could advertise a single export name to clients, via a now deprecated side effect of the -e option. In nbdkit 1.15.2, plugins could read the client requested export name using nbdkit_export_name(3) and serve different content. In nbdkit 1.21.22, plugins could implement .list_exports
to answer NBD_OPT_LIST
queries.
NBD_FLAG_NO_ZEROES
Supported in nbdkit ≥ 1.1.13.
This protocol optimization avoids sending a useless block of zero bytes during protocol negotiation.
NBD_CMD_WRITE_ZEROES
Supported in nbdkit ≥ 1.1.13.
Supported in nbdkit ≥ 1.1.15.
Supported in nbdkit ≥ 1.4.0.
NBD_OPT_GO
Supported in nbdkit ≥ 1.5.3.
This protocol enhancement allows the server to return errors when negotiating the export name.
NBD_OPT_INFO
Supported in nbdkit ≥ 1.9.3.
This protocol enhancement allows a client to inspect details about the export without actually connecting.
NBD_FLAG_CAN_MULTI_CONN
Supported in nbdkit ≥ 1.9.9.
Supported in nbdkit ≥ 1.11.8.
However we don’t expose the capability to send structured replies to plugins yet, nor do we send human-readable error messages using this facility.
In nbdkit ≥ 1.13.9, the command-line option --no-sr can be used to disable server support for structured replies, for testing client fallbacks; disabling structured replies also disables features like block status queries that depend on structured replies.
Supported in nbdkit ≥ 1.11.8.
In nbdkit ≥ 1.37.9, the command-line option --no-meta-contexts can be used to disable server support for meta contexts, for testing client fallbacks.
Supported in nbdkit ≥ 1.11.10.
Only base:allocation
(ie. querying which parts of an image are sparse) is supported.
Sparse reads (using NBD_REPLY_TYPE_OFFSET_HOLE
are not directly supported, but a client can use block status to infer which portions of the export do not need to be read.
NBD_FLAG_DF
Supported in nbdkit ≥ 1.11.11.
This protocol extension allows a client to force an all-or-none read when structured replies are in effect. However, the flag is a no-op until we extend the plugin API to allow a fragmented read in the first place.
NBD_CMD_CACHE
Supported in nbdkit ≥ 1.13.4.
This protocol extension allows a client to inform the server about intent to access a portion of the export, to allow the server an opportunity to cache things appropriately.
NBD_CMD_FLAG_FAST_ZERO
Supported in nbdkit ≥ 1.15.0.
This protocol extension allows a server to advertise that it can rank all zero requests as fast or slow, at which point the client can make fast zero requests which fail immediately with ENOTSUP
if the request is no faster than a counterpart write would be, while normal zero requests still benefit from compressed network traffic regardless of the time taken.
NBD_INFO_NAME
, NBD_INFO_DESCRIPTION
Supported in nbdkit ≥ 1.23.6.
These protocol extensions allow a client to learn more information about an export during NBD_OPT_GO
. The .default_export
callback can inform a client of a canonical non-empty name in place of the default export ""
, and the .export_description
callback can give a client more details about the export.
NBD_INFO_BLOCK_SIZE
Supported in nbdkit ≥ 1.30.
Not supported.
Not supported.
nbdkit(1), https://github.com/NetworkBlockDevice/nbd/blob/master/doc/proto.md, https://nbd.sourceforge.io/.
Eric Blake
Richard W.M. Jones
Pino Toscano
Copyright Red Hat
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
Neither the name of Red Hat nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY RED HAT AND CONTRIBUTORS ''AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL RED HAT OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.