If you have been given an nbdkit Python plugin¶

Assuming you have a Python script which is an nbdkit plugin, you run it like this:

 nbdkit python /path/to/plugin.py

You may have to add further "key=value" arguments to the command line. Read the Python script to see if it requires any.

Python versions¶

Since nbdkit ≥ 1.16 only Python 3 is supported. If you wish to continue using nbdkit plugins written in Python 2 then you must use nbdkit ≤ 1.14, but we advise you to update your plugins.

The version of Python 3 is chosen when nbdkit is built. This is compiled in and can't be changed at runtime. "./configure" looks for (in order):

• the "PYTHON" variable (eg "./configure PYTHON=/usr/bin/python3.9")
• python3 on $PATH
• python on $PATH

"./configure" will fail if the first interpreter found is a Python 2 interpreter.

To find out which version of Python "nbdkit-python-plugin" was compiled for, use the --dump-plugin option:

 $ nbdkit python --dump-plugin
 ...
 python_version=3.7.0
 python_pep_384_abi_version=3

API versions¶

The nbdkit API has evolved and new versions are released periodically. To ensure backwards compatibility plugins have to opt in to the new version. From Python you do this by declaring a constant in your module:

 API_VERSION = 2

(where 2 is the latest version at the time this documentation was written). All newly written Python modules must have this constant.

Executable script¶

If you want you can make the script executable and include a "shebang" at the top:

 #!/usr/sbin/nbdkit python

See also "Shebang scripts" in nbdkit(1).

These scripts can also be installed in the $plugindir. See "WRITING PLUGINS IN OTHER PROGRAMMING LANGUAGES" in nbdkit-plugin(3).

Module functions¶

Your script may use "import nbdkit" to have access to the following methods in the "nbdkit" module:

"nbdkit.debug(msg)"

Send a debug message to stderr or syslog if verbose messages are enabled.

"nbdkit.disconnect(force)"

Disconnect from the client. If "force" is "True" then nbdkit will disconnect the client immediately.

"nbdkit.export_name()"

Return the export name negotiated with the client as a Unicode string. Note this should not be trusted because the client can send whatever it wants.

"nbdkit.is_tls()"

Returns "True" if the client completed TLS authentication, or "False" if the connection is plaintext.

"nbdkit.nanosleep(secs, nsecs)"

Sleep for seconds and nanoseconds.

"nbdkit.parse_delay(what, str)"

Parse a delay or sleep (such as "10ms") into a pair (sec, nsec). Wraps the nbdkit_parse_delay(3) function.

"nbdkit.parse_size(str)"

Parse a string (such as "100M") into a size in bytes. Wraps the nbdkit_parse_size(3) C function.

"nbdkit.parse_probability(what, str)"

Parse a string (such as "100%") into a probability, returning a floating point number. Wraps the nbdkit_parse_probability(3) function.

"nbdkit.peer_pid()", "nbdkit.peer_uid()", "nbdkit.peer_gid()", "nbdkit.peer_security_context()"

Return the client process ID, user ID, group ID or security context. The PID, UID and GID are only available when the client connects by Unix domain socket, and then only on some operating systems. The security context is usually the SELinux label, IPSEC label or NetLabel.

"nbdkit.peer_tls_dn()"

Return the client TLS Distinguished Name. See nbdkit_peer_tls_dn(3).

"nbdkit.peer_tls_issuer_dn()"

Return the client certificate issuer's TLS Distinguished Name. See nbdkit_peer_tls_issuer_dn(3).

"nbdkit.read_password(value)"

Read a password from a config parameter. This returns the password as a Python "bytes" object. See nbdkit_read_password(3) for more information on the different ways that the "value" parameter can be parsed.

"nbdkit.set_error(err)"

Throwing a Python exception from a callback causes an error message to be sent back to the NBD client. The NBD protocol allows an error code (ie. errno) to be sent to the client, but by default the Python plugin always sends "EIO". To control what error code is sent call "nbdkit.set_error":

 def pread(h, buf, offset):
   if access_denied:
     nbdkit.set_error(errno.EPERM)
     raise RuntimeError()

"nbdkit.shutdown()"

Request asynchronous server shutdown.

"nbdkit.stdio_safe()"

Returns "True" if it is safe to interact with stdin and stdout during the configuration phase.

Module constants¶

After "import nbdkit" the following constants are available. These are used in the callbacks below.

"nbdkit.THREAD_MODEL_SERIALIZE_CONNECTIONS"

"nbdkit.THREAD_MODEL_SERIALIZE_ALL_REQUESTS"

"nbdkit.THREAD_MODEL_SERIALIZE_REQUESTS"

"nbdkit.THREAD_MODEL_PARALLEL"

Possible return values from thread_model().

"nbdkit.FLAG_MAY_TRIM"

"nbdkit.FLAG_FUA"

"nbdkit.FLAG_REQ_ONE"

"nbdkit.FLAG_FAST_ZERO"

Flags bitmap passed to certain plugin callbacks. Not all callbacks with a flags parameter use all of these flags, consult the documentation below and nbdkit-plugin(3).

"nbdkit.FUA_NONE"

"nbdkit.FUA_EMULATE"

"nbdkit.FUA_NATIVE"

Possible return values from can_fua().

"nbdkit.CACHE_NONE"

"nbdkit.CACHE_EMULATE"

"nbdkit.CACHE_NATIVE"

Possible return values from can_cache().

"nbdkit.EXTENT_HOLE"

"nbdkit.EXTENT_ZERO"

Used in the "type" field returned by extents().

Threads¶

The thread model for Python callbacks defaults to "nbdkit.THREAD_MODEL_SERIALIZE_ALL_REQUESTS".

Since nbdkit 1.22 it has been possible to set this by implementing a thread_model() function which returns one of the constants "nbdkit.THREAD_MODEL_*".

The Python Global Interpreter Lock (GIL) is still used, so Python code does not run in parallel. However if a plugin callback calls a library which blocks (eg. to make an HTTP request), then another callback might be executed in parallel. Plugins which use "nbdkit.THREAD_MODEL_SERIALIZE_REQUESTS" or "nbdkit.THREAD_MODEL_PARALLEL" may need to use locks on shared data.

Exceptions¶

Python callbacks should throw exceptions to indicate errors. Remember to use "nbdkit.set_error" if you need to control which error is sent back to the client; if omitted, the client will see an error of "EIO".

Python callbacks¶

This just documents the arguments to the callbacks in Python, and any way that they differ from the C callbacks. In all other respects they work the same way as the C callbacks, so you should go and read nbdkit-plugin(3).

"dump_plugin"

(Optional)

There are no arguments or return value.

"config"

(Optional)

 def config(key, value):
   # no return value
    

"config_complete"

(Optional)

There are no arguments or return value.

"thread_model"

(Optional, nbdkit ≥ 1.22)

 def thread_model():
   return nbdkit.THEAD_MODEL_SERIALIZE_ALL_REQUESTS
    

See "Threads" above.

"get_ready"

(Optional)

There are no arguments or return value.

"after_fork"

(Optional, nbdkit ≥ 1.26)

There are no arguments or return value.

"cleanup"

(Optional, nbdkit ≥ 1.28)

There are no arguments or return value.

"list_exports"

(Optional)

 def list_exports(readonly, is_tls):
   # return an iterable object (eg. list) of
   # (name, description) tuples or bare names:
   return [ (name1, desc1), name2, (name3, desc3), ... ]
    

"default_export"

(Optional)

 def default_export(readonly, is_tls):
   # return a string
   return "name"
    

"preconnect"

(Optional, nbdkit ≥ 1.26)

 def preconnect(readonly):
   # no return value
    

"open"

(Required)

 def open(readonly):
   # return handle
    

You can return any Python value (even "None") as the handle. It is passed back as the first arg 'h' in subsequent calls. To return an error from this method you must throw an exception.

"close"

(Optional)

 def close(h):
   # no return value
    

After "close" returns, the reference count of the handle is decremented in the C part, which usually means that the handle and its contents will be garbage collected.

"export_description"

(Optional)

 def export_description(h):
   # return a string
   return "description"
    

"get_size"

(Required)

 def get_size(h):
   # return the size of the disk
    

"block_size"

(Option)

 def block_size(h):
   # return triple (minimum, preferred, maximum) block size
    

"is_rotational"

(Optional)

 def is_rotational(h):
   # return a boolean
    

"can_multi_conn"

(Optional)

 def can_multi_conn(h):
   # return a boolean
    

"can_write"

(Optional)

 def can_write(h):
   # return a boolean
    

"can_flush"

(Optional)

 def can_flush(h):
   # return a boolean
    

"can_trim"

(Optional)

 def can_trim(h):
   # return a boolean
    

"can_zero"

(Optional)

 def can_zero(h):
   # return a boolean
    

"can_fast_zero"

(Optional)

 def can_fast_zero(h):
   # return a boolean
    

"can_fua"

(Optional)

 def can_fua(h):
   # return nbdkit.FUA_NONE or nbdkit.FUA_EMULATE
   # or nbdkit.FUA_NATIVE
    

"can_cache"

(Optional)

 def can_cache(h):
   # return nbdkit.CACHE_NONE or nbdkit.CACHE_EMULATE
   # or nbdkit.CACHE_NATIVE
    

"can_extents"

(Optional)

 def can_extents(h):
   # return a boolean
    

"pread"

(Required)

 def pread(h, buf, offset, flags):
   # read into the buffer
    

The body of your "pread" function should read exactly len(buf) bytes of data starting at disk "offset" and write it into the buffer "buf". "flags" is always 0.

NBD only supports whole reads, so your function should try to read the whole region (perhaps requiring a loop). If the read fails or is partial, your function should throw an exception, optionally using "nbdkit.set_error" first.

"pwrite"

(Optional)

 def pwrite(h, buf, offset, flags):
   length = len(buf)
   # no return value
    

The body of your "pwrite" function should write the buffer "buf" to the disk. You should write "count" bytes to the disk starting at "offset". "flags" may contain "nbdkit.FLAG_FUA".

NBD only supports whole writes, so your function should try to write the whole region (perhaps requiring a loop). If the write fails or is partial, your function should throw an exception,
optionally using "nbdkit.set_error" first.

"flush"

(Optional)

 def flush(h, flags):
   # no return value
    

The body of your "flush" function should do a sync(2) or fdatasync(2) or equivalent on the backing store. "flags" is always 0.

If the flush fails, your function should throw an exception, optionally using "nbdkit.set_error" first.

"trim"

(Optional)

 def trim(h, count, offset, flags):
   # no return value
    

The body of your "trim" function should "punch a hole" in the backing store. "flags" may contain "nbdkit.FLAG_FUA". If the trim fails, your function should throw an exception, optionally using "nbdkit.set_error" first.

"zero"

(Optional)

 def zero(h, count, offset, flags):
   # no return value
    

The body of your "zero" function should ensure that "count" bytes of the disk, starting at "offset", will read back as zero. "flags" is a bitmask which may include "nbdkit.FLAG_MAY_TRIM", "nbdkit.FLAG_FUA", "nbdkit.FLAG_FAST_ZERO".

NBD only supports whole writes, so your function should try to write the whole region (perhaps requiring a loop).

If the write fails or is partial, your function should throw an exception, optionally using "nbdkit.set_error" first. In particular, if you would like to automatically fall back to "pwrite" (perhaps because there is nothing to optimize if "flags & nbdkit.FLAG_MAY_TRIM" is false), use "nbdkit.set_error(errno.EOPNOTSUPP)".

"cache"

(Optional)

 def cache(h, count, offset, flags):
   # no return value
    

The body of your "cache" function should prefetch data in the indicated range.

If the cache operation fails, your function should throw an exception, optionally using "nbdkit.set_error" first.

"extents"

(Optional)

 def extents(h, count, offset, flags):
   # return an iterable object (eg. list) of
   # (offset, length, type) tuples:
   return [ (off1, len1, type1), (off2, len2, type2), ... ]
    

Missing callbacks¶

Missing: "load"

This is not needed since you can use regular Python mechanisms like top level statements to run code when the module is loaded.

Missing: "unload"

This is missing, but in nbdkit ≥ 1.28 you can put code in the cleanup() function to have it run when nbdkit exits. In earlier versions of nbdkit, using a Python atexit handler is recommended.

Missing: "name", "version", "longname", "description", "config_help", "magic_config_key".

These are not yet supported.