plugins: Wire up nbd plugin support for NBD_INFO_INIT_STATE

[nbdkit/ericb.git] / docs / nbdkit-filter.pod

=head1 NAME

nbdkit-filter - how to write nbdkit filters

=head1 SYNOPSIS

 #include <nbdkit-filter.h>
 
 static int
 myfilter_config (nbdkit_next_config *next, void *nxdata,
                  const char *key, const char *value)
 {
   if (strcmp (key, "myparameter") == 0) {
     // ...
     return 0;
   }
   else {
     // pass through to next filter or plugin
     return next (nxdata, key, value);
   }
 }
 
 static struct nbdkit_filter filter = {
   .name              = "filter",
   .config            = myfilter_config,
   /* etc */
 };
 
 NBDKIT_REGISTER_FILTER(filter)

When this has been compiled to a shared library, do:

 nbdkit [--args ...] --filter=./myfilter.so plugin [key=value ...]

When debugging, use the I<-fv> options:

 nbdkit -fv --filter=./myfilter.so plugin [key=value ...]

=head1 DESCRIPTION

One or more nbdkit filters can be placed in front of an nbdkit plugin
to modify the behaviour of the plugin.  This manual page describes how
to create an nbdkit filter.

Filters can be used for example to limit requests to an offset/limit,
add copy-on-write support, or inject delays or errors (for testing).

Different filters can be stacked:

     NBD     ┌─────────┐    ┌─────────┐          ┌────────┐
  client ───▶│ filter1 │───▶│ filter2 │── ─ ─ ──▶│ plugin │
 request     └─────────┘    └─────────┘          └────────┘

Each filter intercepts plugin functions (see L<nbdkit-plugin(3)>) and
can call the next filter or plugin in the chain, modifying parameters,
calling before the filter function, in the middle or after.  Filters
may even short-cut the chain.  As an example, to process its own
parameters the filter can intercept the C<.config> method:

 static int
 myfilter_config (nbdkit_next_config *next, void *nxdata,
                  const char *key, const char *value)
 {
   if (strcmp (key, "myparameter") == 0) {
     // ...
     // here you would handle this key, value
     // ...
     return 0;
   }
   else {
     // pass through to next filter or plugin
     return next (nxdata, key, value);
   }
 }
 
 static struct nbdkit_filter filter = {
   // ...
   .config            = myfilter_config,
   // ...
 };

The call to C<next (nxdata, ...)> calls the C<.config> method of the
next filter or plugin in the chain.  In the example above any
instances of C<myparameter=...> on the command line would not be seen
by the plugin.

To see example filters:
L<https://github.com/libguestfs/nbdkit/tree/master/filters>

Filters must be written in C.

Unlike plugins, where we provide a stable ABI guarantee that permits
operation across version differences, filters can only be run with the
same version of nbdkit that they were compiled with.  The reason for
this is two-fold: the filter API includes access to struct
nbdkit_next_ops that is likely to change if new callbacks are added
(old nbdkit cannot safely run new filters that access new methods);
and if we added new methods then an old filter would not see them and
so they would be passed unmodified through the filter, and in some
cases that leads to data corruption (new nbdkit cannot safely run old
filters unaware of new methods).  Therefore, unlike plugins, you
should not expect to distribute filters separately from nbdkit.

=head1 C<#include E<lt>nbdkit-filter.hE<gt>>

All filters should start by including this header file.

=head1 C<struct nbdkit_filter>

All filters must define and register one C<struct nbdkit_filter>,
which contains the name of the filter and pointers to plugin methods
that the filter wants to intercept.

 static struct nbdkit_filter filter = {
   .name              = "filter",
   .longname          = "My Filter",
   .description       = "This is my great filter for nbdkit",
   .config            = myfilter_config,
   /* etc */
 };
 
 NBDKIT_REGISTER_FILTER(filter)

The C<.name> field is the name of the filter.  This is the only field
which is required.

=head1 NEXT PLUGIN

F<nbdkit-filter.h> defines some function types (C<nbdkit_next_config>,
C<nbdkit_next_config_complete>, C<nbdkit_next_preconnect>,
C<nbdkit_next_open>) and a structure called C<struct nbdkit_next_ops>.
These abstract the next plugin or filter in the chain.  There is also
an opaque pointer C<nxdata> which must be passed along when calling
these functions.

=head2 Next config, open and close

The filter’s C<.config>, C<.config_complete> and C<.open> methods may
only call the next C<.config>, C<.config_complete> and C<.open> method
in the chain (optionally for C<.config>).

The filter’s C<.close> method is called when an old connection closed,
and this has no C<next> parameter because it cannot be
short-circuited.

=head2 C<next_ops>

The filter’s other methods like C<.prepare>, C<.get_size>, C<.pread>
etc ― always called in the context of a connection ― are passed a
pointer to C<struct nbdkit_next_ops> which contains a comparable set
of accessors to plugin methods that can be called during a connection.

It is possible for a filter to issue (for example) extra
C<next_ops-E<gt>pread> calls in response to a single C<.pwrite> call.

Note that the semantics of the functions in C<struct nbdkit_next_ops>
are slightly different from what a plugin implements: for example,
when a plugin's C<.pread> returns -1 on error, the error value to
advertise to the client is implicit (via the plugin calling
C<nbdkit_set_error> or setting C<errno>), whereas
C<next_ops-E<gt>pread> exposes this via an explicit parameter,
allowing a filter to learn or modify this error if desired.

There is also a C<next_ops-E<gt>reopen> function which is used by
L<nbdkit-retry-filter(3)> to close and reopen the underlying plugin.
It should be used with caution because it is difficult to use safely.

=head2 Other considerations

You can modify parameters when you call the C<next> function.  However
be careful when modifying strings because for some methods
(eg. C<.config>) the plugin may save the string pointer that you pass
along.  So you may have to ensure that the string is not freed for the
lifetime of the server.

Note that if your filter registers a callback but in that callback it
doesn't call the C<next> function then the corresponding method in the
plugin will never be called.  In particular, your C<.open> method, if
you have one, B<must> call the C<.next> method.

=head1 CALLBACKS

C<struct nbdkit_filter> has some static fields describing the filter
and optional callback functions which can be used to intercept plugin
methods.

=head2 C<.name>

 const char *name;

This field (a string) is required, and B<must> contain only ASCII
alphanumeric characters and be unique amongst all filters.

=head2 C<.longname>

 const char *longname;

An optional free text name of the filter.  This field is used in error
messages.

=head2 C<.description>

 const char *description;

An optional multi-line description of the filter.

=head2 C<.load>

 void load (void);

This is called once just after the filter is loaded into memory.  You
can use this to perform any global initialization needed by the
filter.

=head2 C<.unload>

 void unload (void);

This may be called once just before the filter is unloaded from
memory.  Note that it's not guaranteed that C<.unload> will always be
called (eg. the server might be killed or segfault), so you should try
to make the filter as robust as possible by not requiring cleanup.
See also L<nbdkit-plugin(3)/SHUTDOWN>.

=head2 C<.config>

 int (*config) (nbdkit_next_config *next, void *nxdata,
                const char *key, const char *value);

This intercepts the plugin C<.config> method and can be used by the
filter to parse its own command line parameters.  You should try to
make sure that command line parameter keys that the filter uses do not
conflict with ones that could be used by a plugin.

If there is an error, C<.config> should call C<nbdkit_error> with an
error message and return C<-1>.

=head2 C<.config_complete>

 int (*config_complete) (nbdkit_next_config_complete *next, void *nxdata);

This intercepts the plugin C<.config_complete> method and can be used
to ensure that all parameters needed by the filter were supplied on
the command line.

If there is an error, C<.config_complete> should call C<nbdkit_error>
with an error message and return C<-1>.

=head2 C<.config_help>

 const char *config_help;

This optional multi-line help message should summarize any
C<key=value> parameters that it takes.  It does I<not> need to repeat
what already appears in C<.description>.

If the filter doesn't take any config parameters you should probably
omit this.

=head2 C<.thread_model>

 int (*thread_model) (void);

Filters may tighten (but not relax) the thread model of the plugin, by
defining this callback.  Note that while plugins use a compile-time
definition of C<THREAD_MODEL>, filters do not need to declare a model
at compile time; instead, this callback is called after
C<.config_complete> and before any connections are created.  See
L<nbdkit-plugin(3)/THREADS> for a discussion of thread models.

The final thread model used by nbdkit is the smallest (ie. most
serialized) out of all the filters and the plugin, and applies for all
connections.  Requests for a model larger than permitted by the plugin
are silently ignored. It is acceptable for decisions made during
C<.config> and C<.config_complete> to determine which model to
request.

This callback is optional; if it is not present, the filter must be
written to handle fully parallel requests, including when multiple
requests are issued in parallel on the same connection, similar to a
plugin requesting C<NBDKIT_THREAD_MODEL_PARALLEL>.  This ensures the
filter doesn't slow down other filters or plugins.

If there is an error, C<.thread_model> should call C<nbdkit_error>
with an error message and return C<-1>.

=head2 C<.preconnect>

 int (*preconnect) (nbdkit_next_preconnect *next, void *nxdata,
                    int readonly);

This intercepts the plugin C<.preconnect> method and can be used to
filter access to the server.

If there is an error, C<.preconnect> should call C<nbdkit_error> with
an error message and return C<-1>.

=head2 C<.open>

 void * (*open) (nbdkit_next_open *next, void *nxdata,
                 int readonly);

This is called when a new client connection is opened and can be used
to allocate any per-connection data structures needed by the filter.
The handle (which is not the same as the plugin handle) is passed back
to other filter callbacks and could be freed in the C<.close>
callback.

Note that the handle is completely opaque to nbdkit, but it must not
be NULL.  If you don't need to use a handle, return
C<NBDKIT_HANDLE_NOT_NEEDED> which is a static non-NULL pointer.

If there is an error, C<.open> should call C<nbdkit_error> with an
error message and return C<NULL>.

This callback is optional, but if provided, it must call C<next>,
passing a value for C<readonly> according to how the filter plans to
use the plugin.  Typically, the filter passes the same value as it
received, or passes true to provide a writable layer on top of a
read-only backend.  However, it is also acceptable to attempt write
access to the plugin even if this filter is readonly, such as when a
file system mounted read-only still requires write access to the
underlying device in case a journal needs to be replayed for
consistency as part of the mounting process.  The filter should
generally call C<next> as its first step, to allocate from the plugin
outwards, so that C<.close> running from the outer filter to the
plugin will be in reverse.

=head2 C<.close>

 void (*close) (void *handle);

This is called when the client closes the connection.  It should clean
up any per-connection resources used by the filter.  It is called
beginning with the outermost filter and ending with the plugin (the
opposite order of C<.open> if all filters call C<next> first),
although this order technically does not matter since the callback
cannot report failures or access the underlying plugin.

=head2 C<.prepare>

=head2 C<.finalize>

  int (*prepare) (struct nbdkit_next_ops *next_ops, void *nxdata,
                  void *handle, int readonly);
  int (*finalize) (struct nbdkit_next_ops *next_ops, void *nxdata,
                   void *handle);

These two methods can be used to perform any necessary operations just
after opening the connection (C<.prepare>) or just before closing the
connection (C<.finalize>).

For example if you need to scan the underlying disk to check for a
partition table, you could do it in your C<.prepare> method (calling
the plugin's C<.pread> method via C<next_ops>).  Or if you need to
cleanly update superblock data in the image on close you can do it in
your C<.finalize> method (calling the plugin's C<.pwrite> method).
Doing these things in the filter's C<.open> or C<.close> method is not
possible.

For C<.prepare>, the value of C<readonly> is the same as was passed to
C<.open>, declaring how this filter will be used.

There is no C<next_ops-E<gt>prepare> or C<next_ops-E<gt>finalize>.
Unlike other filter methods, prepare and finalize are not chained
through the C<next_ops> structure.  Instead the core nbdkit server
calls the prepare and finalize methods of all filters.  Prepare
methods are called starting with the filter closest to the plugin and
proceeding outwards (matching the order of C<.open> if all filters
call C<next> before doing anything locally).  Finalize methods are
called in the reverse order of prepare methods, with the outermost
filter first (and matching the order of C<.close>), and only if the
prepare method succeeded.

If there is an error, both callbacks should call C<nbdkit_error> with
an error message and return C<-1>.  An error in C<.prepare> is
reported to the client, but leaves the connection open (a client may
try again with a different export name, for example); while an error
in C<.finalize> forces the client to disconnect.

=head2 C<.get_size>

 int64_t (*get_size) (struct nbdkit_next_ops *next_ops, void *nxdata,
                      void *handle);

This intercepts the plugin C<.get_size> method and can be used to read
or modify the apparent size of the block device that the NBD client
will see.

The returned size must be E<ge> 0.  If there is an error, C<.get_size>
should call C<nbdkit_error> with an error message and return C<-1>.
This function is only called once per connection and cached by nbdkit.
Similarly, repeated calls to C<next_ops-E<gt>get_size> will return a
cached value.

=head2 C<.can_write>

=head2 C<.can_flush>

=head2 C<.is_rotational>

=head2 C<.can_trim>

=head2 C<.can_zero>

=head2 C<.can_fast_zero>

=head2 C<.can_extents>

=head2 C<.can_fua>

=head2 C<.can_multi_conn>

=head2 C<.can_cache>

=head2 C<.init_sparse>

=head2 C<.init_zero>

 int (*can_write) (struct nbdkit_next_ops *next_ops, void *nxdata,
                   void *handle);
 int (*can_flush) (struct nbdkit_next_ops *next_ops, void *nxdata,
                   void *handle);
 int (*is_rotational) (struct nbdkit_next_ops *next_ops,
                       void *nxdata,
                       void *handle);
 int (*can_trim) (struct nbdkit_next_ops *next_ops, void *nxdata,
                  void *handle);
 int (*can_zero) (struct nbdkit_next_ops *next_ops, void *nxdata,
                  void *handle);
 int (*can_fast_zero) (struct nbdkit_next_ops *next_ops, void *nxdata,
                       void *handle);
 int (*can_extents) (struct nbdkit_next_ops *next_ops, void *nxdata,
                     void *handle);
 int (*can_fua) (struct nbdkit_next_ops *next_ops, void *nxdata,
                 void *handle);
 int (*can_multi_conn) (struct nbdkit_next_ops *next_ops, void *nxdata,
                        void *handle);
 int (*can_cache) (struct nbdkit_next_ops *next_ops, void *nxdata,
                   void *handle);
 int (*init_sparse) (struct nbdkit_next_ops *next_ops, void *nxdata,
                     void *handle);
 int (*init_zero) (struct nbdkit_next_ops *next_ops, void *nxdata,
                   void *handle);

These intercept the corresponding plugin methods, and control feature
bits advertised to the client.

Of note, the semantics of C<.can_zero> callback in the filter are
slightly different from the plugin, and must be one of three success
values visible only to filters:

=over 4

=item C<NBDKIT_ZERO_NONE>

Completely suppress advertisement of write zero support (this can only
be done from filters, not plugins).

=item C<NBDKIT_ZERO_EMULATE>

Inform nbdkit that write zeroes should immediately fall back to
C<.pwrite> emulation without trying C<.zero> (this value is returned
by C<next_ops-E<gt>can_zero> if the plugin returned false in its
C<.can_zero>).

=item C<NBDKIT_ZERO_NATIVE>

Inform nbdkit that write zeroes should attempt to use C<.zero>,
although it may still fall back to C<.pwrite> emulation for C<ENOTSUP>
or C<EOPNOTSUPP> failures (this value is returned by
C<next_ops-E<gt>can_zero> if the plugin returned true in its
C<.can_zero>).

=back

Remember that most of the feature check functions return merely a
boolean success value, while C<.can_zero>, C<.can_fua> and
C<.can_cache> have three success values.

The difference between C<.can_fua> values may affect choices made in
the filter: when splitting a write request that requested FUA from the
client, if C<next_ops-E<gt>can_fua> returns C<NBDKIT_FUA_NATIVE>, then
the filter should pass the FUA flag on to each sub-request; while if
it is known that FUA is emulated by a flush because of a return of
C<NBDKIT_FUA_EMULATE>, it is more efficient to only flush once after
all sub-requests have completed (often by passing C<NBDKIT_FLAG_FUA>
on to only the final sub-request, or by dropping the flag and ending
with a direct call to C<next_ops-E<gt>flush>).

If there is an error, the callback should call C<nbdkit_error> with an
error message and return C<-1>.  These functions are called at most
once per connection and cached by nbdkit. Similarly, repeated calls to
any of the C<next_ops> counterparts will return a cached value; by
calling into the plugin during C<.prepare>, you can ensure that later
use of the cached values during data commands like <.pwrite> will not
fail.

=head2 C<.pread>

 int (*pread) (struct nbdkit_next_ops *next_ops, void *nxdata,
               void *handle, void *buf, uint32_t count, uint64_t offset,
               uint32_t flags, int *err);

This intercepts the plugin C<.pread> method and can be used to read or
modify data read by the plugin.

The parameter C<flags> exists in case of future NBD protocol
extensions; at this time, it will be 0 on input, and the filter should
not pass any flags to C<next_ops-E<gt>pread>.

If there is an error (including a short read which couldn't be
recovered from), C<.pread> should call C<nbdkit_error> with an error
message B<and> return -1 with C<err> set to the positive errno value
to return to the client.

=head2 C<.pwrite>

 int (*pwrite) (struct nbdkit_next_ops *next_ops, void *nxdata,
                void *handle,
                const void *buf, uint32_t count, uint64_t offset,
                uint32_t flags, int *err);

This intercepts the plugin C<.pwrite> method and can be used to modify
data written by the plugin.

This function will not be called if C<.can_write> returned false; in
turn, the filter should not call C<next_ops-E<gt>pwrite> if
C<next_ops-E<gt>can_write> did not return true.

The parameter C<flags> may include C<NBDKIT_FLAG_FUA> on input based
on the result of C<.can_fua>.  In turn, the filter should only pass
C<NBDKIT_FLAG_FUA> on to C<next_ops-E<gt>pwrite> if
C<next_ops-E<gt>can_fua> returned a positive value.

If there is an error (including a short write which couldn't be
recovered from), C<.pwrite> should call C<nbdkit_error> with an error
message B<and> return -1 with C<err> set to the positive errno value
to return to the client.

=head2 C<.flush>

 int (*flush) (struct nbdkit_next_ops *next_ops, void *nxdata,
               void *handle, uint32_t flags, int *err);

This intercepts the plugin C<.flush> method and can be used to modify
flush requests.

This function will not be called if C<.can_flush> returned false; in
turn, the filter should not call C<next_ops-E<gt>flush> if
C<next_ops-E<gt>can_flush> did not return true.

The parameter C<flags> exists in case of future NBD protocol
extensions; at this time, it will be 0 on input, and the filter should
not pass any flags to C<next_ops-E<gt>flush>.

If there is an error, C<.flush> should call C<nbdkit_error> with an
error message B<and> return -1 with C<err> set to the positive errno
value to return to the client.

=head2 C<.trim>

 int (*trim) (struct nbdkit_next_ops *next_ops, void *nxdata,
              void *handle, uint32_t count, uint64_t offset,
              uint32_t flags, int *err);

This intercepts the plugin C<.trim> method and can be used to modify
trim requests.

This function will not be called if C<.can_trim> returned false; in
turn, the filter should not call C<next_ops-E<gt>trim> if
C<next_ops-E<gt>can_trim> did not return true.

The parameter C<flags> may include C<NBDKIT_FLAG_FUA> on input based
on the result of C<.can_fua>.  In turn, the filter should only pass
C<NBDKIT_FLAG_FUA> on to C<next_ops-E<gt>trim> if
C<next_ops-E<gt>can_fua> returned a positive value.

If there is an error, C<.trim> should call C<nbdkit_error> with an
error message B<and> return -1 with C<err> set to the positive errno
value to return to the client.

=head2 C<.zero>

 int (*zero) (struct nbdkit_next_ops *next_ops, void *nxdata,
              void *handle, uint32_t count, uint64_t offset, uint32_t flags,
              int *err);

This intercepts the plugin C<.zero> method and can be used to modify
zero requests.

This function will not be called if C<.can_zero> returned
C<NBDKIT_ZERO_NONE>; in turn, the filter should not call
C<next_ops-E<gt>zero> if C<next_ops-E<gt>can_zero> returned
C<NBDKIT_ZERO_NONE>.

On input, the parameter C<flags> may include C<NBDKIT_FLAG_MAY_TRIM>
unconditionally, C<NBDKIT_FLAG_FUA> based on the result of
C<.can_fua>, and C<NBDKIT_FLAG_FAST_ZERO> based on the result of
C<.can_fast_zero>.  In turn, the filter may pass
C<NBDKIT_FLAG_MAY_TRIM> unconditionally, but should only pass
C<NBDKIT_FLAG_FUA> or C<NBDKIT_FLAG_FAST_ZERO> on to
C<next_ops-E<gt>zero> if the corresponding C<next_ops-E<gt>can_fua> or
C<next_ops-E<gt>can_fast_zero> returned a positive value.

Note that unlike the plugin C<.zero> which is permitted to fail with
C<ENOTSUP> or C<EOPNOTSUPP> to force a fallback to C<.pwrite>, the
function C<next_ops-E<gt>zero> will not fail with C<err> set to
C<ENOTSUP> or C<EOPNOTSUPP> unless C<NBDKIT_FLAG_FAST_ZERO> was used,
because otherwise the fallback has already taken place.

If there is an error, C<.zero> should call C<nbdkit_error> with an
error message B<and> return -1 with C<err> set to the positive errno
value to return to the client.  The filter should not fail with
C<ENOTSUP> or C<EOPNOTSUPP> unless C<flags> includes
C<NBDKIT_FLAG_FAST_ZERO> (while plugins have automatic fallback to
C<.pwrite>, filters do not).

=head2 C<.extents>

 int (*extents) (struct nbdkit_next_ops *next_ops, void *nxdata,
                 void *handle, uint32_t count, uint64_t offset, uint32_t flags,
                 struct nbdkit_extents *extents,
                 int *err);

This intercepts the plugin C<.extents> method and can be used to
modify extent requests.

This function will not be called if C<.can_extents> returned false; in
turn, the filter should not call C<next_ops-E<gt>extents> if
C<next_ops-E<gt>can_extents> did not return true.

It is possible for filters to transform the extents list received back
from the layer below.  Without error checking it would look like this:

 myfilter_extents (..., uint32_t count, uint64_t offset, ...)
 {
   size_t i;
   struct nbdkit_extents *extents2;
   struct nbdkit_extent e;
   int64_t size;

   size = next_ops->get_size (nxdata);
   extents2 = nbdkit_extents_new (offset + shift, size);
   next_ops->extents (nxdata, count, offset + shift, flags, extents2, err);
   for (i = 0; i < nbdkit_extents_count (extents2); ++i) {
     e = nbdkit_get_extent (extents2, i);
     e.offset -= shift;
     nbdkit_add_extent (extents, e.offset, e.length, e.type);
   }
   nbdkit_extents_free (extents2);
 }

If there is an error, C<.extents> should call C<nbdkit_error> with an
error message B<and> return -1 with C<err> set to the positive errno
value to return to the client.

=head3 Allocating and freeing nbdkit_extents list

Two functions are provided to filters only for allocating and freeing
the map:

 struct nbdkit_extents *nbdkit_extents_new (uint64_t start, uint64_t end);

Allocates and returns a new, empty extents list.  The C<start>
parameter is the start of the range described in the list, and the
C<end> parameter is the offset of the byte beyond the end.  Normally
you would pass in C<offset> as the start and the size of the plugin as
the end, but for filters which adjust offsets, they should pass in the
adjusted offset.

On error this function can return C<NULL>.  In this case it calls
C<nbdkit_error> and/or C<nbdkit_set_error> as required.  C<errno> will
be set to a suitable value.

 void nbdkit_extents_free (struct nbdkit_extents *);

Frees an existing extents list.

=head3 Iterating over nbdkit_extents list

Two functions are provided to filters only to iterate over the extents
in order:

 size_t nbdkit_extents_count (const struct nbdkit_extents *);

Returns the number of extents in the list.

 struct nbdkit_extent {
   uint64_t offset;
   uint64_t length;
   uint32_t type;
 };
 struct nbdkit_extent nbdkit_get_extent (const struct nbdkit_extents *,
                                         size_t i);

Returns a copy of the C<i>'th extent.

=head2 C<.cache>

 int (*cache) (struct nbdkit_next_ops *next_ops, void *nxdata,
               void *handle, uint32_t count, uint64_t offset,
               uint32_t flags, int *err);

This intercepts the plugin C<.cache> method and can be used to modify
cache requests.

This function will not be called if C<.can_cache> returned
C<NBDKIT_CACHE_NONE> or C<NBDKIT_CACHE_EMULATE>; in turn, the filter
should not call C<next_ops-E<gt>cache> unless
C<next_ops-E<gt>can_cache> returned C<NBDKIT_CACHE_NATIVE>.

The parameter C<flags> exists in case of future NBD protocol
extensions; at this time, it will be 0 on input, and the filter should
not pass any flags to C<next_ops-E<gt>cache>.

If there is an error, C<.cache> should call C<nbdkit_error> with an
error message B<and> return -1 with C<err> set to the positive errno
value to return to the client.

=head1 ERROR HANDLING

If there is an error in the filter itself, the filter should call
C<nbdkit_error> to report an error message.  If the callback is
involved in serving data, the explicit C<err> parameter determines the
error code that will be sent to the client; other callbacks should
return the appropriate error indication, eg. C<NULL> or C<-1>.

C<nbdkit_error> has the following prototype and works like
L<printf(3)>:

 void nbdkit_error (const char *fs, ...);
 void nbdkit_verror (const char *fs, va_list args);

For convenience, C<nbdkit_error> preserves the value of C<errno>, and
also supports the glibc extension of a single C<%m> in a format string
expanding to C<strerror(errno)>, even on platforms that don't support
that natively.

=head1 DEBUGGING

Run the server with I<-f> and I<-v> options so it doesn't fork and you
can see debugging information:

 nbdkit -fv --filter=./myfilter.so plugin [key=value [key=value [...]]]

To print debugging information from within the filter, call
C<nbdkit_debug>, which has the following prototype and works like
L<printf(3)>:

 void nbdkit_debug (const char *fs, ...);
 void nbdkit_vdebug (const char *fs, va_list args);

For convenience, C<nbdkit_debug> preserves the value of C<errno>, and
also supports the glibc extension of a single C<%m> in a format string
expanding to C<strerror(errno)>, even on platforms that don't support
that natively.  Note that C<nbdkit_debug> only prints things when the
server is in verbose mode (I<-v> option).

=head2 Debug Flags

Debug Flags in filters work exactly the same way as plugins.  See
L<nbdkit-plugin(3)/Debug Flags>.

=head1 INSTALLING THE FILTER

The filter is a C<*.so> file and possibly a manual page.  You can of
course install the filter C<*.so> file wherever you want, and users
will be able to use it by running:

 nbdkit --filter=/path/to/filter.so plugin [args]

However B<if> the shared library has a name of the form
C<nbdkit-I<name>-filter.so> B<and if> the library is installed in the
C<$filterdir> directory, then users can be run it by only typing:

 nbdkit --filter=name plugin [args]

The location of the C<$filterdir> directory is set when nbdkit is
compiled and can be found by doing:

 nbdkit --dump-config

If using the pkg-config/pkgconf system then you can also find the
filter directory at compile time by doing:

 pkgconf nbdkit --variable=filterdir

=head1 PKG-CONFIG/PKGCONF

nbdkit provides a pkg-config/pkgconf file called C<nbdkit.pc> which
should be installed on the correct path when the nbdkit development
environment is installed.  You can use this in autoconf
F<configure.ac> scripts to test for the development environment:

 PKG_CHECK_MODULES([NBDKIT], [nbdkit >= 1.2.3])

The above will fail unless nbdkit E<ge> 1.2.3 and the header file is
installed, and will set C<NBDKIT_CFLAGS> and C<NBDKIT_LIBS>
appropriately for compiling filters.

You can also run pkg-config/pkgconf directly, for example:

 if ! pkgconf nbdkit --exists; then
   echo "you must install the nbdkit development environment"
   exit 1
 fi

You can also substitute the filterdir variable by doing:

 PKG_CHECK_VAR([NBDKIT_FILTERDIR], [nbdkit], [filterdir])

which defines C<$(NBDKIT_FILTERDIR)> in automake-generated Makefiles.

=head1 SEE ALSO

L<nbdkit(1)>,
L<nbdkit-plugin(3)>.

Standard filters provided by nbdkit:

__FILTER_LINKS__.

=head1 AUTHORS

Eric Blake

Richard W.M. Jones

=head1 COPYRIGHT

Copyright (C) 2013-2020 Red Hat Inc.