1nbdkit-filter(3) NBDKIT nbdkit-filter(3)
2
3
4
6 nbdkit-filter - how to write nbdkit filters
7
9 #include <nbdkit-filter.h>
10
11 static int
12 myfilter_config (nbdkit_next_config *next, void *nxdata,
13 const char *key, const char *value)
14 {
15 if (strcmp (key, "myparameter") == 0) {
16 // ...
17 return 0;
18 }
19 else {
20 // pass through to next filter or plugin
21 return next (nxdata, key, value);
22 }
23 }
24
25 static struct nbdkit_filter filter = {
26 .name = "filter",
27 .config = myfilter_config,
28 /* etc */
29 };
30
31 NBDKIT_REGISTER_FILTER(filter)
32
33 When this has been compiled to a shared library, do:
34
35 nbdkit [--args ...] --filter=./myfilter.so plugin [key=value ...]
36
37 When debugging, use the -fv options:
38
39 nbdkit -fv --filter=./myfilter.so plugin [key=value ...]
40
42 One or more nbdkit filters can be placed in front of an nbdkit plugin
43 to modify the behaviour of the plugin. This manual page describes how
44 to create an nbdkit filter.
45
46 Filters can be used for example to limit requests to an offset/limit,
47 add copy-on-write support, or inject delays or errors (for testing).
48
49 Different filters can be stacked:
50
51 NBD ┌─────────┐ ┌─────────┐ ┌────────┐
52 client ───▶│ filter1 │───▶│ filter2 │── ─ ─ ──▶│ plugin │
53 request └─────────┘ └─────────┘ └────────┘
54
55 Each filter intercepts plugin functions (see nbdkit-plugin(3)) and can
56 call the next filter or plugin in the chain, modifying parameters,
57 calling before the filter function, in the middle or after. Filters
58 may even short-cut the chain. As an example, to process its own
59 parameters the filter can intercept the ".config" method:
60
61 static int
62 myfilter_config (nbdkit_next_config *next, void *nxdata,
63 const char *key, const char *value)
64 {
65 if (strcmp (key, "myparameter") == 0) {
66 // ...
67 // here you would handle this key, value
68 // ...
69 return 0;
70 }
71 else {
72 // pass through to next filter or plugin
73 return next (nxdata, key, value);
74 }
75 }
76
77 static struct nbdkit_filter filter = {
78 // ...
79 .config = myfilter_config,
80 // ...
81 };
82
83 The call to "next (nxdata, ...)" calls the ".config" method of the next
84 filter or plugin in the chain. In the example above any instances of
85 "myparameter=..." on the command line would not be seen by the plugin.
86
87 To see example filters:
88 https://gitlab.com/nbdkit/nbdkit/tree/master/filters
89
90 Filters must be written in C.
91
92 Unlike plugins, where we provide a stable ABI guarantee that permits
93 operation across version differences, filters can only be run with the
94 same version of nbdkit that they were compiled with. The reason for
95 this is two-fold: the filter API includes access to struct
96 nbdkit_next_ops that is likely to change if new callbacks are added
97 (old nbdkit cannot safely run new filters that access new methods); and
98 if we added new methods then an old filter would not see them and so
99 they would be passed unmodified through the filter, and in some cases
100 that leads to data corruption (new nbdkit cannot safely run old filters
101 unaware of new methods). Therefore, unlike plugins, you should not
102 expect to distribute filters separately from nbdkit.
103
105 All filters should start by including this header file.
106
108 All filters must define and register one "struct nbdkit_filter", which
109 contains the name of the filter and pointers to plugin methods that the
110 filter wants to intercept.
111
112 static struct nbdkit_filter filter = {
113 .name = "filter",
114 .longname = "My Filter",
115 .description = "This is my great filter for nbdkit",
116 .config = myfilter_config,
117 /* etc */
118 };
119
120 NBDKIT_REGISTER_FILTER(filter)
121
122 The ".name" field is the name of the filter. This is the only field
123 which is required.
124
126 nbdkit-filter.h defines some function types ("nbdkit_next_config",
127 "nbdkit_next_config_complete", "nbdkit_next_preconnect",
128 "nbdkit_next_list_exports", "nbdkit_next_default_export",
129 "nbdkit_next_open") and a structure called "struct nbdkit_next_ops".
130 These abstract the next plugin or filter in the chain. There is also
131 an opaque pointer "backend", "context" or "nxdata" which must be passed
132 along when calling these functions. The value of "backend" is stable
133 between ".after_fork", ".preconnect", ".list_exports", and
134 ".default_export", and can also be obtained by using
135 "nbdkit_context_get_backend" on the "context" parameter to ".open".
136
137 Meanwhile, if the filter does not use "nbdkit_context_set_next", the
138 value of "next" passed to ".prepare" has a stable lifetime that lasts
139 to the corresponding ".finalize", with all intermediate functions (such
140 as ".pread") receiving the same value for convenience. Functions where
141 "nxdata" is not reused are ".config", ".config_complete", and
142 ".get_ready", which are all called during initialization outside any
143 connections. The value of "backend" passed to ".after_fork" also
144 occurs without connections, but is shared with ".preconnect",
145 ".list_exports", and ".default_export", and can also be obtained from
146 the "context" passed to ".open", and has a lifetime that lasts to
147 ".cleanup" for use by "nbdkit_next_context_open". In turn, the value
148 of "context" passed to ".open" has a lifetime that lasts until the
149 matching ".close" for use by "nbdkit_context_get_backend" and
150 "nbdkit_context_set_next".
151
152 Next config, open and close
153 The filter’s ".config", ".config_complete", ".get_ready",
154 ".after_fork", ".preconnect", ".list_exports", ".default_export" and
155 ".open" methods may only call the next ".config", ".config_complete",
156 ".get_ready", ".after_fork", ".preconnect", ".list_exports",
157 ".default_export" and ".open" method in the chain (optionally for
158 ".config" and ".open").
159
160 The filter’s ".close" method is called when an old connection closed,
161 and this has no "next" parameter because it cannot be short-circuited.
162
163 "nbdkit_next"
164 The filter generally needs to call into the underlying plugin, which is
165 done via a pointer to "struct nbdkit_next_ops", also available as the
166 typedef "nbdkit_next". The most common behavior is to create a next
167 context per connection by calling the "next_open" parameter during
168 ".open", at which point the next context will be automatically provided
169 to the filter’s other methods like ".prepare", ".get_size", ".pread"
170 etc. The "nbdkit_next" struct contains a comparable set of accessors
171 to plugin methods that can be called during a connection. When using
172 automatic registration, the "next" parameter is stable between
173 ".prepare" and ".finalize", and nbdkit automatically prepares,
174 finalizes, and closes the next context at the right point in the filter
175 connection lifecycle.
176
177 Alternatively, the filter can manage plugin contexts manually, whether
178 to multiplex multiple client connections through a single context into
179 the plugin, or to open multiple plugin contexts to perform retries or
180 otherwise service a single client connection more efficiently. In this
181 mode of operation, the filter uses "nbdkit_next_context_open" to open a
182 plugin context using the "backend" parameter passed to ".after_fork",
183 ".preconnect", ".list_exports", ".default_export", or obtained from
184 using "nbdkit_context_get_backend" on the "context" parameter to
185 ".open". The resulting next context has a lifecycle under manual
186 control, where the filter must use "next->prepare (next)" before using
187 any other function pointers within the next context, and must reclaim
188 the memory using "next->finalize (next)" and
189 "nbdkit_next_context_close" when done. A filter using manual lifecycle
190 management may use "nbdkit_context_set_next" to associate the next
191 context into the current connection, which lets nbdkit then pass that
192 context as the "next" parameter to future connection-related functions
193 like ".pread" and take over lifecycle responsibility.
194
195 "nbdkit_context_get_backend"
196
197 "nbdkit_next_context_open"
198
199 "nbdkit_next_context_close"
200
201 "nbdkit_context_set_next"
202
203 nbdkit_backend *nbdkit_context_get_backend (nbdkit_context *context);
204
205 Obtains the backend pointer from the "context" parameter to ".open",
206 matching the backend pointer available to ".after_fork", ".preconnect",
207 ".list_exports", and ".default_export". This backend pointer has a
208 stable lifetime from the time of ".after_fork" until ".cleanup".
209
210 nbdkit_next *nbdkit_next_context_open (nbdkit_backend *backend,
211 int readonly, const char *exportname,
212 int shared);
213
214 This function attempts to open a new context into the plugin in
215 relation to the filter's current "backend". The "readonly" and
216 "exportname" parameters behave the same as documented in ".open". The
217 resulting context will be under the filter's manual lifecycle control
218 unless the filter associates it into the connection with
219 "nbdkit_context_set_next". The filter should be careful to not violate
220 any threading model restrictions of the plugin if it opens more than
221 one context.
222
223 If "shared" is false, this function must be called while servicing an
224 existing client connection, and the new context will share the same
225 connection details (export name, tls status, and shorter interned
226 string lifetimes) as the current connection, and thus should not be
227 used after the client connection ends. Conversely, if "shared" is
228 true, this function may be called outside of a current client
229 connection (such as during ".after_fork"), and the resulting context
230 may be freely shared among multiple client connections. In shared
231 mode, it will not be possible for the plugin to differentiate content
232 based on the client export name, the result of the plugin calling
233 nbdkit_is_tls() will depend solely whether --tls=require was on the
234 command line, the lifetime of interned strings (via
235 "nbdkit_strdup_intern" and friends) lasts for the life of the filter,
236 and the filter must take care to not expose potentially-secure
237 information from the backend to an insecure client.
238
239 void nbdkit_next_context_close (nbdkit_next *next);
240
241 This function closes a context into the plugin. If the context has
242 previously been prepared, it should first be finalized before using
243 this function. This function does not need to be called for a plugin
244 context that has been associated with the filter connection via
245 "nbdkit_context_set_next" prior to the ".close" callback.
246
247 nbdkit_next *nbdkit_context_set_next (nbdkit_context *context,
248 nbdkit_next *next);
249
250 This function associates a plugin context with the filter's current
251 connection context, given by the "context" parameter to ".open". Once
252 associated, this plugin context will be given as the "next" parameter
253 to all other connection-specific callbacks. If associated during
254 ".open", nbdkit will take care of preparing the context prior to
255 ".prepare"; if still associated before ".finalize", nbdkit will take
256 care of finalizing the context, and also for closing it. A filter may
257 also pass "NULL" for "next", to remove any association; if no plugin
258 context is associated with the connection, then filter callbacks such
259 as ".pread" will receive "NULL" for their "next" parameter.
260
261 This function returns the previous context that had been associated
262 with the connection prior to switching the association to "next"; this
263 result will be "NULL" if there was no previous association. The filter
264 assumes manual responsibility for any remaining lifecycle functions
265 that must be called on the returned context.
266
267 Using "nbdkit_next"
268 Regardless of whether the plugin context is managed automatically or
269 manually, it is possible for a filter to issue (for example) extra
270 "next->pread" calls in response to a single ".pwrite" call.
271
272 The "next" parameter serves two purposes: it serves as the struct to
273 access the pointers to all the plugin connection functions, and it
274 serves as the opaque data that must be passed as the first parameter to
275 those functions. For example, calling the plugin's can_flush
276 functionality would be done via
277
278 next->can_flush (next)
279
280 Note that the semantics of the functions in "struct nbdkit_next_ops"
281 are slightly different from what a plugin implements: for example, when
282 a plugin's ".pread" returns -1 on error, the error value to advertise
283 to the client is implicit (via the plugin calling "nbdkit_set_error" or
284 setting "errno"), whereas "next->pread" exposes this via an explicit
285 parameter, allowing a filter to learn or modify this error if desired.
286
287 Use of "next->prepare" and "next->finalize" is only needed when
288 manually managing the plugin context lifetime.
289
290 Other considerations
291 You can modify parameters when you call the "next" function. However
292 be careful when modifying strings because for some methods (eg.
293 ".config") the plugin may save the string pointer that you pass along.
294 So you may have to ensure that the string is not freed for the lifetime
295 of the server; you may find "nbdkit_strdup_intern" helpful for avoiding
296 a memory leak while still obeying lifecycle constraints.
297
298 Note that if your filter registers a callback but in that callback it
299 doesn't call the "next" function then the corresponding method in the
300 plugin will never be called. In particular, your ".open" method, if
301 you have one, must call the "next" method if you want the underlying
302 plugin to be available to all further "nbdkit_next" use.
303
305 "struct nbdkit_filter" has some static fields describing the filter and
306 optional callback functions which can be used to intercept plugin
307 methods.
308
309 ".name"
310 const char *name;
311
312 This field (a string) is required, and must contain only ASCII
313 alphanumeric characters or non-leading dashes, and be unique amongst
314 all filters.
315
316 ".longname"
317 const char *longname;
318
319 An optional free text name of the filter. This field is used in error
320 messages.
321
322 ".description"
323 const char *description;
324
325 An optional multi-line description of the filter.
326
327 ".load"
328 void load (void);
329
330 This is called once just after the filter is loaded into memory. You
331 can use this to perform any global initialization needed by the filter.
332
333 ".unload"
334 void unload (void);
335
336 This may be called once just before the filter is unloaded from memory.
337 Note that it's not guaranteed that ".unload" will always be called (eg.
338 the server might be killed or segfault), so you should try to make the
339 filter as robust as possible by not requiring cleanup. See also
340 "SHUTDOWN" in nbdkit-plugin(3).
341
342 ".config"
343 int (*config) (nbdkit_next_config *next, void *nxdata,
344 const char *key, const char *value);
345
346 This intercepts the plugin ".config" method and can be used by the
347 filter to parse its own command line parameters. You should try to
348 make sure that command line parameter keys that the filter uses do not
349 conflict with ones that could be used by a plugin.
350
351 If there is an error, ".config" should call "nbdkit_error" with an
352 error message and return "-1".
353
354 ".config_complete"
355 int (*config_complete) (nbdkit_next_config_complete *next, void *nxdata);
356
357 This intercepts the plugin ".config_complete" method and can be used to
358 ensure that all parameters needed by the filter were supplied on the
359 command line.
360
361 If there is an error, ".config_complete" should call "nbdkit_error"
362 with an error message and return "-1".
363
364 ".config_help"
365 const char *config_help;
366
367 This optional multi-line help message should summarize any "key=value"
368 parameters that it takes. It does not need to repeat what already
369 appears in ".description".
370
371 If the filter doesn't take any config parameters you should probably
372 omit this.
373
374 ".thread_model"
375 int (*thread_model) (void);
376
377 Filters may tighten (but not relax) the thread model of the plugin, by
378 defining this callback. Note that while plugins use a compile-time
379 definition of "THREAD_MODEL", filters do not need to declare a model at
380 compile time; instead, this callback is called after ".config_complete"
381 and before any connections are created. See "THREADS" in
382 nbdkit-plugin(3) for a discussion of thread models.
383
384 The final thread model used by nbdkit is the smallest (ie. most
385 serialized) out of all the filters and the plugin, and applies for all
386 connections. Requests for a model larger than permitted by the plugin
387 are silently ignored. It is acceptable for decisions made during
388 ".config" and ".config_complete" to determine which model to request.
389
390 This callback is optional; if it is not present, the filter must be
391 written to handle fully parallel requests, including when multiple
392 requests are issued in parallel on the same connection, similar to a
393 plugin requesting "NBDKIT_THREAD_MODEL_PARALLEL". This ensures the
394 filter doesn't slow down other filters or plugins.
395
396 If there is an error, ".thread_model" should call "nbdkit_error" with
397 an error message and return "-1".
398
399 ".get_ready"
400 int (*get_ready) (int thread_model);
401
402 This optional callback is reached if the plugin ".get_ready" method
403 succeeded (if the plugin failed, nbdkit has already exited), and can be
404 used by the filter to get ready to serve requests.
405
406 The "thread_model" parameter informs the filter about the final thread
407 model chosen by nbdkit after considering the results of ".thread_model"
408 of all filters in the chain after ".config_complete".
409
410 If there is an error, ".get_ready" should call "nbdkit_error" with an
411 error message and return "-1".
412
413 ".after_fork"
414 int (*after_fork) (nbdkit_backend *backend);
415
416 This optional callback is reached after the plugin ".after_fork" method
417 has succeeded (if the plugin failed, nbdkit has already exited), and
418 can be used by the filter to start background threads. The "backend"
419 parameter is valid until ".cleanup", for creating manual contexts into
420 the backend with "nbdkit_next_context_open".
421
422 If there is an error, ".after_fork" should call "nbdkit_error" with an
423 error message and return "-1".
424
425 ".cleanup"
426 int (cleanup) (nbdkit_backend *backend);
427
428 This optional callback is reached once after all client connections
429 have been closed, but before the underlying plugin ".cleanup" or any
430 ".unload" callbacks. It can be used by the filter to gracefully close
431 any background threads created during ".after_fork", as well as close
432 any manual contexts into "backend" previously opened with
433 "nbdkit_next_context_open".
434
435 Note that it's not guaranteed that ".cleanup" will always be called
436 (eg. the server might be killed or segfault), so you should try to make
437 the filter as robust as possible by not requiring cleanup. See also
438 "SHUTDOWN" in nbdkit-plugin(3).
439
440 ".preconnect"
441 int (*preconnect) (nbdkit_next_preconnect *next, nbdkit_backend *nxdata,
442 int readonly);
443
444 This intercepts the plugin ".preconnect" method and can be used to
445 filter access to the server.
446
447 If there is an error, ".preconnect" should call "nbdkit_error" with an
448 error message and return "-1".
449
450 ".list_exports"
451 int (*list_exports) (nbdkit_next_list_exports *next, nbdkit_backend *nxdata,
452 int readonly, int is_tls,
453 struct nbdkit_exports *exports);
454
455 This intercepts the plugin ".list_exports" method and can be used to
456 filter which exports are advertised.
457
458 The "readonly" parameter matches what is passed to <.preconnect> and
459 ".open", and may be changed by the filter when calling into the plugin.
460 The "is_tls" parameter informs the filter whether TLS negotiation has
461 been completed by the client, but is not passed on to "next" because it
462 cannot be altered.
463
464 It is possible for filters to transform the exports list received back
465 from the layer below. Without error checking it would look like this:
466
467 myfilter_list_exports (...)
468 {
469 size_t i;
470 struct nbdkit_exports *exports2;
471 struct nbdkit_export e;
472 char *name, *desc;
473
474 exports2 = nbdkit_exports_new ();
475 next_list_exports (nxdata, readonly, exports);
476 for (i = 0; i < nbdkit_exports_count (exports2); ++i) {
477 e = nbdkit_get_export (exports2, i);
478 name = adjust (e.name);
479 desc = adjust (e.desc);
480 nbdkit_add_export (exports, name, desc);
481 free (name);
482 free (desc);
483 }
484 nbdkit_exports_free (exports2);
485 }
486
487 If there is an error, ".list_exports" should call "nbdkit_error" with
488 an error message and return "-1".
489
490 Allocating and freeing nbdkit_exports list
491
492 Two functions are provided to filters only for allocating and freeing
493 the list:
494
495 struct nbdkit_exports *nbdkit_exports_new (void);
496
497 Allocates and returns a new, empty exports list.
498
499 On error this function can return "NULL". In this case it calls
500 "nbdkit_error" as required. "errno" will be set to a suitable value.
501
502 void nbdkit_exports_free (struct nbdkit_exports *);
503
504 Frees an existing exports list.
505
506 Iterating over nbdkit_exports list
507
508 Two functions are provided to filters only to iterate over the exports
509 in order:
510
511 size_t nbdkit_exports_count (const struct nbdkit_exports *);
512
513 Returns the number of exports in the list.
514
515 struct nbdkit_export {
516 char *name;
517 char *description;
518 };
519 const struct nbdkit_export nbdkit_get_export (const struct nbdkit_exports *,
520 size_t i);
521
522 Returns a copy of the "i"'th export.
523
524 ".default_export"
525 const char *default_export (nbdkit_next_default_export *next,
526 nbdkit_backend *nxdata,
527 int readonly, int is_tls)
528
529 This intercepts the plugin ".default_export" method and can be used to
530 alter the canonical export name used in place of the default "".
531
532 The "readonly" parameter matches what is passed to <.preconnect> and
533 ".open", and may be changed by the filter when calling into the plugin.
534 The "is_tls" parameter informs the filter whether TLS negotiation has
535 been completed by the client, but is not passed on to "next" because it
536 cannot be altered.
537
538 ".open"
539 void * (*open) (nbdkit_next_open *next, nbdkit_context *context,
540 int readonly, const char *exportname, int is_tls);
541
542 This is called when a new client connection is opened and can be used
543 to allocate any per-connection data structures needed by the filter.
544 The handle (which is not the same as the plugin handle) is passed back
545 to other filter callbacks and could be freed in the ".close" callback.
546
547 Note that the handle is completely opaque to nbdkit, but it must not be
548 NULL. If you don't need to use a handle, return
549 "NBDKIT_HANDLE_NOT_NEEDED" which is a static non-NULL pointer.
550
551 If there is an error, ".open" should call "nbdkit_error" with an error
552 message and return "NULL".
553
554 This callback is optional, but if provided, it should call "next",
555 passing "readonly" and "exportname" possibly modified according to how
556 the filter plans to use the plugin ("is_tls" is not passed, because a
557 filter cannot modify it). Typically, the filter passes the same values
558 as it received, or passes readonly=true to provide a writable layer on
559 top of a read-only backend. However, it is also acceptable to attempt
560 write access to the plugin even if this filter is readonly, such as
561 when a file system mounted read-only still requires write access to the
562 underlying device in case a journal needs to be replayed for
563 consistency as part of the mounting process.
564
565 The "exportname" string is only guaranteed to be available during the
566 call (different than the lifetime for the return of
567 "nbdkit_export_name" used by plugins). If the filter needs to use it
568 (other than immediately passing it down to the next layer) it must take
569 a copy, although "nbdkit_strdup_intern" is useful for this task. The
570 "exportname" and "is_tls" parameters are provided so that filters do
571 not need to use the plugin-only interfaces of "nbdkit_export_name" and
572 "nbdkit_is_tls".
573
574 The filter should generally call "next" as its first step, to allocate
575 from the plugin outwards, so that ".close" running from the outer
576 filter to the plugin will be in reverse. Skipping a call to "next" is
577 acceptable if the filter will not access "nbdkit_next" during any of
578 the remaining callbacks reached on the same connection. The "next"
579 function is provided for convenience; the same functionality can be
580 obtained manually (other than error checking) by using the following:
581
582 nbdkit_context_set_next (context, nbdkit_next_context_open
583 (nbdkit_context_get_backend (context), readonly, exportname, false));
584
585 The value of "context" in this call has a lifetime that lasts until the
586 counterpart ".close", and it is this value that may be passed to
587 "nbdkit_context_get_backend" to obtain the "backend" parameter used to
588 open a plugin context with "nbdkit_next_context_open", as well as the
589 "context" parameter used to associate a plugin context into the current
590 connection with "nbdkit_context_set_next".
591
592 ".close"
593 void (*close) (void *handle);
594
595 This is called when the client closes the connection. It should clean
596 up any per-connection resources used by the filter. It is called
597 beginning with the outermost filter and ending with the plugin (the
598 opposite order of ".open" if all filters call "next" first), although
599 this order technically does not matter since the callback cannot report
600 failures or access the underlying plugin.
601
602 ".prepare"
603 ".finalize"
604 int (*prepare) (nbdkit_next *next, void *handle, int readonly);
605 int (*finalize) (nbdkit_next *next, void *handle);
606
607 These two methods can be used to perform any necessary operations just
608 after opening the connection (".prepare") or just before closing the
609 connection (".finalize").
610
611 For example if you need to scan the underlying disk to check for a
612 partition table, you could do it in your ".prepare" method (calling the
613 plugin's ".get_size" and ".pread" methods via "next"). Or if you need
614 to cleanly update superblock data in the image on close you can do it
615 in your ".finalize" method (calling the plugin's ".pwrite" method).
616 Doing these things in the filter's ".open" or ".close" method is not
617 possible without using manual context lifecycle management.
618
619 For ".prepare", the value of "readonly" is the same as was passed to
620 ".open", declaring how this filter will be used.
621
622 Note that nbdkit performs sanity checking on requests made to the
623 underlying plugin; for example, "next->pread" cannot be called on a
624 given connection unless "next->get_size" has first been called at least
625 once in the same connection (to ensure the read requests are in
626 bounds), and "next->pwrite" further requires an earlier successful call
627 to "next->can_write". In many filters, these prerequisites will be
628 automatically called during the client negotiation phase, but there are
629 cases where a filter overrides query functions or makes I/O calls into
630 the plugin before handshaking is complete, where the filter needs to
631 make those prerequisite calls manually during ".prepare".
632
633 While there are "next->prepare" and "next->finalize" functions, these
634 are different from other filter methods, in that any plugin context
635 associated with the current connection (via the "next" parameter to
636 ".open", or via "nbdkit_context_set_next", is prepared and finalized
637 automatically by nbdkit, so they are only used during manual lifecycle
638 management. Prepare methods are called starting with the filter
639 closest to the plugin and proceeding outwards (matching the order of
640 ".open" if all filters call "next" before doing anything locally), and
641 only when an outer filter did not skip the "next" call during ".open".
642 Finalize methods are called in the reverse order of prepare methods,
643 with the outermost filter first (and matching the order of ".close"),
644 and only if the prepare method succeeded.
645
646 If there is an error, both callbacks should call "nbdkit_error" with an
647 error message and return "-1". An error in ".prepare" is reported to
648 the client, but leaves the connection open (a client may try again with
649 a different export name, for example); while an error in ".finalize"
650 forces the client to disconnect.
651
652 ".get_size"
653 int64_t (*get_size) (nbdkit_next *next, void *handle);
654
655 This intercepts the plugin ".get_size" method and can be used to read
656 or modify the apparent size of the block device that the NBD client
657 will see.
658
659 The returned size must be ≥ 0. If there is an error, ".get_size"
660 should call "nbdkit_error" with an error message and return "-1". This
661 function is only called once per connection and cached by nbdkit.
662 Similarly, repeated calls to "next->get_size" will return a cached
663 value.
664
665 ".export_description"
666 const char *export_description (nbdkit_next *next, void *handle);
667
668 This intercepts the plugin ".export_description" method and can be used
669 to read or modify the export description that the NBD client will see.
670
671 ".block_size"
672 int block_size (nbdkit_next *next, void *handle, uint32_t *minimum,
673 uint32_t *preferred, uint32_t *maximum);
674
675 This intercepts the plugin ".block_size" method and can be used to read
676 or modify the block size constraints that the NBD client will see.
677
678 ".can_write"
679 ".can_flush"
680 ".is_rotational"
681 ".can_trim"
682 ".can_zero"
683 ".can_fast_zero"
684 ".can_extents"
685 ".can_fua"
686 ".can_multi_conn"
687 ".can_cache"
688 int (*can_write) (nbdkit_next *next, void *handle);
689 int (*can_flush) (nbdkit_next *next, void *handle);
690 int (*is_rotational) (nbdkit_next *next, void *handle);
691 int (*can_trim) (nbdkit_next *next, void *handle);
692 int (*can_zero) (nbdkit_next *next, void *handle);
693 int (*can_fast_zero) (nbdkit_next *next, void *handle);
694 int (*can_extents) (nbdkit_next *next, void *handle);
695 int (*can_fua) (nbdkit_next *next, void *handle);
696 int (*can_multi_conn) (nbdkit_next *next, void *handle);
697 int (*can_cache) (nbdkit_next *next, void *handle);
698
699 These intercept the corresponding plugin methods, and control feature
700 bits advertised to the client.
701
702 Of note, the semantics of ".can_zero" callback in the filter are
703 slightly different from the plugin, and must be one of three success
704 values visible only to filters:
705
706 "NBDKIT_ZERO_NONE"
707 Completely suppress advertisement of write zero support (this can
708 only be done from filters, not plugins).
709
710 "NBDKIT_ZERO_EMULATE"
711 Inform nbdkit that write zeroes should immediately fall back to
712 ".pwrite" emulation without trying ".zero" (this value is returned
713 by "next->can_zero" if the plugin returned false in its
714 ".can_zero").
715
716 "NBDKIT_ZERO_NATIVE"
717 Inform nbdkit that write zeroes should attempt to use ".zero",
718 although it may still fall back to ".pwrite" emulation for
719 "ENOTSUP" or "EOPNOTSUPP" failures (this value is returned by
720 "next->can_zero" if the plugin returned true in its ".can_zero").
721
722 Remember that most of the feature check functions return merely a
723 boolean success value, while ".can_zero", ".can_fua" and ".can_cache"
724 have three success values.
725
726 The difference between ".can_fua" values may affect choices made in the
727 filter: when splitting a write request that requested FUA from the
728 client, if "next->can_fua" returns "NBDKIT_FUA_NATIVE", then the filter
729 should pass the FUA flag on to each sub-request; while if it is known
730 that FUA is emulated by a flush because of a return of
731 "NBDKIT_FUA_EMULATE", it is more efficient to only flush once after all
732 sub-requests have completed (often by passing "NBDKIT_FLAG_FUA" on to
733 only the final sub-request, or by dropping the flag and ending with a
734 direct call to "next->flush").
735
736 If there is an error, the callback should call "nbdkit_error" with an
737 error message and return "-1". These functions are called at most once
738 per connection and cached by nbdkit. Similarly, repeated calls to any
739 of the "nbdkit_next" counterparts will return a cached value; by
740 calling into the plugin during ".prepare", you can ensure that later
741 use of the cached values during data commands like <.pwrite> will not
742 fail.
743
744 ".pread"
745 int (*pread) (nbdkit_next *next,
746 void *handle, void *buf, uint32_t count, uint64_t offset,
747 uint32_t flags, int *err);
748
749 This intercepts the plugin ".pread" method and can be used to read or
750 modify data read by the plugin.
751
752 The parameter "flags" exists in case of future NBD protocol extensions;
753 at this time, it will be 0 on input, and the filter should not pass any
754 flags to "next->pread".
755
756 If there is an error (including a short read which couldn't be
757 recovered from), ".pread" should call "nbdkit_error" with an error
758 message and return -1 with "err" set to the positive errno value to
759 return to the client.
760
761 ".pwrite"
762 int (*pwrite) (nbdkit_next *next,
763 void *handle,
764 const void *buf, uint32_t count, uint64_t offset,
765 uint32_t flags, int *err);
766
767 This intercepts the plugin ".pwrite" method and can be used to modify
768 data written by the plugin.
769
770 This function will not be called if ".can_write" returned false; in
771 turn, the filter should not call "next->pwrite" if "next->can_write"
772 did not return true.
773
774 The parameter "flags" may include "NBDKIT_FLAG_FUA" on input based on
775 the result of ".can_fua". In turn, the filter should only pass
776 "NBDKIT_FLAG_FUA" on to "next->pwrite" if "next->can_fua" returned a
777 positive value.
778
779 If there is an error (including a short write which couldn't be
780 recovered from), ".pwrite" should call "nbdkit_error" with an error
781 message and return -1 with "err" set to the positive errno value to
782 return to the client.
783
784 ".flush"
785 int (*flush) (nbdkit_next *next,
786 void *handle, uint32_t flags, int *err);
787
788 This intercepts the plugin ".flush" method and can be used to modify
789 flush requests.
790
791 This function will not be called if ".can_flush" returned false; in
792 turn, the filter should not call "next->flush" if "next->can_flush" did
793 not return true.
794
795 The parameter "flags" exists in case of future NBD protocol extensions;
796 at this time, it will be 0 on input, and the filter should not pass any
797 flags to "next->flush".
798
799 If there is an error, ".flush" should call "nbdkit_error" with an error
800 message and return -1 with "err" set to the positive errno value to
801 return to the client.
802
803 ".trim"
804 int (*trim) (nbdkit_next *next,
805 void *handle, uint32_t count, uint64_t offset,
806 uint32_t flags, int *err);
807
808 This intercepts the plugin ".trim" method and can be used to modify
809 trim requests.
810
811 This function will not be called if ".can_trim" returned false; in
812 turn, the filter should not call "next->trim" if "next->can_trim" did
813 not return true.
814
815 The parameter "flags" may include "NBDKIT_FLAG_FUA" on input based on
816 the result of ".can_fua". In turn, the filter should only pass
817 "NBDKIT_FLAG_FUA" on to "next->trim" if "next->can_fua" returned a
818 positive value.
819
820 If there is an error, ".trim" should call "nbdkit_error" with an error
821 message and return -1 with "err" set to the positive errno value to
822 return to the client.
823
824 ".zero"
825 int (*zero) (nbdkit_next *next,
826 void *handle, uint32_t count, uint64_t offset, uint32_t flags,
827 int *err);
828
829 This intercepts the plugin ".zero" method and can be used to modify
830 zero requests.
831
832 This function will not be called if ".can_zero" returned
833 "NBDKIT_ZERO_NONE" or "NBDKIT_ZERO_EMULATE"; in turn, the filter should
834 not call "next->zero" if "next->can_zero" returned "NBDKIT_ZERO_NONE".
835
836 On input, the parameter "flags" may include "NBDKIT_FLAG_MAY_TRIM"
837 unconditionally, "NBDKIT_FLAG_FUA" based on the result of ".can_fua",
838 and "NBDKIT_FLAG_FAST_ZERO" based on the result of ".can_fast_zero".
839 In turn, the filter may pass "NBDKIT_FLAG_MAY_TRIM" unconditionally,
840 but should only pass "NBDKIT_FLAG_FUA" or "NBDKIT_FLAG_FAST_ZERO" on to
841 "next->zero" if the corresponding "next->can_fua" or
842 "next->can_fast_zero" returned a positive value.
843
844 Note that unlike the plugin ".zero" which is permitted to fail with
845 "ENOTSUP" or "EOPNOTSUPP" to force a fallback to ".pwrite", the
846 function "next->zero" will not fail with "err" set to "ENOTSUP" or
847 "EOPNOTSUPP" unless "NBDKIT_FLAG_FAST_ZERO" was used, because otherwise
848 the fallback has already taken place.
849
850 If there is an error, ".zero" should call "nbdkit_error" with an error
851 message and return -1 with "err" set to the positive errno value to
852 return to the client. The filter should not fail with "ENOTSUP" or
853 "EOPNOTSUPP" unless "flags" includes "NBDKIT_FLAG_FAST_ZERO" (while
854 plugins have automatic fallback to ".pwrite", filters do not).
855
856 ".extents"
857 int (*extents) (nbdkit_next *next,
858 void *handle, uint32_t count, uint64_t offset, uint32_t flags,
859 struct nbdkit_extents *extents,
860 int *err);
861
862 This intercepts the plugin ".extents" method and can be used to modify
863 extent requests.
864
865 This function will not be called if ".can_extents" returned false; in
866 turn, the filter should not call "next->extents" if "next->can_extents"
867 did not return true.
868
869 It is possible for filters to transform the extents list received back
870 from the layer below. Without error checking it would look like this:
871
872 myfilter_extents (..., uint32_t count, uint64_t offset, ...)
873 {
874 size_t i;
875 struct nbdkit_extents *extents2;
876 struct nbdkit_extent e;
877 int64_t size;
878
879 size = next->get_size (next);
880 extents2 = nbdkit_extents_new (offset + shift, size);
881 next->extents (next, count, offset + shift, flags, extents2, err);
882 for (i = 0; i < nbdkit_extents_count (extents2); ++i) {
883 e = nbdkit_get_extent (extents2, i);
884 e.offset -= shift;
885 nbdkit_add_extent (extents, e.offset, e.length, e.type);
886 }
887 nbdkit_extents_free (extents2);
888 }
889
890 If there is an error, ".extents" should call "nbdkit_error" with an
891 error message and return -1 with "err" set to the positive errno value
892 to return to the client.
893
894 Allocating and freeing nbdkit_extents list
895
896 Two functions are provided to filters only for allocating and freeing
897 the map:
898
899 struct nbdkit_extents *nbdkit_extents_new (uint64_t start, uint64_t end);
900
901 Allocates and returns a new, empty extents list. The "start" parameter
902 is the start of the range described in the list, and the "end"
903 parameter is the offset of the byte beyond the end. Normally you would
904 pass in "offset" as the start and the size of the plugin as the end,
905 but for filters which adjust offsets, they should pass in the adjusted
906 offset.
907
908 On error this function can return "NULL". In this case it calls
909 "nbdkit_error" and/or "nbdkit_set_error" as required. "errno" will be
910 set to a suitable value.
911
912 void nbdkit_extents_free (struct nbdkit_extents *);
913
914 Frees an existing extents list.
915
916 Iterating over nbdkit_extents list
917
918 Two functions are provided to filters only to iterate over the extents
919 in order:
920
921 size_t nbdkit_extents_count (const struct nbdkit_extents *);
922
923 Returns the number of extents in the list.
924
925 struct nbdkit_extent {
926 uint64_t offset;
927 uint64_t length;
928 uint32_t type;
929 };
930 struct nbdkit_extent nbdkit_get_extent (const struct nbdkit_extents *,
931 size_t i);
932
933 Returns a copy of the "i"'th extent.
934
935 Reading the full extents from the plugin
936
937 A convenience function is provided to filters only which makes one or
938 more requests to the underlying plugin until we have a full set of
939 extents covering the region "[offset..offset+count-1]".
940
941 struct nbdkit_extents *nbdkit_extents_full (
942 nbdkit_next *next,
943 uint32_t count, uint64_t offset,
944 uint32_t flags, int *err);
945
946 Note this allocates a new "struct nbdkit_extents" which the caller must
947 free. "flags" is passed through to the underlying plugin, but
948 "NBDKIT_FLAG_REQ_ONE" is removed from the set of flags so that the
949 plugin returns as much information as possible (this is usually what
950 you want).
951
952 On error this function can return "NULL". In this case it calls
953 "nbdkit_error" and/or "nbdkit_set_error" as required. *err will be set
954 to a suitable value.
955
956 Enforcing alignment of an nbdkit_extents list
957
958 A convenience function is provided to filters only which makes it
959 easier to ensure that the client only encounters aligned extents.
960
961 int nbdkit_extents_aligned (nbdkit_next *next,
962 uint32_t count, uint64_t offset,
963 uint32_t flags, uint32_t align,
964 struct nbdkit_extents *extents, int *err);
965
966 Calls "next->extents" as needed until at least "align" bytes are
967 obtained, where "align" is a power of 2. Anywhere the underlying
968 plugin returns differing extents within "align" bytes, this function
969 treats that portion of the disk as a single extent with zero and sparse
970 status bits determined by the intersection of all underlying extents.
971 It is an error to call this function with "count" or "offset" that is
972 not already aligned.
973
974 ".cache"
975 int (*cache) (nbdkit_next *next,
976 void *handle, uint32_t count, uint64_t offset,
977 uint32_t flags, int *err);
978
979 This intercepts the plugin ".cache" method and can be used to modify
980 cache requests.
981
982 This function will not be called if ".can_cache" returned
983 "NBDKIT_CACHE_NONE" or "NBDKIT_CACHE_EMULATE"; in turn, the filter
984 should not call "next->cache" if "next->can_cache" returned
985 "NBDKIT_CACHE_NONE".
986
987 The parameter "flags" exists in case of future NBD protocol extensions;
988 at this time, it will be 0 on input, and the filter should not pass any
989 flags to "next->cache".
990
991 If there is an error, ".cache" should call "nbdkit_error" with an error
992 message and return -1 with "err" set to the positive errno value to
993 return to the client.
994
996 If there is an error in the filter itself, the filter should call
997 "nbdkit_error" to report an error message. If the callback is involved
998 in serving data, the explicit "err" parameter determines the error code
999 that will be sent to the client; other callbacks should return the
1000 appropriate error indication, eg. "NULL" or "-1".
1001
1002 "nbdkit_error" has the following prototype and works like printf(3):
1003
1004 void nbdkit_error (const char *fs, ...);
1005 void nbdkit_verror (const char *fs, va_list args);
1006
1007 For convenience, "nbdkit_error" preserves the value of "errno", and
1008 also supports the glibc extension of a single %m in a format string
1009 expanding to "strerror(errno)", even on platforms that don't support
1010 that natively.
1011
1013 Run the server with -f and -v options so it doesn't fork and you can
1014 see debugging information:
1015
1016 nbdkit -fv --filter=./myfilter.so plugin [key=value [key=value [...]]]
1017
1018 To print debugging information from within the filter, call
1019 "nbdkit_debug", which has the following prototype and works like
1020 printf(3):
1021
1022 void nbdkit_debug (const char *fs, ...);
1023 void nbdkit_vdebug (const char *fs, va_list args);
1024
1025 For convenience, "nbdkit_debug" preserves the value of "errno", and
1026 also supports the glibc extension of a single %m in a format string
1027 expanding to "strerror(errno)", even on platforms that don't support
1028 that natively. Note that "nbdkit_debug" only prints things when the
1029 server is in verbose mode (-v option).
1030
1031 Debug Flags
1032 Debug Flags in filters work exactly the same way as plugins. See
1033 "Debug Flags" in nbdkit-plugin(3).
1034
1036 The filter is a "*.so" file and possibly a manual page. You can of
1037 course install the filter "*.so" file wherever you want, and users will
1038 be able to use it by running:
1039
1040 nbdkit --filter=/path/to/filter.so plugin [args]
1041
1042 However if the shared library has a name of the form
1043 "nbdkit-name-filter.so" and if the library is installed in the
1044 $filterdir directory, then users can be run it by only typing:
1045
1046 nbdkit --filter=name plugin [args]
1047
1048 The location of the $filterdir directory is set when nbdkit is compiled
1049 and can be found by doing:
1050
1051 nbdkit --dump-config
1052
1053 If using the pkg-config/pkgconf system then you can also find the
1054 filter directory at compile time by doing:
1055
1056 pkg-config nbdkit --variable=filterdir
1057
1059 nbdkit provides a pkg-config/pkgconf file called "nbdkit.pc" which
1060 should be installed on the correct path when the nbdkit development
1061 environment is installed. You can use this in autoconf configure.ac
1062 scripts to test for the development environment:
1063
1064 PKG_CHECK_MODULES([NBDKIT], [nbdkit >= 1.2.3])
1065
1066 The above will fail unless nbdkit ≥ 1.2.3 and the header file is
1067 installed, and will set "NBDKIT_CFLAGS" and "NBDKIT_LIBS" appropriately
1068 for compiling filters.
1069
1070 You can also run pkg-config/pkgconf directly, for example:
1071
1072 if ! pkg-config nbdkit --exists; then
1073 echo "you must install the nbdkit development environment"
1074 exit 1
1075 fi
1076
1077 You can also substitute the filterdir variable by doing:
1078
1079 PKG_CHECK_VAR([NBDKIT_FILTERDIR], [nbdkit], [filterdir])
1080
1081 which defines "$(NBDKIT_FILTERDIR)" in automake-generated Makefiles.
1082
1084 Instead of using C, it is possible to write filters in C++. However
1085 for inclusion in upstream nbdkit we would generally prefer filters
1086 written in C.
1087
1088 Filters in C++ work almost exactly like those in C, but the way you
1089 define the "nbdkit_filter" struct is slightly different:
1090
1091 namespace {
1092 nbdkit_filter create_filter() {
1093 nbdkit_filter filter = nbdkit_filter ();
1094 filter.name = "myfilter";
1095 filter.config = myfilter_config;
1096 return filter;
1097 }
1098 }
1099 static struct nbdkit_filter filter = create_filter ();
1100 NBDKIT_REGISTER_FILTER(filter)
1101
1103 nbdkit(1), nbdkit-plugin(3).
1104
1105 Standard filters provided by nbdkit:
1106
1107 nbdkit-blocksize-filter(1), nbdkit-blocksize-policy-filter(1),
1108 nbdkit-cache-filter(1), nbdkit-cacheextents-filter(1),
1109 nbdkit-checkwrite-filter(1), nbdkit-cow-filter(1),
1110 nbdkit-ddrescue-filter(1), nbdkit-delay-filter(1),
1111 nbdkit-error-filter(1), nbdkit-exitlast-filter(1),
1112 nbdkit-exitwhen-filter(1), nbdkit-exportname-filter(1),
1113 nbdkit-ext2-filter(1), nbdkit-extentlist-filter(1),
1114 nbdkit-fua-filter(1), nbdkit-gzip-filter(1), nbdkit-ip-filter(1),
1115 nbdkit-limit-filter(1), nbdkit-log-filter(1), nbdkit-luks-filter(1),
1116 nbdkit-multi-conn-filter(1), nbdkit-nocache-filter(1),
1117 nbdkit-noextents-filter(1), nbdkit-nofilter-filter(1),
1118 nbdkit-noparallel-filter(1), nbdkit-nozero-filter(1),
1119 nbdkit-offset-filter(1), nbdkit-partition-filter(1),
1120 nbdkit-pause-filter(1), nbdkit-protect-filter(1),
1121 nbdkit-rate-filter(1), nbdkit-readahead-filter(1),
1122 nbdkit-retry-filter(1), nbdkit-retry-request-filter(1),
1123 nbdkit-scan-filter(1), nbdkit-stats-filter(1), nbdkit-swab-filter(1),
1124 nbdkit-tar-filter(1), nbdkit-tls-fallback-filter(1),
1125 nbdkit-truncate-filter(1), nbdkit-xz-filter(1) .
1126
1128 Eric Blake
1129
1130 Richard W.M. Jones
1131
1133 Copyright (C) 2013-2020 Red Hat Inc.
1134
1136 Redistribution and use in source and binary forms, with or without
1137 modification, are permitted provided that the following conditions are
1138 met:
1139
1140 • Redistributions of source code must retain the above copyright
1141 notice, this list of conditions and the following disclaimer.
1142
1143 • Redistributions in binary form must reproduce the above copyright
1144 notice, this list of conditions and the following disclaimer in the
1145 documentation and/or other materials provided with the
1146 distribution.
1147
1148 • Neither the name of Red Hat nor the names of its contributors may
1149 be used to endorse or promote products derived from this software
1150 without specific prior written permission.
1151
1152 THIS SOFTWARE IS PROVIDED BY RED HAT AND CONTRIBUTORS ''AS IS'' AND ANY
1153 EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
1154 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
1155 PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL RED HAT OR CONTRIBUTORS BE
1156 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
1157 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
1158 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
1159 BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
1160 WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
1161 OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
1162 ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
1163
1164
1165
1166nbdkit-1.32.5 2023-01-03 nbdkit-filter(3)