xref: /spdk/module/bdev/ocf/vbdev_ocf.c (revision 7392cdeff7d8d8d30ec1d8f24a3eebe6f762cde4)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <ocf/ocf.h>
35 #include <ocf/ocf_types.h>
36 #include <ocf/ocf_mngt.h>
37 
38 #include "ctx.h"
39 #include "data.h"
40 #include "volume.h"
41 #include "utils.h"
42 #include "vbdev_ocf.h"
43 
44 #include "spdk/bdev_module.h"
45 #include "spdk/conf.h"
46 #include "spdk/io_channel.h"
47 #include "spdk/string.h"
48 #include "spdk_internal/log.h"
49 #include "spdk/cpuset.h"
50 
51 static struct spdk_bdev_module ocf_if;
52 
53 static TAILQ_HEAD(, vbdev_ocf) g_ocf_vbdev_head
54 	= TAILQ_HEAD_INITIALIZER(g_ocf_vbdev_head);
55 
56 static TAILQ_HEAD(, examining_bdev) g_ocf_examining_bdevs_head
57 	= TAILQ_HEAD_INITIALIZER(g_ocf_examining_bdevs_head);
58 
59 bool g_fini_started = false;
60 
61 /* Structure for keeping list of bdevs that are claimed but not used yet */
62 struct examining_bdev {
63 	struct spdk_bdev           *bdev;
64 	TAILQ_ENTRY(examining_bdev) tailq;
65 };
66 
67 /* Add bdev to list of claimed */
68 static void
69 examine_start(struct spdk_bdev *bdev)
70 {
71 	struct examining_bdev *entry = malloc(sizeof(*entry));
72 
73 	assert(entry);
74 	entry->bdev = bdev;
75 	TAILQ_INSERT_TAIL(&g_ocf_examining_bdevs_head, entry, tailq);
76 }
77 
78 /* Find bdev on list of claimed bdevs, then remove it,
79  * if it was the last one on list then report examine done */
80 static void
81 examine_done(int status, struct vbdev_ocf *vbdev, void *cb_arg)
82 {
83 	struct spdk_bdev *bdev = cb_arg;
84 	struct examining_bdev *entry, *safe, *found = NULL;
85 
86 	TAILQ_FOREACH_SAFE(entry, &g_ocf_examining_bdevs_head, tailq, safe) {
87 		if (entry->bdev == bdev) {
88 			if (found) {
89 				goto remove;
90 			} else {
91 				found = entry;
92 			}
93 		}
94 	}
95 
96 	assert(found);
97 	spdk_bdev_module_examine_done(&ocf_if);
98 
99 remove:
100 	TAILQ_REMOVE(&g_ocf_examining_bdevs_head, found, tailq);
101 	free(found);
102 }
103 
104 /* Free allocated strings and structure itself
105  * Used at shutdown only */
106 static void
107 free_vbdev(struct vbdev_ocf *vbdev)
108 {
109 	if (!vbdev) {
110 		return;
111 	}
112 
113 	free(vbdev->name);
114 	free(vbdev->cache.name);
115 	free(vbdev->core.name);
116 	free(vbdev);
117 }
118 
119 /* Get existing cache base
120  * that is attached to other vbdev */
121 static struct vbdev_ocf_base *
122 get_other_cache_base(struct vbdev_ocf_base *base)
123 {
124 	struct vbdev_ocf *vbdev;
125 
126 	TAILQ_FOREACH(vbdev, &g_ocf_vbdev_head, tailq) {
127 		if (&vbdev->cache == base || !vbdev->cache.attached) {
128 			continue;
129 		}
130 		if (!strcmp(vbdev->cache.name, base->name)) {
131 			return &vbdev->cache;
132 		}
133 	}
134 
135 	return NULL;
136 }
137 
138 /* Get existing OCF cache instance
139  * that is started by other vbdev */
140 static ocf_cache_t
141 get_other_cache_instance(struct vbdev_ocf *vbdev)
142 {
143 	struct vbdev_ocf *cmp;
144 
145 	TAILQ_FOREACH(cmp, &g_ocf_vbdev_head, tailq) {
146 		if (cmp->state.doing_finish || cmp == vbdev) {
147 			continue;
148 		}
149 		if (strcmp(cmp->cache.name, vbdev->cache.name)) {
150 			continue;
151 		}
152 		if (cmp->ocf_cache) {
153 			return cmp->ocf_cache;
154 		}
155 	}
156 
157 	return NULL;
158 }
159 
160 /* Close and unclaim base bdev */
161 static void
162 remove_base_bdev(struct vbdev_ocf_base *base)
163 {
164 	if (base->attached) {
165 		spdk_bdev_module_release_bdev(base->bdev);
166 		spdk_bdev_close(base->desc);
167 		base->attached = false;
168 
169 		if (base->management_channel && !base->is_cache) {
170 			spdk_put_io_channel(base->management_channel);
171 		}
172 	}
173 }
174 
175 /* Finish unregister operation */
176 static void
177 unregister_finish(struct vbdev_ocf *vbdev)
178 {
179 	spdk_bdev_destruct_done(&vbdev->exp_bdev, vbdev->state.stop_status);
180 	vbdev_ocf_cache_ctx_put(vbdev->cache_ctx);
181 	vbdev_ocf_mngt_continue(vbdev, 0);
182 }
183 
184 static void
185 close_core_bdev(struct vbdev_ocf *vbdev)
186 {
187 	remove_base_bdev(&vbdev->core);
188 	vbdev_ocf_mngt_continue(vbdev, 0);
189 }
190 
191 static void
192 remove_core_cmpl(void *priv, int error)
193 {
194 	struct vbdev_ocf *vbdev = priv;
195 
196 	ocf_mngt_cache_unlock(vbdev->ocf_cache);
197 	vbdev_ocf_mngt_continue(vbdev, error);
198 }
199 
200 /* Try to lock cache, then remove core */
201 static void
202 remove_core_poll(struct vbdev_ocf *vbdev)
203 {
204 	int rc;
205 
206 	rc = ocf_mngt_cache_trylock(vbdev->ocf_cache);
207 	if (rc) {
208 		return;
209 	}
210 
211 	vbdev_ocf_mngt_poll(vbdev, NULL);
212 	ocf_mngt_cache_remove_core(vbdev->ocf_core, remove_core_cmpl, vbdev);
213 }
214 
215 /* Detach core base */
216 static void
217 detach_core(struct vbdev_ocf *vbdev)
218 {
219 	if (vbdev->ocf_cache && ocf_cache_is_running(vbdev->ocf_cache)) {
220 		vbdev_ocf_mngt_poll(vbdev, remove_core_poll);
221 	} else {
222 		vbdev_ocf_mngt_continue(vbdev, 0);
223 	}
224 }
225 
226 static void
227 close_cache_bdev(struct vbdev_ocf *vbdev)
228 {
229 	remove_base_bdev(&vbdev->cache);
230 	vbdev_ocf_mngt_continue(vbdev, 0);
231 }
232 
233 /* Detach cache base */
234 static void
235 detach_cache(struct vbdev_ocf *vbdev)
236 {
237 	vbdev->state.stop_status = vbdev->mngt_ctx.status;
238 
239 	/* If some other vbdev references this cache bdev,
240 	 * we detach this only by changing the flag, without actual close */
241 	if (get_other_cache_base(&vbdev->cache)) {
242 		vbdev->cache.attached = false;
243 	}
244 
245 	vbdev_ocf_mngt_continue(vbdev, 0);
246 }
247 
248 static void
249 stop_vbdev_cmpl(ocf_cache_t cache, void *priv, int error)
250 {
251 	struct vbdev_ocf *vbdev = priv;
252 
253 	vbdev_ocf_queue_put(vbdev->cache_ctx->mngt_queue);
254 	ocf_mngt_cache_unlock(cache);
255 	spdk_put_io_channel(vbdev->cache.management_channel);
256 
257 	vbdev_ocf_mngt_continue(vbdev, error);
258 }
259 
260 /* Try to lock cache, then stop it */
261 static void
262 stop_vbdev_poll(struct vbdev_ocf *vbdev)
263 {
264 	if (!ocf_cache_is_running(vbdev->ocf_cache)) {
265 		vbdev_ocf_mngt_continue(vbdev, 0);
266 		return;
267 	}
268 
269 	if (!g_fini_started && get_other_cache_instance(vbdev)) {
270 		SPDK_NOTICELOG("Not stopping cache instance '%s'"
271 			       " because it is referenced by other OCF bdev\n",
272 			       vbdev->cache.name);
273 		vbdev_ocf_mngt_continue(vbdev, 0);
274 		return;
275 	}
276 
277 	if (ocf_mngt_cache_trylock(vbdev->ocf_cache)) {
278 		return;
279 	}
280 
281 	vbdev_ocf_mngt_poll(vbdev, NULL);
282 	ocf_mngt_cache_stop(vbdev->ocf_cache, stop_vbdev_cmpl, vbdev);
283 }
284 
285 /* Stop OCF cache object
286  * vbdev_ocf is not operational after this */
287 static void
288 stop_vbdev(struct vbdev_ocf *vbdev)
289 {
290 	vbdev_ocf_mngt_poll(vbdev, stop_vbdev_poll);
291 }
292 
293 /* Wait for all OCF requests to finish */
294 static void
295 wait_for_requests_poll(struct vbdev_ocf *vbdev)
296 {
297 	if (ocf_cache_has_pending_requests(vbdev->ocf_cache)) {
298 		return;
299 	}
300 
301 	vbdev_ocf_mngt_continue(vbdev, 0);
302 }
303 
304 /* Start waiting for OCF requests to finish */
305 static void
306 wait_for_requests(struct vbdev_ocf *vbdev)
307 {
308 	vbdev_ocf_mngt_poll(vbdev, wait_for_requests_poll);
309 }
310 
311 static void
312 flush_vbdev_cmpl(ocf_cache_t cache, void *priv, int error)
313 {
314 	struct vbdev_ocf *vbdev = priv;
315 
316 	ocf_mngt_cache_unlock(cache);
317 	vbdev_ocf_mngt_continue(vbdev, error);
318 }
319 
320 static void
321 flush_vbdev_poll(struct vbdev_ocf *vbdev)
322 {
323 	if (!ocf_cache_is_running(vbdev->ocf_cache)) {
324 		vbdev_ocf_mngt_continue(vbdev, -EINVAL);
325 		return;
326 	}
327 
328 	if (ocf_mngt_cache_trylock(vbdev->ocf_cache)) {
329 		return;
330 	}
331 
332 	vbdev_ocf_mngt_poll(vbdev, NULL);
333 	ocf_mngt_cache_flush(vbdev->ocf_cache, false, flush_vbdev_cmpl, vbdev);
334 }
335 
336 static void
337 flush_vbdev(struct vbdev_ocf *vbdev)
338 {
339 	vbdev_ocf_mngt_poll(vbdev, flush_vbdev_poll);
340 }
341 
342 /* Procedures called during unregister */
343 vbdev_ocf_mngt_fn unregister_path[] = {
344 	flush_vbdev,
345 	wait_for_requests,
346 	stop_vbdev,
347 	detach_cache,
348 	close_cache_bdev,
349 	detach_core,
350 	close_core_bdev,
351 	unregister_finish,
352 	NULL
353 };
354 
355 /* Start asynchronous management operation using unregister_path */
356 static void
357 unregister_cb(void *opaque)
358 {
359 	struct vbdev_ocf *vbdev = opaque;
360 	int rc;
361 
362 	rc = vbdev_ocf_mngt_start(vbdev, unregister_path, NULL, NULL);
363 	if (rc) {
364 		SPDK_ERRLOG("Unable to unregister OCF bdev: %d\n", rc);
365 		spdk_bdev_destruct_done(&vbdev->exp_bdev, rc);
366 	}
367 }
368 
369 /* Unregister io device with callback to unregister_cb
370  * This function is called during spdk_bdev_unregister */
371 static int
372 vbdev_ocf_destruct(void *opaque)
373 {
374 	struct vbdev_ocf *vbdev = opaque;
375 
376 	if (vbdev->state.doing_finish) {
377 		return -EALREADY;
378 	}
379 	vbdev->state.doing_finish = true;
380 
381 	if (vbdev->state.started) {
382 		spdk_io_device_unregister(vbdev, unregister_cb);
383 		/* Return 1 because unregister is delayed */
384 		return 1;
385 	}
386 
387 	if (vbdev->cache.attached) {
388 		detach_cache(vbdev);
389 		close_cache_bdev(vbdev);
390 	}
391 	if (vbdev->core.attached) {
392 		detach_core(vbdev);
393 		close_core_bdev(vbdev);
394 	}
395 
396 	return 0;
397 }
398 
399 /* Stop OCF cache and unregister SPDK bdev */
400 int
401 vbdev_ocf_delete(struct vbdev_ocf *vbdev, void (*cb)(void *, int), void *cb_arg)
402 {
403 	int rc = 0;
404 
405 	if (vbdev->state.started) {
406 		spdk_bdev_unregister(&vbdev->exp_bdev, cb, cb_arg);
407 	} else {
408 		rc = vbdev_ocf_destruct(vbdev);
409 		if (rc == 0 && cb) {
410 			cb(cb_arg, 0);
411 		}
412 	}
413 
414 	return rc;
415 }
416 
417 /* If vbdev is online, return its object */
418 struct vbdev_ocf *
419 vbdev_ocf_get_by_name(const char *name)
420 {
421 	struct vbdev_ocf *vbdev;
422 
423 	if (name == NULL) {
424 		assert(false);
425 		return NULL;
426 	}
427 
428 	TAILQ_FOREACH(vbdev, &g_ocf_vbdev_head, tailq) {
429 		if (vbdev->name == NULL || vbdev->state.doing_finish) {
430 			continue;
431 		}
432 		if (strcmp(vbdev->name, name) == 0) {
433 			return vbdev;
434 		}
435 	}
436 	return NULL;
437 }
438 
439 /* Return matching base if parent vbdev is online */
440 struct vbdev_ocf_base *
441 vbdev_ocf_get_base_by_name(const char *name)
442 {
443 	struct vbdev_ocf *vbdev;
444 
445 	if (name == NULL) {
446 		assert(false);
447 		return NULL;
448 	}
449 
450 	TAILQ_FOREACH(vbdev, &g_ocf_vbdev_head, tailq) {
451 		if (vbdev->state.doing_finish) {
452 			continue;
453 		}
454 
455 		if (vbdev->cache.name && strcmp(vbdev->cache.name, name) == 0) {
456 			return &vbdev->cache;
457 		}
458 		if (vbdev->core.name && strcmp(vbdev->core.name, name) == 0) {
459 			return &vbdev->core;
460 		}
461 	}
462 	return NULL;
463 }
464 
465 /* Execute fn for each OCF device that is online or waits for base devices */
466 void
467 vbdev_ocf_foreach(vbdev_ocf_foreach_fn fn, void *ctx)
468 {
469 	struct vbdev_ocf *vbdev;
470 
471 	assert(fn != NULL);
472 
473 	TAILQ_FOREACH(vbdev, &g_ocf_vbdev_head, tailq) {
474 		if (!vbdev->state.doing_finish) {
475 			fn(vbdev, ctx);
476 		}
477 	}
478 }
479 
480 /* Called from OCF when SPDK_IO is completed */
481 static void
482 vbdev_ocf_io_submit_cb(struct ocf_io *io, int error)
483 {
484 	struct spdk_bdev_io *bdev_io = io->priv1;
485 
486 	if (error == 0) {
487 		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
488 	} else if (error == -ENOMEM) {
489 		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_NOMEM);
490 	} else {
491 		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
492 	}
493 
494 	ocf_io_put(io);
495 }
496 
497 /* Configure io parameters and send it to OCF */
498 static int
499 io_submit_to_ocf(struct spdk_bdev_io *bdev_io, struct ocf_io *io)
500 {
501 	int dir;
502 	uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
503 	uint64_t offset = bdev_io->u.bdev.offset_blocks * bdev_io->bdev->blocklen;
504 
505 	switch (bdev_io->type) {
506 	case SPDK_BDEV_IO_TYPE_WRITE:
507 	case SPDK_BDEV_IO_TYPE_READ:
508 		dir = OCF_READ;
509 		if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
510 			dir = OCF_WRITE;
511 		}
512 		ocf_io_configure(io, offset, len, dir, 0, 0);
513 		ocf_core_submit_io(io);
514 		return 0;
515 	case SPDK_BDEV_IO_TYPE_FLUSH:
516 		ocf_io_configure(io, offset, len, OCF_WRITE, 0, OCF_WRITE_FLUSH);
517 		ocf_core_submit_flush(io);
518 		return 0;
519 	case SPDK_BDEV_IO_TYPE_UNMAP:
520 		ocf_io_configure(io, offset, len, 0, 0, 0);
521 		ocf_core_submit_discard(io);
522 		return 0;
523 	case SPDK_BDEV_IO_TYPE_RESET:
524 	case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
525 	default:
526 		SPDK_ERRLOG("Unsupported IO type: %d\n", bdev_io->type);
527 		return -EINVAL;
528 	}
529 }
530 
531 /* Submit SPDK-IO to OCF */
532 static void
533 io_handle(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io)
534 {
535 	struct vbdev_ocf *vbdev = bdev_io->bdev->ctxt;
536 	struct ocf_io *io = NULL;
537 	struct bdev_ocf_data *data = NULL;
538 	struct vbdev_ocf_qcxt *qctx = spdk_io_channel_get_ctx(ch);
539 	int err;
540 
541 	io = ocf_core_new_io(vbdev->ocf_core);
542 	if (!io) {
543 		err = -ENOMEM;
544 		goto fail;
545 	}
546 
547 	ocf_io_set_queue(io, qctx->queue);
548 
549 	data = vbdev_ocf_data_from_spdk_io(bdev_io);
550 	if (!data) {
551 		err = -ENOMEM;
552 		goto fail;
553 	}
554 
555 	err = ocf_io_set_data(io, data, 0);
556 	if (err) {
557 		goto fail;
558 	}
559 
560 	ocf_io_set_cmpl(io, bdev_io, NULL, vbdev_ocf_io_submit_cb);
561 
562 	err = io_submit_to_ocf(bdev_io, io);
563 	if (err) {
564 		goto fail;
565 	}
566 
567 	return;
568 
569 fail:
570 	if (io) {
571 		ocf_io_put(io);
572 	}
573 
574 	if (err == -ENOMEM) {
575 		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_NOMEM);
576 	} else {
577 		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
578 	}
579 }
580 
581 static void
582 vbdev_ocf_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
583 		     bool success)
584 {
585 	if (!success) {
586 		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
587 		return;
588 	}
589 
590 	io_handle(ch, bdev_io);
591 }
592 
593 /* Called from bdev layer when an io to Cache vbdev is submitted */
594 static void
595 vbdev_ocf_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io)
596 {
597 	switch (bdev_io->type) {
598 	case SPDK_BDEV_IO_TYPE_READ:
599 		/* User does not have to allocate io vectors for the request,
600 		 * so in case they are not allocated, we allocate them here */
601 		spdk_bdev_io_get_buf(bdev_io, vbdev_ocf_get_buf_cb,
602 				     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
603 		break;
604 	case SPDK_BDEV_IO_TYPE_WRITE:
605 	case SPDK_BDEV_IO_TYPE_FLUSH:
606 	case SPDK_BDEV_IO_TYPE_UNMAP:
607 		io_handle(ch, bdev_io);
608 		break;
609 	case SPDK_BDEV_IO_TYPE_RESET:
610 	case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
611 	default:
612 		SPDK_ERRLOG("Unknown I/O type %d\n", bdev_io->type);
613 		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
614 		break;
615 	}
616 }
617 
618 /* Called from bdev layer */
619 static bool
620 vbdev_ocf_io_type_supported(void *opaque, enum spdk_bdev_io_type io_type)
621 {
622 	struct vbdev_ocf *vbdev = opaque;
623 
624 	switch (io_type) {
625 	case SPDK_BDEV_IO_TYPE_READ:
626 	case SPDK_BDEV_IO_TYPE_WRITE:
627 	case SPDK_BDEV_IO_TYPE_FLUSH:
628 	case SPDK_BDEV_IO_TYPE_UNMAP:
629 		return spdk_bdev_io_type_supported(vbdev->core.bdev, io_type);
630 	case SPDK_BDEV_IO_TYPE_RESET:
631 	case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
632 	default:
633 		return false;
634 	}
635 }
636 
637 /* Called from bdev layer */
638 static struct spdk_io_channel *
639 vbdev_ocf_get_io_channel(void *opaque)
640 {
641 	struct vbdev_ocf *bdev = opaque;
642 
643 	return spdk_get_io_channel(bdev);
644 }
645 
646 static int
647 vbdev_ocf_dump_info_json(void *opaque, struct spdk_json_write_ctx *w)
648 {
649 	struct vbdev_ocf *vbdev = opaque;
650 
651 	spdk_json_write_named_string(w, "cache_device", vbdev->cache.name);
652 	spdk_json_write_named_string(w, "core_device", vbdev->core.name);
653 
654 	spdk_json_write_named_string(w, "mode",
655 				     ocf_get_cache_modename(ocf_cache_get_mode(vbdev->ocf_cache)));
656 	spdk_json_write_named_uint32(w, "cache_line_size",
657 				     ocf_cache_get_line_size(vbdev->ocf_cache));
658 	spdk_json_write_named_bool(w, "metadata_volatile",
659 				   vbdev->cfg.cache.metadata_volatile);
660 
661 	return 0;
662 }
663 
664 static void
665 vbdev_ocf_write_json_config(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
666 {
667 	struct vbdev_ocf *vbdev = bdev->ctxt;
668 
669 	spdk_json_write_object_begin(w);
670 
671 	spdk_json_write_named_string(w, "method", "construct_ocf_bdev");
672 
673 	spdk_json_write_named_object_begin(w, "params");
674 	spdk_json_write_named_string(w, "name", vbdev->name);
675 	spdk_json_write_named_string(w, "mode",
676 				     ocf_get_cache_modename(ocf_cache_get_mode(vbdev->ocf_cache)));
677 	spdk_json_write_named_string(w, "cache_bdev_name", vbdev->cache.name);
678 	spdk_json_write_named_string(w, "core_bdev_name", vbdev->core.name);
679 	spdk_json_write_object_end(w);
680 
681 	spdk_json_write_object_end(w);
682 }
683 
684 /* Cache vbdev function table
685  * Used by bdev layer */
686 static struct spdk_bdev_fn_table cache_dev_fn_table = {
687 	.destruct = vbdev_ocf_destruct,
688 	.io_type_supported = vbdev_ocf_io_type_supported,
689 	.submit_request	= vbdev_ocf_submit_request,
690 	.get_io_channel	= vbdev_ocf_get_io_channel,
691 	.write_config_json = vbdev_ocf_write_json_config,
692 	.dump_info_json = vbdev_ocf_dump_info_json,
693 };
694 
695 /* Poller function for the OCF queue
696  * We execute OCF requests here synchronously */
697 static int
698 queue_poll(void *opaque)
699 {
700 	struct vbdev_ocf_qcxt *qctx = opaque;
701 	uint32_t iono = ocf_queue_pending_io(qctx->queue);
702 	int i, max = spdk_min(32, iono);
703 
704 	for (i = 0; i < max; i++) {
705 		ocf_queue_run_single(qctx->queue);
706 	}
707 
708 	if (iono > 0) {
709 		return 1;
710 	} else {
711 		return 0;
712 	}
713 }
714 
715 /* Called during ocf_submit_io, ocf_purge*
716  * and any other requests that need to submit io */
717 static void
718 vbdev_ocf_ctx_queue_kick(ocf_queue_t q)
719 {
720 }
721 
722 /* OCF queue deinitialization
723  * Called at ocf_cache_stop */
724 static void
725 vbdev_ocf_ctx_queue_stop(ocf_queue_t q)
726 {
727 	struct vbdev_ocf_qcxt *qctx = ocf_queue_get_priv(q);
728 
729 	if (qctx) {
730 		spdk_put_io_channel(qctx->cache_ch);
731 		spdk_put_io_channel(qctx->core_ch);
732 		spdk_poller_unregister(&qctx->poller);
733 	}
734 }
735 
736 /* Queue ops is an interface for running queue thread
737  * stop() operation in called just before queue gets destroyed */
738 const struct ocf_queue_ops queue_ops = {
739 	.kick_sync = vbdev_ocf_ctx_queue_kick,
740 	.kick = vbdev_ocf_ctx_queue_kick,
741 	.stop = vbdev_ocf_ctx_queue_stop,
742 };
743 
744 /* Called on cache vbdev creation at every thread
745  * We allocate OCF queues here and SPDK poller for it */
746 static int
747 io_device_create_cb(void *io_device, void *ctx_buf)
748 {
749 	struct vbdev_ocf *vbdev = io_device;
750 	struct vbdev_ocf_qcxt *qctx = ctx_buf;
751 	int rc;
752 
753 	rc = vbdev_ocf_queue_create(vbdev->ocf_cache, &qctx->queue, &queue_ops);
754 	if (rc) {
755 		return rc;
756 	}
757 
758 	ocf_queue_set_priv(qctx->queue, qctx);
759 
760 	qctx->vbdev      = vbdev;
761 	qctx->cache_ch   = spdk_bdev_get_io_channel(vbdev->cache.desc);
762 	qctx->core_ch    = spdk_bdev_get_io_channel(vbdev->core.desc);
763 	qctx->poller     = spdk_poller_register(queue_poll, qctx, 0);
764 
765 	return rc;
766 }
767 
768 /* Called per thread
769  * Put OCF queue and relaunch poller with new context to finish pending requests */
770 static void
771 io_device_destroy_cb(void *io_device, void *ctx_buf)
772 {
773 	/* Making a copy of context to use it after io channel will be destroyed */
774 	struct vbdev_ocf_qcxt *copy = malloc(sizeof(*copy));
775 	struct vbdev_ocf_qcxt *qctx = ctx_buf;
776 
777 	if (copy) {
778 		ocf_queue_set_priv(qctx->queue, copy);
779 		memcpy(copy, qctx, sizeof(*copy));
780 		spdk_poller_unregister(&qctx->poller);
781 		copy->poller = spdk_poller_register(queue_poll, copy, 0);
782 	} else {
783 		SPDK_ERRLOG("Unable to stop OCF queue properly: %s\n",
784 			    spdk_strerror(ENOMEM));
785 	}
786 
787 	vbdev_ocf_queue_put(qctx->queue);
788 }
789 
790 /* OCF management queue deinitialization */
791 static void
792 vbdev_ocf_ctx_mngt_queue_stop(ocf_queue_t q)
793 {
794 	struct spdk_poller *poller = ocf_queue_get_priv(q);
795 
796 	if (poller) {
797 		spdk_poller_unregister(&poller);
798 	}
799 }
800 
801 static int
802 mngt_queue_poll(void *opaque)
803 {
804 	ocf_queue_t q = opaque;
805 	uint32_t iono = ocf_queue_pending_io(q);
806 	int i, max = spdk_min(32, iono);
807 
808 	for (i = 0; i < max; i++) {
809 		ocf_queue_run_single(q);
810 	}
811 
812 	if (iono > 0) {
813 		return 1;
814 	} else {
815 		return 0;
816 	}
817 }
818 
819 static void
820 vbdev_ocf_ctx_mngt_queue_kick(ocf_queue_t q)
821 {
822 }
823 
824 /* Queue ops is an interface for running queue thread
825  * stop() operation in called just before queue gets destroyed */
826 const struct ocf_queue_ops mngt_queue_ops = {
827 	.kick_sync = NULL,
828 	.kick = vbdev_ocf_ctx_mngt_queue_kick,
829 	.stop = vbdev_ocf_ctx_mngt_queue_stop,
830 };
831 
832 /* Create exported spdk object */
833 static void
834 finish_register(struct vbdev_ocf *vbdev)
835 {
836 	int result;
837 
838 	vbdev->cache.management_channel = vbdev->cache_ctx->management_channel;
839 
840 	/* Copy properties of the base bdev */
841 	vbdev->exp_bdev.blocklen = vbdev->core.bdev->blocklen;
842 	vbdev->exp_bdev.write_cache = vbdev->core.bdev->write_cache;
843 	vbdev->exp_bdev.required_alignment = vbdev->core.bdev->required_alignment;
844 
845 	vbdev->exp_bdev.name = vbdev->name;
846 	vbdev->exp_bdev.product_name = "SPDK OCF";
847 
848 	vbdev->exp_bdev.blockcnt = vbdev->core.bdev->blockcnt;
849 	vbdev->exp_bdev.ctxt = vbdev;
850 	vbdev->exp_bdev.fn_table = &cache_dev_fn_table;
851 	vbdev->exp_bdev.module = &ocf_if;
852 
853 	/* Finally register vbdev in SPDK */
854 	spdk_io_device_register(vbdev, io_device_create_cb, io_device_destroy_cb,
855 				sizeof(struct vbdev_ocf_qcxt), vbdev->name);
856 	result = spdk_bdev_register(&vbdev->exp_bdev);
857 	if (result) {
858 		SPDK_ERRLOG("Could not register exposed bdev\n");
859 	} else {
860 		vbdev->state.started = true;
861 	}
862 
863 	vbdev_ocf_mngt_continue(vbdev, result);
864 }
865 
866 static void
867 add_core_cmpl(ocf_cache_t cache, ocf_core_t core, void *priv, int error)
868 {
869 	struct vbdev_ocf *vbdev = priv;
870 
871 	ocf_mngt_cache_unlock(cache);
872 
873 	if (error) {
874 		SPDK_ERRLOG("Failed to add core device to cache instance\n");
875 	} else {
876 		vbdev->ocf_core = core;
877 		vbdev->core.id  = ocf_core_get_id(core);
878 	}
879 
880 	vbdev->core.management_channel = spdk_bdev_get_io_channel(vbdev->core.desc);
881 	vbdev_ocf_mngt_continue(vbdev, error);
882 }
883 
884 /* Try to lock cache, then add core */
885 static void
886 add_core_poll(struct vbdev_ocf *vbdev)
887 {
888 	if (ocf_mngt_cache_trylock(vbdev->ocf_cache)) {
889 		return;
890 	}
891 
892 	vbdev_ocf_mngt_poll(vbdev, NULL);
893 	ocf_mngt_cache_add_core(vbdev->ocf_cache, &vbdev->cfg.core, add_core_cmpl, vbdev);
894 }
895 
896 /* Add core for existing OCF cache instance */
897 static void
898 add_core(struct vbdev_ocf *vbdev)
899 {
900 	vbdev_ocf_mngt_poll(vbdev, add_core_poll);
901 }
902 
903 static void
904 start_cache_cmpl(ocf_cache_t cache, void *priv, int error)
905 {
906 	struct vbdev_ocf *vbdev = priv;
907 
908 	ocf_mngt_cache_unlock(cache);
909 
910 	vbdev_ocf_mngt_continue(vbdev, error);
911 }
912 
913 static int
914 create_management_queue(struct vbdev_ocf *vbdev)
915 {
916 	struct spdk_poller *mngt_poller;
917 	int rc;
918 
919 	rc = vbdev_ocf_queue_create(vbdev->ocf_cache, &vbdev->cache_ctx->mngt_queue, &mngt_queue_ops);
920 	if (rc) {
921 		SPDK_ERRLOG("Unable to create mngt_queue: %d\n", rc);
922 		return rc;
923 	}
924 
925 	mngt_poller = spdk_poller_register(mngt_queue_poll, vbdev->cache_ctx->mngt_queue, 100);
926 	if (mngt_poller == NULL) {
927 		SPDK_ERRLOG("Unable to initiate mngt request: %s", spdk_strerror(ENOMEM));
928 		return -ENOMEM;
929 	}
930 
931 	ocf_queue_set_priv(vbdev->cache_ctx->mngt_queue, mngt_poller);
932 	ocf_mngt_cache_set_mngt_queue(vbdev->ocf_cache, vbdev->cache_ctx->mngt_queue);
933 
934 	return 0;
935 }
936 
937 /* Start OCF cache, attach caching device */
938 static void
939 start_cache(struct vbdev_ocf *vbdev)
940 {
941 	ocf_cache_t existing;
942 	int rc;
943 
944 	if (vbdev->ocf_cache) {
945 		vbdev->mngt_ctx.status = -EALREADY;
946 		vbdev_ocf_mngt_stop(vbdev);
947 		return;
948 	}
949 
950 	existing = get_other_cache_instance(vbdev);
951 	if (existing) {
952 		SPDK_NOTICELOG("OCF bdev %s connects to existing cache device %s\n",
953 			       vbdev->name, vbdev->cache.name);
954 		vbdev->ocf_cache = existing;
955 		vbdev->cache.id = ocf_cache_get_id(existing);
956 		vbdev->cache_ctx = ocf_cache_get_priv(existing);
957 		vbdev_ocf_cache_ctx_get(vbdev->cache_ctx);
958 		vbdev_ocf_mngt_continue(vbdev, 0);
959 		return;
960 	}
961 
962 	vbdev->cache_ctx = calloc(1, sizeof(struct vbdev_ocf_cache_ctx));
963 	if (vbdev->cache_ctx == NULL) {
964 		vbdev->mngt_ctx.status = -ENOMEM;
965 		vbdev_ocf_mngt_stop(vbdev);
966 		return;
967 	}
968 
969 	vbdev_ocf_cache_ctx_get(vbdev->cache_ctx);
970 	pthread_mutex_init(&vbdev->cache_ctx->lock, NULL);
971 
972 	rc = ocf_mngt_cache_start(vbdev_ocf_ctx, &vbdev->ocf_cache, &vbdev->cfg.cache);
973 	if (rc) {
974 		vbdev_ocf_cache_ctx_put(vbdev->cache_ctx);
975 		vbdev->mngt_ctx.status = rc;
976 		vbdev_ocf_mngt_stop(vbdev);
977 		return;
978 	}
979 
980 	vbdev->cache.id = ocf_cache_get_id(vbdev->ocf_cache);
981 	ocf_cache_set_priv(vbdev->ocf_cache, vbdev->cache_ctx);
982 
983 	rc = create_management_queue(vbdev);
984 	if (rc) {
985 		SPDK_ERRLOG("Unable to create mngt_queue: %d\n", rc);
986 		vbdev_ocf_cache_ctx_put(vbdev->cache_ctx);
987 		vbdev->mngt_ctx.status = rc;
988 		vbdev_ocf_mngt_stop(vbdev);
989 		return;
990 	}
991 
992 	vbdev->cache_ctx->management_channel = spdk_bdev_get_io_channel(vbdev->cache.desc);
993 	vbdev->cache.management_channel = vbdev->cache_ctx->management_channel;
994 
995 	if (vbdev->cfg.loadq) {
996 		ocf_mngt_cache_load(vbdev->ocf_cache, &vbdev->cfg.device, start_cache_cmpl, vbdev);
997 	} else {
998 		ocf_mngt_cache_attach(vbdev->ocf_cache, &vbdev->cfg.device, start_cache_cmpl, vbdev);
999 	}
1000 }
1001 
1002 /* Procedures called during register operation */
1003 vbdev_ocf_mngt_fn register_path[] = {
1004 	start_cache,
1005 	add_core,
1006 	finish_register,
1007 	NULL
1008 };
1009 
1010 /* Start cache instance and register OCF bdev */
1011 static void
1012 register_vbdev(struct vbdev_ocf *vbdev, vbdev_ocf_mngt_callback cb, void *cb_arg)
1013 {
1014 	int rc;
1015 
1016 	if (!(vbdev->core.attached && vbdev->cache.attached) || vbdev->state.started) {
1017 		cb(-EPERM, vbdev, cb_arg);
1018 		return;
1019 	}
1020 
1021 	rc = vbdev_ocf_mngt_start(vbdev, register_path, cb, cb_arg);
1022 	if (rc) {
1023 		cb(rc, vbdev, cb_arg);
1024 	}
1025 }
1026 
1027 /* Init OCF configuration options
1028  * for core and cache devices */
1029 static void
1030 init_vbdev_config(struct vbdev_ocf *vbdev)
1031 {
1032 	struct vbdev_ocf_config *cfg = &vbdev->cfg;
1033 
1034 	/* Id 0 means OCF decides the id */
1035 	cfg->cache.id = 0;
1036 	cfg->cache.name = vbdev->name;
1037 
1038 	/* TODO [metadata]: make configurable with persistent
1039 	 * metadata support */
1040 	cfg->cache.metadata_volatile = false;
1041 
1042 	/* TODO [cache line size]: make cache line size configurable
1043 	 * Using standard 4KiB for now */
1044 	cfg->cache.cache_line_size = ocf_cache_line_size_4;
1045 
1046 	/* This are suggested values that
1047 	 * should be sufficient for most use cases */
1048 	cfg->cache.backfill.max_queue_size = 65536;
1049 	cfg->cache.backfill.queue_unblock_size = 60000;
1050 
1051 	/* TODO [cache line size] */
1052 	cfg->device.cache_line_size = ocf_cache_line_size_4;
1053 	cfg->device.force = true;
1054 	cfg->device.min_free_ram = 0;
1055 	cfg->device.perform_test = false;
1056 	cfg->device.discard_on_start = false;
1057 
1058 	vbdev->cfg.cache.locked = true;
1059 
1060 	cfg->core.volume_type = SPDK_OBJECT;
1061 	cfg->device.volume_type = SPDK_OBJECT;
1062 	cfg->core.core_id = OCF_CORE_MAX;
1063 
1064 	if (vbdev->cfg.loadq) {
1065 		/* When doing cache_load(), we need to set try_add to true,
1066 		 * otherwise OCF will interpret this core as new
1067 		 * instead of the inactive one */
1068 		vbdev->cfg.core.try_add = true;
1069 	}
1070 
1071 	/* Serialize bdev names in OCF UUID to interpret on future loads
1072 	 * Core UUID is pair of (core bdev name, cache bdev name)
1073 	 * Cache UUID is cache bdev name */
1074 	cfg->device.uuid.size = strlen(vbdev->cache.name) + 1;
1075 	cfg->device.uuid.data = vbdev->cache.name;
1076 
1077 	snprintf(vbdev->uuid, VBDEV_OCF_MD_MAX_LEN, "%s %s",
1078 		 vbdev->core.name, vbdev->name);
1079 	cfg->core.uuid.size = strlen(vbdev->uuid) + 1;
1080 	cfg->core.uuid.data = vbdev->uuid;
1081 	vbdev->uuid[strlen(vbdev->core.name)] = 0;
1082 }
1083 
1084 /* Allocate vbdev structure object and add it to the global list */
1085 static int
1086 init_vbdev(const char *vbdev_name,
1087 	   const char *cache_mode_name,
1088 	   const char *cache_name,
1089 	   const char *core_name,
1090 	   bool loadq)
1091 {
1092 	struct vbdev_ocf *vbdev;
1093 	int rc = 0;
1094 
1095 	if (spdk_bdev_get_by_name(vbdev_name) || vbdev_ocf_get_by_name(vbdev_name)) {
1096 		SPDK_ERRLOG("Device with name '%s' already exists\n", vbdev_name);
1097 		return -EPERM;
1098 	}
1099 
1100 	vbdev = calloc(1, sizeof(*vbdev));
1101 	if (!vbdev) {
1102 		goto error_mem;
1103 	}
1104 
1105 	vbdev->cache.parent = vbdev;
1106 	vbdev->core.parent = vbdev;
1107 	vbdev->cache.is_cache = true;
1108 	vbdev->core.is_cache = false;
1109 
1110 	if (cache_mode_name) {
1111 		vbdev->cfg.cache.cache_mode
1112 			= ocf_get_cache_mode(cache_mode_name);
1113 	} else if (!loadq) { /* In load path it is OK to pass NULL as cache mode */
1114 		SPDK_ERRLOG("No cache mode specified\n");
1115 		rc = -EINVAL;
1116 		goto error_free;
1117 	}
1118 	if (vbdev->cfg.cache.cache_mode < 0) {
1119 		SPDK_ERRLOG("Incorrect cache mode '%s'\n", cache_mode_name);
1120 		rc = -EINVAL;
1121 		goto error_free;
1122 	}
1123 
1124 	vbdev->name = strdup(vbdev_name);
1125 	if (!vbdev->name) {
1126 		goto error_mem;
1127 	}
1128 
1129 	vbdev->cache.name = strdup(cache_name);
1130 	if (!vbdev->cache.name) {
1131 		goto error_mem;
1132 	}
1133 
1134 	vbdev->core.name = strdup(core_name);
1135 	if (!vbdev->core.name) {
1136 		goto error_mem;
1137 	}
1138 
1139 	vbdev->cfg.loadq = loadq;
1140 	init_vbdev_config(vbdev);
1141 	TAILQ_INSERT_TAIL(&g_ocf_vbdev_head, vbdev, tailq);
1142 	return rc;
1143 
1144 error_mem:
1145 	rc = -ENOMEM;
1146 error_free:
1147 	free_vbdev(vbdev);
1148 	return rc;
1149 }
1150 
1151 /* Read configuration file at the start of SPDK application
1152  * This adds vbdevs to global list if some mentioned in config */
1153 static int
1154 vbdev_ocf_init(void)
1155 {
1156 	const char *vbdev_name, *modename, *cache_name, *core_name;
1157 	struct spdk_conf_section *sp;
1158 	int status;
1159 
1160 	status = vbdev_ocf_ctx_init();
1161 	if (status) {
1162 		SPDK_ERRLOG("OCF ctx initialization failed with=%d\n", status);
1163 		return status;
1164 	}
1165 
1166 	status = vbdev_ocf_volume_init();
1167 	if (status) {
1168 		vbdev_ocf_ctx_cleanup();
1169 		SPDK_ERRLOG("OCF volume initialization failed with=%d\n", status);
1170 		return status;
1171 	}
1172 
1173 	sp = spdk_conf_find_section(NULL, "OCF");
1174 	if (sp == NULL) {
1175 		return 0;
1176 	}
1177 
1178 	for (int i = 0; ; i++) {
1179 		if (!spdk_conf_section_get_nval(sp, "OCF", i)) {
1180 			break;
1181 		}
1182 
1183 		vbdev_name = spdk_conf_section_get_nmval(sp, "OCF", i, 0);
1184 		if (!vbdev_name) {
1185 			SPDK_ERRLOG("No vbdev name specified\n");
1186 			continue;
1187 		}
1188 
1189 		modename = spdk_conf_section_get_nmval(sp, "OCF", i, 1);
1190 		if (!modename) {
1191 			SPDK_ERRLOG("No modename specified for OCF vbdev '%s'\n", vbdev_name);
1192 			continue;
1193 		}
1194 
1195 		cache_name = spdk_conf_section_get_nmval(sp, "OCF", i, 2);
1196 		if (!cache_name) {
1197 			SPDK_ERRLOG("No cache device specified for OCF vbdev '%s'\n", vbdev_name);
1198 			continue;
1199 		}
1200 
1201 		core_name = spdk_conf_section_get_nmval(sp, "OCF", i, 3);
1202 		if (!core_name) {
1203 			SPDK_ERRLOG("No core devices specified for OCF vbdev '%s'\n", vbdev_name);
1204 			continue;
1205 		}
1206 
1207 		status = init_vbdev(vbdev_name, modename, cache_name, core_name, false);
1208 		if (status) {
1209 			SPDK_ERRLOG("Config initialization failed with code: %d\n", status);
1210 		}
1211 	}
1212 
1213 	return status;
1214 }
1215 
1216 /* Called after application shutdown started
1217  * Release memory of allocated structures here */
1218 static void
1219 vbdev_ocf_module_fini(void)
1220 {
1221 	struct vbdev_ocf *vbdev;
1222 
1223 	while ((vbdev = TAILQ_FIRST(&g_ocf_vbdev_head))) {
1224 		TAILQ_REMOVE(&g_ocf_vbdev_head, vbdev, tailq);
1225 		free_vbdev(vbdev);
1226 	}
1227 
1228 	vbdev_ocf_volume_cleanup();
1229 	vbdev_ocf_ctx_cleanup();
1230 }
1231 
1232 /* When base device gets unpluged this is called
1233  * We will unregister cache vbdev here
1234  * When cache device is removed, we delete every OCF bdev that used it */
1235 static void
1236 hotremove_cb(void *ctx)
1237 {
1238 	struct vbdev_ocf_base *base = ctx;
1239 	struct vbdev_ocf *vbdev;
1240 
1241 	if (!base->is_cache) {
1242 		if (base->parent->state.doing_finish) {
1243 			return;
1244 		}
1245 
1246 		SPDK_NOTICELOG("Deinitializing '%s' because its core device '%s' was removed\n",
1247 			       base->parent->name, base->name);
1248 		vbdev_ocf_delete(base->parent, NULL, NULL);
1249 		return;
1250 	}
1251 
1252 	TAILQ_FOREACH(vbdev, &g_ocf_vbdev_head, tailq) {
1253 		if (vbdev->state.doing_finish) {
1254 			continue;
1255 		}
1256 		if (strcmp(base->name, vbdev->cache.name) == 0) {
1257 			SPDK_NOTICELOG("Deinitializing '%s' because"
1258 				       " its cache device '%s' was removed\n",
1259 				       vbdev->name, base->name);
1260 			vbdev_ocf_delete(vbdev, NULL, NULL);
1261 		}
1262 	}
1263 }
1264 
1265 /* Open base SPDK bdev and claim it */
1266 static int
1267 attach_base(struct vbdev_ocf_base *base)
1268 {
1269 	int status;
1270 
1271 	if (base->attached) {
1272 		return -EALREADY;
1273 	}
1274 
1275 	/* If base cache bdev was already opened by other vbdev,
1276 	 * we just copy its descriptor here */
1277 	if (base->is_cache) {
1278 		struct vbdev_ocf_base *existing = get_other_cache_base(base);
1279 		if (existing) {
1280 			base->desc = existing->desc;
1281 			base->attached = true;
1282 			return 0;
1283 		}
1284 	}
1285 
1286 	status = spdk_bdev_open(base->bdev, true, hotremove_cb, base, &base->desc);
1287 	if (status) {
1288 		SPDK_ERRLOG("Unable to open device '%s' for writing\n", base->name);
1289 		return status;
1290 	}
1291 
1292 	status = spdk_bdev_module_claim_bdev(base->bdev, base->desc,
1293 					     &ocf_if);
1294 	if (status) {
1295 		SPDK_ERRLOG("Unable to claim device '%s'\n", base->name);
1296 		spdk_bdev_close(base->desc);
1297 		return status;
1298 	}
1299 
1300 	base->attached = true;
1301 	return status;
1302 }
1303 
1304 /* Attach base bdevs */
1305 static int
1306 attach_base_bdevs(struct vbdev_ocf *vbdev,
1307 		  struct spdk_bdev *cache_bdev,
1308 		  struct spdk_bdev *core_bdev)
1309 {
1310 	int rc = 0;
1311 
1312 	if (cache_bdev) {
1313 		vbdev->cache.bdev = cache_bdev;
1314 		rc |= attach_base(&vbdev->cache);
1315 	}
1316 
1317 	if (core_bdev) {
1318 		vbdev->core.bdev = core_bdev;
1319 		rc |= attach_base(&vbdev->core);
1320 	}
1321 
1322 	return rc;
1323 }
1324 
1325 /* Init and then start vbdev if all base devices are present */
1326 void
1327 vbdev_ocf_construct(const char *vbdev_name,
1328 		    const char *cache_mode_name,
1329 		    const char *cache_name,
1330 		    const char *core_name,
1331 		    bool loadq,
1332 		    void (*cb)(int, struct vbdev_ocf *, void *),
1333 		    void *cb_arg)
1334 {
1335 	int rc;
1336 	struct spdk_bdev *cache_bdev = spdk_bdev_get_by_name(cache_name);
1337 	struct spdk_bdev *core_bdev = spdk_bdev_get_by_name(core_name);
1338 	struct vbdev_ocf *vbdev;
1339 
1340 	rc = init_vbdev(vbdev_name, cache_mode_name, cache_name, core_name, loadq);
1341 	if (rc) {
1342 		cb(rc, NULL, cb_arg);
1343 		return;
1344 	}
1345 
1346 	vbdev = vbdev_ocf_get_by_name(vbdev_name);
1347 	if (vbdev == NULL) {
1348 		cb(-ENODEV, NULL, cb_arg);
1349 		return;
1350 	}
1351 
1352 	if (cache_bdev == NULL) {
1353 		SPDK_NOTICELOG("OCF bdev '%s' is waiting for cache device '%s' to connect\n",
1354 			       vbdev->name, cache_name);
1355 	}
1356 	if (core_bdev == NULL) {
1357 		SPDK_NOTICELOG("OCF bdev '%s' is waiting for core device '%s' to connect\n",
1358 			       vbdev->name, core_name);
1359 	}
1360 
1361 	rc = attach_base_bdevs(vbdev, cache_bdev, core_bdev);
1362 	if (rc) {
1363 		cb(rc, vbdev, cb_arg);
1364 		return;
1365 	}
1366 
1367 	if (core_bdev && cache_bdev) {
1368 		register_vbdev(vbdev, cb, cb_arg);
1369 	} else {
1370 		cb(0, vbdev, cb_arg);
1371 	}
1372 }
1373 
1374 /* This called if new device is created in SPDK application
1375  * If that device named as one of base bdevs of OCF vbdev,
1376  * claim and open them */
1377 static void
1378 vbdev_ocf_examine(struct spdk_bdev *bdev)
1379 {
1380 	const char *bdev_name = spdk_bdev_get_name(bdev);
1381 	struct vbdev_ocf *vbdev;
1382 
1383 	TAILQ_FOREACH(vbdev, &g_ocf_vbdev_head, tailq) {
1384 		if (vbdev->state.doing_finish) {
1385 			continue;
1386 		}
1387 
1388 		if (!strcmp(bdev_name, vbdev->cache.name)) {
1389 			attach_base_bdevs(vbdev, bdev, NULL);
1390 			continue;
1391 		}
1392 		if (!strcmp(bdev_name, vbdev->core.name)) {
1393 			attach_base_bdevs(vbdev, NULL, bdev);
1394 			break;
1395 		}
1396 	}
1397 	spdk_bdev_module_examine_done(&ocf_if);
1398 }
1399 
1400 struct metadata_probe_ctx {
1401 	struct vbdev_ocf_base base;
1402 	ocf_volume_t volume;
1403 
1404 	struct ocf_volume_uuid *core_uuids;
1405 	unsigned int uuid_count;
1406 
1407 	int result;
1408 	int refcnt;
1409 };
1410 
1411 static void
1412 examine_ctx_put(struct metadata_probe_ctx *ctx)
1413 {
1414 	unsigned int i;
1415 
1416 	ctx->refcnt--;
1417 	if (ctx->refcnt > 0) {
1418 		return;
1419 	}
1420 
1421 	if (ctx->result) {
1422 		SPDK_ERRLOG("OCF metadata probe for bdev '%s' failed with %d\n",
1423 			    spdk_bdev_get_name(ctx->base.bdev), ctx->result);
1424 	}
1425 
1426 	if (ctx->base.desc) {
1427 		spdk_bdev_close(ctx->base.desc);
1428 	}
1429 
1430 	if (ctx->volume) {
1431 		ocf_volume_destroy(ctx->volume);
1432 	}
1433 
1434 	if (ctx->core_uuids) {
1435 		for (i = 0; i < ctx->uuid_count; i++) {
1436 			free(ctx->core_uuids[i].data);
1437 		}
1438 	}
1439 	free(ctx->core_uuids);
1440 
1441 	examine_done(ctx->result, NULL, ctx->base.bdev);
1442 	free(ctx);
1443 }
1444 
1445 static void
1446 metadata_probe_construct_cb(int rc, struct vbdev_ocf *vbdev, void *vctx)
1447 {
1448 	struct metadata_probe_ctx *ctx = vctx;
1449 
1450 	examine_ctx_put(ctx);
1451 }
1452 
1453 /* This is second callback for ocf_metadata_probe_cores()
1454  * Here we create vbdev configurations based on UUIDs */
1455 static void
1456 metadata_probe_cores_construct(void *priv, int error, unsigned int num_cores)
1457 {
1458 	struct metadata_probe_ctx *ctx = priv;
1459 	const char *vbdev_name;
1460 	const char *core_name;
1461 	unsigned int i;
1462 
1463 	if (error) {
1464 		ctx->result = error;
1465 		examine_ctx_put(ctx);
1466 		return;
1467 	}
1468 
1469 	for (i = 0; i < num_cores; i++) {
1470 		core_name = ocf_uuid_to_str(&ctx->core_uuids[i]);
1471 		vbdev_name = core_name + strlen(core_name) + 1;
1472 		ctx->refcnt++;
1473 		vbdev_ocf_construct(vbdev_name, NULL, ctx->base.bdev->name, core_name, true,
1474 				    metadata_probe_construct_cb, ctx);
1475 	}
1476 
1477 	examine_ctx_put(ctx);
1478 }
1479 
1480 /* This callback is called after OCF reads cores UUIDs from cache metadata
1481  * Here we allocate memory for those UUIDs and call ocf_metadata_probe_cores() again */
1482 static void
1483 metadata_probe_cores_get_num(void *priv, int error, unsigned int num_cores)
1484 {
1485 	struct metadata_probe_ctx *ctx = priv;
1486 	unsigned int i;
1487 
1488 	if (error) {
1489 		ctx->result = error;
1490 		examine_ctx_put(ctx);
1491 		return;
1492 	}
1493 
1494 	ctx->uuid_count = num_cores;
1495 	ctx->core_uuids = calloc(num_cores, sizeof(struct ocf_volume_uuid));
1496 	if (!ctx->core_uuids) {
1497 		ctx->result = -ENOMEM;
1498 		examine_ctx_put(ctx);
1499 		return;
1500 	}
1501 
1502 	for (i = 0; i < ctx->uuid_count; i++) {
1503 		ctx->core_uuids[i].size = OCF_VOLUME_UUID_MAX_SIZE;
1504 		ctx->core_uuids[i].data = malloc(OCF_VOLUME_UUID_MAX_SIZE);
1505 		if (!ctx->core_uuids[i].data) {
1506 			ctx->result = -ENOMEM;
1507 			examine_ctx_put(ctx);
1508 			return;
1509 		}
1510 	}
1511 
1512 	ocf_metadata_probe_cores(vbdev_ocf_ctx, ctx->volume, ctx->core_uuids, ctx->uuid_count,
1513 				 metadata_probe_cores_construct, ctx);
1514 }
1515 
1516 static void
1517 metadata_probe_cb(void *priv, int rc,
1518 		  struct ocf_metadata_probe_status *status)
1519 {
1520 	struct metadata_probe_ctx *ctx = priv;
1521 
1522 	if (rc) {
1523 		/* -ENODATA means device does not have cache metadata on it */
1524 		if (rc != -ENODATA) {
1525 			ctx->result = rc;
1526 		}
1527 		examine_ctx_put(ctx);
1528 		return;
1529 	}
1530 
1531 	ocf_metadata_probe_cores(vbdev_ocf_ctx, ctx->volume, NULL, 0,
1532 				 metadata_probe_cores_get_num, ctx);
1533 }
1534 
1535 /* This is called after vbdev_ocf_examine
1536  * It allows to delay application initialization
1537  * until all OCF bdevs get registered
1538  * If vbdev has all of its base devices it starts asynchronously here
1539  * We first check if bdev appears in configuration,
1540  * if not we do metadata_probe() to create its configuration from bdev metadata */
1541 static void
1542 vbdev_ocf_examine_disk(struct spdk_bdev *bdev)
1543 {
1544 	const char *bdev_name = spdk_bdev_get_name(bdev);
1545 	struct vbdev_ocf *vbdev;
1546 	struct metadata_probe_ctx *ctx;
1547 	bool created_from_config = false;
1548 	int rc;
1549 
1550 	examine_start(bdev);
1551 
1552 	TAILQ_FOREACH(vbdev, &g_ocf_vbdev_head, tailq) {
1553 		if (vbdev->state.doing_finish || vbdev->state.started) {
1554 			continue;
1555 		}
1556 
1557 		if (!strcmp(bdev_name, vbdev->cache.name)) {
1558 			examine_start(bdev);
1559 			register_vbdev(vbdev, examine_done, bdev);
1560 			created_from_config = true;
1561 			continue;
1562 		}
1563 		if (!strcmp(bdev_name, vbdev->core.name)) {
1564 			examine_start(bdev);
1565 			register_vbdev(vbdev, examine_done, bdev);
1566 			examine_done(0, NULL, bdev);
1567 			return;
1568 		}
1569 	}
1570 
1571 	/* If devices is discovered during config we do not check for metadata */
1572 	if (created_from_config) {
1573 		examine_done(0, NULL, bdev);
1574 		return;
1575 	}
1576 
1577 	/* Metadata probe path
1578 	 * We create temporary OCF volume and a temporary base structure
1579 	 * to use them for ocf_metadata_probe() and for bottom adapter IOs
1580 	 * Then we get UUIDs of core devices an create configurations based on them */
1581 	ctx = calloc(1, sizeof(*ctx));
1582 	if (!ctx) {
1583 		examine_done(-ENOMEM, NULL, bdev);
1584 		return;
1585 	}
1586 
1587 	ctx->base.bdev = bdev;
1588 	ctx->refcnt = 1;
1589 
1590 	rc = spdk_bdev_open(ctx->base.bdev, true, NULL, NULL, &ctx->base.desc);
1591 	if (rc) {
1592 		ctx->result = rc;
1593 		examine_ctx_put(ctx);
1594 		return;
1595 	}
1596 
1597 	rc = ocf_ctx_volume_create(vbdev_ocf_ctx, &ctx->volume, NULL, SPDK_OBJECT);
1598 	if (rc) {
1599 		ctx->result = rc;
1600 		examine_ctx_put(ctx);
1601 		return;
1602 	}
1603 
1604 	rc = ocf_volume_open(ctx->volume, &ctx->base);
1605 	if (rc) {
1606 		ctx->result = rc;
1607 		examine_ctx_put(ctx);
1608 		return;
1609 	}
1610 
1611 	ocf_metadata_probe(vbdev_ocf_ctx, ctx->volume, metadata_probe_cb, ctx);
1612 }
1613 
1614 static int
1615 vbdev_ocf_get_ctx_size(void)
1616 {
1617 	return sizeof(struct bdev_ocf_data);
1618 }
1619 
1620 static void
1621 fini_start(void)
1622 {
1623 	g_fini_started = true;
1624 }
1625 
1626 /* Module-global function table
1627  * Does not relate to vbdev instances */
1628 static struct spdk_bdev_module ocf_if = {
1629 	.name = "ocf",
1630 	.module_init = vbdev_ocf_init,
1631 	.fini_start = fini_start,
1632 	.module_fini = vbdev_ocf_module_fini,
1633 	.config_text = NULL,
1634 	.get_ctx_size = vbdev_ocf_get_ctx_size,
1635 	.examine_config = vbdev_ocf_examine,
1636 	.examine_disk   = vbdev_ocf_examine_disk,
1637 };
1638 SPDK_BDEV_MODULE_REGISTER(ocf, &ocf_if);
1639 
1640 SPDK_LOG_REGISTER_COMPONENT("vbdev_ocf", SPDK_TRACE_VBDEV_OCF)
1641