Line data Source code
1 : /* SPDX-License-Identifier: BSD-3-Clause
2 : * Copyright (C) 2017 Intel Corporation.
3 : * All rights reserved.
4 : * Copyright (c) 2021-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5 : */
6 :
7 : #include "spdk/stdinc.h"
8 :
9 : #include "spdk/blob.h"
10 : #include "spdk/crc32.h"
11 : #include "spdk/env.h"
12 : #include "spdk/queue.h"
13 : #include "spdk/thread.h"
14 : #include "spdk/bit_array.h"
15 : #include "spdk/bit_pool.h"
16 : #include "spdk/likely.h"
17 : #include "spdk/util.h"
18 : #include "spdk/string.h"
19 : #include "spdk/trace.h"
20 :
21 : #include "spdk_internal/assert.h"
22 : #include "spdk_internal/trace_defs.h"
23 : #include "spdk/log.h"
24 :
25 : #include "blobstore.h"
26 :
27 : #define BLOB_CRC32C_INITIAL 0xffffffffUL
28 :
29 : static int bs_register_md_thread(struct spdk_blob_store *bs);
30 : static int bs_unregister_md_thread(struct spdk_blob_store *bs);
31 : static void blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno);
32 : static void blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
33 : uint64_t cluster, uint32_t extent, struct spdk_blob_md_page *page,
34 : spdk_blob_op_complete cb_fn, void *cb_arg);
35 : static void blob_free_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
36 : uint32_t extent_page, struct spdk_blob_md_page *page, spdk_blob_op_complete cb_fn, void *cb_arg);
37 :
38 : static int blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
39 : uint16_t value_len, bool internal);
40 : static int blob_get_xattr_value(struct spdk_blob *blob, const char *name,
41 : const void **value, size_t *value_len, bool internal);
42 : static int blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal);
43 :
44 : static void blob_write_extent_page(struct spdk_blob *blob, uint32_t extent, uint64_t cluster_num,
45 : struct spdk_blob_md_page *page, spdk_blob_op_complete cb_fn, void *cb_arg);
46 : static void blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg);
47 :
48 : static void bs_shallow_copy_cluster_find_next(void *cb_arg);
49 :
50 : /*
51 : * External snapshots require a channel per thread per esnap bdev. The tree
52 : * is populated lazily as blob IOs are handled by the back_bs_dev. When this
53 : * channel is destroyed, all the channels in the tree are destroyed.
54 : */
55 :
56 : struct blob_esnap_channel {
57 : RB_ENTRY(blob_esnap_channel) node;
58 : spdk_blob_id blob_id;
59 : struct spdk_io_channel *channel;
60 : };
61 :
62 : static int blob_esnap_channel_compare(struct blob_esnap_channel *c1, struct blob_esnap_channel *c2);
63 : static void blob_esnap_destroy_bs_dev_channels(struct spdk_blob *blob, bool abort_io,
64 : spdk_blob_op_with_handle_complete cb_fn, void *cb_arg);
65 : static void blob_esnap_destroy_bs_channel(struct spdk_bs_channel *ch);
66 : static void blob_set_back_bs_dev_frozen(void *_ctx, int bserrno);
67 19201 : RB_GENERATE_STATIC(blob_esnap_channel_tree, blob_esnap_channel, node, blob_esnap_channel_compare)
68 :
69 : static inline bool
70 68456 : blob_is_esnap_clone(const struct spdk_blob *blob)
71 : {
72 68456 : assert(blob != NULL);
73 68456 : return !!(blob->invalid_flags & SPDK_BLOB_EXTERNAL_SNAPSHOT);
74 : }
75 :
76 : static int
77 2875 : blob_id_cmp(struct spdk_blob *blob1, struct spdk_blob *blob2)
78 : {
79 2875 : assert(blob1 != NULL && blob2 != NULL);
80 2875 : return (blob1->id < blob2->id ? -1 : blob1->id > blob2->id);
81 : }
82 :
83 18769 : RB_GENERATE_STATIC(spdk_blob_tree, spdk_blob, link, blob_id_cmp);
84 :
85 : static void
86 46177 : blob_verify_md_op(struct spdk_blob *blob)
87 : {
88 46177 : assert(blob != NULL);
89 46177 : assert(spdk_get_thread() == blob->bs->md_thread);
90 46177 : assert(blob->state != SPDK_BLOB_STATE_LOADING);
91 46177 : }
92 :
93 : static struct spdk_blob_list *
94 4783 : bs_get_snapshot_entry(struct spdk_blob_store *bs, spdk_blob_id blobid)
95 : {
96 4783 : struct spdk_blob_list *snapshot_entry = NULL;
97 :
98 6018 : TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
99 2207 : if (snapshot_entry->id == blobid) {
100 972 : break;
101 : }
102 : }
103 :
104 4783 : return snapshot_entry;
105 : }
106 :
107 : static void
108 3807 : bs_claim_md_page(struct spdk_blob_store *bs, uint32_t page)
109 : {
110 3807 : assert(spdk_spin_held(&bs->used_lock));
111 3807 : assert(page < spdk_bit_array_capacity(bs->used_md_pages));
112 3807 : assert(spdk_bit_array_get(bs->used_md_pages, page) == false);
113 :
114 3807 : spdk_bit_array_set(bs->used_md_pages, page);
115 3807 : }
116 :
117 : static void
118 2901 : bs_release_md_page(struct spdk_blob_store *bs, uint32_t page)
119 : {
120 2901 : assert(spdk_spin_held(&bs->used_lock));
121 2901 : assert(page < spdk_bit_array_capacity(bs->used_md_pages));
122 2901 : assert(spdk_bit_array_get(bs->used_md_pages, page) == true);
123 :
124 2901 : spdk_bit_array_clear(bs->used_md_pages, page);
125 2901 : }
126 :
127 : static uint32_t
128 10283 : bs_claim_cluster(struct spdk_blob_store *bs)
129 : {
130 : uint32_t cluster_num;
131 :
132 10283 : assert(spdk_spin_held(&bs->used_lock));
133 :
134 10283 : cluster_num = spdk_bit_pool_allocate_bit(bs->used_clusters);
135 10283 : if (cluster_num == UINT32_MAX) {
136 0 : return UINT32_MAX;
137 : }
138 :
139 10283 : SPDK_DEBUGLOG(blob, "Claiming cluster %u\n", cluster_num);
140 10283 : bs->num_free_clusters--;
141 :
142 10283 : return cluster_num;
143 : }
144 :
145 : static void
146 2996 : bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
147 : {
148 2996 : assert(spdk_spin_held(&bs->used_lock));
149 2996 : assert(cluster_num < spdk_bit_pool_capacity(bs->used_clusters));
150 2996 : assert(spdk_bit_pool_is_allocated(bs->used_clusters, cluster_num) == true);
151 2996 : assert(bs->num_free_clusters < bs->total_clusters);
152 :
153 2996 : SPDK_DEBUGLOG(blob, "Releasing cluster %u\n", cluster_num);
154 :
155 2996 : spdk_bit_pool_free_bit(bs->used_clusters, cluster_num);
156 2996 : bs->num_free_clusters++;
157 2996 : }
158 :
159 : static int
160 10283 : blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster)
161 : {
162 10283 : uint64_t *cluster_lba = &blob->active.clusters[cluster_num];
163 :
164 10283 : blob_verify_md_op(blob);
165 :
166 10283 : if (*cluster_lba != 0) {
167 5 : return -EEXIST;
168 : }
169 :
170 10278 : *cluster_lba = bs_cluster_to_lba(blob->bs, cluster);
171 10278 : blob->active.num_allocated_clusters++;
172 :
173 10278 : return 0;
174 : }
175 :
176 : static int
177 10283 : bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num,
178 : uint64_t *cluster, uint32_t *lowest_free_md_page, bool update_map)
179 : {
180 10283 : uint32_t *extent_page = 0;
181 :
182 10283 : assert(spdk_spin_held(&blob->bs->used_lock));
183 :
184 10283 : *cluster = bs_claim_cluster(blob->bs);
185 10283 : if (*cluster == UINT32_MAX) {
186 : /* No more free clusters. Cannot satisfy the request */
187 0 : return -ENOSPC;
188 : }
189 :
190 10283 : if (blob->use_extent_table) {
191 6227 : extent_page = bs_cluster_to_extent_page(blob, cluster_num);
192 6227 : if (*extent_page == 0) {
193 : /* Extent page shall never occupy md_page so start the search from 1 */
194 1087 : if (*lowest_free_md_page == 0) {
195 1084 : *lowest_free_md_page = 1;
196 : }
197 : /* No extent_page is allocated for the cluster */
198 1087 : *lowest_free_md_page = spdk_bit_array_find_first_clear(blob->bs->used_md_pages,
199 : *lowest_free_md_page);
200 1087 : if (*lowest_free_md_page == UINT32_MAX) {
201 : /* No more free md pages. Cannot satisfy the request */
202 0 : bs_release_cluster(blob->bs, *cluster);
203 0 : return -ENOSPC;
204 : }
205 1087 : bs_claim_md_page(blob->bs, *lowest_free_md_page);
206 : }
207 : }
208 :
209 10283 : SPDK_DEBUGLOG(blob, "Claiming cluster %" PRIu64 " for blob 0x%" PRIx64 "\n", *cluster,
210 : blob->id);
211 :
212 10283 : if (update_map) {
213 9253 : blob_insert_cluster(blob, cluster_num, *cluster);
214 9253 : if (blob->use_extent_table && *extent_page == 0) {
215 958 : *extent_page = *lowest_free_md_page;
216 : }
217 : }
218 :
219 10283 : return 0;
220 : }
221 :
222 : static void
223 6977 : blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs)
224 : {
225 6977 : xattrs->count = 0;
226 6977 : xattrs->names = NULL;
227 6977 : xattrs->ctx = NULL;
228 6977 : xattrs->get_value = NULL;
229 6977 : }
230 :
231 : void
232 4611 : spdk_blob_opts_init(struct spdk_blob_opts *opts, size_t opts_size)
233 : {
234 4611 : if (!opts) {
235 0 : SPDK_ERRLOG("opts should not be NULL\n");
236 0 : return;
237 : }
238 :
239 4611 : if (!opts_size) {
240 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
241 0 : return;
242 : }
243 :
244 4611 : memset(opts, 0, opts_size);
245 4611 : opts->opts_size = opts_size;
246 :
247 : #define FIELD_OK(field) \
248 : offsetof(struct spdk_blob_opts, field) + sizeof(opts->field) <= opts_size
249 :
250 : #define SET_FIELD(field, value) \
251 : if (FIELD_OK(field)) { \
252 : opts->field = value; \
253 : } \
254 :
255 4611 : SET_FIELD(num_clusters, 0);
256 4611 : SET_FIELD(thin_provision, false);
257 4611 : SET_FIELD(clear_method, BLOB_CLEAR_WITH_DEFAULT);
258 :
259 4611 : if (FIELD_OK(xattrs)) {
260 4611 : blob_xattrs_init(&opts->xattrs);
261 : }
262 :
263 4611 : SET_FIELD(use_extent_table, true);
264 :
265 : #undef FIELD_OK
266 : #undef SET_FIELD
267 : }
268 :
269 : void
270 4346 : spdk_blob_open_opts_init(struct spdk_blob_open_opts *opts, size_t opts_size)
271 : {
272 4346 : if (!opts) {
273 0 : SPDK_ERRLOG("opts should not be NULL\n");
274 0 : return;
275 : }
276 :
277 4346 : if (!opts_size) {
278 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
279 0 : return;
280 : }
281 :
282 4346 : memset(opts, 0, opts_size);
283 4346 : opts->opts_size = opts_size;
284 :
285 : #define FIELD_OK(field) \
286 : offsetof(struct spdk_blob_open_opts, field) + sizeof(opts->field) <= opts_size
287 :
288 : #define SET_FIELD(field, value) \
289 : if (FIELD_OK(field)) { \
290 : opts->field = value; \
291 : } \
292 :
293 4346 : SET_FIELD(clear_method, BLOB_CLEAR_WITH_DEFAULT);
294 :
295 : #undef FIELD_OK
296 : #undef SET_FILED
297 : }
298 :
299 : static struct spdk_blob *
300 6707 : blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id)
301 : {
302 : struct spdk_blob *blob;
303 :
304 6707 : blob = calloc(1, sizeof(*blob));
305 6707 : if (!blob) {
306 0 : return NULL;
307 : }
308 :
309 6707 : blob->id = id;
310 6707 : blob->bs = bs;
311 :
312 6707 : blob->parent_id = SPDK_BLOBID_INVALID;
313 :
314 6707 : blob->state = SPDK_BLOB_STATE_DIRTY;
315 6707 : blob->extent_rle_found = false;
316 6707 : blob->extent_table_found = false;
317 6707 : blob->active.num_pages = 1;
318 6707 : blob->active.pages = calloc(1, sizeof(*blob->active.pages));
319 6707 : if (!blob->active.pages) {
320 0 : free(blob);
321 0 : return NULL;
322 : }
323 :
324 6707 : blob->active.pages[0] = bs_blobid_to_page(id);
325 :
326 6707 : TAILQ_INIT(&blob->xattrs);
327 6707 : TAILQ_INIT(&blob->xattrs_internal);
328 6707 : TAILQ_INIT(&blob->pending_persists);
329 6707 : TAILQ_INIT(&blob->persists_to_complete);
330 :
331 6707 : return blob;
332 : }
333 :
334 : static void
335 13414 : xattrs_free(struct spdk_xattr_tailq *xattrs)
336 : {
337 : struct spdk_xattr *xattr, *xattr_tmp;
338 :
339 15601 : TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) {
340 2187 : TAILQ_REMOVE(xattrs, xattr, link);
341 2187 : free(xattr->name);
342 2187 : free(xattr->value);
343 2187 : free(xattr);
344 : }
345 13414 : }
346 :
347 : static void
348 1398 : blob_unref_back_bs_dev(struct spdk_blob *blob)
349 : {
350 1398 : blob->back_bs_dev->destroy(blob->back_bs_dev);
351 1398 : blob->back_bs_dev = NULL;
352 1398 : }
353 :
354 : static void
355 6707 : blob_free(struct spdk_blob *blob)
356 : {
357 6707 : assert(blob != NULL);
358 6707 : assert(TAILQ_EMPTY(&blob->pending_persists));
359 6707 : assert(TAILQ_EMPTY(&blob->persists_to_complete));
360 :
361 6707 : free(blob->active.extent_pages);
362 6707 : free(blob->clean.extent_pages);
363 6707 : free(blob->active.clusters);
364 6707 : free(blob->clean.clusters);
365 6707 : free(blob->active.pages);
366 6707 : free(blob->clean.pages);
367 :
368 6707 : xattrs_free(&blob->xattrs);
369 6707 : xattrs_free(&blob->xattrs_internal);
370 :
371 6707 : if (blob->back_bs_dev) {
372 1363 : blob_unref_back_bs_dev(blob);
373 : }
374 :
375 6707 : free(blob);
376 6707 : }
377 :
378 : static void
379 406 : blob_back_bs_destroy_esnap_done(void *ctx, struct spdk_blob *blob, int bserrno)
380 : {
381 406 : struct spdk_bs_dev *bs_dev = ctx;
382 :
383 406 : if (bserrno != 0) {
384 : /*
385 : * This is probably due to a memory allocation failure when creating the
386 : * blob_esnap_destroy_ctx before iterating threads.
387 : */
388 0 : SPDK_ERRLOG("blob 0x%" PRIx64 ": Unable to destroy bs dev channels: error %d\n",
389 : blob->id, bserrno);
390 0 : assert(false);
391 : }
392 :
393 406 : if (bs_dev == NULL) {
394 : /*
395 : * This check exists to make scanbuild happy.
396 : *
397 : * blob->back_bs_dev for an esnap is NULL during the first iteration of blobs while
398 : * the blobstore is being loaded. It could also be NULL if there was an error
399 : * opening the esnap device. In each of these cases, no channels could have been
400 : * created because back_bs_dev->create_channel() would have led to a NULL pointer
401 : * deref.
402 : */
403 0 : assert(false);
404 : return;
405 : }
406 :
407 406 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64 ": calling destroy on back_bs_dev\n", blob->id);
408 406 : bs_dev->destroy(bs_dev);
409 : }
410 :
411 : static void
412 406 : blob_back_bs_destroy(struct spdk_blob *blob)
413 : {
414 406 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64 ": preparing to destroy back_bs_dev\n",
415 : blob->id);
416 :
417 406 : blob_esnap_destroy_bs_dev_channels(blob, false, blob_back_bs_destroy_esnap_done,
418 406 : blob->back_bs_dev);
419 406 : blob->back_bs_dev = NULL;
420 406 : }
421 :
422 : struct blob_parent {
423 : union {
424 : struct {
425 : spdk_blob_id id;
426 : struct spdk_blob *blob;
427 : } snapshot;
428 :
429 : struct {
430 : void *id;
431 : uint32_t id_len;
432 : struct spdk_bs_dev *back_bs_dev;
433 : } esnap;
434 : } u;
435 : };
436 :
437 : typedef int (*set_parent_refs_cb)(struct spdk_blob *blob, struct blob_parent *parent);
438 :
439 : struct set_bs_dev_ctx {
440 : struct spdk_blob *blob;
441 : struct spdk_bs_dev *back_bs_dev;
442 :
443 : /*
444 : * This callback is used during a set parent operation to change the references
445 : * to the parent of the blob.
446 : */
447 : set_parent_refs_cb parent_refs_cb_fn;
448 : struct blob_parent *parent_refs_cb_arg;
449 :
450 : spdk_blob_op_complete cb_fn;
451 : void *cb_arg;
452 : int bserrno;
453 : };
454 :
455 : static void
456 35 : blob_set_back_bs_dev(struct spdk_blob *blob, struct spdk_bs_dev *back_bs_dev,
457 : set_parent_refs_cb parent_refs_cb_fn, struct blob_parent *parent_refs_cb_arg,
458 : spdk_blob_op_complete cb_fn, void *cb_arg)
459 : {
460 : struct set_bs_dev_ctx *ctx;
461 :
462 35 : ctx = calloc(1, sizeof(*ctx));
463 35 : if (ctx == NULL) {
464 0 : SPDK_ERRLOG("blob 0x%" PRIx64 ": out of memory while setting back_bs_dev\n",
465 : blob->id);
466 0 : cb_fn(cb_arg, -ENOMEM);
467 0 : return;
468 : }
469 :
470 35 : ctx->parent_refs_cb_fn = parent_refs_cb_fn;
471 35 : ctx->parent_refs_cb_arg = parent_refs_cb_arg;
472 35 : ctx->cb_fn = cb_fn;
473 35 : ctx->cb_arg = cb_arg;
474 35 : ctx->back_bs_dev = back_bs_dev;
475 35 : ctx->blob = blob;
476 :
477 35 : blob_freeze_io(blob, blob_set_back_bs_dev_frozen, ctx);
478 : }
479 :
480 : struct freeze_io_ctx {
481 : struct spdk_bs_cpl cpl;
482 : struct spdk_blob *blob;
483 : };
484 :
485 : static void
486 663 : blob_io_sync(struct spdk_io_channel_iter *i)
487 : {
488 663 : spdk_for_each_channel_continue(i, 0);
489 663 : }
490 :
491 : static void
492 648 : blob_execute_queued_io(struct spdk_io_channel_iter *i)
493 : {
494 648 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
495 648 : struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch);
496 648 : struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
497 : struct spdk_bs_request_set *set;
498 : struct spdk_bs_user_op_args *args;
499 : spdk_bs_user_op_t *op, *tmp;
500 :
501 653 : TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) {
502 5 : set = (struct spdk_bs_request_set *)op;
503 5 : args = &set->u.user_op;
504 :
505 5 : if (args->blob == ctx->blob) {
506 5 : TAILQ_REMOVE(&ch->queued_io, op, link);
507 5 : bs_user_op_execute(op);
508 : }
509 : }
510 :
511 648 : spdk_for_each_channel_continue(i, 0);
512 648 : }
513 :
514 : static void
515 1271 : blob_io_cpl(struct spdk_io_channel_iter *i, int status)
516 : {
517 1271 : struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
518 :
519 1271 : ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0);
520 :
521 1271 : free(ctx);
522 1271 : }
523 :
524 : static void
525 643 : blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
526 : {
527 : struct freeze_io_ctx *ctx;
528 :
529 643 : blob_verify_md_op(blob);
530 :
531 643 : ctx = calloc(1, sizeof(*ctx));
532 643 : if (!ctx) {
533 0 : cb_fn(cb_arg, -ENOMEM);
534 0 : return;
535 : }
536 :
537 643 : ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
538 643 : ctx->cpl.u.blob_basic.cb_fn = cb_fn;
539 643 : ctx->cpl.u.blob_basic.cb_arg = cb_arg;
540 643 : ctx->blob = blob;
541 :
542 : /* Freeze I/O on blob */
543 643 : blob->frozen_refcnt++;
544 :
545 643 : spdk_for_each_channel(blob->bs, blob_io_sync, ctx, blob_io_cpl);
546 : }
547 :
548 : static void
549 628 : blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
550 : {
551 : struct freeze_io_ctx *ctx;
552 :
553 628 : blob_verify_md_op(blob);
554 :
555 628 : ctx = calloc(1, sizeof(*ctx));
556 628 : if (!ctx) {
557 0 : cb_fn(cb_arg, -ENOMEM);
558 0 : return;
559 : }
560 :
561 628 : ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
562 628 : ctx->cpl.u.blob_basic.cb_fn = cb_fn;
563 628 : ctx->cpl.u.blob_basic.cb_arg = cb_arg;
564 628 : ctx->blob = blob;
565 :
566 628 : assert(blob->frozen_refcnt > 0);
567 :
568 628 : blob->frozen_refcnt--;
569 :
570 628 : spdk_for_each_channel(blob->bs, blob_execute_queued_io, ctx, blob_io_cpl);
571 : }
572 :
573 : static int
574 10498 : blob_mark_clean(struct spdk_blob *blob)
575 : {
576 10498 : uint32_t *extent_pages = NULL;
577 10498 : uint64_t *clusters = NULL;
578 10498 : uint32_t *pages = NULL;
579 :
580 10498 : assert(blob != NULL);
581 :
582 10498 : if (blob->active.num_extent_pages) {
583 4258 : assert(blob->active.extent_pages);
584 4258 : extent_pages = calloc(blob->active.num_extent_pages, sizeof(*blob->active.extent_pages));
585 4258 : if (!extent_pages) {
586 0 : return -ENOMEM;
587 : }
588 4258 : memcpy(extent_pages, blob->active.extent_pages,
589 4258 : blob->active.num_extent_pages * sizeof(*extent_pages));
590 : }
591 :
592 10498 : if (blob->active.num_clusters) {
593 7349 : assert(blob->active.clusters);
594 7349 : clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters));
595 7349 : if (!clusters) {
596 0 : free(extent_pages);
597 0 : return -ENOMEM;
598 : }
599 7349 : memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*blob->active.clusters));
600 : }
601 :
602 10498 : if (blob->active.num_pages) {
603 8641 : assert(blob->active.pages);
604 8641 : pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages));
605 8641 : if (!pages) {
606 0 : free(extent_pages);
607 0 : free(clusters);
608 0 : return -ENOMEM;
609 : }
610 8641 : memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
611 : }
612 :
613 10498 : free(blob->clean.extent_pages);
614 10498 : free(blob->clean.clusters);
615 10498 : free(blob->clean.pages);
616 :
617 10498 : blob->clean.num_extent_pages = blob->active.num_extent_pages;
618 10498 : blob->clean.extent_pages = blob->active.extent_pages;
619 10498 : blob->clean.num_clusters = blob->active.num_clusters;
620 10498 : blob->clean.clusters = blob->active.clusters;
621 10498 : blob->clean.num_allocated_clusters = blob->active.num_allocated_clusters;
622 10498 : blob->clean.num_pages = blob->active.num_pages;
623 10498 : blob->clean.pages = blob->active.pages;
624 :
625 10498 : blob->active.extent_pages = extent_pages;
626 10498 : blob->active.clusters = clusters;
627 10498 : blob->active.pages = pages;
628 :
629 : /* If the metadata was dirtied again while the metadata was being written to disk,
630 : * we do not want to revert the DIRTY state back to CLEAN here.
631 : */
632 10498 : if (blob->state == SPDK_BLOB_STATE_LOADING) {
633 4259 : blob->state = SPDK_BLOB_STATE_CLEAN;
634 : }
635 :
636 10498 : return 0;
637 : }
638 :
639 : static int
640 1592 : blob_deserialize_xattr(struct spdk_blob *blob,
641 : struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal)
642 : {
643 : struct spdk_xattr *xattr;
644 :
645 1592 : if (desc_xattr->length != sizeof(desc_xattr->name_length) +
646 : sizeof(desc_xattr->value_length) +
647 1592 : desc_xattr->name_length + desc_xattr->value_length) {
648 0 : return -EINVAL;
649 : }
650 :
651 1592 : xattr = calloc(1, sizeof(*xattr));
652 1592 : if (xattr == NULL) {
653 0 : return -ENOMEM;
654 : }
655 :
656 1592 : xattr->name = malloc(desc_xattr->name_length + 1);
657 1592 : if (xattr->name == NULL) {
658 0 : free(xattr);
659 0 : return -ENOMEM;
660 : }
661 :
662 1592 : xattr->value = malloc(desc_xattr->value_length);
663 1592 : if (xattr->value == NULL) {
664 0 : free(xattr->name);
665 0 : free(xattr);
666 0 : return -ENOMEM;
667 : }
668 :
669 1592 : memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length);
670 1592 : xattr->name[desc_xattr->name_length] = '\0';
671 1592 : xattr->value_len = desc_xattr->value_length;
672 1592 : memcpy(xattr->value,
673 1592 : (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
674 1592 : desc_xattr->value_length);
675 :
676 1592 : TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link);
677 :
678 1592 : return 0;
679 : }
680 :
681 :
682 : static int
683 5980 : blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob)
684 : {
685 : struct spdk_blob_md_descriptor *desc;
686 5980 : size_t cur_desc = 0;
687 : void *tmp;
688 :
689 5980 : desc = (struct spdk_blob_md_descriptor *)page->descriptors;
690 17389 : while (cur_desc < sizeof(page->descriptors)) {
691 17389 : if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
692 5920 : if (desc->length == 0) {
693 : /* If padding and length are 0, this terminates the page */
694 5920 : break;
695 : }
696 11469 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
697 : struct spdk_blob_md_descriptor_flags *desc_flags;
698 :
699 4301 : desc_flags = (struct spdk_blob_md_descriptor_flags *)desc;
700 :
701 4301 : if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) {
702 0 : return -EINVAL;
703 : }
704 :
705 4301 : if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) !=
706 : SPDK_BLOB_INVALID_FLAGS_MASK) {
707 10 : return -EINVAL;
708 : }
709 :
710 4291 : if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) !=
711 : SPDK_BLOB_DATA_RO_FLAGS_MASK) {
712 15 : blob->data_ro = true;
713 15 : blob->md_ro = true;
714 : }
715 :
716 4291 : if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) !=
717 : SPDK_BLOB_MD_RO_FLAGS_MASK) {
718 15 : blob->md_ro = true;
719 : }
720 :
721 4291 : if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
722 712 : blob->data_ro = true;
723 712 : blob->md_ro = true;
724 : }
725 :
726 4291 : blob->invalid_flags = desc_flags->invalid_flags;
727 4291 : blob->data_ro_flags = desc_flags->data_ro_flags;
728 4291 : blob->md_ro_flags = desc_flags->md_ro_flags;
729 :
730 7168 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
731 : struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle;
732 : unsigned int i, j;
733 1396 : unsigned int cluster_count = blob->active.num_clusters;
734 :
735 1396 : if (blob->extent_table_found) {
736 : /* Extent Table already present in the md,
737 : * both descriptors should never be at the same time. */
738 0 : return -EINVAL;
739 : }
740 1396 : blob->extent_rle_found = true;
741 :
742 1396 : desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
743 :
744 1396 : if (desc_extent_rle->length == 0 ||
745 1396 : (desc_extent_rle->length % sizeof(desc_extent_rle->extents[0]) != 0)) {
746 0 : return -EINVAL;
747 : }
748 :
749 2968 : for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
750 21240 : for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
751 19668 : if (desc_extent_rle->extents[i].cluster_idx != 0) {
752 6692 : if (!spdk_bit_pool_is_allocated(blob->bs->used_clusters,
753 6692 : desc_extent_rle->extents[i].cluster_idx + j)) {
754 0 : return -EINVAL;
755 : }
756 : }
757 19668 : cluster_count++;
758 : }
759 : }
760 :
761 1396 : if (cluster_count == 0) {
762 0 : return -EINVAL;
763 : }
764 1396 : tmp = realloc(blob->active.clusters, cluster_count * sizeof(*blob->active.clusters));
765 1396 : if (tmp == NULL) {
766 0 : return -ENOMEM;
767 : }
768 1396 : blob->active.clusters = tmp;
769 1396 : blob->active.cluster_array_size = cluster_count;
770 :
771 2968 : for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
772 21240 : for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
773 19668 : if (desc_extent_rle->extents[i].cluster_idx != 0) {
774 6692 : blob->active.clusters[blob->active.num_clusters++] = bs_cluster_to_lba(blob->bs,
775 6692 : desc_extent_rle->extents[i].cluster_idx + j);
776 6692 : blob->active.num_allocated_clusters++;
777 12976 : } else if (spdk_blob_is_thin_provisioned(blob)) {
778 12976 : blob->active.clusters[blob->active.num_clusters++] = 0;
779 : } else {
780 0 : return -EINVAL;
781 : }
782 : }
783 : }
784 5772 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) {
785 : struct spdk_blob_md_descriptor_extent_table *desc_extent_table;
786 2624 : uint32_t num_extent_pages = blob->active.num_extent_pages;
787 : uint32_t i, j;
788 : size_t extent_pages_length;
789 :
790 2624 : desc_extent_table = (struct spdk_blob_md_descriptor_extent_table *)desc;
791 2624 : extent_pages_length = desc_extent_table->length - sizeof(desc_extent_table->num_clusters);
792 :
793 2624 : if (blob->extent_rle_found) {
794 : /* This means that Extent RLE is present in MD,
795 : * both should never be at the same time. */
796 0 : return -EINVAL;
797 2624 : } else if (blob->extent_table_found &&
798 0 : desc_extent_table->num_clusters != blob->remaining_clusters_in_et) {
799 : /* Number of clusters in this ET does not match number
800 : * from previously read EXTENT_TABLE. */
801 0 : return -EINVAL;
802 : }
803 :
804 2624 : if (desc_extent_table->length == 0 ||
805 2624 : (extent_pages_length % sizeof(desc_extent_table->extent_page[0]) != 0)) {
806 0 : return -EINVAL;
807 : }
808 :
809 2624 : blob->extent_table_found = true;
810 :
811 4825 : for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
812 2201 : num_extent_pages += desc_extent_table->extent_page[i].num_pages;
813 : }
814 :
815 2624 : if (num_extent_pages > 0) {
816 2177 : tmp = realloc(blob->active.extent_pages, num_extent_pages * sizeof(uint32_t));
817 2177 : if (tmp == NULL) {
818 0 : return -ENOMEM;
819 : }
820 2177 : blob->active.extent_pages = tmp;
821 : }
822 2624 : blob->active.extent_pages_array_size = num_extent_pages;
823 :
824 2624 : blob->remaining_clusters_in_et = desc_extent_table->num_clusters;
825 :
826 : /* Extent table entries contain md page numbers for extent pages.
827 : * Zeroes represent unallocated extent pages, those are run-length-encoded.
828 : */
829 4825 : for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
830 2201 : if (desc_extent_table->extent_page[i].page_idx != 0) {
831 1565 : assert(desc_extent_table->extent_page[i].num_pages == 1);
832 3130 : blob->active.extent_pages[blob->active.num_extent_pages++] =
833 1565 : desc_extent_table->extent_page[i].page_idx;
834 636 : } else if (spdk_blob_is_thin_provisioned(blob)) {
835 1272 : for (j = 0; j < desc_extent_table->extent_page[i].num_pages; j++) {
836 636 : blob->active.extent_pages[blob->active.num_extent_pages++] = 0;
837 : }
838 : } else {
839 0 : return -EINVAL;
840 : }
841 : }
842 3148 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
843 : struct spdk_blob_md_descriptor_extent_page *desc_extent;
844 : unsigned int i;
845 1556 : unsigned int cluster_count = 0;
846 : size_t cluster_idx_length;
847 :
848 1556 : if (blob->extent_rle_found) {
849 : /* This means that Extent RLE is present in MD,
850 : * both should never be at the same time. */
851 0 : return -EINVAL;
852 : }
853 :
854 1556 : desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc;
855 1556 : cluster_idx_length = desc_extent->length - sizeof(desc_extent->start_cluster_idx);
856 :
857 1556 : if (desc_extent->length <= sizeof(desc_extent->start_cluster_idx) ||
858 1556 : (cluster_idx_length % sizeof(desc_extent->cluster_idx[0]) != 0)) {
859 0 : return -EINVAL;
860 : }
861 :
862 24472 : for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) {
863 22916 : if (desc_extent->cluster_idx[i] != 0) {
864 10415 : if (!spdk_bit_pool_is_allocated(blob->bs->used_clusters, desc_extent->cluster_idx[i])) {
865 0 : return -EINVAL;
866 : }
867 : }
868 22916 : cluster_count++;
869 : }
870 :
871 1556 : if (cluster_count == 0) {
872 0 : return -EINVAL;
873 : }
874 :
875 : /* When reading extent pages sequentially starting cluster idx should match
876 : * current size of a blob.
877 : * If changed to batch reading, this check shall be removed. */
878 1556 : if (desc_extent->start_cluster_idx != blob->active.num_clusters) {
879 0 : return -EINVAL;
880 : }
881 :
882 1556 : tmp = realloc(blob->active.clusters,
883 1556 : (cluster_count + blob->active.num_clusters) * sizeof(*blob->active.clusters));
884 1556 : if (tmp == NULL) {
885 0 : return -ENOMEM;
886 : }
887 1556 : blob->active.clusters = tmp;
888 1556 : blob->active.cluster_array_size = (cluster_count + blob->active.num_clusters);
889 :
890 24472 : for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) {
891 22916 : if (desc_extent->cluster_idx[i] != 0) {
892 10415 : blob->active.clusters[blob->active.num_clusters++] = bs_cluster_to_lba(blob->bs,
893 : desc_extent->cluster_idx[i]);
894 10415 : blob->active.num_allocated_clusters++;
895 12501 : } else if (spdk_blob_is_thin_provisioned(blob)) {
896 12501 : blob->active.clusters[blob->active.num_clusters++] = 0;
897 : } else {
898 0 : return -EINVAL;
899 : }
900 : }
901 1556 : assert(desc_extent->start_cluster_idx + cluster_count == blob->active.num_clusters);
902 1556 : assert(blob->remaining_clusters_in_et >= cluster_count);
903 1556 : blob->remaining_clusters_in_et -= cluster_count;
904 1592 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
905 : int rc;
906 :
907 476 : rc = blob_deserialize_xattr(blob,
908 : (struct spdk_blob_md_descriptor_xattr *) desc, false);
909 476 : if (rc != 0) {
910 0 : return rc;
911 : }
912 1116 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
913 : int rc;
914 :
915 1116 : rc = blob_deserialize_xattr(blob,
916 : (struct spdk_blob_md_descriptor_xattr *) desc, true);
917 1116 : if (rc != 0) {
918 0 : return rc;
919 : }
920 : } else {
921 : /* Unrecognized descriptor type. Do not fail - just continue to the
922 : * next descriptor. If this descriptor is associated with some feature
923 : * defined in a newer version of blobstore, that version of blobstore
924 : * should create and set an associated feature flag to specify if this
925 : * blob can be loaded or not.
926 : */
927 : }
928 :
929 : /* Advance to the next descriptor */
930 11459 : cur_desc += sizeof(*desc) + desc->length;
931 11459 : if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
932 50 : break;
933 : }
934 11409 : desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
935 : }
936 :
937 5970 : return 0;
938 : }
939 :
940 : static bool bs_load_cur_extent_page_valid(struct spdk_blob_md_page *page);
941 :
942 : static int
943 1556 : blob_parse_extent_page(struct spdk_blob_md_page *extent_page, struct spdk_blob *blob)
944 : {
945 1556 : assert(blob != NULL);
946 1556 : assert(blob->state == SPDK_BLOB_STATE_LOADING);
947 :
948 1556 : if (bs_load_cur_extent_page_valid(extent_page) == false) {
949 0 : return -ENOENT;
950 : }
951 :
952 1556 : return blob_parse_page(extent_page, blob);
953 : }
954 :
955 : static int
956 4306 : blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count,
957 : struct spdk_blob *blob)
958 : {
959 : const struct spdk_blob_md_page *page;
960 : uint32_t i;
961 : int rc;
962 : void *tmp;
963 :
964 4306 : assert(page_count > 0);
965 4306 : assert(pages[0].sequence_num == 0);
966 4306 : assert(blob != NULL);
967 4306 : assert(blob->state == SPDK_BLOB_STATE_LOADING);
968 4306 : assert(blob->active.clusters == NULL);
969 :
970 : /* The blobid provided doesn't match what's in the MD, this can
971 : * happen for example if a bogus blobid is passed in through open.
972 : */
973 4306 : if (blob->id != pages[0].id) {
974 5 : SPDK_ERRLOG("Blobid (0x%" PRIx64 ") doesn't match what's in metadata "
975 : "(0x%" PRIx64 ")\n", blob->id, pages[0].id);
976 5 : return -ENOENT;
977 : }
978 :
979 4301 : tmp = realloc(blob->active.pages, page_count * sizeof(*blob->active.pages));
980 4301 : if (!tmp) {
981 0 : return -ENOMEM;
982 : }
983 4301 : blob->active.pages = tmp;
984 :
985 4301 : blob->active.pages[0] = pages[0].id;
986 :
987 4424 : for (i = 1; i < page_count; i++) {
988 123 : assert(spdk_bit_array_get(blob->bs->used_md_pages, pages[i - 1].next));
989 123 : blob->active.pages[i] = pages[i - 1].next;
990 : }
991 4301 : blob->active.num_pages = page_count;
992 :
993 8715 : for (i = 0; i < page_count; i++) {
994 4424 : page = &pages[i];
995 :
996 4424 : assert(page->id == blob->id);
997 4424 : assert(page->sequence_num == i);
998 :
999 4424 : rc = blob_parse_page(page, blob);
1000 4424 : if (rc != 0) {
1001 10 : return rc;
1002 : }
1003 : }
1004 :
1005 4291 : return 0;
1006 : }
1007 :
1008 : static int
1009 5532 : blob_serialize_add_page(const struct spdk_blob *blob,
1010 : struct spdk_blob_md_page **pages,
1011 : uint32_t *page_count,
1012 : struct spdk_blob_md_page **last_page)
1013 : {
1014 : struct spdk_blob_md_page *page, *tmp_pages;
1015 :
1016 5532 : assert(pages != NULL);
1017 5532 : assert(page_count != NULL);
1018 :
1019 5532 : *last_page = NULL;
1020 5532 : if (*page_count == 0) {
1021 5423 : assert(*pages == NULL);
1022 5423 : *pages = spdk_malloc(blob->bs->md_page_size, 0,
1023 : NULL, SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
1024 5423 : if (*pages == NULL) {
1025 0 : return -ENOMEM;
1026 : }
1027 5423 : *page_count = 1;
1028 : } else {
1029 109 : assert(*pages != NULL);
1030 109 : tmp_pages = spdk_realloc(*pages, blob->bs->md_page_size * (*page_count + 1), 0);
1031 109 : if (tmp_pages == NULL) {
1032 0 : return -ENOMEM;
1033 : }
1034 109 : (*page_count)++;
1035 109 : *pages = tmp_pages;
1036 : }
1037 :
1038 5532 : page = &(*pages)[*page_count - 1];
1039 5532 : memset(page, 0, sizeof(*page));
1040 5532 : page->id = blob->id;
1041 5532 : page->sequence_num = *page_count - 1;
1042 5532 : page->next = SPDK_INVALID_MD_PAGE;
1043 5532 : *last_page = page;
1044 :
1045 5532 : return 0;
1046 : }
1047 :
1048 : /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor.
1049 : * Update required_sz on both success and failure.
1050 : *
1051 : */
1052 : static int
1053 2160 : blob_serialize_xattr(const struct spdk_xattr *xattr,
1054 : uint8_t *buf, size_t buf_sz,
1055 : size_t *required_sz, bool internal)
1056 : {
1057 : struct spdk_blob_md_descriptor_xattr *desc;
1058 :
1059 4320 : *required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) +
1060 2160 : strlen(xattr->name) +
1061 2160 : xattr->value_len;
1062 :
1063 2160 : if (buf_sz < *required_sz) {
1064 60 : return -1;
1065 : }
1066 :
1067 2100 : desc = (struct spdk_blob_md_descriptor_xattr *)buf;
1068 :
1069 2100 : desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR;
1070 2100 : desc->length = sizeof(desc->name_length) +
1071 : sizeof(desc->value_length) +
1072 2100 : strlen(xattr->name) +
1073 2100 : xattr->value_len;
1074 2100 : desc->name_length = strlen(xattr->name);
1075 2100 : desc->value_length = xattr->value_len;
1076 :
1077 2100 : memcpy(desc->name, xattr->name, desc->name_length);
1078 2100 : memcpy((void *)((uintptr_t)desc->name + desc->name_length),
1079 2100 : xattr->value,
1080 2100 : desc->value_length);
1081 :
1082 2100 : return 0;
1083 : }
1084 :
1085 : static void
1086 2516 : blob_serialize_extent_table_entry(const struct spdk_blob *blob,
1087 : uint64_t start_ep, uint64_t *next_ep,
1088 : uint8_t **buf, size_t *remaining_sz)
1089 : {
1090 : struct spdk_blob_md_descriptor_extent_table *desc;
1091 : size_t cur_sz;
1092 : uint64_t i, et_idx;
1093 : uint32_t extent_page, ep_len;
1094 :
1095 : /* The buffer must have room for at least num_clusters entry */
1096 2516 : cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->num_clusters);
1097 2516 : if (*remaining_sz < cur_sz) {
1098 30 : *next_ep = start_ep;
1099 30 : return;
1100 : }
1101 :
1102 2486 : desc = (struct spdk_blob_md_descriptor_extent_table *)*buf;
1103 2486 : desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE;
1104 :
1105 2486 : desc->num_clusters = blob->active.num_clusters;
1106 :
1107 2486 : ep_len = 1;
1108 2486 : et_idx = 0;
1109 6343 : for (i = start_ep; i < blob->active.num_extent_pages; i++) {
1110 3857 : if (*remaining_sz < cur_sz + sizeof(desc->extent_page[0])) {
1111 : /* If we ran out of buffer space, return */
1112 0 : break;
1113 : }
1114 :
1115 3857 : extent_page = blob->active.extent_pages[i];
1116 : /* Verify that next extent_page is unallocated */
1117 3857 : if (extent_page == 0 &&
1118 2283 : (i + 1 < blob->active.num_extent_pages && blob->active.extent_pages[i + 1] == 0)) {
1119 1617 : ep_len++;
1120 1617 : continue;
1121 : }
1122 2240 : desc->extent_page[et_idx].page_idx = extent_page;
1123 2240 : desc->extent_page[et_idx].num_pages = ep_len;
1124 2240 : et_idx++;
1125 :
1126 2240 : ep_len = 1;
1127 2240 : cur_sz += sizeof(desc->extent_page[et_idx]);
1128 : }
1129 2486 : *next_ep = i;
1130 :
1131 2486 : desc->length = sizeof(desc->num_clusters) + sizeof(desc->extent_page[0]) * et_idx;
1132 2486 : *remaining_sz -= sizeof(struct spdk_blob_md_descriptor) + desc->length;
1133 2486 : *buf += sizeof(struct spdk_blob_md_descriptor) + desc->length;
1134 : }
1135 :
1136 : static int
1137 2489 : blob_serialize_extent_table(const struct spdk_blob *blob,
1138 : struct spdk_blob_md_page **pages,
1139 : struct spdk_blob_md_page *cur_page,
1140 : uint32_t *page_count, uint8_t **buf,
1141 : size_t *remaining_sz)
1142 : {
1143 : uint64_t last_extent_page;
1144 : int rc;
1145 :
1146 2489 : last_extent_page = 0;
1147 : /* At least single extent table entry has to be always persisted.
1148 : * Such case occurs with num_extent_pages == 0. */
1149 2516 : while (last_extent_page <= blob->active.num_extent_pages) {
1150 2516 : blob_serialize_extent_table_entry(blob, last_extent_page, &last_extent_page, buf,
1151 : remaining_sz);
1152 :
1153 2516 : if (last_extent_page == blob->active.num_extent_pages) {
1154 2489 : break;
1155 : }
1156 :
1157 27 : rc = blob_serialize_add_page(blob, pages, page_count, &cur_page);
1158 27 : if (rc < 0) {
1159 0 : return rc;
1160 : }
1161 :
1162 27 : *buf = (uint8_t *)cur_page->descriptors;
1163 27 : *remaining_sz = sizeof(cur_page->descriptors);
1164 : }
1165 :
1166 2489 : return 0;
1167 : }
1168 :
1169 : static void
1170 1751 : blob_serialize_extent_rle(const struct spdk_blob *blob,
1171 : uint64_t start_cluster, uint64_t *next_cluster,
1172 : uint8_t **buf, size_t *buf_sz)
1173 : {
1174 : struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle;
1175 : size_t cur_sz;
1176 : uint64_t i, extent_idx;
1177 : uint64_t lba, lba_per_cluster, lba_count;
1178 :
1179 : /* The buffer must have room for at least one extent */
1180 1751 : cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc_extent_rle->extents[0]);
1181 1751 : if (*buf_sz < cur_sz) {
1182 18 : *next_cluster = start_cluster;
1183 18 : return;
1184 : }
1185 :
1186 1733 : desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)*buf;
1187 1733 : desc_extent_rle->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE;
1188 :
1189 1733 : lba_per_cluster = bs_cluster_to_lba(blob->bs, 1);
1190 : /* Assert for scan-build false positive */
1191 1733 : assert(lba_per_cluster > 0);
1192 :
1193 1733 : lba = blob->active.clusters[start_cluster];
1194 1733 : lba_count = lba_per_cluster;
1195 1733 : extent_idx = 0;
1196 810464 : for (i = start_cluster + 1; i < blob->active.num_clusters; i++) {
1197 808735 : if ((lba + lba_count) == blob->active.clusters[i] && lba != 0) {
1198 : /* Run-length encode sequential non-zero LBA */
1199 7276 : lba_count += lba_per_cluster;
1200 7276 : continue;
1201 801459 : } else if (lba == 0 && blob->active.clusters[i] == 0) {
1202 : /* Run-length encode unallocated clusters */
1203 800266 : lba_count += lba_per_cluster;
1204 800266 : continue;
1205 : }
1206 1193 : desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
1207 1193 : desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster;
1208 1193 : extent_idx++;
1209 :
1210 1193 : cur_sz += sizeof(desc_extent_rle->extents[extent_idx]);
1211 :
1212 1193 : if (*buf_sz < cur_sz) {
1213 : /* If we ran out of buffer space, return */
1214 4 : *next_cluster = i;
1215 4 : break;
1216 : }
1217 :
1218 1189 : lba = blob->active.clusters[i];
1219 1189 : lba_count = lba_per_cluster;
1220 : }
1221 :
1222 1733 : if (*buf_sz >= cur_sz) {
1223 1729 : desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
1224 1729 : desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster;
1225 1729 : extent_idx++;
1226 :
1227 1729 : *next_cluster = blob->active.num_clusters;
1228 : }
1229 :
1230 1733 : desc_extent_rle->length = sizeof(desc_extent_rle->extents[0]) * extent_idx;
1231 1733 : *buf_sz -= sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length;
1232 1733 : *buf += sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length;
1233 : }
1234 :
1235 : static int
1236 1943 : blob_serialize_extents_rle(const struct spdk_blob *blob,
1237 : struct spdk_blob_md_page **pages,
1238 : struct spdk_blob_md_page *cur_page,
1239 : uint32_t *page_count, uint8_t **buf,
1240 : size_t *remaining_sz)
1241 : {
1242 : uint64_t last_cluster;
1243 : int rc;
1244 :
1245 1943 : last_cluster = 0;
1246 1965 : while (last_cluster < blob->active.num_clusters) {
1247 1751 : blob_serialize_extent_rle(blob, last_cluster, &last_cluster, buf, remaining_sz);
1248 :
1249 1751 : if (last_cluster == blob->active.num_clusters) {
1250 1729 : break;
1251 : }
1252 :
1253 22 : rc = blob_serialize_add_page(blob, pages, page_count, &cur_page);
1254 22 : if (rc < 0) {
1255 0 : return rc;
1256 : }
1257 :
1258 22 : *buf = (uint8_t *)cur_page->descriptors;
1259 22 : *remaining_sz = sizeof(cur_page->descriptors);
1260 : }
1261 :
1262 1943 : return 0;
1263 : }
1264 :
1265 : static void
1266 1648 : blob_serialize_extent_page(const struct spdk_blob *blob,
1267 : uint64_t cluster, struct spdk_blob_md_page *page)
1268 : {
1269 : struct spdk_blob_md_descriptor_extent_page *desc_extent;
1270 : uint64_t i, extent_idx;
1271 : uint64_t lba, lba_per_cluster;
1272 1648 : uint64_t start_cluster_idx = (cluster / SPDK_EXTENTS_PER_EP) * SPDK_EXTENTS_PER_EP;
1273 :
1274 1648 : desc_extent = (struct spdk_blob_md_descriptor_extent_page *) page->descriptors;
1275 1648 : desc_extent->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE;
1276 :
1277 1648 : lba_per_cluster = bs_cluster_to_lba(blob->bs, 1);
1278 :
1279 1648 : desc_extent->start_cluster_idx = start_cluster_idx;
1280 1648 : extent_idx = 0;
1281 63582 : for (i = start_cluster_idx; i < blob->active.num_clusters; i++) {
1282 62033 : lba = blob->active.clusters[i];
1283 62033 : desc_extent->cluster_idx[extent_idx++] = lba / lba_per_cluster;
1284 62033 : if (extent_idx >= SPDK_EXTENTS_PER_EP) {
1285 99 : break;
1286 : }
1287 : }
1288 1648 : desc_extent->length = sizeof(desc_extent->start_cluster_idx) +
1289 : sizeof(desc_extent->cluster_idx[0]) * extent_idx;
1290 1648 : }
1291 :
1292 : static void
1293 4432 : blob_serialize_flags(const struct spdk_blob *blob,
1294 : uint8_t *buf, size_t *buf_sz)
1295 : {
1296 : struct spdk_blob_md_descriptor_flags *desc;
1297 :
1298 : /*
1299 : * Flags get serialized first, so we should always have room for the flags
1300 : * descriptor.
1301 : */
1302 4432 : assert(*buf_sz >= sizeof(*desc));
1303 :
1304 4432 : desc = (struct spdk_blob_md_descriptor_flags *)buf;
1305 4432 : desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS;
1306 4432 : desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor);
1307 4432 : desc->invalid_flags = blob->invalid_flags;
1308 4432 : desc->data_ro_flags = blob->data_ro_flags;
1309 4432 : desc->md_ro_flags = blob->md_ro_flags;
1310 :
1311 4432 : *buf_sz -= sizeof(*desc);
1312 4432 : }
1313 :
1314 : static int
1315 8864 : blob_serialize_xattrs(const struct spdk_blob *blob,
1316 : const struct spdk_xattr_tailq *xattrs, bool internal,
1317 : struct spdk_blob_md_page **pages,
1318 : struct spdk_blob_md_page *cur_page,
1319 : uint32_t *page_count, uint8_t **buf,
1320 : size_t *remaining_sz)
1321 : {
1322 : const struct spdk_xattr *xattr;
1323 : int rc;
1324 :
1325 10964 : TAILQ_FOREACH(xattr, xattrs, link) {
1326 2100 : size_t required_sz = 0;
1327 :
1328 2100 : rc = blob_serialize_xattr(xattr,
1329 : *buf, *remaining_sz,
1330 : &required_sz, internal);
1331 2100 : if (rc < 0) {
1332 : /* Need to add a new page to the chain */
1333 60 : rc = blob_serialize_add_page(blob, pages, page_count,
1334 : &cur_page);
1335 60 : if (rc < 0) {
1336 0 : spdk_free(*pages);
1337 0 : *pages = NULL;
1338 0 : *page_count = 0;
1339 0 : return rc;
1340 : }
1341 :
1342 60 : *buf = (uint8_t *)cur_page->descriptors;
1343 60 : *remaining_sz = sizeof(cur_page->descriptors);
1344 :
1345 : /* Try again */
1346 60 : required_sz = 0;
1347 60 : rc = blob_serialize_xattr(xattr,
1348 : *buf, *remaining_sz,
1349 : &required_sz, internal);
1350 :
1351 60 : if (rc < 0) {
1352 0 : spdk_free(*pages);
1353 0 : *pages = NULL;
1354 0 : *page_count = 0;
1355 0 : return rc;
1356 : }
1357 : }
1358 :
1359 2100 : *remaining_sz -= required_sz;
1360 2100 : *buf += required_sz;
1361 : }
1362 :
1363 8864 : return 0;
1364 : }
1365 :
1366 : static int
1367 4432 : blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages,
1368 : uint32_t *page_count)
1369 : {
1370 : struct spdk_blob_md_page *cur_page;
1371 : int rc;
1372 : uint8_t *buf;
1373 : size_t remaining_sz;
1374 :
1375 4432 : assert(pages != NULL);
1376 4432 : assert(page_count != NULL);
1377 4432 : assert(blob != NULL);
1378 4432 : assert(blob->state == SPDK_BLOB_STATE_DIRTY);
1379 :
1380 4432 : *pages = NULL;
1381 4432 : *page_count = 0;
1382 :
1383 : /* A blob always has at least 1 page, even if it has no descriptors */
1384 4432 : rc = blob_serialize_add_page(blob, pages, page_count, &cur_page);
1385 4432 : if (rc < 0) {
1386 0 : return rc;
1387 : }
1388 :
1389 4432 : buf = (uint8_t *)cur_page->descriptors;
1390 4432 : remaining_sz = sizeof(cur_page->descriptors);
1391 :
1392 : /* Serialize flags */
1393 4432 : blob_serialize_flags(blob, buf, &remaining_sz);
1394 4432 : buf += sizeof(struct spdk_blob_md_descriptor_flags);
1395 :
1396 : /* Serialize xattrs */
1397 4432 : rc = blob_serialize_xattrs(blob, &blob->xattrs, false,
1398 : pages, cur_page, page_count, &buf, &remaining_sz);
1399 4432 : if (rc < 0) {
1400 0 : return rc;
1401 : }
1402 :
1403 : /* Serialize internal xattrs */
1404 4432 : rc = blob_serialize_xattrs(blob, &blob->xattrs_internal, true,
1405 : pages, cur_page, page_count, &buf, &remaining_sz);
1406 4432 : if (rc < 0) {
1407 0 : return rc;
1408 : }
1409 :
1410 4432 : if (blob->use_extent_table) {
1411 : /* Serialize extent table */
1412 2489 : rc = blob_serialize_extent_table(blob, pages, cur_page, page_count, &buf, &remaining_sz);
1413 : } else {
1414 : /* Serialize extents */
1415 1943 : rc = blob_serialize_extents_rle(blob, pages, cur_page, page_count, &buf, &remaining_sz);
1416 : }
1417 :
1418 4432 : return rc;
1419 : }
1420 :
1421 : struct spdk_blob_load_ctx {
1422 : struct spdk_blob *blob;
1423 :
1424 : struct spdk_blob_md_page *pages;
1425 : uint32_t num_pages;
1426 : uint32_t next_extent_page;
1427 : spdk_bs_sequence_t *seq;
1428 :
1429 : spdk_bs_sequence_cpl cb_fn;
1430 : void *cb_arg;
1431 : };
1432 :
1433 : static uint32_t
1434 25719 : blob_md_page_calc_crc(void *page)
1435 : {
1436 : uint32_t crc;
1437 :
1438 25719 : crc = BLOB_CRC32C_INITIAL;
1439 25719 : crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc);
1440 25719 : crc ^= BLOB_CRC32C_INITIAL;
1441 :
1442 25719 : return crc;
1443 :
1444 : }
1445 :
1446 : static void
1447 4341 : blob_load_final(struct spdk_blob_load_ctx *ctx, int bserrno)
1448 : {
1449 4341 : struct spdk_blob *blob = ctx->blob;
1450 :
1451 4341 : if (bserrno == 0) {
1452 4259 : blob_mark_clean(blob);
1453 : }
1454 :
1455 4341 : ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno);
1456 :
1457 : /* Free the memory */
1458 4341 : spdk_free(ctx->pages);
1459 4341 : free(ctx);
1460 4341 : }
1461 :
1462 : static void
1463 575 : blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
1464 : {
1465 575 : struct spdk_blob_load_ctx *ctx = cb_arg;
1466 575 : struct spdk_blob *blob = ctx->blob;
1467 :
1468 575 : if (bserrno == 0) {
1469 567 : blob->back_bs_dev = bs_create_blob_bs_dev(snapshot);
1470 567 : if (blob->back_bs_dev == NULL) {
1471 0 : bserrno = -ENOMEM;
1472 : }
1473 : }
1474 575 : if (bserrno != 0) {
1475 8 : SPDK_ERRLOG("Snapshot fail\n");
1476 : }
1477 :
1478 575 : blob_load_final(ctx, bserrno);
1479 575 : }
1480 :
1481 : static void blob_update_clear_method(struct spdk_blob *blob);
1482 :
1483 : static int
1484 150 : blob_load_esnap(struct spdk_blob *blob, void *blob_ctx)
1485 : {
1486 150 : struct spdk_blob_store *bs = blob->bs;
1487 150 : struct spdk_bs_dev *bs_dev = NULL;
1488 150 : const void *esnap_id = NULL;
1489 150 : size_t id_len = 0;
1490 : int rc;
1491 :
1492 150 : if (bs->esnap_bs_dev_create == NULL) {
1493 10 : SPDK_NOTICELOG("blob 0x%" PRIx64 " is an esnap clone but the blobstore was opened "
1494 : "without support for esnap clones\n", blob->id);
1495 10 : return -ENOTSUP;
1496 : }
1497 140 : assert(blob->back_bs_dev == NULL);
1498 :
1499 140 : rc = blob_get_xattr_value(blob, BLOB_EXTERNAL_SNAPSHOT_ID, &esnap_id, &id_len, true);
1500 140 : if (rc != 0) {
1501 0 : SPDK_ERRLOG("blob 0x%" PRIx64 " is an esnap clone but has no esnap ID\n", blob->id);
1502 0 : return -EINVAL;
1503 : }
1504 140 : assert(id_len > 0 && id_len < UINT32_MAX);
1505 :
1506 140 : SPDK_INFOLOG(blob, "Creating external snapshot device\n");
1507 :
1508 140 : rc = bs->esnap_bs_dev_create(bs->esnap_ctx, blob_ctx, blob, esnap_id, (uint32_t)id_len,
1509 : &bs_dev);
1510 140 : if (rc != 0) {
1511 0 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64 ": failed to load back_bs_dev "
1512 : "with error %d\n", blob->id, rc);
1513 0 : return rc;
1514 : }
1515 :
1516 : /*
1517 : * Note: bs_dev might be NULL if the consumer chose to not open the external snapshot.
1518 : * This especially might happen during spdk_bs_load() iteration.
1519 : */
1520 140 : if (bs_dev != NULL) {
1521 140 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64 ": loaded back_bs_dev\n", blob->id);
1522 140 : if ((bs->io_unit_size % bs_dev->blocklen) != 0) {
1523 5 : SPDK_NOTICELOG("blob 0x%" PRIx64 " external snapshot device block size %u "
1524 : "is not compatible with blobstore block size %u\n",
1525 : blob->id, bs_dev->blocklen, bs->io_unit_size);
1526 5 : bs_dev->destroy(bs_dev);
1527 5 : return -EINVAL;
1528 : }
1529 : }
1530 :
1531 135 : blob->back_bs_dev = bs_dev;
1532 135 : blob->parent_id = SPDK_BLOBID_EXTERNAL_SNAPSHOT;
1533 :
1534 135 : return 0;
1535 : }
1536 :
1537 : static void
1538 4282 : blob_load_backing_dev(spdk_bs_sequence_t *seq, void *cb_arg)
1539 : {
1540 4282 : struct spdk_blob_load_ctx *ctx = cb_arg;
1541 4282 : struct spdk_blob *blob = ctx->blob;
1542 : const void *value;
1543 : size_t len;
1544 : int rc;
1545 :
1546 4282 : if (blob_is_esnap_clone(blob)) {
1547 150 : rc = blob_load_esnap(blob, seq->cpl.u.blob_handle.esnap_ctx);
1548 150 : blob_load_final(ctx, rc);
1549 725 : return;
1550 : }
1551 :
1552 4132 : if (spdk_blob_is_thin_provisioned(blob)) {
1553 1301 : rc = blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true);
1554 1301 : if (rc == 0) {
1555 575 : if (len != sizeof(spdk_blob_id)) {
1556 0 : blob_load_final(ctx, -EINVAL);
1557 0 : return;
1558 : }
1559 : /* open snapshot blob and continue in the callback function */
1560 575 : blob->parent_id = *(spdk_blob_id *)value;
1561 575 : spdk_bs_open_blob(blob->bs, blob->parent_id,
1562 : blob_load_snapshot_cpl, ctx);
1563 575 : return;
1564 : } else {
1565 : /* add zeroes_dev for thin provisioned blob */
1566 726 : blob->back_bs_dev = bs_create_zeroes_dev();
1567 : }
1568 : } else {
1569 : /* standard blob */
1570 2831 : blob->back_bs_dev = NULL;
1571 : }
1572 3557 : blob_load_final(ctx, 0);
1573 : }
1574 :
1575 : static void
1576 4189 : blob_load_cpl_extents_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1577 : {
1578 4189 : struct spdk_blob_load_ctx *ctx = cb_arg;
1579 4189 : struct spdk_blob *blob = ctx->blob;
1580 : struct spdk_blob_md_page *page;
1581 : uint64_t i;
1582 : uint32_t crc;
1583 : uint64_t lba;
1584 : void *tmp;
1585 : uint64_t sz;
1586 :
1587 4189 : if (bserrno) {
1588 9 : SPDK_ERRLOG("Extent page read failed: %d\n", bserrno);
1589 9 : blob_load_final(ctx, bserrno);
1590 9 : return;
1591 : }
1592 :
1593 4180 : if (ctx->pages == NULL) {
1594 : /* First iteration of this function, allocate buffer for single EXTENT_PAGE */
1595 2624 : ctx->pages = spdk_zmalloc(blob->bs->md_page_size, 0,
1596 : NULL, SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
1597 2624 : if (!ctx->pages) {
1598 0 : blob_load_final(ctx, -ENOMEM);
1599 0 : return;
1600 : }
1601 2624 : ctx->num_pages = 1;
1602 2624 : ctx->next_extent_page = 0;
1603 : } else {
1604 1556 : page = &ctx->pages[0];
1605 1556 : crc = blob_md_page_calc_crc(page);
1606 1556 : if (crc != page->crc) {
1607 0 : blob_load_final(ctx, -EINVAL);
1608 0 : return;
1609 : }
1610 :
1611 1556 : if (page->next != SPDK_INVALID_MD_PAGE) {
1612 0 : blob_load_final(ctx, -EINVAL);
1613 0 : return;
1614 : }
1615 :
1616 1556 : bserrno = blob_parse_extent_page(page, blob);
1617 1556 : if (bserrno) {
1618 0 : blob_load_final(ctx, bserrno);
1619 0 : return;
1620 : }
1621 : }
1622 :
1623 4816 : for (i = ctx->next_extent_page; i < blob->active.num_extent_pages; i++) {
1624 2201 : if (blob->active.extent_pages[i] != 0) {
1625 : /* Extent page was allocated, read and parse it. */
1626 1565 : lba = bs_md_page_to_lba(blob->bs, blob->active.extent_pages[i]);
1627 1565 : ctx->next_extent_page = i + 1;
1628 :
1629 1565 : bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1630 1565 : bs_byte_to_lba(blob->bs, blob->bs->md_page_size),
1631 : blob_load_cpl_extents_cpl, ctx);
1632 1565 : return;
1633 : } else {
1634 : /* Thin provisioned blobs can point to unallocated extent pages.
1635 : * In this case blob size should be increased by up to the amount left in remaining_clusters_in_et. */
1636 :
1637 636 : sz = spdk_min(blob->remaining_clusters_in_et, SPDK_EXTENTS_PER_EP);
1638 636 : blob->active.num_clusters += sz;
1639 636 : blob->remaining_clusters_in_et -= sz;
1640 :
1641 636 : assert(spdk_blob_is_thin_provisioned(blob));
1642 636 : assert(i + 1 < blob->active.num_extent_pages || blob->remaining_clusters_in_et == 0);
1643 :
1644 636 : tmp = realloc(blob->active.clusters, blob->active.num_clusters * sizeof(*blob->active.clusters));
1645 636 : if (tmp == NULL) {
1646 0 : blob_load_final(ctx, -ENOMEM);
1647 0 : return;
1648 : }
1649 636 : memset(tmp + sizeof(*blob->active.clusters) * blob->active.cluster_array_size, 0,
1650 636 : sizeof(*blob->active.clusters) * (blob->active.num_clusters - blob->active.cluster_array_size));
1651 636 : blob->active.clusters = tmp;
1652 636 : blob->active.cluster_array_size = blob->active.num_clusters;
1653 : }
1654 : }
1655 :
1656 2615 : blob_load_backing_dev(seq, ctx);
1657 : }
1658 :
1659 : static void
1660 4464 : blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1661 : {
1662 4464 : struct spdk_blob_load_ctx *ctx = cb_arg;
1663 4464 : struct spdk_blob *blob = ctx->blob;
1664 : struct spdk_blob_md_page *page;
1665 : int rc;
1666 : uint32_t crc;
1667 : uint32_t current_page;
1668 :
1669 4464 : if (ctx->num_pages == 1) {
1670 4341 : current_page = bs_blobid_to_page(blob->id);
1671 : } else {
1672 123 : assert(ctx->num_pages != 0);
1673 123 : page = &ctx->pages[ctx->num_pages - 2];
1674 123 : current_page = page->next;
1675 : }
1676 :
1677 4464 : if (bserrno) {
1678 25 : SPDK_ERRLOG("Metadata page %d read failed for blobid 0x%" PRIx64 ": %d\n",
1679 : current_page, blob->id, bserrno);
1680 25 : blob_load_final(ctx, bserrno);
1681 25 : return;
1682 : }
1683 :
1684 4439 : page = &ctx->pages[ctx->num_pages - 1];
1685 4439 : crc = blob_md_page_calc_crc(page);
1686 4439 : if (crc != page->crc) {
1687 10 : SPDK_ERRLOG("Metadata page %d crc mismatch for blobid 0x%" PRIx64 "\n",
1688 : current_page, blob->id);
1689 10 : blob_load_final(ctx, -EINVAL);
1690 10 : return;
1691 : }
1692 :
1693 4429 : if (page->next != SPDK_INVALID_MD_PAGE) {
1694 : struct spdk_blob_md_page *tmp_pages;
1695 123 : uint32_t next_page = page->next;
1696 123 : uint64_t next_lba = bs_md_page_to_lba(blob->bs, next_page);
1697 :
1698 : /* Read the next page */
1699 123 : tmp_pages = spdk_realloc(ctx->pages, (sizeof(*page) * (ctx->num_pages + 1)), 0);
1700 123 : if (tmp_pages == NULL) {
1701 0 : blob_load_final(ctx, -ENOMEM);
1702 0 : return;
1703 : }
1704 123 : ctx->num_pages++;
1705 123 : ctx->pages = tmp_pages;
1706 :
1707 123 : bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1],
1708 : next_lba,
1709 123 : bs_byte_to_lba(blob->bs, sizeof(*page)),
1710 : blob_load_cpl, ctx);
1711 123 : return;
1712 : }
1713 :
1714 : /* Parse the pages */
1715 4306 : rc = blob_parse(ctx->pages, ctx->num_pages, blob);
1716 4306 : if (rc) {
1717 15 : blob_load_final(ctx, rc);
1718 15 : return;
1719 : }
1720 :
1721 4291 : if (blob->extent_table_found == true) {
1722 : /* If EXTENT_TABLE was found, that means support for it should be enabled. */
1723 2624 : assert(blob->extent_rle_found == false);
1724 2624 : blob->use_extent_table = true;
1725 : } else {
1726 : /* If EXTENT_RLE or no extent_* descriptor was found disable support
1727 : * for extent table. No extent_* descriptors means that blob has length of 0
1728 : * and no extent_rle descriptors were persisted for it.
1729 : * EXTENT_TABLE if used, is always present in metadata regardless of length. */
1730 1667 : blob->use_extent_table = false;
1731 : }
1732 :
1733 : /* Check the clear_method stored in metadata vs what may have been passed
1734 : * via spdk_bs_open_blob_ext() and update accordingly.
1735 : */
1736 4291 : blob_update_clear_method(blob);
1737 :
1738 4291 : spdk_free(ctx->pages);
1739 4291 : ctx->pages = NULL;
1740 :
1741 4291 : if (blob->extent_table_found) {
1742 2624 : blob_load_cpl_extents_cpl(seq, ctx, 0);
1743 : } else {
1744 1667 : blob_load_backing_dev(seq, ctx);
1745 : }
1746 : }
1747 :
1748 : /* Load a blob from disk given a blobid */
1749 : static void
1750 4341 : blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1751 : spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1752 : {
1753 : struct spdk_blob_load_ctx *ctx;
1754 : struct spdk_blob_store *bs;
1755 : uint32_t page_num;
1756 : uint64_t lba;
1757 :
1758 4341 : blob_verify_md_op(blob);
1759 :
1760 4341 : bs = blob->bs;
1761 :
1762 4341 : ctx = calloc(1, sizeof(*ctx));
1763 4341 : if (!ctx) {
1764 0 : cb_fn(seq, cb_arg, -ENOMEM);
1765 0 : return;
1766 : }
1767 :
1768 4341 : ctx->blob = blob;
1769 4341 : ctx->pages = spdk_realloc(ctx->pages, bs->md_page_size, 0);
1770 4341 : if (!ctx->pages) {
1771 0 : free(ctx);
1772 0 : cb_fn(seq, cb_arg, -ENOMEM);
1773 0 : return;
1774 : }
1775 4341 : ctx->num_pages = 1;
1776 4341 : ctx->cb_fn = cb_fn;
1777 4341 : ctx->cb_arg = cb_arg;
1778 4341 : ctx->seq = seq;
1779 :
1780 4341 : page_num = bs_blobid_to_page(blob->id);
1781 4341 : lba = bs_md_page_to_lba(blob->bs, page_num);
1782 :
1783 4341 : blob->state = SPDK_BLOB_STATE_LOADING;
1784 :
1785 4341 : bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1786 4341 : bs_byte_to_lba(bs, bs->md_page_size),
1787 : blob_load_cpl, ctx);
1788 : }
1789 :
1790 : struct spdk_blob_persist_ctx {
1791 : struct spdk_blob *blob;
1792 :
1793 : struct spdk_blob_md_page *pages;
1794 : uint32_t next_extent_page;
1795 : struct spdk_blob_md_page *extent_page;
1796 :
1797 : spdk_bs_sequence_t *seq;
1798 : spdk_bs_sequence_cpl cb_fn;
1799 : void *cb_arg;
1800 : TAILQ_ENTRY(spdk_blob_persist_ctx) link;
1801 : };
1802 :
1803 : static void
1804 1584 : bs_batch_clear_dev(struct spdk_blob *blob, spdk_bs_batch_t *batch, uint64_t lba,
1805 : uint64_t lba_count)
1806 : {
1807 1584 : switch (blob->clear_method) {
1808 1584 : case BLOB_CLEAR_WITH_DEFAULT:
1809 : case BLOB_CLEAR_WITH_UNMAP:
1810 1584 : bs_batch_unmap_dev(batch, lba, lba_count);
1811 1584 : break;
1812 0 : case BLOB_CLEAR_WITH_WRITE_ZEROES:
1813 0 : bs_batch_write_zeroes_dev(batch, lba, lba_count);
1814 0 : break;
1815 0 : case BLOB_CLEAR_WITH_NONE:
1816 : default:
1817 0 : break;
1818 : }
1819 1584 : }
1820 :
1821 : static int
1822 1437 : bs_super_validate(struct spdk_bs_super_block *super, struct spdk_blob_store *bs)
1823 : {
1824 : uint32_t crc;
1825 : static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
1826 :
1827 1437 : if (super->version > SPDK_BS_VERSION ||
1828 1432 : super->version < SPDK_BS_INITIAL_VERSION) {
1829 10 : return -EILSEQ;
1830 : }
1831 :
1832 1427 : if (memcmp(super->signature, SPDK_BS_SUPER_BLOCK_SIG,
1833 : sizeof(super->signature)) != 0) {
1834 0 : return -EILSEQ;
1835 : }
1836 :
1837 1427 : crc = blob_md_page_calc_crc(super);
1838 1427 : if (crc != super->crc) {
1839 5 : return -EILSEQ;
1840 : }
1841 :
1842 1422 : if (memcmp(&bs->bstype, &super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
1843 1405 : SPDK_DEBUGLOG(blob, "Bstype matched - loading blobstore\n");
1844 17 : } else if (memcmp(&bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
1845 7 : SPDK_DEBUGLOG(blob, "Bstype wildcard used - loading blobstore regardless bstype\n");
1846 : } else {
1847 10 : SPDK_DEBUGLOG(blob, "Unexpected bstype\n");
1848 10 : SPDK_LOGDUMP(blob, "Expected:", bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
1849 10 : SPDK_LOGDUMP(blob, "Found:", super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
1850 10 : return -ENXIO;
1851 : }
1852 :
1853 1412 : if (super->size > bs->dev->blockcnt * bs->dev->blocklen) {
1854 10 : SPDK_NOTICELOG("Size mismatch, dev size: %" PRIu64 ", blobstore size: %" PRIu64 "\n",
1855 : bs->dev->blockcnt * bs->dev->blocklen, super->size);
1856 10 : return -EILSEQ;
1857 : }
1858 :
1859 1402 : return 0;
1860 : }
1861 :
1862 : static void bs_mark_dirty(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
1863 : spdk_bs_sequence_cpl cb_fn, void *cb_arg);
1864 :
1865 : static void
1866 6304 : blob_persist_complete_cb(void *arg)
1867 : {
1868 6304 : struct spdk_blob_persist_ctx *ctx = arg;
1869 :
1870 : /* Call user callback */
1871 6304 : ctx->cb_fn(ctx->seq, ctx->cb_arg, 0);
1872 :
1873 : /* Free the memory */
1874 6304 : spdk_free(ctx->pages);
1875 6304 : free(ctx);
1876 6304 : }
1877 :
1878 : static void blob_persist_start(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno);
1879 :
1880 : static void
1881 6304 : blob_persist_complete(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx, int bserrno)
1882 : {
1883 : struct spdk_blob_persist_ctx *next_persist, *tmp;
1884 6304 : struct spdk_blob *blob = ctx->blob;
1885 :
1886 6304 : if (bserrno == 0) {
1887 6239 : blob_mark_clean(blob);
1888 : }
1889 :
1890 6304 : assert(ctx == TAILQ_FIRST(&blob->persists_to_complete));
1891 :
1892 : /* Complete all persists that were pending when the current persist started */
1893 12608 : TAILQ_FOREACH_SAFE(next_persist, &blob->persists_to_complete, link, tmp) {
1894 6304 : TAILQ_REMOVE(&blob->persists_to_complete, next_persist, link);
1895 6304 : spdk_thread_send_msg(spdk_get_thread(), blob_persist_complete_cb, next_persist);
1896 : }
1897 :
1898 6304 : if (TAILQ_EMPTY(&blob->pending_persists)) {
1899 6276 : return;
1900 : }
1901 :
1902 : /* Queue up all pending persists for completion and start blob persist with first one */
1903 28 : TAILQ_SWAP(&blob->persists_to_complete, &blob->pending_persists, spdk_blob_persist_ctx, link);
1904 28 : next_persist = TAILQ_FIRST(&blob->persists_to_complete);
1905 :
1906 28 : blob->state = SPDK_BLOB_STATE_DIRTY;
1907 28 : bs_mark_dirty(seq, blob->bs, blob_persist_start, next_persist);
1908 : }
1909 :
1910 : static void
1911 6239 : blob_persist_clear_extents_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1912 : {
1913 6239 : struct spdk_blob_persist_ctx *ctx = cb_arg;
1914 6239 : struct spdk_blob *blob = ctx->blob;
1915 6239 : struct spdk_blob_store *bs = blob->bs;
1916 : size_t i;
1917 :
1918 6239 : if (bserrno != 0) {
1919 0 : blob_persist_complete(seq, ctx, bserrno);
1920 0 : return;
1921 : }
1922 :
1923 6239 : spdk_spin_lock(&bs->used_lock);
1924 :
1925 : /* Release all extent_pages that were truncated */
1926 8837 : for (i = blob->active.num_extent_pages; i < blob->active.extent_pages_array_size; i++) {
1927 : /* Nothing to release if it was not allocated */
1928 2598 : if (blob->active.extent_pages[i] != 0) {
1929 936 : bs_release_md_page(bs, blob->active.extent_pages[i]);
1930 : }
1931 : }
1932 :
1933 6239 : spdk_spin_unlock(&bs->used_lock);
1934 :
1935 6239 : if (blob->active.num_extent_pages == 0) {
1936 4134 : free(blob->active.extent_pages);
1937 4134 : blob->active.extent_pages = NULL;
1938 4134 : blob->active.extent_pages_array_size = 0;
1939 2105 : } else if (blob->active.num_extent_pages != blob->active.extent_pages_array_size) {
1940 : #ifndef __clang_analyzer__
1941 : void *tmp;
1942 :
1943 : /* scan-build really can't figure reallocs, workaround it */
1944 3 : tmp = realloc(blob->active.extent_pages, sizeof(uint32_t) * blob->active.num_extent_pages);
1945 3 : assert(tmp != NULL);
1946 3 : blob->active.extent_pages = tmp;
1947 : #endif
1948 3 : blob->active.extent_pages_array_size = blob->active.num_extent_pages;
1949 : }
1950 :
1951 6239 : blob_persist_complete(seq, ctx, bserrno);
1952 : }
1953 :
1954 : static void
1955 6239 : blob_persist_clear_extents(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx)
1956 : {
1957 6239 : struct spdk_blob *blob = ctx->blob;
1958 6239 : struct spdk_blob_store *bs = blob->bs;
1959 : size_t i;
1960 : uint64_t lba;
1961 : uint64_t lba_count;
1962 : spdk_bs_batch_t *batch;
1963 :
1964 6239 : batch = bs_sequence_to_batch(seq, blob_persist_clear_extents_cpl, ctx);
1965 6239 : lba_count = bs_byte_to_lba(bs, bs->md_page_size);
1966 :
1967 : /* Clear all extent_pages that were truncated */
1968 8837 : for (i = blob->active.num_extent_pages; i < blob->active.extent_pages_array_size; i++) {
1969 : /* Nothing to clear if it was not allocated */
1970 2598 : if (blob->active.extent_pages[i] != 0) {
1971 936 : lba = bs_md_page_to_lba(bs, blob->active.extent_pages[i]);
1972 936 : bs_batch_write_zeroes_dev(batch, lba, lba_count);
1973 : }
1974 : }
1975 :
1976 6239 : bs_batch_close(batch);
1977 6239 : }
1978 :
1979 : static void
1980 6239 : blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1981 : {
1982 6239 : struct spdk_blob_persist_ctx *ctx = cb_arg;
1983 6239 : struct spdk_blob *blob = ctx->blob;
1984 6239 : struct spdk_blob_store *bs = blob->bs;
1985 : size_t i;
1986 :
1987 6239 : if (bserrno != 0) {
1988 0 : blob_persist_complete(seq, ctx, bserrno);
1989 0 : return;
1990 : }
1991 :
1992 6239 : spdk_spin_lock(&bs->used_lock);
1993 : /* Release all clusters that were truncated */
1994 1342490 : for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1995 1336251 : uint32_t cluster_num = bs_lba_to_cluster(bs, blob->active.clusters[i]);
1996 :
1997 : /* Nothing to release if it was not allocated */
1998 1336251 : if (blob->active.clusters[i] != 0) {
1999 2926 : bs_release_cluster(bs, cluster_num);
2000 : }
2001 : }
2002 6239 : spdk_spin_unlock(&bs->used_lock);
2003 :
2004 6239 : if (blob->active.num_clusters == 0) {
2005 2423 : free(blob->active.clusters);
2006 2423 : blob->active.clusters = NULL;
2007 2423 : blob->active.cluster_array_size = 0;
2008 3816 : } else if (blob->active.num_clusters != blob->active.cluster_array_size) {
2009 : #ifndef __clang_analyzer__
2010 : void *tmp;
2011 :
2012 : /* scan-build really can't figure reallocs, workaround it */
2013 22 : tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * blob->active.num_clusters);
2014 22 : assert(tmp != NULL);
2015 22 : blob->active.clusters = tmp;
2016 :
2017 : #endif
2018 22 : blob->active.cluster_array_size = blob->active.num_clusters;
2019 : }
2020 :
2021 : /* Move on to clearing extent pages */
2022 6239 : blob_persist_clear_extents(seq, ctx);
2023 : }
2024 :
2025 : static void
2026 6239 : blob_persist_clear_clusters(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx)
2027 : {
2028 6239 : struct spdk_blob *blob = ctx->blob;
2029 6239 : struct spdk_blob_store *bs = blob->bs;
2030 : spdk_bs_batch_t *batch;
2031 : size_t i;
2032 : uint64_t lba;
2033 : uint64_t lba_count;
2034 :
2035 : /* Clusters don't move around in blobs. The list shrinks or grows
2036 : * at the end, but no changes ever occur in the middle of the list.
2037 : */
2038 :
2039 6239 : batch = bs_sequence_to_batch(seq, blob_persist_clear_clusters_cpl, ctx);
2040 :
2041 : /* Clear all clusters that were truncated */
2042 6239 : lba = 0;
2043 6239 : lba_count = 0;
2044 1342490 : for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
2045 1336251 : uint64_t next_lba = blob->active.clusters[i];
2046 1336251 : uint64_t next_lba_count = bs_cluster_to_lba(bs, 1);
2047 :
2048 1336251 : if (next_lba > 0 && (lba + lba_count) == next_lba) {
2049 : /* This cluster is contiguous with the previous one. */
2050 1347 : lba_count += next_lba_count;
2051 1347 : continue;
2052 1334904 : } else if (next_lba == 0) {
2053 1333325 : continue;
2054 : }
2055 :
2056 : /* This cluster is not contiguous with the previous one. */
2057 :
2058 : /* If a run of LBAs previously existing, clear them now */
2059 1579 : if (lba_count > 0) {
2060 45 : bs_batch_clear_dev(ctx->blob, batch, lba, lba_count);
2061 : }
2062 :
2063 : /* Start building the next batch */
2064 1579 : lba = next_lba;
2065 1579 : if (next_lba > 0) {
2066 1579 : lba_count = next_lba_count;
2067 : } else {
2068 0 : lba_count = 0;
2069 : }
2070 : }
2071 :
2072 : /* If we ended with a contiguous set of LBAs, clear them now */
2073 6239 : if (lba_count > 0) {
2074 1534 : bs_batch_clear_dev(ctx->blob, batch, lba, lba_count);
2075 : }
2076 :
2077 6239 : bs_batch_close(batch);
2078 6239 : }
2079 :
2080 : static void
2081 6244 : blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2082 : {
2083 6244 : struct spdk_blob_persist_ctx *ctx = cb_arg;
2084 6244 : struct spdk_blob *blob = ctx->blob;
2085 6244 : struct spdk_blob_store *bs = blob->bs;
2086 : size_t i;
2087 :
2088 6244 : if (bserrno != 0) {
2089 5 : blob_persist_complete(seq, ctx, bserrno);
2090 5 : return;
2091 : }
2092 :
2093 6239 : spdk_spin_lock(&bs->used_lock);
2094 :
2095 : /* This loop starts at 1 because the first page is special and handled
2096 : * below. The pages (except the first) are never written in place,
2097 : * so any pages in the clean list must be zeroed.
2098 : */
2099 6324 : for (i = 1; i < blob->clean.num_pages; i++) {
2100 85 : bs_release_md_page(bs, blob->clean.pages[i]);
2101 : }
2102 :
2103 6239 : if (blob->active.num_pages == 0) {
2104 : uint32_t page_num;
2105 :
2106 1857 : page_num = bs_blobid_to_page(blob->id);
2107 1857 : bs_release_md_page(bs, page_num);
2108 : }
2109 :
2110 6239 : spdk_spin_unlock(&bs->used_lock);
2111 :
2112 : /* Move on to clearing clusters */
2113 6239 : blob_persist_clear_clusters(seq, ctx);
2114 : }
2115 :
2116 : static void
2117 6294 : blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2118 : {
2119 6294 : struct spdk_blob_persist_ctx *ctx = cb_arg;
2120 6294 : struct spdk_blob *blob = ctx->blob;
2121 6294 : struct spdk_blob_store *bs = blob->bs;
2122 : uint64_t lba;
2123 : uint64_t lba_count;
2124 : spdk_bs_batch_t *batch;
2125 : size_t i;
2126 :
2127 6294 : if (bserrno != 0) {
2128 50 : blob_persist_complete(seq, ctx, bserrno);
2129 50 : return;
2130 : }
2131 :
2132 6244 : batch = bs_sequence_to_batch(seq, blob_persist_zero_pages_cpl, ctx);
2133 :
2134 6244 : lba_count = bs_byte_to_lba(bs, bs->md_page_size);
2135 :
2136 : /* This loop starts at 1 because the first page is special and handled
2137 : * below. The pages (except the first) are never written in place,
2138 : * so any pages in the clean list must be zeroed.
2139 : */
2140 6329 : for (i = 1; i < blob->clean.num_pages; i++) {
2141 85 : lba = bs_md_page_to_lba(bs, blob->clean.pages[i]);
2142 :
2143 85 : bs_batch_write_zeroes_dev(batch, lba, lba_count);
2144 : }
2145 :
2146 : /* The first page will only be zeroed if this is a delete. */
2147 6244 : if (blob->active.num_pages == 0) {
2148 : uint32_t page_num;
2149 :
2150 : /* The first page in the metadata goes where the blobid indicates */
2151 1862 : page_num = bs_blobid_to_page(blob->id);
2152 1862 : lba = bs_md_page_to_lba(bs, page_num);
2153 :
2154 1862 : bs_batch_write_zeroes_dev(batch, lba, lba_count);
2155 : }
2156 :
2157 6244 : bs_batch_close(batch);
2158 : }
2159 :
2160 : static void
2161 4432 : blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2162 : {
2163 4432 : struct spdk_blob_persist_ctx *ctx = cb_arg;
2164 4432 : struct spdk_blob *blob = ctx->blob;
2165 4432 : struct spdk_blob_store *bs = blob->bs;
2166 : uint64_t lba;
2167 : uint32_t lba_count;
2168 : struct spdk_blob_md_page *page;
2169 :
2170 4432 : if (bserrno != 0) {
2171 0 : blob_persist_complete(seq, ctx, bserrno);
2172 0 : return;
2173 : }
2174 :
2175 4432 : if (blob->active.num_pages == 0) {
2176 : /* Move on to the next step */
2177 0 : blob_persist_zero_pages(seq, ctx, 0);
2178 0 : return;
2179 : }
2180 :
2181 4432 : lba_count = bs_byte_to_lba(bs, bs->md_page_size);
2182 :
2183 4432 : page = &ctx->pages[0];
2184 : /* The first page in the metadata goes where the blobid indicates */
2185 4432 : lba = bs_md_page_to_lba(bs, bs_blobid_to_page(blob->id));
2186 :
2187 4432 : bs_sequence_write_dev(seq, page, lba, lba_count,
2188 : blob_persist_zero_pages, ctx);
2189 : }
2190 :
2191 : static void
2192 4432 : blob_persist_write_page_chain(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx)
2193 : {
2194 4432 : struct spdk_blob *blob = ctx->blob;
2195 4432 : struct spdk_blob_store *bs = blob->bs;
2196 : uint64_t lba;
2197 : uint32_t lba_count;
2198 : struct spdk_blob_md_page *page;
2199 : spdk_bs_batch_t *batch;
2200 : size_t i;
2201 :
2202 : /* Clusters don't move around in blobs. The list shrinks or grows
2203 : * at the end, but no changes ever occur in the middle of the list.
2204 : */
2205 :
2206 4432 : lba_count = bs_byte_to_lba(bs, sizeof(*page));
2207 :
2208 4432 : batch = bs_sequence_to_batch(seq, blob_persist_write_page_root, ctx);
2209 :
2210 : /* This starts at 1. The root page is not written until
2211 : * all of the others are finished
2212 : */
2213 4541 : for (i = 1; i < blob->active.num_pages; i++) {
2214 109 : page = &ctx->pages[i];
2215 109 : assert(page->sequence_num == i);
2216 :
2217 109 : lba = bs_md_page_to_lba(bs, blob->active.pages[i]);
2218 :
2219 109 : bs_batch_write_dev(batch, page, lba, lba_count);
2220 : }
2221 :
2222 4432 : bs_batch_close(batch);
2223 4432 : }
2224 :
2225 : static int
2226 4465 : blob_resize(struct spdk_blob *blob, uint64_t sz)
2227 : {
2228 : uint64_t i;
2229 : uint64_t *tmp;
2230 : uint64_t cluster;
2231 : uint32_t lfmd; /* lowest free md page */
2232 : uint64_t num_clusters;
2233 : uint32_t *ep_tmp;
2234 4465 : uint64_t new_num_ep = 0, current_num_ep = 0;
2235 : struct spdk_blob_store *bs;
2236 : int rc;
2237 :
2238 4465 : bs = blob->bs;
2239 :
2240 4465 : blob_verify_md_op(blob);
2241 :
2242 4465 : if (blob->active.num_clusters == sz) {
2243 566 : return 0;
2244 : }
2245 :
2246 3899 : if (blob->active.num_clusters < blob->active.cluster_array_size) {
2247 : /* If this blob was resized to be larger, then smaller, then
2248 : * larger without syncing, then the cluster array already
2249 : * contains spare assigned clusters we can use.
2250 : */
2251 0 : num_clusters = spdk_min(blob->active.cluster_array_size,
2252 : sz);
2253 : } else {
2254 3899 : num_clusters = blob->active.num_clusters;
2255 : }
2256 :
2257 3899 : if (blob->use_extent_table) {
2258 : /* Round up since every cluster beyond current Extent Table size,
2259 : * requires new extent page. */
2260 2359 : new_num_ep = spdk_divide_round_up(sz, SPDK_EXTENTS_PER_EP);
2261 2359 : current_num_ep = spdk_divide_round_up(num_clusters, SPDK_EXTENTS_PER_EP);
2262 : }
2263 :
2264 3899 : assert(!spdk_spin_held(&bs->used_lock));
2265 :
2266 : /* Check first that we have enough clusters and md pages before we start claiming them.
2267 : * bs->used_lock is held to ensure that clusters we think are free are still free when we go
2268 : * to claim them later in this function.
2269 : */
2270 3899 : if (sz > num_clusters && spdk_blob_is_thin_provisioned(blob) == false) {
2271 1624 : spdk_spin_lock(&bs->used_lock);
2272 1624 : if ((sz - num_clusters) > bs->num_free_clusters) {
2273 10 : rc = -ENOSPC;
2274 10 : goto out;
2275 : }
2276 1614 : lfmd = 0;
2277 2572 : for (i = current_num_ep; i < new_num_ep ; i++) {
2278 958 : lfmd = spdk_bit_array_find_first_clear(blob->bs->used_md_pages, lfmd);
2279 958 : if (lfmd == UINT32_MAX) {
2280 : /* No more free md pages. Cannot satisfy the request */
2281 0 : rc = -ENOSPC;
2282 0 : goto out;
2283 : }
2284 : }
2285 : }
2286 :
2287 3889 : if (sz > num_clusters) {
2288 : /* Expand the cluster array if necessary.
2289 : * We only shrink the array when persisting.
2290 : */
2291 2130 : tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * sz);
2292 2130 : if (sz > 0 && tmp == NULL) {
2293 0 : rc = -ENOMEM;
2294 0 : goto out;
2295 : }
2296 2130 : memset(tmp + blob->active.cluster_array_size, 0,
2297 2130 : sizeof(*blob->active.clusters) * (sz - blob->active.cluster_array_size));
2298 2130 : blob->active.clusters = tmp;
2299 2130 : blob->active.cluster_array_size = sz;
2300 :
2301 : /* Expand the extents table, only if enough clusters were added */
2302 2130 : if (new_num_ep > current_num_ep && blob->use_extent_table) {
2303 1255 : ep_tmp = realloc(blob->active.extent_pages, sizeof(*blob->active.extent_pages) * new_num_ep);
2304 1255 : if (new_num_ep > 0 && ep_tmp == NULL) {
2305 0 : rc = -ENOMEM;
2306 0 : goto out;
2307 : }
2308 1255 : memset(ep_tmp + blob->active.extent_pages_array_size, 0,
2309 1255 : sizeof(*blob->active.extent_pages) * (new_num_ep - blob->active.extent_pages_array_size));
2310 1255 : blob->active.extent_pages = ep_tmp;
2311 1255 : blob->active.extent_pages_array_size = new_num_ep;
2312 : }
2313 : }
2314 :
2315 3889 : blob->state = SPDK_BLOB_STATE_DIRTY;
2316 :
2317 3889 : if (spdk_blob_is_thin_provisioned(blob) == false) {
2318 3028 : cluster = 0;
2319 3028 : lfmd = 0;
2320 12281 : for (i = num_clusters; i < sz; i++) {
2321 9253 : bs_allocate_cluster(blob, i, &cluster, &lfmd, true);
2322 : /* Do not increment lfmd here. lfmd will get updated
2323 : * to the md_page allocated (if any) when a new extent
2324 : * page is needed. Just pass that value again,
2325 : * bs_allocate_cluster will just start at that index
2326 : * to find the next free md_page when needed.
2327 : */
2328 : }
2329 : }
2330 :
2331 : /* If we are shrinking the blob, we must adjust num_allocated_clusters */
2332 1340190 : for (i = sz; i < num_clusters; i++) {
2333 1336301 : if (blob->active.clusters[i] != 0) {
2334 2926 : blob->active.num_allocated_clusters--;
2335 : }
2336 : }
2337 :
2338 3889 : blob->active.num_clusters = sz;
2339 3889 : blob->active.num_extent_pages = new_num_ep;
2340 :
2341 3889 : rc = 0;
2342 3899 : out:
2343 3899 : if (spdk_spin_held(&bs->used_lock)) {
2344 1624 : spdk_spin_unlock(&bs->used_lock);
2345 : }
2346 :
2347 3899 : return rc;
2348 : }
2349 :
2350 : static void
2351 4432 : blob_persist_generate_new_md(struct spdk_blob_persist_ctx *ctx)
2352 : {
2353 4432 : spdk_bs_sequence_t *seq = ctx->seq;
2354 4432 : struct spdk_blob *blob = ctx->blob;
2355 4432 : struct spdk_blob_store *bs = blob->bs;
2356 : uint64_t i;
2357 : uint32_t page_num;
2358 : void *tmp;
2359 : int rc;
2360 :
2361 : /* Generate the new metadata */
2362 4432 : rc = blob_serialize(blob, &ctx->pages, &blob->active.num_pages);
2363 4432 : if (rc < 0) {
2364 0 : blob_persist_complete(seq, ctx, rc);
2365 0 : return;
2366 : }
2367 :
2368 4432 : assert(blob->active.num_pages >= 1);
2369 :
2370 : /* Resize the cache of page indices */
2371 4432 : tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
2372 4432 : if (!tmp) {
2373 0 : blob_persist_complete(seq, ctx, -ENOMEM);
2374 0 : return;
2375 : }
2376 4432 : blob->active.pages = tmp;
2377 :
2378 : /* Assign this metadata to pages. This requires two passes - one to verify that there are
2379 : * enough pages and a second to actually claim them. The used_lock is held across
2380 : * both passes to ensure things don't change in the middle.
2381 : */
2382 4432 : spdk_spin_lock(&bs->used_lock);
2383 4432 : page_num = 0;
2384 : /* Note that this loop starts at one. The first page location is fixed by the blobid. */
2385 4541 : for (i = 1; i < blob->active.num_pages; i++) {
2386 109 : page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
2387 109 : if (page_num == UINT32_MAX) {
2388 0 : spdk_spin_unlock(&bs->used_lock);
2389 0 : blob_persist_complete(seq, ctx, -ENOMEM);
2390 0 : return;
2391 : }
2392 109 : page_num++;
2393 : }
2394 :
2395 4432 : page_num = 0;
2396 4432 : blob->active.pages[0] = bs_blobid_to_page(blob->id);
2397 4541 : for (i = 1; i < blob->active.num_pages; i++) {
2398 109 : page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
2399 109 : ctx->pages[i - 1].next = page_num;
2400 : /* Now that previous metadata page is complete, calculate the crc for it. */
2401 109 : ctx->pages[i - 1].crc = blob_md_page_calc_crc(&ctx->pages[i - 1]);
2402 109 : blob->active.pages[i] = page_num;
2403 109 : bs_claim_md_page(bs, page_num);
2404 109 : SPDK_DEBUGLOG(blob, "Claiming page %u for blob 0x%" PRIx64 "\n", page_num,
2405 : blob->id);
2406 109 : page_num++;
2407 : }
2408 4432 : spdk_spin_unlock(&bs->used_lock);
2409 4432 : ctx->pages[i - 1].crc = blob_md_page_calc_crc(&ctx->pages[i - 1]);
2410 : /* Start writing the metadata from last page to first */
2411 4432 : blob->state = SPDK_BLOB_STATE_CLEAN;
2412 4432 : blob_persist_write_page_chain(seq, ctx);
2413 : }
2414 :
2415 : static void
2416 3108 : blob_persist_write_extent_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2417 : {
2418 3108 : struct spdk_blob_persist_ctx *ctx = cb_arg;
2419 3108 : struct spdk_blob *blob = ctx->blob;
2420 : size_t i;
2421 : uint32_t extent_page_id;
2422 3108 : uint32_t page_count = 0;
2423 : int rc;
2424 :
2425 3108 : if (ctx->extent_page != NULL) {
2426 991 : spdk_free(ctx->extent_page);
2427 991 : ctx->extent_page = NULL;
2428 : }
2429 :
2430 3108 : if (bserrno != 0) {
2431 0 : blob_persist_complete(seq, ctx, bserrno);
2432 991 : return;
2433 : }
2434 :
2435 : /* Only write out Extent Pages when blob was resized. */
2436 6492 : for (i = ctx->next_extent_page; i < blob->active.extent_pages_array_size; i++) {
2437 4375 : extent_page_id = blob->active.extent_pages[i];
2438 4375 : if (extent_page_id == 0) {
2439 : /* No Extent Page to persist */
2440 3384 : assert(spdk_blob_is_thin_provisioned(blob));
2441 3384 : continue;
2442 : }
2443 991 : assert(spdk_bit_array_get(blob->bs->used_md_pages, extent_page_id));
2444 991 : ctx->next_extent_page = i + 1;
2445 991 : rc = blob_serialize_add_page(ctx->blob, &ctx->extent_page, &page_count, &ctx->extent_page);
2446 991 : if (rc < 0) {
2447 0 : blob_persist_complete(seq, ctx, rc);
2448 0 : return;
2449 : }
2450 :
2451 991 : blob->state = SPDK_BLOB_STATE_DIRTY;
2452 991 : blob_serialize_extent_page(blob, i * SPDK_EXTENTS_PER_EP, ctx->extent_page);
2453 :
2454 991 : ctx->extent_page->crc = blob_md_page_calc_crc(ctx->extent_page);
2455 :
2456 991 : bs_sequence_write_dev(seq, ctx->extent_page, bs_md_page_to_lba(blob->bs, extent_page_id),
2457 991 : bs_byte_to_lba(blob->bs, blob->bs->md_page_size),
2458 : blob_persist_write_extent_pages, ctx);
2459 991 : return;
2460 : }
2461 :
2462 2117 : blob_persist_generate_new_md(ctx);
2463 : }
2464 :
2465 : static void
2466 6304 : blob_persist_start(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2467 : {
2468 6304 : struct spdk_blob_persist_ctx *ctx = cb_arg;
2469 6304 : struct spdk_blob *blob = ctx->blob;
2470 :
2471 6304 : if (bserrno != 0) {
2472 10 : blob_persist_complete(seq, ctx, bserrno);
2473 10 : return;
2474 : }
2475 :
2476 6294 : if (blob->active.num_pages == 0) {
2477 : /* This is the signal that the blob should be deleted.
2478 : * Immediately jump to the clean up routine. */
2479 1862 : assert(blob->clean.num_pages > 0);
2480 1862 : blob->state = SPDK_BLOB_STATE_CLEAN;
2481 1862 : blob_persist_zero_pages(seq, ctx, 0);
2482 1862 : return;
2483 :
2484 : }
2485 :
2486 4432 : if (blob->clean.num_clusters < blob->active.num_clusters) {
2487 : /* Blob was resized up */
2488 2095 : assert(blob->clean.num_extent_pages <= blob->active.num_extent_pages);
2489 2095 : ctx->next_extent_page = spdk_max(1, blob->clean.num_extent_pages) - 1;
2490 2337 : } else if (blob->active.num_clusters < blob->active.cluster_array_size) {
2491 : /* Blob was resized down */
2492 22 : assert(blob->clean.num_extent_pages >= blob->active.num_extent_pages);
2493 22 : ctx->next_extent_page = spdk_max(1, blob->active.num_extent_pages) - 1;
2494 : } else {
2495 : /* No change in size occurred */
2496 2315 : blob_persist_generate_new_md(ctx);
2497 2315 : return;
2498 : }
2499 :
2500 2117 : blob_persist_write_extent_pages(seq, ctx, 0);
2501 : }
2502 :
2503 : struct spdk_bs_mark_dirty {
2504 : struct spdk_blob_store *bs;
2505 : struct spdk_bs_super_block *super;
2506 : spdk_bs_sequence_cpl cb_fn;
2507 : void *cb_arg;
2508 : };
2509 :
2510 : static void
2511 197 : bs_mark_dirty_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2512 : {
2513 197 : struct spdk_bs_mark_dirty *ctx = cb_arg;
2514 :
2515 197 : if (bserrno == 0) {
2516 187 : ctx->bs->clean = 0;
2517 : }
2518 :
2519 197 : ctx->cb_fn(seq, ctx->cb_arg, bserrno);
2520 :
2521 197 : spdk_free(ctx->super);
2522 197 : free(ctx);
2523 197 : }
2524 :
2525 : static void bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2526 : struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg);
2527 :
2528 :
2529 : static void
2530 197 : bs_mark_dirty_write(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2531 : {
2532 197 : struct spdk_bs_mark_dirty *ctx = cb_arg;
2533 : int rc;
2534 :
2535 197 : if (bserrno != 0) {
2536 5 : bs_mark_dirty_write_cpl(seq, ctx, bserrno);
2537 5 : return;
2538 : }
2539 :
2540 192 : rc = bs_super_validate(ctx->super, ctx->bs);
2541 192 : if (rc != 0) {
2542 0 : bs_mark_dirty_write_cpl(seq, ctx, rc);
2543 0 : return;
2544 : }
2545 :
2546 192 : ctx->super->clean = 0;
2547 192 : if (ctx->super->size == 0) {
2548 5 : ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
2549 : }
2550 :
2551 192 : bs_write_super(seq, ctx->bs, ctx->super, bs_mark_dirty_write_cpl, ctx);
2552 : }
2553 :
2554 : static void
2555 6961 : bs_mark_dirty(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2556 : spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2557 : {
2558 : struct spdk_bs_mark_dirty *ctx;
2559 :
2560 : /* Blobstore is already marked dirty */
2561 6961 : if (bs->clean == 0) {
2562 6764 : cb_fn(seq, cb_arg, 0);
2563 6764 : return;
2564 : }
2565 :
2566 197 : ctx = calloc(1, sizeof(*ctx));
2567 197 : if (!ctx) {
2568 0 : cb_fn(seq, cb_arg, -ENOMEM);
2569 0 : return;
2570 : }
2571 197 : ctx->bs = bs;
2572 197 : ctx->cb_fn = cb_fn;
2573 197 : ctx->cb_arg = cb_arg;
2574 :
2575 197 : ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
2576 : SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
2577 197 : if (!ctx->super) {
2578 0 : free(ctx);
2579 0 : cb_fn(seq, cb_arg, -ENOMEM);
2580 0 : return;
2581 : }
2582 :
2583 197 : bs_sequence_read_dev(seq, ctx->super, bs_page_to_lba(bs, 0),
2584 197 : bs_byte_to_lba(bs, sizeof(*ctx->super)),
2585 : bs_mark_dirty_write, ctx);
2586 : }
2587 :
2588 : /* Write a blob to disk */
2589 : static void
2590 11346 : blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
2591 : spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2592 : {
2593 : struct spdk_blob_persist_ctx *ctx;
2594 :
2595 11346 : blob_verify_md_op(blob);
2596 :
2597 11346 : if (blob->state == SPDK_BLOB_STATE_CLEAN && TAILQ_EMPTY(&blob->persists_to_complete)) {
2598 5042 : cb_fn(seq, cb_arg, 0);
2599 5042 : return;
2600 : }
2601 :
2602 6304 : ctx = calloc(1, sizeof(*ctx));
2603 6304 : if (!ctx) {
2604 0 : cb_fn(seq, cb_arg, -ENOMEM);
2605 0 : return;
2606 : }
2607 6304 : ctx->blob = blob;
2608 6304 : ctx->seq = seq;
2609 6304 : ctx->cb_fn = cb_fn;
2610 6304 : ctx->cb_arg = cb_arg;
2611 :
2612 : /* Multiple blob persists can affect one another, via blob->state or
2613 : * blob mutable data changes. To prevent it, queue up the persists. */
2614 6304 : if (!TAILQ_EMPTY(&blob->persists_to_complete)) {
2615 28 : TAILQ_INSERT_TAIL(&blob->pending_persists, ctx, link);
2616 28 : return;
2617 : }
2618 6276 : TAILQ_INSERT_HEAD(&blob->persists_to_complete, ctx, link);
2619 :
2620 6276 : bs_mark_dirty(seq, blob->bs, blob_persist_start, ctx);
2621 : }
2622 :
2623 : struct spdk_blob_copy_cluster_ctx {
2624 : struct spdk_blob *blob;
2625 : uint8_t *buf;
2626 : uint64_t io_unit;
2627 : uint64_t new_cluster;
2628 : uint32_t new_extent_page;
2629 : spdk_bs_sequence_t *seq;
2630 : struct spdk_blob_md_page *new_cluster_page;
2631 : };
2632 :
2633 : struct spdk_blob_free_cluster_ctx {
2634 : struct spdk_blob *blob;
2635 : uint64_t page;
2636 : struct spdk_blob_md_page *md_page;
2637 : uint64_t cluster_num;
2638 : uint32_t extent_page;
2639 : spdk_bs_sequence_t *seq;
2640 : };
2641 :
2642 : static void
2643 1025 : blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno)
2644 : {
2645 1025 : struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2646 1025 : struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq;
2647 : TAILQ_HEAD(, spdk_bs_request_set) requests;
2648 : spdk_bs_user_op_t *op;
2649 :
2650 1025 : TAILQ_INIT(&requests);
2651 1025 : TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link);
2652 :
2653 2050 : while (!TAILQ_EMPTY(&requests)) {
2654 1025 : op = TAILQ_FIRST(&requests);
2655 1025 : TAILQ_REMOVE(&requests, op, link);
2656 1025 : if (bserrno == 0) {
2657 1025 : bs_user_op_execute(op);
2658 : } else {
2659 0 : bs_user_op_abort(op, bserrno);
2660 : }
2661 : }
2662 :
2663 1025 : spdk_free(ctx->buf);
2664 1025 : free(ctx);
2665 1025 : }
2666 :
2667 : static void
2668 75 : blob_free_cluster_cpl(void *cb_arg, int bserrno)
2669 : {
2670 75 : struct spdk_blob_free_cluster_ctx *ctx = cb_arg;
2671 75 : spdk_bs_sequence_t *seq = ctx->seq;
2672 :
2673 75 : bs_sequence_finish(seq, bserrno);
2674 :
2675 75 : free(ctx);
2676 75 : }
2677 :
2678 : static void
2679 5 : blob_insert_cluster_revert(struct spdk_blob_copy_cluster_ctx *ctx)
2680 : {
2681 5 : spdk_spin_lock(&ctx->blob->bs->used_lock);
2682 5 : bs_release_cluster(ctx->blob->bs, ctx->new_cluster);
2683 5 : if (ctx->new_extent_page != 0) {
2684 3 : bs_release_md_page(ctx->blob->bs, ctx->new_extent_page);
2685 : }
2686 5 : spdk_spin_unlock(&ctx->blob->bs->used_lock);
2687 5 : }
2688 :
2689 : static void
2690 5 : blob_insert_cluster_clear_cpl(void *cb_arg, int bserrno)
2691 : {
2692 5 : struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2693 :
2694 5 : if (bserrno) {
2695 0 : SPDK_WARNLOG("Failed to clear cluster: %d\n", bserrno);
2696 : }
2697 :
2698 5 : blob_insert_cluster_revert(ctx);
2699 5 : bs_sequence_finish(ctx->seq, bserrno);
2700 5 : }
2701 :
2702 : static void
2703 5 : blob_insert_cluster_clear(struct spdk_blob_copy_cluster_ctx *ctx)
2704 : {
2705 : struct spdk_bs_cpl cpl;
2706 : spdk_bs_batch_t *batch;
2707 5 : struct spdk_io_channel *ch = spdk_io_channel_from_ctx(ctx->seq->channel);
2708 :
2709 : /*
2710 : * We allocated a cluster and we copied data to it. But now, we realized that we don't need
2711 : * this cluster and we want to release it. We must ensure that we clear the data on this
2712 : * cluster.
2713 : * The cluster may later be re-allocated by a thick-provisioned blob for example. When
2714 : * reading from this thick-provisioned blob before writing data, we should read zeroes.
2715 : */
2716 :
2717 5 : cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2718 5 : cpl.u.blob_basic.cb_fn = blob_insert_cluster_clear_cpl;
2719 5 : cpl.u.blob_basic.cb_arg = ctx;
2720 :
2721 5 : batch = bs_batch_open(ch, &cpl, ctx->blob);
2722 5 : if (!batch) {
2723 0 : blob_insert_cluster_clear_cpl(ctx, -ENOMEM);
2724 0 : return;
2725 : }
2726 :
2727 5 : bs_batch_clear_dev(ctx->blob, batch, bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
2728 5 : bs_cluster_to_lba(ctx->blob->bs, 1));
2729 5 : bs_batch_close(batch);
2730 : }
2731 :
2732 : static void
2733 1025 : blob_insert_cluster_cpl(void *cb_arg, int bserrno)
2734 : {
2735 1025 : struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2736 :
2737 1025 : if (bserrno) {
2738 5 : if (bserrno == -EEXIST) {
2739 : /* The metadata insert failed because another thread
2740 : * allocated the cluster first. Clear and free our cluster
2741 : * but continue without error. */
2742 5 : blob_insert_cluster_clear(ctx);
2743 5 : return;
2744 : }
2745 :
2746 0 : blob_insert_cluster_revert(ctx);
2747 : }
2748 :
2749 1020 : bs_sequence_finish(ctx->seq, bserrno);
2750 : }
2751 :
2752 : static void
2753 515 : blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2754 : {
2755 515 : struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2756 : uint32_t cluster_number;
2757 :
2758 515 : if (bserrno) {
2759 : /* The write failed, so jump to the final completion handler */
2760 0 : bs_sequence_finish(seq, bserrno);
2761 0 : return;
2762 : }
2763 :
2764 515 : cluster_number = bs_io_unit_to_cluster(ctx->blob->bs, ctx->io_unit);
2765 :
2766 515 : blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
2767 : ctx->new_extent_page, ctx->new_cluster_page, blob_insert_cluster_cpl, ctx);
2768 : }
2769 :
2770 : static void
2771 385 : blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2772 : {
2773 385 : struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2774 :
2775 385 : if (bserrno != 0) {
2776 : /* The read failed, so jump to the final completion handler */
2777 0 : bs_sequence_finish(seq, bserrno);
2778 0 : return;
2779 : }
2780 :
2781 : /* Write whole cluster */
2782 385 : bs_sequence_write_dev(seq, ctx->buf,
2783 385 : bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
2784 385 : bs_cluster_to_lba(ctx->blob->bs, 1),
2785 : blob_write_copy_cpl, ctx);
2786 : }
2787 :
2788 : static bool
2789 1005 : blob_can_copy(struct spdk_blob *blob, uint64_t cluster_start_io_unit, uint64_t *base_lba)
2790 : {
2791 1005 : uint64_t lba = bs_dev_io_unit_to_lba(blob, blob->back_bs_dev, cluster_start_io_unit);
2792 :
2793 1359 : return (!blob_is_esnap_clone(blob) && blob->bs->dev->copy != NULL) &&
2794 354 : blob->back_bs_dev->translate_lba(blob->back_bs_dev, lba, base_lba);
2795 : }
2796 :
2797 : static void
2798 130 : blob_copy(struct spdk_blob_copy_cluster_ctx *ctx, spdk_bs_user_op_t *op, uint64_t src_lba)
2799 : {
2800 130 : struct spdk_blob *blob = ctx->blob;
2801 130 : uint64_t lba_count = bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz);
2802 :
2803 130 : bs_sequence_copy_dev(ctx->seq,
2804 130 : bs_cluster_to_lba(blob->bs, ctx->new_cluster),
2805 : src_lba,
2806 : lba_count,
2807 : blob_write_copy_cpl, ctx);
2808 130 : }
2809 :
2810 : static void
2811 1025 : bs_allocate_and_copy_cluster(struct spdk_blob *blob,
2812 : struct spdk_io_channel *_ch,
2813 : uint64_t io_unit, spdk_bs_user_op_t *op)
2814 : {
2815 : struct spdk_bs_cpl cpl;
2816 : struct spdk_bs_channel *ch;
2817 : struct spdk_blob_copy_cluster_ctx *ctx;
2818 : uint64_t cluster_start_io_unit;
2819 : uint32_t cluster_number;
2820 : bool is_zeroes;
2821 : bool can_copy;
2822 : bool is_valid_range;
2823 : uint64_t copy_src_lba;
2824 : int rc;
2825 :
2826 1025 : ch = spdk_io_channel_get_ctx(_ch);
2827 :
2828 1025 : if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) {
2829 : /* There are already operations pending. Queue this user op
2830 : * and return because it will be re-executed when the outstanding
2831 : * cluster allocation completes. */
2832 0 : TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
2833 0 : return;
2834 : }
2835 :
2836 : /* Round the io_unit offset down to the first io_unit in the cluster */
2837 1025 : cluster_start_io_unit = bs_io_unit_to_cluster_start(blob, io_unit);
2838 :
2839 : /* Calculate which index in the metadata cluster array the corresponding
2840 : * cluster is supposed to be at. */
2841 1025 : cluster_number = bs_io_unit_to_cluster_number(blob, io_unit);
2842 :
2843 1025 : ctx = calloc(1, sizeof(*ctx));
2844 1025 : if (!ctx) {
2845 0 : bs_user_op_abort(op, -ENOMEM);
2846 0 : return;
2847 : }
2848 :
2849 1025 : assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0);
2850 :
2851 1025 : ctx->blob = blob;
2852 1025 : ctx->io_unit = cluster_start_io_unit;
2853 1025 : ctx->new_cluster_page = ch->new_cluster_page;
2854 1025 : memset(ctx->new_cluster_page, 0, blob->bs->md_page_size);
2855 :
2856 : /* Check if the cluster that we intend to do CoW for is valid for
2857 : * the backing dev. For zeroes backing dev, it'll be always valid.
2858 : * For other backing dev e.g. a snapshot, it could be invalid if
2859 : * the blob has been resized after snapshot was taken. */
2860 2050 : is_valid_range = blob->back_bs_dev->is_range_valid(blob->back_bs_dev,
2861 : bs_dev_io_unit_to_lba(blob, blob->back_bs_dev, cluster_start_io_unit),
2862 1025 : bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz));
2863 :
2864 1025 : can_copy = is_valid_range && blob_can_copy(blob, cluster_start_io_unit, ©_src_lba);
2865 :
2866 2030 : is_zeroes = is_valid_range && blob->back_bs_dev->is_zeroes(blob->back_bs_dev,
2867 : bs_dev_io_unit_to_lba(blob, blob->back_bs_dev, cluster_start_io_unit),
2868 1005 : bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz));
2869 1025 : if (blob->parent_id != SPDK_BLOBID_INVALID && !is_zeroes && !can_copy) {
2870 385 : ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen,
2871 : NULL, SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
2872 385 : if (!ctx->buf) {
2873 0 : SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n",
2874 : blob->bs->cluster_sz);
2875 0 : free(ctx);
2876 0 : bs_user_op_abort(op, -ENOMEM);
2877 0 : return;
2878 : }
2879 : }
2880 :
2881 1025 : spdk_spin_lock(&blob->bs->used_lock);
2882 1025 : rc = bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, &ctx->new_extent_page,
2883 : false);
2884 1025 : spdk_spin_unlock(&blob->bs->used_lock);
2885 1025 : if (rc != 0) {
2886 0 : spdk_free(ctx->buf);
2887 0 : free(ctx);
2888 0 : bs_user_op_abort(op, rc);
2889 0 : return;
2890 : }
2891 :
2892 1025 : cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2893 1025 : cpl.u.blob_basic.cb_fn = blob_allocate_and_copy_cluster_cpl;
2894 1025 : cpl.u.blob_basic.cb_arg = ctx;
2895 :
2896 1025 : ctx->seq = bs_sequence_start_blob(_ch, &cpl, blob);
2897 1025 : if (!ctx->seq) {
2898 0 : spdk_spin_lock(&blob->bs->used_lock);
2899 0 : bs_release_cluster(blob->bs, ctx->new_cluster);
2900 0 : spdk_spin_unlock(&blob->bs->used_lock);
2901 0 : spdk_free(ctx->buf);
2902 0 : free(ctx);
2903 0 : bs_user_op_abort(op, -ENOMEM);
2904 0 : return;
2905 : }
2906 :
2907 : /* Queue the user op to block other incoming operations */
2908 1025 : TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
2909 :
2910 1025 : if (blob->parent_id != SPDK_BLOBID_INVALID && !is_zeroes) {
2911 515 : if (can_copy) {
2912 130 : blob_copy(ctx, op, copy_src_lba);
2913 : } else {
2914 : /* Read cluster from backing device */
2915 385 : bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
2916 : bs_dev_io_unit_to_lba(blob, blob->back_bs_dev, cluster_start_io_unit),
2917 385 : bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz),
2918 : blob_write_copy, ctx);
2919 : }
2920 :
2921 : } else {
2922 510 : blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
2923 : ctx->new_extent_page, ctx->new_cluster_page, blob_insert_cluster_cpl, ctx);
2924 : }
2925 : }
2926 :
2927 : static inline bool
2928 56755 : blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length,
2929 : uint64_t *lba, uint64_t *lba_count)
2930 : {
2931 56755 : *lba_count = length;
2932 :
2933 56755 : if (!bs_io_unit_is_allocated(blob, io_unit)) {
2934 5196 : assert(blob->back_bs_dev != NULL);
2935 5196 : *lba = bs_io_unit_to_back_dev_lba(blob, io_unit);
2936 5196 : *lba_count = bs_io_unit_to_back_dev_lba(blob, *lba_count);
2937 5196 : return false;
2938 : } else {
2939 51559 : *lba = bs_blob_io_unit_to_lba(blob, io_unit);
2940 51559 : return true;
2941 : }
2942 : }
2943 :
2944 : struct op_split_ctx {
2945 : struct spdk_blob *blob;
2946 : struct spdk_io_channel *channel;
2947 : uint64_t io_unit_offset;
2948 : uint64_t io_units_remaining;
2949 : void *curr_payload;
2950 : enum spdk_blob_op_type op_type;
2951 : spdk_bs_sequence_t *seq;
2952 : bool in_submit_ctx;
2953 : bool completed_in_submit_ctx;
2954 : bool done;
2955 : };
2956 :
2957 : static void
2958 966 : blob_request_submit_op_split_next(void *cb_arg, int bserrno)
2959 : {
2960 966 : struct op_split_ctx *ctx = cb_arg;
2961 966 : struct spdk_blob *blob = ctx->blob;
2962 966 : struct spdk_io_channel *ch = ctx->channel;
2963 966 : enum spdk_blob_op_type op_type = ctx->op_type;
2964 : uint8_t *buf;
2965 : uint64_t offset;
2966 : uint64_t length;
2967 : uint64_t op_length;
2968 :
2969 966 : if (bserrno != 0 || ctx->io_units_remaining == 0) {
2970 222 : bs_sequence_finish(ctx->seq, bserrno);
2971 222 : if (ctx->in_submit_ctx) {
2972 : /* Defer freeing of the ctx object, since it will be
2973 : * accessed when this unwinds back to the submission
2974 : * context.
2975 : */
2976 50 : ctx->done = true;
2977 : } else {
2978 172 : free(ctx);
2979 : }
2980 222 : return;
2981 : }
2982 :
2983 744 : if (ctx->in_submit_ctx) {
2984 : /* If this split operation completed in the context
2985 : * of its submission, mark the flag and return immediately
2986 : * to avoid recursion.
2987 : */
2988 85 : ctx->completed_in_submit_ctx = true;
2989 85 : return;
2990 : }
2991 :
2992 : while (true) {
2993 744 : ctx->completed_in_submit_ctx = false;
2994 :
2995 744 : offset = ctx->io_unit_offset;
2996 744 : length = ctx->io_units_remaining;
2997 744 : buf = ctx->curr_payload;
2998 744 : op_length = spdk_min(length, bs_num_io_units_to_cluster_boundary(blob,
2999 : offset));
3000 :
3001 : /* Update length and payload for next operation */
3002 744 : ctx->io_units_remaining -= op_length;
3003 744 : ctx->io_unit_offset += op_length;
3004 744 : if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
3005 659 : ctx->curr_payload += op_length * blob->bs->io_unit_size;
3006 : }
3007 :
3008 744 : assert(!ctx->in_submit_ctx);
3009 744 : ctx->in_submit_ctx = true;
3010 :
3011 744 : switch (op_type) {
3012 522 : case SPDK_BLOB_READ:
3013 522 : spdk_blob_io_read(blob, ch, buf, offset, op_length,
3014 : blob_request_submit_op_split_next, ctx);
3015 522 : break;
3016 137 : case SPDK_BLOB_WRITE:
3017 137 : spdk_blob_io_write(blob, ch, buf, offset, op_length,
3018 : blob_request_submit_op_split_next, ctx);
3019 137 : break;
3020 45 : case SPDK_BLOB_UNMAP:
3021 45 : spdk_blob_io_unmap(blob, ch, offset, op_length,
3022 : blob_request_submit_op_split_next, ctx);
3023 45 : break;
3024 40 : case SPDK_BLOB_WRITE_ZEROES:
3025 40 : spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
3026 : blob_request_submit_op_split_next, ctx);
3027 40 : break;
3028 0 : case SPDK_BLOB_READV:
3029 : case SPDK_BLOB_WRITEV:
3030 0 : SPDK_ERRLOG("readv/write not valid\n");
3031 0 : bs_sequence_finish(ctx->seq, -EINVAL);
3032 0 : free(ctx);
3033 0 : return;
3034 : }
3035 :
3036 : #ifndef __clang_analyzer__
3037 : /* scan-build reports a false positive around accessing the ctx here. It
3038 : * forms a path that recursively calls this function, but then says
3039 : * "assuming ctx->in_submit_ctx is false", when that isn't possible.
3040 : * This path does free(ctx), returns to here, and reports a use-after-free
3041 : * bug. Wrapping this bit of code so that scan-build doesn't see it
3042 : * works around the scan-build bug.
3043 : */
3044 744 : assert(ctx->in_submit_ctx);
3045 744 : ctx->in_submit_ctx = false;
3046 :
3047 : /* If the operation completed immediately, loop back and submit the
3048 : * next operation. Otherwise we can return and the next split
3049 : * operation will get submitted when this current operation is
3050 : * later completed asynchronously.
3051 : */
3052 744 : if (ctx->completed_in_submit_ctx) {
3053 85 : continue;
3054 659 : } else if (ctx->done) {
3055 50 : free(ctx);
3056 : }
3057 : #endif
3058 659 : break;
3059 : }
3060 : }
3061 :
3062 : static void
3063 222 : blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
3064 : void *payload, uint64_t offset, uint64_t length,
3065 : spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
3066 : {
3067 : struct op_split_ctx *ctx;
3068 : spdk_bs_sequence_t *seq;
3069 : struct spdk_bs_cpl cpl;
3070 :
3071 222 : assert(blob != NULL);
3072 :
3073 222 : ctx = calloc(1, sizeof(struct op_split_ctx));
3074 222 : if (ctx == NULL) {
3075 0 : cb_fn(cb_arg, -ENOMEM);
3076 0 : return;
3077 : }
3078 :
3079 222 : cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
3080 222 : cpl.u.blob_basic.cb_fn = cb_fn;
3081 222 : cpl.u.blob_basic.cb_arg = cb_arg;
3082 :
3083 222 : seq = bs_sequence_start_blob(ch, &cpl, blob);
3084 222 : if (!seq) {
3085 0 : free(ctx);
3086 0 : cb_fn(cb_arg, -ENOMEM);
3087 0 : return;
3088 : }
3089 :
3090 222 : ctx->blob = blob;
3091 222 : ctx->channel = ch;
3092 222 : ctx->curr_payload = payload;
3093 222 : ctx->io_unit_offset = offset;
3094 222 : ctx->io_units_remaining = length;
3095 222 : ctx->op_type = op_type;
3096 222 : ctx->seq = seq;
3097 :
3098 222 : blob_request_submit_op_split_next(ctx, 0);
3099 : }
3100 :
3101 : static void
3102 75 : spdk_free_cluster_unmap_complete(void *cb_arg, int bserrno)
3103 : {
3104 75 : struct spdk_blob_free_cluster_ctx *ctx = cb_arg;
3105 :
3106 75 : if (bserrno) {
3107 0 : bs_sequence_finish(ctx->seq, bserrno);
3108 0 : free(ctx);
3109 0 : return;
3110 : }
3111 :
3112 75 : blob_free_cluster_on_md_thread(ctx->blob, ctx->cluster_num,
3113 : ctx->extent_page, ctx->md_page, blob_free_cluster_cpl, ctx);
3114 : }
3115 :
3116 : static void
3117 52830 : blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
3118 : void *payload, uint64_t offset, uint64_t length,
3119 : spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
3120 : {
3121 : struct spdk_bs_cpl cpl;
3122 : uint64_t lba;
3123 : uint64_t lba_count;
3124 : bool is_allocated;
3125 :
3126 52830 : assert(blob != NULL);
3127 :
3128 52830 : cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
3129 52830 : cpl.u.blob_basic.cb_fn = cb_fn;
3130 52830 : cpl.u.blob_basic.cb_arg = cb_arg;
3131 :
3132 52830 : if (blob->frozen_refcnt) {
3133 : /* This blob I/O is frozen */
3134 : spdk_bs_user_op_t *op;
3135 5 : struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
3136 :
3137 5 : op = bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
3138 5 : if (!op) {
3139 0 : cb_fn(cb_arg, -ENOMEM);
3140 5 : return;
3141 : }
3142 :
3143 5 : TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
3144 :
3145 5 : return;
3146 : }
3147 :
3148 52825 : is_allocated = blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
3149 :
3150 52825 : switch (op_type) {
3151 25377 : case SPDK_BLOB_READ: {
3152 : spdk_bs_batch_t *batch;
3153 :
3154 25377 : batch = bs_batch_open(_ch, &cpl, blob);
3155 25377 : if (!batch) {
3156 0 : cb_fn(cb_arg, -ENOMEM);
3157 0 : return;
3158 : }
3159 :
3160 25377 : if (is_allocated) {
3161 : /* Read from the blob */
3162 23531 : bs_batch_read_dev(batch, payload, lba, lba_count);
3163 : } else {
3164 : /* Read from the backing block device */
3165 1846 : bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
3166 : }
3167 :
3168 25377 : bs_batch_close(batch);
3169 25377 : break;
3170 : }
3171 27328 : case SPDK_BLOB_WRITE:
3172 : case SPDK_BLOB_WRITE_ZEROES: {
3173 27328 : if (is_allocated) {
3174 : /* Write to the blob */
3175 : spdk_bs_batch_t *batch;
3176 :
3177 26888 : if (lba_count == 0) {
3178 0 : cb_fn(cb_arg, 0);
3179 0 : return;
3180 : }
3181 :
3182 26888 : batch = bs_batch_open(_ch, &cpl, blob);
3183 26888 : if (!batch) {
3184 0 : cb_fn(cb_arg, -ENOMEM);
3185 0 : return;
3186 : }
3187 :
3188 26888 : if (op_type == SPDK_BLOB_WRITE) {
3189 26848 : bs_batch_write_dev(batch, payload, lba, lba_count);
3190 : } else {
3191 40 : bs_batch_write_zeroes_dev(batch, lba, lba_count);
3192 : }
3193 :
3194 26888 : bs_batch_close(batch);
3195 : } else {
3196 : /* Queue this operation and allocate the cluster */
3197 : spdk_bs_user_op_t *op;
3198 :
3199 440 : op = bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
3200 440 : if (!op) {
3201 0 : cb_fn(cb_arg, -ENOMEM);
3202 0 : return;
3203 : }
3204 :
3205 440 : bs_allocate_and_copy_cluster(blob, _ch, offset, op);
3206 : }
3207 27328 : break;
3208 : }
3209 120 : case SPDK_BLOB_UNMAP: {
3210 120 : struct spdk_blob_free_cluster_ctx *ctx = NULL;
3211 : spdk_bs_batch_t *batch;
3212 :
3213 : /* if aligned with cluster release cluster */
3214 210 : if (spdk_blob_is_thin_provisioned(blob) && is_allocated &&
3215 175 : blob_backed_with_zeroes_dev(blob) &&
3216 85 : bs_io_units_per_cluster(blob) == length) {
3217 75 : struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
3218 : uint64_t cluster_start_page;
3219 : uint32_t cluster_number;
3220 :
3221 75 : assert(offset % bs_io_units_per_cluster(blob) == 0);
3222 :
3223 : /* Round the io_unit offset down to the first page in the cluster */
3224 75 : cluster_start_page = bs_io_unit_to_cluster_start(blob, offset);
3225 :
3226 : /* Calculate which index in the metadata cluster array the corresponding
3227 : * cluster is supposed to be at. */
3228 75 : cluster_number = bs_io_unit_to_cluster_number(blob, offset);
3229 :
3230 75 : ctx = calloc(1, sizeof(*ctx));
3231 75 : if (!ctx) {
3232 0 : cb_fn(cb_arg, -ENOMEM);
3233 0 : return;
3234 : }
3235 : /* When freeing a cluster the flow should be (in order):
3236 : * 1. Unmap the underlying area (so if the cluster is reclaimed in the future, it won't leak
3237 : * old data)
3238 : * 2. Once the unmap completes (to avoid any races with incoming writes that may claim the
3239 : * cluster), update and sync metadata freeing the cluster
3240 : * 3. Once metadata update is done, complete the user unmap request
3241 : */
3242 75 : ctx->blob = blob;
3243 75 : ctx->page = cluster_start_page;
3244 75 : ctx->cluster_num = cluster_number;
3245 75 : ctx->md_page = bs_channel->new_cluster_page;
3246 75 : ctx->seq = bs_sequence_start_bs(_ch, &cpl);
3247 75 : if (!ctx->seq) {
3248 0 : free(ctx);
3249 0 : cb_fn(cb_arg, -ENOMEM);
3250 0 : return;
3251 : }
3252 :
3253 75 : if (blob->use_extent_table) {
3254 45 : ctx->extent_page = *bs_cluster_to_extent_page(blob, cluster_number);
3255 : }
3256 :
3257 75 : cpl.u.blob_basic.cb_fn = spdk_free_cluster_unmap_complete;
3258 75 : cpl.u.blob_basic.cb_arg = ctx;
3259 : }
3260 :
3261 120 : batch = bs_batch_open(_ch, &cpl, blob);
3262 120 : if (!batch) {
3263 0 : free(ctx);
3264 0 : cb_fn(cb_arg, -ENOMEM);
3265 0 : return;
3266 : }
3267 :
3268 120 : if (is_allocated) {
3269 120 : bs_batch_unmap_dev(batch, lba, lba_count);
3270 : }
3271 :
3272 120 : bs_batch_close(batch);
3273 120 : break;
3274 : }
3275 0 : case SPDK_BLOB_READV:
3276 : case SPDK_BLOB_WRITEV:
3277 0 : SPDK_ERRLOG("readv/write not valid\n");
3278 0 : cb_fn(cb_arg, -EINVAL);
3279 0 : break;
3280 : }
3281 : }
3282 :
3283 : static void
3284 53692 : blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
3285 : void *payload, uint64_t offset, uint64_t length,
3286 : spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
3287 : {
3288 53692 : assert(blob != NULL);
3289 :
3290 53692 : if (blob->data_ro && op_type != SPDK_BLOB_READ) {
3291 5 : cb_fn(cb_arg, -EPERM);
3292 5 : return;
3293 : }
3294 :
3295 53687 : if (length == 0) {
3296 615 : cb_fn(cb_arg, 0);
3297 615 : return;
3298 : }
3299 :
3300 53072 : if (offset + length > bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
3301 30 : cb_fn(cb_arg, -EINVAL);
3302 30 : return;
3303 : }
3304 53042 : if (length <= bs_num_io_units_to_cluster_boundary(blob, offset)) {
3305 52820 : blob_request_submit_op_single(_channel, blob, payload, offset, length,
3306 : cb_fn, cb_arg, op_type);
3307 : } else {
3308 222 : blob_request_submit_op_split(_channel, blob, payload, offset, length,
3309 : cb_fn, cb_arg, op_type);
3310 : }
3311 : }
3312 :
3313 : struct rw_iov_ctx {
3314 : struct spdk_blob *blob;
3315 : struct spdk_io_channel *channel;
3316 : spdk_blob_op_complete cb_fn;
3317 : void *cb_arg;
3318 : bool read;
3319 : int iovcnt;
3320 : struct iovec *orig_iov;
3321 : uint64_t io_unit_offset;
3322 : uint64_t io_units_remaining;
3323 : uint64_t io_units_done;
3324 : struct spdk_blob_ext_io_opts *ext_io_opts;
3325 : struct iovec iov[0];
3326 : };
3327 :
3328 : static void
3329 3910 : rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3330 : {
3331 3910 : assert(cb_arg == NULL);
3332 3910 : bs_sequence_finish(seq, bserrno);
3333 3910 : }
3334 :
3335 : static void
3336 930 : rw_iov_split_next(void *cb_arg, int bserrno)
3337 : {
3338 930 : struct rw_iov_ctx *ctx = cb_arg;
3339 930 : struct spdk_blob *blob = ctx->blob;
3340 : struct iovec *iov, *orig_iov;
3341 : int iovcnt;
3342 : size_t orig_iovoff;
3343 : uint64_t io_units_count, io_units_to_boundary, io_unit_offset;
3344 : uint64_t byte_count;
3345 :
3346 930 : if (bserrno != 0 || ctx->io_units_remaining == 0) {
3347 255 : ctx->cb_fn(ctx->cb_arg, bserrno);
3348 255 : free(ctx);
3349 255 : return;
3350 : }
3351 :
3352 675 : io_unit_offset = ctx->io_unit_offset;
3353 675 : io_units_to_boundary = bs_num_io_units_to_cluster_boundary(blob, io_unit_offset);
3354 675 : io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary);
3355 : /*
3356 : * Get index and offset into the original iov array for our current position in the I/O sequence.
3357 : * byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
3358 : * point to the current position in the I/O sequence.
3359 : */
3360 675 : byte_count = ctx->io_units_done * blob->bs->io_unit_size;
3361 675 : orig_iov = &ctx->orig_iov[0];
3362 675 : orig_iovoff = 0;
3363 1435 : while (byte_count > 0) {
3364 760 : if (byte_count >= orig_iov->iov_len) {
3365 440 : byte_count -= orig_iov->iov_len;
3366 440 : orig_iov++;
3367 : } else {
3368 320 : orig_iovoff = byte_count;
3369 320 : byte_count = 0;
3370 : }
3371 : }
3372 :
3373 : /*
3374 : * Build an iov array for the next I/O in the sequence. byte_count will keep track of how many
3375 : * bytes of this next I/O remain to be accounted for in the new iov array.
3376 : */
3377 675 : byte_count = io_units_count * blob->bs->io_unit_size;
3378 675 : iov = &ctx->iov[0];
3379 675 : iovcnt = 0;
3380 1725 : while (byte_count > 0) {
3381 1050 : assert(iovcnt < ctx->iovcnt);
3382 1050 : iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
3383 1050 : iov->iov_base = orig_iov->iov_base + orig_iovoff;
3384 1050 : byte_count -= iov->iov_len;
3385 1050 : orig_iovoff = 0;
3386 1050 : orig_iov++;
3387 1050 : iov++;
3388 1050 : iovcnt++;
3389 : }
3390 :
3391 675 : ctx->io_unit_offset += io_units_count;
3392 675 : ctx->io_units_remaining -= io_units_count;
3393 675 : ctx->io_units_done += io_units_count;
3394 675 : iov = &ctx->iov[0];
3395 :
3396 675 : if (ctx->read) {
3397 510 : spdk_blob_io_readv_ext(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
3398 : io_units_count, rw_iov_split_next, ctx, ctx->ext_io_opts);
3399 : } else {
3400 165 : spdk_blob_io_writev_ext(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
3401 : io_units_count, rw_iov_split_next, ctx, ctx->ext_io_opts);
3402 : }
3403 : }
3404 :
3405 : static void
3406 4195 : blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
3407 : struct iovec *iov, int iovcnt,
3408 : uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg, bool read,
3409 : struct spdk_blob_ext_io_opts *ext_io_opts)
3410 : {
3411 : struct spdk_bs_cpl cpl;
3412 :
3413 4195 : assert(blob != NULL);
3414 :
3415 4195 : if (!read && blob->data_ro) {
3416 5 : cb_fn(cb_arg, -EPERM);
3417 10 : return;
3418 : }
3419 :
3420 4190 : if (length == 0) {
3421 0 : cb_fn(cb_arg, 0);
3422 0 : return;
3423 : }
3424 :
3425 4190 : if (offset + length > bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
3426 0 : cb_fn(cb_arg, -EINVAL);
3427 0 : return;
3428 : }
3429 :
3430 : /*
3431 : * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
3432 : * to split a request that spans a cluster boundary. For I/O that do not span a cluster boundary,
3433 : * there will be no noticeable difference compared to using a batch. For I/O that do span a cluster
3434 : * boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
3435 : * to allocate a separate iov array and split the I/O such that none of the resulting
3436 : * smaller I/O cross a cluster boundary. These smaller I/O will be issued in sequence (not in parallel)
3437 : * but since this case happens very infrequently, any performance impact will be negligible.
3438 : *
3439 : * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
3440 : * for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
3441 : * in a batch. That would also require creating an intermediate spdk_bs_cpl that would get called
3442 : * when the batch was completed, to allow for freeing the memory for the iov arrays.
3443 : */
3444 4190 : if (spdk_likely(length <= bs_num_io_units_to_cluster_boundary(blob, offset))) {
3445 : uint64_t lba_count;
3446 : uint64_t lba;
3447 : bool is_allocated;
3448 :
3449 3930 : cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
3450 3930 : cpl.u.blob_basic.cb_fn = cb_fn;
3451 3930 : cpl.u.blob_basic.cb_arg = cb_arg;
3452 :
3453 3930 : if (blob->frozen_refcnt) {
3454 : /* This blob I/O is frozen */
3455 : enum spdk_blob_op_type op_type;
3456 : spdk_bs_user_op_t *op;
3457 0 : struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
3458 :
3459 0 : op_type = read ? SPDK_BLOB_READV : SPDK_BLOB_WRITEV;
3460 0 : op = bs_user_op_alloc(_channel, &cpl, op_type, blob, iov, iovcnt, offset, length);
3461 0 : if (!op) {
3462 0 : cb_fn(cb_arg, -ENOMEM);
3463 0 : return;
3464 : }
3465 :
3466 0 : TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
3467 :
3468 0 : return;
3469 : }
3470 :
3471 3930 : is_allocated = blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
3472 :
3473 3930 : if (read) {
3474 : spdk_bs_sequence_t *seq;
3475 :
3476 3565 : seq = bs_sequence_start_blob(_channel, &cpl, blob);
3477 3565 : if (!seq) {
3478 0 : cb_fn(cb_arg, -ENOMEM);
3479 0 : return;
3480 : }
3481 :
3482 3565 : seq->ext_io_opts = ext_io_opts;
3483 :
3484 3565 : if (is_allocated) {
3485 675 : bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, rw_iov_done, NULL);
3486 : } else {
3487 2890 : bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
3488 : rw_iov_done, NULL);
3489 : }
3490 : } else {
3491 365 : if (is_allocated) {
3492 : spdk_bs_sequence_t *seq;
3493 :
3494 345 : seq = bs_sequence_start_blob(_channel, &cpl, blob);
3495 345 : if (!seq) {
3496 0 : cb_fn(cb_arg, -ENOMEM);
3497 0 : return;
3498 : }
3499 :
3500 345 : seq->ext_io_opts = ext_io_opts;
3501 :
3502 345 : bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, rw_iov_done, NULL);
3503 : } else {
3504 : /* Queue this operation and allocate the cluster */
3505 : spdk_bs_user_op_t *op;
3506 :
3507 20 : op = bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset,
3508 : length);
3509 20 : if (!op) {
3510 0 : cb_fn(cb_arg, -ENOMEM);
3511 0 : return;
3512 : }
3513 :
3514 20 : op->ext_io_opts = ext_io_opts;
3515 :
3516 20 : bs_allocate_and_copy_cluster(blob, _channel, offset, op);
3517 : }
3518 : }
3519 : } else {
3520 : struct rw_iov_ctx *ctx;
3521 :
3522 260 : ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
3523 260 : if (ctx == NULL) {
3524 5 : cb_fn(cb_arg, -ENOMEM);
3525 5 : return;
3526 : }
3527 :
3528 255 : ctx->blob = blob;
3529 255 : ctx->channel = _channel;
3530 255 : ctx->cb_fn = cb_fn;
3531 255 : ctx->cb_arg = cb_arg;
3532 255 : ctx->read = read;
3533 255 : ctx->orig_iov = iov;
3534 255 : ctx->iovcnt = iovcnt;
3535 255 : ctx->io_unit_offset = offset;
3536 255 : ctx->io_units_remaining = length;
3537 255 : ctx->io_units_done = 0;
3538 255 : ctx->ext_io_opts = ext_io_opts;
3539 :
3540 255 : rw_iov_split_next(ctx, 0);
3541 : }
3542 : }
3543 :
3544 : static struct spdk_blob *
3545 9668 : blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
3546 : {
3547 : struct spdk_blob find;
3548 :
3549 9668 : if (spdk_bit_array_get(bs->open_blobids, blobid) == 0) {
3550 8681 : return NULL;
3551 : }
3552 :
3553 987 : find.id = blobid;
3554 987 : return RB_FIND(spdk_blob_tree, &bs->open_blobs, &find);
3555 : }
3556 :
3557 : static void
3558 2256 : blob_get_snapshot_and_clone_entries(struct spdk_blob *blob,
3559 : struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry)
3560 : {
3561 2256 : assert(blob != NULL);
3562 2256 : *snapshot_entry = NULL;
3563 2256 : *clone_entry = NULL;
3564 :
3565 2256 : if (blob->parent_id == SPDK_BLOBID_INVALID) {
3566 1901 : return;
3567 : }
3568 :
3569 535 : TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) {
3570 470 : if ((*snapshot_entry)->id == blob->parent_id) {
3571 290 : break;
3572 : }
3573 : }
3574 :
3575 355 : if (*snapshot_entry != NULL) {
3576 345 : TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) {
3577 345 : if ((*clone_entry)->id == blob->id) {
3578 290 : break;
3579 : }
3580 : }
3581 :
3582 290 : assert(*clone_entry != NULL);
3583 : }
3584 : }
3585 :
3586 : static int
3587 993 : bs_channel_create(void *io_device, void *ctx_buf)
3588 : {
3589 993 : struct spdk_blob_store *bs = io_device;
3590 993 : struct spdk_bs_channel *channel = ctx_buf;
3591 : struct spdk_bs_dev *dev;
3592 993 : uint32_t max_ops = bs->max_channel_ops;
3593 : uint32_t i;
3594 :
3595 993 : dev = bs->dev;
3596 :
3597 993 : channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
3598 993 : if (!channel->req_mem) {
3599 0 : return -1;
3600 : }
3601 :
3602 993 : TAILQ_INIT(&channel->reqs);
3603 :
3604 509409 : for (i = 0; i < max_ops; i++) {
3605 508416 : TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
3606 : }
3607 :
3608 993 : channel->bs = bs;
3609 993 : channel->dev = dev;
3610 993 : channel->dev_channel = dev->create_channel(dev);
3611 :
3612 993 : if (!channel->dev_channel) {
3613 0 : SPDK_ERRLOG("Failed to create device channel.\n");
3614 0 : free(channel->req_mem);
3615 0 : return -1;
3616 : }
3617 :
3618 993 : channel->new_cluster_page = spdk_zmalloc(bs->md_page_size, 0, NULL, SPDK_ENV_NUMA_ID_ANY,
3619 : SPDK_MALLOC_DMA);
3620 993 : if (!channel->new_cluster_page) {
3621 0 : SPDK_ERRLOG("Failed to allocate new cluster page\n");
3622 0 : free(channel->req_mem);
3623 0 : channel->dev->destroy_channel(channel->dev, channel->dev_channel);
3624 0 : return -1;
3625 : }
3626 :
3627 993 : TAILQ_INIT(&channel->need_cluster_alloc);
3628 993 : TAILQ_INIT(&channel->queued_io);
3629 993 : RB_INIT(&channel->esnap_channels);
3630 :
3631 993 : return 0;
3632 : }
3633 :
3634 : static void
3635 993 : bs_channel_destroy(void *io_device, void *ctx_buf)
3636 : {
3637 993 : struct spdk_bs_channel *channel = ctx_buf;
3638 : spdk_bs_user_op_t *op;
3639 :
3640 993 : while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
3641 0 : op = TAILQ_FIRST(&channel->need_cluster_alloc);
3642 0 : TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
3643 0 : bs_user_op_abort(op, -EIO);
3644 : }
3645 :
3646 993 : while (!TAILQ_EMPTY(&channel->queued_io)) {
3647 0 : op = TAILQ_FIRST(&channel->queued_io);
3648 0 : TAILQ_REMOVE(&channel->queued_io, op, link);
3649 0 : bs_user_op_abort(op, -EIO);
3650 : }
3651 :
3652 993 : blob_esnap_destroy_bs_channel(channel);
3653 :
3654 993 : free(channel->req_mem);
3655 993 : spdk_free(channel->new_cluster_page);
3656 993 : channel->dev->destroy_channel(channel->dev, channel->dev_channel);
3657 993 : }
3658 :
3659 : static void
3660 973 : bs_dev_destroy(void *io_device)
3661 : {
3662 973 : struct spdk_blob_store *bs = io_device;
3663 : struct spdk_blob *blob, *blob_tmp;
3664 :
3665 973 : bs->dev->destroy(bs->dev);
3666 :
3667 973 : RB_FOREACH_SAFE(blob, spdk_blob_tree, &bs->open_blobs, blob_tmp) {
3668 0 : RB_REMOVE(spdk_blob_tree, &bs->open_blobs, blob);
3669 0 : spdk_bit_array_clear(bs->open_blobids, blob->id);
3670 0 : blob_free(blob);
3671 : }
3672 :
3673 973 : spdk_spin_destroy(&bs->used_lock);
3674 :
3675 973 : spdk_bit_array_free(&bs->open_blobids);
3676 973 : spdk_bit_array_free(&bs->used_blobids);
3677 973 : spdk_bit_array_free(&bs->used_md_pages);
3678 973 : spdk_bit_pool_free(&bs->used_clusters);
3679 : /*
3680 : * If this function is called for any reason except a successful unload,
3681 : * the unload_cpl type will be NONE and this will be a nop.
3682 : */
3683 973 : bs_call_cpl(&bs->unload_cpl, bs->unload_err);
3684 :
3685 973 : free(bs);
3686 973 : }
3687 :
3688 : static int
3689 1139 : bs_blob_list_add(struct spdk_blob *blob)
3690 : {
3691 : spdk_blob_id snapshot_id;
3692 1139 : struct spdk_blob_list *snapshot_entry = NULL;
3693 1139 : struct spdk_blob_list *clone_entry = NULL;
3694 :
3695 1139 : assert(blob != NULL);
3696 :
3697 1139 : snapshot_id = blob->parent_id;
3698 1139 : if (snapshot_id == SPDK_BLOBID_INVALID ||
3699 : snapshot_id == SPDK_BLOBID_EXTERNAL_SNAPSHOT) {
3700 617 : return 0;
3701 : }
3702 :
3703 522 : snapshot_entry = bs_get_snapshot_entry(blob->bs, snapshot_id);
3704 522 : if (snapshot_entry == NULL) {
3705 : /* Snapshot not found */
3706 362 : snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
3707 362 : if (snapshot_entry == NULL) {
3708 0 : return -ENOMEM;
3709 : }
3710 362 : snapshot_entry->id = snapshot_id;
3711 362 : TAILQ_INIT(&snapshot_entry->clones);
3712 362 : TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
3713 : } else {
3714 255 : TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
3715 95 : if (clone_entry->id == blob->id) {
3716 0 : break;
3717 : }
3718 : }
3719 : }
3720 :
3721 522 : if (clone_entry == NULL) {
3722 : /* Clone not found */
3723 522 : clone_entry = calloc(1, sizeof(struct spdk_blob_list));
3724 522 : if (clone_entry == NULL) {
3725 0 : return -ENOMEM;
3726 : }
3727 522 : clone_entry->id = blob->id;
3728 522 : TAILQ_INIT(&clone_entry->clones);
3729 522 : TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
3730 522 : snapshot_entry->clone_count++;
3731 : }
3732 :
3733 522 : return 0;
3734 : }
3735 :
3736 : static void
3737 2158 : bs_blob_list_remove(struct spdk_blob *blob)
3738 : {
3739 2158 : struct spdk_blob_list *snapshot_entry = NULL;
3740 2158 : struct spdk_blob_list *clone_entry = NULL;
3741 :
3742 2158 : blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry);
3743 :
3744 2158 : if (snapshot_entry == NULL) {
3745 1888 : return;
3746 : }
3747 :
3748 270 : blob->parent_id = SPDK_BLOBID_INVALID;
3749 270 : TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
3750 270 : free(clone_entry);
3751 :
3752 270 : snapshot_entry->clone_count--;
3753 : }
3754 :
3755 : static int
3756 973 : bs_blob_list_free(struct spdk_blob_store *bs)
3757 : {
3758 : struct spdk_blob_list *snapshot_entry;
3759 : struct spdk_blob_list *snapshot_entry_tmp;
3760 : struct spdk_blob_list *clone_entry;
3761 : struct spdk_blob_list *clone_entry_tmp;
3762 :
3763 1155 : TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
3764 374 : TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
3765 192 : TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
3766 192 : free(clone_entry);
3767 : }
3768 182 : TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
3769 182 : free(snapshot_entry);
3770 : }
3771 :
3772 973 : return 0;
3773 : }
3774 :
3775 : static void
3776 973 : bs_free(struct spdk_blob_store *bs)
3777 : {
3778 973 : bs_blob_list_free(bs);
3779 :
3780 973 : bs_unregister_md_thread(bs);
3781 973 : spdk_io_device_unregister(bs, bs_dev_destroy);
3782 973 : }
3783 :
3784 : void
3785 1303 : spdk_bs_opts_init(struct spdk_bs_opts *opts, size_t opts_size)
3786 : {
3787 :
3788 1303 : if (!opts) {
3789 0 : SPDK_ERRLOG("opts should not be NULL\n");
3790 0 : return;
3791 : }
3792 :
3793 1303 : if (!opts_size) {
3794 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
3795 0 : return;
3796 : }
3797 :
3798 1303 : memset(opts, 0, opts_size);
3799 1303 : opts->opts_size = opts_size;
3800 :
3801 : #define FIELD_OK(field) \
3802 : offsetof(struct spdk_bs_opts, field) + sizeof(opts->field) <= opts_size
3803 :
3804 : #define SET_FIELD(field, value) \
3805 : if (FIELD_OK(field)) { \
3806 : opts->field = value; \
3807 : } \
3808 :
3809 1303 : SET_FIELD(cluster_sz, SPDK_BLOB_OPTS_CLUSTER_SZ);
3810 1303 : SET_FIELD(num_md_pages, SPDK_BLOB_OPTS_NUM_MD_PAGES);
3811 1303 : SET_FIELD(max_md_ops, SPDK_BLOB_OPTS_NUM_MD_PAGES);
3812 1303 : SET_FIELD(max_channel_ops, SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS);
3813 1303 : SET_FIELD(clear_method, BS_CLEAR_WITH_UNMAP);
3814 :
3815 1303 : if (FIELD_OK(bstype)) {
3816 1303 : memset(&opts->bstype, 0, sizeof(opts->bstype));
3817 : }
3818 :
3819 1303 : SET_FIELD(iter_cb_fn, NULL);
3820 1303 : SET_FIELD(iter_cb_arg, NULL);
3821 1303 : SET_FIELD(force_recover, false);
3822 1303 : SET_FIELD(esnap_bs_dev_create, NULL);
3823 1303 : SET_FIELD(esnap_ctx, NULL);
3824 :
3825 : #undef FIELD_OK
3826 : #undef SET_FIELD
3827 : }
3828 :
3829 : static int
3830 602 : bs_opts_verify(struct spdk_bs_opts *opts)
3831 : {
3832 602 : if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
3833 597 : opts->max_channel_ops == 0) {
3834 5 : SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
3835 5 : return -1;
3836 : }
3837 :
3838 597 : if ((opts->cluster_sz % SPDK_BS_PAGE_SIZE) != 0) {
3839 5 : SPDK_ERRLOG("Cluster size %" PRIu32 " is not an integral multiple of blocklen %" PRIu32"\n",
3840 : opts->cluster_sz, SPDK_BS_PAGE_SIZE);
3841 5 : return -1;
3842 : }
3843 :
3844 592 : return 0;
3845 : }
3846 :
3847 : /* START spdk_bs_load */
3848 :
3849 : /* spdk_bs_load_ctx is used for init, load, unload and dump code paths. */
3850 :
3851 : struct spdk_bs_load_ctx {
3852 : struct spdk_blob_store *bs;
3853 : struct spdk_bs_super_block *super;
3854 :
3855 : struct spdk_bs_md_mask *mask;
3856 : bool in_page_chain;
3857 : uint32_t page_index;
3858 : uint32_t cur_page;
3859 : struct spdk_blob_md_page *page;
3860 :
3861 : uint64_t num_extent_pages;
3862 : uint32_t *extent_page_num;
3863 : struct spdk_blob_md_page *extent_pages;
3864 : struct spdk_bit_array *used_clusters;
3865 :
3866 : spdk_bs_sequence_t *seq;
3867 : spdk_blob_op_with_handle_complete iter_cb_fn;
3868 : void *iter_cb_arg;
3869 : struct spdk_blob *blob;
3870 : spdk_blob_id blobid;
3871 :
3872 : bool force_recover;
3873 :
3874 : /* These fields are used in the spdk_bs_dump path. */
3875 : bool dumping;
3876 : FILE *fp;
3877 : spdk_bs_dump_print_xattr print_xattr_fn;
3878 : char xattr_name[4096];
3879 : };
3880 :
3881 : static void
3882 1324 : bs_init_per_cluster_fields(struct spdk_blob_store *bs)
3883 : {
3884 1324 : bs->pages_per_cluster = bs->cluster_sz / bs->md_page_size;
3885 1324 : if (spdk_u32_is_pow2(bs->pages_per_cluster)) {
3886 1324 : bs->pages_per_cluster_shift = spdk_u32log2(bs->pages_per_cluster);
3887 : }
3888 1324 : bs->io_units_per_cluster = bs->cluster_sz / bs->io_unit_size;
3889 1324 : if (spdk_u32_is_pow2(bs->io_units_per_cluster)) {
3890 1324 : bs->io_units_per_cluster_shift = spdk_u32log2(bs->io_units_per_cluster);
3891 : }
3892 1324 : }
3893 :
3894 : static int
3895 973 : bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs,
3896 : struct spdk_bs_load_ctx **_ctx)
3897 : {
3898 : struct spdk_blob_store *bs;
3899 : struct spdk_bs_load_ctx *ctx;
3900 : uint64_t dev_size;
3901 : uint32_t md_page_size;
3902 : int rc;
3903 :
3904 973 : dev_size = dev->blocklen * dev->blockcnt;
3905 973 : if (dev_size < opts->cluster_sz) {
3906 : /* Device size cannot be smaller than cluster size of blobstore */
3907 0 : SPDK_INFOLOG(blob, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
3908 : dev_size, opts->cluster_sz);
3909 0 : return -ENOSPC;
3910 : }
3911 :
3912 973 : md_page_size = spdk_max(spdk_max(dev->phys_blocklen, SPDK_BS_PAGE_SIZE),
3913 : opts->md_page_size);
3914 973 : if (opts->cluster_sz < md_page_size) {
3915 : /* Cluster size cannot be smaller than page size */
3916 0 : SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
3917 : opts->cluster_sz, md_page_size);
3918 0 : return -EINVAL;
3919 : }
3920 973 : bs = calloc(1, sizeof(struct spdk_blob_store));
3921 973 : if (!bs) {
3922 0 : return -ENOMEM;
3923 : }
3924 :
3925 973 : ctx = calloc(1, sizeof(struct spdk_bs_load_ctx));
3926 973 : if (!ctx) {
3927 0 : free(bs);
3928 0 : return -ENOMEM;
3929 : }
3930 :
3931 973 : ctx->bs = bs;
3932 973 : ctx->iter_cb_fn = opts->iter_cb_fn;
3933 973 : ctx->iter_cb_arg = opts->iter_cb_arg;
3934 973 : ctx->force_recover = opts->force_recover;
3935 :
3936 973 : ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3937 : SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
3938 973 : if (!ctx->super) {
3939 0 : free(ctx);
3940 0 : free(bs);
3941 0 : return -ENOMEM;
3942 : }
3943 :
3944 973 : RB_INIT(&bs->open_blobs);
3945 973 : TAILQ_INIT(&bs->snapshots);
3946 973 : bs->dev = dev;
3947 973 : bs->md_page_size = md_page_size;
3948 973 : bs->md_thread = spdk_get_thread();
3949 973 : assert(bs->md_thread != NULL);
3950 :
3951 : /*
3952 : * Do not use bs_lba_to_cluster() here since blockcnt may not be an
3953 : * even multiple of the cluster size.
3954 : */
3955 973 : bs->cluster_sz = opts->cluster_sz;
3956 973 : bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
3957 973 : ctx->used_clusters = spdk_bit_array_create(bs->total_clusters);
3958 973 : if (!ctx->used_clusters) {
3959 0 : spdk_free(ctx->super);
3960 0 : free(ctx);
3961 0 : free(bs);
3962 0 : return -ENOMEM;
3963 : }
3964 :
3965 973 : bs->num_free_clusters = bs->total_clusters;
3966 973 : bs->io_unit_size = dev->blocklen;
3967 973 : bs_init_per_cluster_fields(bs);
3968 :
3969 973 : bs->max_channel_ops = opts->max_channel_ops;
3970 973 : bs->super_blob = SPDK_BLOBID_INVALID;
3971 973 : memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
3972 973 : bs->esnap_bs_dev_create = opts->esnap_bs_dev_create;
3973 973 : bs->esnap_ctx = opts->esnap_ctx;
3974 :
3975 : /* The metadata is assumed to be at least 1 page */
3976 973 : bs->used_md_pages = spdk_bit_array_create(1);
3977 973 : bs->used_blobids = spdk_bit_array_create(0);
3978 973 : bs->open_blobids = spdk_bit_array_create(0);
3979 :
3980 973 : spdk_spin_init(&bs->used_lock);
3981 :
3982 973 : spdk_io_device_register(bs, bs_channel_create, bs_channel_destroy,
3983 : sizeof(struct spdk_bs_channel), "blobstore");
3984 973 : rc = bs_register_md_thread(bs);
3985 973 : if (rc == -1) {
3986 0 : spdk_io_device_unregister(bs, NULL);
3987 0 : spdk_spin_destroy(&bs->used_lock);
3988 0 : spdk_bit_array_free(&bs->open_blobids);
3989 0 : spdk_bit_array_free(&bs->used_blobids);
3990 0 : spdk_bit_array_free(&bs->used_md_pages);
3991 0 : spdk_bit_array_free(&ctx->used_clusters);
3992 0 : spdk_free(ctx->super);
3993 0 : free(ctx);
3994 0 : free(bs);
3995 : /* FIXME: this is a lie but don't know how to get a proper error code here */
3996 0 : return -ENOMEM;
3997 : }
3998 :
3999 973 : *_ctx = ctx;
4000 973 : *_bs = bs;
4001 973 : return 0;
4002 : }
4003 :
4004 : static void
4005 30 : bs_load_ctx_fail(struct spdk_bs_load_ctx *ctx, int bserrno)
4006 : {
4007 30 : assert(bserrno != 0);
4008 :
4009 30 : spdk_free(ctx->super);
4010 30 : bs_sequence_finish(ctx->seq, bserrno);
4011 30 : bs_free(ctx->bs);
4012 30 : spdk_bit_array_free(&ctx->used_clusters);
4013 30 : free(ctx);
4014 30 : }
4015 :
4016 : static void
4017 1026 : bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
4018 : struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
4019 : {
4020 : /* Update the values in the super block */
4021 1026 : super->super_blob = bs->super_blob;
4022 1026 : memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
4023 1026 : super->crc = blob_md_page_calc_crc(super);
4024 1026 : bs_sequence_write_dev(seq, super, bs_page_to_lba(bs, 0),
4025 1026 : bs_byte_to_lba(bs, sizeof(*super)),
4026 : cb_fn, cb_arg);
4027 1026 : }
4028 :
4029 : static void
4030 948 : bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
4031 : {
4032 948 : struct spdk_bs_load_ctx *ctx = arg;
4033 : uint64_t mask_size, lba, lba_count;
4034 :
4035 : /* Write out the used clusters mask */
4036 948 : mask_size = ctx->super->used_cluster_mask_len * ctx->bs->md_page_size;
4037 948 : ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
4038 : SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
4039 948 : if (!ctx->mask) {
4040 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4041 0 : return;
4042 : }
4043 :
4044 948 : ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
4045 948 : ctx->mask->length = ctx->bs->total_clusters;
4046 : /* We could get here through the normal unload path, or through dirty
4047 : * shutdown recovery. For the normal unload path, we use the mask from
4048 : * the bit pool. For dirty shutdown recovery, we don't have a bit pool yet -
4049 : * only the bit array from the load ctx.
4050 : */
4051 948 : if (ctx->bs->used_clusters) {
4052 814 : assert(ctx->mask->length == spdk_bit_pool_capacity(ctx->bs->used_clusters));
4053 814 : spdk_bit_pool_store_mask(ctx->bs->used_clusters, ctx->mask->mask);
4054 : } else {
4055 134 : assert(ctx->mask->length == spdk_bit_array_capacity(ctx->used_clusters));
4056 134 : spdk_bit_array_store_mask(ctx->used_clusters, ctx->mask->mask);
4057 : }
4058 948 : lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
4059 948 : lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
4060 948 : bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
4061 : }
4062 :
4063 : static void
4064 948 : bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
4065 : {
4066 948 : struct spdk_bs_load_ctx *ctx = arg;
4067 : uint64_t mask_size, lba, lba_count;
4068 :
4069 948 : mask_size = ctx->super->used_page_mask_len * ctx->bs->md_page_size;
4070 948 : ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
4071 : SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
4072 948 : if (!ctx->mask) {
4073 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4074 0 : return;
4075 : }
4076 :
4077 948 : ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
4078 948 : ctx->mask->length = ctx->super->md_len;
4079 948 : assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
4080 :
4081 948 : spdk_bit_array_store_mask(ctx->bs->used_md_pages, ctx->mask->mask);
4082 948 : lba = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
4083 948 : lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
4084 948 : bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
4085 : }
4086 :
4087 : static void
4088 948 : bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
4089 : {
4090 948 : struct spdk_bs_load_ctx *ctx = arg;
4091 : uint64_t mask_size, lba, lba_count;
4092 :
4093 948 : if (ctx->super->used_blobid_mask_len == 0) {
4094 : /*
4095 : * This is a pre-v3 on-disk format where the blobid mask does not get
4096 : * written to disk.
4097 : */
4098 30 : cb_fn(seq, arg, 0);
4099 30 : return;
4100 : }
4101 :
4102 918 : mask_size = ctx->super->used_blobid_mask_len * ctx->bs->md_page_size;
4103 918 : ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_NUMA_ID_ANY,
4104 : SPDK_MALLOC_DMA);
4105 918 : if (!ctx->mask) {
4106 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4107 0 : return;
4108 : }
4109 :
4110 918 : ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
4111 918 : ctx->mask->length = ctx->super->md_len;
4112 918 : assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
4113 :
4114 918 : spdk_bit_array_store_mask(ctx->bs->used_blobids, ctx->mask->mask);
4115 918 : lba = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
4116 918 : lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
4117 918 : bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
4118 : }
4119 :
4120 : static void
4121 882 : blob_set_thin_provision(struct spdk_blob *blob)
4122 : {
4123 882 : blob_verify_md_op(blob);
4124 882 : blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
4125 882 : blob->state = SPDK_BLOB_STATE_DIRTY;
4126 882 : }
4127 :
4128 : static void
4129 2617 : blob_set_clear_method(struct spdk_blob *blob, enum blob_clear_method clear_method)
4130 : {
4131 2617 : blob_verify_md_op(blob);
4132 2617 : blob->clear_method = clear_method;
4133 2617 : blob->md_ro_flags |= (clear_method << SPDK_BLOB_CLEAR_METHOD_SHIFT);
4134 2617 : blob->state = SPDK_BLOB_STATE_DIRTY;
4135 2617 : }
4136 :
4137 : static void bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno);
4138 :
4139 : static void
4140 30 : bs_delete_corrupted_blob_cpl(void *cb_arg, int bserrno)
4141 : {
4142 30 : struct spdk_bs_load_ctx *ctx = cb_arg;
4143 : spdk_blob_id id;
4144 : int64_t page_num;
4145 :
4146 : /* Iterate to next blob (we can't use spdk_bs_iter_next function as our
4147 : * last blob has been removed */
4148 30 : page_num = bs_blobid_to_page(ctx->blobid);
4149 30 : page_num++;
4150 30 : page_num = spdk_bit_array_find_first_set(ctx->bs->used_blobids, page_num);
4151 30 : if (page_num >= spdk_bit_array_capacity(ctx->bs->used_blobids)) {
4152 30 : bs_load_iter(ctx, NULL, -ENOENT);
4153 30 : return;
4154 : }
4155 :
4156 0 : id = bs_page_to_blobid(page_num);
4157 :
4158 0 : spdk_bs_open_blob(ctx->bs, id, bs_load_iter, ctx);
4159 : }
4160 :
4161 : static void
4162 30 : bs_delete_corrupted_close_cb(void *cb_arg, int bserrno)
4163 : {
4164 30 : struct spdk_bs_load_ctx *ctx = cb_arg;
4165 :
4166 30 : if (bserrno != 0) {
4167 0 : SPDK_ERRLOG("Failed to close corrupted blob\n");
4168 0 : spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
4169 0 : return;
4170 : }
4171 :
4172 30 : spdk_bs_delete_blob(ctx->bs, ctx->blobid, bs_delete_corrupted_blob_cpl, ctx);
4173 : }
4174 :
4175 : static void
4176 30 : bs_delete_corrupted_blob(void *cb_arg, int bserrno)
4177 : {
4178 30 : struct spdk_bs_load_ctx *ctx = cb_arg;
4179 : uint64_t i;
4180 :
4181 30 : if (bserrno != 0) {
4182 0 : SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
4183 0 : spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
4184 0 : return;
4185 : }
4186 :
4187 : /* Snapshot and clone have the same copy of cluster map and extent pages
4188 : * at this point. Let's clear both for snapshot now,
4189 : * so that it won't be cleared for clone later when we remove snapshot.
4190 : * Also set thin provision to pass data corruption check */
4191 330 : for (i = 0; i < ctx->blob->active.num_clusters; i++) {
4192 300 : ctx->blob->active.clusters[i] = 0;
4193 : }
4194 48 : for (i = 0; i < ctx->blob->active.num_extent_pages; i++) {
4195 18 : ctx->blob->active.extent_pages[i] = 0;
4196 : }
4197 :
4198 30 : ctx->blob->active.num_allocated_clusters = 0;
4199 :
4200 30 : ctx->blob->md_ro = false;
4201 :
4202 30 : blob_set_thin_provision(ctx->blob);
4203 :
4204 30 : ctx->blobid = ctx->blob->id;
4205 :
4206 30 : spdk_blob_close(ctx->blob, bs_delete_corrupted_close_cb, ctx);
4207 : }
4208 :
4209 : static void
4210 15 : bs_update_corrupted_blob(void *cb_arg, int bserrno)
4211 : {
4212 15 : struct spdk_bs_load_ctx *ctx = cb_arg;
4213 :
4214 15 : if (bserrno != 0) {
4215 0 : SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
4216 0 : spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
4217 0 : return;
4218 : }
4219 :
4220 15 : ctx->blob->md_ro = false;
4221 15 : blob_remove_xattr(ctx->blob, SNAPSHOT_PENDING_REMOVAL, true);
4222 15 : blob_remove_xattr(ctx->blob, SNAPSHOT_IN_PROGRESS, true);
4223 15 : spdk_blob_set_read_only(ctx->blob);
4224 :
4225 15 : if (ctx->iter_cb_fn) {
4226 0 : ctx->iter_cb_fn(ctx->iter_cb_arg, ctx->blob, 0);
4227 : }
4228 15 : bs_blob_list_add(ctx->blob);
4229 :
4230 15 : spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
4231 : }
4232 :
4233 : static void
4234 45 : bs_examine_clone(void *cb_arg, struct spdk_blob *blob, int bserrno)
4235 : {
4236 45 : struct spdk_bs_load_ctx *ctx = cb_arg;
4237 :
4238 45 : if (bserrno != 0) {
4239 0 : SPDK_ERRLOG("Failed to open clone of a corrupted blob\n");
4240 0 : spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
4241 0 : return;
4242 : }
4243 :
4244 45 : if (blob->parent_id == ctx->blob->id) {
4245 : /* Power failure occurred before updating clone (snapshot delete case)
4246 : * or after updating clone (creating snapshot case) - keep snapshot */
4247 15 : spdk_blob_close(blob, bs_update_corrupted_blob, ctx);
4248 : } else {
4249 : /* Power failure occurred after updating clone (snapshot delete case)
4250 : * or before updating clone (creating snapshot case) - remove snapshot */
4251 30 : spdk_blob_close(blob, bs_delete_corrupted_blob, ctx);
4252 : }
4253 : }
4254 :
4255 : static void
4256 903 : bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
4257 : {
4258 903 : struct spdk_bs_load_ctx *ctx = arg;
4259 : const void *value;
4260 : size_t len;
4261 903 : int rc = 0;
4262 :
4263 903 : if (bserrno == 0) {
4264 : /* Examine blob if it is corrupted after power failure. Fix
4265 : * the ones that can be fixed and remove any other corrupted
4266 : * ones. If it is not corrupted just process it */
4267 552 : rc = blob_get_xattr_value(blob, SNAPSHOT_PENDING_REMOVAL, &value, &len, true);
4268 552 : if (rc != 0) {
4269 527 : rc = blob_get_xattr_value(blob, SNAPSHOT_IN_PROGRESS, &value, &len, true);
4270 527 : if (rc != 0) {
4271 : /* Not corrupted - process it and continue with iterating through blobs */
4272 507 : if (ctx->iter_cb_fn) {
4273 42 : ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
4274 : }
4275 507 : bs_blob_list_add(blob);
4276 507 : spdk_bs_iter_next(ctx->bs, blob, bs_load_iter, ctx);
4277 552 : return;
4278 : }
4279 :
4280 : }
4281 :
4282 45 : assert(len == sizeof(spdk_blob_id));
4283 :
4284 45 : ctx->blob = blob;
4285 :
4286 : /* Open clone to check if we are able to fix this blob or should we remove it */
4287 45 : spdk_bs_open_blob(ctx->bs, *(spdk_blob_id *)value, bs_examine_clone, ctx);
4288 45 : return;
4289 351 : } else if (bserrno == -ENOENT) {
4290 351 : bserrno = 0;
4291 : } else {
4292 : /*
4293 : * This case needs to be looked at further. Same problem
4294 : * exists with applications that rely on explicit blob
4295 : * iteration. We should just skip the blob that failed
4296 : * to load and continue on to the next one.
4297 : */
4298 0 : SPDK_ERRLOG("Error in iterating blobs\n");
4299 : }
4300 :
4301 351 : ctx->iter_cb_fn = NULL;
4302 :
4303 351 : spdk_free(ctx->super);
4304 351 : spdk_free(ctx->mask);
4305 351 : bs_sequence_finish(ctx->seq, bserrno);
4306 351 : free(ctx);
4307 : }
4308 :
4309 : static void bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
4310 :
4311 : static void
4312 351 : bs_load_complete(struct spdk_bs_load_ctx *ctx)
4313 : {
4314 351 : ctx->bs->used_clusters = spdk_bit_pool_create_from_array(ctx->used_clusters);
4315 351 : if (ctx->dumping) {
4316 0 : bs_dump_read_md_page(ctx->seq, ctx);
4317 0 : return;
4318 : }
4319 351 : spdk_bs_iter_first(ctx->bs, bs_load_iter, ctx);
4320 : }
4321 :
4322 : static void
4323 217 : bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4324 : {
4325 217 : struct spdk_bs_load_ctx *ctx = cb_arg;
4326 : int rc;
4327 :
4328 : /* The type must be correct */
4329 217 : assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
4330 :
4331 : /* The length of the mask (in bits) must not be greater than
4332 : * the length of the buffer (converted to bits) */
4333 217 : assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * ctx->super->md_page_size * 8));
4334 :
4335 : /* The length of the mask must be exactly equal to the size
4336 : * (in pages) of the metadata region */
4337 217 : assert(ctx->mask->length == ctx->super->md_len);
4338 :
4339 217 : rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->mask->length);
4340 217 : if (rc < 0) {
4341 0 : spdk_free(ctx->mask);
4342 0 : bs_load_ctx_fail(ctx, rc);
4343 0 : return;
4344 : }
4345 :
4346 217 : spdk_bit_array_load_mask(ctx->bs->used_blobids, ctx->mask->mask);
4347 217 : bs_load_complete(ctx);
4348 : }
4349 :
4350 : static void
4351 217 : bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4352 : {
4353 217 : struct spdk_bs_load_ctx *ctx = cb_arg;
4354 : uint64_t lba, lba_count, mask_size;
4355 : int rc;
4356 :
4357 217 : if (bserrno != 0) {
4358 0 : bs_load_ctx_fail(ctx, bserrno);
4359 0 : return;
4360 : }
4361 :
4362 : /* The type must be correct */
4363 217 : assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
4364 : /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
4365 217 : assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
4366 : struct spdk_blob_md_page) * 8));
4367 : /*
4368 : * The length of the mask must be equal to or larger than the total number of clusters. It may be
4369 : * larger than the total number of clusters due to a failure spdk_bs_grow.
4370 : */
4371 217 : assert(ctx->mask->length >= ctx->bs->total_clusters);
4372 217 : if (ctx->mask->length > ctx->bs->total_clusters) {
4373 5 : SPDK_WARNLOG("Shrink the used_custers mask length to total_clusters");
4374 5 : ctx->mask->length = ctx->bs->total_clusters;
4375 : }
4376 :
4377 217 : rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->mask->length);
4378 217 : if (rc < 0) {
4379 0 : spdk_free(ctx->mask);
4380 0 : bs_load_ctx_fail(ctx, rc);
4381 0 : return;
4382 : }
4383 :
4384 217 : spdk_bit_array_load_mask(ctx->used_clusters, ctx->mask->mask);
4385 217 : ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->used_clusters);
4386 217 : assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
4387 :
4388 217 : spdk_free(ctx->mask);
4389 :
4390 : /* Read the used blobids mask */
4391 217 : mask_size = ctx->super->used_blobid_mask_len * ctx->super->md_page_size;
4392 217 : ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_NUMA_ID_ANY,
4393 : SPDK_MALLOC_DMA);
4394 217 : if (!ctx->mask) {
4395 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4396 0 : return;
4397 : }
4398 217 : lba = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
4399 217 : lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
4400 217 : bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
4401 : bs_load_used_blobids_cpl, ctx);
4402 : }
4403 :
4404 : static void
4405 217 : bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4406 : {
4407 217 : struct spdk_bs_load_ctx *ctx = cb_arg;
4408 : uint64_t lba, lba_count, mask_size;
4409 : int rc;
4410 :
4411 217 : if (bserrno != 0) {
4412 0 : bs_load_ctx_fail(ctx, bserrno);
4413 0 : return;
4414 : }
4415 :
4416 : /* The type must be correct */
4417 217 : assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
4418 : /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
4419 217 : assert(ctx->mask->length <= (ctx->super->used_page_mask_len * ctx->super->md_page_size *
4420 : 8));
4421 : /* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
4422 217 : if (ctx->mask->length != ctx->super->md_len) {
4423 0 : SPDK_ERRLOG("mismatched md_len in used_pages mask: "
4424 : "mask->length=%" PRIu32 " super->md_len=%" PRIu32 "\n",
4425 : ctx->mask->length, ctx->super->md_len);
4426 0 : assert(false);
4427 : }
4428 :
4429 217 : rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->mask->length);
4430 217 : if (rc < 0) {
4431 0 : spdk_free(ctx->mask);
4432 0 : bs_load_ctx_fail(ctx, rc);
4433 0 : return;
4434 : }
4435 :
4436 217 : spdk_bit_array_load_mask(ctx->bs->used_md_pages, ctx->mask->mask);
4437 217 : spdk_free(ctx->mask);
4438 :
4439 : /* Read the used clusters mask */
4440 217 : mask_size = ctx->super->used_cluster_mask_len * ctx->super->md_page_size;
4441 217 : ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_NUMA_ID_ANY,
4442 : SPDK_MALLOC_DMA);
4443 217 : if (!ctx->mask) {
4444 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4445 0 : return;
4446 : }
4447 217 : lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
4448 217 : lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
4449 217 : bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
4450 : bs_load_used_clusters_cpl, ctx);
4451 : }
4452 :
4453 : static void
4454 217 : bs_load_read_used_pages(struct spdk_bs_load_ctx *ctx)
4455 : {
4456 : uint64_t lba, lba_count, mask_size;
4457 :
4458 : /* Read the used pages mask */
4459 217 : mask_size = ctx->super->used_page_mask_len * ctx->super->md_page_size;
4460 217 : ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
4461 : SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
4462 217 : if (!ctx->mask) {
4463 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4464 0 : return;
4465 : }
4466 :
4467 217 : lba = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
4468 217 : lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
4469 217 : bs_sequence_read_dev(ctx->seq, ctx->mask, lba, lba_count,
4470 : bs_load_used_pages_cpl, ctx);
4471 : }
4472 :
4473 : static int
4474 323 : bs_load_replay_md_parse_page(struct spdk_bs_load_ctx *ctx, struct spdk_blob_md_page *page)
4475 : {
4476 323 : struct spdk_blob_store *bs = ctx->bs;
4477 : struct spdk_blob_md_descriptor *desc;
4478 323 : size_t cur_desc = 0;
4479 :
4480 323 : desc = (struct spdk_blob_md_descriptor *)page->descriptors;
4481 933 : while (cur_desc < sizeof(page->descriptors)) {
4482 933 : if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
4483 298 : if (desc->length == 0) {
4484 : /* If padding and length are 0, this terminates the page */
4485 298 : break;
4486 : }
4487 635 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
4488 : struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle;
4489 : unsigned int i, j;
4490 68 : unsigned int cluster_count = 0;
4491 : uint32_t cluster_idx;
4492 :
4493 68 : desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
4494 :
4495 136 : for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
4496 828 : for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
4497 760 : cluster_idx = desc_extent_rle->extents[i].cluster_idx;
4498 : /*
4499 : * cluster_idx = 0 means an unallocated cluster - don't mark that
4500 : * in the used cluster map.
4501 : */
4502 760 : if (cluster_idx != 0) {
4503 540 : SPDK_NOTICELOG("Recover: cluster %" PRIu32 "\n", cluster_idx + j);
4504 540 : spdk_bit_array_set(ctx->used_clusters, cluster_idx + j);
4505 540 : if (bs->num_free_clusters == 0) {
4506 0 : return -ENOSPC;
4507 : }
4508 540 : bs->num_free_clusters--;
4509 : }
4510 760 : cluster_count++;
4511 : }
4512 : }
4513 68 : if (cluster_count == 0) {
4514 0 : return -EINVAL;
4515 : }
4516 567 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
4517 : struct spdk_blob_md_descriptor_extent_page *desc_extent;
4518 : uint32_t i;
4519 78 : uint32_t cluster_count = 0;
4520 : uint32_t cluster_idx;
4521 : size_t cluster_idx_length;
4522 :
4523 78 : desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc;
4524 78 : cluster_idx_length = desc_extent->length - sizeof(desc_extent->start_cluster_idx);
4525 :
4526 78 : if (desc_extent->length <= sizeof(desc_extent->start_cluster_idx) ||
4527 78 : (cluster_idx_length % sizeof(desc_extent->cluster_idx[0]) != 0)) {
4528 0 : return -EINVAL;
4529 : }
4530 :
4531 978 : for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) {
4532 900 : cluster_idx = desc_extent->cluster_idx[i];
4533 : /*
4534 : * cluster_idx = 0 means an unallocated cluster - don't mark that
4535 : * in the used cluster map.
4536 : */
4537 900 : if (cluster_idx != 0) {
4538 900 : if (cluster_idx < desc_extent->start_cluster_idx &&
4539 0 : cluster_idx >= desc_extent->start_cluster_idx + cluster_count) {
4540 0 : return -EINVAL;
4541 : }
4542 900 : spdk_bit_array_set(ctx->used_clusters, cluster_idx);
4543 900 : if (bs->num_free_clusters == 0) {
4544 0 : return -ENOSPC;
4545 : }
4546 900 : bs->num_free_clusters--;
4547 : }
4548 900 : cluster_count++;
4549 : }
4550 :
4551 78 : if (cluster_count == 0) {
4552 0 : return -EINVAL;
4553 : }
4554 489 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
4555 : /* Skip this item */
4556 394 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
4557 : /* Skip this item */
4558 318 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
4559 : /* Skip this item */
4560 123 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) {
4561 : struct spdk_blob_md_descriptor_extent_table *desc_extent_table;
4562 123 : uint32_t num_extent_pages = ctx->num_extent_pages;
4563 : uint32_t i;
4564 : size_t extent_pages_length;
4565 : void *tmp;
4566 :
4567 123 : desc_extent_table = (struct spdk_blob_md_descriptor_extent_table *)desc;
4568 123 : extent_pages_length = desc_extent_table->length - sizeof(desc_extent_table->num_clusters);
4569 :
4570 123 : if (desc_extent_table->length == 0 ||
4571 123 : (extent_pages_length % sizeof(desc_extent_table->extent_page[0]) != 0)) {
4572 0 : return -EINVAL;
4573 : }
4574 :
4575 240 : for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
4576 117 : if (desc_extent_table->extent_page[i].page_idx != 0) {
4577 78 : if (desc_extent_table->extent_page[i].num_pages != 1) {
4578 0 : return -EINVAL;
4579 : }
4580 78 : num_extent_pages += 1;
4581 : }
4582 : }
4583 :
4584 123 : if (num_extent_pages > 0) {
4585 78 : tmp = realloc(ctx->extent_page_num, num_extent_pages * sizeof(uint32_t));
4586 78 : if (tmp == NULL) {
4587 0 : return -ENOMEM;
4588 : }
4589 78 : ctx->extent_page_num = tmp;
4590 :
4591 : /* Extent table entries contain md page numbers for extent pages.
4592 : * Zeroes represent unallocated extent pages, those are run-length-encoded.
4593 : */
4594 156 : for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
4595 78 : if (desc_extent_table->extent_page[i].page_idx != 0) {
4596 78 : ctx->extent_page_num[ctx->num_extent_pages] = desc_extent_table->extent_page[i].page_idx;
4597 78 : ctx->num_extent_pages += 1;
4598 : }
4599 : }
4600 : }
4601 : } else {
4602 : /* Error */
4603 0 : return -EINVAL;
4604 : }
4605 : /* Advance to the next descriptor */
4606 635 : cur_desc += sizeof(*desc) + desc->length;
4607 635 : if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
4608 25 : break;
4609 : }
4610 610 : desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
4611 : }
4612 323 : return 0;
4613 : }
4614 :
4615 : static bool
4616 1884 : bs_load_cur_extent_page_valid(struct spdk_blob_md_page *page)
4617 : {
4618 : uint32_t crc;
4619 1884 : struct spdk_blob_md_descriptor *desc = (struct spdk_blob_md_descriptor *)page->descriptors;
4620 : size_t desc_len;
4621 :
4622 1884 : crc = blob_md_page_calc_crc(page);
4623 1884 : if (crc != page->crc) {
4624 0 : return false;
4625 : }
4626 :
4627 : /* Extent page should always be of sequence num 0. */
4628 1884 : if (page->sequence_num != 0) {
4629 55 : return false;
4630 : }
4631 :
4632 : /* Descriptor type must be EXTENT_PAGE. */
4633 1829 : if (desc->type != SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
4634 195 : return false;
4635 : }
4636 :
4637 : /* Descriptor length cannot exceed the page. */
4638 1634 : desc_len = sizeof(*desc) + desc->length;
4639 1634 : if (desc_len > sizeof(page->descriptors)) {
4640 0 : return false;
4641 : }
4642 :
4643 : /* It has to be the only descriptor in the page. */
4644 1634 : if (desc_len + sizeof(*desc) <= sizeof(page->descriptors)) {
4645 1634 : desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + desc_len);
4646 1634 : if (desc->length != 0) {
4647 0 : return false;
4648 : }
4649 : }
4650 :
4651 1634 : return true;
4652 : }
4653 :
4654 : static bool
4655 8531 : bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
4656 : {
4657 : uint32_t crc;
4658 8531 : struct spdk_blob_md_page *page = ctx->page;
4659 :
4660 8531 : crc = blob_md_page_calc_crc(page);
4661 8531 : if (crc != page->crc) {
4662 8254 : return false;
4663 : }
4664 :
4665 : /* First page of a sequence should match the blobid. */
4666 277 : if (page->sequence_num == 0 &&
4667 222 : bs_page_to_blobid(ctx->cur_page) != page->id) {
4668 27 : return false;
4669 : }
4670 250 : assert(bs_load_cur_extent_page_valid(page) == false);
4671 :
4672 250 : return true;
4673 : }
4674 :
4675 : static void bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx);
4676 :
4677 : static void
4678 134 : bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4679 : {
4680 134 : struct spdk_bs_load_ctx *ctx = cb_arg;
4681 :
4682 134 : if (bserrno != 0) {
4683 0 : bs_load_ctx_fail(ctx, bserrno);
4684 0 : return;
4685 : }
4686 :
4687 134 : bs_load_complete(ctx);
4688 : }
4689 :
4690 : static void
4691 134 : bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4692 : {
4693 134 : struct spdk_bs_load_ctx *ctx = cb_arg;
4694 :
4695 134 : spdk_free(ctx->mask);
4696 134 : ctx->mask = NULL;
4697 :
4698 134 : if (bserrno != 0) {
4699 0 : bs_load_ctx_fail(ctx, bserrno);
4700 0 : return;
4701 : }
4702 :
4703 134 : bs_write_used_clusters(seq, ctx, bs_load_write_used_clusters_cpl);
4704 : }
4705 :
4706 : static void
4707 134 : bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4708 : {
4709 134 : struct spdk_bs_load_ctx *ctx = cb_arg;
4710 :
4711 134 : spdk_free(ctx->mask);
4712 134 : ctx->mask = NULL;
4713 :
4714 134 : if (bserrno != 0) {
4715 0 : bs_load_ctx_fail(ctx, bserrno);
4716 0 : return;
4717 : }
4718 :
4719 134 : bs_write_used_blobids(seq, ctx, bs_load_write_used_blobids_cpl);
4720 : }
4721 :
4722 : static void
4723 134 : bs_load_write_used_md(struct spdk_bs_load_ctx *ctx)
4724 : {
4725 134 : bs_write_used_md(ctx->seq, ctx, bs_load_write_used_pages_cpl);
4726 134 : }
4727 :
4728 : static void
4729 8481 : bs_load_replay_md_chain_cpl(struct spdk_bs_load_ctx *ctx)
4730 : {
4731 : uint64_t num_md_clusters;
4732 : uint64_t i;
4733 :
4734 8481 : ctx->in_page_chain = false;
4735 :
4736 : do {
4737 8576 : ctx->page_index++;
4738 8576 : } while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
4739 :
4740 8481 : if (ctx->page_index < ctx->super->md_len) {
4741 8347 : ctx->cur_page = ctx->page_index;
4742 8347 : bs_load_replay_cur_md_page(ctx);
4743 : } else {
4744 : /* Claim all of the clusters used by the metadata */
4745 134 : num_md_clusters = spdk_divide_round_up(
4746 268 : ctx->super->md_start + ctx->super->md_len, ctx->bs->pages_per_cluster);
4747 629 : for (i = 0; i < num_md_clusters; i++) {
4748 495 : spdk_bit_array_set(ctx->used_clusters, i);
4749 : }
4750 134 : ctx->bs->num_free_clusters -= num_md_clusters;
4751 134 : spdk_free(ctx->page);
4752 134 : bs_load_write_used_md(ctx);
4753 : }
4754 8481 : }
4755 :
4756 : static void
4757 78 : bs_load_replay_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4758 : {
4759 78 : struct spdk_bs_load_ctx *ctx = cb_arg;
4760 : uint32_t page_num;
4761 : uint64_t i;
4762 :
4763 78 : if (bserrno != 0) {
4764 0 : spdk_free(ctx->extent_pages);
4765 0 : bs_load_ctx_fail(ctx, bserrno);
4766 0 : return;
4767 : }
4768 :
4769 156 : for (i = 0; i < ctx->num_extent_pages; i++) {
4770 : /* Extent pages are only read when present within in chain md.
4771 : * Integrity of md is not right if that page was not a valid extent page. */
4772 78 : if (bs_load_cur_extent_page_valid(&ctx->extent_pages[i]) != true) {
4773 0 : spdk_free(ctx->extent_pages);
4774 0 : bs_load_ctx_fail(ctx, -EILSEQ);
4775 0 : return;
4776 : }
4777 :
4778 78 : page_num = ctx->extent_page_num[i];
4779 78 : spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
4780 78 : if (bs_load_replay_md_parse_page(ctx, &ctx->extent_pages[i])) {
4781 0 : spdk_free(ctx->extent_pages);
4782 0 : bs_load_ctx_fail(ctx, -EILSEQ);
4783 0 : return;
4784 : }
4785 : }
4786 :
4787 78 : spdk_free(ctx->extent_pages);
4788 78 : free(ctx->extent_page_num);
4789 78 : ctx->extent_page_num = NULL;
4790 78 : ctx->num_extent_pages = 0;
4791 :
4792 78 : bs_load_replay_md_chain_cpl(ctx);
4793 : }
4794 :
4795 : static void
4796 78 : bs_load_replay_extent_pages(struct spdk_bs_load_ctx *ctx)
4797 : {
4798 : spdk_bs_batch_t *batch;
4799 : uint32_t page;
4800 : uint64_t lba;
4801 : uint64_t i;
4802 :
4803 78 : ctx->extent_pages = spdk_zmalloc(ctx->super->md_page_size * ctx->num_extent_pages, 0,
4804 : NULL, SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
4805 78 : if (!ctx->extent_pages) {
4806 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4807 0 : return;
4808 : }
4809 :
4810 78 : batch = bs_sequence_to_batch(ctx->seq, bs_load_replay_extent_page_cpl, ctx);
4811 :
4812 156 : for (i = 0; i < ctx->num_extent_pages; i++) {
4813 78 : page = ctx->extent_page_num[i];
4814 78 : assert(page < ctx->super->md_len);
4815 78 : lba = bs_md_page_to_lba(ctx->bs, page);
4816 78 : bs_batch_read_dev(batch, &ctx->extent_pages[i], lba,
4817 78 : bs_byte_to_lba(ctx->bs, ctx->super->md_page_size));
4818 : }
4819 :
4820 78 : bs_batch_close(batch);
4821 : }
4822 :
4823 : static void
4824 8531 : bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4825 : {
4826 8531 : struct spdk_bs_load_ctx *ctx = cb_arg;
4827 : uint32_t page_num;
4828 : struct spdk_blob_md_page *page;
4829 :
4830 8531 : if (bserrno != 0) {
4831 0 : bs_load_ctx_fail(ctx, bserrno);
4832 0 : return;
4833 : }
4834 :
4835 8531 : page_num = ctx->cur_page;
4836 8531 : page = ctx->page;
4837 8531 : if (bs_load_cur_md_page_valid(ctx) == true) {
4838 250 : if (page->sequence_num == 0 || ctx->in_page_chain == true) {
4839 245 : spdk_spin_lock(&ctx->bs->used_lock);
4840 245 : bs_claim_md_page(ctx->bs, page_num);
4841 245 : spdk_spin_unlock(&ctx->bs->used_lock);
4842 245 : if (page->sequence_num == 0) {
4843 195 : SPDK_NOTICELOG("Recover: blob 0x%" PRIx32 "\n", page_num);
4844 195 : spdk_bit_array_set(ctx->bs->used_blobids, page_num);
4845 : }
4846 245 : if (bs_load_replay_md_parse_page(ctx, page)) {
4847 0 : bs_load_ctx_fail(ctx, -EILSEQ);
4848 0 : return;
4849 : }
4850 245 : if (page->next != SPDK_INVALID_MD_PAGE) {
4851 50 : ctx->in_page_chain = true;
4852 50 : ctx->cur_page = page->next;
4853 50 : bs_load_replay_cur_md_page(ctx);
4854 50 : return;
4855 : }
4856 195 : if (ctx->num_extent_pages != 0) {
4857 78 : bs_load_replay_extent_pages(ctx);
4858 78 : return;
4859 : }
4860 : }
4861 : }
4862 8403 : bs_load_replay_md_chain_cpl(ctx);
4863 : }
4864 :
4865 : static void
4866 8531 : bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx)
4867 : {
4868 : uint64_t lba;
4869 :
4870 8531 : assert(ctx->cur_page < ctx->super->md_len);
4871 8531 : lba = bs_md_page_to_lba(ctx->bs, ctx->cur_page);
4872 8531 : bs_sequence_read_dev(ctx->seq, ctx->page, lba,
4873 8531 : bs_byte_to_lba(ctx->bs, ctx->super->md_page_size),
4874 : bs_load_replay_md_cpl, ctx);
4875 8531 : }
4876 :
4877 : static void
4878 134 : bs_load_replay_md(struct spdk_bs_load_ctx *ctx)
4879 : {
4880 134 : ctx->page_index = 0;
4881 134 : ctx->cur_page = 0;
4882 134 : ctx->page = spdk_zmalloc(ctx->bs->md_page_size, 0,
4883 : NULL, SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
4884 134 : if (!ctx->page) {
4885 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4886 0 : return;
4887 : }
4888 134 : bs_load_replay_cur_md_page(ctx);
4889 : }
4890 :
4891 : static void
4892 134 : bs_recover(struct spdk_bs_load_ctx *ctx)
4893 : {
4894 : int rc;
4895 :
4896 134 : SPDK_NOTICELOG("Performing recovery on blobstore\n");
4897 134 : rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
4898 134 : if (rc < 0) {
4899 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4900 0 : return;
4901 : }
4902 :
4903 134 : rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
4904 134 : if (rc < 0) {
4905 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4906 0 : return;
4907 : }
4908 :
4909 134 : rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->bs->total_clusters);
4910 134 : if (rc < 0) {
4911 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4912 0 : return;
4913 : }
4914 :
4915 134 : rc = spdk_bit_array_resize(&ctx->bs->open_blobids, ctx->super->md_len);
4916 134 : if (rc < 0) {
4917 0 : bs_load_ctx_fail(ctx, -ENOMEM);
4918 0 : return;
4919 : }
4920 :
4921 134 : ctx->bs->num_free_clusters = ctx->bs->total_clusters;
4922 134 : bs_load_replay_md(ctx);
4923 : }
4924 :
4925 : static int
4926 346 : bs_parse_super(struct spdk_bs_load_ctx *ctx)
4927 : {
4928 : int rc;
4929 :
4930 346 : if (ctx->super->size == 0) {
4931 10 : ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
4932 : }
4933 :
4934 346 : if (ctx->super->io_unit_size == 0) {
4935 10 : ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE;
4936 : }
4937 346 : if (ctx->super->md_page_size == 0) {
4938 5 : ctx->super->md_page_size = SPDK_BS_PAGE_SIZE;
4939 : }
4940 :
4941 346 : ctx->bs->clean = 1;
4942 346 : ctx->bs->cluster_sz = ctx->super->cluster_size;
4943 346 : ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
4944 346 : ctx->bs->io_unit_size = ctx->super->io_unit_size;
4945 346 : ctx->bs->md_page_size = ctx->super->md_page_size;
4946 346 : bs_init_per_cluster_fields(ctx->bs);
4947 346 : rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->bs->total_clusters);
4948 346 : if (rc < 0) {
4949 0 : return -ENOMEM;
4950 : }
4951 346 : ctx->bs->md_start = ctx->super->md_start;
4952 346 : ctx->bs->md_len = ctx->super->md_len;
4953 346 : rc = spdk_bit_array_resize(&ctx->bs->open_blobids, ctx->bs->md_len);
4954 346 : if (rc < 0) {
4955 0 : return -ENOMEM;
4956 : }
4957 :
4958 1038 : ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up(
4959 692 : ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
4960 346 : ctx->bs->super_blob = ctx->super->super_blob;
4961 346 : memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
4962 :
4963 346 : return 0;
4964 : }
4965 :
4966 : static void
4967 376 : bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4968 : {
4969 376 : struct spdk_bs_load_ctx *ctx = cb_arg;
4970 : int rc;
4971 :
4972 376 : rc = bs_super_validate(ctx->super, ctx->bs);
4973 376 : if (rc != 0) {
4974 30 : bs_load_ctx_fail(ctx, rc);
4975 30 : return;
4976 : }
4977 :
4978 346 : rc = bs_parse_super(ctx);
4979 346 : if (rc < 0) {
4980 0 : bs_load_ctx_fail(ctx, rc);
4981 0 : return;
4982 : }
4983 :
4984 346 : if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0 || ctx->force_recover) {
4985 134 : bs_recover(ctx);
4986 : } else {
4987 212 : bs_load_read_used_pages(ctx);
4988 : }
4989 : }
4990 :
4991 : static inline int
4992 380 : bs_opts_copy(struct spdk_bs_opts *src, struct spdk_bs_opts *dst)
4993 : {
4994 :
4995 380 : if (!src->opts_size) {
4996 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
4997 0 : return -1;
4998 : }
4999 :
5000 : #define FIELD_OK(field) \
5001 : offsetof(struct spdk_bs_opts, field) + sizeof(src->field) <= src->opts_size
5002 :
5003 : #define SET_FIELD(field) \
5004 : if (FIELD_OK(field)) { \
5005 : dst->field = src->field; \
5006 : } \
5007 :
5008 380 : SET_FIELD(cluster_sz);
5009 380 : SET_FIELD(num_md_pages);
5010 380 : SET_FIELD(max_md_ops);
5011 380 : SET_FIELD(max_channel_ops);
5012 380 : SET_FIELD(clear_method);
5013 :
5014 380 : if (FIELD_OK(bstype)) {
5015 380 : memcpy(&dst->bstype, &src->bstype, sizeof(dst->bstype));
5016 : }
5017 380 : SET_FIELD(md_page_size);
5018 380 : SET_FIELD(iter_cb_fn);
5019 380 : SET_FIELD(iter_cb_arg);
5020 380 : SET_FIELD(force_recover);
5021 380 : SET_FIELD(esnap_bs_dev_create);
5022 380 : SET_FIELD(esnap_ctx);
5023 :
5024 380 : dst->opts_size = src->opts_size;
5025 :
5026 : /* You should not remove this statement, but need to update the assert statement
5027 : * if you add a new field, and also add a corresponding SET_FIELD statement */
5028 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bs_opts) == 88, "Incorrect size");
5029 :
5030 : #undef FIELD_OK
5031 : #undef SET_FIELD
5032 :
5033 380 : return 0;
5034 : }
5035 :
5036 : void
5037 391 : spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
5038 : spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
5039 : {
5040 : struct spdk_blob_store *bs;
5041 : struct spdk_bs_cpl cpl;
5042 : struct spdk_bs_load_ctx *ctx;
5043 391 : struct spdk_bs_opts opts = {};
5044 : int err;
5045 :
5046 391 : SPDK_DEBUGLOG(blob, "Loading blobstore from dev %p\n", dev);
5047 :
5048 391 : if ((dev->phys_blocklen % dev->blocklen) != 0) {
5049 5 : SPDK_DEBUGLOG(blob, "unsupported dev block length of %d\n", dev->blocklen);
5050 5 : dev->destroy(dev);
5051 5 : cb_fn(cb_arg, NULL, -EINVAL);
5052 15 : return;
5053 : }
5054 :
5055 386 : spdk_bs_opts_init(&opts, sizeof(opts));
5056 386 : if (o) {
5057 152 : if (bs_opts_copy(o, &opts)) {
5058 0 : return;
5059 : }
5060 : }
5061 :
5062 386 : if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
5063 10 : dev->destroy(dev);
5064 10 : cb_fn(cb_arg, NULL, -EINVAL);
5065 10 : return;
5066 : }
5067 :
5068 376 : err = bs_alloc(dev, &opts, &bs, &ctx);
5069 376 : if (err) {
5070 0 : dev->destroy(dev);
5071 0 : cb_fn(cb_arg, NULL, err);
5072 0 : return;
5073 : }
5074 :
5075 376 : cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
5076 376 : cpl.u.bs_handle.cb_fn = cb_fn;
5077 376 : cpl.u.bs_handle.cb_arg = cb_arg;
5078 376 : cpl.u.bs_handle.bs = bs;
5079 :
5080 376 : ctx->seq = bs_sequence_start_bs(bs->md_channel, &cpl);
5081 376 : if (!ctx->seq) {
5082 0 : spdk_free(ctx->super);
5083 0 : free(ctx);
5084 0 : bs_free(bs);
5085 0 : cb_fn(cb_arg, NULL, -ENOMEM);
5086 0 : return;
5087 : }
5088 :
5089 : /* Read the super block */
5090 376 : bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0),
5091 376 : bs_byte_to_lba(bs, sizeof(*ctx->super)),
5092 : bs_load_super_cpl, ctx);
5093 : }
5094 :
5095 : /* END spdk_bs_load */
5096 :
5097 : /* START spdk_bs_dump */
5098 :
5099 : static void
5100 0 : bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
5101 : {
5102 0 : spdk_free(ctx->super);
5103 :
5104 : /*
5105 : * We need to defer calling bs_call_cpl() until after
5106 : * dev destruction, so tuck these away for later use.
5107 : */
5108 0 : ctx->bs->unload_err = bserrno;
5109 0 : memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
5110 0 : seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
5111 :
5112 0 : bs_sequence_finish(seq, 0);
5113 0 : bs_free(ctx->bs);
5114 0 : free(ctx);
5115 0 : }
5116 :
5117 : static void
5118 0 : bs_dump_print_xattr(struct spdk_bs_load_ctx *ctx, struct spdk_blob_md_descriptor *desc)
5119 : {
5120 : struct spdk_blob_md_descriptor_xattr *desc_xattr;
5121 : uint32_t i;
5122 : const char *type;
5123 :
5124 0 : desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
5125 :
5126 0 : if (desc_xattr->length !=
5127 : sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
5128 0 : desc_xattr->name_length + desc_xattr->value_length) {
5129 : }
5130 :
5131 0 : memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
5132 0 : ctx->xattr_name[desc_xattr->name_length] = '\0';
5133 0 : if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
5134 0 : type = "XATTR";
5135 0 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
5136 0 : type = "XATTR_INTERNAL";
5137 : } else {
5138 0 : assert(false);
5139 : type = "XATTR_?";
5140 : }
5141 0 : fprintf(ctx->fp, "%s: name = \"%s\"\n", type, ctx->xattr_name);
5142 0 : fprintf(ctx->fp, " value = \"");
5143 0 : ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
5144 0 : (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
5145 0 : desc_xattr->value_length);
5146 0 : fprintf(ctx->fp, "\"\n");
5147 0 : for (i = 0; i < desc_xattr->value_length; i++) {
5148 0 : if (i % 16 == 0) {
5149 0 : fprintf(ctx->fp, " ");
5150 : }
5151 0 : fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
5152 0 : if ((i + 1) % 16 == 0) {
5153 0 : fprintf(ctx->fp, "\n");
5154 : }
5155 : }
5156 0 : if (i % 16 != 0) {
5157 0 : fprintf(ctx->fp, "\n");
5158 : }
5159 0 : }
5160 :
5161 : struct type_flag_desc {
5162 : uint64_t mask;
5163 : uint64_t val;
5164 : const char *name;
5165 : };
5166 :
5167 : static void
5168 0 : bs_dump_print_type_bits(struct spdk_bs_load_ctx *ctx, uint64_t flags,
5169 : struct type_flag_desc *desc, size_t numflags)
5170 : {
5171 0 : uint64_t covered = 0;
5172 : size_t i;
5173 :
5174 0 : for (i = 0; i < numflags; i++) {
5175 0 : if ((desc[i].mask & flags) != desc[i].val) {
5176 0 : continue;
5177 : }
5178 0 : fprintf(ctx->fp, "\t\t 0x%016" PRIx64 " %s", desc[i].val, desc[i].name);
5179 0 : if (desc[i].mask != desc[i].val) {
5180 0 : fprintf(ctx->fp, " (mask 0x%" PRIx64 " value 0x%" PRIx64 ")",
5181 0 : desc[i].mask, desc[i].val);
5182 : }
5183 0 : fprintf(ctx->fp, "\n");
5184 0 : covered |= desc[i].mask;
5185 : }
5186 0 : if ((flags & ~covered) != 0) {
5187 0 : fprintf(ctx->fp, "\t\t 0x%016" PRIx64 " Unknown\n", flags & ~covered);
5188 : }
5189 0 : }
5190 :
5191 : static void
5192 0 : bs_dump_print_type_flags(struct spdk_bs_load_ctx *ctx, struct spdk_blob_md_descriptor *desc)
5193 : {
5194 : struct spdk_blob_md_descriptor_flags *type_desc;
5195 : #define ADD_FLAG(f) { f, f, #f }
5196 : #define ADD_MASK_VAL(m, v) { m, v, #v }
5197 : static struct type_flag_desc invalid[] = {
5198 : ADD_FLAG(SPDK_BLOB_THIN_PROV),
5199 : ADD_FLAG(SPDK_BLOB_INTERNAL_XATTR),
5200 : ADD_FLAG(SPDK_BLOB_EXTENT_TABLE),
5201 : };
5202 : static struct type_flag_desc data_ro[] = {
5203 : ADD_FLAG(SPDK_BLOB_READ_ONLY),
5204 : };
5205 : static struct type_flag_desc md_ro[] = {
5206 : ADD_MASK_VAL(SPDK_BLOB_MD_RO_FLAGS_MASK, BLOB_CLEAR_WITH_DEFAULT),
5207 : ADD_MASK_VAL(SPDK_BLOB_MD_RO_FLAGS_MASK, BLOB_CLEAR_WITH_NONE),
5208 : ADD_MASK_VAL(SPDK_BLOB_MD_RO_FLAGS_MASK, BLOB_CLEAR_WITH_UNMAP),
5209 : ADD_MASK_VAL(SPDK_BLOB_MD_RO_FLAGS_MASK, BLOB_CLEAR_WITH_WRITE_ZEROES),
5210 : };
5211 : #undef ADD_FLAG
5212 : #undef ADD_MASK_VAL
5213 :
5214 0 : type_desc = (struct spdk_blob_md_descriptor_flags *)desc;
5215 0 : fprintf(ctx->fp, "Flags:\n");
5216 0 : fprintf(ctx->fp, "\tinvalid: 0x%016" PRIx64 "\n", type_desc->invalid_flags);
5217 0 : bs_dump_print_type_bits(ctx, type_desc->invalid_flags, invalid,
5218 : SPDK_COUNTOF(invalid));
5219 0 : fprintf(ctx->fp, "\tdata_ro: 0x%016" PRIx64 "\n", type_desc->data_ro_flags);
5220 0 : bs_dump_print_type_bits(ctx, type_desc->data_ro_flags, data_ro,
5221 : SPDK_COUNTOF(data_ro));
5222 0 : fprintf(ctx->fp, "\t md_ro: 0x%016" PRIx64 "\n", type_desc->md_ro_flags);
5223 0 : bs_dump_print_type_bits(ctx, type_desc->md_ro_flags, md_ro,
5224 : SPDK_COUNTOF(md_ro));
5225 0 : }
5226 :
5227 : static void
5228 0 : bs_dump_print_extent_table(struct spdk_bs_load_ctx *ctx, struct spdk_blob_md_descriptor *desc)
5229 : {
5230 : struct spdk_blob_md_descriptor_extent_table *et_desc;
5231 : uint64_t num_extent_pages;
5232 : uint32_t et_idx;
5233 :
5234 0 : et_desc = (struct spdk_blob_md_descriptor_extent_table *)desc;
5235 0 : num_extent_pages = (et_desc->length - sizeof(et_desc->num_clusters)) /
5236 : sizeof(et_desc->extent_page[0]);
5237 :
5238 0 : fprintf(ctx->fp, "Extent table:\n");
5239 0 : for (et_idx = 0; et_idx < num_extent_pages; et_idx++) {
5240 0 : if (et_desc->extent_page[et_idx].page_idx == 0) {
5241 : /* Zeroes represent unallocated extent pages. */
5242 0 : continue;
5243 : }
5244 0 : fprintf(ctx->fp, "\tExtent page: %5" PRIu32 " length %3" PRIu32
5245 : " at LBA %" PRIu64 "\n", et_desc->extent_page[et_idx].page_idx,
5246 : et_desc->extent_page[et_idx].num_pages,
5247 : bs_md_page_to_lba(ctx->bs, et_desc->extent_page[et_idx].page_idx));
5248 : }
5249 0 : }
5250 :
5251 : static void
5252 0 : bs_dump_print_md_page(struct spdk_bs_load_ctx *ctx)
5253 : {
5254 0 : uint32_t page_idx = ctx->cur_page;
5255 0 : struct spdk_blob_md_page *page = ctx->page;
5256 : struct spdk_blob_md_descriptor *desc;
5257 0 : size_t cur_desc = 0;
5258 : uint32_t crc;
5259 :
5260 0 : fprintf(ctx->fp, "=========\n");
5261 0 : fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
5262 0 : fprintf(ctx->fp, "Start LBA: %" PRIu64 "\n", bs_md_page_to_lba(ctx->bs, page_idx));
5263 0 : fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
5264 0 : fprintf(ctx->fp, "Sequence: %" PRIu32 "\n", page->sequence_num);
5265 0 : if (page->next == SPDK_INVALID_MD_PAGE) {
5266 0 : fprintf(ctx->fp, "Next: None\n");
5267 : } else {
5268 0 : fprintf(ctx->fp, "Next: %" PRIu32 "\n", page->next);
5269 : }
5270 0 : fprintf(ctx->fp, "In used bit array%s:", ctx->super->clean ? "" : " (not clean: dubious)");
5271 0 : if (spdk_bit_array_get(ctx->bs->used_md_pages, page_idx)) {
5272 0 : fprintf(ctx->fp, " md");
5273 : }
5274 0 : if (spdk_bit_array_get(ctx->bs->used_blobids, page_idx)) {
5275 0 : fprintf(ctx->fp, " blob");
5276 : }
5277 0 : fprintf(ctx->fp, "\n");
5278 :
5279 0 : crc = blob_md_page_calc_crc(page);
5280 0 : fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
5281 :
5282 0 : desc = (struct spdk_blob_md_descriptor *)page->descriptors;
5283 0 : while (cur_desc < sizeof(page->descriptors)) {
5284 0 : if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
5285 0 : if (desc->length == 0) {
5286 : /* If padding and length are 0, this terminates the page */
5287 0 : break;
5288 : }
5289 0 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
5290 : struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle;
5291 : unsigned int i;
5292 :
5293 0 : desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
5294 :
5295 0 : for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
5296 0 : if (desc_extent_rle->extents[i].cluster_idx != 0) {
5297 0 : fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
5298 : desc_extent_rle->extents[i].cluster_idx);
5299 : } else {
5300 0 : fprintf(ctx->fp, "Unallocated Extent - ");
5301 : }
5302 0 : fprintf(ctx->fp, " Length: %" PRIu32, desc_extent_rle->extents[i].length);
5303 0 : fprintf(ctx->fp, "\n");
5304 : }
5305 0 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
5306 : struct spdk_blob_md_descriptor_extent_page *desc_extent;
5307 : unsigned int i;
5308 :
5309 0 : desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc;
5310 :
5311 0 : for (i = 0; i < desc_extent->length / sizeof(desc_extent->cluster_idx[0]); i++) {
5312 0 : if (desc_extent->cluster_idx[i] != 0) {
5313 0 : fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
5314 : desc_extent->cluster_idx[i]);
5315 : } else {
5316 0 : fprintf(ctx->fp, "Unallocated Extent");
5317 : }
5318 0 : fprintf(ctx->fp, "\n");
5319 : }
5320 0 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
5321 0 : bs_dump_print_xattr(ctx, desc);
5322 0 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
5323 0 : bs_dump_print_xattr(ctx, desc);
5324 0 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
5325 0 : bs_dump_print_type_flags(ctx, desc);
5326 0 : } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) {
5327 0 : bs_dump_print_extent_table(ctx, desc);
5328 : } else {
5329 : /* Error */
5330 0 : fprintf(ctx->fp, "Unknown descriptor type %" PRIu8 "\n", desc->type);
5331 : }
5332 : /* Advance to the next descriptor */
5333 0 : cur_desc += sizeof(*desc) + desc->length;
5334 0 : if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
5335 0 : break;
5336 : }
5337 0 : desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
5338 : }
5339 0 : }
5340 :
5341 : static void
5342 0 : bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5343 : {
5344 0 : struct spdk_bs_load_ctx *ctx = cb_arg;
5345 :
5346 0 : if (bserrno != 0) {
5347 0 : bs_dump_finish(seq, ctx, bserrno);
5348 0 : return;
5349 : }
5350 :
5351 0 : if (ctx->page->id != 0) {
5352 0 : bs_dump_print_md_page(ctx);
5353 : }
5354 :
5355 0 : ctx->cur_page++;
5356 :
5357 0 : if (ctx->cur_page < ctx->super->md_len) {
5358 0 : bs_dump_read_md_page(seq, ctx);
5359 : } else {
5360 0 : spdk_free(ctx->page);
5361 0 : bs_dump_finish(seq, ctx, 0);
5362 : }
5363 : }
5364 :
5365 : static void
5366 0 : bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
5367 : {
5368 0 : struct spdk_bs_load_ctx *ctx = cb_arg;
5369 : uint64_t lba;
5370 :
5371 0 : assert(ctx->cur_page < ctx->super->md_len);
5372 0 : lba = bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
5373 0 : bs_sequence_read_dev(seq, ctx->page, lba,
5374 0 : bs_byte_to_lba(ctx->bs, ctx->super->md_page_size),
5375 : bs_dump_read_md_page_cpl, ctx);
5376 0 : }
5377 :
5378 : static void
5379 0 : bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5380 : {
5381 0 : struct spdk_bs_load_ctx *ctx = cb_arg;
5382 : int rc;
5383 :
5384 0 : fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
5385 0 : if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
5386 : sizeof(ctx->super->signature)) != 0) {
5387 0 : fprintf(ctx->fp, "(Mismatch)\n");
5388 0 : bs_dump_finish(seq, ctx, bserrno);
5389 0 : return;
5390 : } else {
5391 0 : fprintf(ctx->fp, "(OK)\n");
5392 : }
5393 0 : fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
5394 0 : fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
5395 0 : (ctx->super->crc == blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
5396 0 : fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
5397 0 : fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
5398 0 : fprintf(ctx->fp, "Super Blob ID: ");
5399 0 : if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
5400 0 : fprintf(ctx->fp, "(None)\n");
5401 : } else {
5402 0 : fprintf(ctx->fp, "0x%" PRIx64 "\n", ctx->super->super_blob);
5403 : }
5404 0 : fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
5405 0 : fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
5406 0 : fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
5407 0 : fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
5408 0 : fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
5409 0 : fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
5410 0 : fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
5411 0 : fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
5412 0 : fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
5413 :
5414 0 : ctx->cur_page = 0;
5415 0 : ctx->page = spdk_zmalloc(ctx->super->md_page_size, 0,
5416 : NULL, SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
5417 0 : if (!ctx->page) {
5418 0 : bs_dump_finish(seq, ctx, -ENOMEM);
5419 0 : return;
5420 : }
5421 :
5422 0 : rc = bs_parse_super(ctx);
5423 0 : if (rc < 0) {
5424 0 : bs_load_ctx_fail(ctx, rc);
5425 0 : return;
5426 : }
5427 :
5428 0 : bs_load_read_used_pages(ctx);
5429 : }
5430 :
5431 : void
5432 0 : spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
5433 : spdk_bs_op_complete cb_fn, void *cb_arg)
5434 : {
5435 : struct spdk_blob_store *bs;
5436 : struct spdk_bs_cpl cpl;
5437 : struct spdk_bs_load_ctx *ctx;
5438 0 : struct spdk_bs_opts opts = {};
5439 : int err;
5440 :
5441 0 : SPDK_DEBUGLOG(blob, "Dumping blobstore from dev %p\n", dev);
5442 :
5443 0 : spdk_bs_opts_init(&opts, sizeof(opts));
5444 :
5445 0 : err = bs_alloc(dev, &opts, &bs, &ctx);
5446 0 : if (err) {
5447 0 : dev->destroy(dev);
5448 0 : cb_fn(cb_arg, err);
5449 0 : return;
5450 : }
5451 :
5452 0 : ctx->dumping = true;
5453 0 : ctx->fp = fp;
5454 0 : ctx->print_xattr_fn = print_xattr_fn;
5455 :
5456 0 : cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
5457 0 : cpl.u.bs_basic.cb_fn = cb_fn;
5458 0 : cpl.u.bs_basic.cb_arg = cb_arg;
5459 :
5460 0 : ctx->seq = bs_sequence_start_bs(bs->md_channel, &cpl);
5461 0 : if (!ctx->seq) {
5462 0 : spdk_free(ctx->super);
5463 0 : free(ctx);
5464 0 : bs_free(bs);
5465 0 : cb_fn(cb_arg, -ENOMEM);
5466 0 : return;
5467 : }
5468 :
5469 : /* Read the super block */
5470 0 : bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0),
5471 0 : bs_byte_to_lba(bs, sizeof(*ctx->super)),
5472 : bs_dump_super_cpl, ctx);
5473 : }
5474 :
5475 : /* END spdk_bs_dump */
5476 :
5477 : /* START spdk_bs_init */
5478 :
5479 : static void
5480 587 : bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5481 : {
5482 587 : struct spdk_bs_load_ctx *ctx = cb_arg;
5483 :
5484 587 : ctx->bs->used_clusters = spdk_bit_pool_create_from_array(ctx->used_clusters);
5485 587 : spdk_free(ctx->super);
5486 587 : free(ctx);
5487 :
5488 587 : bs_sequence_finish(seq, bserrno);
5489 587 : }
5490 :
5491 : static void
5492 587 : bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5493 : {
5494 587 : struct spdk_bs_load_ctx *ctx = cb_arg;
5495 :
5496 : /* Write super block */
5497 587 : bs_sequence_write_dev(seq, ctx->super, bs_page_to_lba(ctx->bs, 0),
5498 587 : bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
5499 : bs_init_persist_super_cpl, ctx);
5500 587 : }
5501 :
5502 : void
5503 607 : spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
5504 : spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
5505 : {
5506 : struct spdk_bs_load_ctx *ctx;
5507 : struct spdk_blob_store *bs;
5508 : struct spdk_bs_cpl cpl;
5509 : spdk_bs_sequence_t *seq;
5510 : spdk_bs_batch_t *batch;
5511 : uint64_t num_md_lba;
5512 : uint64_t num_md_pages;
5513 : uint64_t num_md_clusters;
5514 : uint64_t max_used_cluster_mask_len;
5515 : uint32_t i;
5516 607 : struct spdk_bs_opts opts = {};
5517 : int rc;
5518 : uint64_t lba, lba_count;
5519 :
5520 607 : SPDK_DEBUGLOG(blob, "Initializing blobstore on dev %p\n", dev);
5521 607 : if ((dev->phys_blocklen % dev->blocklen) != 0) {
5522 5 : SPDK_ERRLOG("unsupported dev block length of %d\n",
5523 : dev->blocklen);
5524 5 : dev->destroy(dev);
5525 5 : cb_fn(cb_arg, NULL, -EINVAL);
5526 20 : return;
5527 : }
5528 :
5529 602 : spdk_bs_opts_init(&opts, sizeof(opts));
5530 602 : if (o) {
5531 223 : if (bs_opts_copy(o, &opts)) {
5532 0 : return;
5533 : }
5534 : }
5535 :
5536 602 : if (bs_opts_verify(&opts) != 0) {
5537 10 : dev->destroy(dev);
5538 10 : cb_fn(cb_arg, NULL, -EINVAL);
5539 10 : return;
5540 : }
5541 :
5542 592 : rc = bs_alloc(dev, &opts, &bs, &ctx);
5543 592 : if (rc) {
5544 0 : dev->destroy(dev);
5545 0 : cb_fn(cb_arg, NULL, rc);
5546 0 : return;
5547 : }
5548 :
5549 592 : if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
5550 : /* By default, allocate 1 page per cluster.
5551 : * Technically, this over-allocates metadata
5552 : * because more metadata will reduce the number
5553 : * of usable clusters. This can be addressed with
5554 : * more complex math in the future.
5555 : */
5556 582 : bs->md_len = bs->total_clusters;
5557 : } else {
5558 10 : bs->md_len = opts.num_md_pages;
5559 : }
5560 592 : rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
5561 592 : if (rc < 0) {
5562 0 : spdk_free(ctx->super);
5563 0 : free(ctx);
5564 0 : bs_free(bs);
5565 0 : cb_fn(cb_arg, NULL, -ENOMEM);
5566 0 : return;
5567 : }
5568 :
5569 592 : rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
5570 592 : if (rc < 0) {
5571 0 : spdk_free(ctx->super);
5572 0 : free(ctx);
5573 0 : bs_free(bs);
5574 0 : cb_fn(cb_arg, NULL, -ENOMEM);
5575 0 : return;
5576 : }
5577 :
5578 592 : rc = spdk_bit_array_resize(&bs->open_blobids, bs->md_len);
5579 592 : if (rc < 0) {
5580 0 : spdk_free(ctx->super);
5581 0 : free(ctx);
5582 0 : bs_free(bs);
5583 0 : cb_fn(cb_arg, NULL, -ENOMEM);
5584 0 : return;
5585 : }
5586 :
5587 592 : memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
5588 : sizeof(ctx->super->signature));
5589 592 : ctx->super->version = SPDK_BS_VERSION;
5590 592 : ctx->super->length = sizeof(*ctx->super);
5591 592 : ctx->super->super_blob = bs->super_blob;
5592 592 : ctx->super->clean = 0;
5593 592 : ctx->super->cluster_size = bs->cluster_sz;
5594 592 : ctx->super->io_unit_size = bs->io_unit_size;
5595 592 : ctx->super->md_page_size = bs->md_page_size;
5596 592 : memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
5597 :
5598 : /* Calculate how many pages the metadata consumes at the front
5599 : * of the disk.
5600 : */
5601 :
5602 : /* The super block uses 1 page */
5603 592 : num_md_pages = 1;
5604 :
5605 : /* The used_md_pages mask requires 1 bit per metadata page, rounded
5606 : * up to the nearest page, plus a header.
5607 : */
5608 592 : ctx->super->used_page_mask_start = num_md_pages;
5609 592 : ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
5610 592 : spdk_divide_round_up(bs->md_len, 8),
5611 592 : ctx->super->md_page_size);
5612 592 : num_md_pages += ctx->super->used_page_mask_len;
5613 :
5614 : /* The used_clusters mask requires 1 bit per cluster, rounded
5615 : * up to the nearest page, plus a header.
5616 : */
5617 592 : ctx->super->used_cluster_mask_start = num_md_pages;
5618 592 : ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
5619 592 : spdk_divide_round_up(bs->total_clusters, 8),
5620 592 : ctx->super->md_page_size);
5621 : /* The blobstore might be extended, then the used_cluster bitmap will need more space.
5622 : * Here we calculate the max clusters we can support according to the
5623 : * num_md_pages (bs->md_len).
5624 : */
5625 592 : max_used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
5626 592 : spdk_divide_round_up(bs->md_len, 8),
5627 592 : ctx->super->md_page_size);
5628 592 : max_used_cluster_mask_len = spdk_max(max_used_cluster_mask_len,
5629 : ctx->super->used_cluster_mask_len);
5630 592 : num_md_pages += max_used_cluster_mask_len;
5631 :
5632 : /* The used_blobids mask requires 1 bit per metadata page, rounded
5633 : * up to the nearest page, plus a header.
5634 : */
5635 592 : ctx->super->used_blobid_mask_start = num_md_pages;
5636 592 : ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
5637 592 : spdk_divide_round_up(bs->md_len, 8),
5638 592 : ctx->super->md_page_size);
5639 592 : num_md_pages += ctx->super->used_blobid_mask_len;
5640 :
5641 : /* The metadata region size was chosen above */
5642 592 : ctx->super->md_start = bs->md_start = num_md_pages;
5643 592 : ctx->super->md_len = bs->md_len;
5644 592 : num_md_pages += bs->md_len;
5645 :
5646 592 : num_md_lba = bs_page_to_lba(bs, num_md_pages);
5647 :
5648 592 : ctx->super->size = dev->blockcnt * dev->blocklen;
5649 :
5650 592 : ctx->super->crc = blob_md_page_calc_crc(ctx->super);
5651 :
5652 592 : num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster);
5653 592 : if (num_md_clusters > bs->total_clusters) {
5654 5 : SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
5655 : "please decrease number of pages reserved for metadata "
5656 : "or increase cluster size.\n");
5657 5 : spdk_free(ctx->super);
5658 5 : spdk_bit_array_free(&ctx->used_clusters);
5659 5 : free(ctx);
5660 5 : bs_free(bs);
5661 5 : cb_fn(cb_arg, NULL, -ENOMEM);
5662 5 : return;
5663 : }
5664 : /* Claim all of the clusters used by the metadata */
5665 79357 : for (i = 0; i < num_md_clusters; i++) {
5666 78770 : spdk_bit_array_set(ctx->used_clusters, i);
5667 : }
5668 :
5669 587 : bs->num_free_clusters -= num_md_clusters;
5670 587 : bs->total_data_clusters = bs->num_free_clusters;
5671 :
5672 587 : cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
5673 587 : cpl.u.bs_handle.cb_fn = cb_fn;
5674 587 : cpl.u.bs_handle.cb_arg = cb_arg;
5675 587 : cpl.u.bs_handle.bs = bs;
5676 :
5677 587 : seq = bs_sequence_start_bs(bs->md_channel, &cpl);
5678 587 : if (!seq) {
5679 0 : spdk_free(ctx->super);
5680 0 : free(ctx);
5681 0 : bs_free(bs);
5682 0 : cb_fn(cb_arg, NULL, -ENOMEM);
5683 0 : return;
5684 : }
5685 :
5686 587 : batch = bs_sequence_to_batch(seq, bs_init_trim_cpl, ctx);
5687 :
5688 : /* Clear metadata space */
5689 587 : bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
5690 :
5691 587 : lba = num_md_lba;
5692 587 : lba_count = ctx->bs->dev->blockcnt - lba;
5693 587 : switch (opts.clear_method) {
5694 567 : case BS_CLEAR_WITH_UNMAP:
5695 : /* Trim data clusters */
5696 567 : bs_batch_unmap_dev(batch, lba, lba_count);
5697 567 : break;
5698 0 : case BS_CLEAR_WITH_WRITE_ZEROES:
5699 : /* Write_zeroes to data clusters */
5700 0 : bs_batch_write_zeroes_dev(batch, lba, lba_count);
5701 0 : break;
5702 20 : case BS_CLEAR_WITH_NONE:
5703 : default:
5704 20 : break;
5705 : }
5706 :
5707 587 : bs_batch_close(batch);
5708 : }
5709 :
5710 : /* END spdk_bs_init */
5711 :
5712 : /* START spdk_bs_destroy */
5713 :
5714 : static void
5715 5 : bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5716 : {
5717 5 : struct spdk_bs_load_ctx *ctx = cb_arg;
5718 5 : struct spdk_blob_store *bs = ctx->bs;
5719 :
5720 : /*
5721 : * We need to defer calling bs_call_cpl() until after
5722 : * dev destruction, so tuck these away for later use.
5723 : */
5724 5 : bs->unload_err = bserrno;
5725 5 : memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
5726 5 : seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
5727 :
5728 5 : bs_sequence_finish(seq, bserrno);
5729 :
5730 5 : bs_free(bs);
5731 5 : free(ctx);
5732 5 : }
5733 :
5734 : void
5735 5 : spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
5736 : void *cb_arg)
5737 : {
5738 : struct spdk_bs_cpl cpl;
5739 : spdk_bs_sequence_t *seq;
5740 : struct spdk_bs_load_ctx *ctx;
5741 :
5742 5 : SPDK_DEBUGLOG(blob, "Destroying blobstore\n");
5743 :
5744 5 : if (!RB_EMPTY(&bs->open_blobs)) {
5745 0 : SPDK_ERRLOG("Blobstore still has open blobs\n");
5746 0 : cb_fn(cb_arg, -EBUSY);
5747 0 : return;
5748 : }
5749 :
5750 5 : cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
5751 5 : cpl.u.bs_basic.cb_fn = cb_fn;
5752 5 : cpl.u.bs_basic.cb_arg = cb_arg;
5753 :
5754 5 : ctx = calloc(1, sizeof(*ctx));
5755 5 : if (!ctx) {
5756 0 : cb_fn(cb_arg, -ENOMEM);
5757 0 : return;
5758 : }
5759 :
5760 5 : ctx->bs = bs;
5761 :
5762 5 : seq = bs_sequence_start_bs(bs->md_channel, &cpl);
5763 5 : if (!seq) {
5764 0 : free(ctx);
5765 0 : cb_fn(cb_arg, -ENOMEM);
5766 0 : return;
5767 : }
5768 :
5769 : /* Write zeroes to the super block */
5770 5 : bs_sequence_write_zeroes_dev(seq,
5771 : bs_page_to_lba(bs, 0),
5772 : bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
5773 : bs_destroy_trim_cpl, ctx);
5774 : }
5775 :
5776 : /* END spdk_bs_destroy */
5777 :
5778 : /* START spdk_bs_unload */
5779 :
5780 : static void
5781 814 : bs_unload_finish(struct spdk_bs_load_ctx *ctx, int bserrno)
5782 : {
5783 814 : spdk_bs_sequence_t *seq = ctx->seq;
5784 :
5785 814 : spdk_free(ctx->super);
5786 :
5787 : /*
5788 : * We need to defer calling bs_call_cpl() until after
5789 : * dev destruction, so tuck these away for later use.
5790 : */
5791 814 : ctx->bs->unload_err = bserrno;
5792 814 : memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
5793 814 : seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
5794 :
5795 814 : bs_sequence_finish(seq, bserrno);
5796 :
5797 814 : bs_free(ctx->bs);
5798 814 : free(ctx);
5799 814 : }
5800 :
5801 : static void
5802 814 : bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5803 : {
5804 814 : struct spdk_bs_load_ctx *ctx = cb_arg;
5805 :
5806 814 : bs_unload_finish(ctx, bserrno);
5807 814 : }
5808 :
5809 : static void
5810 814 : bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5811 : {
5812 814 : struct spdk_bs_load_ctx *ctx = cb_arg;
5813 :
5814 814 : spdk_free(ctx->mask);
5815 :
5816 814 : if (bserrno != 0) {
5817 0 : bs_unload_finish(ctx, bserrno);
5818 0 : return;
5819 : }
5820 :
5821 814 : ctx->super->clean = 1;
5822 :
5823 814 : bs_write_super(seq, ctx->bs, ctx->super, bs_unload_write_super_cpl, ctx);
5824 : }
5825 :
5826 : static void
5827 814 : bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5828 : {
5829 814 : struct spdk_bs_load_ctx *ctx = cb_arg;
5830 :
5831 814 : spdk_free(ctx->mask);
5832 814 : ctx->mask = NULL;
5833 :
5834 814 : if (bserrno != 0) {
5835 0 : bs_unload_finish(ctx, bserrno);
5836 0 : return;
5837 : }
5838 :
5839 814 : bs_write_used_clusters(seq, ctx, bs_unload_write_used_clusters_cpl);
5840 : }
5841 :
5842 : static void
5843 814 : bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5844 : {
5845 814 : struct spdk_bs_load_ctx *ctx = cb_arg;
5846 :
5847 814 : spdk_free(ctx->mask);
5848 814 : ctx->mask = NULL;
5849 :
5850 814 : if (bserrno != 0) {
5851 0 : bs_unload_finish(ctx, bserrno);
5852 0 : return;
5853 : }
5854 :
5855 814 : bs_write_used_blobids(seq, ctx, bs_unload_write_used_blobids_cpl);
5856 : }
5857 :
5858 : static void
5859 814 : bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5860 : {
5861 814 : struct spdk_bs_load_ctx *ctx = cb_arg;
5862 : int rc;
5863 :
5864 814 : if (bserrno != 0) {
5865 0 : bs_unload_finish(ctx, bserrno);
5866 0 : return;
5867 : }
5868 :
5869 814 : rc = bs_super_validate(ctx->super, ctx->bs);
5870 814 : if (rc != 0) {
5871 0 : bs_unload_finish(ctx, rc);
5872 0 : return;
5873 : }
5874 :
5875 814 : bs_write_used_md(seq, cb_arg, bs_unload_write_used_pages_cpl);
5876 : }
5877 :
5878 : void
5879 824 : spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
5880 : {
5881 : struct spdk_bs_cpl cpl;
5882 : struct spdk_bs_load_ctx *ctx;
5883 :
5884 824 : SPDK_DEBUGLOG(blob, "Syncing blobstore\n");
5885 :
5886 : /*
5887 : * If external snapshot channels are being destroyed while the blobstore is unloaded, the
5888 : * unload is deferred until after the channel destruction completes.
5889 : */
5890 824 : if (bs->esnap_channels_unloading != 0) {
5891 5 : if (bs->esnap_unload_cb_fn != NULL) {
5892 0 : SPDK_ERRLOG("Blobstore unload in progress\n");
5893 0 : cb_fn(cb_arg, -EBUSY);
5894 10 : return;
5895 : }
5896 5 : SPDK_DEBUGLOG(blob_esnap, "Blobstore unload deferred: %" PRIu32
5897 : " esnap clones are unloading\n", bs->esnap_channels_unloading);
5898 5 : bs->esnap_unload_cb_fn = cb_fn;
5899 5 : bs->esnap_unload_cb_arg = cb_arg;
5900 5 : return;
5901 : }
5902 819 : if (bs->esnap_unload_cb_fn != NULL) {
5903 5 : SPDK_DEBUGLOG(blob_esnap, "Blobstore deferred unload progressing\n");
5904 5 : assert(bs->esnap_unload_cb_fn == cb_fn);
5905 5 : assert(bs->esnap_unload_cb_arg == cb_arg);
5906 5 : bs->esnap_unload_cb_fn = NULL;
5907 5 : bs->esnap_unload_cb_arg = NULL;
5908 : }
5909 :
5910 819 : if (!RB_EMPTY(&bs->open_blobs)) {
5911 5 : SPDK_ERRLOG("Blobstore still has open blobs\n");
5912 5 : cb_fn(cb_arg, -EBUSY);
5913 5 : return;
5914 : }
5915 :
5916 814 : ctx = calloc(1, sizeof(*ctx));
5917 814 : if (!ctx) {
5918 0 : cb_fn(cb_arg, -ENOMEM);
5919 0 : return;
5920 : }
5921 :
5922 814 : ctx->bs = bs;
5923 :
5924 814 : ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
5925 : SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
5926 814 : if (!ctx->super) {
5927 0 : free(ctx);
5928 0 : cb_fn(cb_arg, -ENOMEM);
5929 0 : return;
5930 : }
5931 :
5932 814 : cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
5933 814 : cpl.u.bs_basic.cb_fn = cb_fn;
5934 814 : cpl.u.bs_basic.cb_arg = cb_arg;
5935 :
5936 814 : ctx->seq = bs_sequence_start_bs(bs->md_channel, &cpl);
5937 814 : if (!ctx->seq) {
5938 0 : spdk_free(ctx->super);
5939 0 : free(ctx);
5940 0 : cb_fn(cb_arg, -ENOMEM);
5941 0 : return;
5942 : }
5943 :
5944 : /* Read super block */
5945 814 : bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0),
5946 814 : bs_byte_to_lba(bs, sizeof(*ctx->super)),
5947 : bs_unload_read_super_cpl, ctx);
5948 : }
5949 :
5950 : /* END spdk_bs_unload */
5951 :
5952 : /* START spdk_bs_set_super */
5953 :
5954 : struct spdk_bs_set_super_ctx {
5955 : struct spdk_blob_store *bs;
5956 : struct spdk_bs_super_block *super;
5957 : };
5958 :
5959 : static void
5960 10 : bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5961 : {
5962 10 : struct spdk_bs_set_super_ctx *ctx = cb_arg;
5963 :
5964 10 : if (bserrno != 0) {
5965 0 : SPDK_ERRLOG("Unable to write to super block of blobstore\n");
5966 : }
5967 :
5968 10 : spdk_free(ctx->super);
5969 :
5970 10 : bs_sequence_finish(seq, bserrno);
5971 :
5972 10 : free(ctx);
5973 10 : }
5974 :
5975 : static void
5976 10 : bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5977 : {
5978 10 : struct spdk_bs_set_super_ctx *ctx = cb_arg;
5979 : int rc;
5980 :
5981 10 : if (bserrno != 0) {
5982 0 : SPDK_ERRLOG("Unable to read super block of blobstore\n");
5983 0 : spdk_free(ctx->super);
5984 0 : bs_sequence_finish(seq, bserrno);
5985 0 : free(ctx);
5986 0 : return;
5987 : }
5988 :
5989 10 : rc = bs_super_validate(ctx->super, ctx->bs);
5990 10 : if (rc != 0) {
5991 0 : SPDK_ERRLOG("Not a valid super block\n");
5992 0 : spdk_free(ctx->super);
5993 0 : bs_sequence_finish(seq, rc);
5994 0 : free(ctx);
5995 0 : return;
5996 : }
5997 :
5998 10 : bs_write_super(seq, ctx->bs, ctx->super, bs_set_super_write_cpl, ctx);
5999 : }
6000 :
6001 : void
6002 10 : spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
6003 : spdk_bs_op_complete cb_fn, void *cb_arg)
6004 : {
6005 : struct spdk_bs_cpl cpl;
6006 : spdk_bs_sequence_t *seq;
6007 : struct spdk_bs_set_super_ctx *ctx;
6008 :
6009 10 : SPDK_DEBUGLOG(blob, "Setting super blob id on blobstore\n");
6010 :
6011 10 : ctx = calloc(1, sizeof(*ctx));
6012 10 : if (!ctx) {
6013 0 : cb_fn(cb_arg, -ENOMEM);
6014 0 : return;
6015 : }
6016 :
6017 10 : ctx->bs = bs;
6018 :
6019 10 : ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
6020 : SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
6021 10 : if (!ctx->super) {
6022 0 : free(ctx);
6023 0 : cb_fn(cb_arg, -ENOMEM);
6024 0 : return;
6025 : }
6026 :
6027 10 : cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
6028 10 : cpl.u.bs_basic.cb_fn = cb_fn;
6029 10 : cpl.u.bs_basic.cb_arg = cb_arg;
6030 :
6031 10 : seq = bs_sequence_start_bs(bs->md_channel, &cpl);
6032 10 : if (!seq) {
6033 0 : spdk_free(ctx->super);
6034 0 : free(ctx);
6035 0 : cb_fn(cb_arg, -ENOMEM);
6036 0 : return;
6037 : }
6038 :
6039 10 : bs->super_blob = blobid;
6040 :
6041 : /* Read super block */
6042 10 : bs_sequence_read_dev(seq, ctx->super, bs_page_to_lba(bs, 0),
6043 10 : bs_byte_to_lba(bs, sizeof(*ctx->super)),
6044 : bs_set_super_read_cpl, ctx);
6045 : }
6046 :
6047 : /* END spdk_bs_set_super */
6048 :
6049 : void
6050 15 : spdk_bs_get_super(struct spdk_blob_store *bs,
6051 : spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
6052 : {
6053 15 : if (bs->super_blob == SPDK_BLOBID_INVALID) {
6054 5 : cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
6055 : } else {
6056 10 : cb_fn(cb_arg, bs->super_blob, 0);
6057 : }
6058 15 : }
6059 :
6060 : uint64_t
6061 254 : spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
6062 : {
6063 254 : return bs->cluster_sz;
6064 : }
6065 :
6066 : uint64_t
6067 114 : spdk_bs_get_page_size(struct spdk_blob_store *bs)
6068 : {
6069 114 : return bs->md_page_size;
6070 : }
6071 :
6072 : uint64_t
6073 1006 : spdk_bs_get_io_unit_size(struct spdk_blob_store *bs)
6074 : {
6075 1006 : return bs->io_unit_size;
6076 : }
6077 :
6078 : uint64_t
6079 700 : spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
6080 : {
6081 700 : return bs->num_free_clusters;
6082 : }
6083 :
6084 : uint64_t
6085 194 : spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
6086 : {
6087 194 : return bs->total_data_clusters;
6088 : }
6089 :
6090 : static int
6091 973 : bs_register_md_thread(struct spdk_blob_store *bs)
6092 : {
6093 973 : bs->md_channel = spdk_get_io_channel(bs);
6094 973 : if (!bs->md_channel) {
6095 0 : SPDK_ERRLOG("Failed to get IO channel.\n");
6096 0 : return -1;
6097 : }
6098 :
6099 973 : return 0;
6100 : }
6101 :
6102 : static int
6103 973 : bs_unregister_md_thread(struct spdk_blob_store *bs)
6104 : {
6105 973 : spdk_put_io_channel(bs->md_channel);
6106 :
6107 973 : return 0;
6108 : }
6109 :
6110 : spdk_blob_id
6111 712 : spdk_blob_get_id(struct spdk_blob *blob)
6112 : {
6113 712 : assert(blob != NULL);
6114 :
6115 712 : return blob->id;
6116 : }
6117 :
6118 : uint64_t
6119 30 : spdk_blob_get_num_io_units(struct spdk_blob *blob)
6120 : {
6121 30 : assert(blob != NULL);
6122 :
6123 30 : return bs_cluster_to_io_unit(blob->bs, blob->active.num_clusters);
6124 : }
6125 :
6126 : uint64_t
6127 707 : spdk_blob_get_num_clusters(struct spdk_blob *blob)
6128 : {
6129 707 : assert(blob != NULL);
6130 :
6131 707 : return blob->active.num_clusters;
6132 : }
6133 :
6134 : uint64_t
6135 415 : spdk_blob_get_num_allocated_clusters(struct spdk_blob *blob)
6136 : {
6137 415 : assert(blob != NULL);
6138 :
6139 415 : return blob->active.num_allocated_clusters;
6140 : }
6141 :
6142 : static uint64_t
6143 30 : blob_find_io_unit(struct spdk_blob *blob, uint64_t offset, bool is_allocated)
6144 : {
6145 30 : uint64_t blob_io_unit_num = spdk_blob_get_num_io_units(blob);
6146 :
6147 55 : while (offset < blob_io_unit_num) {
6148 50 : if (bs_io_unit_is_allocated(blob, offset) == is_allocated) {
6149 25 : return offset;
6150 : }
6151 :
6152 25 : offset += bs_num_io_units_to_cluster_boundary(blob, offset);
6153 : }
6154 :
6155 5 : return UINT64_MAX;
6156 : }
6157 :
6158 : uint64_t
6159 15 : spdk_blob_get_next_allocated_io_unit(struct spdk_blob *blob, uint64_t offset)
6160 : {
6161 15 : return blob_find_io_unit(blob, offset, true);
6162 : }
6163 :
6164 : uint64_t
6165 15 : spdk_blob_get_next_unallocated_io_unit(struct spdk_blob *blob, uint64_t offset)
6166 : {
6167 15 : return blob_find_io_unit(blob, offset, false);
6168 : }
6169 :
6170 : /* START spdk_bs_create_blob */
6171 :
6172 : static void
6173 2346 : bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
6174 : {
6175 2346 : struct spdk_blob *blob = cb_arg;
6176 2346 : uint32_t page_idx = bs_blobid_to_page(blob->id);
6177 :
6178 2346 : if (bserrno != 0) {
6179 0 : spdk_spin_lock(&blob->bs->used_lock);
6180 0 : spdk_bit_array_clear(blob->bs->used_blobids, page_idx);
6181 0 : bs_release_md_page(blob->bs, page_idx);
6182 0 : spdk_spin_unlock(&blob->bs->used_lock);
6183 : }
6184 :
6185 2346 : blob_free(blob);
6186 :
6187 2346 : bs_sequence_finish(seq, bserrno);
6188 2346 : }
6189 :
6190 : static int
6191 4717 : blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
6192 : bool internal)
6193 : {
6194 : uint64_t i;
6195 4717 : size_t value_len = 0;
6196 : int rc;
6197 4717 : const void *value = NULL;
6198 4717 : if (xattrs->count > 0 && xattrs->get_value == NULL) {
6199 10 : return -EINVAL;
6200 : }
6201 5103 : for (i = 0; i < xattrs->count; i++) {
6202 401 : xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
6203 401 : if (value == NULL || value_len == 0) {
6204 5 : return -EINVAL;
6205 : }
6206 396 : rc = blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
6207 396 : if (rc < 0) {
6208 0 : return rc;
6209 : }
6210 : }
6211 4702 : return 0;
6212 : }
6213 :
6214 : static void
6215 2330 : blob_opts_copy(const struct spdk_blob_opts *src, struct spdk_blob_opts *dst)
6216 : {
6217 : #define FIELD_OK(field) \
6218 : offsetof(struct spdk_blob_opts, field) + sizeof(src->field) <= src->opts_size
6219 :
6220 : #define SET_FIELD(field) \
6221 : if (FIELD_OK(field)) { \
6222 : dst->field = src->field; \
6223 : } \
6224 :
6225 2330 : SET_FIELD(num_clusters);
6226 2330 : SET_FIELD(thin_provision);
6227 2330 : SET_FIELD(clear_method);
6228 :
6229 2330 : if (FIELD_OK(xattrs)) {
6230 2330 : memcpy(&dst->xattrs, &src->xattrs, sizeof(src->xattrs));
6231 : }
6232 :
6233 2330 : SET_FIELD(use_extent_table);
6234 2330 : SET_FIELD(esnap_id);
6235 2330 : SET_FIELD(esnap_id_len);
6236 :
6237 2330 : dst->opts_size = src->opts_size;
6238 :
6239 : /* You should not remove this statement, but need to update the assert statement
6240 : * if you add a new field, and also add a corresponding SET_FIELD statement */
6241 : SPDK_STATIC_ASSERT(sizeof(struct spdk_blob_opts) == 80, "Incorrect size");
6242 :
6243 : #undef FIELD_OK
6244 : #undef SET_FIELD
6245 2330 : }
6246 :
6247 : static void
6248 2366 : bs_create_blob(struct spdk_blob_store *bs,
6249 : const struct spdk_blob_opts *opts,
6250 : const struct spdk_blob_xattr_opts *internal_xattrs,
6251 : spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
6252 : {
6253 : struct spdk_blob *blob;
6254 : uint32_t page_idx;
6255 : struct spdk_bs_cpl cpl;
6256 : struct spdk_blob_opts opts_local;
6257 : struct spdk_blob_xattr_opts internal_xattrs_default;
6258 : spdk_bs_sequence_t *seq;
6259 : spdk_blob_id id;
6260 : int rc;
6261 :
6262 2366 : assert(spdk_get_thread() == bs->md_thread);
6263 :
6264 2366 : spdk_spin_lock(&bs->used_lock);
6265 2366 : page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
6266 2366 : if (page_idx == UINT32_MAX) {
6267 0 : spdk_spin_unlock(&bs->used_lock);
6268 0 : cb_fn(cb_arg, 0, -ENOMEM);
6269 2346 : return;
6270 : }
6271 2366 : spdk_bit_array_set(bs->used_blobids, page_idx);
6272 2366 : bs_claim_md_page(bs, page_idx);
6273 2366 : spdk_spin_unlock(&bs->used_lock);
6274 :
6275 2366 : id = bs_page_to_blobid(page_idx);
6276 :
6277 2366 : SPDK_DEBUGLOG(blob, "Creating blob with id 0x%" PRIx64 " at page %u\n", id, page_idx);
6278 :
6279 2366 : spdk_blob_opts_init(&opts_local, sizeof(opts_local));
6280 2366 : if (opts) {
6281 2330 : blob_opts_copy(opts, &opts_local);
6282 : }
6283 :
6284 2366 : blob = blob_alloc(bs, id);
6285 2366 : if (!blob) {
6286 0 : rc = -ENOMEM;
6287 0 : goto error;
6288 : }
6289 :
6290 2366 : blob->use_extent_table = opts_local.use_extent_table;
6291 2366 : if (blob->use_extent_table) {
6292 1440 : blob->invalid_flags |= SPDK_BLOB_EXTENT_TABLE;
6293 : }
6294 :
6295 2366 : if (!internal_xattrs) {
6296 2025 : blob_xattrs_init(&internal_xattrs_default);
6297 2025 : internal_xattrs = &internal_xattrs_default;
6298 : }
6299 :
6300 2366 : rc = blob_set_xattrs(blob, &opts_local.xattrs, false);
6301 2366 : if (rc < 0) {
6302 15 : goto error;
6303 : }
6304 :
6305 2351 : rc = blob_set_xattrs(blob, internal_xattrs, true);
6306 2351 : if (rc < 0) {
6307 0 : goto error;
6308 : }
6309 :
6310 2351 : if (opts_local.thin_provision) {
6311 446 : blob_set_thin_provision(blob);
6312 : }
6313 :
6314 2351 : blob_set_clear_method(blob, opts_local.clear_method);
6315 :
6316 2351 : if (opts_local.esnap_id != NULL) {
6317 75 : if (opts_local.esnap_id_len > UINT16_MAX) {
6318 0 : SPDK_ERRLOG("esnap id length %" PRIu64 "is too long\n",
6319 : opts_local.esnap_id_len);
6320 0 : rc = -EINVAL;
6321 0 : goto error;
6322 :
6323 : }
6324 75 : blob_set_thin_provision(blob);
6325 75 : blob->invalid_flags |= SPDK_BLOB_EXTERNAL_SNAPSHOT;
6326 75 : rc = blob_set_xattr(blob, BLOB_EXTERNAL_SNAPSHOT_ID,
6327 75 : opts_local.esnap_id, opts_local.esnap_id_len, true);
6328 75 : if (rc != 0) {
6329 0 : goto error;
6330 : }
6331 : }
6332 :
6333 2351 : rc = blob_resize(blob, opts_local.num_clusters);
6334 2351 : if (rc < 0) {
6335 5 : goto error;
6336 : }
6337 2346 : cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
6338 2346 : cpl.u.blobid.cb_fn = cb_fn;
6339 2346 : cpl.u.blobid.cb_arg = cb_arg;
6340 2346 : cpl.u.blobid.blobid = blob->id;
6341 :
6342 2346 : seq = bs_sequence_start_bs(bs->md_channel, &cpl);
6343 2346 : if (!seq) {
6344 0 : rc = -ENOMEM;
6345 0 : goto error;
6346 : }
6347 :
6348 2346 : blob_persist(seq, blob, bs_create_blob_cpl, blob);
6349 2346 : return;
6350 :
6351 20 : error:
6352 20 : SPDK_ERRLOG("Failed to create blob: %s, size in clusters/size: %lu (clusters)\n",
6353 : spdk_strerror(rc), opts_local.num_clusters);
6354 20 : if (blob != NULL) {
6355 20 : blob_free(blob);
6356 : }
6357 20 : spdk_spin_lock(&bs->used_lock);
6358 20 : spdk_bit_array_clear(bs->used_blobids, page_idx);
6359 20 : bs_release_md_page(bs, page_idx);
6360 20 : spdk_spin_unlock(&bs->used_lock);
6361 20 : cb_fn(cb_arg, 0, rc);
6362 : }
6363 :
6364 : void
6365 16 : spdk_bs_create_blob(struct spdk_blob_store *bs,
6366 : spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
6367 : {
6368 16 : bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
6369 16 : }
6370 :
6371 : void
6372 1999 : spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
6373 : spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
6374 : {
6375 1999 : bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
6376 1999 : }
6377 :
6378 : /* END spdk_bs_create_blob */
6379 :
6380 : /* START blob_cleanup */
6381 :
6382 : struct spdk_clone_snapshot_ctx {
6383 : struct spdk_bs_cpl cpl;
6384 : int bserrno;
6385 : bool frozen;
6386 :
6387 : struct spdk_io_channel *channel;
6388 :
6389 : /* Current cluster for inflate operation */
6390 : uint64_t cluster;
6391 :
6392 : /* For inflation force allocation of all unallocated clusters and remove
6393 : * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
6394 : bool allocate_all;
6395 :
6396 : struct {
6397 : spdk_blob_id id;
6398 : struct spdk_blob *blob;
6399 : bool md_ro;
6400 : } original;
6401 : struct {
6402 : spdk_blob_id id;
6403 : struct spdk_blob *blob;
6404 : } new;
6405 :
6406 : /* xattrs specified for snapshot/clones only. They have no impact on
6407 : * the original blobs xattrs. */
6408 : const struct spdk_blob_xattr_opts *xattrs;
6409 : };
6410 :
6411 : static void
6412 429 : bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
6413 : {
6414 429 : struct spdk_clone_snapshot_ctx *ctx = cb_arg;
6415 429 : struct spdk_bs_cpl *cpl = &ctx->cpl;
6416 :
6417 429 : if (bserrno != 0) {
6418 8 : if (ctx->bserrno != 0) {
6419 0 : SPDK_ERRLOG("Cleanup error %d\n", bserrno);
6420 : } else {
6421 8 : ctx->bserrno = bserrno;
6422 : }
6423 : }
6424 :
6425 429 : switch (cpl->type) {
6426 354 : case SPDK_BS_CPL_TYPE_BLOBID:
6427 354 : cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
6428 354 : break;
6429 75 : case SPDK_BS_CPL_TYPE_BLOB_BASIC:
6430 75 : cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
6431 75 : break;
6432 0 : default:
6433 0 : SPDK_UNREACHABLE();
6434 : break;
6435 : }
6436 :
6437 429 : free(ctx);
6438 429 : }
6439 :
6440 : static void
6441 411 : bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
6442 : {
6443 411 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6444 411 : struct spdk_blob *origblob = ctx->original.blob;
6445 :
6446 411 : if (bserrno != 0) {
6447 0 : if (ctx->bserrno != 0) {
6448 0 : SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
6449 : } else {
6450 0 : ctx->bserrno = bserrno;
6451 : }
6452 : }
6453 :
6454 411 : ctx->original.id = origblob->id;
6455 411 : origblob->locked_operation_in_progress = false;
6456 :
6457 : /* Revert md_ro to original state */
6458 411 : origblob->md_ro = ctx->original.md_ro;
6459 :
6460 411 : spdk_blob_close(origblob, bs_clone_snapshot_cleanup_finish, ctx);
6461 411 : }
6462 :
6463 : static void
6464 411 : bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
6465 : {
6466 411 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6467 411 : struct spdk_blob *origblob = ctx->original.blob;
6468 :
6469 411 : if (bserrno != 0) {
6470 30 : if (ctx->bserrno != 0) {
6471 5 : SPDK_ERRLOG("Cleanup error %d\n", bserrno);
6472 : } else {
6473 25 : ctx->bserrno = bserrno;
6474 : }
6475 : }
6476 :
6477 411 : if (ctx->frozen) {
6478 : /* Unfreeze any outstanding I/O */
6479 266 : blob_unfreeze_io(origblob, bs_snapshot_unfreeze_cpl, ctx);
6480 : } else {
6481 145 : bs_snapshot_unfreeze_cpl(ctx, 0);
6482 : }
6483 :
6484 411 : }
6485 :
6486 : static void
6487 5 : bs_clone_snapshot_newblob_cleanup(struct spdk_clone_snapshot_ctx *ctx, int bserrno)
6488 : {
6489 5 : struct spdk_blob *newblob = ctx->new.blob;
6490 :
6491 5 : if (bserrno != 0) {
6492 5 : if (ctx->bserrno != 0) {
6493 0 : SPDK_ERRLOG("Cleanup error %d\n", bserrno);
6494 : } else {
6495 5 : ctx->bserrno = bserrno;
6496 : }
6497 : }
6498 :
6499 5 : ctx->new.id = newblob->id;
6500 5 : spdk_blob_close(newblob, bs_clone_snapshot_origblob_cleanup, ctx);
6501 5 : }
6502 :
6503 : /* END blob_cleanup */
6504 :
6505 : /* START spdk_bs_create_snapshot */
6506 :
6507 : static void
6508 276 : bs_snapshot_swap_cluster_maps(struct spdk_blob *blob1, struct spdk_blob *blob2)
6509 : {
6510 : uint64_t *cluster_temp;
6511 : uint64_t num_allocated_clusters_temp;
6512 : uint32_t *extent_page_temp;
6513 :
6514 276 : cluster_temp = blob1->active.clusters;
6515 276 : blob1->active.clusters = blob2->active.clusters;
6516 276 : blob2->active.clusters = cluster_temp;
6517 :
6518 276 : num_allocated_clusters_temp = blob1->active.num_allocated_clusters;
6519 276 : blob1->active.num_allocated_clusters = blob2->active.num_allocated_clusters;
6520 276 : blob2->active.num_allocated_clusters = num_allocated_clusters_temp;
6521 :
6522 276 : extent_page_temp = blob1->active.extent_pages;
6523 276 : blob1->active.extent_pages = blob2->active.extent_pages;
6524 276 : blob2->active.extent_pages = extent_page_temp;
6525 276 : }
6526 :
6527 : /* Copies an internal xattr */
6528 : static int
6529 25 : bs_snapshot_copy_xattr(struct spdk_blob *toblob, struct spdk_blob *fromblob, const char *name)
6530 : {
6531 25 : const void *val = NULL;
6532 : size_t len;
6533 : int bserrno;
6534 :
6535 25 : bserrno = blob_get_xattr_value(fromblob, name, &val, &len, true);
6536 25 : if (bserrno != 0) {
6537 0 : SPDK_ERRLOG("blob 0x%" PRIx64 " missing %s XATTR\n", fromblob->id, name);
6538 0 : return bserrno;
6539 : }
6540 :
6541 25 : bserrno = blob_set_xattr(toblob, name, val, len, true);
6542 25 : if (bserrno != 0) {
6543 0 : SPDK_ERRLOG("could not set %s XATTR on blob 0x%" PRIx64 "\n",
6544 : name, toblob->id);
6545 0 : return bserrno;
6546 : }
6547 25 : return 0;
6548 : }
6549 :
6550 : static void
6551 261 : bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
6552 : {
6553 261 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6554 261 : struct spdk_blob *origblob = ctx->original.blob;
6555 261 : struct spdk_blob *newblob = ctx->new.blob;
6556 :
6557 261 : if (bserrno != 0) {
6558 5 : bs_snapshot_swap_cluster_maps(newblob, origblob);
6559 5 : if (blob_is_esnap_clone(newblob)) {
6560 0 : bs_snapshot_copy_xattr(origblob, newblob, BLOB_EXTERNAL_SNAPSHOT_ID);
6561 0 : origblob->invalid_flags |= SPDK_BLOB_EXTERNAL_SNAPSHOT;
6562 : }
6563 5 : bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6564 5 : return;
6565 : }
6566 :
6567 : /* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
6568 256 : bserrno = blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
6569 256 : if (bserrno != 0) {
6570 0 : bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6571 0 : return;
6572 : }
6573 :
6574 256 : bs_blob_list_add(ctx->original.blob);
6575 :
6576 256 : spdk_blob_set_read_only(newblob);
6577 :
6578 : /* sync snapshot metadata */
6579 256 : spdk_blob_sync_md(newblob, bs_clone_snapshot_origblob_cleanup, ctx);
6580 : }
6581 :
6582 : static void
6583 266 : bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
6584 : {
6585 266 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6586 266 : struct spdk_blob *origblob = ctx->original.blob;
6587 266 : struct spdk_blob *newblob = ctx->new.blob;
6588 :
6589 266 : if (bserrno != 0) {
6590 : /* return cluster map back to original */
6591 5 : bs_snapshot_swap_cluster_maps(newblob, origblob);
6592 :
6593 : /* Newblob md sync failed. Valid clusters are only present in origblob.
6594 : * Since I/O is frozen on origblob, not changes to zeroed out cluster map should have occurred.
6595 : * Newblob needs to be reverted to thin_provisioned state at creation to properly close. */
6596 5 : blob_set_thin_provision(newblob);
6597 5 : assert(spdk_mem_all_zero(newblob->active.clusters,
6598 : newblob->active.num_clusters * sizeof(*newblob->active.clusters)));
6599 5 : assert(spdk_mem_all_zero(newblob->active.extent_pages,
6600 : newblob->active.num_extent_pages * sizeof(*newblob->active.extent_pages)));
6601 :
6602 5 : bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
6603 5 : return;
6604 : }
6605 :
6606 : /* Set internal xattr for snapshot id */
6607 261 : bserrno = blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
6608 261 : if (bserrno != 0) {
6609 : /* return cluster map back to original */
6610 0 : bs_snapshot_swap_cluster_maps(newblob, origblob);
6611 0 : blob_set_thin_provision(newblob);
6612 0 : bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
6613 0 : return;
6614 : }
6615 :
6616 : /* Create new back_bs_dev for snapshot */
6617 261 : origblob->back_bs_dev = bs_create_blob_bs_dev(newblob);
6618 261 : if (origblob->back_bs_dev == NULL) {
6619 : /* return cluster map back to original */
6620 0 : bs_snapshot_swap_cluster_maps(newblob, origblob);
6621 0 : blob_set_thin_provision(newblob);
6622 0 : bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
6623 0 : return;
6624 : }
6625 :
6626 : /* Remove the xattr that references an external snapshot */
6627 261 : if (blob_is_esnap_clone(origblob)) {
6628 15 : origblob->invalid_flags &= ~SPDK_BLOB_EXTERNAL_SNAPSHOT;
6629 15 : bserrno = blob_remove_xattr(origblob, BLOB_EXTERNAL_SNAPSHOT_ID, true);
6630 15 : if (bserrno != 0) {
6631 0 : if (bserrno == -ENOENT) {
6632 0 : SPDK_ERRLOG("blob 0x%" PRIx64 " has no " BLOB_EXTERNAL_SNAPSHOT_ID
6633 : " xattr to remove\n", origblob->id);
6634 0 : assert(false);
6635 : } else {
6636 : /* return cluster map back to original */
6637 0 : bs_snapshot_swap_cluster_maps(newblob, origblob);
6638 0 : blob_set_thin_provision(newblob);
6639 0 : bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
6640 0 : return;
6641 : }
6642 : }
6643 : }
6644 :
6645 261 : bs_blob_list_remove(origblob);
6646 261 : origblob->parent_id = newblob->id;
6647 : /* set clone blob as thin provisioned */
6648 261 : blob_set_thin_provision(origblob);
6649 :
6650 261 : bs_blob_list_add(newblob);
6651 :
6652 : /* sync clone metadata */
6653 261 : spdk_blob_sync_md(origblob, bs_snapshot_origblob_sync_cpl, ctx);
6654 : }
6655 :
6656 : static void
6657 266 : bs_snapshot_freeze_cpl(void *cb_arg, int rc)
6658 : {
6659 266 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6660 266 : struct spdk_blob *origblob = ctx->original.blob;
6661 266 : struct spdk_blob *newblob = ctx->new.blob;
6662 : int bserrno;
6663 :
6664 266 : if (rc != 0) {
6665 0 : bs_clone_snapshot_newblob_cleanup(ctx, rc);
6666 0 : return;
6667 : }
6668 :
6669 266 : ctx->frozen = true;
6670 :
6671 266 : if (blob_is_esnap_clone(origblob)) {
6672 : /* Clean up any channels associated with the original blob id because future IO will
6673 : * perform IO using the snapshot blob_id.
6674 : */
6675 15 : blob_esnap_destroy_bs_dev_channels(origblob, false, NULL, NULL);
6676 : }
6677 266 : if (newblob->back_bs_dev) {
6678 266 : blob_back_bs_destroy(newblob);
6679 : }
6680 : /* set new back_bs_dev for snapshot */
6681 266 : newblob->back_bs_dev = origblob->back_bs_dev;
6682 : /* Set invalid flags from origblob */
6683 266 : newblob->invalid_flags = origblob->invalid_flags;
6684 :
6685 : /* inherit parent from original blob if set */
6686 266 : newblob->parent_id = origblob->parent_id;
6687 266 : switch (origblob->parent_id) {
6688 15 : case SPDK_BLOBID_EXTERNAL_SNAPSHOT:
6689 15 : bserrno = bs_snapshot_copy_xattr(newblob, origblob, BLOB_EXTERNAL_SNAPSHOT_ID);
6690 15 : if (bserrno != 0) {
6691 0 : bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
6692 0 : return;
6693 : }
6694 15 : break;
6695 186 : case SPDK_BLOBID_INVALID:
6696 186 : break;
6697 65 : default:
6698 : /* Set internal xattr for snapshot id */
6699 65 : bserrno = blob_set_xattr(newblob, BLOB_SNAPSHOT,
6700 65 : &origblob->parent_id, sizeof(spdk_blob_id), true);
6701 65 : if (bserrno != 0) {
6702 0 : bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
6703 0 : return;
6704 : }
6705 : }
6706 :
6707 : /* swap cluster maps */
6708 266 : bs_snapshot_swap_cluster_maps(newblob, origblob);
6709 :
6710 : /* Set the clear method on the new blob to match the original. */
6711 266 : blob_set_clear_method(newblob, origblob->clear_method);
6712 :
6713 : /* sync snapshot metadata */
6714 266 : spdk_blob_sync_md(newblob, bs_snapshot_newblob_sync_cpl, ctx);
6715 : }
6716 :
6717 : static void
6718 271 : bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6719 : {
6720 271 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6721 271 : struct spdk_blob *origblob = ctx->original.blob;
6722 271 : struct spdk_blob *newblob = _blob;
6723 :
6724 271 : if (bserrno != 0) {
6725 5 : bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6726 5 : return;
6727 : }
6728 :
6729 266 : ctx->new.blob = newblob;
6730 266 : assert(spdk_blob_is_thin_provisioned(newblob));
6731 266 : assert(spdk_mem_all_zero(newblob->active.clusters,
6732 : newblob->active.num_clusters * sizeof(*newblob->active.clusters)));
6733 266 : assert(spdk_mem_all_zero(newblob->active.extent_pages,
6734 : newblob->active.num_extent_pages * sizeof(*newblob->active.extent_pages)));
6735 :
6736 266 : blob_freeze_io(origblob, bs_snapshot_freeze_cpl, ctx);
6737 : }
6738 :
6739 : static void
6740 276 : bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
6741 : {
6742 276 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6743 276 : struct spdk_blob *origblob = ctx->original.blob;
6744 :
6745 276 : if (bserrno != 0) {
6746 5 : bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6747 5 : return;
6748 : }
6749 :
6750 271 : ctx->new.id = blobid;
6751 271 : ctx->cpl.u.blobid.blobid = blobid;
6752 :
6753 271 : spdk_bs_open_blob(origblob->bs, ctx->new.id, bs_snapshot_newblob_open_cpl, ctx);
6754 : }
6755 :
6756 :
6757 : static void
6758 276 : bs_xattr_snapshot(void *arg, const char *name,
6759 : const void **value, size_t *value_len)
6760 : {
6761 276 : assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
6762 :
6763 276 : struct spdk_blob *blob = (struct spdk_blob *)arg;
6764 276 : *value = &blob->id;
6765 276 : *value_len = sizeof(blob->id);
6766 276 : }
6767 :
6768 : static void
6769 289 : bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6770 : {
6771 289 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6772 : struct spdk_blob_opts opts;
6773 : struct spdk_blob_xattr_opts internal_xattrs;
6774 289 : char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
6775 :
6776 289 : if (bserrno != 0) {
6777 8 : bs_clone_snapshot_cleanup_finish(ctx, bserrno);
6778 13 : return;
6779 : }
6780 :
6781 281 : ctx->original.blob = _blob;
6782 :
6783 281 : if (_blob->data_ro || _blob->md_ro) {
6784 5 : SPDK_DEBUGLOG(blob, "Cannot create snapshot from read only blob with id 0x%"
6785 : PRIx64 "\n", _blob->id);
6786 5 : ctx->bserrno = -EINVAL;
6787 5 : spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6788 5 : return;
6789 : }
6790 :
6791 276 : if (_blob->locked_operation_in_progress) {
6792 0 : SPDK_DEBUGLOG(blob, "Cannot create snapshot - another operation in progress\n");
6793 0 : ctx->bserrno = -EBUSY;
6794 0 : spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6795 0 : return;
6796 : }
6797 :
6798 276 : _blob->locked_operation_in_progress = true;
6799 :
6800 276 : spdk_blob_opts_init(&opts, sizeof(opts));
6801 276 : blob_xattrs_init(&internal_xattrs);
6802 :
6803 : /* Change the size of new blob to the same as in original blob,
6804 : * but do not allocate clusters */
6805 276 : opts.thin_provision = true;
6806 276 : opts.num_clusters = spdk_blob_get_num_clusters(_blob);
6807 276 : opts.use_extent_table = _blob->use_extent_table;
6808 :
6809 : /* If there are any xattrs specified for snapshot, set them now */
6810 276 : if (ctx->xattrs) {
6811 5 : memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
6812 : }
6813 : /* Set internal xattr SNAPSHOT_IN_PROGRESS */
6814 276 : internal_xattrs.count = 1;
6815 276 : internal_xattrs.ctx = _blob;
6816 276 : internal_xattrs.names = xattrs_names;
6817 276 : internal_xattrs.get_value = bs_xattr_snapshot;
6818 :
6819 276 : bs_create_blob(_blob->bs, &opts, &internal_xattrs,
6820 : bs_snapshot_newblob_create_cpl, ctx);
6821 : }
6822 :
6823 : void
6824 289 : spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
6825 : const struct spdk_blob_xattr_opts *snapshot_xattrs,
6826 : spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
6827 : {
6828 289 : struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
6829 :
6830 289 : if (!ctx) {
6831 0 : cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
6832 0 : return;
6833 : }
6834 289 : ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
6835 289 : ctx->cpl.u.blobid.cb_fn = cb_fn;
6836 289 : ctx->cpl.u.blobid.cb_arg = cb_arg;
6837 289 : ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
6838 289 : ctx->bserrno = 0;
6839 289 : ctx->frozen = false;
6840 289 : ctx->original.id = blobid;
6841 289 : ctx->xattrs = snapshot_xattrs;
6842 :
6843 289 : spdk_bs_open_blob(bs, ctx->original.id, bs_snapshot_origblob_open_cpl, ctx);
6844 : }
6845 : /* END spdk_bs_create_snapshot */
6846 :
6847 : /* START spdk_bs_create_clone */
6848 :
6849 : static void
6850 60 : bs_xattr_clone(void *arg, const char *name,
6851 : const void **value, size_t *value_len)
6852 : {
6853 60 : assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
6854 :
6855 60 : struct spdk_blob *blob = (struct spdk_blob *)arg;
6856 60 : *value = &blob->id;
6857 60 : *value_len = sizeof(blob->id);
6858 60 : }
6859 :
6860 : static void
6861 60 : bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6862 : {
6863 60 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6864 60 : struct spdk_blob *clone = _blob;
6865 :
6866 60 : ctx->new.blob = clone;
6867 60 : bs_blob_list_add(clone);
6868 :
6869 60 : spdk_blob_close(clone, bs_clone_snapshot_origblob_cleanup, ctx);
6870 60 : }
6871 :
6872 : static void
6873 60 : bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
6874 : {
6875 60 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6876 :
6877 60 : ctx->cpl.u.blobid.blobid = blobid;
6878 60 : spdk_bs_open_blob(ctx->original.blob->bs, blobid, bs_clone_newblob_open_cpl, ctx);
6879 60 : }
6880 :
6881 : static void
6882 65 : bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6883 : {
6884 65 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6885 : struct spdk_blob_opts opts;
6886 : struct spdk_blob_xattr_opts internal_xattrs;
6887 65 : char *xattr_names[] = { BLOB_SNAPSHOT };
6888 :
6889 65 : if (bserrno != 0) {
6890 0 : bs_clone_snapshot_cleanup_finish(ctx, bserrno);
6891 5 : return;
6892 : }
6893 :
6894 65 : ctx->original.blob = _blob;
6895 65 : ctx->original.md_ro = _blob->md_ro;
6896 :
6897 65 : if (!_blob->data_ro || !_blob->md_ro) {
6898 5 : SPDK_DEBUGLOG(blob, "Clone not from read-only blob\n");
6899 5 : ctx->bserrno = -EINVAL;
6900 5 : spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6901 5 : return;
6902 : }
6903 :
6904 60 : if (_blob->locked_operation_in_progress) {
6905 0 : SPDK_DEBUGLOG(blob, "Cannot create clone - another operation in progress\n");
6906 0 : ctx->bserrno = -EBUSY;
6907 0 : spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6908 0 : return;
6909 : }
6910 :
6911 60 : _blob->locked_operation_in_progress = true;
6912 :
6913 60 : spdk_blob_opts_init(&opts, sizeof(opts));
6914 60 : blob_xattrs_init(&internal_xattrs);
6915 :
6916 60 : opts.thin_provision = true;
6917 60 : opts.num_clusters = spdk_blob_get_num_clusters(_blob);
6918 60 : opts.use_extent_table = _blob->use_extent_table;
6919 60 : if (ctx->xattrs) {
6920 5 : memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
6921 : }
6922 :
6923 : /* Set internal xattr BLOB_SNAPSHOT */
6924 60 : internal_xattrs.count = 1;
6925 60 : internal_xattrs.ctx = _blob;
6926 60 : internal_xattrs.names = xattr_names;
6927 60 : internal_xattrs.get_value = bs_xattr_clone;
6928 :
6929 60 : bs_create_blob(_blob->bs, &opts, &internal_xattrs,
6930 : bs_clone_newblob_create_cpl, ctx);
6931 : }
6932 :
6933 : void
6934 65 : spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
6935 : const struct spdk_blob_xattr_opts *clone_xattrs,
6936 : spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
6937 : {
6938 65 : struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
6939 :
6940 65 : if (!ctx) {
6941 0 : cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
6942 0 : return;
6943 : }
6944 :
6945 65 : ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
6946 65 : ctx->cpl.u.blobid.cb_fn = cb_fn;
6947 65 : ctx->cpl.u.blobid.cb_arg = cb_arg;
6948 65 : ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
6949 65 : ctx->bserrno = 0;
6950 65 : ctx->xattrs = clone_xattrs;
6951 65 : ctx->original.id = blobid;
6952 :
6953 65 : spdk_bs_open_blob(bs, ctx->original.id, bs_clone_origblob_open_cpl, ctx);
6954 : }
6955 :
6956 : /* END spdk_bs_create_clone */
6957 :
6958 : /* START spdk_bs_inflate_blob */
6959 :
6960 : static void
6961 15 : bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
6962 : {
6963 15 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6964 15 : struct spdk_blob *_blob = ctx->original.blob;
6965 :
6966 15 : if (bserrno != 0) {
6967 0 : bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6968 0 : return;
6969 : }
6970 :
6971 : /* Temporarily override md_ro flag for MD modification */
6972 15 : _blob->md_ro = false;
6973 :
6974 15 : bserrno = blob_set_xattr(_blob, BLOB_SNAPSHOT, &_parent->id, sizeof(spdk_blob_id), true);
6975 15 : if (bserrno != 0) {
6976 0 : bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6977 0 : return;
6978 : }
6979 :
6980 15 : assert(_parent != NULL);
6981 :
6982 15 : bs_blob_list_remove(_blob);
6983 15 : _blob->parent_id = _parent->id;
6984 :
6985 15 : blob_back_bs_destroy(_blob);
6986 15 : _blob->back_bs_dev = bs_create_blob_bs_dev(_parent);
6987 15 : bs_blob_list_add(_blob);
6988 :
6989 15 : spdk_blob_sync_md(_blob, bs_clone_snapshot_origblob_cleanup, ctx);
6990 : }
6991 :
6992 : static void
6993 70 : bs_inflate_blob_done(struct spdk_clone_snapshot_ctx *ctx)
6994 : {
6995 70 : struct spdk_blob *_blob = ctx->original.blob;
6996 : struct spdk_blob *_parent;
6997 :
6998 70 : if (ctx->allocate_all) {
6999 : /* remove thin provisioning */
7000 40 : bs_blob_list_remove(_blob);
7001 40 : if (_blob->parent_id == SPDK_BLOBID_EXTERNAL_SNAPSHOT) {
7002 10 : blob_remove_xattr(_blob, BLOB_EXTERNAL_SNAPSHOT_ID, true);
7003 10 : _blob->invalid_flags &= ~SPDK_BLOB_EXTERNAL_SNAPSHOT;
7004 : } else {
7005 30 : blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
7006 : }
7007 40 : _blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
7008 40 : blob_back_bs_destroy(_blob);
7009 40 : _blob->parent_id = SPDK_BLOBID_INVALID;
7010 : } else {
7011 : /* For now, esnap clones always have allocate_all set. */
7012 30 : assert(!blob_is_esnap_clone(_blob));
7013 :
7014 30 : _parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
7015 30 : if (_parent->parent_id != SPDK_BLOBID_INVALID) {
7016 : /* We must change the parent of the inflated blob */
7017 15 : spdk_bs_open_blob(_blob->bs, _parent->parent_id,
7018 : bs_inflate_blob_set_parent_cpl, ctx);
7019 15 : return;
7020 : }
7021 :
7022 15 : bs_blob_list_remove(_blob);
7023 15 : _blob->parent_id = SPDK_BLOBID_INVALID;
7024 15 : blob_back_bs_destroy(_blob);
7025 15 : _blob->back_bs_dev = bs_create_zeroes_dev();
7026 : }
7027 :
7028 : /* Temporarily override md_ro flag for MD modification */
7029 55 : _blob->md_ro = false;
7030 55 : blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
7031 55 : _blob->state = SPDK_BLOB_STATE_DIRTY;
7032 :
7033 55 : spdk_blob_sync_md(_blob, bs_clone_snapshot_origblob_cleanup, ctx);
7034 : }
7035 :
7036 : /* Check if cluster needs allocation */
7037 : static inline bool
7038 1500 : bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
7039 : {
7040 : struct spdk_blob_bs_dev *b;
7041 :
7042 1500 : assert(blob != NULL);
7043 :
7044 1500 : if (blob->active.clusters[cluster] != 0) {
7045 : /* Cluster is already allocated */
7046 40 : return false;
7047 : }
7048 :
7049 1460 : if (blob->parent_id == SPDK_BLOBID_INVALID) {
7050 : /* Blob have no parent blob */
7051 100 : return allocate_all;
7052 : }
7053 :
7054 1360 : if (blob->parent_id == SPDK_BLOBID_EXTERNAL_SNAPSHOT) {
7055 80 : return true;
7056 : }
7057 :
7058 1280 : b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
7059 1280 : return (allocate_all || b->blob->active.clusters[cluster] != 0);
7060 : }
7061 :
7062 : static void
7063 635 : bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
7064 : {
7065 635 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
7066 635 : struct spdk_blob *_blob = ctx->original.blob;
7067 : struct spdk_bs_cpl cpl;
7068 : spdk_bs_user_op_t *op;
7069 : uint64_t offset;
7070 :
7071 635 : if (bserrno != 0) {
7072 0 : bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
7073 0 : return;
7074 : }
7075 :
7076 820 : for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
7077 750 : if (bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
7078 565 : break;
7079 : }
7080 : }
7081 :
7082 635 : if (ctx->cluster < _blob->active.num_clusters) {
7083 565 : offset = bs_cluster_to_lba(_blob->bs, ctx->cluster);
7084 :
7085 : /* We may safely increment a cluster before copying */
7086 565 : ctx->cluster++;
7087 :
7088 : /* Use a dummy 0B read as a context for cluster copy */
7089 565 : cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
7090 565 : cpl.u.blob_basic.cb_fn = bs_inflate_blob_touch_next;
7091 565 : cpl.u.blob_basic.cb_arg = ctx;
7092 :
7093 565 : op = bs_user_op_alloc(ctx->channel, &cpl, SPDK_BLOB_READ, _blob,
7094 : NULL, 0, offset, 0);
7095 565 : if (!op) {
7096 0 : bs_clone_snapshot_origblob_cleanup(ctx, -ENOMEM);
7097 0 : return;
7098 : }
7099 :
7100 565 : bs_allocate_and_copy_cluster(_blob, ctx->channel, offset, op);
7101 : } else {
7102 70 : bs_inflate_blob_done(ctx);
7103 : }
7104 : }
7105 :
7106 : static void
7107 75 : bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
7108 : {
7109 75 : struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
7110 : uint64_t clusters_needed;
7111 : uint64_t i;
7112 :
7113 75 : if (bserrno != 0) {
7114 0 : bs_clone_snapshot_cleanup_finish(ctx, bserrno);
7115 0 : return;
7116 : }
7117 :
7118 75 : ctx->original.blob = _blob;
7119 75 : ctx->original.md_ro = _blob->md_ro;
7120 :
7121 75 : if (_blob->locked_operation_in_progress) {
7122 0 : SPDK_DEBUGLOG(blob, "Cannot inflate blob - another operation in progress\n");
7123 0 : ctx->bserrno = -EBUSY;
7124 0 : spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
7125 0 : return;
7126 : }
7127 :
7128 75 : _blob->locked_operation_in_progress = true;
7129 :
7130 75 : switch (_blob->parent_id) {
7131 10 : case SPDK_BLOBID_INVALID:
7132 10 : if (!ctx->allocate_all) {
7133 : /* This blob has no parent, so we cannot decouple it. */
7134 5 : SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
7135 5 : bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
7136 5 : return;
7137 : }
7138 5 : break;
7139 10 : case SPDK_BLOBID_EXTERNAL_SNAPSHOT:
7140 : /*
7141 : * It would be better to rely on back_bs_dev->is_zeroes(), to determine which
7142 : * clusters require allocation. Until there is a blobstore consumer that
7143 : * uses esnaps with an spdk_bs_dev that implements a useful is_zeroes() it is not
7144 : * worth the effort.
7145 : */
7146 10 : ctx->allocate_all = true;
7147 10 : break;
7148 55 : default:
7149 55 : break;
7150 : }
7151 :
7152 70 : if (spdk_blob_is_thin_provisioned(_blob) == false) {
7153 : /* This is not thin provisioned blob. No need to inflate. */
7154 0 : bs_clone_snapshot_origblob_cleanup(ctx, 0);
7155 0 : return;
7156 : }
7157 :
7158 : /* Do two passes - one to verify that we can obtain enough clusters
7159 : * and another to actually claim them.
7160 : */
7161 70 : clusters_needed = 0;
7162 820 : for (i = 0; i < _blob->active.num_clusters; i++) {
7163 750 : if (bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
7164 565 : clusters_needed++;
7165 : }
7166 : }
7167 :
7168 70 : if (clusters_needed > _blob->bs->num_free_clusters) {
7169 : /* Not enough free clusters. Cannot satisfy the request. */
7170 0 : bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
7171 0 : return;
7172 : }
7173 :
7174 70 : ctx->cluster = 0;
7175 70 : bs_inflate_blob_touch_next(ctx, 0);
7176 : }
7177 :
7178 : static void
7179 75 : bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
7180 : spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
7181 : {
7182 75 : struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
7183 :
7184 75 : if (!ctx) {
7185 0 : cb_fn(cb_arg, -ENOMEM);
7186 0 : return;
7187 : }
7188 75 : ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
7189 75 : ctx->cpl.u.bs_basic.cb_fn = cb_fn;
7190 75 : ctx->cpl.u.bs_basic.cb_arg = cb_arg;
7191 75 : ctx->bserrno = 0;
7192 75 : ctx->original.id = blobid;
7193 75 : ctx->channel = channel;
7194 75 : ctx->allocate_all = allocate_all;
7195 :
7196 75 : spdk_bs_open_blob(bs, ctx->original.id, bs_inflate_blob_open_cpl, ctx);
7197 : }
7198 :
7199 : void
7200 35 : spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
7201 : spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
7202 : {
7203 35 : bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
7204 35 : }
7205 :
7206 : void
7207 40 : spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
7208 : spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
7209 : {
7210 40 : bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
7211 40 : }
7212 : /* END spdk_bs_inflate_blob */
7213 :
7214 : /* START spdk_bs_blob_shallow_copy */
7215 :
7216 : struct shallow_copy_ctx {
7217 : struct spdk_bs_cpl cpl;
7218 : int bserrno;
7219 :
7220 : /* Blob source for copy */
7221 : struct spdk_blob_store *bs;
7222 : spdk_blob_id blobid;
7223 : struct spdk_blob *blob;
7224 : struct spdk_io_channel *blob_channel;
7225 :
7226 : /* Destination device for copy */
7227 : struct spdk_bs_dev *ext_dev;
7228 : struct spdk_io_channel *ext_channel;
7229 :
7230 : /* Current cluster for copy operation */
7231 : uint64_t cluster;
7232 :
7233 : /* Buffer for blob reading */
7234 : uint8_t *read_buff;
7235 :
7236 : /* Struct for external device writing */
7237 : struct spdk_bs_dev_cb_args ext_args;
7238 :
7239 : /* Actual number of copied clusters */
7240 : uint64_t copied_clusters_count;
7241 :
7242 : /* Status callback for updates about the ongoing operation */
7243 : spdk_blob_shallow_copy_status status_cb;
7244 :
7245 : /* Argument passed to function status_cb */
7246 : void *status_cb_arg;
7247 : };
7248 :
7249 : static void
7250 20 : bs_shallow_copy_cleanup_finish(void *cb_arg, int bserrno)
7251 : {
7252 20 : struct shallow_copy_ctx *ctx = cb_arg;
7253 20 : struct spdk_bs_cpl *cpl = &ctx->cpl;
7254 :
7255 20 : if (bserrno != 0) {
7256 0 : SPDK_ERRLOG("blob 0x%" PRIx64 " shallow copy, cleanup error %d\n", ctx->blob->id, bserrno);
7257 0 : ctx->bserrno = bserrno;
7258 : }
7259 :
7260 20 : ctx->ext_dev->destroy_channel(ctx->ext_dev, ctx->ext_channel);
7261 20 : spdk_free(ctx->read_buff);
7262 :
7263 20 : cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
7264 :
7265 20 : free(ctx);
7266 20 : }
7267 :
7268 : static void
7269 10 : bs_shallow_copy_bdev_write_cpl(struct spdk_io_channel *channel, void *cb_arg, int bserrno)
7270 : {
7271 10 : struct shallow_copy_ctx *ctx = cb_arg;
7272 10 : struct spdk_blob *_blob = ctx->blob;
7273 :
7274 10 : if (bserrno != 0) {
7275 0 : SPDK_ERRLOG("blob 0x%" PRIx64 " shallow copy, ext dev write error %d\n", ctx->blob->id, bserrno);
7276 0 : ctx->bserrno = bserrno;
7277 0 : _blob->locked_operation_in_progress = false;
7278 0 : spdk_blob_close(_blob, bs_shallow_copy_cleanup_finish, ctx);
7279 0 : return;
7280 : }
7281 :
7282 10 : ctx->cluster++;
7283 10 : if (ctx->status_cb) {
7284 10 : ctx->copied_clusters_count++;
7285 10 : ctx->status_cb(ctx->copied_clusters_count, ctx->status_cb_arg);
7286 : }
7287 :
7288 10 : bs_shallow_copy_cluster_find_next(ctx);
7289 : }
7290 :
7291 : static void
7292 10 : bs_shallow_copy_blob_read_cpl(void *cb_arg, int bserrno)
7293 : {
7294 10 : struct shallow_copy_ctx *ctx = cb_arg;
7295 10 : struct spdk_bs_dev *ext_dev = ctx->ext_dev;
7296 10 : struct spdk_blob *_blob = ctx->blob;
7297 :
7298 10 : if (bserrno != 0) {
7299 0 : SPDK_ERRLOG("blob 0x%" PRIx64 " shallow copy, blob read error %d\n", ctx->blob->id, bserrno);
7300 0 : ctx->bserrno = bserrno;
7301 0 : _blob->locked_operation_in_progress = false;
7302 0 : spdk_blob_close(_blob, bs_shallow_copy_cleanup_finish, ctx);
7303 0 : return;
7304 : }
7305 :
7306 10 : ctx->ext_args.channel = ctx->ext_channel;
7307 10 : ctx->ext_args.cb_fn = bs_shallow_copy_bdev_write_cpl;
7308 10 : ctx->ext_args.cb_arg = ctx;
7309 :
7310 20 : ext_dev->write(ext_dev, ctx->ext_channel, ctx->read_buff,
7311 10 : bs_cluster_to_lba(_blob->bs, ctx->cluster),
7312 10 : bs_dev_byte_to_lba(_blob->bs->dev, _blob->bs->cluster_sz),
7313 : &ctx->ext_args);
7314 : }
7315 :
7316 : static void
7317 15 : bs_shallow_copy_cluster_find_next(void *cb_arg)
7318 : {
7319 15 : struct shallow_copy_ctx *ctx = cb_arg;
7320 15 : struct spdk_blob *_blob = ctx->blob;
7321 :
7322 25 : while (ctx->cluster < _blob->active.num_clusters) {
7323 20 : if (_blob->active.clusters[ctx->cluster] != 0) {
7324 10 : break;
7325 : }
7326 :
7327 10 : ctx->cluster++;
7328 : }
7329 :
7330 15 : if (ctx->cluster < _blob->active.num_clusters) {
7331 20 : blob_request_submit_op_single(ctx->blob_channel, _blob, ctx->read_buff,
7332 10 : bs_cluster_to_lba(_blob->bs, ctx->cluster),
7333 10 : bs_dev_byte_to_lba(_blob->bs->dev, _blob->bs->cluster_sz),
7334 : bs_shallow_copy_blob_read_cpl, ctx, SPDK_BLOB_READ);
7335 : } else {
7336 5 : _blob->locked_operation_in_progress = false;
7337 5 : spdk_blob_close(_blob, bs_shallow_copy_cleanup_finish, ctx);
7338 : }
7339 15 : }
7340 :
7341 : static void
7342 20 : bs_shallow_copy_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
7343 : {
7344 20 : struct shallow_copy_ctx *ctx = cb_arg;
7345 20 : struct spdk_bs_dev *ext_dev = ctx->ext_dev;
7346 : uint32_t blob_block_size;
7347 : uint64_t blob_total_size;
7348 :
7349 20 : if (bserrno != 0) {
7350 0 : SPDK_ERRLOG("Shallow copy blob open error %d\n", bserrno);
7351 0 : ctx->bserrno = bserrno;
7352 0 : bs_shallow_copy_cleanup_finish(ctx, 0);
7353 0 : return;
7354 : }
7355 :
7356 20 : if (!spdk_blob_is_read_only(_blob)) {
7357 5 : SPDK_ERRLOG("blob 0x%" PRIx64 " shallow copy, blob must be read only\n", _blob->id);
7358 5 : ctx->bserrno = -EPERM;
7359 5 : spdk_blob_close(_blob, bs_shallow_copy_cleanup_finish, ctx);
7360 5 : return;
7361 : }
7362 :
7363 15 : blob_block_size = _blob->bs->dev->blocklen;
7364 15 : blob_total_size = spdk_blob_get_num_clusters(_blob) * spdk_bs_get_cluster_size(_blob->bs);
7365 :
7366 15 : if (blob_total_size > ext_dev->blockcnt * ext_dev->blocklen) {
7367 5 : SPDK_ERRLOG("blob 0x%" PRIx64 " shallow copy, external device must have at least blob size\n",
7368 : _blob->id);
7369 5 : ctx->bserrno = -EINVAL;
7370 5 : spdk_blob_close(_blob, bs_shallow_copy_cleanup_finish, ctx);
7371 5 : return;
7372 : }
7373 :
7374 10 : if (blob_block_size % ext_dev->blocklen != 0) {
7375 5 : SPDK_ERRLOG("blob 0x%" PRIx64 " shallow copy, external device block size is not compatible with \
7376 : blobstore block size\n", _blob->id);
7377 5 : ctx->bserrno = -EINVAL;
7378 5 : spdk_blob_close(_blob, bs_shallow_copy_cleanup_finish, ctx);
7379 5 : return;
7380 : }
7381 :
7382 5 : ctx->blob = _blob;
7383 :
7384 5 : if (_blob->locked_operation_in_progress) {
7385 0 : SPDK_DEBUGLOG(blob, "blob 0x%" PRIx64 " shallow copy - another operation in progress\n", _blob->id);
7386 0 : ctx->bserrno = -EBUSY;
7387 0 : spdk_blob_close(_blob, bs_shallow_copy_cleanup_finish, ctx);
7388 0 : return;
7389 : }
7390 :
7391 5 : _blob->locked_operation_in_progress = true;
7392 :
7393 5 : ctx->cluster = 0;
7394 5 : bs_shallow_copy_cluster_find_next(ctx);
7395 : }
7396 :
7397 : int
7398 20 : spdk_bs_blob_shallow_copy(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
7399 : spdk_blob_id blobid, struct spdk_bs_dev *ext_dev,
7400 : spdk_blob_shallow_copy_status status_cb_fn, void *status_cb_arg,
7401 : spdk_blob_op_complete cb_fn, void *cb_arg)
7402 : {
7403 : struct shallow_copy_ctx *ctx;
7404 : struct spdk_io_channel *ext_channel;
7405 :
7406 20 : ctx = calloc(1, sizeof(*ctx));
7407 20 : if (!ctx) {
7408 0 : return -ENOMEM;
7409 : }
7410 :
7411 20 : ctx->bs = bs;
7412 20 : ctx->blobid = blobid;
7413 20 : ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
7414 20 : ctx->cpl.u.bs_basic.cb_fn = cb_fn;
7415 20 : ctx->cpl.u.bs_basic.cb_arg = cb_arg;
7416 20 : ctx->bserrno = 0;
7417 20 : ctx->blob_channel = channel;
7418 20 : ctx->status_cb = status_cb_fn;
7419 20 : ctx->status_cb_arg = status_cb_arg;
7420 20 : ctx->read_buff = spdk_malloc(bs->cluster_sz, bs->dev->blocklen, NULL,
7421 : SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
7422 20 : if (!ctx->read_buff) {
7423 0 : free(ctx);
7424 0 : return -ENOMEM;
7425 : }
7426 :
7427 20 : ext_channel = ext_dev->create_channel(ext_dev);
7428 20 : if (!ext_channel) {
7429 0 : spdk_free(ctx->read_buff);
7430 0 : free(ctx);
7431 0 : return -ENOMEM;
7432 : }
7433 20 : ctx->ext_dev = ext_dev;
7434 20 : ctx->ext_channel = ext_channel;
7435 :
7436 20 : spdk_bs_open_blob(ctx->bs, ctx->blobid, bs_shallow_copy_blob_open_cpl, ctx);
7437 :
7438 20 : return 0;
7439 : }
7440 : /* END spdk_bs_blob_shallow_copy */
7441 :
7442 : /* START spdk_bs_blob_set_parent */
7443 :
7444 : struct set_parent_ctx {
7445 : struct spdk_blob_store *bs;
7446 : int bserrno;
7447 : spdk_bs_op_complete cb_fn;
7448 : void *cb_arg;
7449 :
7450 : struct spdk_blob *blob;
7451 : bool blob_md_ro;
7452 :
7453 : struct blob_parent parent;
7454 : };
7455 :
7456 : static void
7457 30 : bs_set_parent_cleanup_finish(void *cb_arg, int bserrno)
7458 : {
7459 30 : struct set_parent_ctx *ctx = cb_arg;
7460 :
7461 30 : assert(ctx != NULL);
7462 :
7463 30 : if (bserrno != 0) {
7464 0 : SPDK_ERRLOG("blob set parent finish error %d\n", bserrno);
7465 0 : if (ctx->bserrno == 0) {
7466 0 : ctx->bserrno = bserrno;
7467 : }
7468 : }
7469 :
7470 30 : ctx->cb_fn(ctx->cb_arg, ctx->bserrno);
7471 :
7472 30 : free(ctx);
7473 30 : }
7474 :
7475 : static void
7476 25 : bs_set_parent_close_snapshot(void *cb_arg, int bserrno)
7477 : {
7478 25 : struct set_parent_ctx *ctx = cb_arg;
7479 :
7480 25 : if (ctx->bserrno != 0) {
7481 10 : spdk_blob_close(ctx->parent.u.snapshot.blob, bs_set_parent_cleanup_finish, ctx);
7482 10 : return;
7483 : }
7484 :
7485 15 : if (bserrno != 0) {
7486 0 : SPDK_ERRLOG("blob close error %d\n", bserrno);
7487 0 : ctx->bserrno = bserrno;
7488 : }
7489 :
7490 15 : bs_set_parent_cleanup_finish(ctx, ctx->bserrno);
7491 : }
7492 :
7493 : static void
7494 15 : bs_set_parent_close_blob(void *cb_arg, int bserrno)
7495 : {
7496 15 : struct set_parent_ctx *ctx = cb_arg;
7497 15 : struct spdk_blob *blob = ctx->blob;
7498 15 : struct spdk_blob *snapshot = ctx->parent.u.snapshot.blob;
7499 :
7500 15 : if (bserrno != 0 && ctx->bserrno == 0) {
7501 0 : SPDK_ERRLOG("error %d in metadata sync\n", bserrno);
7502 0 : ctx->bserrno = bserrno;
7503 : }
7504 :
7505 : /* Revert md_ro to original state */
7506 15 : blob->md_ro = ctx->blob_md_ro;
7507 :
7508 15 : blob->locked_operation_in_progress = false;
7509 15 : snapshot->locked_operation_in_progress = false;
7510 :
7511 15 : spdk_blob_close(blob, bs_set_parent_close_snapshot, ctx);
7512 15 : }
7513 :
7514 : static void
7515 15 : bs_set_parent_set_back_bs_dev_done(void *cb_arg, int bserrno)
7516 : {
7517 15 : struct set_parent_ctx *ctx = cb_arg;
7518 15 : struct spdk_blob *blob = ctx->blob;
7519 :
7520 15 : if (bserrno != 0) {
7521 0 : SPDK_ERRLOG("error %d setting back_bs_dev\n", bserrno);
7522 0 : ctx->bserrno = bserrno;
7523 0 : bs_set_parent_close_blob(ctx, bserrno);
7524 0 : return;
7525 : }
7526 :
7527 15 : spdk_blob_sync_md(blob, bs_set_parent_close_blob, ctx);
7528 : }
7529 :
7530 : static int
7531 15 : bs_set_parent_refs(struct spdk_blob *blob, struct blob_parent *parent)
7532 : {
7533 : int rc;
7534 :
7535 15 : bs_blob_list_remove(blob);
7536 :
7537 15 : rc = blob_set_xattr(blob, BLOB_SNAPSHOT, &parent->u.snapshot.id, sizeof(spdk_blob_id), true);
7538 15 : if (rc != 0) {
7539 0 : SPDK_ERRLOG("error %d setting snapshot xattr\n", rc);
7540 0 : return rc;
7541 : }
7542 15 : blob->parent_id = parent->u.snapshot.id;
7543 :
7544 15 : if (blob_is_esnap_clone(blob)) {
7545 : /* Remove the xattr that references the external snapshot */
7546 5 : blob->invalid_flags &= ~SPDK_BLOB_EXTERNAL_SNAPSHOT;
7547 5 : blob_remove_xattr(blob, BLOB_EXTERNAL_SNAPSHOT_ID, true);
7548 : }
7549 :
7550 15 : bs_blob_list_add(blob);
7551 :
7552 15 : return 0;
7553 : }
7554 :
7555 : static void
7556 25 : bs_set_parent_snapshot_open_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
7557 : {
7558 25 : struct set_parent_ctx *ctx = cb_arg;
7559 25 : struct spdk_blob *blob = ctx->blob;
7560 : struct spdk_bs_dev *back_bs_dev;
7561 :
7562 25 : if (bserrno != 0) {
7563 0 : SPDK_ERRLOG("snapshot open error %d\n", bserrno);
7564 0 : ctx->bserrno = bserrno;
7565 0 : spdk_blob_close(blob, bs_set_parent_cleanup_finish, ctx);
7566 0 : return;
7567 : }
7568 :
7569 25 : ctx->parent.u.snapshot.blob = snapshot;
7570 25 : ctx->parent.u.snapshot.id = snapshot->id;
7571 :
7572 25 : if (!spdk_blob_is_snapshot(snapshot)) {
7573 5 : SPDK_ERRLOG("parent blob is not a snapshot\n");
7574 5 : ctx->bserrno = -EINVAL;
7575 5 : spdk_blob_close(blob, bs_set_parent_close_snapshot, ctx);
7576 5 : return;
7577 : }
7578 :
7579 20 : if (blob->active.num_clusters != snapshot->active.num_clusters) {
7580 5 : SPDK_ERRLOG("parent blob has a number of clusters different from child's ones\n");
7581 5 : ctx->bserrno = -EINVAL;
7582 5 : spdk_blob_close(blob, bs_set_parent_close_snapshot, ctx);
7583 5 : return;
7584 : }
7585 :
7586 15 : if (blob->locked_operation_in_progress || snapshot->locked_operation_in_progress) {
7587 0 : SPDK_ERRLOG("cannot set parent of blob, another operation in progress\n");
7588 0 : ctx->bserrno = -EBUSY;
7589 0 : spdk_blob_close(blob, bs_set_parent_close_snapshot, ctx);
7590 0 : return;
7591 : }
7592 :
7593 15 : blob->locked_operation_in_progress = true;
7594 15 : snapshot->locked_operation_in_progress = true;
7595 :
7596 : /* Temporarily override md_ro flag for MD modification */
7597 15 : blob->md_ro = false;
7598 :
7599 15 : back_bs_dev = bs_create_blob_bs_dev(snapshot);
7600 :
7601 15 : blob_set_back_bs_dev(blob, back_bs_dev, bs_set_parent_refs, &ctx->parent,
7602 : bs_set_parent_set_back_bs_dev_done,
7603 : ctx);
7604 : }
7605 :
7606 : static void
7607 30 : bs_set_parent_blob_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
7608 : {
7609 30 : struct set_parent_ctx *ctx = cb_arg;
7610 :
7611 30 : if (bserrno != 0) {
7612 0 : SPDK_ERRLOG("blob open error %d\n", bserrno);
7613 0 : ctx->bserrno = bserrno;
7614 0 : bs_set_parent_cleanup_finish(ctx, 0);
7615 0 : return;
7616 : }
7617 :
7618 30 : if (!spdk_blob_is_thin_provisioned(blob)) {
7619 5 : SPDK_ERRLOG("blob is not thin-provisioned\n");
7620 5 : ctx->bserrno = -EINVAL;
7621 5 : spdk_blob_close(blob, bs_set_parent_cleanup_finish, ctx);
7622 5 : return;
7623 : }
7624 :
7625 25 : ctx->blob = blob;
7626 25 : ctx->blob_md_ro = blob->md_ro;
7627 :
7628 25 : spdk_bs_open_blob(ctx->bs, ctx->parent.u.snapshot.id, bs_set_parent_snapshot_open_cpl, ctx);
7629 : }
7630 :
7631 : void
7632 45 : spdk_bs_blob_set_parent(struct spdk_blob_store *bs, spdk_blob_id blob_id,
7633 : spdk_blob_id snapshot_id, spdk_blob_op_complete cb_fn, void *cb_arg)
7634 : {
7635 : struct set_parent_ctx *ctx;
7636 :
7637 45 : if (snapshot_id == SPDK_BLOBID_INVALID) {
7638 5 : SPDK_ERRLOG("snapshot id not valid\n");
7639 5 : cb_fn(cb_arg, -EINVAL);
7640 5 : return;
7641 : }
7642 :
7643 40 : if (blob_id == snapshot_id) {
7644 5 : SPDK_ERRLOG("blob id and snapshot id cannot be the same\n");
7645 5 : cb_fn(cb_arg, -EINVAL);
7646 5 : return;
7647 : }
7648 :
7649 35 : if (spdk_blob_get_parent_snapshot(bs, blob_id) == snapshot_id) {
7650 5 : SPDK_NOTICELOG("snapshot is already the parent of blob\n");
7651 5 : cb_fn(cb_arg, -EEXIST);
7652 5 : return;
7653 : }
7654 :
7655 30 : ctx = calloc(1, sizeof(*ctx));
7656 30 : if (!ctx) {
7657 0 : cb_fn(cb_arg, -ENOMEM);
7658 0 : return;
7659 : }
7660 :
7661 30 : ctx->bs = bs;
7662 30 : ctx->parent.u.snapshot.id = snapshot_id;
7663 30 : ctx->cb_fn = cb_fn;
7664 30 : ctx->cb_arg = cb_arg;
7665 30 : ctx->bserrno = 0;
7666 :
7667 30 : spdk_bs_open_blob(bs, blob_id, bs_set_parent_blob_open_cpl, ctx);
7668 : }
7669 : /* END spdk_bs_blob_set_parent */
7670 :
7671 : /* START spdk_bs_blob_set_external_parent */
7672 :
7673 : static void
7674 20 : bs_set_external_parent_cleanup_finish(void *cb_arg, int bserrno)
7675 : {
7676 20 : struct set_parent_ctx *ctx = cb_arg;
7677 :
7678 20 : if (bserrno != 0) {
7679 0 : SPDK_ERRLOG("blob set external parent finish error %d\n", bserrno);
7680 0 : if (ctx->bserrno == 0) {
7681 0 : ctx->bserrno = bserrno;
7682 : }
7683 : }
7684 :
7685 20 : ctx->cb_fn(ctx->cb_arg, ctx->bserrno);
7686 :
7687 20 : free(ctx->parent.u.esnap.id);
7688 20 : free(ctx);
7689 20 : }
7690 :
7691 : static void
7692 10 : bs_set_external_parent_close_blob(void *cb_arg, int bserrno)
7693 : {
7694 10 : struct set_parent_ctx *ctx = cb_arg;
7695 10 : struct spdk_blob *blob = ctx->blob;
7696 :
7697 10 : if (bserrno != 0 && ctx->bserrno == 0) {
7698 0 : SPDK_ERRLOG("error %d in metadata sync\n", bserrno);
7699 0 : ctx->bserrno = bserrno;
7700 : }
7701 :
7702 : /* Revert md_ro to original state */
7703 10 : blob->md_ro = ctx->blob_md_ro;
7704 :
7705 10 : blob->locked_operation_in_progress = false;
7706 :
7707 10 : spdk_blob_close(blob, bs_set_external_parent_cleanup_finish, ctx);
7708 10 : }
7709 :
7710 : static void
7711 10 : bs_set_external_parent_unfrozen(void *cb_arg, int bserrno)
7712 : {
7713 10 : struct set_parent_ctx *ctx = cb_arg;
7714 10 : struct spdk_blob *blob = ctx->blob;
7715 :
7716 10 : if (bserrno != 0) {
7717 0 : SPDK_ERRLOG("error %d setting back_bs_dev\n", bserrno);
7718 0 : ctx->bserrno = bserrno;
7719 0 : bs_set_external_parent_close_blob(ctx, bserrno);
7720 0 : return;
7721 : }
7722 :
7723 10 : spdk_blob_sync_md(blob, bs_set_external_parent_close_blob, ctx);
7724 : }
7725 :
7726 : static int
7727 10 : bs_set_external_parent_refs(struct spdk_blob *blob, struct blob_parent *parent)
7728 : {
7729 : int rc;
7730 :
7731 10 : bs_blob_list_remove(blob);
7732 :
7733 10 : if (spdk_blob_is_clone(blob)) {
7734 : /* Remove the xattr that references the snapshot */
7735 0 : blob->parent_id = SPDK_BLOBID_INVALID;
7736 0 : blob_remove_xattr(blob, BLOB_SNAPSHOT, true);
7737 : }
7738 :
7739 10 : rc = blob_set_xattr(blob, BLOB_EXTERNAL_SNAPSHOT_ID, parent->u.esnap.id,
7740 10 : parent->u.esnap.id_len, true);
7741 10 : if (rc != 0) {
7742 0 : SPDK_ERRLOG("error %d setting external snapshot xattr\n", rc);
7743 0 : return rc;
7744 : }
7745 10 : blob->invalid_flags |= SPDK_BLOB_EXTERNAL_SNAPSHOT;
7746 10 : blob->parent_id = SPDK_BLOBID_EXTERNAL_SNAPSHOT;
7747 :
7748 10 : bs_blob_list_add(blob);
7749 :
7750 10 : return 0;
7751 : }
7752 :
7753 : static void
7754 20 : bs_set_external_parent_blob_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
7755 : {
7756 20 : struct set_parent_ctx *ctx = cb_arg;
7757 : const void *esnap_id;
7758 : size_t esnap_id_len;
7759 : int rc;
7760 :
7761 20 : if (bserrno != 0) {
7762 0 : SPDK_ERRLOG("blob open error %d\n", bserrno);
7763 0 : ctx->bserrno = bserrno;
7764 0 : bs_set_parent_cleanup_finish(ctx, 0);
7765 10 : return;
7766 : }
7767 :
7768 20 : ctx->blob = blob;
7769 20 : ctx->blob_md_ro = blob->md_ro;
7770 :
7771 20 : rc = spdk_blob_get_esnap_id(blob, &esnap_id, &esnap_id_len);
7772 20 : if (rc == 0 && esnap_id != NULL && esnap_id_len == ctx->parent.u.esnap.id_len &&
7773 5 : memcmp(esnap_id, ctx->parent.u.esnap.id, esnap_id_len) == 0) {
7774 5 : SPDK_ERRLOG("external snapshot is already the parent of blob\n");
7775 5 : ctx->bserrno = -EEXIST;
7776 5 : goto error;
7777 : }
7778 :
7779 15 : if (!spdk_blob_is_thin_provisioned(blob)) {
7780 5 : SPDK_ERRLOG("blob is not thin-provisioned\n");
7781 5 : ctx->bserrno = -EINVAL;
7782 5 : goto error;
7783 : }
7784 :
7785 10 : if (blob->locked_operation_in_progress) {
7786 0 : SPDK_ERRLOG("cannot set external parent of blob, another operation in progress\n");
7787 0 : ctx->bserrno = -EBUSY;
7788 0 : goto error;
7789 : }
7790 :
7791 10 : blob->locked_operation_in_progress = true;
7792 :
7793 : /* Temporarily override md_ro flag for MD modification */
7794 10 : blob->md_ro = false;
7795 :
7796 10 : blob_set_back_bs_dev(blob, ctx->parent.u.esnap.back_bs_dev, bs_set_external_parent_refs,
7797 : &ctx->parent, bs_set_external_parent_unfrozen, ctx);
7798 10 : return;
7799 :
7800 10 : error:
7801 10 : spdk_blob_close(blob, bs_set_external_parent_cleanup_finish, ctx);
7802 : }
7803 :
7804 : void
7805 30 : spdk_bs_blob_set_external_parent(struct spdk_blob_store *bs, spdk_blob_id blob_id,
7806 : struct spdk_bs_dev *esnap_bs_dev, const void *esnap_id,
7807 : uint32_t esnap_id_len, spdk_blob_op_complete cb_fn, void *cb_arg)
7808 : {
7809 : struct set_parent_ctx *ctx;
7810 : uint64_t esnap_dev_size, cluster_sz;
7811 :
7812 30 : if (sizeof(blob_id) == esnap_id_len && memcmp(&blob_id, esnap_id, sizeof(blob_id)) == 0) {
7813 5 : SPDK_ERRLOG("blob id and external snapshot id cannot be the same\n");
7814 5 : cb_fn(cb_arg, -EINVAL);
7815 5 : return;
7816 : }
7817 :
7818 25 : esnap_dev_size = esnap_bs_dev->blockcnt * esnap_bs_dev->blocklen;
7819 25 : cluster_sz = spdk_bs_get_cluster_size(bs);
7820 25 : if ((esnap_dev_size % cluster_sz) != 0) {
7821 5 : SPDK_ERRLOG("Esnap device size %" PRIu64 " is not an integer multiple of "
7822 : "cluster size %" PRIu64 "\n", esnap_dev_size, cluster_sz);
7823 5 : cb_fn(cb_arg, -EINVAL);
7824 5 : return;
7825 : }
7826 :
7827 20 : ctx = calloc(1, sizeof(*ctx));
7828 20 : if (!ctx) {
7829 0 : cb_fn(cb_arg, -ENOMEM);
7830 0 : return;
7831 : }
7832 :
7833 20 : ctx->parent.u.esnap.id = calloc(1, esnap_id_len);
7834 20 : if (!ctx->parent.u.esnap.id) {
7835 0 : free(ctx);
7836 0 : cb_fn(cb_arg, -ENOMEM);
7837 0 : return;
7838 : }
7839 :
7840 20 : ctx->bs = bs;
7841 20 : ctx->parent.u.esnap.back_bs_dev = esnap_bs_dev;
7842 20 : memcpy(ctx->parent.u.esnap.id, esnap_id, esnap_id_len);
7843 20 : ctx->parent.u.esnap.id_len = esnap_id_len;
7844 20 : ctx->cb_fn = cb_fn;
7845 20 : ctx->cb_arg = cb_arg;
7846 20 : ctx->bserrno = 0;
7847 :
7848 20 : spdk_bs_open_blob(bs, blob_id, bs_set_external_parent_blob_open_cpl, ctx);
7849 : }
7850 : /* END spdk_bs_blob_set_external_parent */
7851 :
7852 : /* START spdk_blob_resize */
7853 : struct spdk_bs_resize_ctx {
7854 : spdk_blob_op_complete cb_fn;
7855 : void *cb_arg;
7856 : struct spdk_blob *blob;
7857 : uint64_t sz;
7858 : int rc;
7859 : };
7860 :
7861 : static void
7862 252 : bs_resize_unfreeze_cpl(void *cb_arg, int rc)
7863 : {
7864 252 : struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
7865 :
7866 252 : if (rc != 0) {
7867 0 : SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
7868 : }
7869 :
7870 252 : if (ctx->rc != 0) {
7871 5 : SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
7872 5 : rc = ctx->rc;
7873 : }
7874 :
7875 252 : ctx->blob->locked_operation_in_progress = false;
7876 :
7877 252 : ctx->cb_fn(ctx->cb_arg, rc);
7878 252 : free(ctx);
7879 252 : }
7880 :
7881 : static void
7882 252 : bs_resize_freeze_cpl(void *cb_arg, int rc)
7883 : {
7884 252 : struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
7885 :
7886 252 : if (rc != 0) {
7887 0 : ctx->blob->locked_operation_in_progress = false;
7888 0 : ctx->cb_fn(ctx->cb_arg, rc);
7889 0 : free(ctx);
7890 0 : return;
7891 : }
7892 :
7893 252 : ctx->rc = blob_resize(ctx->blob, ctx->sz);
7894 :
7895 252 : blob_unfreeze_io(ctx->blob, bs_resize_unfreeze_cpl, ctx);
7896 : }
7897 :
7898 : void
7899 269 : spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
7900 : {
7901 : struct spdk_bs_resize_ctx *ctx;
7902 :
7903 269 : blob_verify_md_op(blob);
7904 :
7905 269 : SPDK_DEBUGLOG(blob, "Resizing blob 0x%" PRIx64 " to %" PRIu64 " clusters\n", blob->id, sz);
7906 :
7907 269 : if (blob->md_ro) {
7908 5 : cb_fn(cb_arg, -EPERM);
7909 5 : return;
7910 : }
7911 :
7912 264 : if (sz == blob->active.num_clusters) {
7913 12 : cb_fn(cb_arg, 0);
7914 12 : return;
7915 : }
7916 :
7917 252 : if (blob->locked_operation_in_progress) {
7918 0 : cb_fn(cb_arg, -EBUSY);
7919 0 : return;
7920 : }
7921 :
7922 252 : ctx = calloc(1, sizeof(*ctx));
7923 252 : if (!ctx) {
7924 0 : cb_fn(cb_arg, -ENOMEM);
7925 0 : return;
7926 : }
7927 :
7928 252 : blob->locked_operation_in_progress = true;
7929 252 : ctx->cb_fn = cb_fn;
7930 252 : ctx->cb_arg = cb_arg;
7931 252 : ctx->blob = blob;
7932 252 : ctx->sz = sz;
7933 252 : blob_freeze_io(blob, bs_resize_freeze_cpl, ctx);
7934 : }
7935 :
7936 : /* END spdk_blob_resize */
7937 :
7938 :
7939 : /* START spdk_bs_delete_blob */
7940 :
7941 : static void
7942 1862 : bs_delete_close_cpl(void *cb_arg, int bserrno)
7943 : {
7944 1862 : spdk_bs_sequence_t *seq = cb_arg;
7945 :
7946 1862 : bs_sequence_finish(seq, bserrno);
7947 1862 : }
7948 :
7949 : static void
7950 1862 : bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
7951 : {
7952 1862 : struct spdk_blob *blob = cb_arg;
7953 :
7954 1862 : if (bserrno != 0) {
7955 : /*
7956 : * We already removed this blob from the blobstore tailq, so
7957 : * we need to free it here since this is the last reference
7958 : * to it.
7959 : */
7960 0 : blob_free(blob);
7961 0 : bs_delete_close_cpl(seq, bserrno);
7962 0 : return;
7963 : }
7964 :
7965 : /*
7966 : * This will immediately decrement the ref_count and call
7967 : * the completion routine since the metadata state is clean.
7968 : * By calling spdk_blob_close, we reduce the number of call
7969 : * points into code that touches the blob->open_ref count
7970 : * and the blobstore's blob list.
7971 : */
7972 1862 : spdk_blob_close(blob, bs_delete_close_cpl, seq);
7973 : }
7974 :
7975 : struct delete_snapshot_ctx {
7976 : struct spdk_blob_list *parent_snapshot_entry;
7977 : struct spdk_blob *snapshot;
7978 : struct spdk_blob_md_page *page;
7979 : bool snapshot_md_ro;
7980 : struct spdk_blob *clone;
7981 : bool clone_md_ro;
7982 : spdk_blob_op_with_handle_complete cb_fn;
7983 : void *cb_arg;
7984 : int bserrno;
7985 : uint32_t next_extent_page;
7986 : };
7987 :
7988 : static void
7989 138 : delete_blob_cleanup_finish(void *cb_arg, int bserrno)
7990 : {
7991 138 : struct delete_snapshot_ctx *ctx = cb_arg;
7992 :
7993 138 : if (bserrno != 0) {
7994 0 : SPDK_ERRLOG("Snapshot cleanup error %d\n", bserrno);
7995 : }
7996 :
7997 138 : assert(ctx != NULL);
7998 :
7999 138 : if (bserrno != 0 && ctx->bserrno == 0) {
8000 0 : ctx->bserrno = bserrno;
8001 : }
8002 :
8003 138 : ctx->cb_fn(ctx->cb_arg, ctx->snapshot, ctx->bserrno);
8004 138 : spdk_free(ctx->page);
8005 138 : free(ctx);
8006 138 : }
8007 :
8008 : static void
8009 28 : delete_snapshot_cleanup_snapshot(void *cb_arg, int bserrno)
8010 : {
8011 28 : struct delete_snapshot_ctx *ctx = cb_arg;
8012 :
8013 28 : if (bserrno != 0) {
8014 0 : ctx->bserrno = bserrno;
8015 0 : SPDK_ERRLOG("Clone cleanup error %d\n", bserrno);
8016 : }
8017 :
8018 28 : if (ctx->bserrno != 0) {
8019 28 : assert(blob_lookup(ctx->snapshot->bs, ctx->snapshot->id) == NULL);
8020 28 : RB_INSERT(spdk_blob_tree, &ctx->snapshot->bs->open_blobs, ctx->snapshot);
8021 28 : spdk_bit_array_set(ctx->snapshot->bs->open_blobids, ctx->snapshot->id);
8022 : }
8023 :
8024 28 : ctx->snapshot->locked_operation_in_progress = false;
8025 28 : ctx->snapshot->md_ro = ctx->snapshot_md_ro;
8026 :
8027 28 : spdk_blob_close(ctx->snapshot, delete_blob_cleanup_finish, ctx);
8028 28 : }
8029 :
8030 : static void
8031 15 : delete_snapshot_cleanup_clone(void *cb_arg, int bserrno)
8032 : {
8033 15 : struct delete_snapshot_ctx *ctx = cb_arg;
8034 :
8035 15 : ctx->clone->locked_operation_in_progress = false;
8036 15 : ctx->clone->md_ro = ctx->clone_md_ro;
8037 :
8038 15 : spdk_blob_close(ctx->clone, delete_snapshot_cleanup_snapshot, ctx);
8039 15 : }
8040 :
8041 : static void
8042 60 : delete_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
8043 : {
8044 60 : struct delete_snapshot_ctx *ctx = cb_arg;
8045 :
8046 60 : if (bserrno) {
8047 0 : ctx->bserrno = bserrno;
8048 0 : delete_snapshot_cleanup_clone(ctx, 0);
8049 0 : return;
8050 : }
8051 :
8052 60 : ctx->clone->locked_operation_in_progress = false;
8053 60 : spdk_blob_close(ctx->clone, delete_blob_cleanup_finish, ctx);
8054 : }
8055 :
8056 : static void
8057 65 : delete_snapshot_sync_snapshot_cpl(void *cb_arg, int bserrno)
8058 : {
8059 65 : struct delete_snapshot_ctx *ctx = cb_arg;
8060 65 : struct spdk_blob_list *parent_snapshot_entry = NULL;
8061 65 : struct spdk_blob_list *snapshot_entry = NULL;
8062 65 : struct spdk_blob_list *clone_entry = NULL;
8063 65 : struct spdk_blob_list *snapshot_clone_entry = NULL;
8064 :
8065 65 : if (bserrno) {
8066 5 : SPDK_ERRLOG("Failed to sync MD on blob\n");
8067 5 : ctx->bserrno = bserrno;
8068 5 : delete_snapshot_cleanup_clone(ctx, 0);
8069 5 : return;
8070 : }
8071 :
8072 : /* Get snapshot entry for the snapshot we want to remove */
8073 60 : snapshot_entry = bs_get_snapshot_entry(ctx->snapshot->bs, ctx->snapshot->id);
8074 :
8075 60 : assert(snapshot_entry != NULL);
8076 :
8077 : /* Remove clone entry in this snapshot (at this point there can be only one clone) */
8078 60 : clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
8079 60 : assert(clone_entry != NULL);
8080 60 : TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
8081 60 : snapshot_entry->clone_count--;
8082 60 : assert(TAILQ_EMPTY(&snapshot_entry->clones));
8083 :
8084 60 : switch (ctx->snapshot->parent_id) {
8085 50 : case SPDK_BLOBID_INVALID:
8086 : case SPDK_BLOBID_EXTERNAL_SNAPSHOT:
8087 : /* No parent snapshot - just remove clone entry */
8088 50 : free(clone_entry);
8089 50 : break;
8090 10 : default:
8091 : /* This snapshot is at the same time a clone of another snapshot - we need to
8092 : * update parent snapshot (remove current clone, add new one inherited from
8093 : * the snapshot that is being removed) */
8094 :
8095 : /* Get snapshot entry for parent snapshot and clone entry within that snapshot for
8096 : * snapshot that we are removing */
8097 10 : blob_get_snapshot_and_clone_entries(ctx->snapshot, &parent_snapshot_entry,
8098 : &snapshot_clone_entry);
8099 :
8100 : /* Switch clone entry in parent snapshot */
8101 10 : TAILQ_INSERT_TAIL(&parent_snapshot_entry->clones, clone_entry, link);
8102 10 : TAILQ_REMOVE(&parent_snapshot_entry->clones, snapshot_clone_entry, link);
8103 10 : free(snapshot_clone_entry);
8104 : }
8105 :
8106 : /* Restore md_ro flags */
8107 60 : ctx->clone->md_ro = ctx->clone_md_ro;
8108 60 : ctx->snapshot->md_ro = ctx->snapshot_md_ro;
8109 :
8110 60 : blob_unfreeze_io(ctx->clone, delete_snapshot_unfreeze_cpl, ctx);
8111 : }
8112 :
8113 : static void
8114 70 : delete_snapshot_sync_clone_cpl(void *cb_arg, int bserrno)
8115 : {
8116 70 : struct delete_snapshot_ctx *ctx = cb_arg;
8117 : uint64_t i;
8118 :
8119 70 : ctx->snapshot->md_ro = false;
8120 :
8121 70 : if (bserrno) {
8122 5 : SPDK_ERRLOG("Failed to sync MD on clone\n");
8123 5 : ctx->bserrno = bserrno;
8124 :
8125 : /* Restore snapshot to previous state */
8126 5 : bserrno = blob_remove_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, true);
8127 5 : if (bserrno != 0) {
8128 0 : delete_snapshot_cleanup_clone(ctx, bserrno);
8129 0 : return;
8130 : }
8131 :
8132 5 : spdk_blob_sync_md(ctx->snapshot, delete_snapshot_cleanup_clone, ctx);
8133 5 : return;
8134 : }
8135 :
8136 : /* Clear cluster map entries for snapshot */
8137 690 : for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
8138 625 : if (ctx->clone->active.clusters[i] == ctx->snapshot->active.clusters[i]) {
8139 615 : if (ctx->snapshot->active.clusters[i] != 0) {
8140 410 : ctx->snapshot->active.num_allocated_clusters--;
8141 : }
8142 615 : ctx->snapshot->active.clusters[i] = 0;
8143 : }
8144 : }
8145 104 : for (i = 0; i < ctx->snapshot->active.num_extent_pages &&
8146 78 : i < ctx->clone->active.num_extent_pages; i++) {
8147 39 : if (ctx->clone->active.extent_pages[i] == ctx->snapshot->active.extent_pages[i]) {
8148 36 : ctx->snapshot->active.extent_pages[i] = 0;
8149 : }
8150 : }
8151 :
8152 65 : blob_set_thin_provision(ctx->snapshot);
8153 65 : ctx->snapshot->state = SPDK_BLOB_STATE_DIRTY;
8154 :
8155 65 : if (ctx->parent_snapshot_entry != NULL) {
8156 10 : ctx->snapshot->back_bs_dev = NULL;
8157 : }
8158 :
8159 65 : spdk_blob_sync_md(ctx->snapshot, delete_snapshot_sync_snapshot_cpl, ctx);
8160 : }
8161 :
8162 : static void
8163 70 : delete_snapshot_update_extent_pages_cpl(struct delete_snapshot_ctx *ctx)
8164 : {
8165 : int bserrno;
8166 :
8167 : /* Delete old backing bs_dev from clone (related to snapshot that will be removed) */
8168 70 : blob_back_bs_destroy(ctx->clone);
8169 :
8170 : /* Set/remove snapshot xattr and switch parent ID and backing bs_dev on clone... */
8171 70 : if (ctx->snapshot->parent_id == SPDK_BLOBID_EXTERNAL_SNAPSHOT) {
8172 10 : bserrno = bs_snapshot_copy_xattr(ctx->clone, ctx->snapshot,
8173 : BLOB_EXTERNAL_SNAPSHOT_ID);
8174 10 : if (bserrno != 0) {
8175 0 : ctx->bserrno = bserrno;
8176 :
8177 : /* Restore snapshot to previous state */
8178 0 : bserrno = blob_remove_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, true);
8179 0 : if (bserrno != 0) {
8180 0 : delete_snapshot_cleanup_clone(ctx, bserrno);
8181 0 : return;
8182 : }
8183 :
8184 0 : spdk_blob_sync_md(ctx->snapshot, delete_snapshot_cleanup_clone, ctx);
8185 0 : return;
8186 : }
8187 10 : ctx->clone->parent_id = SPDK_BLOBID_EXTERNAL_SNAPSHOT;
8188 10 : ctx->clone->back_bs_dev = ctx->snapshot->back_bs_dev;
8189 : /* Do not delete the external snapshot along with this snapshot */
8190 10 : ctx->snapshot->back_bs_dev = NULL;
8191 10 : ctx->clone->invalid_flags |= SPDK_BLOB_EXTERNAL_SNAPSHOT;
8192 60 : } else if (ctx->parent_snapshot_entry != NULL) {
8193 : /* ...to parent snapshot */
8194 10 : ctx->clone->parent_id = ctx->parent_snapshot_entry->id;
8195 10 : ctx->clone->back_bs_dev = ctx->snapshot->back_bs_dev;
8196 10 : blob_set_xattr(ctx->clone, BLOB_SNAPSHOT, &ctx->parent_snapshot_entry->id,
8197 : sizeof(spdk_blob_id),
8198 : true);
8199 : } else {
8200 : /* ...to blobid invalid and zeroes dev */
8201 50 : ctx->clone->parent_id = SPDK_BLOBID_INVALID;
8202 50 : ctx->clone->back_bs_dev = bs_create_zeroes_dev();
8203 50 : blob_remove_xattr(ctx->clone, BLOB_SNAPSHOT, true);
8204 : }
8205 :
8206 70 : spdk_blob_sync_md(ctx->clone, delete_snapshot_sync_clone_cpl, ctx);
8207 : }
8208 :
8209 : static void
8210 73 : delete_snapshot_update_extent_pages(void *cb_arg, int bserrno)
8211 : {
8212 73 : struct delete_snapshot_ctx *ctx = cb_arg;
8213 : uint32_t *extent_page;
8214 : uint64_t i;
8215 :
8216 112 : for (i = ctx->next_extent_page; i < ctx->snapshot->active.num_extent_pages &&
8217 81 : i < ctx->clone->active.num_extent_pages; i++) {
8218 42 : if (ctx->snapshot->active.extent_pages[i] == 0) {
8219 : /* No extent page to use from snapshot */
8220 12 : continue;
8221 : }
8222 :
8223 30 : extent_page = &ctx->clone->active.extent_pages[i];
8224 30 : if (*extent_page == 0) {
8225 : /* Copy extent page from snapshot when clone did not have a matching one */
8226 27 : *extent_page = ctx->snapshot->active.extent_pages[i];
8227 27 : continue;
8228 : }
8229 :
8230 : /* Clone and snapshot both contain partially filled matching extent pages.
8231 : * Update the clone extent page in place with cluster map containing the mix of both. */
8232 3 : ctx->next_extent_page = i + 1;
8233 3 : memset(ctx->page, 0, SPDK_BS_PAGE_SIZE);
8234 :
8235 3 : blob_write_extent_page(ctx->clone, *extent_page, i * SPDK_EXTENTS_PER_EP, ctx->page,
8236 : delete_snapshot_update_extent_pages, ctx);
8237 3 : return;
8238 : }
8239 70 : delete_snapshot_update_extent_pages_cpl(ctx);
8240 : }
8241 :
8242 : static void
8243 75 : delete_snapshot_sync_snapshot_xattr_cpl(void *cb_arg, int bserrno)
8244 : {
8245 75 : struct delete_snapshot_ctx *ctx = cb_arg;
8246 : uint64_t i;
8247 :
8248 : /* Temporarily override md_ro flag for clone for MD modification */
8249 75 : ctx->clone_md_ro = ctx->clone->md_ro;
8250 75 : ctx->clone->md_ro = false;
8251 :
8252 75 : if (bserrno) {
8253 5 : SPDK_ERRLOG("Failed to sync MD with xattr on blob\n");
8254 5 : ctx->bserrno = bserrno;
8255 5 : delete_snapshot_cleanup_clone(ctx, 0);
8256 5 : return;
8257 : }
8258 :
8259 : /* Copy snapshot map to clone map (only unallocated clusters in clone) */
8260 745 : for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
8261 675 : if (ctx->clone->active.clusters[i] == 0) {
8262 665 : ctx->clone->active.clusters[i] = ctx->snapshot->active.clusters[i];
8263 665 : if (ctx->clone->active.clusters[i] != 0) {
8264 460 : ctx->clone->active.num_allocated_clusters++;
8265 : }
8266 : }
8267 : }
8268 70 : ctx->next_extent_page = 0;
8269 70 : delete_snapshot_update_extent_pages(ctx, 0);
8270 : }
8271 :
8272 : static void
8273 10 : delete_snapshot_esnap_channels_destroyed_cb(void *cb_arg, struct spdk_blob *blob, int bserrno)
8274 : {
8275 10 : struct delete_snapshot_ctx *ctx = cb_arg;
8276 :
8277 10 : if (bserrno != 0) {
8278 0 : SPDK_ERRLOG("blob 0x%" PRIx64 ": failed to destroy esnap channels: %d\n",
8279 : blob->id, bserrno);
8280 : /* That error should not stop us from syncing metadata. */
8281 : }
8282 :
8283 10 : spdk_blob_sync_md(ctx->snapshot, delete_snapshot_sync_snapshot_xattr_cpl, ctx);
8284 10 : }
8285 :
8286 : static void
8287 75 : delete_snapshot_freeze_io_cb(void *cb_arg, int bserrno)
8288 : {
8289 75 : struct delete_snapshot_ctx *ctx = cb_arg;
8290 :
8291 75 : if (bserrno) {
8292 0 : SPDK_ERRLOG("Failed to freeze I/O on clone\n");
8293 0 : ctx->bserrno = bserrno;
8294 0 : delete_snapshot_cleanup_clone(ctx, 0);
8295 0 : return;
8296 : }
8297 :
8298 : /* Temporarily override md_ro flag for snapshot for MD modification */
8299 75 : ctx->snapshot_md_ro = ctx->snapshot->md_ro;
8300 75 : ctx->snapshot->md_ro = false;
8301 :
8302 : /* Mark blob as pending for removal for power failure safety, use clone id for recovery */
8303 75 : ctx->bserrno = blob_set_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, &ctx->clone->id,
8304 : sizeof(spdk_blob_id), true);
8305 75 : if (ctx->bserrno != 0) {
8306 0 : delete_snapshot_cleanup_clone(ctx, 0);
8307 0 : return;
8308 : }
8309 :
8310 75 : if (blob_is_esnap_clone(ctx->snapshot)) {
8311 10 : blob_esnap_destroy_bs_dev_channels(ctx->snapshot, false,
8312 : delete_snapshot_esnap_channels_destroyed_cb,
8313 : ctx);
8314 10 : return;
8315 : }
8316 :
8317 65 : spdk_blob_sync_md(ctx->snapshot, delete_snapshot_sync_snapshot_xattr_cpl, ctx);
8318 : }
8319 :
8320 : static void
8321 88 : delete_snapshot_open_clone_cb(void *cb_arg, struct spdk_blob *clone, int bserrno)
8322 : {
8323 88 : struct delete_snapshot_ctx *ctx = cb_arg;
8324 :
8325 88 : if (bserrno) {
8326 13 : SPDK_ERRLOG("Failed to open clone\n");
8327 13 : ctx->bserrno = bserrno;
8328 13 : delete_snapshot_cleanup_snapshot(ctx, 0);
8329 13 : return;
8330 : }
8331 :
8332 75 : ctx->clone = clone;
8333 :
8334 75 : if (clone->locked_operation_in_progress) {
8335 0 : SPDK_DEBUGLOG(blob, "Cannot remove blob - another operation in progress on its clone\n");
8336 0 : ctx->bserrno = -EBUSY;
8337 0 : spdk_blob_close(ctx->clone, delete_snapshot_cleanup_snapshot, ctx);
8338 0 : return;
8339 : }
8340 :
8341 75 : clone->locked_operation_in_progress = true;
8342 :
8343 75 : blob_freeze_io(clone, delete_snapshot_freeze_io_cb, ctx);
8344 : }
8345 :
8346 : static void
8347 88 : update_clone_on_snapshot_deletion(struct spdk_blob *snapshot, struct delete_snapshot_ctx *ctx)
8348 : {
8349 88 : struct spdk_blob_list *snapshot_entry = NULL;
8350 88 : struct spdk_blob_list *clone_entry = NULL;
8351 88 : struct spdk_blob_list *snapshot_clone_entry = NULL;
8352 :
8353 : /* Get snapshot entry for the snapshot we want to remove */
8354 88 : snapshot_entry = bs_get_snapshot_entry(snapshot->bs, snapshot->id);
8355 :
8356 88 : assert(snapshot_entry != NULL);
8357 :
8358 : /* Get clone of the snapshot (at this point there can be only one clone) */
8359 88 : clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
8360 88 : assert(snapshot_entry->clone_count == 1);
8361 88 : assert(clone_entry != NULL);
8362 :
8363 : /* Get snapshot entry for parent snapshot and clone entry within that snapshot for
8364 : * snapshot that we are removing */
8365 88 : blob_get_snapshot_and_clone_entries(snapshot, &ctx->parent_snapshot_entry,
8366 : &snapshot_clone_entry);
8367 :
8368 88 : spdk_bs_open_blob(snapshot->bs, clone_entry->id, delete_snapshot_open_clone_cb, ctx);
8369 88 : }
8370 :
8371 : static void
8372 1940 : bs_delete_blob_finish(void *cb_arg, struct spdk_blob *blob, int bserrno)
8373 : {
8374 1940 : spdk_bs_sequence_t *seq = cb_arg;
8375 1940 : struct spdk_blob_list *snapshot_entry = NULL;
8376 : uint32_t page_num;
8377 :
8378 1940 : if (bserrno) {
8379 78 : SPDK_ERRLOG("Failed to remove blob\n");
8380 78 : bs_sequence_finish(seq, bserrno);
8381 78 : return;
8382 : }
8383 :
8384 : /* Remove snapshot from the list */
8385 1862 : snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id);
8386 1862 : if (snapshot_entry != NULL) {
8387 180 : TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link);
8388 180 : free(snapshot_entry);
8389 : }
8390 :
8391 1862 : page_num = bs_blobid_to_page(blob->id);
8392 1862 : spdk_bit_array_clear(blob->bs->used_blobids, page_num);
8393 1862 : blob->state = SPDK_BLOB_STATE_DIRTY;
8394 1862 : blob->active.num_pages = 0;
8395 1862 : blob_resize(blob, 0);
8396 :
8397 1862 : blob_persist(seq, blob, bs_delete_persist_cpl, blob);
8398 : }
8399 :
8400 : static int
8401 1940 : bs_is_blob_deletable(struct spdk_blob *blob, bool *update_clone)
8402 : {
8403 1940 : struct spdk_blob_list *snapshot_entry = NULL;
8404 1940 : struct spdk_blob_list *clone_entry = NULL;
8405 1940 : struct spdk_blob *clone = NULL;
8406 1940 : bool has_one_clone = false;
8407 :
8408 : /* Check if this is a snapshot with clones */
8409 1940 : snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id);
8410 1940 : if (snapshot_entry != NULL) {
8411 243 : if (snapshot_entry->clone_count > 1) {
8412 30 : SPDK_ERRLOG("Cannot remove snapshot with more than one clone\n");
8413 30 : return -EBUSY;
8414 213 : } else if (snapshot_entry->clone_count == 1) {
8415 88 : has_one_clone = true;
8416 : }
8417 : }
8418 :
8419 : /* Check if someone has this blob open (besides this delete context):
8420 : * - open_ref = 1 - only this context opened blob, so it is ok to remove it
8421 : * - open_ref <= 2 && has_one_clone = true - clone is holding snapshot
8422 : * and that is ok, because we will update it accordingly */
8423 1910 : if (blob->open_ref <= 2 && has_one_clone) {
8424 88 : clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
8425 88 : assert(clone_entry != NULL);
8426 88 : clone = blob_lookup(blob->bs, clone_entry->id);
8427 :
8428 88 : if (blob->open_ref == 2 && clone == NULL) {
8429 : /* Clone is closed and someone else opened this blob */
8430 0 : SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
8431 0 : return -EBUSY;
8432 : }
8433 :
8434 88 : *update_clone = true;
8435 88 : return 0;
8436 : }
8437 :
8438 1822 : if (blob->open_ref > 1) {
8439 20 : SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
8440 20 : return -EBUSY;
8441 : }
8442 :
8443 1802 : assert(has_one_clone == false);
8444 1802 : *update_clone = false;
8445 1802 : return 0;
8446 : }
8447 :
8448 : static void
8449 0 : bs_delete_enomem_close_cpl(void *cb_arg, int bserrno)
8450 : {
8451 0 : spdk_bs_sequence_t *seq = cb_arg;
8452 :
8453 0 : bs_sequence_finish(seq, -ENOMEM);
8454 0 : }
8455 :
8456 : static void
8457 1953 : bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
8458 : {
8459 1953 : spdk_bs_sequence_t *seq = cb_arg;
8460 : struct delete_snapshot_ctx *ctx;
8461 1953 : bool update_clone = false;
8462 :
8463 1953 : if (bserrno != 0) {
8464 13 : bs_sequence_finish(seq, bserrno);
8465 63 : return;
8466 : }
8467 :
8468 1940 : blob_verify_md_op(blob);
8469 :
8470 1940 : ctx = calloc(1, sizeof(*ctx));
8471 1940 : if (ctx == NULL) {
8472 0 : spdk_blob_close(blob, bs_delete_enomem_close_cpl, seq);
8473 0 : return;
8474 : }
8475 :
8476 1940 : ctx->snapshot = blob;
8477 1940 : ctx->cb_fn = bs_delete_blob_finish;
8478 1940 : ctx->cb_arg = seq;
8479 :
8480 : /* Check if blob can be removed and if it is a snapshot with clone on top of it */
8481 1940 : ctx->bserrno = bs_is_blob_deletable(blob, &update_clone);
8482 1940 : if (ctx->bserrno) {
8483 50 : spdk_blob_close(blob, delete_blob_cleanup_finish, ctx);
8484 50 : return;
8485 : }
8486 :
8487 1890 : if (blob->locked_operation_in_progress) {
8488 0 : SPDK_DEBUGLOG(blob, "Cannot remove blob - another operation in progress\n");
8489 0 : ctx->bserrno = -EBUSY;
8490 0 : spdk_blob_close(blob, delete_blob_cleanup_finish, ctx);
8491 0 : return;
8492 : }
8493 :
8494 1890 : blob->locked_operation_in_progress = true;
8495 :
8496 : /*
8497 : * Remove the blob from the blob_store list now, to ensure it does not
8498 : * get returned after this point by blob_lookup().
8499 : */
8500 1890 : spdk_bit_array_clear(blob->bs->open_blobids, blob->id);
8501 1890 : RB_REMOVE(spdk_blob_tree, &blob->bs->open_blobs, blob);
8502 :
8503 1890 : if (update_clone) {
8504 88 : ctx->page = spdk_zmalloc(blob->bs->md_page_size, 0, NULL, SPDK_ENV_NUMA_ID_ANY,
8505 : SPDK_MALLOC_DMA);
8506 88 : if (!ctx->page) {
8507 0 : ctx->bserrno = -ENOMEM;
8508 0 : spdk_blob_close(blob, delete_blob_cleanup_finish, ctx);
8509 0 : return;
8510 : }
8511 : /* This blob is a snapshot with active clone - update clone first */
8512 88 : update_clone_on_snapshot_deletion(blob, ctx);
8513 : } else {
8514 : /* This blob does not have any clones - just remove it */
8515 1802 : bs_blob_list_remove(blob);
8516 1802 : bs_delete_blob_finish(seq, blob, 0);
8517 1802 : free(ctx);
8518 : }
8519 : }
8520 :
8521 : void
8522 1953 : spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
8523 : spdk_blob_op_complete cb_fn, void *cb_arg)
8524 : {
8525 : struct spdk_bs_cpl cpl;
8526 : spdk_bs_sequence_t *seq;
8527 :
8528 1953 : SPDK_DEBUGLOG(blob, "Deleting blob 0x%" PRIx64 "\n", blobid);
8529 :
8530 1953 : assert(spdk_get_thread() == bs->md_thread);
8531 :
8532 1953 : cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
8533 1953 : cpl.u.blob_basic.cb_fn = cb_fn;
8534 1953 : cpl.u.blob_basic.cb_arg = cb_arg;
8535 :
8536 1953 : seq = bs_sequence_start_bs(bs->md_channel, &cpl);
8537 1953 : if (!seq) {
8538 0 : cb_fn(cb_arg, -ENOMEM);
8539 0 : return;
8540 : }
8541 :
8542 1953 : spdk_bs_open_blob(bs, blobid, bs_delete_open_cpl, seq);
8543 : }
8544 :
8545 : /* END spdk_bs_delete_blob */
8546 :
8547 : /* START spdk_bs_open_blob */
8548 :
8549 : static void
8550 4341 : bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
8551 : {
8552 4341 : struct spdk_blob *blob = cb_arg;
8553 : struct spdk_blob *existing;
8554 :
8555 4341 : if (bserrno != 0) {
8556 82 : blob_free(blob);
8557 82 : seq->cpl.u.blob_handle.blob = NULL;
8558 82 : bs_sequence_finish(seq, bserrno);
8559 82 : return;
8560 : }
8561 :
8562 4259 : existing = blob_lookup(blob->bs, blob->id);
8563 4259 : if (existing) {
8564 5 : blob_free(blob);
8565 5 : existing->open_ref++;
8566 5 : seq->cpl.u.blob_handle.blob = existing;
8567 5 : bs_sequence_finish(seq, 0);
8568 5 : return;
8569 : }
8570 :
8571 4254 : blob->open_ref++;
8572 :
8573 4254 : spdk_bit_array_set(blob->bs->open_blobids, blob->id);
8574 4254 : RB_INSERT(spdk_blob_tree, &blob->bs->open_blobs, blob);
8575 :
8576 4254 : bs_sequence_finish(seq, bserrno);
8577 : }
8578 :
8579 : static inline void
8580 5 : blob_open_opts_copy(const struct spdk_blob_open_opts *src, struct spdk_blob_open_opts *dst)
8581 : {
8582 : #define FIELD_OK(field) \
8583 : offsetof(struct spdk_blob_open_opts, field) + sizeof(src->field) <= src->opts_size
8584 :
8585 : #define SET_FIELD(field) \
8586 : if (FIELD_OK(field)) { \
8587 : dst->field = src->field; \
8588 : } \
8589 :
8590 5 : SET_FIELD(clear_method);
8591 5 : SET_FIELD(esnap_ctx);
8592 :
8593 5 : dst->opts_size = src->opts_size;
8594 :
8595 : /* You should not remove this statement, but need to update the assert statement
8596 : * if you add a new field, and also add a corresponding SET_FIELD statement */
8597 : SPDK_STATIC_ASSERT(sizeof(struct spdk_blob_open_opts) == 24, "Incorrect size");
8598 :
8599 : #undef FIELD_OK
8600 : #undef SET_FIELD
8601 5 : }
8602 :
8603 : static void
8604 5353 : bs_open_blob(struct spdk_blob_store *bs,
8605 : spdk_blob_id blobid,
8606 : struct spdk_blob_open_opts *opts,
8607 : spdk_blob_op_with_handle_complete cb_fn,
8608 : void *cb_arg)
8609 : {
8610 : struct spdk_blob *blob;
8611 : struct spdk_bs_cpl cpl;
8612 : struct spdk_blob_open_opts opts_local;
8613 : spdk_bs_sequence_t *seq;
8614 : uint32_t page_num;
8615 :
8616 5353 : SPDK_DEBUGLOG(blob, "Opening blob 0x%" PRIx64 "\n", blobid);
8617 5353 : assert(spdk_get_thread() == bs->md_thread);
8618 :
8619 5353 : page_num = bs_blobid_to_page(blobid);
8620 5353 : if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
8621 : /* Invalid blobid */
8622 60 : cb_fn(cb_arg, NULL, -ENOENT);
8623 1012 : return;
8624 : }
8625 :
8626 5293 : blob = blob_lookup(bs, blobid);
8627 5293 : if (blob) {
8628 952 : blob->open_ref++;
8629 952 : cb_fn(cb_arg, blob, 0);
8630 952 : return;
8631 : }
8632 :
8633 4341 : blob = blob_alloc(bs, blobid);
8634 4341 : if (!blob) {
8635 0 : cb_fn(cb_arg, NULL, -ENOMEM);
8636 0 : return;
8637 : }
8638 :
8639 4341 : spdk_blob_open_opts_init(&opts_local, sizeof(opts_local));
8640 4341 : if (opts) {
8641 5 : blob_open_opts_copy(opts, &opts_local);
8642 : }
8643 :
8644 4341 : blob->clear_method = opts_local.clear_method;
8645 :
8646 4341 : cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
8647 4341 : cpl.u.blob_handle.cb_fn = cb_fn;
8648 4341 : cpl.u.blob_handle.cb_arg = cb_arg;
8649 4341 : cpl.u.blob_handle.blob = blob;
8650 4341 : cpl.u.blob_handle.esnap_ctx = opts_local.esnap_ctx;
8651 :
8652 4341 : seq = bs_sequence_start_bs(bs->md_channel, &cpl);
8653 4341 : if (!seq) {
8654 0 : blob_free(blob);
8655 0 : cb_fn(cb_arg, NULL, -ENOMEM);
8656 0 : return;
8657 : }
8658 :
8659 4341 : blob_load(seq, blob, bs_open_blob_cpl, blob);
8660 : }
8661 :
8662 : void
8663 5348 : spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
8664 : spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
8665 : {
8666 5348 : bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg);
8667 5348 : }
8668 :
8669 : void
8670 5 : spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid,
8671 : struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
8672 : {
8673 5 : bs_open_blob(bs, blobid, opts, cb_fn, cb_arg);
8674 5 : }
8675 :
8676 : /* END spdk_bs_open_blob */
8677 :
8678 : /* START spdk_blob_set_read_only */
8679 : int
8680 296 : spdk_blob_set_read_only(struct spdk_blob *blob)
8681 : {
8682 296 : blob_verify_md_op(blob);
8683 :
8684 296 : blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
8685 :
8686 296 : blob->state = SPDK_BLOB_STATE_DIRTY;
8687 296 : return 0;
8688 : }
8689 : /* END spdk_blob_set_read_only */
8690 :
8691 : /* START spdk_blob_sync_md */
8692 :
8693 : static void
8694 1927 : blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
8695 : {
8696 1927 : struct spdk_blob *blob = cb_arg;
8697 :
8698 1927 : if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
8699 497 : blob->data_ro = true;
8700 497 : blob->md_ro = true;
8701 : }
8702 :
8703 1927 : bs_sequence_finish(seq, bserrno);
8704 1927 : }
8705 :
8706 : static void
8707 1927 : blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
8708 : {
8709 : struct spdk_bs_cpl cpl;
8710 : spdk_bs_sequence_t *seq;
8711 :
8712 1927 : cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
8713 1927 : cpl.u.blob_basic.cb_fn = cb_fn;
8714 1927 : cpl.u.blob_basic.cb_arg = cb_arg;
8715 :
8716 1927 : seq = bs_sequence_start_bs(blob->bs->md_channel, &cpl);
8717 1927 : if (!seq) {
8718 0 : cb_fn(cb_arg, -ENOMEM);
8719 0 : return;
8720 : }
8721 :
8722 1927 : blob_persist(seq, blob, blob_sync_md_cpl, blob);
8723 : }
8724 :
8725 : void
8726 1370 : spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
8727 : {
8728 1370 : blob_verify_md_op(blob);
8729 :
8730 1370 : SPDK_DEBUGLOG(blob, "Syncing blob 0x%" PRIx64 "\n", blob->id);
8731 :
8732 1370 : if (blob->md_ro) {
8733 5 : assert(blob->state == SPDK_BLOB_STATE_CLEAN);
8734 5 : cb_fn(cb_arg, 0);
8735 5 : return;
8736 : }
8737 :
8738 1365 : blob_sync_md(blob, cb_fn, cb_arg);
8739 : }
8740 :
8741 : /* END spdk_blob_sync_md */
8742 :
8743 : struct spdk_blob_cluster_op_ctx {
8744 : struct spdk_thread *thread;
8745 : struct spdk_blob *blob;
8746 : uint32_t cluster_num; /* cluster index in blob */
8747 : uint32_t cluster; /* cluster on disk */
8748 : uint32_t extent_page; /* extent page on disk */
8749 : struct spdk_blob_md_page *page; /* preallocated extent page */
8750 : int rc;
8751 : spdk_blob_op_complete cb_fn;
8752 : void *cb_arg;
8753 : };
8754 :
8755 : static void
8756 1105 : blob_op_cluster_msg_cpl(void *arg)
8757 : {
8758 1105 : struct spdk_blob_cluster_op_ctx *ctx = arg;
8759 :
8760 1105 : ctx->cb_fn(ctx->cb_arg, ctx->rc);
8761 1105 : free(ctx);
8762 1105 : }
8763 :
8764 : static void
8765 1061 : blob_op_cluster_msg_cb(void *arg, int bserrno)
8766 : {
8767 1061 : struct spdk_blob_cluster_op_ctx *ctx = arg;
8768 :
8769 1061 : ctx->rc = bserrno;
8770 1061 : spdk_thread_send_msg(ctx->thread, blob_op_cluster_msg_cpl, ctx);
8771 1061 : }
8772 :
8773 : static void
8774 126 : blob_insert_new_ep_cb(void *arg, int bserrno)
8775 : {
8776 126 : struct spdk_blob_cluster_op_ctx *ctx = arg;
8777 : uint32_t *extent_page;
8778 :
8779 126 : extent_page = bs_cluster_to_extent_page(ctx->blob, ctx->cluster_num);
8780 126 : *extent_page = ctx->extent_page;
8781 126 : ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
8782 126 : blob_sync_md(ctx->blob, blob_op_cluster_msg_cb, ctx);
8783 126 : }
8784 :
8785 : struct spdk_blob_write_extent_page_ctx {
8786 : struct spdk_blob_store *bs;
8787 :
8788 : uint32_t extent;
8789 : struct spdk_blob_md_page *page;
8790 : };
8791 :
8792 : static void
8793 39 : blob_free_cluster_msg_cb(void *arg, int bserrno)
8794 : {
8795 39 : struct spdk_blob_cluster_op_ctx *ctx = arg;
8796 :
8797 39 : spdk_spin_lock(&ctx->blob->bs->used_lock);
8798 39 : bs_release_cluster(ctx->blob->bs, ctx->cluster);
8799 39 : spdk_spin_unlock(&ctx->blob->bs->used_lock);
8800 :
8801 39 : ctx->rc = bserrno;
8802 39 : spdk_thread_send_msg(ctx->thread, blob_op_cluster_msg_cpl, ctx);
8803 39 : }
8804 :
8805 : static void
8806 39 : blob_free_cluster_update_ep_cb(void *arg, int bserrno)
8807 : {
8808 39 : struct spdk_blob_cluster_op_ctx *ctx = arg;
8809 :
8810 39 : if (bserrno != 0 || ctx->blob->bs->clean == 0) {
8811 39 : blob_free_cluster_msg_cb(ctx, bserrno);
8812 39 : return;
8813 : }
8814 :
8815 0 : ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
8816 0 : blob_sync_md(ctx->blob, blob_free_cluster_msg_cb, ctx);
8817 : }
8818 :
8819 : static void
8820 0 : blob_free_cluster_free_ep_cb(void *arg, int bserrno)
8821 : {
8822 0 : struct spdk_blob_cluster_op_ctx *ctx = arg;
8823 :
8824 0 : spdk_spin_lock(&ctx->blob->bs->used_lock);
8825 0 : assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true);
8826 0 : bs_release_md_page(ctx->blob->bs, ctx->extent_page);
8827 0 : spdk_spin_unlock(&ctx->blob->bs->used_lock);
8828 0 : ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
8829 0 : blob_sync_md(ctx->blob, blob_free_cluster_msg_cb, ctx);
8830 0 : }
8831 :
8832 : static void
8833 657 : blob_persist_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
8834 : {
8835 657 : struct spdk_blob_write_extent_page_ctx *ctx = cb_arg;
8836 :
8837 657 : free(ctx);
8838 657 : bs_sequence_finish(seq, bserrno);
8839 657 : }
8840 :
8841 : static void
8842 657 : blob_write_extent_page_ready(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
8843 : {
8844 657 : struct spdk_blob_write_extent_page_ctx *ctx = cb_arg;
8845 :
8846 657 : if (bserrno != 0) {
8847 0 : blob_persist_extent_page_cpl(seq, ctx, bserrno);
8848 0 : return;
8849 : }
8850 657 : bs_sequence_write_dev(seq, ctx->page, bs_md_page_to_lba(ctx->bs, ctx->extent),
8851 657 : bs_byte_to_lba(ctx->bs, ctx->bs->md_page_size),
8852 : blob_persist_extent_page_cpl, ctx);
8853 : }
8854 :
8855 : static void
8856 657 : blob_write_extent_page(struct spdk_blob *blob, uint32_t extent, uint64_t cluster_num,
8857 : struct spdk_blob_md_page *page, spdk_blob_op_complete cb_fn, void *cb_arg)
8858 : {
8859 : struct spdk_blob_write_extent_page_ctx *ctx;
8860 : spdk_bs_sequence_t *seq;
8861 : struct spdk_bs_cpl cpl;
8862 :
8863 657 : ctx = calloc(1, sizeof(*ctx));
8864 657 : if (!ctx) {
8865 0 : cb_fn(cb_arg, -ENOMEM);
8866 0 : return;
8867 : }
8868 657 : ctx->bs = blob->bs;
8869 657 : ctx->extent = extent;
8870 657 : ctx->page = page;
8871 :
8872 657 : cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
8873 657 : cpl.u.blob_basic.cb_fn = cb_fn;
8874 657 : cpl.u.blob_basic.cb_arg = cb_arg;
8875 :
8876 657 : seq = bs_sequence_start_bs(blob->bs->md_channel, &cpl);
8877 657 : if (!seq) {
8878 0 : free(ctx);
8879 0 : cb_fn(cb_arg, -ENOMEM);
8880 0 : return;
8881 : }
8882 :
8883 657 : assert(page);
8884 657 : page->next = SPDK_INVALID_MD_PAGE;
8885 657 : page->id = blob->id;
8886 657 : page->sequence_num = 0;
8887 :
8888 657 : blob_serialize_extent_page(blob, cluster_num, page);
8889 :
8890 657 : page->crc = blob_md_page_calc_crc(page);
8891 :
8892 657 : assert(spdk_bit_array_get(blob->bs->used_md_pages, extent) == true);
8893 :
8894 657 : bs_mark_dirty(seq, blob->bs, blob_write_extent_page_ready, ctx);
8895 : }
8896 :
8897 : static void
8898 1030 : blob_insert_cluster_msg(void *arg)
8899 : {
8900 1030 : struct spdk_blob_cluster_op_ctx *ctx = arg;
8901 : uint32_t *extent_page;
8902 :
8903 1030 : ctx->rc = blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
8904 1030 : if (ctx->rc != 0) {
8905 5 : spdk_thread_send_msg(ctx->thread, blob_op_cluster_msg_cpl, ctx);
8906 5 : return;
8907 : }
8908 :
8909 1025 : if (ctx->blob->use_extent_table == false) {
8910 : /* Extent table is not used, proceed with sync of md that will only use extents_rle. */
8911 410 : ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
8912 410 : blob_sync_md(ctx->blob, blob_op_cluster_msg_cb, ctx);
8913 410 : return;
8914 : }
8915 :
8916 615 : extent_page = bs_cluster_to_extent_page(ctx->blob, ctx->cluster_num);
8917 615 : if (*extent_page == 0) {
8918 : /* Extent page requires allocation.
8919 : * It was already claimed in the used_md_pages map and placed in ctx. */
8920 126 : assert(ctx->extent_page != 0);
8921 126 : assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true);
8922 126 : blob_write_extent_page(ctx->blob, ctx->extent_page, ctx->cluster_num, ctx->page,
8923 : blob_insert_new_ep_cb, ctx);
8924 : } else {
8925 : /* It is possible for original thread to allocate extent page for
8926 : * different cluster in the same extent page. In such case proceed with
8927 : * updating the existing extent page, but release the additional one. */
8928 489 : if (ctx->extent_page != 0) {
8929 0 : spdk_spin_lock(&ctx->blob->bs->used_lock);
8930 0 : assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true);
8931 0 : bs_release_md_page(ctx->blob->bs, ctx->extent_page);
8932 0 : spdk_spin_unlock(&ctx->blob->bs->used_lock);
8933 0 : ctx->extent_page = 0;
8934 : }
8935 : /* Extent page already allocated.
8936 : * Every cluster allocation, requires just an update of single extent page. */
8937 489 : blob_write_extent_page(ctx->blob, *extent_page, ctx->cluster_num, ctx->page,
8938 : blob_op_cluster_msg_cb, ctx);
8939 : }
8940 : }
8941 :
8942 : static void
8943 1030 : blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
8944 : uint64_t cluster, uint32_t extent_page, struct spdk_blob_md_page *page,
8945 : spdk_blob_op_complete cb_fn, void *cb_arg)
8946 : {
8947 : struct spdk_blob_cluster_op_ctx *ctx;
8948 :
8949 1030 : ctx = calloc(1, sizeof(*ctx));
8950 1030 : if (ctx == NULL) {
8951 0 : cb_fn(cb_arg, -ENOMEM);
8952 0 : return;
8953 : }
8954 :
8955 1030 : ctx->thread = spdk_get_thread();
8956 1030 : ctx->blob = blob;
8957 1030 : ctx->cluster_num = cluster_num;
8958 1030 : ctx->cluster = cluster;
8959 1030 : ctx->extent_page = extent_page;
8960 1030 : ctx->page = page;
8961 1030 : ctx->cb_fn = cb_fn;
8962 1030 : ctx->cb_arg = cb_arg;
8963 :
8964 1030 : spdk_thread_send_msg(blob->bs->md_thread, blob_insert_cluster_msg, ctx);
8965 : }
8966 :
8967 : static void
8968 75 : blob_free_cluster_msg(void *arg)
8969 : {
8970 75 : struct spdk_blob_cluster_op_ctx *ctx = arg;
8971 : uint32_t *extent_page;
8972 : uint32_t start_cluster_idx;
8973 75 : bool free_extent_page = true;
8974 : size_t i;
8975 :
8976 75 : ctx->cluster = bs_lba_to_cluster(ctx->blob->bs, ctx->blob->active.clusters[ctx->cluster_num]);
8977 :
8978 : /* There were concurrent unmaps to the same cluster, only release the cluster on the first one */
8979 75 : if (ctx->cluster == 0) {
8980 10 : blob_op_cluster_msg_cb(ctx, 0);
8981 10 : return;
8982 : }
8983 :
8984 65 : ctx->blob->active.clusters[ctx->cluster_num] = 0;
8985 65 : if (ctx->cluster != 0) {
8986 65 : ctx->blob->active.num_allocated_clusters--;
8987 : }
8988 :
8989 65 : if (ctx->blob->use_extent_table == false) {
8990 : /* Extent table is not used, proceed with sync of md that will only use extents_rle. */
8991 26 : spdk_spin_lock(&ctx->blob->bs->used_lock);
8992 26 : bs_release_cluster(ctx->blob->bs, ctx->cluster);
8993 26 : spdk_spin_unlock(&ctx->blob->bs->used_lock);
8994 26 : ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
8995 26 : blob_sync_md(ctx->blob, blob_op_cluster_msg_cb, ctx);
8996 26 : return;
8997 : }
8998 :
8999 39 : extent_page = bs_cluster_to_extent_page(ctx->blob, ctx->cluster_num);
9000 :
9001 : /* There shouldn't be parallel release operations on same cluster */
9002 39 : assert(*extent_page == ctx->extent_page);
9003 :
9004 39 : start_cluster_idx = (ctx->cluster_num / SPDK_EXTENTS_PER_EP) * SPDK_EXTENTS_PER_EP;
9005 72 : for (i = 0; i < SPDK_EXTENTS_PER_EP; ++i) {
9006 72 : if (ctx->blob->active.clusters[start_cluster_idx + i] != 0) {
9007 39 : free_extent_page = false;
9008 39 : break;
9009 : }
9010 : }
9011 :
9012 39 : if (free_extent_page) {
9013 0 : assert(ctx->extent_page != 0);
9014 0 : assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true);
9015 0 : ctx->blob->active.extent_pages[bs_cluster_to_extent_table_id(ctx->cluster_num)] = 0;
9016 0 : blob_write_extent_page(ctx->blob, ctx->extent_page, ctx->cluster_num, ctx->page,
9017 : blob_free_cluster_free_ep_cb, ctx);
9018 : } else {
9019 39 : blob_write_extent_page(ctx->blob, *extent_page, ctx->cluster_num, ctx->page,
9020 : blob_free_cluster_update_ep_cb, ctx);
9021 : }
9022 : }
9023 :
9024 :
9025 : static void
9026 75 : blob_free_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num, uint32_t extent_page,
9027 : struct spdk_blob_md_page *page, spdk_blob_op_complete cb_fn, void *cb_arg)
9028 : {
9029 : struct spdk_blob_cluster_op_ctx *ctx;
9030 :
9031 75 : ctx = calloc(1, sizeof(*ctx));
9032 75 : if (ctx == NULL) {
9033 0 : cb_fn(cb_arg, -ENOMEM);
9034 0 : return;
9035 : }
9036 :
9037 75 : ctx->thread = spdk_get_thread();
9038 75 : ctx->blob = blob;
9039 75 : ctx->cluster_num = cluster_num;
9040 75 : ctx->extent_page = extent_page;
9041 75 : ctx->page = page;
9042 75 : ctx->cb_fn = cb_fn;
9043 75 : ctx->cb_arg = cb_arg;
9044 :
9045 75 : spdk_thread_send_msg(blob->bs->md_thread, blob_free_cluster_msg, ctx);
9046 : }
9047 :
9048 : /* START spdk_blob_close */
9049 :
9050 : static void
9051 5211 : blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
9052 : {
9053 5211 : struct spdk_blob *blob = cb_arg;
9054 :
9055 5211 : if (bserrno == 0) {
9056 5211 : blob->open_ref--;
9057 5211 : if (blob->open_ref == 0) {
9058 : /*
9059 : * Blobs with active.num_pages == 0 are deleted blobs.
9060 : * these blobs are removed from the blob_store list
9061 : * when the deletion process starts - so don't try to
9062 : * remove them again.
9063 : */
9064 4254 : if (blob->active.num_pages > 0) {
9065 2392 : spdk_bit_array_clear(blob->bs->open_blobids, blob->id);
9066 2392 : RB_REMOVE(spdk_blob_tree, &blob->bs->open_blobs, blob);
9067 : }
9068 4254 : blob_free(blob);
9069 : }
9070 : }
9071 :
9072 5211 : bs_sequence_finish(seq, bserrno);
9073 5211 : }
9074 :
9075 : static void
9076 140 : blob_close_esnap_done(void *cb_arg, struct spdk_blob *blob, int bserrno)
9077 : {
9078 140 : spdk_bs_sequence_t *seq = cb_arg;
9079 :
9080 140 : if (bserrno != 0) {
9081 0 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64 ": close failed with error %d\n",
9082 : blob->id, bserrno);
9083 0 : bs_sequence_finish(seq, bserrno);
9084 0 : return;
9085 : }
9086 :
9087 140 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64 ": closed, syncing metadata on thread %s\n",
9088 : blob->id, spdk_thread_get_name(spdk_get_thread()));
9089 :
9090 : /* Sync metadata */
9091 140 : blob_persist(seq, blob, blob_close_cpl, blob);
9092 : }
9093 :
9094 : void
9095 5211 : spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
9096 : {
9097 : struct spdk_bs_cpl cpl;
9098 : spdk_bs_sequence_t *seq;
9099 :
9100 5211 : blob_verify_md_op(blob);
9101 :
9102 5211 : SPDK_DEBUGLOG(blob, "Closing blob 0x%" PRIx64 "\n", blob->id);
9103 :
9104 5211 : if (blob->open_ref == 0) {
9105 0 : cb_fn(cb_arg, -EBADF);
9106 140 : return;
9107 : }
9108 :
9109 5211 : cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
9110 5211 : cpl.u.blob_basic.cb_fn = cb_fn;
9111 5211 : cpl.u.blob_basic.cb_arg = cb_arg;
9112 :
9113 5211 : seq = bs_sequence_start_bs(blob->bs->md_channel, &cpl);
9114 5211 : if (!seq) {
9115 0 : cb_fn(cb_arg, -ENOMEM);
9116 0 : return;
9117 : }
9118 :
9119 5211 : if (blob->open_ref == 1 && blob_is_esnap_clone(blob)) {
9120 140 : blob_esnap_destroy_bs_dev_channels(blob, false, blob_close_esnap_done, seq);
9121 140 : return;
9122 : }
9123 :
9124 : /* Sync metadata */
9125 5071 : blob_persist(seq, blob, blob_close_cpl, blob);
9126 : }
9127 :
9128 : /* END spdk_blob_close */
9129 :
9130 276 : struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
9131 : {
9132 276 : return spdk_get_io_channel(bs);
9133 : }
9134 :
9135 : void
9136 276 : spdk_bs_free_io_channel(struct spdk_io_channel *channel)
9137 : {
9138 276 : blob_esnap_destroy_bs_channel(spdk_io_channel_get_ctx(channel));
9139 276 : spdk_put_io_channel(channel);
9140 276 : }
9141 :
9142 : void
9143 140 : spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
9144 : uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
9145 : {
9146 140 : blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
9147 : SPDK_BLOB_UNMAP);
9148 140 : }
9149 :
9150 : void
9151 60 : spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
9152 : uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
9153 : {
9154 60 : blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
9155 : SPDK_BLOB_WRITE_ZEROES);
9156 60 : }
9157 :
9158 : void
9159 27349 : spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
9160 : void *payload, uint64_t offset, uint64_t length,
9161 : spdk_blob_op_complete cb_fn, void *cb_arg)
9162 : {
9163 27349 : blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
9164 : SPDK_BLOB_WRITE);
9165 27349 : }
9166 :
9167 : void
9168 26143 : spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
9169 : void *payload, uint64_t offset, uint64_t length,
9170 : spdk_blob_op_complete cb_fn, void *cb_arg)
9171 : {
9172 26143 : blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
9173 : SPDK_BLOB_READ);
9174 26143 : }
9175 :
9176 : void
9177 175 : spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
9178 : struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
9179 : spdk_blob_op_complete cb_fn, void *cb_arg)
9180 : {
9181 175 : blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false, NULL);
9182 175 : }
9183 :
9184 : void
9185 1655 : spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
9186 : struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
9187 : spdk_blob_op_complete cb_fn, void *cb_arg)
9188 : {
9189 1655 : blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true, NULL);
9190 1655 : }
9191 :
9192 : void
9193 260 : spdk_blob_io_writev_ext(struct spdk_blob *blob, struct spdk_io_channel *channel,
9194 : struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
9195 : spdk_blob_op_complete cb_fn, void *cb_arg, struct spdk_blob_ext_io_opts *io_opts)
9196 : {
9197 260 : blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false,
9198 : io_opts);
9199 260 : }
9200 :
9201 : void
9202 2105 : spdk_blob_io_readv_ext(struct spdk_blob *blob, struct spdk_io_channel *channel,
9203 : struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
9204 : spdk_blob_op_complete cb_fn, void *cb_arg, struct spdk_blob_ext_io_opts *io_opts)
9205 : {
9206 2105 : blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true,
9207 : io_opts);
9208 2105 : }
9209 :
9210 : struct spdk_bs_iter_ctx {
9211 : int64_t page_num;
9212 : struct spdk_blob_store *bs;
9213 :
9214 : spdk_blob_op_with_handle_complete cb_fn;
9215 : void *cb_arg;
9216 : };
9217 :
9218 : static void
9219 1460 : bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
9220 : {
9221 1460 : struct spdk_bs_iter_ctx *ctx = cb_arg;
9222 1460 : struct spdk_blob_store *bs = ctx->bs;
9223 : spdk_blob_id id;
9224 :
9225 1460 : if (bserrno == 0) {
9226 557 : ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
9227 557 : free(ctx);
9228 557 : return;
9229 : }
9230 :
9231 903 : ctx->page_num++;
9232 903 : ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
9233 903 : if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
9234 336 : ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
9235 336 : free(ctx);
9236 336 : return;
9237 : }
9238 :
9239 567 : id = bs_page_to_blobid(ctx->page_num);
9240 :
9241 567 : spdk_bs_open_blob(bs, id, bs_iter_cpl, ctx);
9242 : }
9243 :
9244 : void
9245 366 : spdk_bs_iter_first(struct spdk_blob_store *bs,
9246 : spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
9247 : {
9248 : struct spdk_bs_iter_ctx *ctx;
9249 :
9250 366 : ctx = calloc(1, sizeof(*ctx));
9251 366 : if (!ctx) {
9252 0 : cb_fn(cb_arg, NULL, -ENOMEM);
9253 0 : return;
9254 : }
9255 :
9256 366 : ctx->page_num = -1;
9257 366 : ctx->bs = bs;
9258 366 : ctx->cb_fn = cb_fn;
9259 366 : ctx->cb_arg = cb_arg;
9260 :
9261 366 : bs_iter_cpl(ctx, NULL, -1);
9262 : }
9263 :
9264 : static void
9265 527 : bs_iter_close_cpl(void *cb_arg, int bserrno)
9266 : {
9267 527 : struct spdk_bs_iter_ctx *ctx = cb_arg;
9268 :
9269 527 : bs_iter_cpl(ctx, NULL, -1);
9270 527 : }
9271 :
9272 : void
9273 527 : spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
9274 : spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
9275 : {
9276 : struct spdk_bs_iter_ctx *ctx;
9277 :
9278 527 : assert(blob != NULL);
9279 :
9280 527 : ctx = calloc(1, sizeof(*ctx));
9281 527 : if (!ctx) {
9282 0 : cb_fn(cb_arg, NULL, -ENOMEM);
9283 0 : return;
9284 : }
9285 :
9286 527 : ctx->page_num = bs_blobid_to_page(blob->id);
9287 527 : ctx->bs = bs;
9288 527 : ctx->cb_fn = cb_fn;
9289 527 : ctx->cb_arg = cb_arg;
9290 :
9291 : /* Close the existing blob */
9292 527 : spdk_blob_close(blob, bs_iter_close_cpl, ctx);
9293 : }
9294 :
9295 : static int
9296 1178 : blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
9297 : uint16_t value_len, bool internal)
9298 : {
9299 : struct spdk_xattr_tailq *xattrs;
9300 : struct spdk_xattr *xattr;
9301 : size_t desc_size;
9302 : void *tmp;
9303 :
9304 1178 : blob_verify_md_op(blob);
9305 :
9306 1178 : if (blob->md_ro) {
9307 5 : return -EPERM;
9308 : }
9309 :
9310 1173 : desc_size = sizeof(struct spdk_blob_md_descriptor_xattr) + strlen(name) + value_len;
9311 1173 : if (desc_size > SPDK_BS_MAX_DESC_SIZE) {
9312 5 : SPDK_DEBUGLOG(blob, "Xattr '%s' of size %zu does not fix into single page %zu\n", name,
9313 : desc_size, SPDK_BS_MAX_DESC_SIZE);
9314 5 : return -ENOMEM;
9315 : }
9316 :
9317 1168 : if (internal) {
9318 917 : xattrs = &blob->xattrs_internal;
9319 917 : blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
9320 : } else {
9321 251 : xattrs = &blob->xattrs;
9322 : }
9323 :
9324 1438 : TAILQ_FOREACH(xattr, xattrs, link) {
9325 402 : if (!strcmp(name, xattr->name)) {
9326 132 : tmp = malloc(value_len);
9327 132 : if (!tmp) {
9328 0 : return -ENOMEM;
9329 : }
9330 :
9331 132 : free(xattr->value);
9332 132 : xattr->value_len = value_len;
9333 132 : xattr->value = tmp;
9334 132 : memcpy(xattr->value, value, value_len);
9335 :
9336 132 : blob->state = SPDK_BLOB_STATE_DIRTY;
9337 :
9338 132 : return 0;
9339 : }
9340 : }
9341 :
9342 1036 : xattr = calloc(1, sizeof(*xattr));
9343 1036 : if (!xattr) {
9344 0 : return -ENOMEM;
9345 : }
9346 :
9347 1036 : xattr->name = strdup(name);
9348 1036 : if (!xattr->name) {
9349 0 : free(xattr);
9350 0 : return -ENOMEM;
9351 : }
9352 :
9353 1036 : xattr->value_len = value_len;
9354 1036 : xattr->value = malloc(value_len);
9355 1036 : if (!xattr->value) {
9356 0 : free(xattr->name);
9357 0 : free(xattr);
9358 0 : return -ENOMEM;
9359 : }
9360 1036 : memcpy(xattr->value, value, value_len);
9361 1036 : TAILQ_INSERT_TAIL(xattrs, xattr, link);
9362 :
9363 1036 : blob->state = SPDK_BLOB_STATE_DIRTY;
9364 :
9365 1036 : return 0;
9366 : }
9367 :
9368 : int
9369 216 : spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
9370 : uint16_t value_len)
9371 : {
9372 216 : return blob_set_xattr(blob, name, value, value_len, false);
9373 : }
9374 :
9375 : static int
9376 511 : blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
9377 : {
9378 : struct spdk_xattr_tailq *xattrs;
9379 : struct spdk_xattr *xattr;
9380 :
9381 511 : blob_verify_md_op(blob);
9382 :
9383 511 : if (blob->md_ro) {
9384 5 : return -EPERM;
9385 : }
9386 506 : xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
9387 :
9388 521 : TAILQ_FOREACH(xattr, xattrs, link) {
9389 456 : if (!strcmp(name, xattr->name)) {
9390 441 : TAILQ_REMOVE(xattrs, xattr, link);
9391 441 : free(xattr->value);
9392 441 : free(xattr->name);
9393 441 : free(xattr);
9394 :
9395 441 : if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
9396 306 : blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
9397 : }
9398 441 : blob->state = SPDK_BLOB_STATE_DIRTY;
9399 :
9400 441 : return 0;
9401 : }
9402 : }
9403 :
9404 65 : return -ENOENT;
9405 : }
9406 :
9407 : int
9408 45 : spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
9409 : {
9410 45 : return blob_remove_xattr(blob, name, false);
9411 : }
9412 :
9413 : static int
9414 2852 : blob_get_xattr_value(struct spdk_blob *blob, const char *name,
9415 : const void **value, size_t *value_len, bool internal)
9416 : {
9417 : struct spdk_xattr *xattr;
9418 : struct spdk_xattr_tailq *xattrs;
9419 :
9420 2852 : xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
9421 :
9422 3636 : TAILQ_FOREACH(xattr, xattrs, link) {
9423 1728 : if (!strcmp(name, xattr->name)) {
9424 944 : *value = xattr->value;
9425 944 : *value_len = xattr->value_len;
9426 944 : return 0;
9427 : }
9428 : }
9429 1908 : return -ENOENT;
9430 : }
9431 :
9432 : int
9433 192 : spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
9434 : const void **value, size_t *value_len)
9435 : {
9436 192 : blob_verify_md_op(blob);
9437 :
9438 192 : return blob_get_xattr_value(blob, name, value, value_len, false);
9439 : }
9440 :
9441 : struct spdk_xattr_names {
9442 : uint32_t count;
9443 : const char *names[0];
9444 : };
9445 :
9446 : static int
9447 5 : blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
9448 : {
9449 : struct spdk_xattr *xattr;
9450 5 : int count = 0;
9451 :
9452 15 : TAILQ_FOREACH(xattr, xattrs, link) {
9453 10 : count++;
9454 : }
9455 :
9456 5 : *names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
9457 5 : if (*names == NULL) {
9458 0 : return -ENOMEM;
9459 : }
9460 :
9461 15 : TAILQ_FOREACH(xattr, xattrs, link) {
9462 10 : (*names)->names[(*names)->count++] = xattr->name;
9463 : }
9464 :
9465 5 : return 0;
9466 : }
9467 :
9468 : int
9469 5 : spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
9470 : {
9471 5 : blob_verify_md_op(blob);
9472 :
9473 5 : return blob_get_xattr_names(&blob->xattrs, names);
9474 : }
9475 :
9476 : uint32_t
9477 5 : spdk_xattr_names_get_count(struct spdk_xattr_names *names)
9478 : {
9479 5 : assert(names != NULL);
9480 :
9481 5 : return names->count;
9482 : }
9483 :
9484 : const char *
9485 10 : spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
9486 : {
9487 10 : if (index >= names->count) {
9488 0 : return NULL;
9489 : }
9490 :
9491 10 : return names->names[index];
9492 : }
9493 :
9494 : void
9495 5 : spdk_xattr_names_free(struct spdk_xattr_names *names)
9496 : {
9497 5 : free(names);
9498 5 : }
9499 :
9500 : struct spdk_bs_type
9501 2 : spdk_bs_get_bstype(struct spdk_blob_store *bs)
9502 : {
9503 2 : return bs->bstype;
9504 : }
9505 :
9506 : void
9507 0 : spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
9508 : {
9509 0 : memcpy(&bs->bstype, &bstype, sizeof(bstype));
9510 0 : }
9511 :
9512 : bool
9513 60 : spdk_blob_is_read_only(struct spdk_blob *blob)
9514 : {
9515 60 : assert(blob != NULL);
9516 60 : return (blob->data_ro || blob->md_ro);
9517 : }
9518 :
9519 : bool
9520 65 : spdk_blob_is_snapshot(struct spdk_blob *blob)
9521 : {
9522 : struct spdk_blob_list *snapshot_entry;
9523 :
9524 65 : assert(blob != NULL);
9525 :
9526 65 : snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id);
9527 65 : if (snapshot_entry == NULL) {
9528 35 : return false;
9529 : }
9530 :
9531 30 : return true;
9532 : }
9533 :
9534 : bool
9535 85 : spdk_blob_is_clone(struct spdk_blob *blob)
9536 : {
9537 85 : assert(blob != NULL);
9538 :
9539 85 : if (blob->parent_id != SPDK_BLOBID_INVALID &&
9540 65 : blob->parent_id != SPDK_BLOBID_EXTERNAL_SNAPSHOT) {
9541 50 : assert(spdk_blob_is_thin_provisioned(blob));
9542 50 : return true;
9543 : }
9544 :
9545 35 : return false;
9546 : }
9547 :
9548 : bool
9549 46657 : spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
9550 : {
9551 46657 : assert(blob != NULL);
9552 46657 : return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
9553 : }
9554 :
9555 : bool
9556 57607 : spdk_blob_is_esnap_clone(const struct spdk_blob *blob)
9557 : {
9558 57607 : return blob_is_esnap_clone(blob);
9559 : }
9560 :
9561 : static void
9562 4291 : blob_update_clear_method(struct spdk_blob *blob)
9563 : {
9564 : enum blob_clear_method stored_cm;
9565 :
9566 4291 : assert(blob != NULL);
9567 :
9568 : /* If BLOB_CLEAR_WITH_DEFAULT was passed in, use the setting stored
9569 : * in metadata previously. If something other than the default was
9570 : * specified, ignore stored value and used what was passed in.
9571 : */
9572 4291 : stored_cm = ((blob->md_ro_flags & SPDK_BLOB_CLEAR_METHOD) >> SPDK_BLOB_CLEAR_METHOD_SHIFT);
9573 :
9574 4291 : if (blob->clear_method == BLOB_CLEAR_WITH_DEFAULT) {
9575 4291 : blob->clear_method = stored_cm;
9576 0 : } else if (blob->clear_method != stored_cm) {
9577 0 : SPDK_WARNLOG("Using passed in clear method 0x%x instead of stored value of 0x%x\n",
9578 : blob->clear_method, stored_cm);
9579 : }
9580 4291 : }
9581 :
9582 : spdk_blob_id
9583 324 : spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
9584 : {
9585 324 : struct spdk_blob_list *snapshot_entry = NULL;
9586 324 : struct spdk_blob_list *clone_entry = NULL;
9587 :
9588 619 : TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
9589 916 : TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
9590 621 : if (clone_entry->id == blob_id) {
9591 211 : return snapshot_entry->id;
9592 : }
9593 : }
9594 : }
9595 :
9596 113 : return SPDK_BLOBID_INVALID;
9597 : }
9598 :
9599 : int
9600 246 : spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
9601 : size_t *count)
9602 : {
9603 : struct spdk_blob_list *snapshot_entry, *clone_entry;
9604 : size_t n;
9605 :
9606 246 : snapshot_entry = bs_get_snapshot_entry(bs, blobid);
9607 246 : if (snapshot_entry == NULL) {
9608 35 : *count = 0;
9609 35 : return 0;
9610 : }
9611 :
9612 211 : if (ids == NULL || *count < snapshot_entry->clone_count) {
9613 10 : *count = snapshot_entry->clone_count;
9614 10 : return -ENOMEM;
9615 : }
9616 201 : *count = snapshot_entry->clone_count;
9617 :
9618 201 : n = 0;
9619 427 : TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
9620 226 : ids[n++] = clone_entry->id;
9621 : }
9622 :
9623 201 : return 0;
9624 : }
9625 :
9626 : static void
9627 5 : bs_load_grow_continue(struct spdk_bs_load_ctx *ctx)
9628 : {
9629 : int rc;
9630 :
9631 5 : if (ctx->super->size == 0) {
9632 0 : ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
9633 : }
9634 :
9635 5 : if (ctx->super->io_unit_size == 0) {
9636 0 : ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE;
9637 : }
9638 5 : if (ctx->super->md_page_size == 0) {
9639 0 : ctx->super->md_page_size = SPDK_BS_PAGE_SIZE;
9640 : }
9641 :
9642 : /* Parse the super block */
9643 5 : ctx->bs->clean = 1;
9644 5 : ctx->bs->cluster_sz = ctx->super->cluster_size;
9645 5 : ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
9646 5 : ctx->bs->md_page_size = ctx->super->md_page_size;
9647 5 : ctx->bs->io_unit_size = ctx->super->io_unit_size;
9648 5 : bs_init_per_cluster_fields(ctx->bs);
9649 5 : rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->bs->total_clusters);
9650 5 : if (rc < 0) {
9651 0 : bs_load_ctx_fail(ctx, -ENOMEM);
9652 0 : return;
9653 : }
9654 5 : ctx->bs->md_start = ctx->super->md_start;
9655 5 : ctx->bs->md_len = ctx->super->md_len;
9656 5 : rc = spdk_bit_array_resize(&ctx->bs->open_blobids, ctx->bs->md_len);
9657 5 : if (rc < 0) {
9658 0 : bs_load_ctx_fail(ctx, -ENOMEM);
9659 0 : return;
9660 : }
9661 :
9662 15 : ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up(
9663 10 : ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
9664 5 : ctx->bs->super_blob = ctx->super->super_blob;
9665 5 : memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
9666 :
9667 5 : if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
9668 0 : SPDK_ERRLOG("Can not grow an unclean blobstore, please load it normally to clean it.\n");
9669 0 : bs_load_ctx_fail(ctx, -EIO);
9670 0 : return;
9671 : } else {
9672 5 : bs_load_read_used_pages(ctx);
9673 : }
9674 : }
9675 :
9676 : static void
9677 5 : bs_load_grow_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
9678 : {
9679 5 : struct spdk_bs_load_ctx *ctx = cb_arg;
9680 :
9681 5 : if (bserrno != 0) {
9682 0 : bs_load_ctx_fail(ctx, bserrno);
9683 0 : return;
9684 : }
9685 5 : bs_load_grow_continue(ctx);
9686 : }
9687 :
9688 : static void
9689 5 : bs_load_grow_used_clusters_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
9690 : {
9691 5 : struct spdk_bs_load_ctx *ctx = cb_arg;
9692 :
9693 5 : if (bserrno != 0) {
9694 0 : bs_load_ctx_fail(ctx, bserrno);
9695 0 : return;
9696 : }
9697 :
9698 5 : spdk_free(ctx->mask);
9699 :
9700 5 : bs_sequence_write_dev(ctx->seq, ctx->super, bs_page_to_lba(ctx->bs, 0),
9701 5 : bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
9702 : bs_load_grow_super_write_cpl, ctx);
9703 : }
9704 :
9705 : static void
9706 5 : bs_load_grow_used_clusters_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
9707 : {
9708 5 : struct spdk_bs_load_ctx *ctx = cb_arg;
9709 : uint64_t lba, lba_count;
9710 : uint64_t dev_size;
9711 : uint64_t total_clusters;
9712 :
9713 5 : if (bserrno != 0) {
9714 0 : bs_load_ctx_fail(ctx, bserrno);
9715 0 : return;
9716 : }
9717 :
9718 : /* The type must be correct */
9719 5 : assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
9720 : /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
9721 5 : assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
9722 : struct spdk_blob_md_page) * 8));
9723 5 : dev_size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
9724 5 : total_clusters = dev_size / ctx->super->cluster_size;
9725 5 : ctx->mask->length = total_clusters;
9726 :
9727 5 : lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
9728 5 : lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
9729 5 : bs_sequence_write_dev(ctx->seq, ctx->mask, lba, lba_count,
9730 : bs_load_grow_used_clusters_write_cpl, ctx);
9731 : }
9732 :
9733 : static void
9734 5 : bs_load_try_to_grow(struct spdk_bs_load_ctx *ctx)
9735 : {
9736 : uint64_t dev_size, total_clusters, used_cluster_mask_len, max_used_cluster_mask;
9737 : uint64_t lba, lba_count, mask_size;
9738 :
9739 5 : dev_size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
9740 5 : total_clusters = dev_size / ctx->super->cluster_size;
9741 5 : used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
9742 5 : spdk_divide_round_up(total_clusters, 8),
9743 5 : ctx->super->md_page_size);
9744 5 : max_used_cluster_mask = ctx->super->used_blobid_mask_start - ctx->super->used_cluster_mask_start;
9745 : /* No necessary to grow or no space to grow */
9746 5 : if (ctx->super->size >= dev_size || used_cluster_mask_len > max_used_cluster_mask) {
9747 0 : SPDK_DEBUGLOG(blob, "No grow\n");
9748 0 : bs_load_grow_continue(ctx);
9749 0 : return;
9750 : }
9751 :
9752 5 : SPDK_DEBUGLOG(blob, "Resize blobstore\n");
9753 :
9754 5 : ctx->super->size = dev_size;
9755 5 : ctx->super->used_cluster_mask_len = used_cluster_mask_len;
9756 5 : ctx->super->crc = blob_md_page_calc_crc(ctx->super);
9757 :
9758 5 : mask_size = used_cluster_mask_len * ctx->super->md_page_size;
9759 5 : ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_NUMA_ID_ANY,
9760 : SPDK_MALLOC_DMA);
9761 5 : if (!ctx->mask) {
9762 0 : bs_load_ctx_fail(ctx, -ENOMEM);
9763 0 : return;
9764 : }
9765 5 : lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
9766 5 : lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
9767 5 : bs_sequence_read_dev(ctx->seq, ctx->mask, lba, lba_count,
9768 : bs_load_grow_used_clusters_read_cpl, ctx);
9769 : }
9770 :
9771 : static void
9772 5 : bs_grow_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
9773 : {
9774 5 : struct spdk_bs_load_ctx *ctx = cb_arg;
9775 : int rc;
9776 :
9777 5 : rc = bs_super_validate(ctx->super, ctx->bs);
9778 5 : if (rc != 0) {
9779 0 : bs_load_ctx_fail(ctx, rc);
9780 0 : return;
9781 : }
9782 :
9783 5 : bs_load_try_to_grow(ctx);
9784 : }
9785 :
9786 : struct spdk_bs_grow_ctx {
9787 : struct spdk_blob_store *bs;
9788 : struct spdk_bs_super_block *super;
9789 :
9790 : struct spdk_bit_pool *new_used_clusters;
9791 : struct spdk_bs_md_mask *new_used_clusters_mask;
9792 :
9793 : spdk_bs_sequence_t *seq;
9794 : };
9795 :
9796 : static void
9797 40 : bs_grow_live_done(struct spdk_bs_grow_ctx *ctx, int bserrno)
9798 : {
9799 40 : if (bserrno != 0) {
9800 10 : spdk_bit_pool_free(&ctx->new_used_clusters);
9801 : }
9802 :
9803 40 : bs_sequence_finish(ctx->seq, bserrno);
9804 40 : free(ctx->new_used_clusters_mask);
9805 40 : spdk_free(ctx->super);
9806 40 : free(ctx);
9807 40 : }
9808 :
9809 : static void
9810 10 : bs_grow_live_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
9811 : {
9812 10 : struct spdk_bs_grow_ctx *ctx = cb_arg;
9813 10 : struct spdk_blob_store *bs = ctx->bs;
9814 : uint64_t total_clusters;
9815 :
9816 10 : if (bserrno != 0) {
9817 0 : bs_grow_live_done(ctx, bserrno);
9818 0 : return;
9819 : }
9820 :
9821 : /*
9822 : * Blobstore is not clean until unload, for now only the super block is up to date.
9823 : * This is similar to state right after blobstore init, when bs_write_used_md() didn't
9824 : * yet execute.
9825 : * When cleanly unloaded, the used md pages will be written out.
9826 : * In case of unclean shutdown, loading blobstore will go through recovery path correctly
9827 : * filling out the used_clusters with new size and writing it out.
9828 : */
9829 10 : bs->clean = 0;
9830 :
9831 : /* Reverting the super->size past this point is complex, avoid any error paths
9832 : * that require to do so. */
9833 10 : spdk_spin_lock(&bs->used_lock);
9834 :
9835 10 : total_clusters = ctx->super->size / ctx->super->cluster_size;
9836 :
9837 10 : assert(total_clusters >= spdk_bit_pool_capacity(bs->used_clusters));
9838 10 : spdk_bit_pool_store_mask(bs->used_clusters, ctx->new_used_clusters_mask);
9839 :
9840 10 : assert(total_clusters == spdk_bit_pool_capacity(ctx->new_used_clusters));
9841 10 : spdk_bit_pool_load_mask(ctx->new_used_clusters, ctx->new_used_clusters_mask);
9842 :
9843 10 : spdk_bit_pool_free(&bs->used_clusters);
9844 10 : bs->used_clusters = ctx->new_used_clusters;
9845 :
9846 10 : bs->total_clusters = total_clusters;
9847 20 : bs->total_data_clusters = bs->total_clusters - spdk_divide_round_up(
9848 10 : bs->md_start + bs->md_len, bs->pages_per_cluster);
9849 :
9850 10 : bs->num_free_clusters = spdk_bit_pool_count_free(bs->used_clusters);
9851 10 : assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
9852 10 : spdk_spin_unlock(&bs->used_lock);
9853 :
9854 10 : bs_grow_live_done(ctx, 0);
9855 : }
9856 :
9857 : static void
9858 40 : bs_grow_live_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
9859 : {
9860 40 : struct spdk_bs_grow_ctx *ctx = cb_arg;
9861 : uint64_t dev_size, total_clusters, used_cluster_mask_len, max_used_cluster_mask;
9862 : int rc;
9863 :
9864 40 : if (bserrno != 0) {
9865 0 : bs_grow_live_done(ctx, bserrno);
9866 0 : return;
9867 : }
9868 :
9869 40 : rc = bs_super_validate(ctx->super, ctx->bs);
9870 40 : if (rc != 0) {
9871 5 : bs_grow_live_done(ctx, rc);
9872 5 : return;
9873 : }
9874 :
9875 35 : dev_size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
9876 35 : total_clusters = dev_size / ctx->super->cluster_size;
9877 35 : used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
9878 35 : spdk_divide_round_up(total_clusters, 8),
9879 35 : ctx->super->md_page_size);
9880 35 : max_used_cluster_mask = ctx->super->used_blobid_mask_start - ctx->super->used_cluster_mask_start;
9881 : /* Only checking dev_size. Since it can change, but total_clusters remain the same. */
9882 35 : if (dev_size == ctx->super->size) {
9883 20 : SPDK_DEBUGLOG(blob, "No need to grow blobstore\n");
9884 20 : bs_grow_live_done(ctx, 0);
9885 20 : return;
9886 : }
9887 : /*
9888 : * Blobstore cannot be shrunk, so check before if:
9889 : * - new size of the device is smaller than size in super_block
9890 : * - new total number of clusters is smaller than used_clusters bit_pool
9891 : * - there is enough space in metadata for used_cluster_mask to be written out
9892 : */
9893 15 : if (dev_size < ctx->super->size ||
9894 15 : total_clusters < spdk_bit_pool_capacity(ctx->bs->used_clusters) ||
9895 : used_cluster_mask_len > max_used_cluster_mask) {
9896 5 : SPDK_DEBUGLOG(blob, "No space to grow blobstore\n");
9897 5 : bs_grow_live_done(ctx, -ENOSPC);
9898 5 : return;
9899 : }
9900 :
9901 10 : SPDK_DEBUGLOG(blob, "Resizing blobstore\n");
9902 :
9903 10 : ctx->new_used_clusters_mask = calloc(1, total_clusters);
9904 10 : if (!ctx->new_used_clusters_mask) {
9905 0 : bs_grow_live_done(ctx, -ENOMEM);
9906 0 : return;
9907 : }
9908 10 : ctx->new_used_clusters = spdk_bit_pool_create(total_clusters);
9909 10 : if (!ctx->new_used_clusters) {
9910 0 : bs_grow_live_done(ctx, -ENOMEM);
9911 0 : return;
9912 : }
9913 :
9914 10 : ctx->super->clean = 0;
9915 10 : ctx->super->size = dev_size;
9916 10 : ctx->super->used_cluster_mask_len = used_cluster_mask_len;
9917 10 : bs_write_super(seq, ctx->bs, ctx->super, bs_grow_live_super_write_cpl, ctx);
9918 : }
9919 :
9920 : void
9921 40 : spdk_bs_grow_live(struct spdk_blob_store *bs,
9922 : spdk_bs_op_complete cb_fn, void *cb_arg)
9923 : {
9924 : struct spdk_bs_cpl cpl;
9925 : struct spdk_bs_grow_ctx *ctx;
9926 :
9927 40 : assert(spdk_get_thread() == bs->md_thread);
9928 :
9929 40 : SPDK_DEBUGLOG(blob, "Growing blobstore on dev %p\n", bs->dev);
9930 :
9931 40 : cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
9932 40 : cpl.u.bs_basic.cb_fn = cb_fn;
9933 40 : cpl.u.bs_basic.cb_arg = cb_arg;
9934 :
9935 40 : ctx = calloc(1, sizeof(struct spdk_bs_grow_ctx));
9936 40 : if (!ctx) {
9937 0 : cb_fn(cb_arg, -ENOMEM);
9938 0 : return;
9939 : }
9940 40 : ctx->bs = bs;
9941 :
9942 40 : ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
9943 : SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
9944 40 : if (!ctx->super) {
9945 0 : free(ctx);
9946 0 : cb_fn(cb_arg, -ENOMEM);
9947 0 : return;
9948 : }
9949 :
9950 40 : ctx->seq = bs_sequence_start_bs(bs->md_channel, &cpl);
9951 40 : if (!ctx->seq) {
9952 0 : spdk_free(ctx->super);
9953 0 : free(ctx);
9954 0 : cb_fn(cb_arg, -ENOMEM);
9955 0 : return;
9956 : }
9957 :
9958 : /* Read the super block */
9959 40 : bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0),
9960 40 : bs_byte_to_lba(bs, sizeof(*ctx->super)),
9961 : bs_grow_live_load_super_cpl, ctx);
9962 : }
9963 :
9964 : void
9965 5 : spdk_bs_grow(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
9966 : spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
9967 : {
9968 : struct spdk_blob_store *bs;
9969 : struct spdk_bs_cpl cpl;
9970 : struct spdk_bs_load_ctx *ctx;
9971 5 : struct spdk_bs_opts opts = {};
9972 : int err;
9973 :
9974 5 : SPDK_DEBUGLOG(blob, "Loading blobstore from dev %p\n", dev);
9975 :
9976 5 : if ((dev->phys_blocklen % dev->blocklen) != 0) {
9977 0 : SPDK_DEBUGLOG(blob, "unsupported dev block length of %d\n", dev->blocklen);
9978 0 : dev->destroy(dev);
9979 0 : cb_fn(cb_arg, NULL, -EINVAL);
9980 0 : return;
9981 : }
9982 :
9983 5 : spdk_bs_opts_init(&opts, sizeof(opts));
9984 5 : if (o) {
9985 5 : if (bs_opts_copy(o, &opts)) {
9986 0 : return;
9987 : }
9988 : }
9989 :
9990 5 : if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
9991 0 : dev->destroy(dev);
9992 0 : cb_fn(cb_arg, NULL, -EINVAL);
9993 0 : return;
9994 : }
9995 :
9996 5 : err = bs_alloc(dev, &opts, &bs, &ctx);
9997 5 : if (err) {
9998 0 : dev->destroy(dev);
9999 0 : cb_fn(cb_arg, NULL, err);
10000 0 : return;
10001 : }
10002 :
10003 5 : cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
10004 5 : cpl.u.bs_handle.cb_fn = cb_fn;
10005 5 : cpl.u.bs_handle.cb_arg = cb_arg;
10006 5 : cpl.u.bs_handle.bs = bs;
10007 :
10008 5 : ctx->seq = bs_sequence_start_bs(bs->md_channel, &cpl);
10009 5 : if (!ctx->seq) {
10010 0 : spdk_free(ctx->super);
10011 0 : free(ctx);
10012 0 : bs_free(bs);
10013 0 : cb_fn(cb_arg, NULL, -ENOMEM);
10014 0 : return;
10015 : }
10016 :
10017 : /* Read the super block */
10018 5 : bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0),
10019 5 : bs_byte_to_lba(bs, sizeof(*ctx->super)),
10020 : bs_grow_load_super_cpl, ctx);
10021 : }
10022 :
10023 : int
10024 30 : spdk_blob_get_esnap_id(struct spdk_blob *blob, const void **id, size_t *len)
10025 : {
10026 30 : if (!blob_is_esnap_clone(blob)) {
10027 15 : return -EINVAL;
10028 : }
10029 :
10030 15 : return blob_get_xattr_value(blob, BLOB_EXTERNAL_SNAPSHOT_ID, id, len, true);
10031 : }
10032 :
10033 : struct spdk_io_channel *
10034 17482 : blob_esnap_get_io_channel(struct spdk_io_channel *ch, struct spdk_blob *blob)
10035 : {
10036 17482 : struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(ch);
10037 17482 : struct spdk_bs_dev *bs_dev = blob->back_bs_dev;
10038 17482 : struct blob_esnap_channel find = {};
10039 : struct blob_esnap_channel *esnap_channel, *existing;
10040 :
10041 17482 : find.blob_id = blob->id;
10042 17482 : esnap_channel = RB_FIND(blob_esnap_channel_tree, &bs_channel->esnap_channels, &find);
10043 17482 : if (spdk_likely(esnap_channel != NULL)) {
10044 17427 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64 ": using cached channel on thread %s\n",
10045 : blob->id, spdk_thread_get_name(spdk_get_thread()));
10046 17427 : return esnap_channel->channel;
10047 : }
10048 :
10049 55 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64 ": allocating channel on thread %s\n",
10050 : blob->id, spdk_thread_get_name(spdk_get_thread()));
10051 :
10052 55 : esnap_channel = calloc(1, sizeof(*esnap_channel));
10053 55 : if (esnap_channel == NULL) {
10054 0 : SPDK_NOTICELOG("blob 0x%" PRIx64 " channel allocation failed: no memory\n",
10055 : find.blob_id);
10056 0 : return NULL;
10057 : }
10058 55 : esnap_channel->channel = bs_dev->create_channel(bs_dev);
10059 55 : if (esnap_channel->channel == NULL) {
10060 0 : SPDK_NOTICELOG("blob 0x%" PRIx64 " back channel allocation failed\n", blob->id);
10061 0 : free(esnap_channel);
10062 0 : return NULL;
10063 : }
10064 55 : esnap_channel->blob_id = find.blob_id;
10065 55 : existing = RB_INSERT(blob_esnap_channel_tree, &bs_channel->esnap_channels, esnap_channel);
10066 55 : if (spdk_unlikely(existing != NULL)) {
10067 : /*
10068 : * This should be unreachable: all modifications to this tree happen on this thread.
10069 : */
10070 0 : SPDK_ERRLOG("blob 0x%" PRIx64 "lost race to allocate a channel\n", find.blob_id);
10071 0 : assert(false);
10072 :
10073 : bs_dev->destroy_channel(bs_dev, esnap_channel->channel);
10074 : free(esnap_channel);
10075 :
10076 : return existing->channel;
10077 : }
10078 :
10079 55 : return esnap_channel->channel;
10080 : }
10081 :
10082 : static int
10083 17452 : blob_esnap_channel_compare(struct blob_esnap_channel *c1, struct blob_esnap_channel *c2)
10084 : {
10085 17452 : return (c1->blob_id < c2->blob_id ? -1 : c1->blob_id > c2->blob_id);
10086 : }
10087 :
10088 : struct blob_esnap_destroy_ctx {
10089 : spdk_blob_op_with_handle_complete cb_fn;
10090 : void *cb_arg;
10091 : struct spdk_blob *blob;
10092 : struct spdk_bs_dev *back_bs_dev;
10093 : bool abort_io;
10094 : };
10095 :
10096 : static void
10097 170 : blob_esnap_destroy_channels_done(struct spdk_io_channel_iter *i, int status)
10098 : {
10099 170 : struct blob_esnap_destroy_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
10100 170 : struct spdk_blob *blob = ctx->blob;
10101 170 : struct spdk_blob_store *bs = blob->bs;
10102 :
10103 170 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64 ": done destroying channels for this blob\n",
10104 : blob->id);
10105 :
10106 170 : if (ctx->cb_fn != NULL) {
10107 155 : ctx->cb_fn(ctx->cb_arg, blob, status);
10108 : }
10109 170 : free(ctx);
10110 :
10111 170 : bs->esnap_channels_unloading--;
10112 170 : if (bs->esnap_channels_unloading == 0 && bs->esnap_unload_cb_fn != NULL) {
10113 5 : spdk_bs_unload(bs, bs->esnap_unload_cb_fn, bs->esnap_unload_cb_arg);
10114 : }
10115 170 : }
10116 :
10117 : static void
10118 180 : blob_esnap_destroy_one_channel(struct spdk_io_channel_iter *i)
10119 : {
10120 180 : struct blob_esnap_destroy_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
10121 180 : struct spdk_blob *blob = ctx->blob;
10122 180 : struct spdk_bs_dev *bs_dev = ctx->back_bs_dev;
10123 180 : struct spdk_io_channel *channel = spdk_io_channel_iter_get_channel(i);
10124 180 : struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(channel);
10125 : struct blob_esnap_channel *esnap_channel;
10126 180 : struct blob_esnap_channel find = {};
10127 :
10128 180 : assert(spdk_get_thread() == spdk_io_channel_get_thread(channel));
10129 :
10130 180 : find.blob_id = blob->id;
10131 180 : esnap_channel = RB_FIND(blob_esnap_channel_tree, &bs_channel->esnap_channels, &find);
10132 180 : if (esnap_channel != NULL) {
10133 15 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64 ": destroying channel on thread %s\n",
10134 : blob->id, spdk_thread_get_name(spdk_get_thread()));
10135 15 : RB_REMOVE(blob_esnap_channel_tree, &bs_channel->esnap_channels, esnap_channel);
10136 :
10137 15 : if (ctx->abort_io) {
10138 : spdk_bs_user_op_t *op, *tmp;
10139 :
10140 10 : TAILQ_FOREACH_SAFE(op, &bs_channel->queued_io, link, tmp) {
10141 0 : if (op->back_channel == esnap_channel->channel) {
10142 0 : TAILQ_REMOVE(&bs_channel->queued_io, op, link);
10143 0 : bs_user_op_abort(op, -EIO);
10144 : }
10145 : }
10146 : }
10147 :
10148 15 : bs_dev->destroy_channel(bs_dev, esnap_channel->channel);
10149 15 : free(esnap_channel);
10150 : }
10151 :
10152 180 : spdk_for_each_channel_continue(i, 0);
10153 180 : }
10154 :
10155 : /*
10156 : * Destroy the channels for a specific blob on each thread with a blobstore channel. This should be
10157 : * used when closing an esnap clone blob and after decoupling from the parent.
10158 : */
10159 : static void
10160 606 : blob_esnap_destroy_bs_dev_channels(struct spdk_blob *blob, bool abort_io,
10161 : spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
10162 : {
10163 : struct blob_esnap_destroy_ctx *ctx;
10164 :
10165 606 : if (!blob_is_esnap_clone(blob) || blob->back_bs_dev == NULL) {
10166 436 : if (cb_fn != NULL) {
10167 436 : cb_fn(cb_arg, blob, 0);
10168 : }
10169 436 : return;
10170 : }
10171 :
10172 170 : ctx = calloc(1, sizeof(*ctx));
10173 170 : if (ctx == NULL) {
10174 0 : if (cb_fn != NULL) {
10175 0 : cb_fn(cb_arg, blob, -ENOMEM);
10176 : }
10177 0 : return;
10178 : }
10179 170 : ctx->cb_fn = cb_fn;
10180 170 : ctx->cb_arg = cb_arg;
10181 170 : ctx->blob = blob;
10182 170 : ctx->back_bs_dev = blob->back_bs_dev;
10183 170 : ctx->abort_io = abort_io;
10184 :
10185 170 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64 ": destroying channels for this blob\n",
10186 : blob->id);
10187 :
10188 170 : blob->bs->esnap_channels_unloading++;
10189 170 : spdk_for_each_channel(blob->bs, blob_esnap_destroy_one_channel, ctx,
10190 : blob_esnap_destroy_channels_done);
10191 : }
10192 :
10193 : /*
10194 : * Destroy all bs_dev channels on a specific blobstore channel. This should be used when a
10195 : * bs_channel is destroyed.
10196 : */
10197 : static void
10198 1269 : blob_esnap_destroy_bs_channel(struct spdk_bs_channel *ch)
10199 : {
10200 : struct blob_esnap_channel *esnap_channel, *esnap_channel_tmp;
10201 :
10202 1269 : assert(spdk_get_thread() == spdk_io_channel_get_thread(spdk_io_channel_from_ctx(ch)));
10203 :
10204 1269 : SPDK_DEBUGLOG(blob_esnap, "destroying channels on thread %s\n",
10205 : spdk_thread_get_name(spdk_get_thread()));
10206 1309 : RB_FOREACH_SAFE(esnap_channel, blob_esnap_channel_tree, &ch->esnap_channels,
10207 : esnap_channel_tmp) {
10208 40 : SPDK_DEBUGLOG(blob_esnap, "blob 0x%" PRIx64
10209 : ": destroying one channel in thread %s\n",
10210 : esnap_channel->blob_id, spdk_thread_get_name(spdk_get_thread()));
10211 40 : RB_REMOVE(blob_esnap_channel_tree, &ch->esnap_channels, esnap_channel);
10212 40 : spdk_put_io_channel(esnap_channel->channel);
10213 40 : free(esnap_channel);
10214 : }
10215 1269 : SPDK_DEBUGLOG(blob_esnap, "done destroying channels on thread %s\n",
10216 : spdk_thread_get_name(spdk_get_thread()));
10217 1269 : }
10218 :
10219 : static void
10220 35 : blob_set_back_bs_dev_done(void *_ctx, int bserrno)
10221 : {
10222 35 : struct set_bs_dev_ctx *ctx = _ctx;
10223 :
10224 35 : if (bserrno != 0) {
10225 : /* Even though the unfreeze failed, the update may have succeed. */
10226 0 : SPDK_ERRLOG("blob 0x%" PRIx64 ": unfreeze failed with error %d\n", ctx->blob->id,
10227 : bserrno);
10228 : }
10229 35 : ctx->cb_fn(ctx->cb_arg, ctx->bserrno);
10230 35 : free(ctx);
10231 35 : }
10232 :
10233 : static void
10234 35 : blob_frozen_set_back_bs_dev(void *_ctx, struct spdk_blob *blob, int bserrno)
10235 : {
10236 35 : struct set_bs_dev_ctx *ctx = _ctx;
10237 : int rc;
10238 :
10239 35 : if (bserrno != 0) {
10240 0 : SPDK_ERRLOG("blob 0x%" PRIx64 ": failed to release old back_bs_dev with error %d\n",
10241 : blob->id, bserrno);
10242 0 : ctx->bserrno = bserrno;
10243 0 : blob_unfreeze_io(blob, blob_set_back_bs_dev_done, ctx);
10244 0 : return;
10245 : }
10246 :
10247 35 : if (blob->back_bs_dev != NULL) {
10248 35 : blob_unref_back_bs_dev(blob);
10249 : }
10250 :
10251 35 : if (ctx->parent_refs_cb_fn) {
10252 25 : rc = ctx->parent_refs_cb_fn(blob, ctx->parent_refs_cb_arg);
10253 25 : if (rc != 0) {
10254 0 : ctx->bserrno = rc;
10255 0 : blob_unfreeze_io(blob, blob_set_back_bs_dev_done, ctx);
10256 0 : return;
10257 : }
10258 : }
10259 :
10260 35 : SPDK_NOTICELOG("blob 0x%" PRIx64 ": hotplugged back_bs_dev\n", blob->id);
10261 35 : blob->back_bs_dev = ctx->back_bs_dev;
10262 35 : ctx->bserrno = 0;
10263 :
10264 35 : blob_unfreeze_io(blob, blob_set_back_bs_dev_done, ctx);
10265 : }
10266 :
10267 : static void
10268 35 : blob_set_back_bs_dev_frozen(void *_ctx, int bserrno)
10269 : {
10270 35 : struct set_bs_dev_ctx *ctx = _ctx;
10271 35 : struct spdk_blob *blob = ctx->blob;
10272 :
10273 35 : if (bserrno != 0) {
10274 0 : SPDK_ERRLOG("blob 0x%" PRIx64 ": failed to freeze with error %d\n", blob->id,
10275 : bserrno);
10276 0 : ctx->cb_fn(ctx->cb_arg, bserrno);
10277 0 : free(ctx);
10278 0 : return;
10279 : }
10280 :
10281 : /*
10282 : * This does not prevent future reads from the esnap device because any future IO will
10283 : * lazily create a new esnap IO channel.
10284 : */
10285 35 : blob_esnap_destroy_bs_dev_channels(blob, true, blob_frozen_set_back_bs_dev, ctx);
10286 : }
10287 :
10288 : void
10289 10 : spdk_blob_set_esnap_bs_dev(struct spdk_blob *blob, struct spdk_bs_dev *back_bs_dev,
10290 : spdk_blob_op_complete cb_fn, void *cb_arg)
10291 : {
10292 10 : if (!blob_is_esnap_clone(blob)) {
10293 0 : SPDK_ERRLOG("blob 0x%" PRIx64 ": not an esnap clone\n", blob->id);
10294 0 : cb_fn(cb_arg, -EINVAL);
10295 0 : return;
10296 : }
10297 :
10298 10 : blob_set_back_bs_dev(blob, back_bs_dev, NULL, NULL, cb_fn, cb_arg);
10299 : }
10300 :
10301 : struct spdk_bs_dev *
10302 5 : spdk_blob_get_esnap_bs_dev(const struct spdk_blob *blob)
10303 : {
10304 5 : if (!blob_is_esnap_clone(blob)) {
10305 0 : SPDK_ERRLOG("blob 0x%" PRIx64 ": not an esnap clone\n", blob->id);
10306 0 : return NULL;
10307 : }
10308 :
10309 5 : return blob->back_bs_dev;
10310 : }
10311 :
10312 : bool
10313 35 : spdk_blob_is_degraded(const struct spdk_blob *blob)
10314 : {
10315 35 : if (blob->bs->dev->is_degraded != NULL && blob->bs->dev->is_degraded(blob->bs->dev)) {
10316 5 : return true;
10317 : }
10318 30 : if (blob->back_bs_dev == NULL || blob->back_bs_dev->is_degraded == NULL) {
10319 15 : return false;
10320 : }
10321 :
10322 15 : return blob->back_bs_dev->is_degraded(blob->back_bs_dev);
10323 : }
10324 :
10325 3 : SPDK_LOG_REGISTER_COMPONENT(blob)
10326 3 : SPDK_LOG_REGISTER_COMPONENT(blob_esnap)
10327 :
10328 : static void
10329 0 : blob_trace(void)
10330 : {
10331 0 : struct spdk_trace_tpoint_opts opts[] = {
10332 : {
10333 : "BLOB_REQ_SET_START", TRACE_BLOB_REQ_SET_START,
10334 : OWNER_TYPE_NONE, OBJECT_BLOB_CB_ARG, 1,
10335 : {
10336 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 }
10337 : }
10338 : },
10339 : {
10340 : "BLOB_REQ_SET_COMPLETE", TRACE_BLOB_REQ_SET_COMPLETE,
10341 : OWNER_TYPE_NONE, OBJECT_BLOB_CB_ARG, 0,
10342 : {
10343 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 }
10344 : }
10345 : },
10346 : };
10347 :
10348 0 : spdk_trace_register_object(OBJECT_BLOB_CB_ARG, 'a');
10349 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
10350 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_START, OBJECT_BLOB_CB_ARG, 1);
10351 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_DONE, OBJECT_BLOB_CB_ARG, 0);
10352 0 : }
10353 3 : SPDK_TRACE_REGISTER_FN(blob_trace, "blob", TRACE_GROUP_BLOB)
|
