xref: /spdk/lib/blob/blobstore.h (revision 01a9118d0c29e76df08581c2978d7f82e2b3e508)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #ifndef SPDK_BLOBSTORE_H
35 #define SPDK_BLOBSTORE_H
36 
37 #include "spdk/assert.h"
38 #include "spdk/blob.h"
39 #include "spdk/queue.h"
40 #include "spdk/util.h"
41 
42 #include "request.h"
43 
44 /* In Memory Data Structures
45  *
46  * The following data structures exist only in memory.
47  */
48 
49 #define SPDK_BLOB_OPTS_CLUSTER_SZ (1024 * 1024)
50 #define SPDK_BLOB_OPTS_NUM_MD_PAGES UINT32_MAX
51 #define SPDK_BLOB_OPTS_MAX_MD_OPS 32
52 #define SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS 512
53 #define SPDK_BLOB_BLOBID_HIGH_BIT (1ULL << 32)
54 
55 struct spdk_xattr {
56 	/* TODO: reorder for best packing */
57 	uint32_t	index;
58 	char		*name;
59 	void		*value;
60 	uint16_t	value_len;
61 	TAILQ_ENTRY(spdk_xattr)	link;
62 };
63 
64 /* The mutable part of the blob data that is sync'd to
65  * disk. The data in here is both mutable and persistent.
66  */
67 struct spdk_blob_mut_data {
68 	/* Number of data clusters in the blob */
69 	uint64_t	num_clusters;
70 
71 	/* Array LBAs that are the beginning of a cluster, in
72 	 * the order they appear in the blob.
73 	 */
74 	uint64_t	*clusters;
75 
76 	/* The size of the clusters array. This is greater than or
77 	 * equal to 'num_clusters'.
78 	 */
79 	size_t		cluster_array_size;
80 
81 	/* Number of metadata pages */
82 	uint32_t	num_pages;
83 
84 	/* Array of page offsets into the metadata region, in
85 	 * the order of the metadata page sequence.
86 	 */
87 	uint32_t	*pages;
88 };
89 
90 enum spdk_blob_state {
91 	/* The blob in-memory version does not match the on-disk
92 	 * version.
93 	 */
94 	SPDK_BLOB_STATE_DIRTY,
95 
96 	/* The blob in memory version of the blob matches the on disk
97 	 * version.
98 	 */
99 	SPDK_BLOB_STATE_CLEAN,
100 
101 	/* The in-memory state being synchronized with the on-disk
102 	 * blob state. */
103 	SPDK_BLOB_STATE_LOADING,
104 };
105 
106 TAILQ_HEAD(spdk_xattr_tailq, spdk_xattr);
107 
108 struct spdk_blob_list {
109 	spdk_blob_id id;
110 	size_t clone_count;
111 	TAILQ_HEAD(, spdk_blob_list) clones;
112 	TAILQ_ENTRY(spdk_blob_list) link;
113 };
114 
115 struct spdk_blob {
116 	struct spdk_blob_store *bs;
117 
118 	uint32_t	open_ref;
119 
120 	spdk_blob_id	id;
121 	spdk_blob_id	parent_id;
122 
123 	enum spdk_blob_state		state;
124 
125 	/* Two copies of the mutable data. One is a version
126 	 * that matches the last known data on disk (clean).
127 	 * The other (active) is the current data. Syncing
128 	 * a blob makes the clean match the active.
129 	 */
130 	struct spdk_blob_mut_data	clean;
131 	struct spdk_blob_mut_data	active;
132 
133 	bool		invalid;
134 	bool		data_ro;
135 	bool		md_ro;
136 
137 	uint64_t	invalid_flags;
138 	uint64_t	data_ro_flags;
139 	uint64_t	md_ro_flags;
140 
141 	struct spdk_bs_dev *back_bs_dev;
142 
143 	/* TODO: The xattrs are mutable, but we don't want to be
144 	 * copying them unecessarily. Figure this out.
145 	 */
146 	struct spdk_xattr_tailq xattrs;
147 	struct spdk_xattr_tailq xattrs_internal;
148 
149 	TAILQ_ENTRY(spdk_blob) link;
150 
151 	uint32_t frozen_refcnt;
152 };
153 
154 struct spdk_blob_store {
155 	uint64_t			md_start; /* Offset from beginning of disk, in pages */
156 	uint32_t			md_len; /* Count, in pages */
157 
158 	struct spdk_io_channel		*md_channel;
159 	uint32_t			max_channel_ops;
160 
161 	struct spdk_thread		*md_thread;
162 
163 	struct spdk_bs_dev		*dev;
164 
165 	struct spdk_bit_array		*used_md_pages;
166 	struct spdk_bit_array		*used_clusters;
167 	struct spdk_bit_array		*used_blobids;
168 
169 	pthread_mutex_t			used_clusters_mutex;
170 
171 	uint32_t			cluster_sz;
172 	uint64_t			total_clusters;
173 	uint64_t			total_data_clusters;
174 	uint64_t			num_free_clusters;
175 	uint32_t			pages_per_cluster;
176 
177 	spdk_blob_id			super_blob;
178 	struct spdk_bs_type		bstype;
179 
180 	struct spdk_bs_cpl		unload_cpl;
181 	int				unload_err;
182 
183 	TAILQ_HEAD(, spdk_blob)		blobs;
184 	TAILQ_HEAD(, spdk_blob_list)	snapshots;
185 
186 	bool                            clean;
187 };
188 
189 struct spdk_bs_channel {
190 	struct spdk_bs_request_set	*req_mem;
191 	TAILQ_HEAD(, spdk_bs_request_set) reqs;
192 
193 	struct spdk_blob_store		*bs;
194 
195 	struct spdk_bs_dev		*dev;
196 	struct spdk_io_channel		*dev_channel;
197 
198 	TAILQ_HEAD(, spdk_bs_request_set) need_cluster_alloc;
199 	TAILQ_HEAD(, spdk_bs_request_set) queued_io;
200 };
201 
202 /** operation type */
203 enum spdk_blob_op_type {
204 	SPDK_BLOB_WRITE,
205 	SPDK_BLOB_READ,
206 	SPDK_BLOB_UNMAP,
207 	SPDK_BLOB_WRITE_ZEROES,
208 	SPDK_BLOB_WRITEV,
209 	SPDK_BLOB_READV,
210 };
211 
212 /* back bs_dev */
213 
214 #define BLOB_SNAPSHOT "SNAP"
215 #define SNAPSHOT_IN_PROGRESS "SNAPTMP"
216 
217 struct spdk_blob_bs_dev {
218 	struct spdk_bs_dev bs_dev;
219 	struct spdk_blob *blob;
220 };
221 
222 /* On-Disk Data Structures
223  *
224  * The following data structures exist on disk.
225  */
226 #define SPDK_BS_INITIAL_VERSION 1
227 #define SPDK_BS_VERSION 3 /* current version */
228 
229 #pragma pack(push, 1)
230 
231 #define SPDK_MD_MASK_TYPE_USED_PAGES 0
232 #define SPDK_MD_MASK_TYPE_USED_CLUSTERS 1
233 #define SPDK_MD_MASK_TYPE_USED_BLOBIDS 2
234 
235 struct spdk_bs_md_mask {
236 	uint8_t		type;
237 	uint32_t	length; /* In bits */
238 	uint8_t		mask[0];
239 };
240 
241 #define SPDK_MD_DESCRIPTOR_TYPE_PADDING 0
242 #define SPDK_MD_DESCRIPTOR_TYPE_EXTENT 1
243 #define SPDK_MD_DESCRIPTOR_TYPE_XATTR 2
244 #define SPDK_MD_DESCRIPTOR_TYPE_FLAGS 3
245 #define SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL 4
246 
247 struct spdk_blob_md_descriptor_xattr {
248 	uint8_t		type;
249 	uint32_t	length;
250 
251 	uint16_t	name_length;
252 	uint16_t	value_length;
253 
254 	char		name[0];
255 	/* String name immediately followed by string value. */
256 };
257 
258 struct spdk_blob_md_descriptor_extent {
259 	uint8_t		type;
260 	uint32_t	length;
261 
262 	struct {
263 		uint32_t        cluster_idx;
264 		uint32_t        length; /* In units of clusters */
265 	} extents[0];
266 };
267 
268 #define SPDK_BLOB_THIN_PROV (1ULL << 0)
269 #define SPDK_BLOB_INTERNAL_XATTR (1ULL << 1)
270 #define SPDK_BLOB_INVALID_FLAGS_MASK	(SPDK_BLOB_THIN_PROV | SPDK_BLOB_INTERNAL_XATTR)
271 
272 #define SPDK_BLOB_READ_ONLY (1ULL << 0)
273 #define SPDK_BLOB_DATA_RO_FLAGS_MASK	SPDK_BLOB_READ_ONLY
274 #define SPDK_BLOB_MD_RO_FLAGS_MASK	0
275 
276 struct spdk_blob_md_descriptor_flags {
277 	uint8_t		type;
278 	uint32_t	length;
279 
280 	/*
281 	 * If a flag in invalid_flags is set that the application is not aware of,
282 	 *  it will not allow the blob to be opened.
283 	 */
284 	uint64_t	invalid_flags;
285 
286 	/*
287 	 * If a flag in data_ro_flags is set that the application is not aware of,
288 	 *  allow the blob to be opened in data_read_only and md_read_only mode.
289 	 */
290 	uint64_t	data_ro_flags;
291 
292 	/*
293 	 * If a flag in md_ro_flags is set the the application is not aware of,
294 	 *  allow the blob to be opened in md_read_only mode.
295 	 */
296 	uint64_t	md_ro_flags;
297 };
298 
299 struct spdk_blob_md_descriptor {
300 	uint8_t		type;
301 	uint32_t	length;
302 };
303 
304 #define SPDK_INVALID_MD_PAGE UINT32_MAX
305 
306 struct spdk_blob_md_page {
307 	spdk_blob_id     id;
308 
309 	uint32_t        sequence_num;
310 	uint32_t	reserved0;
311 
312 	/* Descriptors here */
313 	uint8_t		descriptors[4072];
314 
315 	uint32_t	next;
316 	uint32_t	crc;
317 };
318 #define SPDK_BS_PAGE_SIZE 0x1000
319 SPDK_STATIC_ASSERT(SPDK_BS_PAGE_SIZE == sizeof(struct spdk_blob_md_page), "Invalid md page size");
320 
321 #define SPDK_BS_SUPER_BLOCK_SIG "SPDKBLOB"
322 
323 struct spdk_bs_super_block {
324 	uint8_t		signature[8];
325 	uint32_t        version;
326 	uint32_t        length;
327 	uint32_t	clean; /* If there was a clean shutdown, this is 1. */
328 	spdk_blob_id	super_blob;
329 
330 	uint32_t	cluster_size; /* In bytes */
331 
332 	uint32_t	used_page_mask_start; /* Offset from beginning of disk, in pages */
333 	uint32_t	used_page_mask_len; /* Count, in pages */
334 
335 	uint32_t	used_cluster_mask_start; /* Offset from beginning of disk, in pages */
336 	uint32_t	used_cluster_mask_len; /* Count, in pages */
337 
338 	uint32_t	md_start; /* Offset from beginning of disk, in pages */
339 	uint32_t	md_len; /* Count, in pages */
340 
341 	struct spdk_bs_type	bstype; /* blobstore type */
342 
343 	uint32_t	used_blobid_mask_start; /* Offset from beginning of disk, in pages */
344 	uint32_t	used_blobid_mask_len; /* Count, in pages */
345 
346 	uint8_t		reserved[4012];
347 	uint32_t	crc;
348 };
349 SPDK_STATIC_ASSERT(sizeof(struct spdk_bs_super_block) == 0x1000, "Invalid super block size");
350 
351 #pragma pack(pop)
352 
353 struct spdk_bs_dev *spdk_bs_create_zeroes_dev(void);
354 struct spdk_bs_dev *spdk_bs_create_blob_bs_dev(struct spdk_blob *blob);
355 
356 /* Unit Conversions
357  *
358  * The blobstore works with several different units:
359  * - Byte: Self explanatory
360  * - LBA: The logical blocks on the backing storage device.
361  * - Page: The read/write units of blobs and metadata. This is
362  *         an offset into a blob in units of 4KiB.
363  * - Cluster Index: The disk is broken into a sequential list of
364  *		    clusters. This is the offset from the beginning.
365  *
366  * NOTE: These conversions all act on simple magnitudes, not with any sort
367  *        of knowledge about the blobs themselves. For instance, converting
368  *        a page to an lba with the conversion function below simply converts
369  *        a number of pages to an equivalent number of lbas, but that
370  *        lba certainly isn't the right lba that corresponds to a page offset
371  *        for a particular blob.
372  */
373 static inline uint64_t
374 _spdk_bs_byte_to_lba(struct spdk_blob_store *bs, uint64_t length)
375 {
376 	assert(length % bs->dev->blocklen == 0);
377 
378 	return length / bs->dev->blocklen;
379 }
380 
381 static inline uint64_t
382 _spdk_bs_dev_byte_to_lba(struct spdk_bs_dev *bs_dev, uint64_t length)
383 {
384 	assert(length % bs_dev->blocklen == 0);
385 
386 	return length / bs_dev->blocklen;
387 }
388 
389 static inline uint64_t
390 _spdk_bs_lba_to_byte(struct spdk_blob_store *bs, uint64_t lba)
391 {
392 	return lba * bs->dev->blocklen;
393 }
394 
395 static inline uint64_t
396 _spdk_bs_page_to_lba(struct spdk_blob_store *bs, uint64_t page)
397 {
398 	return page * SPDK_BS_PAGE_SIZE / bs->dev->blocklen;
399 }
400 
401 static inline uint64_t
402 _spdk_bs_dev_page_to_lba(struct spdk_bs_dev *bs_dev, uint64_t page)
403 {
404 	return page * SPDK_BS_PAGE_SIZE / bs_dev->blocklen;
405 }
406 
407 static inline uint32_t
408 _spdk_bs_lba_to_page(struct spdk_blob_store *bs, uint64_t lba)
409 {
410 	uint64_t	lbas_per_page;
411 
412 	lbas_per_page = SPDK_BS_PAGE_SIZE / bs->dev->blocklen;
413 
414 	assert(lba % lbas_per_page == 0);
415 
416 	return lba / lbas_per_page;
417 }
418 
419 static inline uint64_t
420 _spdk_bs_dev_lba_to_page(struct spdk_bs_dev *bs_dev, uint64_t lba)
421 {
422 	uint64_t	lbas_per_page;
423 
424 	lbas_per_page = SPDK_BS_PAGE_SIZE / bs_dev->blocklen;
425 
426 	assert(lba % lbas_per_page == 0);
427 
428 	return lba / lbas_per_page;
429 }
430 
431 static inline uint64_t
432 _spdk_bs_cluster_to_page(struct spdk_blob_store *bs, uint32_t cluster)
433 {
434 	return cluster * bs->pages_per_cluster;
435 }
436 
437 static inline uint32_t
438 _spdk_bs_page_to_cluster(struct spdk_blob_store *bs, uint64_t page)
439 {
440 	assert(page % bs->pages_per_cluster == 0);
441 
442 	return page / bs->pages_per_cluster;
443 }
444 
445 static inline uint64_t
446 _spdk_bs_cluster_to_lba(struct spdk_blob_store *bs, uint32_t cluster)
447 {
448 	return cluster * (bs->cluster_sz / bs->dev->blocklen);
449 }
450 
451 static inline uint32_t
452 _spdk_bs_lba_to_cluster(struct spdk_blob_store *bs, uint64_t lba)
453 {
454 	assert(lba % (bs->cluster_sz / bs->dev->blocklen) == 0);
455 
456 	return lba / (bs->cluster_sz / bs->dev->blocklen);
457 }
458 
459 static inline uint64_t
460 _spdk_bs_blob_lba_to_back_dev_lba(struct spdk_blob *blob, uint64_t lba)
461 {
462 	return lba * blob->bs->dev->blocklen / blob->back_bs_dev->blocklen;
463 }
464 
465 static inline uint64_t
466 _spdk_bs_blob_lba_from_back_dev_lba(struct spdk_blob *blob, uint64_t lba)
467 {
468 	return lba * blob->back_bs_dev->blocklen / blob->bs->dev->blocklen;
469 }
470 
471 /* End basic conversions */
472 
473 static inline uint32_t
474 _spdk_bs_blobid_to_page(spdk_blob_id id)
475 {
476 	return id & 0xFFFFFFFF;
477 }
478 
479 /* The blob id is a 64 bit number. The lower 32 bits are the page_idx. The upper
480  * 32 bits are not currently used. Stick a 1 there just to catch bugs where the
481  * code assumes blob id == page_idx.
482  */
483 static inline spdk_blob_id
484 _spdk_bs_page_to_blobid(uint32_t page_idx)
485 {
486 	return SPDK_BLOB_BLOBID_HIGH_BIT | page_idx;
487 }
488 
489 /* Given a page offset into a blob, look up the LBA for the
490  * start of that page.
491  */
492 static inline uint64_t
493 _spdk_bs_blob_page_to_lba(struct spdk_blob *blob, uint32_t page)
494 {
495 	uint64_t	lba;
496 	uint32_t	pages_per_cluster;
497 
498 	pages_per_cluster = blob->bs->pages_per_cluster;
499 
500 	assert(page < blob->active.num_clusters * pages_per_cluster);
501 
502 	lba = blob->active.clusters[page / pages_per_cluster];
503 	lba += _spdk_bs_page_to_lba(blob->bs, page % pages_per_cluster);
504 
505 	return lba;
506 }
507 
508 /* Given a page offset into a blob, look up the number of pages until the
509  * next cluster boundary.
510  */
511 static inline uint32_t
512 _spdk_bs_num_pages_to_cluster_boundary(struct spdk_blob *blob, uint32_t page)
513 {
514 	uint32_t	pages_per_cluster;
515 
516 	pages_per_cluster = blob->bs->pages_per_cluster;
517 
518 	return pages_per_cluster - (page % pages_per_cluster);
519 }
520 
521 /* Given a page offset into a blob, look up the number of pages into blob to beginning of current cluster */
522 static inline uint32_t
523 _spdk_bs_page_to_cluster_start(struct spdk_blob *blob, uint32_t page)
524 {
525 	uint32_t	pages_per_cluster;
526 
527 	pages_per_cluster = blob->bs->pages_per_cluster;
528 
529 	return page - (page % pages_per_cluster);
530 }
531 
532 /* Given a page offset into a blob, look up if it is from allocated cluster. */
533 static inline bool
534 _spdk_bs_page_is_allocated(struct spdk_blob *blob, uint32_t page)
535 {
536 	uint64_t	lba;
537 	uint32_t	pages_per_cluster;
538 
539 	pages_per_cluster = blob->bs->pages_per_cluster;
540 
541 	assert(page < blob->active.num_clusters * pages_per_cluster);
542 
543 	lba = blob->active.clusters[page / pages_per_cluster];
544 
545 	if (lba == 0) {
546 		assert(spdk_blob_is_thin_provisioned(blob));
547 		return false;
548 	} else {
549 		return true;
550 	}
551 }
552 
553 #endif
554